Notes

Aspects of Modified Gravity
A thesis submitted to the University of Manchester

for the degree of Doctor of Philosophy
in the faculty of Engineering and Physical Sciences
2013
By
Edward Reeves
School of Physics & Astronomy
Jodrell Bank Centre for Astrophysics
2
Contents
Abstract 9
Declaration 11
Copyright statement 13
The Author 15
Acknowledgements 17
Glossary 19
Useful Expressions 23
1 Introduction to Gravitational Theories and Cosmology 27
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.2 General Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.2.1 General Relativity, covariant derviatives, energy-momentum

tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.2.2 The Riemann tensor and curvature and the Einstein tensor 29
3
4 CONTENTS
1.2.3 Formulation of the Einstein equations . . . . . . . . . . . . 30
1.2.4 The FRW metric and the Ricci scalar . . . . . . . . . . . . 32
1.2.5 Density, pressure and equation of state . . . . . . . . . . . 33
1.2.6 The Friedmann equations . . . . . . . . . . . . . . . . . . 34
1.2.7 Einstein equations derived from the action . . . . . . . . . 34
1.3 A brief history of the Universe . . . . . . . . . . . . . . . . . . . . 36
1.3.1 The radiation era . . . . . . . . . . . . . . . . . . . . . . . 37
1.3.2 The matter dominated era and the CMB . . . . . . . . . . 37
1.3.3 The vacuum era . . . . . . . . . . . . . . . . . . . . . . . . 38
1.4 Modifying gravity with F (R) gravity models . . . . . . . . . . . . 38
1.4.1 A special case . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.4.2 A vacuum solution . . . . . . . . . . . . . . . . . . . . . . 42
1.4.3 A modified expression for the Hubble parameter and the

effective equation of state . . . . . . . . . . . . . . . . . . 42
1.4.4 Dark energy . . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.4.4.1 The cosmological constant . . . . . . . . . . . . . 44
1.4.4.2 Quintessence . . . . . . . . . . . . . . . . . . . . 45
1.4.4.3 Effective dark energy . . . . . . . . . . . . . . . . 45
1.4.5 The modified gravity particle . . . . . . . . . . . . . . . . 45
1.4.6 The coincidence problem . . . . . . . . . . . . . . . . . . . 46
1.4.7 The growth of matter and perturbations . . . . . . . . . . 47
1.4.8 Constraints on F (R) . . . . . . . . . . . . . . . . . . . . . 47

CONTENTS 5
1.4.9 Conditions for an accelerating Universe . . . . . . . . . . . 49
1.4.10 Probing the Growth of Large Scale Structure . . . . . . . . 49
2 Gravitation Models Which Mimic ΛCDM 53
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.2 Finding all models which mimic ΛCDM . . . . . . . . . . . . . . . 55
2.2.1 Algebraic solution for R > 4Λ . . . . . . . . . . . . . . . . 56
2.2.2 Algebraic solution which includes R = 4Λ . . . . . . . . . 58
2.2.3 Matching conditions for the two expressions for F (R) . . . 59
2.2.4 Application of constraints to the solutions . . . . . . . . . 64
2.2.4.1 The condition that FR (R) > 0 . . . . . . . . . . . 65
2.2.4.2 The condition that FRR (R) ≥ 0 . . . . . . . . . . 65
2.2.4.3 The local gravity constraint . . . . . . . . . . . . 65
2.2.5 The matter growth index, γ . . . . . . . . . . . . . . . . . 67
2.3 In the radiation era . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3 Non-ΛCDM Gravitation Models 75
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.2 Three models compared . . . . . . . . . . . . . . . . . . . . . . . 78
3.2.1 The Erf model . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.2.1.1 The dark energy equation of state . . . . . . . . . 80
3.2.1.2 Late-time behaviour . . . . . . . . . . . . . . . . 84

6 CONTENTS
3.2.1.3 The matter growth index . . . . . . . . . . . . . 87
3.2.2 The AB model . . . . . . . . . . . . . . . . . . . . . . . . 89
3.2.3 The HSS model . . . . . . . . . . . . . . . . . . . . . . . . 92
3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4 The weff < −1 Theorem for F (R) Models 101
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.2 The theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5 Evolution of Perturbations 105
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.2 Matter perturbation equations in k-space for the Jordan Frame . 107
5.2.1 The ΛCDM model . . . . . . . . . . . . . . . . . . . . . . 109
5.3 Approximating algebraic solutions . . . . . . . . . . . . . . . . . . 110
5.3.1 Oscillatory solution . . . . . . . . . . . . . . . . . . . . . . 112
5.3.2 Non-oscillatory solution . . . . . . . . . . . . . . . . . . . 116
5.3.3 The combined algebraic solution . . . . . . . . . . . . . . . 117
5.3.4 Applying initial conditions to the algebraic approximation 119
5.4 Algebraic and numerical solutions compared . . . . . . . . . . . . 120
5.4.1 k = 0.1h Mpc−1 . . . . . . . . . . . . . . . . . . . . . . . . 121
5.4.2 k = 0.01h Mpc−1 . . . . . . . . . . . . . . . . . . . . . . . 127
5.4.3 k = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.4.4 Algebraic and numeric solutions for x(N ) compared for

other models of chapter 3. . . . . . . . . . . . . . . . . . . 133
CONTENTS 7
5.4.4.1 Model 2 . . . . . . . . . . . . . . . . . . . . . . . 133
5.4.4.2 Model 4 . . . . . . . . . . . . . . . . . . . . . . . 134
5.4.4.3 Model 5 . . . . . . . . . . . . . . . . . . . . . . . 135
5.4.5 Comment . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
5.5.1 Oscillations in δR . . . . . . . . . . . . . . . . . . . . . . . 139
6 Conclusions 141
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.1.1 Models with standard Einstein Gravity expansion history . 142
6.1.2 Extreme values generated by F (R) models . . . . . . . . . 143
6.1.3 Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
6.1.4 Oscillations of the potentials of the perturbed FRW metric 144
6.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Bibliography 149
8 CONTENTS
Abstract
Algebraic expressions are determined for all F (R) gravity models which mimic
standard Einstein Gravity. We find that they are severely constrained by local
gravity in the Solar System which makes them virtually indistinguishable from
standard Einstein Gravity. In addition, we find that instead of F (R) → R as
R → ∞, there is the possibility that F (R) ∝ R4/3 .
Aspects of two well-known F (R) models are discussed together with a new
model. For the three models, the evolution of the effective equation of state
parameter, weff , is followed as well as the effective density parameter, Ωeff , and the
relationship between them. Also considered are the dependence of the potential
on its defining field and the evolution of the matter growth index, which is also
compared with that of standard Einstein Gravity. The problems that occur, the
constraints that act in order to make these parameters extreme today, and the
effect that a late-time de Sitter attractor has and how this can be evaded to give
large deviations from weff,0 = −1 are discussed.
It is the case that in the radiation and matter eras, for F (R) models that tend
to standard Einstein Gravity as R → ∞, weff < −1 and decreases with time. A
proof of this is given as a theorem.
Matter oscillations of the perturbed, weak matter fields are considered and the
equations solved algebraically to show how they evolve in time both in amplitude
and frequency. They are compared with the numeric solutions of the equations
and, under certain circumstances, they compare favourably.
9
10
Declaration
No portion of the work referred to in this thesis has been

submitted in support of an application for another degree
or qualification of this or any other university or other
institution of learning.
11
12
Copyright statement
1. The author of this thesis (including any appendices and/or schedules to

this thesis) owns certain copyright or related rights in it (the Copyright)
and he has given The University of Manchester certain rights to use such
Copyright, including for administrative purposes.
2. Copies of this thesis, either in full or in extracts and whether in hard or

electronic copy, may be made only in accordance with the Copyright, De-
signs and Patents Act 1988 (as amended) and regulations issued under it
or, where appropriate, in accordance with licensing agreements which the
University has from time to time. This page must form part of any such
copies made.
3. The ownership of certain Copyright, patents, designs, trade marks and other
intellectual property (the Intellectual Property) and any reproductions of
copyright works in the thesis, for example graphs and tables (Reproduc-
tions), which may be described in this thesis, may not be owned by the
author and may be owned by third parties. Such Intellectual Property and
Reproductions cannot and must not be made available for use without the
prior written permission of the owner(s) of the relevant Intellectual Property
and/or Reproductions.
4. Further information on the conditions under which disclosure, publication

and commercialisation of this thesis, the Copyright and any Intellectual
13
Property and/or Reproductions described in it may take place is available
in the University IP Policy1 , in any relevant Thesis restriction declarations
deposited in the University Library, The University Library’s regulations2
and in The University’s policy on presentation of theses.
1
http://documents.manchester.ac.uk/display.aspx?DocID=487
2
http://www.manchester.ac.uk/library/aboutus/regulations
14
The Author
As a Senior Exhibitioner in Mathematics at Peterhouse, Cambridge, the au-

thor was awarded Senior Optime (Class II honours) in Part 2 of the Math-
ematical Tripos in 1968. On going down, he served in the Royal Navy as
an Instructor Officer specialising in teaching Mathematics to skilled ratings
and working on Programmed Learning techniques and also on computers
and radar. He then taught Mathematics at Wycliffe College, Stonehouse,
Glos, before going on to The King’s School, Worcester where he was Head
of the Mathematics Department for 24 years. He retired from teaching in
2006 when he joined the Particle Physics Group in the School of Physics
and Astronomy at the University of Manchester as a part-time graduate
student.
He is married to Ruth and has two married sons and four grandchildren.
15
16
Acknowledgements
I should like to thank all those in the Particle Physics Group and those
at the Jodrell Bank Centre for Astrophysics for the help they have given
me in their various ways. Especially to be mentioned must be Professor
Fred Loebinger who initially made me very welcome and is instrumental,
I believe, in making the Particle Physics Group very friendly. Even when
he had not, he always had time to answer my mundane questions. Being
part-time and remote in Worcester, for most of the time, means I have not
mixed at all with other graduate students so he and others have been the
butt of my questions when a full-time student would normally have asked
a student colleague. I suppose it has helped that we are virtually the same
age.
Also to be thanked are two very important people: Ann Morrow, Particle
Physics Secretary, who has been helpful to me, and Sabah Salih, the Group’s
incredible computing man, who has helped on many occasions to sort out
my computing woes. Both these people are treasures; thank you.
Then there are my two supervisors, Professors Jeff Forshaw and Richard
Battye, two able and talented people who helped lift me from the abyss of
not knowing anything about Cosmology to knowing a little more. While it
has not been easy, I have learned much from them and I owe them a debt
of gratitude.
17
I must also thank my Rotary club, the Rotary Club of Worcester South,
for allowing me leave of absence for the last few months when I was writing
this thesis. Especially, I must thank the two gentlemen of the club who
took over my major responsibilities for the duration.
I could not bring this to a close if I did not thank my dear wife, Ruth.
She has been unstinting in allowing me to work over the past seven years,
uninterrupted, in my study.
18
Glossary
AU – Astronomical Unit = 1.5 × 108 km.
BCS – Blanco Cosmology Survey.
CDM – Cold dark matter.
COBE – Cosmic Background Explorer satellite. It was launched in Novem-

ber 1989 to measure infra-red and microwave radiation from the early Uni-
verse. It carried three instruments, a Diffuse Infrared Background Exper-
iment (DIRBE) to search for the cosmic infrared background radiation, a
Differential Microwave Radiometer (DMR) to map the cosmic radiation
sensitively, and a Far Infrared Absolute Spectrophotometer (FIRAS) to
compare the spectrum of the cosmic microwave background radiation with
a precise blackbody [1, 2].
Expansion scale factor – the dimensionless parameter, a, which measures

the expansion of the Universe. Today, a is defined as 1; the history of the
Universe, to date, is described by 0 ≤ a ≤ 1. Its natural logarithm is
denoted by N .
Ghost – particle whose Hamiltonian contains negative energy terms.
GR – General Relativity represented in modified gravity by F (R) = R.
Hubble constant – today’s value of the Hubble parameter. The Planck

best fit value is H0 = 67.11 km s−1 Mpc−1 .
19
Hubble parameter, H, – defined as ȧ/a, which measures the relative rate
of expansion of the Universe. It has dimensions [T −1 ].
ΛCDM – cold dark matter with constant dark energy density represented
in modified gravity by F (R) = R − 2Λ. Λ is termed the cosmological
constant which could explain the accelerated expansion of the Universe. It
is sometimes termed standard Einstein Gravity.
LSS– Large Scale Structure. On scales smaller than the scale of the Uni-
verse, galaxies and clusters of galaxies formed as a result of gravitational
collapse in regions of higher average density which were, themselves, caused
by perturbations early in the history of the Universe.
MACS – Massive Cluster Survey.
Newtonian gauge – a perturbation of the FRW metric. See equation

(5.1).
The phantom divide – the line weff = −1.
Planck – a space observatory operated by the European Space Agency

(ESA), and designed to observe anisotropies of the cosmic microwave back-
ground (CMB) at microwave and infra-red frequencies, with high sensitiv-
ity and small angular resolution. It began operation in 2009 and comple-
mented and improved on WMAP measurements. Its results were published
in February 2013 [3].
REFLEX – ROSAT-ESO Flux-Limited X-Ray galaxy cluster survey.
Ricci or curvature scalar, R, – a measure of the curvature of space-time

due to energy sources in the Universe. It has dimensions [T −2 ].
Req – the value of the Ricci scalar at equality. This is when the radiation
density equals the matter density. Req ≈ 3 × 1010 H02 .
20
RΛ – the value of the Ricci scalar under ΛCDM.
Rs – the value of the Ricci scalar at the Solar System. In this thesis it is
taken to have the value 106 H02 .
ROSAT – the Röntgen Satellite, was an X-ray observatory developed

through a cooperative program between Germany, the United States, and
the United Kingdom.
FR (R)−RFRR (R) 1
Scalaron Mass – defined as M via M 2 = 3FRR (R)
≈ 3FRR (R)
. M has
dimensions [T −1 ].
Solar System constraint – the constraint on the parameters of a model

by the need for Rs FRR (Rs ) 10−23 .
Today – in cosmological terms, the current time denoted by suffix 0 so that

X0 is the value of the variable X, today. The value of Ωm , today, is denoted
by Ωm,0 . Today, a = 1 and N = 0.
WMAP – the NASA Wilkinson Microwave Anisotropy Probe launched in

2001 with the first results in February 2003 [4, 5] with a nine year update
published in June 2013 [6, 7].
21
22
Useful Expressions
In a flat FRW Universe:
ȧ
H =
a
N = log a
0 d
=
dN
d d
= H
dt dN
3H 2
ρcrit =
8πG
= ρr + ρm + ρeff
3H 2 = 8πGρr + 8πGρm + 8πGρeff

ρr 8πGρr
Ωr = =
ρcrit 3H 2
ρm 8πGρm
Ωm = =
ρcrit 3H 2
Ωeff = 1 − Ωm − Ωr
8πGρeff
=
3H 2
ρ0eff
weff = −1 −
3ρeff
2H 0 H02 e−4N Ωr,0
weff Ωeff = −1 − −
3H 3H 2
8πGρeff
H 2 = H02 e−4N Ωr,0 + H02 e−3N Ωm,0 +
3
23
R = 12H 2 + 6Ḣ
= 12H 2 + 6HH 0
= 3H02 e−3N Ωm,0 + 8πG (4ρeff + ρ0eff )
= 3H02 e−3N Ωm,0 + 8πGρeff (1 − 3weff )
= 3H 2 (1 − 3weff Ωeff ) − 3H02 e−4N Ωr,0

ä 4πG
= − (2ρr + ρm + (1 + 3weff ) ρeff )
a 3
1 + Ωr
ä > 0 ⇒ weff < −
3Ωeff
24
For Ruth and our grandchildren,
Charlotte, Hannah, Danny and Sophie
25
26
Chapter 1
Introduction to Gravitational
Theories and Cosmology
1.1 Introduction
This thesis is about some aspects of modifying the gravitational theory of General
Relativity (GR) in the light of some well known problems. GR is not a particle
theory but makes the suggestion that matter and other energy sources curve
space-time which puts it at variance with traditional, Galilean and Newtonian
mechanics. However, low energy, non-relativistic sources must and do, in their
effects, tend to their Newtonian counterparts.
What we consider in this thesis are some aspects of what is termed F (R) gravity
but first, in this chapter, we review GR and consider a very brief history of the
Universe. We then look at what is termed dark energy, consider a couple of
perceived problems, consider matter perturbations and list the constraints which
act on F (R) models.
27
28 Chapter 1. Introduction to Gravitational Theories and Cosmology
1.2 General Relativity
1.2.1 General Relativity, covariant derviatives, energy-

momentum tensor
Our starting point [8–11] is to generalise Poisson’s equation ∇2 Φ = 4πGρ for

mass density ρ; Φ is the equivalent of the Newtonian potential such that, for a
single mass m, it produces a gravitational potential a distance r away from it
of Φ = −Gm/r, G being Newton’s gravitational constant. We wish to find the
General Relativity (GR) equivalent of Poisson’s equation. It is a second order
differential equation which, in GR, should be a tensorial equation. It will become
a second order differential equation involving the covariant derivative, ∇µ . As is
usual, let the metric for GR be gµν such that
ds2 = gµν dxµ dxν , (1.1)
with x0 representing time and xi being the space components. We use the con-
vention (−1, 1, 1, 1) for the signature of the metric and put c = 1.
The covariant derivative is defined on a vector V µ via the Levi-Civita connection
∇µ V ν = ∂µ V ν + Γνµλ V λ , (1.2)
where Γνµλ are termed the Christoffel symbols of the second kind. On a scalar,
φ, this reduces to
∇µ φ = ∂µ φ. (1.3)
Applied to tensor T µν , it gives
∇µ T νρ = ∂µ T νρ + Γνµλ T λρ + Γρµλ T µλ . (1.4)
A property we use is that the covariant derivative of the metric is zero, i.e.,
∇ρ gµν = 0. This means that ∇ρ g µν = 0 and that the Levi-Civita connection can,
1.2. General Relativity 29
as a result, be defined by
1 σρ
Γσµν = g (∂µ gνρ + ∂ν gρµ − ∂ρ gµν ) . (1.5)
2
In (1.4), we have used the torsion-free requirement that Γλνµ = Γλµν . This
being the case, we describe the connection as being torsion free. If this were not
true we could form the “torsion tensor”, Γλµν − Γλνµ , which is anti-symmetric in
its lower indices [11].
Now the tensor equivalent to the mass density ρ is the energy-momentum tensor,
Tµν , mass and energy being synonymous in Special Relativity. It is defined by
√
2 δ ( −g ) LM
Tµν = − √ , (1.6)
−g δg µν
in which LM is the Lagrangian for all the physical fields contributing energy to
the Universe and g is the determinant of gµν .
In cosmology we assume the expansion of the Universe in controlled by the

presence of perfect fluids. A perfect fluid is completely specified by its rest-frame
density ρ and isotropic rest-frame pressure P . When the fluid has four-velocity
U µ the energy-momentum tensor can be expressed as
Tµν = (ρ + P ) Uµ Uν + P gµν . (1.7)
With respect to comoving coordinates, U µ = (1, 0, 0, 0), T µ ν = diag (−ρ, P, P, P ).

In Special Relativity, the energy-momentum conservation equation is ∂µ T µν = 0.
In GR, this becomes ∇µ T µν = 0 from which also follows ∇µ Tµν = 0.
1.2.2 The Riemann tensor and curvature and the Einstein

tensor
The Riemann tensor quantifies the curvature of space-time. It is Rρσµν = gρλ Rλσµν
where
Rρσµν = ∂µ Γρσν − ∂ν Γρσµ + Γρλµ Γλσν − Γρλν Γλσµ . (1.8)

It has the following properties:
Rρσµν = −Rσρµν , (1.9)
Rρσµν = −Rρσνµ , (1.10)
Rρσµν = Rµνρσ , (1.11)
Rρσµν + Rρµνσ + Rρνσµ = 0. (1.12)
The Bianchi identity holds:
∇λ Rρσµν + ∇ρ Rσλµν + ∇σ Rλρµν = 0. (1.13)
By contracting the Riemann tensor, the Ricci tensor is formed:
Rµν = Rλµλν , (1.14)
and the trace of this is the Ricci or curvature scalar:
R = g µν Rµν . (1.15)
Contracting the Bianchi identity twice:
g µλ g νσ (∇λ Rρσµν + ∇ρ Rσλµν + ∇σ Rλρµν ) = 0 (1.16)
⇒ ∇µ Rρµ − ∇ρ R + ∇ν Rρν = 0 (1.17)

1
⇐⇒ ∇µ Rρµ = ∇ρ R. (1.18)
2
From the Ricci tensor and the Ricci scalar is defined the Einstein tensor:
1
Gµν = Rµν − gµν R. (1.19)
2
From (1.18), we see that ∇µ Gµν = 0, just as ∇µ Tµν = 0.
1.2.3 Formulation of the Einstein equations
In view of the property expressed in (1.19), and that we want a tensor indicating
the curvature of space-time to relate to the energy-momentum tensor, we consider
the equality
Gµν = κTµν (1.20)

where κ = 8πG, and see if this is viable as a law of Physics. Let us consider our
local, Minkowskian, frame of reference with metric ηµν which the gravitational
effects of static, local masses have altered to gµν . It will be a small effect so we
shall consider only first-order changes to ηµν . For matter, by which we mean
baryons and dark matter, P = 0 and the energy-momentum tensor is given by
T µν = diag (−ρ, 0, 0, 0) and the trace of (1.20) gives R = −κT = κρ.
The 00 component of (1.20) gives:
1
R00 + R = κρ (1.21)
2
1
⇐⇒ R00 = κρ (1.22)
2
Now, and since we can ignore time derivatives in a static field,
R00 = Rλ0λ0 (1.23)
= Ri 0i0 (1.24)
= δii Ri 0j0 (1.25)
= δij ∂ j Γi 00 , to first order, (1.26)

1
= − ∂ i ∂i g00 . (1.27)
2
g00 = −1 + h00 , say, means that, to first order small quantities, R00 = −∇2 h00 /2.
This gives, as the 00 component of (1.20),
∇2 h00 = −κρ. (1.28)
Thus, identifying −h00 /2 with Φ gives Newtonian gravity, i.e., the Einstein equa-
tion reduces in the weak field limit to
κ
∇2 Φ = ρ (1.29)
2
= 4πGρ. (1.30)
We also have the possibility that, more generally,
1
Gµν = Rµν − gµν R = 8πGTµν . (1.31)
2
These are Einstein’s equations. The trace of them gives
R = −8πGT (1.32)
= 8πG (ρ − 3P ) . (1.33)
1.2.4 The FRW metric and the Ricci scalar
Observation of the Universe suggests that a good, large-scale model is to be had

by postulating it to be expanding and, on average, spatially homogeneous and
isotropic, that is, that matter is spread out uniformly throughout the Universe in
all directions no matter from which point of the manifold one is looking. Observa-
tions support the idea that space-time has no intrinsic curvature of its own which
is thus described as being flat, which will be assumed throughout this thesis. In
the past this has been perceived as being a problem and is discussed in [12]. The
assumptions that the Universe is flat, homogeneous, isotropic and expanding is
is incorporated in the Friedmann-Lemaı̂tre-Robertson-Walker (FRW) metric de-
fined by gµν = diag(−1, a2 , a2 , a2 ). a is a parameter termed the expansion scale
factor of the Universe. Initially, at the Big Bang, a = 0 and it increases with time
to become, by definition, a = 1, today. For reviews see, for example, [13, 14].
The relative expansion rate of the Universe is the Hubble parameter, H = ȧ/a.
If H is known, we can compute the age of the Universe. Today’s value is H0
determined in the recent Planck Collaboration as H0 = 100 h kms−1 Mpc−1 =
67.4 ± 1.4 kms−1 Mpc−1 [15–17].
Denoting da/dt by ȧ, gives the following as the only non-zero Christoffel sym-
bols:
Γ011 = Γ022 = Γ033 = aȧ. (1.34)

ȧ
Γ110 = Γ101 = Γ220 = Γ202 = Γ330 = Γ303 = . (1.35)
a
Replacing ȧ by aH, these Christoffel symbols lead to the Riemann tensor as

3ä 2 2 2
Rµν = diag − , aä + 2ȧ , aä + 2ȧ , aä + 2ȧ (1.36)
a

2 2 2 2 2 2 2
= diag −3 H + Ḣ , a 3H + Ḣ , a 3H + Ḣ , a 3H + Ḣ
(1.37)
and the Ricci scalar as R = 12H 2 + 6Ḣ.
1.2.5 Density, pressure and equation of state
In the expression for the energy-momentum tensor in subsection 1.2.1, mention

is made of the energy density, ρ, and associated pressure, P , of the perfect fluids
which are present in the Universe at any time. These fluids are radiation, in the
form of photons, and other relativistic species, with density ρr , and matter, with
density ρm . Some authors also admit to a vacuum or dark energy of density ρvac .
Dark energy is discussed in subsection 1.4.4. Each fluid may exert a pressure
and pressure is connected to fluid density by an equation of state, P = wρ. w
is termed the equation of state parameter and could vary with time, as we shall
see. For radiation, w = 1/3 so that Pr = ρr /3, for matter w = 0 so Pm = 0, while
for the vacuum, w = −1 so that Pvac = −ρvac .
Looking at the 0 component the conservation of energy equation, ∇µ Tµν = 0,

gives
∇µ Tµ0 = 0, (1.38)
⇒ ∂µ T µ0 + Γµµλ T λ0 − Γλµ0 T µλ = 0, (1.39)
⇐⇒ −∂0 ρ − 3H (ρ + P ) = 0. (1.40)
using (1.34) and (1.35). (1.40) is true for each fluid in the energy-momentum
tensor so that we have, for each,
ρ̇ = −3H (1 + w) ρ. (1.41)
If w is constant for any fluid then we can integrate (1.41) to
ρ = ρ0 a−3(1+w) , (1.42)
where ρ0 is a constant of integration. Thus for radiation, ρr ∝ a−4 , for matter,

ρm ∝ a−3 , and the density of the vacuum, ρvac , is constant.
1.2.6 The Friedmann equations
If we say that the densities and pressures in the energy-momentum tensor are
given by ρ and by P , respectively, solving the Einstein equations gives, from the
00 component
3H 2 = 8πGρ, (1.43)
2
ȧ 8πG
⇐⇒ = ρ, (1.44)
a 3
which is referred to as the Friedmann equation. The ij components give
3H 2 + 2Ḣ = −8πGP, (1.45)

ä 4πG
⇒ = − (ρ + 3P ) . (1.46)
a 3
This last equation is referred to as the second Friedmann equation.
1.2.7 Einstein equations derived from the action
There is an alternative way, proposed by Hilbert [10, 18], of deriving the Einstein
equations and that is by considering the total action of the gravitational field
and the action due to sources of energy, of whatever kind, which we call matter
fields. Let the total action be S = SG + SM . The gravitational part of the action
is defined to be
√
Z
1
SG = d4 x −g R, (1.47)
2κ
where κ = 8πG, as in section 1.2.3, above. The action for the matter fields is
√
Z
1
SM = d4 x −g LM , (1.48)
2κ
where LM is the Lagrangian for the matter fields and it is connected to the
energy-momentum tensor by equation (1.6). Varying SG with respect to the
inverse metric, g µν ,
√
Z
1
d4 x δ −g R

δSG = (1.49)
2κ
√
Z
1
d4 x δ −g g µν Rµν

= (1.50)
2κ
√ √ √
Z
1
d4 x δ −g g µν Rµν + −g δ (g µν ) Rµν + −g g µν δRµν .

=
2κ
(1.51)
Without going into too much detail,
√ 1
δ −g = − √ δg (1.52)
2 −g
g
= − √ g µν δgµν (1.53)
2 −g
√
−g
= − gµν δg µν ). (1.54)
2
δRµν = δRλµλν (1.55)
= ∇λ δΓλνµ − ∇ν δΓλλµ , (1.56)
⇒ g µν δRµν = ∇λ g µν δΓλνµ − ∇ν g µν Γλλµ

(1.57)
= ∇σ g µν δΓσνµ − g µσ δΓλλµ .

