Master Thesis

Gravitational Redshift of Clusters in the Symmetron

and Chameleon f (R) Models

Max Grönke

June 2013

Institute of Theoretical Astrophysics

University of Oslo
Abstract

The gravitational redshift in clusters of galaxies can be measured to probe gravity as was
first done by Wojtak et al. (2011). Using N -body simulations we are able to analyze the
gravitational redshift profiles for the symmetron and the Hu-Sawicky f (R) models. The
characteristic feature of these models is the screening mechanism that hides the fifth force
in dense environments recovering general relativity. We find that due to the nature of the
screening, the deviation with respect to ΛCDM is highly dependent on the halo mass. So the
f (R) parameters |fR0 − 1| = 10−5 , n = 1 cause an enhancement of the gravitational signal
by up to 60% for halos with masses between 3 × 1012 M⊙ h−1 and 1014 M⊙ h−1 . However, for
both larger and smaller halos there is hardly any deviation present. The characteristic mass
range, where the fifth force is most active, varies with the model parameters. Additionally, we
observe not only a stronger but also a possible weaker gravitational redshift. This contradicts
the usual assumption that the presence of a fifth force leads to a deeper potential well but may
happen if screening in the central regions is active which results in a shift of the additional
clustering to larger radii.
Acknowledgment

First, I would like to thank my two supervisors, David F. Mota and Claudio Llinares, for
their valuable help and their support throughout my thesis. I am thankful to David F. Mota
for the trust he put in me by giving me an exciting and challenging thesis topic.
A special thanks goes to Claudio Llinares, who devoted his time and energy to help me
proceeding with my work, even whilst abroad. He also read through my thesis and gave me
very constructive feedback.
I would also like to thank all the people I have been working with throughout this thesis
and, especially, to Hans A. Winther for the numerous discussions and answers he provided
to my stupid questions.
In addition, I am grateful to Rahel Frick, Andres Spicher and Michael Küffmeier for reading
through my thesis and giving me feedback.
Moreover, thanks to my family for supporting me and helping me financially to pursue my
studies in Oslo. Thanks also to my friends and fellow students, especially Elisabeth Jordahl
and Eunseong Lee, for making this time particularly pleasant.
Contents

Introduction 1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Notation and conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1 Preliminaries 5
1.1 General relativity in a nutshell . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 The Einstein-Hilbert action . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.2 Gravitational redshift . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.3 The Friedmann equations . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 Dark energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.1 ΛCDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.2 Quintessence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.3 Einstein and Jordan frame . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.4 Scalar-tensor theories . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.5 f (R)-gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.6 f (R)-gravity as a scalar-tensor theory . . . . . . . . . . . . . . . . . . 16

2 Review of the Chameleon and Symmetron Models 18

2.1 Introduction to the chameleon mechanism . . . . . . . . . . . . . . . . . . . . 19
2.1.1 The chameleon in outer space . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.2 The chameleon close to masses . . . . . . . . . . . . . . . . . . . . . . 22
2.2 The symmetron model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.1 Cosmology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2.2 Spherical solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3 The Hu-Sawicky f (R) model . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Gravitational redshift in the symmetron and chameleon models . . . . . . . . 30

3 The N -body Code and Simulation Parameters 32

3.1 Introduction to N -body simulations . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.1 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.2 Gravity calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2 The simulation code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2.1 RAMSES code structure . . . . . . . . . . . . . . . . . . . . . . . . . . 37
ii Contents

3.2.2 The ISIS code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.3 Simulation parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.4 Post-processing: Identifying the halos . . . . . . . . . . . . . . . . . . . . . . 40

4 Spherical Collapse and Virialization 43

4.1 Derivation of the virialization equation . . . . . . . . . . . . . . . . . . . . . . 43
4.2 Spherical collapse in standard models . . . . . . . . . . . . . . . . . . . . . . 47
4.2.1 Einstein-de Sitter universe . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2.2 ΛCDM model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3 Spherical collapse in screened modified gravity . . . . . . . . . . . . . . . . . 51
4.3.1 Symmetron implementation . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3.2 f (R) implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3.3 Conclusions of the spherical collapse model . . . . . . . . . . . . . . . 61
4.4 Virialization state in N -body simulations . . . . . . . . . . . . . . . . . . . . 62
4.4.1 Virialization parameter in scalar-tensor theories . . . . . . . . . . . . . 68

5 Gravitational Redshift in Screened Modified Gravity 69

5.1 Obtaining the gravitational redshift . . . . . . . . . . . . . . . . . . . . . . . 69
5.2 The NFW profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.3 Uncertainty estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.4 Cluster profiles in the ΛCDM model . . . . . . . . . . . . . . . . . . . . . . . 74
5.5 The f (R) models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.5.1 Matter power spectra and halo mass functions . . . . . . . . . . . . . 75
5.5.2 Chameleons in the wild: The fifth force unbound . . . . . . . . . . . . 76
5.5.3 Gravitational redshift profiles . . . . . . . . . . . . . . . . . . . . . . . 79
5.5.4 The halo density profiles . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.5.5 Velocity profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.6 The symmetron models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.6.1 Matter power spectra and halo mass functions . . . . . . . . . . . . . 85
5.6.2 Symmetron additional force . . . . . . . . . . . . . . . . . . . . . . . . 86
5.6.3 Gravitational redshift profiles . . . . . . . . . . . . . . . . . . . . . . . 88
5.6.4 Halo density profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.6.5 Velocity profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.7 Connection to observational data . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.8 Systematic effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.9 Summary of results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

6 Conclusions 98
6.1 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

A Appendix 101
A.1 Derivation of the Klein-Gordon equation . . . . . . . . . . . . . . . . . . . . . 101
A.2 Gauss Seidel method with successive overrelaxation . . . . . . . . . . . . . . . 101

Bibliography 104
Introduction

Although there is overwhelming evidence that dark energy exists (see Amendola & Tsujikawa
(2010, Chap. 5) or Wang (2009, Chap. 4-7) for observational aspects), its nature is still
a mystery. There are two possible explanations for this: Either dark energy is some type
of yet unknown matter or general relativity is not fully correct and therefore requires some
modification which can account for the phenomena attributed to dark energy.
In the first approach, the new matter would have to possess an equation of state equal or close
to −1 and its detection would be a milestone of particle physics. In this case, the presently
well-known problems of such a cosmological constant solution, namely the coincidence and
fine-tuning problem, would be uncovered as midleading hints.
Pursuing the second approach, often titled the modification of gravity, intense efforts were
made in order to establish new theories (see e.g. Clifton et al. (2012) for a theoretical
overview). The main problems concerning the development of new gravitational theories are
of theoretical, like the existence of ghosts, as well as observational kind, meaning the exper-
imental constraints from local experiments are very tight. One explanation for the possibly
undiscovered variations are so-called screening mechanisms, where the local constraints are
fulfilled because of the additional force is hidden in dense environments such as the solar
system.
In this work, two models with screening are discussed: The symmetron model (Hinterbichler
& Khoury, 2010) and an f (R)-gravity model that uses chameleon screening (Hu & Sawicki,
2007). The first model works through the dependency of the vacuum expectation value of
the scalar field on the local matter density, whereas in the latter, the mass of the scalar
field is linked to the matter density. Both mechanisms result in the same outcome: In low-
density regions, a fifth force is mediated and in high-density regions this force is suppressed.
This behavior requires astrophysical tests in order to examine the existence of these scalar
fields.
Due to the non-linearity of the screening, N -body simulations are a preferable tool to make
observational predictions. For both models, this has been done before (in the symmetron
case by Brax et al. (2012), Davis et al. (2011) and Winther et al. (2011) and in the chameleon
case by Brax et al. (2013), Li et al. (2012), Li & Hu (2011) and Zhao et al. (2011)) with a
focus on large scale imprints (i.e. mostly the deviation in the matter power spectra and halo
mass functions were analyzed). In this work, the ISIS code (Llinares et al., 2013) was used
2 Introduction

to perform a total of 8 simulation runs over a variety of model parameters. The resolution
obtained through 5123 particles in a 256 Mpc h−1 box allows us to analyze deviations in
halo properties such as the density or velocity dispersion profiles. These observables can be
analyzed using lensing (Johnston & Sheldon, 2007; Oguri et al., 2012; Okabe et al., 2013).
Possible constraints for the f (R) model have been established by Lombriser et al. (2012)
and Schmidt (2010).
Another possible test of gravity scales below 1 Mpc is the use of gravitational redshift, as
the wavelength shift of light is directly proportional to the depth of the potential well of the
clusters of galaxies. This has been observed for the SDSS survey data by Wojtak et al. (2011)
who compared the data points to analytical prediction for general relativity based on the
NFW-profile (Navarro et al., 1995), an analytical TeVeS prediction1 and a semi-analytical
profile from Schmidt (2010). For the latter, the NFW prediction has been boosted by the
factor 4/3 which is the maximum enhancement of gravitational strength according to f (R)
gravity.
We simulate gravitational redshift profiles of clusters in the three gravitational models and
show that such comparisons and predictions have to be made with extreme caution when
dealing with screening models. The main reason for this is that the deviation may not occur
with the same strength in all mass and length scales with the same strength. In particular,
three kind of effects were observed: (i) An enhanced gravitational redshift profile, which
is linked to additional clustering in the center of halos; (ii) a weakened effect, pointing
to clustering in the outskirts, or (iii) no deviation from ΛCDM if the additional force is
screened in this regime.
When doing a comparison of cluster properties, it is essential we only study dynamical relaxed
objects. Therefore, a big part of this work is about sample selection and virialization, as for
scalar-tensor theories the potential energy of the scalar field has to be taken into account as
well when determining the virialization state.

Overview

The first part of this work is meant to give a rough introduction to the subject: First, the
theory of general relativity and the possibility of its derivation from the Einstein-Hilbert
action are revised. Then, the “dark energy problem” is briefly summarized and scalar fields
as a possible solution as well as the link to f (R) models are presented. Here, focus is put on
the the two screening mechanisms analyzed in this work: The chameleon and the symmetron
models, where insights are given as to how the screening in these two cases works. Also, the
Hu-Sawicky model is shown to be a chameleon theory.
In the second part, the approach chosen for this work, namely the use of N -body simula-
tions in order to find possible imprints of the modified gravity theories on smaller scales, is

1 The analytical TeVeS profile used was later shown to be based on inappropriate assumptions (Bekenstein
& Sanders, 2012).
Notation and conventions 3

discussed. In particular, the sample selection and the problematics of virialization of halos is
laid out. This is specifically done by applying the spherical collapse model to the symmetron
and f (R) case.
Finally, the output from the N -body simulation is presented and discussed with a focus on
the variation in the gravitational redshift profiles. The results obtained are also compared
against current measurements.

Notation and conventions

Throughout this work units were used so that c = ~ = 1, except in some cases when the
symbols are added for clarity. This means all the units can be expressed in terms of energy
and the following conversions from SI units apply:

1 meter = 5.068 × 10−39 GeV −1

1 second = 1.519 × 1024 GeV
1 kilogram = 5.610 × 1026 GeV

The metric signature adopted is (−, + , + ,+) and the Einstein summation convention applies
with Greek indices running from 0 to 3 and Latin indices from 1 to 3. In addition, a subscript
0 usually refers to the value today.

List of frequently used symbols

Symbol Name Definition or Value

G Gravitational constant × 10−39 GeV−2
G = 6.71 √
MPl Reduced Planck mass MPl = 1/ 8πG = 2.44 × 1018 GeV
M⊙ Mass of Sun M⊙ = 1.99 × 1030 kg
c Speed of light c=1
~ Reduced Planck constant ~=1
a Cosmic scale factor, normalized to unity
today
H Hubble parameter H = ȧ/a
ρ (Energy) density
p pressure
w Equation of state w = p/ρ
ρc Critical density of the universe ρc = 3H 2 MPl2 = 3H 2 /(8πG)

Ωi Relative density Ωi = ρi /ρc,0

Φ Newtonian gravitational potential
zg , vg Gravitational redshift vg = zg c
φ Scalar field
β Dimensionless coupling constant
4 Introduction

List of frequently used symbols – continued from previous page

Symbol Name Definition or Value
˙ Derivative with respect to time
,µ , ∂ µ Partial derivative with respect to xµ ∂ µ = ∂ /∂ xµ
,φ , ∂ φ Partial derivative with respect to φ ∂ φ = ∂ /∂ φ
;µ , ∇µ Covariant derivative with respect to xµ
 D’Alembert operator  = ∇µ ∇µ
∇, ∇ ~ Del- or Nabla-Operator in three spatial
dimensions
gµν Metric tensor
ηµν Minkowsky tensor η = diag(−1,1,1,1)
α
Γµν Christoffel symbols
Rµν Ricci tensor
R Ricci scalar
Gµν Einstein Tensor
Tµν Energy momentum tensor
W Potential energy
T Kinetic energy
ES Surface pressure term
βvir Virialization parameter βvir ≡ (2T − Es )/W + 1
Rvir , Rv Virial radius
1

Preliminaries

This chapter gives a short introduction to the mathematical basics needed further on. First,
the theory of relativity is discussed in brief and the possibility of its derivation from the
Einstein-Hilbert action is mentioned as the same procedure is followed later on for modified
actions. Also, the cause of gravitational redshift and the Friedmann equations are given as a
bonus. Secondly, some dark energy models are presented with focus on scalar-tensor theories
and f (R) gravity since the models analyzed in this work are of this type.

1.1 General relativity in a nutshell

It is not the aim of this work to have the character of a book. Therefore, this section is
merely a listing of equations used in a later section for completeness. The books of Carroll
(2004), Hartle (2003) and Misner et al. (1973) were used as reference and are recommended
as a proper introduction to the general theory of relativity.
Starting with a metric tensor gµν describing the line element in a 4-dimensional spacetime
ds2 = gµν xµ xν , the Christoffel symbols are defined as

α gασ
Γµν ≡ (gµσ,ν + gνσ,µ − gµν,σ ) . (1.1)
2
The Ricci tensor can be obtained:
α α α β α β
Rµν = Γµν,α − Γµα,ν + Γµν Γαβ − Γµβ Γαν . (1.2)

The contraction thereof gives the Ricci scalar:

R = Rµµ = gµν Rµν . (1.3)

The Einstein equations can then be written as

gµν
Rµν − R = 8πGTµν (1.4)
2
6 1 Preliminaries

or, by defining the Einstein tensor as Gµν ≡ Rµν − gµν R/2 and κ = 8πG one can obtain a
more compact version:

Gµν = κTµν (1.5)

The tensor Tµν on the right hand side of the equation is the the energy-momentum tensor,
which for a perfect fluid takes the form

Tµν = (ρ + p)uµ uν + pgµν . (1.6)

Here p is the pressure, ρ the density and uµ the 4-velocity of the fluid.

1.1.1 The Einstein-Hilbert action

Equation (1.5) can also be recovered by varying the Einstein-Hilbert action given by
√
−g
Z
4
SEH = d x R (1.7)
2κ
and the action of the matter fields Sm with respect to the metric gµν :

1 4  √ √
Z
d x δ( −g)R + −g (δgµν Rµν + gµν δRµν ) + δSm

δS = (1.8)
2κ
√ √
Note that g is the determinant of gµν and therefore δ( −g) = − 12 −ggµν δgµν . This can be
derived by writing the determinant as g = gµν M(µν) , with the determinant of the cofactor
matrix M(µν) not depending on gµν .
The variation of the Ricci tensor (defined in equation (1.2)) yields
α α α β α β α β α β
δRµν = δΓµν,α − δΓµα,ν + δΓµν Γαβ + Γµν δΓαβ − δΓµβ Γαν − Γµβ δΓαν . (1.9)

Comparing this term with the covariant derivative of δΓµν α , which is – as the difference

between two connection symbols – a tensor and, therefore, given by

α α α σ σ α σ α
δΓµν;γ = δΓµν,γ + Γσγ δΓµν − Γµγ δΓσν − Γγν δΓµσ (1.10)

a much simpler expression for δRµν can be found:

α α
δRµν = δΓµν;α − δΓµα;ν . (1.11)

Thus, the whole variation of the Ricci scalar is

δR = Rµν δgµν + gµν δRµν (1.12)

 
= Rµν δgµν + gµν δΓµν;α
α α
− δΓαµ;ν (1.13)
 
= Rµν δgµν + gµν δΓµν
α
− gµα δΓµν
ν
. (1.14)
;α
1.1 General relativity in a nutshell 7

In the last equality the fact that the covariant derivative of the metric is zero (gµν;α = 0)
has been applied.

Consequently, using the Gauss’ theorem, the last term of (1.8) vanishes if the variation of
the metric goes to zero at infinity and we are left with

1 √ 1
Z  
4
δS = dx −g Rµν − Rgµν − κTµν δgµν . (1.15)
2κ 2

Here the definition of the energy-momentum tensor

1 √
Z
4
δSm =− dx −gTµν δgµν (1.16)
2
has been used. Using the Lagrangian density, Eq. (1.16) can be rewritten as

2 δLm
Tµν = − √ . (1.17)
−g δgµν

Demanding the action principle δS = 0, results in Eq. (1.5).

1.1.2 Gravitational redshift

The picture often referred to is that photons lose energy as they have to “climb out” of
potential wells caused by the bending of space-time due to the existence of matter. The
following section provides the mathematical background to this.

First note that in general the quantity

ξ · p = gµν ξ µ pν (1.18)

is conserved (Hartle, 2003, Chap. 8) with ξ being the Killing vector and p the four-
momentum. The four-momentum consists of the energy and the classical three-momentum,
i.e.: pµ = (E,~
p) and the Killing vector is a normalized vector that points in the direction in
which the metric is preserved. In the following we will assume a time-independent metric
and therefore the Killing vector is ξ µ = (1,0,0,0). In addition, the assumption g0i = 0 is
made.

As the four-velocity is normalized (uµ uµ = −1) and we assume a stationary observer (ui = 0),
we obtain for the time component:
1
u0obs = √ . (1.19)
−g00

This means, the energy of a photon received by the observer, which is in general given by
E = −p · u, becomes

Eobs = (−g00 )−1/2 (ξ · p)obs . (1.20)

8 1 Preliminaries

Using the conservation of ξ · p stated above and E = h/λ, we obtain for the redshift:
s
λobs (g00 )⋆
z≡ −1= −1. (1.21)
λ⋆ (g00 )obs

Here the subscripts ⋆ and obs were used for quantities to be evaluated at the emitters and
observers position, respectively.

The metric usually used to quantify this effect is the Schwarzschild metric, which describes
the consistence of space-time around a spherically symmetric, non-rotating object. The line
element is given by:
h  i
ds2 = − e2Φ dt2 + e−2µ dr 2 + r 2 dθ + sin2 θ dϕ2 , (1.22)

with the constants Φ and µ reducing to the gravitational potential in the Newtonian limit
(Misner et al., 1973, Chap. 23).

As this metric fulfills the requirements of the assumptions made above, the redshift is given
by (1.21) and reads:

z = eΦ⋆ −Φobs −1 ≈ Φ⋆ − Φobs . (1.23)

1.1.3 The Friedmann equations

Using the Friedmann-Lemaître-Robertson-Walker (FLRW) spacetime, which describes a ho-

mogeneous and isotropic universe, the line-element is given by
!
2 µ ν 2 dr 2
2
ds = gµν dx dx = − dt + a (t) + r 2 dΩ 2 (1.24)
1 − kr 2

with the curvature k, the expansion factor a(t) and the differential solid angle dΩ 2 =
dθ 2 + sin2 θ dϕ2 . The values 1, 0 and −1 for k correspond to an open, flat and closed
universe respectively.

The energy-momentum tensor for a perfect fluid takes the form of Eq. (1.6). In comoving
coordinates uµ is simply (−1,0,0,0) and gµν = ηµν = diag(−1,1,1,1). Consequently, in this
case, the non-zero elements of the energy-momentum tensor are T(00) = −ρ and T(ii) = p.

Inserting these two ingredients into Eq. (1.5) leads to

8πG k
H2 = ρ− 2 (1.25)
3 a
for the (00) component and

ä k
2 + H 2 = −8πGp + 2 (1.26)
a a
1.1 General relativity in a nutshell 9

for the spacial components. The Hubble parameter is defined by

ȧ
H≡ . (1.27)
a
Rewriting the two equations and using w ≡ p/ρ one can obtain the second Friedmann
equation (which can also be found directly using the trace of the Einstein equations):
ä 4 k
= − πGρ(3w + 1) + 2 (1.28)
a 3 a
and the continuity equation
ρ̇ + 3Hρ(1 + w) = 0 . (1.29)
The solution hereof is
a0 3(1+w)
 
ρ = ρ0 . (1.30)
a
Quantities with a subscript 0 denote the values today and the expansion factor is normalized
to unity today.
Now these solutions can be rewritten in a more compact form with the help of some useful
definitions. First, the critical density is
3H 2
ρc = . (1.31)
8πG
Using this relation, the relative density of the fluid components can be expressed as:
ρi
Ωi = . (1.32)
ρc
In a similar way the relative curvature is
k
Ωk = − . (1.33)
(aH)2

Finally, the evolution of a universe filled with different matter i with relative densities today
Ωi,0 and equation of state wi is given as:
H2 X
= Ωi,0 a−3(1+wi ) + Ωk0 a−2 . (1.34)
H02 i
From this equation a general analytic solution for a(t) cannot be found. Nevertheless, under
the assumption of a unique fluid with equation of state w dominating in a flat universe
(Ωk = 0), integration yields:

eH0 t if w = −1
a(t) =  3 2/(3(1+w)) (1.35)
 (1 + w)H0 t otherwise
2
√
This means, a flat universe filled with radiation (wr = 1/3) evolves as a(t) ∝ t or ordinary
matter (wm = 1) as a ∝ t2/3 .
See for instance Hartle (2003, Chap. 18) or Carroll (2004, Chap. 8) for a detailed description
about the evolution of different universe models.
10 1 Preliminaries

1.2 Dark energy

Supernovae observations (Perlmutter et al., 1999; Riess et al., 1998) showed that the expan-
sion of the universe is accelerating. From the second Friedmann equation (1.28) it can be
seen that such a behavior would be possible with a new component of the energy-momentum
tensor with an effective equation of state w < −1/3. This could be new a form of matter
that behaves totally differently to any known form of matter or it could be a modification
of the field equations of general relativity.
What this component is, is absolutely unclear but it is called dark energy and the additional
observational evidence about its existence is overwhelming (see for example Amendola &
Tsujikawa (2010, Chap. 5)). Some of the evidence are:
• The age of the universe without dark energy is smaller than the age of globular clusters
measured by Carretta et al. (2000) or Hansen et al. (2002).
• Measurements of the cosmic microwave background as done by the WMAP (Larson
et al., 2011) or Planck (Planck Collaboration et al., 2013) project estimate the amount
of dark energy of around 73% and restrain w very close to −1.
• Other phenomena such as Barionic Acoustic Oscillations and development of Large-
Scale Structures.
The field of dark energy research is huge and various books and reviews have been published
(e.g. Amendola & Tsujikawa (2010) and Clifton et al. (2012) which have mainly been used
as reference for this section). This section tries to give a brief argument on how the analyzed
models fit into the bigger picture of the dark energy sector. In order to do this, first the
standard model – ΛCDM – is introduced, then the “Quintessence” part show how scalar field
can play a role and at last scalar-tensor theories, f (R)-models and their possible equivalence
are shown. For the last point a short intermezzo, where the concept of conformal frames is
explained, was also added.

1.2.1 ΛCDM

The easiest model that accounts for these points is based on the introducing a new term
√
−2 −gΛ/16πG with the cosmological constant Λ into the action Eq. (1.7). The new field
equations then reads:
gµν
Rµν − R + Λgµν = 8πGTµν . (1.36)
2
This model is called ΛCDM because of its main components: Dark energy (Λ) and cold dark
matter (CDM).
To redefine the Einstein or the energy-momentum tensor in order to obtain the same form
as in (1.5) is a matter of taste and interpretation: Is gravity itself modified or is another
type of matter included in the universe? Although this is often just two sides of the same
coin, it is very common to follow the latter path and assign the density ρΛ = Λ/8πG and its
1.2 Dark energy 11

equation of state wΛ = pΛ /ρΛ = −1. Inserting these in Eq. (1.6) leaves the initial additional
term of −Λgµν as stated above.

An equation of state equal to minus one leads to interesting behavior of the fluid: As can be
seen from the continuity equation Eq. (1.29), the density stays constant. This means dark
energy gains more importance since the density of the other forms of matter decreases in an
expanding space. Another fact is the exponential expansion, stated already in Eq. (1.35).

There are two major problems with the ΛCDM model: Firstly, although the density of
matter ρm decreases with a−3 and the density of dark energy ρΛ stays constant, these two
quantities are roughly the same just today. Because this happens only once in the whole
history of the universe, it seems to be a very big coincidence this event happened just recently.
Therefore, this problem is called “coincidence problem”. The second famous problem with
the cosmological constant is the so called “fine tuning problem”: The observed value of Λ
is around 120 orders of magnitude smaller than the vacuum energy found with a cut-off at
the Planck scale. This discrepancy can be lowered to around 50 orders of magnitude with
certain super-symmetric theories (Martin, 2012), but it is still too large to “fine tune” it
away.

Other problems including missing satellite galaxies (Mateo, 1998) and non-matching of ob-
served and simulated galaxy profiles (de Blok, 2010) are more related to CDM.

1.2.2 Quintessence

A hypothetical form of matter that could account for the dark energy phenomena is a simple
scalar field. In an FLRW metric the pressure and density of a scalar field are given as usual
by the time and space components of the energy-momentum tensor (Fujii & Maeda, 2007,
Sec. 5.2):

(φ) 1
pφ = Tii = φ̇2 − V (φ)
2 (1.37)
(φ) 1
ρφ = −T00 = φ2 + V (φ) .
2
Consequently, the equation of state is
1 2
pφ 2 φ̇ − V (φ)
wφ = = 1 2 . (1.38)
ρφ 2 φ̇ + V (φ)

This makes a scalar field a dark energy candidate because if φ̇2 /2 ≪ V (φ) (“slowly rolling”)
the matter of state is wφ ≈ −1. However, in opposition to the equation of state of a
cosmological constant wφ is changing in time.

In order to study the behavior of the quintessence model, the same steps as for the ΛCDM
model can be done: Vary the action and demand that this variation is zero. So let’s start
12 1 Preliminaries

with the action, which has an extra term for the scalar field φ and is given(Fujii & Maeda,
2007) by
√ 1
Z  
4
Sφ = − dx −g ∂ µ φ∂ µ φ + V (φ) (1.39)
2

with the potential V (φ). Variation of the action with respect to φ results in the Klein-Gordon
equation

φ − V ′ (φ) = 0 . (1.40)

This can be derived using δSφ = Sφ (φ + δφ) − Sφ (φ). Assuming the variation δφ vanishes at
R 4 √
the boundary, the first term becomes then − d x −g∂ µ φ∂ µ δφ. After partial integration
we obtain the form above. For the full derivation see Appendix A.1
Assuming a homogeneous scalar field and a surrounding FLRW metric the equation turns
into:

φ̈ + 3H φ̇ + V ′ = 0 . (1.41)

In quintessence models, a wide range of initial conditions may converge quickly to the same
cosmological evolution, so called “tracker solutions” (Steinhardt et al., 1999). As these
tracker solutions may solve partly the coincidence problem, they still need fine tuning (e.g. of
the potential) in order to explain why the accelerated expansion happens today (Armendariz-
Picon et al., 2000).

1.2.3 Einstein and Jordan frame

In modified gravity models the Einstein-Hilbert action Eq. (1.7) is usually modified. Never-
theless, because general relativity can explain a large number of observations, the modified
actions are often constructed in a way that the Einstein-Hilbert action can be recovered in
some limit.

Name Equation
Metric gµν = Cg̃µν
Determinant g = C 4 g̃  
Christoffel Symbol α
Γµν = Γ̃µν α ˜ µ C + δα ∇
+ C −1 δνα ∇ ˜ ν C − g̃µν ∇
˜ α C /2
 µ 
Ricci Tensor Rµν = R̃µν + C −2 3∇ ˜ µC∇ ˜ ν C − Cg̃µν C
˜ − 2C ∇ ˜ µ∇
˜ ν C /2
 
Ricci Scalar R = R̃/C + 3C −3 ∇ ˜ αC ∇
˜ α C − 2C C
˜ /2
 
D’Alembert Operator φ = C −2 ∇˜ α φ∇
˜ α C + C φ
˜

Table 1.1: Table of some important quantities expressed in terms of the metric gµν under the
conformal transformation gµν → g̃µν as defined in equation (1.42).
1.2 Dark energy 13

It is possible to conformally transform the metric (meaning scales are altered, but angles are
unchanged) and state the theory in a different “frame”. Mathematically, this transformation
can be stated as:

gµν = C(x)g̃µν (1.42)

with C(x) being a positive function of the spacetime coordinate xµ . The line elements of
the two frames are therefore related by ds2 = C(x) ds̃2 . Physically, this change is possible
because causality is linked to angles in spacetime, i.e. the metrics g̃µν and gµν contain the
same causal structure, although the norm of the vectors can be changed.
The relations occurring between some quantities of the conformally related frames will be
important later on and are therefore summarized in the table 1.1.
As C(x) is an arbitrary positive function, the theory can be stated in an infinite number of
frames. Two of these frames are very important and frequently used: The Einstein and the
Jordan frame.
In the Jordan frame the energy-momentum tensor is covariantly conserved (T;νµν = 0) and
massless test particles follow the geodesic equation in this metric (Clifton et al., 2012).
In the Einstein frame the action takes the same form as the original Einstein-Hilbert action
(linear in the Ricci scalar) with the addition of some extra matter fields that arouse in the
transformation.
Which of the frames is the “physical” one is subject of discussion (see e.g. Faraoni et al.
(1998)).
From now on, because the two frames will be used frequently, the function C(x) will be used
to describe the transformation between them and not as a general transformation function.

