Shinji Tsujikawa - Modified Gravity Models of Dark Energy

Modified gravity models of dark energy
Shinji Tsujikawa1
1
Department of Physics, Faculty of Science, Tokyo University of Science,
1-3, Kagurazaka, Shinjuku-ku, Tokyo 162-8601, Japan
(Dated: January 4, 2011)
We review recent progress of modified gravity models of dark energy–based on f (R) gravity,
scalar-tensor theories, braneworld gravity, Galileon gravity, and other theories. In f (R) gravity
it is possible to design viable models consistent with local gravity constraints under a chameleon
mechanism, while satisfying conditions for the cosmological viability. We also construct a class
of scalar-tensor dark energy models based on Brans-Dicke theory in the presence of a scalar-field
arXiv:1101.0191v1 [gr-qc] 31 Dec 2010
potential with a large coupling strength Q between the field and non-relativistic matter in the
Einstein frame. We study the evolution of matter density perturbations in f (R) and Brans-Dicke
theories to place observational constraints on model parameters from the power spectra of galaxy
clustering and Cosmic Microwave Background (CMB).
The Dvali-Gabadazde-Porrati braneworld model can be compatible with local gravity constraints
through a nonlinear field self-interaction φ(∂µ φ∂ µ φ) arising from a brane-bending mode, but the
self-accelerating solution contains a ghost mode in addition to the tension with the combined data
analysis of Supernovae Ia (SN Ia) and Baryon Acoustic Oscillations (BAO). The extension of the
field self-interaction to more general forms satisfying the Galilean symmetry ∂µ φ → ∂µ φ + bµ in
the flat space-time allows a possibility to avoid the appearance of ghosts and instabilities, while the
late-time cosmic acceleration can be realized by the field kinetic energy. We study observational
constraints on such Galileon models by using the data of SN Ia, BAO, and CMB shift parameters.
We also briefly review other modified gravitational models of dark energy–such as those based on
Gauss-Bonnet gravity and Lorentz-violating theories.
Contents
I. Introduction 2
II. f (R) gravity 3

A. Cosmological dynamics in f (R) gravity 4
B. Conditions for the avoidance of ghosts and tachyonic instabilities 7
C. Local gravity constraints on f (R) gravity models 9
III. Scalar-tensor gravity 12

A. Scalar-tensor theories and the matter coupling in the Einstein frame 12
B. Cosmological dynamics in Brans-Dicke theory 13
C. Local gravity constraints on Brans-Dicke theory 16
IV. DGP model 18

A. Self-accelerating solution 18
B. Observational constraints on the DGP model and other aspects of the model 20
V. Galileon gravity 22
A. Cosmology of a covariant Galileon field 22
1. Tracker solution (r1 = 1) 24
2. Solutions in the regime r1 ≪ 1 25
3. Conditions for the avoidance of ghosts and Laplacian instabilities 26
4. Observational constraints on Galileon cosmology from the background cosmic expansion history 28
B. Generalized Galileon gravity 28
VI. Other modified gravity models of dark energy 30

A. Gauss-Bonnet gravity with a scalar coupling 30
B. R/2 + f (G) gravity 30
C. Lorentz violating models 31
2
VII. Observational signatures of modified gravity 33

A. f (R) gravity 33
1. Evolution of perturbations in the regime: M 2 ≫ k 2 /a2 34
2. Evolution of perturbations in the regime: M 2 ≪ k 2 /a2 35
3. Matter power spectra and the ISW effect 35
B. Brans-Dicke theory 37
C. DGP model 40
D. Galileon gravity 42
1. Covariant Galileon gravity 42
2. Modified Galileon gravity 44
E. Observables in weak lensing 45
VIII. Conclusions 46
Acknowledgments 47
References 47
I. INTRODUCTION
The cosmic acceleration today has been supported by independent observational data such as the Supernovae-type
Ia (SN Ia) [1, 2], the Cosmic Microwave Background (CMB) temperature anisotropies measured by WMAP [3, 4],
and Baryon Acoustic Oscillations [5, 6]. The origin of dark energy responsible for this cosmic acceleration is one of
the most serious problems in modern cosmology [7–20]. The cosmological constant is one of the simplest candidates
for dark energy, but it is plagued by a severe energy scale problem if it originates from the vacuum energy appearing
in particle physics [21].
The first step toward understanding the nature of dark energy is to clarify whether it is a simple cosmological
constant or it originates from other sources that dynamically change in time. The dynamical models can be distin-
guished from the cosmological constant by considering the evolution of the equation of state of dark energy (= wDE ).
The scalar field models of dark energy such as quintessence [22–26] and k-essence [27, 28] predict a wide variety of
the variation of wDE , but still the current observational data are not sufficient to distinguish those models from the
Λ-Cold-Dark-Matter (ΛCDM) model. Moreover it is generally difficult to construct viable scalar-field models in the
framework of particle physics because of a very tiny mass (mφ . 10−33 eV) required for the cosmic acceleration today
[29, 30].
There exists another class of dynamical dark energy models based on the large-distance modification of gravity. The
models that belong to this class are f (R) gravity [31–34] (f is function of the Ricci scalar R), scalar-tensor theories
[35–40], braneworld models [41], Galileon gravity [42], Gauss-Bonnet gravity [43, 44], and so on. An attractive feature
of these models is that the cosmic acceleration can be realized without recourse to a dark energy matter component.
If we modify gravity from General Relativity (GR), however, there are tight constraints coming from local gravity
tests as well as a number of observational constraints. Hence the restriction on modified gravity models is in general
stringent compared to modified matter models (such as quintessence and k-essence).
For example, an f (R) model of the form f (R) = R − µ2(n+1) /Rn (n > 0) was proposed to explain the late-time
cosmic acceleration [32, 33] (see also Refs. [45–50] for early works). However this model suffers from a number of
problems such as the incompatibility with local gravity constraints [51–54], the instability of density perturbations
[55–58], and the absence of a matter-dominated epoch [59, 60]. As we will see in this review there are a number of
conditions required for the viability of f (R) dark energy models [55–58, 61–67], which stimulated to propose viable
models [68–72].
The simplest version of scalar-tensor theories is so-called Brans-Dicke theory in which a scalar field ϕ couples to
the Ricci scalar R with the Lagrangian density L = ϕR/2 − (ωBD /2ϕ) (∇ϕ)2 , where ωBD is a so-called Brans-Dicke
parameter [73]. GR can be recovered by taking the limit ωBD → ∞. If we allow the presence of the field potential
U (ϕ) in Brans-Dicke theory, f (R) theory in the metric formalism is equivalent to this generalized Brans-Dicke theory
with the parameter ωBD = 0 [51, 74]. By transforming the action in generalized Brans-Dicke theory (“Jordan frame”)
to an “Einstein frame” action by a conformal transformation, the theory in the Einstein frame is equivalent to a
coupled quintessence scenario [75] with a constant coupling Q satisfying the relation 1/(2Q2)√ = 3 + 2ωBD [76]. For
example, f (R) theory in the metric formalism corresponds to the constant coupling Q = −1/ 6, i.e. ωBD = 0. For
|Q| of the order of unity it is generally difficult to satisfy local gravity constraints unless some mechanism can be at
work to suppress the propagation of the fifth force between the field and non-relativistic matter. It is possible for
3
such large-coupling models to be consistent with local gravity constraints [61, 64, 68, 77–80] through the so-called
chameleon mechanism [81, 82], provided that a spherically symmetric body has a thin-shell around its surface.
A braneworld model of dark energy was proposed by Dvali, Gabadadze, and Porrati (DGP) by embedding a 3-
brane in the 5-dimensional Minkowski bulk spacetime [41]. In this scenario the gravitational leakage to the extra
dimension leads to a self-acceleration of the Universe on the 3-brane. Moreover a longitudinal graviton (i.e. a brane-
bending mode φ) gives rise to a nonlinear self-interaction of the form (rc2 /mpl ) φ(∂ µ φ∂µ φ) through the mixing with a
transverse graviton, where rc is a cross-over scale (of the order of the Hubble radius H0−1 today) and mpl is the Planck
mass [83, 84]. In the local region where the energy density ρ is much larger than rc−2 m2pl the nonlinear self-interaction
can lead to the decoupling of the field from matter through the so-called Vainshtein mechanism [85], which allows a
possibility for the consistency with local gravity constraints. However the DGP model suffers from a ghost problem
[86–88], in addition to the difficulty for satisfying the combined observational constraints of SN Ia and BAO [89–94].
The equations of motion following from the self-interacting Lagrangian φ(∂ µ φ∂µ φ) present in the DGP model are
invariant under the Galilean shift ∂µ φ → ∂µ φ + bµ in the Minkowski background. While the DGP model is plagued by
the ghost problem, the extension of the field self-interaction to more general forms satisfying the Galilean symmetry
may allow us to avoid the appearance of ghosts. Nicolis et al. [42] showed that there are only five field Lagrangians
Li (i = 1, · · · , 5) that respect the Galilean symmetry in the Minkowski background. In Refs. [95, 96] these terms were
extended to covariant forms in the curved space-time. In addition one can keep the equations of motion up to the
second-order, while recovering the Galileon Lagrangian in the limit of the Minkowski space-time. This property is
welcome to avoid the appearance of an extra degree of freedom associated with ghosts. In fact, Refs. [97, 98] derived
the viable model parameter space in which the appearance of ghosts and instabilities associated with scalar and tensor
perturbations can be avoided. Moreover the late-time cosmic acceleration is realized by the existence of a stable de
Sitter solution. We shall review the cosmological dynamics of Galileon gravity as well as conditions for the avoidance
of ghosts and instabilities.
In order to distinguish between different models of dark energy based on modified gravitational theories, it is
important to study the evolution of cosmological perturbations as well as the background expansion history of the
Universe. In particular, the modified growth of matter perturbations δm relative to the ΛCDM model changes the
matter power spectrum of large-scale structures (LSS) as well as the weak lensing spectrum [99]-[126]. Moreover the
modification of gravity manifests itself for the evolution of the effective gravitational potential ψ related with the
Integrated-Sachs-Wolfe (ISW) effect in CMB anisotropies. We shall review a number of observational signatures for
the modified gravitational models of dark energy.
This review is organized as follows. In Sec. II we construct viable dark energy models based on f (R) theories after
discussing conditions for the cosmological viability as well as for the consistency with local gravity tests. In Sec. III we
show that, in Brans-Dicke theories with large matter couplings, it is possible to design the field potential consistent
with both cosmological and local gravity constraints. In Sec. IV we derive the field equations in the DGP model
and confront the model with observations at the background level. In Sec. V we review the cosmological dynamics
based on Galileon gravity as well as conditions for the avoidance of ghosts and Laplacian instabilities. In Sec. VI we
briefly mention other modified gravity models of dark energy based on Gauss-Bonnet gravity and Lorentz-violating
theories. In Sec. VII we study observational signatures of dark energy models based on f (R) gravity, Brans-Dicke
theory, DGP model, and Galileon gravity, in order to confront them with the observations of LSS, CMB, and weak
lensing. Sec. VIII is devoted to conclusions.
Throughout the review we use the units such that c = ~ = kB = 1, where c is the speed of light, ~ is reduced
Planck’s constant, and kB is Boltzmann’s constant. We also adopt the metric signature (−, +, +, +).
II. f (R) GRAVITY
We start with the action in f (R) gravity:
1 4 √
Z Z
S = 2 d x −gf (R) + d4 x LM (gµν , ΨM ) , (1)
2κ
where κ2 = 8πG (G is a bare gravitational constant), g is a determinant of the metric gµν , f (R) is an arbitrary
function in terms of the Ricci scalar R, and LM is a matter action with matter fields ΨM . Variation of the action (1)
with respect to gµν leads to the following field equation
1
F (R)Rµν (g) − f (R)gµν − ∇µ ∇ν F (R) + gµν F (R) = κ2 Tµν , (2)
2
4
√
where F (R) ≡ f,R = ∂f /∂R, Rµν is a Ricci tensor, and Tµν = −(2/ −g)δLM /δg µν is an energy-momentum tensor
of matter. The trace of Eq. (2) gives
3 F (R) + F (R)R − 2f (R) = κ2 T , (3)
where T = g µν Tµν = −ρ + 3P . Here ρ and P are the energy density and the pressure of matter, respectively.
Regarding the variation of the action (1), there is another approach called the Palatini formalism [127] in which gµν
and the affine connection Γα 1
βγ are treated as independent variables. The resulting field equations are second-order
[132–137] and the cosmological dynamics of dark energy models have been studied by a number of authors [138–145].
However f (R) theory in the Palatini formalism gives rise to a large coupling between a scalar field degree of freedom
and ordinary matter [135–137, 146–150], which implies difficulty for compatibility with standard models of particle
physics. This large coupling also leads to significant growth of matter density perturbations, unless the models are
very close to the ΛCDM model [151–155].
In the following we focus on the variational approach (so called the metric formalism) given above. The Einstein
gravity without a cosmological constant corresponds to f (R) = R and F (R) = 1, so that the term F (R) in Eq. (3)
vanishes. Since in this case R = −κ2 T = κ2 (ρ − 3P ), the Ricci scalar R is directly determined by matter. In f (R)
gravity with a non-linear term in R, F (R) does not vanish in Eq. (3). Hence there is a propagating scalar degree of
freedom, ψ ≡ F (R), dubbed “scalaron” in Ref. [156]. The trace equation (3) allows the dynamics of the scalar field
ψ.
The de Sitter point corresponds to a vacuum solution with constant R. Since F (R) = 0 at this point, we obtain
F (R)R − 2f (R) = 0 . (4)
Since the quadratic model f (R) = αR2 satisfies this condition, it gives rise to an exact de Sitter solution. In the
inflation model f (R) = R + αR2 proposed by Starobinsky [156], the accelerated cosmic expansion ends when the term
αR2 becomes smaller than the linear term R. It is possible to construct such f (R) inflation models in the framework
of supergravity [157, 158].
A. Cosmological dynamics in f (R) gravity
We first study cosmological dynamics for the models based on f (R) theories in the metric formalism. In order to
derive conditions for the cosmological viability of f (R) models we shall carry out general analysis without specifying
the form of f (R). We consider a flat Friedmann-Lemaître-Robertson-Walker (FLRW) background with the line
element
ds2 = −dt2 + a(t)2 dx2 , (5)
where a(t) is a scale factor. For the matter Lagrangian LM in Eq. (1) we take into account non-relativistic matter and
radiation, whose energy densities ρm and ρr satisfy the usual continuity equations ρ̇m + 3Hρm = 0 and ρ̇r + 4Hρr = 0
respectively. Here H ≡ ȧ/a is the Hubble parameter and a dot represents a derivative with respect to cosmic time t.
From Eqs. (2) and (3) we obtain
3F H 2 = κ2 (ρm + ρr ) + (F R − f )/2 − 3H Ḟ , (6)

2
2F Ḣ = −κ [ρm + (4/3)ρr ] − F̈ + H Ḟ , (7)
where the Ricci scalar is given by
R = 6(2H 2 + Ḣ) . (8)
Let us introduce the following dimensionless variables:
Ḟ f R κ 2 ρr
x1 ≡ − , x2 ≡ − , x3 ≡ , x4 ≡ , (9)
HF 6F H 2 6H 2 3F H 2
1 We also note that there is another approach for the variational principle–known as the metric-affine formalism–in which the matter
Lagrangian LM depends not only on the metric gµν but also on the connection Γα βγ [128–131].
5
together with the density parameters
κ 2 ρm
Ωm ≡ = 1 − x1 − x2 − x3 − x4 , Ωr ≡ x4 , ΩDE ≡ x1 + x2 + x3 . (10)
3F H 2
It is straightforward to derive the following equations [62]:
dx1
= −1 − x3 − 3x2 + x21 − x1 x3 + x4 , (11)
dN
dx2 x1 x3
= − x2 (2x3 − 4 − x1 ) , (12)
dN m
dx3 x1 x3
= − − 2x3 (x3 − 2) , (13)
dN m
dx4
= −2x3 x4 + x1 x4 , (14)
dN
where N = ln a and
d ln F Rf,RR
m ≡ = , (15)
d ln R f,R
d ln f Rf,R x3
r ≡ − =− = . (16)
d ln R f x2
From Eq. (16) one can express R as a function of x3 /x2 . Since m is a function of R, it follows that m is a function
of r, i.e., m = m(r). The ΛCDM model, f (R) = R − 2Λ, corresponds to m = 0. Then the quantity m characterizes
the deviation from the ΛCDM model.
The effective equation of state of the system is given by
2Ḣ 1
weff ≡ −1 − 2
= − (2x3 − 1) . (17)
3H 3
In the absence of radiation (x4 = 0) the fixed points for the dynamical system (11)-(14) are
P1 : (x1 , x2 , x3 ) = (0, −1, 2), Ωm = 0, weff = −1 , (18)

P2 : (x1 , x2 , x3 ) = (−1, 0, 0), Ωm = 2, weff = 1/3 , (19)
P3 : (x1 , x2 , x3 ) = (1, 0, 0), Ωm = 0, weff = 1/3 , (20)
P4 : (x1 , x2 , x3 ) = (−4, 5, 0), Ωm = 0, weff = 1/3 , (21)

3m 1 + 4m 1 + 4m m(7 + 10m) m
P5 : (x1 , x2 , x3 ) = ,− , , Ωm = 1 − , weff = − , (22)
1 + m 2(1 + m)2 2(1 + m) 2(1 + m)2 1+m
2 − 5m − 6m2

2(1 − m) 1 − 4m (1 − 4m)(1 + m)
P6 : (x1 , x2 , x3 ) = , ,− , Ωm = 0, weff = . (23)
1 + 2m m(1 + 2m) m(1 + 2m) 3m(1 + 2m)
The points P5 and P6 are on the line m(r) = −r − 1 in the (r, m) plane.
Only the point P5 can be responsible for the matter-dominated epoch (Ωm ≃ 1 and weff ≃ 0). This is realized
provided m is close to 0. In the (r, m) plane the matter point P5 exists around (r, m) = (−1, 0). Either the point P1
or P6 can lead to the late-time cosmic acceleration. The former corresponds to a de Sitter point (weff = −1) with
r = −2, in which case the condition (4) is satisfied. Depending on the values of m, the point P6 can be responsible
for the cosmic acceleration [62]. In the following we shall focus on the case in which the matter point P5 is followed
by the de Sitter point P1 .
The stability of the fixed points is known by considering small perturbations δxi (i = 1, 2, 3) around them [62]. For
the point P5 the eigenvalues for the 3 × 3 Jacobian matrix of perturbations are
p
′ −3m5 ± m5 (256m35 + 160m25 − 31m5 − 16)
3(1 + m5 ), , (24)
4m5 (m5 + 1)
where m5 ≡ m(r5 ) and m′5 ≡ dm dr (r5 ) with r5 ≈ −1. In the limit |m5 | ≪ 1 the latter two eigenvalues reduce to
p
−3/4 ± −1/m5. The f (R) models with m5 < 0 show a divergence of the eigenvalues as m5 → −0, in which case
the system cannot remain for a long time around the point P5 . For example the model f (R) = R − α/Rn with n > 0
6
1.2
1.0
PB
.
0.80
m = - r -1
(iv)
0.60
m
r=-2
. . .
0.40
PA (iii)
0.20
(ii)
.
0.0
(i) PM
-0.20
-2.2 -2 -1.8 -1.6 -1.4 -1.2 -1 -0.8
r
Figure 1: Four trajectories in the (r, m) plane. Each trajectory corresponds to the models: (i) ΛCDM, (ii) f (R) = (Rb − Λ)c ,
(iii) f (R) = R − αRn with α > 0, 0 < n < 1, and (iv) m(r) = −C(r + 1)(r 2 + ar + b). Here PM , PA and PB are the matter
point P5 , the de Sitter point P1 , and the accelerated point P6 , respectively. From Ref. [64].
and α > 0 falls into this category. On the other hand, if 0 < m5 < 0.327, the latter two eigenvalues in Eq. (24) are
complex with negative real parts. Then, provided that m′5 > −1, the point P5 corresponds to a saddle point with
a damped oscillation. Hence the Universe can evolve toward the point P5 from the radiation era and leave for the
late-time acceleration. Then the condition for the existence of the saddle matter era is
dm
m(r) ≈ +0 , > −1 , at r = −1 . (25)
dr
The first condition implies that the f (R) models need to be close to the ΛCDM model during the matter era.
The eigenvalues for the Jacobian matrix of perturbations about the point P1 are
p
3 25 − 16/m1
− 3, − ± , (26)
2 2
where m1 = m(r = −2). This shows that the condition for the stability of the de Sitter point P1 is [62, 159–161]
0 < m(r = −2) ≤ 1 . (27)
The trajectories that start from the saddle matter point P5 with the condition (25) and then approach the stable de
Sitter point P1 with the condition (27) are cosmologically viable.
Let us consider a couple of viable f (R) models in the (r, m) plane. The ΛCDM model, f (R) = R−2Λ, corresponds to
m = 0, in which case the trajectory is a straight line from P5 : (r, m) = (−1, 0) to P1 : (r, m) = (−2, 0). The trajectory
(ii) in Fig. 1 represents the model f (R) = (Rb −Λ)c [64], which corresponds to the straight line m(r) = [(1−c)/c]r+b−1
in the (r, m) plane. The existence of a saddle matter epoch requires the condition c ≥ 1 and bc ≈ 1. The trajectory
(iii) represents the model [62, 63]
f (R) = R − αRn (α > 0, 0 < n < 1), (28)

2
which corresponds to the curve m = n(1 + r)/r. The trajectory (iv) in Fig. 1 shows
√ the model m(r) = −C(r + 1)(r +
ar + b), in which case the late-time accelerated attractor is the point P6 with ( 3 − 1)/2 < m < 1.
In Ref. [62] it was shown that the variable m needs to be close to 0 during the radiation-dominated epoch as well.
Hence the viable f (R) models are close to the ΛCDM model, f (R) = R − 2Λ, in the region R ≫ R0 (where R0 is
the present cosmological Ricci scalar). The Ricci scalar R given in Eq. (8) remains positive from the radiation era to
the present epoch, as long as the it does not oscillate. As we will see in Sec. II B, we require the condition f,R > 0
7
to avoid ghosts. Then the condition m > 0 for the presence of the matter-dominated epoch translates to f,RR > 0.
The model f (R) = R − α/Rn (α > 0, n > 0) is not viable because the condition f,RR > 0 is violated. We also note
that the power-law models with f (R) = Rn do not give rise to a successful cosmological trajectory [59, 60] (unlike
the claims in Ref. [162]).
In order to derive the equation of state of dark energy to confront with SN Ia observations for the cosmologically
viable models, we rewrite Eqs. (6) and (7) as follows2 :
3AH 2 = κ2 (ρm + ρr + ρDE ) , (29)

2
−2AḢ = κ [ρm + (4/3)ρr + ρDE + PDE ] , (30)
where A is some constant and
κ2 ρDE ≡ (1/2)(F R − f ) − 3H Ḟ + 3H 2 (A − F ) , (31)

2 2
κ PDE ≡ F̈ + 2H Ḟ − (1/2)(F R − f ) − (3H + 2Ḣ)(A − F ) . (32)
Defining ρDE and PDE in this way, one can show that these satisfy the usual continuity equation
ρ̇DE + 3H(ρDE + PDE ) = 0 . (33)
The dark energy equation of state related with SN Ia observations is given by wDE ≡ PDE /ρDE . From Eqs. (29)
and (30) it follows that
2AḢ + 3AH 2 + κ2 ρr /3 weff

wDE = − 2 2
≃ , (34)
3AH − κ (ρm + ρr ) 1 − (F/A)Ωm
where the last approximate equality in Eq. (34) is valid in the regime where the radiation density ρr is negligible
relative to the matter density. The viable f (R) models approach the ΛCDM model in the past, i.e. F → 1 as R → ∞.
In order to reproduce the standard matter era for the redshifts z ≫ 1, one can choose A = 1 in Eqs. (29) and (30).
Another possible choice is A = F0 , where F0 is the present value of F . This choice is suitable if the deviation of F0
from 1 is small (as in the scalar-tensor theory with a massless scalar field [164–166]). In both cases the equation of
state wDE can be smaller than −1 before reaching the de Sitter attractor [64, 71, 167–169]. Thus f (R) gravity models
give rise to a phantom equation of state without violating stability conditions of the system.
As we see in Eq. (34), the presence of non-relativistic matter is important to lead to the apparent phantom behavior.
We wish to stress here that for viable f (R) models constructed to satisfy all required conditions [such as the models
(45), (46), and (48) we will discuss later] the ghosts are not present even if wDE < −1. A number of authors proposed
some models to realize wDE < −1 without including non-relativistic matter [170, 171], which means that wDE = weff
from Eq. (34). However, such f (R) models usually imply the presence of ghosts3 , because weff < −1 corresponds to
Ḣ > 0.
The observational constraints on specific f (R) models have been carried out in Refs. [172–175] from the background
expansion history of the Universe (see also Refs. [176–180] for the reconstruction of f (R) models from observations).
Since the deviation of wDE from that in the ΛCDM model (wDE = −1) is not so significant [68, 167], the viable models
such as (45), (46), and (48) can be consistent with the data fairly easily. In other words we do not obtain very tight
bounds on model parameters from the information of the background expansion history only. However, the models
can be more strongly constrained at the level of perturbations, as we will see in Sec. VII A.
B. Conditions for the avoidance of ghosts and tachyonic instabilities
In this subsection we shall derive conditions for the avoidance of ghosts and tachyonic instabilities in f (R) theories.
In doing so we expand the action (1) up to the second-order by considering the following perturbed metric about the
FLRW background
ds2 = −(1 + 2α)dt2 − 2a(t)∂i βdtdxi + a2 (t) (δij + 2ψδij + 2∂i ∂j γ) dxi dxj , (35)
2 If the field equations are written in this form, we can also show that the background cosmological dynamics has a correspondence with
equilibrium thermodynamics on the apparent horizon [163].
3 If a late-time de Sitter solution is a stable spiral, it happens that wDE oscillates around −1 with a small amplitude, even for viable f (R)
models. Here we are discussing the real ghosts out of this regime.
8
where α, β, ψ, γ are scalar metric perturbations [181].

Introducing the perturbation δF for the quantity F = ∂f /∂R, one can construct the gauge-invariant curvature
perturbation
H
R≡ψ− δF . (36)
Ḟ
Expanding the action (1) without the matter source, we obtain the second-order action for the curvature perturbation
[19, 182]

1 2 1 1
Z
(2) 3 3 2
δS = dt d x a Qs Ṙ − (∇R) , (37)
2 2 a2
where
3Ḟ 2
Qs ≡ . (38)
2κ2 F [H + Ḟ /(2F )]2
The negative sign of Qs corresponds to a ghost field because of the negative kinetic energy. Hence the condition for
the avoidance of ghosts is given by
F > 0. (39)
For the matter sector the ghost does not appear for ρM (1 + wM )/wM > 0 (where wM is the equation of state for the
matter fluid) [183], which is satisfied for radiation (wM = 1/3) and non-relativistic matter (wM ≃ +0).
If Qs is √
positive, the action (37) can be written in the following form by introducing the new variables u = zs R
and zs = a Qs :
1 zs′′ 2

1 ′2 1
Z
(2) 3 2
δS = dη d x u − (∇u) + u , (40)
2 2 2 zs
R
where a prime represents a derivative with respect to the conformal time τ = a−1 dt. Equation (40) shows that the
scalar degree of freedom has the effective mass
1 zs′′ Q̇2s Q̈s 3H Q̇s

Ms2 ≡ − 2
= 2
− −
a zs 4Qs 2Qs 2Qs
!
2
72F 2 H 4 1 288H 3 − 12HR 1 f,RR Ṙ2 7
= − + F + + − 24H 2 + R , (41)
(2F H + f,RR Ṙ)2 3 2F H + f,RR Ṙ f ,RR 4F 2 6
where we have eliminated the term Ḣ by using the background equations.

