Home Search Collections Journals About Contact us My IOPscience

Introduction to the application of dynamical systems theory in the study of the dynamics of

cosmological models of dark energy

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2015 Eur. J. Phys. 36 025008

(http://iopscience.iop.org/0143-0807/36/2/025008)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 193.255.248.150
This content was downloaded on 30/01/2015 at 11:15

Please note that terms and conditions apply.

 European Journal of Physics
Eur. J. Phys. 36 (2015) 025008 (23pp) doi:10.1088/0143-0807/36/2/025008

Introduction to the application of dynamical

systems theory in the study of the dynamics
of cosmological models of dark energy
Ricardo García-Salcedo1, Tame Gonzalez2,
Francisco A Horta-Rangel2, Israel Quiros2,3 and
Daniel Sanchez-Guzmán1
1
CICATA— Legaria del Instituto Politécnico Nacional, 11500, México, D.F., Mexico
2
Departamento de Ingeniería Civil, División de Ingeniería, Universidad de Guanajuato,
Gto., Mexico

E-mail: rigarcias@ipn.mx, tamegc72@gmail.com, anthort@hotmail.com,

iquiros6403@gmail.com and dsanchezgzm@gmail.com

Received 13 August 2014, revised 14 December 2014

Accepted for publication 16 December 2014
Published 19 January 2015

Abstract
The theory of dynamical systems is a very complex subject that has produced
several surprises in the recent past in connection with the theory of chaos and
fractals. The application of the tools of dynamical systems in cosmological
settings is less known, in spite of the number of published scientific papers on
this subject. In this paper, a mostly pedagogical introduction to the cosmo-
logical application of the basic tools of dynamical systems theory is presented.
It is shown that, in spite of their amazing simplicity, these tools allow us to
extract essential information on the asymptotic dynamics of a wide variety of
cosmological models. The power of these tools is illustrated within the context
of the so-called Λ-cold dark matter (ΛCDM) and scalar field models of dark
energy. This paper is suitable for teachers, undergraduate students, and post-
graduate students in the disciplines of physics and mathematics.

Keywords: science teaching, cosmology, nonlinear dynamics and chaos

1. Introduction

Cosmology is the study of the structure and evolution of the universe as a whole. Einsteinʼs
general relativity (GR) stands as its mathematical basis. In general, Einsteinʼs GR equations
are a very complex system of coupled, nonlinear differential equations, but the system is
3
Author to whom any correspondence should be addressed.

0143-0807/15/025008+23$33.00 © 2015 IOP Publishing Ltd Printed in the UK 1

Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

simplified by underlying spacetime symmetries. The Friedmann−Robertson−Walker (FRW)

spacetime model is depicted by the line element4
⎡ dr 2 ⎤
ds 2 = −dt 2 + a(t )2 ⎢
⎣ 1 − k r2
(
+ r 2 dθ 2 + sin2 θ dϕ 2 ⎥ ,
⎦
)
where t is the cosmic time, r, θ, and ϕ are comoving spherical coordinates, a(t) is the
cosmological scale factor, and the constant, k = −1, 0, +1, parametrizes the curvature of the
spatial sections. This model is based on what is known as the cosmological principle: at large
scales, the Universe is homogeneous and isotropic. Homogeneity and isotropy imply that
there is no preferred place and no preferred direction in the universe5. These observations
confirm the validity of the cosmological principle at large scales (∼100 Mpc ≈ 10 26 cm ). The
dynamics of the universe is fully given by the explicit form of the scale factor, which depends
on the symmetry properties and the matter content of the Universe.
To determine the scale factor, it is necessary to solve Einsteinʼs cosmological field
equations, which relate the spacetime geometry with the distribution of matter in the Uni-
verse. The maximal symmetry implied by the cosmological principle allows for great sim-
plification of Einsteinʼs equations. In this paper, for further simplification, we shall consider
FRW spacetimes with flat spatial sections (k = 0). The resulting cosmological equations are
written as follows6:

1⎛ ⎞
3H 2 = ∑ρx + ρ X , H˙ = − ⎜⎜ ∑γx ρx + γX ρ X ⎟⎟ ,
x
2⎝ x ⎠
ρ˙x + 3γx Hρx = 0, ρ˙ X + 3γX Hρ X = 0, (1)
where H ≡ a˙ a is the Hubble parameter, ρx and ρX are the energy densities of the gravitating
matter and of the X-fluid (the dark energy in this paper), respectively, and the sum is over all
of the gravitating matter species living in the FRW spacetime (dark matter, baryons, radiation,
etc.). The last two equations above are the conservation equations for the gravitating matter
degrees of freedom and for the X-fluid, respectively7. In addition, it is adopted that the
following equations of state are obeyed:
( ) (
px = γx − 1 ρx , p X = γX − 1 ρ X , )
where px is the barotropic pressure of the x-matter fluid and pX is the parametric pressure of
the X-component8, while γx and γX are their barotropic parameters9.
According to our current understanding of expansion history, the Universe was born out
of a set of initial conditions known as the big bang. Further evolution led to a stage of
primeval inflation driven by a scalar field [4]. The inflation era was followed by a stage
dominated by radiation, which was followed in turn, by an intermediate stage dominated by

4
For a pedagogical exposition of the geometrical and physical significance of the FRW metric, see [1].
5
Here we adopt the units where 8πG = c = 1.
6
For a didactic derivation of the cosmological equations not based on GR, see [1–3].
7
It should be pointed out that one equation in (1), say the second (Raychaudhuri) equation, is redundant.
8
By parametric pressure, we understand that it might not correspond to standard barotropic pressure with the usual
thermodynamic properties, but that it is rather a conveniently defined parameter that obeys similar equations to its
matter counterpart.
9
In several places in this paper, we use the so-called equation of state (EOS) parameter, ω X = p X ρ X = γX − 1,
interchangeably with the barotropic parameter, γX .

2
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

nonrelativistic cold dark matter (CDM)10. During this stage of the cosmic expansion, most of
the structure we observe was formed. The end of the matter-domination era can be traced back
to a recent moment of the cosmic expansion when a very peculiar form of matter known as
dark energy (DE), which does not interact with baryons, radiation, or any other form of
‘visible’ matter, started dominating [5]. This alien form of matter antigravitates and causes the
present Universe to expand at an accelerating pace instead of decelerating, as it would for a
matter-dominated (either CDM or baryons) or radiation-dominated Universe. Actually, if one
substitutes
a¨
= H˙ + H 2
a
back into the second equation in (1) and takes into account the Friedmann equation (the first
equation in (1)) one is left with the following equation:
a¨ 1 ⎛ 2⎞ 1⎛ 2⎞
= − ∑ ⎜ γx − ⎟ ρx − ⎜ γX − ⎟ ρ X . (2)
a 2 x ⎝ 3 ⎠ 2 ⎝ 3⎠
It is known that for standard (gravitating) forms of matter, the barotropic pressure is a
non-negative quantity: px = (γx − 1) ρx ⩾ 0 ⇒ γx ⩾ 1. In particular, for CDM and baryons,
the (constant) barotropic index γm = 1, while for radiation γr = 4 3. Hence, since obviously
γx > 2 3—if we forget for awhile about the second term on the right-hand side (rhs) of (2)—
one can see that a cosmic background composed of standard matter will expand at a decel-
erated pace (ä < 0 ). In contrast, for unconventional matter with γX < 2 3 (the parametric
pressure pX is obviously negative), the second term on the rhs of equation (2) contributes to
accelerating the pace of the cosmological expansion. In particular, if we choose the X-
component in the form of vacuum energy, ( γvac, ρvac ) → ( γX , ρX ), usually identified with the
energy density of the cosmological constant (see section 3), since pvac = −ρvac , then
γvac = 0 < 2 3. Due to the known fact that the vacuum energy is not diluted by the cosmic
expansion11, eventually this component of the cosmic budget will dominate the cosmological
dynamics, leading to late-time acceleration of the expansion
a¨ 1
≈ ρvac > 0.
a 3
A related (dimensionless) parameter that measures whether the expansion is accelerated
or decelerated is the so-called deceleration parameter, which is defined as
H˙ aa¨
q ≡ −1 − 2 = − 2 . (3)
H a˙
The pace of the expansion accelerates if q < 0 . On the contrary, positive q > 0 means that the
expansion is decelerating. This parameter will be useful in the discussion in the following
sections.
In spite of the apparent simplicity of equation (1), as with any system of nonlinear
second-order differential equations, it is a very difficult (and perhaps unsuccessful) task to
find exact solutions. Even when an analytic solution can be found, it will not be unique, but
rather just one in a large set of solutions [6]. This is in addition to the question about the
10
Dark matter is a conventional form of matter in the sense that it gravitates just like any other known sort of
standard matter such as radiation, baryons, etc. However, unlike the latter forms of matter, CDM does not interact
with radiation, which is why it is called ‘dark matter.’
11
We recall that, in contrast to the vacuum energy density, which does not dilute with the expansion
ρvac = Λ = const , the energy density of CDM and baryons dilutes like ∝ a−3, while the density of radiation decays
very quickly ∝ a−4.

