Sie sind auf Seite 1von 32

Chapter 4

Systems of equations
In this chapter we examine methods for analysing systems of odes. We shall examine both exact
and graphical solutions of the problem. We shall examine the circumstances under which the
solutions of the systems of odes tend to limits and the stability of those limiting solutions.

4.1

Linear systems of ODES

Consider two populations of animals in an ecosystem with populations x(t) and y(t) as a function
of time t. Suppose that population y feeds on population x. What are the differential equations
governing the rate of change of populations x(t)

and y(t).

A very simple model could be as


follows:
First consider the birth rates. Clearly the larger the population x(t), the larger the possibility
of reproduction, so its a good starting bet that x is proportional to x(t). Similarly y will be
proportional to y(t). Hence we have
x = ax(t) + ,
y = cy(t) +
where a and c are constants.
Now consider the death rates. Clearly the number of each population dying from natural
deaths will also be proportional to the respective populations. Natural deaths thus do not change
the form of the above equations, it only lowers the values of the constants a and c.
However if y feeds on x, the larger the population of y, the greater the death rate of x (yum
yum!). Conversely if y depends on x to survive, the larger the supply of food x, the easier the
169

170

CHAPTER 4. SYSTEMS OF EQUATIONS

life for y and so the greater the potential for population y to grow in numbers. Hence we could
add the following terms to each equation:
x = ax(t) + by(t),
y = cy(t) + dx(y)
where a, b, c, d are constants. We now have a set of two coupled linear ordinary differential
equations involving two unknown functions x(t) and y(t). At first sight, it seems that to solve
the x equation to find x(t) we have to first know y(t), but conversely to solve the y equation to
find y(t) we have to know x(t).
What should we do? Such equations occur all the time, not just in predator-prey population
models, but also in areas such asmodels of war and insurgency, coupled mechanical systems and
climate change.
Dont panic! There are several ways to attack this problem and the following sections show
you how to do it.

4.2
4.2.1

Analytical solution techniques


Solution of linear system by conversion to a higher order ode

We first consider linear systems of ODEs of the form


dx
= a x (t) + b y (t)
dt
dy
= c x (t) + d y (t)
dt
where the two unknown functions x(t) and y(t) depend on time and a, b, c and d are constants.
The first thing to note is that this system can be related to higher order constant coefficient
ODEs by eliminating y between the two equations as follows. The second equation gives
y=

cx
,
d

d 6= 0.

Substituting this into the first equation to eliminate y we obtain

y c x
x = a x + b
) by = d x (ad
d

bc) x.

4.2. ANALYTICAL SOLUTION TECHNIQUES

x = a x + b y
Thus substituting for y we have
x = a x + d x

(ad

bc) x

171

x = a x + b y.

(a + d) x + (ad

bc) x = 0

This is a second order constant coefficient equation which can be solved as above. Once the
solution for x is known, that for y can be found by back substituting in either of the original
differential equations and solving the resulting ODE for y.

4.2.2

Solution of linear system by matrix methods

There is however a more direct method that solves for x and y simultaneously and works for
such 2 by 2 systems linear ODEs but can also be generalised easily to n by n systems with
n unknown functions and n equations.
First write the problem in matrix form


a b
x
x
=
y
c d
y
and define the matrix and column vector as

x
x
,
y

M=

a b
c d

so that
x = Mx.
If M and x were not matrices we would have an equation that looks rather like the simplest
interest rate/population growth model. These equations have solutions that look like A exp ( t),
so lets try that here.
We guess that there is a solution of the form
t

x (t) = x0 e =

x0
y0

where x0 is an arbitrary constant vector and is a constant. As with the constant-coefficient


linear equations we find these constants by substituting this guessed solution into the equation.
We note that x can be found to be
t

x (t) = x0 e =

x0
y0

e t.

172

CHAPTER 4. SYSTEMS OF EQUATIONS

Substituting into the equation we obtain


x0 e t = Mx0 e

Mx0 = x0 .

This last equation always has the trivial solution x0 0, but this is not very interesting. There
are however non-trivial solutions to the equation and we can find these if we note that the equation
is none other than the equation for the eigenvalues of the matrix M with eigenvectors x0 . Hence
we know from matrix theory that is given by the solutions of the characteristic polynomial
given by
det (M
I) = 0.
The eigenvectors x0 are then found by substituting the calculated values of .
The general solution is then the sum of arbitrary constants times each eigenvector times the
exponential of the corresponding eigenvalue times t.
Example:
Solve

x = x + y
y = 4 x + y

We re-write this in the form


x = Mx,

M=

and then substituting the guessed solution


)
into the equation to obtain

1 1
x0 =
x0
4 1

This has non-trivial solutions when det (M


1
4

1
1

=0

)2

(1

Now substitute in for each value of

=3:

1 1
4 1

1 0
0 1

1 1
4 1

x = x0 e

1 1
4 1

x0 = 0.