In Fourier space the perturbation u satisfies the equation of motion
u′′ + k 2 + Ms2 a2 u = 0 .

(42)
For k 2 /a2 ≫ Ms2 , the propagation speed cs of the field u is equivalent to the speed of light c. Hence, in f (R) gravity,
the gradient instability associated with negative c2s is absent. For small k satisfying k 2 /a2 ≪ Ms2 , we require that
Ms2 > 0 to avoid the tachyonic instability of perturbations. The viable dark energy models based on f (R) theories
need to satisfy the condition Rf,RR ≪ F (i.e. m = Rf,RR /f,R ≪ 1) at early cosmological epochs in order to have
successful cosmological evolution from radiation domination till matter domination. At these epochs the mass squared
is approximately given by
F
Ms2 ≃ . (43)
3f,RR
Under the no-ghost condition (39) the tachyonic instability is absent for
f,RR > 0 . (44)
The viable f (R) dark energy models have been constructed to satisfy the conditions (39) and (44) in the regime
R ≥ R1 , where R1 is the Ricci scalar at the late-time de Sitter point. Moreover we require that the models are
9
consistent with the conditions (25) and (27). The model (28) can be consistent with all these conditions, but the
local gravity constraints demand that the variable m is very much smaller than 1 in the regions of high density (i.e.
R ≫ R0 , where R0 is the cosmological Ricci scalar today). In the model (28) one has m ≃ n(−r − 1) around r ≃ −1.
For the consistency with local gravity constraints we require that n . 10−10 [78], but in this case the deviation from
the ΛCDM model around the present epoch (R ≈ R0 ) is very small.
If the variable m behaves as m = C(−r − 1)p with p > 1 in the region R ≫ R0 , then it is possible to satisfy local
gravity constraints (i.e. m ≪ 0.01 ∼ 0.1 for R ≫ R0 ) while at the same time showing deviations from the ΛCDM
(m & 0.01 ∼ 0.1 for R ≃ R0 ). The models constructed in this vein are
(R/Rc )2n
(A) f (R) = R − µRc with µ > 0, n > 0 and Rc > 0 , (45)
(R/Rc )2n + 1
h −n i
(B) f (R) = R − µRc 1 − 1 + R2 /Rc2 with µ > 0, n > 0 and Rc > 0 , (46)
which were proposed by Hu and Sawicki [68] and Starobinsky [69], respectively. Rc is roughly of the order of the
present cosmological Ricci scalar R0 for µ and n of the order of unity. The models (A) and (B) asymptotically behave
as
f (R) ≃ R − µRc [1 − (R2 /Rc2 )−n ] for R ≫ Rc , (47)
which gives m(r) = C(−r − 1)2n+1 .

Another viable model that leads to the even rapid decrease of m toward the past is [71]
(C) f (R) = R − µRc tanh (R/Rc ) with µ > 0, Rc > 0 . (48)
Other similar models were proposed by Appleby and Battye [70] and Linder [72].
In what follows we shall discuss local gravity constraints on the above models.
C. Local gravity constraints on f (R) gravity models
Let us proceed to discuss local gravity constraints on f (R) gravity models. In the region of high density like Earth
or Sun, the Ricci scalar R is much larger than the background cosmological value R0 . In this case the linear expansion
of R = R0 + δR cannot be justified. In such a non-linear regime the chameleon mechanism [81, 82] plays an important
role for the f (R) models to satisfy local gravity constraints [61, 64, 68, 77–80] (see also Refs. [184–190]).
To discuss the chameleon mechanism in f (R) gravity, it is convenient to transform the action (1) to the so-called
Einstein frame action via the conformal transformation [191]:
g̃µν = Ω2 gµν , Ω2 = F . (49)
The action in the Einstein frame includespa linear term in R̃, where the tilde represents quantities in the Einstein
frame. Introducing a new scalar field φ = 3/2κ2 ln F , we obtain the action in the Einstein frame, as [19, 191]
Z
1 1 µν
Z
4
R̃ − g̃ ∂µ φ∂ν φ − V (φ) + d4 x LM (F −1 g̃µν , ΨM ) ,
p
SE = d x −g̃ (50)
2κ2 2
where
RF − f
V (φ) = . (51)
2κ2 F 2
In the Einstein frame the scalar field
√φ directly couples with non-relativistic matter. The strength of this coupling
depends on the conformal factor Ω = F . We define the coupling Q as
Ω,φ F,φ 1
Q≡− =− = −√ , (52)
Ω 2F 6
which is of the order of unity in f (R) gravity. If the field potential V (φ) is absent, the field propagates freely with a
large coupling Q. Since a potential (51) with a gravitational origin is present in f (R) gravity, it is possible for f (R)
dark energy models to satisfy local gravity constraints through the chameleon mechanism [61, 68, 78, 80].
10
In a spherically symmetric space-time under a weak gravitational background (i.e. neglecting the backreaction of
gravitational potentials), variation of the action (50) with respect to the scalar field φ leads to
d2 φ 2 dφ dVeff
+ = , (53)
dr̃2 r̃ dr̃ dφ
where r̃ is a distance from the center of symmetry, and Veff (φ) is an effective potential defined by
Veff (φ) = V (φ) + eQφ ρ∗ . (54)
Here ρ∗ is a conserved quantity in the Einstein frame, which is related to the density ρ̃ in the Jordan frame via the
relation ρ∗ = e3Qφ ρ̃. By the end of this section we use the unit κ2 = 1.
We assume that a spherically symmetric body has a constant density ρ∗ = ρA inside the body (r̃ < r̃c ) and that
the density outside the body (r̃ > r̃c ) is ρ∗ = ρB . The mass Mc of the body and the gravitational potential Φc at
the radius r̃c are given by Mc = (4π/3)r̃c3 ρA and Φc = Mc /8πr̃c , respectively. The effective potential Veff (φ) has two
′ ′
minima at the field values φA and φB satisfying Veff (φA ) = 0 and Veff (φB ) = 0, respectively (here a prime represents a
derivative with respect to φ). The former corresponds to the region with a high density that gives rise to a large mass
squared m2A ≡ Veff ′′
(φA ), whereas the latter to the lower density region with a smaller mass squared m2B ≡ Veff ′′
(φB ).
When the “dynamics” of the field φ with the field equation (53) is studied, we need to consider the effective potential
(−Veff ) so that it has two maxima at φ = φA and φ = φB .
We impose the two boundary conditions (dφ/dr̃)(r̃ = 0) = 0 and φ(r̃ → ∞) = φB . The field φ is at rest at r̃ = 0
and begins to roll down the potential when the matter-coupling term QρA eQφ becomes important at a radius r̃1 in
Eq. (53). As long as r̃1 is close to r̃c such that ∆r̃c ≡ r̃c − r̃1 ≪ r̃c , the body has a thin-shell inside the body. The field
acquires a sufficient kinetic energy in the thin-shell regime (r̃1 < r̃ < r̃c ) and hence the field climbs up the potential
hill outside the body (r̃ > r̃c ).
The field profile can be obtained by matching the solutions of Eq. (53) at the radius r̃ = r̃1 and r̃ = r̃c . Neglecting
the mass term mB , the thin-shell field profile outside the body is given by [192]
GMc
φ(r̃) = φB − 2Qeff , (55)
r̃
where
φB − φA
Qeff ≃ 3Qǫth , ǫth ≡ . (56)
6QΦc
Here ǫth is called a thin-shell parameter. Under the conditions ∆r̃c /r̃c ≪ 1 and 1/(mA r̃c ) ≪ 1, the thin-shell
parameter is approximately given by [192]
∆r̃c 1
ǫth ≃ + . (57)
r̃c mA r̃c
Provided that ǫth ≪ 1, the amplitude
√ of the effective coupling Qeff can be much smaller than 1. It is then possible
for the f (R) models (|Q| = 1/ 6) to be consistent with local gravity experiments. Originally the thin-shell solution
was derived by assuming that the field is frozen in the region 0 < r̃ < r̃1 [81, 82]. In this case the thin-shell parameter
is given by ǫth ≃ ∆r̃c /r̃c , which is different from Eq. (57). However, this difference is not important because the
condition ∆r̃c /r̃c ≫ 1/(mA r̃c ) is satisfied for most of viable models [192].
Consider the bound on the thin-shell parameter from the possible violation of equivalence principle (EP). The
tightest bound comes from the solar system tests of weak EP using the free-fall acceleration of Moon (aMoon ) and
Earth (a⊕ ) toward Sun [82]. The experimental bound on the difference of two accelerations is given by [193]
|aMoon − a⊕ |
< 10−13 . (58)
(aMoon + a⊕ )/2
Provided that Earth, Sun, and Moon have thin-shells, the field profiles outside the bodies are given by Eq. (55) with
the replacement of corresponding quantities. The acceleration induced by a fifth force with the field profile φ(r̃) and
the effective coupling Qeff is afifth = |Qeff ∇φ(r̃)|. Using the thin-shell parameter ǫth,⊕ for Earth, the accelerations a⊕
and aMoon toward Sun (mass M⊙ ) are
Φ2⊕

GM⊙ 2 2 Φ⊕ GM⊙ 2 2
a⊕ ≃ 1 + 18Q ǫth,⊕ , aMoon ≃ 1 + 18Q ǫth,⊕ , (59)
r̃2 Φ⊙ r̃2 Φ⊙ ΦMoon
11
where Φ⊙ ≃ 2.1 × 10−6 , Φ⊕ ≃ 7.0 × 10−10 , and ΦMoon ≃ 3.1 × 10−11 are the gravitational potentials of Sun, Earth
and Moon, respectively. Then the condition (58) translates to
ǫth,⊕ < 8.8 × 10−7 /|Q| . (60)
Since the condition |φB | ≫ |φA | is satisfied for viable f (R) models (as we will see below), we have ǫth,⊕ ≃ φB /(6QΦ⊕ )
from Eq. (56). Hence the condition (60) corresponds to
|φB,⊕ | < 3.7 × 10−15 . (61)
Let us consider local gravity constraints on the f (R) models given in Eqs. (45) and (46). In the region of high
density where local gravity experiments are carried out, it is sufficient to use the asymptotic form given in Eq. (47).
In order for these models to be responsible for the present cosmic acceleration, Rc is roughly the same order as the
cosmological Ricci scalar R0 today for µ and n of the order of unity. For the functional form (47) we have the following
relations
√
F = e2φ/ 6
= 1 − 2nµ(R/Rc )−(2n+1) , (62)
" 2n/(2n+1) #
√ √

1 −4φ/ 6 −φ
Veff (φ) ≃ µRc e 1 − (2n + 1) √ + ρ∗ e−φ/ 6 . (63)
2 6nµ
Inside and outside the body the effective potential (63) has minima at
√ √
φA ≃ − 6nµ(Rc /ρA )2n+1 , φB ≃ − 6nµ(Rc /ρB )2n+1 . (64)
If ρA ≫ ρB , then one has |φB | ≫ |φA |.

The bound (61) translates into
2n+1
nµ R1
< 1.5 × 10−15 . (65)
x2n+1
1 ρB
Here x1 is defined by x1 ≡ R1 /Rc , where R1 is the Ricci scalar at the late-time de Sitter fixed point P1 given in
Eq. (18). Let us consider the model described by the Lagrangian density (47) for R ≥ R1 . If we use the models (45)
and (46), then there are some modifications for the estimation of R1 . However this change is not significant when we
place constraints on model parameters.
The de Sitter solution for the model (47) satisfies µ = x2n+1
1 /[2(x2n
1 − n − 1)]. Substituting this relation into
Eq. (65), it follows that
2n+1
n R1
< 1.5 × 10−15 . (66)
2(x2n
1 − n − 1) ρB
For the stability of the de Sitter point we require that m(R1 ) < 1, which translates into the condition x2n 2
1 > 2n +3n+1.
2n
Hence the term n/[2(x1 − n − 1)] in Eq. (66) is smaller than 0.25 for n > 0.
Let us use the simple approximation that R1 and ρB are of the orders of the present cosmological density 10−29
g/cm3 and the baryonic/dark matter density 10−24 g/cm3 in our galaxy, respectively. From Eq. (66) we obtain the
constraint [78]
n > 0.9 . (67)
Thus n is not required to be much larger than unity. Under the condition (67), as R decreases to the order of Rc ,
one can cosmologically see an appreciable deviation from the ΛCDM model. The deviation from the ΛCDM model
appears when R decreases to the order of Rc . The model (48) also shows similar behavior. If we consider the model
(28), it was shown in Ref. [78] that the bound (61) gives the constraint n < 3 × 10−10 . Hence the deviation from the
ΛCDM model is very small. The models (45) and (46) are carefully constructed to satisfy local gravity constraints,
while at the same time the deviation from the ΛCDM model appears even for n = O(1). Note that the model (48)
can easily satisfy local gravity constraints because of the rapid approach to the ΛCDM in the regime R ≫ Rc .
In the strong gravitational background (such as neutron stars), Kobayashi and Maeda [194, 195] pointed out that
for the model (46) it is difficult to obtain thin-shell solutions inside a spherically symmetric body with constant
density. For chameleon models with general couplings Q, a thin-shell field profile was analytically derived in Ref. [196]
by employing a linear expansion in terms of the gravitational potential Φc at the surface of a compact object with
12
constant density. Using the boundary condition set by analytic solutions, Ref. [196] also numerically confirmed the
existence of thin-shell solutions for Φc . 0.3 in the case of inverse power-law potentials V (φ) = M 4+n φ−n . Ref. [197]
also showed that static relativistic stars with constant density exists for the model (46). The effect of the relativistic
pressure is important around the center of the body, so that the field tends to roll down the potential quickly unless
the boundary condition is carefully chosen. Realistic stars have densities ρA (r) that globally decrease as a function of
r. The numerical simulation of Refs. [198, 199] showed that thin-shell solutions are present for the f (R) model (46)
by considering a polytropic equation of state even in the strong gravitational background (see also Ref. [200]).
III. SCALAR-TENSOR GRAVITY
There is another class of modified gravity called scalar-tensor theories in which the Ricci scalar R is coupled to a
scalar field ϕ. One of the simplest examples is the so-called Brans-Dicke theory with the action
√
Z
1 ωBD
Z
S = d4 x −g ϕR − (∇ϕ)2 − U (ϕ) + d4 x LM (gµν , ΨM ) , (68)
2 2ϕ
where ωBD is a constant (called the Brans-Dicke parameter), U (ϕ) is a field potential, and LM is a matter Lagrangian
that depends on the metric gµν and matter fields Ψm . The original Brans-Dicke theory [73] does not have the field
potential. As we will see below, metric f (R) gravity discussed in Sec. II is equivalent to the Brans-Dicke theory with
ωBD = 0.
A. Scalar-tensor theories and the matter coupling in the Einstein frame
The general action for scalar-tensor theories can be written as
4 √
Z
1 1
Z
S = d x −g f (ϕ, R) − ζ(ϕ)(∇ϕ) + d4 x LM (gµν , ΨM ) ,
2
(69)
2 2
where f depends on the scalar field ϕ and the Ricci scalar R, ζ is a function of ϕ. We choose the unit κ2 = 8πG = 1.
The action (69) covers a wide variety of theories such as f (R) gravity (f (ϕ, R) = f (R), ζ = 0), Brans-Dicke theory
(f = ϕR and ζ = ωBD /ϕ), and dilaton gravity (f = e−ϕ R and ζ = −e−ϕ ).
Let us consider theories of the type
f (ϕ, R) = F (ϕ)R − 2U (ϕ) . (70)
In order to avoid the appearance
√ of ghosts we require that F (ϕ) > 0. Under the conformal transformation (49) with
the conformal factor Ω = F , the action (69) can be transformed to that in the Einstein frame:
Z
1 1 ˜ 2
Z p
SE = d4 x −g̃ R̃ − (∇φ) − V (φ) + d4 x LM (g̃µν F −1 (φ), ΨM ) , (71)
2 2
where
V = U/F 2 . (72)
We have introduced a new scalar field φ in order to make the kinetic term canonical:
s 2
3 F,ϕ ζ
Z
φ ≡ dϕ + , (73)
2 F F
We define the coupling between dark energy and non-relativistic matter in the Einstein frame:
" 2 #−1/2
F,φ F,ϕ 3 F,ϕ ζ
Q≡− =− + . (74)
2F F 2 F F
√
Recall that in metric f (R) gravity we have that Q = −1/ 6. If Q is a constant, the following relations hold from
Eqs. (73) and (74):
2
−2Qφ 2 dφ
F =e , ζ = (1 − 6Q )F . (75)
dϕ
13
Then the action (69) in the Jordan frame can be written as [76]
" # Z
4 √ 1 1
Z
S = d x −g F (φ)R − (1 − 6Q )F (φ)(∇φ) − U (φ) + d4 x LM (gµν , ΨM ) .
2 2
(76)
2 2
In the limit that Q → 0, the action (76) reduces to the one for a minimally coupled scalar field φ with the potential
U (φ). The transformation of the Jordan frame action (76) via a conformal transformation g̃µν = F (φ)gµν gives rise
to the Einstein frame action (71) with a constant coupling Q. The action (71) is equivalent to the action (50) with
g̃µν = e−2Qφ gµν .
One can compare (76) with the action (68) in Brans-Dicke theory. Setting ϕ = F = e−2Qφ , one finds that two
actions are equivalent if the parameter ωBD is related to Q via the relation [76, 82]
1
3 + 2ωBD = . (77)
2Q2
Using this relation, we find
√ that the General Relativistic limit (ωBD → ∞) corresponds to the vanishing coupling
(Q → 0). Since Q = −1/ 6 in metric f (R) gravity, this corresponds to the Brans-Dicke parameter ωBD = 0 [51, 74].
The experimental bound on ωBD for a massless scalar field is given by ωBD > 40000 [193, 201], which translates into
the condition
|Q| < 2.5 × 10−3 (for a massless field). (78)
In such cases it is difficult to find a large difference relative to the uncoupled quintessence model. In the presence of
the field potential, however, it is possible for large coupling models (|Q| ∼ 1) to satisfy local gravity constraints via
the chameleon mechanism [76].
The above Brans-Dicke theory is one of the examples in scalar-tensor theories. In general the coupling Q is field-
dependent apart from Brans-Dicke theory. If we consider a nonminimally coupled scalar field with F (ϕ) = 1 − ξϕ2
and ζ(ϕ) = 1, then it follows that Q(φ) = ξφ/[1 − ξφ2 (1 − 6ξ)]1/2 . The cosmological dynamics in such a theory have
been studied by a number of authors [35–40, 202–205, 207–211]. If the field is nearly massless during most of the
cosmological epochs, the coupling Q needs to be suppressed to avoid the propagation of the fifth force.
In the following we shall study the cosmological dynamics and local gravity constraints on the constant coupling
models based on the action (76) with F (φ) = e−2Qφ .
B. Cosmological dynamics in Brans-Dicke theory
We study the cosmological dynamics for the Jordan frame action (76) in the presence of a non-relativistic fluid with
energy density ρm and a radiation fluid with energy density ρr . We regard the Jordan frame as a physical frame due
to the usual conservation of non-relativistic matter (ρm ∝ a−3 ). In the flat FLRW background variation of the action
(76) with respect to gµν and φ gives the following equations of motion
3F H 2 = (1 − 6Q2 )F φ̇2 /2 + U − 3H Ḟ + ρm + ρr , (79)
2 2
2F Ḣ = −(1 − 6Q )F φ̇ − F̈ + H Ḟ − ρm − (4/3)ρr , (80)
2
(1 − 6Q )F [φ̈ + 3H φ̇ + Ḟ /(2F )φ̇] + U,φ + QF R = 0 . (81)
Let us introduce the following variables
r r
φ̇ 1 U 1 ρr
x1 ≡ √ , x2 ≡ , x3 ≡ , (82)
6H H 3F H 3F
and
ρm √
Ωm ≡ 2
, Ωrad ≡ x23 , ΩDE ≡ (1 − 6Q2 )x21 + x22 + 2 6Qx1 . (83)
3F H
These satisfy the relation Ωm + Ωrad + ΩDE = 1 from Eq. (79). Using Eqs. (79)-(81), we obtain the differential
equations for x1 , x2 and x3 :
√ √
dx1 6 2
√ 6Q h √ i Ḣ
= (λx2 − 6x1 ) + (5 − 6Q2 )x21 + 2 6Qx1 − 3x22 + x23 − 1 − x1 2 , (84)
dN √2 2 H
dx2 6 Ḣ
= (2Q − λ)x1 x2 − x2 2 , (85)
dN 2 H
dx3 √ Ḣ
= 6Qx1 x3 − 2x3 − x3 2 , (86)
dN H
14
where N = ln a, λ = −U,φ /U , and
Ḣ 1 − 6Q2 2 2 2 2 2
√
= − 3 + 3x1 − 3x2 + x3 − 6Q x1 + 2 6Qx1 + 3Q(λx22 − 4Q) . (87)
H2 2
The effective equation of state of the system is given by by weff = −1 − 2Ḣ/(3H 2 ).

If λ is a constant, one can derive the fixed points of the system (84)-(86) in the absence of radiation (x3 = 0) [76]:
• (a)
√ !
6Q 3 − 2Q2 4Q2
(x1 , x2 ) = ,0 , Ωm = , weff = . (88)
3(2Q2 − 1) 3(1 − 2Q2 )2 3(1 − 2Q2 )
• (b)
√
1 3 ∓ 6Q
(x1 , x2 ) = √ ,0 , Ωm = 0 , weff = √ . (89)
6Q ± 1 3(1 ± 6Q)
• (c)
√ 1/2 !
6 − λ2 + 8Qλ − 16Q2 20Q2 − 9Qλ − 3 + λ2

6(4Q − λ)
(x1 , x2 ) = , , Ωm = 0 , weff = − .
6(4Q2 − Qλ − 1) 6(4Q2 − Qλ − 1)2 3(4Q2 − Qλ − 1)
(90)
• (d)
√ r !
6 3 + 2Qλ − 6Q2 3 − 12Q2 + 7Qλ 2Q
(x1 , x2 ) = , , Ωm = 1 − , weff = − . (91)
2λ 2λ2 λ2 λ
• (e)
(x1 , x2 ) = (0, 1) , Ωm = 0 , weff = −1 . (92)
The point (e) corresponds to the de Sitter point, which exists only for λ = 4Q [this can be confirmed by setting φ̇ = 0
in Eqs. (79)-(81)].
We first study the case of non-zero values of Q with constant λ, i.e. for the exponential potential U (φ) = U0 e−λφ .
We do not consider the special case of λ = 4Q. The matter-dominated era can be realized either by the point (a) or
by the point (d). If the point (a) is responsible for the matter era, the condition Q2 ≪ 1 is required. We then have
Ωm ≃ 1 + 10Q2/3 > 1 and weff ≃ 4Q2 /3. √When Q2 ≪ 1 √ the scalar-field dominated point (c) yields an accelerated
expansion of the Universe provided that − 2 + 4Q < λ < 2 + 4Q. Under these conditions the point (a) is followed
by the late-time cosmic acceleration. The scaling solution (d) can give rise to the equation of state, weff ≃ 0 for
|Q| ≪ |λ|. In this case, however, the condition weff < −1/3 for the point (c) gives λ2 < 2. Then the energy fraction of
the pressureless matter for the point (d) does not satisfy the condition Ωm ≃ 1. From the above discussion the viable
cosmological trajectory for constant λ corresponds
√ to the sequence
√ from the point (a) to the scalar-field dominated
2
point (c) under the conditions Q ≪ 1 and − 2 + 4Q < λ < 2 + 4Q.
We shall proceed to the case where λ varies with time. The fixed points derived above for constant λ can be
regarded as the “instantaneous” fixed points, provided that the time scale of the variation of λ is smaller than that of
the cosmic expansion. The matter era can be realized by the point (d) with |Q| ≪ |λ|. The solutions finally approach
either the de Sitter point (e) with λ = 4Q or the accelerated point (c).
In the following we focus on the case in which the matter era with the point (d) is followed by the accelerated
epoch with the de Sitter solution (e). To study the stability of the point (e) we define a variable x4 ≡ F , satisfying
the following equation
dx4 √
= −2 6Qx1 x4 . (93)
dN
15
Considering the 3 × 3 matrix for perturbations δx1 , δx2 and δx4 around the point (e), we obtain the eigenvalues
" r #
3 8 dλ
− 3 , − 1 ± 1 − F1 Q (F1 ) , (94)
2 3 dF
where F1 ≡ F (φ1 ) is the value of F at the de Sitter point with the field value φ1 . Since F1 > 0, we find that the de
Sitter point is stable under the condition
dλ dλ
Q (F1 ) ≥ 0 , i.e., (φ1 ) ≤ 0 . (95)
dF dφ
Let us consider the f (R) model (47) in which the models (45) and (46) are recovered in the regime R ≫ Rc . Since
√
2φ/ 6
e = 1 − 2nµ(R/Rc )−(2n+1) , the potential U = (F R − f )/2 is given by
√ 2n/(2n+1)

µRc 2n + 1
2φ/ 6
U (φ) = 1− 1 − e , (96)
2 (2nµ)2n/(2n+1)
In this case the slope of the potential, λ = −U,φ /U , is

√
4ne2φ/ 6 √ −2n/(2n+1) √ −1/(2n+1)

2n + 1
2φ/ 6 2φ/ 6
λ = −√ 1− 1 − e 1 − e . (97)
6(2nµ)2n/(2n+1) (2nµ)2n/(2n+1)
In the deep matter-dominated epoch during which the condition R/Rc ≫ 1 is satisfied, the field φ is very close to
zero. For n and µ of the order of unity, we have |λ| ≫ 1 at this stage. Hence the matter era can be realized by the
instantaneous fixed point
√ (d). As R/Rc gets smaller, |λ| decreases to the order of unity. If the solutions reach the
point λ = 4Q = −4/ 6 and satisfy the stability condition dλ/dF ≤ 0, then the final attractor corresponds to the de
Sitter fixed point (e).
For the theories with general couplings Q, it is possible to construct a scalar-field potential that is the generalization
of (51). One example is [76]
U (φ) = U0 1 − C(1 − e−2Qφ )p

(U0 > 0, C > 0, 0 < p < 1) . (98)
√
The f (R) model (47) corresponds to Q = −1/ 6 and p = 2n/(2n + 1). The slope of the potential is given by
2Cp Qe−2Qφ (1 − e−2Qφ )p−1