3
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

stability of given solutions. In this case, the dynamical systems tools come to our rescue.
These very simple tools allow us to correlate important concepts like past and future attractors
(and saddle equilibrium points) in the phase space, with generic solutions to the set of
equation (1) without the need to analytically solve them.
In correspondence with the previously mentioned stages of the cosmic expansion, one
expects that the state space and the phase space of any feasible cosmological model should be
characterized by at least one of the following equilibrium configurations in the space of
cosmological states: (i) a scalar-field-dominated (inflationary) past attractor, (ii) saddle critical
points associated with radiation-domination, radiation-matter scaling, and matter-domination,
and either (iii) a matter/DE scaling or (iv) a de Sitter12 future attractor. Any heteroclinic orbit
joining these critical points then represents a feasible image of the featured cosmic evolution
in the phase space13.
While the theory of dynamical systems has produced several surprises in the recent past
in connection with the theory of chaos [7, 8] and fractals [9], the application of the tools of
dynamical systems in cosmological settings is less known, in spite of the number of published
scientific papers on this subject (to cite few, see [2, 5, 10–19] and the related references
therein).
This paper presents an introduction to the application of the simplest tools of the theory
of dynamical systems—those that can be explained with only previous knowledge of the
fundamentals of linear algebra and of ordinary differential equations—within the context of
the so-called ‘concordance’ or ΛCDM (section 3) and scalar field models (section 4) of dark
energy [5, 11–16, 18–22]. The dynamical systems tools play a central role in the under-
standing of the asymptotic structure of these models. These tools are helpful in looking for an
answer to questions like where does our Universe come from? What would be its fate? As a
specific example, the cosh potential is explored in section 5, to show the power of these tools
in the search for generic dynamical behaviour.
We want to stress that there are very good introductory and review papers on the
application of dynamical systems in cosmology—see, for instance, [5, 10–12, 16]—so, what
is the point of writing another introductory paper on the issue? We want to point out that in
the present paper, we aim not only to provide another introductory exposition of the essentials
of the subject, but also to fill several gaps regarding specific topics not usually covered in
similar publications. Here we pay special attention to the necessary (yet usually forgotten)
definition of the physically meaningful phase space (i.e., the region of the phase space that
contains physically meaningful cosmological solutions). We also include the derivation of
relevant formulas to pave the way for beginners in their progress towards concrete compu-
tations. We shall explain, in particular, a method developed in [23] to consider arbitrary
potentials within the dynamical systems study of scalar field cosmological models. This
subject is not usually covered in the existing introductory literature. For completeness, at the
end of the paper we include an appendix (section appendix) where an elementary introduction
to the theory of the dynamical systems is given and useful comments on the interplay between
the cosmological field equations and the equivalent phase space are also provided. Only
knowledge of the fundamentals of linear algebra and of ordinary differential equations is
required to understand the material in the appendix. The exposition in the appendix is
12
A de Sitter cosmological phase is a stage of the cosmological expansion whose dynamics is described by the line
element ds 2 = −dt 2 + e H0 t dx 2 , where H0 is the Hubble constant. Since the deceleration parameter,
q ≡ −1 − H˙ H 2 = −1, is a negative quantity, the de Sitter expansion is inflationary.
13
For very concise introductory information on the theory of dynamical systems and related concepts like critical
points (attractors and saddle points), heteroclinic orbits, etc, see the appendix.

4
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

complete enough to allow one to understand how the computations in sections 3 and 4 are
done. We recommend that readers who are not familiar with dynamical systems theory start
with this section.
Our goal is to keep the discussion as general as possible while presenting the exposition
in as pedagogical a way as possible. The paper contains many footnotes with comments and
definitions that complement the main text. Since the subject is covered with some degree of
technical detail, and since a basic knowledge of cosmology is assumed, this paper is suitable
for teachers, undergraduate students, and postgraduate students, from the disciplines of
physics and mathematics.

2. Asymptotic cosmological dynamics: a model-independent analysis

In this section, the asymptotic structure in the phase space of a general DE model will be
discussed in detail. For simplicity and compactness of the exposition, we shall assume only a
two-component cosmological fluid composed of CDM, which is labelled here by the index
‘m,’ (i.e., (ρm , pm ) → (ρx , px )) and of DE (the X-component).
To derive an autonomous system of ordinary differential equations (ASODE) out of (1),
it is useful to introduce the so-called dimensionless energy density parameters of matter and
of the DE
ρm ρX
Ωm = , ΩX = , (4)
3H 2 3H 2
respectively, which are always non-negative quantities14. In terms of these parameters, the
Friedmann equation—the first equation in (1)—can be written as the following constraint:
Ω m + Ω X = 1, (5)
which entails that none of the non-negative dimensionless energy density components alone
may exceed unity: 0 ⩽ Ωm ⩽ 1, 0 ⩽ Ω X ⩽ 1.
Given the above constraints, if one thinks of the dimensionless density parameters as
variables of some state space, only one of them is linearly independent, say Ω X . Let us write a
dynamical equation for Ω X , where we rewrite the Raychaudhuri equation—the second
equation in (1)—and the conservation equation for the X-component, in terms of Ω X :
H˙
2 = −3γm Ω m − 3γX Ω X
H2
= −3γm + 3( γm − γX ) Ω X ,
ρ˙ X
= −3γX HΩ X . (6)
3H 2
Then, if we substitute (6) into
ρ˙ X H˙
Ω˙ X = −2 ΩX,
2 H
3H
and consider constraint (5), we obtain
Ω X′ = 3( γm − γX ) Ω X ( 1 − Ω X ), (7)

14
In agreement with the conventional non-negativity of energy, we consider non-negative energy densities
exclusively.

5
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

where the tilde denotes the derivative with respect to the new variable, τ ≡ ln a (Ω˙ = HΩ′,
etc).
In general, for varying γX , another ordinary differential equation (ODE) for γX is needed
to have a closed system of differential equations (the barotropic index of matter, γm, is usually
set to a constant). The problem is that, unlike the ODE (7), which is model-independent, the
ODE for γX = (ρX + pX ) ρX requires certain specifications that depend on the chosen model
(see below).
Certain important results may be extracted from (7) under certain assumptions. For
instance, if we assume a constant γX , the previously mentioned ODE is enough to uncover the
asymptotic structure of the model (1) in the phase space. In this case, the phase space is the
one-dimensional segment, Ω X ∈ [0, 1]. The critical points are

Ω X ( 1 − Ω X ) = 0 ⇒ Ω X = 1, Ω X = 0,

where we assume that γX ≠ γm . The first fixed point, Ω X = 1, is correlated with DE

domination, while the second one, Ω X = 0 (Ωm = 1), is associated with CDM-dominated
expansion.
If we linearize equation (7) around the critical points, 1 − δ1 → 1 and 0 + δ 0 → 0 (δ1 and
δ 0 are small perturbations), one obtains