3=0

1 0
0 1

I) = 0, i.e., when

4=0

= 3 or

1.

to find the relevant x0

x0 = 0

2 1
4

x0
y0

=0

2x0 + y0 = 0
4x0 y0 = 0

4.2. ANALYTICAL SOLUTION TECHNIQUES

173

from which we deduce that x0 is any vector that takes the form:

x0 = D
1
) x0 = D
.
y0 = 2D
2

where D is an arbitrary constant. Hence we have one solution that is



1
x=D
e3t .
2
The other eigenvalue generates the following eigenvector

1:

1 1
4 1

( 1)

1 0
0 1

x0 = 0

2 1
4 2

x0
y0

from which we deduce that x0 is any vector that takes the form:

x0 = E
1
) x0 = E
.
y0 = 2E
2

=0

2x0 + y0 = 0
4x0 + 2y0 = 0

where E is an arbitrary constant. Hence we have another solution

1
x=E
e t.
2

We now have two possible solutions to the original equation. What do we do? Well the
original equation was linear and we know that any linear combination of solutions of a
linear equation is also a solution, so we just add these to get the total solution

1
1
3t
x=D
e +E
e t
2
2
which can be written as

or

x
y

De3t + Ee t
2De3t 2Ee

x (t) = De3t + Ee t ,

y (t) = 2De3t

2Ee

Beware: You cannot rewrite the 2D and 2E in the y solution as new arbitrary constants
F and G say so that
x (t) = De3t + Ee t ;

y (t) = F e3t + Ge t .

This is because the solutions x and y were coupled by the original system of ODEs. What
happens in one solution affects what happens in the other via the coupling of the differential
equations. Consequently the arbitrary coefficients in one solution are slaved to those in the
other.

174

CHAPTER 4. SYSTEMS OF EQUATIONS

The arbitrary constants D and E are found from initial or boundary data. Two such conditions are required as we have two arbitrary constants.
Example: Solve the above system subject to the initial data x (0) = 3, y (0) = 2.
From the solutions we have:
x (0) = 3 = De0 + Ee0
y (0) = 2 = 2De0 2Ee0

3=D+E
1=D E

x (t) = 2e3t + e t ,

y (t) = 4e3t

)
2e

D=2
E=1

xHtL, yHtL
yHtL

40

40
20

-3

-2

20

-1

-20

(a) Values of x (t) = 2e + e , y (t) = 4e


plotted as a function of t.
t

20

30

40

xHtL

-20

-40
3t

10

-40
3t

2e

(b) Plot of y(t) against x(t) for

3t1

Figure 4.1: Different ways of representing the solutions of a 2 2 set of coupled equations.
We can chose how we plot the solutions. We can obviously plot x(t) and y(t) as explicit
functions of t. However, since x(t) and y(t) were coupled by the original differential equations,
it often makes sense to plot y(t) against x(t) in the (x, y) phase plane. Here we treat t as a
parameter: you give me a value of t and I use x(t) and y(t) to generate the values of x and y
at that value of t. I plot this point in the (x, y) plane. Then you give me another value of t, I
generate and plot another pair (x, y) etc. The result is a curve, or trajectory, in the (x, y) plane
that should how y behaves for each value of x. There is a direction to each trajectory, determined
by the direction the system travels along the trajectory as t increases (cf. the phase plane we
studied in the last chapter.)
Even though the matrix equation was only first order, there are two arbitrary constants. This
seems strange since you should by now know that the number of arbitrary constants in a linear
ODE is equal to the order of the equation. However the matrix form of the equation disguises that

4.2. ANALYTICAL SOLUTION TECHNIQUES

175

both x and y can be found from solutions of second order linear constant coefficient equations,
as detailed above. Hence we expect 2 arbitrary constants.
The above method will work on a system of n first order coupled constant coefficient equations. You obtain (at most) n eigenvalues, n solutions and n arbitrary constants.
As in higher order linear constant coefficient equations, it is entirely possible to obtain complex or repeated eigenvalues. These are dealt with in ways similar to the methods of higher order
constant coefficient equations.
The following gives some idea of how the ideas can be extended to consdier special cases
where the eigenvalues are not simple distinct real values.

