NS66CH05-Schmidt ARI 5 September 2016 11:58

ANNUAL
REVIEWS Further
Click here to view this article's
online features:
• Download figures as PPT slides
• Navigate linked references
• Download citations Dark Energy Versus
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

• Explore related articles

• Search keywords
Modified Gravity
Austin Joyce,1 Lucas Lombriser,2 and Fabian Schmidt3
1
Enrico Fermi Institute and Kavli Institute for Cosmological Physics, University of Chicago,
Chicago, Illinois 60637; email: [email protected]
2
Institute for Astronomy, University of Edinburgh, Edinburgh EH9 3HJ, United Kingdom;
email: [email protected]
3
Max Planck Institute for Astrophysics, D-85748 Garching, Germany;
email: [email protected]

Annu. Rev. Nucl. Part. Sci. 2016. 66:95–122 Keywords

First published online as a Review in Advance on cosmology, dark energy, modified gravity, structure formation, large-scale
June 17, 2016
structure
The Annual Review of Nuclear and Particle Science
is online at nucl.annualreviews.org Abstract
This article’s doi: Understanding the reason for the observed accelerated expansion of the
10.1146/annurev-nucl-102115-044553
Universe represents one of the fundamental open questions in physics. In
Copyright  c 2016 by Annual Reviews. cosmology, a classification has emerged among physical models for this ac-
All rights reserved
celeration, distinguishing between dark energy and modified gravity. In this
review, we provide a brief overview of models in both categories as well as
their phenomenology and characteristic observable signatures in cosmology.
We also introduce a rigorous distinction between dark energy and modified
gravity based on the strong and weak equivalence principles.

95
NS66CH05-Schmidt ARI 5 September 2016 11:58

Contents
1. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2. OVERVIEW: DARK ENERGY AND MODIFIED GRAVITY . . . . . . . . . . . . . . . . . . 97
2.1. Dark Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
2.2. Modified Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
2.3. Drawing a Boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3. COSMOLOGICAL PHENOMENOLOGY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

3.1. Background Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

3.2. Structure Formation with Quintessence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.3. Beyond Quintessence Dark Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.4. Structure Formation in Modified Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4. COSMOLOGICAL TESTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.1. Observables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.2. Consistency Tests: Geometry and Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.3. Parameterized Tests of Gravity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.4. Cosmic Degeneracies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.5. Targeted Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
5. OUTLOOK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

1. INTRODUCTION
The accelerated expansion of the Universe, discovered in 1998 (1, 2), has raised fascinating ques-
tions for cosmology and physics as a whole. Elucidating the physics behind the acceleration is one
of the main science drivers for a major experimental effort using large-scale structure (LSS) surveys.
All current observations are consistent with a cosmological constant (CC); whereas the CC is in
some sense the most economical possibility, it has its own theoretical and naturalness problems (3,
4), so it is worthwhile to consider alternatives. A convenient classification scheme has emerged
in the field of cosmology, separating different physical scenarios of accelerating cosmologies into
the categories of dark energy and modified gravity. Essentially, dark energy models modify the
stress-energy content of the Universe, adding another component with equation of state w  −1.
That is, we modify the right-hand side of the Einstein equations. The modified gravity category
corresponds to modifying the left-hand side—that is, general relativity (GR) itself—for example,
by modifying the Einstein–Hilbert action. This review intends to provide an overview of these
two categories, and describe the cosmological observables and tests that are able to distinguish
experimentally between these two scenarios of physics driving the acceleration.
As discussed below, although there are models that unambiguously belong to one category or
the other, in reality there is a continuum of models between the two extremes of “pure” dark
energy and modified gravity such that a strict division into these two categories is to some extent
a matter of personal preference. In this review, we base the distinction in theory space upon
the strong equivalence principle (SEP). A compatible, slightly less rigorous, but more practical
observational distinction in cosmology is to call modified gravity those models that feature a light
universally coupled degree of freedom mediating a fifth force.
Throughout this review, we restrict ourselves to observations on extragalactic scales, which
translates to scales of order 1 Mpc and above. Furthermore, we consider only scenarios for which

96 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

effects become relevant at late times. That is, we disregard the possible presence of such effects in
the early Universe (often referred to as early dark energy).
We begin in Section 2 with an overview of the landscape of dark energy and modified gravity
Quintessence
models. Section 3 then describes the cosmological phenomenlogy of these models. Section 4 models: the
provides a summary of the observables and experimental methods used in cosmology to test dark phenomenology of
energy and modified gravity. We conclude with an outlook in Section 5. Throughout, we work in these models is rich;
units where c =  = 1. however, their
qualitative behavior
can be of two
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

types (11): models that

2. OVERVIEW: DARK ENERGY AND MODIFIED GRAVITY are just starting to
In this section, we briefly review dark energy and modified gravity models before turning to deviate from w  −1
(thawing) and models
the tricky issue of drawing a boundary between these two paradigms. We do not attempt to be
that are approaching
comprehensive—there exist many excellent reviews on both subjects (see, e.g., References 5 and w  −1 today
6)—but rather want to give a flavor of the types of models and their phenomenology. (freezing)
We note at the outset that the models we describe are not per se solutions of the CC problem.
Generally, it is still necessary to invoke some other physics or symmetry principle to explain why
the expected contribution to the CC from Standard Model (SM) fields is absent. Despite lacking
a complete solution of the CC problem, the general approaches considered here are promising
and represent reasonable possibilities for how physics in the gravitational sector may behave.

2.1. Dark Energy

Perhaps the most natural direction to explore to explain the observed value of the CC is to posit that
the CC is itself a dynamical field, which relaxes to its present value through some mechanism (7–9).
There are many possible incarnations of this idea; what they share is the presence of a degree of
freedom which develops a condensate profile that drives the background cosmological evolution.
The simplest one is to imagine that the CC is a canonical scalar field with a potential1
  2 
√ M Pl R 1
S = d4 x −g − (∂φ)2 − V(φ) . 1.
2 2
We refer to this model as quintessence2 (10). A homogeneous condensate of this field, φ = φ(t),
behaves as a perfect fluid, with equation of state w = P/ρ, given by
φ̇
1 2
− V(φ)
w= 2
, 2.
φ̇
1 2
2
+ V(φ)
where (here and throughout) dots denote derivatives with respect to time. Observations restrict
w to be very close to −1 in the present Universe (note that this need not be the case for all times).
This is equivalent to the requirement that the evolution of the field be potential dominated. In
this sense, the present stage of cosmic acceleration in quintessence models is similar to a period
of very low scale inflation. It is also important to note that the quintessence equation of state is
generally time dependent.
An intriguing aspect of quintessence models is that a large class of them exhibits so-called
tracking behavior (12–15), where the energy density in the field closely traces the energy density

1 2
We define the reduced Planck mass as M Pl = (8π G)−1 .
2
In his influential review, Weinberg (3) showed that models of quintessence necessarily have to be tuned in order to make the
CC small.

www.annualreviews.org • Dark Energy Versus Modified Gravity 97

NS66CH05-Schmidt ARI 5 September 2016 11:58

DARK ENERGY VERSUS MODIFIED GRAVITY

In the text, we draw a distinction between dark energy and modified gravity by means of the strong equivalence
principle (SEP). We classify any theory which obeys the SEP as dark energy, and any theory which violates it as
modified gravity (see Section 2.3). Heuristically, the SEP forbids the presence of fifth forces.
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

in dark matter until the near past, providing a possible explanation of the coincidence problem—
that is, why the current density of dark energy and dark matter are comparable. The canonical
example is the Ratra–Peebles potential (9),
M 4+n
V(φ) = , 3.
φn
with n > 0.
Another consideration to keep in mind with dark energy (and even to some extent modified
gravity) models is to be sure that we are not secretly reintroducing a CC. As a simple example, a
scalar field with an exactly flat potential could plausibly be thought of as dark energy, but in this
case we should really absorb this constant potential into the bare CC in the Lagrangian. Another
compelling possibility to consider is that dark energy and dark matter may be unified into a single
component. However, this turns out to be so fine-tuned that one effectively returns to a two-fluid
model (16).

2.1.1. More general dark energy models. Even within the paradigm of the CC as a dynamical
scalar field, there is a plethora of different models, with varied phenomenology. Perhaps the most
obvious generalization away from Equation 1 is to consider a noncanonical scalar field—one that
possesses derivative self-interactions. Concretely, these models may be written as

√
S = d4 x −g4 K (X ), 4.

where we have defined X ≡ − 21 4 (∂φ)2 and K (X ) is an arbitrary function. Models of this type
are often referred to as k-essence (17, 18). Similar to the canonical case, for homogeneous field
profiles, φ = φ(t), these models behave as a perfect fluid, though of a more general type, where
now the equation-of-state parameter w is given by
K
w= . 5.
2XK ,X − K
By suitably choosing a functional form for K , one can reproduce the cosmic expansion history.
Field perturbations away from the background evolution no longer propagate luminally in k-
essence models; instead, they propagate with a speed of sound given by
K ,X
c s2 = . 6.
K ,X + 2XK ,X X
This allows structure formation to be noticeably different with respect to canonical dark energy
models, as we discuss in Section 3.3.
Dark energy models of the late-time acceleration are similar to inflation, and there has been an
effort to adapt the effective field theory of inflation formalism (19) to the problems of dark energy
and modified gravity (20–24).

98 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

2.2. Modified Gravity

We begin this section by reviewing some commonly studied IR modifications of GR. These
examples allow us to draw some more general lessons: Generic modifications lead to new physics
on small scales (25). Indeed, all known modifications to Einstein gravity introduce new degrees
of freedom in the gravitational sector.3 These new particles mediate a fifth force, so the theory
must employ some screening mechanism in order to evade local tests of gravity, which are very
constraining.
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

2.2.1. Scalar–tensor theories. Scalar–tensor theories are probably the best-studied modifica-
tions to Einstein gravity. Restricting to canonical scalar fields for the moment, these theories can
be cast in Einstein frame as
  2 
√ M Pl R 1  
S = d4 x −g − (∂φ)2 − V(φ) + Smatter A2 (φ)g μν , ψ . 7.
2 2
Here, R is the Ricci scalar constructed from the metric g μν . Matter fields, ψ, couple to the Jordan-
frame metric g̃ μν = A2 (φ)g μν . This coupling induces interactions between matter fields and φ,
causing test particles in the Newtonian regime to feel a force,
 
a = −∇ + ln A(φ) , 8.

sourced by both the Einstein-frame potential, , and the scalar, φ.

The prototypical scalar–tensor theory is Brans–Dicke theory (26), which corresponds to V(φ) =
0 and A2 (φ) = exp[−φ/(M Pl 3/2 + ω)] in Equation 7. This theory is usually cast in Jordan frame,
g μν → g̃ μν , where it takes the form
2   
M Pl ω  
S= d x −g̃ φ R̃ − (∂φ) + Smatter g̃ μν , ψ ,
4 2
9.
2 φ

with ω a constant parameter. Solar System tests of gravity place the bound ω > 4
∼ 4×10 (6, 27). This
bound corresponds to taking the coupling to matter to be very weak, making the Brans–Dicke
theory essentially equivalent to a dark energy model.
In more general scalar–tensor theories—by suitably choosing V(φ) and A(φ)—it is possible to
evade Solar System constraints, while still having interesting phenomenology for the scalar. To
see how this can work, we focus on the usual case where the scalar field φ couples to the trace of
the Jordan-frame stress tensor, T̃ = A−4 T , where T is the trace of the stress tensor in Einstein
frame. If we consider nonrelativistic sources, T = −ρ, and define ρ̄ ≡ A−1 ρ, the scalar obeys the
equation of motion
dVeff
φ = , Veff (φ) = V(φ) + A(φ)ρ̄. 10.
dφ
The key point is that φ responds to an effective potential, which depends on the external matter
sources. This makes it possible to engineer situations where the field behaves differently in different
environments.