δ1′ = −3( γm − γX ) δ1 ⇒ δ1 (τ ) = δ¯1 e −3(γm− γX ) τ ,

δ 0′ = 3( γm − γX ) δ 0 ⇒ δ 0 (τ ) = δ¯0 e 3(γm− γX ) τ . (8)

As seen in the DE-dominated solution,

Ω X = 1 ⇒ 3H 2 = ρ X ∝ a−3γX ⇒ a (t ) ∝ t 2 3γX

is a stable (isolated) equilibrium point whenever γm > γX since, in this case, the perturbation

δ1 (τ ) ∝ e −3(γm− γX ) τ ,

decreases exponentially. Hence, given that γm > γX , the DE-dominated solution, Ω X = 1, is

the future attractor solution that describes the fate of the Universe. Besides, if γm > γX , the
matter-dominated solution, Ωm = 1 (Ω X = 0): a (t ) ∝ t 2 3γm, is the past attractor since the
initially small perturbation, δ 0 (τ ) ∝ e 3(γm − γX ) τ , increases exponentially with τ , thus taking the
system away from the condition Ωm = 1. Otherwise, if γm < γX , the X-dominated solution is
the past attractor while the matter-dominated solution is the future attractor. However, this last
situation is not consistent with the known cosmic history of our Universe.
The interesting thing about the above result is that it is model-independent (i.e., no matter
which model we adopt for the DE (X-component), given that γX is a constant, the above
mentioned critical points and their stability properties are obtained). Besides, since a non-
constant, γX , obeys an autonomous ODE of the general form γX′ = f (γX , Ω X ), at any equili-
brium point where, necessarily

Ω X′ = 0, γX′ = 0 ⇒ γX = γ¯X = const

provided that γ¯X ≠ γm , either Ω X = 1, or Ω X = 0 (Ωm = 1). Hence, independent of the model
adopted to account for the X-component, matter-dominated and X-fluid-dominated solutions
are always equilibrium points in the phase space of the two-fluids cosmological model
depicted by equations (1). This is a generic result.

6
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

3. ΛCDM model

With the help of a specific cosmological model of DE, we will now show how it is possible
that a two-dimensional system of ODEs may come out of the three-dimensional system of
second-order field equations (1) where, we recall, one of these equations is redundant. For this
purpose, we will focus on the so-called Λ-cold dark matter (ΛCDM) model [24], whose phase
space is a subspace of the phase plane15. Here, and only for the extent of this section, in
addition to the CDM and to the DE (the cosmological constant), we will consider a radiation
fluid with energy density ρr and pressure pr = ρr 3 (γr = 4 3). The conservation equation for
radiation reads
ρ˙r + 4Hρr = 0 ⇒ ρr ∝ a−4 .
Meanwhile, for the cosmological constant term, as for any vacuum fluid, we have pΛ = −ρΛ
(γΛ = 0 ), which means that ρ˙Λ = 0 , so this term does not evolve during the course of the
cosmic expansion.
In what follows, since we adopted the units system with 8πG = c = 1, we write ρΛ = Λ.
Besides, since we deal here with CDM, which is modelled by a pressureless dust, we set
γm = 1. After these assumptions, and identifying ρX ≡ ρΛ = Λ in equation (1), the resulting
cosmological equations are
1⎛4 ⎞
3H 2 = ρr + ρm + Λ , H˙ = − ⎜ ρr + ρm ⎟ ,
2⎝3 ⎠
ρ˙m + 3Hρm = 0, ρ˙r + 4Hρr = 0. (9)
At this point, let us explore how the cosmological constant can explain the present speed-
up of the cosmic expansion. If we conveniently combine the first two equations above, taking
into account that straightforward integration of the last equations in (9) yields ρm = 6Cm a3
and ρr = 3Cr a 4 , respectively (Cm and Cr are constants), we get
a¨ Cr Cm Λ
= H˙ + H 2 = − 4 − 3 + .
a a a 3
We see that acceleration of the expansion ä > 0 may occur in this model thanks to the
third term above (the DE, i.e., the cosmological constant). Besides, the accelerated pace of the
expansion is a recent phenomenon: during the course of the cosmic evolution, the initially
dominating radiation component dilutes very quickly, ∝ a−4, until the matter component starts
dominating. As the cosmic expansion further proceeds, the matter component also dilutes,
∝ a−3, until very recently, when the cosmological constant started to dominate to yield to
positive ä > 0 .
Following the same procedure as in the preceding section, we write the Friedmann
constraint

Ωr + Ω m + Ω Λ = 1, (10)

where Ωr = ρr 3H 2 and Ω Λ = Λ 3H 2 . In what follows, we choose the following variables of

the phase space:
x ≡ Ωr , y ≡ Ω Λ, (11)

15
For a dynamical systems study of the FRW cosmological model with spatial curvature k ≠ 0 and with a
cosmological constant, see [2].

7
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

Figure 1. Phase portrait of the ASODE (13) corresponding to the ΛCDM model (top
panel). The orbits of (13) emerge from the past attractor, P: r (1, 0)—the radiation
domination state—and end up in the future attractor, PdS : (0, 1) (de Sitter phase). For a
non-empty set of initial conditions, the corresponding orbits approach close enough to
the matter-dominated saddle point at the origin, Pm: (0, 0). The bottom panel shows the
phase portrait of the ASODE (31), which presents two orbits generated by two different
sets of initial conditions. Here, the free parameter, μ, has been arbitrarily set at μ = 10 .

where, for the sake of simplicity of writing, we adopt x and y in place of the dimensionless
energy densities of radiation, Ωr , and of the cosmological constant, Ω Λ, respectively. But we
warn the reader that the same symbols, x and y, will be used in the next sections to represent
different variables of the phase space.
We have that Ωm = 1 − x − y , and, since 0 ⩽ x ⩽ 1, 0 ⩽ y ⩽ 1, and 0 ⩽ Ωm ⩽ 1, the
physically relevant phase space is defined as the following two-dimensional triangular region
(see the top panel of figure 1):

ΨΛ = { (x , y): x + y ⩽ 1, 0 ⩽ x ⩽ 1, 0 ⩽ y ⩽ 1} . (12)

As before, to derive the autonomous ODEs, we first take the derivative of the variables x
and y with respect to the cosmic time, t:

8
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

ρ˙r ρr H˙ Λ H˙
x˙ = −2 , y˙ = −2 ,
2 2 H
3H 3H 3H 2 H

where we have to keep in mind the Raychaudhuri equation and the conservation equation for
radiation in (9), written in terms of the variables of the phase space

2
H
H˙ = − ( 3 + x − 3y ) ,
2
and then we have to go to derivatives with respect to the variable τ = ln a (t ) (ξ′ → H −1ξ˙).
We obtain
x ′ = − x ( 1 − x + 3y ) , y ′ = ( 3 + x − 3 y ) y . (13)
The simplicity of the system of two ordinary differential equation (13) is remarkable
when compared with the system of three second-order cosmological equation (9). The critical
points of (13) in ΨΛ are easily found by solving the following system of algebraic equations:
x (1 − x + 3y ) = 0, (3 + x − 3y ) y = 0 .
In the present case, in terms of the variables x, y, the deceleration parameter,
q = −1 − H˙ H 2 (equation (3)), can be written as
q = (1 + x − 3y) 2,
which means that those fixed points that fall in the region of the phase space lying above the
line, y = 1 3 + x 3, correspond to cosmological solutions where the expansion is
accelerating.