4.2.3

Solutions involving complex eigenvalues:

Example: Solve
x = x + 5y
y = x + 3y
The method of solution proceeds as above: x = Mx where

x =

5
1 3

5
x ) x = x0 e t
1 3

1 0
1
=0 )
0 1
1

= 2 + 2i :

1
5
1 3

(2 + 2i)

1 0
0 1

det (A

5
3

=0

1
1

2i 5
1

I) = 0
2

4 +8=0

= 2 2i.

x0 = 0

x0 = D
y0 = 1+2i
D
5

1
x = D 1+2i e(2+2i)t .
)

2i

x0
y0

=0

176

CHAPTER 4. SYSTEMS OF EQUATIONS

=2

2i :
5
1 3

(2

2i)

1 0
0 1

x0 = 0

x=E

1 + 2i 5
1
1 + 2i

x0
y0

=0

x0 = E
y0 = 1 52i E

1
1 2i
5

e(2

2i)t

Hence the general solution is


x=A

5
1 + 2i

(2+2i)t

+B

5
1

2i

e(2

2i)t

Suppose we have the initial data x(0) = 1, y(0) = 0. Then we can solve for A and B
above as

1
5
5
x(0) =
=A
+B
.
0
1 + 2i
1 2i
Then we have
1 = 5A + 5B
0 = (1 + 2i)A + (1
which can be solved (exercise for you!) to give
1
(2 + i)
20
1
B =
(2 i)
20
A =

2i)B

4.2. ANALYTICAL SOLUTION TECHNIQUES

177

Substituting these values into the general solution we obtain


1
x =
(2 + i)
20

5
1 + 2i

(2+2i)t

1
+ (2
20

e2t
=
4

2+i
i

e2t
=
4

2+i
i

e2t
=
4

(2 + i + 2 i) cos 2t + (2i 1
(i i) cos 2t + ( 1 1) sin 2t

e2t
=
4

4 cos 2t 2 sin 2t
2 sin 2t

e2t
=
2

2 cos 2t
sin 2t

(2+2i)t

2it

1
+
20

i)

1
=
20

10 + 5i
5i

i
i

10 5i
5i

(cos 2t + i sin 2t) +

sin 2t

5
1

e(2

2i

e(2

2i)t

2i)t

2it

i
i

2i

(cos 2t + i sin 2t)

1) sin 2t

e2t
(2 cos 2t
2
e2t
y(t) =
sin 2t.
2

x(t) =

sin 2t) ,

The moral of this example is that, subject to the appropriate initial conditions, solutions associated with complex eigenvalues can be converted into real exponentials multiplying oscillatory
behaviour involving cosines and sines.

178

CHAPTER 4. SYSTEMS OF EQUATIONS

xHtL, yHtL

yHtL

20

4
10

-1.0

-0.5

0.5

1.0

1.5

t
2.0

-10

-25

-20

-15

-10

xHtL

-5
-2

-20

(a) Values of x (t) = e2t (2 cos 2t sin 2t) /2,, (b) Plot of y(t) against x(t) for 1 t 2. The
y (t) = e2t sin 2t/2 plotted as a function of t.
curve is a spiral, exponentially increasing in distance
from the origin.

Figure 4.2: The functions x(t), y(t) and their trajectory in the (x, y) plane.

4.2.4

Repeated roots

Repeated eigenvalues can lead to two different cases


1. M has a complete set of eigenvectors,
2. M has an incomplete set.

Example (repeated roots, complete set of eigenvectors):


x = 3x
Solve the (trivial, since they are decoupled) set of equations
y = 3y
t
x = x0 e
) ... )
= 3, twice.

x =

3 0
0 3

The eigenvector equation is thus the same for both eigenvalues


x0
0 0
x0
0
(M 3I)
=0 )
=
y0
0 0
y0
0
Thus x0 and y0 can be anything! Hence we can have two distinct eigenvectors just as long as the
entries make them linearly independent, say

x0
0
B
=
and
.
y0
A
0

4.2. ANALYTICAL SOLUTION TECHNIQUES

179

This gives us the general solution of


x=A
or

0
1

3t

e +B

x = Be3t ,

1
0

e3t ,

y = Ae3t .

Note here that A and B are independent constants, due to the fact that the original odes were
decoupled.
Example (repeated roots, incomplete set of eigenvectors):
Solve the set of equations
x = 3x + y
y = 3y

x =

3 1
0 3

x = x0 e

...