3
In this review, we focus on modified gravity applications to cosmic acceleration. There has also been research on modifying
gravity to obviate the need for dark matter, but we do not discuss that topic here.

www.annualreviews.org • Dark Energy Versus Modified Gravity 99

NS66CH05-Schmidt ARI 5 September 2016 11:58

The best-known example is the chameleon field (28, 29), where the potential and matter
coupling are chosen so that the effective mass of the scalar field,
d2 Veff d2 V d2 A
m2eff (φ) = 2
= 2
+ ρ̄, 11.
dφ dφ dφ 2
increases in regions of high density, like the Solar System. A concrete example is the Ratra–Peebles
potential and a linear coupling (28, 29):
M 4+n φ
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

V(φ) = , A(φ)  1 + ξ . 12.

φn M Pl
Another popular scalar–tensor model with a screening mechanism is the symmetron model (30),
which can also be cast as Equation 7, but with
μ2 2 λ 4 φ2
V(φ) = − φ + φ , A(φ)  1 + . 13.
2 4 2M 2
In regions of low density, the Z2 symmetry of the model is spontaneously broken and φ mediates
a fifth force; in regions of high density, the symmetry is restored and the additional force turns
off.
More precisely, in both the chameleon and the symmetron a screened field is sourced only by
a thin shell near the surface of a dense object. Whether or not the screening operates depends on
the Newtonian potential of the object exceeding a certain threshold (29, 30); see Section 2.2.4.
One of the most-studied scalar–tensor theories is so-called f (R) gravity, which on its face
appears not to be a scalar–tensor theory at all. This model consists of higher-curvature corrections
to the Einstein–Hilbert action
2   
M Pl √
S= d4 x −g R + f (R) + Smatter [g μν , ψ], 14.
2
where f (R) is a function only of the Ricci scalar, chosen to become significant in the low-curvature
regime R → 0. f (R) models have been used to drive cosmic acceleration (31–33). However, these
original models are actually in conflict with precision tests of gravity (34), essentially because
these models are scalar–tensor theories in disguise (35, 36). One can see this by performing a field
redefinition and conformal transformation (see, e.g., Reference 37 for details) to cast the theory
as in Equation 7 with

M Pl φ f,φ − f √ φ
V(φ) = , A(φ) = e 6M Pl
, 15.
2 1 + f,φ 2

which is equivalent to the Brans–Dicke theory in Equation 9 with ω = 0 and a potential.

f (R) models compatible with local tests of gravity were constructed in References 38 and 39.
For example, in Reference 38 f (R) is given by
n  −n
c 1 mR2 R
f (R) = −m2 n ≈ ρ ,eff + f R0 , 16.
c 2 m2 + 1
R
R̄0

where the second form is the leading expression for R/m2 1, which is required by cosmological
and local tests (Section 4). The parameters c 1 and c 2 can be adjusted so that ρ,eff supplies a CC
consistent with observations, leaving f R0 as a free parameter. R̄0 is the background Ricci scalar
today, so that f R0 corresponds to the field value in the background today. In scalar–tensor language,
the model is consistent with experiment because of the chameleon mechanism. f (R) gravity has
been explored in a variety of contexts, and we refer the reader to Reference 40 for more details.

100 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

2.2.2. More general scalar–tensor theories: Horndeski and generalizations. The space of
scalar–tensor theories is far broader than the example presented in Equation 7. In fact, in recent
years, there has been an effort to map out the most general consistent theory of a metric interacting Ghost: a quantum
with a scalar. In this context, consistency requires that the theory be absent of ghosts. Typically, with either negative
theories with equations of motion that are higher than second order in time derivatives have a energy density or
negative norm. The
ghost.4 Therefore, theories that have second-order equations of motion have attracted a great deal
existence of
of interest. propagating ghosts in a
The most general theory of a scalar field interacting with a metric that has second-order theory poses problems
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

equations of motion has been derived (42); it corresponds to the Lagrangian for its quantization; in
particular, the vacuum
Lgen.gal. = K (φ, X ) − G3 (φ, X )φ + G4 (φ, X )R is unstable to decay
+ G4,X (φ, X )[(φ)2 − (∇μ ∇ν φ)2 ] + G5 (φ, X )Gμν ∇ μ ∇ ν φ 17. into ghosts and healthy
particles (43, 44)
1
− G5,X (φ, X )[(φ)3 − 3(φ)(∇μ ∇ν )2 + 2∇ μ ∇α φ∇ α ∇β φ∇ β ∇μ φ].
6
This theory has four arbitrary functions, K , G3 , G4 , and G5 , and X is defined as in Equation 4.
Interestingly, an equivalent theory had been derived much earlier by Horndeski (45, 46), but had
gone essentially unnoticed in the literature.
A particularly well studied limit of this theory is the so-called galileon (47). The simplest version
of this theory—the cubic galileon—is a special case of Equation 17:
2
M Pl R 1 1
Lgal. = − (∂φ)2 − 3 φ(∂φ)2 . 18.
2 2 
In the limit where gravity is taken to be nondynamical, this theory has the symmetry δφ = c +b μ x μ .
In general dimensions, there are only a finite number of terms invariant under this symmetry,
which also have second-order equations of motion. The galileon appears in many constructions;
in particular, it describes the decoupling limits both of massive gravity (48) and of the Dvali–
Gabadadze–Porrati (DGP) model (49, 50). For a review of many interesting properties of the
galileon, see Reference 51.
Recently, an effort has been made to relax the assumption of second-order equations of motion
in order to construct the most general scalar–tensor type theory that propagates three degrees of
freedom nonlinearly. In such theories, the expected ghostly degrees of freedom from higher-order
equations of motion are projected out by constraints (see Reference 52 and references therein).

2.2.3. Massive gravity and brane-worlds. Einstein gravity is the theory of a massless spin-2
particle. A natural question to ask is whether it is possible for the graviton to be a massive particle.
The intuition behind this explanation for cosmic acceleration is that massive fields induce potentials
of the Yukawa type,
e −mr
V (r) ∼ , 19.
r
which shut off at distances of order m−1 . The idea is that the presently observed cosmic acceleration
could be due to a weakening of gravity at large scales, with m ∼ H 0 .
Massive gravity has been studied since the initial investigations of Fierz & Pauli (53), but
much of the modern interest is due to a ghost-free theory by de Rham, Gabadadze, and Tolley
(dRGT) (48, 54), built on earlier research (55, 56), which showed how to overcome the traditional
difficulties associated with nonlinearly completing the Fierz–Pauli theory. In particular, generic
nonlinear theories propagate a ghost in addition to the graviton degrees of freedom (57). The

4
This is the content of Ostrogradsky’s theorem (41).

www.annualreviews.org • Dark Energy Versus Modified Gravity 101

NS66CH05-Schmidt ARI 5 September 2016 11:58

absence of this Boulware–Deser ghost in the dRGT theory was shown in Reference 58. Recovery
of GR in the massless limit relies on the Vainshtein mechanism (see Section 2.2.4). For reviews
of the theoretical background and recent developments, see References 59 and 60.
φ and N :
on subhorizon scales Another avenue to massive gravity is as a resonance (as opposed to the hard mass above). This
in cosmology, and idea has been very influential in cosmology; the most-studied example is the DGP model (49).
within the Compton The setup consists of a four-dimensional (4D) brane (on which matter fields live) embedded in an
wavelength of the infinitely large five-dimensional (5D) bulk space-time. The model consists of dynamical gravity
field, φ, and the
both on the brane and in the bulk:
Newtonian potential,
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

N , are proportional
 
in the absence of M 53 √ M 43 √
screening, as may be S= d5 X −GR + d4 x −g R, 20.
2 2
deduced using the
scalar equation of
motion and the where M 5 and M 4 are the 5D and 4D Planck masses; G and R are the 5D metric and Ricci scalar;
Poisson equation
and g and R are the 4D metric and Ricci scalar. From the perspective of an observer living on the
brane, gravity is mediated by a continuum of gravitons, and at short distances, gravity appears to
be 4D, but at large distances, the gravitational potential “leaks” off the brane and is that of a 5D
theory. The potential due to a mass M source behaves as (59, 61)

⎧
⎪ M
⎪
⎪ r rc
⎨M 2r
4
V (r)  , 21.
⎪
⎪ M rc
⎪
⎩ 2 r rc
M4r r

where rc = M 42 /2M 53 . Therefore, the potential is weaker at large distances than it would be in
pure GR. This weakening allows for de Sitter solutions on the brane absent a bare CC (61, 62).
Unfortunately, this self-accelerating branch of solutions is unstable—perturbations are ghostly (50,
63). The DGP model remains quite interesting as a benchmark model; it is one of the better-studied
modified gravity models in the literature. For a review of DGP, see Reference 64.
Interestingly, a massive graviton could address the CC problem by means of degravitation (65,
66): A large CC nevertheless does not gravitate, because a massive graviton behaves as a high-pass
filter (67). However, a realistic degravitating solution has not yet been found in either dRGT or
brane-world.

2.2.4. Screening mechanisms. In all of the models we have discussed, new degrees of freedom
appear in the gravitational sector. This is in fact a general feature: GR is the unique low-energy
theory of a massless spin-2 particle (68, 69), so essentially any departure from GR introduces new
degrees of freedom, typically with a mass of order the Hubble parameter today: m ∼ H 0 . Because
they couple to the SM, these light fields mediate a long-range force between matter sources.
The presence of additional forces is strongly constrained by laboratory and Solar System
tests (27). Therefore, the additional degrees of freedom must hide themselves locally. The screen-
ing mechanisms by which this is achieved fall into several broad classes, activating in regions where
the Newtonian potential N or successive gradients become large. N is defined as the potential
that solves the Poisson equation (Equation 26). For concreteness, we assume that the additional
degree of freedom is a scalar in the following discussion, but the general philosophy is much
broader. For a more complete discussion, see Reference 37.

102 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

These classes of models are described as follows.