3.1. Critical points

2
3.1.1. Radiation-dominated critical point. r (1, 0), Ωr = 1 ⇒ 3H = ρr
The critical point, P:
⇒ a (t ) ∝ t , corresponds to the radiation-dominated cosmic phase. This solution depicts
decelerated expansion, since q = 1. The linearization matrix for the system of ODE (13) (for
details, see the appendix) is
⎛ −1 + 2x − 3y −3x ⎞
J=⎜ ⎟.
⎝ y 3 + x − 6y ⎠
The roots of the algebraic equation

det J ( Pr ) − λU = det 1 − λ −3
0 (
4−λ
= 0, )
are the eigenvalues λ1 = 1 and λ 2 = 4. Since both eigenvalues are positive reals, then Pr is a
source point (past attractor). This means that in the ΛCDM model described by the
cosmological equation (9), the radiation-dominated solution, a (t ) ∝ t , is a privileged
solution.
From the inspection of the phase portrait of (13), seen in the top panel of figure 1, it is
apparent that any viable pattern of cosmological evolution should start in a state where the
matter content of the Universe is dominated by radiation. Of course this is a drawback of our
classical model, which cannot explain the very early stages of the cosmic evolution, where the
quantum effects of gravity play a role. This includes the early inflation. In a model that would
account for this early period of the cosmic dynamics, inflation should be the past attractor in
the equivalent phase space.

9
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

3.1.2. Matter-dominated critical point. The critical point, Pm: (0, 0), Ωm = 1 ⇒ 3H 2 = ρm
⇒ a (t ) ∝ t 2 3, corresponds to the matter-dominated solution, which is associated with
decelerated expansion (q = 1 2). Following the same procedure above (see the appendix) one
finds the following eigenvalues of the linearization matrix, J (Pm ), evaluated at the hyperbolic
equilibrium point, Pm: λ1 = 1, λ 2 = −3. Hence, this is a saddle critical point. As seen in the
phase portrait in the top panel of figure 1, only with conveniently chosen initial conditions do
the given orbits in ΨΛ approach close enough to Pm. Since this is an unstable (metastable)
critical point, it can only be associated with a transient stage of the cosmic expansion. This is
good, since a stage dominated by dark matter can only be a transient state, lasting for enough
time to account for the observed amount of cosmic structure. A drawback is that we need to
fine tune the initial conditions in order for feasible orbits in the phase space to get close
enough to Pm.

3.1.3. de Sitter phase. The critical point, PdS : (0, 1), Ω Λ = 1 ⇒ 3H 2 = Λ ⇒ a (t ) ∝ e Λ 3e t ,

corresponds to a stage of inflationary (q = −1) de Sitter expansion. The eigenvalues of the
matrix J (PdS ) are λ1 = −3 and λ 2 = −4. This means that PdS is a future attractor (see the
classification of isolated equilibrium points in the appendix). This entails that, independent of
the initial conditions chosen, Ωr0 , Ω Λ0 , the orbits in ΨΛ are always attracted to the de Sitter
state, which explains the actual accelerated pace of the cosmic expansion in perfect fit with
the observational data. This is why, in spite of the serious drawbacks in connection with the
cosmological constant problem [25, 26], the very simple ΛCDM model represents such a
successful description of the present cosmological paradigm and therefore is called the
‘concordance model.’ Any other DE model, regardless of its nature, ought to be compared to
the ΛCDM predictions as a first viability test.
Additional refinement of the above model is achieved if we add the energy density of
baryons.

4. Scalar field models of dark energy: the dynamical systems perspective

The cosmological constant problem can be split into two questions [25]. (i) Why isn’t the
vacuum energy ρvac = Λ much larger? (ii) Why it is of the same order of magnitude as the
present mass density of the Universe? The first question is an old cosmological constant
problem, and the second is acknowledged as the new cosmological constant problem. To
avoid the old cosmological constant problem, which is exclusive to the ΛCDM model, scalar
field models of dark energy are invoked16. In this last case, an effective ‘dynamical’ cos-
mological constant is described by the scalar fieldʼs X self-interaction potential, V(X). This
may be arranged so that at early times, the vacuum energy, ρvac,0 = V (X0 ) ( Ẋ = 0), is large
enough to produce the desired amount of inflation, while at late times, ρvac, f = V (X f ) is of
the same order of the CDM energy density, ρm .
In this section, we will explore the asymptotic structure of general scalar field models of
DE [5, 11–16, 18–22, 27, 28]. These structures represent a viable alternative to the ΛCDM
model explored above. Among them, ‘exponential quintessence,’ V (X ) = M exp (−μX )
[11, 29], is one of the most popular scalar field models of DE.
The specification of a scalar field model for the DE means that the energy density and the
parametric pressure in the cosmological field equation (1) are given by
16
The new cosmological problem is also inherent in several scalar field models of dark energy.

10
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

2 2
ρ X = X˙ 2 + V (X ), p X = X˙ 2 − V (X ), (14)

respectively. In these equations, V = V (X ) is the self-interacting potential of the scalar field,

X. In addition, for the equation of state parameter ω X = γX − 1, one has
2
pX X˙ − 2V
ωX ≡ = . (15)
ρX X˙ 2 + 2V
After the above choice, the conservation equation (1) can be written in the form of the
following Klein-Gordon equation:

X¨ + 3HX˙ = −V, X . (16)

It happens that deriving an autonomous ODE for the variable ω X (see section 2) can be a
very difficult task. Hence, it could be better to choose a different set of phase space variables
to do the job. Here, instead of the phase space variables Ω X and ω X , we choose the variables
[11]
X˙ V
x≡ , y≡ , (17)
6H 3H
so that the dimensionless energy density of the scalar field, X, is given by Ω X = x 2 + y 2 , and
the Friedmann constraint (10) can be written as

Ωm = 1 − x 2 − y2. (18)

4.1. Determination of the physical phase space

The first step toward a complete study of the asymptotic structure of a given cosmological
model is the rigorous determination of the phase space where we look for relevant equilibrium
points. In the present model, there are several constraints on the physical parameters that help
us define the physically meaningful phase space.
One of these constraints is given by (18)

0 ⩽ Ω m ⩽ 1 ⇒ 0 ⩽ x 2 + y 2 ⩽ 1.

Also, since 0 ⩽ Ω X ⩽ 1, then ∣ x ∣ ⩽ 1. Additionally, we will be interested in cosmic

expansion exclusively, H ⩾ 0, so that y ⩾ 0 . For instance, for potentials of one of the
following kinds (see below), V = V0 , V = V0 e±μX , the physical phase plane is defined as the
upper semi-disc,

{
ΨX := (x , y): 0 ⩽ x 2 + y 2 ⩽ 1, x ⩽ 1, y ⩾ 0 .} (19)

However, as we will see, in general (arbitrary potentials) the phase space is of a dimension
higher than 2 [20].

4.2. Autonomous system of ODEs

To derive the autonomous ordinary equations for the phase space variables x, y, one proceeds
in a similar fashion as in section 2. First, we write the Raychaudhuri equation in (1)

11
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

1
(
H˙ = − γm ρm + ρ X + p X
2
)
2
H˙ X˙
⇒ −2 = 3γm Ω m + ,
H2 H2
in terms of the new variables:

H˙
−2
H2
(
= 3γm 1 − x 2 − y 2 + 6x 2 , ) (20)

where we have taken into account the Friedmann constraint (18). Then, given definition (17),
let us find

X¨ X˙ H˙
x˙ = − , (21)
6H 6H H
or if one substitutes (16) and (20) back into (21), then
3γm 3 V, X 2
(
x ′ = − 3x 1 − x 2 + ) 2
(
x 1 − x 2 − y2 − ) 2 V
y , (22)

where we replaced derivatives with respect to the cosmic time, t, with derivatives with respect
to the variable τ ≡ ln a (t ). Applying the same procedure with the variable y, one obtains
⎛ 1 V, X X˙ H˙ ⎞
y′ = y ⎜ − ⎟,
⎝2 V H H2⎠
where we have taken into account that V˙ = V, X X˙ ; hence
⎡ 3γm ⎤ 3 V, X
y′ = y ⎢
⎣ 2
(
1 − x 2 − y 2 + 3x 2 ⎥ +
⎦
) 2 V
xy . (23)

Unless V = V0 ⇒ V, X V = 0, or V = V0 e±μX ⇒ V, X V = ±μ = const , the ASODEs (22)

and (23) are not a closed system of equations since one equation involving the derivative of
V, X V with respect to τ is lacking.
One example where the ASODEs (22) and (23) are indeed closed systems of ODEs is
given by the so-called ‘exponential quintessence’. This case has been investigated in detail in
[11], so we point the reader to that reference to look for a very interesting example of the
application of the dynamical systems tools in cosmology. As a matter of fact, reproducing the
results of the dynamical systems study in [11] is a very useful exercise for those who want to
learn how to apply dynamical systems tools in cosmological settings.
To be able to consider self-interaction potentials beyond the constant and the exponential
potentials, in addition to variables x, y, one needs to adopt a new variable [14, 23, 30]
z ≡ V, X V, (24)

so that z = 0 corresponds to the constant potential, while z = ±μ for the exponential

potential17. By taking the derivative of the new variable, z, with respect to τ, one obtains
17
For non-exponential potentials, in addition to the increase in the dimensionality of the phase space, there is
another problem. Typically the phase space becomes unbounded, so that two different sets of variables are required to
cover the entire phase space [20]. In the present paper, however, the chosen example of the cosh-like potential does
not lead to unbounded phase space. In consequence, a single set of phase space variables is enough to describe the
whole asymptotic (and intermediate) dynamics.