= 3,

twice.
The eigenvector equation is thus

x0
(M 3I)
=0
y0

0 1
0 0

x0
y0

0
0

This set of equations has the solution y0 = 0 and x0 undertermined. Hence we can write

x0
D
=
.
y0
0
where D is an arbitrary constant. Here is thus only one eigenvector

D
e3t
0
Thus there is only one solution of this exponential type, when we were expecting two. Since we
started with a 2x2 system we therefore say this is an incomplete set of eigenvectors.
The resolution is to proceed as in higher order constant coefficient equations and look for
solutions of the form
x = (x1 + x0 t) e t
where x0 and x1 are constant 2 1 column vectors.
The solution procedure is as follows:
x =

(x1 + x0 t) e t + x0 e

180

CHAPTER 4. SYSTEMS OF EQUATIONS

Hence in x = Mx we have

(x1 + x0 t) e t + x0 e t = Mx1 e t + Mx0 te t .


Collecting powers of t we get

(M

I) x1

x0 + t (M

I) x0 = 0.

For this to be valid for all t, we compare powers of t on each side of the equation. We will
thus certainly need (M
I) x0 = 0,which is the same equation as above. Hence, as before,
D
we deduce that = 3 twice and x0 =
.
0
We now back substitute for x0 to obtain the equation

(M

I) x1 x0 = 0

0 1
0 0

x1
y1

D
0

0
0

0 1
0 0

x1
y1

with solution y1 = D and x1 undetermined = C, say.

x1 =

x1
y1

C
D

where D is the same D as in x0 and C is a second arbitrary constant (the one we were looking
for all along!). Hence the general solution is

x = (x1 + x0 t) e =

C
D

D
0

This can be rewritten as


x = (C + Dt) e3t ,

y = De3t .


t e3t .

D
0

4.3. COUPLED EQUATIONS, PHASE PLANE AND DYNAMICAL SYSTEMS

4.3

181

Coupled Equations, Phase Plane and Dynamical Systems

We have shown above how to find, explicit, specific solutions of coupled linear equations of the
form
dx
= a x (t) + b y (t)
dt
dy
= c x (t) + d y (t)
dt
(where a, b, c and d are constants). Here we show how the general behaviour of the solution
trajectories in the phase plane (x(t), y(t)) can be deduced locally from knowledge of the location
and nature of the equilibrium points of the system.
We will also demonstrate the geometric role played by the eigenvalues and eigenvectors of
the solutions.
More generally a similar approach can be applied to nonlinear coupled systems of the form
dx
= f (x(t), y(t))
dt
dy
= g(x(t), y(t))
dt
Systems of this form where there is no explicit dependence on the independent variable (here
t1 ) are called autonomous systems. Such systems form the basis of a field of study called
dynamical systems. The solution trajectories (x(t), y(t)) describe the state of the dynamical
system.

4.3.1

Equilibrium Points

Fixed points of dynamical systems are occur when


dx
= 0
dt
dy
= 0
dt
simultaneously. At an equilibrium point the solutions are stationary with respect to t.
In general, the solution trajectories (x(t), y(t)) are either attached to, or repulse from such
points. For example if the solution trajectory is attracted to such a point, when it encounters it,
1

The system x = 2xy, y = sin(x + y) is autonomous since the t dependence on the right hand side is only
appearing implicitly in the functions x(t) and y(t). The system x = 2xy + t, y = sin(x + y) is not autonomous
since in the first equation there is a separate t dependence that is present that is not part of x(t) or y(t).

182

CHAPTER 4. SYSTEMS OF EQUATIONS

there is no rate of change of x(t) or y(t) at that point, and so the system remains in that state and
remains fixed there and does not change.
Such points are also called equilibrium points.
Any solution that starts at an equilibrium point will stay at that point and not change (the
derivatives for both x and y are zero simultaneously at an equilibrium point and so the solutions
for x and y simultaneously cannot change there, and so the system may be stuck there). For
this reason, equilibrium points are also sometimes called fixed points. (We examine below what
happens to solutions that start near to the equilibrium points.)
For the linear systems considered above, equilibrium points are given by:
ax + by = 0
cx + dy = 0

(x, y) = (0, 0)

Hence unless ad = bc the equilibrium point is unique and centred at the origin (0,0). (This can
always be arranged for any linear system by a constant translation in the dependent variables.)
More generally, equilibrium points for nonlinear equations
dx
= f (x(t), y(t))
dt
dy
= g(x(t), y(t))
dt
are given by the simultaneous solution in x and y of the nonlinear equations
f (x, y) = 0
g(x, y) = 0
Hence for nonlinear systems, more than one fixed point may occur.
Note also that for autonomous systems we do not need to solve the differential system before
we can find the equilibrium point: the equilibrium points in the (x, y) phase plane are fixed for all
t by the simultaneous solution of f (x, y) = 0, and g(x, y) = 0. For this reason the equilibrium
points of autonomous systems can also be referred to as fixed points.
Example (equilibrium points of nonlinear system):
The coupled set of differential equations
dx
= y3 + x
dt
dy
= x y2
dt

4.3. COUPLED EQUATIONS, PHASE PLANE AND DYNAMICAL SYSTEMS

183

is a nonlinear system. This is because the equations contain powers (greater than one) of the
unknown functions x(t) and/or y(t).
It has equilibrium points at the simultaneous solution of
y3 + x = 0
x y2 = 0

(x, y) = (0, 0) or (1, 1).