1.2.4 Scalar-tensor theories

Probably one of the best studied modifications of General relativity fall under the group of
scalar-tensor theories, amongst which the most prominent is the Brans-Dicke theory (Brans &
Dicke, 1961). Scalar-tensor theories can be understood as a generalization of the quintessence
model described before. Instead of having a completely free scalar field, a coupling to
matter of order unity is introduced. This stronger coupling is on the one hand much more
natural (Carroll, 1998) but on the other hand it gives rise to a fifth force which would be
observationally detectable.
In the Einstein frame, the action for scalar-tensor theories is given (Clifton et al., 2012) by
( )
R̃ 1 ˜ ˜µ
Z
4 p
S= dx −g̃ − ∇ µ φ∇ φ − V (φ) + Sm (gµν ,Ψm ) . (1.43)
2κ 2
Here, the additional scalar field component Sφ as used in Eq. (1.39) is added to the Einstein-
Hilbert action as an additional component. Keep in mind that particles (here represented
14 1 Preliminaries

by a matter field Ψm ) will follow geodesics of their Jordan frame metric gµν .1 Therefore, in
the Einstein frame particles feel an additional “fifth” force.
Variation of this action with respect to the metric g̃µν leads to
1
   
˜ µ φ∇
G̃µν = κ T̃µν + ∇ ˜ ν φ − κg̃µν ∇α φ∇α φ + V (φ) . (1.44)
2
Since φ is a second degree of freedom, the action can also be varied with respect to φ which
results in the equation of motion of the scalar field
˜ − V ′ (φ) = V+′ (T ,C) .
φ (1.45)

The right hand side depends on the trace of the energy-momentum tensor T = Tµν gµν , which
was defined in Eq. (1.17) and the definition of the function C, which links the Einstein and
the Jordan frame. In later sections, the function C is fixed and the behavior of the scalar field
will be studied. What we can notice already now comparing this result with the equation
of motion obtained for the quintessence model (see Eq. (1.40)), is that φ behaves like a free
scalar field in an effective potential Vef f = V+ + V .

1.2.5 f (R)-gravity

In the f (R) gravity the Ricci scalar R in the Einstein-Hilbert action (1.7) is replaced by a
general function f (R) and the new action reads:
1 √
Z
4
S= dx −gf (R) . (1.46)
2κ
The motivation behind this approach is to assume that f (R) = R is valid for small ranges
where a higher order is to be used for larger scales. Unfortunately while a simple model like
f (R) = R + αR2 can be responsible for inflation, it can not explain the late time acceleration
(Amendola & Tsujikawa, 2010, Chap. 9).
In addition to the late time acceleration, the model has to fulfill also local gravity constraints.
Therefore, only more complex models are not ruled out. Three examples are:
• The model of Hu & Sawicki (2007) given by
µRc
f (R) = R − (1.47)
1 + (R/Rc )−2n

• Starobinsky (2007) come up with

 !−n 
R2
f (R) = R − µRc 1 − 1 +  (1.48)
Rc2

1 The assumption was made that the scalar field couples to all the matter types with the same strength.
(i)
Otherwise there would be several Jordan-frame metrics gµν .
1.2 Dark energy 15

• And the approach from Appleby & Battye (2007):

 
f (R) = R + Rc log e−µ +(1 − e−µ ) e−R/Rc (1.49)

In all examples n, µ and Rc are positive constants.

Without specifying the shape of f (R) further, one can carry out the variation of the action
(1.46) as done before for the Einstein-Hilbert action:

1 √ 1
Z  
4
δS = dx −g − f (R)gµν δgµν + fR (Rµν δgµν + gµν δRµν ) + δSm (1.50)
2κ 2
where the commonly used definition fR ≡ ∂ f /∂ R has been applied. Note that in this case,
the last term inside the parenthesis does not vanish as the multiplication with fR may lead
to nonzero boundary terms. Therefore, the full variation of Rµν has to be analyzed. Recall
that
 
δR = Rµν δgµν + gµν δΓµν;α
α α
− δΓαµ;ν (1.51)

as stated in Eq. (1.13). Varying the Christoffel symbols defined in Eq. (1.1) results in

α gασ
δΓµν = (δgµσ;ν + δgνσ;µ − δgµν;σ ) . (1.52)
2
This can be derived by using the fact that δgµν,α = δgµν;α + Γνα σ δg σ
µσ + Γµα δgνσ and the
non-tensorial terms (i.e. the ones including a connection symbol) have to cancel because as
α is a tensor.
stated before δΓµν
Rewriting Eq. (1.51) using Eq. (1.52) leads to:

δR =Rµν δgµν + ∇µ ∇ν δgµν − gµν δgµν

=Rµν δgµν − ∇µ ∇ν δgµν − gµν δgµν . (1.53)

Now, the total variation of the action can be written as

√ 
−g f
Z
4
δS = d x − gµν δgµν
2κ 2
 (1.54)
+ fR (Rµν δgµν − ∇µ ∇ν δgµν + gµν δgµν ) + δSm
√
−g µν f
Z  
4
= dx δg − gµν + fR Rµν − ∇µ ∇ν fR + gµν fR + δSm , (1.55)
2κ 2

where in the last equality integration by parts (with the assumption that the boundary
terms are negligible) has been performed twice. Furthermore, as done before, by demanding
δS = 0 and using the definition of the energy-momentum tensor (1.16) we obtain:

f
fR Rµν − gµν − ∇µ ∇ν f + gµν fR = κTµν (1.56)
2
16 1 Preliminaries

1.2.6 f (R)-gravity as a scalar-tensor theory

In the previous sections it was presented that the framework of a scalar-tensor theory is
very convenient to find viable models that help solving the known ΛCDM problems. On the
other hand, the f (R) framework can be a very natural extension to general relativity. In
this section, it is shown that the two frameworks are under just two sides of the same coin,
i.e. that under certain conditions an f (R) theory can be rephrased as a scalar-tensor theory
and the other way around.
In order to show the possible equivalence between the f (R)-models and scalar-tensor theories,
we start with the “arbitrary” action
√
−g 
Z
4
S = dx f (χ) + f ′ (χ) [R − χ] , (1.57)
2κ
where χ is a new field that depends on xµ just as R.
Variation with respect to χ yields
δS 1 √
Z
4
= dx −gf ′′ (χ) [R − χ] . (1.58)
δχ 2κ
This means that χ = R given f ′′ (χ) 6= 0. Plugging this result back into the starting equation
(1.57), we obtain
1 √
Z
4
S= dx −gf (R) , (1.59)
2κ
which is the same as the action of the f (R)-gravity Eq. (1.46). Consequently, if the action
(1.57) can be brought to the same form as Eq. (1.43), the equality between the theories has
been proven.
Starting point is the transformation of this action to the Einstein frame. Using the general
transformation rules summarized in table 1.1, the equation can be written as
˜ αC ∇
˜ α C − 6C C
˜
( !)
R̃ 3∇
Z
4 p 2 ′
S= dx −g̃C f +f + 3
−χ . (1.60)
C 2C

Here and for the rest of this section, units such that κ = 1 have been chosen. From this
intermediate result we can see in order to obtain a linear term in R̃, the relation C = 1/f ′
has to hold.
With the transformation function C specified, we obtain after some rearranging:
( 2 )
R̃ f − f ′ χ 3 ˜ α f ′ + 3 f
Z 
4 p
˜ αf ′∇ ˜ ′
S= dx −g̃ + − ∇ (1.61)
2 2f ′2 2f ′ 2f ′

Now, we can introduce the scalar field φ and identify the second term of the integrand with
its potential V :
f − f ′χ
f ′ (χ) ≡ e2βφ/MPl , V =− . (1.62)
2f ′2
1.2 Dark energy 17

Here, β is the dimensionless coupling constant. The factor 2 in the exponent is introduced
in accordance with the literature. An assumptionR to be made at this point is that φ vanishes
4 √
at the boundary, so the total divergence terms ( d x −gφ) disappear, which leaves
( )
R̃
Z
4 ˜ α φ∇
˜ αφ − V
− 3β 2 ∇
p
S= dx −g̃ . (1.63)
2
√
Comparing this result with the canonical form of (1.43), it is clear that β = 1/ 6. If the
coupling parameter takes a different value in a scalar-tensor theory, this theory cannot be
formulated as an f (R)-model.
To summarize the results of this section: An f (R) model can be treated as a scalar-tensor
theory, with the scalar field appearing through the conformal transformation of the metric
given by
1
gµν = e2βφ/MPl g̃µν with β = √ . (1.64)
6
The potential of this scalar field is given by (1.62), which can be rewritten as χ = R resulting
in
f − fR R
V =− . (1.65)
2fR2

Here the notation introduced in the section about f (R) models, 1.2.5, namely fR ≡ df / dR
has been restored.
2

Review of the Chameleon and Symmetron Models

As the matter proportion in the universe is similar to the one of dark energy, there might be
some relation between those. This motivation supports “coupled dark energy” models like
the Brans-Dicke theory (Brans & Dicke, 1961) or Dilaton (Brax et al., 2010a).

A simple coupled dark energy model is the “coupled quintessence” model, in which a scalar
field is coupled to matter with the coupling strength β. This can potentially solve the
coincidence problem, as the energy ratio of the two components dark energy and dark matter
was constant or similar throughout the whole history of the universe and did not change
dramatically in the recent past (Wetterich, 1995).

However, as shown in section 1.2.4, there is an additional force acting on the particles in
scalar-tensor theories. If the coupling constant is around unity, this fifth force should have
been detected.

A possible solution to this problem are so called “screening mechanisms” that lead to a
suppression of this additional force in the solar system and therefore avoiding detection
through local experiments.

In this chapter the two screening mechanisms that this work is related to are reviewed:
The chameleon and the Symmetron models. In both models the screening is environment
dependent but works through different mechanisms:

• The chameleon Mechanism (Brax et al., 2004; Khoury & Weltman, 2004a,b) the
effective mass of the field depends on the local matter density in such a way that in
outer space the field is very light and a fifth force is mediated. If the matter density is
high, the field becomes heavy and the fifth force is suppressed. The chameleon model
is reviewed in section 2.1 and the connection to f (R) theory is made in section 2.3.

• In the Symmetron Model (Hinterbichler & Khoury, 2010; Hinterbichler et al., 2011;
Olive & Pospelov, 2008) the vacuum-expectation value (VEV) of the scalar field de-
pends on the local matter density. In high density regions the VEV is close to zero
but when the matter density is lower, the symmetry of the potential is broken and the
2.1 Introduction to the chameleon mechanism 19

VEV is higher, leading to an additional force in these regions. The symmetron model
in its implications are discussed in section 2.2.

2.1 Introduction to the chameleon mechanism

In the following a short introduction to the chameleon theory is given. Better, more in-depth
reviews, which were also used as reference are for example Khoury & Weltman (2004a) and
Waterhouse (2006).
The equation to start is the general action of a scalar-tensor theory in the Einstein frame
describing a scalar field φ with potential V (φ) as stated already in (1.43):
√ R 1
Z  
4
S= d x −g − ∇µ φ∇µ φ − V (φ) + Sm (gµν
(i)
,Ψm ) . (2.1)
2κ 2
(i)
Here, each of the matter fields couples to its Jordan frame metric gµν which is linked to the
Einstein frame metric via1
(i)
gµν ≡ e2βi φ/MPl gµν (2.2)

with the dimensionless coupling constants βi . This conformal transformation is of the same
kind as the one obtained in the previous section 1.2.6 where the equivalence of the f (R)-
model and the Scalar-tensor theories was shown. However, leaving the coupling βi free means
that the chameleon mechanism chapter cannot necessarily
√ be described in the framework of
the f (R)-models. This is only the case if βi = 1/ 6 for all βi .
After variation of Eq. (2.1) we obtain
X 1 ∂ L 2βi (i)
φ − V,φ (φ) − √ (i) M
gµν = 0 . (2.3)
i
−g ∂ gµν pl

This equation is similar to the equation of motion for the scalar field Eq. (1.45) with the
exception that we can specify the effective potential more precisely. Here the action of matter
(i)
fields ψm has been expressed with the Lagrangian density:
Z  
4 (i) (i)
Sm = − d x L ψm ,gµν . (2.4)

Using the energy-momentum tensor for the ith matter component

(i) 2 ∂L
Tµν =√ µν (2.5)
g(i) ∂ g(i)

1 In order to be in consistence with the literature, the quantities in the Einstein frame will be without and
the ones describing the Jordan frame with an embraced indice or a tilde. As this is just the opposite as
used in the previous chapter, hopefully nothing is too confusing.
20 2 Review of the Chameleon and Symmetron Models

(i)
and the assumption that the matter is a perfect fluid gµν Tiµν = −ρ(i) + 3p(i) = −(1 − 3wi )ρ(i)
one can rewrite the equation of motion to
X 2βi
φ − V,φ (φ) + e4βi φ/MPl (1 − 3wi )ρ̃(i) = 0 , (2.6)
i
MPl

where the exponential term arises from the fact that g = det(g) = exp(−8βi φ/MPl )det(g(i) ).
Please note that for radiation (w = 1/3), the equation of motion is not dependent on the
surrounding density. In Section 2.4 we will see that for this case no fifth force arises.
It would be very convenient to express the equation of motion in terms of some energy density
instead of the rather abstract energy-momentum tensor. Therefore, we look for a quantity
ρ(i) obeying the continuity equation ρ(i) ∝ a−3(1+wi ) in the Einstein frame. Starting from
the conservation of the energy-momentum tensor in the Jordan frame ∇ ˜ ν T̃ µν = 0 (Fujii &
Maeda, 2007), with the covariant derivative ∇ ˜ µ corresponding to the metric g̃µν , one finds
this quantity is

ρ(i) = e3(1+wi )βi /MPl ρ̃(i) . (2.7)

This simplifies the equation of motion to

X βi (i) (1−3wi )βi φ/MPl
φ − V,φ (φ) − (1 − 3wi ) ρ e =0. (2.8)
i
MPl

Introducing an effective potential so that φ = Veff,φ (φ) results in

ρ(i) e(1−3wi )βi φ/MPl .

X
Veff (φ) = V (φ) + (2.9)
i

The shape of the potential V (φ) has to be in a way that φ acts as a cosmological constant,
i.e. φ has to slowly roll down the potential (see section 1.2.2 for an introduction of the
slow-roll mechanism). In order to solve the coincidence problem the scalar field should not
have began to start rolling recently, but instead should be always rolling down the potential.
Therefore, V (φ) should be monotonically decreasing. Furthermore, the following should be
fulfilled for all i ∈ N (Khoury & Weltman, 2004a):

∂ (i) V ∂ (i−1) V ∂ (i) V ∂ (i−1) V

lim V (φ) = 0, lim / = 0 and lim / =∞ (2.10)
φ→∞ φ→∞ ∂ φ(i) ∂ φ(i−1) φ→0 ∂ φ(i) ∂ φ(i−1)

Common choices for the potential are, e.g. the inverse power-law potential

M 4+n
V (φ) = (2.11)
φn
or the exponential potential given by
n /φn
V (φ) = M 4 eM . (2.12)

In both cases M and n are positive constants and M has the units of mass.
2.1 Introduction to the chameleon mechanism 21
VHΦL

Φ Φ
Figure 2.1: chameleon mechanism at work: The solid, dashed and dotted lines show respec-
tively the effective potential Vef f , the bare potential V and the density dependent component.
The black dot illustrates the minimum of the effective potential. On the left side the situation
is shown in a high-density region: The effective minimum is very close to φ = 0 and the fifth
force is screened. In the right graph the field is in a lower density region. φ can be at much
higher values and therefore a non-neglectable fifth force may be present.

As shown in Fig. 2.1 the effective potential Veff can have a minimum in a region where the
matter density ρi is not zero and if V (φ) and βi have the same sign. The value where the
scalar field settles φmin is therefore given by
βi
ρi e(1−3wi )βi φmin /MPl = 0
X
V,φ (φmin ) + (1 − 3wi ) (2.13)
i
MPl

and the mass of the field is defined as

βi2
m2min ≡ Veff,φφ (φmin ) = V,φφ (φmin ) + (1 − 3wi )2 (1−3wi )βi φmin /MPl
X
2 ρi e
MPl
. (2.14)
i

This term is important as the mass is the inverse of the characteristic range of the fifth force
(Khoury & Weltman, 2004a).

2.1.1 The chameleon in outer space

Using the standard FLRW metric introduced in section 1.1.3 and assuming spatial homo-
geneity of the scalar field (∂ i φi = 0) the first term of the equation of motion (2.8) turns
into
φ = gµν ∇µ ∇ν φ (2.15)
00 ii
= g ∂ 0∂ 0φ − g Γii0 φ,0 (2.16)
= −φ̈ − 3H φ̇ . (2.17)
Consequently, we obtain the usual solution for a homogeneous scalar field in the potential
Veff :
φ̈ + 3H φ̇ = −Veff,φ (φ) (2.18)
22 2 Review of the Chameleon and Symmetron Models

2.1.2 The chameleon close to masses

In the following only one pressureless particle species is considered that couples to the metric
g̃µν with coupling parameter β. That is i is fixed and wi is taken to be zero.
The motion of a particle in the Jordan frame is given by the geodesics of the transformed
metric:

ẍα + Γ̃µν
α µ ν
ẋ ẋ = 0 (2.19)
α are the Christoffel symbols of the metric g̃ . Calculating the Christoffel symbols
Here Γ̃µν µν
as in section 1.1 leads to
β
ẍα + Γµν
α µ ν
ẋ ẋ + (2φ,µ ẋµ ẋα + gασ φ,σ ) = 0 . (2.20)
MPl
Taking now the non-relativistic limit (ẋµ ≪ c, gµν ≈ ηµν ) and assuming a static field
(φ,0 = 0) we obtain:

β ~
ẍ = − ∇φ (2.21)
MPl
This shows that φ can be interpreted as the potential giving rise to the chameleon force.
Commonly (as in Khoury & Weltman (2004a)) the variation of the fifth force close to a
spherical body with radius Rc and density ρc embedded in a medium with density ρ∞ is
discussed. The solution is sketched here in brief so that the qualitative behavior of the fifth
force is clear. For a full derivation see Khoury & Weltman (2004a) or Waterhouse (2006).
The equation to solve is the equation of motion of the field (2.8). Expressed in polar coordi-
nates including the assumptions that gµν ≈ ηµν and the non relativistic matter introduced
above this reads
d2 φ 2 dφ β
2
+ = V,φ + ρ(r) eβφ/MPl (2.22)
dr r dr MPl
with the density given as
(
ρc if r < Rc
ρ(r) = (2.23)
ρ∞ if r > Rc .

In both regions the effective potential Veff has a minimum whose positions are given by φc
and φ∞ , respectively. The mass of the field at both positions will be denoted as mc and
m∞ .
In order to solve this equation the following boundary conditions are used
dφ
= 0 at r = 0 (2.24)
dr
φ → φ∞ as r → ∞ . (2.25)
2.1 Introduction to the chameleon mechanism 23

Both make sense from a physical perspective: Firstly,a singularity at the origin is unphysical
and secondly the fifth force should vanish very far away from the object.

In the literature two analytical solutions are usually given: One called the “thin-shell” or
“screened” solution is an approximation for relatively large objects. Here, the potential takes
the value φc for regions that are ∆RC below the surface of the sphere. In this “core” region
the chameleon is fully screened and there is not only no fifth force within this area but the
contribution of volume elements lying within this region is negligible for the φ value outside
the sphere. Far away from the object the value φ ≈ φ∞ is fixed. Consequently, the solution
outside the body is an interpolation between φc and φ∞ :

3β ∆Rc Mc e−m∞ r
φ(s) (r) ≈ − + φ∞ (2.26)
4πMPl Rc r

with r being the distance from the center of the sphere, Mc the mass of the sphere and the
thickness of the shell being defined via

∆Rc |φ∞ − φc |
= (2.27)
Rc 6βMPl Φc

where Φc = Mc /(8πMPl 2 R ) denotes the Newtonian potential of the object. In order that
c
this approximation can be used, this ratio has to be much smaller than one: ∆Rc /Rc ≪ 1.

In the other approximation, not surprisingly referred to as “thick-shell” and “unscreened”,

the object is smaller and ∆Rc /Rc > 1. Here the whole body contributes to the φ-field
outside the object and not just the “shell” as before. The exterior solution is then given by

β Mc e−m∞ r
φ(u) (r) ≈ − + φ∞ . (2.28)
4πMPl r
The strength of the fifth force in these two cases can be approximated using Eq. (2.21).
With the reasonable assumption that m∞ is small, the additional acceleration for the thin
shell – that is the screened – solution is

(s) 6GMc β 2 β 2 ∆Rc ∆Rc

ẍFifth ≈ − êr = 6β 2 ẍN (2.29)
r2 Rc Rc

with ẍN = −GMc /r 2 êr being the acceleration due to the Newtonian force.

For the thick shell, i.e. the unscreened solution, the result obtained is

(u) 2GMc β 2
ẍFifth ≈ − êr = 2β 2 ẍN . (2.30)
r2

This means, the maximum enhancement of the gravitational force is 2β 2 which is equal to
1/3 for f (R)-models as shown in Section 1.2.6.

The results derived here will be used when studying spherical collapse in screened modified
gravity models in Section 4.2.
24 2 Review of the Chameleon and Symmetron Models

2.2 The symmetron model

The symmetron was presented in Hinterbichler & Khoury (2010) and Hinterbichler et al.
(2011), articles which outline this section will follow.
As in the chameleon theory presented in the previous section, the action in the Einstein
frame is given by S = SEH + Sφ + Smatter :
( ! )
Z
√ 2
MPl 1
4
R − ∇µ φ∇µ φ − V (φ) + −g̃Lm (ψ (i) ,g̃µν
p
S= dx −g (2.31)
2 2

where the matter fields ψ (i) couple to the Jordan frame metric given by1

g̃µν ≡ A2 (φ)gµν . (2.32)

This leads to a equation of motion for the scalar field given by

φ − V,φ + A3 (φ)A,φ (φ)T̃ = 0 (2.33)

with the trace of the energy-momentum tensor is

2g̃µν δLm
T̃ = g̃µν T̃µν = − √ . (2.34)
−g̃ δg̃µν
The coupling to matter A(φ) as well as the potential V (φ) have to be of a form so that they
are symmetric under φ → −φ. A commonly considered coupling is
!
φ2 φ4
A(φ) = 1 + +O (2.35)
2M 2 M4

and the simplest potential can be stated as

1 1
V (φ) = − µ2 φ2 + λφ4 . (2.36)
2 4
Here, the variables µ and M have the unit of mass and λ is a positive dimensionless coupling
constant.
This leaves the effective potential from Eq. (2.33) using non relativistic matter (T̃ ≈ −ρ̃)
as

Veff = V (φ) + A3 (φ)A,φ (φ)ρ̃ (2.37)

1 ρ 1
 
2
= 2
− µ φ2 + φ4 , (2.38)
2 M 4
where the density ρ = A3 (φ)ρ̃, which is conserved in the Einstein frame, has been used. In
this equation non-φ dependent terms have been neglected as they do not change the behavior
of the scalar field.

1 This conformal transformation is the equivalent to C = A2 with the function C introduced in section 1.2.3.
2.2 The symmetron model 25

As done in the previous section, one can write the Christoffel symbols in the Jordan frame
as
α
Γ̃µν = Γµν + δνα ∂ µ (log(A)) + δµα ∂ ν (log(A)) − gµν ∂ α (log(A)) . (2.39)

Therefore, the geodesic equation is

ẍα + Γµν
α µ ν
ẋ ẋ + 2∂ σ (log(A)) ẋσ ẋα + gασ ∂ σ (log(A)) = 0 . (2.40)

Here, the fact that gµν ẋµ ẋν = −1 has been used for the last term. Taking now the non-
relativistic limit in flat spacetime, we obtain for the spacial coordinates:
~ (log A) .
ẍ = −∇ (2.41)

This means one can interpret log(A) as an additional potential, giving rise to an fifth force.
As Fig. 2.2 shows, the minimum of this effective potential is located at φmin = 0 if ρ > M 2 µ2
(left panel) and different
√ from zero if the density falls below that threshold (right panel).
Especially φmin = ±µ/ λ if ρ = 0. Consequently, the fifth force is screened in regions with
high density.

2.2.1 Cosmology

As in the section about the chameleon model the behavior of the symmetron model in outer
space is studied using the FLRW metric and non-interacting perfect fluids (T̃ = i (−1 +
P

3wi )ρ̃i ). Then the equation of motion (2.33) becomes

φ̈ + 3H φ̇ + V,φ + A3 (φ)A,φ
X
(1 − 3wi )ρ̃i = 0 (2.42)
i
VHΦL

Φ Φ
Figure 2.2: The symmetron screening effect: The solid, dashed and dotted lines show respec-
tively the effective potential Vef f , the bare potential V and the density dependent component.
The black dot illustrates a minimum of the effective potential. On the left side the situation
is shown in a high-density region: The effective minimum is at φ = 0 and the fifth force is
screened. In the right graph the field is in a lower density region. φ can be at much higher val-
ues and therefore a non-neglectable fifth force may be present.
26 2 Review of the Chameleon and Symmetron Models

or using the definition of ρ introduced above:

X A,φ
φ̈ + 3H φ̇ + V,φ + (1 − 3wi )ρi = 0 (2.43)
i
A3wi

This is the same result as obtained for the chameleon model but with changed effective
potential:

A1−3wi (φ)ρi
X
Veff = V (φ) + (2.44)
i

The geodesic equation (2.40) for non-relativistic particles (v ≪ 1) will become in FLRW
metric
1 i
ẍi + 2H ẋi + 2∂ 0 (log A)ẋi + ∂ (log A) = 0 . (2.45)
a2
Often, the time derivative term is neglected as it is assumed that A changes slowly.

2.2.2 Spherical solution

Similarly as done in the previous section a case were a sphere of density ρ and radius R
in vacuum is studied. Therefore, the equation of motion (2.33) in the Newtonian limit is
written in spherical coordinates and the definition of ρ = A3 ρ̃ has been used:

d2 φ 2 dφ
+ = V,φ + A,φ ρ (2.46)
dr 2 r dr
The boundary conditions are that the solution has to smooth at the origin and take a fixed
value at infinity:
dφ
=0 (2.47)
dr r=0
lim = φ∞ (2.48)
r→∞

For the interior solution the right hand side of (2.46), namely the derivative of the effective
(in)
potential Vef f ,φ can be approximated by Vef f ,φ (φ) = ρφM 2 because ρ ≫ µ2 M 2 . Using this
assumption the solution, which also satisfies the boundary condition Eq. (2.47), is
R ρ
 
(in)
φ (r) = C1 sinh r . (2.49)
r M
The constant C1 will be fixed by demanding a smooth solution at the boundary between the
interior and the exterior solution.
In order to find the exterior solution, the effective potential outside the sphere is approxi-
mated as
(out)
Vef f = µ2 (φ − φ∞ )2 . (2.50)
2.3 The Hu-Sawicky f (R) model 27

This is a reasonable assumption since the potential only gets important near φ∞ . Under
this simplification the solution of Eq. (2.46) which fulfills Eq. (2.48) is
R −√2µ(r−R)
φ(out) (r) = C2 e +φ∞ . (2.51)
r
Fixing the two integration constants C1 and C2 is done by demanding a smooth solution at
the boundary r = R. This results in:
C1 = φ∞ u sechu (2.52)
1
 
C2 = φ∞ u tanh −1 (2.53)
u
Here u2 is a thin-shell factor and can be interpreted the same way as in the chameleon model

∆Rs M2
u2 ≡ = 2 Φ (2.54)
Rs 6MPl
with the Newtonian gravitational potential Φ = ρR2 /6MPl
2 .

In order to complete the connection to the chameleon mechanism to simplify matters for the
future treatment, it is possible (Hinterbichler & Khoury, 2010; Hinterbichler et al., 2011)
to consider the additional force on a test particle around the spherical object. As the mass
of the scalar field in vacuum should be very light, Eq. (2.41) can be expanded around µ in
order to get a approximation for the behavior of the fifth force:
µ ∂φ
ẍFifth ≈ − √ . (2.55)
λM 2 ∂ r
As done in the chameleon section before, two kinds of solutions can be considered in order
to expand Eq. (2.51), the screened (∆R/R ≪ 1) and unscreened (∆R/R ≫ 1) one. In the
screened case the additional acceleration compared to the Newtonian one is
(s) ∆R µ2 MPl2
ẍFifth ≈ 6 , (2.56)
R λM 4
and for unscreened objects this becomes
(u) µ2 MPl
2
ẍFifth ≈ 2 . (2.57)
λM 4
µ2 MPl
2
Note, if we identify λM 4 with β the result is the same as in the chameleon case.
This relations will be used for the spherical collapse model in Section 4.2.