λ= . (99)
1 − C(1 − e−2Qφ )p
We have U (φ) → U0 for φ → 0 and U (φ) → U0 (1 − C) in the limits φ → ∞ (for Q > 0) and φ → −∞ (for Q < 0).
The field is nearly frozen around the value φ = 0 during the deep radiation and matter epochs. In these epochs we
have R ≃ ρm /F from Eqs. (79)-(81) by noting that U0 is negligibly small compared to ρm or ρr . Using Eq. (81), it
follows that U,φ + Qρm ≃ 0. Hence, in the high-curvature region, the field φ evolves along the instantaneous minima
given by
1/(1−p)
1 2U0 pC
φm ≃ . (100)
2Q ρm
The field value |φm | increases for decreasing ρm . As long as the condition ρm ≫ 2U0 pC is satisfied, we have |φm | ≪ 1
from Eq. (100).
For field values around φ = 0 one has |λ| ≫ 1 from Eq. (99). Hence the instantaneous fixed point (d) can be
responsible for the matter-dominated epoch provided that |Q| ≪ |λ|. The variable F = e−2Qφ decreases in time
irrespective of the sign of the coupling Q and hence 0 < F < 1. The de Sitter solution corresponds to λ = 4Q, that is
2
C= . (101)
(1 − F1 )p−1 [2 + (p − 2)F1 ]
This solution is present as long as the solution of this equation exists in the region 0 < F1 < 1.
From Eq. (99) the derivative of λ with respect to φ is
dλ 4CpQ2 F (1 − F )p−2 [1 − pF − C(1 − F )p ]

=− . (102)
dφ [1 − C(1 − F )p ]2
16
WDE
1.0
Wm
0.50
Wrad
0.0
-0.50
w eff
-1.0
0 2 4 6 8 10
N
Figure 2: The evolution of ΩDE , Ωm , Ωrad and weff in Brans-Dicke theory with the potential (98). The model parameters are
Q = 0.01, p = 0.2 and C = 0.7 with the initial conditions x1 = 0, x2 = 2.27 × 10−7 , x3 = 0.7, and x4 − 1 = −5.0 × 10−13 . From
Ref. [76].
The de Sitter point is stable under the condition 1 − pF1 > C(1 − F1 )p . Using Eq. (101) this translates into
F1 > 1/(2 − p) . (103)
When 0 < C < 1 one can show that dλ/dφ < 0 is always satisfied. Hence the solutions approach the de Sitter
attractor after the end of the matter era. When C > 1, the de Sitter point is stable under the condition (103). If this
condition is violated, the solutions choose another stable fixed point [such as the point (c)] as an attractor.
The above discussion shows that, if 0 < C < 1, the matter point (d) can be followed by the stable de Sitter solution
(e). In Fig. 2 we plot the evolution of ΩDE , Ωm , Ωrad , and weff for Q = 0.01, p = 0.2 and C = 0.7. This shows that
the viable cosmological trajectory can be realized for the potential (98). In order to confront with SN Ia observations,
it is possible to rewrite Eqs. (79) and (80) in the forms of Eqs. (29) and (30) by defining the dark energy density
ρDE and the pressure PDE in the similar way. It was shown in Ref. [76] that the phantom equation of state as well
as the cosmological constant boundary crossing can be realized for the field potentials U (φ) satisfying local gravity
constraints.
C. Local gravity constraints on Brans-Dicke theory
We study local gravity constraints on Brans-Dicke theory described by the action (76). In the absence of the
potential U (φ) we already mentioned that the Brans-Dicke parameter ωBD is constrained to be ωBD > 4.0 × 104 from
solar-system experiments. This gives the upper bound (78) on the coupling Q between the field φ and non-relativistic
matter in the Einstein frame. This bound also applies to the case of a nearly massless field with the potential U (φ) in
which the Yukawa correction e−Mr is close to unity (where M is the scalar field mass and r is an interaction length).
In the presence of the field-potential it is possible for large coupling models (|Q| ∼ 1) to satisfy local gravity
constraints provided that the mass M of the field φ is sufficiently large in the region of high density. In fact,
the potential (98) is designed to have a large mass in the high-density region, so that it can be compatible with
experimental tests of gravity through the chameleon mechanism. In the following we study the model (98) and derive
the conditions under which local gravity constraints can be satisfied. If we make a conformal transformation for the
action (98), the action in the Einstein frame is given by (76) with F (φ) = e−2Qφ . We can use the results obtained in
Sec. II C, because thin-shell solutions have been derived for the general coupling Q.
17
As in the case of f (R) gravity, we consider a configuration in which a spherically symmetric body has a constant
density ρA inside the body and that the density outside the body is given by ρ = ρB (≪ ρA ). Under the condition
|Qφ| ≪ 1, we have V,φ ≃ −2U0 QpC(2Qφ)p−1 for the potential V = U/F 2 in the Einstein frame. Then the field values
at the potential minima inside and outside the body are
1/(1−p) 1/(1−p)
1 2U0 p C 1 2U0 p C
φA ≃ , φB ≃ . (104)
2Q ρA 2Q ρB
In order to realize the accelerated expansion today, the energy scale U0 is required to be the same order as the square
of the present Hubble parameter H0 , i.e. U0 ∼ H02 ∼ ρ0 , where ρ0 ≃ 10−29 g/cm3 is the cosmological density today.
The baryonic/dark matter density in our galaxy corresponds to ρB ≃ 10−24 g/cm3 . Hence the conditions |QφA | ≪ 1
and |QφB | ≪ 1 are in fact satisfied unless C ≫ 1. The field mass squared m2A ≡ V,φφ at φ = φA is approximately
given by
(2−p)/(1−p)
1−p ρA
m2A ≃ Q2 U0 , (105)
(2p pC)1/(1−p) U0
which means that mA can be much larger than H0 because of the condition ρA ≫ U0 . This large mass allows the
chameleon mechanism to work, because the condition 1/(mA r̃c ) ≪ 1 is satisfied.
The bound (61) coming from the violation of equivalence principle in the solar system translates into
1/(1−p)
(2U0 pC/ρB ) < 7.4 × 10−15 |Q| . (106)
Let us consider the case in which the solutions finally approach the de Sitter solution (e). At the point (e), one has
3F1 H12 = U0 [1 − C(1 − F1 )p ] with C given in Eq. (101). Hence we get the following relation
U0 = 3H12 [2 + (p − 2)F1 ] /p . (107)
Plugging this into Eq. (106), it follows that

1/(1−p)
(R1 /ρB ) (1 − F1 ) < 7.4 × 10−15 |Q| , (108)
where R1 = 12H12 is the Ricci scalar at the de Sitter point. Since the term (1 − F1 ) is smaller than 1/2 from the
condition (103), we obtain the inequality (R1 /ρB )1/(1−p) < 1.5 × 10−14 |Q|. Using the values R1 = 10−29 g/cm3 and
ρB = 10−24 g/cm3 , we obtain the following bound
5
p>1− . (109)
13.8 − log10 |Q|
When |Q| = 10−1 and |Q| = 1 we have p > 0.66 and p > 0.64, respectively. Thus the model can be compatible with
local gravity experiments even for |Q| = O(1).
In Ref. [79] it was shown that in order to satisfy both local gravity and cosmological constraints the chameleon
potentials in the Einstein frame need to be of the form V (φ) = M 4 [1 + f (φ)], where the function f (φ) is smaller
than 1 today and M is a mass that corresponds to the dark energy scale (M ∼ 10−12 GeV). The potential V (φ) =
M 4 exp[µ(M/φ)n ] [82, 212] is one of those viable candidates, but the allowed model parameter space is severely
constrained by the 2006 Eöt-Wash experiment [213]. Unless the parameter µ is unnaturally small (µ . 10−5 ), this
potential is incompatible with local gravity constraints for {n, Q} = O(1).
On the other hand, the chameleon potential V (φ) = V0 [1 − µ(1 − e−φ )n ] (0 < n < 1) can satisfy both local gravity
and cosmological constraints4. In Ref. [79] this potential is consistent with the constraint coming from 2006 Eöt-
Wash experiments as well as the WMAP bound on the variation of the field-dependent mass [214] for natural model
parameters.
4 In the Einstein frame the potential (98) takes the form V (φ) = U/F 2 = U0 e4Qφ 1 − C(1 − e−2Qφ )p , so in the region |Qφ| ≪ 1 this

n
potential is similar to V (φ) = V0 [1 − µ(1 − e ) ].
−φ
18
IV. DGP MODEL
In this section we review braneworld models of dark energy motivated by string theory. In braneworlds standard
model particles are confined on a 3-dimensional (3D) brane embedded in 5-dimensional bulk with large extra dimen-
sions [215, 216]. Dvali, Gabadadze, and Porrati (DGP) [41] proposed a braneworld model in which the 3-brane is
embedded in a Minkowski bulk with infinitely large extra dimensions. One can recover Newton’s law by adding a 4D
Einstein-Hilbert action sourced by the brane curvature to the 5D action [217]. The presence of such a 4D term may
be induced by quantum corrections coming from the bulk gravity and its coupling with matter on the brane. In the
DGP model the standard 4D gravity is recovered at small distances, whereas the effect from the 5D gravity manifests
itself for large distances. Interestingly one can realize the self cosmic acceleration without introducing a dark energy
component [218, 219] (see also Ref. [220]).
A. Self-accelerating solution
The action of the DGP model is given by

1 1 √
Z Z Z
d5 X −g̃ R̃ + 2 d4 X −gR − d5 X −g̃ LM ,
p p
S= 2 (110)
2κ(5) 2κ(4)
where g̃AB is the metric in the 5D bulk and gµν = ∂µ X A ∂ν X B g̃AB is the induced metric on the brane with X A (xc )
being the coordinates of an event on the brane labelled by xc . The 5D and 4D gravitational constants, κ2(5) and κ2(4) ,
are related with the 5D and 4D Planck masses, M(5) and M(4) , via
κ2(5) = 1/M(5)
3
, κ2(4) = 1/M(4)
2
. (111)
The first and second terms in Eq. (110) correspond to Einstein-Hilbert actions in the 5D bulk and on the brane,
respectively. There is no contribution to the Lagrangian LM from the bulk because we are considering √ a Minkowski
bulk. Then the matter action consists of a brane-localized matter whose action is given by d4 x −g (σ + Lbrane
R
M ),
where σ is the 3-brane tension and Lbrane
M is the Lagrangian density on the brane. Since the tension is unrelated to
the Ricci scalar R, it can be adjusted to be zero (as we do in the following).
In order to study the cosmological dynamics on the brane (located at y = 0), we take a metric of the form:
ds2 = −n2 (τ, y)dτ 2 + a2 (τ, y) γij dxi dxj + dy 2 , (112)
where γij represents a maximally symmetric space-time with a constant curvature K. The 5D Einstein equations are
1
G̃AB ≡ R̃AB − R̃ g̃AB = κ2(5) T̃AB , (113)
2
(brane)
where R̃AB is the 5D Ricci tensor, T̃AB is the sum of the energy momentum tensor TAB on the brane and the
contribution ŨAB coming from the scalar curvature of the brane:
(brane)
T̃AB = TAB + ŨAB . (114)
(brane)
Since we are considering a homogeneous and isotropic Universe on the brane, one can write TBA in the form
(brane)
TBA = δ(y) diag(−ρM , PM , PM , PM , 0) , (115)
where ρM and PM are functions of τ only. The non-vanishing components coming from the Ricci scalar R of the
brane are
2
n2

3 ȧ
Ũ00 = − 2 + K 2 δ(y) , (116)
κ(4) a2 a
2 2
1 a ȧ ȧ ṅ ä
Ũij = − 2 − 2 +2 −2 − K γij δ(y) , (117)
κ(4) n2 a an a
19
where a dot represents a derivative with respect to τ . The non-vanishing components of the 5D Einstein tensor G̃AB
are [218, 221, 222]
2
n2
′′
a′2

ȧ a
G̃00 = 3 2 − n2 + 2 +K 2 , (118)
a a a a
a2
′′
n′′ a′2 a′ n′ ä a′2

a ȧṅ
G̃ij = a2 2 + + 2 +2 + 2 −2 − 2 + 2 − K γij , (119)
a n a an n a a an
′
ȧ′

ȧn
G̃05 = 3 − , (120)
an a
3 ä ȧ2
′2
a′ n′

a ȧṅ K
G̃55 = 3 + − + − −3 2 , (121)
a2 an n2 a a2 an a
where a prime represents a derivative with respect to y.

Assuming no flow of matter along the 5-th dimensions, we have T̃05 = 0 and hence G̃05 = 0. Then Eqs. (118) and
(121) can be written as
3n2 ′ 3 ˙
G̃00 = − I , G̃55 = − I, (122)
2a3 a′ 2a3 ȧ
where
(ȧa)2
I ≡ (a′ a)2 − − Ka2 . (123)
n2
Since we are considering the Minkowski bulk, we have G̃00 = 0 and G̃55 = 0 locally in the bulk. This gives I ′ = 0 and
I˙ = 0. Integrations of these equations lead to
(ȧa)2
(a′ a)2 − − Ka2 + C = 0 , (124)
n2
where C is a constant independent of τ and y.
We shall find solutions of the Einstein equations (113) in the vicinity of y = 0. The metric needs to be continuous
across the brane in order to have a well-defined geometry. However, its derivatives with respect to y can be discon-
tinuous at y = 0. The Einstein tensor is made of the metric up to the second derivatives with respect to y, so the
Einstein equations with a distributional source are written in the form [218, 221, 222]
g ′′ = T δ(y) , (125)
where δ(y) is a Dirac’s delta function. Integrating this equation across the brane gives
[g ′ ] = T , where [g ′ ] ≡ g ′ (0+ ) − g ′ (0− ) . (126)
The jump of the first derivative of the metric is equivalent to the energy-momentum tensor on the brane.
Equations (118) and (119) include the derivatives a′′ and n′′ of the metric. Integrating the Einstein equations
G̃00 = κ2(5) T̃00 and G̃ij = κ2(5) T̃ij across the brane, we obtain
[a′ ] κ2(5) κ2(5) ȧ2b n2b

=− ρM + 2 2 +K 2 , (127)
ab 3 κ(4) nb a2b ab
[n′ ] κ2(5) κ2(5) ȧ2b n2b

ȧb ṅb äb
= (3PM + 2ρM ) − 2 2 +2 −2 +K 2 , (128)
nb 3 κ(4) nb a2b ab nb ab ab
where the subscript “b” represents the quantities on the brane.

We assume the symmetry y ↔ −y, in which case [a′ ] = 2a′ (0+ ) and [n′ ] = 2n′ (0+ ). Substituting Eq. (127) into
Eq. (124), we obtain the modified Friedmann equation on the brane:
s
κ2(5) κ2(5)

2
K C 2 K
ǫ H + 2− 4 = 2 H + 2 − ρM , (129)
ab ab 2κ(4) ab 6
20
where H ≡ ȧb /(ab nb ) is the Hubble parameter and ǫ = ±1 is the sign of [a′ ]. The constant C can be interpreted as
the term coming from the 5D bulk Weyl tensor [218, 219, 223]. Since the Weyl tensor vanishes for the Minkowski
bulk, we set C = 0 in the following discussion. We introduce a length scale
κ2(5) 2
M(4)
rc ≡ = . (130)
2κ2(4) 3
2M(5)
Then Eq. (129) can be written as
κ2(4)
r
ǫ K 2 K
H2 + = H + − ρM , (131)
rc a2 a2 3
where we have omitted the subscript “b” for the quantities at y = 0.
Plugging the junction conditions (127) and (128) into the (05) component of the Einstein equations, G̃05 = 0, the
following matter conservation equation holds on the brane:
dρM
+ 3H(ρM + PM ) = 0 , (132)
dt
where t is the cosmic time related to the time τ via the relation dt = nb dτ . If the equation of state, wM = PM /ρM ,
is specified, the cosmological evolution is known by solving Eqs. (131) and (132).
For a flat geometry (K = 0), Eq. (131) reduces to
2 ǫ κ2(4)
H − H= ρM . (133)
rc 3
If the crossover scale rc is much larger than the Hubble radius H −1 , the first term in Eq. (133) dominates over the
second one. In this case the standard Friedmann equation, H 2 = κ2(4) ρM /3, is recovered. On the other hand, in the
regime rc < H −1 , the presence of the second term in Eq. (133) leads to a modification to the standard Friedmann
equation. In the Universe dominated by non-relativistic matter (ρM ∝ a−3 ), the Universe approaches a de Sitter
solution for the branch ǫ = +1:
H → HdS = 1/rc . (134)
We can realize the cosmic acceleration today provided that rc is of the order of the present Hubble radius H0−1 .
B. Observational constraints on the DGP model and other aspects of the model
Equation (131) can be written as

s 2
κ 2
K (4) 1 1
H2 + 2 =  ρM + 2 +  . (135)
a 3 4rc 2rc
For the matter on the brane, we consider non-relativistic matter with the energy density ρm and the equation of state
(0)
wm = 0. We then have ρm = ρm (1 + z)3 from Eq. (132). Let us introduce the following density parameters
(0)
(0) K 1 κ2(4) ρm
ΩK = − , Ω(0)
rc = , Ω(0)
m = . (136)
a0 H02
2 4rc H02
2 3H02
Then Eq. (135) reads

" q q 2 #
2 (0) (0) (0) (0)
H (z) = H02 ΩK (1 2
+ z) + Ωm (1 + z)3 + Ωrc + Ωrc . (137)
The normalization condition at z = 0 is given by

q q
(0) (0) (0)
Ω(0)
m + Ω K + 2 1 − Ω K Ωr c = 1 . (138)
21
0.225
0.2
0.175
0.15
Wrc
0.125
0.1
0.075
0.1 0.2 0.3 0.4 0.5

Wm
Figure 3: Observational constraints on the DGP model from the SNLS data [225] (solid thin), the BAO [5] (dotted), and the
CMB shift parameter from the WMAP 3-year data [226] (dot-dashed). The thick line represents the curve (139) for the flat
(0) (0) (0)
model (ΩK = 0). The figure labels Ωm and Ωrc correspond to Ωm and Ωrc , respectively. From Ref. [91].
For the flat universe (K = 0) this relation corresponds to
Ω(0) (0) 2
rc = (1 − Ωm ) /4 . (139)
The parametrization (137) of the Hubble parameter together with the normalization (138) can be used to place
observational constraints on the DGP model at the background level [89–93]. In Ref. [89] the authors found a
significantly worse fit to Supernova Ia (SN Ia) data and the distance to the last-scattering surface [3] relative to the
ΛCDM model. In Refs. [90] and [92] the authors showed that the flat DGP model is disfavored from the combined
data analysis of SN Ia [224, 225] and BAO [5]. In Fig. 3 we show the joint observational constraints [91] from the
data of SNLS [225], BAO [5], and the CMB shift parameter [226]. While the flat DGP model can be consistent with
the SN Ia data, it is under strong observational pressure by adding the data of BAO and the CMB shift parameter.
The open DGP model gives a slightly better fit relative to the flat model [91, 93]. The joint analysis using the data of
SN Ia, BAO, CMB, gamma ray bursts, and the linear growth factor of matter perturbations show that the flat DGP
model is incompatible with current observations [94].
In the DGP model a brane-bending mode φ (i.e. longitudinal graviton) gives rise to a field self-interaction of the
form φ(∂ µ φ∂µ φ) through a mixing with the transverse graviton [83, 84]. This can lead to the decoupling of the field
φ from gravitational dynamics in the local region by the so-called Vainshtein mechanism [85]. The General Relativistic
behavior can be recovered within a radius r∗ = (rg rc2 )1/3 , where rg is the Schwarzschild radius of a source. Since
r∗ is larger than the solar-system scales, the DGP model can evade local gravity constraints [83, 84, 228]. However
the DGP model is plagued by a strong coupling problem for typical distances smaller than 1000 km [229]. Some
regularization methods have been proposed to avoid the strong coupling problem, such as smoothing out the delta
profile on the brane [230, 231] or re-using the delta function profile but in a higher-dimensional brane [232–234].
As we will see in VII C, the analysis of 5D cosmological perturbations on the scales larger than r∗ shows that the
DGP model contains a ghost mode in the scalar sector of the gravitational field [86–88]. There are several ways of
the generalization of the DGP model to avoid the appearance of ghosts. One way is to consider the 6D braneworld
set-up as in the Cascading gravity [235]. Another is to generalize the field self-interaction term φ(∂ µ φ∂µ φ) to more
general forms in the 4D gravity [42]. In Sec. V we shall discuss the latter approach (“Galileon gravity”) in detail.
22
V. GALILEON GRAVITY
In the DGP model the field derivative self-coupling φ(∂µ φ∂ µ φ), arising from a brane-bending mode, allows the
decoupling of the field from matter within a Vainshtein radius. In the local regions where solar-system experiments
are carried out, the field is nearly frozen through the non-linear self-interaction. This is different from the chameleon
mechanism in which the presence of the field potential with a matter coupling gives rise to a minimum with a large
mass in the regions of high density.
Under the Galilean shift ∂µ φ → ∂µ φ+bµ , the field equation following from the Lagrangian φ(∂ µ φ∂µ φ) is unchanged
in the Minkowski space-time. The generalization of the nonlinear field Lagrangian to more general cases may be useful,
e.g., to overcome the ghost problem associated with the DGP model. In fact Nicolis et al. [42] derived five Lagrangians
that lead to the field equations invariant under the Galilean shift ∂µ φ → ∂µ φ + bµ in the Minkowski space-time. The
scalar field respecting the Galilean symmetry is dubbed “Galileon”. Each of the five terms only leads to second-order
differential equations, keeping the theory free from unstable spin-2 ghost degrees of freedom.
If we extend the analysis in Ref. [42] to that in the curved space-time, the Lagrangians should be promoted to
covariant forms. Deffayet et al. [95, 96] derived covariant Lagrangians Li (i = 1, · · · , 5) that keep the field equations
up to second-order, while recovering the five Lagrangians derived by Nicolis et al. in the Minkowski space-time. This
can be achieved by introducing field-derivative couplings with the Ricci scalar R and the Einstein tensor Gνρ in the
expression of L4,5 . Since the existence of those terms affects the effective gravitational coupling, the Galileon gravity
based on the covariant Lagrangians Li (i = 1, · · · , 5) can be classified as one of modified gravitational theories.
The cosmological dynamics including the terms up to L4 and L5 have been studied by a number of authors [97, 236].
In particular Refs. [97, 98] have shown that, for the covariant Galileon theory having de Sitter attractors, cosmological
solutions with different initial conditions converge to a common trajectory– a tracker solution. Moreover there is a
viable parameter space in which the conditions for the avoidance of ghosts and Laplacian instabilities of scalar and
tensor perturbations are satisfied.
The generalization of Galileon gravity, which mostly corresponds to the modification of the term L3 = φ(∂ µ φ∂µ φ),
has been also extensively studied recently [237]-[258]. One application is to introduce the non-linear field self-
interaction of the form ξ(φ)φ(∂µ φ∂ µ φ) in the action of (generalized) Brans-Dicke theories [183, 238–240, 242],
where ξ is a function of φ. For suitable choices of the function ξ(φ), there exist de Sitter (dS) solutions responsible
for dark energy even in the absence of the field potential. The cosmology based on a further general term G(φ, X)φ
has been discussed in the context of either dark energy and inflation [246, 247, 252, 253, 256–258].
In the following we review the cosmological dynamics in the Galileon dark energy model based on the covariant
Lagrangians Li (i = 1, · · · , 5) and study observational constraints on the model. We will also discuss the modified
version of Galileon gravity in which the term L3 = φ(∂ µ φ∂µ φ) is generalized.
A. Cosmology of a covariant Galileon field
We start with the covariant Galileon gravity described by the action [95, 96]
5
" # Z
2
4 √
Mpl 1X
Z
S = d x −g R+ ci Li + d4 x LM , (140)
2 2 i=1
where Mpl = (8πG)−1/2 is the reduced Planck mass, and ci ’s are constants. The five Lagrangians Li (i = 1, · · · , 5)
satisfying the Galilean symmetry in the limit of the Minkowski space-time are given by
L1 = M 3 φ , L2 = (∇φ)2 , L3 = (φ)(∇φ)2 /M 3 , L4 = (∇φ)2 2(φ)2 − 2φ;µν φ;µν − R(∇φ)2 /2 /M 6 ,

L5 = (∇φ)2 [(φ)3 − 3(φ) φ;µν φ;µν + 2φ;µ ν φ;ν ρ φ;ρ µ − 6φ;µ φ;µν φ;ρ Gνρ ]/M 9 , (141)
where M is a constant having a dimension of mass, and Gνρ is the Einstein tensor. For the matter Lagrangian LM we
take into account perfect fluids of non-relativistic matter (energy density ρm , equation of state wm = 0) and radiation
(energy density ρr , equation of state wr = 1/3).
Let us consider the FLRW metric with the cosmic curvature K:
dr2

2 2 2 2 2 2 2
ds = −dt + a (t) + r (dθ + sin θ dφ ) . (142)
1 − Kr2
Variation of the action (140) with respect to gµν leads to the following equations of motion
2
3Mpl H 2 = ρDE + ρm + ρr + ρK , (143)
2
3Mpl H2 + 2
2Mpl Ḣ = −PDE − ρr /3 + ρK /3 , (144)
23
2
where ρK ≡ −3KMpl /a2 , and
ρDE ≡ −c1 M 3 φ/2 − c2 φ̇2 /2 + 3c3 H φ̇3 /M 3 − 45c4 H 2 φ̇4 /(2M 6 ) + 21c5 H 3 φ̇5 /M 9 , (145)
PDE ≡ c1 M 3 φ/2 − c2 φ̇2 /2 − c3 φ̇2 φ̈/M 3 + 3c4 φ̇3 [8H φ̈ + (3H 2 + 2Ḣ)φ̇]/(2M 6 )
−3c5 H φ̇4 [5H φ̈ + 2(H 2 + Ḣ)φ̇]/M 9 . (146)
The matter fluids satisfy the continuity equations ρ̇m + 3Hρm = 0 and ρ̇r + 4Hρr = 0. We define the dark energy
equation of state wDE and the effective equation of state weff , as
PDE 2Ḣ
wDE ≡ , weff ≡ −1 − . (147)
ρDE 3H 2
Using the continuity equation ρ̇DE + 3H(ρDE + PDE ) = 0, it follows that wDE = weff − Ω̇DE /(3HΩDE ).
Since we are interested in the case where the late-time cosmic acceleration is realized by the field kinetic energy,
we set c1 = 0 in the following discussion5 . Then the de Sitter solution (H = HdS = constant) can be present for
φ̇ = φ̇dS = constant. We normalize the mass M to be M 3 = Mpl HdS 2
, which gives M ≈ 10−40 Mpl for HdS ≈ 10−60 Mpl .
Defining xdS ≡ φ̇dS /(HdS Mpl ), Eqs. (143) and (144) lead to the following relations at the de Sitter solution:
c2 x2dS = 6 + 9α − 12β , c3 x3dS = 2 + 9α − 9β , (148)
where
α ≡ c4 x4dS , β ≡ c5 x5dS . (149)
It is convenient to use the variables α and β, because the coefficients of physical quantities and dynamical equations
can be expressed by α and β. The relations (148) do not change under the rescaling xdS → γxdS and ci → ci /γ i ,
where γ is a real constant. Then the rescaled choice of ci can provide the same physics.
In order to study the cosmological dynamics, we introduce the following dimensionless variables:
!4
φ̇dS HdS 1 φ̇
r1 ≡ , r2 ≡ . (150)
φ̇H r1 φ̇dS
At the de Sitter solution r1 = 1 and r2 = 1. We define the dark energy density parameter
ρDE 3 2
ΩDE ≡ 2 H 2 = −(2 + 3α − 4β)r1 r2 /2 + (2 + 9α − 9β)r1 r2 − 15αr1 r2 /2 + 7βr2 . (151)
3Mpl
2
Then Eq. (143) can be written as ΩDE + Ωm + Ωr + ΩK = 1, where Ωm ≡ ρm /(3Mpl H 2 ), Ωr ≡ ρr /(3Mpl
2
H 2 ), and
2 2 2
ΩK ≡ ρK /(3MplH ) = −K/(aH) .
The autonomous equations for the variables r1 , r2 , Ωr , and ΩK are given by [98, 259]:
1
r1′ = (r1 − 1) r1 [r1 (r1 (−3α + 4β − 2) + 6α − 5β) − 5β]
∆
× 2 (Ωr − ΩK + 9) + 3r2 r13 (−3α + 4β − 2) + 2r12 (9α − 9β + 2) − 15r1 α + 14β ,