Similarly the following set of coupled equations is also a nonlinear system:


dx
= x y
dt
dy
= y(x 1)
dt
This is because the second equation contains products of the unknown functions x(t) and y(t).
It has equilibrium points at the simultaneous solution of
x y = 0
y(x 1) = 0

(x, y) = (0, 0) or (1, 1).

Since these are nonlinear equations, the superposition of solutions of this type of equation does
not lead to another solution of the equation. Clearly it is not possible to represent such as system
in the form x = Mx, where M is a constant coefficient matrix. Hence nonlinear systems are,
in general far, more difficult to solve than the linear coupled equations. This is one of the main
reasons why the theory of dynamical systems was developed.
For now now we will explain how to analyse the local behaviour of equilibrium points in the
context of linear systems. We shall see an example of how to apply this to a nonlinear system
later on.

4.3.2

Role of eigenvectors and eigenvalues

For the linear system


dx
= ax+by
dt
dy
= cx+dy
dt
we have seen that the general solution is given by


x(t)

x(t) =
=A
e
y(t)

1t

+B

2t

184

CHAPTER 4. SYSTEMS OF EQUATIONS

where the i are the eigenvalues of the coefficient matrix and the column vectors are the corresponding eigenvectors. Suppose for the moment that the i are both real.
Here A and B are just arbitrary constants. Each pair of values (A, B) corresponds to a unique
solution trajectory in the phase plane (x, y).
Consider what happens when the initial conditions are such that, e.g., B = 0. The solution is
then given by


x(t)

x(t) =
=A
e 1t
y(t)

Extracting this specific solution from the matrix we just have x(t) = e 1 t , y(t) = e 1 t . Eliminating the t, this just gives the relationship between the x and y co-ordinates in the (x, y) phase
plane as
y=

x.

For this specific solution, the e 1 t dependence just determines how far each of the x and y coordinates are away from the origin on this line.
Likewise, if we consider the specific solution where A = 0 and carrying out a similar procedure, we find that this has a trajectory in the (x, y) phase plane given by
y=

x.

A simple sketch of the these solutions in the (x, y) phase plane shows that they are both straight
lines passing though the origin, or equilibrium point.
The direction of flow of the solutions along the trajectories as t increases is given by the sign
of the i :
If i > 0 the solutions moves away from the equilibrium point as t increases (x and y
increase with t).
If i > 0 the solutions moves towards the equilibrium point as t increases (x and y
decrease with t).
For other arbitrary values of A and B we have other trajectories, one each for each pair of (A, B).
These can be sketched from the direction field
dy
(dy/dt)
cx + dy
=
=
.
dx
(dx/dt)
ax + by
In general, due to the autonomy of the equations and the uniqueness of initial value problems,
these other trajectories will only intersect at equilibrium points. Hence the equilibrium points
will give the local behaviour around which the rest of the trajectory geometry is organised.

4.3. COUPLED EQUATIONS, PHASE PLANE AND DYNAMICAL SYSTEMS

185

The direction of flow of the solutions along the trajectories can be deduced from continuity
near the trajectories through the equilibrium point. If the flow along nearby trajectories were not
in the same sense as the flow along the lines through the equilibrium point, then there would be
have to be a discontinuity in the differential equations (which here there is not).
In this case we can see that the equilibrium point is an unstable one. Almost all the trajectories that are initially close to the equilibrium point at the origin do not remain so as t ! +1.
Hence solutions that start a small distance from the equilibrium point will almost always diverge
to infinity.

<0: y=

0.4

0.2

y(t)

0.0

>0:

y=

-0.2

-0.4

-0.4

-0.2

0.0

0.2

0.4

x(t)
Figure 4.3: An example of the geometry of a (x, y) phase plane plot in the vicinity of a saddle
equilibrium point at (x, y) = (0, 0). The flow of the solutions along two special trajectories
y = x/ and y = x/ (thick black lines) given by the eigenvectors is determined in this case
to the sign of the corresponding real eigenvalues. For negative real eigenvalues the solutions are
attracted along the eigenvalue trajectory towards the equilibrium saddle point. For positive real
eigenvalues the solutions are repulsed along the eigenvalue trajectory away from the equilibrium
saddle point. The trajectories of solutions that start at other initial values (x(0), y(0)) can be
sketched in locally by continuity, as can the flow directors along them (so that away from the
equilibrium point, the arrow on adjacent trajectories always point in the same direction.