2.3 The Hu-Sawicky f (R) model

As we have seen in section 1.2.6, f (R)-models can be treated as scalar-tensor theories, if we

redefine the metric the following way:
1
g̃µν = e2βφ/MPl gµν with β = √ . (2.58)
6
28 2 Review of the Chameleon and Symmetron Models

Under this transformation to the Einstein frame metric gµν , a scalar field appears with

fR = e−2βφ/MPl (2.59)

and a potential

2 fR R − f
V (φ) = MPl . (2.60)
2fR2

as stated in Eq. (1.62). This means, certain f (R)-models have a resulting potential V (φ),
which fulfills the condition Eq. (2.10) and fall, therefore, in the category of chameleon mod-
els.

One of these is the previously introduced model by Hu & Sawicki (2007), in which

c1 (R/m2 )n
f (R) = R − m2 (2.61)
1 + c2 (R/m2 )n

where c1 , c2 , m and n are positive constants. Furthermore, m is given by

H02
m2 = MPl
2
ρ̄m0 /3 = (2.62)
Ωm0

with ρ̄m0 being the average matter density today.

Taking the high curvature limit, that is expanding Eq. (2.61) around m2 /R → 0 results in
!n
c1 c1 m2
f (R) ≈ R − m2 + 2 m2 . (2.63)
c2 c2 R
Notice how the constant second term plays the role of a cosmological constant and in the
case c1 /c22 → 0 and c1 /c2 m2 → 2Λ we recover the standard ΛCDM case.

As we can see the model has three free parameters: c1 , c2 and n. In order to obtain
background ΛCDM evolution we set
c1 2
m = 2Λ = 16πGρΛ . (2.64)
c2

Furthermore, it is convenient to introduce the background curvature today fR0 ≡ df / dR a=a0

and write Eq. (2.63) as

fR0 − 1 R0n+1
f (R) = R − 16πGρΛ − . (2.65)
n Rn
The two remaining parameters n and fR0 specify the evolution of the Compton wavelength
of the scalar field and the deviation from ΛCDM respectively. The constraints get weaker
with increasing n because the screening mechanism works more efficiently (Ferraro et al.,
2011) and for fR0 → 1 ΛCDM is recovered.
2.3 The Hu-Sawicky f (R) model 29

Using now (2.59) and (2.65) we obtain in leading order

2βφ Rn+1
= (fR0 − 1) 0n+1 . (2.66)
MPl R

Therefore, the potential (2.60) can be written as

n/(n+1)
4βφ n+1 2 2βφ
  
V (φ) = 1 + ρΛ + MPl (fR0 − 1)R0 . (2.67)
MPl 2n MPl (1 − fR0 )

According to Brax et al. (2008) the condition for an f (R) model to have chameleon like
behavior is

V ′ (φ) < 0, V ′′ (φ) > 0, V ′′′ (φ) < 0 (2.68)

which is fulfilled for the potential given in the region

!n+1
1 − fR0 2
R0 MPl
0 < φ < MPl (2.69)
2β 4ρΛ

The curvature today can be found using the Einstein equations Eq. (1.5) with a FLRW
metric and is given by
−2
X
R0 = MPl ρi (1 − 3wi )
i
= 3H02 (Ωm,0 + 4ΩΛ,0 ) (2.70)

where in the first line the sum is over all matter species and in the second line radiation is
ignored as it has a neglectable energy content today. This transforms Eq. (2.69) into
!n+1
1 − fR0 Ωm,0
0 < φ < MPl +1 . (2.71)
2β 4ΩΛ,0

This means, the Hu-Sawicky model has a chameleon mechanism if the free parameter fR0 is
between 0 and 1. This can be also shown by analyzing the effective potential used in Section
2.1. In first order this reads
βφ
Vef f = V (φ) + ρ
MPl
′
and the minimum is given by Vef = 0 and can be written as
f φ=φmin
" #n+1
MPl |1 − fR0 | MPl2 R
0
φmin =
2β ρ + 4ρΛ
" #n+1
MPl |1 − fR0 | Ωm,0 + 4ΩΛ,0
= (2.72)
2β ∆ρ + 4ΩΛ,0
30 2 Review of the Chameleon and Symmetron Models

VeffHΦL

Φ Φ
Figure 2.3: The chameleon effect in the Hu-Sawicky model. The left panel shows a lower
density region (∆ρ = 170) and the right panel a higher density region (∆ρ = 500) where the
minima is closer to φ = 0. The solid black, dashed gray and dotted gray lines represent the
values |fR0 − 1| = (10−6 , 10−5 , 10−4 ), respectively. This behavior is the same as observed in
Fig. 2.1.

with ∆ρ = ρ/ρc,0 . It is to note that φmin approaches zero with a growing ρ which causes the
screening of the fifth force as seen before.
The screening effect is illustrated in Fig. 2.3 for three different values of fR0 . While for the
low density region (∆ρ = 1) shown in the left panel no screening is active, on the right panel
only the 1 − fR0 = 10−4 model is not fully screened. Therefore, not only the additional force
is weaker for values of fR0 closer to one – which can be noted through the slope of the curve,
that is not directly proportional to the fifth force but related to it1 – but also the screening
kicks in for a lower densities.

2.4 Gravitational redshift in the symmetron and chameleon models

Already Eq. (2.22) in the chameleon case and Eq. 2.33) in the symmetron case show that
photons with an equation of state wγ = 1/3 and therefore traceless energy-momentum tensor
do not “feel” the influence of the fifth force. Another way of seeing this is by transforming the
geodesics of a photon with a conformal transformation as introduced in Sec. 1.2.3. Photons
travel on null-geodesics (Hartle, 2003, Sec. 5.5), meaning if uµ is the 4-velocity of a photon
then the equation

gµν uµ uν = 0 (2.73)

holds. In terms of the conformally transformed metric, this equation becomes

g̃µν uµ uν = 0 , (2.74)

1 ~ as shown in Sec. 2.1

Since the fifth force is given by FFifth ∝ ∇φ,
2.4 Gravitational redshift in the symmetron and chameleon models 31

which is exactly the same equation as above. Physically speaking: The lightcone is preserved
under a conformal transformation. This result is not too suprising as angles and causality
are per definition preserved.
Consequently, gravitational redshift is only influenced by the Newtonian potential and not by
Veff (φ). Nevertheless, the gravitational redshift profile of clusters of galaxies can be different
in these screened modified gravities as the underlying density distribution is affected by the
additional force and therefore indirectly also the gravitational redshift profile.
3

The N -body Code and Simulation Parameters

In this chapter a short introduction in the tool of choice for this work is given. It is split
into three parts: In the first part, a short general introduction to cosmological N -body
simulations and their key concepts is given. The description of the simulation code used for
this work is presented in the second part. Finally, in the last part, the runtime parameters
for the simulation runs are provided and the halo-finding procedure is explained.

3.1 Introduction to N -body simulations

Steadily increasing computational power and the highly non-linear nature of the problem
have made N -body simulations a very popular tool for studying large scale structure forma-
tion in the universe.

In the last decade a number of simulations have been performed and some of them included
an impressive number of particles (see e.g. Kuhlen et al. (2012) for an overview). Different
algorithms have been developed to optimize this computationally expensive task.

In the following section a brief overview over existing algorithms is given with focus on the
code used in this project.

3.1.1 Concepts

N -body simulations can be split in two categories: Collisional and collisionless N -body
codes. In order to determine which algorithm is more appropriate to model a physical
system, a range of interactions can be defined. According to Hockney & Eastwood (1988),
for gravitational systems this is given by

Gm2
λi = 1 2
. (3.1)
2 mv
3.1 Introduction to N -body simulations 33

If the average of λi is small compared to the mean inter-particle spacing, a computationally

much less expensive collisionless N -body code can be used instead of the collisional algorithm.
Fortunately the galaxy density in the universe is very low which allows us to use collisionless
N -body codes for cosmological applications.

A collisionless system is described by the collisionless Boltzmann equation (CBE)

df ∂f ∂f ∂f
= + · ẋ + · ẍ = 0, (3.2)
dt ∂t ∂x ∂ ẋ

where f (x,ẋ,t) is the probability density function giving the probability that inside the
3 3
six-dimensional differential volume d x d ẋ at time t dN particles are located.
3
f (x,ẋ,t)d x d3 ẋ = dN (3.3)

For gravitational systems Eq. (3.2) can be simplified using the equation ẍ = −∇Φ:

∂f ∂f ∂f ∂Φ
+ · ẋ − · = 0. (3.4)
∂t ∂x ∂ ẋ ∂ x

The resulting expression is a non linear partial differential equation in seven dimensions.
This means it is not directly numerically solvable if it is not in or close to equilibrium state
( ∂∂ ft = 0). Consequently a different method is needed.

N -body simulations are based on the idea to model only N different points in phase space and
to obtain the observables afterwards as an ensemble average in a Monte-Carlo like fashion
via
1
Z
3
hOi (t) = d x d3 ẋ O(x,ẋ,t)f (x,ẋ,t) (3.5)
N
N
1 X
≈ O(xi ,ẋi ,t)f (xi ,ẋi ,t) , (3.6)
N i=1

here xi and ẋi are the position and the speed of the ith particle respectively.
df
Because of dt = 0 as stated in the CBE Eq. (3.2), Eq. (3.6) can be rewritten as

N
1 X
hOi (t) ≈ O(xi ,ẋi ,t)Wi , (3.7)
N i=1

where Wi = f (xi ,ẋi ,t)|t=t0 is the weight of the particles representing the distribution function
at the initial time t0 .

The actual initial conditions (xi ,ẋi )|t=t0 are chosen by the user. Thus, a lot of particles may
be placed in an area of particular interest to achieve a higher resolution locally (Bertschinger,
2011).
34 3 The N -body Code and Simulation Parameters

3.1.2 Gravity calculation

To obtain the wanted particle trajectories, for each particle i the equations
v̇i = −∇Φ(xi ) (3.8)
ẋi = vi (3.9)
∇2 Φ = 4πGρ (3.10)
have to be solved (Teyssier, 2002). For cosmological applications this is done in comoving
coordinates, meaning that xi′ and v′i are taken with respect to the expanding background
spacetime:
xi
xi′ = , (3.11)
a(t)
where a(t) denotes the cosmological expansion factor.
Expressing v̇i in the comoving coordinates and using this in Eq. (3.8) results in:
1 ä
v̇′i + 2v′i H = − ∇Φ − xi′ , (3.12)
a a
where H is the Hubble constant introduced Section 1.1.3.
By defining the spatial derivative with respect to the comoving coordinates ∇′ ≡ a∇ and
1
Φ′ ≡ aΦ + äa2 xi′2 , (3.13)
2
Poisson’s equation (3.10) can also be expressed in the comoving coordinate set
2
∇′ Φ′ = 4πGρa3 + 3äa2 . (3.14)
This equation can be further simplified using the result from the second Friedmann equation
(1.28) and defining the comoving density ρ′ ≡ ρa3 which leads (Hockney & Eastwood, 1988)
to the set of equations:
1
v̇′i + 2Hv′i = − 3 ∇′ Φ′ (3.15)
a
ẋi′ = v′i (3.16)
2
∇′ Φ′ = 4πG ρ′ − ρ0


(3.17)
These equations can be derived in an alternative way, taking the Newtonian gauge metric1
given by
ds2 = −(1 + 2Φ) dt2 + a(t)2 (1 − 2Φ)δij dxi dxj (3.18)
as a basis. Then, the geodesic equation for particle i becomes
1
ẍi + 2H ẋi + 2 ∇Φ = 0 (3.19)
a
which is identical to Eq. (3.15).

1 The Newtonian gauge has strictly speaking two scalar degrees of freedom often denoted as Φ and Ψ .
However, without anisotropic stress the Einstein equations give Φ = −Ψ (Dodelson, 2003).
3.1 Introduction to N -body simulations 35

Obtaining the density

In order to solve these equations it is crucial to obtain the particle density at a given position
ρ(x). Simply utilizing the Monte Carlo estimate (3.7) will just lead to a sum of delta
functions. A commonly used solution is to widen the delta peaks which leads to the density
estimate (Silverman, 1986)
X mi 
|x − xi |

ρ≈ K̃ , (3.20)
i
ε3 ε

with the normalized function K̃(x) being the kernel function, mi the mass of the particle i
and ε the softening length that controls how fast the kernel function falls off.
Applying (3.17) to (3.20) results in an estimate for the potential Φ (Dehnen & Read, 2011):
X Gmi 
|x − xi |

Φ(x) ≈ − K̂ (3.21)
i
ε ε
The choice of the kernel function and the softening length is not trivial. E.g. the Plummer
softening is used widely (Dehnen & Read, 2011) where
 −5/2  −1/2
K̃(x) = 3/(4π) x2 + 1 and K̂(x) = x2 + 1 . (3.22)

Therefore, the estimate for the potential becomes

X mi
Φ(x) ≈ −G p . (3.23)
i |x − xi |2 + ε2

The advantages of this (as in most other kernel functions used) is the non-singularity at x = 0
and the reduction of numerical artifacts due to close encounters which are not described by
collisionless but instead by collisional N -body codes (Dehnen & Read, 2011).
Another choice often used in grid-based force calculation methods (see below) is the cloud
in cell (CIC) scheme (Birdsall & Fuss, 1969). Here the kernel function in equation (3.20) is
given by a linear interpolation:
(
1−x if |x| < 1
K̃(x) = (3.24)
0 otherwise

Force calculation methods

In theory it is possible to carry out the sum of Eq. (3.23) (i.e. its derivative) in order to
obtain the force. In practice this direct summation is too slow and a number of more rapid
algorithms were developed.
The simplest grid-based method is the particle-mesh (PM) algorithm (Hockney & East-
wood, 1988; Klypin & Shandarin, 1983). The density is calculated at every node of a regular
36 3 The N -body Code and Simulation Parameters

grid. This allows to solve the equations (3.17) and (3.15) for the potential and the accelera-
tion respectively at every grid point using the computationally cheap fast Fourier transform
(FFT) (Cooley & Tukey, 1965). Afterwards the acceleration can be interpolated back to
the particles’ positions and their coordinates updated accordingly. The automatic periodic
boundary conditions that come with the FFT and the high performance (O(N )+O(N log N )
(Bertschinger, 1998)) made this algorithm popular for cosmological usage. The main dis-
advantage of this algorithm is the poorly modeled force for particles that are closer than
several grid spacings. This means in order to obtain good results the global grid resolution
has to be increased which is computationally very expensive.
A logical extension to the PM method is therefore to calculate the forces for particles closer
than two or three grid spacings via direct summation. This idea leads to the particle-
particle/particle-mesh (P3 M) algorithm (Efstathiou & Eastwood, 1981; Hockney et al.,
1974). The relative high precision achieved through the direct summation stands against
the high computational cost if dense regions occur.
One possible solution to this is to use multiple grids: In the coarse grid Poisson’s equation is
solved with periodic boundary conditions whereas in high-density regions a finer grid will be
added with fixed (Dirichlet) boundary conditions. This adaptive particle-particle/par-
ticle-mesh (AP3 ) method (Couchman, 1991) still uses direct summation for particles closer
than several grid spacings but lowers the computational cost by using locally the PM method
in the finer grids.
Another approach is to stick to direct summation but to group close particles together and
treat them as one particle if necessary. This means for a particle further away from a given
group of particles, it is then not necessary to interact with each particle belonging to the
group but instead with the group as a whole. To achieve this the TREE algorithm (Barnes
& Hut, 1986) starts also with a regular grid. Note that this grid is not used to solve the
Fourier transformed Poisson’s equation as the methods described before but that each cell
represents a group of particles. If a group contains more particles than some threshold, the
cell is split with the newly created cells being the children of the original cell. The result is a
tree structure of cells with each “leaf” (a cell that has no children cells) having roughly the
same number of particles. The further apart a particle is to a given point, the higher up the
tree is the node with which the particle can interact. This leaves a direct interaction only for
very few particles. A big advantage of this approach is the speed (O(N ) log N ) by keeping
the errors low (Hernquist, 1987). To achieve periodic boundary conditions important for
cosmological usage, various TREE codes (i.e. Bouchet & Hernquist (1988)) use a method
named Ewald (1921) summation.
A unification of the TREE and the AP3 M methods is found in the adaptive mesh re-
finement (AMR) algorithm (Jessop et al., 1994; Kravtsov et al., 1997). Instead of simply
adding finer grids onto the coarse grid at various regions as done in the AP3 M method, a
tree of subgrids is built. It is therefore possible to use very fine grids locally which makes
direct summation obsolete. This has two main advantages: First the force resolution is fully
spatially adaptive and not uniform as in the TREE or (A)P3 M methods. Knebe et al. (2000)
show that this approach preserves halo properties better. Secondly, leaving out the direct
particle-particle interaction simplifies the usage of a modified Lagrangian.
3.2 The simulation code 37

3.2 The simulation code

The simulation code used called ISIS (Llinares et al., 2013) is based on the RAMSES code
by Teyssier (2002). The AMR method (see section 3.1 for an introduction to the various
N -body methods) used allows to include directly modified Lagrangian as mentioned before.
In the following the structure of the original RAMSES code is summarized and the scalar field
equations are introduced.

3.2.1 RAMSES code structure

The RAMSES code (Teyssier, 2002) is a heavily parallelized multi-grid algorithms and performs
for every timestep the following sub-steps:

1. Compute the density ρ on the mesh using the cloud in cell scheme described above.

2. Solve the Poisson’s equations in order to get the potential Φ. In order to achieve
this, the Gauss Seidel method with successive overrelaxation and chessboard ordering
is employed. Additionally, Dirichlet boundary conditions are used. This step will be
explained in more detail below.

3. Compute acceleration on the mesh.

4. Compute the particles’ acceleration using an inverse CIC.

5. Update the velocity of each particle using the acceleration.

6. Update the position of each particle using the velocity.

In order to obtain the potential for the finer grids in step 2, the boundary conditions have
to be fixed first. Therefore, the potential already obtained for the grid next in cell size is
linearly interpolated to a temporary buffer region used as the boundary values. The next
step is to solve the Poisson’s equation with these boundary conditions. To achieve this the
Gauss Seidel method with successive overrelazation and cheeseboard ordering (Press et al.,
1992) is used. The concept of the Gauss Seidel method with successive overrelaxation is
explained in the appendix in Sec. A.2.

3.2.2 The ISIS code

The simulation code used in this work is ISIS (Llinares et al., 2013), which is a modification
of RAMSES (Teyssier, 2002), that includes the development of a scalar field φ. Hence, the
equation (3.15) is altered to

1 ′ ′
v̇′i + 2Hv′i = − ∇ Φ − F(t,φ,...) (3.25)
a3
38 3 The N -body Code and Simulation Parameters

where F describes the supplementary influence of the scalar field on the particles’ trajectories
and, consequently, is a function of time, φ and its derivatives. In addition, the equation
describing the development of the scalar field itself is solved

∇2 φ = S(t, ρ, φ) (3.26)

with the source function of the scalar field on the right hand side.
For scalar-tensor theories as defined in Sec. 2.2, i.e. introduced via g̃µν = A2 (φ)gµν , the
functions F and S are given by
∂ 1 ~
F=2 (log A)x + 2 ∇(log A) (3.27)
∂t a
S = V,φ − A,φ ρ . (3.28)

These equations where derived in a general form in Section 2.2 (see Eqs. (2.37) and (2.40)).
The Newtonian gauge metric (3.18) can then be used as in Sec. 3.1.2 to bring them in the
above form. It is to note, that the usual approximation (Oyaizu, 2008), that is to work in
the quasi-static limit, i.e. to drop the terms including time-derivatives, has been applied.
In a similar form these equations where also implemented in the ECOSMOG code (Li & Hu,
2011), which is also based on the RAMSES code, an N -body code by Zhao et al. (2011) based
on MLAPM (Knebe et al., 2001) and in a code by Oyaizu (2008).

Symmetron implementation

The symmetron implementation in ISIS uses the same functions V (φ) and A(φ) introduced
in Sec. 2.2. Equation (3.25) turns for this model into
1 1
ẍ + 2H ẋ + 2
∇Φ + φ∇φ = 0 . (3.29)
a (M a)2
The same result as could be obtained using the general modified geodesics (2.40) using the
Newtonian gauge metric.
Similar to Winther et al. (2011), instead of the original parameters (µ, M , λ), more physical
parameters where introduced which are all linked to the properties of the scalar field in
vacuum. Firstly, the range of the field
1
L= √ , (3.30)
2µ
secondly, the expansion factor for which the symmetry is broken in the background level
Ωm,0 ρc,0
a3ssb = (3.31)
µ2 M 2
and last a dimensionless coupling constant
µMPl
β= . (3.32)
M 2λ
3.3 Simulation parameters 39

In addition, the scalar field itself is normalized to its vacuum expectation value:

φa3ssb
χ ≡ φ/φ0 = (3.33)
6H02 MPl L2 βΩm,0

With these redefinitions Eq. (3.29) can be rewritten as

1 6H02 L2 β 2
ẍ + 2H ẋ + ∇Φ + χ∇χ = 0 . (3.34)
a2 a2 a3ssb

The equation of motion of the scalar field (3.26) is in this case

" 3 #
2 a2 assb 3
∇ χ= ηχ − χ + χ (3.35)
2L2 a

with η ≡ ρm a3 /(Ωm,0 ρc,0 ). This means η describes the local density normalized to the
average matter density.

f (R) implementation

The Hu-Sawicky model was discussed in Sec. 2.3, where the parameters used in the sim-
ulation, namely n and fR0 , are also introduced. Basically, the effective potential and the
conformal transformation described there can be used directly to solve the modified geodesics
and the scalar field evolution numerically, as done for the symmetron model. The equation
tho solve would be
" 1 #
1 2 ΩΛ fR0 − 1 ΩΛ
   
n+1
∇ fR = Ωm H02 1+4 − a −3
+4 −3
− ηa . (3.36)
a2 Ωm fR − 1 Ωm

Practically, solving this equation numerically is not possible due to the singularity at fR = 0.
Therefore, a field redefinition

fR ∝ e u (3.37)

is a better choice. This leads to the code formulas described in detail in Llinares et al. (2013),
which are similar but not identical to the implementations by Li et al. (2012) and Oyaizu
(2008).

3.3 Simulation parameters

All simulation runs contained 5123 particles with a mass of 9.26138 × 109 M⊙ /h which
represent the total matter distribution. This corresponds to a matter content of 26.7%
(Ωb,0 = 0.045, Ωcdm,0 = 0.222 and ΩΛ0 = 0.733). The simulation box has a side-length of
256 h−1 Mpc and periodic boundary conditions. The data sets used are the snapshots taken
at z = 0.
40 3 The N -body Code and Simulation Parameters

Name zssb β L in Mpc h−1

Name |fR0 − 1| n
symm_A 1 1 1
fofr4 10−4 1
symm_B 2 1 1
fofr5 10−5 1
symm_C 1 2 1
fofr6 10−6 1
symm_D 3 1 1
(a) f (R) runs
(b) symmetron runs

Table 3.1: Parameters of the different simulation runs. Left: The f (R) parameters are n and
the value of df / dR today. Right: The symmetron parameters are the redshift of the symme-
try breaking zssb = 1/assb − 1, the coupling β and the length scale of the fifth force L.

All simulation runs were using the same initial conditions, which were generated by the
MPGRAFIC (Prunet et al., 2008) program. The parameters for the f (R) and the symmetron
models are summarized in table 3.1.

In the case of the f (R)-gravity the parameter n was fixed to 1 while fR0 took values from
10−6 , which resulted in hardly any deviation from ΛCDM, to 10−4 , which is on the border of
violating cluster abundance constraints (Ferraro et al., 2011). The runs were named fofr4,
fofr5 and fofr6 for the values 10−4 , 10−5 and 10−6 , respectively.

For the symmetron model mainly the time of the symmetry breaking was varied while leaving
the others constant: assb was set to the values 1/2, 1/3 and 1/4 for the simulation runs
symm_A, symm_B and symm_D, respectively. For each of these models β and L/(Mpc/h)
were chosen to be unity. The run named symm_C stands out as it has the same symmetry
breaking time as symm_A but an altered β value of 2.

3.4 Post-processing: Identifying the halos

The output of an N -body simulation is usually limited to the position and a few other
attributes of the single particles. In order to determine the position and other attributes of
gravitationally bound objects – halos or clusters – the N -body simulation output needs to
be post-processed with a so-called halo finder.

To achieve this mainly two different approaches are used: In the spherical overdensity (SO)
algorithms the particle density peaks are found and then particles around them are being
defined as part of the halo until the particle density falls below a certain threshold. This top-
down design is implemented in AHF (Knollmann & Knebe, 2009), BDM (Klypin et al., 1999),
SKID (Stadel, 2001) and others. Alternatively, nearby particles can be grouped together
either using a linking length in the friends-of-friends (FOF) algorithms or by “hopping”
from a particle to the densest neighbor (HOP). These down-top designs are used in the FOF
case e.g. by SUBFIND (Springel et al., 2001), FOF (Davis et al., 1985) or pFOF (Habib et
al., 2009) and AdaptaHOP (Tweed et al., 2009) in the HOP case. Both of these approaches
can be extended to phase-space, as it is done e.g. in 6DFOF (Diemand et al., 2006). For
3.4 Post-processing: Identifying the halos 41

106 106
fofr4 symmA
fofr5 symmB
105 fofr6 105 symmC
ΛCDM symmD
ΛCDM
104 104
N(>M)

N(>M)
103 103

102 102

101 101

100 100
1011 1012 1013 1014 1015 1016 1011 1012 1013 1014 1015 1016
Halo mass in M⊙/h Halo mass in M⊙/h

Figure 3.1: Halo mass functions of the data sets. Left panel: The ΛCDM and the f (R) data
sets. The total number of halos is 115 858, 120 203, 124 404 and 133 241 for the ΛCDM, fofr4,
fofr5 and fofr6 data sets, respectively. Right panel: The ΛCDM and the symmetron data sets.
The total number of halos for the symmetron models is 141 131, 152 307, 193 814 and 141 100
in alphabetic order.

a more detailed overview of current halo finding algorithms as well as their comparison see
Knebe et al. (2011).

The halo finder chosen for this work is Rockstar (Behroozi et al., 2013), a publicly available
FOF code. Through its ability to work in six dimensions (even seven if simulation snapshots
for multiple timesteps are available) and adaptive linking length, Rockstar shows excellent
results in halo finding as well as in calculating the halo properties (Knebe et al., 2011).
Additionally, Rockstar is a highly efficient code with only 5 − 10 CPU hours and 60 GB of
memory required per billion particles analyzed (at z = 0).

The algorithm works roughly as follows: First the simulation box is divided into 3D FOF
groups. These groups are then spread over the available processors. In each group the
nearest neighbors are found using the phase-space metric by Gottloeber (1998). The distance
between two particles with positions x1,2 and velocities v1,2 is given by:
s
|x1 − x2 |2 |v1 − v2 |2
d1,2 = + (3.38)
σx2 σv2

Consequently, the definition of distance within each subgroup depends on the velocity dis-
persion σv and position dispersion σx . The linking length is then chosen so that 70% of
all particles in a subgroup are linked together. This results in a high precision of finding
substructures. Afterwards the last two steps (namely defining the metric and linking the
particles together) are repeated over each new subgroup. Finally, when a minimum number
of particles are left in a subgroup, this group is defined as seed of a halo and each particle is
defined as part of the closest halo. Eventually, gravitationally unbound particles are removed
and halo properties calculated. A more detailed explanation about the Rockstar halo finder
can be found in Behroozi et al. (2013).
42 3 The N -body Code and Simulation Parameters

In our case Rockstar was run with the virial radius being the radius R200c , meaning the
boundary of the halo is the threshold of 200 times the critical density today. This number
was initially chosen to be able to better compare the results with other papers as often
the same criteria is picked. There were, however, tests made with the criteria from Bryan
& Norman (1998), which corresponds roughly to 360 times the background density today
(Behroozi et al., 2013), that lead to the same results.
The proper treatment of halo finding in the self-screening models considered is, however, not
to use a constant like 200, but instead adapt this number depending on the halo mass (see
section 4.2). This treatment needs a modification of the halo finding code which has not
yet been done (see for a summary of the current state in halo finding Knebe et al. (2013))
and goes beyond the scope of this work. Instead, the results obtained with Rv = R200c from
ΛCDM were also compared with Rv = R160c for the most extreme models without finding a
big difference.
As Rockstar returns a halo-subhalo tree structure but the gravitational well of a halo is due
to the host halo, the next step taken was to flatten the hierarchy, i.e. to attach each particle
to its “mother halo”. This means all particles inside the radius R200c or are member of a
halo which has its center inside this radius are defined to be part of a halo. For each of this
newly organized halos the properties like mass are updated. Special attention was in this
respect given to the calculation of the energies as these were used to find out whether a halo
is in a virialized state or not. This subject is covered in the next section.
Our samples contain each around 120,000 clusters. For the ΛCDM run around 73,000 of
them contain more than 100 particles and circa 9200 more than 1000 particles. In Fig. 3.1
the halo mass function, i.e. the number of halos bigger than a given mass, is shown. An
in-depth interpretation on the change of the halo mass functions in the symmetron and
chameleon model was done by Brax et al. (2012, 2013) and is not subject of this work.
4

Spherical Collapse and Virialization

In order to study halo properties it is crucial not to mix clusters that are dynamically relaxed
and those which have not reached such an equilibrium state yet as the two groups have a
different distribution of halo properties (Shaw et al., 2006). Using the whole cluster sample
and not separating these two groups may therefore lead to a skewed mean and higher noise.
Especially when comparing the effect of different gravitational theories below Mpc scale –
as done in this work – a combination of the two groups may lead to a false interpretation.
This misinterpretation can for example occur since it might not be clear if the change is
happening on an actual halo property or if the number of unrelaxed objects change.
Measuring the level of relaxation – or virialization – can be done using the virial theorem,
which is derived in the first section of this chapter. This level of relaxation is also important
in defining what a halo actually is or where its boundaries are. This question is pursuit in
the second section using the tool of spherical collapse which is extended to screened modified
gravity models. The last section is dedicated to the implementation of these results, i.e. how
the energy quantities are to be calculated using discrete particles like from the output of an
N -body simulation.