12
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

z′ = 6 xf (z ), (25)
where we have defined (as before, take into account that V˙ = V, X X˙ and V˙, X = V, XX X˙ , etc)
f (z ) ≡ z 2 [Γ − 1], Γ ≡ VV, XX V,2X , (26)
and the main assumption has been that the above Γ is a function of the variable z, Γ = Γ (z ).
Note that, since
VV, XX V, XX V 1 V, XX
Γ≡ = = ,
V,2X ( V,X V ) 2
z2 V

the left-hand equation in (26) can also be rewritten as

V, XX
f (z ) = − z 2. (27)
V
In the event that Γ cannot be explicitly written as a function of z, then an additional ODE,
Γ′ = … , is to be considered. However, this case is far more complex and does not frequently
arise.
Before going further, we want to point out that the function Γ and the variable z were first
identified in [30]. In addition, the dynamical system (22), (23) and (25) was explored in [14]
several years before it was studied in [23]. However, it was in [23] where the possibility that
for several specific self-interaction potentials, Γ can be explicitly written as a function of z,
was explored for the first time. In [14], although the correct dynamical system was identified,
the authors were not interested in specific self-interaction potentials. The cost of the achieved
generality of the analysis was that the authors had to rely on the obscure notion of ‘instan-
taneous critical points’.

5. An example: the cosh-like potential

To provide a concrete example of how the function f (z) can be obtained for an specific
potential, let us choose the cosh potential [31, 32]
V = V0 cosh (μX ). (28)
This potential, as well as a small variation of it in the last row of table 1, has a very interesting
behaviour near the minimum at X = 0. Actually, in the neighbourhood of the minimum,
potential (28) approaches
m2 2
V (X ) ≈ V0 + X , m2 = V0 μ2 .
2
The quadratic term is responsible for oscillations of the scalar field around the minimum,
which play the role of CDM; hence, at late times when the system is going to stabilize in the
minimum of V(X), the cosh potential makes the scalar field model of DE indistinguishable
from the standard ΛCDM model (see the properties of the equivalent model of [15]).
Taking derivatives of (28) with respect to the scalar field X, we have that
V, X = μV0 sinh (μX ), and V, XX = μ2 V . Hence,
z = V, X V = μ tanh (μX ), f (z ) = μ2 − z 2. (29)
Notice that the z, which solves f (z ) = 0 (i.e., z = ±μ, which corresponds to the exponential
potentials), are the critical points of (25). In table 1, the functions f (z) for several well-known
potentials are displayed.

13
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

Table 1. The functions, f (z), in equation (27) for different self-interaction potentials.

Correspondence f (z ) ⇔ V (X )
Function f (z) z Self-interaction potential V(X) Reference

f (z ) = 0 ±μ V = V0 e±μX [11]
α + β e (β − α ) X
f (z ) = −αβ + (α + β ) z − z 2
1 + e (β − α ) X
(
V = V0 eαX + e βX ) [21]
f (z ) = (p 2 μ 2 − z 2 ) p pμ cotanh(μX ) V = V0 sinh p (μX ) [31]
f (z ) = (p 2 μ 2 − z 2 ) p pμ tanh (μX ) V = V0 cosh p (μX ) [31]
sinh (μX )
f (z ) = (p 2 μ2 − z 2 ) 2p pμ cosh (μX ) − 1 V = V0 [cosh (μX ) − 1] p [32]

Let us collect all of the already-found autonomous ODEs that correspond to a scalar field
model of DE with arbitrary self-interaction potential:
3γm 3 2
(
x′ = −3x 1 − x 2 +
2
) (
x 1 − x 2 − y2 − ) 2
y z,

⎡ 3γm ⎤ 3
y′ = y ⎢
⎣ 2
(
1 − x 2 − y 2 + 3x 2⎥ +
⎦
) 2
xyz ,

z′ = 6 xf (z ), (30)
where it is evident that the equations for x and y are independent of the self-interaction
potential, and that details of the given model are encoded in the function f (z) (equation (27)),
which depends on the concrete form of the potential (see table 1).

5.1. Critical points

If we apply the simplest tools of dynamical systems theory to the the above example with the
cosh potential, the corresponding asymptotic dynamics in the phase space is revealed. Given
that, at the minimum of the cosh potential, the model behaves as ΛCDM, here we shall
assume that the matter fluid is composed mainly of baryons so that it behaves like pressureless
dust (γm = 1).
The following closed system of autonomous ODEs is obtained (it is simply equation (30)
with the appropriate substitutions):

3 3 2
x′ = −
2
(
x 1 − x 2 + y2 − ) 2
y z,

3 3
y′ =
2
(
y 1 + x 2 − y2 + ) 2
xyz ,

z′ = 6x(μ 2
− z2 .) (31)

Since z = μ tanh (μX ) ⇒ −μ ⩽ z ⩽ μ, the phase space where we look for critical points of
(31) is the bounded semi-cylinder (see equation (19)):

{
Ψcosh := (x , y , z ): 0 ⩽ x 2 + y 2 ⩽ 1,
x ⩽ 1, y ⩾ 0, z < μ} . (32)
Nine critical points, Pi : (xi , yi , zi ), of the autonomous system of the ODE (31) and one
critical manifold are found in Ψcosh . These, together with their main properties, are shown in

14
 Eur. J. Phys. 36 (2015) 025008
Table 2. Critical points of the autonomous system of ODE (31) and their properties.

Critical points x y z Existence ΩX Ωm λ1 λ2 λ3

 0 0 z Always 0 1 32 −3 2 0
PdS 0 1 0 ″ 1 0 3
−2 +
3
1−
4μ2 3
−2 −
3
1−
4μ2 −3
2 3 2 3
15

PK± ±1 0 μ ″ 1 0 3± 3
μ ∓2 6 μ 3
2
±
P̄K ±1 0 −μ ″ 1 0 3∓ 3
μ ±2 6 μ 3
2
3 1 3 1 3 3 3 3 3 24 3 3 24
PX± m ± ∓μ μ2 ⩾ 1− −4 + −7 −4 − −7 6
2 μ 2 μ 2 μ2 μ2 4 μ2 4 μ2
μ μ2
PX± ± 1−
μ2 ∓μ μ2 ⩽ 6 1 0 − 3 + μ2 −3 + 2 2μ2
6 6