This leads us to some of the key points in the theory of dynamical systems:

186

CHAPTER 4. SYSTEMS OF EQUATIONS

The nature of an equilibrium point determines the local flow of trajectories in the
vicinity of it.
It is possible to sketch the trajectories in the phase plane qualitatively in the vicinity
of the an equilibrium point using the direction field and continuity of the flow along
the solution curves.
The local flow will give an indication of both the effect of perturbing solutions from
the equilibrium points, as well as the asymptotic behaviour in t of couple solutions
originating in a particular region of the phase plane.
In systems of nonlinear differential equations, if there is more than one equilibrium point, then
continuity of flow between the local regions is used to patch the local behaviours together (see
example below).
Thus the types of fixed point and their stability are determined by the values of the eigenvalues i . Looking at the diagram above we see that if
If Re( 1 ) < 0 and Re( 2 ) < 0, then |x| ! 0 as t ! +1: the equilibrium point is stable.
If Re( i ) > 0 for any i, then |x| ! 1 as t ! +1: the equilibrium point is unstable.

4.3.3

General Classification of trajectories in neighbourhood of equilibrium points

For linear systems x = Mx with constants coefficients, we have:

a b
x =
x ) x = x0 e t ) det (A
c d
a
c

b
d

=0

(a + d) + (ad

I) = 0

bc) = 0

In general the quadratic equation will have two distinct roots 1 and 2 provided that (a + d)2 6=
4(ad bc). The roots may be real or complex. In the latter case, for real a, b, c, d the roots will
be complex conjugates.

4.4. LINEAR EXAMPLES

187

The classification is then as follows:


1.

1,

2.

1,

2 R:
1

< 0 and

< 0: stable node;

> 0 and

> 0: unstable node;

1 2
2

< 0: (unstable) saddle.

2 C with

= + i!,

i!:

> 0: unstable spiral;


< 0: stable spiral;
= 0: centre.

Generalisations to higher dimensional systems involving n-coupled equations, with solution


trajectories (x(t), y(t), z(t), ) etc. follow on from the 2-D case by obvious generalisation.

4.4

Linear Examples

4.4.1

Example: stable node

The system,
x = Mx,
has general solutions:
x=A

M=

p
2

4t

p3
2

+B

p
2
2
1p

Hence from above we have 1 = 4, 2 = 1 so 1 < 0, 2 < 0 and we have stable node at
the origin. The node is organised around the straight lines passing through the centre given by
the associated relevant eigenvectors:

4:

With B = 0 we eliminate t between x and y to obtain x/(


The solution decreases like e
a stable direction.

4t

2) = y/1 ) y =

p
x/ 2.

along this trajectory as t ! +1, i.e., towards the origin -

188

CHAPTER 4. SYSTEMS OF EQUATIONS

1:

p
p
With A = 0 we eliminate t between x and y to obtain x/1 = y/ 2 ) y = 2x.
The solution decreases like e
direction.

along trajectory as t ! +1, i.e., towards origin - a stable

The rest of the trajectories can be sketched in locally using the direction field (or continuity).
Continuity in the neighbourhood of the special eigenvector solutions also gives the flow along
the trajectories of the solutions. All the solutions thus flow into the origin, no matter where they
start from. Hence the origin is overall, a stable node.
3

y(t)

-1

-2

-3
-3

-2

-1

x(t)
Figure 4.4: Phase plane plot of the trajectories of solutions x(t), y(t) near to the stable
p node
at (x,py) = (0, 0). The thick lines denote the eigenvector directions along y = x/ 2 and
y = 2x that pass through the equilibrium point. The sign of the arrows on these particular lines
are determined by the eigenvalues associated with the eigenvector. Negative eigenvalues mean
the solutions decrease towards the equilibrium point (0,0) as t ! +1. By continuity along
adjacent trajectories all the other arrows must also point towards the node at the origin and so the
trajectories eventually all pass through the origin. The trajectories can be plotted using the local
direction field. Note that the curvature of the trajectories is to become tangential at the node to
the special eigenvector direction corresponding to the least negative eigenvalue.

4.4. LINEAR EXAMPLES

4.4.2

189

Example: (unstable) saddle

The system,
x = Mx,
has general solutions:
x=A

1
2

M=

3t

e +B

1 1
4 1
1
2

Hence from above we have 1 = +3, 2 = 1 so 1 2 < 0 and we have a saddle point at the
origin. The saddle is organised around the straight lines passing through the centre given by the
associated relevant eigenvectors:

= +3:

With B = 0 we eliminate t between x and y to obtain x/1 = y/2 ) y = 2x.