(1.58)
Thus the right hand term of (1.51) gives, by Stokes’s Theorem, a boundary term.
We can set the boundary at infinity so this term vanishes. This leaves
√ √
Z
1
d4 x δ −g g µν Rµν + −g δ (g µν ) Rµν

δSG = (1.59)
2κ
√
Z
1 1
= d x −g − R gµν + Rµν δg µν .
4
(1.60)
2κ 2
For the matter fields,
√
δ ( −g LM ) µν
Z
4
δSM = dx δg (1.61)
δg µν
√
Z
1
= − d4 x −g Tµν δg µν . (1.62)
2
so,
δS
= 0, (1.63)
δg µν
1
⇒ Rµν − gµν R − κTµν = 0, (1.64)
2
1
⇐⇒ Rµν − gµν R = 8πGTµν . (1.65)
2
Varying the action in this way enables us to produce our own versions of Ein-
stein’s equations, which is what we do when we consider actions which modify
gravity in section 1.4.
1.3 A brief history of the Universe
When considering the large scale structure of the Universe, as we do with modified
gravity, we simply split up its history into the eras when the various energies
represented in the energy-momentum tensor dominate. A detailed and scholarly
account is given in Kolb and Turner [19]. Initially, there was the radiation era
which was followed by the matter era. We include a third, the vacuum era,
because the densities of radiation and matter eventually become negligible leaving
just the vacuum. Conjectures on the past and future of the Universe are discussed
in [20].
1.3. A brief history of the Universe 37
1.3.1 The radiation era
It is believed that immediately after the Universe came into being it underwent
a brief period of very rapid expansion which smoothed out any original uneven-
ness there might have been. This period is termed inflation and lasted until
the Universe was about 10−32 s old. The radiation era stretched from just after
inflation, to around 1011 s ≈ 3000 years during which time the temperature fell
from 1013 GeV to 10eV. It ended when the radiation density, which we have seen
is proportional to a−4 , fell to such an extent that it equated to the matter density
which was falling at a slower rate. The point at which this happened is termed
equality. The following two epochs formed part of the radiation era.
1. Reheating occurred between 10−32 s to 1s as the temperature fell from

1013 GeV to 10MeV. It saw the creation of a dense soup of quarks, photons,
gluons and leptons. Quarks and gluons formed hadrons which decayed to
protons, electrons, photons and neutrinos. There was a slight imbalance of
matter to antimatter of the order of 1 + 10−10 : 1. This annihilated to leave
a small residue of matter plus photons and neutrinos.
2. Big Bang Nucleosynthesis (BBN) then followed in which hydrogen and some
of the lighter elements such as deuterium, helium and lithium formed. This
epoch lasted until 200s when the temperature decreased to 0.1MeV.
1.3.2 The matter dominated era and the CMB
This era stretched from when the Universe was 3000 years old until it was 109
years old. During that time, the temperature fell from 10eV to 10meV. Initially,
the Universe was opaque to light but, as cooling progressed, in the process known
as recombination, when electrons and protons formed the light atoms, photons
became decoupled to make the Universe transparent to light. We observe this
today as the Cosmic Microwave Background (CMB) at z ≈ 103 [21, 22]. The
CMB was studied in the recent Planck survey [3] following earlier measurements
by COBE [23] and WMAP [6, 7]. From the Integrated Sachs-Wolfe (ISW) plateau
of the CMB, [24], is determined the value of the dark energy density parameter
and its equation of state parameter, [25]. With respect to the potentials, Φ and
Ψ in the Newtonian gauge, of chapter 5, Amendola et al., in [26], state that Φ+Ψ
could be measured using the ISW effect and from weak lensing while Ψ might
be measured via measurements of the velocities of galaxies. The acoustic peaks
of the CMB give us information about the various density parameters. From the
first acoustic peak, it is found that the Universe is almost spatially flat, which we
assume in this thesis. The second acoustic peak gives constraints on the baryon
density and gives a value to the baryon density parameter while the third acoustic
peak gives a value for the total matter density parameter.
1.3.3 The vacuum era
This is the era in which we now find ourselves. It commenced at around z = 9,

109 years after the Big Bang, when stars and galaxies were formed as were the
heavier elements via supernovae. In this epoch the expansion of the Universe
began to accelerate [27–29].
1.4 Modifying gravity with F (R) gravity models
It is mainly because of the cosmological constant problem (see subsection 1.4.1)

but also because observations do not confirm that the effective, or dark energy,
equation of state parameter is necessarily equal to −1, that F (R) gravity models
have been studied. It is shown that by exchanging R for F (R) in the gravita-
tional action then, without resorting to any vacuum energy whatsoever, the late,
accelerated expansion of the Universe can be accounted for. F (R) gravity models
yield an effective energy which replaces the phenomenon known as dark energy.
1.4. Modifying gravity with F (R) gravity models 39
For a review of F (R) gravity, see [30–34].
In modified F (R) gravity, we obtain the equivalent of the Einstein equations

by replacing R in (1.47) with F (R) in the gravitation action. Varying the total
action with respect to g µν gives
1
gµν FR (R) − ∇µ ∇ν FR (R) + Rµν FR (R) − gµν F (R) = 8πGTµν , (1.66)
2
where FR (R) denotes dF (R)/dR and Tµν contains only contributions from ra-
diation and matter. These are the field equations for F (R) and are the mod-
ified gravity equivalent to the Einstein equations. It can be seen that putting
F (R) = R reduces (1.66) to the Einstein equations, (1.31).
Equation (1.66) can be derived as follows. Using the notation of subsection

1.2.7,
√
Z
1
SG = d4 x −g F (R), (1.67)
2κ
√ √
Z
1
d4 x F (R)δ −g + −g δF (R)

⇒ δSG = (1.68)
2κ
√
Z
1 4 1 µν µν
= d x −g − gµν F (R)δg + FR (R)δ (g Rµν ) (1.69)
2κ 2
√
Z
1 4 1 µν µν µν
= d x −g − gµν F (R)δg + FR (R)Rµν δg + FR (R)g δRµν ,
2κ 2
(1.70)
using (1.54) and (1.15). The right hand term of (1.70) is
√ √
Z Z
1 1
d x −g FR (R)g µν δRµν =
4
d4 x −g FR (R)∇σ g µν δΓσµν − g µσ δΓλλµ

2κ 2κ
√
Z
1
= d4 x −g FR (R) [gµν δg µν − ∇µ ∇ν δg µν ]
2κ
√
Z
1
= d4 x −g [gµν FR (R) − ∇µ ∇ν FR (R)] δg µν ,
2κ
(1.71)
using (1.58) and after some manipulation involving integration by parts and em-
ploying Stoke’s Theorem [35, 36]. Thus,
√
Z
1 1
δSG = d x −g − gµν F (R) + Rµν FR (R) + gµν FR (R) − ∇µ ∇ν FR (R) δg µν
4
2κ 2
(1.72)
from which, using (1.63), (1.66) follows.
Taking the 00 component of (1.66) gives

d2 FR (R) 1
−FR (R) − 2
+ R00 FR (R) + F (R) = 8πGT00 . (1.73)
dt 2
Now, for the FRW metric of subsection 1.2.4, = −∂02 − 3H∂0 , as there is no
spatial dependence. Thus,
d 1
3H FR (R) − 3 H 2 + Ḣ FR (R) + F (R) = 8πG (ρr + ρm ) ,
dt 2
(1.74)
1
⇐⇒ −3H ṘFRR (R) + 3 H 2 + Ḣ FR (R) − F (R) = −8πG (ρr + ρm ) ,
2
(1.75)
where FRR (R) = d2 F (R)/dR2 . Taking the trace of (1.66), gives
3FR (R) + RFR (R) − 2F (R) = 8πGT
= 8πG (−ρ + 3P ) , (1.76)
in which, of course, the only pressure term on the right hand side is due to
radiation. Observe that (1.75) is a second order differential equation in F (R)
while (1.76) is third order. When we come to solving such equations we shall
generally solve (1.75) which will be loosely referred to as the field equation.
1.4.1 A special case
If we put F (R) = R − 2Λ, where Λ is some constant, into (1.66) we have

1
Rµν − gµν R + Λgµν = 8πGTµν . (1.77)
2
It is noticed that Einstein could have had these as his original equations in place
of (1.19) because the covariant derivative of the left hand side is zero just as it is
for the left hand side of (1.19).
If we subtract Λgµν from both sides and absorb the term into the energy-
momentum tensor, comparing the new equation with (1.19) would suggest that
Λ could be thought of as contributing to the density of the vacuum, i.e., Λ =
8πGρvac . That being the case, substituting F (R) = R−2Λ into the field equation,
(1.75) gives 3H 2 = 8πG (ρr + ρm + ρvac ), the Friedmann equation. But there is a
problem with this.
At late time when the effects of radiation and matter may be considered to
p
be negligible, H = ȧ/a ' 8πGρvac /3 , which is constant. Solution of this
yields a solution for a as an exponential function of time with, ä = aH 2 ; i.e., an
accelerating, expanding Universe. If we equate 8πGρvac with 3H02 Ωvac ≈ 2.1H02 ,
we find that ρvac = 5 × 10−47 GeV4 . The problem is that, if we consider all the
energy available in the vacuum, it is of the order of 10120 times this [37–39] so
why is only a small faction of the available energy being used to accelerate the
expansion? See [40–42] for the history of this. This is a reason why do not accept
any vacuum energy in the energy-momentum tensor when we modify gravity.
This is not to decry attempts to come to terms with this cosmological problem
[43].
Historically, Λ, originally denoted λ, was introduced into the Einstein equations

[44] to solve a problem perceived by Lemaı̂tre that without it the Universe would
expand while, at the time (1916-1919), Einstein thought that the Universe was
static. It was abandoned, however, when Hubble discovered, in 1929, that the
Universe was expanding. Today, we still make use of Λ and, when we do, we term
it the cosmological constant. The problem referred to in the previous paragraph
is termed the cosmological constant problem. We term F (R) = R − 2Λ ΛCDM,
which is based on GR, with Λ 6= 0.
1.4.2 A vacuum solution
It is noticed that when all the energy content of the energy-momentum tensor
has run out, at the end of the vacuum era, (1.33) gives R = 0. This, of course,
assumes that there is no residual vacuum energy, then. It is desired that this,
too, should be a solution of (1.76). If there is a time independent solution,
then we have R FR (R) − 2F (R) = 0 and so F (0) = 0. It is also possible that
R FR (R) − 2F (R) = 0 has an attractor solution R > 0. This is termed the de
Sitter solution [45, 46]. There could, instead, be a time-dependent solution which
would yield oscillations [47]. See subsection 3.2.1.2 for an example of this.
1.4.3 A modified expression for the Hubble parameter

and the effective equation of state
It is clear by comparing (1.75) with (1.77) that modifying gravity will produce
a term in the expression for H 2 which might be taken to be a form of energy
missing from the energy-momentum tensor. So, rather than using the term,
ρvac , which suggests it represents the vacuum, let us term it an effective energy
density, ρeff . We can assume it to have the properties of a perfect fluid with
pressure, Peff , and equation of state parameter, weff , such that Peff = weff ρeff and
ρ̇eff = −3H (1 + weff ) ρeff . In allusion to the earlier expression we had for H 2 ,
namely, 3H 2 = 8πG (ρr + ρm + ρvac ) let
8πG
H2 = (ρr + ρm + ρeff ) , (1.78)
3
⇐⇒ ρcrit = ρr + ρm + ρeff . (1.79)
where 8πGρcrit = 3H 2 . Define also the density parameters for radiation and
matter by Ωr = ρr /ρcrit and Ωm = ρm /ρcrit , respectively, and replace ρeff /ρcrit by
Ωeff , and we have
1 = Ωr + Ωm + Ωeff . (1.80)
We define Ωeff from (1.80) and it equals 8πGρeff / (3H 2 ). We know from subsec-
tion 1.2.5 that ρr = ρr,0 a−4 and ρm = ρm,0 a−3 , where ρr,0 and ρm,0 are today’s
values of ρr and ρm , respectively, because we define today by putting a = 1. This
leads us to
8πG
H2 = ρr,0 a−4 + ρm,0 a−3 + ρeff

(1.81)
3
8πG
= H02 Ωr,0 a−4 + Ωm,0 a−3 + ρeff , (1.82)
3
where H0 is today’s value of the Hubble parameter, termed the Hubble constant.
In this thesis, rather than taking time to be our universal parameter, we use
N = log a, which is dimensionless. If we denote differentiation with respect to N
by 0 , then d/dt = Hd/dN so that ġ = H g 0 , for generic function g(t). Then
ρ̇eff = −3H (1 + weff ) ρeff (1.83)

ρ0
⇒ weff = −1 − eff (1.84)
3ρeff
from which follows
2H 0 H02 e−4N Ωr,0

weff Ωeff = −1 − − (1.85)
3H 3H 2
0
2H
≈ −1 − , in the matter and vacuum eras. (1.86)
3H
In F (R) models, it is a fact that weff decreases very slowly from being −1 at the
start of the radiation era and continues to become more negative into the matter
era. This is proved in Chapter 4. We say that, at early times, weff lies below the
phantom divide so that weff < −1 for much of its history and until recent times
[48–50].
1.4.4 Dark energy
Dark energy [37, 51–53] is the name given to what might be vacuum energy or
the effective energy derived from the geometrical nature of modifying gravity.
It was mooted, in 1998, after the discovery that the Universe’s expansion was
accelerating [40, 54, 55]. Without dark energy [56], the expansion would slow
down and eventually come to a halt as matter runs out. Dark energy is defined
to be the fluid which causes the expansion of the Universe to accelerate. Theories
suggest that it was dark energy which was responsible for the initial inflation
of the Universe [12] but we are concerned with what happened after inflation
and since the beginning of the radiation era. It has been known for a long time
that the equation of state parameter for dark energy, which we dub, weff , is close
to −1 suggesting that ρeff , if it varies, varies slowly and eventually becomes the
dominant energy in the Universe as the matter density, being proportional to a−3 ,
continues to decrease [57].
Possible ways of accounting for dark energy are contained in [37, 52, 58–63].
In this thesis the F (R) gravity option has been studied but there are two other
main contenders.
1.4.4.1 The cosmological constant
This is a strong contender to explain the accelerated expansion of the Universe as

even the latest measurements [3, 17] do not rule it out. The measurements are not
definitive in making the equation of state parameter tightly constrained around
−1 but have wvac,0 = −1.04+0.72
−0.69 at the 95% significance level and dwvac,0 /da <
1.32, at the same level when pure ΛCDM would give dwvac,0 /da = 0.
There is, however, the cosmological constant problem to worry about as men-
tioned in subsection 1.4.1.
1.4.4.2 Quintessence
Briefly, in this theory, dark energy is modelled by a slow-rolling scalar field φ

coupled to potential V (φ) [64–67]. The equation of state parameter is given by
φ̇2
1 + wvac = . (1.87)
φ̇2 /2 + V (φ)
Initially, at high z, φ̇ ∼ 0, because we have ΛCDM. As time progresses, φ̇2
increases and, providing V (φ) > 0, so does 1 + wvac , that is 1 + wvac > 0, for
all time. Some quintessence models have a phantom potential, V (φ) < 0, which
must decrease such that V (φ) < −φ̇2 /2. In this case, 1 + wvac < 0 for all time. So
one feature of quintessence models is that wvac remains firmly on one side of the
phantom divide. This is unlike the case for F (R) models for which the effective
equivalent to wvac is less than −1 before the present day [68].
1.4.4.3 Effective dark energy
If we let F (R) = R + f (R), substitute it into equation (1.66) and rearrange it so

that it resembles the form of (1.65), where the left hand sides of both equations
agree, we could say that F (R) gravity creates a fluid; effective dark energy with
density ρeff . It also has equation of state parameter, weff , and density parameter,
Ωeff as have been mentioned at the beginning of this subsection. One aspect of
effective dark energy is that, for models which tend to ΛCDM as R → ∞, weff
starts from being −1, decreases into the phantom zone and continues to do so
until a local minimum is reached when it returns to be closer to −1 today, [48, 69].
1.4.5 The modified gravity particle
Associated with F (R) gravity is a massive particle dubbed the scalaron [70] which
has mass M defined by
FR (R) − RFRR (R)
M2 = . (1.88)
3FRR (R)
Owing to the various constraints on F (R), this is often approximated to M 2 =

1/ [3FRR (R)]. Generally, F (R) does not deviate greatly from ΛCDM, for which
F (R) = R − 2Λ, so the mass of the scalaron is generally large. For models where
the deviation from ΛCDM increases with time, then the mass of the scalaron
decreases with time.
Where there are oscillations of the weak field potentials (Chapter 5), the fre-
quency of oscillation is shown to be proportional to 1/M at high z, in the early
stages of the matter era, which increases as z increases backwards in time.
1.4.6 The coincidence problem
This is simply the question as to why is Ωeff,0 ≈ 0.7 of the same order of magnitude
as Ωm . If we look at a graph of Ω0eff against N , Figure 1.1, we see that Ω0eff is
close to zero for much of the history and future, so why does “now”, i.e., a = 1,
N = 0, occur when Ω0eff is close to its maximum?
Weff HNL, Wm HNL W¢ eff HNL

1.0
0.8 0.6
0.6
0.4
0.4
0.2
0.2
N N
-10 -5 5 10 -10 -5 5 10
(a) (b)
Figure 1.1: The coincidence problem. (a) Evolution of the matter and effective density
parameters. (b) Ω0eff against N . The local maximum in Ω0eff occurs at around z = 0.3.
1.4.7 The growth of matter and perturbations
In Chapter 3, subsection 3.2.1.3, the standard matter perturbation equation is

solved numerically for the various models used in that chapter. It yields the time-
dependent parameter, γ, [56, 71–75] which indicates the expansion growth history
of matter. γ is termed the matter growth index and is very sensitive to models in
that different models representing the same expansion history will yield different
values of γ at any given stage in the history of the Universe. This is illustrated
especially in Figure 2.6 in which the models illustrated share the same ΛCDM
history. The matter growth indices for F (R) models show features which are
not seen in the matter growth index for ΛCDM. How the ΛCDM matter growth
index evolves can be see in Figure 3.9(b). Ascertaining the value of γ0 would be
evidence in helping decide whether or not ΛCDM is or is not, realistically, the
current theory of gravity.
In Chapter 5, we consider how non-relativistic matter sources perturb the FRW

metric in the Jordan frame, the frame with FRW metric gµν , to produce matter
perturbation potentials. Algebraic approximations, appropriate for all time up
to the present, to the defining equations for these potentials are found and com-
pared with numerical solutions to the same equations. Providing the co-moving
wavenumber, k, is large enough in the sub-horizon regime, we shall see that there
is good agreement.
1.4.8 Constraints on F (R)
Constraints on the parameters of F (R) models are explained in [76–79] as follows:
1. FR (R) > 0, otherwise there would be ghosts, that is, particles with negative
kinetic energy, [80–83].
2. FRR (R) > 0, otherwise there would be generic instabilities, and the scalaron
mass would be negative or zero, [84–86].
If it is desired to have F (R) → R − 2Λ, as R → ∞ then we must have [76]:
3. FR (R) < 1.
4. FR (R) → 1 as R → ∞.
Then there are local gravity constraints, discussed in [33], of which the severest
is the Solar System constraint. Local gravity constraints are derived from the
effective Newtonian gravitation constant which is derived under a weak field ap-
proximation by considering a spherical mass of constant density, ρ, surrounded by
matter of very small average density, effectively zero. As shown in [33, 49], using
linear perturbation theory in a Minkowskian background with a perturbation hµν
such that F perturbs to F + δF , FR perturbs to FR + δFR , etc, this gives for the
effective Newtonian gravitation constant
e−M l

G
Geff = 1+ (1.89)
FR (R) 3
where l is the scale of the experiment and M is the scalaron mass as defined in
(1.88), which can be approximated by M 2 ≈ 1/ (3FRR (R)) when FR (R) ≈ 1 and
RFRR (R) 1, which will be true in the high curvature regimes. M l 1 (see
[49]) which translates in [87] to
2
ρ l
m (1.90)
ρ0 H0−1
where m = RFRR (R)/FR (R) ≈ RFRR (R).
For the Solar System, the scale of the experiment is 1AU = 1.5 × 1013 cm,
ρ ∼ 10−23 g/cm3 and (1.90) gives m 10−23 . Using R = 8πGρ, this density
corresponds to R ∼ 106 H02 . Weaker constraints come from the Shapiro time-
delay effect and laboratory, Cavendish-type experiments to establish the value
of G. In the Shapiro case, l ∼ 7 × 1010 cm and ρ ∼ 10−15 g/cm3 which give
m 10−20 and R ∼ 1014 H02 . In Cavendish-type experiments, l ∼ 10−2 cm and
ρ ∼ 10−12 g/cm3 to give m 10−43 and R ∼ 1017 H02 .
To see which is of these constraints is the strongest, we model f (R) as

−n
2 R
f (R) = −2Λ + µ H0 (1.91)
H02
where n is very large and positive because, at these values of R, F (R) very closely
approximates to ΛCDM. Keeping n constant, in each of the three cases, let us
see what constraint the condition on m places on µ.
The Solar System constraint, for which m 10−23 , gives n (n + 1) µ 106n−17 .

The Shapiro case gives n (n + 1) µ 1014n−6 while the Cavendish case gives
n (n + 1) µ 1017n−26 . Thus, providing, n > 9/11, the tightest constraint put
upon µ is from the Solar System constraint.
All viable F (R) models that tend to standard Einstein Gravity should auto-
matically, by their construction, satisfy constraints 1. to 4. Some F (R) models
need only satisfy constraints 1. and 2. but all must suffer the application of the
local gravity constraint which will constrain the model’s parameters.
1.4.9 Conditions for an accelerating Universe
From (1.46), the expansion of the Universe continues to accelerate provided ρ +

3P < 0. Ignoring radiation, this gives us in F (R) gravity, 3weff Ωeff < −1. This
means, from Useful Expressions on page 24, that R > 6H 2 . Thus any solution
which has R < 6H 2 or R oscillating and going below 6H 2 , as with the example
of Figure 3.8, represents a more complicated Universe than one in which the rate
of expansion is monotonic.
1.4.10 Probing the Growth of Large Scale Structure
Despite the Universe being being isotropic on very large scales, perturbations,
which owe their origin to times before recombination (section 1.3.2), caused mat-
ter, after recombination, to clump into galaxies and clusters of galaxies. Thereby
also formed voids, to give the structure we see today from surveys such as the
Two Degree Field Galaxy Redshift Survey (2dfGRS) [88] and the Sloan Digital
Sky Survey (SDSS) [89]. Measurements from these surveys provide tests for the
existence of effective dark energy. These and other surveys probe at redshifts
lower, often considerably lower, than that at which the CMB was formed. This is
useful because effective dark energy makes its effects known during the vacuum
and later matter eras.
Large scale structures (LSS) consist mainly of dark matter and observational
probes consist of weak lensing surveys, cluster counts and measurements of baryon
acoustic oscillations (BAO). The results are compared with numerical simulations
for various models of effective dark energy. Ongoing surveys include SDSS-III
[90] and BOSS (Baryon-Oscillation Spectroscopic Survey) [91]. BOSS will supply
stand-alone constraints on the properties of effective dark energy, Quasi-Stellar
Objects (QSO), H(z) and provide insights into the matter content of the Universe.
BOSS operates at two ranges of redshift. First, 0.2 < z < 0.8 on 1.5 million
luminous red galaxies and, secondly, 2.3 < z < 2.8 on 160, 000 QSOs at precisions
in excess of 1.5%.
For any large scale structure being probed, using the acoustic peak as a standard
ruler, two ratios are measured which give the angular diameter distance, dA (z)
[52], and H(z), independently of each other. Also can be found, by observing the
power spectrum of the structure, is the wavenumber at equality, keq , given by
s
2Ωm,0
keq = H0 (1.92)
aeq
where aeq is the scale factor at equality. In these cases, simple scenarios can
be tested for dark energy, e.g., the CDM model, in which Ωm,0 ∼ 1.0, or the
ΛCDM model, for which, Ωm,0 ∼ 0.3, with the result that dark energy models are
favoured above CDM models.
Ongoing is also the Dark Energy Survey (DES)1 which will combine probes of
Type 1a supernovae, BAO, Galaxy Clusters and Weak Lensing. For the future
are planned the EUCLID satellite2 , which will, from 2020, look at the distance –
redshift relationship and the evolution of cosmic structures up to z ∼ 2, and the
wide aperture Large Synoptic Survey Telescope (LSST)3 in Chile which, from
2022, will map over a period of ten years the sky and the Galaxy and detect
transient events.
1
http://www.darkenergysurvey.org
2
http://sci.esa.int/euclid
3
http://www.lsst.org/lsst
Chapter 2
Gravitation Models Which

Mimic ΛCDM
2.1 Introduction
The standard flat ΛCDM model has F (R) = R − 2Λ, as discussed in Chapter 1.
All F (R) models must solve the underlying field equation, (1.74). If 3Ḣ is replaced
by R/2 − 6H 2 and the dependence of ρr and ρm on the expansion scale factor, a,
introduced, using (1.78) and (1.82), equation (1.74) becomes

d R 2 1 2 Ωr,0 Ωm,0
−3H FR (R) + − 3H FR (R) − F (R) = −3H0 + 3 .
dt 2 2 a4 a
(2.1)
Generally, when solving this equation for a specific model, F (R), we re-write it as
a relationship between H(N ) and N = log a, eliminating R via R = 12H 2 +6HH 0 .
When solving for those F (R) with a given ΛCDM history, we can simplify it to
give a relationship between F (R) and R, as in (2.6).
If F (R) = R − 2Λ were inserted into (2.1), the following expression for H 2
53
54 Chapter 2. Gravitation Models Which Mimic ΛCDM
would be obtained:
H02 Ωr,0 H02 Ωm,0 Λ
H2 = + + , (2.2)
a4 a3 3
for which, using R = 12H 2 + 6H H 0 ,
3H02 Ωm,0
R = + 4Λ. (2.3)
a3
Today, we have Ωr,0 + Ωm,0 + Ωeff,0 = 1 from which, using (2.2), we find H02 Ωeff,0 =
Λ/3.
Models which do not have a ΛCDM history have a Hubble parameter given by
H02 Ωr,0 H02 Ωm,0
H2 = + + H02 Ωeff,0 χeff (a), (2.4)
a4 a3
where χeff (a) characterises how ρeff (a) scales with a such that χeff (1) = 1.
In this chapter, we aim to find algebraic forms for F (R) for all models with a
ΛCDM history such that (2.2) holds. Clearly, F (R) = R − 2Λ is one solution but
we show it is not unique. Solving the field equation, algebraically, through all
eras, viz, radiation, matter and vacuum, would seem to be impossible but, if we
restrict ourselves to the later two eras, then an algebraic solution is attainable,
as is shown below.
There has been some discussion, for example in [92], as to whether or not
solutions other than F (R) = R − 2Λ exist in all eras, with the result that there
are. Fay et al., in [92], used the concept of critical points to suggest simple,
approximating expressions for F (R). It has even been suggested, for example
in [93], that there is no real algebraic solution other than F (R) = R − 2Λ but
this is not true. The mistake that Dunsby et al. make in [93] is to choose the
wrong variable with which to express the series solution for F (R). The expression
they find is divergent in terms of this variable. Further, they say that at values
of R ≥ 4Λ, the series makes F (R) complex so the only option is to choose the
constants of integration to be zero leaving the particular integral, F (R) = R−2Λ,
as the only possibility.
2.2. Finding all models which mimic ΛCDM 55
Other than the standard F (R) = R − 2Λ solution, the solutions we find are infi-
nite power series which fall into two categories. First, we solve the field equation
when R > 4Λ. This solution is highly convergent at large R but the convergence
is very slow as R = 4Λ is approached. Secondly, we find a power series solution
for which convergence does include R = 4Λ but the convergence only extends as
far as R = 5Λ. That there is some overlap between these two solutions means
that we must reconcile them with regard to the constants of integration and this
is done next.
Finally, we investigate algebraic solutions to the field equation in the radiation

era, that is large R. Here, we find that, as R increases, F (R) may be approximated
by R or by R4/3 , depending on the parameter choice.
In this chapter, Ωm,0 , will be be taken to be 0.3. If we ignore radiation, this

means that Ωeff,0 = 0.7 and R0 = 9.3H02 .
2.2 Finding all models which mimic ΛCDM
Our goal is to establish all functions, F (R), which lead to the same expansion
history as F (R) = R − 2Λ. We demand that the Universe evolves according to
H 2 = Ωr,0 H02 e−4N + Ωm,0 H02 e−3N + Λ/3, i.e., exactly as for ΛCDM. For any F (R),
R = 12H 2 + 6HH 0 and R0 = −3(R − 4Λ) are determined uniquely once the scale
factor is fixed. This means that equation (2.1) can be written as:

2 R 2 1
9H (R − 4Λ) FRR (R) − 3H − FR (R) − F (R) = −3H 2 + Λ. (2.5)
2 2
In the matter and vacuum eras, we can make the approximation that R ≈ 3H 2 +
3Λ, which allows us to write (2.5) as a differential equation in R, i.e.

