Dynamical systems in cosmology

Christian G. Böhmer∗1 and Nyein Chan†2,3

1
Department of Mathematics, University College London
Gower Street, London, WC1E 6BT, UK
2
Faculty of Engineering, Computing and Science
arXiv:1409.5585v2 [gr-qc] 13 Oct 2014

Swinburne University of Technology, Sarawak Campus

Jalan Simpang Tiga, 93350 Kuching, Sarawak, Malaysia
3
International School Yangon
Shwe Taungyar Street, Yangon, Myanmar

Abstract
Cosmology is a well established research area in physics while dy-
namical systems are well established in mathematics. It turns out that
dynamical system techniques are very well suited to study many aspects
of cosmology. The aim of this book chapter is to provide the reader with a
concise introduction to both cosmology and dynamical system. The ma-
terial is self-contained with references to more detailed work. It is aimed
at applied mathematics and theoretical physics graduate level students
who have an interest in this exciting topic.

These lecture notes are based on the thesis of NC and on lectures given by CGB
at the London Taught Course Centre (LTCC).

∗ [email protected] (corresponding author)

† [email protected]

1
Contents
1 A brief introduction to cosmology 3
1.1 The basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Cosmological solutions . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 A very brief history of the universe . . . . . . . . . . . . . . . . . 5
1.4 A first taste of dynamical systems . . . . . . . . . . . . . . . . . 6

2 Some aspects of dynamical systems 7

2.1 Linear stability theory . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Lyapunov functions . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Centre manifold theory . . . . . . . . . . . . . . . . . . . . . . . 16

3 Cosmology using dynamical systems 19

3.1 Cosmology with matter and scalar field . . . . . . . . . . . . . . 20
3.2 Cosmology with matter and scalar field and interactions . . . . . 25

4 Final remarks 27

2
1 A brief introduction to cosmology
1.1 The basics
Cosmology is the study of the universe as a whole, and its aim is to understand
the origin of the universe and its evolution. The study of the cosmos is as old as
humanity and has always been fascinating. Physical cosmology1 is the scientific
study of the universe as a whole based on the laws of physics. The dominant
interaction between macroscopic objects is the gravitational force. Therefore,
we must study the dynamics of the universe within the framework of Einstein’s
theory of General Relativity which was formulated in 1916. In simple terms,
the main concept of general relativity is the following equation

geometry = κ × matter (1)

where κ = 8πG/c4 is a coupling constant which determines the strength of the

gravitational force. G is Newton’s gravitational constant and c is the speed of
light.
The Einstein field equations are a set of 10 coupled non-linear PDEs, or in
other words, very difficult equations to deal with in general [1]. However, these
equations can be simplified considerably by making some suitable assumptions.
In cosmology [2] this is known as the cosmological principle. It is an axiom
which states that the universe is homogeneous and isotropic when viewed over
large enough scales.
These scales are of the order of 100 − 1000 MPc. To translate this into more
practical units, we note that 1 pc ≈ 3.26 ly ≈ 1012 km. This means 100 MPc ≈
326 Mly ≈ 1018 km, and has the simple practical implication that we cannot
test the cosmological principle directly by making observations at two points
in the universe separated by cosmologically significant scales. However, there
are other possibilities of testing the cosmological principle. For instance, if
we were to observe a very large structure in the universe which is bigger than
100 MPc, say, then this would force us to revise this number upwards. It turns
out that such very large structure have already been observed, see the Clowes–
Campusano Large Quasar Group for one such example.
Henceforth we assume that the cosmological principle is valid for some suit-
able length scale. A homogeneous and isotropic 4-dimensional Lorentzian man-
ifold is characterised by only one function which is usually denoted by a(t) and
one constant k = (±1, 0). Such models were studied independently by Fried-
mann, Lemaı̂tre, and Robertson & Walker. The function a(t) is called the scale
factor and is the only dynamical degree of freedom in the cosmological Einstein
field equations. The constant k characterises the curvature of the so-called con-
stant time hypersurfaces, k = 0 corresponds to a Euclidean space, k = +1 to
a 3-sphere and k = −1 to hyperbolic space. The cosmological Einstein field
equations are given by
ȧ2 k
3 + 3 2 − Λ = κρ (2a)
a2 a
ä ȧ2 k
−2 − 2 − 2 + Λ = κ p. (2b)
a a a
1 We will drop the word physical soon. It is used here to emphasise the scientific aspect of

cosmology opposed to the philosophical or religious studies.

3
Here Λ is the so-called cosmological constant, ρ and p are the energy density
and pressure of some matter components, respectively. This matter could be
a perfect fluid with prescribed equation of state, or a scalar field for instance.
More complicated forms of matter can also be included. One can verify by direct
calculation that these two equations imply the energy-conservation equation
ȧ
ρ̇ + 3 (ρ + p) = 0. (3)
a
In cosmology one assumes that every matter component satisfies its own con-
servation equation, which does not follow from the field equations but must be
assumed or derived separately. Inspection of Eqs. (2) shows that we have two
equations but 3 functions to be found, namely a(t), ρ(t) and p(t). This system
of equations is under-determined. In order to close it, we will assume a linear
equation of state between the pressure and the energy density p = wρ, where
the equation of state parameter w ∈ (−1, 1].
The scale factor a(t) is a measure of the size of the universe at time t.
However, since we do not have an absolute length scale, the numerical value of
a(t) can be rescaled. One convention is to choose a(ttoday ) = 1 and compare
the universe’s size with its current value. Moreover, it turns out to be useful to
introduce the Hubble function H(t) := ȧ/a which is a measure of the universe’s
expansion rate at time t. A positive value for this quantity was first observed by
Edwin Hubble in 1929, thereby giving experimental evidence to an expanding
universe. Today’s value Htoday is of the order of 70km/s/Mpc.
Let us now rewrite the field equations using the Hubble parameter. Firstly,
we need the relation
ä ȧ2 ä
Ḣ = − 2 = − H2 (4)
a a a
which allows us to write (2) in the following form
k
3H 2 + 3 − Λ = κρ (5a)
a2
k
−2Ḣ − 3H 2 − 2 + Λ = κ p. (5b)
a
Equation (5a) is of particular interest to us. By dividing the entire equation by
3H 2 we arrive at
κρ Λ k
1= + − 2 2 (6)
3H 2 3H 2 a H
and observe that each of the three terms is dimensionless.
It is common to introduce the following dimensionless density parameters
κρ Λ
Ω= , ΩΛ = . (7)
3H 2 3H 2
Note that Ω may contain different forms of matter, the total matter content
might contain a pressure-less perfect fluid (standard matter or sometimes called
dust) and radiation, in which case one would write Ω = Ωm + Ωr . Before getting
started with dynamical systems and their application to cosmology, we need to
discuss some of the well known solutions in cosmology.