R García-Salcedo et al
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

table 2. For conciseness, here we will concentrate only on those that have some particular
interest.
The de Sitter critical point, PdS : (0, 1, 0), for which Ω X = 1 (Ωm = 0 ), is associated with
accelerating expansion, since the deceleration parameter
H˙ 1 3 2
q = −1 − 2 = +
H 2 2
(
x − y 2 = −1, ) (33)

is a negative quantity. Since z = 0 ⇒ V = V0 , and since x = 0 ⇒ X˙ = 0 , and

V0
y=1 ⇒ H= ⇒ a (t ) ∝ eHt ,
3
this point corresponds to a de Sitter (inflationary) phase of the cosmic expansion. We
underline the fact that to a given point in the phase space, in this case PdS ∈ Ψcosh, it
corresponds to specific cosmological dynamics (de Sitter expansion in the present case).
Next, we have the scalar field X-dominated equilibrium points
⎛ ⎞
μ μ2 μ2
PX±: ⎜⎜ ± , 1− , ∓μ⎟⎟ , Ω X = 1, q = −1 + .
⎝ 6 6 ⎠ 2

The first thing one has to pay attention to is the existence of the critical point. By existence,
we mean that the given point actually belongs in the phase space (Ψcosh in the present case).
Hence, ∣ x ∣ ⩽ 1 ⇒ μ2 ⩽ 6. This is precisely the condition for the existence of the points PX±:
μ2 ⩽ 6. In addition, this warrants that y = 1 − μ2 6 be a real number. If one substitutes
(17) back into (15), one gets
x 2 − y2
ωX = ,
x 2 + y2
so that for the present case, the EOS parameter of the X-field is
ω X = μ2 3 − 1 ⇒ γX = μ2 3. Then the continuity equation (1) can be readily integrated at
2
PX± to obtain ρX = M a−3γX = M a−μ , where M is an integration constant. After this, the
Friedmann constraint
a˙ M −μ2
Ω X = 1 ⇒ 3H 2 = ρ X ⇒ = a ,
a 3
2
can also be integrated to obtain the following cosmological dynamics a (t ) ∝ t 2 μ , which is to
be associated with the critical points, PX±. Note that whenever 2 < μ2 ⩽ 6 (q > 0 ), the present
scalar-field-dominated solution represents decelerated expansion.
In the case of the matter-dominated critical manifold18
 = {(0, 0, z ): z ∈ R}, Ω m = 1, q = 1 2,
instead of an isolated equilibrium point, this is actually a one-dimensional manifold, which is
extended along the z-direction. This means, in turn, that the matter-dominated solution,
Ωm = 1 ⇒
3H 2 = ρm = Ma−3 ⇒ a (t ) ∝ t 2 3,
may coexist with the scalar field domination no matter whether V ∝ cosh (μX ) → intermediate
X, V = V0 → X = 0, or V ∝ e±μX → very large X (∣ X ∣ → ∞). This means that in the phase
18
Here, by critical manifold we understand a curve in the phase space, all of whose points are critical points.

16
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

portrait, a heteroclinic orbit (an orbit connecting two or more different equilibrium points) can
be found that joins the matter-dominated equilibrium state with a scalar field domination critical
point, originated either by a cosmological constant or by exponential or cosh-like potentials.

5.2. Stability

To judge the stability of given hyperbolic equilibrium points (as explained in the appendix)
one needs to linearize the system of ODEs (in this case (31)) which means, in the end, that we
have to find the eigenvalues of the linearization or Jacobian matrix, J (see equation (A.2) in
the appendix)

⎛ ∂x′ ∂x ∂x′ ∂y ∂x′ ∂z ⎞

⎜ ⎟
J = ⎜ ∂y′ ∂x ∂y′ ∂y ∂y′ ∂z ⎟ .
⎜ ⎟
⎝ ∂z′ ∂x ∂z′ ∂y ∂z′ ∂z ⎠

Next, J should be evaluated at each non-hyperbolic critical point, such as J (PdS ), J (PX ), etc.
The eigenvalues of the resulting numerical matrices are what we use to judge the stability of
the given points.
Take, for instance, the de Sitter point, PdS : (0, 1, 0). In this case, one must find the roots
of the following algebraic equation for the unknown λ :

⎛ −3 − λ 0 3 2 μ⎞
⎜ ⎟
det ⎜ 0 −3 − λ 0 ⎟ = 0.
⎜ ⎟
⎝ − 6μ 0 −λ ⎠

The resulting eigenvalues are

3 3 4
λ1,2 = − ± 1 − μ2, λ3 = −3.
2 2 3

For μ2 ⩽ 3 4, the critical point is an isolated focus, while for μ2 > 3 4, since the eigenvalues
λ1 and λ 2 are complex numbers with negative real parts, the point PdS is a stable spiral. Hence,
the de Sitter solution is always a future attractor (see the bottom panel of figure 1). This is
consistent with the fact that the cosh potential is a minimum at X = 0 ⇒ V = V0 .
The eigenvalues of the linearization matrix, J (PX±), are

λ1 = −3 + μ2 2, λ 2 = −3 + μ2 , λ3 = μ2 .

Hence, as required by the existence of μ2 < 6, PX± are always saddle critical points (at least
one of the eigenvalues is of a different sign).
If one computes the eigenvalues of J (), one finds λ1,2 = ±3 2, λ3 = 0. The vanishing
eigenvalue is due to the fact that the one-dimensional manifold extends along the z-direc-
tion19. The differing signs of the eigenvalues λ1 and λ 2 mean that each point in  is a saddle.
It might be a nice exercise for interested readers to judge the stability of the remaining critical
points of (31) (see table 2).
19
In general, a critical point for which at least one of the real parts of the eigenvalues of the Jacobian matrix is
vanishing is called a non-hyperbolic critical point. In this case, the Hartman−Grobman theorem [33] cannot be
applied. As a consequence, one must go beyond the linear stability theory. The centre manifold theory is useful in this
case. For the application of the centre manifold theory in a case of cosmological interest, see [18, 19].

17
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

6. Conclusion

In this paper, we have shown in as pedagogical a way as possible how the application of the
basic tools of dynamical systems theory can help us understand the dynamics of cosmological
models of DE. Our main aim has been to clarify how much useful information about
asymptotic cosmological dynamics, which decides the origin and the fate of our Universe, can
be extracted by means of these simple tools. We have concentrated our exposition on scalar
field cosmological models of DE, and the analysis was kept as general as possible. Special
attention has been paid to the derivation of relevant equations and formulas to facilitate the
path for beginners to get to concrete computations.
It has been demonstrated that certain generic asymptotic behaviour arises that is quite
independent of the concrete model of DE: matter dominance transient phase and dark energy
dominance at late times. Just for illustration, we have chosen a concrete potential (the cosh
potential), which at late times approaches the ΛCDM model. The approach undertaken here
can be easily applied to other self-interaction potentials (see table 1).
In the present paper, for completeness, we have included a very concise and simple
exposition of the fundamentals of the theory of dynamical systems, found in the appendix,
which may be considered as material for an introductory course on this subject.

Acknowledgments

The authors thank Fernando R González-Díaz and Ricardo Medel-Esquivel for useful com-
ments on the original version of the manuscript, and we also thank the SNI of Mexico for
support. The work of RGS was partly supported by SIP20131811, SIP20140318, COFAA-
IPN, and EDI-IPN grants. The research of TG, FA, HR and IQ was partially supported by
Fondo Mixto 2012-03, GTO-2012-C03-194941. IQ was also supported by CONACyT of
Mexico. Last but not least, their authors are indebted to the anonymous referees for their
constructive criticism.

Appendix. Dynamical systems in cosmology

In classical mechanics, the state of a given physical system composed of, say, N particles, is
completely specified by the knowledge of the N generalized coordinates, qi (i = 1, … , N ),
and the N conjugated momenta, pi, which satisfy Hamiltonʼs canonical equations of motion
q˙i = ∂ ∂pi , p˙i = −∂ ∂qi ,

where  =  (q1, … , qN , p1 , … , pN , t ) is the Hamiltonian of the physical system. Hence, a

given state of this system corresponds to a point in the 2N -dimensional space of states
spanned by the 2N coordinates, q1, … , qN , p1 , …, pN , also known as the ‘phase space’.
As for any other physical system, the possible states of a cosmological system may also
be correlated with the points in an equivalent state space. However, unlike in the classical
mechanics case, the choice of the variables—generalized coordinates and their conjugate
momenta—for a cosmological model is not a trivial issue. In this case, a certain degree of
uncertainty in the choice of an appropriate set of variables of the phase space arises. There
are, however, certain unwritten rules one follows when choosing appropriate variables of the
phase space: (i) the variables should be dimensionless, and (ii) whenever possible, they
should be bounded. The latter requirement is necessary to have a bounded phase space where
all of the existing equilibrium points are ‘visible’ (i.e., none of them goes to infinity).