(152)
1
r2′ = − [r2 (6r12 (r2 (45α2 − 4(9α + 2)β + 36β 2 ) − (Ωr − ΩK − 7)(9α − 9β + 2)) + r13 (−2(Ωr − ΩK + 33)
∆
×(3α − 4β + 2) − 3r2 (−2(201α + 89)β + 15α(9α + 2) + 356β 2 )) − 3r1 α(−28Ωr + 28ΩK + 123r2 β + 36)
+10β(−11Ωr + 11ΩK + 21r2 β − 3) + 3r14 r2 (9α2 − 30α(4β + 1) + 2(2 − 9β)2 ) + 3r16 r2 (3α − 4β + 2)2
+3r15 r2 (9α − 9β + 2)(3α − 4β + 2))] , (153)
Ω′r ′
= Ωr (−4 − 2H /H) , (154)
Ω′K = ΩK (−2 − 2H ′ /H) , (155)
5 In this case the only solution in the Minkowski background (H = 0) corresponds to φ̇ = 0 for c2 6= 0.
24
where a prime represents a derivative with respect to N = ln a, and
∆ ≡ 2r14 r2 [72α2 + 30α(1 − 5β) + (2 − 9β)2 ] + 4r12 [9r2 (5α2 + 9αβ + (2 − 9β)β) + 2(9α − 9β + 2)]
+4r13 [−3r2 −2(15α + 1)β + 3α(9α + 2) + 4β 2 − 3α + 4β − 2] − 24r1 α(16r2 β + 3) + 10β(21r2 β + 8) . (156)

The Hubble parameter follows from the equation H ′ /H = −5r1′ /(4r1 ) − r2′ /(4r2 ). The solutions to Eqs. (154) and
(0) (0)
(155) are given by Ωr (N ) = Ωr e−4N H02 /H 2 (N ) and ΩK (N ) = ΩK e−2N H02 /H 2 (N ) respectively, where the subscript
“(0)” represents the values today (N = 0).
From Eqs. (152) and (153) we find that there are three distinct fixed points: (A) (r1 , r2 ) = (0, 0), (B) (r1 , r2 ) = (1, 0),
and (C) (r1 , r2 ) = (1, 1). As we have already mentioned, the point (C) corresponds to the de Sitter solution. By
considering homogeneous perturbations about this point, we can show that the de Sitter solution (C) is always
classically stable [98]. The point (B) is a tracker solution found in Ref. [97], along which the field velocity evolves
as φ̇ ∝ 1/H. During the radiation and matter eras the variable r2 is much smaller than 1. The fixed point (B) is
followed by the stable de Sitter point (C) once r2 grows to the order of 1. If the initial conditions of both r1 and r2
in the radiation era are much smaller than 1, then the solutions are close to the point (A) at the initial stage. At
late times the solutions approach the tracker at r1 = 1. Depending on the initial values of r1 , the epoch at which
the solutions reach the tracker is different. In the following we consider the background evolution in two regimes: (i)
r1 = 1 and (ii) r1 ≪ 1 in more detail.
1. Tracker solution (r1 = 1)
Along the tracker (r1 = 1) the dark energy density parameter (151) is given by
ΩDE = r2 , (157)
which is much smaller than 1 during the radiation and matter eras. From Eqs. (154) and (155) we obtain r2′ /r2 =
8 + 2Ω′r /Ωr . This is integrated to give
r2 = d1 a8 Ω2r , (158)
where d1 is a constant. From Eqs. (153) and (154) we have Ω′K /ΩK − Ω′r /Ωr = 2, which is integrated to give
(0)
ΩK ΩK
= d2 a2 , with d2 = (0)
. (159)
Ωr Ωr
Substituting Eqs. (158) and (159) into Eq. (154), we obtain the cosmologically viable solution to Ωr , as
p
−1 + d3 a − d2 a2 + 4d1 a8 + (−1 + d3 a − d2 a2 )2
Ωr = , (160)
2d1 a8
where d3 is another constant. Since the density parameter (160) evolves as Ωr ≃ 1 + d3 a in the early time (a ≪ 1),
this demands the condition d3 < 0 (provided ΩDE > 0). Using the density parameters today, the constants d1 and
(0) (0) (0) (0) (0) (0)
d3 can be expressed as d1 = [1 − Ωm − Ωr − ΩK ]/(Ωr )2 and d3 = −Ωm /Ωr . Since Ωr ∝ ρr /H 2 ∝ 1/(a4 H 2 ),
(0)
we have H 2 /H02 = (Ωr /Ωr )(1/a4 ). Using Eq. (160), the Hubble parameter can be expressed in terms of the redshift
z = 1/a − 1:
2
H(z) 1 (0) 1 1
= Ω (1 + z)2 + Ω(0) (1 + z)3 + Ω(0) (1 + z)4
H0 2 K 2 m 2 r
r
(0) (0) (0) (1 + z)4 h (0) (0) (0)
i2
+ 1 − Ωm − Ωr − ΩK + ΩK + Ωm (1 + z) + Ωr (1 + z)2 , (161)
4
which is useful to test the viability of the tracker solution from observations.
On the tracker, the equations of state defined in Eq. (147) are given by
Ωr − ΩK + 6 Ωr − ΩK − 6r2
wDE = − , weff = . (162)
3(r2 + 1) 3(r2 + 1)
25
0.0
-0.50 (a)
(b)
(c)
(d)
(e)
-1.0
wDE -1.5
-2.0
-2.5
0 2 4 6 8
log10 (1+z)
Figure 4: Evolution of wDE versus z for α = 0.3 and β = 0.14 [cases (a)-(d)] in the flat FLRW Universe (K = 0). We
choose four different initial conditions: (a) r1 = 5.000 × 10−11 , r2 = 8.000 × 10−12 , and Ωr = 0.999995 at z = 5.89 × 108 , (b)
r1 = 1.500 × 10−10 , r2 = 2.667 × 10−12 , and Ωr = 0.999992 at z = 3.63 × 108 , (c) r1 = 5.000 × 10−9 , r2 = 8.000 × 10−14 , and
Ωr = 0.99995 at z = 6.72 × 107 , (d) r1 = 5.000 × 10−6 , r2 = 8.000 × 10−17 , and Ωr = 0.9986 at z = 2.04 × 106 . The case (e)
corresponds to the tracker solution. From Ref. [98].
During the cosmological sequence of radiation (Ωr ≃ 1, |ΩK | ≪ 1, r2 ≪ 1), matter (Ωr ≪ 1, |ΩK | ≪ 1, r2 ≪ 1), and
de Sitter (Ωr ≪ 1, |ΩK | ≪ 1, r2 = 1) eras, the dark energy equation of state evolves as wDE = −7/3 → −2 → −1,
whereas the effective equation of state evolves as weff = 1/3 → 0 → −1. This peculiar evolution of wDE for the tracker
corresponds to the case (e) in Fig. 4. Although the effect of the cosmic curvature does not affect the dynamics of wDE
significantly, it can change the diameter distance as well as the luminosity distance relative to the flat Universe.
The epoch at which the solutions reach the tracking regime r1 ≃ 1 depends on model parameters and initial
conditions. The approach to this regime occurs later for smaller initial values of r1 , see Fig. 4. In Ref. [98] it was
shown that the tracker is stable in the direction of r1 by considering a homogeneous perturbation δr1 . This means
that once the solutions reach the tracker the variable r1 does not repel away from 1. If r1 . 2 initially, numerical
simulations show that the solutions approach the tracker with the late-time cosmic acceleration. Meanwhile, for the
initial conditions with r1 & 2, the dominant contribution to ΩDE comes from the Lagrangian L2 , so that the field
energy density decreases rapidly as in the standard massless scalar field.
2. Solutions in the regime r1 ≪ 1
There is another case in which the solutions start to evolve from the regime r1 ≪ 1 (where the term L5 gives the
dominant contribution to the field dynamics). In this regime, the variables r1 and r2 satisfy the following approximate
equations
9 + Ωr − ΩK + 21βr2 3 + 11Ωr − 11ΩK − 21βr2
r1′ ≃ r1 , r2′ ≃ r2 . (163)
8 + 21βr2 8 + 21βr2
As long as {βr2 , |ΩK |} ≪ 1, the evolution of r1 and r2 during the radiation (matter) era is given by r1 ∝ a5/4 and
r2 ∝ a7/4 (r1 ∝ a9/8 and r2 ∝ a3/8 ). Then the field velocity grows as φ̇ ∝ t3/8 during the radiation era and φ̇ ∝ t1/4
during the matter era. The evolution of φ̇ is slower than that for the tracker (i.e. φ̇ ∝ t).
In the regime r1 ≪ 1 the equations of state are
1 + Ωr − ΩK 8Ωr − 8ΩK − 21βr2
wDE ≃ − , weff ≃ . (164)
8 + 21βr2 3(8 + 21βr2 )
26
Provided that {βr2 , |ΩK |} ≪ 1, one has wDE ≃ −1/4, weff ≃ 1/3 during the radiation era and wDE ≃ −1/8, weff ≃ 0
during the matter era. This evolution of wDE is quite different from that for the tracker solution.
In Fig. 4 the variation of wDE is plotted for a number of different initial conditions with r1 ≪ 1 [which correspond
to the cases (a)-(d)]. As expected, the solutions start to evolve from the value wDE ≃ −1/4 in the radiation era. For
larger initial values of r1 they approach the tracker earlier. This tracking behavior also occurs in the presence of the
cosmic curvature K [259].
3. Conditions for the avoidance of ghosts and Laplacian instabilities
Let us find a model parameter space in which the appearance of ghosts and instabilities can be avoided in covariant
Galileon gravity. In doing so, we need to study a linear perturbation theory on the FLRW background. For simplicity
we focus on the flat Universe with K = 0. Let us consider the perturbed metric
ds2 = −[1 + 2Ψ(t, x)] dt2 + ∂i χ(t, x) dt dxi + a2 (t)[1 + 2Φ(t, x)] dx2 , (165)
where Ψ, Φ, and χ are scalar metric perturbations. We have chosen the gauge δφ = 0 without a non-diagonal scalar
perturbation in the spatial part of the metric, i.e. ∂ij γ = 0 [181]. Taking into account two perfect fluids with the
equations of state wi = Pi /ρi (i = 1, 2), there are three propagating scalar degrees of freedom. The velocity potentials
(i) (i)
vi (i = 1, 2) of perfect fluids are related with the energy-momentum tensor Tj0 , as Tj0 = −(ρi + Pi )∂j vi (i = 1, 2).
Introducing the vector Q ~ = (v1 , v2 , Φ) and expanding the action (140) up to the second-order, we obtain the
second-order action for scalar perturbations [98] (see also Refs. [183, 260, 261]):

1 1 ~t
Z
(2) 3 3 ~˙ t ~˙ ~ ˙t ~
~ ~ t ~
δSS = dt d x a Q AQ − 2 ∇Q C ∇Q − Q B Q − Q D Q , (166)
2 a
where the fields Ψ and χ are integrated out. A, C and D are 3 × 3 symmetric matrices and B is an antisymmetric
matrix (for which we do not write explicit forms).
In order to avoid the appearance of ghosts we require that the matrix A is positive definite. This corresponds to
the conditions (1 + w1 )ρ1 /w1 > 0, (1 + w2 )ρ2 /w2 > 0, and
QS 6(1 + µ1 )(µ1 + µ2 + µ1 µ2 − 2µ3 − µ23 )

2 ≡ − > 0, (167)
Mpl (1 + µ3 )2
where
µ1 ≡ 3αr1 r2 /2 − 3βr2 , (168)

µ2 ≡ (3α − 4β + 2)r13 r2 /2 − 2(9α − 9β + 2)r12 r2 + 45αr1 r2 /2 − 28βr2 , (169)
µ3 ≡ −(9α − 9β + 2)r12 r2 /2 + 15αr1 r2 /2 − 21βr2 /2 . (170)
The propagation speeds cS of three scalar degrees of freedom are known by solving the equation
det(c2S A − C) = 0 . (171)
For the two perfect fluids we have c2S = w1 and c2S = w2 , which are are positive for both radiation and non-relativistic
matter. The third stability condition associated with another scalar degree of freedom is given by
(1 + µ1 )2 [2µ′3 − (1 + µ3 )(5 + 3weff ) + 4Ωr + 3Ωm ] − 4µ′1 (1 + µ1 )(1 + µ2 ) + 2(1 + µ3 )2 (1 + µ4 )
c2S ≡ > 0, (172)
6(1 + µ1 )(µ1 + µ2 + µ1 µ2 − 2µ3 − µ23 )
where
µ4 ≡ −αr1 r2 /2 − 3βr2 (r1′ /r1 + r2′ /r2 )/4 . (173)
Let us consider tensor perturbations with δgij = a2 hij , where hij is traceless (hi i = 0) and divergence-free (hij ,j =
0). We expand the action (140) at second-order in terms of the two polarization modes, hij = h⊕ ǫ⊕ ⊗
ij + h⊗ ǫij , where
ǫ⊕ ⊗
ij and ǫij are the polarization tensors. For the polarization mode h⊕ , the second-order action is given by
c2T

1
Z
(2) 3 3 2 2
δST = dt d x a QT ḣ⊕ − 2 (∇h⊕ ) . (174)
2 a
27
¢
Α = 12 Β12 -9 Β - 2, cT 2 = 0 at IcT 2M = 0
1
2
Α = 2 Β + , HQSLr1=1 = 0
0 3
-1
Α = 2 Β, IcS2MdS = 0
12 10
Α= Β - , IcS2Mr =1,W =1 = 0
13 13 1 r
-2
-1.0 -0.5 0.0 0.5
Β
Figure 5: The viable parameter region (blue colored region) along the tracker solution (r1 = 1) in the covariant Galileon model
(where r2 > 0). If the solutions start from the regime r1 ≪ 1, we also require that β > 0. From Ref. [97].
The conditions for the avoidance of ghosts and Laplacian instabilities of tensor perturbations correspond, respectively,
to
QT 1 3 3
Mpl2 ≡ 2 + 4 αr1 r2 − 2 βr2 > 0 , (175)
2r1 (2 − αr1 r2 ) − 3β(r2 r1′ + r1 r2′ )

c2T ≡ > 0. (176)
2r1 (2 + 3αr1 r2 − 6βr2 )
The same conditions also follow from h⊗ .
2
In the regime r1 ≪ 1 and r2 ≪ 1 one has QS /Mpl ≃ 60βr2 and QT ≃ 1/2. For the initial conditions with r2 > 0
we require that β > 0 to avoid the scalar ghost. Since c2S ≃ (1 + Ωr )/40 and c2T ≃ 1 + 3βr2 (5 − 3Ωr )/8 ≃ 1, there are
no Laplacian instabilities of scalar and tensor perturbations in this regime.
In the tracking regime characterized by r1 = 1 (either r2 ≪ 1 or r2 = 1), the conditions (167), (172), (175), and
(176) give the bounds on the parameters α and β. In the regime r2 ≪ 1 these conditions translate to
QS ≃ 3(2 − 3α + 6β)r2 > 0 , (177)
8 + 10α − 9β + Ωr (2 + 3α − 3β)
c2S ≃ > 0. (178)
3(2 − 3α + 6β)
For the branch r2 > 0 the first condition reduces to 2 − 3α + 6β > 0. Since c2T ≃ 1 − r2 (4α + 3β + 3βΩr )/2 ≃ 1 and
2
QT /Mpl = 1/2 + 3(α − 2β)r2 /4 > 0, the tensor modes do not provide additional constraints. At the de Sitter point
(r2 = 1) we require that
QS 4 − 9(α − 2β)2 QT 1
2 = > 0, 2 = (2 + 3α − 6β) > 0 , (179)
Mpl 3(α − 2β)2 Mpl 4
(α − 2β)(4 + 15α2 − 48αβ + 36β 2 ) 2−α
c2S = > 0, c2T = > 0. (180)
2[4 − 9(α − 2β)2 ] 2 + 3α − 6β
If β > 0, it can happen that c2T has a minimum during the transition from the regime r2 ≪ 1 to r2 ≃ 1 [97, 98].
This value tends to decrease as β approaches 1. Imposing that c2T > 0 at the minimum, we obtain the bound
p
α < 12 β − 9β − 2 . (181)
In Fig. 5 we plot the parameter space in the (α, β) plane constrained by the conditions (177)-(181). Clearly there are
viable model parameters satisfying all the theoretical constraints.
28
4. Observational constraints on Galileon cosmology from the background cosmic expansion history
Since the evolution of the dark energy equation of state in covariant Galileon gravity is rather peculiar, the ob-
servational data related with the background cosmic expansion history may place tight constraints on the model.
Especially the analytic formula (161) for the tracker is useful for such a purpose. In Ref. [259] the authors confronted
the Galileon model by using the observational data of SN Ia (Constitution [262] and Union2 sets [263]), the CMB
(WMAP7) shift parameters [4], and BAO (SDSS7) [6].
If either of the SN Ia data (Constitution or Union2) is used in the data analysis, the χ2 for the tracker is similar
to that in the ΛCDM model. In the presence of the cosmic curvature K, the tracker solution is compatible with
the individual observational bound constrained from either CMB or BAO. However, the combined data analysis of
Constitution+BAO+CMB shows that the difference of χ2 between the tracker and the ΛCDM is δχ2 ∼ 22 (or ∼ 4.3σ).
This means that the tracker is severely disfavored with respect to the ΛCDM. A similar conclusion was reached from
the combined data analysis of Union2+BAO+CMB. The reason for this incompatibility is that the SN Ia data favor
(0) (0)
the large values of Ωm (& 0.32), whereas the CMB and BAO data constrain smaller values of Ωm (. 0.27).
The general solutions starting from the regime r1 ≪ 1 finally approach the tracker as r1 grows to 1. In Ref. [259]
the authors carried out the likelihood analysis for such general solutions and found that the solutions approaching the
tracker at late times (such as the case (a) in Fig. 4) are favored from the combined data analysis. In the flat FLRW
background the best-fit model parameters are α = 1.411 ± 0.056, β = 0.422 ± 0.022 (Constitution+CMB+BAO, 68 %
CL), and α = 1.404 ± 0.057, β = 0.419 ± 0.023 (Union2+CMB+BAO, 68 % CL).
(0)
For several fixed values of ΩK it was shown that the late-time tracking solutions can be consistent with the data,
(0) (0)
apart from the models with largely negative ΩK such as ΩK . −0.01. For example, the general solutions with
(0)
ΩK = 0.01 and the model parameters (α, β) = (1.862, 0.607) give the similar value of χ2 to that in the the ΛCDM.
In this case the Akaike-Information-Criteiron (AIC) statistics [264] also have the same support for the two models
(see also Ref. [265])
The Bayesian-Information-Criterion (BIC) statistics [266] show that the general solutions, with all 4 parameters
(0) (0)
(α, β, Ωm , ΩK ) are varied, are not particularly favored over the ΛCDM model. This mainly comes from the statistical
property that the numbers of model parameters are larger than those in the flat ΛCDM. In fact the late-time tracking
solutions with a non-zero cosmic curvature can be well consistent with the combined data analysis at the background
level.
B. Generalized Galileon gravity
In Sec. III we showed that in Brans-Dicke theory with the coupling Q of the order of unity the presence of the field
potential allows a possibility for the consistency with local gravity constraints through the chameleon mechanism.
Another way to recover the General Relativistic behavior in the regions of high density is to introduce the Galileon-like
field self-interaction. Silva and Koyama [238] studied Brans-Dicke theory in the presence of the term ξ(φ)φ(∂µ φ∂ µ φ)
[which is the generalization of the term L3 = φ(∂µ φ∂ µ φ)]. The action of this theory is given by
4 √
Z
1 ωBD
Z
S = d x −g φR − (∇φ) + ξ(φ)φ(∂µ φ∂ φ) + d4 x LM .
2 µ
(182)
2 2φ
If ξ(φ) ∝ φ−2 , there exists a de Sitter solution that can be responsible for the late-time acceleration. As in the Galileon
model discussed in Sec. V A the field is nearly frozen during the radiation and matter eras through the cosmological
Vainshtein mechanism, but it finally approaches the de Sitter solution characterized by φ̇ = constant. Moreover, as in
the DGP model, the Vainshtein radius can be much larger than the solar system scale, so that the General Relativistic
behavior can be recovered in the local region [238].
We may consider more general theories described by the action [242]
√
Z
1
Z
S = d4 x −g F (φ)R + B(φ)X + ξ(φ)φ(∂ µ φ∂µ φ) + d4 xLM , (183)
2
where X = −g µν ∂µ φ∂ν φ/2, and F (φ), B(φ), ξ(φ) are functions of φ. From the requirement of having de Sitter
solutions responsible for dark energy, it is possible to restrict the functional forms of F (φ), B(φ), and ξ(φ). In the
presence of non-relativistic matter (energy density ρm ) and radiation (energy density ρr ), the field equations are given
29
by
Bφ2 2 φF,φ 2ξφ3 2 3

φξ,φ ρm ρr
1= x − x+ H x 1− x + 2
+ , (184)
6F F F 6ξ 3F H 3F H 2
!
2ξφ3 2 2 Bφ2 F,φφ φ2 6ξφ3 2 2

Ḣ φF,φ ẋ Ḣ 2 φF,φ φξ,φ
−2 2 = − H x +x 2 +x + x+ x− + H x 1− x x
H F F H H F F F F 3ξ
ρm 4ρr
+ + , (185)
F H2 3F H 2
where x ≡ φ̇/(Hφ). Let us search for de Sitter solutions at which H and x are constants. If F (φ) and ξ(φ) are
power-law functions of φ, the quantities such as φF,φ /F , F,φφ φ2 /F , and φξ,φ /ξ remain constants. Provided that
B/F ∝ φ−2 and ξ/F ∝ φ−3 , we can solve Eqs. (184) and (185) for x and H at the de Sitter point. These conditions
are satisfied for the following functions
2
F (φ) = Mpl (φ/Mpl )3−n , B(φ) = ω(φ/Mpl )1−n , ξ(φ) = (λ/µ3 )(φ/Mpl )−n , (186)
where Mpl ≃ 1018 GeV is the reduced Planck mass, µ (> 0) is a constant having a dimension of mass, and ω and λ are
dimensionless constants. One can show that the coupling λ must be positive for the consistency of theories [242]. The
Brans-Dicke theory described by the action (182) corresponds to n = 2 with the Brans-Dicke parameter ωBD = ω.
Since F (φ) is constant for n = 3, the theory with n = 3 corresponds to k-essence minimally coupled gravity.
From Eqs. (184) and (185) we obtain the following algebraic equations at the de Sitter fixed point:
n(n − 3)2 x3dS + (n − 3)(n − 12)x2dS − 6(n − 5)xdS + 18