1. Screening at large field values. The first type of screening we consider activates in regions
where the Newtonian potential exceeds some threshold value: N > . Generally, this leads Test bodies: objects
to the additional degree of freedom itself developing a large vacuum expectation value— that are sufficiently
which causes the coupling to matter to weaken, the mass of the field to increase, or the small that we may
self-coupling of the field to become large—leading to a diminishing of the force mediated. neglect gravitational
tidal forces in deriving
Notable examples are the chameleon (28, 29), symmetron (30, 70, 71), and dilaton (72,
their motion; a more
73). From a phenomenological viewpoint, these mechanisms activate in regions of large precise definition can
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

potential, N , so regions of small Newtonian potential should exhibit the largest deviations be found in
from GR. Reference 82
2. Screening with first derivatives. This mechanism turns on when the local gravitational ac-
celeration exceeds some critical value: ∇ N > 2 . This condition roughly corresponds
to first gradients of the scalar field becoming large; screening occurs due to kinetic self-
interactions: ∂φ/2 1. General P(X ) models can display this kind of screening (74); a
broad class of these models is referred to as k-mouflage (75). Because these models screen
in regions of large acceleration, it is intriguing to imagine that they could be relevant for
reproducing the phenomenological successes of MOND (76) in a more complete framework
(77, 78).
3. Screening with second derivatives. The last category of screening mechanisms we discuss
is those that become active in regions of large curvature: ∇ 2 N > 3 , which is equiva-
lent to high density. These mechanisms rely on nonlinearities in the second derivatives:
φ/3 1. The most commonly studied example in this class is the Vainshtein mech-
anism (79), which operates in the galileon and in massive gravity. In these models, the
largest deviations from GR are expected in low-curvature regimes. Additionally, there is
evidence that Vainshtein screening is less efficient than naı̈ve estimates in time-dependent
scenarios (80, 81), making this a promising avenue to search for deviations from GR.

2.3. Drawing a Boundary

In order to meaningfully discuss observationally discriminating between modified gravity and
dark energy, we must draw a distinction between the two scenarios. This is a somewhat aesthetic
choice, but nonetheless it helps us organize our thinking about exploring and testing the various
possibilities.
The distinction we make relies essentially upon the motion of bodies in the theory. To begin,
recall the weak equivalence principle (WEP). The WEP is the statement that there exists some
(usually Jordan-frame) metric to which all matter species couple universally. Then, test bodies—
regardless of their composition—fall along geodesics of this metric. This is usually stated as the
equivalence of inertial mass and gravitational mass. In this review, we focus on theories that satisfy
the WEP at the level of the action.
To distinguish dark energy versus modified gravity, we further invoke the SEP. The SEP ex-
tends the universality of free fall to be completely independent of a body’s composition, including
gravitational binding energy, so compact objects like black holes also follow geodesics (83). We
call anything that obeys the SEP dark energy, and anything that does not, modified gravity. The
motivation for this definition is to classify models that influence ordinary matter only gravita-
tionally as dark energy. In these models, the force felt between two bodies is only that of GR
(and possibly other SM forces). However, in models of modified gravity two isolated bodies may
carry additional charges (e.g., scalar charge), which causes them to experience an additional force
beyond that of gravity. The appeal to the SEP is an attempt to make this intuition precise. A

www.annualreviews.org • Dark Energy Versus Modified Gravity 103

NS66CH05-Schmidt ARI 5 September 2016 11:58

YES Strong YES YES

equivalence Dynamical Dark
principle? dark energy? energy
Weak
equivalence
principle?
Interacting Modified Cosmological
NO dark sector NO gravity NO constant

Figure 1
A flowchart of the distinctions made in the text.
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

theoretical motivation for this distinction based upon the SEP is that it is believed—although not
proven—that GR is the only metric theory that obeys the SEP (27, 83).
The preceding discussion is somewhat abstract, so it is useful to illustrate the main points with
scalar–tensor theory. We again consider the action Equation 7, but allow for each matter species
to have a different coupling to φ:
  2 
√ M Pl 1  
S = d4 x −g R − (∂φ)2 − V(φ) + Smatter Ai2 (φ)g μν , ψi . 22.
2 2
Here, the notation Ai2 (φ) captures the fact that the individual matter fields, ψi , do not necessarily
all couple to the same Jordan-frame metric. The first restriction we make is to demand that the
model obey the WEP5 (Figure 1), which restricts the couplings to be the same: Ai2 (φ) ≡ A2 (φ).
Models with A2 (φ) = 1 satisfy the SEP and, hence, are models of dark energy. In these models,
the scalar field, φ, affects the motion of matter only through its gravitational influence; it is a
decoupled source of stress energy. Cases where A2 (φ) is some nontrivial function are models of
modified gravity. In modified gravity models, the force mediated by φ does not affect all objects
universally. As an extreme example, consider the motion of some diffuse object compared with that
of a black hole. Due to the no-hair theorem (85), black holes carry no scalar charge and therefore
feel no fifth force from φ. However, a more diffuse object, such as a star or planet, will feel such a
fifth force, leading to a large violation of the SEP.
Note that although the theories we consider satisfy the WEP at the microscopic level (all matter
couples to the same metric), it is nevertheless possible to have apparent violations of the WEP
for macroscopic objects (86, 87). This occurs in theories with screening mechanisms; whether or
not the mechanisms operate depends mainly on the masses of the objects involved, so large-mass
objects can fall at a different rate than small-mass objects. This is essentially because there is
not necessarily a relationship between the scalar charge carried by objects and their gravitational
potential, as there is in scalar–tensor theories without screening mechanisms, which generally
exhibit violations of the equivalence principle only when N is large (86).
Although the distinction based on the SEP is theoretically clean and satisfying, in practice it is
not very useful phenomenologically. A more pragmatic distinction (relevant for the tests discussed
below) is to call anything which has a fifth force modified gravity, and anything else, dark energy.

3. COSMOLOGICAL PHENOMENOLOGY
In order to describe the phenomenology of dark energy and modified gravity models, it is conve-
nient to first consider the unperturbed universe, described by the scale factor a(t), before turning

5
Interactions in the dark sector—that is, a coupling between dark matter and dark energy—are a further scenario to consider.
Because dark energy cannot strongly interact with visible matter, this approach violates the WEP. We do not consider this
possibility in depth, but see Reference 84 for a review of the phenomenology.

104 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

to the equations governing the evolution of cosmological perturbations. The phenomenology of

the latter is much richer, and more likely to yield insights distinguishing modified gravity and dark
energy.

3.1. Background Expansion

In GR, applying the Einstein equations to the Friedmann–Robertson–Walker (FRW) metric yields
the Friedmann equations
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

1  1
H2 = 2
ρi = 2
[ρ m + ρ DE ] ,
3M Pl i 3M Pl
2  2
Ḣ + H 2 = − 2
[ρ̄i + 3 p̄ i ] = − 2
[ρ m + (1 + 3wDE )ρ DE ]. 23.
3M Pl i
3M Pl

Here, we have specialized to the case of a spatially flat universe and two stress-energy components,
nonrelativistic matter [dominated by cold dark matter (CDM)] and dark energy with equation of
state w(t) (note that w is time dependent), which is the relevant case in the late Universe.
Modified gravity models lead to correspondingly modified Friedmann equations when applied
to a homogeneous cosmology. As an example, consider the DGP model (see Section 2.2.3), where
the first Friedmann equation is modified to
H 1
H2 ± = 2 m
ρ 24.
rc 3M Pl

in the absence of dark energy. The negative sign corresponds to the self-accelerating branch, which
admits a de Sitter solution H = constant at late times. However, this expansion history is easily
mimicked in GR by a suitably chosen quintessence potential that yields ρ DE (t)/3M Pl 2
= H(t)/rc .
This clearly illustrates that, while constraints on the background expansion are crucial in order to
interpret LSS observables, they do not allow us to cleanly distinguish between modified gravity
and dark energy.

3.2. Structure Formation with Quintessence

We now turn to the growth of structure in the context of a quintessence dark energy model, that
is, a canonical light scalar field with an effective sound speed of one. The CC, and thus lambda
cold dark matter (CDM), is included here as a limiting case. Current constraints already imply
that, for z <
∼ 1, the equation of state has to be close to −1. In these simplest dark energy models,
density and pressure perturbations of the dark energy component are of order (1 + w) , where
typical cosmological potential perturbations are of order ∼ 10−4 . Furthermore, anisotropic
stress is negligible in these models. Thus, dark energy perturbations have a very small effect on
LSS.
We continue to assume a spatially flat background and work in the conformal-Newtonian
gauge, so that the metric is given by

ds 2 = −(1 + 2)dt 2 + 2a(t)Bi dtdx i + a 2 (t)[(1 − 2 )δi j + γi j ]dx i + dx j . 25.

Assuming that the scalar potential perturbations  and are much smaller than one, we can
work to linear order in them. This is accurate almost everywhere in the universe. The potential
 governs the dynamics of nonrelativistic objects, whereas the combination γ ≡ ( + )/2
determines the null geodesics, namely light propagation. One way to see this is to note that null

www.annualreviews.org • Dark Energy Versus Modified Gravity 105

NS66CH05-Schmidt ARI 5 September 2016 11:58

geodesics are conformally invariant. Then, consider a conformal transformation g μν → e 2ω g μν .

At linear order, this transformation corresponds to {, } → { + ω, − ω}, so that γ is the
unique linear combination of potentials that is conformally invariant.
Subhorizon scales
k  aH: most LSS In Equation 25, Bi is a transverse vector capturing the vector modes, whereas γi j is a transverse
measurements are on trace-free tensor corresponding to gravitational waves (tensor modes). We neglect the vector
these scales; in this modes throughout, as they decay and do not propagate. We briefly discuss tensor modes in
limit, time derivatives Section 3.4.3.
are of order H , such
In GR, the difference between the potentials, − , is sourced by the anisotropic stress,
that ˙ ∼ H , can be
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

neglected in namely the trace-free part of T i j . In the models considered in this section, this part is negligible,
comparison to spatial so we can set =  (see Section 3.3 for generalizations). On subhorizon scales k a H , the 00
derivatives component of the Einstein equations reduces to the Poisson equation
3
∇2 = 4π G ρ m a 2 δ, or ∇2 = m (a) (aH)2 δ, 26.
2
where δ = ρ/ρ m − 1 is the fractional matter overdensity, and m (a) = ρ m (a)/3H 2 MPl2
.
For later reference, we define the solution to Equation 26 as Newtonian potential N . The
continuity and Euler equations for the collisionless dark matter fluid are
1
δ̇ + ∂k [(1 + δ)v k ] = 0,
a
1 1
v̇ i + Hv i + v k ∂k v i = − ∂ i . 27.
a a
−1
Here, v = adx/dt is the fluid peculiar velocity. On large scales, for Fourier modes k <
∼ 0.05 h Mpc ,
both the density perturbation δ and the peculiar velocities |v| are much less than one, and we can
linearize these equations. They can then be combined to yield the linear growth equation
3
δ̈(k, t) + 2H δ̇(k, t) = −k2 (k, t) = m (a)H 2 δ(k, t), 28.
2
while the velocities obey v = −iak/k2 δ̇. Thus, individual Fourier modes of the density field
evolve independently and equally on different scales. Furthermore, the evolution of density and
velocity (given initial conditions) is completely controlled by the expansion history H(t) of the
universe. Thus, quintessence predicts a definite relation between the observed expansion history
and growth of structure, allowing for consistency tests (see Section 4.2). Generally, for fixed initial
conditions and an approximately constant equation of state w, the amount of late-time structure
monotonically decreases with increasing w (Figure 2), as the accelerated expansion sets in earlier
for less-negative w.
The same conclusions remain valid when considering the full Euler–Poisson system
(Equations 26–27) and, indeed, the exact Vlasov (collisionless Boltzmann) equation for dark mat-
ter. The only impact of quintessence dark energy is through the background expansion history.
The nonlinear regime of LSS is frequently described by the halo model reviewed in
Reference 88, which is motivated by the fact that a significant fraction of the dark matter re-
sides in self-gravitating collapsed structures called halos. Moreover, most observed LSS tracers,
such as galaxies and clusters, are found in these halos. In contrast to the complicated processes gov-
erning galaxy formation, collisionless N-body simulations are able to accurately predict abundance
and clustering of halos.
The mass function of halos—that is, their number density in a logarithmic mass interval—is
described to within ∼ 5% [in the context of CDM (90)] by the universal form
 
ρ  d ln σ (M, z) 
dn(M, Z)
= m f (ν)   , ν ≡ δc (M) , 29.
d ln M M d ln M  σ (M)

106 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

2 × 104

Pm (h–3 Mpc3)
1 × 104
ΛCDM
w = –0.7, cs = 1
w = –0.7, cs = 0
5,000
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

w = –1.3, cs = 1
w = –1.3, cs = 0

0.01 0.05 0.10 0.15

k (h Mpc–1)
Figure 2
Linear matter power spectrum in quintessence dark energy with different values of equation of state w and
speed of sound c s . For the values of w shown, clustering dark energy with vanishing speed of sound modifies
the matter power spectrum by a few percent. It can be probed by weak lensing measurements of the cosmic
microwave background and galaxies. Modified from Reference 96.

where σ (M, z) is the variance of the linear density field at redshift z when smoothed with a top-hat
filter of radius R(M), which contains the mass M at the background density ρ m . f (ν) is in general a
free function, and δc is the initial overdensity of a spherical top-hat overdensity that collapses (i.e.,
reaches radius Rth = 0) at a given redshift z, extrapolated from the initial time to redshift z through
linear growth. For quintessence dark energy, the top-hat radius obeys (see, e.g., Reference 89).