R 1
3 (R − 3Λ) (R − 4Λ) FRR (R) − − 3Λ FR (R) − F (R) = −R + 4Λ.
2 2
(2.6)
This has an obvious particular integral, F (R) = R − 2Λ, which will be utilised in
the series solutions, below.
2.2.1 Algebraic solution for R > 4Λ
Define the dimensionless parameter p = Λ/ (R − 3Λ) = Λ/ (3H 2 ) = H02 Ωeff,0 /H 2 .

Given in terms of N = log a, it is p = Ωeff,0 / Ωm,0 e−3N + Ωeff,0 . It obviously

takes finite positive values in the range 0 < p < 1. To start with we work in the
matter era, i.e., we take z < 1000 which means, when we take Ωm,0 = 0.3, that
R < 4 × 108 Λ ≈ 109 H02 , N > −6.9 and p > 2 × 10−9 . Figure 2.1(a) shows how p
varies with N . We shall return to earlier times in section 2.3.
p q
1.0 1.0
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
N N
-4 -3 -2 -1 1 2 0.5 1.0 1.5 2.0
(a) (b)
Figure 2.1: (a) p against N = log a and (b) q against N . q is defined in subsec-
tion 2.2.2. Axes cross at the current time when p = Ωeff,0 = 0.7 and when q ≈ 0.43.
This substitution for R in terms of p converts the homogeneous part of equation

(2.6) into
−6p2 (1 − p) Fpp + p (15p − 13) Fp + F = 0. (2.7)
Looking for solutions of the form
X
F = pn ai p i , (2.8)
i=0
where the coefficients ai are constant, and putting this into (2.7) gives the solu-
tions
F = C pn1 α0 + α1 p + α2 p2 + . . . + D p−n2 β0 + β1 p + β2 p2 + . . . ,(2.9)

√
in which C, D are constants of integration, and n1 = −7 + 73 /12 ≈ 0.129
√
and n2 = 7 + 73 /12 ≈ 1.295. n1 and n2 are found, using the Frobenius
Method, by forming the indicial equation, 6n2 + 7n − 1 = 0. This has roots n1
and −n2 .
The coefficients αi and βi satisfy
(α0 , β0 ) = (1, 1) , (2.10)

3 [n + i − 1] [2 (n + i) + 1]
(αi , βi ) = (αi−1 , βi−1 ), (2.11)
6 (n + i)2 + 7 (n + i) − 1

2 [5 (n + i) + 1]
= 1− (αi−1 , βi−1 ), (2.12)
6 (n + i)2 + 7 (n + i) − 1
for i ≥ 1 and where n = n1 for the αi and n = −n2 for the βi . Table 2.1 gives
the first six values of αi and βi .
i αi βi
0 1 1
1 0.0865 0.625
2 0.0375 −0.193
3 0.0218 −0.0635
4 0.0146 −0.0336
5 0.0105 −0.0214
Table 2.1: Values of αi and βi for solution (2.9), to 3 significant figures.
The complete solution of equation (2.6) is thus
F (R) = R − 2Λ + CΛf (R) + DΛg(R) (2.13)

−n1 X ∞ −i
R − 3Λ R − 3Λ
where f (R) = αi (2.14)
Λ i=0
Λ
n2 X∞ −i
R − 3Λ R − 3Λ
and g(R) = βi (2.15)
Λ i=0
Λ
It can seen from (2.12) that, as i → ∞, |αn+1 /αi | → 1 and |βn+1 /βi | → 1 so that
convergence of the series is only guaranteed when |p| < 1, that is, R > 4Λ. Only
when R is close to 4Λ is a large number of terms in (2.14) and (2.15) required
but, for R 4Λ, a very good approximation is
−n1 n
R − 3Λ R − 3Λ 2
F (R) = R − 2Λ + CΛ + DΛ (2.16)
Λ Λ
" n2 −1 #
R
' R 1+D , (2.17)
Λ
for sufficiently large R.
Solutions (2.13) for typical values of C and D are shown in Figure 2.2. It is
clear that only when D = 0 does the solution tend to R − 2Λ, the standard
Einstein Gravity solution, as R increases.
2.2.2 Algebraic solution which includes R = 4Λ
This is achieved by setting q = (R − 4Λ) /Λ. How q varies with N is shown in

Figure 2.1(b), above. Equation (2.6), in terms of q, is
−6q (1 + q) Fqq − (2 − q) Fq + F = 2q. (2.18)
The homogeneous form of (2.6) has indicial equation with roots n = 2/3 and
n = 0 so that (2.6) has series solution
" #
X X
F (R) = Λ q + 2 + C 0 γi q i+2/3 + D0 δi q i . (2.19)
i=0 i=0
The first six values of γi and δi are listed in Table 2.2 and are such that their
sums to infinity are convergent. They satisfy
(γ0 , δ0 ) = (1, 1) (2.20)

6i2 − 11i + 2 6i2 − 19i + 12

(γi , δi ) = − γi−1 , − δi−1 , (2.21)
2i (3i + 2) 2i (3i − 2)
for i ≥ 1. In the form of (2.11) they are
" #
6 (n + i)2 − 19 (n + i) + 12
(γi , δi ) = − (γi−1 , δi−1 ) , (2.22)
2 (n + i) (3 (n + i) − 2)
for i ≥ 1 where n = 2/3 for the γi and n = 0 for the δi .
i γi δi
0 1 1
1 3/10 1/2
2 −0.0375 0.0625
3 0.01307 −0.01339
4 −0.00630 0.00536
5 0.00360 −0.00276
Table 2.2: Values of γi and δi for solution (2.19), to 3 significant figures.
In terms of R, (2.19) is
" i+2/3 i #
X R − 4Λ X R − 4Λ
F (R) = R − 2Λ + Λ C 0 γi + D0 δi .
i=0
Λ i=0
Λ
(2.23)
Convergence of (2.19) is valid for |q| ≤ 1, that is 4Λ ≤ R ≤ 5Λ and F (4Λ) =

(2 + D0 ) Λ. Note that if D0 = 0, F (4Λ) = 2Λ, the same as the late time de Sitter
limit for standard ΛCDM. If C 0 6= 0 then FR (4Λ) is unbounded and, as R → 4Λ,
FR (R) takes the sign of C 0 . However, if C 0 = 0, FR (R) → 1 as R → 4Λ, just
as it does in standard cosmology. Two curves for which C 0 = 0 are shown in
Figure 2.4.
2.2.3 Matching conditions for the two expressions for F (R)
The expression for F (R) of section 2.2.1 must be the same as the expression for
F (R) from subsection 2.2.2 where they have common values of R, i.e., 4Λ < R ≤
5Λ. In terms of q 6= 0, where p = 1/ (1 + q), they are
" #
X X
F (R) = Λ q + 2 + C αi (1 + q)−(i+n1 ) + D βi (1 + q)−(i−n2 ) ,
i=0 i=0
(2.24)
" #
X X
= Λ q + 2 + C0 γi q i+2/3 + D0 δi q i , (2.25)
i=0 i=0
which are the same when 0 < q ≤ 1. Note that (2.24) does not have a Maclaurin
expansion in terms of q.
Numerically, C 0 and D0 are linearly dependent on C and D according to the

approximate relations,
C 0 = −0.774 C + 0.970 D, (2.26)
D0 = 1.256 C + 1.187 D. (2.27)
This correspondence between coefficients C and D and coefficients C 0 and D0 was

obtained by equating (2.24) with (2.25), and their derivatives with respect to q,
at some value of q between 0 and 1. In fact, the parameter choice we made was
q = 1, as convergence of the series is fastest the further one is away from R = 4Λ.
Figure 2.2 shows graphs of F (R)/ (R − 2Λ) against p for various values of C
and D and with the corresponding values of C 0 and D0 as defined by (2.26) and
(2.27) listed in Table 2.3.
To demonstrate the overlap between the solutions of (2.13) and (2.23), curves
2 and 3 of Figure 2.2 and their counterparts as defined from (2.23), using the
relations, (2.26) and (2.27), are shown in Figure 2.3. Also shown is how quickly
each curve diverges from its counterpart.
Figure 2.4 shows two examples with C 0 = 0, as has already been mentioned.
From (2.26) we see that, in this case, D 6= 0, for any given non-zero C, so that
the F (R) is bound to diverge from R − 2Λ at large values of R, as p approaches
zero. From (2.23), the limit of F (R) as R → 4Λ is (2 + D0 ) Λ where D0 = 2.23C.
FHRLHR-2LL
1.3 FHRLHR-2LL
1.004
1.2
1
1.1 1.002
2
3 6
4 1.0
5 4
1.000
6
0.9 2
7
0.998
0.8
p 0.996 p
0.0 0.2 0.4 0.6 0.8 1.0 0.990 0.992 0.994 0.996 0.998 1.000
(a) (b)
Figure 2.2: F (R)/ (R − 2Λ) against p = Λ/ (R − 3Λ). (a) shows seven numbered
curves for values of C and D. In (a), axes cross at the current time, “today”, when
p = 0.7. In (b), axes cross at some time in the future. (b) shows detail very close to
p = 1 for curves 2, 4 and 6 to indicate their gradients as p → 1. Values of the pairs
(C, D) for each of the seven curves are: 1 (dot-dashed, maroon), (0.1, 0.1); 2 (dashed,
dark green), (−0.1, 0.1); 3 (continuous, dark blue), (0.1, 0); 4 (continuous, red), (0, 0),
standard Einstein Gravity; 5 (continuous, royal blue), (−0.1, 0); 6 (dashed, green),
(0.1, −0.1); 7 (dot-dashed, pink), (−0.1, −0.1).
Curve No. C D C0 D0 FR (R) as p → 1
1 0.1 0.1 0.0196 0.244 +∞
2 −0.1 0.1 0.174 −0.00691 +∞
3 0.1 0 −0.0773 0.126 −∞
4 0 0 0 0 1
5 −0.1 0 0.0773 −0.126 +∞
6 0.1 −0.1 −0.174 0.00691 −∞
7 −0.1 −0.1 −0.0196 −0.244 −∞
Table 2.3: Values of C, D, C 0 and D0 for each of the solution curves in the left hand
pane of Figure 2.2 and the behaviour of FR (R) as R → 4Λ (p → 1). Curve No. 4 is
standard Einstein Gravity. The values of C, D, C 0 and D0 chosen are for representative
purposes only.
FHRLHR-2LL FHRLHR-2LL
1.30 1.07
1.25 1.06
1.05
1.20
1.04
1.15
2 1.03
1.10 3 1.02
1.05 1.01
p p
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) (b)
Figure 2.3: Two example curves from Figure 2.2 and subsets of their counterparts
as defined from (2.23). (a) shows curve 2 (dashed, dark green) and its counterpart
(continuous, pink) while (b) shows curve 3 (continuous, dark blue) and its counterpart
(dashed, orange). The dotted, black line is at p = 0.5. Each pair of curves has common
points when 4Λ < R ≤ 5Λ which is 0.5 ≤ p < 1. Axes cross at the current time,
p = 0.7.
Thus, the limits, as p → 1 of F (R)/ (R − 2Λ) of the two examples of Figure 2.4,
are 1.11 when C 0 = 0 and C = 10−1 , and 1.01 when C 0 = 0 and C = 10−2 .
FHRLHR-2LL
1.3
1.2 log10 FHRLL

10
C ¢ = 0, C = -1
10
8
1.1
6
C¢ = 0, C = 10-2
1.0 4
p log10 HR-2LLL
0.0 0.2 0.4 0.6 0.8 1.0 0 2 4 6 8
(a) (b)
Figure 2.4: F (R)/ (R − 2Λ) against p, and log10 [F (R)/Λ] against log10 [(R − 2Λ) /Λ]
for two solutions for which C 0 = 0. The straight lines (in red) are the standard Einstein
Gravity curves for comparison. Notice how, the smaller C is, the more the solution
curve “hugs” the standard ΛCDM curve for a longer time before finally veering away
at large R. For both curves, D > 0. Time goes from left to right in (a) and from right
to left in (b). Axes cross at the current time.
From (2.2), the matter density equals the “effective” vacuum density when
H02 Ωm,0 /a3 = Λ/3 to give, using (2.3), R = 5Λ. 5Λ thus represents the value of R
at what one might describe as equality between matter and “effective” vacuum
densities, in analogy to equality between radiation and matter densities.
We note that, providing D 6= 0, F (R) deviates significantly from R − 2Λ at

early times in the matter era. At late times, in the vacuum era, the behaviour of
F (R) is governed by the values of both C 0 and D0 and only when D0 = 0 will the
limit of F (R) coincide with the limit of R − 2Λ as R → 4Λ.
2.2.4 Application of constraints to the solutions
Apart from local tests of gravity, which we shall reconsider later, the general
requirements of F (R) are taken to be FR (R) > 0 and FRR (R) ≥ 0 [76, 77]. They
have been discussed in subsection 1.4.8. The two quantities are connected via the
mass, M , of the scalaron, viz
FR (R) − RFRR (R)

M2 = , (2.28)
3FRR (R)
⇐⇒ FR (R) = 3M 2 + R FRR (R).

(2.29)
Thus, except when M → ∞, FR (R) > 0 ⇐⇒ FRR (R) > 0, and FR (R) <
0 ⇐⇒ FRR (R) < 0. In standard Einstein Gravity, FRR (R) = 0 and M 2 → ∞.
If M → ∞, FRR (R) = 0 and F (R) is a linear function of R which could be
thought of as rescaled standard Einstein Gravity.
Defining the two variables m and r, as is standard in texts on modified gravity

[49, 52, 77, 87, 92], by m = RFRR (R)/FR (R) and r = −RFR (R)/F (R), it is
noted in [49] that “viable cosmological trajectories are restricted to be in the
range m > 0 and r < 0.”
We shall define R1 to be the smallest value of R at which we shall apply FR (R) >
0 and FRR (R) ≥ 0 to the solution (2.13). Today’s value of R = R0 = 4.4Λ so
we choose R1 ≥ 4.4Λ. R2 is the largest value of R at which we can apply these
criteria. R2 ∼ Req , the curvature at the time of equal matter and radiation.
Generally, we take R2 = 1010 Λ.
2.2.4.1 The condition that FR (R) > 0
FR (R) > 0 gives 1 + CΛfR (R) + DΛgR (R) > 0 which rearranges at R = R2 and
R = R1 = 4.4Λ as
1 gR (R2 )
C < − 0
− D, (2.30)
Λf (R2 ) fR (R2 )
n +1 n +n
1 R2 1 n2 R2 1 2
= + D, (2.31)
n1 Λ n1 Λ
≈ 1.5 × 10 + 1.8 × 1015 D,
12
(2.32)
1 gR (R1 )
and C < − 0 − D, (2.33)
Λf (R1 ) fR (R1 )
≈ 3.2 + 4.9 D. (2.34)
Boundaries for the regions defined in (2.30) and in (2.33) meet at C = 3.2 and
at D = −8.6 × 10−4 . The boundaries are shown in Figure 2.5 as FR (R2 ) = 0 and
FR (R1 ) = 0, respectively.
2.2.4.2 The condition that FRR (R) ≥ 0
FRR (R) ≥ 0 is CΛfRR (R) + DΛgRR (R) ≥ 0 which rearranges to

n +n
n2 (n2 − 1) R2 1 2
C ≥ − D, (2.35)
n1 (n1 + 1) Λ
≈ −4.6 × 1014 D, (2.36)
gRR (R1 )
and C ≥ − D, (2.37)
fRR (R1 )
≈ 0.58 D, (2.38)
since −gRR (R1 )/fRR (R1 ) has a maximum value of approximately 0.58 at R1 = R0 .
The two boundaries of this region are shown in Figure 2.5 as FRR (R1 ) = 0 and
FRR (R2 ) = 0.
2.2.4.3 The local gravity constraint
As discussed in subsection 1.4.8, the local gravity constraint implies m 10−23

[33, 49, 94] at R = Rs ≈ 106 Λ, which is well inside the matter era. In terms of the
scalaron mass, via (2.29), this translates to 3Ms2 /Rs 1023 FR (Rs ), where Ms is
the value of M at R = Rs . The smalless of FRR (R) and the closeness of FR (R)
to unity mean that the scalaron mass, defined from (2.28), is approximated by
3Ms2 = 1/FRR (Rs ). Thus, on rearrangement, we have
n1 +1 n +n
R2 Rs n2 (n2 − 1) Rs 1 2
C = − D, (2.39)
3Ms2 n1 (n1 + 1) Λ n1 (n1 + 1) Λ
1.36 × 1013 Λ
≈ − 9.2 × 108 D, (2.40)
Ms2
4.1 × 10−16 − 9.2 × 108 D. (2.41)
The boundary of the region represented by (2.41) meets the boundary of the
region defined by (2.35) at (D2 , C2 ), where
n1 +1 " n2 +n1 #−1
Rs Rs Rs
C2 = 1− , (2.42)
3Ms2 n1 (n1 + 1) Λ R2
≈ 4.1 × 10−16 , (2.43)
−n2 +1 " n2 +k1 #−1
Rs Rs R2
D2 = − −1 , (2.44)
3Ms2 n2(n2 − 1) Λ Rs
≈ −8.9 × 10−31 . (2.45)
The boundary of the region represented by (2.41) meets the boundary of the
region defined by (2.37) at (D0 , C0 ) = (4.4 × 10−25 , 4.0 × 10−26 ). It is noted that
C2 {−D2 , C0 , D0 }. The boundaries (2.32), (2.34), (2.36) and (2.38) and the
line (2.39) are shown schematically in Fig. 2.5.
If we require FRR (R) not to be singular as R → 4Λ then we must set C 0 = 0

in equation (2.26). This adds the further constraint C = 1.25D, which is shown
as the short green line in Figure 2.5. This meets the Solar System constraint
boundary (2.39) at (D3 , C3 ) = (4.4 × 10−25 , 5.5 × 10−25 ).
To see which terms of (2.16) for the expression for F (R) are relevant, with the
above constraints imposed, near equality D ∼ 10−25 , R ∼ 1010 Λ give
C
FR HR1L = 0
=0
2L
HR
FR
HD2 , C2 L
F RR
HR 2L
HD3 , C3 L
=0
=0
C
¢
HD0 , C0 L FRR HR1L = 0
0 D
Figure 2.5: Schematic diagram showing the constraints from F (R) > 0 and FRR (R) ≥
0 and from local gravity. The axes are labelled 0D and 0C with 0 being the origin.
The solid, diagonal, red line on which lie the points (D0 , C0 ), (D2 , C2 ) and (D3 , C3 ), is
the line represented by (2.39) when Ms takes its smallest value. The allowed region is
thus the unshaded triangular region with the origin as one vertex.
DΛ (R/Λ)n2 ∼ 10−12 Λ. From this we conclude that the range of values of (C, D)
which are compatible with FR (R) > 0, FRR (R) > 0 and the local gravity con-
straint is very restricted to being close to (0, 0). Therefore, it appears that the
only models meeting these criteria are effectively standard Einstein Gravity.
2.2.5 The matter growth index, γ
We investigate what bearing various solutions have on the matter growth index,
γ, as introduced in Chapter 1 and expanded upon in Chapter 3. We define the
matter growth index, γ, by [74]
d log δm
(Ωm )γ = (2.46)
d log a
δ0
= m. (2.47)
δm
where δm , known as the matter density contrast, is defined in terms of perturba-

tions in the matter density as δm = δρm /ρm .
We shall ignore the severe parameter constraints of subsection 2.2.4, for illus-
trative reasons, because to include them we should not be able to differentiate
between them and standard Einstein Gravity. We show it for completeness and
to illustrate how the same a(t) can give very different γ.
We shall use the expression of equation (2.13). Clearly if F (R) is to tend to

standard Einstein Gravity at high R we must let D = 0. Thus, we are dealing
with a one-parameter set of solutions. We choose C = 0, 10−5 , 10−4 and 10−3 .
The first of these is, of course standard Einstein Gravity. We cannot allow C to
be negative when D = 0 as this would result in FRR (R) being negative for all
values of R greater than some minimum value. The results acquired by numerical
integration are shown against redshift z in Figure 2.6 and the trend as C varies
is seen clearly. At high z, there is ΛCDM, γ ≈ 6/11 = 0.545 while, as z → −1,
γ tends to 2/3 for all models, again, as for ΛCDM. This is discussed further in
Chapter 3. With C as small as in the allowed triangle of Figure 2.5, the result
is indistinguishable from ΛCDM and if, further, we must have C 0 = 0, then the
only solution is C = D = 0, i.e., ΛCDM.
2.3 In the radiation era
It can be seen in Figure 2.2 that curves 1, 2, 6 and 7, curves for which D 6= 0, are
not tending to standard Einstein Gravity at high curvature, that is as p → 0. It
is thus worth investigating what happens to solutions like these in the radiation
era, which we have ignored so far.
In the radiation era, we may take R ≈ 3H02 Ωm,0 e−3N and H 2 ≈ H02 Ωr,0 e−4N ,
1/3
i.e., H 2 = κR4/3 , where κ = Ωr,0 / 81H02 Ω4m,0 . Since H 2 Λ, we may rewrite
2.3. In the radiation era 69
Γ
0.7
0.6
standard Einstein Gravity
0.5 10-5
10-4
0.4
0.3
0.2
10-3
0.1
z
2 4 6 8 10
(a)
Figure 2.6: Curves of the matter growth index, γ, against z for four examples of
the solution to equation (2.13). The values of C are indicated except when C = 0
when the curve is labelled standard Einstein Gravity. Today’s values of γ for each
curve are as follows: C = 0, γ0 = 0.555; C = 10−5 , γ0 = 0.533; C = 10−4 , γ0 = 0.475;
C = 10−3 , γ0 = 0.430. In all cases, D = 0.
the field equation, (2.5), as
R−1/3 R−4/3

3RFRR − 1 − FR − F + 1 = 0. (2.48)
6κ 6κ
This has solution F (R) = R and asymptotic solution F (R) = α R4/3 , where α
is constant. We see that as R increases without limit, either F (R)/R → 1 or
F (R)/R4/3 → constant. Notice that the matter era solution, (2.16), has n2 =
1.295 which is close to 4/3, so the two solutions match quite nicely.
If we want a power series solution to (2.48), it is helpful to rewrite it in terms

of the scale factor a as
3 Ωm,0 9H 2 Ωm,0
aFaa + 5Fa − F = − 04 , (2.49)
2 Ωr,0 a
where 0 < a 1.
Equation (2.48) has particular integral F (R) = R while we look for solutions to
the homogeneous form of (2.49) of the form F (R) = an i=0 ηi ai . The indicial
P
equation is n(n − 1) + 5n = 0 giving n = 0 or −4 which leads to the general

solution
4/3 X −i/3 X R −i/3

R R
F (R) = R + AΛ ρi + (A log a + B) Λ σi ,
Λ i=0
Λ i=0
Λ
(2.50)
where a = (Ωm,0 /Ωeff,0 )1/3 (R/Λ)−1/3 and ρi , σi are constants, the first six of
which are given in Table 2.4, and A and B are constants of integration. The
introduction of the log term has been necessitated by the roots of the indicial
equation differing by an integer. See, for example, [95].
i ρi σi
0 −1.09 × 10−12 1
1 1.23 × 10−9 6.79 × 102
2 −1.04 × 10−6 1.91 × 105
3 1.18 × 10−3 3.10 × 107
4 1 3.29 × 109
5 −1.36 × 102 2.48 × 1011
Table 2.4: The first six coefficients of equation (2.50). The fact that these coeffi-
cients are becoming numerically larger with increasing i does not matter as each is also
multiplied by (R/Λ)−i/3 where R/Λ is large.
For R/Λ sufficiently large, F (R) = R+AΛρ0 (R/Λ)4/3 is a good approximation.

By sufficiently large, in this context, we mean that R & 1016 Λ which is before
equality, well into the radiation era. The point at which AΛρ0 (R/Λ)4/3 overtakes
R depends on the value of AΛρ0 .
Only if A = 0 is the large R solution equal to that of standard Einstein

Gravity, otherwise it is given by F (R) ∼ R4/3 . Of course, as we have seen,
2.4. Conclusion 71
the Solar System constraint must make Aρ0 very small; we have seen its coun-
terpart, D, in the matter era has to be of order 10−25 or smaller. We expect
AΛρ0 (R/Λ)4/3 ∼ DΛ (R/Λ)n2 at equality which means that the R4/3 term will
only become dominant when R & 1076 Λ, when z & 1025 .
2.4 Conclusion
In this chapter we have shown that there are functions, F (R), which satisfy the
field equation

2 0 R 1
−3H R FRR (R) + − 3H FR (R) − F (R) = −3H 2 + Λ, (2.51)
2
2 2
other than the standard ΛCDM model, F (R) = R−2Λ. These solutions share the
same history as defined by the Hubble parameter, H, where H 2 = Ωr,0 H02 e−4N +
Ωm,0 H02 e−3N + Λ/3 and N = log a. In the matter and vacuum eras, when the
effects of radiation may be neglected, algebraic expressions have been found which
fall into two groups:
I. There is an infinite series solution, expressed as a power series in Λ/ (R − 3Λ),

which is convergent for R > 4Λ. The significance of 4Λ is that it is the lim-
iting, de Sitter, value of R at the end of the vacuum era, when 3H 2 = Λ.
II. There is also an infinite series solution, this time expressed as a power series
in (R − 4Λ) /Λ, but which is only convergent in the range 4Λ ≤ R ≤ 5Λ.
Where the convergence ranges of these two groups overlap, i.e., when 4Λ < R ≤
5Λ, the solutions have been matched in terms of their particular constants of
integration showing one set to be a linear combination of the other set. The rela-
tionship between the two sets is given in equations (2.26) and (2.27) with graphical
correspondence between two pairs of examples being shown in Figure 2.3.
The late time value of the standard Einstein Gravity model is F (4Λ) = 2Λ.
A feature of these more general solutions is that at late times it is possible for
F (4Λ) to equal 2Λ by choosing D0 to be zero. Recall that C 0 and D0 are defined
in (2.19). In no other sense does the general F (R) tend to R − 2Λ because for
FR (R) to tend to a common value of unity, C 0 must be zero, as well, and if both
are zero we have F (R) = R − 2Λ, for all R.
At early times in the matter era, the general F (R) can tend to R − 2Λ which
requires D of (2.13) to be equal to zero. With D not being zero F (R) diverges
away from R − 2Λ as R increases. When R 4Λ, we find F (R) ∝ Rn2 where
n2 ≈ 1.3.
Then we considered what could happen in the radiation era by solving a sim-
plified field equation applicable when R Req . Here, we were only concerned
with the leading terms of any series solution but found that either F (R) → R, as
R increases, or F (R)/R4/3 → constant.
For none of the solutions illustrated in Figures 2.2, 2.3 and 2.4 were local
gravity constraints applied. When they were applied it was seen that the solutions
are extremely close to standard Einstein Gravity; the coefficients C, D, C 0 and
D0 were in the range O(10−16 ) to O(10−25 ), or smaller. Further, if there is to
be no singularity in FR (R) as R → 4Λ, then either the solution is standard
Einstein Gravity or solutions must run backwards through time very close to
F (R) = R − 2Λ until well past Rs when they diverge such that F (R) ∝ R4/3 as
R increases further. In this case, C, D, C 0 and D0 are of order 10−25 , or smaller.
As an extension of this work, we could specify the history of H in some other

way or specify the history of weff , of Ωeff or of R. Generally these would all be
equivalent to specifying the history of H by use of the following relationships,
2.4. Conclusion 73
again ignoring the radiation era although this could be included if necessary:
H02 e−3N Ωm,0

Ωeff = 1 − (2.52)
H2
3H 2 + 2HH 0
weff = − 2 (2.53)
3H − 3H02 e−3N Ωm,0
R = 12H 2 + 6HH 0 (2.54)
We would then be left with a differential equation for F (N ) in terms of N .