ω = − , (187)
x2dS (xdS + 3)
µ3 [(n − 3)xdS − 2][(n − 3)xdS − 3]
λ = 2 , (188)
Mpl HdS 2x3dS (xdS + 3)
where xdS and HdS are the values of x and H at the de Sitter point, respectively. We fix the mass scale µ to be
2 1/3
µ = (Mpl HdS ) , where we have used HdS ≃ 10−60 Mpl . For given ω and n, the quantity xdS is determined by solving
Eq. (187). Then the dimensionless constant λ is known from Eq. (188).
In order to recover the General Relativistic behavior in the early cosmological epoch we require that the field initial
value φi is close to Mpl from Eq. (186). The quantity x is much smaller than 1 in the early cosmological epoch, so that
the field is nearly frozen during the radiation and matter eras. The field starts to evolve at the late cosmological epoch
in which x grows to the order of unity. Introducing the dimensionless quantities y ≡ λx2 H 2 /HdS 2
and Ωr ≡ ρr /(3F H 2 ),
one can show that the fixed point corresponding to the matter era corresponds to (x, y, Ωr ) = (0, (3 − n)/6, 0) [242].
Since y is positive definite, it follows that n ≤ 3.
The conditions for the avoidance of ghosts and Laplacian instabilities are known by employing the method presented
in Sec. V A 3. Provided F (φ) > 0, we require that x > 0 to avoid ghosts during the cosmological evolution from the
radiation era to the epoch of cosmic acceleration [242]. The stability of the de Sitter point is automatically ensured
for n ≤ 3 and xdS > 0. From Eq. (187) the parameter ω is restricted in the range
ω < −n(n − 3)2 . (189)
The field propagation speed squared during the radiation and matter eras is given by c2S ≃ 6/5 and c2S ≃ 2/3,
respectively in which case no instabilities of linear perturbations are present. Meanwhile, at the de Sitter solution,
we have [242]
(n − 2)[(n − 3)(n − 4)x2dS − 8(n − 3)xdS + 6]xdS

c2s = − , (190)
(n − 3)(3n2 − 10n + 12)x3dS + 12(n − 5)x2dS + 18(n − 8)xdS − 108
which is positive for n ≥ 2. Hence the parameter n is restricted in the range
2 ≤ n ≤ 3, (191)
which includes Brans-Dicke theory with the action (182) as a specific case (n = 2).
In Ref. [239] the authors studied the evolution of matter density perturbations and showed that, for the model with
n = 2, there is an anti-correlation between the cross-correlation of large scale structure and the integrated Sachs-Wolfe
effect in CMB anisotropies. We shall discuss the main reason of this anti-correlation in Sec. VII D. This property will
be useful to distinguish the above model from the ΛCDM in future observations.
30
VI. OTHER MODIFIED GRAVITY MODELS OF DARK ENERGY
In this section we briefly discuss other classes of modified gravity models of dark energy. These include (i) Gauss-
Bonnet gravity with a scalar coupling f (φ)G, (ii) R/2 + f (G) gravity, and (iii) Lorentz-violating models.
A. Gauss-Bonnet gravity with a scalar coupling
In addition to the Ricci scalar R, we can construct other scalar quantities coming from the Ricci tensor Rµν and
the Riemann tensor Rµναβ , i.e. P ≡ Rµν Rµν and Q ≡ Rµναβ Rµναβ [267]. It is possible to avoid the appearance of
spurious spin-2 ghosts by taking a Gauss-Bonnet (GB) combination [268, 269, 271], defined by
G ≡ R2 − 4Rµν Rµν + Rµναβ Rµναβ . (192)
A simple model that can be responsible for the cosmic acceleration today is [43]
√
Z
1 1
Z
S = d4 x −g R − (∇φ)2 − V (φ) − f (φ)G + d4 x LM , (193)
2 2
where V (φ) and f (φ) are functions of a scalar field φ. The coupling of the field with the GB term appears in low-
energy effective string theory [270]. For the exponential potential V (φ) = V0 e−λφ and the coupling f (φ) = (f0 /µ)eµφ ,
the cosmological dynamics were studied in Refs. [43, 282–287]. In this model there exists a de Sitter solution due to
the presence of the GB term. In Refs. [282, 283] it was also found that the late-time de-Sitter solution is preceded by
a scaling matter era.
Koivisto and Mota [282] placed observational constraints on the model (193) with the exponential potential V (φ) =
V0 e−λφ , by using the Gold data set of SN Ia together with the CMB shift parameter data of WMAP. The parameter
λ is constrained to be 3.5 < λ < 4.5 at the 95% confidence level. In Ref. [284], they also included the constraints
coming from the BAO, LSS, big bang nucleosynthesis, and solar system data. This joint analysis showed that the
model is strongly disfavored by the data. In Ref. [283] it was also found that tensor perturbations are subject to
negative instabilities when the GB term dominates the dynamics (see also Refs. [288, 289] for related works).
Amendola et al. [290] studied local gravity constraints on the above model and showed that the density parameter
ΩGB coming from the GB term is required to be strongly suppressed for the compatibility with solar-system experi-
ments (which is typically of the order of ΩGB < 10−30 ). The above discussion indicates that the GB term with the
scalar-field coupling f (φ)G can hardly be the source for dark energy.
B. R/2 + f (G) gravity
The general Lagrangian including the scalar quantities constructed from the Ricci scalar R, the Ricci tensor Rµν
and the Riemann tensor Rµναβ is given by L = f (R, P, Q). The dark energy models based on these theories have been
studied in Refs. [272–279]. In order to avoid spurious spin-2 ghosts we need to choose the GB combination (192), i.e.
L = f (R, Q − 4P ) [280, 281].
The cosmological dynamics based on the action,
4 √
Z
1
Z
S = d x −g R + f (G) + d4 x LM , (194)
2
have been studied by a number of authors [44, 291–296]. In order to ensure the stability of radiation/matter solutions,
we need to satisfy the condition f,GG > 0 for all G. We also require the regularities of the functions f , f,G , and
f,GG [293]. There exists a de Sitter solution responsible for dark energy, whose stability requires the condition
0 < HdS f,GG (HdS ) < 1/384. Moreover f,GG must approach +0 in the limit |G| → ∞.
In Ref. [293] the authors proposed a number of f (G) models satisfying these conditions (see also Ref. [294]). One
of such models is given by

G G p
f (G) = λ √ arctan − αλ G∗ , (195)
G∗ G∗
where α, λ and G∗ are constants. The numerical simulation of Ref. [293] shows that the model (195) is cosmologically
viable at least at the background level. Moreover it can be consistent with solar-system constraints for a wide range
of model parameters [297].
31
In order to study the viability of the theories described by the action (194) further, let us consider the evolution of
matter density perturbations in the presence of a perfect fluid with the barotropic equation of state wM = PM /ρM .
In Ref. [298] it was shown that, for small scales (i.e. for large momenta k), there are two different scalar propagation
speeds. One of them corresponds to the mode for the fluid, i.e. c21 = wM , whereas another is given by
2Ḣ 1 + wM κ2 ρM
c22 = 1 + 2
+ , (196)
H 1 + 4µ 3H 2
where
µ = H Ġf,GG . (197)
The parameter µ characterizes the deviation from the ΛCDM model (note that the linear term f = c G does not give
rise to any contribution to the field equation). For viable f (G) models we have |µ| ≪ 1 at high redshifts [293]. Since the
background evolution during the radiation/matter domination is given by 3H 2 ≃ κ2 ρM and Ḣ/H 2 ≃ −(3/2)(1+wM ),
it follows that
c22 ≃ −1 − 2wM . (198)
The Laplacian instability at small scales is absent only for wM < −1/2. Since wM = 1/3 and wM = 0 during the
radiation and matter eras, respectively, the perturbations with large momentum modes are unstable. This leads to
violent growth of matter density perturbations incompatible with the observations of large-scale structure [292, 298].
By considering the full perturbation equations, one can show that the onset of the negative instability corresponds
to [298]
µ ≈ (aH/k)2 . (199)
Even when µ is much smaller than 1, we can always find a wave number k (≫ aH) satisfying the condition (199).
For the scales smaller than that determined by the wave number in Eq. (199), the linear perturbation theory breaks
down. Hence the background solutions cannot be trusted for those scales, which makes the theory unpredictable. The
Laplacian instability can be avoided only for µ = 0, which corresponds to the ΛCDM model. The above property
persists irrespective of the forms of f (G).
For more general theories described by the Lagrangian density f (R, G) it is possible to avoid such Laplacian
instabilities [299], depending on the models [300] (see also Refs. [301–304]). It may be of interest to construct some
viable dark energy models in such theories.
C. Lorentz violating models
The modified gravity models such as f (R) gravity and Galileon gravity can give rise to the phantom dark energy
equation of state wDE < −1 without violating the conditions for the appearance of ghosts and instabilities. In the
models with a broken Lorentz invariance it is also possible to realize wDE < −1 without pathological behavior in the
Ultra-Violet (UV) region [305–307] (see also Ref. [308] for a review). In order to construct Lorentz violating models
without pathological behavior of phantoms, one may start with a field theory consistent at energy scales from zero
to the UV cutoff scale M and then deform the theory in the Infra-Red (IR) in such a way that its behavior at high
energies remains healthy. Although the weak energy condition is violated in the homogenous background, pathological
states are present below a certain low scale ǫ only. Provided that ǫ is close to the Hubble scale, a theory of this sort
should be acceptable.
Let us consider a Lorentz violating model with two-derivative kinetic terms with healthy behavior below the scale
M [309] plus one-derivative term suppressed by the small parameter ǫ [310]. The model has a vector field Bµ and a
scalar field Φ with a potential V (B, Φ). The Lagrangian is given by [306, 307]
L = L(2) + L(1) + L(0) + LM , (200)

1 1 Bν Bλ 1
L(2) = − α(Ξ)g νλ Dµ Bν Dµ Bλ + β(Ξ)Dµ Bν Dµ Bλ 2
+ ∂µ Φ∂ µ Φ , (201)
2 2 M 2
L(1) = ǫ∂µ ΦB µ , L(0) = −V (B, Φ) , (202)
where Ξ = Bµ B µ /M2 (M is the UV cut-off scale). The dimensionless parameters α and β are the functions of Ξ,
and ǫ is a free positive parameter that characterizes an IR scale. The Lorentz invariance is broken for Ξ 6= 0.
32
1.5
Wr Wm
WDE
1.0
0.50 weff
0.0
-0.50
w
-1.0
-1.5
0 5 10 15
ln a
Figure 6: Evolution of wDE (denoted as w in the figures) and weff = 1 − 2Ḣ/(3H 2 ) together with the density parameters
of dark energy (ΩDE ), non-relativistic matter (Ωm ), and radiation (Ωr ) for the Lorentz-violating model (200) with V =
m2 φ2 /2 − M 2 X 2 /2. The model parameters are chosen to be α = 1, γ = 1/2, ǫ/m√= 3 and M/m = 1 with the initial conditions
X = Ẋ = φ̇ = 0, φ = 0.5φA , and Ωr = 0.99, Ωm = 0.01 [where φA = M mpl /( 4παm)]. The present epoch corresponds to
Ωm = 0.3 and ΩDE = 0.7, which is denoted by a vertical line. After the cosmological constant boundary crossing, wDE reaches
a minimum −1.19 and then it increases toward the de Sitter value −1 from the phantom side. From Ref. [307].
In spatially homogeneous background we have
B0 = X , Bi = 0 , Φ = φ, (203)
where B0 and Bi are time and space components of Bµ . In the flat FLRW background with non-relativistic matter
and radiation for the matter Lagrangian LM , we obtain the following equations of motion

2 8πG 1 2 3α 2 2 1 2
H = γ Ẋ − H X + φ̇ + W (φ) + U (X) + ρm + ρr , (204)
3 2 2 2
1 3
γ Ẍ + 3H Ẋ + γ,X Ẋ 2 + α,X H 2 X 2 + 3αH 2 X − ǫφ̇ + V,X = 0 , (205)
2 2
φ̈ + 3H φ̇ + ǫ(Ẋ + 3HX) + V,φ = 0 , (206)
where γ(X) = X 2 β(X)/M2 − α(X).

For the separable potential V = m2 φ2 /2−M 2 X 2 /2 with constant X the cosmological dynamics of the above system
have been
p studied in detail in Refs. [306, 307]. In what follows we assume that√ both α and γ are constants. Provided
ǫ/m > 2α/3 one can show that there is a de Sitter solution with H = M/ 3α, at which φ and X are frozen. In the
early cosmological epoch the field X is close to 0, whereas the field φ slowly rolls down its potential (with the energy
density dominating over that of X).
Prior to
√ the epoch of de Sitter cosmic acceleration there is a transient phantom regime (wDE < −1) characterized by
H < M/ 3α in which√ the field φ rolls up its potential. The Hubble parameter slowly increases toward the de Sitter
value HdS = M/ 3α. In Fig. 6 we plot the evolution of the dark energy equation of state wDE = PDE /ρDE and the
effective equation of state weff = 1 − 2Ḣ/(3H 2 ) together with the density parameters of dark energy, non-relativistic
matter, and radiation6 . Clearly the solution undergoes the period with wDE < −1 before reaching the de Sitter
attractor.
6 Here the definition of ρDE and PDE is ρDE = γ Ẋ 2 /2 + φ̇2 /2 + V − 3αH 2 X 2 /2 and PDE = γ Ẋ 2 /2 + φ̇2 /2 − V + ǫφ̇X + αḢX 2 +
2αHX Ẋ + 3αH 2 X 2 /2.
33
The phantom equation of state can be realized without having ghosts, tachyons or superluminal modes in the
UV region. In pthe IR region characterized by p . ǫ, either tachyons or ghosts appear for the spatial momenta p
smaller than (ǫ2 − M 2 )/α − m2 [306, 307]. The presence of tachyons at IR scales leads to the amplification of
large-scale field perturbations whose wavelengths are roughly comparable to the present Hubble radius. There are
two tachyonic regions of spatial momenta in this model: (a) one is sub-horizon and its momenta are characterized by
M 2 /γ < p2 < (ǫ2 − M 2 )/α − m2 ; (b) another is super-horizon with the momenta 0 < p2 < m2 M 2 /ǫ2 . In the region
(a) there is a parameter space in which the perturbations always remain smaller than the homogenous fields. In the
region (b) the growth of the perturbations is suppressed by the factor m2 /ǫ2 .
There are other classes of Lorentz violating models such as ghost condensate [305] and Horava-Lifshitz gravity [311]
(see also Refs. [312]-[317]). The application of such scenarios to dark energy has been studied by a number of authors,
see e.g., [318–320]. It remains to to see whether such Lorentz-violating models can be observationally distinguished
from other dark energy models.
VII. OBSERVATIONAL SIGNATURES OF MODIFIED GRAVITY
In order to confront modified gravity models with the observations of large-scale structure and CMB, we discuss
the evolution of density perturbations in four modified gravity models: (i) f (R) gravity, (ii) scalar-tensor gravity,
(iii) DGP braneworld model, and (iv) Galileon gravity. We also discuss observables to confront with weak lensing
observations.
A. f (R) gravity
Let us first consider metric f (R) gravity in the presence of non-relativistic matter. We take the following perturbed
metric in a longitudinal gauge about the flat FLRW background with scalar metric perturbations Φ and Ψ [181]
ds2 = −(1 + 2Ψ)dt2 + a2 (1 + 2Φ)δij dxi dxj . (207)
The energy momentum tensors of a non-relativistic perfect fluid are decomposed into background and perturbed parts,
as T00 = −(ρm + δρm ) and Tα0 = −ρm vm,α (vm is a velocity potential).
The equations for matter perturbations, in the Fourier space, are given by [321–323]
h i
δ ρ̇m + 3Hδρm = −ρm 3Φ̇ + (k 2 /a)vm , (208)
v̇m + Hvm = Ψ/a , (209)
where k is a comoving wave number. We define the gauge-invariant matter density perturbation δm , as
δm ≡ δρm /ρm + 3Hv , where v ≡ avm . (210)
Then Eqs. (208) and (209) yield
δ̇m = −(k 2 /a2 )v − 3(Φ − Hv)· , (211)

v̇ = Ψ , (212)
from which we obtain
δ̈m + 2H δ̇m + (k 2 /a2 )Ψ = 3B̈ + 6H Ḃ , (213)
where B ≡ −Φ + Hv.
In f (R) gravity the quantity F (R) = ∂f /∂R has a perturbation δF . In the following we use the unit κ2 = 8πG = 1,
but we restore gravitational constant G when it is required. For the action given in Eq. (2), we obtain the linearized
perturbation equations in Fourier space [324–326]
k2 k2

1 ˙ 2
− 2 Φ + 3H(HΨ − Φ̇) = 3H δF − 3Ḣ + 3H − 2 δF − 3H Ḟ Ψ − 3Ḟ (HΨ − Φ̇) − δρm , (214)
a 2F a
2
¨ + 3H δF
δF ˙ + k + M 2 δF = 1 δρm + Ḟ (3HΨ + Ψ̇ − 3Φ̇) + (2F̈ + 3H Ḟ )Ψ , (215)
a2 3
Ψ + Φ = −δF/F . (216)
34
In Eq. (215) we have introduced the mass term

1 F
M2 ≡ −R . (217)
3 f,RR
For viable dark energy models the condition F/f,RR ≫ R is satisfied during most of the cosmological epoch, so that
M 2 ≃ F/(3f,RR ) [69, 71]. This is equivalent to the mass squared Ms2 introduced in Eq. (43), which is required to be
positive to avoid the tachyonic instability.
For the observations of large-scale structure and weak lensing we are interested in the modes deep inside the Hubble
radius (k ≫ aH). In the following we employ the quasi-static approximation under which the dominant terms in
Eqs. (213)-(216) correspond to those including k 2 /a2 , δρm (or δm ) and M 2 [12, 155, 165, 327–329]. We then obtain
the following approximate relations from Eqs. (213)-(216) :
δ̈m + 2H δ̇m + (k 2 /a2 )Ψ = 0 , (218)

2 2
1 a 1 a
Φ= δρm − δF , Ψ = − δρm + δF , (219)
2F k 2 2F k 2
¨ + 3H δF˙ + k 2 /a2 + M 2 δF = δρm /3 .

δF (220)
The evolution of perturbations is different depending on whether M 2 is larger or smaller than k 2 /a2 . We shall discuss
two cases: (A) M 2 ≫ k 2 /a2 and (B) M 2 ≪ k 2 /a2 , separately. For viable f (R) models the mass squared M 2 is large
in the past and it gradually decreases with time. Hence the transition from the regime (A) to the regime (B) can
occur in the past, depending on the wave numbers k.
1. Evolution of perturbations in the regime: M 2 ≫ k2 /a2
The solutions to Eq. (220) are given by the sum of the oscillating solution δFosc obtained by setting δρm = 0 and
the special solution δFind of Eq. (220) induced by the presence of matter perturbations δρm . The oscillating part
δFosc satisfies the equation (a3/2 δFosc )·· +M 2 (a3/2 δFosc ) ≃ 0. Using the WKB approximation, we obtain the following
solution [69]
!
1
Z
δFosc ∝ a−3/2 f,RR 1/4 cos p dt , (221)
3f,RR
where we have used the approximation F ≃ 1.

For the analytic estimation of the oscillating mode we take the model (47), which corresponds to the asymptotic
form of the models (45) and (46) in the region R ≫ Rc . During the matter era in which the background Ricci scalar
evolves as R(0) = 4/(3t2 ), the quantity f,RR has a dependence f,RR ∝ R−2(n+1) ∝ t4(n+1) for the model (47). Then
the evolution of the perturbation, δRosc = δFosc /f,RR , is given by
δRosc ≃ c t−(3n+4) cos(c0 t−2(n+1) ) , (222)
where c and c0 are constants. Unless the coefficient c is chosen to be very small, as we go back to the past, the
perturbation δRosc dominates over the background value R(0) (∝ t−2 ). Since the Ricci scalar can be negative, this
leads to the violation of the stability conditions (f,RR > 0 and F > 0).
The special solution δFind to Eq. (220) can be derived by neglecting the first and second terms relative to others,
giving
δFind ≃ δρm /(3M 2 ) , δRind ≃ δρm . (223)
Under the condition |δFosc | ≪ |δFind |, one has δF ≃ δρm /(3M 2 ) and hence Ψ = −Φ = −(a2 /k 2 )δρm /(2F ). Then the
matter perturbation equation (218) reduces to
δ̈m + 2H δ̇m − 4πGρm δm /F = 0 , (224)
where we have reproduced the gravitational constant G. In Refs. [330, 331] the perturbation equations have been
derived without neglecting the oscillating mode.
35
During the matter-dominated epoch (Ωm = ρm /(3F H 2 ) ≃ 1), Eq. (224) has the growing-mode solution
δm ∝ t2/3 . (225)
This is the same evolution as that in standard General Relativity. From Eq. (223) the matter-induced mode evolves
as δFind ∝ t4(n+2/3) and δRind ∝ t−4/3 . Compared to the oscillating mode (222), δRind decreases more slowly and
hence it dominates over δRosc at late times. The evolution of the perturbation δR = δRosc + δRind relative to the
background value R(0) is given by
δR/R(0) ≃ c1 t−(3n+2) cos(c0 t−p ) + c2 t2/3 , (226)
where c1 and c2 are constants. In order to avoid the dominance of the oscillating mode at the early cosmological
epoch, we require that the coefficient c1 is suppressed relative to c2 [69, 71, 332].
This
p fine-tuning of initial conditions is related with the singularity problem raised by Frolov [333]. The field
φ = 3/2κ2 ln F in the Einstein frame has a weak singularity at φ = 0 (at which the curvature R and the mass
M go to infinity with a finite potential V ). Unless the oscillating mode of the field perturbation δφ is strongly
suppressed relative to the background field φ(0) , the system can access the curvature singularity. This past singularity
can be cured by taking into account the R2 term [334]. Note that the f (R) models proposed in Ref. [335] [e.g.,
f (R) = R − αRc ln(1 + R/Rc )] to cure the singularity problem satisfy neither the local gravity constraints [336] nor
observational constraints of large-scale structure [337]. There are some works for the construction of unified models
of inflation and dark energy based on f (R) theories [34, 338, 339], but the smooth transition between two accelerated
epochs without crossing the point f,RR = 0 is not easy unless the forms of f (R) are carefully chosen [334].
2. Evolution of perturbations in the regime: M 2 ≪ k2 /a2
Since the mass M decreases as M ∝ t−2(n+1) , the modes initially in the region M 2 ≫ k 2 /a2 can enter the regime
M 2 ≪ k 2 /a2 during the matter-dominated epoch. It is sufficient to consider the matter-induced mode because the
oscillating mode is already suppressed during the evolution in the regime M 2 ≫ k 2 /a2 . The matter-induced special
solution of Eq. (220) in the regime M 2 ≪ k 2 /a2 is approximately given by
a2
δFind ≃ δρm . (227)
3k 2
From Eq. (219) the gravitational potentials satisfy
4 1 a2 2 1 a2
Ψ=− · δρm , Φ= · δρm . (228)
3 2F k 2 3 2F k 2
Plugging Eq. (228) into Eq. (218), the matter perturbation obeys the following equation
4
δ̈m + 2H δ̇m − · 4πGρm δm /F = 0 . (229)
3
During the matter-dominated epoch (Ωm ≃ 1 and a ∝ t2/3 ), we obtain the following evolution
√
δ m ∝ t( 33−1)/6
. (230)
The growth rate of δm gets larger compared to that in the regime M 2 ≫ k 2 /a2 .
3. Matter power spectra and the ISW effect
If the transition from the regime M 2 ≫ k 2 /a2 to the regime M 2 ≪ √

k 2 /a2 occurs during the matter era, the
2/3 ( 33−1)/6
evolution of matter perturbations changes from δm ∝ t to δm ∝ t . We use the subscript “k” for the
quantities at which k is equal to aM , whereas the subscript “Λ” is used at which the accelerated expansion starts
(ä = 0). While the redshift zΛ is independent of k, zk depend on k and also on the mass M .
For the model (47) the variable m = Rf,RR /f,R can grow fast from the regime m ≪ (aH/k)2 (i.e., M 2 ≫ k 2 /a2 ) to
the regime m ≫ (aH/k)2 (i.e., M 2 ≪ k 2 /a2 ). In fact, m can grow to the order of 0.1 even if m is much smaller than
10−6 in the deep matter era. For the sub-horizon modes relevant to the galaxy power spectrum, the transition at
36
(a) Q=0.7, p=0.6

(b) Q=-(1/6)1/2, p=0.6
(c) ΛCDM, linear
(d) ΛCDM, nonlinear
105
Pδ [h Mpc ]
3
4
10
-3
m
3
10
2
10
0.001 0.01 0.1 1
-1
k [h Mpc ]
Figure 7: The matter power spectra Pδm√(k) in Brans-Dicke theory with the potential (98). This analysis covers the f (R) model
(47) with√the correspondence Q = −1/ 6 and p = 2n/(2n + 1). Each case corresponds to (a) Q = 0.7, p = 0.6, C = 0.9, (b)
Q = −1/ 6, p = 0.6, C = 0.9, (c) the ΛCDM model, and (d) the ΛCDM model with a nonlinear halo-fitting (σ8 = 0.78 and
(0) 2
shape parameter Γ = 0.2). The model parameters are Ωm = 0.28, H0 = 3.34 × 10−4 h Mpc−1 , ns = 1 and δH = 3.2 × 10−10 .
From Ref. [340].
M 2 = k 2 /a2 typically occurs at the redshift zk larger than 1 (provided that n = O(1)). For the mode k/(a0 H0 ) = 300
one has zk = 4.83 for n = 1 and zk = 2.49 for n = 2. As n gets larger, the period of non-standard evolution of δm
becomes shorter because zk tends to be smaller. Since the scalaron mass evolves as M ∝ t−2(n+1) for the model (47),
the time tk has a scale-dependence tk ∝ k −3/(6n+4) . This means that the smaller-scale modes cross the transition
point earlier. The matter power spectrum Pδm = |δm |2 at the time tΛ shows a difference compared to the case of the
ΛCDM model:
√
33−1
2 − 32
Pδm (tΛ ) tΛ 6 √
33−5
= ∝k 6n+4 . (231)
Pδm ΛCDM (tΛ ) tk
From Fig. 7 we find that the matter power spectrum in the f (R) model (47) with n = 3/4 is in fact larger than that
in the ΛCDM model on smaller scales.
The galaxy matter power spectrum is modified by this effect. Meanwhile the CMB spectrum is hardly affected
except for very large scales (for the multipoles ℓ = O(1)) at which the Integrated Sachs-Wolfe (ISW) effect becomes
important. Hence there is a difference for the spectral indices of two power spectra, i.e.
√
33 − 5
∆n(tΛ ) = . (232)
6n + 4
For larger n the redshift zk can be as close as zΛ , which means that the estimation (232) is not necessarily valid in
such cases. Moreover the estimation (232) does not take into account the evolution of δm after z = zΛ to the present
epoch (z = 0). It was found in Ref. [71] that the estimation (232) agrees well with the numerically obtained ∆n(tΛ )
for n ≤ 2.
In order to discuss the growth rate of matter perturbations, it is customary to introduce the growth index [341]
δ̇m
fδ ≡ = (Ω̃m )γ , (233)
Hδm
where Ω̃m = κ2 ρm /(3H 2 ). In the ΛCDM model γ is nearly constant for the redshifts 0 < z < 1, i.e. γ ≃ 0.55
[342, 343]. In f (R) gravity, if the perturbations are in the GR regime (M 2 ≫ k 2 /a2 ) today, γ is close to 0.55. On
the other hand, if the transition to the scalar-tensor regime occurs at the redshift zk larger than 1, the growth index
tends to be smaller than 0.55 [344, 345]. Since 0 < Ω̃m < 1, the smaller γ implies a larger growth rate.
37
For the wave numbers relevant to the linear regime of the matter power spectra (0.01 h Mpc−1 . k . 0.2 h Mpc−1 ,
where h ≈ 0.7) the viable f (R) models (45)-(48) can give rise to the growth index γ0 = 0.40 today. Depending on the
wave numbers k, γ0 can be dispersed in the regime 0.40 . γ0 . 0.55, or γ0 can show the convergence in the regime
0.40 . γ0 . 0.43 [344, 345]. Moreover, when γ0 is small, the growth index exhibits a large variation even for low
redshifts (0 < z < 1). Although the present observational constraints on γ are quite weak, the unusual evolution of γ
can be useful to distinguish the f (R) models from the ΛCDM in future observations.
After the system enters the epoch of cosmic acceleration, the wave number k can again become smaller than aM .
Hence the k-dependence is not necessarily negligible even for z < zΛ . However numerical simulations show that ∆n(t0 )
is not much different from ∆n(tΛ ) derived by Eq. (232) [71]. Thus the analytic estimation (232) is certainly reliable
to place constraints on model parameters except for n ≫ 1. Observationally we do not find any strong difference for
the slopes of the spectra of LSS and CMB. If we take the mild bound ∆n(tΛ ) < 0.05, we obtain the constraint n ≥ 2
[69]. In this case the local gravity constraint (67) is also satisfied.
For the wave numbers k & 0.2 h Mpc−1 we need to take into account the non-linear effect of density perturbations.
In Refs. [346–349] the authors carried out N -body simulations for the f (R) model (46) (see also Refs. [350, 351]).
Hu and Sawicki (HS) [352] proposed a fitting formula to describe the non-linear power spectrum based on the halo
model. Koyama et al. [353] studied the validity of the HS fitting formula by comparing it with the results of N -body
simulations and showed that the HS fitting formula can reproduce the power spectrum in N -body simulations for the
3 2 2
scales k < 0.5h Mpc−1 . In the quasi non-linear regime a normalized skewness, S3 = hδm i/hδm i , has been evaluated
in f (R) gravity and in Brans-Dicke theories [354]. The skewness in f (R) dark energy models differs only by a few
percent relative to the value S3 = 34/7 in the ΛCDM model.
The modified growth of matter perturbations also affects the evolution of the gravitational potentials Ψ and Φ.
The effective potential ψ ≡ Φ − Ψ is important to discuss the ISW effect on the CMB as well as the weak lensing
observations [355]. For the modes deep inside the Hubble radius, Eq. (219) gives
2
aH
ψ = 3Ωm δm . (234)
k
Note that for the large-scale modes relevant to the ISW effect in CMB we need to solve the full perturbation equations
without using the quasi-static approximation on sub-horizon scales. For viable f (R) models, however, numerical
simulations show that the result (234) can be trustable even for the wave numbers close to the Hubble radius today
(i.e. k & a0 H0 ) [155]. In the ΛCDM model the potential ψ remains constant during the standard matter era, but
it decays after the system enters the accelerated epoch, producing the ISW contribution for low multipoles on the
CMB power spectrum. In f (R) gravity the additional growth of matter perturbations in the region z < zk changes
the evolution of ψ.
From the observation of the CMB angular power spectrum, the constraint on the deviation parameter B ≡
(ṘH/RḢ)m from the ΛCDM is weak. The value of B today is constrained to be B0 < 4.3 [356]. There is an-
other observational constraint coming from the angular correlation between the CMB temperature field and the
galaxy number density field induced by the ISW effect. The avoidance of a large anti-correlation between the obser-
vational data of CMB and LSS places an upper bound of B0 . 1. Since this roughy corresponds to m(z = 0) . 1, the
CMB observations do not provide tight constraints on f (R) models relative to the matter power spectrum of LSS. In
weak lensing observations, the modified evolution of the lensing potential ψ directly leads to the change even for the
small-scale shear power spectrum [340, 355, 357]. Hence this can be a powerful tool to constrain f (R) gravity models
from future observations.
B. Brans-Dicke theory
Let us proceed to discuss the evolution of matter perturbations in Bran-Dicke theory described by the action (76)
with the potential U (φ) and the coupling F (φ) = e−2Qφ . We are mainly interested in large coupling models with |Q|
of the order of unity [76, 81, 82], as this gives rise to significant deviation from the ΛCDM. We define the field mass
squared to be M 2 ≡ U,φφ . If the scalar field is light such that the condition M < H0 is always satisfied irrespective of
high or low density regions, the coupling Q is constrained to be |Q| . 10−3 from local gravity constraints. Meanwhile,
if the mass M in the region of high density is much larger than that on cosmological scales, the model can satisfy local
gravity constraints under the chameleon mechanism even if |Q| is of the order of unity. Cosmologically the mass M
can decrease from the past to the present, which can allow the transition from the “GR regime” to the “scalar-tensor
regime” as it happens in f (R) gravity. An example of the field potential showing this behavior is given by Eq. (98).
As in f (R) gravity, the matter perturbation δm obeys Eq. (213). The difference appears in the expression of the
38
gravitational potential Ψ. In Fourier space the scalar metric perturbations obey the following equations [76, 325, 326]
"
k2 1 1
− 2 Φ + 3H(HΨ − Φ̇) = − ω φ̇δ φ̇ + (ω,φ φ̇2 − F,φ R + 2V,φ )δφ
a 2F 2
#
2

k
+ 3Ḣ + 3H 2 − 2 δF − 3Hδ Ḟ + (3H Ḟ − ω φ̇2 )Ψ + 3Ḟ (HΨ − Φ̇) + δρm , (235)
a
" #
k 2 ω,φ φ̇2