The solution increases like e+3t along trajectory as t ! +1, i.e., away from the origin an unstable direction.

1:

With A = 0 we eliminate t between x and y to obtain x/1 = y/( 2) ) y =


The solution decreases like e
a stable direction.

2x.

along this trajectory as t ! +1, i.e., towards the origin -

The rest of the trajectories can be sketched in locally using the direction field (or continuity).
Continuity in the neighbourhood of the special eigenvector solutions also gives the flow along
the trajectories of the solutions.
Because there is one unstable direction through the equilibrium point, the saddle is overall
classified as an unstable equilibrium point, even though solutions are attracted to the equilibrium
point along the other trajectory passing through it.

190

CHAPTER 4. SYSTEMS OF EQUATIONS

y(t)

-1

-2

-3
-3

-2

-1

x(t)
Figure 4.5: Phase plane plot of the trajectories of solutions x(t), y(t) near to the (unstable)
saddle at (x, y) = (0, 0). The thick lines denote the eigenvector directions along y = 2x
that pass through the equilibrium point. The sign of the arrows on these particular lines are
determined by the eigenvalues associated with the eigenvector. Negative eigenvalues mean the
solutions decrease towards the equilibrium point (0,0) as t ! +1. Positive eigenvalues mean
the solutions increase away from the equilibrium point (0,0) as t ! +1

4.4. LINEAR EXAMPLES

4.4.3

191

Example: unstable spiral

The system,
x = Mx,

M=

has general solutions:


x=A

5
1 + 2i

(2+2i)t

+B

5
1 3

5
1

2i

e(2

2i)t

Hence we have 1 = 2 + 2i, 2 = 2 2i so i = i!, with = +2 and ! + 2. From above


we have > 0 and thus we have an unstable spiral away the node at the origin.
The eigenvectors here do not explicitly play such an obvious role, since they involve (as
first sight) complex numbers. The sense of the spiral is given by considering a test point and
computing the direction of the tangent vector, e.g., at (x, y) = (1, 0):

x
1
5
1
1
=
=
y
1 3
0
1
Thus, since we know that the motion along the trajectory is (a) unstable, so away from the
origin and (b) tangent to the vector (1, 1) at (1, 0), the spiral must be travelling in a clockwise
direction.
The rest of the trajectories can be sketched in locally using the direction field and continuity
of the flows along the trajectory.

192

CHAPTER 4. SYSTEMS OF EQUATIONS

y(t)

-1

-2

-3
-3

-2

-1

x(t)
Figure 4.6: Phase plane plot of the trajectories of solutions x(t), y(t) near to the unstable node
at (x, y) = (0, 0). The direction of the spiral is deduced as in the main text. The arrows are then
added to all the trajectories by using continuity of the directions of flow between them.

4.4. LINEAR EXAMPLES

4.4.4

193

Example: centre

The system,
x = Mx,

M=

2
1 0

where 2 R. This has general solutions:

i
i
it
x=A
e +B
e
1
1

it

Hence we have 1 = i, 2 = i so i = i!, with = 0 and ! + . From above, since


we have complex eigenvalues with = 0 we thus have a centre.
The sense of travel around the centre is given by considering a test point and computing the
direction of the tangent vector, e.g., at (x, y) = (1, 0):

x
0
2
1
0
=
=
y
1 0
0
1
Thus the direction of travel is tangent to the vector (0, 1) at (1, 0), the trajectories around the
centre must be traveling in a clockwise direction. The rest of the trajectories can be sketched in
locally using continuity of the direction flow.
(If the principal axes of the elliptical centres are not parallel to the coordinate axes, then
further analysis of their eigenvalues can identify their directions.)

194

CHAPTER 4. SYSTEMS OF EQUATIONS

y(t)

-1

-2

-3
-3

-2

-1

x(t)
Figure 4.7: Phase plane plot of the trajectories of solutions x(t), y(t) near to the (neutrally stable)
centre at (x, y) = (0, 0). The direction of the spiral is deduced as in the main text. The arrows
are then added to all the trajectories by using continuity of the directions of flow between them.