18
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

The interplay between a cosmological model and the corresponding phase space is
possible because of an existing one-to-one correspondence between exact solutions of the
cosmological field equation (1) and points in the phase space spanned by given variables, x,
y20:

( ) (
x = x H , ρx , ρ X , y = y H , ρx , ρ X . )
When we replace the original field variables H, ρx , and ρX by the phase space variables x, y,
we have to keep in mind that, at the same time, we trade the original set of nonlinear second-
order differential equations with respect to the cosmological time, t (equation (1)), for a set of
first-order ODEs):
x′ = f (x , y), y′ = g (x , y), (A.1)
where the tilde denotes a derivative with respect to the dimensionless ‘time’ parameter,
τ ≡ ∫ da a ⇒ dτ = dln a = H dt . The introduction of τ instead of the cosmic time t is
dictated by simplicity and compactness of writing. In addition, this warrants that the first-
order differential equation (A.1) are a closed system of equations.
The most important feature of the system of ODEs (A.1) is that the functions f (x, y ) and
g (x, y ) do not depend explicitly on the parameter τ . This is why (A.1) is called an autono-
mous system of ODEs. The images of the integral curves of (A.1) in the phase space are
called ‘orbits’ of the system of autonomous ODEs. The critical points of (A.1), which are also
fixed or stationary points, Pcr : (xcr , ycr ), are those for which the ‘velocity’ vector, v = (x′, y′),
vanishes (see below):
( )
v ( Pcr ) = x′ ( Pcr ), y′ ( Pcr ) = 0.
To judge the stability of given equilibrium points, one needs to linearly expand (A.1) in
the neighbourhood of each hyperbolic critical point:

( ) (
f x cr + δx , ycr + δy , g x cr + δx , ycr + δy . )
Depending on whether the linear perturbations δx = δx (τ ), δy = δy (τ ) decay (grow) with
time τ or decay in one direction while they grow in the other, the critical point can be a future
(past) attractor or a saddle point.
Knowledge of the critical, equilibrium, or fixed points in the phase space corresponding
to a given cosmological model is very important. In particular, the existence of attractors can
be correlated with generic cosmological solutions that might really decide the fate and/or the
origin of the cosmic evolution. The ‘meta-stable’ saddle equilibrium points—stationary points
that are not local extrema—can be associated with (no less important) transient cosmological
solutions. Hence, in the end we trade the study of the cosmological dynamics depicted by
H = H (t ), ρx = ρx (t ), ρX = ρX (t ), for the study of the properties of the equilibrium points of
the equivalent autonomous system of ODE (A.1) in the phase space, Ψ , of the cosmological
model, which, we recall, differs from the usual definition in classical mechanics.

A.1. Remarks on phase space analysis

In this subsection, we provide a very simplified exposition of the fundamentals of dynamical

systems theory, which are useful in most cosmological applications [5, 11, 12]. Only
knowledge of elementary linear algebra and of ODEs is required to understand this material.
20
Here, for simplicity of the exposition, we consider a two-dimensional phase space.

19
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

For a more formal and complete introduction to this subject, we recommend the well-known
texts [8, 10, 34–37].
The critical points of the system of ODE (A.1) Pi : (xi ; yi ) (i.e., the roots of the system of
algebraic equations)
f (x , y) = 0, g (x , y) = 0,
correspond to privileged or generic solutions of the original system of cosmological
equations. To judge their stability properties, it is first necessary to linearize (A.1) around the
hyperbolic equilibrium points21. To linearize around a given hyperbolic equilibrium point, Pi,
amounts to considering small linear perturbations
x → xi + δx (τ ), y → yi + δy (τ ).
These perturbations would obey the following system of coupled ODE (here we use matrix
notation):
⎛ ∂f ∂f ⎞
⎜ ⎟
⎛ δx ⎞ ∂x ∂y ⎟
δ x′ = J ( Pi ) · δ x , δ x = ⎜ ⎟ , J=⎜ , (A.2)
⎝ δy ⎠ ⎜ ∂g ∂g ⎟
⎜ ⎟
⎝ ∂x ∂y ⎠
where J, the Jacobian (also linearization) matrix, is to be evaluated at Pi. Thanks to the
Hartman−Grobman theorem [33], which basically states that the behaviour of a dynamical
system in the neighbourhood of each hyperbolic equilibrium point is qualitatively the same as
the behaviour of its linearization, we can safely replace the study of the dynamics of (A.1)
with the corresponding study of its linearization (A.2).
We assume that J can be diagonalized (i.e., JD = M −1JM ), where M is the diag-
onalization matrix and
⎛ λ1 0 ⎞
JD = ⎜ ⎟
⎝ 0 λ2 ⎠
is the diagonal matrix whose non-vanishing components are the eigenvalues of the Jacobian
matrix, J:

det J − λU = 0, U = ( )
1 0 .
0 1
(A.3)

After diagonalization, the coupled system of ODEs (A.2) is decoupled:

δ x¯ ′ = JD · δ x¯ , δ x¯ = M −1δ x , (A.4)
where we have to recall that a different matrix, JD, corresponds to each equilibrium point. The
decoupled system of ODEs (A.4) is easily integrated:
δx¯ (τ ) = δx¯ (0)e λ1 τ , δy¯ (τ ) = δy¯ (0)e λ2 τ .
Since the diagonal perturbations, δx̄ and δ, y¯ are linear combinations of the perturbations δx ,
δy :
δx¯ = c11 δx + c12 δy , δy¯ = c21 δx + c22 δy ,

21
Ph is a hyperbolic critical point if the real parts of all of the eigenvalues of the linearization matrix around Ph are
necessarily non-vanishing. In particular, a hyperbolic point cannot be a centre. In more technical words, a hyperbolic
critical point is a fixed point that does not have any centre manifolds.

20
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

where the constants cij are the components of the matrix M −1, then
δx (τ ) = c¯11 e λ1 τ + c¯12 e λ2 τ ,
δy (τ ) = c¯21 e λ1 τ + c¯22 e λ2 τ , (A.5)
where
c22 δx¯ (0) c12 δy¯ (0)
c¯11 = , c¯12 = − ,
c22 c11 − c12 c21 c22 c11 − c12 c21
c21 δx¯ (0) c11 δy¯ (0)
c¯21 = − , c¯22 = .
c22 c11 − c12 c21 c22 c11 − c12 c21
As a matter of fact, we do not need to compute the coefficients c̄ij ; instead, the structure
of the eigenvalues, λi , is the only thing we need to judge regarding the stability of given
(hyperbolic) equilibrium points of (A.1). If the eigenvalues are complex numbers,
λ± = ν ± iω, the perturbations (A.5) do oscillations with frequency ω. If the real part, ν , is
positive, the oscillations are enhanced, while if ν < 0 , the oscillations are damped. The case
ν = 0 is associated with a centre (harmonic oscillations) and is not frequently encountered in
cosmological applications. This latter kind of point in the phase space—an eigenvalue with
vanishing real part—is called a non-hyperbolic critical point.

A.2. Taxonomy of isolated equilibrium points

Here we summarize the basic classification of isolated equilibrium points in the phase
plane. For simplicity, we assume that λ1 ⩽ λ 2 , but nothing changes if one assumes
that λ1 ⩾ λ 2 , or if one makes no assumption at all. In figure A1 , the illustrative phase
portraits—the drawing of the trajectories of a dynamical system in the phase plane—are
shown.