Whether or not this is soluble, algebraically, would depend upon circumstances.
Chapter 3
Non-ΛCDM Gravitation Models
3.1 Introduction
Currently, there is some doubt as to today’s value of what we term the “effective”
equation of state, weff,0 . While the latest results from the Planck Collaboration
[17] do not rule out ΛCDM as a valid model describing the history of the Universe,
they do not confirm weff,0 to be −1. Instead Planck gives a broad range of values
at the 95% level of significance, namely, weff,0 = −1.04+0.72
−0.69 . It should be noted
that quintessence models [96] have weff ≥ −1 while F (R) models have weff close
to −1 but always below it, at high redshift, which becomes more negative until
a local minimum is reached when weff increases to cross the phantom boundary,
weff = −1, some time later [49, 68]. With a particular F (R) model it is possible
that 1 + weff,0 is positive, zero or negative [97–100].
In this chapter we look at three models which, while they cleave to ΛCDM at
high curvature in the matter era, veer away from it in the late matter or the early
vacuum eras to give values of weff,0 differerent from −1. The three models chosen
will be:
I. the HSS (Hu-Sawicki-Starobinsky) model, [101]. It is a modified version of
75
76 Chapter 3. Non-ΛCDM Gravitation Models
the Hu-Sawicki model described in [97],
II. the AB (Appleby-Battye) model, also described in [101] but first introduced
in [76],
III. a new model, defined in subsection 3.2.1 of this thesis, which we call the Erf
model because it is based on the error function, erf(x).
The AB model is a two-parameter model while the Erf and HSS models, have
three parameters. For all of them, we see how weff,0 and Ωeff,0 vary as their
parameters are varied. Similarly, we see the relation between weff,0 and Ωeff,0 and
0
between weff,0 and weff,0
For each of the Erf, AB and HSS models, two examples are chosen and, for
each, it is seen how f (R), fR (R), weff and Ωeff vary with time. For these six
examples it is arranged that Ωeff,0 = 0.7. This is just a round figure as it is noted
that Planck [17, 102] gives its value as 0.686 ± 0.020 at the 68% significance
level. This result, which is numerically smaller than previously measured [5], was
found using the Doran and Roberts model for dark energy [103] in the context of
ΛCDM. The ΛCDM model is used as a background model because it has proved
to be so successful. If some other model were substituted then the reults of
Planck and previous attempts such as WMAP might change. A quote from the
introductory paper to the Planck results [3]: . . . the Planck data are in remarkable
accord with a flat ΛCDM model; however, there are tantalizing hints of tensions
both internal to the Planck data and with other data sets. While such tensions
are model-dependent, none of the extensions of the ΛCDM cosmology we explored
resolve them. It is to be hoped that more data and further analysis will shed light
on these areas of tension. Along these lines, we expect significant improvement
in data quality and the level of systematic error control, plus the addition of
polarization data, from Planck in the near future.
One of the demands of General Relativity is that F (0) = 0, that is, letting
3.1. Introduction 77
F (R) = R + f (R), f (R) → 0 as R → 0. Just because f (0) = 0 does not

necessarily mean that R can become zero, however. This is because there could
be a de Sitter attractor [46, 104], RdS , which prevents this so that R → RdS
as N = log a → ∞. How this happens, whether R gradually decreases to RdS
or whether it oscillates around it with ever-decreasing size, is discussed. The
parameter space of the contour plots that we present in this chapter is generically
divided into two regions; one region where there would be a de Sitter attractor,
RdS ≥ 0, and another where there is a Minkowski solution R = 0, as N → ∞
[105]. In fact, the de Sitter region is split up into two; a region in which R
oscillates on its way to the attractor, we call this the de Sitter oscillating region,
and another region in which there is a gradual movement towards the de Sitter
attractor with no oscillation, we call this the de Sitter non-oscillating region.
What happens in the Minkowski region can be more complicated especially when
R oscillates as, when R is close to zero, at some point, it will go negative. See
subsection 3.2.1.2.
We use the phrase would be in the previous paragraph because the purpose of
this exercise is to see how large we can make |1 + weff,0 | without any regard to the
future. If we were restricted to the de Sitter region, we would not see the greater
possibilities of straying into the Minkowski region. We find that for examples in
the Minkowski region, if left unaltered, give future values of weff and Ωeff which
are unacceptable. See, for example, subsection 3.2.1.2.
Note, there are some graphs in this chapter with an axis marked as f 0 (R).
This means fR (R) = df (R)/dR and not df (R)/dN . Where some graphs are
plotted against N = log a, today is always represented by N = 0 so that redshift
z = e−N − 1.
3.2 Three models compared
Three models are compared; The AB and HSS models and also a new function
which we dub the Erf function. The first two have been extensively studied and
are represented in [78, 97, 101].
As has been explained elsewhere in this thesis, all successful models approxi-
mate to ΛCDM at high curvature. This requirement might not be strictly neces-
sary but it tends to be assumed because of the severe constraint imposed by the
Solar system [49] at R = Rs = 106 H02 . Also, using data from measurements of
supernovae type Ia and from the CMB, m = R FRR (R)/FR (R) [49, 77, 106] must
be less than O(0.1)[49, 87]. In effect, this means that 0 < FRR (R) < 0.1/R which
means that when R is very large, as in the early matter era just past equality,
deviations for F (R) = R + f (R) from ΛCDM, that is F (R) = R − 2Λ, must be
very small indeed. Thus it is assumed that F (R) → R − 2Λ as R → ∞, as a
practical convenience.
We also demand that at the end of time if R → 0 so should F (R). This is to

be compatible with GR, i.e., F (0) = 0.
As always in this thesis, F (R) = R + f (R) so that f (R) denotes deviations of

F (R) from GR. As R → ∞, f (R) → −2Λ∞ ; Λ∞ being an effective cosmological
constant.
An intrinsic feature of all models is that R0 , weff,0 and Ωeff,0 are connected
via R0 = 3H02 (1 − 3weff,0 Ωeff,0 ), if we neglect the tiny effect of radiation; see
Useful Expressions on page 24. This poses a problem if we wish to have Ωeff,0
equal to a fixed value like 0.7, say, because R0 is constrained to be within a
certain range. Thus, large differences in weff,0 between models are bound to be
impossible. Applying data from Planck, which assumed GR, gives an idea that
R0 lies in the approximate range 7 H02 < R0 < 12 H02 .
3.2. Three models compared 79
3.2.1 The Erf model
This model was derived in an attempt to see if the values of weff,0 it produced
could deviate more from −1 than the values delivered by other models. It is
based on the error function erf(x) which is defined as
Z x
2 2
erf(x) = √ e−t dt. (3.1)
π 0
It is an odd function in that erf(−x) = −erf(x) and continuously takes all values
between −1, when x → −∞, and 1, when x → ∞.
For general positive parameter b, the Erf model is
erf [c + b R/H02 ] − erf [c]

f (R) = −2Λ∞ ; (3.2)
1 − erf [c]
" 2 #
4Λ∞ b R
fR (R) = − √ exp − c + b 2 . (3.3)
π H02 (1 − erf [c]) H0
It is a three-parameter model; b, c and Λ∞ being the parameters.
For given values of b and c the model approaches ΛCDM as R increases so it is

able to pass the Solar System test (3.5) for suitable values of b and c. Typically,
R0 ' 10H02 ; ΛCDM gives R0 = 9.3 H02 . For R0 of this order (R0 changes slightly
as the parameters change), decreasing the value of b from a large positive value
to a smaller, positive one effects a swing away from ΛCDM. fR (R) changes from
being numerically very small to being numerically larger, as can be seen from
(3.3). The negative, squared exponent in (3.3) ensures that F (R) adheres very
2
closely to ΛCDM until it swings away when it does so rapidly, just as e−x clings
to zero until x becomes sufficiently close to zero. Some other models may not have
the facility to do this because local gravity tests constrain model parameters to
such an extent that they are not able to deviate sufficiently from ΛCDM, today.
For numerically extremely small b, f (R) approximates to

2
4Λ∞ b e−c R
f (R) ∼ − √ (3.4)
π (1 − erf [c]) H02
which can be considered to give re-scaled GR, but we do not get this far as we
are limited as to how small R can go. As can be seen from (3.2), increasing the
value of c + b R/H02 , takes the Erf model closer to ΛCDM, so increasing the value
of c does the same thing as can be seen in the contour plots of Figure 3.4.
The behaviours of f (R) and its derivative with respect to R, when plotted
against N , are illustrated in Figure 3.1. We choose two representative points in
parameter space, both of which satisfy the Solar System constraint [49], namely,
Rs fRR (Rs ) 10−23 , (3.5)
where Rs ≈ 106 H02 . For the general Erf model this is

" 2 #
8Λ∞ Rs b2 (c + b Rs /H02 ) Rs
√ exp − c + b 2 10−23 . (3.6)
H04 π (1 − erf [c]) H0
Because of the negative squared exponent, this does not place practical con-
straints on the parameters. It is sufficient that b & 10−5 . When c = 1, b 10−5.2
which falls, when c = 6, to b 10−6.3 .
The model for which c = 1.5, log10 b = −1.517 and Λ∞ = 2 H02 , which will
represented by continuous, pink curves, below, we shall term Model 1 while that
for which c = 1.5, log10 b = −0.914 and Λ∞ = 2 H02 , represented by dashed,
mauve curves, we shall term Model 2.
3.2.1.1 The dark energy equation of state
For the parameter values of Figure 3.1, the corresponding values of weff and Ωeff
over the history of the Universe are shown in Figure 3.3. They were derived by
solving the field equation (1.75) for H(N ), for given F (R), on the assumption that
at the initial value of N , Ni , the theory is ΛCDM, that is, H 2 (Ni ) = H02 Ωm,0 e−3Ni +
Λ∞ /3. weff Ωeff is then found from (1.86) while Ωeff is defined as 8πGρeff / (3H 2 )
f HRLH0 2 f ¢ HRL
N N
-1.5 -1.0 -0.5 -1.5 -1.0 -0.5 Model 2
-1 -0.05
-0.10
-2 Model 1
-0.15
Model 1
-3
-0.20
Model 2
-4
(a) (b)
Figure 3.1: f (R)/H02 and f 0 (R) against N = log a.
RH0 2
100
80
60
Model 1 40
20
Model 2
N
-1.5 -1.0 -0.5
(a)
Figure 3.2: R/H02 against N . For Model 1, R0 = 6.2 H02 while for Model 2, R0 =
9.1 H02 .
using (1.82), assuming the effects of radiation are negligible. This gives
H02 e−3N Ωm,0
Ωeff (N ) = 1 − , (3.7)
3H(N )2
3H(N )2 + 2H(N )H 0 (N )
weff (N ) = − . (3.8)
3H(N )2 − 3H02 e−3N Ωm,0
Ωm,0 is taken to be 0.3 to give Ωeff,0 = 0.7. The slight mismatch at N = Ni
between what H(N ) should be and the ΛCDM value we use manifests itself in
graphs of weff as small but rapid oscillations which die out as N increases. This
can be seen in the Erf Model 1 graph of Figure 3.3(a). These oscillations can be
reduced by starting the numerical integration to find H(N ) at a slightly smaller
value of Ni , as can be seen in the Erf Model 2 graph.
Weff
weff 0.7
0.6
-0.6
0.5
-0.8 Model 1 0.4
Model 1
0.3
z
0.5 1.0 1.5 2.0 2.5 3.0 3.5
Model 2
0.2 Model 2
-1.2
0.1
-1.4 0.0 z
0 1 2 3 4
(a) (b)
Figure 3.3: weff and Ωeff against z for the curves of Figure 3.1.
Figure 3.4 shows how today’s values of weff and Ωeff vary with the parameters
c and log10 b, for the case when Λ∞ = 2 H02 . The points in parameter space
representing Models 1 and 2 are indicated by spots, pink for Model 1 and mauve
for Model 2. The large c region corresponds to GR while c = 0, and large b
coincides with ΛCDM, with Ωeff,0 tending to Λ∞ /3. In the limit b → 0, f (R) → 0
which would give GR if it were not for the fact that before that could happen R0
becomes zero.
0
In Figure 3.5, we plot weff against Ωeff,0 and weff,0 against weff,0 for various
values of Λ∞ as indicated when c = 1.5. We see that along each curve, b increases
towards the fixed point at which Ωeff,0 = Λ∞ /(3H02 ).
weff,0 Weff,0
0.67
-0.6 -0.6
de Sitter oscillating de Sitter oscillating
-0.8 -0.8
0.75
Model 2 0.7 Model 2
de S de S
-1.0 -0.72 itter -1.0 itter
non non
- os - os
cilla 0.78 cilla 0.72
ting ting
log10 b
log10 b
-0.9
-1.2 Minkowski -1.2
Minkowski
0.67
-0.18
-0.54 0.7
-1.4 -1.4 0.72 0.75
0.56
0.36 Model 1 Model 1
0.59
-1.6 0 -0.36
-1.6 0.64
0.54 0.62
-1.8 0.54 -1.8
0.18
0.72
-2.0 -2.0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
c c
(a) (b)
Figure 3.4: For Λ∞ = 2 H02 , contour plots of weff,0 and Ωeff,0 as functions of c and
log10 b. The lower boundary is where R0 comes close to being zero.
w eff,0 w¢ eff,0
-0.2 0.6
0.5
-0.4
0.4
1.5
0.3
-0.6
1.8
0.2
1.5 2
-0.8 1.8 0.1 2.4
2 2.8
2.4
2.8 weff,0
Weff,0 -0.9 -0.8 -0.7 -0.6 -0.5
-1.0
0.6 0.7 0.8 0.9 1.0 -0.1
(a) (b)
0
Figure 3.5: For c = 1.5, weff,0 as a function of Ωeff,0 and weff,0 as a function of
weff,0 . The fixed points are in (a) (Λ∞ / 3H02 , −1) and in (b) (−1, 0). Each curve

is represented by a value of Λ∞ /H02 which is indicated. The spots represent Model 1

which is at (0.7, −0.50) in (a) and at (−0.50, 0.79) (not shown owing to the scale of the
diagram) in (b), and also by Model 2 at (0.7, −0.97) in (a) and at (−0.97, 0.10) in (b).
3.2.1.2 Late-time behaviour
This thesis is concerned with the the history of the Universe up to the present
time. How do parameters such as H(N ), R(N ), weff (N ), Ωeff (N ) etc develop? The
thesis is not concerned with the future, as has been discussed in the introduction
to this chapter. As has been said, if our models give unacceptable problems after
today, such as can be seen in Figure 3.8, that can be remedied simply by altering
the model so that an appropriate change can be put into effect, then, to keep the
model well behaved. However it is instructive to gain an insight into what would
happen if the models were not altered.
It will be noticed in Figure 3.4 that three regions, viz, de Sitter oscillating, de
Sitter non-oscillating and Minkowski, covering the whole parameter plane, have
been marked separated by boundaries coloured red. These regions represent the
late-time, vacuum state behaviour of the model. Following Frolov in [107], we
identify the scalar field, φ(R) = FR (R) − 1 = fR (R) with associated potential
V (φ(R)) given by
Z R
1
V (φ(R)) = (r + 2f (r) − r fR (r)) fRR (r) dr. (3.9)
∞ 3
As has already been said, the late-time value of R is not necessarily equal to
zero but is defined where V (φ) is a local minimum. The resulting equation to be
satisfied is
R + 2f (R) − RfR (R) = 0. (3.10)
If such a stable solution exists, R = RdS is termed the late-time de Sitter

attractor solution. Putting all the time derivatives (equivalently, derivatives with
respect to N = log a) and densities in the field equation (q.v.) equal to zero
gives equation (3.10). There is always a solution for which R = 0 when space is
completely flat and this solution is thus termed the GR or Minkowski solution
[108]. If this solution is not a de Sitter solution, it may be unacceptable because
R will tend to oscillate around zero as it decays to zero. However, as can be seen
from Figure 3.4, the Minkowski region has the potential to give larger values of
|1 + weff,0 | than would otherwise be the case for the de Sitter region. The de Sitter
solutions fall into two groups; oscillating, for which weff , R and H exhibit late-time
oscillations, and non-oscillating, for which these variables increase or decrease to
their final values. An example of the latter is ΛCDM where weff = −1, Ωeff → 1,
R → 4Λ and H 2 → Λ/3 with no oscillations.
Two examples when Λ∞ = 2H02 for the Erf model are shown in Figures 3.6
and 3.7. One is in the oscillating de Sitter region of Figure 3.4 and one is in the
non-oscillating de Sitter region. The differing ways in which R/H02 and weff vary
with time in the future are clearly shown.
RH0 2 RH0 2
6.5
7.52
6.0
7.51 5.5
5.0
7.50 4.5
4.0
7.49
3.5
N N
2 3 4 5 6 7 8 2 4 6 8 10
(a) (b)
Figure 3.6: R/H02 against N = log a for two examples of the Erf model at times well
after today. (a) shows Model 2 with Λ∞ = 2H02 , c = 1.5 and log10 b = −0.914, which is
in the de Sitter oscillating region, and (b) shows an example with Λ∞ = 2H02 , c = 1.5
and log10 b = −1.12, which is in the de Sitter non-oscillating region. Solving (3.10) for
R, gives the de Sitter values of R as (a) 7.486 H02 and (b) 3.309 H02 , respectively.
It is also instructive to see what happens in the Minkowski region and here
we shall consider Model 1. Just like Model 2 there are oscillations but there
are also oscillations in Ωeff (N ). For example, in the de Sitter region, Ωeff (N )
simply keeps growing until it tends to unity as N → ∞. We can see that from
the expressions for weff and weff Ωeff in the glossary entitled Useful Expressions,
weff
-0.85 weff
-0.90 -0.9985
-0.9990
-0.95
-0.9995
z N
-1.0 -0.5 0.5 1.0 1.5 2.0 2.5 2 3 4 5 6 7 8
-1.05 -1.0005
-1.0010
-1.10
-1.0015
-1.15 -1.0020
(a) (b)
Figure 3.7: weff for the two examples of the Erf model in Figure 3.6. (a) shows both
examples plotted against z. (b) shows the late-time oscillations of Model 2 when weff is
plotted against N . Note the severe gradient in (a) of weff for the non-oscillating curve
(continuous, green) as z → −1. While dweff /dN → 0+ , dz/dN → 0− .
putting the derivative with respect to N equal to zero gives final de Sitter values
of weff, dS = −1 and Ωeff, dS = 1.
For Model 1, however, R/H02 oscillates between positive and negative values
of ever-decreasing amplitude while Ωeff (N ) oscillates between fixed negative and
positive values as shown in Figure 3.8. This means, that weff would tend to
infinity at some point in the future. This is because weff Ωeff = −1 − 2H 0 / (3H)
(1.86) with H 0 /H, though it oscillates, remaining finite.
RH0 2 Weff
1.5 0.8
1.0 0.6
0.5 0.4
0.2
N
0.5 1.0 1.5 2.0 2.5 3.0
N
-0.5 0.5 1.0 1.5 2.0 2.5 3.0
-0.2
-1.0
(a) (b)
Figure 3.8: Showing the late-time oscillations in R/H02 and Ωeff for Model 1 in the
Minkowski region.
3.2.1.3 The matter growth index
Starting from the standard, linearised, matter perturbation equation in the sub-
horizon regime [74, 109–113], which is derived in [112],
!
G 3 + 4k 2 / (aM (a))2
δ̈m + 2H δ̇m − 4π ρm δm = 0, (3.11)
FR (R) 3 + 3k 2 / (aM (a))2
where k is the comoving wavenumber and M is the scalaron mass defined by

equation (1.88) with approximation given by
1
M2 ≈ , (3.12)
3FRR (R)
we can define the matter growth index, γ, as a function of the matter density
contrast, as already stated in section 2.2.5, by
d log δm
(Ωm )γ = (3.13)
d log a
0
δm
= . (3.14)
δm
Equation (3.11) can be re-cast as
Ω0m H0 3
γ 0 log Ωm + γ + (Ωm )γ + 2 + = η(a)(Ωm )1−γ , (3.15)
Ωm H 2
in which
2
1 + 4 ka2 FFRR (R) #
"
1 R (R)
η(a) = . (3.16)
FR (R) 1 + 3 k22 FRR (R)
a FR (R)
The factor in square brackets on the right hand side of equation (3.16) increases
from being 1 at early times to being 4/3 at late times. Equation (3.15) can be
solved numerically, with the initial condition at large R being that γ = 6/11 (as
appropriate for ΛCDM, [111, 113]). The ΛCDM value of γ at high R can be
deduced from (3.15) by varying Ωm with respect to Λ/H 2 .
The solution for γ, again corresponding to the parameters used in Figure 3.1
are shown in Figure 3.9(a) for k = 0.14 h Mpc−1 . This value of k is also used
in the examples, where relevant, of sections 3.2.2 and 3.2.3. For comparison, in
Figure 3.9(b) we show how γ varies in ΛCDM. Clearly the range of values for γ0
can, under certain circumstances, be larger than has been previously supposed
[74, 114].
Ignoring those cases when γ(N ) has been taken to be constant or linear, authors
have commented on the value of γ0 only in respect of particular models. In [114],
in which five models are considered, it is found that all of these models have
0.40 . γ0 . 0.55. We find that both the AB and the Erf models can give values
of γ0 lower that 0.40. Measurements of galaxy clusters by Rapetti et al. [73],
using data from ROSAT, BCS, REFLEX and Bright MACS, with a background
ΛCDM model, give γ0 = 0.55+0.13
−0.10 . Using XLF data, they find a γ0 with a mean
value of 0.38. See also see [110].
Γ
Γ
0.5 0.65
Model 2
0.0 0.60
z
1 2 3 4 5 6
0.55
-0.5
Model 1
z
-1.0 -1 1 2 3 4 5 6
(a) (b)
Figure 3.9: Erf model: γ against z for the examples of Figure 3.1. The dotted, red
line represents the asymptotic ΛCDM value 6/11. The general behaviour of γ, as z
decreases, is to increase initially away from 6/11, as it would in ΛCDM, to a local
maximum before decreasing to a local minimum from which it rises to today’s value.
The value of γ0 for curve for Model 1 is 0.362 while for Model 2 it is 0.426. (b) shows γ
for the ΛCDM model which has Ωm,0 = 0.3. At early times γ is asymptotic to 6/11. As
z decreases, γ increases to approach the value 2/3 as γ 0 → 0 and z 0 → 0 making dγ/dz
undefined. Solution of the ΛCDM equivalent to (3.15) shows that, for Ωm,0 = 0.3, then
γ0 = 0.55472 ± 8 × 10−6 and γ00 = 0.01688 ± 4 × 10−6 .
3.2.2 The AB model
The AB model [101] is

R AB cosh (R/AB − b)
f (R) = − + log , (3.17)
2 2 cosh b
Rvac
AB = , (3.18)
b + log (2 cosh b)
Rvac = 4Λ∞ . (3.19)
It is a two-parameter model and it, like the Erf model, similarly adheres closely
to ΛCDM until comparatively late in the history of the Universe. The Solar
System constraint means that b −4, although very large values of Λ∞ reduce
this limit slightly, e.g., Λ∞ = 10 means b −3. Positive values of b means
the Solar System test is passed with ease. For example, Λ∞ = 2H02 , b = 0
gives Rs fRR (Rs ) = 5 × 10−75,253 . In the paper where it was introduced [76],
Appleby and Battye take the parameter b & 1.2. This was done because current
data suggests that we are approaching a late time de Sitter vacuum state (not
Minkowski space). Since ΛCDM tends to a late-time de Sitter vacuum state,
Appleby and Battye wanted to find models that behaved like ΛCDM (because the
data are consistent with ΛCDM), but which had no true cosmological constant.
This, however, restricts the value of |1 + weff,0 | to be less than 0.04 [101].
Figure 3.11 shows how the fixed points are approached, as b increases, for a
range of Λ∞ .
For Figures 3.12, 3.13 and 3.14, two examples have been chosen, one with
b = 0.2 and Λ∞ = 1.65H02 , which we term Model 3, and one with b = 1.2 and
Λ∞ = 1.92H02 , which we term Model 4. Both give Ωeff,0 = 0.7 and are indicated
by spots in Figures 3.10 and 3.11.
weff,0 Weff,0
2.0 -0.99 2.0 0.54 0.75
1.5 de Sitter oscillating 1.5 de Sitter oscillating

0.91
0.64
de Sitter non oscillating de Sitter non oscillating
1.0 1.0
0.59 0.81
b
b
Minkowski
0.5 0.5
Minkowski
0.7
-0.85 -0.7 0.86
0.0 0.0
-0.57
0.64
-0.29
-0.43
-0.15 0.59
0.14 0.54
-0.5 0.42 -0.5
0 0.28 0.48
1.6 1.8 2.0 2.2 2.4 2.6 2.8 1.6 1.8 2.0 2.2 2.4 2.6 2.8
L¥ H0 2 L¥ H0 2
(a) (b)
Figure 3.10: Contour plots for the AB model. Though they do not show up at all
well, in (a) all values weff,0 are negative. The two spots represent Models 3 and 4, see
below. Note the inclusion of the late-time de Sitter and Minkowski regions.
w¢ eff,0
w eff,0
-0.2 0.5
0.4
-0.4
0.3 1.5
1.8
-0.6 0.2
2
0.1 2.4
-0.8 1.5 2.8
1.8 2 2.4 2.8 weff,0
-0.9 -0.8 -0.7 -0.6
-1.0 Weff,0
0.6 0.7 0.8 0.9 1.0 -0.1
(a) (b)
0
Figure 3.11: AB model: weff,0 as a function of Ωeff,0 and weff,0 as a function of weff,0
for the values of Λ∞ /H02 indicated. The fixed points are in (a) (Λ∞ /(3H02 ), −1) and
in (b) (−1, 0). Parameter b starts at the top of each curve and increases towards the
fixed points.
f HRLH0 2 f ¢ HRL
N N
-2.0 -1.5 -1.0 -0.5 -1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2
Model 4
-0.02
-1 -0.04
Model 3
-0.06
-2
-0.08
-3 -0.10
Model 3
-0.12
Model 4
-4 -0.14
(a) (b)
Figure 3.12: AB model: f (R)/H02 and fR (R) against N .
RH0 2
70
60
50
40
30
Model 3
20
Model 4
10
N
-1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2
Figure 3.13: AB model: R/H02 against N for the curves of Figure 3.12. For Model 3,
R0 = 8.1 H02 and, for Model 4, R0 = 9.1 H02 .
Weff
weff 0.8
-0.8
0.6
-0.9 Model 3
z 0.4 Model 3
0.5 1.0 1.5 2.0 2.5 3.0
Model 4 Model 4
-1.1 0.2
-1.2
0.0 z
1 2 3 4
(a) (b)
Figure 3.14: AB model: weff and Ωeff against z for the curves of Figure 3.12.
Γ
0.6
0.4
Model 4
0.2
z
1 2 3 4
-0.2
Model 3
-0.4
-0.6
Figure 3.15: AB model: Evolution of γ against z for the examples of Figure 3.12.
The values of γ0 are 0.39 for Model 3 and 0.42 for Model 4.
3.2.3 The HSS model
Like the Erf model, the HSS model is a three parameter model:
Rvac c (R/Rvac )2n

f (R) = − , (3.20)
2 1 + c (R/Rvac )2n
Rvac = 4Λ∞ , (3.21)
with n > 0. In fact, for this model to pass the Solar System test, we must have
n > 1. For n < 2, when Λ∞ = 2H02 , for instance, it passes this test only for
some values of Rvac , as c varies, so it is safest to have n ≥ 2. For very large but
unrealistic values of Λ∞ the lower limit on n may have to be increased to 3. The
limit as R → ∞ is f (∞) = −Rvac /2 which is ΛCDM with cosmological constant
equal to Rvac /4, as for the AB model. It should be noted that up to the present
time R > R0 . If R0 /Rvac > 1 this model tends to ΛCDM if either of c or n should
tend to infinity.
Graphical examples below are for n = 2 and n = 6. The blank ‘triangular’ re-
gion at the bottom right hand corner of the contour plots is owing to the fact that,
for small enough c, fRR (R0 ) will become zero (see, for example, Figure 3.22(b)).
The value of c for which this happens is given by
−2n
2n − 1 R0
c = . (3.22)
2n + 1 Rvac
weff,0 Weff,0
3.0 3.0 0.37 0.79
2.5 2.5 0.56
-0.99
2.0 de Sitter oscillating 2.0