R̈th 4π G 4π G
=− [ρ m + (1 + 3w)ρ DE ] − δρm , 30.
Rth 3 3
where δρm = ρm (<Rth ) − ρ m is the overdensity in the interior of the top hat. Equation 29 is
inspired by the excursion set approach (91), and the well-known Press–Schechter mass function
(92) is a special case. Massive halos correspond to small variances σ (M) and, hence, to high ν.
In the high-ν limit, f(ν) asymptotes to exp(−q ν 2 ), q = O(1), corresponding to exponentially rare
high-mass halos. The abundance of halos depends on the growth history through both σ (M) and
δc , although the dependence of δc on cosmology is quite weak. The ansatz (Equation 29) is not
guaranteed to be correct, even for a free function f (ν). Nevertheless, it is accurate at the ∼ 10%
level for CDM (93).
The large-scale distribution of halos, described statistically by their N-point functions, can
also be derived in this picture. The most important statistic on large scales is the two-point
function or power spectrum P(k). On scales where linear perturbation theory applies, the halo
two-point function is related to that of matter by Ph (k, z) = b 12 (z)Pm (k, z)+C, where C is a constant
corresponding to white noise, and the linear bias parameter b 1 can be derived from the halo mass
function (Equation 29). b 1 describes the response of the abundance of halos to a long-wavelength
density perturbation δl or, equivalently, a change in the background density ρ m → ρ m (1 + δl ).
This corresponds to lowering the threshold δc → δc − δl , so that the linear halo bias becomes (94)
ρ m ∂n(M, z) 1 d f (ν)
b 1 (M, z) = =− . 31.
n(M, z) ∂ρ m σ (M ) f (ν) dν
Finally, assuming a universal density profile for halos ρ(r|M, z), such as the NFW profile
ρ(r|M, z) ∝ [r(1 + r/rs )2 ]−1 (95), the halo model provides a description of the statistics of the
matter density field on nonlinear scales.

www.annualreviews.org • Dark Energy Versus Modified Gravity 107

NS66CH05-Schmidt ARI 5 September 2016 11:58

3.3. Beyond Quintessence Dark Energy

We now consider three relevant generalizations of the quintessence dark energy case considered
in the previous section: small speed of sound c s = (δp/δρ)1/2 1, anisotropic stress, and an
interacting dark sector. Dark energy density perturbations are negligible on scales that are much
smaller than the sound horizon of the dark energy: Rs ,DE  c s H −1 in physical units. Let us now
consider the opposite case, namely what happens on scales much greater than Rs ,DE . On those
scales, the pressure perturbations of the dark energy component are negligible compared with
its density perturbations. This means that the dark energy fluid moves on the same trajectories
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

(geodesics) as the pressureless CDM component. Thus, density perturbations increase on small
scales in both components, although the dark energy perturbations are suppressed by 1 + w
(on intermediate scales, c s k/a H ∼ 1, dark energy pressure perturbations need to be taken into
account). Specifically, the Euler–Poisson system is augmented by the continuity equation for the
dark energy component and becomes
1
δ̇DE − 3Hw δDE + ∂k [(1 + w + δ)v k ] = 0; ∇2 = 4π G(ρ m δ + ρ DE δDE ).
a
It is also possible to extend the spherical collapse Equation 30 to this case (96):

R̈th 4π G 4π G
=− [ρ m + (1 + 3w)ρ DE ] − (δρ + δρDE ), 32.
Rth 3 3
Ṙth
ρ̇DE (<Rth ) = −3 [ρDE (<Rth ) + wρ DE ] . 33.
Rth
Again, we see that if 1 + w 1, then initially small perturbations in the dark energy density
will stay small. Figure 2 illustrates the effects of c s = 0 dark energy for the linear matter power
spectrum. For w = −0.9, the effect on the matter power spectrum is of order 1% at z = 0 (and
smaller at higher redshifts). By contrast, the halo abundance, calculated using Equation 29 with
the modified σ (M ) and δc , is larger by 5–10% at the very high mass end for c s = 0 as compared
with c s = 1 (96). The quantitative effect of a small dark energy sound speed can be increased if
there is a significant “early dark energy” component (97).
We now turn to the case of a dark energy component with scalar anisotropic stress iDE j ≡
T i j − δi j T klDE δ kl /3. iDE
DE
j sources a difference between the two metric potentials following
 
1
∂i ∂ j − δi j ∇ 2 ( − ) = 8π G a 2 iDE j . 34.
3
The fact that =  is relevant to the cause of distinguishing between modified gravity and dark
energy, as the difference between and  is otherwise a distinctive feature of modified gravity
(see the following section). Dark energy models that involve a single scalar field do not lead to
anisotropic stress, whereas models involving vector fields generally do (98, 99). Koivisto & Mota
(100) and Mota et al. (101) investigated the impact on structure formation of dark energy with
anisotropic stress on the basis of the effective parameterization by Hu (102) via a viscous parameter
c vis . The authors showed that the effect of dark energy anisotropic stress observable via − 
is significant only for horizon-scale perturbations. By contrast, modified gravity generally leads
to a significant −  on all scales, so that one can still hope to effectively distinguish between
modified gravity and dark energy via this probe.
Finally, one can also couple the dark energy component to the nonbaryonic components, such
as CDM or neutrinos (103, 104). Following our distinction in Section 2.3, these are not modified
gravity models, given that the fifth force does not obey the WEP. The absence of coupling to
baryons and radiation lets these models evade local tests, so the effects on the LSS can become

108 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

quite significant (105). Interestingly, a generic signature of this model is a different clustering
amplitude (bias) of baryons relative to the dark matter on large scales, which can be used to place
constraints on this type of model. Parameterized
post-Friedmann
(PPF) approach:
3.4. Structure Formation in Modified Gravity on linear scales and for
In this section, we briefly review the rich phenomenology of structure growth in modified gravity, adiabatic initial
conditions, any theory
separately for the linear and nonlinear regimes.
of gravity can be
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

parameterized by two
3.4.1. Scalar perturbations on linear scales. We begin with structure formation in modified
free functions of scale
gravity theories on sufficiently large scales, where linear perturbation theory applies, and assume and time
that stress-energy perturbations are only due to a CDM component. It is convenient to work (Equation 36), with an
in Fourier space, where individual modes evolve independently at linear order. Then, given that additional constraint
space-time is described by two potentials, (k, t) and (k, t), and the matter sector is characterized on superhorizon scales
(Equation 35)
completely by δ(k, t) (the velocity is determined via the continuity equation), any theory of gravity
can be described in this regime by two free functions of k and t, which parameterize the relation
between the three fields , , and δ.
On superhorizon scales k a H and for adiabatic perturbations, further constraints are posed
by diffeomorphism invariance (106–108). Specifically, a superhorizon adiabatic mode behaves as
a separate, curved universe, so that its evolution is constrained by the solution to the background
expansion in the given theory. Bertschinger (106) derived
 
¨ − Ḧ ˙ + H ˙ + 2 Ḣ − Ḧ H = 0, 35.
Ḣ H Ḣ

which, given one function of time specifying the relation between and  for k a H , determines
the superhorizon evolution of the potentials.
We now turn again to the subhorizon limit and assume adiabatic initial conditions. In the
absence of preferred directions, we can parameterize the relation between δ and ,  via
3
− k2 (k, t) = m (a)(aH)2 μ(k, t)δ,
2
(k, t)
= γ (k, t). 36.
(k, t)
These equations reduce to GR for μ = γ = 1. Given adiabatic initial conditions and assuming
the WEP, , , and δ are, at linear order, all related to the initial conditions (i.e., a single
stochastic variable) through transfer functions T , T  , and T δ (k, t), so that μ ∼ (k/a H )2 T  /T δ
and γ = T /T  . Parameterizations based on Equations 35 and 36 are referred to as the PPF
approach. Moreover, for a local 4D theory of gravity, T i are functions of k2 only, so that μ and
γ reduce to rational functions of k2 . If we further assume that no higher than second derivatives
appear in the equations for and , then μ and γ are completely described by five functions of
time only (109):

1 + p 3 (t)k2 p 1 (t) + p 2 (t)k2

μ(k, t) = , γ (k, t) = . 37.
p 4 (t) + p 5 (t)k2 1 + p 3 (t)k2
Note that higher powers of k can appear in the equations for and  after the additional degrees
of freedom are integrated out, even if the fundamental equations are all second order in derivatives.
Using suitable interpolation, one can connect these subhorizon limit results to superhorizon scales
in a parameterization that enforces Equation 35 (107, 110, 111), further reducing the number

www.annualreviews.org • Dark Energy Versus Modified Gravity 109

NS66CH05-Schmidt ARI 5 September 2016 11:58

of free functions. Finally, Gleyzes et al. (23) and Bellini & Sawicki (24) found that four free
functions of time, along with a free background expansion history, completely describe the linear
growth of perturbations for the Horndeski Lagrangian (Equation 17). While different equivalent
parameterizations exist, the specific choice of μ and γ adopted here is motivated by the fact that μ
can be directly constrained by the growth of structure, as  is the potential governing the motion
of matter. From Equation 28, the growth equation is modified to
3
δ̈(k, t) + 2H δ̇(k, t) = m (a)H 2 μ(k, t)δ(k, t). 38.
2
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

Specifically, in modified gravity, the linear growth is in general scale (k) dependent, unlike the
quintessence case or the case for dark energy with c s = 0. Measuring the growth history as a
function of k and t in principle enables a measurement of μ(k, t).
By contrast, γ , also referred to as gravitational slip, quantifies the departure between the two
space-time potentials. The propagation of photons is, in the coordinate frame of Equation 25,
governed by the combination of potentials
1 1
γ = ( + )= (1 + γ ). 39.
2 2
This is in exact analogy to the γPPN parameter in the parameterized post-Newtonian (PPN)
framework (27), which is similarly constrained by combining photon propagation (Shapiro delay)
with dynamics (the Earth’s orbit). The fact that Solar System tests constrain |γPPN − 1| ∼ < 10−5
clearly shows that a consistent modification of gravity in cosmology has to be scale or environment
dependent. Gravitational lensing observables in cosmology are then approximately related to the
matter density through the combination (1 + γ )μ of the modified gravity parameterization (see
Section 4).