4
1.2 Cosmological solutions
We will now discuss the most important solutions of the field equations (2).
This is needed in order to understand and interpret the solutions encountered
later using dynamical systems techniques.
In order to simplify the equations, we will assume that the spatial curvature
parameter vanishes, i.e. k = 0 and we will also neglect the cosmological term
Λ = 0. Let us firstly assume that the equation of state parameter w = 0. This
corresponds to a matter dominated universe. One can immediately integrate
the conservation equation (3) and find that
ρ ∝ a−3 . (8)
This result is not unexpected since we find that density is inversely proportional
to volume. Using this result in the field equation (2a) yields the solutions
a(t) ∝ t2/3 .
Secondly, we consider w = 1/3 which corresponds to a radiation dominated
universe. In that case, the conservation equation gives
ρ ∝ a−4 . (9)
and the remaining field equations can be solved to find a(t) ∝ t1/2 .
Lastly, we consider the case where ρ = p = 0, however, we assume Λ > 0.
Then, we can integrate (5a) and find
p 
a(t) ∝ exp Λ/3 t . (10)

This solution is generally called the de Sitter solution and corresponds to a

universe which undergoes an accelerated expansion.

1.3 A very brief history of the universe

Based on a variety of observations, the evolution of the universe can be re-
constructed fairly accurately. We are currently living in a matter dominated
universe w = 0, and there is strong evidence for the presence of a positive
cosmological constant Λ > 0. Moreover, the spatial curvature of the universe
appears to be zero k = 0. There are some highly restrictive conditions in k 6= 0
models.
Since the universe is currently expanding, it must have been smaller and
denser in the past. From equations (8) and (9) we see that radiation decays
faster than matter in an expanding universe. Therefore, at some point in the
past, the universe was dominated by radiation. Going back in time further, the
universe was very dense and therefore hot and relatively small. The ‘beginning’
of the universe is often referred to as the big bang, giving the image of a vast
explosion from which the evolution of the universe started.
It appears very likely that the universe also underwent a period of accelerated
expansion at its very early stages, similar to the late time acceleration due to
the cosmological term. The reasons for this are beyond the scope of this short
introduction, however, we note that this epoch is called inflation.
Very roughly speaking, the standard model of cosmology can be summarised
by the succession of the following dominated eras
inflation −→ radiation −→ matter −→ cosmological term (11)

5
and a good cosmological model should be able to reproduce (parts of) this
pattern.

1.4 A first taste of dynamical systems

In order to get a first taste of the usefulness of dynamical systems techniques in
cosmology [3, 4], let us consider a universe which is spatially flat k = 0, and its
matter content is radiation ρr with w = 1/3, and a perfect fluid (dust) ρm with
w = 0. The following four equations completely determine the dynamics of the
system
3H 2 − Λ = κ (ρm + ρr ) (12a)
1
−2Ḣ − 3H 2 + Λ = κ ρr (12b)
3
ρ̇r + 4Hρr = 0 (12c)
ρ̇m + 3Hρm = 0. (12d)
Using the dimensionless density parameters Ωm , Ωr and ΩΛ , we find that equa-
tion (12a) becomes the constraint
1 = Ωm + Ωr + ΩΛ (13)
which means that we have two independent quantities, and choose to work with
Ωm and Ωr . Moreover, since we expect energy densities to be positive we also
have the conditions 0 ≤ Ωm ≤ 1 and 0 ≤ Ωr ≤ 1. Therefore, also ΩΛ ≤ 1 is
needed to satisfy equation (13).
The solution to the system (12) at any given time t will correspond to a
point in the (Ωm , Ωr ) plane. The constraint equation (13) together with the
aforementioned inequalities reduces the allowed (Ωm , Ωr ) plane to the triangle2
defined by ∆ = {(Ωm , Ωr ) | 0 ≤ Ωm + Ωr ≤ 1 ∩ 0 ≤ Ωm ≤ 1 ∩ 0 ≤ Ωr ≤ 1}, see
also Figure 1.
Next, we wish to find the dynamical equations for the dimensionless vari-
ables Ωm and Ωr . This requires a slightly lengthy but otherwise straightforward
calculation of which we will show some details. We start with
!
d d  κ ρm  κ ρ̇m H 2 − ρm 2H Ḣ κ ρ̇m Ḣ
Ωm = = = − 2ρm 2 . (14)
dt dt 3H 2 3 H4 3H H H

From (12d) we get an expression for ρ̇m /H, while (12b) can be solved for Ḣ/H 2 .
This yields
 
1 d κ Λ κ ρr

Ωm = −3ρm + 3ρm 1 − + (15)
H dt 3H 2 3H 2 9H 2
= −3Ωm + 3Ωm (1 − ΩΛ + Ωr /3). (16)
The last step is to eliminate ΩΛ using (13) which gives the equation
1 d
Ωm = −3Ωm + 3Ωm (Ωm + Ωr + Ωr /3) (17)
H dt
= −3Ωm + 3Ωm (Ωm + 4Ωr /3) (18)
= Ωm (3Ωm + 4Ωr − 3). (19)
2 To the best of our knowledge this idea goes back to Nicola Tamanini.

6
Figure 1: This figure shows the triangle defined by {(Ωm , Ωr ) | 0 ≤ Ωm + Ωr ≤
1 ∩ 0 ≤ Ωm ≤ 1 ∩ 0 ≤ Ωr ≤ 1}. Every solution to the field equations (12)
corresponds to a trajectory inside this triangle, one calls this region the phase
space of the system.

We now note that

ȧ
d log(a) = dt = Hdt (20)
a
which means that by introducing the new independent variable N = log(a) and
denoting differentiation with respect to N by a prime, we finally arrive at
Ω′m = Ωm (3Ωm + 4Ωr − 3). (21)
Following a similar calculation one can find the corresponding equation for Ωr
which is given by
Ω′r = Ωr (3Ωm + 4Ωr − 4). (22)
For any set of initial conditions (Ωm (Ni ), Ωr (Ni )) with initial ‘time’ Ni in the
triangle ∆, the equations (21) and (22) will determine a trajectory which de-
scribes the dynamical behaviour of the cosmological model we are studying. It
should be noted that equations (21) and (22) do not depend explicitly on the
‘time’ parameter N , such a system is called an autonomous system of equations,
or a dynamical system. Equations of this type can be studied using particu-
lar methods developed for such systems. In the next Section we will give a
brief introduction to dynamical systems and the most common methods used
to analyse them.

2 Some aspects of dynamical systems

What is a dynamical system? It can be anything ranging from something as
simple as a single pendulum to something as complex as the human brain and