0.47 de Sitter oscillating 0.7
-0.97
1.5 1.5
log10 c
log10 c
-1.03 0.61 0.84
Minkowski -0.93
1.0 1.0
Minkowski
0.42
-0.95
0.5 0.5
-0.87 0.75
-1.01
0.0 -0.89
-0.82 -0.76 0.0 0.51
0.89
-0.8
-0.78
-0.5 -0.85 -0.5
-0.91 -0.74 0.65
1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4
2 2
L¥ H0 L¥ H0
(a) (b)
Figure 3.16: Contour plots for the HSS model of weff,0 and Ωeff,0 against Λ∞ /H02 and
log10 c when n = 2. The pink spot represents Model 6, see below. The narrow late-time
de Sitter non-oscillating region has not been labelled.
weff,0 Weff,0
3 3 0.4
-1.01 de Sitter oscillating 0.7
de Sitter oscillating
2 Minkowski 2 0.6
-0.99
-1.06
Minkowski 0.75
1 1 0.5
-1.11 0.8
log10 c
log10 c
0 -1.04 0
0.65
-0.94
-1.16
-1 -1 0.9
-0.96
0.45
-1.08 0.85
-2 -2
0.55
-1.13
0.95
-3 -3
1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4
L¥ H0 2 L¥ H0 2
(a) (b)
Figure 3.17: Contour plots for the HSS model of weff,0 and Ωeff,0 against Λ∞ /H02 and
log10 c when n = 6. The mauve spot locates Model 5, see below.
Notice, from Figures 3.16 and 3.17, how restricting solutions to the de Sitter
regions gives values of |1 + weff,0 | . 0.04. How the fixed points are approached,
for a given value of Λ∞ /H02 , is shown in Figures 3.18 and 3.19.
w¢ eff,0
w eff,0
-0.80 0.5
1.2
-0.85 1.5 0.4
1.8 2
2.4 0.3 1.2
-0.90
0.2 1.5
-0.95 1.8
0.1 2
2.4
-1.00 Weff,0 weff,0
0.5 0.6 0.7 0.8 0.9 1.0 -1.05 -0.95 -0.90 -0.85 -0.80
-1.05 -0.1
(a) (b)
0
Figure 3.18: HSS model: weff,0 as a function of Ωeff,0 and weff,0 as a function of weff,0
when n = 2. The fixed points are in (a) (Λ∞ / 3H02 , −1) and in (b) (−1, 0). Each

curve is labelled with its value of Λ∞ /H02 ; parameter c increases towards the fixed
point. The pink spot locates the example we term Model 6.
Of the plethora of possibilities, we choose just two to illustrate the nature and
properties of the HSS f (R). The chosen examples, both with Λ∞ = 1.2 H02 have
n = 6, which we call Model 5, or n = 2, which is Model 6, and appropriate values
of c to give Ωeff,0 = 0.7.
w¢ eff,0
0.4
w eff,0 1.2
-0.95
0.3
2.4 1.5
-1.00 Weff,0 0.2
0.5 0.6 0.7 2 0.8 0.9 1.0 1.8
1.8
2
-1.05 0.1
1.5
2.4
weff,0
-1.10 1.2 -1.10 -1.05 -0.95
-0.1
(a) (b)
0
Figure 3.19: HSS model: weff,0 as a function of Ωeff,0 and weff,0 as a function of weff,0
when n = 6. The fixed points are in (a) (Λ∞ / 3H02 , −1) and in (b) (−1, 0). Each

curve is labelled with its value of Λ∞ /H02 ; parameter c increases towards the fixed
point. The mauve spot locates the example we term Model 5.
f HRLH0 2 f ¢ HRL
N N
-1.5 -1.0 -0.5 -1.5 -1.0 -0.5
Model 5
-0.5
-0.05
-1.0
Model 6
-0.10
-1.5
Model 6 -2.0 -0.15

Model 5
-2.5
(a) (b)
Figure 3.20: HSS model: f (R)/H02 and fR (R) against N when Λ∞ = 1.2 H02 and for
two values of n. Model 5 has n = 6 and log10 c = −2.34 while Model 6 has n = 2 and
log10 c = −0.496.
RH0 2
350
300
250
200
150
100
Model 6 50
Model 5 N
-2.0 -1.5 -1.0 -0.5
Figure 3.21: HSS model: R/H02 against N for the curves of Figure 3.20. Model 5 has
R0 = 9.8 H02 while Model 6 has R0 = 8.7 H02 .
f ¢ HRL f ¢ HRL
0.0 RH0 2 RH0 2
6 8 10 12 14 10 20 30 40
-0.2 Model 6 -0.1
-0.4 -0.2
-0.3
-0.6
-0.4
-0.8 Model 5
-0.5
-1.0
(a) (b)
Figure 3.22: HSS model: fR (R) against R for Models 5 and 6, and three examples
when n = 2 with parameters close to the lower boundary of Figure 3.16. (a) shows how
far the values of R0 , denoted by spots, are from the turning points. In (b) the curves
have parameters (Λ∞ /H02 , log10 c), as follows: dot-dashed, pink curve, (1.6, −0.58);
blue, (2, −0.28); dashed, green (2.5, 0.05). The spots indicate the respective values of
R0 and indicate how close they are to the problematic values of R for which fRR (R) = 0.
Weff
weff 0.8
-0.9
0.6
-1.0 z
1 2 3 4 5 6
0.4
-1.1 Model 5 Model 6
Model 6 0.2 Model 5

-1.2
0.0 z
-1.3 1 2 3 4
(a) (b)
Figure 3.23: HSS model: weff and Ωeff against z for the curves of Figure 3.20.
Γ
0.5 Model 5
z
1 2 3 4 5 6 7
-0.5
-1.0
-1.5 Model 6
-2.0
Figure 3.24: HSS model: Evolution of γ against z for the examples of Figure 3.20.
Model 5 gives γ0 = 0.40 while Model 6 gives γ0 = 0.37.
3.3 Conclusion
The results of Planck do not conclusively show that we are living in a ΛCDM
Universe, though they do not rule that out. Essentially, there is still some doubt
as to today’s value of the effective dark energy equation of state parameter,
which we have called weff . It was decided to review the properties of some F (R)
models and to produce one of our own in order to see how large a deviation
from −1 we could obtain for weff,0 . Constraining weff,0 , constrains R0 for, using
R = 3H 2 (1 − 3weff Ωeff ) (from Useful Expressions on page 24), |1 + weff,0 | < δ
means that |R0 − RΛ,0 | < 9H02 Ωeff,0 δ. In this regard we went beyond the bounds
set by Planck and its predecessor, WMAP, to see what would happen but ensuring
that Ωeff,0 was always the same.
All F (R) = R+f (R) models are constrained quite severely by the Solar System
which has the effect of ensuring that f (R) lies close to −2Λ∞ until comparatively
recently. Until then, weff always lies close to −1. When f (R) then swings away
from −2Λ∞ then fR (R) swings away from being close to zero, though still neg-
ative, and weff swings away from −1, first in a negative sense, then reaching a
local minimum in the recent past before returning to give a value for weff,0 .
For each generic model, we have seen how varying the parameters of those
0
models varies the values of weff,0 and Ωeff,0 , how weff,0 and Ωeff,0 and how weff,0 and
weff,0 are connected when one specific parameter changes. We have also seen how,
if left unmodified, what the future of each model might be. In some instances, the
potential of the scalar field, φ = fR (R), reaches a stable minimum, at RdS , as a de
Sitter attractor which can be reached by R oscillating with decaying amplitude
about RdS or by simply decreasing to it without any oscillations. It could be that
RdS = 0. Alternatively, R tends to zero by wild oscillations with Ωeff oscillating
between positive and negative values thereby driving weff to infinity and back
again, repeatedly, with ever-increasing frequency.
3.3. Conclusion 99
The matter growth index was studied for all six examples which showed that,
while those models which kept close to ΛCDM at the present time gave values
of γ, the matter growth index, of 0.4 or slightly higher, those which swung away
markedly from ΛCDM gave values of γ0 less than 0.4. The perceived wisdom
to date is that F (R) models give γ0 in the approximate range 0.4 to 0.55. The
γ for ΛCDM, itself, increases from the value 6/11 = 0.545 at high curvature to
give γ0 = 0.555. It is also noticed that some of the values of γ go negative as
the Universe evolves. This is illustrated, but not remarked upon, in Appleby
and Weller’s paper [113]. The effect of letting γ go negative is to allow Ωγm and,
0
hence, δm /δm to be larger than they would otherwise be. Thus, whatever value
δm starts with at some earlier stage, it has a larger value today than it otherwise
would have. This is illustrated in Figure 3.25 for three models: standard Einstein
gravity, Erf Model 1, for which γ goes negative, and Erf Model 2, for which γ > 0.
Wm Γ ∆m ∆i
1 1000
1.0
2
0.8 800
g
sE
0.6
600
0.4
400
0.2
0.0 z 200 z
1 2 3 4 5 0.5 1.0 1.5 2.0 2.5 3.0
(a)
Figure 3.25: Comparing the effect of γ going negative as opposed to its always being
positive. Illustrated are standard Einstein gravity (sEg, dotted green), Erf Model 1
(1, continuous pink) and Erf Model 2 (2, dashed mauve). The evolution of γ for these
models is shown in Figure 3.9. Pane (a), above, shows how Ωγm varies with z while
(b) shows late detail of the evolution of the ratio δm /δi , where δi is the value of δm at
z = 1095, the point at which evolution of γ was started. For z & 3, in these cases the
values of δm /δi are indistinguishable. At z = 0, the respective values of δm /δi are 859
for standard Einstein gravity, 919 for Erf Model 2 and 953 for Erf Model 1.
The graphs were obtained by numerically solving (3.15) for γ and then integrat-
ing (3.14) for δm . The initial value of z was taken to be 1095, at which γ = 6/11
and δm = δi . Assuming Erf Model 1 to be an extreme case, we see that F (R)
models give a deviation, today, for γ of up to around 10% from the standard Ein-
stein gravity case but that cases in which γ does not go negative achieve almost
as much.
The next step could be to try to construct an F (R) model which could give
realistic values within their range of uncertainty to all the cosmological parameters
currently measured.
Chapter 4
The weff < −1 Theorem for F (R)

Models
4.1 Introduction
Excluding ΛCDM, all practical F (R) = R + f (R) models have the property that
f (R) → −2Λ, as R → ∞ and they all have graphs similar to Figure 3.1 in the
matter and radiation eras. Further, fR (R) < 0, fRR (R) > 0 such that fR (R) → 0
and fRR (R) → 0, as R → ∞. All instances of graphs of weff for these F (R)
models show that it is slightly less than −1 at high z becoming more negative
with time until it reaches a local minimum [47, 68, 97, 101, 115]. In this chapter,
we prove that in the radiation and matter eras, weff < −1 and decreases.
4.2 The theorem
Theorem: For F (R) models which tend to ΛCDM as R → ∞, which is a → 0,

weff decreases from −1 as a increases from 0 through the radiation and matter
eras.
101
102 Chapter 4. The weff < −1 Theorem for F (R) Models
Proof : To prove this we use weff = −1 − ρ0eff / (3ρeff ), which is (1.84), and show
that ρ0eff (N ) > 0 in the radiation and early matter eras. A difficulty arises in
that, at early times, f (R) lies extremely close to −2Λ, while being above it, for
a very long time; see Figure 3.1(a). ρ0eff (N ) will be numerically very small indeed
for much of the history of the Universe up to the present time. Figure 4.1, which
compares 8πGρeff /H02 with corresponding weff , shows one example.
8ΠGΡeff H0 2 ,weff

4
N
-2.0 -1.5 -1.0 -0.5
-1
Figure 4.1: A combined graph showing how 8πGρeff /H02 (continuous) and correspond-
ing weff (dashed) vary with N . Data are from Erf Model 1 of subsection 3.2.1.
As yet, making no assumptions, let our variable be N = log a, as usual, and

introduce a new parameter, χ = 8πGρeff /3. Thus
8πGρr 8πGρm 8πGρeff

H2 = + + (4.1)
3 3 3
−4N −3N
= αe + βe + χ, (4.2)
0
χ
⇒ weff = −1 − , (4.3)
3χ
where α and β are constants. χ will tend to Λ/3 as R → ∞. From this we find
2HH 0 = −3H 2 + 3χ + χ0 − αe−4N . (4.4)
R = 3H 2 + 9χ + 3χ0 − 8πGρr . (4.5)

4.2. The theorem 103
From the field equation (1.75),

2 R 1
3H∂0 FR (R) + 3H − FR (R) + F (R) = 3H 2 − 3χ (4.6)
2 2

R 1
⇒ 3H∂0 fR (R) + 3H 2 − fR (R) + f (R) = −3χ. (4.7)
2 2
From the trace, (1.76),
3FR (R) + RFR (R) − 2F (R) = 8πG (−ρ + 3P ) (4.8)
⇒ −3∂02 fR (R) − 9H∂0 fR (R) − R + RfR (R) − 2f (R) = −3H 2 + 3χ + 8πGρr
(4.9)
⇒ −3∂02 fR (R) − 9H∂0 fR (R) + RfR (R) − 2f (R) = 12χ + 3χ0 . (4.10)
Adding 4×(4.7) to (4.10) and then dividing by 3 gives

R
2
−∂0 fR (R) + H∂0 fR (R) + 4H − 2
fR (R) = χ0 . (4.11)
3
In the regions we are concerned with, χ0 and fR (R) are first order small quan-
tities. In expressing the values of the terms on the left hand side of (4.11) to
form (4.21), we make use the following zero-th order approximations, 2HH 0 =
− (4 − θ) H 2 , R = 3θ H 2 and R0 = −3R. The parameter θ has been introduced to
indicate where in the radiation and matter eras we are operating. In the matter
era, θ = 1 while, in the radiation era, θ = 0, except we must be careful not to let
R = 0 where so doing would be incorrect. θ takes values between 0 and 1.
We can express f (R) as a convergent power series in a but because it is so flat

in the radiation era and most of, if not all of, the matter era, we can express f (R)
as
∞
X
n
f (R) = −2Λ + a µ i ai , (4.12)
i=0
P∞
where the µi are constants, µ0 > 0, an i=0 µi ai > 0 and n will be very large. In
view of the flatness of f (R) and that 0 < a 1, at early times, we need only
104 Chapter 4. The weff < −1 Theorem for F (R) Models
consider the leading term, so let
f (R) = −2Λ + µ0 an , (4.13)
= −2Λ + µ0 enN , (4.14)
using N = log a. We now express fR (R) in terms of N and evaluate ∂0 fR (R) and
∂02 fR (R).
f 0 (R)
fR (R) = (4.15)
R0
nµ0 nN
= − e . (4.16)
3R
0
Hnµ0 enN

∂0 fR (R) = − (4.17)
3 R
n (n + 3) µ0 H nN
= − e . (4.18)
3R
n (n + 3) µ0 H 2 nN
H∂0 fR (R) = − e . (4.19)
3R
n (n + 3) (2n + 2 + θ) µ0 H 2 nN
∂02 fR (R) = − e . (4.20)
6R
Putting these into (4.11) gives
n [2n2 + (6 + θ) n − 8 + 5θ] µ0 H 2 nN
χ0 = e (4.21)
6R
n3 µ0 H 2 nN
' e > 0. (4.22)
3R
Initially, when a = 0 and N → −∞, χ0 will be zero because, although H 2 /R is

proportional to e−N in the radiation era, χ0 is proportional to e(n−1)N which tends
to zero as N → −∞. So starting from f (R) = −2Λ, where weff = −1, e(n−1)N
increases. That is, χ0 increases with N while, to very good approximation, χ
remains constant. χ0 /χ increases so that weff decreases into the phantom region.
It continues to do so until the proposed expression for f (R), (4.13), fails to be a
good approximation.
Chapter 5
Evolution of Perturbations
5.1 Introduction
In this chapter we consider matter and metric perturbations of the line element
in an FRW Universe. In the Newtonian gauge these can be encoded via
ds2 = − (1 + 2Ψ) dt2 + a2 (1 − 2Φ) δij dxi dxj , (5.1)
where Φ and Ψ are small, dimensionless perturbations to the FRW metric. We

attempt to produce algebraic solutions to the evolution equations (A1) and (A2)
and the linear Einstein equations (A3) and (A4) of Pogosian and Silvestri [79].
They are:

0 k 0 δP
δ + V − 3 (1 + w) Φ + 3 − w δ = 0, (5.2)
aH δρ

0 k δP Π k
V + (1 − 3w) V − − δ− (1 + w) Ψ = 0. (5.3)
aH δρ δ aH
k2

0 3Ei
(1 + fR ) 6Ψ + 6Φ + 2 2 2 Φ = − 2 δi + 3fRR δR0 − fR0 (6Ψ + 3Φ0 ) ,
aH H
H0 k2

0
− 3fRR 1 + − 2 2 fRR − 3fRR δR.
H aH
(5.4)
105
106 Chapter 5. Evolution of Perturbations
3a 1
(1 + fR ) (kΨ + kΦ0 ) = Ei Vi + k (fRR δR)0
2H 2
1 1
− kfRR δR − kfR0 Ψ. (5.5)
2 2
These equations are in Fourier space with δ ≡ δρ/ρ, the density contrast,
V ≡ (1 + w) v, where v is the scalar component of the velocity, and E ≡ H 2 /H02 .
Ei and δi are the various energy components of E and δ, respectively, so that as far
as matter perturbations are concerned, Ei δi = Em δm = 8πGρm δm / (3H02 ). δP is

the pressure perturbation which, for matter, is zero and ρ Π ≡ k̂ j k̂i − δ j i /3 π i j
is the anisotropic stress which is zero because π i j is the traceless component of
the energy-momentum tensor [79]. In particular, Πm = 0.
(5.4) is a corrected version of (A3) with three typographical errors removed. As

a check they were compared to the equivalent equations in Ma and Bertschinger
[116]. 0 means differentiation with respect to N = log a. We solve these equations,
as applied to matter perturbations only, for a given co-moving k.
Some work, in this area, had already been done on this by Starobinsky [78],
by Appelby et al. [101] and by Elizalde et al. [117], but they only considered
what happens at early times in the matter era and the approach was different
from ours. No initial conditions were applied. Tsujikawa [112] adopted different
approach and found expressions for the perturbed potentials as powers of t. Our
challenge was to produce a solution which remained viable up to the present time.
In trying to produce an algebraic solution to these equations, a series approach

in terms of a single variable was adopted. This variable was chosen to remain
as small as possible throughout the history of the Universe. The four equations
above, were reduced, with the aid of the field equation and an expression for δR,
to two coupled, second order, homogeneous, linear differential equations in Φ and
Ψ. Approximate algebraic expressions in terms of this variable were then found
for Φ and Ψ which contained both oscillating and non-oscillating components.
In order to illustrate the oscillations, graphically, it was necessary to define as

5.2. Matter perturbation equations in k-space for the Jordan Frame 107
x(N ) a “difference” in Φ and Ψ, as defined in (5.54) different from Φ − Ψ. To

complement this, the sum, y(N ), was also constructed, (5.55). Initial conditions
were applied to Φ and Ψ. Using the same initial conditions, the two differen-
tial equations were integrated, numerically, and the functions corresponding to
the algebraic solutions for x(N ) and y(N ) produced and compared for differing
models and for different values of k.
It was found that, while the algebraic and numeric solutions were in general
agreement when k & 0.1h Mpc−1 , smaller values of k produced less agreement in
the later stages of the Universe’s history. Also noticed in the solution for x(N )
was a “jump” which is due to the non-oscillating component of x(N ).
5.2 Matter perturbation equations in k-space

for the Jordan Frame
The equations we solve are those of Pogosian and Silvestri [79]. Subscripts “m”
refer to the matter component and the anisotropic stress is taken to be zero.
The anisotropy equation with no anisotropic contribution from matter, equation

(30) of [79], is
FR (Φ − Ψ) = FRR δR. (5.6)
while the evolution equations for matter density perturbations and matter fluid
velocity are
0 k
δm + Vm = 3Φ0 , (5.7)
aH
k
Vm0 + Vm = Ψ. (5.8)
aH
These are equations (A1) and (A2) of [79] with w = 0 and δP/δρ set to zero since
P , the pressure of matter, is zero at all times. Π is also set to zero since we take
π i j of equations (26) of [79] to be zero.
The 00 and 0, i components of the Einstein equations, perturbed to first order,

give, with some manipulation,
2
k 2 R
FR 2 (Φ + Ψ) − 6H − (Φ − Ψ)
a 2

2 0 0 3aH
+ 3H R FRR (Φ + Ψ) = −8πGρm δm + Vm ,(5.9)
k
8πGρm a
FR (Φ0 + Ψ0 + Φ + Ψ) − R0 FRR (Φ − 2Ψ) = Vm . (5.10)
Hk
(5.9) is derived from (A3) and (A4) by eliminating (FRR δR)0 , (5.10) is (A4), (5.7)
is (A1) and (5.8) is (A2). ρ0m = −3ρm .
The perturbed Ricci scalar in k-space is [112, 118]
4k 2 2k 2 2 00 0 2
0
δR = − Φ + Ψ − 6H (Φ + Ψ ) − 12H + R Φ − 2RΨ. (5.11)
a2 a2
It should be borne in mind that, when converting a derivative with respect to

time to one with respect to N = log a, d/dt = H d/dN , so that, for example,
Φ̈ = HH 0 Φ0 + H 2 Φ00 .
If we can ignore radiation then the 00 component and the trace of the field
equation give
6H 2 R02 FRRR = −3F + R + 6H 2 FR − R R0 + 6H 2 R00 FRR . (5.12)

This will be needed when differentiating (5.10) and using (5.8) to eliminate Vm .
Introducing two new variables
3H 2 FRR
ξ = , (5.13)
FR
k
λ = , (5.14)
aH
and eliminating δR from (5.6) and (5.11) gives
2λ2

00 R 0 1
Φ + 2+ Φ + + Φ
6H 2 2ξ 3
λ2

0 1 R
+Ψ − + − Ψ = 0, (5.15)
2ξ 3 3H 2
5.2. Matter perturbation equations in k-space for the Jordan Frame 109
and, from (5.8), (5.10) and (5.12),

00R 0 F
Φ + 2+ Φ + Φ
6H 2 2H 2 FR
R0 ξ

00 R 0 R 3F
+ Ψ + 2+ + 2 Ψ + − Ψ = 0. (5.16)
6H 2 H H 2 2H 2 FR
Now using R = 12H 2 + 6HH 0 and F = R + f these two equations become
H0 2λ2

00 0 1
Φ + 4+ Φ + + Φ
H 2ξ 3
λ2 H0

0 1
+Ψ − + −1− Ψ = 0, (5.17)
2ξ 3 2H
H0 3H 0

00 0 f
Φ + 4+ Φ + 6+ + Φ
H H 2H 2 (1 + fR )
H 0 R0 ξ 3H 0

00 0 3f
+ Ψ + 4+ + 2 Ψ − 6+ + Ψ = 0. (5.18)
H H H 2H 2 (1 + fR )
In solving (5.17) and (5.18), it is assumed that F (R) is ΛCDM at an early value
of a in the matter era [79]. Numerical solution of these equations is relatively
straightforward. Before attempting to solve them algebraically, we look at the
ΛCDM case.
5.2.1 The ΛCDM model
In ΛCDM, F (R) = R − 2Λ and ξ = 0. Equation (5.6) gives Φ = Ψ and (5.16)

becomes
5R − 12Λ 0 3Λ
Φ00 + Φ + Φ = 0. (5.19)
2 (R − 3Λ) R − 3Λ
Following Pogosian et al. in [79], the initial conditions are Φ = −1 and Φ0 = 0.

