You are on page 1of 8

6

Stability

There are two different ways to think about the stability for a system of nonlinear equations. Firstly, we could ask whether if a point is in a particular region of
phase space whether it stays near this region. Secondly we can consider making
a small change to the system and ask whether this results in a significant change
in the solutions. The latter in called structural stability. We will consider both
types in turn.

6.1

Stability of a Point in Phase Space

There are two main definitions of stability in phase space x. One is when
trajectories starting at nearby points flow to a point xf . This is asymptotic
stability. The other is when trajectories starting near a point xf stay near this
point. This is called Liapunov stability.
We can be more precise in these definitions. Suppose x(t) is some trajectory
and x(t = 0) = x0 is near xf :
Liapunov stability. A fixed point xf is Liapunov stable if for all possible
> 0 there exists a such that for all x0 satisfying |x0 xf | < we have
|x(t) xf | < for all t.
Asymptotic stability. A fixed point xf is asymptotically stable if it
is Liapunov stable and in addition there exists a such that for all x0
satisfying |x0 xf | < we have x(t) xf for all t.
The important point to grasp about Liapunov stability is that can be made
as small as one likes, i.e. stability only occurs if by an appropriate choice of
one can remain arbitrarily close to xf . Asymptotic stability is clearly a subclass
of Liapunov stability. It is also possible that trajectories may flow ultimately
to a point but without Liapunov stability. This is known as quasi-asymptotic
stability. An illustration of all three cases are illustrated in fig. 13.
In two dimensions a stable star, node or focus is asymptotically stable,
whereas a centre is Liapunov stable only. The region where all trajectories flow
to a point is the domain of asymptotic stability. The largest region is the basin
of attraction. We have already seen an example of this when considering the
phase portrait for rabbits and sheep, where the phase space separated into a
basin of attraction for rabbits only and for sheep only separated by the line
known as the stable manifold.
Quasi-asymptotic stability is demonstrated by the set of equations
= sin2 (/2),

r = r(1 r2 ),

(256)

where r and represent polar coordinates. This system of equations has a fixed
point at
r = 1, = 0
(x = 1, y = 0).
(257)
49

Figure 13: An illustration of (a) Liapunov stability, (b) asymptotic stability,


(c) quasi-asymptotic stability.
This is a stable fixed point. This is clear by considering perturbations about
r = 1. If we let = we are clearly attracted directly back to the fixed point.
However, if = 2 we only get back to the fixed point by travelling in a
circuit encompassing the origin and reaching a distance up to 2 from the fixed
point no matter how close we start.

6.2

Limit Cycles

We can also sometimes find asymptotic solutions corresponding to a continuous


region rather than a single point. Periodic orbits sometimes attract (or repel)
nearby points. We can define the stability of such orbits in a similar manner
as for fixed points. Define the set of points on the orbital trajectory as , and
define a neighbourhood of by N (, ) which contains points y where we can
find points x in such that |y x| < . We define
50

Liapunov orbital stability. For all > 0 there exists a such that for
all y0 N (, ) then for all t we have y(t) N (, ).
Asymptotic stability. We require that is Liapunov stable and that
there is a such that for all y0 N (, ) then we have y(t) approaches
as t .
Asymptotically stable orbits are called limit cycles. These are important in
populations, economics and in physiology. They do not arise in purely linear
systems. Consider the linear equations x = Gx, where G is a constant matrix.
If x(t) is a solution to this equation then so is cx(t), where c is a constant.
Hence, any orbital solution has indeterminate size.
A simple example of a limit cycle is a van der Pol oscillator, described by
x + (x2 1)x + x = 0,

(258)

where is positive. If the oscillation amplitude exceeds 1 there is damping of


the oscillation, while if the amplitude is less than 1 there is negative damping
or enhancement. Another simple example is given in circular polars by the
equations
r = r(1 r2 ),
= 1.
(259)
The stable solution for r is clearly r = 1, while a constant rotation about the
origin occurs for any value of r.

6.3

Structural Stability of Solutions

This considers changes to the system of equations. Consider the equations


x = g(x).

(260)

This system of equations is said to be structurally stable if for small we can add
h(x) to the right-hand side such that solutions remain qualitatively equivalent
to the original solutions, i.e. there is a one-to-one mapping between the two
sets of solutions. This is a complicated topic, and we can only treat it in a
qualitative manner. We can consider the stability by looking at the form of
solutions in the linear case.

6.4

The Linearisation Theorem

This states that for any simple fixed point of a system of nonlinear differential
equations the phase space portrait close to the fixed point remains qualitatively
the same (one-to-one mapping) as for the linearised form of the equations provided the fixed point is not a centre. Hence, an attracting or repelling fixed
point retains its form, but a star may become a focus (or vice versa), for example. Centres are unstable in this sense because they rely on a precise fine-tuning
such that there is no growth or decay at all in the solutions (trG = 0 precisely)
which can easily be disturbed by nonlinear terms.
51

Figure 14: (a) The phase portrait for a non-simple fixed point obtained in
the linear limit. (b) the distortion of this phase portrait by the addition of a
non-linear term.
A simple fixed point means that there is not a zero eigenvalue for the linear system. A zero eigenvalue again relies on a very precisely defined system
and small perturbations can remove a line along which there is no evolution.
Consider the linear equations
x = x,

y = 0.