7
the entire universe itself. In general, a dynamical system can be thought of as
any abstract system consisting of
1. a space (state space or phase space), and
2. a mathematical rule describing the evolution of any point in that space.
The second point is crucial. Finding a mathematical rule which, for instance,
describes the evolution of information at any neuron in the human brain is
probably impossible. So, we need a mathematical rule as an input and finding
one might be very difficult indeed.
The state of the system we are interested in is described by a set of quantities
which are considered important about the system, and the state space is the
set of all possible values of these quantities. In the case of the pendulum, the
position of the mass and its momentum are natural quantities to specify the
state of the system. For more complicated systems like the universe as a whole,
the choice of good quantities is not at all obvious and it turns out to be useful
to choose convenient variables. It is possible to analyse the same dynamical
system with different sets of variables, either of which might be more suitable
to a particular question.
There are two main types of dynamical systems: The first are continuous
dynamical systems whose evolution is defined by a set of ordinary differential
equations (ODEs) and the other ones are called time-discrete dynamical systems
which are defined by a map or difference equations. In the context of cosmol-
ogy we are studying the Einstein field equations which for a homogeneous and
isotropic space result in a system of ODEs. Thus we are only interested in con-
tinuous dynamical systems and will not discuss time-discrete dynamical systems
in the remainder.
Let us denote x = (x1 , x2 , . . . , xn ) ∈ X to be an element of the state space
X ⊆ Rn . The standard form of a dynamical system is usually expressed as [5]
ẋ = f (x) (23)
where the function f : X → X and where the dot denotes differentiation with
respect to some suitable time parameter. We view the function f as a vector
field on Rn such that
f (x) = (f1 (x), · · · , fn (x)). (24)
The ODEs (23) define the vector fields of the system. At any point x ∈ X
and any particular time t, f (x) defines a vector field in Rn . When discussing
a particular solution to (23) this will often be denoted by ψ(t) to simplify the
notation. We restrict ourselves to systems which are are finite dimensional and
continuous. In fact, we will require the function f to be at least differentiable
in X.
Definition 1 (Critical point or fixed point). The autonomous equation ẋ = f (x)
is said to have a critical point or fixed point at x = x0 if and only if f (x0 ) = 0.
There is an easy way to justify this definition. Let us consider a one dimen-
sional mechanical system with force F . Newton’s equation for such a system
is
mẍ = F (x). (25)

8
Let us introduce a second variable p = mẋ such that the single second order
ODE (25) becomes a system of two first order equations

ẋ = p/m (26)
ṗ = F (x). (27)

Therefore, according to Definition 1, the critical points of system (27) corre-

spond to those points x where the force vanishes F (x) = 0. At these points,
there is no force acting on the particle and the system could, in principle, remain
in this (steady) state indefinitely.
This leads to the question of stability of a critical point or fixed point. The
following two definitions will clarify what is meant by stable and asymptotically
stable. In simple words a fixed point x0 of the system (23) is called stable if all
solutions x(t) starting near x0 stay close to it.
Definition 2 (Stable fixed point). Let x0 be a fixed point of system (23). It is
called stable if for every ε > 0 we can find a δ such that if ψ(t) is any solution
of (23) satisfying kψ(t0 ) − x0 k < δ, then the solution ψ(t) exists for all t ≥ t0
and it will satisfy kψ(t) − x0 k < ε for all t ≥ t0 .
The point is called asymptotically stable if it is stable and the solutions
approach the critical point for all nearby initial conditions.
Definition 3 (Asymptotically stable fixed point). Let x0 be a stable fixed
point of system (23). It is called asymptotically stable if there exists a number
δ such that if ψ(t) is any solution of (23) satisfying kψ(t0 ) − x0 k < δ, then
limt→∞ ψ(t) = x0 .
The main difference is simply that all trajectories near an asymptotically
stable fixed point will eventually reach that point while trajectories near a stable
point could for instance circle around that point. If the point is unstable then
solutions will move away from it.
We will not encounter fixed points which are stable but not asymptotically
stable when studying cosmological dynamical systems.
Having defined a concept of stability, we will now discuss methods which
can be used to analyse the stability properties of critical points.

2.1 Linear stability theory

The basic idea of linear stability theory can be explained neatly using the above
one dimensional mechanical system mẍ = F (x). Let us assume that there is
a point x0 where the force vanishes F (x0 ) = 0. Can we find the behaviour of
the particle near this point? We set x(t) = x0 + δx(t) and assume δx(t) to
be small. Then ẍ(t) = δx(t) ¨ and F (x) = F (x0 + δx) ≈ F (x0 ) + F ′ (x0 )δx +
′
. . . = F (x0 )δx + . . . (recall F (x0 ) = 0) so that Newton’s equations near the
critical point becomes mδx ¨ = F ′ (x0 )δx where F ′ (x0 ) is a constant. This is a
linear second order constant coefficient ODE, its auxiliary equation is simply
λ2 = F ′ (x0 )/m. Therefore, the sign of F ′ (x0 ) determines the stability properties
of the point x0 . If F ′ (x0 ) < 0 the solution involves trigonometric functions and
we would speak of a stable point, for F ′ (x0 ) > 0 the solution would involve
exponentials and we would refer to this point as unstable.

9
 Exactly the same ideas can be utilised when studying an arbitrary dynam-
ical system. Let ẋ = f (x) be a given dynamical system with fixed point at
x0 . We will now linearise the system around its critical point. Since f (x) =
(f1 (x), . . . , fn (x)), we can Taylor expand each fi (x1 , x2 , . . . , xn ) near x0
n n
X ∂fi 1 X ∂ 2 fi
fi (x) = fi (x0 ) + (x0 )yj + (x0 )yj yk + . . . (28)
j=1
∂xj 2! ∂xj ∂xk
j,k=1

where the vector y is defined by y = x−x0 . Note that in what follows we are only
interested in the first partial derivatives. Therefore, of particular importance is
the object ∂fi /∂xj which if interpreted as a matrix is the Jacobian matrix of
vector calculus of the vector valued function f . We define
 
∂f1 ∂f1
∂x1 . . . ∂x n
∂fi  . .. .. 
J= = .. . . 
 (29)
∂xj 
∂fn ∂fn
∂x1 . . . ∂x n

It is the eigenvalues of the Jacobian matrix J, evaluated at the critical points x0 ,

which contain the information about stability. In this context J is sometimes
referred to as the stability matrix of the system. As J is an n × n matrix,
it will have n, possibly complex, eigenvalues (counting repeated eigenvalues
accordingly). Recalling the example of the one dimensional mechanical system
at the beginning, it is clear that this approach might encounter problems if one
or more of the eigenvalues are zero. This motivates the following definition [5].
Definition 4 (Hyperbolic point). Let x = x0 ∈ X ⊂ Rn be a fixed point
(critical point) of the system ẋ = f (x). Then x0 is said to be hyperbolic if none
of the eigenvalues of the Jacobian matrix J(x0 ) have zero real part. Otherwise
the point is called non-hyperbolic.
Linear stability theory fails for non-hyperbolic points and other methods
have to be employed to study the stability properties.
Roughly speaking we are distinguishing three broad cases: If all eigenvalues
have negative real parts, then we can regard the point as stable. If at least one
eigenvalues has a positive real part, then the corresponding fixed point would
not be stable and correspond to a saddle point which attracts trajectories in
some directions but repels them along others. Lastly, all eigenvalues could have
a positive real part, in which case all trajectories would be repelled.
In more than 3 dimensions it becomes very difficult to classify all possible
critical points based on their eigenvalues. However, in dimensions 2 and 3 this
can be done. In the following we present all possible cases for two dimensional
autonomous systems.
Let us consider the two dimensional autonomous system given by

ẋ = f (x, y) (30a)
ẏ = g(x, y) (30b)

where f and g are (smooth) functions of x and y. We assume that there exits a
hyperbolic critical point at (x0 , y0 ) so that f (x0 , y0 ) = 0 and g(x0 , y0 ) = 0. The

10
Jacobian matrix of the system is given by
!
f,x f,y
J= (31)
g,x g,y

where the f,x means differentiation with respect to x. Its two eigenvalues λ1,2
are given by
1 1
q
λ1 = (f,x + g,y ) + (f,x − g,y )2 + 4f,y g,x (32a)
2 2
1 1
q
λ2 = (f,x + g,y ) − (f,x − g,y )2 + 4f,y g,x (32b)
2 2
and be evaluated at any fixed point (x0 , y0 ).
Table 1 contains all possible cases in order to understand the stability or
instability properties of the critical point (x0 , y0 ) based on the two eigenvalues
λ1 and λ2 .