4.5. NONLINEAR SYSTEMS

4.5

195

Nonlinear Systems

Consider the nonlinear system we looked at above


dx
= x y,
dt
dy
= y(x 1).
dt
This has two sets of equilibrium points:
(x, y) = (0, 0) or (1, 1).
To explore the behaviour of the trajectories near to an equilibrium point C at say, (xc , yc ) it
is usual to shift the origin to C. This can be done by a linear translation of variables to local
coordinates (, ) centred on C:
x = xc +
y = yc +
for ! 0 and ! 0.
We proceed by substituting this into the equations of the full system then linearising the
equations in terms of the (assumed small) displacements and about C, i.e., neglecting the
quadratic and higher terms in and .
Applying this to the above system we have the following:

4.5.1

Behaviour near to (0, 0)

Here we can just linearise around (xc , yc ) = (0, 0) by assuming that we are so close to the origin,
that any powers of x, y greater than 1, or products of x and y, can be neglected with respect to
terms that are just linear in x and y.
To that end we have
dx
= =x
dt
dy
= y(x
dt

y
1) = xy

y = O(xy)

Thus near to the equilibrium point at (0, 0), the system behaves like:


x
1
1
x
=
y
0
1
y

196

CHAPTER 4. SYSTEMS OF EQUATIONS

This local system has eigenvalues: 1 = 1, 2 = +1. Thus we have 1 < 0 and 2 > 0. Hence
after comparing with the classification above, we see that (0, 0) is a locally like a saddle.
A further bit of analysis gives the corresponding eigenvectors as

1
1:
)
y = 2x is stable axis of the saddle.
1 =
2

1
)
y = 0 is unstable axis of the saddle.
2 = +1 :
0

0.4

0.2

y(t)

0.0

-0.2

-0.4

-0.4

-0.2

0.0

0.2

0.4

x(t)
Figure 4.8: Phase plane plot of the trajectories of linearised solutions x(t), y(t) of the full nonlinear system near to the (unstable) saddle at (x, y) = (0, 0).

4.5. NONLINEAR SYSTEMS

4.5.2

197

Behaviour near to (1, 1)

Here we linearise around (xc , yc ) = (1, 1) by setting:


x = 1+
y = 1+
for ! 0 and ! 0.
Substituting this into the system of equations we have:
d(1 + )
= (1 + ) (1 + )
dt
d(1 + )
= (1 + )(1 + 1).
dt
Hence we have:
d
=
dt
d
= (1 + ) = + O() = .
dt
Thus near to the equilibrium point at (1,1), the system behaves like:


1
1

=
1 0

p
p
This local system has eigenvalues: 1 = (1 + i 3)/2, 2 = (1 i 3)/2. Thus we have
eigenvalues of the form 1 = + i! and 2 = i!, i.e., complex conjugates with = +1.
Hence after comparing with the classification above, we see that (1, 1) is locally like an unstable
spiral.

198

CHAPTER 4. SYSTEMS OF EQUATIONS

0.4

0.2

(t)

0.0

-0.2

-0.4

-0.4

-0.2

0.0

0.2

0.4

(t)
Figure 4.9: Phase plane plot of the trajectories of linearised solutions (t), (t) of the full nonlinear system near to the (unstable) saddle at (, ) = (0, 0). Note that the original and local
chordates are created by the linear translations x = + xc = + 1 and y = + yc = + 1.

4.5. NONLINEAR SYSTEMS

4.5.3

199

Patching up the local behaviours

Remember that the two linear systems above are the approximate behaviours of the original
nonlinear system in the vicinity of the two equilibrium points at (x, y) = (0, 0) and (1, 1). To get
the global behaviour of the full nonlinear solution we now have to patch the two local behaviours
to obtain an approximate representation of the global behaviour in the (x, y) phase plane. In the
absence of a full numerical (or analytical) solution, this is often done graphically, by placing the
two local behaviours in the correct relative position and joining the curves up using continuity
of the trajectories and flows.

y(t)

(t)

(1, 1)

(0, 0)

(t)

x(t)

Figure 4.10: Phase plane plots of the trajectories of linearised solutions in the vicinities of the
two equilibrium points, but plotted in their relative positions on the global phase plane (x, y).
Also shown are the axes of the local coordinates (, ) centred on the unstable spiral at (x, y) =
(1, 1). The final task is to patch the two together using continuity of the trajectories and flows by
sketching.

200

CHAPTER 4. SYSTEMS OF EQUATIONS

In this case it is not too difficult to do and we can arrive at the final sketch of the approximate
behaviour of the global phase plane of the full nonlinear solution as below. The final result shows
that the locally stable trajectory entering the saddle at (0, 0) is actually becomes one of the locally
unstable trajectories emerging from the node at (1, 1).
3

y(t)

-1

-2

-3
-3

-2

-1

x(t)
Figure 4.11: Phase plane plot of the trajectories of the solution of the full non-linear system
obtained byjoining up the the trajectories and observing the continuity of the flows in Figure
4.8.

Das könnte Ihnen auch gefallen