3.4.2. Scalar perturbations on nonlinear scales. The description of nonlinear structure forma-
tion in the context of modified gravity is complicated by the necessity of screening mechanisms.
In other words, although we can in principle extend the parameterization of Equation 36 to small,
nonlinear scales, that will in general violate Solar System constraints on gravity, which tightly con-
strain μ and γ . The most interesting modified gravity models can circumvent these constraints
by employing nonlinear screening mechanisms. By definition, they are not captured by a linear
parameterization of the form of Equation 36. For this reason, cosmological constraints on modi-
fied gravity from nonlinear scales are generically model dependent. In this context, it is useful to
classify models by their screening mechanism (see Section 2.2.4).
In models of the chameleon or symmetron/dilaton type, a certain depth of the gravitational
potential is necessary to pull the field away from its background value ϕ̄ and activate the screening
mechanism. Specifically, if the Newtonian potential N , which is the solution of Equation 26,
satisfies − N > ∼ C ϕ̄/M Pl , then ϕ becomes locally suppressed compared with ϕ̄, and the deep
potential region is screened. The constant C depends on the model but is typically of order unity;
for f (R) gravity, C = 3/2. In order for the Solar System to be screened, C ϕ̄/M Pl < −5
∼ 10 , which
can be shown to place an upper limit on the mass m̄ϕ H of the field in the background (112).
The result is that viable models with this screening mechanism modify structure formation only
on scales below ∼30 Mpc.
This screening behavior further means that there will be a characteristic threshold halo mass
M scr (ϕ̄); halos above this mass will be screened, whereas halos below this threshold will be un-
screened (because halo profiles are nearly universal, there is a well-defined mapping between
central potential and halo mass). This transition effect can be clearly observed in N-body sim-
ulations of chameleon-type models; Figure 3a shows the gravitational acceleration within dark

110 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

a b

1.2
1.3
|fR0| = 10–4
|fR0| = 10–5
dN < 1
|fR0| = 10–6 nDGP-1

vir, DGP
vir, f(R)

1.2
1.1 nDGP-2
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

sDGP

1.1

1.0

1.0

1014 1015 1014 1015

M300 (M☉/h) M200 (M☉/h)

Figure 3
Mean mass-weighted gravitational acceleration (relative to general relativity) ḡ within dark matter halos, as
measured in (a) modified gravity simulations for f (R), corresponding to chameleon screening, and (b) the
Dvali–Gabadadze–Porrati model (DGP), showing Vainshtein screening. The lines show the expectations
based on a simple spherical model of the halos (114). The circled points show halos that are screened due to
the potential of a massive halo in the vicinity, rather than their own potential well. The ratio of dynamical to
lensing mass of these halos is given by M D /M L = ḡ3/5 . It can be probed by comparing dynamical mass
estimates of galaxy clusters with their gravitational lensing signal. The f (R) model is of the Hu–Sawicki type
(Equation 16) (38). The sDGP model corresponds to the self-accelerating branch where gravity is weakened,
whereas the nDGP models show normal-branch DGP with a quintessence component added to produce an
expansion history identical to CDM (with rc = 500 or 3,000 Mpc for nDGP-1 or nDGP-2, respectively)
(115). Modified from Reference 114.

matter halos in f (R) simulations, which consistently include the chameleon mechanism (113).
Halos with a potential depth larger than the field value − N > ∼ 2| f R0 |/3 become screened, whereas
lower-mass halos are unscreened. Because unscreened halos accrete mass at a rate higher than in
GR, this transition effect is also manifested in the halo mass function (89). The halo of our own
Galaxy has to be screened in order for a chameleon/symmetron model to satisfy Solar System tests,
so these models can be expected to show modified gravity effects only in lower-mass halos. This
effect motivates the search for modified gravity effects in nearby dwarf galaxies (see Section 4.1.3).
We now turn to the Vainshtein screening mechanism. For simplicity, we restrict the discussion
to the cubic galileon interaction term (Equation 18). In the case of spherical symmetry, the equation
of motion in the subhorizon limit can be integrated once to yield
   
dϕ M (<r) r
∝ g , where g(ξ ) = ξ 3
1 + ξ −3 − 1 . 40.
dr M Pl r 2 r∗ (r)
Here, M (<r) is the mass (over the background density) enclosed within r. The scale r∗ (r) ∝
[M (<r)/M Pl 3 ]1/3 is the r-dependent Vainshtein radius (this scaling also holds for higher-order
galileon interactions). In the case of the DGP model, the scale is given by
 1/3  
16M (<r)rc2 Ḣ
r∗ (r) = , β = 1 ± 2Hr c 1 + . 41.
9M Pl β
2 2 3H 2

www.annualreviews.org • Dark Energy Versus Modified Gravity 111

NS66CH05-Schmidt ARI 5 September 2016 11:58

We see that dϕ/dr is suppressed in comparison to the Newtonian gradient M Pl d N /dr for
r/r∗ 1, where the quantity (r/r∗ )3 is directly proportional to the inverse of the interior density
−1
CMB: cosmic ρ (<r). Specifically, Vainshtein screening occurs at fixed interior density, in contrast to the
microwave chameleon type, which occurs at fixed mass. The threshold density for screening is of order the
background background matter density today for natural model parameters (in the case of DGP, rc ∼ H 0−1 ),
BAO: baryon acoustic which for nonlinear LSS leads to a qualitatively different behavior compared with the chameleon
oscillations case (Figure 3b): Halos of all masses are screened within a fixed fraction of their scale radius.
Thus, Solar System constraints do not force us to look to certain types of objects for interesting
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

signatures of Vainshtein-type models.

3.4.3. Tensor perturbations. In addition to modifying the scalar perturbations, modified gravity
generally alters the propagation of tensor modes. More specifically, a running of the gravitational
coupling causes a change in the decay of the gravitational wave amplitude in an expanding universe.
Furthermore, the G4 and G5 terms of a Horndeski scalar–tensor theory alter the propagation speed
of tensor modes (23, 116).

4. COSMOLOGICAL TESTS
In this section, we provide an overview of the observables and experimental methods that are used
to probe gravity and dark energy on cosmological scales.

4.1. Observables
Modified gravity and dynamical dark energy models in general change the background expansion
history of our Universe with respect to CDM. The predicted deviation can be tested by measuring
distances to astronomical objects and cosmic “standard rulers.” We refer to these measurements as
geometric probes. For instance, a comparison between the magnitudes of high-redshift and low-
redshift Type Ia supernovae, serving as standard candles, yields a measurement of the evolution
of the luminosity distance d L (z) = (1 + z)r(z), where r(z) ≡ (1 + z)d A(z) is the comoving angular
diameter distance. For a spatially flat universe, r(z) is simply the comoving radial distance χ (z) =
z 
0
d z /H(z ). This is a relative distance measurement, although calibration with the low-redshift
“distance ladder” also yields an absolute distance measurement. Observations of Cepheid stars in
supernova host galaxies yield H 0 = (73.8 ± 2.4) km s−1 Mpc−1 (117) for the Hubble constant
today. Measurements of the acoustic peaks in the cosmic microwave background (CMB), and their
imprint on LSS—the baryon acoustic oscillations (BAO) observed in galaxy clustering—serve as
a standard ruler, and provide complementary information about the absolute distance scale to
z> ∼ 0.1. These geometric probes, together with recent Planck measurements, constrain a constant
dark energy equation of state in a spatially flat universe at w = −1.006 ± 0.045 (118), consistent
with the CC in CDM.
When dark energy is evolving or gravity is modified, it leaves imprints in the formation of
the LSS, as shown by Equation 28. Modified gravity models can match the CDM expansion
history at an observationally indistinguishable level, whereas the growth of cosmic structure may
still be significantly modified, rendering the latter a vital probe for modified gravity. Nonetheless,
geometric probes provide important constraints on cosmological parameters, which limit degen-
eracies in the growth of structure constraints. In the following subsections, we review some of the
important observables of cosmological structures that are used to test modified gravity and dark
energy. Figure 4 summarizes some of the cosmological constraints that have been inferred on
f (R) gravity, which serves here as a representative for typical modified gravity models that has

112 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

EG*

0 ISW: ∂tΦγ

Galaxy-ISW: Ψ, ∂tΦγ

CMB lensing: Φγ
–2 Density profile: Φγ
log10 |fR0|
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

P(k): clustering: Ψ;
RSD: Ψ; lensing: Φγ
Gas*
–4
Abundance: Ψ

* Ψ versus Φγ
Solar System
–6
Isolated dwarfs Unscreened P(k)

0.0 0.5 1.0 1.5 2.0 2.5 3.0

log10 (r h/Mpc)

Figure 4
Current and forecasted constraints on f (R) gravity, in particular the parameter f R0 of the Hu–Sawicki model
(Equation 16), and potentials tested. Linear constraints: integrated Sachs–Wolfe (ISW) and galaxy–ISW
cross-correlations (120); cosmic microwave background (CMB) lensing (121), the E G probe (122), and power
spectrum constraints from galaxy clustering; redshift-space distortions (RSD); and weak gravitational lensing
(WL) (123). Nonlinear constraints: cluster abundance (120, 124, 125), cluster density profiles (119), and the
comparison of cluster gas and weak lensing measurements (126). Also indicated are the currently tightest
constraints on | f R0 | from Solar System tests (38) and distance indicators in isolated dwarf galaxies (127).
Future cosmological surveys measuring the power spectrum P (k < 0.3 h/Mpc) in unscreened regions at the
1% level will be able to outperform the astrophysical constraints (128). Modified from Reference 119.

also been particularly well tested. We show the range of scales covered by different cosmological
observables and indicate which of the metric potentials in Equation 25 they probe. Table 1 pro-
vides a summary of the current observational constraints on the modified gravity models discussed
in Section 2.