Eigenvalues Description
λ1 < 0, λ2 < 0 the fixed point is asymptotically stable and trajec-
tories starting near that point will approach that
point limt→∞ (x(t), y(t)) = (x0 , y0 )
λ1 > 0, λ2 > 0 the fixed point is unstable and trajectories will
be repelled from the point limt→−∞ (x(t), y(t)) =
(x0 , y0 ). We can speak of (x0 , y0 ) as the past time
attractor
λ1 < 0, λ2 > 0 the fixed point is a saddle point. Some trajectories
will be repelled, others will be attracted
λ1 = 0, λ2 > 0 the point is unstable. The positive eigenvalues en-
sures that there is at least one unstable direction
λ1 = 0, λ2 < 0 linear stability theory fails to determine stability.
The point is non-hyperbolic and other methods are
needed to study the behaviour of trajectories near
that point
λ1 = α + iβ, λ2 = with α > 0 and β 6= 0 the fixed point is an unstable
α − iβ spiral
λ1 = α + iβ, λ2 = with α < 0 and β 6= 0 the fixed point is a stable
α − iβ spiral
λ1 = iβ , λ2 = −iβ solutions are oscillatory and the point is called a
centre. Note that a critical point being a centre is
not related to centre manifolds.

Table 1: Stability or instability properties of the critical point (x0 , y0 ) based on

the two eigenvalues λ1 and λ2 .

11
Example – Cosmology with matter, radiation and cosmo-
logical term
Recall the cosmological dynamical system (21) and (22) which will be our base
model henceforth. The equations read
Ω′m = Ωm (3Ωm + 4Ωr − 3) (33a)
Ω′r = Ωr (3Ωm + 4Ωr − 4) (33b)
1 = Ωm + Ωr + ΩΛ . (33c)
We can find the fixed points of this system by solving the simultaneous equa-
tions Ω′m = 0 and Ω′r = 0 for the pair (Ωm , Ωr ). We find three fixed points,
namely O = (0, 0), R = (0, 1) and M = (1, 0). As we use the relative energy
densities Ωi as our dynamical variables, it is easy to interpret those fixed points.
At R, the radiation dominates and normal matter is absent. Likewise, at M ,
the normal matter dominates while radiation is absent. The point O contains
neither radiation nor matter, and is therefore dominated by the cosmological
term because of (33c).
The Jacobian matrix of system (33) is computed straightforwardly. Evalu-
ated at the three fixed points, we find
! ! !
−3 0 1 0 3 4
J(O) = , J(R) = , J(M ) = , (34)
0 −4 3 4 0 −1
respectively. The corresponding eigenvalues of the stability matrix are given by
O: λ1 = −3, λ2 = −4 (35a)
R: λ1 = 1, λ2 = 4 (35b)
M: λ1 = −1, λ2 = 3 (35c)
which implies that that O is the only attractor of the system. Therefore, all
trajectories will eventually approach O. R is unstable, however, since both
eigenvalues are positive, we can think of R as the only past time attractor. This
means all trajectories will have ‘started’ at R. Lastly, M is a saddle point.
This means that some trajectories are attracted towards M but are eventually
repelled to move towards O. The phase space diagram Fig. 2 clearly shows these
features.
In the cosmological context this has the following interpretation. Consider
a spatially flat universe filled with normal matter and radiation, and with a
very small cosmological term3 . Such a universe will generically be dominated by
radiation at early times, then it will undergo a period where matter dominates its
energy contents. Eventually it will evolve to a state where the cosmological term
dominates. This result is in line with our expectation of a good cosmological
model, see (11).

2.2 Lyapunov functions

The following methods of studying the stability of a fixed point goes back to
Lyapunov. It is completely different to linear stability and can be applied di-
3 Ifthe cosmological term happens to be ‘large’ then matter will never dominate and one
obtains an almost direct transition from radiation to a state where the cosmological term
dominates.

12
 Figure 2: Phase space diagram of system (33).

rectly to the system in question. The main problem with this approach is that
one has to be able to guess the Lyapunov function since there is no systematic
way of doing so. Let us start by defining what a Lyapunov function is and its
relation to stability of an autonomous system of equations.
Definition 5 (Lyapunov function). Let ẋ = f (x) with x ∈ X ⊂ Rn be a
smooth autonomous system of equations with fixed point x0 . Let V : Rn → R
be a continuous function in a neighbourhood U of x0 , the V is called a Lyapunov
function for the point x0 if
1. V is differentiable in U \ {x0 }
2. V (x) > V (x0 )
3. V̇ ≤ 0 ∀x ∈ U \ {x0 }.
Note that the third requirement is the crucial one. It implies
d ∂V ∂V
V (x1 , x2 , . . . , xn ) = ẋ1 + . . . + ẋn
dt ∂x1 ∂xn
∂V ∂V
= f1 + . . . + fn ≤ 0 (36)
∂x1 ∂xn
which required repeated use of the chain rule and substitution of the autonomous
system equations to eliminate the terms ẋi for i = 1, . . . , n.

13
 One can conveniently write dV /dt using vector calculus notation

d
V (x1 , x2 , . . . , xn ) = grad V · ẋ = grad V · f (x). (37)
dt
Let us now state the main theorem which connects a Lyapunov function to the
stability of a fixed point of a dynamical system.
Theorem 1 (Lyapunov stability). Let x0 be a critical point of the system ẋ =
f (x), and let U be a domain containing x0 . If there exists a Lyapunov function
V (x) for which V̇ ≤ 0, then x0 is a stable fixed point. If there exists a Lyapunov
function V (x) for which V̇ < 0, then then x0 is a asymptotically stable fixed
point.
Furthermore, if kxk → ∞ and V (x) → ∞ for all x, then x0 is said to be
globally stable or globally asymptotically stable, respectively.
One can also find some instability results, see e.g. [6], which will also depend
on our ability to find a suitable Lyapunov function. However, we will not use
results along those lines since we are mainly concerned about the stability of
certain fixed points in the context of cosmology.
Should we be able to find a Lyapunov function satisfying the criteria of the
Lyapunov stability theorem, we could establish (asymptotic) stability without
any reference to a solution of the ODEs. However, just because we failed in
finding a Lyapunov function at a particular point does not necessarily imply
that such a point is unstable. Since there is no systematic way of constructing a
function, it is possible that we were simply not clever enough to find a Lyapunov
function for the critical point concerned.