4.1 Derivation of the virialization equation

The derivation of the virialization equation presented in this section follows roughly Chan-
drasekhar (1981). Starting point is the collisionless Boltzmann equation already mentioned
in Sec. 3.1:
∂f ∂f ∂f
+ ẋ + ẍ = 0 . (4.1)
∂t ∂x ∂ ẋ
By multiplying this equation with ẋ and integrating over velocity space one obtains for the
first term
∂ ∂(uj n)
Z
d3 vvj f = , (4.2)
∂t ∂t
44 4 Spherical Collapse and Virialization

where n is the particle density. This means, the quantity u(x,t) can be interpreted as the
mean speed of the particles inside a volume element dV centered at x at time t.
For the second term one obtains
∂ ∂
Z Z
d3 vvj vi = d3 v [(vj − uj ) + uj ] [(vi − ui ) + ui ] (4.3)
∂xi ∂xi
∂ 1
 
= Pji + uj ui n (4.4)
∂xi m
where the pressure tensor
Z
Pji ≡ d3 vm(vj − uj )(vi − ui )f (4.5)

has been introduced. In the following, we consider a diagonal pressure tensor with isotropic
pressure, Pij = δij p.
Applying the same procedure on the third term produces

∂ ∂
Z Z Z
3 3
d vvj (ai f ) = d v (vj ai f ) − d3 vai f δij = −nhaj i . (4.6)
∂vi ∂vi
If the force acting on the particles is independent of the particles’ velocities, as it is the case
for gravity, then simply haj i = aj .
Putting the three terms together results in

∂ 1
(nu) + ∇p + ∇ · (uu) − an = 0 . (4.7)
∂t m
Using the conservation of the number of particles

∂n
+ ∇ · (nu) = 0 (4.8)
∂t
this identity can be rewritten as

du
ρ = −∇p + ρa (4.9)
dt
where
d ∂
= +u·∇ (4.10)
dt ∂t
is the total time derivative. This equation is the equation of motion for a fluid with flow
velocity u, pressure p and density ρ.
If only gravity acts on the fluid this can be written in the familiar form

du
ρ = −∇p − ρ∇Φ (4.11)
dt
4.1 Derivation of the virialization equation 45

where Φ is the gravitational potential. For reasons that will be clear later on, the general
form with the acceleration a is kept instead.
Moving on by multiplying Eq. (4.9) by x and integrating over a volume V one obtains on
the left hand side
du d2 x
Z Z
3 3
d x ρx · = d x ρx ·
dt dt2
V V
( 2 )
d dx dx
Z   
3
= dxρ x· −
dt dt dt
V
d dx
Z    
3
= dxρ x· − u2 (4.12)
dt dt
V

and on the right hand side

Z Z
3 3
d x {ρx · a − x · ∇p} = W − d x {∇ · (xp) − p∇ · x} (4.13)
V V
Z Z
3
=W− dS · (xp) + 3 dxp (4.14)
S V

respectively.
In (4.13) the relation
Z
3
W = d x ρx · a (4.15)
V

was used. The implementation of the virialization criterion will play a big role for the results.
Therefore, the equality of W as given in Eq. (4.15) and the maybe more common form of
the gravitational potential energy of a massive body with volume V given by
G ρ(x′ )ρ(x)
ZZ
3 3
W̃ = − d x d x′ (4.16)
2 |x − x′ |
V

will be briefly shown here:

∂φ
Z
3
W =− d x ρ(x)xi
∂ xi
V
∂ ρ(x′ )
Z Z
3 3
=G d x ρ(x)xi d x′
∂ xi |x − x′ |
V V
xi (xi − x′i )
ZZ
3 3
= −G d x d x′ ρ(x)ρ(x′ )
|x − x′ |3
V
G ρ(x)ρ(x′ ) h i
ZZ i
3 3
=− d x d x′ x (x i − x′
i ) + x′i ′
(xi − x i ) (4.17)
2 |x − x′ |3
V
46 4 Spherical Collapse and Virialization

G ρ(x)ρ(x′ ) h
ZZ i
3 3 i
=− d x d x′ (x i − x′
i )(x − x ′i
)
2 |x − x′ |3
V
= W̃

where in the first line the relation a = −∇Φ was used and in Eq. (4.17) an identical integral,
with the variables of integration x and x′ swapped, was added.
Considering now a coordinate system moving with the center of mass of the object (u =
ρẋ = 0), the term in Eq. (4.12) often labeled (e.g. in fluid mechanics) as “kinetic energy”
R

(Chandrasekhar, 1981) disappears. Instead the otherwise named “internal energy” plays a
crucial role and is named “kinetic energy”. Inserting the definition of the pressure tensor
above and assuming isotropy, this term can be written as
3 1
Z Z
3 3
T ≡ dxp = d x ρ(v (x) − u)2 . (4.18)
2 2
V V

Instead of pressure, one speaks sometimes about the velocity dispersion σ 2 ≡ p and in the
case of isotropy σ 2 /3 = σx2 = σy2 = σz2 .
Going back to the left hand side of Eq. (4.11), the first term of Eq. (4.12) can be rewritten:
d dx 1 d2  2 
Z   Z
3 3
dxρ x· = dxρ x (4.19)
dt dt 2 dt2
V V
1 d2 I
= , (4.20)
2 dt2
where the moment of inertia
Z
3
I≡ d x ρx2 (4.21)
V

has been introduced.

The remaining term of Eq. (4.14) can be identified with the surface pressure term
Z
ES = dS · (xp) (4.22)
S

which is zero for a halo positioned in empty space (Shaw et al., 2006).
Collecting all the terms, that is Eqs. (4.15), (4.18), (4.20) and (4.22), gives the well known
virial equation
1 d2 I
= 2T + W − ES . (4.23)
2 dt2
A fully virialized object is defined if the left hand side is zero, leaving

2T − ES = −W . (4.24)

This relationship is know as the virial theorem.

4.2 Spherical collapse in standard models 47

4.2 Spherical collapse in standard models

Now, it is clear what virialized objects are, the next obvious question is how they form. In
other words: What initial overdensity is necessary for structure formation to happen? And,
what density do these structures have today?
These questions will be addressed in this section through the well-established model of
spherical collapse: A spherical overdensity is placed in an expanding space in order to find
out how big this overdensity is required to be in order to have a total collapse today, i.e. the
radius of the object would be zero at a = 1. This model allows to find a rough estimate for
the density contrast between the virialized structure and its surrounding.
The spherical collapse model is analytically solvable for an Einstein-de Sitter background,
which will be discussed in Sec. 4.2.1. A numerical approach is necessary in order to find the
development in a more complex background evolution, which is done here for ΛCDM. The
code developed was implemented in Wolfram Mathematica®1 and is presented in Sec. 4.2.2.

4.2.1 Einstein-de Sitter universe

For a first approximation we consider the simplest possible model: A spherical overdensity is
placed in an otherwise empty Einstein-de Sitter universe (Ωk = 0, Ωm = 1). This overdensity
will expand first – due to the expansion of space – and then collapse under the influence of
its own gravity, given the overdensity was big enough. In this simplified model, the sphere
will then collapse until the radius reaches zero, resulting in an infinity density of the sphere.
For real systems, however, the collapse will cease at virialization (see Eq. (4.19)). This major
discrepancy is due to the fact that in real systems matter is not distributed homogeneously
in spherical shells and also not all the movement is in radial direction2 . To overcome this
problem, the collapse has to be halted manually when 2T = −W is fulfilled.
In order to study the evolution of the sphere, we consider a “sub-universe” with higher
density than the surrounding universe Ωms > Ω
m = 1. In this case, the sub-universe is
positively curved (Ωk < 0) and develops according to the Friedmann equations introduced
s

in Sec. 1.1.3 in the following way:

s
as0 Ωm,0
as (ϕ) = s (1 − cos ϕ) (4.25)
2 Ωm,0 −1
s
Ωm,0  −3/2
ts (ϕ) = s
Ωm,0 −1 (ϕ − sin ϕ) . (4.26)
2H0

The superscript s denotes the quantities related to the sub-universe, e.g. the radius of the
overdensity as . The parameter ϕ goes from 0 to 2π and, hence, the radius has a maximum of

1 www.wolfram.com/mathematica
2 Engineer et al. (2000) introduced additional shear and angular momenta terms in the spherical collapse
equations which lead to natural ceasing of the collapse.
48 4 Spherical Collapse and Virialization

as (π) = asmax at tmax . At this point all the energy of the system is potential energy (Wmax ).
So in general we can write

T = Wmax − W (4.27)
3 1 1
 
2
= GM − . (4.28)
5 as asmax
At time of virialization – denoted by tvir and ϕvir – the condition 2T = −W is fulfilled
leaving us with
1 s
asvir = a . (4.29)
2 max
This relation is fulfilled twice in Eq. (4.25), at ϕ = π/2 and ϕ = 3/2π. A solution after the
begin of collapse (at ϕta = π) is wanted and thus ϕvir = 3/2π.
This solution can now be used to find the density relation between the density of the sphere
at time of virialization and the background density:
s (as )−3
ρs Ωm,0 9
= = (2 + 3π)2 ≈ 147 (4.30)
ρ ϕvir (3/2H0 ts )2 ϕvir 8

Hereby, the solution of the scale factor a(t) = a0 (3/2H0 t)2/3 derived in Sec. 1.1.3 has been
used.
Another number often found in literature is 178. This results can be found if the ratio
ρs (ϕvir )/ρs (2π) is taken instead:

ρs (ϕvir )
∆c = = 18π 2 ≈ 178 (4.31)
ρ(2π)

Both results mean the same: Structures with a density 100 − 200 times the background
density fulfill the virial theorem and are therefore gravitationally bound objects.

4.2.2 ΛCDM model

For the case including a cosmological constant Λ, the solution cannot be found analytically
anymore. Therefore, the considerations in this section are geared towards finding a numerical
solution. The reason for this section to be included in full length, is that the equations derived
here can be used for the modified gravity cases with only slight modifications.
The equation governing a collapse of a sphere with radius R in an expanding space is (Gunn
& Gott, 1972)

R̈ 4 X
=− G (ρi (1 + 3wi )) , (4.32)
R 3 i

which is basically the second Friedmann equation Eq. (1.28) without curvature.
4.2 Spherical collapse in standard models 49

Assuming only dark energy in a form of a cosmological constant and ordinary matter, the
equation can be rewritten as

R̈ 4
= − G (−2ρΛ + ρ)
R 3 !
3
Ri

= −2H02 −2ΩΛ,0 + Ωm,0 (1 + δi ) . (4.33)
Rai

In the last line the matter density in the sphere ρ has been expressed in terms of the initial
overdensity δi and the initial radius of the sphere Ri at time a(t) = ai via
3 3
Ri Ri
 
ρm = (1 + δi )ρm,0 a−3
i = (1 + δi )ρc,0 Ωm,0 a−3
i . (4.34)
R R

Using the dimensionless variable R̃ ≡ Rai /Ri with the initial radius Ri at scale factor ai
Eq. (4.33) becomes
¨
R̃ H2  
= − 0 −2ΩΛ,0 + R̃−3 γΩm,0 (1 + δi ) . (4.35)
R̃ 2

It is convenient for numerical calculations to use x ≡ ln(a) as time variable. Then, the
equation becomes

d2 R̃ ä dR̃ H0 
  
H2 + − H2 = − R̃ −2Ωφ,0 + R̃−3 Ωm,0 (1 + δi ) , (4.36)
dx2 a dx 2

which is the differential equation as implemented in the Mathematica® code.

The boundary conditions to solve this equations are:

1. R ai
= Ri . Therefore, R̃ = ai
ai

dR̃ R̃˙
2. Ṙ ai
=H ai
Ri . Consequently, dx ai = H = ai .
ai

Note that the solution is not dependent on the initial radius Ri , which means that the
evolution of the collapse does not depend on the size of the halo. The value of the initial
overdensity is chosen in such a way that the collapse would happen today, at a(t) = 1. It is
found via simple iteration through various values of δi .
As in the case of the simple model before, the sphere would not fully collapse, but stop the
collapse when the sphere is virialized, i.e. fulfills the virial equation (4.24). A common way
of finding when the sphere is virialized is by integrating the Tolman-Bondi equation (Bondi,
1947; Tolman, 1934). This was done for example by Lahav et al. (1991). The approach
chosen here is focusing on the acceleration, so that the virialization criterion does not have
to be altered for other forms of dark energy, that is when an additional force is acting on
the sphere besides gravity.
50 4 Spherical Collapse and Virialization

The kinetic energy is simply given by

ρ 3
Z
T = d3 x ẋ2 = M Ṙ2 . (4.37)
2 10

Expressed in the numerical variables defined above this becomes

!2
3 ai dR̃
T = H . (4.38)
10 Ri dx

The total potential energy is given by Eq. (4.15) which becomes in this application

ZR
3
W = 4π drρrr̈ = M R̈ . (4.39)
5
0

Here the assumption is made that the shells of the sphere expand in a way so that there
is no shell crossing and as a result the density stays uniform. Assuming no dark energy
contribution, the Newtonian gravitational potential energy or WN = −3/5GM 2 /R can be
recovered using R̈ = −GM/R. With dark energy in form of a cosmological constant, however,
there is an additional contribution to the Newtonian gravitational potential energy given by
1
WΛ = ΛR2 , (4.40)
5
which was – as mentioned above – used by Lahav et al. (1991)1 .
In code units Eq. (4.39) becomes:
2 " #
3 Ri dR̃ ä dR̃
  
W = M + − H2 . (4.41)
5 ai dx2 ä dx

Finally, the density contrast is defined as in (4.31):

ρvir −3
∆c = = (1 + δi )R̃vir (4.42)
ρc,0 Ωm,0

with the density of the sphere at virialization ρvir and the critical density today ρc,0 .
The code was tested, by replicating the results from the Einstein-de Sitter model: For
an initial overdensity of δi ≈ 3.316 · 10−4 , ∆c = 177.659 was obtained. The numerical
uncertainties can explain the small discrepancy to the analytical value (4.31). Note that for
this and further runs we set ai = 10−4 and H0 = 71.9 kms−1 Mpc−1 ≈ 1.53 · 10−42 GeV−1 .

1 Strictly speaking the result from the integration of the Tolman-Bondi-equation differs by a factor 1/2.
However, this is compensated by the relation of the kinetic to the potential energy of W ∝ Rm with
T = n/2W used in this context. See Lahav et al. (1991) and Maor & Lahav (2005) for details.
4.3 Spherical collapse in screened modified gravity 51

In the ΛCDM case (Ωm,0 = 0.25, ΩΛ,0 = 0.75) the result is

δi ≈4.250 · 10−4 (4.43)

∆c ≈380.765 . (4.44)

Since the universe expands faster compared to the previous case, the initial overdensity
needed for a collapse is expected to be bigger. Also, the additional potential energy means
that the velocities have to be higher in order for the system to be in equilibrium. Therefore,
the shells have to fall further inwards to acquire the velocity and ∆c is larger.
In order to test our method, the virialization condition

2T + WN + WΛ = 0 (4.45)

used by Lahav et al. (1991) was also implemented giving the same results.

4.3 Spherical collapse in screened modified gravity

In this section the spherical collapse model is studied in the symmetron and Hu-Sawicky
f (R)-gravity case. This simplified model can help to understand the acting of the fifth
force inside clusters of galaxies without employing resource-intensive N -body simulations.
Naturally, the simplification is crude since important factors as the environment of the halo
or inhomogeneous density distributions are neglected. After all, this model can help testing
if the definition of a halo as used in ΛCDM can be directly transferred to screened modified
gravity models. Thus, the aim of this section is to analyze the density contrast value ∆c in
these models and to understand the process behind eventual deviations from the standard
model.
In order to do this the code introduced in the last section was modified. That is, the
Newtonian gravitational constant was replaced by an effective gravitational constant taking
screening effect into account. The virialization condition does not have to be modified as
the acceleration is a sum of the Newtonian acceleration and the additional acceleration due
to the scalar field. Therefore, the energy of the scalar field is automatically included.

In the case of screened modified gravity models, the gravitational constant in Eq. (4.32)
needs to be replaced by effective ones. Several effective constants can be introduced as
there might be a different force acting on distinct matter species. In the example here, the
assumption is made that the additional force only acts on ordinary matter. Consequently,
equation (4.33) turns into
3 !
R̈ Ri

= −2H02 −2ΩΛ,0 + γΩm,0 (1 + δi ) (4.46)
R Rai

with γ = Gef f /G. Note that, the value of this effective gravitational constant Gef f depends
on the environment and hence Gef f is not a constant at all. Furthermore, the assumption
52 4 Spherical Collapse and Virialization

has been made that there is an effective cosmological constant present as it is in most
chameleon-like models.

In code units, the equation of motion then reads

d2 R̃ ä dR̃ H0 
  
2
H + − H2 = − R̃ −2Ωφ,0 + γ R̃−3 Ωm,0 (1 + δi ) . (4.47)
dx2 a dx 2

As stated in the spherical solution sections of Chapter 2, objects can be in any state between
fully screened to unscreened. The difference of the effective gravitational constants of these
two extreme cases is equal to 3∆R/R. Therefore, we approximate γ to be of the form

3∆R
 
2
γ = 1 + 2β T with T ≡ min 1, . (4.48)
R

Here, the same variable transformation as in Sec. 3.2.2 has been performed for the symmetron
model in order to replace the model parameters (µ, M , λ) by (β, assb , L). Notice that the
same set of equations can be found in Brax et al. (2010b) where the spherical collapse in the
chameleon model has been analyzed with a slightly different definition of γ.

The thin shell factor was stated in Eq. (2.27), that is

∆R |φ∞ − φc |
= , (4.49)
R 6βMPl Φ

where φ∞ and φc are the values of the scalar field far away and at the center of the object,
respectively. The Newtonian potential of the sphere Φ in code units is simply given by

R2 ρ Gai Mc
Φ= 2 = . (4.50)
6MPl R̃Ri

This introduces a dependency on the mass of the object, i.e. ∆c and the virial radius will
depend on the mass of the cluster.

In scalar-tensor theories the virialization condition has to take the energy of the scalar field
into account as well. Very elegant solutions are discussed in Maor & Lahav (2005) and
Mota & Bruck (2004). Instead, in this implementation the particles’ acceleration is used
(see Eq. (4.39)) and the scalar field is considered in the equation of motion, the virialization
condition used for ΛCDM needs not to be altered.

In the following the two implementations – the f (R) Hu-Sawicky model and the symmetron
model – are presented, which differ in the value of |φ∞ − φc |. A similar implementation for
a chameleon model was done by Brax et al. (2010b) and in time of writing Lombriser et al.
(2013) published a work for the same f (R) model including excursion sets. The spherical
collapse has not been applied to the symmetron model before.
4.3 Spherical collapse in screened modified gravity 53

0.35 0.35
Log10 HMMŸL
0.30 EdS 0.30 12
0.25 LCDM 0.25 13
0.20 symm_A 0.20 14

R
Ž
R
Ž

0.15 H1014 MŸL 0.15 15

0.10 0.10
0.05 0.05
0.00 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
a a

Figure 4.1: Evolution of the spherical collapse. Left panel: Collapse in the Einstein-de-Sitter
(black line) and ΛCDM (red line) model compared to the evolution of a sphere with mass
M = 1014 M⊙ in the symm_A model. Right panel: Collapse in the symmetron model using
the symm_A parameters for different masses.

4.3.1 Symmetron implementation

In the symmetron model the effective potential is unbroken far away from the sphere before
the time of symmetry breaking assb . Afterwards, the scalar field in the background settles
at
q
assb
3
2L2 βρc0 Ωc0 1 − a
φ∞ = ± . (4.51)
a3ssb MPl

Naturally, inside the sphere the potential is also unbroken for a < assb since the density is
always higher there. Also afterwards, for a > assb , this unbroken state inside the sphere is
expected to persist for most parameters and sphere masses, meaning φc = 0. If, however,
the density inside the sphere drops due to the expansion of space below the threshold value
of ρssb = Ωm0 ρc0 a−3
ssb , then the effective potential inside the sphere can brake.

Hence, the scalar field value inside the sphere is given by

if ρ > ρssb


 0
 q
φc = 2L2 β ρc0 Ωm0 (ρc0 Ωm0 − a3ssb ρ) (4.52)
± otherwise.


a3ssb MPl

Note, that φc as well as φ∞ both possess two solutions after symmetry breaking. This is
because for a local density ρ > ρssb there is a potential minimum in the positive and negative
φ-region (see Fig. 2.2 in Sec. 2.2). As shown by Llinares & Mota (2013a) although it is indeed
possible for the scalar field to settle in both minima, only the solution with the same sign
as the background is stable. If, for example, φ∞ takes a positive value after assb , only the
positive solution of φc is stable. Therefore, we decided to force φc ≥ 0 and φ∞ ≥ 0 for all
the solution discussed at first. In the last part of this section, we also discuss briefly the
possibility when φ∞ ≥ 0 and φc ≤ 0.
54 4 Spherical Collapse and Virialization

In the first, the stable, case the thin shell factor given by Eq. (4.49) is zero for a < assb and
can afterwards take a maximum value of
q
assb
3
∆R 8L2 πRρc0 Ωm0 1 − a
= , (4.53)
R 3a3ssb Mc

if φc = 0 and a > assb .

Fig. 4.1 shows the development of the spherical collapse for the Einstein-de-Sitter, ΛCDM
and symmetron models, including different masses in the latter. As expected, the collapse
in the symmetron case is mass-dependent and thus the resulting quantity ∆c should be,
too. The next step is to analyze the behavior of ∆c with the help of the development of the
virialization time avir , the initial overdensity δi , the evolution of the kinetic and potential
energy and – maybe most importantly – the fifth force which is responsible for the deviation
from ΛCDM.
To quantify the fifth force, the formerly introduced T is used. T takes values from 0 to 1
and can be understood as the normalized γ parameter. Fig. 4.2 shows T as a function of the
expansion factor for the parameter sets symm_A–D 1 . Since symm_C is the only parameter
set where β takes a value of 2 instead of 1, the gravitational force in this model can be
amplified by a factor of 9 instead of only three. Visually, the four panels of Fig. 4.2 can be
split into two categories: The first column with symm_A, C and the second column with
symm_B, D. In the first column, for the smaller masses T rises to its maximum right after
the time of symmetry breaking assb to then – at some point – decline monotonically. Note
that the evolution of the fifth force after assb depends also on the mass of the sphere: For
the smallest spheres the fifth force is acting with full strength on the system for (nearly) the
whole collapse (the virialization times are marked with a round dot on each curve in Fig. 4.2).
Consequently, the collapse evolution for very small spheres is identical to a ΛCDM collapse
with an effective gravitational constant of Gef f = (1 + β 2 )G. For the bigger spheres, T and,
hence, also γ ≡ Gef f /G, might not reach its theoretical maximum at all as the density of
the sphere at assb is already at a point where the self-screening mechanism works. Hence, we
expect the standard ΛCDM collapse to be a limit for very high masses. The second column
of Fig. 4.2 – which contains the evolution of T for symm_B and symm_D – shows an entirely
different behavior: After the time of symmetry breaking (assb = 1/3, 1/4 for symm_B and
symm_D, respectively) γ reaches for all masses its possible maximum, followed shortly after
by a rapid decline of the fifth force. Afterwards, for the symm_D model, the value of T
rises again, in some cases as high as its maximal possible value, to then decline in the same
fashion as can be observed in the symm_A and symm_C case. In the symm_B case no
second incline of the fifth force happens. Instead, T declines monotonically.
These two categories of the various sphere evolutions are due to the two possible regimes of
φc stated in Eq. (4.52): Either the density ρ is always greater than ρssb resulting in a constant
φc = 0 or the effective potential inside the sphere is also broken resulting in a smaller thin
shell factor. These two regimes can be observed in Fig. 4.3, where the matter density inside

1 See Table 3.1 in Sec. 3.3 for a summary of the model parameters.
4.3 Spherical collapse in screened modified gravity 55

1 1
LogHMMŸL
symm_A

symm_B
12.
13.
0.5 .5
14.
15.

0 0
T

1 1
symm_C

symm_D

0.5 .5

0 0
0.2 0.6 1 0.2 0.6 1

Figure 4.2: T , which is defined in Eq. (4.48) and measures the strength of the effective gravi-
tational constant Gef f , as a function of scale factor. As an example, four masses for each sym-
metron model are shown. The points mark the time of virialization for each mass. The differ-
ent values for assb = (1/2, 1/3, 1/2, 1/4) for symm_A–D explains the different positions of the
sharp rise of the fifth force.

the sphere is plotted against time. In the symm_A and symm_C case (left column) ρ & ρssb
for all shown masses, meaning all solutions are member of the first category. In the symm_D
case, on the other hand, the situation is quite the opposite as for all the solutions exist a
time where ρ < ρssb . For the other parameter sets – symm_B – the affiliation depends on
the mass of the sphere. These two categories have to be remembered when interpreting the
following results.

Fig. 4.4 shows the initial overdensity needed for a collapse today, δi , the density contrast as
defined in Eq. (4.42), ∆c , the deviation of the virial radius and the time of virialization as a
function of the mass. For bigger masses, when the self-screening works more effectively, the
values approach the ΛCDM result as expected. Also, not surprising is for smaller masses,
when the self-screening is not yet active, the earlier the symmetry breaking happens, the
56 4 Spherical Collapse and Virialization

symm_A symm_B
1.5 .5
Log10 HѐΡc0 L

LogHMMŸL
11.
1 1 12.
13.
0.5 .5 14.
15.
symm_C symm_D 16.
1.5 .5
Log10 HѐΡc0 L

L


1 1

0.5 .5

0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8

a

Figure 4.3: Evolution of the matter density inside the sphere for the parameter sets
symm_A–D. In each panel the solution is shown for spheres with masses log(M/M⊙ ) =
(11,12,13,14,15,16) (black, red, blue, green, orange, purple) until the virialization time avir .
In addition, the ρssb threshold is marked for each model with a black dashed line.

bigger is the resulting deviation of δi from Λ since the fifth force has more or less time to
act on the object. The results suggests the existence of a mass threshold below which the
measured quantity stays constant. This can be explained via the phenomenon mentioned
before: Too small masses will never reach the density to be screened and hence the collapse is
independent of the initial radius. In addition, the increase of the coupling parameter β seems
have an enormous impact. Comparing the symm_C results with the ones for symm_A, it
can be seen that the deviation from the ΛCDM value is roughly doubled. Not shown are the
results with an increased L. For them the curve seems to be shifted to higher masses, which
is reasonable since the impact of the fifth force is strongest at length scales ∼ L.
Between the two mass regimes mentioned where the values are constant or approaching a
constant value, two features in the ∆c versus mass plot are striking: Firstly, a local maximum
for which is especially present in the symm_D case and, secondly, a local minimum that
 4.3 Spherical collapse in screened modified gravity 57

4
350 symm_A

300 symm_B
symm_C 3
250

104 ´ ∆i
symm_D
Dc

200
2
150

100 1
50
11 12 13 14 15 16 11 12 13 14 15 16
LogHMMŸL LogHMMŸL

1.0
0.96

0.8 0.94
R vir R vir,Lcdm -1

0.92
0.6
avir

0.90
Ž

0.4 0.88
Ž

0.86
0.2
0.84
0.0
11 12 13 14 15 16 11 12 13 14 15 16
LogHMMŸL LogHMMŸL

Figure 4.4: ∆c (top left), δi (top right), the relative deviation of the virial radius compared
to ΛCDM (bottom left) and the time of virialization avir (bottom right) are shown for the sym-
metron model. The symm_A,B,C,D are shown in red, blue, green and orange, respectively.
The gray dotted line shows the ΛCDM value as a comparison.

marks the point of maximum deviation from ΛCDM. The existence for these two features is
explained in the following in detail. The knowledge obtained from the simplified spherical
collapse model about the evolution of a sphere can later on be used for the analysis and
interpretation of the N -body results.

The reason for the characteristic mass range with the most deviation from the ΛCDM virial
radius and therefore also ∆c , i.e. between the low mass and the high mass regime mentioned,
can be seen by studying the time of virialization avir . This point in time seems to be always
relatively close to the ΛCDM value (with the constant shift for small masses), except in a
certain mass range where the virialization occurs earlier. Consequently, the sphere has not
reached the density of the ΛCDM case yet and ∆c differs the most.