ω,φ 2U,φ − F,φ R
δ φ̈ + 3H + φ̇ δ φ̇ + 2 + + δφ
ω a ω ,φ 2 2ω ,φ
ω,φ 2 1
= φ̇Ψ̇ + 2φ̈ + 3H φ̇ + φ̇ Ψ + 3φ̇(HΨ − Φ̇) + F,φ δR , (236)
ω 2ω
δF F,φ
Ψ+Φ=− =− δφ , (237)
F F
where δφ is the perturbation of the field φ, ω = (1 − 6Q2 )F , and
h i
δR = 2 3(Φ̇ − HΨ)· − 12H(HΨ − Φ̇) + k 2 /a2 − 3Ḣ Ψ + 2(k 2 /a2 )Φ . (238)
Provided that the field is sufficiently heavy to satisfy the conditions M 2 ≫ R, one can employ the approximation
[(2U,φ − F,φ R)/2ω],φ ≃ M 2 /ω in Eq. (236). The solution to Eq. (236) consists of the sum of the matter-induced mode
δφind sourced by the matter perturbation and the oscillating mode δφosc , i.e. δφ = δφind + δφosc (as in the case of
f (R) gravity).
In order to know the evolution of the matter-induced mode we employ the quasi-static approximation on sub-horizon
scales. Under this approximation, we have δRind ≃ 2(k 2 /a2 )[Φ − (F,φ /F )δφind ] from Eqs. (235) and (238), where the
subscript “ind” represents the matter-induced mode. Then from Eq. (236) we find
2QF k2
δφind ≃ − Φ. (239)
(k 2 /a2 )(1 − 2Q2 )F + M 2 a2
Using Eqs. (235) and (237) we obtain
k2 δρm (k 2 /a2 )(1 + 2Q2 )F + M 2 k2 δρm (k 2 /a2 )(1 − 2Q2 )F + M 2

Ψ ≃ − , Φ ≃ . (240)
a2 2F (k 2 /a2 )F + M 2 a2 2F (k 2 /a2 )F + M 2
In the massive limit M 2 /F ≫ k 2 /a2 , we recover the standard result of General Relativity. In the massless limit
M 2 /F ≪ k 2 /a2 , it follows that (k 2 /a2 )Ψ ≃ −(δρm /2F )(1 + 2Q2 ) and (k 2 /a2 )Φ ≃ (δρm /2F )(1 − 2Q2 ).
Plugging Eq. (240) into Eq. (218), we obtain the equation for matter perturbations [76]
δ̈m + 2H δ̇m − 4πGeff ρm δm = 0 , (241)
where the effective gravitational coupling is
G (k 2 /a2 )(1 + 2Q2 )F + M 2

Geff = . (242)
F (k 2 /a2 )F + M 2
We have recovered the bare gravitational constant G. In the massless limit (M 2 ≪ k 2 /a2 ) this reduces to
G G 4 + 2ωBD
Geff ≃ (1 + 2Q2 ) = , (243)
F F 3 + 2ωBD
where in the last line we have used the relation (77) between the coupling Q and the Brans-Dicke parameter ωBD . In
f (R) gravity we have ωBD = 0 and hence Geff = 4G/(3F ).
Let us consider the evolution of the oscillating mode of perturbations. Using Eqs. (235) and (236) for sub-horizon
modes (k 2 /a2 ≫ H 2 ), the gravitational potentials can be expressed by δφosc (note that δρm = 0 for the oscillating
mode). From Eq. (238) the perturbation of R corresponding to the oscillating mode is given by
h i
δRosc ≃ 6Q δ φ̈osc + 3Hδ φ̇osc + (k 2 /a2 )δφosc . (244)
39
Substituting Eq. (244) into Eq. (236), it follows that
δ φ̈osc + 3Hδ φ̇osc + k 2 /a2 + M 2 /F δφosc ≃ 0 ,

(245)
which is valid in the regime M 2 ≫ R.

When |Q| = O(1) the field potential U (φ) is required to be heavy in the region of high density for the consistency
with local gravity constraints. We take the potential (98) as an example of a viable model. During the matter era the
1
p−1
field φ settles down at the instantaneous minima characterized by the condition (100). Then we have that φ ∝ ρm
2−p
and M 2 ∝ ρm 1−p
during the matter-dominated epoch. The field φ can initially be heavy to satisfy the condition
M 2 /F ≫ k 2 /a2 for the modes relevant to the galaxy power spectrum. Depending upon the model parameters and
the mode k, the mass squared M 2 can be smaller than k 2 /a2 during the matter era [76].
In the regime M 2 /F ≫ k 2 /a2 the matter perturbation equation (241) reduces to the standard one in Einstein grav-
ity, which gives the evolution δm ∝ t2/3 . For the model (98) the matter-induced mode of the field perturbation
evolves
2(4−p) p
1
as δφind ∝ δρm /M 2 ∝ t 3(1−p) . Meanwhile, the WKB solution to Eq. (245) is given by δφosc ∝ t 2(1−p) cos ct− 1−p ,
2
where c is a constant. Since the background field φ during the matter era evolves as φ ∝ t 1−p , we find
4−p
1

δφ/φ = (δφind + δφosc )/φ ≃ c1 t2/3 + c2 t− 2(1−p) cos ct− 1−p . (246)
As long as the oscillating mode is initially suppressed relative to the matter-induced mode, the latter remains the
dominant contribution in the subsequent cosmic expansion history.
In the regime M 2 /F ≪ k 2 /a2 the effective gravitational coupling is given by Eq. (243). Solving Eq. (241) in this
case, we obtain the solution
√ 2
δm ∝ t( 25+48Q −1)/6 . (247)
√ √
Setting Q = −1/ 6, this recovers the solution δm ∝ t( 33−1)/6 in metric f (R) gravity.
The potential (98) has a heavy mass M much larger than H in the deep matter-dominated epoch, but it gradually
decreases with time. Depending on the modes k, the system crosses the point M 2 /F = k 2 /a2 at t = tk during
2−p
the matter-dominated epoch. Since the field mass evolves as M ∝ t− 1−p during the matter era, the time tk has a
3(1−p)
scale-dependence given by tk ∝ k − 4−p . Since 0 < p < 1 the smaller scale modes (i.e. larger k) cross the transition
point earlier. During the matter era the mass squared is approximately given by
2−p
1−p
2 1−p ρm
M ≃ p Q2 U0 . (248)
(2 p C)1/(1−p) V0
(0)
Using the relation ρm = 3F0 Ωm H02 (1 + z)3 , the critical redshift zk at time tk can be estimated as
" 1
# 4−p
2(1−p)
k 1 2p pC 1 U0
zk ≃ −1, (249)
a 0 H0 Q (1 − p)1−p (3F0 Ω(0) )2−p H02
m
where a0 is the scale factor today. The critical redshift increases for larger k/(a0 H0 ) and for smaller p. If k/(a0 H0 ) =
600 and p = 0.7, Eq. (249) gives zk = 3.9.
Defining the growth rate of matter perturbations as in Eq. p(233), it follows that the asymptotic values of fδ in the
regions t ≪ tk and t ≫ tk are given by fδ = 1 and fδ = ( 25 + 48Q2 − 1)/4, respectively. Numerical simulations
show that the growth rate reaches a maximum value around the end of the matter era and then it starts to decrease
during the epoch of cosmic acceleration [76]. The observational constraint on fδ reported by McDonald et al. [358]
is fδ = 1.46 ± 0.49 around the redshift z = 3, whereas the data reported by Viel and Haehnelt [359] in the redshift
range 2 < z < 4 show that even the value p fδ = 2 can be allowed in some of the observations. If we use the criterion
fδ < 2 with the analytic estimation fδ = ( 25 + 48Q2 − 1)/4, we obtain the bound Q < 1.08. Note that the growth
index today can be smaller than 0.4 for |Q| larger than 0.4, so this will be also useful to place tight bounds on Q in
future observations.
The relative difference of the matter power spectrum Pδm at time t = tΛ (at which ä = 0) from that in the ΛCDM
is given by
√
25+48Q2 −1
2 − 23 √
Pδm (tΛ ) tΛ 6 (1−p)( 25+48Q2 −5)
= ∝k 4−p . (250)
PδΛCDM
m
(tΛ ) tk
40
In Fig. 7 we plot the matter power spectrum for Q = 0.7 and p = 0.6, which deviates from that in the ΛCDM model
on small scales. The estimation (250) shows fairly good agreement with numerical results [340].
From Eqs. (240) we find that the effective gravitational potential ψ = Φ − Ψ satisfies the same equation as (234).
Since the ISW effect induced by the modified evolution of ψ is limited on large-scale CMB perturbations irrelevant
to the galaxy power spectrum, there is a difference between the spectral indices of the matter power spectrum and of
the CMB spectrum on the scales, k > 0.01h Mpc−1 :
p
(1 − p)( 25 + 48Q2 − 5)
∆n(tΛ ) = . (251)
4−p
√
This reproduces the result (232) in f (R) gravity by setting Q = −1/ 6 and p = 2n + 1. If we use the criterion
∆n(tΛ ) < 0.05, as in the case of the f (R) gravity, we obtain the bounds p > 0.957 for Q = 1 and p > 0.855 for
Q = 0.5. As long as p is close to 1, it is possible to satisfy both cosmological and local gravity constraints for |Q| . 1.
C. DGP model
In this section we study the evolution of linear matter perturbations in the DGP braneworld model. The discussion
below is valid for the wavelengths larger than the Vainshtein radius r∗ . For the radius r smaller than r∗ the non-linear
effect coming from the brane-bending mode becomes crucially important. The perturbed metric in the 5-dimensional
longitudinal gauge with four scalar metric perturbations Ψ, Φ, B, E is given by [87, 361]
ds2 = −(1 + 2Ψ)n(t, y)2 dt2 + (1 + 2Φ)A(t, y)2 δij dxi dxj + 2rc B,i dxi dy + (1 + 2E)dy 2 , (252)
where the brane is located at y = 0 in the 5-th dimension characterized by the coordinate y (we are considering a flat
FLRW spacetime on the brane). Note that B can be identified as a brane bending mode describing a perturbation
of the brane location and that rc is the crossover scale defined in Eq. (130). The background solution describing the
self-accelerating Universe is [218]
n(t, y) = 1 + H(1 + Ḣ/H 2 )y , A(t, y) = a(t)(1 + Hy) . (253)
The Hubble parameter H = ȧ/a satisfies Eq. (133) with ǫ = +1.

In what follows we neglect the terms suppressed by the factor aH/k ≪ 1 because we are considering sub-horizon
perturbations. We also ignore the terms such as (A′ /A)Φ′ , where a prime represents a derivative with respect to y.
This comes from the fact that Φ′ is of the order of (k/a)Φ, as we will show later. The time-derivative terms can be
also dropped under the quasi-static approximation on sub-horizon scales. Then the perturbed 5-dimensional Einstein
tensors δ G̃AB obey the following equations locally in the bulk [361]:
∇2 rc A′

0 2 2
′′
δ G̃0 = 3Φ + 2 ∇ Φ + 2 (E − rc B ) − 2 2′
∇2 B = 0 , (254)
A A A A
′
n′

1 rc A
δ G̃ij = − 2 (∇i ∇j − δji ∇2 )(Φ + Ψ + E − rc B ′ ) + δji (Ψ′′ + 2Φ′′ ) + 2 (∇i ∇j − δji ∇2 ) + B = 0 , (255)
A A A n
δ G̃5i = −(Ψ′ + 2Φ′ ),i = 0 , (256)
′
n′

5 1 2 rc A
δ G̃5 = 2 ∇ (Ψ + 2Φ) − 2 2 + ∇2 B = 0 . (257)
A A A n
Taking the divergence of the traceless part of Eq. (255), we obtain
∇2 rc A′ n′

(Φ + Ψ + E − rc B ′
) − + ∇2 B = 0 . (258)
A2 A2 A n
The consistency between Eqs. (256) and (257) requires that
B ′ = 0, Ψ′ + 2Φ′ = 0 . (259)
From Eqs. (257) and (258) we find
∇2 ′ 1 ∇2 rc n′ 2
(E − rc B ) = − Ψ + ∇ B. (260)
A2 2 A2 2A2 n
41
Substituting Eqs (257) and (260) into Eq. (254) together with the use of Eq. (259), we obtain
∇2 n′ rc 2
Ψ′′ + 2
Ψ− ∇ B = 0. (261)
A n A2
Under the sub-horizon approximation (k/aH ≫ 1) the solution to Eq. (261), upon the Fourier transformation, is
given by
n′ h i
Ψ− rc B = c1 (1 + Hy)−k/aH + c2 (1 + Hy)k/aH , (262)
n
where c1 and c2 are integration constants. In order to avoid the divergence of the perturbation in the limit y → ∞
we choose c2 = 0.
The junction condition at the brane can be expressed in terms of an extrinsic curvature Kµν and an energy-
momentum tensor on the brane [362]:
Kµν − Kgµν = −κ2(5) Tµν /2 + rc Gµν , (263)
where K ≡ Kµµ . The extrinsic curvature is defined as Kµν = hλµ ∇λ nν , where nν is the unit vector normal to the
brane and hµν = gµν − nµ nν is the induced metric on the brane. The (0, 0) and spatial components of the junction
condition (263) give
2 2 1 3
∇ Φ = −κ2(4) δρm + 2 ∇2 B − Φ′ , (264)
a2 a rc
Φ + Ψ = B, (265)
Ψ′ + 2Φ′ = 0 , (266)
where δρm is the matter perturbation on the brane. Equation (266) is consistent with the latter of Eq. (259).
From Eq. (262) it follows that Φ′ ∼ (k/a)Φ in Fourier space. For the perturbations whose wavelengths are much
smaller than the cross-over scale rc , i.e., rc k/a ≫ 1, the term (3/rc )Φ′ in Eq. (264) is much smaller than (k 2 /a2 )Φ.
In Fourier space Eq. (264) is approximately given by
2k 2 2 k2
Φ = κ (4) δρ m + B. (267)
a2 a2
Using the projection of Eq. (257) as well as Eqs. (265) and (267), we find that metric perturbations Ψ and Φ obey
the following equations
2
k2 κ(4) k2 κ2(4)

1 1
2
Ψ=− 1+ δρm , Φ = 1− δρm , (268)
a 2 3β a2 2 3β
where
!
A′ n′

2rc Ḣ
β(t) ≡ 1 − 2 + = 1 − 2Hrc 1+ . (269)
3 A n 3H 2
The matter perturbation δm satisfies the same form of equation as given in (218) for the modes deep inside the horizon
[360]. Substituting the former of Eq. (268) into Eq. (218), we find that the matter perturbation obeys the following
equation [87, 360]
δ̈m + 2H δ̇m − 4πGeff ρm δm = 0 , (270)
where

1
Geff = 1+ G. (271)
3β
From Eq. (268) the effective gravitational potential ψ = Φ − Ψ obeys the same equation as (234).
In the deep matter era one has Hrc ≫ 1 and hence β ≃ −Hrc , so that β is largely negative (|β| ≫ 1). In this regime
the evolution of the matter perturbation is similar to that in General Relativity (δm ∝ t2/3 ). The solutions finally
approach the de Sitter attractor characterized by HdS = 1/rc . At the e Sitter solution one has β ≃ 1 − 2Hrc ≃ −1.
42
Since 1 + 1/(3β) ≃ 2/3, the growth rate in this regime is smaller than that in GR. The growth index is approximately
given by γ ≈ 0.68 [343], which is different from the value γ ≃ 0.55 in the ΛCDM model. If the future imaging survey
of galaxies can constrain γ within 20 %, it may be possible to distinguish the DGP model from the ΛCDM model
[363].
Comparing Eq. (271) with the effective gravitational coupling (243) in Brans-Dicke theory with a massless limit (or
the absence of the field potential), we find that the Brans-Dicke parameter ωBD has the following relation with β:
3
ωBD = (β − 1) . (272)
2
Since β < 0 for the self-accelerating DGP solution, this implies that ωBD < −3/2. Since in this case the kinetic energy
of a scalar field degree of freedom is negative in the Einstein frame, the DGP model contains a ghost mode. The
solution in another branch of the DGP model is not plagued by this problem, because the minus sign of Eq. (269) is
replaced by the plus sign. The self accelerating solution in the original DGP model can be realized at the expense of
an appearance of the ghost state.
D. Galileon gravity
1. Covariant Galileon gravity
In covariant Galileon gravity described by the action (140) the evolution of matter density perturbations was studied
in Ref. [364]. In spite of the complexities of full perturbation equations, they are simplified under the quasi-static
approximation on sub-horizon scales. Under this approximation the matter perturbation obeys the same equation as
(270) with a different effective gravitational coupling Geff . In Ref. [364] it was found that Geff is independent of the
wave number k, as in the DGP model. Geff is close to the gravitational constant G in the asymptotic past, but it
starts to deviate at the late cosmological epoch.
We define the effective gravitational potential ψ = Φ − Ψ as well as the anisotropic parameter η = −Φ/Ψ. Under
the quasi-static approximation on sub-horizon scales we obtain [364]
2
Geff 1 + η aH
ψ≃3 Ωm δ m , (273)
G 2 k
where Ωm and δm and the density parameter and the perturbation of non-relativistic matter, respectively. In f (R)
gravity, Brans-Dicke theory, and the DGP model the effective gravitational potential obeys Eq. (234) for the modes
deep inside the Hubble radius. In Galileon gravity the combination (Geff /G)(1 + η)/2 is different from 1. This means
that the effective gravitational potential may acquire some additional growth compared to other models. In Galileon
cosmology there are three different regimes characterized by (i) r1 ≪ 1, r2 ≪ 1, (ii) r1 = 1, r2 ≪ 1, and (iii) r1 = 1,
r2 = 1, where r1 and r2 are defined in Eq. (150). In these regimes we can estimate Geff and η as follows [364]. We
stress that these analytic results are valid for the modes deep inside the Hubble radius.
• (i) r1 ≪ 1, r2 ≪ 1
Expanding Geff and η about r1 = 0, r2 = 0, it follows that

Geff 255 211 2 129 589
=1+ β+ αr1 r2 + O(r2 ) , η =1+ β+ αr1 r2 + O(r22 ) , (274)
G 8 16 8 16
where α and β are defined by Eq. (149). Since β > 0 to avoid ghosts (for the branch r2 > 0), we have Geff > G
and η > 1 in this regime. This means that the growth rates of δm and ψ are larger than those in the ΛCDM
model.
• (ii) r1 = 1, r2 ≪ 1
Expansion of Geff and η about r2 = 0 gives
Geff 291α2 + 702β 2 − 933αβ + 20α − 84β + 4
= 1+ r2 + O(r22 ) , (275)
G 2(10α − 9β + 8)
3(126α2 + 306β 2 − 405αβ + 4α − 30β)
η = 1− r2 + O(r22 ) . (276)
2(10α − 9β + 8)
The evolution of Geff and η depends on both α and β. If α = 1.4 and β = 0.4, for example, we have
Geff /G ≃ 1 + 4.31r2 and η ≃ 1 − 5.11r2 , respectively. In this case Geff > G, but η is smaller than 1.
43
Figure 8: Evolution of the perturbations for α = 1.37 and β = 0.44 with the background initial conditions r1 = 0.03 and
r2 = 0.003. (Left) δm /a versus z for the wave numbers (a) k = 300a0 H0 , (b) k = 30a0 H0 , and (c) k = 5a0 H0 . (Right) ψ versus
z for the wave numbers (a) k = 300a0 H0 , (b) k = 10a0 H0 , and (c) k = 5a0 H0 . Note that δm /a and ψ are divided by their
initial amplitudes δm (ti )/a(ti ) and ψ(ti ), respectively, so that their initial values are normalized to be 1. The bold dotted lines
show the results obtained under the quasi-static approximation on sub-horizon scales. From Ref. [364].
• (iii) r1 = 1, r2 = 1
At the dS point we have
Geff 1
= , η = 1, (277)
G 3(α − 2β)
which means that there is no anisotropic stress.