A first example
This first example is taken from [5]. Suppose that a system is described by the
vector field

ẋ = y (38a)
2
ẏ = −x + ǫx y (38b)

which has one critical point at (x, y) = (0, 0). A candidate Lyapunov’s function
is given by

x2 + y 2
V (x, y) = , (39)
2
satisfying V (0, 0) = 0 and V (x, y) > 0 in the neighbourhood of the fixed point.
This function leads to

V̇ = grad V · (ẋ, ẏ) = ǫx2 y 2 (40)

from which we conclude that the point is globally asymptotically stable if ǫ < 0
since x2 y 2 is positive definite and thus V̇ < 0 in the neighbourhood of the
fixed point. It is important to emphasise, however, that ǫ > 0 does not imply
instability.

14
A second example
Let us consider the system

ẋ = −x3 + xy (41a)
2 2
ẏ = −y − 2x − x y (41b)

which has one fixed point at (x, y) = (0, 0). Computing the eigenvalues of the
Jacobian matrix at the fixed point yields λ1 = −1 and λ2 = 0. Therefore, we
cannot decide, based on linear stability theory, whether the origin is stable or
not. However, starting with the candidate Lyapunov function

V (x, y) = 2x2 + y 2 (42)

leads to

V̇ = −4x4 − 2y 2 − 2x2 y 2 . (43)

Therefore the point is globally asymptotically stable since all terms in V̇ are
negative definite and thus V̇ < 0 in the neighbourhood of the fixed point. This
example has been adapted from a similar one in [7].
Note that the phase space plot is in agreement with our conclusion of sta-
bility, see Fig. 3.

Figure 3: Phase space plot of the system (41).

15
2.3 Centre manifold theory
Centre manifold theory is a method that allows us to simplify dynamical systems
by reducing their dimensionality near fixed points with vanishing eigenvalues
of the Jacobian matrix. It is also central to other elegant concepts such as
bifurcations and the method of normal forms [8]. Here the essential basics of
centre manifold theory are discussed following [5] and [7].
Let us, as above, consider the dynamical system
ẋ = f (x) (44)
n
with x ∈ R and let us assume that it has a fixed point x0 . Near this point we
can linearise the system using (29). Denoting y = x − x0 , we can write (44)
ẏ = Jy (45)
where we emphasise that J is a constant coefficient n × n matrix. As such
it will have n eigenvalues which motivates the following. The space Rn is the
direct sum of three subspaces which are denoted by Es , Eu and Ec , where the
superscripts stand for stable, unstable and centre, respectively. The space Es is
spanned by the eigenvectors of J which have negative real part, Eu is spanned
by the eigenvectors of J which have positive real part, and Ec is spanned by the
eigenvectors of J which have zero real part. Linear stability theory is sufficient
to understand the dynamics of trajectories in Es and Eu . Centre manifold theory
will determine the dynamics of trajectories in Ec .
In the context of centre manifold theory it is useful to write our dynamical
system (44) in the form
ẋ = Ax + f (x, y) (46a)
ẏ = By + g(x, y), (46b)
where (x, y) ∈ Rc × Rs . Moreover, we assume
f (0, 0) = 0, ∇f (0, 0) = 0 (47a)
g(0, 0) = 0, ∇g(0, 0) = 0. (47b)
In the system (46), A is a c × c matrix having eigenvalues with zero real parts,
while B is an s × s matrix whose eigenvalues have negative real parts. Our
aim is to understand the centre manifold of this system in order to investigate
its dynamics. We have suppressed some regularity assumptions on f and g for
simplicity.
Definition 6 (Centre Manifold). A geometrical space is a centre manifold
for (46) if it can be locally represented as
W c (0) = {(x, y) ∈ Rc × Rs |y = h(x), |x| < δ, h(0) = 0, ∇h(0) = 0} (48)
for δ sufficiently small.
The conditions h(0) = 0 and ∇h(0) = 0 from the definition imply that the
space W c (0) is tangent to the eigenspace E c at the critical point (x, y) = (0, 0).
Centre manifold theory is based on three main theorems [5]. The first one
is about the existence of the centre manifold, the second one clarifies the issue
of stability of solution while the last one is about constructing the actual centre
manifold needed to investigate the stability. We will state those theorems but
will not state the proofs, the interested reader is referred to [7].

16
Theorem 2 (Existence). There exists a centre manifold for (46). The dynamics
of the system (46) restricted to the centre manifold is given by

u̇ = Au + f (u, h(u)) (49)

for u ∈ Rc sufficiently small.

Theorem 3 (Stability). Suppose the zero solution of (49) is stable (asymptoti-
cally stable or unstable). Then the zero solution of (49) is also stable (asymptot-
ically stable or unstable). Furthermore, if (x(t), y(t)) is also a solution of (49)
with (x(0), y(0)) sufficiently small, there exists a solution u(t) of (49) such that

x(t) = u(t) + O(e−γt ) (50a)

−γt
y(t) = h(u(t)) + O(e ) (50b)

as t → ∞, where γ > 0 is a constant.

We now know that the centre manifold exists, and we can establish the
stability or instability of a solution. However, our ability to do so depends
on the knowledge of the function h(x) in Definition 6. We will now derive a
differential equation for the function h(x).
Following Definition 6, we have that y = h(x). Let us differentiate this with
respect to time and apply the chain rule. This gives

ẏ = ∇h(x) · ẋ (51)

Since W c (0) is based on the dynamics generated by the system (46), we can sub-
stitute for ẋ the right-hand side of (46a) and for ẏ the right-hand side of (46b).
This yields

Bh(x) + g(x, h(x)) = ∇h(x) · [Ax + f (x, h(x))] (52)

where we also used that y = h(x). The latter equation can be re-arranged into
the quasilinear partial different equation

N (h(x)) := ∇h(x) [Ax + f (x, h(x))] − Bh(x) − g(x, h(x)) = 0 (53)

which must be satisfied by h(x) for it to be the centre manifold. In general, we

cannot find a solution to this equation. Even for relatively simple dynamical
systems it is often impossible to find an exact solution of this equation. It is
the third and last theorem which explain why not all is lost at this point.
Theorem 4 (Approximation). Let φ : Rc → Rs be a mapping with φ(0) =
∇φ(0) = 0 such that N (φ(x)) = O(|x|q ) as x → 0 for some q > 1. Then

|h(x) − φ(x)| = O(|x|q ) as x → 0. (54)

The main point of this theorem is that an approximate knowledge of the

centre manifold returns the same information about stability as the exact solu-
tion of equation (53). It turns out that finding an approximation for the centre
manifold is a fairly doable task in comparison to finding the exact solution. The
centre manifold machinery is best explained with a concrete example.