The dependency of the virialization time on the model parameters is more complicated, as
58 4 Spherical Collapse and Virialization

1.0 1.0

0.9 0.8

0.8

2TW + 1
0.6
WWvir

0.7
0.4
0.6

0.5 0.2

0.4 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.60 0.65 0.70 0.75 0.80 0.85 0.90
Ha-ata Lavir a

Figure 4.5: Evolution of the energy quantities in the symm_A model. The color codes are
indicate the mass of the spheres, with log(M/M⊙ ) = (12.1, 13.1, 14.1) (black, red and blue line,
respectively). Left panel: Potential energy relative to its value at virialization Wvir between
the turnaround ata and virialization point avir . Right panel: 2T /W + 1 – which1 is zero at
virialization – versus a.

many dependencies have to be considered. It is clear, however, that from the turn around
point the kinetic energy T is a positive function and steady increasing from zero whereas the
potential energy W is a negative function with a behavior that influences the outcome of
avir . In order to understand why the time of virialization has this characteristic minimum,
we analyze the evolution of three distinct sphere masses in the symm_A case. The masses of
the spheres has been chosen in way that one mass is in the point of maximum deviation (M =
1013.1 M⊙ ), while the other two are smaller (M = 1012.1 M⊙ ) and bigger (M = 1014.1 M⊙ ).
The analysis is carried out for the parameter set symm_A as in this case ρ > ρssb for
all considered masses. Hence, in this case φc = 0 for the whole collapse and matters are
simplified.

The evolution of the kinetic and potential energies of the spheres is shown in Fig. 4.5. In
the left panel, the potential energy is displayed in the time span of interest – between the
turn around point ata and the point of virialization avir . In the fully screened (Gef f ≈ G,
gray dotted curve) or totally unscreened (Gef f ≈ (1 + β 2 )G, blue curve) cases, W is a steady
decreasing function of the scale factor. In opposition, for a changing fifth force over time the
energy of the scalar field is increasing (red and blue curve). The crucial difference between
the solutions for Mc = 1013.1 M⊙ and Mc = 1014.1 M⊙ (red and blue curve in Fig. 4.5,
respectively) is clear by comparing the evolution of the fifth force in both cases in Fig. 4.2
(top left panel): For the heavier sphere γ is a smooth function between ata and avir but for
the lighter a break between the decreasing and fully unscreened regime exists additionally.
The transition between the two regimes leads to flattening of the W (a) curve in the time
shortly before avir . The “delay” means T reaches faster the threshold of virialization, which
4.3 Spherical collapse in screened modified gravity 59

is best seen in terms of 2T /W + 11 , a quantity that is zero for virialized objects. The right
panel of Fig. 4.5 shows 2T /W + 1 against a and the decrease proceeds in a similar way for all
displayed masses but Mc = 1013.1 M⊙ where instead a double slope behavior is observable.
Due to the steep tail the lower value of avir occurs and therefore the characteristic extremum
in ∆c arises.
The other feature in the ∆c versus mass plot (Fig. 4.4, top left panel) concerns the local
maximum for masses slightly smaller than the just discussed ones. It only exists if the
potential is broken also inside the sphere, i.e. if φc is different from 0 at some stage of the
collapse. As discussed before, this results in a smaller thin shell factor and, consequently,
in a weaker additional force and in less deviation from ΛCDM. So is the maximum most
clearly present for symm_D, where the densities are also much lower than ρssb (compare
with Fig. 4.3). For symm_A and symm_C the feature is non-existent as ρ & ρssb for all
times. Located between these cases is symm_B since here ρ & ρssb some masses existent
valid.
The second solution for the thin shell factor was already mentioned in the beginning of this
section and will be briefly discussed here. The scalar field inside the sphere can settle at
the other minimum than the outside solution, resulting in opposite signs of φc and φ∞ .
This leads to a overall bigger thin-shell factor and therefore to a higher additional force.
The existence of these two different values of the scalar fields in N -body simulations has
been shown by Llinares & Mota (2013a) by including time derivatives of the field, that is,
not assuming the quasi-static approximation. There are still a lot of open questions about
the behavior of the scalar field in this case, e.g. regarding the stability of the scalar field

1.0
350 symm_A
symm_B 0.8 Log(M/M☉)
300
symm_C 15
250 0.6
16
Dc

symm_D
T

200
0.4
150
100 0.2
50 0.0
11 12 13 14 15 16 0.0 0.2 0.4 0.6 0.8 1.0
LogHMMŸL a
Figure 4.6: Comparison between the solution with a the interior (φc ) and the background
solution (φ∞ ) having opposite signs (dashed lines), and the solution showed previously (solid
lines). Left panel: ∆c versus mass of sphere. Right panel: T as a function of time using the
symm_D solution for two different masses parameters shown previously in Fig. 4.2. In this
panel is in addition to the dashed curves unnatural solutions with a forced φc = 0 added with
dotted curves.

1 βvir ≡ (2T − Es )/W + 1 is the virialization parameter and is properly introduced in Sec. 4.4.
60 4 Spherical Collapse and Virialization

value. Thus, this second solution is merely shown here as a mathematical possibility in order
to help understanding better the symmetron breaking process. Fig. 4.6 shows the results
obtained through the second possible solution – that is φ∞ and φc possessing different signs
– with dashed lines. As expected, this solution shows greater deviation from ΛCDM for the
sphere evolution where ρ < ρassb occurs. The simple explanation for that behavior is the
even greater thin-shell factor than in the unbroken interior potential. For bigger spheres a
sudden restoring of the symmetry can happen, which leads to a sudden decrease (as opposed
to a sudden increase for the previously discussed solution) in T . For illustrative purpose a
solution with a forced φc = 0 was added to the right panel of Fig. 4.6. When the three lines
meet, the symmetry inside the sphere is restored, that is ρ > ρssb .

4.3.2 f (R) implementation

The spherical collapse implementation and results in f (R)-gravity is very similar to the one
discussed in the symmetron model and will therefore be only presented in brief.

The thin shell factor depends on the value of the scalar field inside and far away from the
sphere. Therefore, Eq. (2.72) is simply applied for ∆ρ = Ωm,0 a−3 in order to obtain φ∞ and
∆ρ = ρ/ρc,0 = 3Mc /4ρc,0 πRc3 for the center of the halo. This leaves the thin shell factor
as
∆R |φ∞ − φc |
=
R 6βMPl Φ
" #n+1
|1 − fR0 | Ωm,0 + 4ΩΛ,0
= 2
. (4.54)
12β Φ ρ/ρc,0 − Ωm,0 a−3

1.0
1.30
0.8
1.25
2TW + 1

1.20 0.6
Γ

1.15
0.4
1.10
0.2
1.05

1.00 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.6 0.7 0.8 0.9
a a

Figure 4.7: γ (left) and βvir = 2T /W + 1 (right) for the f (R) model with n = 1, |fR0 − 1| =
10−4 (fofr4 ). The black, red, blue and green curve denote the mass of the sphere as 1012 , 1013 ,
1014 and 1015 M⊙ , respectively. A value of βvir = 0 defines the time of virialization avir and is
marked with a dot on the corresponding curve in the left panel.
4.3 Spherical collapse in screened modified gravity 61

4.0
350
3.5

104 ´ ∆i
300 3.0
Dc

250 2.5

2.0
200
1.5
11 12 13 14 15 16 11 12 13 14 15 16
LogHMMŸL LogHMMŸL

Figure 4.8: ∆c (left) and δi (right) for the f (R) model. For all curves n = 1, but |fR0 − 1| =
(10−4 ,10−5 ,10−6 ) (fofr4, fofr5, fofr6 ) are shown in red, blue and green, respectively. The gray
dashed line shows the ΛCDM value as a comparison.

Again, Φ = R2 ρ/6MPl
2 is the Newtonian potential of the sphere.

The results, in the form of ∆c and δi plotted against the halo mass, are shown in Fig. 4.8
and the evolution of a sample collapse in form of βvir and γ are shown in Fig. 4.7. The same
features as in the symmetron case can be spotted and will not be discussed again in detail: (i)
Approach of the ΛCDM value with increasing mass as the screening has a bigger effect, (ii)
constant value for smaller halos and (iii) the characteristic minima in ∆c in the intermediate
regime. However, the local maximum that was present for symm_B–D does not exist as the
field value in the interior of the sphere stays small and does not change dramatically as it is
possible in the symmetron mechanism. Consequently, the f (R) solutions resemble more the
solutions obtained for the symmetron model without a symmetry breaking inside the sphere
as in the symm_A case. Also, the feature (ii) is not exactly the same but altered in two
ways: First, the constant level in ∆c is closer to the ΛCDM level as β has a lower value and
secondly the constant limit does not exist for δi at all. This is due to the fact that there is
no sudden switching on of the fifth force but instead a more continuous evolution.

4.3.3 Conclusions of the spherical collapse model

In this section the spherical collapse model was studied numerically for the symmetron
and the Hu-Sawicky f (R) model. The computer code produced for this purpose was tested
reproducing the analytical results derived for a flat Einstein-de-Sitter universe and obtaining
the ΛCDM values for two different virialization criteria. The main resulting quantity of
interest is the overdensity parameter ∆c as it is important for calculating halo properties
and therefore relevant if these are compared using different theories. In addition, the simple
model allows an understanding of the screening mechanism and its effect on halos. This
qualitative behavior was found to be in agreement with Brax et al. (2010b) and Wintergerst &
62 4 Spherical Collapse and Virialization

Pettorino (2010) where the spherical collapse model has been analyzed for similar underlying
models.

The results show a possible major deviation from the ΛCDM ∆c value which may cause
various definitions of where the boundaries of a halo lies. To correct for this effect a proper
treatment is needed in finding halo attributes, which would be possible to implement e.g. in
the halo finding code. Although the problem is known, until now, no such code exist (Knebe
et al., 2013). When interpreting these results, it is, however, important to keep in mind that
major simplifications have been made:

(i) No shell crossing: As there is no shell crossing allowed, the matter density inside the
sphere stays constant per definition. This is certainly not the case in a dynamical
model as pointed out by Brax et al. (2010b) because of the smooth scalar field solution
and the arising effects in the shell region. Engineer et al. (2000) and Shaw & Mota
(2008) constructed semi-analytical extensions to the spherical collapse model which
take the effects of shell-crossing into account and would be worth including.

(ii) Homogeneous initial density and no pressure: A realistic model would result in a
power-law density profile as known from simulations and observations (Ludlow et al.,
2013; Okabe et al., 2013). This will have various implications on the fifth force, e.g.
lead to a screened and unscreened region (Lombriser et al., 2012). A consequence of
this is that there is no tangential motions and not local perturbations in the sphere.

(iii) No environmental screening: The neglection of the environmental screening, i.e. the
vanishing of the fifth force due to the location of the halo in a denser surrounding,
e.g. inside a filament, is maybe a very crude approximation especially for the smaller
masses, as the environment can have a major effect on a halo (Winther et al., 2011).
A more detailed analysis using an excursion set approach was done by Li & Efstathiou
(2012) and Lombriser et al. (2013).

Therefore, the results obtained in this section give a good guideline for further analysis and
allow to understand the process of halo formation in screened modified gravity cases in a
qualitative way but should not be over-interpreted. Also, worth noticing is that a change
of ∆c does not lead to an equally strong change in the halo mass since the density of real
dark matter halos falls off with ∝ r −3 in the outskirts. Hence, we expect only second order
corrections to the halo properties and leave a detailed study for future work.

4.4 Virialization state in N -body simulations

In the first section of this Chapter, the virial theorem was derived. It was applied to the
spherical collapse model where the matter density is given by a simple function. In this
section, the question about how to define an object as virialized in N -body simulations, that
is when dealing with discrete particles, is studied.
4.4 Virialization state in N -body simulations 63

First, let us recall what a virialized object is mathematically: As was shown in Sec. 4.1, a
virialized object fulfills the condition
2T − ES = −W . (4.55)
Therefore, the potential energy W , the surface term ES and the kinetic energy T have to be
calculated for each halo of the simulation output.
In order to measure the degree of virialization, the virialization parameter βvir can be intro-
duced similarly to Shaw et al. (2006):
2T − Es
βvir ≡ +1. (4.56)
W
This means, one can define a halo being sufficiently relaxed or virialized if |βvir | is smaller
than a certain threshold.
The question now is how to implement the expressions for the energy quantities. Clearly,
the terms derived in Sec. 4.1 cannot be used directly as it was done for the spherical collapse
model in Sec. 4.2 since N -body simulations constist out of discrete particles. Therefore, the
next three subsections are devoted to the answer of that question starting with the kinetic
energy, then the potential energy and last the surface term.

The kinetic energy

The kinetic energy is the most straightforward one, as it is given by

1X
T = mi |vi − vH |2 (4.57)
2 i∈H

with the velocity of particle i, vi , taken relative to the halo velocity vH . This equation is
the discretization (4.18) with the halo velocity
1 X
vH = mi vi . (4.58)
MH NH i∈H

In the case of particles with equal masses this relation becomes simply
1 X
vH = vi (4.59)
NH i∈H

where NH is the number of particles in the halo.

For completeness it is repeated, that the quantity of velocity dispersion σ may be introduced1
in this context with
2 1 X
σH = |vi − vH |2 . (4.60)
NH i∈H

1 Note that this definition differs from the one made in Sec. 4.1, as this one is the velocity dispersion of the
halo H and the previous one the three-dimensional velocity dispersion of a volume element dV .
64 4 Spherical Collapse and Virialization

The potential energy

The potential energy W can be calculated in two different ways, given by (4.15) and (4.16):
Z
3
W = d x ρx · a
V
G ρ(x′ ρ(x))
ZZ
3 3
W̃ = − d x d x′ .
2 |x − x′ |
V

This leaves in general three possibilities of calculating the potential energy of a halo H using
discrete particles:
1. Using a simple direct summation approach:
G X mi mj 1X
WDS = − q = mi ΦDS,i (4.61)
2 i,j∈H |xi − xj |2 + ε2 2 i∈H
i6=j

2. Summing over the gravitational potential at each particle position φi = φ(xi ):

X mi
WP = Φi (4.62)
i∈H
2

The gravitational potential at each particle’s position Φi = Φ(x) was obtained from
the gravitational potential used by the N -body code through interpolation.
3. Utilizing the particles’ acceleration instead:
X mi
Wa = (xi − xH ) · ẍi (4.63)
i∈H
2

Here the position of the halo xH is inserted because of the comoving reference system.
The first possibility is very computational expensive and does not result in exactly the same
value as the second one because the gravitational potential is obtained via summation of
the whole box and not just the other members of the halo. Although the difference is small
– especially near the center of the bigger halos, where most of the contribution comes from
the other members of the halo – there is another problem, namely the normalization of the
gravitational potential. This problem can be overcome by assuming that the central particle
is only affected by the members of the halo and shifting the values of Φi by ∆Φ = ΦDS,0 −Φ0 ,
i.e. the difference between the gravitational potential at the central particles’ position Φ0 ≡
Φ(xH ) obtained directly from the N -body simulation and the value calculated through direct
summation over the members of the halo.
This normalization technique, which can potentially avoid the computationally expensive
task of obtaining the gravitational through direct summation for each particle, is shown in
Fig. 4.9 in the form of Φi , ΦDS,i and Φi + ∆Φ. One can clearly see that although ΦDS,i and
Φi + ∆Φ are normalized in the center of the halo, they do not agree on the outskirts. This
4.4 Virialization state in N -body simulations 65

0
Gravitational potential (1012 m2/s2)

-2

-4

-6

-8
Φi
Φi+∆Φ
-10
ΦDS,i

-6 -4 -2 0 2 4 6
Projected distance in Mpc/h

Figure 4.9: Comparison between the methods obtaining the gravitational potential for the
energy calculation. In black the values obtained from the N -body simulation as an interpo-
lation from the grid cells to the particles’ position are shown (lowest curve). The same data
points shifted by ∆Φ are plotted in light gray. As a comparison are is the gravitational poten-
tial obtained through direct summation, ΦDS,i , shown dark gray. On the x-axis the quantity
∆r · sgn(∆x) is used as projected distance.

is due to nearby halos and unbound objects, which can be inside or close by the halo, that
all have an gravitational effect on the halos’ particles.
The third possibility – using the scalar product between the acceleration and the position of
the particle – has two main advantages: Firstly, it is computationally much less expensive as
not even ∆Φ has to be computed directly. Secondly, in section 4.1 it was shown that W = W̃ ,
but only if gravity is the only force acting on the body. In the case of the modified gravity
models, there is an additional force acting on the particles. This can be also rephrased in
terms of energy: The scalar field which is part of these models, contains energy, too.
In Fig. 4.10, it can be seen that the three methods differ more than expected. It is also
noticeable that without the renormalization of ∆WP = ∆ΦMH /2, the energies obtained
through summation of the gravitational potentials (rightmost panel) differs greatly from
the other two methods. However, with the renormalization the mean, first and third quan-
tile of the distributions are very similar: (0.06, −0.54, −0.30) TJ kg−1 for the distribution
of Wa /MH (third panel), (−0.37, −0.42, −0.26) TJ kg−1 for the direct summation method
(second panel) and (−0.35, −0.43, −0.27) TJ kg−1 for (WP + ∆WP )/MH (first panel from
66 4 Spherical Collapse and Virialization

−1
M☉ h
800
13
Number of halos with M>3 ⋅ 10
600
400
200
0

-1.5 -0.5 0.5 -2.0 -1.0 0.0 -2.0 -1.0 0.0 -3.0 -2.0 -1.0 0.0
12
W/MH (10 ⋅ J/kg)

Figure 4.10: The four ways of calculating the potential energy: Histograms with the poten-
tial energy per halo mass for halos bigger than 3 × 1013 on the x axis. From left to right it is
shown: (i) The normalized WP , that is WP + ∆ΦMH /2, (ii) WDS , (iii) Wa , (iv) WP . Note that
the binsize is doubled in the last panel. All data used is from the ΛCDM run.

the left). The shift in the mean of Wa /MH is due to ∼ 10 outliers with values as high as
185MH TJ kg−1 .

From these techniques the one which is best suited in order to calculate βvir can be seen
by comparing the distribution of virialization parameters shown in Fig. 4.11. While the
βvir -distribution obtained from the direct summation and the particles’ acceleration agree
well, using WP + ∆WP leads to more spread out βvir values. This is due to the crude
approximation of ∆WP : As it was shown in Fig. 4.9, ΦDS,i and Φi + ∆Φ do not agree well
on the halo boundary and consequently WDS and WP + ∆WP can differ substantially.

The surface pressure term

The third quantity, the surface pressure at the boundary ES , is often not stated because for
a system in total isolation this term vanishes. For a halo where a rather arbitrary cut has
been made at Rvir to define the members of a halo this is obviously not true: Particles which
are not member of a given halo but nevertheless gravitationally bound add a contribution
to the overall energy. Consequently, the term stated in Eq. (4.22) cannot be neglected.

Throughout the calculations presented in this thesis, the ES term was calculated following
Shaw et al. (2006) who use an approximation that is motivated by the ideal gas law. In this
the pressure p and the kinetic energy T are connected by

2
pV = T . (4.64)
3
4.4 Virialization state in N -body simulations 67
700
−1

600
M ☉h
13

500
Number of halos with M>3 ⋅ 10

400
300
200
100
0

-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
β vir

Figure 4.11: Distribution of the virialization parameter with (red color, downwards hatch-
ing) and without (green, upwards hatching) the surface pressure term Es . From left to right
for the gravitational potential energy WDS , WP + ∆WP and Wa was used in order to calculate
βvir . The data shown is from the ΛCDM data set.

Using as volume the outermost 20% of the halos particle distribution, the surface pressure
can be approximated by

1 i mi (vi − vH )2
P
pS = 3 − R3 ) . (4.65)
3 4/3π(R1.0 0.8

Here the notation Rx is used to describe the x-percent quantile of the particle distribution.
Therefore, the surface pressure term can be calculated using
3
Es ≈ 4πR0.9 pS . (4.66)

The effect of the surface pressure term are similar to the ones described by Shaw et al. (2006),
i.e. shifting the distribution of βvir towards smaller values as it can be seen in Fig. 4.11, in
which βvir is shown using the data of the N -body simulation.
68 4 Spherical Collapse and Virialization

4.4.1 Virialization parameter in scalar-tensor theories

In the case of scalar-tensor theories, the energy of the scalar field has to be included in
the considerations, as well. This is done automatically in Wa , if the total acceleration is
considered, i.e. ẍi = ẍN ,i + ẍFifth,i . That this additional contribution cannot be neglected
can be seen in Fig. 4.12, where βDS,a = (2T − ES )/WDS,a have been compared: For ΛCDM,
the fofr6 or the symm_A model, the deviation is not significant, but in the more extreme
models (fofr4, symm_D) the direct summation method is clearly not sufficient. Therefore,
in this work βa was used as already done in the spherical collapse model.

ΛCDM fofr4 fofr5 fofr6

βDS
βa
600
400
−1
M☉h
200
13
Number of halos with M>3 × 10
0

symm_A symm_B symm_C symm_D

600
400
200
0

-0.4 0.0 0.4 -0.4 0.0 0.4 -0.4 0.0 0.4 -0.4 0.0 0.4
βvir

Figure 4.12: Distribution of the virialization parameter βvir for halos with mass
M > 3 × 1013 M⊙ h−1 . βDS (using the potential energy WDS obtained through direct sum-
mation) is shown in orange with upward hatching and βa (using Wa ) is shown in blue with
downward hatching.
5

Gravitational Redshift in Screened Modified Gravity

In the following chapter, the main results obtained in this thesis are presented. First, the
extraction of the data is explained and applied to the ΛCDM run as an example. Secondly,
the comparison of the quantities obtained in the f (R) and symmetron runs are compared to
the ΛCDM data and discussed. In order to understand the deviation in gravitational redshift
also the density profiles, the fifth force and the velocity dispersion profiles are analyzed. In
addition, the deviation in the matter power spectra and halo mass functions are shown
for completeness. In the last part, possible sources of error and a brief comparison with
observational data is laid out.

5.1 Obtaining the gravitational redshift

In order to obtain the gravitational redshift, first the gravitational potential at each particle’s
position is needed. To achieve this, an additional modification of the N -body code was made:
If a flag was set the code interpolated the properties (density, gravitational potential, ...)
to the particles’ position via inverse-CIC as done before; but instead of moving them and
continuing with the next time step these properties were written in a file. This allowed us
to use the efficient potential calculation method of RAMSES and to output the following
properties: Position of the particle xi , velocity of the particle vi (both used for the halo
finding process), Newtonian gravitational potential ΦN , Newtonian force per unit mass ẍN ≡
FN /m, fifth force per unit mass ẍFifth ≡ FFifth /m and the matter density ρ. Special attention
was made to output the Newtonian gravitational acceleration separately from the scalar field
as only the former influences photons as explained in Sec. 2.4.

The measured gravitational redshift was then defined as in Kim & Croft (2004) and Wojtak
et al. (2011):

∆Φ Φc − Φ
vg = = , (5.1)
c c
70 5 Gravitational Redshift in Screened Modified Gravity

where Φc is the gravitational potential at the center of the cluster which is – following the
definition of the center of the cluster done in Sec. 3.4 – the minimal gravitational potential
found in the halo. Therefore, vg is always negative and can be interpreted as the gravitational
blueshift of light seen by an observer located at the center of the halo.

Instead of defining a single point as the center of the cluster and calculating the gravitational
redshift with respect to that point, it is also possible to model the central galaxy as being
spread out and take an average over gravitational potentials of the most central particles as
the reference point. This procedure is used by Kim & Croft (2004) and leads to a flattening
of the profile. However, they found out that the effect is rather small for dark-matter-only
simulations. Due to this finding and also the fact that our results are consistent with the
NFW predictions, we do not adopt the alternative procedure. In order to find the deviation
in the gravitational redshift of modified gravity models with respect to ΛCDM – which is
the goal of this work – the choice of technique is not crucial. If, however, the focus is on
comparing the results with observational data, the measure of gravitational redshift is highly
important to maximize the signal-to-noise ratio. Conclusively, for future work the measure
of gravitational redshift might have to be modified.

5.2 The NFW profile

Navarro, Frenk & White (1995; 1996; 1997) found a fitting function to dark matter halos
given by

δchar
ρ(r) = ρc,0 . (5.2)
r
cv rv (1 + cv rrv )2

Here rv is the virial radius, cv the concentration value and δchar the characteristic overdensity
given by

1
δchar = vg(cv )c3v , (5.3)
3
where
1
g(cv ) = . (5.4)
log(1 + cv )cv /(1 + cv )

The virial radius is defined by the parameter v as the border where the halo density drops
below v times the critical density today. In Sec. 4.2 it was shown that this value is 178 for an
Einstein-de Sitter universe 380 in the ΛCDM spherical infall model. As discussed in section
3.4, the results presented in this work are obtained with the value v = 200.

This leaves only one free parameter, cv , for halos of a given size. In spite of its simplicity,
the NFW-profile is still comparable with high resolution simulations (e.g. Ludlow et al.
(2013)).
5.3 Uncertainty estimation 71

The gravitational potential of a density profile given by Eq. (5.2) with a total mass of Mv
inside the virial radius is (Cole & Lacey, 1996)

g(cv ) r
Φ(r) = −Mv G log(1 + cv ) (5.5)
r rv

and hence
GMv rv
 
∆Φ(r) = Φc − Φ(r) = −cv g(cv ) 1+ log(1 + cv r/rv ) (5.6)
rv r

because Φc ≡ limr→0 Φ(r) is finite although Eq. (5.2) is not.

√
For the projected potential at a distance rp = r 2 − ∆x2 from the center, we follow Lokas
& Mamon (2001) and define the 2D mass distribution function:

ρ(r)
f (r,rp ) = (5.7)
ρΣ (rp )

where ρΣ (rp ) is the average density along the line of sight, i.e.

Z∞
rρ(r)
ρΣ (rp ) = 2 q dr (5.8)
rp r 2 − rp2
   
rv
Mv g(c) arccos crp
= 1 − q  . (5.9)
2π rp − (rv /c)2
2 2
(crp /rv ) − 1
√
Note that in the last term arccos(1/x)/ x2 − 1 the nominator as well as the denominator
become purely complex for x = crp /rv < 1, resulting nevertheless in a real result.

So the projected potential difference is

Z∞
2 rρ(r)
∆Φp (rp ) = q ∆Φ(r) dr . (5.10)
ρΣ (rp ) r 2 − rp2
rp

These equations can be used to fit the data. The resulting concentration values found can
be compared to observational data.

5.3 Uncertainty estimation

The question we want to address is: How different is a prediction for an observable given
one theory compared to the same observable in another underlying model? In order to
answer this question, the measure of interest is the ratio of these two quantities. This
means concretely for the observable O and the model i the relative deviation is given by the
72 5 Gravitational Redshift in Screened Modified Gravity

observable in the modified gravity model divided by the same observable obtained in the
ΛCDM run (minus one), i.e.

R(i) = ∆O/OΛCDM = Oi /OΛCDM − 1 . (5.11)

The easiest and most straightforward procedure to obtain an uncertainty estimate for the
relative deviation would be to run the simulation several times. Out of the resulting means,
and their standard deviations it is possible to obtain an overall ratio and error (Lee &
Forthofer, 2006, Chap. 4). This method was chosen by Brax et al. (2012, 2013) for the
matter power spectra and halo mass functions. The aim of this work is instead to emphasize
on the halo structure, which requires a higher dynamical range than the one used by Brax
et al. Hence, the computational power is not available for this procedure and this precision.

Please note, that simply taking the variance of the observed values inside a single simulation
run would not be sufficient because the point of interest is not the confidence interval of an
observable but of its mean. The crude estimation of the individual samples (i.e. the halos)
being independent would lead to the standard error of the mean.

A method that is commonly used to avoid running a large number of computationally ex-
pensive simulations is to create new “fake” data points. The intention is that they are being
drawn from the same (or a very similar) distribution function as the underlying distribution.
One possibility used for projected quantities is to change the projection angle as done for
instance in Forero-Romero et al. (2010). Another possibility would be to divide the simu-
lation box in several sub-samples and obtaining multiple values instead of just one from a
single simulation run, which is the procedure we follow here. The box was split in n = 8
equal sub-boxes, each giving a relative deviation Ri , which results in an overall R̄ mean and
error of this mean is given by (Everitt & Skrondal, 2002)

Pn !1/2
s 2
i=1 (Ri − R̄)
SER̄ = √ = . (5.12)
n n(n − 1)

The error estimation obtained from this method is not as good as for a sample of independent
runs. This is the case especially for larger halos as the long modes are not independent in
each sub-box. Nevertheless, this procedure is sufficient for our purpose.

Another important point in the uncertainty estimation is the consideration of systematical

errors due to the resolution of the N -body simulation. As mentioned in Sec. 3.3, the simu-
lation box had a side-length of 256Mpc h−1 which contained 5123 particles. The maximum
refinement level was chosen to be 16, which is the maximum number of times the box can
be split in a high-density region. Thus, in the highest level of refinement the simulation box
has 216 nodes per dimension, resulting in a minimum cell size of ∼ 4 kpc h−1 . We estimated
the resolution to be twice that minimum cell size, which is ∼ 8 kpc h−1 . In order to show
the capabilities and limits of the available simulation sets, we preferred sometimes to show
data points located below this resolution limit as well.
5.4 Cluster profiles in the ΛCDM model 73

106
1.5 -10
105
1
104

∆y in Mpc/h
0.5 -15

vg in km/s
ρ/ρc0

103
0
2
10 -0.5 -20
1
10 -1

100 -1.5 -25

0.001 0.01 0.1 1 10 -1.5 -1 -0.5 0 0.5 1 1.5
R/Rv
∆x in Mpc/h

Figure 5.1: Left panel: Average density at the particle positions for halos with masses
1014 M⊙ /h < M < 1014.5 M⊙ /h. For the black round points only the particles of the host
halo are included, the gray triangles include the subhalos and the red line is a NFW fit to the
data. Right panel: Projection of a halo. The coloring and the contour lines correspond to the
difference in gravitational potential.