(261)

This results in the phase portrait in fig. 14(a). These can have the nonlinear
extension
x = x,
y = y 2 .
(262)
Even for small the nonlinear term alters the evolution in the y direction
completely and results in the much-altered phase portrait in fig. 14(b).
Centres and non-simple fixed points are called borderline cases, because
they only exist due to very precise eigenvalues, whereas small effective changes
in eigenvalues for other fixed points do not alter the fact that there is growth, or
damping. This suggests that it is unlikely that we will find centres as genuine
solutions to systems of nonlinear equations since their existence is so fragile.
This is generally true. However, they can be protected in special circumstances
where there are conserved quantities, as is often the case for physical systems.
We will consider these conservative systems next.

52

Conservative Systems

Conservative systems form an important class of dynamical systems. They


occur very frequently since physical systems often have one, or more conserved
quantities, e.g. energy, and angular momentum. The existence of this conserved
quantity then often makes it easier to find the form of solutions.

7.1

First Integrals and Conservative Systems

A first integral of a system of differential equations is a function Q(x) that is


constant on a given trajectory. It is called first integral because it often arises
from integration of the equation for the slope of a trajectory, e.g.

dy
f (x)
=
dx
g(y)
Z

dy g(y) =

dx f (x) + c,

(263)

Therefore the first integral is


Q(x, y) =

y0

d
y g(
y)

x0

d
x f (
x),

(264)

where we could add a constant.


We do not allow the function Q(x) to have the same value for a continuous
region of phase space, i.e. for all solutions, since this tends to lead to trivial
results. A conservative system is then defined to be one which contains a first
integral for all possible solutions in phase space.
In D dimensions Q(x) = c gives a manifold of dimension D 1 within which
a solutions trajectory lies. For example, in 3 dimensions the solutions lie on a
surface of constant first integral.

7.2

Examples

There are various examples of conservative systems. The obvious one is a


Hamiltonian System. A simple example of this for a particle moving in
a potential in 1 dimension is
x = p/m,

p =

dV (x)
.
dx

(265)

The trajectory for this is

and so

dp
dV /dx
=
,
dx
p/m

(266)

(267)

dp p/m =

53

dx

dV (x)
+ c,
dx

and so

p2
+ V (x) E,
(268)
2m
i.e. the energy E is the first integral. In fact we know from our results in section
1 that this is a general result, with the Hamiltonian always being a conserved
quantity unless the potential has explicit time dependence, and in most cases
the Hamiltonian is equal to the energy of the system. In this case we have the
general first integral
p2
E=
+ V (r).
(269)
2m
Another example of a conservative system is the Linear Saddle. This has
the equations
x = x,
y = y,
(270)
c=

which gives a saddle point at the origin. The trajectory is


dy
= y/x,
dx

(271)

ln y = ln x + c.

(272)

c Q(x, y) = xy

(273)

which integrates to give


Therefore the first integral is

is a constant on all trajectories and the system is conservative.

7.3

First Integral and Trajectories

Knowing the first integral gives us information about the form of solutions.
In 2 dimensions Q(x, y) = c reveals a trajectory in phase space. However,
different trajectories (with different solutions) can have the same first integral,
e.g. consider the system
x = x(2 y),

y = y(2 y).

(274)

This has the same first integral as the linear saddle. This means that the
trajectories are exactly the same lines as for the linear saddle. However, the
rate of movement along the trajectories is different in this case. In particular
there is a fixed line at y = 2, and for y > 2 the direction of the arrows on the
trajectories is reversed.

7.4

Consequences for Conservative Systems

There are various consequences for and features of conservative systems.


1. As we have seen above, Q(x) = c gives a useful method for finding trajectories.
54

2. There are no purely attracting or repelling fixed points for a conservative


system, i.e. there are no nodes, stars, foci or limit cycles. This is because
if a conserved quantity exists then it has some value at the attracting or
repelling fixed point. But all trajectories are linked by this fixed point and
Q(x) is a constant throughout a continuous region of phase space. This
is ruled out by the definition of a first integral. For example, consider the
stable star:
x = x,
y = y.
(275)
This leads to the trajectory
dy
y
=

ln y = ln x + c
dx
x
Q(x, y) = x/y = c.

(276)

This is ill-defined at the origin and is not valid everywhere in phase space.
It is true that away from the origin a trajectory satisfies x = cy, i.e. a
straight line, but c cannot be interpreted as a conserved quantity.
3. Trajectories are given by contours of constant Q(x). Since Q(x) is the first
integral then a stationary point for Q(x) is a zero for the right-hand side
of the equations and is a fixed point. Maxima and minima are surrounded
by closed orbits, as shown below.
Q(x, y)

y
Qmin

55

Therefore we have nonlinear centres in conservative systems, even though


we saw that these are generally unstable for nonlinear equations. We also
find that a saddle point in Q(x) is a saddle point in the phase portrait.
An example is the linear saddle where Q(x, y) = xy has a saddle point at
the origin.
4. Trajectories that leave a saddle point in a conservative system often return
to the same point. This is because it is a point with the same first integral.
These are called homoclinic trajectories. If two saddles have the same
Q(x) and trajectories leave one and enter the other we have heteroclinic
trajectories. These are rare in systems that have no conserved quantity.
An important class of conservative conservative systems are Hamiltonian
systems, which nearly always have conserved quantities. We will examine some
examples of these next.

56

You might also like