17
Example – a simple two-dimensional model
The following two dimensional example is taken from Wiggins [5]. We consider
the system

ẋ = x2 y − x5 (55a)
2
ẏ = −y + x . (55b)

The origin (x, y) = (0, 0) is a fixed point. The Jacobian matrix of the linearised
system about the origin has eigenvalues of 0 and −1. Since there is a zero eigen-
value, the point is non-hyperbolic and linear stability theory fails to determine
the nature of stability of this point.
By Theorem 2, there exists a centre manifold for the system (55) and it can
be represented locally as

W c (0) = {(x, y) ∈ R2 |y = h(x), |x| < δ, h(0) = Dh(0) = 0} (56)

for δ sufficiently small. Next, we need to compute W c (0). Here we can exploit
Theorem 4 which says that it suffices to approximate the centre manifold to
establish stability properties. Therefore, it is customary to assume an expansion
for h(x) of the form

h(x) = ax2 + bx3 + O(x4 ) (57)

where a and b are constants to be determined. This expression is then substi-

tuted into (53) with the aim of determining those constants.
In this example, the equations (55) yield

A=0 B = −1 (58a)
2 5
f (x, y) = x y − x (58b)
2
g(x, y) = x . (58c)

This, in addition to (57), is substituted into (53) and gives

N = (2ax + 3bx2 + · · · )(ax4 + bx5 − x5 + · · · )

+ ax2 + bx3 − x2 + · · · = 0. (59)

The coefficients of each power of x must be zero so that (59) holds. This provides
us with a set on linear equations in the constants a and b which is solved by

a = 1 b = 0, (60)

where all terms of order O(x4 ) have been ignored. Therefore, the centre manifold
is locally given by

h(x) = x2 + O(x4 ). (61)

Finally, following Theorem 2, the dynamics of the system restricted to the centre
manifold is obtained to be
ẋ = x4 + O(x5 ). (62)
We conclude that for x sufficiently small, x = 0 is unstable. Therefore, the
critical point (0, 0) is unstable. In Fig. 4 we show the phase space for this
system and also indicate the centre manifold.

18
Figure 4: Phase space plot of the system (55). The centre manifold is indicated
by a dashed line and was computed up to terms x13 . One sees very clearly how
the centre manifold attracts the trajectories and how they are repelled from the
origin (along the centre manifold) making this point unstable.

3 Cosmology using dynamical systems

We discussed some aspects of cosmology in the context of dynamical systems,
see [3, 4] for more details, and also [9], or [10] for anisotropic models. Based on
the series of cosmological epochs inflation → radiation → matter → cosmological term
of Section 1.3, which could be called a ‘minimal’ cosmological model, we will now
make links with dynamical systems. A very neat paper studying cosmological
models in the Lotka-Volterra framework is [11].
Let us now consider a generic ‘minimal’ cosmological model described by
an n × n system of autonomous equations. Should this model begin with an
inflationary period, then this should correspond to an early time attractor in the
dynamical system. All eigenvalues of the Jacobian matrix at this point should
be positive in order to ensure that all trajectories evolve away from this point,
this means λi > 0 for i = 1, . . . , n.
In an ideal model we would also have two saddle points (λj > 0, λk < 0 with
j + k = n) which correspond to a radiation dominated and matter dominated
universe, respectively. These epochs being saddle points makes sure that some
trajectories are attracted to these points, however, they will eventually be re-
pelled. In this case the universe will evolve through both epochs. Let us note
here that most models will only contain either matter or radiation, and thus we

19
would be satisfied if there was only one saddle point.
Lastly, we require a late-time attractor (λi < 0 for i = 1, . . . , n) where the
universe is undergoing an accelerated expansion which corresponds to the de Sit-
ter solution. We say the universe is approaching de Sitter space asymptotically.
This can be summarised as follows

inflation −→ radiation/matter −→ de Sitter

(63)
λi > 0 λj > 0, λk < 0 λi < 0

where, for simplicity, we neglected the possibility of some zero eigenvalues.

3.1 Cosmology with matter and scalar field

The cosmological constant Λ has strong observational support [12, 13], but
also leads to a variety of problems which are called the cosmological constant
problems, we refer the reader to [14, 15] and in particular [16]. These problems
can largely be avoided if the constant term Λ is replaced by a dynamically
evolving scalar field ϕ with some given potential V (ϕ). In this case one often
speaks of dark energy. In many models the potential V is assumed to be of
exponential form, V = V0 exp(−λκϕ).
Moreover, instead of writing the equation of state for the matter as p = wρ,
one often encounters a slightly different parametrisation which is given by

pγ = wγ ργ = (γ − 1)ργ (64)

where γ = 1 + wγ is a constant and 0 ≤ γ ≤ 2. Its value is 4/3 when there is

radiation, and is 1 for standard matter or dark matter in this context.
For this setup, the Einstein field equations are

κ2
 
1
H2 = ργ + ϕ̇2 + V (65a)
3 2
κ2
Ḣ = − (ργ + pγ + ϕ̇2 ). (65b)
2
We can interpret ρϕ = ϕ̇2 /2 + V as the energy density of the scalar field and
pϕ = ϕ̇2 /2 − V as its pressure. This also allows us to define an effective equation
of state for the field. The conservation equations for the matter and the scalar
field are given by

ρ̇γ = −3H(ργ + pγ ) (66a)

dV
ϕ̈ = −3H ϕ̇ − = −3H ϕ̇ + λκV (66b)
dϕ
where we used the exponential form of the potential. We follow the approach
outlined in Section 1.4 and rewrite (65) and (66) using more suitable variables.
As before, we start with dividing equation (65a) with H 2 which results in

κ 2 ργ κ2 ϕ̇2 κ2 V
1= 2
+ 2
+ . (67)
3H 6H 3H 2

20
Every term on the right-hand side is positive since V > 0 and ργ > 0, and it
turns out that the following the dimensionless variables [17, 18] are particularly
useful
κ2 ϕ̇2 κ2 V κ 2 ργ
x2 = , y2 = , s2 = (68)
6H 2 3H 2 3H 2
which transform (67) into

1 = x2 + y 2 + s2 . (69)

Therefore, we can choose x, y as two independent variables. This leads to

κ 2 ργ
1 ≥ 1 − x2 − y 2 = s2 = ≥0 (70)
3H 2
implying that 0 ≤ x2 + y 2 ≤ 1 which means that the physical phase space of
this model is contained within the unit circle.
We will introduce three more quantities which are useful in understanding
the physical properties at the fixed points. The dimensionless density parame-
ter (7) of the scalar field ϕ can be expressed in terms of the new variables and
is given by
κ 2 ρϕ
Ωϕ = = x2 + y 2 . (71)
3H 2
Moreover, we define the equation of state for the scalar field by
pϕ 2x2
γϕ = 1 + wϕ = 1 + = 2 . (72)
ρϕ x + y2
Lastly, we define the effective equation of state of the total system by
pγ + pϕ wγ ργ + ϕ̇2 /2 − V
weff = =
ργ + ρϕ ργ + ϕ̇2 /2 + V
= wγ (1 − x2 − y 2 ) + x2 − y 2 . (73)

Now, we are ready to derive a two dimensional dynamical system using the
variables x and y. As before, we will introduce a new ‘time’ variable N = log(a)
so that dN = Hdt, and denote differentiation with respect to N by a prime.
Let us begin by differentiating x with respect to time t
!
κ ϕ̈H − ϕ̇Ḣ κ ϕ̈ Ḣ
ẋ = √ = √ − ϕ̇ 2 . (74)
6 H2 6 H H

Substituting for ϕ̈ using (66b) and for Ḣ using (65b) we arrive at

κ2
 
κ V 2
ẋ = √ −3ϕ̇ + λκ + ϕ̇ (γργ + ϕ̇ ) . (75)
6 H 2H 2
Next, using the variables (68) and the condition (69) we get
" r #
3 2 3
x (1 − x2 − y 2 )γ + 2x2


ẋ = H −3x + λy + (76)
2 2

21
One can now introduce the new ‘time parameter’ N . Following similar steps,
the equation for y ′ can be derived. The final system is
r
3 2 3
′
λy + x 2x2 + γ(1 − x2 − y 2 )


x = −3x + (77a)
2 2
r
3 3
y ′ = −λ xy + y 2x2 + γ(1 − x2 − y 2 ) .