-2

-4
v (km s−1 )

-6

-8
(1012.5 − 1013 )M⊙ h−1
(1013 − 1013.5 )M⊙ h−1
(1013.5 − 1014 )M⊙ h−1
-10 (1014 − 1014.5 )M⊙ h−1
NFW fit
0.01 0.1 1
R in Mpc/h

Figure 5.2: Difference in gravitational redshift of the ΛCDM simulation between the central
galaxy and the individual halos binned in radius and halo mass. The error bars show the stan-
dard deviation of all halos and are therefore not the same as the model comparison. The NFW
fit uses the mean values of the high mass bin.
74 5 Gravitational Redshift in Screened Modified Gravity

-2
v (km s−1 )

-4

-6

(1012.5 − 1013 )M⊙ h−1

(1013 − 1013.5 )M⊙ h−1
-8 (1013.5 − 1014 )M⊙ h−1
(1014 − 1014.5 )M⊙ h−1
NFW fit
0.01 0.1 1
R in Mpc/h

Figure 5.3: Difference in gravitational redshift of the ΛCDM simulation between the central
galaxy and the individual halos binned in radius and halo mass. This time only the virialized
(|β| < 0.2) halos have been included and the radii have been virialized. The NFW fit uses the
mean values of the 1014 − 1014.5 M⊙ h−1 mass bin.

5.4 Cluster profiles in the ΛCDM model

The right panel of Fig. 5.1 shows the projection of a single halo. The color coding corresponds
to the average gravitational redshift associated to the particles in the area. It is possible to
that close to the center of the halo (at ∆x = ∆y = 0) the gravitational redshift reaches its
minimum. It does, however, not reach zero since a projection of the halo is shown. In the left
panel of Fig. 5.1, the mass density for halos with mass 1014 M⊙ h−1 < M < 1014.5 M⊙ h−1 is
plotted against the normalized radius. Subhalos are included in the data points represented
by gray triangles and a bump around R ≈ Rv due to them is clearly visible. The NFW fit
provided follows nicely the density points without subhalos (black dots).

In order to obtain the halo redshift profiles we take the difference in gravitational potential
between the central particle and each member of the halo. The result for the complete
sample – binned in radial bins and halo mass bins – is presented in Fig. 5.2. As expected the
profile is steeper for more massive halos and is consistent with the NFW profile calculated
in Sec. 5.2. This is also consistent with the theoretical predictions from Cappi (1995).

Please note, as consequence of the used binning, each data point is not represented by the
same number of particles or halos: Especially for larger radii only the more massive halos are
contributing. As the lighter halos shown in Fig. 5.2 with halo mass of 12.5 < log(M h/M⊙ ) <
5.5 The f (R) models 75

13, have a typical virial radius of 0.26 Mpc h−1 , only a few particles were taken into account
for larger radii. This and the fact that some of these particles on the far edge are unbound
or part of elliptical halos lead to a wide spread of data far away from the center.
To overcome this problem and to avoid contaminating the sample with major mergers for
which the gravitational redshift is ill defined, we subtract halos with a virialization param-
eter |βvir | < 0.2, where βvir was calculated as described in Sec. 4.4. In other words, the
distribution of the virialization parameter shown in Fig. 4.12 is cut off at βvir = ±0.2.
This removes around half of all the halos from each data sample. Mainly smaller halos
are removed: For the ΛCDM data, the four halo mass bins adopted, i.e. log(M h/M⊙ ) ∈
((12.5, 13), (13, 13.5), (13.5, 14), (14, 14.5)), loose (28, 19, 13, 17) percent of the halos, which
leaves (13118, 5060, 1544, 315) halos. The three halos in the data set with masses larger than
1014.5 M⊙ h−1 are also unvirialized according to our definition. In addition, we normalize the
radii with the virial radius of each halo in order to make a fair comparison between halos
in each mass bin. The normalization allows us also to compare deviations in fixed radial
bins between different mass bins. As Fig. 5.3 shows, these measure decrease the variance of
the data points, especially for the smaller halo mass bins. Now, the trend of blueshifting is
visible for halos with masses 12.5 < log(M h/M⊙ ) < 13, too.

5.5 The f (R) models

As we are interested in how the introduction of a fifth force changes the halo profile, we will
investigate the relative deviation of the quantities vG and ∆ρ, that is ni /nΛCDM − 1 where
ni and nΛCDM are the quantity in the modified gravity data set i and in the ΛCDM data set,
respectively. As discussed in Sec. 5.3, the error was estimated by dividing the simulation box
into 8 sub-boxes and using the variance of the relative deviation calculated in each subset
to fix the error of the mean. The data sample was binned in two ways: First, the same way
as in Fig. 5.2–5.3 with the variation of the normalized radius in the four previously adopted
halo mass bins between 1012.5 M⊙ h−1 and 1014.5 M⊙ h−1 . In addition, the deviation at a
fixed distance as a function of the halo mass was analyzed. For the comparison between
the models only the virialized halos (with |βvir | < 0.2) were used. In all the Figures in
this section, the color coding is consistent and corresponds to the three f (R) parameter sets
summarized in Table 3.1.

5.5.1 Matter power spectra and halo mass functions

The relative deviation in the matter power spectrum ∆P/PΛCDM and in the halo mass
function ∆N (> M )/N (> M )ΛCDM shown in Fig. 5.4 is merely provided for completeness
and overview of the models.
The matter power spectrum for each model was calculated using the publicly available
powmes code (Colombi et al., 2009). The halo mass function N (> M ) is defined by the
number of halos whose masses are bigger than a mass M . The impact on the change of the
76 5 Gravitational Redshift in Screened Modified Gravity

0.5 1
fofr4
fofr5
fofr6
0.4 0.8

0.3 0.6

∆N/NΛCDM
∆P/PΛCDM

0.2 0.4

0.1 0.2

0 0
1012 1013 1014
0.01 0.1 1 10 Halo mass in M⊙/h
k (h/Mpc)

Figure 5.4: Relative deviation in the matter power spectra (left panel) and the halo mass
function (right panel) The error bars indicate the spread of values in a given bin.

halo mass functions as well as the linear matter power spectra in the generalized chameleon
and the f (R) model was already done by Brax et al. (2013) and Li et al. (2012) and is
not subject of this work. Please note, that the error bars provided show only the standard
deviation for one bin, not the deviation between several samples as done in the further
analysis. It can be observed that (i) the higher the |fR0 − 1| value, the higher the deviation
from ΛCDM and (ii) that each of the models has a certain halo mass range of impact. These
features can be seen clearer in the upcoming sections and will be discussed there. The third
point to note is the very high deviation in the fofr4 model not only for small scales, but also
for low values of k which suggest the need of renormalizing σ8 . See Sec. 5.8 for a discussion
about this and other systematic effects.
The variation in the halo mass functions agrees very well with the results from Li & Hu
(2011) who covered the same f (R) parameters in their considerations. They predict a further
growth in the deviation in the |fR0 − 1| = 10−4 case for halos up to M = 1015 M⊙ h−1 , but
no deviation in this mass range if |fR0 − 1| = 10−5 .

5.5.2 Chameleons in the wild: The fifth force unbound

In Fig. 5.5 the absolute value of the directly obtained additional acceleration |ẍFifth | is shown
as a function of halo mass for fixed radial bins (upper panels) and as a function of normalized
radius for fixed halo masses (lower panels).
In the upper row of Fig. 5.5, we can see that for every radial bin, each set of parameters results
in a certain characteristic halo mass range beyond which the fifth force is screened at the
higher end and the scalar field less affected at the lower end. In the fofr5 -model, for example,
the fifth force around 0.1Rv is the strongest for halo with masses 12.5 . log(M h/M⊙ ) . 14
(see upper right panel). In the fofr6 model this range can be seen in the radial bin around
the virial radius Rv : For M & 1014 M⊙ h−1 , the fifth force is completely screened. In
this radial bin, the fofr5 -peak moved to higher masses (∼ 1014 M⊙ h−1 ) since the screening
mechanism depends on the local matter density which declines with increasing radius. A
5.5 The f (R) models 77

120 12
0.09 <R/Rv< 0.11 0.99 <R/Rv< 1.01
100 10
fofr4
xFifth (10-12 m2/s2)

xFifth (10-12 m2/s2)

fofr5
80 fofr6 8

60 6

40 4
..

..
20 2

0 0
1012 1013 1014 1012 1013 1014
Halo mass in M⊙/h Halo mass in M⊙/h
15

40

60

100
1012.5-1013.0 M⊙h-1 1013.0-1013.5 M⊙h-1 1013.5-1014.0 M⊙h-1 1014.0-1014.5 M⊙h-1

80
xFifth (10 m /s )
-12 2 2

40

60
20

40
20

20
..
0

0
0.01 0.1 1 10 0.01 0.1 1 10 0.01 0.1 1 10 0.01 0.1 1 10
R/Rv R/Rv R/Rv R/Rv

Figure 5.5: Additional acceleration ẍFifth due to the presence of a scalar field in the f (R)
model. Upper panel: |ẍFifth | as a function of halo mass for in the radial bin R/Rv ≈ 0.1 (left)
and R/Rv ≈ 1 (right). Lower panel: |ẍFifth | as a function of normalized radius for the halo
mass ranges (log(Mlower h/M⊙ ), log(Mupper h/M⊙ )) = (12.5,13), (13.13.5), (13.5, 14) and
(14,14.5) (from left to right). In all the graphs the red circles mark the fofr4, the blue trian-
gles the fofr5 and the green diamonds the fofr6 run. In addition, the black horizontal line
corresponds approximately to two grid cells in the finest refinement level.

similar screening of the fifth force for the fofr4 model is expected for even greater densities.
This means, we expect the fofr4 curves in the upper panels of Fig. 5.5 to possess a maximum
as higher halo masses than available in the data set.
The lower panels of Fig. 5.5 show the radial dependency of the fifth force. In each panel, a
horizontal black line marks the approximate1 size of two grid cells in the finest refinement
level, which is roughly the resolution limit. A feature of all curves – with the exception of
the fofr6 results for high masses – is their characteristic maximum. Also, the decline of
the fifth force for increasing R seems to be identical for the outer radii. This independence
of the shape from the fR0 value can especially be seen for the fofr4 and fofr5 curves since
ẍF if th is much weaker in the fofr6 case. However, the position of the characteristic maxima
is highly dependent on the fR0 parameter: In the two lower-mass panels the fofr4 and fofr5
curve have their characteristic maximum around R ≈ 0.1Rv , while the fofr6 data points
suggest a maximum around 0.5Rv and Rv for the mass ranges 1012.5 − 1013 M⊙ h−1 and
1013 − 1013.5 M⊙ h−1 , respectively. Also the magnitude of the additional acceleration depends

1 For each mass bin the mean virial radius of the ΛCDM run was used to calculate the approximate grid
cell size.
78 5 Gravitational Redshift in Screened Modified Gravity

on fR0 . The simple relationship – the closer fR0 to unity, the stronger the additional force
– holds mostly. It is not necessarily true, however, in the “preferred mass range” which was
identified in the upper panels of Fig. 5.5. So is, for example, the fifth force stronger with
|fR0 − 1| = 10−5 (fofr5 ) compared to |fR0 − 1| = 10−4 (fofr4 ) for halos with masses between
1012.5 M⊙ h−1 and 1013.5 M⊙ h−1 .

In order to understand the features of Fig. 5.5 it is important to recall how the chameleon
screening works and how a fifth force arises in this context. As previously discussed in
Chapter 2, a fifth force arises if the value of the scalar field φ changes in space. This change
in φ happens, because the scalar field settles in the minimum of the effective potential
φmin whose position depends on the local matter density ρ. The screening nature of this
mechanism is due to the fact that φmin changes only minimally for high density regions.
That is, the fifth force is effectively zero if ρ is greater than a certain density threshold. The
value of this threshold depends on the model parameter |fR0 − 1|. Consequently, there is
two possibilities for a vanishing fifth force: (i) The local matter density ρ does not change
or (ii) ρ is greater than the screening threshold.

Since the halo density decreases with increasing R, the vanishing fifth force for large radii
is due to reason (i) . This means, fifth force declines at larger R because the matter density
does not change as strongly as in the regions around the characteristic peak in ẍFifth . The
decline of the additional acceleration left of the maximum can be due to (i) or (ii) : (i)
considering the density in the inner radii might be greater than the screening threshold
mentioned above or (ii) since the matter density profile flattens in the central region as
a result of the simulation resolution. Hence, the difference between the resolution limit
(marked with a vertical black line in Fig. 5.5) and the position of the characteristic peak
plays an important role. If the peak is very close to the resolution limit, then the decline for
smaller radii is due to reason (ii) . Another criterion for (ii) can be established by including
the mass dependency features of ẍFifth shown in the upper row of Fig. 5.5: If for the smaller
radial bin (left upper panel) the curve is rising – that is the fifth force is larger for more
massive halos and, thus, higher local densities – the screening in that mass range cannot be
active. Hence, the radial decline of the fifth force is due to (i) if the ẍFifth versus M curve is
falling for a given halo mass range. In addition, the characteristic peak in the ẍFifth versus
R data should not be too close to the resolution limit.

Applying these criteria to the data shown in the lower row of Fig. 5.5, we can state:

• The fofr4 result show the inner falling of the fifth force is due to reason (ii) for all
shown mass ranges. Here, the characteristic peak is always very close to the resolution
limit.

• For the fofr5 results (ii) is only the case for masses between 1012.5 and 1013.5 M⊙ h−1
(two leftmost panels). For the two bigger mass ranges, the decline can be accounted for
(i) , that is the screening in the central regions. It is noteworthy that the characteristic
peak for the mass range 1013.5 − 1014 M⊙ h−1 is at ∼ 0.1Rv and for 1014 − 1014.5 M⊙ h−1
around 0.4Rv .
5.5 The f (R) models 79

• The characteristic peak in the fofr6 data is always far away from the resolution limit.
Hence, the decline of the fifth force in the central regions of the halos is due to screening,
i.e. reason (ii) .

These findings are in agreement with the expectations: The lower |fR0 − 1|, the lower is the
density threshold above which the chameleon screening is effective.

5.5.3 Gravitational redshift profiles

Since the measurable signal of the gravitational redshift is proportional to the difference in
gravitational potential, the quantity vg is a direct measurement of the shape of the grav-
itational potential well. Hence, the relative deviation of this quantity, i.e. ∆vg /vg,ΛCDM ,
which is shown in Fig. 5.6–5.7, describes the change in this shape with respect to ΛCDM.
This means, if the potential well is deeper in the center and/or shallower for larger R,
∆vg /vg,ΛCDM is positive. With a fifth force acting on the particles, we expect additional
clustering to happen and, consequently, this is the expected case to happen. If, on the other
hand, the potential well is not as deep in the center and/or deeper in the outskirts than in
the ΛCDM model, the observable ∆vg /vg,ΛCDM will be negative.

In Fig. 5.6 the relative deviation of the gravitational redshift vg with respect to ΛCDM
is shown. We see, the smaller fR0 , the smaller is the deviation from ΛCDM, as expected.
Especially for the |fR0 − 1| = 10−6 case, the result is basically indistinguishable from ΛCDM,
except a small bump for the smallest halos around R ≈ Rvir . On the other hand, for the
more extreme cases of |fR0 − 1| = 10−4 and |fR0 − 1| = 10−5 the relative deviation is much
larger, especially for small radii. In most of the parameters the results show a stronger
gravitational redshift, which is the result of the increased clustering in the center of the
halo due to the fifth force: Stronger clustering in the center means a deeper gravitational
potential well and, hence, a larger gravitational redshift.

The strength of this effect changes with the cluster mass, e.g. for the fofr5 data set the
enhancement at R/Rvir = 0.1 is around 18% for the smallest halo mass range (top left
panel of Fig. 5.6), then increases to 45% for bigger halo masses (top right panel in the
same figure) just to drop then dramatically for even bigger halos. For the model with
|fR0 − 1| = 10−4 we observe a converse feature at the same radius: For the biggest halos
there is a strong enhancement of the gravitational redshift whereas for the smaller halos
the redshift observed is even smaller than in the ΛCDM case. These trends can be best
seen in Fig. 5.7 where for a fixed radial the relative deviation of the gravitational redshift
is shown. In these slices, the bump of deviation for the fofr6 model is much clearer. Due
to its position between 1012 M⊙ h−1 and 1013 M⊙ h−1 , it did not appear as clearly in the
mass bins of Fig. 5.6. One very interesting feature of Fig. 5.7 is the characteristic position
of the maximum deviation from ΛCDM depending on the fR0 value: In the fofr6 data the
maximum deviation is for halos with masses around 2 × 1012 M⊙ h−1 , for |fR0 − 1| = 10−5 it
is around 3 × 1013 M⊙ h−1 whereas |fR0 − 1| = 10−4 results in a change of the biggest halos in
the data set. This maximum is consistent throughout all mass bins, but as seen in Fig. 5.6
it varies in strength.
80 5 Gravitational Redshift in Screened Modified Gravity

fofr4 fofr5 fofr6

0.7
0.2 1012.5-1013.0 M⊙h-1 0.6 1013.0-1013.5 M⊙h-1
0.5
0.1 0.4
0.3
0 0.2
0.1
∆vg/vg,ΛCDM

∆vg/vg,ΛCDM
-0.1 0

0.3 1013.5-1014.0 M⊙h-1 0.5 1014.0-1014.5 M⊙h-1

0.4
0.2 0.3
0.2
0.1
0.1

0 0

0.01 0.1 1 0.01 0.1 1

R/Rvir R/Rvir

Figure 5.6: Relative deviation in gravitational redshift between the f (R) runs and the
ΛCDM run. Here the red circles mark the fofr4, the blue triangles the fofr5 and the green
diamonds the fofr6 run. The four panels show the result obtained for different halo masses
ranging from 1012.5 -1013 M⊙ h−1 (upper left) to 1014 -1014.5 M⊙ h−1 (lower right panel).

As discussed in the previous section, the reason for the existence of such a characteristic
range of halo sizes lies in the nature of the screening mechanism: Halos heavier than a
certain threshold have a big enough density to activate the screening mechanism to kick
in. Hence, the value of the scalar field is very small, there is no fifth force and therefore
no additional clustering in the center. Only halos within this mass range have an altered
structure and as a result, we obtain the characteristic bumps in Fig. 5.7. The position of
the maxima supports the explanation given: The closer the fR0 value to unity, the smaller
are the affected halos. Naturally, these mass ranges correspond to the ones mentioned in the
last section (compare to the upper panels of Fig. 5.5).

The position and magnitude of the characteristic peak in the ẍFifth versus R plots (lower
panels of Fig. 5.5) plays also an important role when interpreting the gravitational redshift
results. If the position of the peak is close to the center of the halo, the fifth force will lead to
a deeper potential well and an enhanced gravitational redshift. The stronger the fifth force
is in this regions, the bigger the enhancement. Comparing the findings of the last section
with the gravitational redshift results, we find:

• In the fofr4 case, the relative deviation is biggest for the most massive halos consid-
ered. This agrees with the strong additional force found in the central regions of these
clusters. Also for halos with mass 1013.5 − 1014 M⊙ h−1 the deviation is overall greater
5.5 The f (R) models 81

fofr4 fofr5 fofr6

0.8 0.049 <R/Rv< 0.051 0.09 <R/Rv< 0.11

0.6
0.6
0.4
0.4
0.2 0.2

0 0
∆vg/vg,ΛCDM

∆vg/vg,ΛCDM
-0.2
-0.2
0.4 0.49 <R/Rv< 0.51 0.4 0.99 <R/Rv< 1.01

0.2 0.2

0 0

-0.2 -0.2
1012 1013 1014 1012 1013 1014
Halo mass in M⊙/h Halo mass in M⊙/h

Figure 5.7: Relative deviation in gravitational redshift of the f (R) models with respect to
ΛCDM plotted against the halo mass for fixed radial bins (R/Rv ≈ (0.05, 0.1, 0.5,1) from left
top to right bottom). Here the red circles mark the fofr4, the blue triangles the fofr5 and the
green diamonds the fofr6 run.

than in the other cases, just as the fifth force was strongest. A strange feature is the
negative deviation found for the smallest masses. This is most probable and an artifact
due to the resolution limit but could also be accounted on the fact that the strongly
increased clustering leads to more small halos being located closer to bigger halos and
therefore in higher density regions.

• The fofr5 data shows also the most deviation when the fifth force was most active,
i.e. for the 1013.5 − 1014 M⊙ h−1 mass range. The fact that for the biggest halos the
screening mechanism suppresses the fifth force in the central regions and, consequently,
the additional clustering takes place around the virial radius, results in a negative
deviation of vg . Although we account the decline of the fifth force in the central
regions of the halos with mass 1013 − 1013.5 M⊙ h−1 also on screening, the position of
the characteristic peak in the ẍFifth versus R plot around 0.1Rv is here much closer to
the center. Thus, the deviation in vg for these halos is positive.

• Since ẍFifth takes very small values for |fR0 − 1| = 10−6 (fofr6 ), the results in gravita-
tional redshift for masses & 1012.5 M⊙ h−1 are basically indistinguishable from ΛCDM.
The characteristic mass range for the fofr6 model is around ∼ 1012.2 M⊙ h−1 (see
82 5 Gravitational Redshift in Screened Modified Gravity

Fig. 5.7). This is limiting the simulation’s resolution limit. Therefore, we cannot
analyze the radial dependency of the redshift deviation in detail.

5.5.4 The halo density profiles

Because the underlying cause for the change of gravitational redshift in the models considered
is the change of the halo density profiles, we have measured these directly in two ways:
Firstly, the matter density ρ obtained at a particle’s position through an inverse cloud-in-
cell algorithm similarly to the gravitational potential studied in the last section. The ratio
of ∆ρ/ρΛCDM is presented in Fig. 5.8. And secondly, the mean number of particles per
halo in a given radial range from the center of its halo. The comparison of this quantity is
shown in Fig. 5.9. Again, both quantities were binned in normalized radii and halo masses
and compared to the ΛCDM values in the same way as used for the gravitational redshift
before.
Both comparisons show the same features as seen previously in the gravitational redshift
(Fig. 5.6). One important difference is, however, the quicker fading away of a deviation:
E.g., the deviation in redshift for halos with masses 1013 − 1013.5 M⊙ h−1 in the fofr5 -model
is clearly visible even for radii R & Rvir (see Fig. 5.6), whereas the deviation in N is already
below 10% at R ≈ 0.02 (see Fig. 5.9). We can conclude, that the effect of gravitational
redshift is highly dependent on the cluster density at the center of the halo. This leaves a

fofr4 fofr5 fofr6

1012.5-1013.0 M⊙h-1 1.4 1013.0-1013.5 M⊙h-1

0.2 1.2
1
0.8
0.6
0 0.4
0.2
∆ρ/ρΛCDM

∆ρ/ρΛCDM

0
13.5 14.0 -1 14.0 14.5 -1
10 -10 M⊙h 10 -10 M⊙h
0.4 0.4

0.2 0.2

0 0

-0.2 -0.2
0.01 0.1 1 0.01 0.1 1
R/Rvir R/Rvir

Figure 5.8: Relative deviation in the matter density ρ plotted against the normalized radius
for the same mass bins as in Fig. 5.6.
5.5 The f (R) models 83

fofr4 fofr5 fofr6

0.4 1.5
1012.5-1013.0 M⊙h-1 1013.0-1013.5 M⊙h-1
0.2 1

0 0.5
-0.2
0
-0.4
-0.5
∆N/NΛCDM

∆N/NΛCDM
-0.6
0.6 1013.5-1014.0 M⊙h-1 0.8 1014.0-1014.5 M⊙h-1
0.6
0.4
0.4
0.2
0.2
0 0
-0.2 -0.2
-0.4
-0.4
-0.6
0.01 0.1 1 10 0.01 0.1 1 10
R/Rvir R/Rvir

Figure 5.9: Relative difference between the mean number of particles found per halo in a
given distance from the center of the halo. The results are presented in the same halo mass
ranges as before, ranging from 1012.5 M⊙ h−1 in the upper-left panel to 1014.5 M⊙ h−1 in the
lower-right one.

variety of possible implications, such as a different location of the densest region, which will
be discusses in Sec. 5.8.

5.5.5 Velocity profiles

An important reason for analyzing the velocity dispersion of halos is to test if the increased
matter density is not due to slower particles at that point. However, since there is an
additional force accelerating the particles, that finding would be very unexpected.
The velocity dispersion as defined in Eq. (4.60) and its deviation from the ΛCDM values is
shown in Fig. 5.10. As expected the dependency of the additional particle speed with the
fifth force (Fig. 5.5) is clearly visible: The fofr6 data only shows for the smallest halo mass
bin some deviation in the region ∼ Rv , where there is also the bump in the fifth acceleration.
The particles are faster in the fofr5 than in the fofr4 data for halo masses . 1013.5 M⊙ h−1 ,
just as the fifth force is higher for the latter. If there is no additional force at a specific
radius, σ 2 goes towards the ΛCDM value rather quick. So does the fofr5 data set for radii
R . 0.1Rv . These results show that the increased particle density is not due to slower
particles but due to additional cluster, as expected.
84 5 Gravitational Redshift in Screened Modified Gravity

0.35 0.8

0.3 0.6
σ2 (106 km2/s2)

0.25

∆σ2/σ2ΛCDM
0.4
0.2
0.2
0.15

0.1 0

0.05 -0.2
0.01 0.1 1 10 0.01 0.1 1
R/Rv R/Rv

0.7 0.8
0.6
0.6
σ2 (106 km2/s2)

0.5
∆σ2/σ2ΛCDM

0.4 0.4

0.3 0.2
0.2
0
0.1
0 -0.2
0.01 0.1 1 10 0.01 0.1 1
R/Rv R/Rv

1.6 0.8
1.4
0.6
σ2 (106 km2/s2)

1.2
∆σ2/σ2ΛCDM

1 0.4

0.8 0.2
0.6
0
0.4
0.2 -0.2
0.01 0.1 1 10 0.01 0.1 1
R/Rv R/Rv

Figure 5.10: Velocity dispersion profiles (left) and relative deviation of velocity disper-
sion (right). Three halos mass bins are shown 13 < log(M h/M⊙ ) < 13.5 (top), 13.5 <
log(M h/M⊙ ) < 14 (top) and 14 < log(M h/M⊙ ) < 14.5 (bottom). The color code is the
same as for the other f (R) plots, i.e. fofr4 in red, fofr5 in blue and fofr6 in green with the
additional data points of ΛCDM in black.
5.6 The symmetron models 85

5.6 The symmetron models

For the analysis of the symmetron results, we follow the same structure as in the last section:
First, the matter power spectra, halo mass functions and fifth force data are shown. Then,
the gravitational redshift results are presented. And lastly, the quantities that can help
explaining these results, that is the matter density and the velocity dispersion are given.
Also here, the radii were normalized with the virial radius Rv = R200 and only halos with
|βvir | < 0.2 were considered

5.6.1 Matter power spectra and halo mass functions

The same way as in the f (R) section, the deviation in the matter power spectra and the halo
mass function shown in Fig. 5.11 is given here merely out of completeness. However, the
key points made in Brax et al. (2012) can be confirmed: Decreasing the symmetry breaking
time assb (symm_A versus symm_B versus symm_D) leaves the fifth force more time to be
active and hence increases the effect on the matter power spectra and halo mass function. A
similar effect is achieved by increasing the coupling constant β (symm_A versus symm_C )
which controls the strength of the fifth force directly.

One striking difference between the symm_D results and the other models is the much
higher number of medium and big halos (see right panel of Fig. 5.11). This increase cannot
be explained in merely the slight mass increase of also in the ΛCDM data set existing halos.
Instead, the creation of fully new, big halos mainly through clustering of many small or
medium sized halos takes place. This phenomenon is also described by Li & Hu (2011) for
their no-chameleon model.

0.4 0.4
symmA
symmB
symmC
0.3 symmD
∆N/NΛCDM
∆P/PΛCDM

0.2 0.2

0.1

0
0

1012 1013 1014

0.01 0.1 1 10 Halo mass in M⊙/h
k (h/Mpc)

Figure 5.11: Relative deviation in the matter power spectra (left panel) and the halo mass
function (right panel). The error bars indicate the spread of values in each bin.
86 5 Gravitational Redshift in Screened Modified Gravity

35 9
0.99 <R/Rv< 1.01
0.09 <R/Rv< 0.11 8
30 symmA
symmB 7
xFifth (10-12 m2/s2)

xFifth (10-12 m2/s2)

25 symmC
symmD 6
20 5
15 4
3
10
..

..
2
5 1
0 0
1012 1013 1014 1012 1013 1014
Halo mass in M⊙/h Halo mass in M⊙/h
30

30

16

8
1012.5-1013.0 M⊙h-1 1013.0-1013.5 M⊙h-1 1013.5-1014.0 M⊙h-1 1014.0-1014.5 M⊙h-1
xFifth (10 m /s )
-12 2 2
15

15

4
..
0

0
0.01 0.1 1 10 0.01 0.1 1 10 0.01 0.1 1 10 0.01 0.1 1 10
R/Rv R/Rv R/Rv R/Rv

Figure 5.12: Additional acceleration ẍFifth due to the presence of a scalar field in the sym-
metron model. Lower panel: |ẍFifth | as a function of normalized radius for the halo mass
ranges (log(Mlower h/M⊙ ), log(Mupper h/M⊙ )) = (12.5, 13), (13, 13.5), (13.5, 14) and (14, 14.5)
(from left to right). Upper panel: |ẍFifth | as a function of halo mass for in the radial bins
R ≈ 0.1Rv (left) and R ≈ Rv (right). In all the graphs the red circles mark the symm_A,
the blue triangles the symm_B, the green diamonds the symm_C and the orange circles the
symm_D run. The black horizontal lines correspond approximately to two grid cells in the
finest refinement level.