Recall that the late-time tracking solutions are favored from observational constraints at the background level. In
this case the solutions start from the regime (i) and finally approach the de Sitter fixed point with a short period of
the regime (ii). In Fig. 8 we plot δm /a and ψ versus the redshift z for α = 1.37 and β = 0.44 with the background
initial conditions r1 = 0.03 and r2 = 0.003. In this case the solutions approach the tracker at late times. The initial
conditions of perturbations are chosen to recover the GR behavior in the asymptotic past. Note that these results
are obtained by numerical integration of the full perturbation equations without quasi-static approximations. For
the mode k = 300a0H0 the numerical result shows excellent agreement with that obtained under the quasi-static
approximation on sub-horizon scales. The difference starts to appear for the modes k/(a0 H0 ) < O(10). From the left
panel of Fig. 8 we find that, on larger scales, the growth of δm tends to be less significant. For the modes k ≫ a0 H0
the matter perturbation evolves faster than a during the matter era.
From the right panel of Fig. 8 we find that, unlike the ΛCDM model, ψ changes in time even during the matter era
for the modes k ≫ aH. Before reaching the tracker we have Geff /G ≃ 1+255βr2/8 > 1 and η ≃ 1+129βr2/8 > 1 from
Eq. (274). Then the growth rates of ψ and δm get larger than those in GR. In particular the term (Geff /G)(1 + η)/2
in Eq. (273) is larger than 1, which leads to the additional growth of ψ to that coming from δm . In Galileon gravity
the unusual behavior of the anisotropic parameter η leads to the non-trivial evolution of perturbations. For the model
parameters α = 1.37 and β = 0.44, Eq. (277) gives Geff ≃ 0.68G at the de Sitter fixed point. Since in this case Geff
is smaller than G, ψ begins to decrease at some point after the matter era.
For the large-scale modes relevant to the ISW effect in CMB anisotropies (k/(a0 H0 ) . 10), ψ is nearly constant in
the early matter-dominated epoch. However, as we see in Fig. 8, ψ exhibits temporal growth during the transition
from the matter era to the epoch of cosmic acceleration. The characteristic variation of ψ in the Galileon model may
leave interesting observational signatures on the large-scale CMB anisotropies.
For the model parameters constrained by SN Ia (Union2)+CMB+BAO data sets, i.e. α = 1.404 ± 0.057 and
β = 0.419 ± 0.023 [259], the effective gravitational coupling at the de Sitter solution is restricted in the range
44
0.5G < Geff < 0.72G. If the model parameters are close to the upper limit α = 2β + 2/3 of the allowed parameter
space at the background level (i.e. Geff is close to 0.5G at the de Sitter point), the parameter η tends to show a
divergence during the transition from the matter era to the epoch of cosmic acceleration. If Geff is larger than 0.66G,
we find that such divergent behavior is typically avoided. For the viable model parameters the evolution of δm and ψ
exhibits more or less the similar property to that shown in Fig. 8.
2. Modified Galileon gravity
Finally we study modified Galileon theories in which the term φ(∂µ φ∂ µ φ) is generalized to ξ(φ)φ(∂µ φ∂ µ φ). Let
us consider general theories described by the action
√
Z
1
Z
S = d4 x −g f (R, φ, X) + ξ(φ)φ(∂µ φ∂ µ φ) + d4 x LM , (278)
2
where
f (R, φ, X) = f1 (R, φ) + f2 (φ, X) . (279)
We introduce two mass scales associates with the field φ and the scalar gravitational degree of freedom, respectively,
as
Mφ2 ≡ −f,φφ /2 , MR2 ≡ F/(3F,R ) , (280)
where F ≡ ∂f /∂R.
The full perturbation equations for the perturbed metric (207) are given in Ref. [183]. Under the quasi-static ap-
proximation on sub-horizon scales the matter perturbation satisfies Eq. (270) with the effective gravitational coupling
[183]
( )
1 1 + 4s1 [F,φ + 2(1 + 4s1 )ξ φ̇2 ]2
Geff = 1+ , (281)
8πF 1 + 3s1 (1 + 4s1 )µF
where
2
µ ≡ (1 + 3s1 )(f,X + 2s2 ) + 3F,φ /F + 2ξ[4(1 + 3s1 )(φ̈ + 2H φ̇) − 2F,φ φ̇2 /F − 2ξ φ̇4 (1 + 4s1 )/F ] , (282)
2 2
s1 ≡ k /(3a MR2 ) , s2 ≡ a 2
Mφ2 /k 2 . (283)
The anisotropic parameter η = −Φ/Ψ is given by

2
(1 + 2s1 )(f,X + 2s2 ) + 2F,φ /F + 4ξ[2(1 + 2s1 )(φ̈ + 2H φ̇) − F,φ φ̇2 /F ]
η= 2 /F + 8ξ(1 + 4s )(φ̈ + 2H φ̇)
. (284)
(1 + 4s1 )(f,X + 2s2 ) + 4F,φ 1
The effective gravitational potential ψ = Φ − Ψ obeys the equation of the form (273), whose explicit form is
2 2
(1 + 3s1 )[f,X + 2s2 + 8ξ(φ̈ + 2H φ̇)] + 3F,φ /F − 2ξF,φ φ̇2 /F

aH
ψ = 3Ωm δm , (285)
k 2 /F − 4ξ[F φ̇2 /F + ξ φ̇4 (1 + 4s )/F ]
(1 + 3s1 )[f,X + 2s2 + 8ξ(φ̈ + 2H φ̇)] + 3F,φ ,φ 1
where Ωm = ρm /(3F H 2 ).
In the massive limits Mφ2 → ∞ and MR2 → ∞, i.e. s1 → 0 and s2 → ∞ we recover the standard results in
2
GR: Geff ≃ 1/(8πF ), η ≃ 1, and ψ ≃ 3Ωm δm (aH/k) . The difference appears in the regimes characterized by the
conditions s1 & 1 and s2 . 1, as it happens for the late-time evolution of perturbations in f (R) gravity and in
Brans-Dicke theory.
2
For the theories with ξ = 0, Eq. (285) gives the standard relation ψ = 3Ωm δm (aH/k) . If ξ 6= 0, then this relation
no longer holds. In this case ψ is not directly related with δm due to the additional contribution from the ξ-dependent
term. This can be regarded as the main reason for the anti-correlation between the ISW effect in CMB and the
large-scale structure found for the model (182) [239].
In Ref. [183] the authors derived conditions for avoiding the appearance of ghosts and Laplacian instabilities.
Provided F > 0 the tensor ghosts do not appear. Since c2T = 1 for the theories (278), tensor perturbations have
45
no Laplacian instabilities. For the theories with f,RR 6= 0 the conditions for the avoidance of ghosts and Laplacian
instabilities of scalar perturbations are given, respectively, by
24ξH φ̇ − 8ξ,φ φ̇2 + f,X + f,XX φ̇2 − F,X

2 2
φ̇ /F,R > 0 , (286)
2 4
f,X + 8(φ̈ + 2H φ̇)ξ − 16ξ φ̇ /(3F ) > 0 . (287)
Similar conditions have been also derived for the theories with f,RR 6= 0. The dark energy models based on the action
(278) need to be constructed to satisfy these conditions.
E. Observables in weak lensing
We have shown that modified gravity models of dark energy generally lead to changes for the growth rate of matter
perturbations compared to the ΛCDM model. Since there are two free functions that determine the first-order metrics
Ψ and Φ, dark energy models can be classified according to how the gravitational potentials are linked to δm . In order
to quantify this, we introduce two quantities q(k, t) and ζ(k, t) defined by
(k 2 /a2 )Φ = 4πGq δm ρm , (288)

(Φ + Ψ)/Φ = ζ , (289)
where G is the 4-dimensional bare gravitational constant. The ΛCDM model corresponds to q = 1 and ζ = 0 (note
that the cosmological constant does not cluster). Modified gravity models give rise to different values of q and ζ relative
to those in the ΛCDM model. Therefore the functions q and ζ characterize gravitational theories for first-order scalar
perturbations on small scales.
In Brans-Dicke theory discussed in Sec. VII B, the gravitational potentials are given by Eq. (240) on sub-horizon
scales. In this case we have
1 (k 2 /a2 )(1 − 2Q2 )F + M 2 4F (k 2 /a2 )Q2
q= , ζ=− , (290)
F (k 2 /a2 )F + M 2 (k 2 /a2 )(1
− 2Q2 )F + M 2
where we have used the unit 8πG = 1. In the regime M 2 /F ≫ k 2 /a2 (and F ≃ 1) it follows that q ≃ 1 and ζ ≃ 0.
In the regime M 2 /F ≪ k 2 /a2 we have q ≃ (1 − 2Q2 )/F and ζ ≃ −4Q2 /(1 − 2Q2 ), so that the√deviation from the
ΛCDM model appears. The expression (290) covers the case of f (R) gravity by setting Q = −1/ 6.
In the DGP model the gravitational potentials obey Eq. (268) and hence
q = 1 − 1/(3β) , ζ = 2/(1 − 3β) . (291)
In the deep matter era one has |β| ≫ 1, so that q ≃ 1 and ζ ≃ 0. The deviation from (q, ζ) = (1, 0) appears when |β|
decreases to the order of unity, i.e., when the Universe enters the epoch of cosmic acceleration.
In order to confront dark energy models with the observations of weak lensing, it may be convenient to introduce
the following quantity [355]
Σ ≡ q(1 − ζ/2) . (292)
From Eqs. (288) and (289) we find that the effective gravitational potential ψ = Φ − Ψ associated with weak lensing
observations can be expressed as
ψ = 8πG(a2 /k 2 )ρm δm Σ . (293)
In the DGP model and in Brans-Dicke theory we have Σ = 1 and Σ = 1/F , respectively. In (modified) Galileon
theories the term Σ is of more complicated forms, see e.g., Eq. (285).
The effect of modified gravity theories manifests itself in weak lensing observations in at least two ways. One is the
multiplication of the term Σ on the r.h.s. of Eq. (293). Another is the modification of the evolution of δm . The latter
depends on two parameters q and ζ, or equivalently, Σ and ζ. Thus two parameters (Σ, ζ) will be useful to detect
signatures of modified gravity theories from future surveys of weak lensing.
46
VIII. CONCLUSIONS
We have reviewed modified gravitational models of dark energy responsible for the cosmic acceleration today. In
addition to cosmological constraints such as the presence of a matter era followed by a stable de Sitter solution, we
require that the models satisfy local gravity constraints. There are two mechanisms for the recovery of GR behavior
in the regions of high density. The first one is the chameleon mechanism in which the mass of a scalar-field degree
of freedom depends on the matter density in the surrounding environment. The chameleon mechanism can be at
work in f (R) gravity and Brans-Dicke theory, as long as the field potential is designed to have a large mass in the
regions of high density. The second one is the Vainshtein mechanism in which the nonlinear effect of scalar-field self
interactions leads to the recovery of GR at small distances. This can be applied to the DGP braneworld model and
Galileon gravity.
The modified gravity models can give rise to the phantom equation of state of dark energy (wDE < −1) without
having ghosts, tachyons, and Laplacian instabilities. The deviation of wDE from that in the ΛCDM model (wDE = −1)
is not so significant in the models based on f (R) gravity and Brans-Dicke theory, so these models can be compatible
with the observational constraints at the background cosmology fairly easily. On the other hand, the tracker solution
in covariant Galileon gravity, which has wDE = −2 during the matter era, is disfavored from the joint data analysis
of SN Ia, BAO, and CMB shift parameters. However the late-time tracking without a significant deviation from
wDE = −1 is allowed observationally.
In order to confront the modified gravity models with the observations of large-scale structure, CMB, and weak
lensing, we have also discussed the evolution of matter density perturbations. In the models based on f (R) and
Brans-Dicke theories there is a “General Relativistic” regime in which the field is heavy such that M 2 ≫ k 2 /a2 .
At late times this is followed by a “scalar-tensor” regime (M 2 ≪ k 2 /a2 ) in which the gravitational law is modified
from that in General Relativity. In√Brans-Dicke theory the evolution of matter perturbations during the matter era
2
changes from δm ∝ t2/3 to δm ∝ t( 25+48Q −1)/6 , where Q is related to the Brans-Dicke parameter via the relation
3 + 2ωBD = 1/(2Q2 ). The effective gravitational couplings Geff in the DGP and Galileon models are independent
of the wave numbers k. This reflects the fact that the field is massless in those models. In the DGP model the
growth rate of δm is smaller than that in the ΛCDM, which is associated with the appearance of ghosts. In (modified)
Galileon gravity, a non-trivial relation between the effective gravitational potential ψ and the matter perturbation δm
leads to the extra growth of ψ.
We summarize the current status of each modified gravity model of dark energy.
• (i) In f (R) gravity there are some viable dark energy models such as (45), (46), and (48), which can be consistent
with both cosmological and local gravity constraints. For the models (45) and (46) the local gravity constraints
can be satisfied for n > 0.9 under the chameleon mechanism. If we use the criterion that the difference between
the spectral indices between the matter power spectrum and the CMB spectrum is smaller than 0.05, then we
obtain the bound n ≥ 2. In these models the initial conditions of perturbations need to be chosen such that
the oscillating mode does not dominate over the matter-induced mode in the early Universe. This is associated
with a weak singularity problem about the divergence of the mass squared M 2 ≃ F/(3f,RR ) for R → ∞, but it
can be circumvented by including higher-curvature terms such as R2 to the Lagrangian.
• (ii) In Brans-Dicke theory, even if the coupling Q is of the order of unity, it is possible to design field potentials
consistent with both cosmological and local gravity constraints. One of the examples is the potential (98),
which is motivated by the f (R) models √ (45) and (46). Depending on the couplings Q the growth of matter
perturbations is different (Q = −1/ 6 in metric f (R) gravity). From the observational constraints on the
growth rate of δm , |Q| is required to be smaller than the order of 1.
• (iii) In the DGP model the self-acceleration is realized through the gravitational leakage to the extra dimension,
but the joint analysis using the data of SN Ia, BAO, and CMB shift parameters shows that the model is in
tension with observations. Moreover the linear perturbation theory beyond the Vainshtein radius shows that
the model contains a ghost mode with the effective Brans-Dicke parameter ωBD smaller than −3/2. However,
the modification of the DGP model like Cascading gravity can alleviate this problem.
• (iv) In covariant Galileon gravity there is a tracker that attracts solutions with different initial conditions to a
common trajectory. The joint observational constraints at the background cosmology shows that the late-time
tracking solutions are favored from the data. Cosmological perturbations in (modified) Galileon theories exhibit
peculiar features because of non-trivial relations between the effective gravitational potential ψ and the matter
perturbation δm .
• (v) The dark energy models described by a Gauss-Bonnet term with a scalar coupling F (φ)G do not satisfy
both cosmological and local gravity constraints. The generalized Gauss-Bonnet model in which the Lagrangian
47
density is given by R/2 + f (G) is plagued by a serious problem of the Laplacian instability in the presence of
matter fluids. There are some viable Lorentz-violating models of dark energy in which the phantom equation
of state can be realized without having ghosts, tachyons, and Laplacian instabilities.
We hope to find some signatures for the modification of gravity in future high-precision observations. This will shed
new light on the nature of dark energy.
Acknowledgments
The author thanks the organizers of Second TRR33 Winter School held in Passo del Tonale (Italy) for inviting him
to have a series of lectures about modified gravity models of dark energy. He is also grateful to Georg Wolschin for
providing him an opportunity to write a lecture note in a book published by Springer. This note is one of the five
chapters in the book “Lectures on Cosmology: Accelerated Expansion of the Universe (Lecture Notes in Physics)”.
[1] A. G. Riess et al., Astron. J. 116, 1009 (1998); Astron. J. 117, 707 (1999).
[2] S. Perlmutter et al., Astrophys. J. 517, 565 (1999).
[3] D. N. Spergel et al., Astrophys. J. Suppl. 148, 175 (2003).
[4] E. Komatsu et al., arXiv:1001.4538 [astro-ph.CO].
[5] D. J. Eisenstein et al. [SDSS Collaboration], Astrophys. J. 633, 560 (2005).
[6] W. J. Percival et al., Mon. Not. Roy. Astron. Soc. 401, 2148 (2010).
[7] V. Sahni and A. A. Starobinsky, Int. J. Mod. Phys. D 9, 373 (2000).
[8] S. M. Carroll, Living Rev. Rel. 4, 1 (2001).
[9] T. Padmanabhan, Phys. Rept. 380, 235 (2003).
[10] P. J. E. Peebles and B. Ratra, Rev. Mod. Phys. 75, 559 (2003).
[11] V. Sahni, Lect. Notes Phys. 653, 141 (2004).
[12] E. J. Copeland, M. Sami and S. Tsujikawa, Int. J. Mod. Phys. D 15, 1753 (2006).
[13] S. Nojiri and S. D. Odintsov, eConf C0602061, 06 (2006) [Int. J. Geom. Meth. Mod. Phys. 4, 115 (2007)].
[14] R. P. Woodard, Lect. Notes Phys. 720, 403 (2007).
[15] R. Durrer and R. Maartens, Gen. Rel. Grav. 40, 301 (2008).
[16] F. S. N. Lobo, arXiv:0807.1640 [gr-qc].
[17] T. P. Sotiriou and V. Faraoni, Rev. Mod. Phys. 82, 451 (2010).
[18] P. Brax, arXiv:0912.3610 [astro-ph.CO].
[19] A. De Felice and S. Tsujikawa, Living Rev. Rel. 13, 3 (2010).
[20] S. Tsujikawa, arXiv:1004.1493 [astro-ph.CO] (Published in ’Dark Matter and Dark Energy: A Challenge for Modern
Cosmology’ Series: Astrophysics and Space Science Library, Vol. 370, ISBN: 978-90-481-8684-6, 2011).
[21] S. Weinberg, Rev. Mod. Phys. 61, 1 (1989).
[22] Y. Fujii, Phys. Rev. D 26, 2580 (1982).
[23] L. H. Ford, Phys. Rev. D 35, 2339 (1987).
[24] C. Wetterich, Nucl. Phys B. 302, 668 (1988).
[25] B. Ratra and J. Peebles, Phys. Rev D 37, 321 (1988).
[26] R. R. Caldwell, R. Dave and P. J. Steinhardt, Phys. Rev. Lett. 80, 1582 (1998).
[27] T. Chiba, T. Okabe and M. Yamaguchi, Phys. Rev. D 62, 023511 (2000).
[28] C. Armendariz-Picon, V. F. Mukhanov and P. J. Steinhardt, Phys. Rev. Lett. 85, 4438 (2000).
[29] S. M. Carroll, Phys. Rev. Lett. 81, 3067 (1998).
[30] C. F. Kolda and D. H. Lyth, Phys. Lett. B 458, 197 (1999).
[31] S. Capozziello, Int. J. Mod. Phys. D 11, 483, (2002).
[32] S. Capozziello, V. F. Cardone, S. Carloni and A. Troisi, Int. J. Mod. Phys. D, 12, 1969 (2003).
[33] S. M. Carroll, V. Duvvuri, M. Trodden and M. S. Turner, Phys. Rev. D 70, 043528 (2004).
[34] S. Nojiri and S. D. Odintsov, Phys. Rev. D 68, 123512 (2003).
[35] L. Amendola, Phys. Rev. D 60, 043501 (1999).
[36] J. P. Uzan, Phys. Rev. D 59, 123510 (1999).
[37] T. Chiba, Phys. Rev. D 60, 083508 (1999).
[38] N. Bartolo and M. Pietroni, Phys. Rev. D 61 023518 (2000).
[39] F. Perrotta, C. Baccigalupi and S. Matarrese, Phys. Rev. D 61, 023507 (2000).
[40] A. Riazuelo and J. P. Uzan, Phys. Rev. D 66, 023525 (2002).
[41] G. R. Dvali, G. Gabadadze and M. Porrati, Phys. Lett. B 485, 208 (2000).
[42] A. Nicolis, R. Rattazzi and E. Trincherini, Phys. Rev. D 79, 064036 (2009).
[43] S. Nojiri, S. D. Odintsov and M. Sasaki, Phys. Rev. D 71, 123509 (2005).
[44] S. Nojiri and S. D. Odintsov, Phys. Lett. B 631, 1 (2005).
48
[45] M. E. Soussa and R. P. Woodard, Gen. Rel. Grav. 36, 855 (2004).
[46] G. Allemandi, A. Borowiec and M. Francaviglia, Phys. Rev. D 70, 103503 (2004).
[47] D. A. Easson, Int. J. Mod. Phys. A 19, 5343 (2004).
[48] R. Dick, Gen. Rel. Grav. 36, 217 (2004).
[49] S. Carloni, P. K. S. Dunsby, S. Capozziello and A. Troisi, Class. Quant. Grav. 22, 4839 (2005).
[50] A. W. Brookfield, C. van de Bruck and L. M. H. Hall, Phys. Rev. D 74, 064028 (2006).
[51] T. Chiba, Phys. Lett. B 575, 1 (2003).
[52] A. D. Dolgov and M. Kawasaki, Phys. Lett. B 573, 1 (2003).
[53] G. J. Olmo, Phys. Rev. D 72, 083505 (2005).
[54] I. Navarro and K. Van Acoleyen, JCAP 0702, 022 (2007).
[55] S. M. Carroll, I. Sawicki, A. Silvestri and M. Trodden, New J. Phys. 8, 323 (2006).
[56] R. Bean, D. Bernat, L. Pogosian, A. Silvestri and M. Trodden, Phys. Rev. D 75, 064020 (2007).
[57] Y. S. Song, W. Hu and I. Sawicki, Phys. Rev. D 75, 044004 (2007).
[58] I. Sawicki and W. Hu, Phys. Rev. D 75, 127502 (2007).
[59] L. Amendola, D. Polarski and S. Tsujikawa, Phys. Rev. Lett. 98, 131302 (2007).
[60] L. Amendola, D. Polarski and S. Tsujikawa, Int. J. Mod. Phys. D 16, 1555 (2007).
[61] T. Faulkner, M. Tegmark, E. F. Bunn and Y. Mao, Phys. Rev. D 76, 063505 (2007).
[62] L. Amendola, R. Gannouji, D. Polarski and S. Tsujikawa, Phys. Rev. D 75, 083504 (2007).
[63] B. Li and J. D. Barrow, Phys. Rev. D 75, 084010 (2007).
[64] L. Amendola and S. Tsujikawa, Phys. Lett. B 660, 125 (2008).
[65] S. Fay, S. Nesseris and L. Perivolaropoulos, Phys. Rev. D 76, 063504 (2007).
[66] L. Pogosian and A. Silvestri, Phys. Rev. D 77, 023503 (2008).
[67] N. Deruelle, M. Sasaki and Y. Sendouda, Phys. Rev. D 77, 124024 (2008).
[68] W. Hu and I. Sawicki, Phys. Rev. D 76, 064004 (2007).
[69] A. A. Starobinsky, JETP Lett. 86, 157 (2007).
[70] S. A. Appleby and R. A. Battye, Phys. Lett. B 654, 7 (2007).
[71] S. Tsujikawa, Phys. Rev. D 77, 023507 (2008).
[72] E. V. Linder, Phys. Rev. D 80, 123528 (2009).
[73] C. Brans and R. H. Dicke, Phys. Rev. 124, 925 (1961).
[74] J. O’Hanlon, Phys. Rev. Lett. 29, 137 (1972).
[75] L. Amendola, Phys. Rev. D 62, 043511 (2000).
[76] S. Tsujikawa, K. Uddin, S. Mizuno, R. Tavakol and J. Yokoyama, Phys. Rev. D 77, 103009 (2008).
[77] J. A. R. Cembranos, Phys. Rev. D 73, 064029 (2006).
[78] S. Capozziello and S. Tsujikawa, Phys. Rev. D 77, 107501 (2008).
[79] R. Gannouji, B. Moraes, D. F. Mota, D. Polarski, S. Tsujikawa and H. A. Winther, Phys. Rev. D 82, 124006 (2010).
[80] P. Brax, C. van de Bruck, A. C. Davis and D. J. Shaw, Phys. Rev. D 78, 104021 (2008).
[81] J. Khoury and A. Weltman, Phys. Rev. Lett. 93, 171104 (2004).
[82] J. Khoury and A. Weltman, Phys. Rev. D 69, 044026 (2004).
[83] C. Deffayet, G. R. Dvali, G. Gabadadze and A. I. Vainshtein, Phys. Rev. D 65, 044026 (2002).
[84] M. Porrati, Phys. Lett. B 534, 209 (2002).
[85] A. I. Vainshtein, Phys. Lett. B 39, 393 (1972).
[86] A. Nicolis and R. Rattazzi, JHEP 0406, 059 (2004).
[87] K. Koyama and R. Maartens, JCAP 0601, 016 (2006).
[88] D. Gorbunov, K. Koyama and S. Sibiryakov, Phys. Rev. D 73, 044016 (2006).
[89] I. Sawicki and S. M. Carroll, arXiv:astro-ph/0510364.
[90] M. Fairbairn and A. Goobar, Phys. Lett. B 642 , 432 (2006).
[91] R. Maartens and E. Majerotto, Phys. Rev. D 74, 023004 (2006).
[92] U. Alam and V. Sahni, Phys. Rev. D 73, 084024 (2006).
[93] Y. S. Song, I. Sawicki and W. Hu, Phys. Rev. D 75, 064003 (2007).
[94] J. Q. Xia, Phys. Rev. D 79, 103527 (2009).
[95] C. Deffayet, G. Esposito-Farese and A. Vikman, Phys. Rev. D 79, 084003 (2009).
[96] C. Deffayet, S. Deser and G. Esposito-Farese, Phys. Rev. D 80, 064015 (2009).
[97] A. De Felice and S. Tsujikawa, Phys. Rev. Lett. 105, 111301 (2010).
[98] A. De Felice and S. Tsujikawa, arXiv:1008.4236 [hep-th].
[99] C. Schimd, J. P. Uzan and A. Riazuelo, Phys. Rev. D 71, 083512 (2005).
[100] M. Ishak, A. Upadhye and D. N. Spergel, Phys. Rev. D 74, 043513 (2006).
[101] L. Knox, Y. S. Song and J. A. Tyson, Phys. Rev. D 74, 023512 (2006).
[102] D. Huterer and E. V. Linder, Phys. Rev. D 75, 023519 (2007).
[103] P. Zhang, M. Liguori, R. Bean and S. Dodelson, Phys. Rev. Lett. 99, 141302 (2007).
[104] S. Wang, L. Hui, M. May and Z. Haiman, Phys. Rev. D 76, 063503 (2007).
[105] B. Jain and P. Zhang, Phys. Rev. D 78, 063503 (2008).
[106] S. F. Daniel, R. R. Caldwell, A. Cooray and A. Melchiorri, Phys. Rev. D 77, 103513 (2008).
[107] E. Bertschinger and P. Zukin, Phys. Rev. D 78, 024015 (2008).
[108] S. Carloni, P. K. S. Dunsby and A. Troisi, Phys. Rev. D 77, 024024 (2008).
49
[109] K. N. Ananda, S. Carloni and P. K. S. Dunsby, Class. Quant. Grav. 26, 235018 (2009).
[110] G. B. Zhao, L. Pogosian, A. Silvestri and J. Zylberberg, Phys. Rev. D 79, 083513 (2009).
[111] Y. S. Song and K. Koyama, JCAP 0901, 048 (2009).
[112] Y. S. Song and O. Dore, JCAP 0903, 025 (2009).
[113] S. A. Thomas, F. B. Abdalla and J. Weller, Mon. Not. Roy. Astron. Soc. 395, 197 (2009).
[114] A. Borisov and B. Jain, Phys. Rev. D 79, 103506 (2009).
[115] J. Guzik, B. Jain and M. Takada, Phys. Rev. D 81, 023503 (2010).
[116] P. Brax, C. van de Bruck, A. C. Davis and D. Shaw, JCAP 1004, 032 (2010).
[117] R. Bean and M. Tangmatitham, Phys. Rev. D 81, 083534 (2010).
[118] R. Reyes, R. Mandelbaum, U. Seljak, T. Baldauf, J. E. Gunn, L. Lombriser and R. E. Smith, Nature 464, 256 (2010).
[119] G. B. Zhao et al., Phys. Rev. D 81, 103510 (2010).
[120] S. F. Daniel et al., Phys. Rev. D 81, 123508 (2010).
[121] S. F. Daniel and E. V. Linder, Phys. Rev. D 82, 103523 (2010).
[122] Z. Girones, A. Marchetti, O. Mena, C. Pena-Garay and N. Rius, JCAP 1011, 004 (2010).
[123] T. Narikawa and K. Yamamoto, Phys. Rev. D 81, 043528 (2010).
[124] K. Yamamoto, G. Nakamura, G. Hutsi, T. Narikawa and T. Sato, Phys. Rev. D 81, 103517 (2010).
[125] L. Lombriser, A. Slosar, U. Seljak and W. Hu, arXiv:1003.3009 [astro-ph.CO].
[126] Y. S. Song, G. B. Zhao, D. Bacon, K. Koyama, R. C. Nichol and L. Pogosian, arXiv:1011.2106 [astro-ph.CO].
[127] A. Palatini, Rend. Circ. Mat. Palermo 43, 203 (1919).
[128] F. W. Hehl and G. D. Kerling, Gen. Rel. Grav. 9, 691 (1978).
[129] T. P. Sotiriou and S. Liberati, Annals Phys. 322, 935 (2007).
[130] T. P. Sotiriou and S. Liberati, J. Phys. Conf. Ser. 68, 012022 (2007).
[131] S. Capozziello, R. Cianci, C. Stornaiolo and S. Vignolo, Class. Quant. Grav. 24, 6417 (2007).
[132] M. Ferraris, M. Francaviglia and I. Volovich, Class. Quant. Grav. 11, 1505 (1994).
[133] D. N. Vollick, Phys. Rev. D 68, 063510 (2003).
[134] D. N. Vollick, Class. Quant. Grav. 21, 3813 (2004).
[135] E. E. Flanagan, Class. Quant. Grav. 21, 417 (2003).
[136] E. E. Flanagan, Phys. Rev. Lett. 92, 071101 (2004).
[137] E. E. Flanagan, Class. Quant. Grav. 21, 3817 (2004).
[138] X. Meng and P. Wang, Class. Quant. Grav. 20, 4949 (2003).
[139] X. Meng and P. Wang, Class. Quant. Grav. 21, 951 (2004).
[140] X. H. Meng and P. Wang, Phys. Lett. B 584, 1 (2004).
[141] S. Nojiri and S. D. Odintsov, Gen. Rel. Grav. 36, 1765 (2004).
[142] T. P. Sotiriou, Class. Quant. Grav. 23, 1253 (2006).
[143] T. P. Sotiriou, Phys. Rev. D 73, 063515 (2006).
[144] M. Amarzguioui, O. Elgaroy, D. F. Mota and T. Multamaki, Astron. Astrophys. 454, 707 (2006).
[145] S. Fay, R. Tavakol and S. Tsujikawa, Phys. Rev. D 75, 063509 (2007).
[146] A. Iglesias, N. Kaloper, A. Padilla and M. Park, Phys. Rev. D 76, 104001 (2007).
[147] G. J. Olmo, Phys. Rev. Lett. 98, 061101 (2007).
[148] G. J. Olmo, Phys. Rev. D 77, 084021 (2008).
[149] G. J. Olmo, arXiv:0910.3734 [gr-qc].
[150] E. Barausse, T. P. Sotiriou and J. C. Miller, Class. Quant. Grav. 25, 105008 (2008).
[151] T. Koivisto and H. Kurki-Suonio, Class. Quant. Grav. 23, 2355 (2006).
[152] T. Koivisto, Phys. Rev. D 73, 083517 (2006).
[153] B. Li and M. C. Chu, Phys. Rev. D 74, 104010 (2006).
[154] B. Li, K. C. Chan and M. C. Chu, Phys. Rev. D 76, 024002 (2007).
[155] S. Tsujikawa, K. Uddin and R. Tavakol, Phys. Rev. D 77, 043007 (2008).
[156] A. A. Starobinsky, Phys. Lett. B 91, 99 (1980).
[157] S. V. Ketov, Phys. Lett. B 692, 272 (2010).
[158] S. V. Ketov and A. A. Starobinsky, arXiv:1011.0240 [hep-th].
[159] V. Muller, H. J. Schmidt and A. A. Starobinsky, Phys. Lett. B 202, 198 (1988).
[160] V. Faraoni, Phys. Rev. D 72, 061501 (2005).
[161] V. Faraoni and S. Nadeau, Phys. Rev. D 72, 124005 (2005).
[162] S. Capozziello, S. Nojiri, S. D. Odintsov and A. Troisi, Phys. Lett. B 639, 135 (2006).
[163] K. Bamba, C. Q. Geng and S. Tsujikawa, Phys. Lett. B 688, 101 (2010).
[164] D. F. Torres, Phys. Rev. D 66, 043522 (2002).
[165] B. Boisseau, G. Esposito-Farese, D. Polarski and A. A. Starobinsky, Phys. Rev. Lett. 85, 2236 (2000).
[166] G. Esposito-Farese and D. Polarski, Phys. Rev. D 63, 063504 (2001).
[167] H. Motohashi, A. A. Starobinsky and J. Yokoyama, Prog. Theor. Phys. 123, 887 (2010).
[168] K. Bamba, C. Q. Geng and C. C. Lee, JCAP 1008, 021 (2010).
[169] K. Bamba, C. Q. Geng and C. C. Lee, JCAP 1011, 001 (2010).
[170] K. Bamba, C. Q. Geng, S. Nojiri and S. D. Odintsov, Phys. Rev. D 79, 083014 (2009).
[171] K. Bamba, Open Astron. J. 3, 13 (2010).
[172] A. Dev, D. Jain, S. Jhingan, S. Nojiri, M. Sami and I. Thongkool, Phys. Rev. D 78, 083515 (2008).
50
[173] M. Martinelli, A. Melchiorri and L. Amendola, Phys. Rev. D 79, 123516 (2009).
[174] V. F. Cardone, A. Diaferio and S. Camera, arXiv:0907.4689 [astro-ph.CO].
[175] A. Ali, R. Gannouji, M. Sami and A. A. Sen, Phys. Rev. D 81, 104029 (2010).
[176] S. Capozziello, V. F. Cardone and A. Troisi, Phys. Rev. D 71, 043503 (2005).
[177] T. Multamaki and I. Vilja, Phys. Rev. D 73, 024018 (2006).
[178] A. de la Cruz-Dombriz and A. Dobado, Phys. Rev. D 74, 087501 (2006).
[179] X. Wu and Z. H. Zhu, Phys. Lett. B 660, 293 (2008).
[180] S. Carloni, R. Goswami and P. K. S. Dunsby, arXiv:1005.1840 [gr-qc].
[181] J. M. Bardeen, Phys. Rev. D 22, 1882 (1980).
[182] J. c. Hwang, Class. Quant. Grav. 14, 3327 (1997).
[183] A. De Felice, S. Mukohyama and S. Tsujikawa, Phys. Rev. D 82, 023524 (2010).
[184] G. J. Olmo, Phys. Rev. Lett. 95, 261102 (2005).
[185] G. J. Olmo, Phys. Rev. D 72, 083505 (2005).
[186] V. Faraoni, Phys. Rev. D 74, 023529 (2006).
[187] A. L. Erickcek, T. L. Smith and M. Kamionkowski, Phys. Rev. D 74, 121501 (2006).
[188] T. Chiba, T. L. Smith and A. L. Erickcek, Phys. Rev. D 75, 124014 (2007).
[189] K. Kainulainen, J. Piilonen, V. Reijonen and D. Sunhede, Phys. Rev. D 76, 024020 (2007).
[190] K. Kainulainen and D. Sunhede, Phys. Rev. D 78, 063511 (2008).
[191] K. i. Maeda, Phys. Rev. D 39, 3159 (1989).
[192] T. Tamaki and S. Tsujikawa, Phys. Rev. D 78, 084028 (2008).
[193] C. M. Will, Living Rev. Rel. 9, 3 (2005).
[194] T. Kobayashi and K. i. Maeda, Phys. Rev. D 78, 064019 (2008).
[195] T. Kobayashi and K. i. Maeda, Phys. Rev. D 79, 024009 (2009).
[196] S. Tsujikawa, T. Tamaki and R. Tavakol, JCAP 0905, 020 (2009).
[197] A. Upadhye and W. Hu, Phys. Rev. D 80, 064002 (2009).
[198] E. Babichev and D. Langlois, Phys. Rev. D 80, 121501 (2009).
[199] E. Babichev and D. Langlois, Phys. Rev. D 81, 124051 (2010).
[200] A. Cooney, S. DeDeo and D. Psaltis, Phys. Rev. D 82, 064033 (2010).
[201] B. Bertotti, L. Iess and P. Tortora, Nature 425, 374 (2003).
[202] L. Perivolaropoulos, JCAP 0510, 001 (2005).
[203] R. Gannouji, D. Polarski, A. Ranquet and A. A. Starobinsky, JCAP 0609, 016 (2006).
[204] S. Nesseris and L. Perivolaropoulos, JCAP 0701, 018 (2007).
[205] J. Martin, C. Schimd and J. P. Uzan, Phys. Rev. Lett. 96, 061303 (2006).
[206] A. Billyard, A. Coley and J. Ibanez, Phys. Rev. D 59, 023507 (1999).
[207] E. Gunzig, V. Faraoni, A. Figueiredo, T. M. Rocha and L. Brenig, Class. Quant. Grav. 17, 1783 (2000).
[208] V. Acquaviva and L. Verde, JCAP 0712, 001 (2007).
[209] N. Agarwal and R. Bean, Class. Quant. Grav. 25, 165001 (2008).
[210] L. Jarv, P. Kuusk and M. Saal, Phys. Rev. D 78, 083530 (2008).
[211] S. Carloni, S. Capozziello, J. A. Leach and P. K. S. Dunsby, Class. Quant. Grav. 25, 035008 (2008).
[212] P. Brax, C. van de Bruck, A. C. Davis, J. Khoury and A. Weltman, Phys. Rev. D 70, 123518 (2004).
[213] D. J. Kapner et al., Phys. Rev. Lett. 98, 021101 (2007).
[214] R. Nagata, T. Chiba and N. Sugiyama, Phys. Rev. D 69, 083512 (2004).
[215] L. Randall and R. Sundrum, Phys. Rev. Lett. 83, 3370 (1999).
[216] L. Randall and R. Sundrum, Phys. Rev. Lett. 83, 4690 (1999).
[217] G. R. Dvali and G. Gabadadze, Phys. Rev. D 63 (2001), 065007.
[218] C. Deffayet, Phys. Lett. B 502 (2001), 199.
[219] C. Deffayet, G. R. Dvali and G. Gabadadze, Phys. Rev. D 65 (2002), 044023.
[220] V. Sahni and Y. Shtanov, JCAP 0311, 014 (2003).
[221] P. Binetruy, C. Deffayet and D. Langlois, Nucl. Phys. B 565, 269 (2000).
[222] P. Binetruy, C. Deffayet, U. Ellwanger and D. Langlois, Phys. Lett. B 477, 285 (2000).
[223] T. Shiromizu, K. i. Maeda and M. Sasaki, Phys. Rev. D 62, 024012 (2000).
[224] A. G. Riess et al. [Supernova Search Team Collaboration], Astrophys. J. 607, 665 (2004).
[225] P. Astier et al. [The SNLS Collaboration], Astron. Astrophys. 447, 31 (2006).
[226] D. N. Spergel et al. [WMAP Collaboration], Astrophys. J. Suppl. 170, 377 (2007).
[227] J. Q. Xia, Phys. Rev. D 79, 103527 (2009).
[228] A. Gruzinov, New Astron. 10, 311 (2005).
[229] M. A. Luty, M. Porrati and R. Rattazzi, JHEP 0309, 029 (2003).
[230] M. Kolanovic, M. Porrati and J. W. Rombouts, Phys. Rev. D 68, 064018 (2003).
[231] M. Kolanovic, Phys. Rev. D 67, 106002 (2003).
[232] C. de Rham, G. Dvali, S. Hofmann, J. Khoury, O. Pujolas, M. Redi and A. J. Tolley, Phys. Rev. Lett. 100, 251603 (2008).
[233] C. de Rham, S. Hofmann, J. Khoury and A. J. Tolley, JCAP 0802, 011 (2008).
[234] N. Agarwal, R. Bean, J. Khoury and M. Trodden, Phys. Rev. D 81, 084020 (2010).
[235] C. de Rham, J. Khoury and A. J. Tolley, Phys. Rev. D 81, 124027 (2010).
[236] R. Gannouji and M. Sami, Phys. Rev. D 82, 024011 (2010).
51
[237] N. Chow and J. Khoury, Phys. Rev. D 80, 024037 (2009).