(77b)
2 2
The complete dynamics of this cosmological model is describe by the two equa-
tions (77).
We noted that the phase space of this system is contained in the unit circle.
Inspection of the dynamical equations shows that system (77) is invariant under
the transformation y 7→ −y and symmetric under time reversal t 7→ −t. This
implies that we can restrict our analysis on the upper half-disk with y > 0.
The lower half-disc of the phase space corresponds to the contracting universe
because H < 0 in this region.
The properties of the dynamical system (77) depend on the values of the
constants λ and γ. Amongst others, they will in particular affect the existence
and stability of the fixed points of the system, see [17]. This can be related to
the theory of bifurcations, something that has not been explored in cosmological
dynamical systems. The following Table 2 contains all critical points of the
system (77).

x y existence
O 0 0 ∀λ and γ
A+ 1 0 ∀λ and γ
A− -1 0 ∀λ and γ
√
B λ/ 6 [1 − λ2 /6]1/2 λ2 < 6
p
C 3/2γ/λ [3(2 − γ)γ/2λ2 ]1/2 λ2 > 3γ

Table 2: Critical point of the system (77).

Having found all the possible fixed points, we can now compute the eigen-
values and determine their stability which is summarised in Table 3, see [17].

Stability Ωϕ γϕ
O saddle point for 0 < γ < 2 0 Undefined
√
A+ unstable
√ node for λ < 6 and saddle point for 1 2
λ> 6
√
A− √ node for λ > − 6 and addle point for
unstable 1 2
λ<− 6
B stable node for λ2 < 3γ and saddle point for 1 λ2 /3
3γ < λ2 < 6
C stable node for 3γ < λ2 < 24γ 2 /(9γ − 2) and 3γ/λ2 γ
stable spiral for λ2 > 24γ 2 /(9γ − 2)

Table 3: Summary of the properties of the critical points.

22
 The three figures Fig. 5–7 show the phase spaces of this model for various
parameter choices.

Figure 5: Phase space plot scalar field cosmology with exponential potential
and matter. Parameter values are γ = 1 and λ = 1.

Figure 6: Phase space plot scalar field cosmology with exponential potential
and matter. Parameter values are γ = 1 and λ = 2.

It should be noted that the inequality signs in Table 3 exclude certain values
from the analysis. For instance, when we choose λ2 = 3γ, the two points B and
C havepthe same coordinates
p (the system has one critical point less), namely
x0 = γ/2 and y0 = 1 − γ/2 so that x20 + y02 = 1 and its eigenvalues are
0, 3/2(γ − 2). Linear stability theory cannot determine the stability of this
point. One could, in principle, apply centre manifold theory. However, this
is problematic as the physical phase space is bounded by the unit circle and
centre manifold theory will take into account the entire phase space. One could
construct the centre manifold and only consider it inside the circle but this also

23
Figure 7: Phase space plot scalar field cosmology with exponential potential
and matter. Parameter values are γ = 1 and λ = 3.

has problems.
√ For concreteness
p we set γ = 1 in the following, which means
λ = 3 and x0 = y0 = 1/2.
The easiest way forward is to use Lyapunov’s method near this point. We
start with the candidate Lyapunov function of the form
 2  2
1 1
V = x− √ +4 y− √ (78)
2 2

and one verify that this function satisfies V̇ < 0 near the critical point. Since
the function is positive definite near that point by construction, we can apply
Theorem 1. Following for instance [6], we can estimate the region of asymptotic
stability. Defining Sδ := {(x, y)|V ≤ δ} for δ ≥ 0, and denoting by Cδ the com-
ponent of Sδ containing the critical point, we have the following statement [6].
Let Ω be the set where V̇ < 0, then the interior of Cδ contained in Ω lies in
the region of asymptotic stability. As mentioned earlier, this approach relies on
our ability to find a suitable Lyapunov function. Different choices can result in
different parts of the region of asymptotic stability being covered and there is
no guarantee that the entire region can be identified by this method alone. In
Fig. 8, we show the region of asymptotic stability based on the Lyapunov func-
tion (78) for model (77). A better Lyapunov function would of course improve
this picture and increase the region.
A detailed and comprehensive phase-space analysis, based on linear stability
theory alone, of this model can be found in [17]. Other methods were explored
in [19]. A complete discussion of all its properties in the context of cosmology is
also given. This model has many interesting features as well as some problems
which motivates various extensions, many of which have been considered in the
literature. In fact, the literature of dynamical systems applications in early-time
and late-time cosmology is so vast, that it could fill several books with ease!
We should point out that this model falls short our wish list (63). The early
time fixed points A± are dominated by the scalar field, however, the effective
equation of state is weff = 1 which is unphysical. It is point C which makes

24
Figure 8: Phase space plot scalar field cosmology√with exponential potential
and matter. Parameter values are γ = 1 and λ = 3. The shaded area shows
part of the region of asymptotic stability of the fixed point. In this region V̇ < 0
and V < 3/2.

this model so interesting because this fixed point is stable and contains both,
a non-vanishing scalar field and matter. One speaks of scaling solutions as the
scalar field energy density is proportional to that of the fluid.

3.2 Cosmology with matter and scalar field and interac-

tions
The models considered so far were all two dimensional. This relied on the fact
that we were able to ‘eliminate’ the Hubble parameter H from the equations due
to a smart choice of variables and a clever choice of ‘time’. However, there are
many known models where this approach does not work and one has to introduce
new variables. In the following we will discuss one such type of models and a
possible choice of a new variable.
The cosmological Einstein field equations (65)–(66) are compatible with the
introduction of an additional interaction term Q, say. This interaction would
allow for an energy transfer from the scalar ϕ to the matter ργ and vice versa.
The introduction of such a term leaves Eqs. (65) unchanged, but (66) becomes

ρ̇γ = −3H(ργ + pγ ) − Q (79a)

dV Q
ϕ̈ = −3H ϕ̇ − + (79b)
dϕ ϕ̇
where we note that the term Qϕ̇ is natural when one computes the conserva-
tion equation ρ̇ϕ = −3H(ρϕ + pϕ ). Various choices for the coupling function Q
were considered in the literature, for instance Q = αHργ or Q = (2/3)κβργ ϕ̇
with α and β being dimensionless constants whose sign determines the direc-
tion of energy transfer from one component to the other [20, 21, 22]. Those two
choices can be motivated physically, however, one of the main motivation is the
fact that the dynamical system with these coupling remains two dimensional