5.6.2 Symmetron additional force

Fig. 5.12 shows the additional acceleration ẍFifth versus the halo mass (upper row) and the
normalized halo radius (lower row). The overall finding is that the symm_D parameter set
results in the strongest fifth force, followed by symm_C and symm_B. The symm_A curve
shows the least amount of additional acceleration. This confirms the expected relationships:
(i) The earlier the symmetry breaking, the stronger the fifth force and (ii) an increase of
the coupling constant also leads to an increased fifth force.
The upper row of Fig. 5.12 shows that in addition to the magnitude of ẍFifth the model
parameters alter also the mass range of the affected halos. So has the symm_D the strongest
impact for halos ∼ 1013 M⊙ h−1 and is only fully screened in the center of halos with masses
& 1014.3 M⊙ h−1 (upper left panel). The later the symmetry breaking time, the smaller are
these two characteristic factors; e.g., the mentioned peak is for symm_B already around
1012.5 M⊙ h−1 and the fifth force in the central region becomes neglectable at masses &
1013.3 M⊙ h−1 . In the symm_A case these values are even more shifted to lower masses.
In order to study the effect of an increased coupling parameter on the affected halo masses,
5.6 The symmetron models 87

the symm_A data (β = 1) has to be compared with the symm_C results (β = 2). Here,
it can be seen that the increase of β does not shift the mass range. So is the mentioned
maximum of the force for both data sets slightly bigger than 1012 M⊙ h−1 at R = Rv (top
right panel). However, as mentioned above, increasing β increased the magnitude of the fifth
force. This leads to the feature that around Rv and for halo masses ∼ 1012 M⊙ h−1 symm_C
results in even a greater fifth force than symm_D.

The lower row of Fig. 5.12 can be analyzed in a similar way to the radial dependency of
the fifth force Sec. 5.5.2. This means, the decline of the additional acceleration for bigger
radii is due to the less changing matter density in the outskirts. Towards the central region,
however, the decline can be also a result of the screening mechanism. This happens if the
density there is above a certain effective threshold density, which for the symmetron model
is equal to ρssb , the average matter density at assb . The criteria established in Sec. 5.5.2 for
the differentiation between these two possibilities were (a) the proximity between the radius
of maximal fifth acceleration and the resolution limit and (b) if ẍFifth is greater for more
massive halos in the central region.

For the symmetron model, we came to the following conclusions about the presence of the
screening mechanism:

• symm_D: The fifth force in the central region is strongest for halos with mass ∼ 1013 M⊙ h−1
(see top left panel of Fig. 5.12). Hence, the smallest mass range analyzed is the only
one where the screening mechanism is not active. This finding is supported by the fact
that for this mass range the radial peak is very close to the resolution limit. However,
for the next bigger halos, this peak is still very close to the central region (∼ 0.1Rv ).
For even bigger halos, this radius of maximal fifth force moves further out until it
reaches ∼ Rv for the biggest halos considered.

• symm_A and symm_C : ẍFifth takes much smaller values than in the symm_D case.
With the halo mass of the maximal affected central regions being ∼ 1012 M⊙ h−1 , the
screening mechanism is active for all the four halo mass ranges.

• symm_B: Since the described peak is around ∼ 1012.8 M⊙ h−1 , for the last three halo
ranges the screening is active in the central region. Also here it is noteworthy that the
magnitude of the additional acceleration is very weak.

Conclusively, this means, while the overall behavior is similar to the chameleon mechanism,
i.e. no fifth force in the center, a force maximum at intermediate radii and a downfall of the
fifth force in outer radii, there are significant differences: In both models, there is a certain
“threshold mass” for each radial bin above that the force is fully screened, as mentioned
before. The results are indicating, while this threshold is very abrupt for the chosen f (R)
parameters, it is more gradiently in the symmetron model. This said, however, only a
small area of the parameter space was probed and the symmetron model can have a similar
behavior implying the right choice of parameters.
88 5 Gravitational Redshift in Screened Modified Gravity

symmA symmB symmC symmD

0.3
1012.5-1013.0 M⊙h-1 1013.0-1013.5 M⊙h-1
0.2 0.1

0.1

0 0
∆vg/vg,ΛCDM

∆vg/vg,ΛCDM
1013.5-1014.0 M⊙h-1 0.1 1014.0-1014.5 M⊙h-1

0
0

-0.1
-0.1

-0.2
0.01 0.1 1 0.01 0.1 1
R/Rvir R/Rvir

Figure 5.13: Relative deviation between the symmetron runs and the ΛCDM run. The red
circles mark the symm_A and the blue triangles the symm_B run. Please note the change in
scale compared to Fig. 5.6

5.6.3 Gravitational redshift profiles

Fig. 5.13 and Fig. 5.14 show the deviation in gravitational redshift with respect to ΛCDM
versus the radius and the halo mass, respectively. Given the overall weaker additional
forces discussed in the last section, it is not surprising the imprint in the variation of the
gravitational redshift profiles is not as strong as it was the case for the f (R) results. The
maximal deviation is around 20% for the symm_B model for halos with masses between
1012.5 and 1013 M⊙ h−1 (top left panel in Fig. 5.13). As expected, especially the symm_A
model with a very late symmetry breaking time is basically indistinguishable from ΛCDM.

One prominent feature that did not occur the same way in the f (R) case is the negative
deviation obtained for high (& 3 × 1013 M⊙ h−1 ) and low (. 3 × 1012 M⊙ h−1 ) masses in the
symm_D model. A similar but fainter imprint can be seen for the symm_B run with the
difference that for intermediate masses (∼ 8 × 1012 M⊙ h−1 ) the positive deviation is much
stronger than in the symm_D case. This fact paired with the very small radial dependency of
the deviation for the bigger symm_D-halos denotes a less deep but wider gravitational well,
i.e. less matter in the central parts and more in the outskirts of the clusters. This weaker
gravitational redshift with respect to ΛCDM is an unexpected result since the additional
force should lead to an enhanced clustering and, thus, a stronger gravitational redshift.
This maybe unexpected feature can be explained with the radial position of the additional
clustering as discussed in the last section. If the central region is screened and the fifth force
5.6 The symmetron models 89

symmA symmB symmC symmD

0.4
0.4 0.049 <R/Rv< 0.051 0.09 <R/Rv< 0.11

0.2 0.2
0
-0.2 0

-0.4
-0.2
-0.6
∆vg/vg,ΛCDM

∆vg/vg,ΛCDM
-0.8
0.49 <R/Rv< 0.51 0.2 0.99 <R/Rv< 1.01
0.2

0
0

-0.2 -0.2

1012 1013 1014 1012 1013 1014

Halo mass in M⊙/h Halo mass in M⊙/h

Figure 5.14: Relative deviation in gravitational redshift of the symmetron models against
ΛCDM plotted against the halo mass for a fixed radial bin. Here, the red circles mark the
symm_A, the blue triangles the symm_B, the green diamonds the symm_C and the orange
circles the symm_D run.

is strongest for larger radii, the additional clustering takes place in the outskirts of the halo
leading to a less deep and wider potential well. Such a shape of the potential well results in
a negative deviation in vg .

5.6.4 Halo density profiles

In Fig. 5.15 the deviation in matter density ρ for the symmetron model is shown. The
trends seen in the gravitational redshift profiles can be rediscovered in this plot: Hardly any
deviation from ΛCDM for symm_A, then more for model “B” followed by “C”, just as the
strength of the fifth force seen in Fig. 5.12.

The presumption for the reason of the constant lower redshift for the bigger halos of the
symm_D model can be confirmed: While the matter density is lower at smaller radii than
in the ΛCDM sample, with bigger radius this value increases by up to twenty percentage
points. Just the opposite case happens for the smaller halos: The deviation in the matter
density seems to be constantly increased around 15 − 20% leaving the characteristic drop in
the gravitational redshift seen in the upper panels of Fig. 5.13.
90 5 Gravitational Redshift in Screened Modified Gravity

symmA symmB symmC symmD

0.4 1012.5-1013.0 M⊙h-1 1013.0-1013.5 M⊙h-1

0.2
0.2

0
0
∆ρ/ρΛCDM

∆ρ/ρΛCDM
0.4 1013.5-1014.0 M⊙h-1 0.4 1014.0-1014.5 M⊙h-1

0.2
0.2

0
0
-0.2
0.01 0.1 1 0.01 0.1 1
R/Rvir R/Rvir

Figure 5.15: Relative deviation in the matter density ρ between the symmetron and the
ΛCDM models. The results are presented in the same halo mass bins as before, ranging from
1012.5 M⊙ h−1 in the upper-left panel to 1014.5 M⊙ h−1 in the lower-right one.

5.6.5 Velocity profiles

In Fig. 5.16 the velocity dispersion profiles and their deviation from ΛCDM for the sym-
metron runs is shown. As expected, the extra speed of the particles is highly dependent on
the strength of the fifth force in that region (compare with Fig. 5.12): The fastest particles
are to be found in the “D” model, slowest in symm_A. In all plots, the deviation curve
consists of a nearly constant part up to ∼ 0.1 Rv and then a linear increase. The value of
the constant part depends on the value of the additional acceleration for small radii: E.g.,
the fifth force was already fully screened in the symm_B case for masses between 1013 and
1013.5 M⊙ h−1 around 0.1 × Rv while in the symm_C case it was not. This results in a
different constant value of the ∆σ 2 /σΛCDM
2 curve.
An important point to note is, that the negative deviation in mass density in the symm_D
model is not due to a higher particle velocity since for the biggest masses all the velocities
below 0.1 Rv are comparable.
5.6 The symmetron models 91

0.35 1.4
1.2
0.3
1
σ2 (106 km2/s2)

0.25

∆σ2/σ2ΛCDM
0.8
0.2 0.6
0.4
0.15
0.2
0.1
0
0.05 -0.2
0.01 0.1 1 10 0.01 0.1 1
R/Rv R/Rv

0.6 1
0.55
0.5 0.8
0.45
σ2 (106 km2/s2)

0.6
∆σ2/σ2ΛCDM

0.4
0.35
0.4
0.3
0.25 0.2
0.2
0.15 0
0.1
0.05 -0.2
0.01 0.1 1 10 0.01 0.1 1
R/Rv R/Rv

1.1 0.6
1
0.9 0.4
σ2 (106 km2/s2)

0.8
∆σ2/σ2ΛCDM

0.7
0.2
0.6
0.5
0.4 0
0.3
0.2 -0.2
0.01 0.1 1 10 0.01 0.1 1
R/Rv R/Rv

Figure 5.16: Velocity dispersion profiles (left) and relative deviation of velocity disper-
sion (right). Three halos mass bins are shown 13 < log(M h/M⊙ ) < 13.5 (top), 13.5 <
log(M h/M⊙ ) < 14 (center) and 14 < log(M h/M⊙ ) < 14.5 (bottom). The red circles mark
the symm_A, the blue triangles the symm_B, the green diamonds the symm_C, the orange
circles the symm_D and the black points the ΛCDM run.
92 5 Gravitational Redshift in Screened Modified Gravity

5.7 Connection to observational data

Measuring the gravitational redshift directly is very hard as it is only possible to measure
the total redshift of an object. The main contributions of this total redshift ztot of an galaxy
which is dlos away from the observer are given by three components:

cztot = H(z)dlos + vpec + vg . (5.13)

In this equation the gravitational redshift vg is about two orders of magnitude smaller than
the Doppler shift due to the peculiar motions of the galaxies in the halo vpec . Nevertheless,
the two components can be separated because the mean of the gravitational redshift has a
radial dependency while the other one has not.
The first detection of gravitational redshift in clusters was done by Wojtak et al. (2011)
using 7800 clusters from the SDSS data set. Several corrections to their predictions have
been made later on: Zhao et al. (2012) pointed out that due to the relative motion of
the galaxies, an additional component because of time-dilation exists. This term is called
transverse Doppler effect and has opposite sign as vg . This and other effects, e.g., a changed
redshift as a result of relativistic beaming, have been analyzed by Kaiser (2013), all of them
should be considered when doing a proper analysis of observational data. As the scope of
this work is, however, the deviation of vg in screened modified gravity models with respect to
ΛCDM, these additional effects have not been studied and have not been included in Fig. 5.17
where our predictions are compared with the data points from Wojtak et al. (2011). The

0
fofr/fofr4
symmetron/symmD
-2 ΛCDM
Wojtak et. al.

-4

-6
∆vg (km s-1)

-8

-10

-12

-14

1 2 3 4 5 6
Projected distance in Mpc

Figure 5.17: Comparison with the data points from Wojtak et al. (2011) of fofr4 and
symm_D.
5.7 Connection to observational data 93

Number of galaxies
Name LRGs ELGs Time span Reference
SDSS/BOSS 1.5 × 106 − 2000 − 2014 Aihara et al. (2011)
BigBOSS 3.5 × 106 15.5 × 106 2017 − 2022 Schlegel et al. (2009)
Euclid − 50 × 106 2020 − 2027 Laureijs et al. (2011)

Table 5.1: Listing of the expected number of spectroscopically measuredluminous red galax-
ies (LRGs) and emission line galaxies (ELGs) in the full SDSS/BOSS, BigBOSS and Euclid
surveys.

main point of this comparison (where even the most extreme models are compatible with
the data), is to illustrate by how much the error bars have to be reduced in order to give
viable constraints on screened modified gravity models. Croft (2013) shows error predictions
for the full SDSS/BOSS (Aihara et al., 2011), BigBOSS (Schlegel et al., 2009) and Euclid
(Laureijs et al., 2011) data sets (see Table 5.2). Accordingly, BigBOSS and Euclid should
be able to map out the amplitude of the vg curve with 6.5% and 4% precision, respectively.
These values were obtained using a different technique than our approach, that is instead of
taking the difference between the gravitational redshift of the central galaxy (BCG) and each
particle in a cluster, the data was compared pairwise. Hence, they cannot be transferred
directly. In addition, the binning of data in halo mass, which is necessary to detect signatures
due to screened gravity, will naturally increase the error. Nevertheless, the planned new sky
surveys will allow a restriction of the parameter space of some modified gravity models.
For the f (R) Hu-Sawicky model the parameter space is already constrained beyond the
parameters used in this work (see Table 5.2, left) and stronger limits obtained through
gravitational redshift data are questionable. However, as the dark energy mystery is far from
solved, it is important to test general relativity on kpc to Mpc scales. Gravitational redshift
in combination with the upcoming sky surveys like Euclid are a potent tool to do that. For
example, the symmetron model, which was also studied in this work, is hardly constrained
yet (see Table 5.2, right) since it has a more efficient screening mechanism. Consequently, in
order to constrain the parameter space of the symmetron model, the probing of cosmological
scales is necessary.

Method |fR0 − 1| . Reference Parameter Citation

Solar system 2 × 10−6 Hu et. al (2007) −3/2
Laassb . 10−3 H0 Hinterbichler et al. (2010)
Strong lensing 2.5 × 10−6 Smith (2009) L ≫ 1 mm Upadhye (2013)
Cepheids 5 × 10−7 Jain et al. (2012) β.1 -1

Table 5.2: Left: Strongest existing constraints on the f (R) Hu-Sawicky model. Shown are
the upper bounds for |fR0 − 1| with n = 1. Right: Order of magnitude “constraints” for the
symmetron model.

1 A coupling varying greatly from unity will lead to no detectable effects at the lower end or too strong
imprints in the mass power spectrum at the higher end.
94 5 Gravitational Redshift in Screened Modified Gravity

0.9 0.9
10 10

0.8 0.8
10 10

0.7 0.7
10 10
cv

cv
0.6 0.6
10 10

0.5 0.5
10 10

0.4 0.4
10 10
11 12 13 14 15 11 12 13 14 15
10 10 10 10 10 10 10 10 10 10
−1 −1
Halo mass in M☉ h Halo mass in M☉ h

Figure 5.18: Mean of concentration value cv versus halo mass with data points from John-
ston & Sheldon (2007) (black points). In both plots, the shaded region shows the standard
deviation and the black line the ΛCDM value. Left panel: fofr4 –6 (red, blue and green lines).
Right panel: symm_A–D (red, blue, green and orange lines).

Of course, there are also other cluster properties that can be observed and used for testing
modified gravity models. Llinares & Mota (2013b) used for example the ellipticity of clusters
to constrain the symmetron and chameleon parameter space. With this observable they
managed to exclude high β regions. Another commonly measured value is the concentration
value of halos cv which describes the breaking point of the radial density profile. In Fig. 5.18
this concentration value obtained by the Rockstar halo finder (Behroozi et al., 2013) through
a NFW-fit on the density is compared against the data points from Johnston & Sheldon
(2007) who used combined SDSS data of 130000 groups and clusters for the density fits. Also
here, more precise measurements are necessary to be able to differ between the considered
models or put additional constraints on the parameter space. A more detailed analysis of the
change in concentration value in screened modified gravity models is left for future studies.

5.8 Systematic effects

The cluster density and redshift profiles obtained through simulations are affected by the
following systematic effects:

(i) Baryons: Including baryonic effects inside clusters can lead to a flattening of the cluster
density profile near the center. This is due to the movement of gas from the center
to larger radii through active galactic nucleus feedback (Martizzi et al., 2011; Teyssier
et al., 2010).

(ii) Halo definition: The definition of halo made in Sec. 3.4 was mainly aimed in getting
the difference in gravitational potential between the brightest cluster galaxy (BCG)
and the rest of the cluster and so imitating an observational standpoint. This is also
why nearby subhalos where included in the definition of a halo. This can, however, be
a problem as the minimum of the gravitational potential is not necessarily the densest
central region. In order to verify the findings, the results were also reproduced by
5.9 Summary of results 95

taking the halo definition from the Rockstar halo finder and only minor deviations
inside the virial radius were found.
(iii) Halo finding process: As discussed in Sec. 3.4 and Sec. 4.2, the definition of the virial
radius depends on the mass of the halo in screened modified gravity. Therefore, the
ΛCDM R200 results were compared against the R160 results of the most extreme (fofr4,
symm_D) runs. The results were only slightly affected.
(iv) Renormalization of the mass power spectra: In order to be in agreement with other
constraints the deviation from the ΛCDM mass power spectrum should only be on
small scales. This was given for the set of parameters used in this work but for further
exploration of the given parameter space the rescaling of σ8 has to be taken into
account.
(v) Simulation systematics: Llinares & Mota (2013a) showed that not taking the quasi-
static approximation can result in different scalar field values and consequently also
in different additional forces. The impact of this approximation is part of future
studies. Against numerical problems, the implementation of the ISIS code was tested
thoroughly in Llinares et al. (2013).
The analysis was performed assuming that these systematic effects are neglectable or beyond
the scope of this work. A more detailed analysis would have to take these effects into
account.

5.9 Summary of results

We performed a set of 8 high resolution simulation runs with 5123 particles and a boxlength
of 256 Mpc h−1 each. The data was used to analyze the density and gravitational redshift
profiles of virialized clusters. For this purpose, first, the virialization parameter was com-
puted for each halo with the inclusion of the scalar field energy. Furthermore, we calculated
the gravitational redshift of all particles the following way: The value of the gravitational
potential at the particle’s position was subtracted from the halo’s minimal gravitational po-
tential. Ultimately, the modified gravity results could be compared to the ΛCDM values.
These steps were not only undertaken for the gravitational redshift but also for the matter
density, particle density, fifth force and velocity dispersion. These quantities allowed us to
analyze and reason the gravitational redshift results. The overall finding from the numeri-
cal results is that in both analyzed models the deviation from ΛCDM can vary enormously
depending on the choice of parameters and the analyzed halo size.

f (R) results

The f (R) results confirmed that lowering |fR0 − 1| reduces the deviation from ΛCDM. In the
data sets we found an enhanced clustering mainly in the central regions of the halos. This
causes an increase in gravitational redshift with respect to the ΛCDM results. The intensity
96 5 Gravitational Redshift in Screened Modified Gravity

of this additional clustering – and, thus, the deviation in gravitational redshift – was shown
to be highly dependent on the halo mass. In addition, a relationship between the size of the
maximal affected clusters and the fR0 parameter could be established: The lower the value of
|fR0 − 1|, the smaller are the affected halos. This is due to the chameleon nature of the fifth
force: Very small halos are deep inside the “thick shell regime” and, therefore, the minimum
of the effective potential is hardly varied. Hence, only a small fifth force arises. On the other
hand, very big halos possess a “thin shell”, that is the fifth force is fully self-screened within
the halo. The characteristic halo mass range is between these two regimes and marks when
the fifth force is strongest.

Symmetron results

In three of the four symmetron runs made assb was changed. This parameter controls the
time of symmetry breaking and the strength of the fifth force. In one run the coupling
constant β was varied instead.
As expected a lower value of assb or a higher value of β results in a generally greater deviation
from the ΛCDM cluster profiles. The effect was in some cases contrary to the expectation:
Instead of stronger clustering in the center of the halo, we found effective negative clustering
leading to a negative deviation in gravitational redshift. This shifted clustering can be
understood by analyzing the dependency of the fifth force versus the halo mass and the
distance from the center of the cluster by keeping in mind that the analysis is made withing
a certain cluster mass bin. The simple explanation attempt made is based on the idea that
through the possible screening of the fifth force in the central regions a de-facto preferred
mass density exists. This means, once the mass density inside the center is higher the
clustering takes place in the outer regions.
There are indications of the same effect in other data sets analyzed during this work. How-
ever, this would have to be further investigated using a wider parameter range and simulation
snapshots at higher redshifts allowing the analysis of the cluster formation process.

Highlights and comparison

All of the analyzed models show an increased clustering rate which can most easily be ob-
served in the matter power spectra (Fig. 5.4 and Fig. 5.11). This effect due to the additional
force is not surprising and was shown already using N -body simulations in previous work
(Brax et al., 2012, 2013; Li et al., 2012).
The most common expectation about the change of the gravitational redshift through this
additional clustering is to observe an enhanced signal. This idea is motivated by the fact
that additional clustering should lead to deeper potential wells and, hence, to a stronger
gravitational redshift signal. This is the major imprint observed in the f (R) models studied
and the same behavior was expected in the symmetron models. However, depending on the
halo mass and model parameters, the matter density can be enhanced in the outskirts of
the halo. In such a way altered halo density profiles in modified gravity N -body simulations
5.9 Summary of results 97

were already observed by Baldi (2011). For some parameters they obtained an increased
clustering in the center of the halo, while for others this additional clustering took place at
larger radii. Keeping in mind the halo mass is effectively fixed in one mass bin, this result
is not surprising as missing mass in one region has to be compensated by additional mass
in another region. Such an altered halo density profile means the potential well becomes
shallower in the central regions and a deeper in the outskirts. Consequently, the resulting
deviation in gravitational redshift is negative.
However, this connection between clustering in the central regions and stronger gravitational
redshift or clustering in the outer regions and a negative deviation does not explain the
contrary imprint obtained for various parameter sets. Therefore, the question we tried to
answer is, what mechanism controls if the clustering happens in the central region or not?
The explanation provided is based on the nature of the screening mechanism: If the fifth
force is screened in the central regions of a halo but not in the outskirts, then this is where
the additional clustering will take place. Another way of phrasing it is that each set of model
parameters results in a “preferred density”, that is a matter density for which the fifth force
is just screened. This reasoning can most easily be tested by observing the dependency of
the fifth force (see Fig. 5.12) and the deviation of the gravitational redshift (Fig. 5.13) for the
symm_D model since the screening is active in the central region. However, the explanation
is supported by several additional features in the results: By lowering assb the effect should
still persist in a weaker form – which it does in the symm_B data set. Here, clustering
the center is effectively weakened for clusters with mass & 1013.5 M⊙ h−1 . Consequently,
gravitational redshift is decreased in these halos, too. For even smaller values of assb , the
additional force is too small and the existence of the same effect cannot be confirmed. A
special position is taken by the fofr4 data: The increase of the number of massive clusters is
immense (Fig. 5.4) suggesting the same or even stronger merging of clusters. On the other
hand, the force enhancement close to the center has not yet reached its maximum, even for
the largest halos of the data sample (top left panel of Fig. 5.5) opposed to the symm_D case,
where this maximum lies around 1013 M⊙ h−1 for R ≈ 0.1 Rv (Fig. 5.12). This means, in the
fofr4 the “effective preferred density” is even higher than the one present in the central parts
of the biggest clusters. Therefore, the clustering is dominant in the central parts, leading
to a higher gravitational redshift as seen in Fig. 5.6. It is more appropriate to compare the
symm_D with the fofr5 results due to their bigger similarity in the fifth force profiles. The
fifth acceleration in the central region is, however, more present for halos in the mass range
1013.5 − 1014 M⊙ h−1 in the fofr5 data. For the next bigger halos, ẍFifth is neglectable in
the central regions in both models. This could explain the resulting positive deviation in
gravitational redshift for the first halo mass bin and the negative deviation in the second
(Fig. 5.6). Of course, the different other characteristics of the symmetron and the chameleon
model (e.g. the time dependence of the fifth force) have to be taken into account if the data
is to be analyzed in more detail. This explanation provided can, after all, clarify the partial
contrary results obtained.
6

Conclusions

The main goal of the thesis was to study gravitational redshift profiles of virialized halos
in both the symmetron model and chameleon f (R)-gravity using N -body simulations. This
required several sub-steps, which can be split into three categories: Firstly, the extraction
of data from the simulation output; secondly, the identification and definition of a halo and
thirdly the question whether a halo is virialized or not.

The gravitational redshift as well as other particle related quantities, such as the acceleration
due to the additional force, that were used to explain the results were calculated directly
from the N -body simulation. This allowed the straightforward extraction of the gravitational
redshift profiles.

In order to obtain the state of virialization, several methods were analyzed and the most
adequate one – using the total acceleration of the particles – was employed. This allowed us
to calculate the virialization parameter taking the energy of the scalar field into account.

Furthermore, the spherical collapse model was utilized to test if the common halo definition
with a fixed density contrast value leading to R200 or similar is applicable for screened mod-
ified gravity models. The conclusion is mixed as, on the one hand, we found the density
contrast value to vary significantly with halo mass. On the other hand, effects like envi-
ronmental screening have not been taken into account. Therefore, we expect this effect to
have a minor impact on the final results. A more detailed analysis of this problem has to be
carried out and is subject of future work.

We found the results to be highly dependent on the halo mass, which means the consideration
of multiple mass ranges is crucial when analyzing observational data. In particular, three
possible regimes were identified:

(i) No deviation from ΛCDM. If the halo is fully screened (either through self-screening or
environmental screening) or is too small to affect the scalar field, no deviation in the
density and, consequently, neither in the gravitational redshift profiles was observed.
6.1 Outlook 99

(ii) Enhanced gravitational redshift. This is the expected result as the fifth force leads to
additional clustering in the center of the halo. Additional matter in the central region
means a deeper potential well and, therefore, a positive deviation compared to ΛCDM.
(iii) Weaker gravitational redshift. If the fifth force is screened in the central region of the
halo but not in the outskirts, a negative deviation can appear. The additional force
leads to a matter overdensity in the outer regions while – with respect to the halo mass
– the inner regions are less dense than in the ΛCDM case.
This shows, that simply assuming the change of the gravitational redshift is equal to the
maximum possible change of the gravitational constant, as done frequently when predictions
are made, is not sufficient. Instead, the prediction obtained by N -body simulations as in
this work should be used.
Finally, the results were compared with existing observational data, mainly to show that only
future surveys may allow the exclusion of certain models or the restriction of the parameter
space. This topic is referred to in the next section.

6.1 Outlook

As previously mentioned, thanks to the planned next generation experiments like BigBOSS
(start in 2017) and Euclid (expected launch 2020), which will cover ten to fifty times more
galaxies than the full SDSS survey (completed in 2014), gravitational redshift will become an
observationally important tool besides lensing. This quantitative and qualitative improve-
ment offered by the new surveys, will automatically lead to a higher signal to noise ratio.
Naturally, the numerous additional effects described by Kaiser (2013) and Zhao et al. (2012)
have to be taken into account as well in order to avoid systematic errors. The third point on
the observational side will be the combination of data with other measurements as pointed
out by Croft (2013) to determine the matter of the object. All these factors will eventu-
ally lead to high-precision measurements that can be used to exclude alternative models of
gravity and establish further constraints.
Before these future goals can be achieved, more work has to be done on the numerical side.
A major part will be to map out the parameter space of various modified gravity models. In
order to reduce the computing cost, this can be done by using the spherical collapse model
to extrapolate data obtained from N -body simulations beyond the simulation parameters.
For this purpose a proper framework using excursion set theory has to be set up as done
by Lombriser et al. (2013) for the Hu-Sawicky model. Another question that has to be
addressed when using mock data sets is finding the optimal gravitational redshift estimator
and binning technique to improve the signal to noise ratio. The last point worth mentioning
is the obviously improvable simulation setup, as a higher resolution and added baryonic
effects can show hidden features in the low-mass end.
A

Appendix

A.1 Derivation of the Klein-Gordon equation

Varying the the action of a free scalar field given by (Fujii & Maeda, 2007)
√ 1
Z  
4
Sφ = − d x −g ∂ µ φ∂ µ φ + V (φ) (A.1)
2
with respect to φ results in
√ 1
Z  
4
δS = − d x −g δ (∇µ φ∇µ φ) + V,φ δφ (A.2)
2
√
Z
= − d4 x −g {∇µ φ∇µ δφ + V,φ δφ} (A.3)
√
Z
=− d4 x −g {−∇µ ∇µ φδφ + ∇µ [∇µ φδφ] + V,φ δφ} (A.4)
√
Z
=− d4 x −g {−φ + V,φ } δφ . (A.5)

Where the second term in (A.4) vanishes due to Gauss’ theorem if δφ is zero at the bound-
aries.
Demanding δS = 0 results in the Klein-Gordon equation

φ − V,φ (φ) = 0 . (A.6)

A.2 Gauss Seidel method with successive overrelaxation

The idea behind mesh relaxation is that a matrix equation of the form

Ay = q , (A.7)
102 A Appendix

with A being a matrix and y and q are vectors is solved using the decomposition of the
matrix A

A =B+R , (A.8)

where B is easily invertible. The equation can then be rearranged resulting in

y = B−1 q + My . (A.9)

The matrix M = −B−1 (A − B) is usually denoted as iteration matrix.