4.1.1. Cosmic microwave background. The CMB radiation is the primary observable for ob-
taining information about our cosmos. Of particular importance for dark energy and modified

Table 1 Current observational constraints on modified gravity models

Screening mechanism Representative model Linear cosmology Nonlinear cosmology Submegaparsec
Chameleon gravitya | f R0 | < −4 | f R0 | < −5 | f R0 | < −7 −6
f(R) ∼ 10 ∼ 10 ∼ (10 –10 )
Symmetron Equation 13b None None < (10−7 –10−6 )
χ∼
Vainshtein nDGPc H 0 rc > H 0 rc > H 0 rc > −4
∼ (0.1–1) ∼ 0.1 ∼ 10
Vainshtein Cubic galileond Incompatible ISW Incompatible voids Compatible

a
From References 38, 123, and 125–127.
b
From References 30 and 127.
c
From References 6 and 149–152.
d
From References 47, 153, and 154. Abbreviations: ISW, integrated Sachs–Wolfe effect; nDGP, normal branch Dvali–Gabadadze–Porrati model.

www.annualreviews.org • Dark Energy Versus Modified Gravity 113

NS66CH05-Schmidt ARI 5 September 2016 11:58

gravity are constraints on the amplitude and adiabaticity of initial perturbations. However, late-
time modifications manifest themselves in the CMB temperature and polarization only via sec-
ISW effect: ondary anisotropies. The presence of dark energy or a CC gives rise to the integrated Sachs–Wolfe
integrated (ISW) effect (129), a fluctuation in the temperature field at the largest scales due to the change
Sachs–Wolfe effect in energy of the CMB photons ∝ ˙ γ when traversing the evolving metric potentials. This effect
SZ effect: is modified in alternative theories of gravity or dynamical dark energy, as shown by Equation 36.
Sunyaev–Zel’dovich A related secondary anisotropy in the CMB is due to the depths of the potentials γ , which
effect
determine the gravitational lensing of the CMB photons by the cosmological structure between
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

WL: weak the observer and the last-scattering surface. The directions of the observed CMB photons are
gravitational lensing
displaced by the lensing deflection angle d. The effect is often described by the convergence field
κ = ∇ θ · d, where ∇ θ is the gradient operator on the sphere. On smaller angular scales, the
Sunyaev–Zel’dovich (SZ) effect, caused by the energy gain of CMB photons by collisions with
electrons in the postreionization universe, provides a complementary signature that probes gravi-
tationally induced motion of matter and, hence, the dynamical potential  rather than the lensing
potential γ . The fairly small kinetic SZ effect (130) measures the line-of-sight component of the
large-scale electron momentum density. The larger thermal SZ effect (131) is induced by virialized
structures. In order to extract these late-time signatures, it can also be useful to consider their
cross-correlation with the foreground structure.
At linear order, the ISW temperature fluctuation, the lensing convergence, and the galaxy
density perturbation are given as functions of the position on the sky θ̂ by
 χ∗
dt ∂
T ISW (θ̂) = −2 dχ γ (χ θ̂, χ ), 42.
dχ ∂t
 χ∗ 0
χ∗ − χ 2
κ(θ̂) = dχ ∇θ γ (χ θ̂, χ ), 43.
 0 χ∗ χ∗ χ
dz
g(θ̂) = dχ b[z(χ )](z)δ(χ θ̂, χ ), 44.
0 dχ
respectively, where χ∗ denotes the conformal radial distance to the last-scattering surface, the
galaxy bias b is scale independent but dependent on redshift, and  is a normalized selection
function. For multipoles  > ∼ 10, one can employ the subhorizon limit and Limber approximation
to write the angular cross-power spectra of these quantities as

H (z) 
CX Y = dz 2 [FX (k, z) FY (k, z) P(k)]k=/χ (z) , where 45.
χ (z)
3H 02 m,0 ∂
FISW (k, z) ≡ T CMB G(k, z),
k2 ∂z
Fg (k, z) ≡ b(z)(z)D(k, z),
3H 02 m,0 χ (χ∗ − χ )
Fκ (k, z) ≡ G(k, z), and
2H (z) χ∗
1
G(k, z) ≡ [1 + γ (k, z)]μ(k, z)(1 + z)D(k, z).
2
Here, we have used Equation 36, defined D(z, k) ≡ δ(z, k)/δ(0, k) as the scale-dependent growth
factor, and denoted P(k) as the matter power spectrum today.

4.1.2. Large-scale structure. The bulk of cosmological information on dynamical dark energy
and modified gravity is contained in the three-dimensional (3D) LSS at low redshifts z < ∼ 5.
Galaxy shape correlations probe the lensing potential γ via WL, specifically the convergence
κ (Equation 43, with χ∗ replaced with the comoving distance to the source galaxies). Similarly,
angular correlations of galaxies in redshift slices probe the matter power spectrum on large

114 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

scales (Equation 45). Furthermore, the 3D power spectrum of galaxies traces the underlying
distribution of matter on large scales via

Pg (k, z) = b 2(z) P(k, z) + 2b(z) k̂2 Pδθ (k, z) + k̂4 Pθθ (k, z), 46. RSD: redshift-space
distortions
where k̂ is the cosine of k with the line of sight. The peculiar motion of galaxies leads to the
terms involving the cross- and autocorrelation of the matter velocity divergence θ [redshift-
space distortions (RSD) (132)], which can be used to measure the growth rate of structure
f ≡ d ln D/d ln a. Note that the galaxy bias b cannot be predicted from first principles, and must
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

be either constrained using lensing or marginalized over (see Section 4.5).

The statistics of matter and γ can be measured well into the nonlinear regime through galaxy
clustering and WL, where Equations 45 and 46 are no longer correct. Nonlinear effects become
important on scales of k > −1
∼ 0.1 h Mpc (depending on redshift), and potentially even larger scales in
modified gravity. Upcoming surveys will measure statistics of galaxy counts and lensing with high
signal-to-noise ratios out to wave numbers k > 1 h Mpc−1 , so that the bulk of information is in
the nonlinear regime. The interpretation of measurements in the nonlinear regime is complicated
by nonlinear gravitational evolution, baryonic feedback effects, and screening in modified gravity.
Because a fundamental description of the nonlinear structure for general dark energy and modified
gravity theories is lacking, observational tests have therefore been limited to specific models.
A further, fully nonlinear probe is the abundance of galaxy clusters, governed by the halo
abundance d/d ln M (Equation 29), which yields tight constraints on the allowed deviations from
CDM (Figure 4) (124); their density profiles have also been used to probe modified gravity
(119). Clusters can be identified via their galaxy content, X-ray emission, or SZ signal. A crucial
ingredient for cosmological constraints is the relation between observable and halo mass. This
relation can be determined through gravitational lensing of background galaxies or the CMB,
which yields the lensing mass M L (<r) ∝ r 2 ∇ γ (r). The gas temperature T gas and pressure Pgas
profiles measured in X-rays and SZ (where SZ measures only Pgas ) yield estimates of the dynamical
mass M D (<r) ∝ r 2 ∇(r). For this purpose, one assumes hydrostatic equilibrium, which in GR
yields
1 dPgas (r) G M D (r)
=− , 47.
ρgas dr r2
where for thermal pressure Pgas ∝ ρgas T gas ∝ Pe . The dominant systematic uncertainty in M D from
these measurements is due to the poorly known nonthermal pressure components. Similarly, the
velocity dispersion of galaxies within clusters probes M D , although the interpretation is hampered
by uncertainties in the velocity bias of galaxies. This uncertainty becomes smaller on larger scales,
where galaxies exhibit coherent inflow motion onto clusters (133, 134).
As noted in Section 3.4, a difference between M L and M D is generic to modified gravity. Along
with the screening effects discussed in Section 3.4.2, this difference clearly needs to be taken into
account when constraining modified gravity using cluster abundance. The discrepancy can also be
used as probe of modified gravity itself (see Section 4.5). The screening of modified gravity effects
in dense regions motivates the use of abundance, clustering, and profiles of underdense regions
(i.e., voids) as a modified gravity probe (86).
Finally, with the advent of gravitational wave astronomy, demonstrated with the first detec-
tion from aLIGO (135), and from aVIRGO and eLISA in the future, it will become feasible to
test the modifications in the propagation of gravitational waves described in Section 3.4.3 (also
see Reference 136). In particular, a direct comparison of the arrival times between gravitational
waves and the electromagnetic emissions from a single event will place tight constraints on de-
viations between the propagation speed of tensor modes and the speed of light. Furthermore,

www.annualreviews.org • Dark Energy Versus Modified Gravity 115

NS66CH05-Schmidt ARI 5 September 2016 11:58

tight constraints on the G4 and G5 terms of Horndeski gravity (137) are placed by binary pulsar
timing, as these are not screened by the Vainshtein mechanism (assuming no additional screening
mechanism is active).

4.1.3. Submegaparsec scales. Although small scales lie beyond the scope of this review, we note
that, in addition to the Solar System constraints, the tightest bounds on chameleon/symmetron
modifications are currently inferred from stellar distance indicators (standard candles) in nearby
dwarf galaxies residing in a low-density region of space (127). This test uses the combination
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

of red giant stars, which are expected to be screened, with Cepheids, whose pulsating envelopes
are still unscreened. The difference in screening introduces a systematic deviation between the
distances inferred. These constraints depend crucially on a correct identification of unscreened
environments for these stars, obtained from a reconstruction of the local density field.

4.2. Consistency Tests: Geometry and Growth

The most common method of obtaining cosmological constraints on dark energy is to combine
various geometric probes of the expansion history and observations of the growth of structure to
obtain joint constraints on the dark energy equation of state w. Because only quintessence unam-
biguously predicts the growth factor D(a) once the expansion history is fixed, these constraints
apply only to this type of dark energy. Moreover, as discussed in Section 2.1, a constant w = −1
is not necessarily a good assumption, so a more general parameterization (138, 139),

w(t) = w0 + wa [1 − a(t)], 48.

is often adopted. Any observational sign for w = −1 or wa = 0 would be evidence against

CDM. In order to test for evidence beyond quintessence, Linder (140) proposed extending this
parameterization by a parameter γ̃ , motivated by the fact that the growth factor in a variety of
quintessence models satisfies f = d ln D(a)/d ln a = m (a)γ̃ , where γ̃ ≈ 0.55. Thus, finding
γ̃ = 0.55 would imply clustering dark energy, modified gravity, or an interacting dark sector
(142). Mortonson et al. (141) and Ruiz & Huterer (143) devised more general consistency tests
between geometry and growth (Figure 5).
Although the consistency tests are model independent, it is generally not true that findings
consistent with CDM rule out any modification of the concordance model, as these parameters do
not encompass the entire space of possible dark energy and modified gravity models. Furthermore,
it is not straightforward to turn a constraint on γ̃ into a constraint on a given modified gravity or

Geometric Parameterized
tests tests
w ≠ −1? Quintessence μ, γ
YES μ≠1

γ≠1
Small cs
Consistency dark energy Targeted
tests tests
YES
Growth
inconsistent Presence of
Modified fifth force?
with H(z)? gravity
YES
Figure 5
Summary of the various types of cosmological tests and their implications. Detection of deviations from
CDM phenomenology in these four classes of tests implies different physics beyond the standard
cosmology (arrows).

116 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

non-quintessence-type dark energy model, as the parameterization is based on linear subhorizon

scales; even if the constraint can be mapped onto a model parameter, it will generally not be an
optimal constraint.