25
as the Hubble parameter can be eliminated from the equations. However, both
choices appear rather arbitrary and one would prefer a choice where the cou-
pling is simply proportional to an energy density, for instance Q = Γργ with Γ
assumed to be small, see [23], or for a further generalisation [24]. In this case
the phase space cannot be represented in the plane and one has to work in a
three dimensional space.
As before, we start with the variables (68) but need a third variable in order
to be able to write the cosmological field equations as an autonomous system
of differential equations. A possible third variable z can be chosen to be
H0
z= (80)
H + H0
where H0 is the Hubble parameter at an arbitrary fixed time. It is convenient
to chose this time to be ‘today’. This variable z ensures that the physical phase-
space is compact. The Hubble parameter H → 0 in the early time universe and
H → ∞ for the late time universe. Therefore

0
 if H = 0
z = 1/2 if H = H0 (81)

1 if H → ∞


and z is bounded by 0 ≤ z ≤ 1. Since the phase-space of system (77) is half

a unit circle, we have that with coupling term Q = Γργ the phase-space now
corresponds to a half-cylinder of unit height and unit radius.
The resulting dynamical system is given by
√
6 2 3 (1 − x2 − y 2 )z
′
x = −3x + λ y + x(1 + x2 − y 2 ) − ζ (82a)
2 2 2x(z − 1)
√
6 3
y ′ = −λ xy + y(1 + x2 − y 2 ) (82b)
2 2
3
z ′ = z(1 − z)(1 + x2 − y 2 ) (82c)
2
where ζ = Γ/H0 . A detailed phase-space analysis of this model can be found
in [23]. However, this model also has some additional interesting features outside
the standard linear stability theory. For instance, there is a point vertically
above point D in Fig. 9 which attract trajectories. The system (82), however,
does not have a critical point there. √
By inspecting equations (82) at x0 = y0 = 6/(2λ) we note that y ′ (x0 , y0 ) =
0 and that
(λ − 3/λ)z
x′ (x0 , y0 ) = −ζ √ (83a)
6(z − 1)
3
z ′ (x0 , y0 ) = z(1 − z) (83b)
2
for some small coupling term ζ ≪ 1. Therefore, as the trajectories approach
the z = 1 plane, we have that also z ′ (x0 , y0 ) → 0. However, the behaviour of
x′ (x0 , y0 ) is more involved as
ζ
x′ (x0 , y0 ) ∝ . (84)
1−z

26
 Figure 9: Phase space plot of system (82) with λ = 4 and ζ = 10−6 .

On the other hand, ζ ≪ 1 but (1 − z) → 0 as z → 1. Therefore, the fraction

ζ/(1 − z) will initially be small. However, eventually the (1 − z) term will
dominate and ζ/(1 − z) will become large, explaining the repeller behaviour of
this point in Fig. 9.

4 Final remarks
It is hoped that this chapter succeeded in giving the reader a useful introduction
into the exciting field of dynamical systems in cosmology. We should remark
that the majority of papers dealing with the subject are confined to linear
stability theory and focus more on the interpretation of results in the context
of cosmology. However, there are many models where a more in depth analysis
is needed to gain a complete understanding of the physics involved. Moreover,
there is no need to select models primarily because of their simpler mathematical
structure since we have all the tools at hand to study the more difficult ones
too. We hope the reader feels encouraged to study all aspects of a cosmological
dynamical system and use a variety of techniques developed by mathematicians,
beyond linear stability theory. As Einstein wrote ‘Everything should be made
as simple as possible, but not simpler.’

27
Acknowledgements
We would like to thank Nicola Tamanini and Matthew Wright for valuable
comments on these notes.

References
[1] Y. Choquet-Bruhat. General Relativity and the Einstein Equations. Oxford
University Press, Oxford, 2008.
[2] S. Dodelson. Modern Cosmology. Academic Press, Elsevier, San Diego,
2003.
[3] J. Wainwright and G. F. R. Ellis. Dynamical Systems in Cosmology. Cam-
bridge University Press, 1997.
[4] A. A. Coley. Dynamical systems and cosmology. Kluwer Academic Pub-
lishers, Dordrecht Boston London, 2003.
[5] S. Wiggins. Introduction to Applied Nonlinear Dynamical Systems and
Chaos. Springer, New York Heidelberg Berlin, 1990.
[6] F. Brauer and J. A. Nohel. The qualitative theory of ordinary differential
equations. Dover Publications, New York, 1989.
[7] J. Carr. Applications of centre manifold theory. Springer, New York Hei-
delberg Berlin, 1981.
[8] A. H. Nayfeh. The Method of Normal Forms. Wiley-VCH, Weinheim, 2011.
[9] A. D. Rendall. Cosmological models and centre manifold theory. Gen. Rel.
Grav., 34:1277–1294, 2002.
[10] A. Coley, and S. Hervik. A Dynamical systems approach to the tilted
Bianchi models of solvable type. Class. Quant. Grav., 22: 597–606, 2005.
[11] J. Perez, A. Füzfa, T. Carletti, L. Mélot, and L. Guedezounme. The Jun-
gle Universe: coupled cosmological models in a Lotka-Volterra framework.
Gen.Rel.Grav., 46:1753, 2014.
[12] S. Perlmutter et al. Measurements of Omega and Lambda from 42 High-
Redshift Supernovae. Astrophys. J., 517:565–586, 1999.
[13] A. G. Riess et al. Observational Evidence from Supernovae for an Accel-
erating Universe and a Cosmological Constant. Astron. J., 116:1009–1038,
1998.
[14] S. Weinberg. The Cosmological Constant Problem. Rev.Mod.Phys., 61:1–
23, 1989.
[15] V. Sahni. The Cosmological constant problem and quintessence.
Class.Quant.Grav., 19:3435–3448, 2002.
[16] J. Martin. Everything You Always Wanted To Know About The Cosmo-
logical Constant Problem (But Were Afraid To Ask). Comptes Rendus
Physique, 13:566–665, 2012.

28
[17] E. J. Copeland, A. R. Liddle, and D. Wands. Exponential potentials and
cosmological scaling solutions. Phys. Rev., D57:4686–4690, 1998.
[18] E. J. Copeland, M. Sami, and S. Tsujikawa. Dynamics of dark energy. Int.
J. Mod. Phys., D15:1753–1936, 2006.
[19] C. G. Böhmer, T. Harko, and S. V. Sabau. Jacobi stability analy-
sis of dynamical systems: Applications in gravitation and cosmology.
Adv.Theor.Math.Phys., 16:1145–1196, 2012.
[20] L. Amendola. Scaling solutions in general nonminimal coupling theories.
Phys.Rev., D60:043501, 1999.
[21] D. J. Holden and D. Wands. Selfsimilar cosmological solutions with a
nonminimally coupled scalar field. Phys.Rev., D61:043506, 2000.
[22] A. P. Billyard and A. A. Coley. Interactions in scalar field cosmology.
Phys.Rev., D61:083503, 2000.
[23] C. G. Böhmer, G. Caldera-Cabral, R. Lazkoz, and R. Maartens. Dynamics
of dark energy with a coupling to dark matter. Phys.Rev., D78:023505,
2008.
[24] C. G. Böhmer, G. Caldera-Cabral, N. Chan, R. Lazkoz, and R. Maartens.
Quintessence with quadratic coupling to dark matter. Phys.Rev.,
D81:083003, 2010.

29