1. The eigenvalues λ1, λ 2 in equation (A.5) are real numbers.

(a) 0 < λ1 < λ 2 . The critical point, Pi, is an unstable node or, also, a source point (past
attractor). This is an unstable critical point which represents the origin of a non-empty
set of orbits in the phase plane (left-hand top panel of figure A1).
(b) λ1 < 0 < λ 2 . Pi is a saddle point. This is an unstable equilibrium point, but it can be
associated with a marginally stable state since the orbits of the system of autonomous
ODEs (A.1) spend some ‘time’ in the neighbourhood, N (ϵ, Pi ), of the critical point
until these leave N (ϵ, Pi ) to go elsewhere in the phase plane (centre top panel of
figure A1).
(c) λ1 < λ 2 < 0 (also λ1 = λ 2 < 0 ). Pi is a stable node (future attractor). This kind of
equilibrium point is associated with an asymptotically stable state. The future attractor
is the end point of a non-empty set of phase space orbits generated by a wide range of
initial conditions (right-hand top panel of figure A1).
2. The eigenvalues λ1, λ 2 in equation (A.5) are complex numbers: λ1,2 = ν ± iω (ω ≠ 0 ).
(a) ν > 0 . The equilibrium point, Pi, is an unstable spiral (past attractor). It is associated
with unstable oscillation, (i.e., with amplified oscillations, as in the left-hand bottom
panel in figure A1).
(b) ν = 0 . Pi is a centre (not frequent). It is associated with free oscillations (centre
bottom panel in figure A1).
(c) ν < 0 . Pi is a stable spiral (future attractor). The system does damped oscillations
until it settles down in the equilibrium state (right-hand bottom panel in figure A1).

21
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

Figure A1. Phase portrait: isolated equilibrium points in the phase space when the
eigenvalues λ1 and λ 2 are real numbers (top panels). The velocity field, represented by
the arrow fields in the figures, shows where the flux of the ODE (A.1) flows. Several
orbits generated by different sets of initial conditions [x (0) = x 0i ; y (0) = y0i ] are
shown in each case. In the left-hand top panel, the origin P: (0; 0) is an unstable node
or past attractor (λ 2 > λ1 > 0 ), while in the centre top panel the origin is a saddle
critical point (λ1 < 0 < λ 2 ), and in the right-hand top panel, P: (0; 0) is a stable node or
future attractor (λ1 < λ 2 < 0 ). In the bottom panels the equilibrium points for cases
when the eigenvalues are complex numbers [λ1,2 = ν ± iω ] are shown. In the left-hand
bottom panel, the origin (x; y ) = (0; 0) is a unstable spiral (ν > 0 ), while in the centre
panel, it is a centre (ν = 0 ). The point (0; 0) is a stable spiral in the right-hand bottom
panel (ν < 0 ).

The above classification of isolated critical points encompasses the kinds of equilibrium
points most frequently encountered in cosmological applications. Any curve in the phase
space that connects critical points (two or more, perhaps all of them) is called a ‘heteroclinic
orbit.’ A curve that joins a critical point with itself is called a ‘homoclinic trajectory’. There
are other useful concepts that are not explained here. The interested reader is referred to the
bibliography [8, 10, 34–37].

References

[1] Sonego S and Talamini V 2012 Am. J. Phys. 80 670–9

[2] Uzan J P and Lehoucq R 2001 Eur. J. Phys. 22 371–84
[3] Jordan T 2005 Am. J. Phys. 73 653–62
Visser M 2005 Gen. Rel. Grav. 37 1541–8
Akridge R 2001 Am. J. Phys. 62 195–200
Lemons D S 1988 Am. J. Phys. 56 502–4
[4] Liddle A 2003 An Introduction to Modern Cosmology 2nd edn (New York: Wiley)

22
Eur. J. Phys. 36 (2015) 025008 R García-Salcedo et al

[5] Copeland E J, Sami M and Tsujikawa S 2006 Int. J. Mod. Phys. D 15 1753–936
[6] Faraoni V 1999 Am. J. Phys. 67 732
[7] Kellert S H 1993 In the Wake of Chaos: Unpredictable Order in Dynamical Systems (Chicago:
University of Chicago Press)
[8] Hirsch M W, Smale S and Devaney R L 2004 Differential Equations, Dynamical Systems, and an
Introduction to Chaos (New York: Elsevier)
[9] Mandelbrot B B 2004 Fractals and Chaos (Berlin: Springer)
[10] Wainwright J and Ellis G F R 1997 Dynamical Systems in Cosmology (Cambridge: Cambridge
University Press)
[11] Copeland E J, Liddle A R and Wands D 1998 Phys. Rev. D 57 4686
Copeland E J, Mizuno S and Shaeri M 2009 Phys. Rev. D 79 103515
[12] Coley A A arXiv:gr-qc/9910074
[13] Amendola L 2000 Phys. Rev. D 62 043511
[14] Ng S C C, Nunes N J and Rosati F 2001 Phys. Rev. D 64 083510
[15] Matos T, Luevano J R, Quiros I, Urena-Lopez L A and Vazquez J A 2009 Phys. Rev. D 80 123521
[16] Urena-Lopez L A 2012 J. Cosmol. Astropart. Phys. JCAP03(2012)035
[17] Garcia-Salcedo R, Gonzalez T, Quiros I and Thompson-Montero M 2013 Phys. Rev. D 88 043008
Avelino A, Garcia-Salcedo R, Gonzalez T, Nucamendi U and Quiros I 2013 J. Cosmol. Astropart.
Phys. JCAP08(2013)012
[18] Boehmer C G, Chan N and Lazkoz R 2012 Phys. Lett. B 714 11–17
[19] Boehmer C G and Chan N arXiv:1409.5585
[20] Nunes A and Mimoso J P 2000 Phys. Lett. B 488 423–7
[21] Barreiro T, Copeland E J and Nunes N J 2000 Phys. Rev. D 61 127301
[22] Linder E V 2008 Am. J. Phys. 76 197–204
[23] Scherrer R J and Sen A A 2008 Phys. Rev. D 77 083515
Fang W, Li Y, Zhang K and Lu H Q 2009 Class. Quantum Grav. 26 155005
[24] Riess A G et al (Supernova Search Team Collaboration) 1998 Astron. J. 116 1009–38
Peebles P J E and Ratra B 2003 Rev. Mod. Phys. 75 559–606
[25] Weinberg S 1989 Rev. Mod. Phys. 61 1–23
[26] Adler R J, Casey B and Jacob O C 1995 Am. J. Phys. 63 620–6
[27] Peebles P J E and Ratra B 2003 Rev. Mod. Phys. 75 559
Nojiri S, Frieman J, Turner M and Huterer D 2008 Ann. Rev. Astron. Astrophys. 46 385
[28] Dungan R and Prosper H B 2011 Am. J. Phys. 79 57–62
[29] Carroll S M 1998 Phys. Rev. Lett. 81 3067–70
Zlatev I, Wang L-M and Steinhardt P J 1999 Phys. Rev. Lett. 82 896–9
Amendola L 2000 Phys. Rev. D 62 043511
[30] Steinhardt P J, Wang L-M and Zlatev I 1999 Phys. Rev. D 59 123504
[31] Sahni V and Starobinsky A 2000 Int. J. Mod. Phys. D 9 373
Urena-Lopez L A and Matos T 2000 Phys. Rev. D 62 081302(R)
Pavluchenko S A 2003 Phys. Rev. D 67 103518
[32] Sahni V and Wang L-M 2000 Phys. Rev. D 62 103517
[33] Hartman P 1960 Proc. Am. Math. Soc. 11 610–20
Grobman D M 1962 Mat. Sb. (N.S.) 56 77–94
[34] Brauer F 1969 The Qualitative Theory of Ordinary Differential Equations (New York: Benjamin)
[35] Arrowsmith D K and Place C M 1990 Introduction to Dynamical Systems (Cambridge: Cambridge
University Press)
[36] Perko L 2001 Differential Equations and Dynamical Systems (USA: Springer-Verlag)
[37] Arnold V I 1973 Ordinary Differential Equations ed R A Silverman (Cambridge, MA: MIT Press)
(translated from Russian)

23