[238] F. P. Silva and K. Koyama, Phys. Rev. D 80, 121301 (2009).
[239] T. Kobayashi, H. Tashiro and D. Suzuki, Phys. Rev. D 81, 063513 (2010).
[240] T. Kobayashi, Phys. Rev. D 81, 103533 (2010).
[241] C. de Rham and A. J. Tolley, JCAP 1005, 015 (2010).
[242] A. De Felice and S. Tsujikawa, JCAP 1007, 024 (2010).
[243] P. Creminelli, A. Nicolis and E. Trincherini, JCAP 1011, 021 (2010).
[244] A. Padilla, P. M. Saffin and S. Y. Zhou, JHEP 1012, 031 (2010); arXiv:1008.0745 [hep-th].
[245] C. Deffayet, S. Deser and G. Esposito-Farese, Phys. Rev. D 82, 061501 (2010).
[246] T. Kobayashi, M. Yamaguchi and J. Yokoyama, Phys. Rev. Lett. 105, 231302 (2010).
[247] C. Deffayet, O. Pujolas, I. Sawicki and A. Vikman, JCAP 1010, 026 (2010).
[248] K. Hinterbichler, M. Trodden and D. Wesley, arXiv:1008.1305 [hep-th].
[249] A. Ali, R. Gannouji and M. Sami, arXiv:1008.1588 [astro-ph.CO].
[250] M. Andrews, K. Hinterbichler, J. Khoury and M. Trodden, arXiv:1008.4128 [hep-th].
[251] G. L. Goon, K. Hinterbichler and M. Trodden, arXiv:1008.4580 [hep-th].
[252] S. Mizuno and K. Koyama, Phys. Rev. D 82, 103518 (2010).
[253] C. Burrage, C. de Rham, D. Seery and A. J. Tolley, arXiv:1009.2497 [hep-th].
[254] E. Babichev, arXiv:1009.2921 [hep-th].
[255] S. Y. Zhou, arXiv:1011.0863 [hep-th].
[256] R. Kimura and K. Yamamoto, arXiv:1011.2006 [astro-ph.CO].
[257] K. Kamada, T. Kobayashi, M. Yamaguchi and J. Yokoyama, arXiv:1012.4238 [astro-ph.CO].
[258] K. Hirano, arXiv: 1012.5451 [astro-ph.CO].
[259] S. Nesseris, A. De Felice and S. Tsujikawa, arXiv:1010.0407 [astro-ph.CO] (Physical Review D, to appear).
[260] L. D. Faddeev and R. Jackiw, Phys. Rev. Lett. 60, 1692 (1988).
[261] A. De Felice and T. Suyama, JCAP 0906, 034 (2009); Phys. Rev. D 80, 083523 (2009).
[262] M. Hicken et al., Astrophys. J. 700, 1097 (2009).
[263] R. Amanullah et al., Astrophys. J. 716, 712 (2010).
[264] H. Akaike, IEEE Trans. Auto. Control, 19, 716 (1974).
[265] A. R. Liddle, Mon. Not. Roy. Astron. Soc. 351, L49 (2004).
[266] G. Schwarz, Annals of Statistics, 5, 461 (1978).
[267] E. Kretschmann, Ann. Phys, 53, 575-614 (1917).
[268] K. S. Stelle, Gen. Rel. Grav. 9, 353 (1978).
[269] N. H. Barth and S. M. Christensen, Phys. Rev. D 28, 1876 (1983).
[270] M. Gasperini and G. Veneziano, Phys. Rept. 373, 1 (2003).
[271] A. Nunez and S. Solganik, Phys. Lett. B 608, 189 (2005).
[272] S. M. Carroll, A. De Felice, V. Duvvuri, D. A. Easson, M. Trodden and M. S. Turner, Phys. Rev. D 71, 063513 (2005).
[273] G. Calcagni, S. Tsujikawa and M. Sami, Class. Quant. Grav. 22, 3977 (2005).
[274] M. Sami, A. Toporensky, P. V. Tretjakov and S. Tsujikawa, Phys. Lett. B 619, 193 (2005).
[275] T. Chiba, JCAP 0503, 008 (2005).
[276] O. Mena, J. Santiago and J. Weller, Phys. Rev. Lett. 96, 041103 (2006).
[277] J. D. Barrow and S. Hervik, Phys. Rev. D 74, 124017 (2006).
[278] G. Cognola, M. Gastaldi and S. Zerbini, Int. J. Theor. Phys. 47, 898 (2008).
[279] L. M. Sokolowski, Class. Quant. Grav. 24, 3391 (2007).
[280] A. De Felice, M. Hindmarsh and M. Trodden, JCAP 0608, 005 (2006).
[281] G. Calcagni, B. de Carlos and A. De Felice, Nucl. Phys. B 752, 404 (2006).
[282] T. Koivisto and D. F. Mota, Phys. Lett. B 644, 104 (2007).
[283] S. Tsujikawa and M. Sami, JCAP 0701, 006 (2007).
[284] T. Koivisto and D. F. Mota, Phys. Rev. D 75, 023518 (2007).
[285] I. P. Neupane, Class. Quant. Grav. 23, 7493 (2006).
[286] B. M. Leith and I. P. Neupane, JCAP 0705, 019 (2007).
[287] A. K. Sanyal, Phys. Lett. B 645, 1 (2007).
[288] S. Kawai, M. a. Sakagami and J. Soda, Phys. Lett. B 437, 284 (1998).
[289] Z. K. Guo, N. Ohta and S. Tsujikawa, Phys. Rev. D 75, 023520 (2007).
[290] L. Amendola, C. Charmousis and S. C. Davis, JCAP 0612, 020 (2006).
[291] G. Cognola, E. Elizalde, S. Nojiri, S. D. Odintsov and S. Zerbini, Phys. Rev. D 73, 084007 (2006).
[292] B. Li, J. D. Barrow and D. F. Mota, Phys. Rev. D 76, 044027 (2007).
[293] A. De Felice and S. Tsujikawa, Phys. Lett. B 675, 1 (2009).
[294] S. Y. Zhou, E. J. Copeland and P. M. Saffin, JCAP 0907, 009 (2009).
[295] M. Mohseni, Phys. Lett. B 682, 89 (2009).
[296] J. Moldenhauer, M. Ishak, J. Thompson and D. A. Easson, Phys. Rev. D 81, 063514 (2010).
[297] A. De Felice and S. Tsujikawa, Phys. Rev. D 80, 063516 (2009).
[298] A. De Felice, D. F. Mota and S. Tsujikawa, Phys. Rev. D 81, 023532 (2010).
[299] A. De Felice, J. M. Gerard and T. Suyama, Phys. Rev. D 82, 063526 (2010).
[300] A. De Felice and T. Suyama, arXiv:1010.3886 [astro-ph.CO].
52
[301] M. Alimohammadi and A. Ghalee, Phys. Rev. D 79, 063006 (2009).

[302] M. Alimohammadi and A. Ghalee, Phys. Rev. D 80, 043006 (2009).
[303] E. Elizalde, R. Myrzakulov, V. V. Obukhov and D. Saez-Gomez, Class. Quant. Grav. 27, 095007 (2010).
[304] A. De Felice and T. Tanaka, Prog. Theor. Phys. 124, 503 (2010).
[305] N. Arkani-Hamed, H. C. Cheng, M. A. Luty and S. Mukohyama, JHEP 0405, 074 (2004).
[306] V. A. Rubakov, Theor. Math. Phys. 149, 1651 (2006) [arXiv:hep-th/0604153].
[307] M. Libanov, V. Rubakov, E. Papantonopoulos, M. Sami and S. Tsujikawa, JCAP 0708, 010 (2007).
[308] V. A. Rubakov and P. G. Tinyakov, Phys. Usp. 51, 759 (2008).
[309] B. M. Gripaios, JHEP 0410, 069 (2004).
[310] M. V. Libanov and V. A. Rubakov, JHEP 0508, 001 (2005).
[311] P. Horava, Phys. Rev. D 79, 084008 (2009).
[312] S. M. Carroll and E. A. Lim, Phys. Rev. D 70, 123525 (2004).
[313] J. M. Cline and L. Valcarcel, JHEP 0403, 032 (2004).
[314] N. Arkani-Hamed, H. C. Cheng, M. Luty and J. Thaler, JHEP 0507, 029 (2005).
[315] H. C. Cheng, M. A. Luty, S. Mukohyama and J. Thaler, JHEP 0605, 076 (2006).
[316] S. Kanno and J. Soda, Phys. Rev. D 74, 063505 (2006).
[317] P. G. Ferreira, B. M. Gripaios, R. Saffari and T. G. Zlosnik, Phys. Rev. D 75, 044014 (2007).
[318] F. Piazza and S. Tsujikawa, JCAP 0407, 004 (2004).
[319] E. N. Saridakis, Eur. Phys. J. C 67, 229 (2010).
[320] M. i. Park, JCAP 1001, 001 (2010).
[321] H. Kodama and M. Sasaki, Prog. Theor. Phys. Suppl. 78, 1 (1984).
[322] V. F. Mukhanov, H. A. Feldman and R. H. Brandenberger, Phys. Rept. 215, 203 (1992).
[323] B. A. Bassett, S. Tsujikawa and D. Wands, Rev. Mod. Phys. 78, 537 (2006).
[324] L. A. Kofman, V. F. Mukhanov and D. Y. Pogosian, Sov. Phys. JETP 66, 433 (1987).
[325] J. c. Hwang and H. r. Noh, Phys. Rev. D 65, 023512 (2002).
[326] J. C. Hwang and H. Noh, Phys. Rev. D 71, 063536 (2005).
[327] A. A. Starobinsky, JETP Lett. 68, 757 (1998) [Pisma Zh. Eksp. Teor. Fiz. 68, 721 (1998)].
[328] S. Tsujikawa, Phys. Rev. D 76, 023514 (2007).
[329] J. c. Hwang, H. Noh and C. G. Park, arXiv:1012.0885 [astro-ph.CO].
[330] A. de la Cruz-Dombriz, A. Dobado and A. L. Maroto, Phys. Rev. D 77, 123515 (2008).
[331] H. Motohashi, A. A. Starobinsky and J. Yokoyama, Int. J. Mod. Phys. D 18, 1731 (2009).
[332] S. A. Appleby and R. A. Battye, JCAP 0805, 019 (2008).
[333] A. V. Frolov, Phys. Rev. Lett. 101, 061103 (2008).
[334] S. A. Appleby, R. A. Battye and A. A. Starobinsky, JCAP 1006, 005 (2010).
[335] V. Miranda, S. E. Joras, I. Waga and M. Quartin, Phys. Rev. Lett. 102, 221101 (2009).
[336] I. Thongkool, M. Sami, R. Gannouji and S. Jhingan, Phys. Rev. D 80, 043523 (2009).
[337] A. de la Cruz-Dombriz, A. Dobado and A. L. Maroto, Phys. Rev. Lett. 103, 179001 (2009).
[338] S. Nojiri and S. D. Odintsov, Phys. Lett. B 657, 238 (2007).
[339] S. Nojiri and S. D. Odintsov, Phys. Rev. D 77, 026007 (2008).
[340] S. Tsujikawa and T. Tatekawa, Phys. Lett. B 665, 325 (2008).
[341] P. J. E. Peebles, Large-Scale Structure of the Universe, Princeton University Press (1980).
[342] L. M. Wang and P. J. Steinhardt, Astrophys. J. 508, 483 (1998).
[343] E. V. Linder, Phys. Rev. D 72, 043529 (2005).
[344] R. Gannouji, B. Moraes and D. Polarski, JCAP 0902, 034 (2009).
[345] S. Tsujikawa, R. Gannouji, B. Moraes and D. Polarski, Phys. Rev. D 80, 084044 (2009).
[346] H. Oyaizu, Phys. Rev. D 78, 123523 (2008).
[347] H. Oyaizu, M. Lima and W. Hu, Phys. Rev. D 78, 123524 (2008).
[348] F. Schmidt, M. V. Lima, H. Oyaizu and W. Hu, Phys. Rev. D 79, 083518 (2009).
[349] G. B. Zhao, B. Li and K. Koyama, arXiv:1011.1257 [astro-ph.CO].
[350] H. F. Stabenau and B. Jain, Phys. Rev. D 74, 084007 (2006).
[351] I. Laszlo and R. Bean, Phys. Rev. D 77, 024048 (2008).
[352] W. Hu and I. Sawicki, Phys. Rev. D 76, 104043 (2007).
[353] K. Koyama, A. Taruya and T. Hiramatsu, Phys. Rev. D 79, 123512 (2009).
[354] T. Tatekawa and S. Tsujikawa, JCAP 0809, 009 (2008).
[355] L. Amendola, M. Kunz and D. Sapone, JCAP 0804, 013 (2008).
[356] Y. S. Song, H. Peiris and W. Hu, Phys. Rev. D 76, 063517 (2007).
[357] F. Schmidt, Phys. Rev. D 78, 043002 (2008).
[358] P. McDonald et al., astro-ph/0407377.
[359] M. Viel and M. G. Haehnelt, Mon. Not. Roy. Astron. Soc. 365, 231 (2006).
[360] A. Lue, R. Scoccimarro and G. D. Starkman, Phys. Rev. D 69, 124015 (2004).
[361] K. Koyama and F. P. Silva, Phys. Rev. D 75, 084040 (2007).
[362] K. Hinterbichler, A. Nicolis and M. Porrati, JHEP 0909, 089 (2009).
[363] K. Yamamoto, D. Parkinson, T. Hamana, R. C. Nichol and Y. Suto, Phys. Rev. D 76, 023504 (2007).
[364] A. De Felice, R. Kase and S. Tsujikawa, arXiv:1011.6132 [astro-ph.CO].

Shinji Tsujikawa - Modified Gravity Models of Dark Energy

Uploaded by

Copyright:

Available Formats

Shinji Tsujikawa - Modified Gravity Models of Dark Energy

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Shinji Tsujikawa - Modified Gravity Models of Dark Energy

Uploaded by

Copyright:

Available Formats

Modified gravity models of dark energy

II. f (R) gravity 3

III. Scalar-tensor gravity 12

IV. DGP model 18

VI. Other modified gravity models of dark energy 30

VII. Observational signatures of modified gravity 33

II. f (R) GRAVITY

We start with the action in f (R) gravity:

3 F (R) + F (R)R − 2f (R) = κ2 T , (3)

F (R)R − 2f (R) = 0 . (4)

A. Cosmological dynamics in f (R) gravity

ds2 = −dt2 + a(t)2 dx2 , (5)

3F H 2 = κ2 (ρm + ρr ) + (F R − f )/2 − 3H Ḟ , (6)

where the Ricci scalar is given by

R = 6(2H 2 + Ḣ) . (8)

Let us introduce the following dimensionless variables:

together with the density parameters

P1 : (x1 , x2 , x3 ) = (0, −1, 2), Ωm = 0, weff = −1 , (18)

0 < m(r = −2) ≤ 1 . (27)

f (R) = R − αRn (α > 0, 0 < n < 1), (28)

3AH 2 = κ2 (ρm + ρr + ρDE ) , (29)

where A is some constant and

κ2 ρDE ≡ (1/2)(F R − f ) − 3H Ḟ + 3H 2 (A − F ) , (31)

ρ̇DE + 3H(ρDE + PDE ) = 0 . (33)

2AḢ + 3AH 2 + κ2 ρr /3 weff

B. Conditions for the avoidance of ghosts and tachyonic instabilities

where α, β, ψ, γ are scalar metric perturbations [181].

1 zs′′ Q̇2s Q̈s 3H Q̇s

where we have eliminated the term Ḣ by using the background equations.

f,RR > 0 . (44)

f (R) ≃ R − µRc [1 − (R2 /Rc2 )−n ] for R ≫ Rc , (47)

which gives m(r) = C(−r − 1)2n+1 .

(C) f (R) = R − µRc tanh (R/Rc ) with µ > 0, Rc > 0 . (48)

C. Local gravity constraints on f (R) gravity models

g̃µν = Ω2 gµν , Ω2 = F . (49)

Veff (φ) = V (φ) + eQφ ρ∗ . (54)

ǫth,⊕ < 8.8 × 10−7 /|Q| . (60)

|φB,⊕ | < 3.7 × 10−15 . (61)

If ρA ≫ ρB , then one has |φB | ≫ |φA |.

n > 0.9 . (67)

III. SCALAR-TENSOR GRAVITY

A. Scalar-tensor theories and the matter coupling in the Einstein frame

The general action for scalar-tensor theories can be written as

B. Cosmological dynamics in Brans-Dicke theory

where N = ln a, λ = −U,φ /U , and

The effective equation of state of the system is given by by weff = −1 − 2Ḣ/(3H 2 ).

(x1 , x2 ) = (0, 1) , Ωm = 0 , weff = −1 . (92)

In this case the slope of the potential, λ = −U,φ /U , is

U (φ) = U0 1 − C(1 − e−2Qφ )p

2Cp Qe−2Qφ (1 − e−2Qφ )p−1

dλ 4CpQ2 F (1 − F )p−2 [1 − pF − C(1 − F )p ]

F1 > 1/(2 − p) . (103)

C. Local gravity constraints on Brans-Dicke theory

U0 = 3H12 [2 + (p − 2)F1 ] /p . (107)

Plugging this into Eq. (106), it follows that

IV. DGP MODEL

The action of the DGP model is given by

ds2 = −n2 (τ, y)dτ 2 + a2 (τ, y) γij dxi dxj + dy 2 , (112)