4.3. Parameterized Tests of Gravity

We can improve several of the disadvantages of the consistency tests discussed above by adopting
the more physical parameterization introduced in Equation 36, which incorporates physical con-
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

straints such as mass and momentum conservation. Quintessence models correspond to μ = γ = 1,

while clustering dark energy can be captured by γ = 1 and μ ∝ 1 + w. Note however that this
parameterization is still restricted to linear scales.
Without further constraints, this parameterization involves two free functions of both time
and scale. Zhao et al. (144) performed a principal component analysis (PCA), for bins in k and z,
of μ and γ to determine how well the data can constrain the eigenmodes of μ and γ along with
redshift bins in w(a). They found that the future LSST survey will be able to constrain the 10
best eigenmodes in μ and γ to ∼3% and ∼10%, respectively. Although PCA is useful to address
the possible modifications in full generality, it may allow for too much freedom to detect a well-
defined deviation predicted by a particular clustering dark energy or modified gravity model. As
discussed in Section 3.4.1, assumptions such as, at most, second-order derivatives or a Horndeski
Lagrangian allow us to significantly reduce the freedom in the parameterization, leading to, for
example, Equation 37. One might further expect the modifications to scale as the effective dark
energy density DE = 1 − K − m , providing an even more restrictive parameterization of the
five functions p 1−5 (t) = p i,0 DE (t)/ DE,0 . Moreover, one could argue that if the gravitational
modifications are to drive cosmic acceleration, then any scale appearing in the equations should
be of order the Hubble scale today. Then, μ and γ are completely determined by one function of
time each on subhorizon scales k a H , which motivates the restriction of cosmological tests of
μ and γ to the two constants p 3,0 / p 5,0 and p 2,0 / p 3,0 . Note that this last restriction fails to capture
f (R) and other chameleon/symmetron models.
The main caveat to this type of parameterization is that it is limited to linear scales. Although
more parameters can be introduced to model a screening-induced suppression of μ and γ on
nonlinear scales, doing so clearly does not realistically model the intrinsically nonlinear screening
effects. Finally, even on linear scales, constraints on μ and γ are weakened when allowing for other
nonstandard ingredients, as discussed in the following section.

4.4. Cosmic Degeneracies

When comparing theoretical predictions of dark energy and modified gravity models against
cosmological observations, it is important to discriminate the effects from other signatures of new
fundamental physics and complex nonlinear processes. In particular, massive neutrinos or baryonic
feedback effects can compensate for the effects of an enhanced growth of structure in the power
spectrum or the halo mass function (145, 146). Similar effects can be produced by non-Gaussian
initial conditions. It has also been pointed out that Horndeski scalar–tensor modifications are
endowed with sufficient freedom to allow the recovery of the background expansion history and
linear LSS of a CDM universe (although these models require some fine-tuning) (147), which
limits a fundamental discrimination between modified gravity and dark energy or a CC based on
cosmological structure on large scales. Importantly, the modified propagation of tensor modes
(Section 3.4.3) allows these models to be distinguished from CDM.

www.annualreviews.org • Dark Energy Versus Modified Gravity 117

NS66CH05-Schmidt ARI 5 September 2016 11:58

4.5. Targeted Tests

In order to avoid known and unknown degeneracies, it is desirable to devise tests specifically
E G : combination of targeting the distinction between dark energy and modified gravity that we have introduced in
galaxy–lensing and
Section 2.3. More precisely, we would like to devise tests that constrain a universally coupled fifth
galaxy–velocity
cross-correlations that force. The most promising such test uses the generically predicted difference between dynamical
is sensitive to a fifth () and lensing ( γ ) signatures. In the absence of a fifth force, this difference could be sourced
force while canceling only through a large anisotropic stress in the dark sector, which is difficult to obtain (Section 3.3).
the unknown galaxy On subhorizon linear scales, such a test can be performed by combining the cross-correlation
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

bias (Equation 49)

between foreground galaxies and lensing, Pg(∇ 2 γ ) , and the galaxy density–velocity cross-
correlation of the same population of galaxies, Pgθ . The latter can be extracted from the 3D
clustering (Equation 46). The ratio (148)
Pg(∇ 2 γ) 1 m
EG ≡  (1 + γ )μ 49.
Pgθ 2 f
cancels the linear galaxy bias and isolates a combination of the μ and γ parameters that are induced
by fifth forces, providing a robust test of gravity on linear scales. Note that f depends on μ as well.
On smaller, mildly nonlinear scales, another comparison between dynamics and lensing is
possible through the use of the infall of galaxies onto massive clusters (133, 134). WL around
clusters constrains the spherically averaged profile of γ . The two-dimensional cross-correlation
between galaxies and clusters on scales ∼5–20 Mpc, which measures a projection of the galaxy
phase space, contains information about the infall velocity and thus . This probe goes beyond E G
in that it is not restricted to linear scales. However, effects such as a bias between galaxy velocities
and those of the underlying matter distribution need to be carefully controlled.
On even smaller scales, one can compare the lensing mass M L with the dynamical mass M D
of clusters (measured using X-rays or SZ; see Section 4.1.2) or galaxies (measured using stellar
velocities). If γ is scale independent, and in the absence of screening, M D /M L = (1 + γ )3/5
for virialized objects (114). For viable models however, taking screening into account is essential
(Figure 3) (114, 126).

5. OUTLOOK
We have provided a brief overview of different physical mechanisms to explain the accelerated
expansion of the universe, and introduced theoretical and phenomenological distinctions between
the two scenarios of dark energy and modified gravity. In the next 15 years, LSS and CMB surveys
(AdvACT, eBOSS, DES, DESI, Euclid, HSC/PFS, LSST, POLARBEAR, SPT-3G, WFIRST,
and others) will have the potential to constrain dynamical dark energy and departures from GR
at the few-percent level. This will either rule out a large swath of the interesting parameter space
of dark energy and modified gravity models or yield another breakthrough in cosmology with the
detection of departures from CDM. Thus, this area of cosmology is certain to yield interesting
results in the near future.

SUMMARY POINTS
1. The weak and strong equivalence principles provide a means to rigorously distinguish
dark energy, modified gravity, and an interacting dark sector at the theory level
(Figure 1).

118 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

2. Quintessence dark energy is completely characterized by its equation of state w(t). Dark
energy physics beyond a canonical scalar field can be probed by searching for an incon-
sistency between geometry of the expanding universe as indicated by H(z) and growth
of the density fluctuation δ(k, z) (Figure 5).
3. At the phenomenological level, modified gravity can be probed by searching for fifth
forces via comparison of dynamics with lensing or targeting screening effects (Figure 3).
These signatures are very difficult to mimic with dark energy.
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

DISCLOSURE STATEMENT
The authors are not aware of any affiliations, memberships, funding, or financial holdings that
might be perceived as affecting the objectivity of this review.

ACKNOWLEDGMENTS
We thank Alex Barreira, Dragan Huterer, Eiichiro Komatsu, Eric Linder, and Filippo Vernizzi
for comments and Paolo Creminelli for permission to use Figure 2. A.J. was supported in part
by the Kavli Institute for Cosmological Physics at the University of Chicago through grant NSF
PHY-1125897, an endowment from the Kavli Foundation and its founder Fred Kavli, and by the
Robert R. McCormick Postdoctoral Fellowship. L.L. was supported by a Science and Technology
Facilities Council (STFC) Consolidated Grant for Astronomy and Astrophysics at the University
of Edinburgh and a Postdoc.Mobility Grant from the Swiss National Science Foundation (SNSF)
(grant 161058). F.S. acknowledges support from the Marie Curie Career Integration Grant (FP7-
PEOPLE-2013-CIG) “FundPhysicsAndLSS.”

LITERATURE CITED
1. Riess AG, et al. Astron. J. 116:1009 (1998)
2. Perlmutter S, et al. Astrophys. J. 517:565 (1999)
3. Weinberg S. Rev. Mod. Phys. 61:1 (1989)
4. Martin J. C. R. Phys. 13:566 (2012)
5. Frieman J, Turner M, Huterer D. Annu. Rev. Astron. Astrophys. 46:385 (2008)
6. Clifton T, Ferreira PG, Padilla A, Skordis C. Phys. Rep. 513:1 (2012)
7. Wetterich C. Nucl. Phys. B 302:668 (1988)
8. Peebles PJE, Ratra B. Astrophys. J. Lett. 325:L17 (1988)
9. Ratra B, Peebles PJE. Phys. Rev. D 37:3406 (1988)
10. Caldwell RR, Dave R, Steinhardt PJ. Phys. Rev. Lett. 80:1582(1998)
11. Caldwell RR, Linder EV. Phys. Rev. Lett. 95:141301 (2005)
12. Wetterich C. Astron. Astrophys. 301:321 (1995)
13. Copeland EJ, Liddle AR, Wands D. Phys. Rev. D 57:4686 (1998)
14. Ferreira PG, Joyce M. Phys. Rev. Lett. 79:4740 (1997)
15. Zlatev I, Wang LM, Steinhardt PJ. Phys. Rev. Lett. 82:896 (1999)
16. Sandvik H, Tegmark M, Zaldarriaga M, Waga I. Phys. Rev. D 62:023511 (2000)
17. Chiba T, Okabe T, Yamaguchi M. Phys. Rev. D 62:023511 (2000)
18. Armendáriz-Picón, Mukhanov VF, Steinhardt PJ. Phys. Rev. D 63:103510 (2001)
19. Cheung C, et al. J. High Energy Phys. 03:014 (2008)
20. Creminelli P, D’Amico G, Norena J, Vernizzi F. J. Cosmol. Astropart. Phys. 0902:018 (2009)
21. Gubitosi G, Piazza F, Vernizzi F. J. Cosmol. Astropart. Phys. 1302:032 (2013)

www.annualreviews.org • Dark Energy Versus Modified Gravity 119

NS66CH05-Schmidt ARI 5 September 2016 11:58

22. Bloomfield JK, Flanagan, Park M, Watson S. J. Cosmol. Astropart. Phys. 1308:010 (2013)
23. Gleyzes J, Langlois D, Piazza F, Vernizzi F. J. Cosmol. Astropart. Phys. 1308:025 (2013)
24. Bellini E, Sawicki I. J. Cosmol. Astropart. Phys. 7:050 (2014)
25. Lue A, Scoccimarro R, Starkman G. Phys. Rev. D 69:044005 (2004)
26. Brans C, Dicke RH. Phys. Rev. 124:925 (1961)
27. Will CM. Living Rev. Relativ. 17:4 (2014)
28. Khoury J, Weltman A. Phys. Rev. Lett. 93:171104 (2004)
29. Khoury J, Weltman A. Phys. Rev. D 69:044026 (2004)
30. Hinterbichler K, Khoury J. Phys. Rev. Lett. 104:231301 (2010)
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