Numerical solution of this is shown in Figure 5.1, where we see that, today, Φ
has risen to around −0.78. Algebraic solution of (5.19) could be attempted. By
substituting for R in terms of N , using R = 3H02 Ωm,0 e−3N + 4Λ, gives Φ =
A + Be−5N/2 as an approximating solution at early times, and Φ = Ce−N at late
times as R → 4Λ. For a better solution than this, a series approach would need
to be taken using a variable such as p = Λ/ (R − 3Λ).
F
F z
N = log a 1 2 3 4
-6 -4 -2 2 4 6 -0.75
-0.2
-0.80
-0.4 -0.85
-0.6 -0.90
-0.8 -0.95
-1.0 -1.00
(a) (b)
Figure 5.1: ΛCDM: Numerical solution of (5.19) shown against (a) N = log a and
also (b) z. The value of Λ has been chosen to give Ωm,0 = 0.3, namely, Λ = 2.1 H02 .
5.3 Approximating algebraic solutions
From now on, the dependence of variables on N will be shown. Subtracting (5.17)
and (5.18) gives
2λ(N )2 3H 0 (N )

1
+ −6− − (N ) Φ(N )
2ξ(N ) 3 H
H 0 (N ) R0 (N )ξ(N )

00
= Ψ (N ) + 3 + + 2
Ψ0 (N )
H(N ) H(N )
2
5H 0 (N )

1 λ(N )
+ + − 10 − Ψ(N ) − 3(N )Ψ(N ) (5.20)
2ξ(N ) 3 H(N )
and equation (5.18) can be re-expressed as
H 0 (N )

00 00
Φ (N ) + Ψ (N ) + 4 + [Φ0 (N ) + Ψ0 (N )]
H(N )
R0 (N )ξ(N ) 0
= − Ψ (N )
H(N )2
3H 0 (N )

− 6+ [Φ(N ) − Ψ(N )] − (N ) [Φ(N ) − 3Ψ(N )] . (5.21)
H(N )
5.3. Approximating algebraic solutions 111
Three parameters have been introduced in order to reduce the apparent complex-
ity of these equations. They are ξ(N ), (N ) and λ(N ) defined as follows:
3H(N )2 FRR [R(N )]
ξ(N ) = , (5.22)
FR [R(N )]
f [R(N )]
(N ) = 2
, (5.23)
2H(N ) (1 + fR [R(N )])
k k e−N
λ(N ) = = . (5.24)
aH(N ) H(N )
In order to introduce a degree of symmetry into equation (5.20) we replace ξ(N )
with ζ(N ) defined by
1 1
= + λ2 (N ). (5.25)
ζ(N ) ξ(N )
This changes (5.20) into

H 0 (N ) R0 (N )ξ(N )

00
Ψ (N ) + 3 + + 2
Ψ0 (N )
H(N ) H(N )
2 0

1 1 λ(N ) 5H (N )
+ − − 10 − − 3(N ) Ψ(N )
2 ζ(N ) 3 H(N )
λ(N )2 3H 0 (N )

1 1
− + −6− − (N ) Φ(N ) = 0, (5.26)
2 ζ(N ) 3 H
while (5.21) is re-written as
H 0 (N )

00 00
Φ (N ) + Ψ (N ) + 4 + [Φ0 (N ) + Ψ0 (N )]
H(N )
3H 0 (N )

+ 6+ [Φ(N ) − Ψ(N )] + (N ) [Φ(N ) − 3Ψ(N )]
H(N )
R0 (N )ξ(N ) 0
= − Ψ (N ). (5.27)
H(N )2
Equations (5.26) and (5.27) are the two differential equations we shall attempt
to solve algebraically. We also notice that they are homogeneous and that at
early times (N ) ∼ 0, λ(N ) ∼ 0 and ζ(N ) ∼ 0, which suggests, by inspection
of equation (5.26), the possibility of high frequency oscillations in Ψ(N ) and,
therefore, also in Φ(N ). If there are oscillations, their frequency could be of
p
order 1/ ζ(N ) . There will also be solutions for Φ(N ) and Ψ(N ) which do not
oscillate.
Because (5.26) and (5.27) are homogeneous, oscillating solutions will be inde-
pendent of non-oscillating solutions. Thus we can set Φ(N ) and Ψ(N ) each to be
the sum of an oscillating component and a non-oscillating component. We shall
express Φ(N ) and Ψ(N ) as follows:
λ(N )2

1
Φ(N ) = Φosc (N ) + − ζ(N )Φ0 (N ) + ζ(N )Φ2 (N ), (5.28)
ζ(N ) 3
λ(N )2

1
Ψ(N ) = Ψosc (N ) + + ζ(N )Ψ0 (N ) + ζ(N )Ψ2 (N ). (5.29)
ζ(N ) 3
Φosc (N ) and Ψosc (N ) are the oscillating components while Φ0 (N ) and Φ2 (N ) are
non-oscillating components of Φ(N ) and similarly for Ψ(N ). The Φ0 (N ) and
Ψ0 (N ) terms are zero-th order in ζ(N ), at early times. Using (5.25), the Φ0 (N )
term migrates to 2/3 × Φ0 (N ) at late times while the Ψ0 (N ) term migrates to
4/3 × Ψ0 (N ). The reason for expressing these terms in this way will become clear
in subsection 5.3.2. The Φ2 (N ) and Ψ2 (N ) terms are first order in ζ(N ).
5.3.1 Oscillatory solution
Wishing to keep the expressions for Φosc (N ) as simple as possible, we write
Ψosc (N ) = Ψ1 (N ) sin ω(N ), (5.30)
where Ψ1 (N ) is a function of N to be determined. Inspection of (5.26) shows that

Φ(N ) is a linear function of Ψ00 (N ), Ψ0 (N ) and Ψ(N ), which means that Φ(N )
could contain both sine and cosine terms. It also means that the frequency of
Φ(N ) will be the same as Ψ(N ). We therefore write
Φosc (N ) = Φ1 (N ) sin ω(N ) + Φ3 (N ) cos ω(N ). (5.31)
The expressions for Φosc (N ) and Ψosc (N ) contain four unknown functions, viz,
Φ1 (N ), Ψ1 (N ), Φ3 (N ) and ω(N ). Taking the sine and cosine components of
equations (5.26) and (5.27) gives four independent equations. Unfortunately,

because of the unknown frequency, and its integral with respect to N , ω(N ),
they are not linear and are, therefore, intractable. Therefore we must adopt a
series approach and find expressions for the unknowns in terms of some variable
which remains small throughout the history of the Universe.
Possibilities for this variable at early times are ζ(N ), ξ(N ) and 1/λ(N )2 but
how small do they remain? At early times in the matter era, λ(N )2 ∝ a
so it increases with time but we are dealing with subhorizon scales for which
k/aH(N ) = λ(N ) 1 which is why 1/λ(N )2 might be a contender. k = 0,
however, means that 1/λ(N ) does not even exist. Figure 5.2 shows an example
using Model 1 of the Erf Model of Chapter 2 with k = 0.1h Mpc−1 . The graphs
of ζ(N ), ξ(N ) and 1/λ(N ) are plotted.
Ζ, Ξ, 1Λ2
Ζ, Ξ, 1Λ2
0.08 0.012
0.010
0.06
0.008
0.04 0.006
0.004
0.02
0.002
N N
-1.5 -1.0 -0.5 -1.5 -1.4 -1.3 -1.2 -1.1
(a) (b)
Figure 5.2: Graphs of ζ(N ) (continuous, blue), ξ(N ) (dashed, green) and 1/λ(N )2
(dot-dashed, mauve) against N . (b) shows early detail.
We see that ζ(N ) is tangential to ξ(N ) at early times, as can be seen also from
(5.22) - (5.24), but at much later times it becomes closer to 1/λ(N )2 . As the
parameters of an f (R) model are changed, or as we change the model itself, the
numerical values of these three variables will change but not their relation with
respect to one another. Thus the variable which remains smallest is ζ(N ). At
early times it is vanishingly small. This is on account of the requirement for
F (R) to pass the Solar System test (see subsection 1.4.8). In the Solar System,
ζs ≈ ξs ≈ Rs FRR (Rs ) < 10−23 . At early times, any series expansion in ζ(N ) will
require very few terms in ζ(N ) to be accurate, whereas, at later times, judging
from Figure 5.2, it is possible that more terms would be required, with k → 0
posing a problem. For any particular model, however, ζ(N ) will be smallest, at
late times, when k is largest.
Therefore, ζ(N ) will be chosen as the variable with which to express Φ1 (N ),

Ψ1 (N ) and Φ3 (N ). Accuracy will reduce, as N increases, unless the number of
terms in ζ(N ) increases. Accuracy also reduces as k becomes smaller.
The sine component of (5.26) gives
λ(N )2 λ(N )2

0 2 1 1 Φ1 (N )
ω (N ) = − − + . (5.32)
2ζ(N ) 6 2ζ(N ) 6 Ψ1 (N )
All the terms on the right hand side of (5.32) are numerically much greater than
unity. In finding (5.32), terms of order 1 or smaller have been ignored. These are
Φ1 (N )/Ψ1 (N ), H 0 (N )/H(N ), Ψ01 (N )/Ψ1 (N ) and Ψ001 (N )/Ψ1 (N ). Similarly, the
sine component of (5.26)+(5.27) gives
λ(N )2 λ(N )2

0 2 1 1 Ψ1 (N )
ω (N ) = + − − , (5.33)
2ζ(N ) 6 2ζ(N ) 6 Φ1 (N )
plus other terms of order 1 and also terms of uncertain order, at this stage, viz,
H 0 (N ) ω 00 (N ) Φ3 (N )ω 0 (N ) ω 0 (N )Φ03 (N )

4+ + +2 . (5.34)
H(N ) ω(N ) Φ1 (N ) Φ1 (N )
The cosine component of (5.26) gives
3 H 0 (N ) ξ(N )R0 (N ) Ψ01 (N ) ω 00 (N )

+ + + +
2 2H(N ) 2H(N )2 Ψ1 (N ) 2ω 0 (N )
λ(N )2

1 1 Φ3 (N )
= + , (5.35)
4 ζ(N ) 3 Ψ1 (N )ω 0 (N )
in which the other terms:
H 0 (N )

(N ) Φ3 (N )
3+ +3 (5.36)
2 H(N ) Ψ1 (N )ω 0 (N )
have been discarded as being of smaller value.

p
Since 1/ω 0 (N ) ∼ ζ(N ) , we can see from (5.35) that Φ3 (N )/Ψ1 (N ) is of order
p
ζ(N )ω 0 (N ) ∼ ζ(N ) so that the terms in (5.34) are of order ζ(N )ω 0 (N )2 ∼ 1
and need not appear in (5.33).
Equating (5.32) and (5.33), gives Φ1 (N ) = −Ψ1 (N ) or

(1/ζ(N ) + λ2 (N )/3) Φ1 (N ) = (1/ζ(N ) − λ2 (N )/3) Ψ1 (N ). The second of these
gives ω 0 (N ) = 0, which we reject, while the former gives
1
ω 0 (N ) = p . (5.37)
ζ(N )
We already have the cosine component of (5.26) as (5.35). The cosine compo-
nent of (5.27) gives,
H 0 (N ) Ψ01 (N ) ω 00 (N ) λ(N )2

3 1 1 0 2 Φ3 (N )
+ + + 0 = + − 2ω (N )
2 2H(N ) Ψ1 (N ) 2ω (N ) 4 ζ(N ) 3 Ψ1 (N )ω 0 (N )
λ(N )2

1 1 Φ3 (N )
= − + . (5.38)
4 ζ(N ) 3 Ψ1 (N )ω 0 (N )
The term ξ(N )R0 (N )/ (2H(N )2 ) of equation (5.35) varies from O(ζ(N )) at early
times to O(10−1 ), today, say, so should not be discarded.
From (5.35) and (5.38), it is deduced that

p ξ(N )R0 (N )
Φ3 (N ) = ζ(N ) Ψ1 (N ) , (5.39)
H(N )2
s 1/4
−3(N −Ni )/2 H(Ni ) ζ(N )
Ψ1 (N ) = Ψ1 (Ni )e e−ν(N ) , (5.40)
H(N ) ζ(Ni )
Z N
λ(n)2 ζ(n) ξ(n)R0 (n)

where ν(N ) = 1− dn. (5.41)
Ni 3 4H(n)2
Ni is some initial value for N . For much of the history of the Universe, e−ν(N )
is close to 1 and increases very slowly. For example, for Erf Model 1, with
k = 0.1h Mpc−1 , it rises from around unity at z = 1 to achieve a value of 1.14,
today. When k = 0, today’s value rises with this model to around 1.22.
Thus, there are oscillations from some early time in the matter era, to date,
with frequency approximated by equation (5.37). Oscillations at early time have
been studied by Song et al. in [119] and Starobinsky in [78]. Starobinsky looked
at the oscillating nature of R, at high curvatures only, and concluded that it
p
possessed an oscillating component with a frequency equivalent to 1/ ξ(N ) but
the amplitude of these oscillations grew into the past. The same effect has been
studied in [120] and is discussed further in Chapter 6.
5.3.2 Non-oscillatory solution
The non-oscillatory terms of Φ(N ) and Ψ(N ) are expressed in terms of two non-
oscillatory functions, respectively, of N as
λ(N )2

1
Φn (N ) = − ζ(N )Φ0 (N ) + ζ(N )Φ2 (N ), (5.42)
ζ(N ) 3
λ(N )2

1
Ψn (N ) = + ζ(N )Ψ0 (N ) + ζ(N )Ψ2 (N ). (5.43)
ζ(N ) 3
The reason for the rather complicated looking coefficients of Φ0 (N ) and Ψ0 (N )

is to simplify the relation between Φ0 (N ) and Ψ0 (N ) , as will be seen.
Applying these expressions to (5.26) and (5.27) and adopting a series approach,
gives Φ0 (N ) = Ψ0 (N ) satisfying
H 0 (N )

Ψ000 (N )
+ 4+ Ψ00 (N )
H(N )
0 2

2 H (N ) 2λ(N ) ζ(N )
− λ(N ) ζ(N ) 2 + + (N ) 1 + Ψ0 (N ) = 0,
H(N ) 3
(5.44)
which is algebraically soluble when |λ(N )2 ζ(N )| 1 and |(N )| 1. Under

ΛCDM, (5.44) is the same as equation (5.19).

λ(N )2 λ(N )2
From (5.26), an expression for 1 + 3 ζ(N ) Φ2 (N )− 1 − 3
ζ(N ) Ψ2 (N )
(see equation (5.52)) is found in terms of Ψ0 (N ) and derivatives. Note that this
approximates to Φ2 (N ) − Ψ2 (N ) at early times but could be markedly different
at late times.
There is another expression, a complicated second order differential equation

for Φ2 (N ) and Ψ2 (N ), which is obtained from (5.27). It is not symmetric in
Φ2 (N ) and Ψ2 (N ) and, even with the use the expression above in terms of Φ2 (N )
and Ψ2 (N ), it could only be solved numerically. If we wanted to solve these
equations for Φ2 (N ) and Ψ2 (N ), separately, we run into the problem that we
can no longer assume ζ(Ni ) = 0 because, in the expressions for Φ(N ) and Ψ(N ),
equations (5.28) and (5.29), Φ2 (N ) and Ψ2 (N ) appear as multiples of ζ(N ). It
therefore seems better, instead of showing x(N ) = Φ(N ) − Ψ(N ), as [79] does,
λ(N )2 λ(N )2
to show x(N ) = (1 + 3
ζ(N ))Φ(N ) − (1 − 3
ζ(N ))Ψ(N ); see (5.54). This
approximates to Φ(N ) − Ψ(N ) at early times and means that contributions from
Φ2 (N ) and Ψ2 (N ) can be shown in the algebraic expression for x(N ).
5.3.3 The combined algebraic solution
Putting the independent solutions from subsections 5.3.1 and 5.3.2 together gives
a suggested algebraic solution to equations (5.26) and (5.27), up to O(ζ(N )), as
Φ(N ) = −Ψ1 (N ) sin ω(N ) + Φ3 (N ) cos ω(N ),

λ(N )2

1
+ − ζ(N )Ψ0 (N ) + ζ(N )Φ2 (N ), (5.45)
ζ(N ) 3
Ψ(N ) = Ψ1 (N ) sin ω(N ),
λ(N )2

1
+ + ζ(N )Ψ0 (N ) + ζ(N )Ψ2 (N ), (5.46)
ζ(N ) 3
Z N
1
ω(N ) = p dn + ω(Ni ), (5.47)
Ni ζ(n)
p ξ(N )R0 (N )
Φ3 (N ) = ζ(N ) Ψ1 (N ) , (5.48)
H(N )2
Z N
λ(n)2 ζ(n) ξ(n)R0 (n)

ν(N ) = 1− dn, (5.49)
Ni 3 4H(n)2
s 1/4
−3(N −Ni )/2 H(Ni ) ζ(N )
Ψ1 (N ) = Ψ1 (Ni )e e−ν(N ) , (5.50)
H(N ) ζ(Ni
in which Ni is a constant. Ψ0 (N ) solves
H 0 (N )

00
Ψ0 (N ) + 4 + Ψ00 (N )
H(N )
H 0 (N ) 2λ(N )2 ζ(N )

2
− λ(N ) ζ(N ) 2 + + (N ) 1 + Ψ0 (N ) = 0.
H(N ) 3
(5.51)

λ(N )2 λ(N )2
The expression for 1 + 3
ζ(N ) Φ2 (N )− 1 − 3
ζ(N ) Ψ2 (N ), denoting
λ(N )2 ζ(N )/3 by lz(N ), is
H 0 (N )

00 0
2 [1 + lz(N )] Ψ0 (N ) + 2 (1 + lz(N )) 3 + + 4lz (N ) Ψ00 (N )
H(N )
H 0 (N )

− 4 (1 + 4lz(N )) 2 + + 4(N ) (1 + 2lz(N )) Ψ0 (N )
H(N )
H 0 (N )

0 00
+ 2lz (N ) 3 + + 2lz (N ) Ψ0 (N ). (5.52)
H(N )
Because this is the only tractable connection between Φ2 (N ) and Ψ2 (N ) and

to emphasise the oscillations by removing Ψ0 (N ), we consider the “difference”
between Φ(N ) and Ψ(N ), x(N ), as already mentioned, above, by defining it as
λ(N )2 λ(N )2

x(N ) = 1+ ζ(N ) Φ(N ) − 1 − ζ(N ) Ψ(N ) (5.53)
3 3
At early times, λ(N )2 ζ(N ) 1, so, then, x(N ) ∼ Φ(N ) − Ψ(N ).
We define y(N ) to be the sum of Φ(N ) and Ψ(N ). In it, we ignore any contri-
bution from Φ2 (N ) and Ψ2 (N ) because y is dominated by Ψ0 (N ), which approx-
imates to −1, and they are suppressed by the factor ζ(N ).
In terms of their various components, x(N ) and y(N ) are given by
λ(N )2

x(N ) = −2Ψ1 (N ) sin ω(N ) + 1 + ζ(N ) ζ(N )Φ2 (N )
3
λ(N )2

− 1− ζ(N ) ζ(N )Ψ2 (N ), (5.54)
3
y(N ) = Φ3 (N ) cos ω(N ) + 2Ψ0 (N ). (5.55)
We now go on to apply initial conditions to these at some value of N we call Ni .

5.3.4 Applying initial conditions to the algebraic approx-

imation
We apply the initial conditions at N = Ni at which time we assume there to be

only slight deviation from ΛCDM, for which ΦΛ (N ) = ΨΛ (N ) and Φ0Λ = 0. Here,
ζ(N ) ∼ 0 so that x(N ) ∼ Φ(N ) − Ψ(N ) and Φ3 (N ) ∼ 0. y(N ) = Φ(N ) + Ψ(N ) ∼
2Ψ0 (N ). Near N = Ni , let Φ(N ) = ΦΛ (N ) + δΦ and Ψ(N ) = ΦΛ (N ) + δΨ. At
N = Ni ,
x(Ni ) = Φ(Ni ) − Ψ(Ni ) (5.56)
= δΦi − δΨi . (5.57)
To keep things simple, we’ll choose the phase of the oscillations so that x(Ni ) = 0
but let us choose Ni to be so early that we may consider , δΦi = δΨi = 0. Let
x0 (Ni ) = x0i . y(Ni ) = 2ΦΛ (Ni ) = 2Φ0 (Ni ) ≈ −2 and y 0 (Ni ) = 0.
x0i = δΦ0i − δΨ0i , (5.58)
yi = 2ΦΛ (Ni ) + δΦi + δΨi , (5.59)
⇒ 0 = 2Φ0Λ (Ni ) + δΦ0i + δΨ0i . (5.60)
Thus, δΦ0i = −Φ0Λ (Ni ) + x0i /2 and δΨ0i = −Φ0Λ (Ni ) − x0i /2 and the initial conditions
are
Φ(Ni ) = ΦΛ (Ni ), (5.61)
Ψ(Ni ) = ΦΛ (Ni ), (5.62)
Φ0 (Ni ) = x0i /2, (5.63)
Ψ0 (Ni ) = −x0i /2. (5.64)
The values for ΦΛ (N ) and ΨΛ (N ) can be found at N = Ni by solving the ΛCDM

equations of subsection 5.2.1.
We choose the value of Ni to be the value at which we can reliably start the
iterations which solve the original field equation, (1.75), for whichever model we
are using. For values of N < Ni , F (R) is so close to ΛCDM that the difference
in not discernible although the oscillations are suppressed into the past. They
increase in frequency but their amplitude, given by Ψ1 (N ) in (5.50), dies away
to zero.
Applying the initial conditions to the algebraic solution gives

p
ω(Ni ) = 0, Ψ1 (Ni ) = −x0i ζ(Ni ) /2, Φ0 (Ni ) = ΦΛ (Ni ) and Φ00 (Ni ) = 0. The
value of Φ3 (Ni ) will be taken to be as defined by (5.48). This makes Ψ1 (N ), as
specified in (5.50), become
s
x0 H(Ni )
Ψ1 (N ) = − i e−3(N −Ni )/2 ζ(Ni )1/4 ζ(N )1/4 e−ν(N ) . (5.65)
2 H(N )
From (5.54), the amplitude of the oscillations in x(N ) is 2Ψ1 (N ).
5.4 Algebraic and numerical solutions compared
In this section we take three values of k for Erf Model 1 (subsection 3.2.1) and
compare the algebraic solutions of (5.54) and (5.55) to the equivalent solutions
found by solving equations (5.26) and (5.27), numerically.
The values of k we choose, to demonstrate the accuracy of the algebraic solu-

tions in relation to the numerical solutions, are k = 0.1h Mpc−1 , k = 0.01h Mpc−1
and k = 0. For demonstration purposes in order to make the oscillations stand
out, x0i is initially taken to be 10 but is reduced to 1 in subsection 5.4.1 to see
what effect that has. In all the graphs it should be remembered that they show
a history and that today is represented by N = 0.
It should also be borne in mind that the amplitude today, for a given model,
depends not only on the value of x0i but also in the the value of Ni at which the ini-
tial conditions are applied. In practice, it is unrealistic to apply initial conditions
at z = 1000 as is stated in [79] “when the deviations from GR are small”. At this
5.4. Algebraic and numerical solutions compared 121
early value of z, however, the oscillation frequency will be extremely high then;
computing facilities would not be able to cope. In this thesis, initial conditions
are applied at a value of Ni so that the algebraic and numeric solutions can be
compared fairly. However, it should be noted that the earlier initial conditions
are applied, the smaller will be the amplitude of the oscillations, today. Putting
N = 0 into the expression for the amplitude of oscillations of x(N ), derived from
(5.65), gives for the amplitude, today, of oscillations
s
H(Ni ) 1/4
x0i e3Ni /2 ζ(Ni )1/4 ζ0 e−ν0 (5.66)
H0
= x0i e3Ni /4 ζ(Ni )1/4 (Ωm,0 ζ0 )1/4 e−ν0 , (5.67)
and e3Ni /4 ζ(Ni )1/4 is smaller the earlier Ni is. This is because, at early times,
3H(N )2 ≈ R(N ) and FR (R(N )) ≈ 1, and, using (5.25) and (5.22), ζ(N ) ≈
ξ(N ) ≈ R(N ) fRR (R(N )) ∝ e−3N fRR (R(N )) so that e3Ni /4 ζ(Ni )1/4 ∝ [fRR (R(Ni ))]1/4 .
An earlier value of Ni gives a smaller value of fRR (R(Ni )). We have discounted
any changes in ν0 , see (5.49), as ν(N ) is close to zero for much of its history.
5.4.1 k = 0.1h Mpc−1
We present the full range solutions, first, the algebraic solution in Figure 5.3 and,
secondly, the numerical solution in Figure 5.4. The corresponding solutions for
y(N ) are given in Figure 5.5.
While Figures 5.3 and 5.4 look very similar there are some differences. The
main difference is that, when trying to plot the algebraic solution for x(N ), it
RN p
was impossible to compute Ni 1/ ζ(n) dn for a sufficient number of values of
N in the range Ni ≤ N ≤ 0, so an interpolating function had to be used. While
RN p
this function agrees with Ni 1/ ζ(n) dn , initially, its derivative does not always
do so which makes the frequencies, and therefore the phase, appear to disagree
slightly as N increases. Fortunately these differences remain relatively small, as
is discussed, below, and illustrated in Figure 5.6.
0.004
algebraic
0.002
N
-1.5 -1.0 -0.5
-0.002
Figure 5.3: Algebraic solution, (5.54), for Erf Model Model 1 for k = 0.1h Mpc−1 .
0.004
numeric
0.002
N
-1.5 -1.0 -0.5
-0.002
Figure 5.4: Numerical solution, as defined by (5.53), of equations (5.26) and (5.27)
for Erf Model 1, for k = 0.1h Mpc−1 .
y
N
-1.5 -1.0 -0.5
-1.85
-1.90
-1.95
-2.00
-2.05
algebraic -2.10
numeric
Figure 5.5: Algebraic and numeric solutions, y(N ) = Φ(N ) + Ψ(N ), where Φ(N ) and
Ψ(N ) solve equations (5.26) and (5.27), for Erf Model 1, for k = 0.1h Mpc−1 . There
is close agreement between the two solutions for much of the history of the Universe
but, at later times, some separation is evident.
x x
0.0015
0.0010 0.004
0.0005
0.002
0.0000 N
-1.55 -1.54 -1.53 -1.52 -1.51 -1.50
-0.0005 0.000 N
-1.4 -1.3 -1.2 -1.1 -1.0
-0.0010
-0.002
-0.0015
(a) (b)
Figure 5.6: Comparison between algebraic and numeric solutions for x(N ) for Erf
Model 1 when k = 0.1h Mpc−1 formed by overlaying them. That is, in (a) and (b),
both algebraic and numeric solutions for x(N ) are plotted. (a) shows −1.56 < N < 1.5
while (b) shows 1.5 < N < 1.0. On these scales, the respective frequencies seem to be
very close, as do the envelopes.
Figure 5.6 shows that the envelopes of the algebraic and numeric solutions are
very similar as are the frequencies. Note that these graphs do not go right down
p
as far as Ni . If we can assume the frequency is represented by 1/ ζ(N ) and the
phases agree at N = N i, then the phase at all times is represented by
Z N
1
ω(N ) = p dn. (5.68)
Ni ζ(n)
p
It is unfortunate that, because of the very large value of 1/ ζ(N ) at early times,
sine and cosine of ω(N ) cannot be fairly computed for all values of N and that
an approximation has to be used in order to plot the algebraic graphs. This
introduces errors as shown in Figure 5.7(a) which shows the fractional difference
p
between the the frequency represented by 1/ ζ(N ) and the approximating func-
tion; showing how closely they agree. Fractional differences when the frequency
is high are to be expected. Fortunately, the disagreement is not large as can be
seen from Figure 5.6. Since the disagreement in Figure 5.7(a) occurs at the same
value of N as that in Figure 5.7(b) and noting the general agreement between the
p
two pairs of curves in Figure 5.6, it seems reasonable to suppose that 1/ ζ(N )
does represent a good approximation to the oscillation frequency at any time.
p
Figure 5.8 shows how the frequency, as represented by 1/ ζ(N ) , changes with
N . Initially, it is very high but falls to a minimum (in this case at around
N = −1.35) which corresponds to z ≈ 2.9 whereafter it increases to a local
maximum at N ≈ −0.7 (z ≈ 1).
The value x0i = 10 was chosen to make the oscillations highly visible. Let us
now see what the effect of reducing its value to unity is. See Figure 5.9. The
amplitude of the oscillations will be reduced but should enable the underlying
structure to be seen more clearly.
We see that the strange “jump” in the region −1.5 < N < −1.4 is still there.
It is clear that it is due to non-oscillating term in (5.54), namely,
λ(N )2 λ(N )2