Mesh relaxation works by starting with an initial guess y (0) and iterating from t = 0 to tc
when some convergence criteria is fulfilled using

y (t+1) = B−1 q + My (t) . (A.10)

Convergence towards the exact solution y can be shown (Hockney & Eastwood, 1988) defin-
ing the error

ε(t) = y (t) − y . (A.11)

Substituting this back in (A.10) and noting that (A.9) is fulfilled leaves

ε(t) = Mε(t−1) = Mt ε(0) . (A.12)

The equation can also be generalized to a variable M by replacing Mt with M(t) :

ε(t) = Mε(t−1) = M(t) ε(0) (A.13)

(t) (t)
The error ||ε(t) || = i εi εi converges iff the greatest absolute eigenvalue of the matrix
P

||M(t) || is smaller than 1 for all times bigger than some arbitrary time. This can be seen by
taking the norm of Eq. (A.13) and noting that the euclidean (L2 ) norm of a matrix is its
greatest absolute eigenvalue (Horn & Johnson, 1985):

||ε(t) || (t)

(t)

≤ ||M || = max λ i . (A.14)
||ε(0) ||

In the algorithm the matrices A, B or M(t) are never actually stored or inverted. For a
differential equation - as the Poisson equation - finite difference aproximations are used.
E.g. a general two dimensional Laplacian can be aproximated as

f (x0 − h,y0 ) + f (x0 ,y0 − h) + f (x0 + h,y0 ) + f (x0 ,y0 − h) − 4f (x0 ,y0 )
∇2 f (x,y) ≈ .
x=x0 h2
y=y0
(A.15)

This results in matrix notation for Poisson equation in comoving coordinates Eq. (3.17) on
a two dimensional uniform mesh:
(t) h2 h (t) (t) (t) (t)
i
ỹi,j = yi−1,j + yi,j−1 + yi+1,j + yi,j+1 − πG(ρ′i,j − ρ0 ) (A.16)
4
A.2 Gauss Seidel method with successive overrelaxation 103

the factors may differ in other applications like non-uniform grids or more dimensional
systems.
For the Gauss Seidel method the original matrix A is decomposed in a lower triangular part
(for B) which is easily invertible and the rest, a stricly upper triangular part (for R). After
each timestep the new value for is simply
(t+1) (t)
yi,j = ỹi,j . (A.17)

Originally this method only converges certainly if the matrix A is diagonally dominant or
symmetric positive definite (but may also converge for other cases) (Hockney & Eastwood,
1988).
Therefore - and to decrease the convergence time dramatically - equation (A.17) is altered.
(t)
Instead of simply taking ỹi,j as the new value this trend is “over relaxed” to anticipate future
changes (Hockney & Eastwood, 1988):

new (t) (t)

yi,j = ω ỹi,j + (1 − ω)yi,j with 1 < ω < 2 (A.18)

This procedure is knows as successive overrelaxation and is controled via the the relaxation
factor ω. A value of ω = 1 corresonds to the simple GS method and a value of
2
ω= p (A.19)
1 + 1 − ρ2

is optimal – given the greatest absolute Eigenvalue (spectral radius) of the Jacobi matrix ρ
(Hockney & Eastwood, 1988). In RAMSES a value of
2
ω= , (A.20)
1 + π/L̄

where L̄ is the average patch size is used in the RAMSES code (Teyssier, 2002).
Bibliography

Aihara, H., Allende Prieto, C., An, D., Anderson, S. F., Aubourg, É., Balbinot, E., Beers,
T. C., Berlind, A. A., Bickerton, S. J., Bizyaev, D., Blanton, M. R., Bochanski, J. J.,
Bolton, A. S., Bovy, J., Brandt, W. N., Brinkmann, J., Brown, P. J., Brownstein, J. R.,
Busca, N. G., et al. Apr. 2011, The Eighth Data Release of the Sloan Digital Sky Survey:
First Data from SDSS-III, ApJS, 193, 29,p. 29
Amendola, L & Tsujikawa, S: Dark energy: Theory and observations. Cambridge University
Press, 2010
Appleby, S. & Battye, R. Oct. 2007, Do consistent F(R)F(R) models mimic general relativity
plus Λ?, Physics Letters B, 654,pp. 7–12
Armendariz-Picon, C, Mukhanov, V & Steinhardt, P. 2000, Dynamical solution to the prob-
lem of a small cosmological constant and late-time cosmic acceleration, Physical Review
Letters
Baldi, M. 2011, Time-dependent couplings in the dark sector: from background evolution
to non-linear structure formation, Monthly Notices of the Royal Astronomical Society,
27.February,pp. 1–27
Barnes, J. & Hut, P. Dec. 1986, A hierarchical O(N log N) force-calculation algorithm, Na-
ture, 324,pp. 446–449
Behroozi, P. S., Wechsler, R. H. & Wu, H.-Y. Jan. 2013, The Rockstar Phase-Space Tem-
poral Halo Finder and the Velocity Offsets of Cluster Cores, The Astrophysical Journal,
762.2,p. 109
Bekenstein, J. D. & Sanders, R. H. Mar. 2012, TeVeS/MOND is in harmony with gravi-
tational redshifts in galaxy clusters, Monthly Notices of the Royal Astronomical Society:
Letters, 421.1,pp. L59–L61
Bertschinger, E. June 2011, GRAFIC-2: Multiscale Gaussian Random Fields for Cosmolog-
ical Simulations, Astrophysics Source Code Library,p. 6008
Bertschinger, E. Sept. 1998, Simulations of Structure Formation in the Universe, Annual
Review of Astronomy and Astrophysics, 36.1,pp. 599–654
Bibliography 105

Birdsall, C. K. & Fuss, D. Apr. 1969, Clouds-in-Clouds, Clouds-in-Cells, Physics for Many-
Body Plasma Simulation, Journal of Computational Physics, 3,p. 494
Bondi, H. 1947, Spherically symmetrical models in general relativity, MNRAS, 107,p. 410
Bouchet, F. R. & Hernquist, L. Dec. 1988, Cosmological simulations using the hierarchical
tree method, ApJS, 68,pp. 521–538
Brans, C. & Dicke, R. H. Nov. 1961, Mach’s Principle and a Relativistic Theory of Gravita-
tion, Phys. Rev. 124, (3),pp. 925–935
Brax, P., Bruck, C van de & Davis, A. 2004, Detecting dark energy in orbit: The cosmological
chameleon, Physical Review D,pp. 1–31
Brax, P., Bruck, C. V. D., Davis, A. & Shaw, D. 2008, f (R) gravity and chameleon theories,
Physical Review D,pp. 1–18
Brax, P., Bruck, C. van de, Davis, A.-C. & Shaw, D. Sept. 2010a, Dilaton and modified
gravity, Physical Review D, 82.6,p. 063519
Brax, P., Rosenfeld, R & Steer, D. Aug. 2010b, Spherical collapse in chameleon models,
Journal of Cosmology and Astroparticle Physics, 2010.08,pp. 033–033
Brax, P., Davis, A.-C., Li, B., Winther, H. A. & Zhao, G.-B. June 2012, Systematic simula-
tions of modified gravity: symmetron and dilaton models, arXiv preprint arXiv:1206.3568,p. 30
Brax, P., Davis, A., Li, B., Winther, H. & Zhao, G. 2013, Systematic simulations of modified
gravity: chameleon models, arXiv preprint arXiv:1303.0007v2,pp. 1–16
Bryan, G. L. & Norman, M. L. Mar. 1998, Statistical Properties of X-Ray Clusters: Analytic
and Numerical Comparisons, ApJ, 495,pp. 80–99
Cappi, A 1995, Gravitational redshift in galaxy clusters., Astronomy and Astrophysics
Carretta, E., Gratton, R. G., Clementini, G. & Fusi Pecci, F. Apr. 2000, Distances, Ages,
and Epoch of Formation of Globular Clusters, ApJ, 533,pp. 215–235
Carroll, S. M. Oct. 1998, Quintessence and the Rest of the World: Suppressing Long-Range
Interactions, Physical Review Letters, 81,pp. 3067–3070
Carroll, S.: Spacetime and geometry: an introduction to general relativity. Addison-Wesley
Longman, Incorporated, 2004
Chandrasekhar, S: Hydrodynamic and hydromagnetic stability. Dover Publications, Incorpo-
rated, 1981
Clifton, T., Ferreira, P. G., Padilla, A. & Skordis, C. Mar. 2012, Modified gravity and cos-
mology, Physics Reports, 513.1-3,pp. 1–189
Cole, S & Lacey, C July 1996, The structure of dark matter haloes in hierarchical clustering
models, Monthly Notices of the Royal Astronomical Society, 281.2,pp. 716–736
106 Bibliography

Colombi, S., Jaffe, A., Novikov, D. & Pichon, C. Feb. 2009, Accurate estimators of power spec-
tra in N -body simulations, Monthly Notices of the Royal Astronomical Society, 393.2,pp. 511–
526
Cooley, J. & Tukey, J. 1965, An algorithm for the machine calculation of complex Fourier
series, Mathematics of computation,pp. 297–301
Couchman, H. M. P. Feb. 1991, Mesh-refined P3M - A fast adaptive N-body algorithm, ApJ,
368,pp. L23–L26
Croft, R. 2013, Gravitational redshifts from large-scale structure, arXiv preprint arXiv:1304.4124,
April
Davis, A., Li, B. & Mota, D. 2011, Structure Formation in the Symmetron Model, Arxiv
preprint arXiv:1108.3081
Davis, M., Efstathiou, G., Frenk, C. S. & White, S. D. M. May 1985, The evolution of
large-scale structure in a universe dominated by cold dark matter, ApJ, 292,pp. 371–394
de Blok, W. J. G. 2010, The Core-Cusp Problem, Advances in Astronomy, 2010, 789293
Dehnen, W. & Read, J. 2011, N-body simulations of gravitational dynamics, The European
Physical Journal Plus,pp. 1–28
Diemand, J., Kuhlen, M. & Madau, P. Sept. 2006, Early Supersymmetric Cold Dark Matter
Substructure, ApJ, 649,pp. 1–13
Dodelson, S.: Modern cosmology. Academic Press. Academic Press, 2003
Efstathiou, G. & Eastwood, J. W. Feb. 1981, On the clustering of particles in an expanding
universe, MNRAS, 194,pp. 503–525
Engineer, S., Kanekar, N. & Padmanabhan, T. May 2000, Non-linear density evolution from
an improved spherical collapse model, Monthly Notices of the Royal Astronomical Society,
314.2,pp. 279–289
Everitt, B & Skrondal, A: The Cambridge dictionary of statistics. Third Edit. Cambridge
University Press, 2002
Ewald, P. P. 1921, Die Berechnung optischer und elektrostatischer Gitterpotentiale, Annalen
der Physik, 369.3,pp. 253–287
Faraoni, V., Gunzig, E. & Nardone, P. 1998, Conformal transformations in classical gravi-
tational theories and in cosmology, arXiv preprint gr-qc/9811047
Ferraro, S., Schmidt, F. & Hu, W. 2011, Cluster abundance in f (R) gravity models, Physical
Review D, 1,pp. 1–8
Forero-Romero, J. E., Gottlöber, S. & Yepes, G. Dec. 2010, Bullet Clusters in the Marenos-
trum Universe, The Astrophysical Journal, 725.1,pp. 598–604
Fujii, Y. & Maeda, K.: The Scalar-Tensor Theory of Gravitation. Cambridge University
Press, 2007
Bibliography 107

Gottloeber, S. 1998, Galaxy Tracers in Cosmological N-Body Simulations. In: Large Scale
Structure: Tracks and Traces. Ed. by V. Mueller, S. Gottloeber, J. P. Muecket & J. Wamb-
sganss. 1998, pp. 43–46
Gunn, J. & Gott, J. 1972, On the infall of matter into clusters of galaxies and some effects
on their evolution, The Astrophysical Journal
Habib, S., Pope, A., Lukić, Z., Daniel, D., Fasel, P., Desai, N., Heitmann, K., Hsu, C.-H.,
Ankeny, L., Mark, G., Bhattacharya, S. & Ahrens, J. July 2009, Hybrid petacomputing
meets cosmology: The Roadrunner Universe project, Journal of Physics Conference Series,
180.1,p. 012019
Hansen, B. M. S., Brewer, J., Fahlman, G. G., Gibson, B. K., Ibata, R., Limongi, M., Rich,
R. M., Richer, H. B., Shara, M. M. & Stetson, P. B. Aug. 2002, The White Dwarf Cooling
Sequence of the Globular Cluster Messier 4, ApJ, 574,pp. L155–L158
Hartle, J.: Gravity: an introduction to Einstein’s general relativity. Addison-Wesley, 2003
Hernquist, L. Aug. 1987, Performance characteristics of tree codes, ApJS, 64,pp. 715–734
Hinterbichler, K. & Khoury, J. June 2010, Screening Long-Range Forces through Local Sym-
metry Restoration, Physical Review Letters, 104.23,p. 231301
Hinterbichler, K., Khoury, J., Levy, A. & Matas, A. 2011, Symmetron cosmology, Physical
Review D
Hockney, R. W., Goel, S. P. & Eastwood, J. W. Feb. 1974, Quiet High-Resolution Computer
Models of a Plasma, Journal of Computational Physics, 14,p. 148
Hockney, R. & Eastwood, J.: Computer simulation using particles. Taylor & Francis, 1988
Horn, R. A. & Johnson, C. R.: Matrix analysis. Cambridge University Press, 1985
Hu, W. & Sawicki, I. 2007, Models of f (R) cosmic acceleration that evade solar system tests,
Physical Review D,pp. 1–13
Jain, B., Vikram, V. & Sakstein, J. Apr. 2012, Astrophysical Tests of Modified Gravity:
Constraints from Distance Indicators in the Nearby Universe,p. 37
Jessop, C., Duncan, M. & Chau, W. Y. Dec. 1994, Multigrid Methods for N-Body Gravita-
tional Systems, Journal of Computational Physics, 115,pp. 339–351
Johnston, D. & Sheldon, E. 2007, Cross-correlation Weak Lensing of SDSS galaxy Clusters II:
Cluster Density Profiles and the Mass–Richness Relation, arXiv preprint arXiv:0709.1159v1
Kaiser, N. 2013, Measuring Gravitational Redshifts in Galaxy Clusters, arXiv preprint arXiv:1303.3663,
000.March
Khoury, J. & Weltman, A. 2004a, Chameleon cosmology, Physical Review D
– 2004b, Chameleon fields: Awaiting surprises for tests of gravity in space, Physical review
letters
108 Bibliography

Kim, Y.-R. Y.-r. & Croft, R. A. C. May 2004, Gravitational Redshifts in Simulated Galaxy
Clusters, The Astrophysical Journal, 607.1,pp. 164–174
Klypin, A., Gottlöber, S., Kravtsov, A. V. & Khokhlov, A. M. May 1999, Galaxies in N-Body
Simulations: Overcoming the Overmerging Problem, ApJ, 516,pp. 530–551
Klypin, A. A. & Shandarin, S. F. Sept. 1983, Three-dimensional numerical model of the
formation of large-scale structure in the Universe, MNRAS, 204,pp. 891–907
Knebe, A., Green, A. & Binney, J. Aug. 2001, Multi-level adaptive particle mesh (MLAPM):
a c code for cosmological simulations, MNRAS, 325,pp. 845–864
Knebe, A., Kravtsov, A. V., Gottlober, S. & Klypin, A. a. Sept. 2000, On the effects of
resolution in dissipationless cosmological simulations, Monthly Notices of the Royal As-
tronomical Society, 317.3,pp. 630–648
Knebe, A., Knollmann, S. R., Muldrew, S. I., Pearce, F. R., Aragon-Calvo, M. A., Ascasibar,
Y., Behroozi, P. S., Ceverino, D., Colombi, S., Diemand, J., Dolag, K., Falck, B. L., Fasel,
P., Gardner, J., Gottlöber, S., Hsu, C.-H., Iannuzzi, F., Klypin, A., Lukić, Z., et al. Aug.
2011, Haloes gone MAD: The Halo-Finder Comparison Project, Monthly Notices of the
Royal Astronomical Society, 415.3,pp. 2293–2318
Knebe, A., Pearce, F. & Lux, H. 2013, Structure Finding in Cosmological Simulations: The
State of Affairs, arXiv preprint arXiv:1304.0585v1, 42.April,pp. 1–42
Knollmann, S. R. & Knebe, A. June 2009, AHF: Amiga’s Halo Finder, ApJS, 182,pp. 608–
624
Kravtsov, A. V., Klypin, A. A. & Khokhlov, A. M. July 1997, Adaptive Refinement Tree:
A New High-Resolution N -Body Code for Cosmological Simulations, The Astrophysical
Journal Supplement Series, 111.1,pp. 73–94
Kuhlen, M., Vogelsberger, M. & Angulo, R. Nov. 2012, Numerical simulations of the dark
universe: State of the art and the next decade, Physics of the Dark Universe, 1.1-2,pp. 50–
93
Lahav, O., Lilje, P. B., Primack, J. R. & Rees, M. J. 1991, Dynamical effects of the cosmo-
logical constant, Monthly Notices of the . . .
Larson, D., Dunkley, J., Hinshaw, G., Komatsu, E., Nolta, M. R., Bennett, C. L., Gold, B.,
Halpern, M., Hill, R. S., Jarosik, N., Kogut, A., Limon, M., Meyer, S. S., Odegard, N.,
Page, L., Smith, K. M., Spergel, D. N., Tucker, G. S., Weiland, J. L., et al. Feb. 2011,
Seven-year Wilkinson Microwave Anisotropy Probe (WMAP) Observations: Power Spectra
and WMAP-derived Parameters, ApJS, 192, 16,p. 16
Laureijs, R, Amiaux, J, Arduini, S, Auguères, J. L., Brinchmann, J., Cole, R., Cropper, M.,
Dabin, C., Duvet, L., Ealet, A., Garilli, B., Gondoin, P., Guzzo, L., Hoar, J., Hoekstra, H.,
Holmes, R., Kitching, T., Maciaszek, T., Mellier, Y., et al. Oct. 2011, Euclid Definition
Study Report, arXiv preprint arXiv:1110.3193, July,p. 116
Bibliography 109

Lee, E. & Forthofer, R.: Analyzing Complex Survey Data. Analyzing Complex Survey Data.
SAGE Publications, 2006
Li, B. & Efstathiou, G. Apr. 2012, An extended excursion set approach to structure formation
in chameleon models, Monthly Notices of the Royal Astronomical Society, 421.2,pp. 1431–
1442
Li, B., Hellwing, W. A., Koyama, K., Zhao, G.-B., Jennings, E. & Baugh, C. M. Oct. 2012,
The non-linear matter and velocity power spectra in f(R) gravity, Monthly Notices of the
Royal Astronomical Society, 428.1,pp. 743–755
Li, Y. & Hu, W. 2011, Chameleon halo modeling in f (R) gravity, Physical Review D, 1,pp. 1–
7
Llinares, C. & Mota, D. F. Apr. 2013a, Releasing Scalar Fields: Cosmological Simulations
of Scalar-Tensor Theories for Gravity Beyond the Static Approximation, Physical Review
Letters, 110.16,p. 161101
– Apr. 2013b, Shape of Clusters of Galaxies as a Probe of Screening Mechanisms in Modified
Gravity, Physical Review Letters, 110.15,p. 151104
Llinares, C., Mota, D. F. & Winther, H. A. 2013, Isis : a new N-body cosmological code with
scalar fields based on Ramses, (in preparation)
Lokas, E. L. & Mamon, G. a. Feb. 2001, Properties of spherical galaxies and clusters with an
NFW density profile, Monthly Notices of the Royal Astronomical Society, 321.1,pp. 155–
166
Lombriser, L., Schmidt, F., Baldauf, T., Mandelbaum, R., Seljak, U. & Smith, R. May 2012,
Cluster density profiles as a test of modified gravity, Physical Review D, 85.10,p. 102001
Lombriser, L., Li, B., Koyama, K. & Zhao, G. 2013, Modeling halo mass functions in
chameleon f (R) gravity, arXiv preprint arXiv:1304.6395,pp. 1–11
Ludlow, A. D., Navarro, J. F., Boylan-Kolchin, M., Bett, P. E., Angulo, R. E., Li, M., White,
S. D. M., Frenk, C. & Springel, V. Feb. 2013, The Mass Profile and Accretion History of
Cold Dark Matter Halos, arXiv preprint arXiv:1302.0288, 000.February,p. 12
Maor, I. & Lahav, O. July 2005, On virialization with dark energy, Journal of Cosmology
and Astroparticle Physics, 2005.07,pp. 003–003
Martin, J. 2012, Everything You Always Wanted To Know About The Cosmological Constant
Problem (But Were Afraid To Ask)
Martizzi, D., Teyssier, R., Moore, B. & Wentz, T. 2011, The effects of baryon physics, black
holes and AGN feedback on the mass distribution in clusters of galaxies, arXiv preprint
arXiv:1112.2752, 000.April
Mateo, M. L. 1998, Dwarf Galaxies of the Local Group, ARA&A, 36,pp. 435–506
Misner, C., Thorne, K. & Wheeler, J.: Gravitation: Physics Series poeng 3. W. H. Freeman,
1973
110 Bibliography

Mota, D. & Bruck, C van de 2004, On the spherical collapse model in dark energy cosmologies,
Astronomy and Astrophysics, 81,pp. 71–81
Navarro, J. F., Frenk, C. S. & White, S. D. M. Dec. 1995, Simulations of X-ray clusters,
Monthly Notices of the Royal Astronomical Society
– May 1996, The Structure of Cold Dark Matter Halos, The Astrophysical Journal, 462,p. 563
– 1997, A Universal density profile from hierarchical clustering, The Astrophysical Journal,
1.1,pp. 493–508
Oguri, M., Bayliss, M. B., Dahle, H. k., Sharon, K., Gladders, M. D., Natarajan, P., Hennawi,
J. F. & Koester, B. P. Mar. 2012, Combined strong and weak lensing analysis of 28 clusters
from the Sloan Giant Arcs Survey, Monthly Notices of the Royal Astronomical Society,
420.4,pp. 3213–3239
Okabe, N., Smith, G. & Umetsu, K. 2013, LoCuSS: The Mass Density Profile of Massive
Galaxy Clusters at z= 0.2, arXiv preprint arXiv:1302.2728v1, 1,pp. 1–6
Olive, K. & Pospelov, M. 2008, Environmental dependence of masses and coupling constants,
Physical Review D, September
Oyaizu, H. 2008, Nonlinear evolution of f (R) cosmologies. I. Methodology, Physical Review
D
Perlmutter, S., Aldering, G., Goldhaber, G., Knop, R. A., Nugent, P., Castro, P. G., Deustua,
S., Fabbro, S., Goobar, A., Groom, D. E., Hook, I. M., Kim, A. G., Kim, M. Y., Lee, J. C.,
Nunes, N. J., Pain, R., Pennypacker, C. R., Quimby, R., Lidman, C., et al. June 1999, Mea-
surements of Omega and Lambda from 42 High-Redshift Supernovae, ApJ, 517,pp. 565–
586
Planck Collaboration, Ade, P. A. R., Aghanim, N., Armitage-Caplan, C., Arnaud, M., Ash-
down, M., Atrio-Barandela, F., Aumont, J., Baccigalupi, C., Banday, A. J. & al., et Mar.
2013, Planck 2013 results. XVI. Cosmological parameters, ArXiv e-prints
Press, W., Flannery, B., Teukolsky, S. & Vetterling, W.: Numerical Recipes in FORTRAN
Example Book: The Art of Scientific Computing. The Art of Scientific Computing. Cam-
bridge University Press, 1992
Prunet, S., Pichon, C., Aubert, D., Pogosyan, D., Teyssier, R. & Gottloeber, S. Oct. 2008,
Initial Conditions For Large Cosmological Simulations, ApJS, 178,pp. 179–188
Riess, A. G., Filippenko, A. V., Challis, P., Clocchiatti, A., Diercks, A., Garnavich, P. M.,
Gilliland, R. L., Hogan, C. J., Jha, S., Kirshner, R. P., Leibundgut, B., Phillips, M. M.,
Reiss, D., Schmidt, B. P., Schommer, R. A., Smith, R. C., Spyromilio, J., Stubbs, C.,
Suntzeff, N. B. & Tonry, J. Sept. 1998, Observational Evidence from Supernovae for an
Accelerating Universe and a Cosmological Constant, AJ, 116,pp. 1009–1038
Bibliography 111

Schlegel, D. J., Bebek, C., Heetderks, H., Ho, S., Lampton, M., Levi, M., Mostek, N., Pad-
manabhan, N., Perlmutter, S., Roe, N., Sholl, M., Smoot, G., White, M., Dey, A., Abra-
ham, T., Jannuzi, B., Joyce, D., Liang, M., Merrill, M., et al. Apr. 2009, BigBOSS: The
Ground-Based Stage IV Dark Energy Experiment, ArXiv e-prints
Schmidt, F. 2010, Dynamical masses in modified gravity, Physical Review D,pp. 1–18
Shaw, D. J. & Mota, D. F. Feb. 2008, An Improved Semianalytical Spherical Collapse
Model for Nonlinear Density Evolution, The Astrophysical Journal Supplement Series,
174.2,pp. 277–281
Shaw, L. D., Weller, J., Ostriker, J. P. & Bode, P. Aug. 2006, Statistics of Physical Properties
of Dark Matter Clusters, The Astrophysical Journal, 646.2,pp. 815–833
Silverman, B. W.: Density estimation for statistics and data analysis. 1986
Smith, T. L. July 2009, Testing gravity on kiloparsec scales with strong gravitational lenses,
arXiv preprint arXiv:0907.4829,p. 13
Springel, V., Yoshida, N. & White, S. D. M. Apr. 2001, GADGET: a code for collisionless
and gasdynamical cosmological simulations, New A, 6,pp. 79–117
Stadel, J. G.: Cosmological N-body simulations and their analysis. PhD thesis. UNIVERSITY
OF WASHINGTON 2001
Starobinsky, A. A. Oct. 2007, Disappearing cosmological constant in f( R) gravity, Soviet
Journal of Experimental and Theoretical Physics Letters, 86,pp. 157–163
Steinhardt, P., Wang, L. & Zlatev, I. 1999, Cosmological tracking solutions, Physical Review
D
Teyssier, R 2002, Cosmological hydrodynamics with adaptive mesh refinement. A new high
resolution code called RAMSES, Astronomy and Astrophysics, 364,pp. 337–364
Teyssier, R., Moore, B. & Martizzi, D. 2010, Mass Distribution in Galaxy Clusters: the Role
of AGN Feedback, arXiv preprint arXiv:1003.4744v2, 000.December
Tolman, R. C. Mar. 1934, Effect of Inhomogeneity on Cosmological Models, Proceedings of
the National Academy of Science, 20,pp. 169–176
Tweed, D., Devriendt, J., Blaizot, J., Colombi, S. & Slyz, A. Nov. 2009, Building merger
trees from cosmological N-body simulations. Towards improving galaxy formation models
using subhaloes, A&A, 506,pp. 647–660
Upadhye, A. 2013, Symmetron dark energy in laboratory experiments, Physical Review Let-
ters, 60439,pp. 2–5
Wang, Y.: Dark Energy. Wiley, 2009
Waterhouse, T. P. Nov. 2006, An Introduction to Chameleon Gravity, arXiv preprint arXiv:0611.816,
November,p. 33
112 Bibliography

Wetterich, C. 1995, An asymptotically vanishing time-dependent cosmological "constant",

Astronomy & Astrophysics, 302,p. 321
Wintergerst, N. & Pettorino, V. 2010, Clarifying spherical collapse in coupled dark energy
cosmologies, Physical Review D,pp. 1–16
Winther, H. A., Mota, D. F. & Li, B. 2011, Environment dependence of dark matter halos
in symmetron modified gravity,pp. 1–10
Wojtak, R., Hansen, S. H. & Hjorth, J. Sept. 2011, Gravitational redshift of galaxies in
clusters as predicted by general relativity, Nature, 477.7366,pp. 567–569
Zhao, G.-B., Li, B. & Koyama, K. Feb. 2011, N-body simulations for f(R) gravity using a
self-adaptive particle-mesh code, Physical Review D, 83.4,p. 044007
Zhao, H., Peacock, J. & Li, B. 2012, Gravity theories, Transverse Doppler and Gravitational
Redshifts in Galaxy Clusters, arXiv preprint arXiv:1206.5032, 2,pp. 1–4