31. Capozziello S. Int. J. Mod. Phys. D 11:483 (2002)

32. Capozziello S, Carloni S, Troisi A. Recent Res. Dev. Astron. Astrophys. 1:625 (2003)
33. Carroll SM, Duvvuri V, Trodden M, Turner MS. Phys. Rev. D 70:043528 (2004)
34. Erickcek AL, Smith TL, Kamionkowski M. Phys. Rev. D 74:121501 (2006)
35. Barrow JD, Cotsakis S. Phys. Lett. B 214:515 (1988)
36. Chiba T. Phys. Lett. B 575:1 (2003)
37. Joyce A, Jain B, Khoury J, Trodden M. Phys. Rep. 568:1 (2015)
38. Hu W, Sawicki I. Phys. Rev. D 76:064004 (2007)
39. Starobinsky AA. J. Exp. Theor. Phys. Lett. 86:157 (2007)
40. De Felice A, Tsujikawa S. Living Rev. Relativ. 13:3 (2010)
41. Woodard RP. Scholarpedia 10(8):32243 (2015)
42. Deffayet C, Gao X, Steer DA, Zahariade G. Phys. Rev. D 84:064039 (2011)
43. Carroll SM, Hoffman M, Trodden M. Phys. Rev. D 68:023509 (2003)
44. Cline JM, Jeon S, Moore GD. Phys. Rev. D 70:043543 (2004)
45. Horndeski GW. Int. J. Theor. Phys. 10:363 (1974)
46. Kobayashi T, Yamaguchi M, Yokoyama J. Prog. Theor. Phys. 126:511 (2011)
47. Nicolis A, Rattazzi R, Trincherini E. Phys. Rev. D 79:064036 (2009)
48. de Rham C, Gabadadze G. Phys. Rev. D 82:044020 (2010)
49. Dvali GR, Gabadadze G, Porrati M. Phys. Lett. B 485:208 (2000)
50. Luty MA, Porrati M, Rattazzi R. J. High Energy Phys. 09:029 (2003)
51. de Rham C. C. R. Phys. 13:666 (2012)
52. Langlois D, Noui K. J. Cosmol. Astropart. Phys. 1602:034 (2016)
53. Fierz M, Pauli W. Proc. R. Soc. A 173:211 (1939)
54. de Rham C, Gabadadze G, Tolley AJ. Phys. Rev. Lett. 106:231101 (2011)
55. Arkani-Hamed N, Georgi H, Schwartz MD. Ann. Phys. 305:96 (2003)
56. Creminelli P, Nicolis A, Papucci M, Trincherini E. J. High Energy Phys. 9:003 (2005)
57. Boulware DG, Deser S. Phys. Rev. D 6:3368 (1972)
58. Hassan SF, Rosen RA. Phys. Rev. Lett. 108:041101 (2012)
59. Hinterbichler K. Rev. Mod. Phys. 84:671 (2012)
60. de Rham C. Living Rev. Relativ. 17:7 (2014)
61. Deffayet C, Dvali GR, Gabadadze G. Phys. Rev. D 65:044023 (2002)
62. Deffayet C. Phys. Lett. B 502:199 (2001)
63. Nicolis A, Rattazzi R. J. High Energy Phys. 6:059 (2004)
64. Lue A. Phys. Rep. 423:1 (2006)
65. Dvali G, Gabadadze G, Shifman M. Phys. Rev. D 67:044020 (2003)
66. Arkani-Hamed N, Dimopoulos S, Dvali G, Gabadadze G. arXiv:hep-th/0209227 (2002)
67. Dvali G, Hofmann S, Khoury J. Phys. Rev. D 76:084006 (2007)
68. Weinberg S. Phys. Rev. B 138:988 (1965)
69. Feynman R, Morinigo F, Wagner W, Hatfield B. Feynman Lectures on Gravitation. Boulder, CO: West-
view (1996)
70. Pietroni M. Phys. Rev. D 72:043535 (2005)
71. Olive KA, Pospelov M. Phys. Rev. D 77:043524 (2008)
72. Damour T, Polyakov AM. Nucl. Phys. B 423:532 (1994)
73. Brax P, van de Bruck C, Davis AC, Shaw D. Phys. Rev. D 82:063519 (2010)

120 Joyce · Lombriser · Schmidt

NS66CH05-Schmidt ARI 5 September 2016 11:58

74. de Rham C, Ribeiro RH. J. Cosmol. Astropart. Phys. 1411:016 (2014)

75. Babichev E, Deffayet C, Ziour R. Int. J. Mod. Phys. D 18:2147 (2009)
76. Milgrom M. Astrophys. J. 270:365 (1983)
77. Babichev E, Deffayet C, Esposito-Farese G. Phys. Rev. D 84:061502 (2011)
78. Khoury J. Phys. Rev. D 91:024022 (2015)
79. Vainshtein AI. Phys. Lett. B 39:393 (1972)
80. de Rham C, Tolley AJ, Wesley DH. Phys. Rev. D 87:044025 (2013)
81. Chu YZ, Trodden M. Phys. Rev. D 87:024011 (2013)
82. Gralla SE. Phys. Rev. D 81:084060 (2010)
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

83. Will CM. Theory and Experiment in Gravitational Physics. Cambridge, UK: Cambridge Univ. Press (1993)
84. Silvestri A, Trodden M. Rep. Prog. Phys. 72:096901 (2009)
85. Bekenstein JD. Phys. Rev. D 51:6608 (1995)
86. Hui L, Nicolis A, Stubbs C. Phys. Rev. D 80:104002 (2009)
87. Hiramatsu T, Hu W, Koyama K, Schmidt F. Phys. Rev. D 87:063525 (2013)
88. Cooray A, Sheth RK. Phys. Rep. 372:1 (2002)
89. Schmidt F, Lima MV, Oyaizu H, Hu W. Phys. Rev. D 79:083518 (2009)
90. Tinker JL, et al. Astrophys. J. 688:709 (2008)
91. Bond JR, Cole S, Efstathiou G, Kaiser N. Astrophys. J. 379:440 (1991)
92. Press WH, Schechter P. Astrophys. J. 187:425 (1974)
93. Tinker JL, et al. Astrophys. J. 688:709 (2008)
94. Mo HJ, White SDM. Mon. Not. R. Astron. Soc. 282:347 (1996)
95. Navarro JF, Frenk CS, White SDM. Astrophys. J. 490:493 (1997)
96. Creminelli P, et al. J. Cosmol. Astropart. Phys. 3:27 (2010)
97. de Putter R, Huterer D, Linder EV. 81:103513 (2010)
98. Armendáriz-Picón C. J. Cosmol. Astropart. Phys. 7:7 (2004)
99. Wei H, Cai RG. Phys. Rev. D 73:083002 (2006)
100. Koivisto T, Mota DF. Phys. Rev. D 73:083502 (2006)
101. Mota DF, Kristiansen JR, Koivisto T, Groeneboom NE. Mon. Not. R. Astron. Soc. 382:793 (2007)
102. Hu W. Astrophys. J. 506:485 (1998)
103. Farrar GR, Peebles PJE. Astrophys. J. 604:1 (2004)
104. Amendola L, Baldi M, Wetterich C. Phys. Rev. D 78:023015 (2008)
105. Baldi M, Pettorino V, Robbers G, Springel V. Mon. Not. R. Astron. Soc. 403:1684 (2010)
106. Bertschinger E. Astrophys. J. 648:797 (2006)
107. Hu W, Sawicki I. Phys. Rev. D 76:104043 (2007)
108. Gleyzes J, Langlois D, Mancarella M, Vernizzi F. J. Cosmol. Astropart. Phys. 8:054 (2015)
109. Silvestri A, Pogosian L, Buniy RV. Phys. Rev. D 87:104015 (2013)
110. Bertschinger E, Zukin P. Phys. Rev. D 78:024015 (2008)
111. Baker T, Ferreira PG, Skordis C. Phys. Rev. D 87:024015 (2013)
112. Wang J, Hui L, Khoury J. Phys. Rev. Lett. 109:241301 (2012)
113. Oyaizu H. Phys. Rev. D. 78:123523 (2008)
114. Schmidt F. Phys. Rev. D. 81:103002 (2010)
115. Schmidt F. Phys. Rev. D. 80:123003 (2009)
116. Saltas ID, Sawicki I, Amendola L, Kunz M. Phys. Rev. Lett. 113:191101 (2014)
117. Riess AG, et al. Astrophys. J. 730:119 (2011); Riess AG, et al. Erratum. Astrophys. J. 732:129 (2011)
118. Planck Collab. arXiv:1502.01589 [astro-ph] (2015)
119. Lombriser L, et al. Phys. Rev. D 85:102001 (2012)
120. Lombriser L, Slosar A, Seljak U, Hu W. Phys. Rev. D 85:124038 (2012)
121. Hu B, Liguori M, Bartolo N, Matarrese S. Phys. Rev. D 88:123514 (2013)
122. Reyes R, et al. Nature 464:256 (2010)
123. Hojjati A, et al. Phys. Rev. D 93:043531 (2016)
124. Schmidt F, Vikhlinin A, Hu W. Phys. Rev. D 80:083505 (2009)
125. Cataneo M, et al. Phys. Rev. D 92:044009 (2015)
126. Terukina A, et al. J. Cosmol. Astropart. Phys. 1404:013 (2014)

www.annualreviews.org • Dark Energy Versus Modified Gravity 121

NS66CH05-Schmidt ARI 5 September 2016 11:58

127. Jain B, Vikram V, Sakstein J. Astrophys. J. 779:39 (2013)

128. Lombriser L, Simpson F, Mead A. Phys. Rev. Lett. 114:251101 (2015)
129. Sachs RK, Wolfe AM. Astrophys. J. 147:73 (1967)
130. Sunyaev RA, Zel’dovich YB. Mon. Not. R. Astron. Soc. 190:413 (1980)
131. Sunyaev RA, Zel’dovich YB. Annu. Rev. Astron. Astrophys. 18:537 (1980)
132. Kaiser N. Mon. Not. R. Astron. Soc. 227:1 (1987)
133. Lam TY, Nishimichi T, Schmidt F, Takada M. Phys. Rev. Lett. 109:051301 (2012)
134. Zu Y, et al. Mon. Not. R. Astron. Soc. 445:1885 (2014)
135. Abbott P, et al. Phys. Rev. Lett. 116:061102 (2016)
Downloaded from www.annualreviews.org. Guest (guest) IP: 90.173.222.136 On: Mon, 31 Mar 2025 16:23:35

136. LIGO Sci. Collab., Virgo Collab. arXiv:1602.03841 [gr-qc] (2016)

137. Jimenez JB, Piazza F, Velten H. Phys. Rev. Lett. 116:061101 (2016)
138. Chevallier M, Polarski D. Int. J. Mod. Phys. D 10:213 (2001)
139. Linder EV. Phys. Rev. Lett. 90:091301 (2003)
140. Linder EV. Phys. Rev. D 72:043529 (2005)
141. Mortonson MJ, Hu W, Huterer D. Phys. Rev. D 79:023004 (2009)
142. Ishak M, Upadhye A, Spergel DN. Phys. Rev. D 74:043513 (2006)
143. Ruiz EJ, Huterer D. Phys. Rev. D. 91:063009 (2015)
144. Zhao GB, Pogosian L, Silvestri A, Zylberberg J. Phys. Rev. Lett. 103:241301 (2009)
145. Baldi M, et al. Mon. Not. R. Astron. Soc. 440:75 (2014)
146. Puchwein E, Baldi M, Springel V. Mon. Not. R. Astron. Soc. 436:348 (2013)
147. Lombriser L, Taylor A. J. Cosmol. Astropart. Phys. 1603:031 (2016)
148. Zhang P, Liguori M, Bean R, Dodelson S. Phys. Rev. Lett. 99:141302 (2007)
149. Lombriser L, Hu W, Fang W, Seljak U. Phys. Rev. D 80:063536 (2009)
150. Raccanelli A, et al. Mon. Not. R. Astron. Soc. 436:89 (2013)
151. Xu L. J. Cosmol. Astropart. Phys. 1402:048 (2014)
152. Wyman M, Khoury J. Phys. Rev. D 82:044032 (2010)
153. Barreira A, Li B, Baugh C, Pascoli S. J. Cosmol. Astropart. Phys. 1408:059 (2014)
154. Barreira A, et al. J. Cosmol. Astropart. Phys. 8:028 (2015)

122 Joyce · Lombriser · Schmidt