ζ(N ) 1 + ζ(N ) Φ2 (N ) − 1 − ζ(N ) Ψ2 (N ) (5.69)
3 3
D Ω¢
Ω¢
Dx
0.01
N 0.0002
-1.4 -1.2 -1.0 -0.8
-0.01
0.0001
-0.02
-0.03 N
-1.5 -1.4 -1.3 -1.2 -1.1 -1.0
-0.04 -0.0001
-0.05 -0.0002
-0.06
(a) (b)
Figure 5.7: (a) The relative difference between frequency, in terms of the simpli-
fying interpolating function used to plot the algebraic curves, and the frequency as
p
represented by 1/ ζ(N ) . (b) The difference between the algebraic solution, using
an interpolating function for the phase of the sinusoidal components, and the numeric
solution against N for Erf Model 1 when k = 0.1h Mpc−1 . The spike in the relative
frequency just before N = −1.4 has introduced a phase difference which makes ∆x
sinusoidal. Fortunately, the effect is small. Ni = −1.609.
-log10 Ζ@ND 1 Ζ@ND
225
4
220
3
215
2
210
1 205
N N
-1.5 -1.0 -0.5 -1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2
(a) (b)
Figure 5.8: How the frequency of the oscillations changes with N for Erf Model 1
when k = 0.1h Mpc−1 .
x x
0.0020 0.0020
0.0015 0.0015
0.0010 0.0010
algebraic numeric
0.0005 0.0005
N N
-1.5 -1.0 -0.5 -1.5 -1.0 -0.5
-0.0005 -0.0005
-0.0010 -0.0010
(a) (b)
Figure 5.9: Model 1: The equivalent of Figures 5.3 and 5.4 with k still being
0.1h Mpc−1 but x0i changed from 10 to 1.
and shows up clearly in Figure 5.11(a). While the jump depends to some extent
on the common factor, ζ(N ), it also depends heavily on the multiple of ζ(N ) in
(5.69), as there is a change of sign involved. As is stated in subsection 5.3.2,
this multiple depends on Ψ0 (N ). Looking at the evolution of Φ(N ) and Ψ(N )
separately gives a jump at the right place in the difference between them, as
shown in Figure 5.10. The difference in Φ(N ) and Ψ(N ) is not maintained in
FHNL, YHNL
-1.5 -1.0 -0.5
N
-0.2
-0.4
F HNL -0.6
-0.8
-1.0
-1.2
Y HNL
-1.4
Figure 5.10: Model 1: The numeric solutions for Φ(N ) and Ψ(N ) when k =
0.1h Mpc−1 and x0i = 10. The scale of the graphs makes the oscillations barely vis-
ible.
x(N ) because, as can be seen from (5.53), while x(N ) ≈ Φ(N ) − Ψ(N ), initially,
it becomes closer to 2 (2Φ(N ) − Ψ(N )) /3 as ζ(N ) ∼ 1/λ(N )2 , at later times.
The oscillating component of x(N ) is shown in Figure 5.11(b) with x0i = 1. It

proportional to x0i and depends on the value of k. The non-oscillating compo-
nent is independent of x0i but does depend on k. Both components are model
dependent.
xnon-os xos
0.0020
0.0002
0.0015
0.0001
0.0010
0.0005 N
-1.5 -1.0 -0.5
N -0.0001
-1.5 -1.0 -0.5
-0.0005 -0.0002
(a) (b)
Figure 5.11: Model 1: For the algebraic solution, (a) shows the non-oscillating com-
ponent of x(N ). (b) shows the oscillating component of x(N ) when x0i = 1. Both are for
Erf Model 1 when k = 0.1h Mpc−1 . The two components combine to give Figure 5.9(a).
5.4.2 k = 0.01h Mpc−1
With a smaller value of k, we see larger discrepancies between the algebraic and
numeric solutions appear than when k was ten times larger. This is as expected
and as discussed in subsection 5.3.1. Algebraic and numeric solution curves for
x(N ) are shown in Figure 5.12 with detail in Figure 5.13. In these graphs we
see how the effect of a much smaller value of k manifests itself by showing the
algebraic solution at later times to be insufficient. The corresponding solutions
for y(N ) are given in Figure 5.14.
p
How the frequencies vary, again as represented by 1/ ζ(N ) and to compare
with Figure 5.8, is shown in Figure 5.15. Cursory measurement of the last four
wavelengths of Figure 5.12 confirms, as N increases above N = −1.2, the fre-
quency continues to fall, then rises a little and then falls, as shown in Figure 5.15,
but that the mean frequency for N > −1.4 is overestimated by around 4%.
x
numeric 0.05
0.04
0.03
algebraic 0.02
0.01
N
-1.5 -1.0 -0.5
-0.01
-0.02
Figure 5.12: Algebraic and numerical solutions for x(N ) for Erf Model 1, when
k = 0.01h Mpc−1 and x0i = 10 . Close agreement between the two solutions is now
p
prior to N = −1.4. 1/ ζ(N ) remains a fair estimate of the oscillation frequency at
later times.
x
0.003
0.002
0.001 numeric
N
-1.55 -1.50 -1.45 -1.40
-0.001
-0.002
-0.003
algebraic
-0.004
Figure 5.13: Detail of Figure 5.12 for N < −1.4, showing good agreement almost up
to then and where the two curves diverge.
y
N
-1.5 -1.0 -0.5
-1.85
-1.90
-1.95
-2.00
algebraic
numeric -2.05
Figure 5.14: Algebraic and numeric solutions for y(N )=Φ(N ) + Ψ(N ), for Erf Model
1, when k = 0.01h Mpc−1 . Compare Figure 5.5.
-log10 Ζ HNL 1 Ζ HNL
4 24
3 23
2 22
1 21
N N
-1.5 -1.0 -0.5 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2
(a) (b)
Figure 5.15: How the frequency of the oscillations changes with N for Erf Model
1 when k = 0.01h Mpc−1 . It can be seen how the local minimum and maximum of
Figure 5.8 are being eroded.
At later times, when N > −1.2, say, the frequency is around two orders of
magnitude lower than when k = 0.1h Mpc−1 and x0i = 10.
The graphs equivalent to those shown in Figure 5.11 are shown in Figure 5.16.
x0i = 1 so as to compare them.
xnon-os xos
0.025
0.0006
0.020
0.0004
0.015
0.0002
0.010
0.005 N
-1.5 -1.0 -0.5
N -0.0002
-1.5 -1.0 -0.5
-0.005 -0.0004
-0.010 -0.0006
(a) (b)
Figure 5.16: For the algebraic soution, (a) shows the non-oscillating component of
x(N ). (b) shows the oscillating component of x(N ) when x0i = 1. Both are for Erf
Model 1 when k = 0.01h Mpc−1 .
5.4.3 k=0
The algebraic and numeric solution curves when k = 0 for Erf Model 1 are
included, as Figures 5.17 and 5.18, just to see how the disagreement between the
two types of solution continues. The corresponding solutions for y(N ) are given
in Figure 5.19.
p
With k = 0, the frequency, as represented by 1/ ζ(N ) , continuously falls as
N increases. How it does so is shown in Figure 5.20.
The graphs equivalent to those shown in Figures 5.11 and 5.16 are shown in
Figure 5.21. Again, x0i = 1.
It is clear from all the graphs for x(N ) that, for values of N less than some value,
say −1.5, for the Erf Model 1, that the expression for x(N ) can be simplified for
x
0.30
numeric
0.25
0.20
algebraic 0.15
0.10
0.05
N
-1.5 -1.0 -0.5
Figure 5.17: Algebraic and numerical solutions for x(N ) for Erf Model 1, when k = 0
and x0i = 10. Again, For N > −1.4, say, the frequency of the oscillations of the algebraic
solution does seem to be slightly larger than for the numeric one.
x
0.004
0.002
N
-1.55 -1.50 -1.45 -1.40 -1.35 -1.30
-0.002 algebraic
-0.004 numeric
Figure 5.18: Detail of Figure 5.17 for N < −1.3.

y
-1.65
-1.70
algebraic -1.75
N
-1.5 -1.0 -0.5
-1.85
numeric
-1.90
-1.95
-2.00
Figure 5.19: Algebraic and numeric solutions for y(N )=Φ(N ) + Ψ(N ), for Erf Model
1, when k = 0.
-log10 Ζ HNL 1 Ζ HNL
12
4
10
3 8
2 6
4
1
2
N N
-1.5 -1.0 -0.5 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2
p
Figure 5.20: How the frequency of the oscillations, as represented by 1/ ζ(N ) ,
changes with N for Erf Model 1 when k = 0. The erosion of the local minimum and
maximum of Figure 5.8 is complete.
all values of k. At early times, the non-oscillating component of x is suppressed

by the factor ζ(N ) leaving x(N ) to be virtually sinusoidal, namely,
x(N ) ≈ −2Ψ1 (N ) sin ω(N ) (5.70)
where Ψ1 (N ) is defined in (5.65).
xnon-os
xos
0.0010
0.15
0.0005
0.10
N
-1.5 -1.0 -0.5
0.05
-0.0005
N
-1.5 -1.0 -0.5 -0.0010
(a) (b)
Model 1 when k = 0.
5.4.4 Algebraic and numeric solutions for x(N ) compared

for other models of chapter 3.
Before we leave this topic we should see how the results compare for the other
models studied in this thesis. Thus we shall consider the comparison between the
algebraic and numeric solutions for Models 2, 4 and 5 of Chapter 3. We shall
assume x0i = 1 and k = 0.1h Mpc−1 for all of them.
5.4.4.1 Model 2
This is from the Erf model with parameters Λ∞ = 2H02 , c = 1.5 and log10 b =
−0.914. It is located in the de Sitter region of the contour plot, Figure 3.4.
x x
0.0008 0.0008
0.0006 0.0006
0.0004 0.0004
0.0002 0.0002
N N
-1.0 -0.8 -0.6 -0.4 -0.2 -1.0 -0.8 -0.6 -0.4 -0.2
-0.0002 -0.0002
-0.0004 -0.0004
(a) (b)
Figure 5.22: Algebraic (a) and Numeric solutions (b) for x(N ) for Model 2 when
k = 0.1h Mpc−1 and x0s = 1. Slight differences in the oscillation amplitude are apparent.
Considering how closely f (R) (Figure 3.1) cleaves to −2Λ∞ until very recently,
it is worth noting that the jump is still present. Some slight differences in oscil-
lation amplitude, especially at around N = −0.8, are becoming visible.
5.4.4.2 Model 4
This is from the AB model with parameters Λ∞ = 1.92H02 , b = 1.2. It is located

in the de Sitter region of the contour plot, Figure 3.10.
x x
0.0006 0.0006
0.0004 0.0004
0.0002 0.0002
N N
-1.2 -1.0 -0.8 -0.6 -0.4 -0.2 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2
-0.0002 -0.0002
(a) (b)
k = 0.1h Mpc−1 and x0s = 1. There is no noticeable difference between the two. The
corresponding graphs for y are also indistinguishable on these plotting scales.
5.4.4.3 Model 5
This is from the HSS model with parameters n = 6, Λ∞ = 1.2 H02 , log10 c = −2.34.
It is located in the Minkowski region of the contour plot, Figure 3.17.
x x
0.0010 0.0010
0.0008
0.0006
0.0005
0.0004
0.0002
N N
-1.2 -1.0 -0.8 -0.6 -0.4 -0.2 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2
-0.0002
-0.0004
-0.0005
(a) (b)
k = 0.1h Mpc−1 and x0s = 1. There is reasonable agreement.
5.4.5 Comment
As is obvious from the graphs, above, there are differences between the algebraic
and numeric solutions but these are small, to first order in ζ(N ). The differences
become more marked the smaller k is. Recall that ζ(N ) = 1/ [1/ξ(N ) + λ(N )2 ].
If ζ(N ) is not sufficiently small then more terms would be needed in the algebraic
expression for x(N ) and y(N ) than are actually used. At early times, ξ(N ) is
very small but rises with N , as we have seen. That leaves λ(N ) to be “large”
when this happens. λ(N ) = k/ (aH(N )) = k e−N /H(N ) so the larger the value
of k the more easily this requirement is fulfilled.
p
It is noticeable from the graphs how reasonably accurate is 1/ ζ(N ) as an
estimator of the true frequency of the oscillations, even at later times. It does
seem to be less accurate, however, the smaller k is. It is also noticeable how
the frequency at later times, say at N = −1, decreases as k decreases. At early
times, the frequency depends mainly on the value of ξ(N ) which is a function
of the model only. In the case of k = 0, since ζ(N ) = ξ(N ), and since ξ(N )
increases with N , then the frequency of oscillations decreases with N , given that
p
1/ ζ(N ) remains a valid estimator of the oscillation frequency.
The jump in x(N ), evident when k 6= 0 for N > −1.5, is caused by the non-
oscillating component of x(N ) and the effect seems to increase in amplitude and
also spread out as k becomes smaller. Figure 5.25 shows the equivalent graphs to
those of Figures 5.11, 5.16 and 5.21 but with k = 0.005h Mpc−1 . One could also
claim, for a particular model, that the jump is moving to smaller values of N .
xnon-os xos
0.04
0.03 0.0005
0.02
N
0.01 -1.5 -1.0 -0.5
N
-1.5 -1.0 -0.5 -0.0005
-0.01
(a) (b)
Model 1 when k = 0.005h Mpc−1 .
5.5 Conclusion
This was an attempt to gain some insight into the size and nature of the os-
cillations in the perturbed FRW metric. In all fairness one could say there has
been partial success with very good accuracy when k is large and of the order
of 0.1h Mpc−1 , or higher. Over a wide range, that is, throughout the matter era
from equality right up to the present time in the vacuum era, the frequency of
p
the oscillations is well represented by 1/ ζ(N ) . This contrasts with Starobinsky
in [78] and also Appleby in [101] who restricted themselves to the early matter
5.5. Conclusion 137
era, where R ∝ a−3 , and considered the oscillations inherent in δR; see (5.11)
which were termed δRosc . It was found that the frequency of these oscillations
p p
was, in effect1 , the same as 1/ ξ(N ) which is what 1/ ζ(N ) reduces to at
early times, as has been discussed. However, the amplitude of the oscillations de-
creases with time but becomes unbounded as R → ∞. However, before that can
happen, as R grows into the past, the conditions for the validity of the defining
differential equation (equation (13) in [78]) are broken. It is clear, from that dif-
ferential equation, that FRR (R) → 0 implies that δRosc → 0. At high curvatures,
FRR (R) → 0 as R increases further, when F (R) gravity becomes more and more
indistinguishable from ΛCDM.
At early times, in the matter era, the amplitude of the oscillations, as rep-
p
resented in (5.45) and (5.46), is proportional to e−3N/2 / H(N ) ξ(N )1/4 , which
itself is proportional to R1/4 [R FRR (R)]1/4 and decreases into the past. In other
words, as N increases from Ni , the amplitude of the oscillations, as specified by
|Ψ1 (N )| in (5.65), increases and continues to do so until λ(N )2 ∼ 1/ξ(N ) when
what happens next depends on the value of k. If k = 0, the amplitude continues to
increase, as is shown in Figure 5.17. For larger k, the amplitude starts to decrease,
as is illustrated in Figure 5.12 when k = 0.01h Mpc−1 and in Figure 5.4 when
k = 0.1h Mpc−1 . This a complicated effect and deserves more study. It depends
p
on the tension between e−3N/2 , which decreases, between e−ν(N ) / H(N ) , which
increases, and between ζ(N ), which rises, falls and rises again, as N increases.
A feature which has not been seen before is the “jump” in x(N ) which origi-
nates from the non-oscillatory solution. As k decreases, the cross-over value of
N 2 increases and the numerical values of the local maximum and minimum on
either side of it also increase. The jump is, however, the result of the way x(N )
is constructed and due to the sudden divergence between Φ(N ) and Ψ(N ), as
1
They considered functions of time.
2
The value of N where x(N ) = 0 between the negative local minimum and positive local
maximum of the non-oscillating component of x(N ).
explained in subsection 5.4.1. Had merely the difference, Φ(N ) − Ψ(N ), been
plotted rather than the expression for x(N ) of (5.53) then this jump, now a step,
would have been missed and the oscillations rendered invisible as Figure 5.26(a)
shows.
FHNL-YHNL
FHNL-YHNL
0.6 0.00007
0.5
0.00006
0.4
0.00005
0.3
0.00004
0.2
0.1 0.00003
-1.0 -0.8 -0.6 -0.4 -0.2

N -1.068 -1.066 -1.064 -1.062
N
-1.060
(a) (b)
Figure 5.26: Erf Model 2: Both graphs show both the numeric and algebraic solutions,
Φ(N ) − Ψ(N ), superimposed on one another, for different ranges of N . Differences
between the algebraic and numeric graphs are indiscernible. In (a), which shows the
full range of N , the oscillations are invisible but the restricted range of (b) shows them
and also how the mean value of Φ(N ) − Ψ(N ) is increasing with N . The range of (b)
is of necessity restricted; a larger range renders the oscillations to be much less visible
because the algebraic and numeric graphs become steeper and, reducing the values of N
shown, means that the frequency becomes too large to be manageable. k = 0.1h Mpc−1
and x0s = 1.
How large the oscillations could be, today, with any model, is debatable. If
the initial conditions are applied at around z = 1000, for which Ni ≈ −7, and
ζ(N ) and the amplitude of the oscillations can be detected if they are of order
α, today, then we must have x0i ζ(Ni )1/4 ∼ α104.5 /ζ(N0 )1/4 ∼ 106 α, say, or larger,
which may or may not be realistic. If not then the oscillations are invisible. In
other words, the smaller ζ(Ni ) is, the larger must be x0i . At the very outside for
all models, because of the Solar System constraint, ζ(Ni ) 10−23 , see Chapter 3,
section 3.2.1 (3.5), so it is quite likely that ζ(Ni ) will be considerably smaller.
5.5. Conclusion 139
Thus a first attempt at quantifying the matter perturbations has been made
for all times in the post-radiation era. A brief comparison between this and work
done on perturbing the Ricci scalar by Starobinsky [78] and by Appleby, Battye
and Starobinsky [101], now follows.
5.5.1 Oscillations in δR
The oscillations in Φ and Ψ are the same as those found by authors who studied
perturbations of the Ricci scalar at high curvature. The result, as quoted in [101]
is
"Z #
−3/2 −3/4 1
δR = Ca [FRR (RGR )] sin p dt , (5.71)
3FRR (RGR )
where C is a constant. This translates, using RGR = R = 3H 2 , R ∝ a−3 ,

ζN ≈ ξ(N ) ≈ 3H 2 FRR (R), so that R now represents the non-oscillating part of
R, for simplicity, and dN/dt = H, into
"Z #
√ −3/4 1
δR = D R [FRR (R)] sin p dN , (5.72)
ζ(N )
where D is a constant. Using just the significant oscillatory components at high
R for Φ(N ) and Ψ(N ), namely,
"Z #
e−3N/2 ζ(N )1/4 1
Ψ(N ) = A p sin p dN (5.73)
H(N ) ζ(N )
= −Φ(N ), (5.74)
where A is a constant, and substituting these expressions for Φ(N ) and Ψ(N )
into the expression for δR in equation (5.11), gives as the most significant term
"Z #
√ 1
δR = −2A R [FRR (R)]−3/4 sin p dN . (5.75)
ζ(N )
on the assumption that (k/a)2 Ψ(N ) is not significant and that 0 < FRR (R)
1/ (n2 R) given that, at high R, fR (R) can be described by (4.16). There is
agreement.
Chapter 6
Conclusions
6.1 Summary
This thesis is concerned with attempting to solve, often algebraically, some of

the equations which result after modifying gravity by replacing the Ricci scalar,
R, by F (R) in the action represented by equation (1.47). There has been a
good deal of success especially when one considers how complicated some of the
equations are. The areas looked at were: finding all F (R) for which the Universe
has a standard Einstein Gravity expansion history; introducing a new model and
exploring, for given Ωeff,0 , how extreme the present value of the effective equation
of state parameter, weff,0 , for this model, the AB model and the HSS model could
become; proving that the effective equation of state parameter for F (R) models
that tend to standard Einstein Gravity at high curvature must lie below the
phantom divide and continue to decrease throughout the radiation era and some
of the matter era; finding expressions for the oscillations of potentials Φ and Ψ
in the Newtonian gauge induced by perturbing the FRW metric.
141
142 Chapter 6. Conclusions
6.1.1 Models with standard Einstein Gravity expansion

history
The expansion of the Universe is measured by the scale factor a for which ȧ > 0
as it expands. The Hubble parameter, H = ȧ/a, the fractional rate of expansion,
is used to define the expansion history of the Universe. Standard Einstein gravity
has an expansion history given by equation (2.2), in which Λ is the cosmological
constant. The corresponding curvature at any value of a is given by equation
(2.3). The complete solutions for F (R) in the matter and vacuum eras are given,
as infinite power series, by equation (2.13), when R > 4Λ, and also at late time,
when 4Λ ≤ R ≤ 5Λ by (2.23); today is at R0 ≈ 4.4Λ. Application of the various
constraints suffered by F (R) models, of which the Solar System constraint is the
severest, restricts the coefficients of these solutions to such an extent that they
are barely discernible from the standard Einstein Gravity model, F (R) = R −2Λ.
In section 2.3, a series solution was found for F (R) in the radiation era but,
because of the very large value of R/Λ, we need only concern ourselves with the
solution F (R) = R + AΛρ0 (R/Λ)4/3 . Constraints limit Aρ0 such that
0 ≤ Aρ0 10−25 .
Concern has been shown in, for example, [78, 101] about the singularity in the
scalaron mass which occurs in cases where F (R) tends to R − 2Λ as R → ∞; the
scalaron mass tends to infinity and can be very large at relatively low values of
R. An attempt has been designed to alleviate this problem by adding the term,
R2 / (6M 2 ), to F (R), where M is some large mass scale. In [101], for example,
it is stated that if the quadratic term is to drive inflation then M ∼ 1012 GeV'
1054 H0 . This term is only significant at very high curvatures but limits the
scalaron mass to being of the order of M when the term is significant. While not
alleviating the problem, the solution F (R) = R + AΛρ0 (R/Λ)4/3 gives a scalaron
mass of order of R at high curvature.
6.1. Summary 143
6.1.2 Extreme values generated by F (R) models
In Chapter 3, a new model, based on the error function and which we have
termed the Erf model, was used. The idea behind it was that we wished to
generate values of weff,0 which were further from the phantom divide than other
models in use. We needed a model which stuck closely to standard Einstein
Gravity as z decreased until comparatively recently, when z is of the order of a
few and when weff would still be very close to −1, and then swing away sharply
to give relatively larger values of |1 + weff,0 | than had been seen before. The close
adherence to the phantom divide was necessitated by the stringent requirement
of the Solar System constraint. The Erf model was compared to the AB and HSS
models. In the past, restrictions on the parameters of a model caused by the
possibility of there being a late time de Sitter attractor solution, RdS , for which
2
RdS = 12HdS , have produced relatively small values of |1 + weff,0 |. This was
discussed for each model and showed how allowing late-time Minkowski solutions
could greatly extend the range of |1 + weff,0 |. We argued that, should there be
a problem beyond today, we could amend the particular F (R), by re-defining it
for the future, so that any problems caused by having a Minkowski solution are
evaded. For example, the HSS model has a problem with FRR (R) becoming zero
and going negative at some time in the future but it is generally saved by having
de Sitter attractor solutions.
One constraint that was applied to all models was that we insisted that to-
day’s value of the effective dark energy density parameter, Ωeff,0 , was the same
token value of 0.7. Because weff,0 and R0 are connected by the relation, R0 =
3H02 (1 − 3weff,0 Ωeff,0 ), extreme values of weff,0 generate extreme and, therefore,
unrealistic values of R0 , that is, outside the range for R0 given in section 3.2.
However, we never allowed R0 to go negative.
This aspect of F (R) gravity, of finding extreme values of measurable parameters

given certain constraints, is most interesting and points a way forward. See
section 6.2.
6.1.3 Theorem
It is noticeable that, for all examples of F (R) models which have the property of
tending to standard Einstein Gravity as R → ∞, graphs show weff falling below
the phantom divide, as z decreases, and they continue to do so until they return
to cross the phantom divide in the recent past, now or, maybe, in the near future.
This has been discussed in relation to specific models in [47, 49, 68, 97]. It is one
property of F (R) gravity which may distinguish it from other types of model and
certainly from quintessence.
By assuming that f (R), where F (R) = R + f (R), can be approximated by

−2Λ + µ0 an where µ0 is a positive constant and n is a very large number, we
prove the first part of this, namely that as a increases from 0, weff decreases
from −1 and continues to do so until the expression for f (R) ceases to be valid.
Because n is very large, which it has to be because of the Solar system constraint,
this will happen at a value of R within the matter or vacuum eras. To show that
weff then reaches a local minimum would require more knowledge of how f (R)
evolves with time.
6.1.4 Oscillations of the potentials of the perturbed FRW

metric
In Chapter 5, the perturbed FRW metric can be written, to first order in the
Newtonian gauge, as:
ds2 = − (1 + 2Ψ) dt2 + a2 (1 − 2Φ) δij dxi dxj . (6.1)

6.1. Summary 145
A total of seven relevant equations were distilled into two to produce a pair of
coupled, second-order, homogeneous, linear, differential equations for the poten-
tials Φ and Ψ in terms of the variable N = log a. These equations were solved
both numerically and algebraically with the same initial conditions. Algebraically,
both of the potentials were shown to consist of an oscillating component and a
non-oscillating component.
It was clear that both oscillating components had the same frequency and an
expression was found which appeared to be robust right from early times to the
present. This is in contrast to previous work ([78, 101, 117]), in which were
considered the oscillations of the perturbed Ricci scalar but only at early times.
When the units were changed so that the frequencies could be compared it was
found that our solutions agreed with them. The expression for the frequency
of our solutions was proportional to the reciprocal of the scalaron mass, at early
times, migrating to k/ (a H), at late times, around today, where k is the co-moving
wavenumber. How the oscillation frequency changed over time for a particular
model, Erf Model 1, was illustrated and how it depended on the value of k.
Approximations to the envelope of each waveform were determined, as were the
non-oscillating components up to order ζ(N ), which is defined in equation (5.25).
Then initial conditions were applied and the phase of the oscillations chosen so
that the algebraic approximations could be given for Φ(N ) and Ψ(N ) in terms of
just one parameter, x0i .
In order to make the oscillations visible when plotting graphs, the variable x(N )
was formed which removes the largest of the non-oscillating terms of Φ(N ) and
Ψ(N ) and which approximates to Φ(N )−Ψ(N ) at high z but to (4Φ(N ) − 2Ψ(N )) /3
at recent times. In order to display the non-oscillating components the sum
y(N ) = Φ(N ) + Ψ(N ) was formed which was dominated by the lowest order
non-oscillating components. To illustrate how accurate the algebraic solutions
were, the algebraic and numeric solutions for x(N ) and y(N ) were compared
graphically.
This exercise was completed for four example models. There was always good
agreement at early times. At later times, near today, there was good agreement
when k was of the order of k = 0.1h Mpc−1 but the algebraic solutions were less
accurate for smaller k with the greatest errors being when k = 0. The reason
for this is that at late times ζ(N ) is proportional to 1/k 2 while, at early times,
it is independent of k. The efficacy of the algebraic solution requires that ζ(N )
is always small, which puts a model-dependent condition on k. This explains
why, for a given value of k, different models gave slightly different disagreements
between the algebraic and corresponding numeric solutions.
One feature of the graphs of x(N ) against N which was unexpected was the
appearance of a “jump” in x(N ) at a value of N where the expression for ζ(N )
changes from being dominated from one expression, ξ(N ), to being dominated
by another, 1/λ(N )2 .
The oscillations in Φ(N ) and Ψ(N ) have been shown to be the same, at high
curvature, as those studied as perturbations in the Riccci scalar in [78, 101].
There remains the problem that the amplitude of the oscillations for Φ(N ) and
Ψ(N ) increase, initially, into the future (see Figure 5.11), whereas for δR the
amplitude of the oscillations increases into the past. The oscillations for Φ(N )
and Ψ(N ) really start before the formation of the CMB so that they start at some
value of z 103 . The same should be true for δR; that they only recede into the
past as far as this point which would give the oscillations a maximum amplitude
which is finite. In Einstein Gravity δR = 0, so there could be no oscillations in
δR until this point when they suddenly “appear” with maximum amplitude to
decay over time. This gives credence to the suggestion, in [26], that all F (R)
functions should be modified by the addition of the R2 / (6M 2 ) term mentioned
in subsection 6.1.1, above, which would at least limit the maximum amplitude
and frequency at high values of z.
6.2. Outlook 147
6.2 Outlook
1. The time has come for researchers to test F (R) models against parameters
determined from observations. Current results from Planck, assuming a
0
ΛCDM background and a flat FRW Universe, in effect constrain weff,0 , weff,0 ,
H0 , Ωeff,0 , Ωm,0 and Ωr,0 , the last being coupled to the previous via the value
of z at equality. It would be interesting to see how the tight variations in
these cosmological parameters constrain the parameters of all the successful
models proposed and what these models would then predict. What are their
similarities and how do they differ?
2. Is it possible to prove, for all viable F (R) models that weff reaches a local
minimum below the phantom divide sometime in the recent past?
3. An attempt could be made to find more complete algebraic solutions which

would represent better what happens today. Are oscillations in potentials,
Φ and Ψ, detectable and, if so, of what order might they be?
Bibliography
[1] J. C. Mather, Cosmic background explorer (COBE) mission, in Infrared

Spaceborne Remote Sensing (M. S. Scholl, ed.), vol. 2019 of Society of
Photo-Optical Instrumentation Engineers (SPIE) Conference Series,
pp. 146–157, Oct., 1993.
[2] N. W. Boggess, J. C. Mather, R. Weiss, C. L. Bennett, E. S. Cheng,

E. Dwek, S. Gulkis, M. G. Hauser, M. A. Janssen, T. Kelsall, S. S. Meyer,
S. H. Moseley, T. L. Murdock, R. A. Shafer, R. F. Silverberg, G. F.
Smoot, D. T. Wilkinson, and E. L. Wright, The COBE mission - Its
design and performance two years after launch, 397 (Oct., 1992) 420–429.
[3] Planck Collaboration Collaboration, P. Ade et. al., Planck 2013

results. I. Overview of products and scientific results, arXiv:1303.5062.
[4] WMAP Collaboration Collaboration, C. Bennett et. al., First year

Wilkinson Microwave Anisotropy Probe (WMAP) observations:
Preliminary maps and basic results, Astrophys.J.Suppl. 148 (2003) 1,
[astro-ph/0302207].
[5] WMAP Collaboration Collaboration, D. Spergel et. al., First year

Wilkinson Microwave Anisotropy Probe (WMAP) observations:
Determination of cosmological parameters, Astrophys.J.Suppl. 148 (2003)
175–194, [astro-ph/0302209].
[6] WMAP Collaboration Collaboration, G. Hinshaw et. al., Nine-Year
149
150 Bibliography
Wilkinson Microwave Anisotropy Probe (WMAP) Observations:

Cosmological Parameter Results, arXiv:1212.5226.
[7] WMAP Collaboration Collaboration, C. Bennett et. al., Nine-Year

Wilkinson Microwave Anisotropy Probe (WMAP) Observations: Final
Maps and Results, arXiv:1212.5225.
[8] A. Einstein, The Foundation of the General Theory of Relativity, Annalen

Phys. 49 (1916) 769–822.
[9] H. Lorentz, The Principle of Relativity. Dover Publications, Inc, 1923.
[10] L. Infeld, Equations of Motion in General Relativity Theory and the

Action Principle, Rev.Mod.Phys. 29 (1957) 398–411.
[11] S. M. Carroll, Spacetime and geometry: An introduction to general

relativity. Adison Wesley, 2004.
[12] A. H. Guth, The Inflationary Universe: A Possible Solution to the

Horizon and Flatness Problems, Phys.Rev. D23 (1981) 347–356.
[13] A. D. Linde, Particle physics and inflationary cosmology,

Contemp.Concepts Phys. 5 (1990) 1–362, [hep-th/0503203].
[14] D. Coule, Canonical measure and the flatness of a FRW universe,

Class.Quant.Grav. 12 (1995) 455–470, [gr-qc/9408026].
[15] G. Lemaı̂tre, Expansion of the universe, A homogeneous universe of

constant mass and increasing radius accounting for the radial velocity of
extra-galactic nebulae, 1931MNRAS..91..483L (1931).
[16] G. Tammann, The Ups and Downs of the Hubble Constant,

Rev.Mod.Astron. 19 (2006) 1, [astro-ph/0512584].
[17] Planck Collaboration Collaboration, P. Ade et. al., Planck 2013

results. XVI. Cosmological parameters, arXiv:1303.5076.
Bibliography 151
[18] R. Feynman, F. Morinigo, W. Wagner, and B. Hatfield, Feynman lectures

on gravitation, .
[19] E. W. Kolb and M. S. Turner, The Early Universe, Front.Phys. 69 (1990)

1–547.
[20] A. A. Starobinsky, Future and origin of our universe: Modern view,

Grav.Cosmol. 6 (2000) 157–163, [astro-ph/9912054].
[21] A. A. Penzias and R. W. Wilson, A Measurement of excess antenna

temperature at 4080-Mc/s, Astrophys.J. 142 (1965) 419–421.
[22] R. Dicke, P. Peebles, P. Roll, and D. Wilkinson, Cosmic Black-Body

Radiation, Astrophys.J. 142 (1965) 414–419.
[23] C. L. Bennett, N. W. Boggess, E. S. Cheng, M. G. Hauser, T. Kelsall,

J. C. Mather, S. H. Moseley, Jr., T. L. Murdock, R. A. Shafer, and R. F.
Silverberg, Scientific results from COBE, Advances in Space Research 13
(Dec., 1993) 409–.
[24] R. Sachs and A. Wolfe, Perturbations of a cosmological model and angular

variations of the microwave background, Astrophys.J. 147 (1967) 73–90.
[25] T. Giannantonio, R. Scranton, R. G. Crittenden, R. C. Nichol, S. P.

Boughn, et. al., Combined analysis of the integrated Sachs-Wolfe effect
and cosmological implications, Phys.Rev. D77 (2008) 123520,
[arXiv:0801.4380].
[26] L. Amendola, M. Kunz, and D. Sapone, Measuring the dark side (with
weak lensing), JCAP 0804 (2008) 013, [arXiv:0704.2421].
[27] Supernova Cosmology Project Collaboration, S. Perlmutter et. al.,

Discovery of a supernova explosion at half the age of the Universe and its
cosmological implications, Nature 391 (1998) 51–54, [astro-ph/9712212].
152 Bibliography
[28] M. S. Turner, Why cosmologists believe the universe is accelerating,

astro-ph/9904049.
[29] W. Chakraborty, Accelerating Expansion of the Universe,

arXiv:1105.1087.
[30] A. De Felice, D. F. Mota, and S. Tsujikawa, Matter instabilities in general

Gauss-Bonnet gravity, Phys.Rev. D81 (2010) 023532, [arXiv:0911.1811].
[31] S. Nojiri and S. D. Odintsov, Unified cosmic history in modified gravity:

from F(R) theory to Lorentz non-invariant models, Phys.Rept. 505 (2011)
59–144, [arXiv:1011.0544].
[32] T. P. Sotiriou and V. Faraoni, f(R) Theories Of Gravity, Rev.Mod.Phys.

82 (2010) 451–497, [arXiv:0805.1726].
[33] A. De Felice and S. Tsujikawa, f(R) theories, Living Rev.Rel. 13 (2010) 3,

[arXiv:1002.4928].
[34] T. Clifton, P. G. Ferreira, A. Padilla, and C. Skordis, Modified Gravity

and Cosmology, Phys.Rept. 513 (2012) 1–189, [arXiv:1106.2476].
[35] A. Guarnizo, L. Castaneda, and J. M. Tejeiro, Boundary Term in Metric

f(R) Gravity: Field Equations in the Metric Formalism, Gen.Rel.Grav. 42
(2010) 2713–2728, [arXiv:1002.0617].
[36] E. Poisson, A relativist’s toolkit : the mathematics of black-hole

mechanics. 2004.
[37] E. J. Copeland, M. Sami, and S. Tsujikawa, Dynamics of dark energy,

Int.J.Mod.Phys. D15 (2006) 1753–1936, [hep-th/0603057].
[38] R. Adler, B. Casey, and O. Jacob, Vacuum catastrophe: An Elementary

exposition of the cosmological constant problem, Am.J.Phys. 63 (1995)
620–626.
Bibliography 153
[39] J. Martin, Everything You Always Wanted To Know About The

Cosmological Constant Problem (But Were Afraid To Ask), Comptes
Rendus Physique 13 (2012) 566–665, [arXiv:1205.3365].
[40] A. Goobar, S. Perlmutter, G. Aldering, G. Goldhaber, R. Knop, et. al.,

The acceleration of the universe: Measurements of cosmological
parameters from type Ia supernovae, Phys.Scripta T85 (2000) 47–58.
[41] S. Weinberg, Anthropic Bound on the Cosmological Constant,

Phys.Rev.Lett. 59 (1987) 2607.
[42] S. Weinberg, The Cosmological Constant Problem, Rev.Mod.Phys. 61

(1989) 1–23.
[43] T. Padmanabhan and H. Padmanabhan, CosMIn: The Solution to the

Cosmological Constant Problem, arXiv:1302.3226.
[44] A. Einstein, Do gravitational fields play an essential part in the structure

of the elementary particles of matter?,
Sitzungsber.Preuss.Akad.Wiss.Berlin (Math.Phys.) 1919 (1919) 433.
[45] A. Dolgov, M. Einhorn, and V. I. Zakharov, The Vacuum of de Sitter

space, Acta Phys.Polon. B26 (1995) 65–90, [gr-qc/9405026].
[46] G. Cognola, E. Elizalde, S. Odintsov, P. Tretyakov, and S. Zerbini, Initial

and final de Sitter universes from modified f(R) gravity, Phys.Rev. D79
(2009) 044001, [arXiv:0810.4989].
[47] H. Motohashi, A. A. Starobinsky, and J. Yokoyama, Future Oscillations

around Phantom Divide in f(R) Gravity, JCAP 1106 (2011) 006,
[arXiv:1101.0744].
[48] W. Hu, Crossing the phantom divide: Dark energy internal degrees of
freedom, Phys.Rev. D71 (2005) 047301, [astro-ph/0410680].
154 Bibliography
[49] L. Amendola and S. Tsujikawa, Phantom crossing, equation-of-state

singularities, and local gravity constraints in f(R) models, Phys.Lett. B660
(2008) 125–132, [arXiv:0705.0396].
[50] K. Bamba, C.-Q. Geng, S. Nojiri, and S. D. Odintsov, Crossing of

Phantom Divide in F (R) Gravity, Mod.Phys.Lett. A25 (2010) 900–908,
[arXiv:1003.0769].
[51] J. Frieman, M. Turner, and D. Huterer, Dark Energy and the Accelerating
Universe, Ann.Rev.Astron.Astrophys. 46 (2008) 385–432,
[arXiv:0803.0982].
[52] L. Amendola and S. Tsujikawa, Dark Energy: Theory and Observations.

2010.
[53] L. Amendola, D. Polarski, and S. Tsujikawa, Are f(R) dark energy models
cosmologically viable ?, Phys.Rev.Lett. 98 (2007) 131302,
[astro-ph/0603703].
[54] Supernova Search Team Collaboration, A. G. Riess et. al.,

Observational evidence from supernovae for an accelerating universe and a
cosmological constant, Astron.J. 116 (1998) 1009–1038,
[astro-ph/9805201].
[55] Supernova Cosmology Project Collaboration, S. Perlmutter et. al.,

Measurements of Omega and Lambda from 42 high redshift supernovae,
Astrophys.J. 517 (1999) 565–586, [astro-ph/9812133].
[56] G. Huey, L.-M. Wang, R. Dave, R. Caldwell, and P. J. Steinhardt,

Resolving the cosmological missing energy problem, Phys.Rev. D59 (1999)
063005, [astro-ph/9804285].
[57] D. Polarski and A. Ranquet, On the equation of state of dark energy,

Phys.Lett. B627 (2005) 1–8, [astro-ph/0507290].
Bibliography 155
[58] V. Sahni and A. A. Starobinsky, The Case for a positive cosmological

Lambda term, Int.J.Mod.Phys. D9 (2000) 373–444, [astro-ph/9904398].
[59] S. M. Carroll, W. H. Press, and E. L. Turner, The Cosmological constant,

Ann.Rev.Astron.Astrophys. 30 (1992) 499–542.
[60] P. Peebles and B. Ratra, The Cosmological constant and dark energy,
Rev.Mod.Phys. 75 (2003) 559–606, [astro-ph/0207347].
[61] V. Sahni and A. Starobinsky, Reconstructing Dark Energy,

Int.J.Mod.Phys. D15 (2006) 2105–2132, [astro-ph/0610026].
[62] T. Padmanabhan, Cosmological constant: The Weight of the vacuum,

Phys.Rept. 380 (2003) 235–320, [hep-th/0212290].
[63] S. Tsujikawa, Modified gravity models of dark energy, Lect.Notes Phys.

800 (2010) 99–145, [arXiv:1101.0191].
[64] E. Di Pietro and J. Demaret, A constant equation of state for

quintessence?, Int.J.Mod.Phys. D10 (2001) 231–237, [gr-qc/9908071].
[65] R. Caldwell, An introduction to quintessence, Braz.J.Phys. 30 (2000)

215–229.
[66] I. Zlatev, L.-M. Wang, and P. J. Steinhardt, Quintessence, cosmic

coincidence, and the cosmological constant, Phys.Rev.Lett. 82 (1999)
896–899, [astro-ph/9807002].
[67] D. Polarski, Dark energy, AIP Conf.Proc. 1514 (2012) 111–117.
[68] K. Bamba, C.-Q. Geng, and C.-C. Lee, Phantom crossing in viable f (R)
theories, Int.J.Mod.Phys. D20 (2011) 1339–1345, [arXiv:1108.2557].
[69] M. Kunz and D. Sapone, Crossing the Phantom Divide, Phys.Rev. D74
(2006) 123503, [astro-ph/0609040].
156 Bibliography
[70] A. A. Starobinsky, A New Type of Isotropic Cosmological Models Without

Singularity, Phys.Lett. B91 (1980) 99–102.
[71] P.J.E.Peebles, Large Scale Structure of the Universe. Princeton University

Press, 1980.
[72] L.-M. Wang and P. J. Steinhardt, Cluster abundance constraints on

quintessence models, Astrophys.J. 508 (1998) 483–490,
[astro-ph/9804015].
[73] D. Rapetti, S. W. Allen, A. Mantz, and H. Ebeling, The Observed Growth

of Massive Galaxy Clusters III: Testing General Relativity on
Cosmological Scales, Mon.Not.Roy.Astron.Soc. 406 (2010) 1796–1804,
[arXiv:0911.1787].
[74] E. V. Linder, Cosmic growth history and expansion history, Phys.Rev.

D72 (2005) 043529, [astro-ph/0507263].
[75] A. J. Lopez-Revelles, Growth of matter perturbations for realistic F (R)

models, Phys.Rev. D87 (2013) 024021, [arXiv:1301.2120].
[76] S. A. Appleby and R. A. Battye, Do consistent F (R) models mimic

General Relativity plus Λ?, Phys.Lett. B654 (2007) 7–12,
[arXiv:0705.3199].
[77] L. Amendola, R. Gannouji, D. Polarski, and S. Tsujikawa, Conditions for

the cosmological viability of f(R) dark energy models, Phys.Rev. D75
(2007) 083504, [gr-qc/0612180].
[78] A. A. Starobinsky, Disappearing cosmological constant in f(R) gravity,

JETP Lett. 86 (2007) 157–163, [arXiv:0706.2041].
[79] L. Pogosian and A. Silvestri, The pattern of growth in viable f(R)

cosmologies, Phys.Rev. D77 (2008) 023503, [arXiv:0709.0296].
Bibliography 157
[80] A. Nunez and S. Solganik, The Content of f(R) gravity, hep-th/0403159.
[81] A. Krause and S.-P. Ng, Ghost cosmology: exact solutions, transitions
between standard cosmologies and ghost dark energy/matter evolution,
Int.J.Mod.Phys. A21 (2006) 1091–1122, [hep-th/0409241].
[82] B. Himmetoglu, C. R. Contaldi, and M. Peloso, Ghost instabilities of

cosmological models with vector fields nonminimally coupled to the
curvature, Phys.Rev. D80 (2009) 123530, [arXiv:0909.3524].
[83] N. Deruelle, M. Sasaki, Y. Sendouda, and A. Youssef, Inflation with a

Weyl term, or ghosts at work, JCAP 1103 (2011) 040,
[arXiv:1012.5202].
[84] A. Dolgov and M. Kawasaki, Can modified gravity explain accelerated

cosmic expansion?, Phys.Lett. B573 (2003) 1–4, [astro-ph/0307285].
[85] G. J. Olmo, Post-Newtonian constraints on f(R) cosmologies in metric

and Palatini formalism, Phys.Rev. D72 (2005) 083505, [gr-qc/0505135].
[86] V. Faraoni, Solar System experiments do not yet veto modified gravity
models, Phys.Rev. D74 (2006) 023529, [gr-qc/0607016].
[87] S. Tsujikawa, Observational signatures of f(R) dark energy models that

satisfy cosmological and local gravity constraints, Phys.Rev. D77 (2008)
023507, [arXiv:0709.1391].
[88] M. Colless, First results from the 2dF galaxy redshift survey,
astro-ph/9804079.
[89] J. Loveday, The Sloan Digital Sky Survey, Contemporary Physics 43

(June, 2002) 437–449, [astro-ph/0207189].
[90] SDSS Collaboration Collaboration, D. J. Eisenstein et. al., SDSS-III:

Massive Spectroscopic Surveys of the Distant Universe, the Milky Way
158 Bibliography
Galaxy, and Extra-Solar Planetary Systems, Astron.J. 142 (2011) 72,

[arXiv:1101.1529].
[91] K. S. Dawson et. al., The Baryon Oscillation Spectroscopic Survey of

SDSS-III, Astron.J. 145 (2013), no. 1 10, [arXiv:1208.0022].
[92] S. Fay, S. Nesseris, and L. Perivolaropoulos, Can f(R) Modified Gravity

Theories Mimic a LCDM Cosmology?, Phys.Rev. D76 (2007) 063504,
[gr-qc/0703006].
[93] P. K. Dunsby, E. Elizalde, R. Goswami, S. Odintsov, and D. S. Gomez,

On the LCDM Universe in f(R) gravity, Phys.Rev. D82 (2010) 023519,
[arXiv:1005.2205].
[94] T. Chiba, T. L. Smith, and A. L. Erickcek, Solar System constraints to

general f(R) gravity, Phys.Rev. D75 (2007) 124014, [astro-ph/0611867].
[95] G. Arfken and H. Weber, Mathematical Methods for Physicists. Elsevier

Academic Press, 2005.
[96] S. M. Carroll, Why is the universe accelerating?, eConf C0307282 (2003)

TTH09, [astro-ph/0310342].
[97] W. Hu and I. Sawicki, Models of f(R) Cosmic Acceleration that Evade

Solar-System Tests, Phys.Rev. D76 (2007) 064004, [arXiv:0705.1158].
[98] E. V. Linder, Exponential Gravity, Phys.Rev. D80 (2009) 123528,

[arXiv:0905.2962].
[99] K. Bamba, C.-Q. Geng, and C.-C. Lee, Cosmological evolution in

exponential gravity, JCAP 1008 (2010) 021, [arXiv:1005.4574].
[100] M. Martinelli, A. Melchiorri, and L. Amendola, Cosmological constraints

on the Hu-Sawicki modified gravity scenario, Phys.Rev. D79 (2009)
123516, [arXiv:0906.2350].
Bibliography 159
[101] S. A. Appleby, R. A. Battye, and A. A. Starobinsky, Curing singularities

in cosmological evolution of F(R) gravity, JCAP 1006 (2010) 005,
[arXiv:0909.1737].
[102] D. K. Hazra and A. Shafieloo, Test of Consistency between Planck and

WMAP, arXiv:1308.2911.
[103] M. Doran and G. Robbers, Early dark energy cosmologies, JCAP 0606
(2006) 026, [astro-ph/0601544].
[104] M. Realdi, The Universe of Willem de Sitter, 2007.
[105] A. de la Cruz-Dombriz and A. Dobado, A f(R) gravity without

cosmological constant, Phys.Rev. D74 (2006) 087501, [gr-qc/0607118].
[106] S. Tsujikawa, K. Uddin, and R. Tavakol, Density perturbations in f(R)

gravity theories in metric and Palatini formalisms, Phys.Rev. D77 (2008)
043007, [arXiv:0712.0082].
[107] A. V. Frolov, A Singularity Problem with f(R) Dark Energy,

Phys.Rev.Lett. 101 (2008) 061103, [arXiv:0803.2500].
[108] V. Faraoni, De Sitter attractors in generalized gravity, Phys.Rev. D70

(2004) 044037, [gr-qc/0407021].
[109] H. Motohashi, A. A. Starobinsky, and J. Yokoyama, Analytic solution for

matter density perturbations in a class of viable cosmological f(R) models,
Int.J.Mod.Phys. D18 (2009) 1731–1740, [arXiv:0905.0730].
[110] D. Polarski and R. Gannouji, On the growth of linear perturbations,

Phys.Lett. B660 (2008) 439–443, [arXiv:0710.1510].
[111] E. V. Linder and R. N. Cahn, Parameterized Beyond-Einstein Growth,

Astropart.Phys. 28 (2007) 481–488, [astro-ph/0701317].
160 Bibliography
[112] S. Tsujikawa, Matter density perturbations and effective gravitational

constant in modified gravity models of dark energy, Phys.Rev. D76 (2007)
023514, [arXiv:0705.1032].
[113] S. A. Appleby and J. Weller, Parameterizing scalar-tensor theories for

cosmological probes, JCAP 1012 (2010) 006, [arXiv:1008.2693].
[114] S. Tsujikawa, R. Gannouji, B. Moraes, and D. Polarski, The dispersion of

growth of matter perturbations in f(R) gravity, Phys.Rev. D80 (2009)
084044, [arXiv:0908.2669].
[115] H. Motohashi, A. A. Starobinsky, and J. Yokoyama, f(R) Gravity and its

Cosmological Implications, Int.J.Mod.Phys. D20 (2011) 1347–1355,
[arXiv:1101.0716].
[116] C.-P. Ma and E. Bertschinger, Cosmological perturbation theory in the

synchronous and conformal Newtonian gauges, Astrophys.J. 455 (1995)
7–25, [astro-ph/9506072].
[117] E. Elizalde, S. Odintsov, L. Sebastiani, and S. Zerbini, Oscillations of the

F(R) dark energy in the accelerating universe, Eur.Phys.J. C72 (2012)
1843, [arXiv:1108.6184].
[118] R. Bean, D. Bernat, L. Pogosian, A. Silvestri, and M. Trodden, Dynamics

of Linear Perturbations in f(R) Gravity, Phys.Rev. D75 (2007) 064020,
[astro-ph/0611321].
[119] Y.-S. Song, W. Hu, and I. Sawicki, The Large Scale Structure of f(R)
Gravity, Phys.Rev. D75 (2007) 044004, [astro-ph/0610532].
[120] S. Appleby and R. Battye, Aspects of cosmological expansion in F(R)

gravity models, JCAP 0805 (2008) 019, [arXiv:0803.1081].

Notes

Uploaded by

Notes

Uploaded by

Aspects of Modified Gravity

A thesis submitted to the University of Manchester

1 Introduction to Gravitational Theories and Cosmology 27

1.2 General Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1.2.1 General Relativity, covariant derviatives, energy-momentum

1.2.3 Formulation of the Einstein equations . . . . . . . . . . . . 30

1.2.4 The FRW metric and the Ricci scalar . . . . . . . . . . . . 32

1.2.5 Density, pressure and equation of state . . . . . . . . . . . 33

1.2.6 The Friedmann equations . . . . . . . . . . . . . . . . . . 34

1.2.7 Einstein equations derived from the action . . . . . . . . . 34

1.3 A brief history of the Universe . . . . . . . . . . . . . . . . . . . . 36

1.3.1 The radiation era . . . . . . . . . . . . . . . . . . . . . . . 37

1.3.2 The matter dominated era and the CMB . . . . . . . . . . 37

1.3.3 The vacuum era . . . . . . . . . . . . . . . . . . . . . . . . 38

1.4 Modifying gravity with F (R) gravity models . . . . . . . . . . . . 38

1.4.1 A special case . . . . . . . . . . . . . . . . . . . . . . . . . 40

1.4.2 A vacuum solution . . . . . . . . . . . . . . . . . . . . . . 42

1.4.3 A modified expression for the Hubble parameter and the

1.4.4 Dark energy . . . . . . . . . . . . . . . . . . . . . . . . . . 43

1.4.4.1 The cosmological constant . . . . . . . . . . . . . 44

1.4.4.3 Effective dark energy . . . . . . . . . . . . . . . . 45

1.4.5 The modified gravity particle . . . . . . . . . . . . . . . . 45

1.4.6 The coincidence problem . . . . . . . . . . . . . . . . . . . 46

1.4.7 The growth of matter and perturbations . . . . . . . . . . 47

1.4.8 Constraints on F (R) . . . . . . . . . . . . . . . . . . . . . 47

1.4.9 Conditions for an accelerating Universe . . . . . . . . . . . 49

1.4.10 Probing the Growth of Large Scale Structure . . . . . . . . 49

2 Gravitation Models Which Mimic ΛCDM 53

2.2 Finding all models which mimic ΛCDM . . . . . . . . . . . . . . . 55

2.2.1 Algebraic solution for R > 4Λ . . . . . . . . . . . . . . . . 56

2.2.2 Algebraic solution which includes R = 4Λ . . . . . . . . . 58

2.2.3 Matching conditions for the two expressions for F (R) . . . 59

2.2.4 Application of constraints to the solutions . . . . . . . . . 64

2.2.4.1 The condition that FR (R) > 0 . . . . . . . . . . . 65

2.2.4.2 The condition that FRR (R) ≥ 0 . . . . . . . . . . 65

2.2.4.3 The local gravity constraint . . . . . . . . . . . . 65

2.2.5 The matter growth index, γ . . . . . . . . . . . . . . . . . 67

2.3 In the radiation era . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3 Non-ΛCDM Gravitation Models 75

3.2 Three models compared . . . . . . . . . . . . . . . . . . . . . . . 78

3.2.1 The Erf model . . . . . . . . . . . . . . . . . . . . . . . . . 79

3.2.1.1 The dark energy equation of state . . . . . . . . . 80

3.2.1.2 Late-time behaviour . . . . . . . . . . . . . . . . 84

3.2.1.3 The matter growth index . . . . . . . . . . . . . 87

3.2.2 The AB model . . . . . . . . . . . . . . . . . . . . . . . . 89

3.2.3 The HSS model . . . . . . . . . . . . . . . . . . . . . . . . 92

4 The weff < −1 Theorem for F (R) Models 101

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

4.2 The theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

5 Evolution of Perturbations 105

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

5.2.1 The ΛCDM model . . . . . . . . . . . . . . . . . . . . . . 109

5.3 Approximating algebraic solutions . . . . . . . . . . . . . . . . . . 110

5.3.1 Oscillatory solution . . . . . . . . . . . . . . . . . . . . . . 112

5.3.2 Non-oscillatory solution . . . . . . . . . . . . . . . . . . . 116

5.3.3 The combined algebraic solution . . . . . . . . . . . . . . . 117

5.3.4 Applying initial conditions to the algebraic approximation 119

5.4 Algebraic and numerical solutions compared . . . . . . . . . . . . 120

5.4.1 k = 0.1h Mpc−1 . . . . . . . . . . . . . . . . . . . . . . . . 121

5.4.2 k = 0.01h Mpc−1 . . . . . . . . . . . . . . . . . . . . . . . 127

5.4.4 Algebraic and numeric solutions for x(N ) compared for

5.4.4.1 Model 2 . . . . . . . . . . . . . . . . . . . . . . . 133

5.4.4.2 Model 4 . . . . . . . . . . . . . . . . . . . . . . . 134

5.4.4.3 Model 5 . . . . . . . . . . . . . . . . . . . . . . . 135

5.4.5 Comment . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

5.5.1 Oscillations in δR . . . . . . . . . . . . . . . . . . . . . . . 139

6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

6.1.1 Models with standard Einstein Gravity expansion history . 142

6.1.2 Extreme values generated by F (R) models . . . . . . . . . 143

6.1.3 Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

6.1.4 Oscillations of the potentials of the perturbed FRW metric 144

6.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

No portion of the work referred to in this thesis has been

1. The author of this thesis (including any appendices and/or schedules to

2. Copies of this thesis, either in full or in extracts and whether in hard or