Non-Linear Dynamics Homework Solutions

Week 7
February 23, 2009
7.3.1 Consider the system

x =
y =

x y x(x2 + 5y 2 )
x + y y(x2 + y 2 ).

a) Classify the fixed point at the origin.

We first compute the Linearization matrix using the Jacobian

J|O =

1 3x2 5y 2
1 2yx

1 10yx
1 x2 3y 2

1
1

1
1

which has T = 2 and D = 2, implying that the origin is an unstable spiral since T 2 4D =
4 < 0.
b) Rewrite the system in polar coordinates using rr = xx + y y and = (xy y x)/r
2.
From the first equation we get
rr

=
=
=
=

x2 xy x2 (x2 + 5y 2 ) + yx + y 2 y 2 (x2 + y 2 )
x2 + y 2 x4 y 4 6x2 y 2
r2 ((x2 + y 2 )2 2x2 y 2 ) 6x2 y 2
r2 r4 4r4 cos2 sin2

Note that in this simplification we have used the facts that x2 + y 2 = r2 , x = r cos and
y = r sin . Dividing this final equation through by r2 we find that
r = r r3 (1 + 4 cos2 sin2 ).
We also find in a similar manner that
r2

= (x2 + yx yx(x2 + y 2 )) (yx y 2 yx(x2 + 5y 2 ))

= r2 + 4xy 3
= 1 + 4r2 cos sin3

2
1.8
1.6
1.4
1.2

Figure 1: Plot of the periodic term from 7.3.1 vs.

c) Determine the circle of maximum radius, r1 , centered on the origin such that all trajectories have a radially outward component on it.
This task is equivalent to finding the maximum radius for which r > 0 for all . If we
look at our equation for r,
we can see that the bigger the periodic term (1 + 4 cos2 sin2 )
is, the harder it will be to make r > 0, so we want to find the maximum of the term
inside the parenthesis, since well our r1 to make r > 0 for that maximum, but will also
be guarenteed that r > 0 for all other values, since they would only make the periodic
term less, and hence, make r even larger. To find this maximal value, we could resort to
analytic techniques, ie. find the derivitive of the periodic term with respect to , set that
equal to zero and then figure out which of the resulting critical points give us maxima and
which give us minima. There may even be an easier analytic method involving geometric
of trigonometric aproaches, but a graphic argument is really easy and quite convincing.
Taking a look at Figure 4, we can se that the maximum value of this term is 2. We
can therefore set that term equal to 2 in our equation and see what values of r makes r
positive. This brings us to solving the inequality

0 < r

=
=

r 2r3
r(1 2r2 )

p
Since this expression for r is a downwardp
facing cubic with x intercepts at r = 1/2
and r = 0 we can deduce that for all r < 1/2 but greater than zero, r > 0. See Figure
5 for p
graphical clearification on this arguement. Thus we get the desired value by setting
r1 = 1/2 for any positive but small as we like .
d) Determine the circle of minimum radius, r2 , centered on the origin such that all trajectories have a radially inward component on it.
We proceed here as we did in part (c), only here we want the smallest r for which r is
negative for all . Consequently, we want to find the minimum of (1 + 4 cos2 sin2 )
instead of the maximum. By applying analytic methods such as those described in part
(c), or looking at Figure 4 again, we can see that the minimum of the term is 1. We use
this in the equation for r as we did above, and find that if r = r(1 r2 ) then we will have
a negative value of r so long as r > 1. Thus we may set r2 = 1 + for some small but
positive .
e) Prove that the system has a limit cycle somewhere in the trapping region r1 < r < r2 .
2

0.2
0.1

0.2

0.4

0.6

0.8

-0.1
-0.2
-0.3

Figure 2: Plot of r versus r

Proof. By our construction we have a trapping region that satisfies the conditions of the
Poincare-Bendixson Theorem. Consequently, there exists a limit cycle in the region in
question.
7.3.7 Consider the system

x =

y + ax(1 2b r2 )

x + ay(1 r2 )

where a and b are parameters (0 < a 1, 0 b < 1/2) and r2 = x2 + y 2 .

a) Rewrite this system in polar coordinates.
We proceed as we did in problem 7.3.1.
rr
r

r2

= xx + y y
= (ax2 + ay 2 )(1 r) 2bax2
= ar(1 r2 2b cos2 )

= x2 + ayx(1 r2 ) y 2 ayx(1 2b r2 )
= 1 + 2ba cos sin

b) Prove that there is at least one limit cycle, and that if there are several, they all have the
same period T (a, b).
As in problem 7.3.1, we find r1 and r2 with the properties we need to have a trapping
region. We want to maximaize the periodic term 2b cos2 , which happens when = k
for some integer k. At these points the term in question equals 2b, so we substitute this
2
2
into our
equation for r and find that r = (1 r 2b) = ar((1 2b) r ), so if we set
r1 = 1 2b , for some small but positive we get what we need.
Since the minimum of the periodic term is 0, we can use this to find a suitable r2 . We
get r = ar(1 r2 ) so if we let r2 = 1 + for some small but positive , then we have our
outer trapping region boundary.
3

By the Poincare-Bendixson Theorem, and the existence of a trapping region for this
system, the system must have at least one limit cycle in the region inbetween r1 and r2 .
Reguarding the period of such trajectories, note that depends only on and not on r.
Consequently, for any two initial conditions with the same initial value, the trajectories
will make a full revolution in the same amount of time, since their angular positions are
governed by the same dynamical equations.
c) Prove that for b = 0 there is only one limit cycle.
Proof. When b = 0 our inner trapping radius will be r1 = 1 and our outer will be
r2 = 1 + . For two limit cycles to exist, there must be a radial distance dr between them
for any given value of , but we can make the values of as low as we like, so low that
2 < dr so that there is no way both could fall inside the trapping region.
7.5.1 For the van der Pol oscillator with >> 1, show that the positive branch of the cubic nullcline
begins at xA = 2 and xB = 1.
The start of the positive branch begins at the local minimum on the positive side of the graph,
and is considered ended at the point along that branch which is the same height as the local
maximum on the negative branch (See Figure 7.5.1 in Strogatz). So, to find xA , we take the
derivative of the cubic nullcline and find where it is equal to zero, since this locates the local
minimum of the nullcline. Since F 0 (x) = x2 1, our critical points are at 1, so xA = 1. Now
we figure out how high the local maximum of the cubic nullcline is: F (1) = 1/3 + 1 = 2/3.
Now we set F (x) = 2/3 and find the other solution (the one that is not x = 1). This
can be done by plugging the equality into mathematica, or by knowing that if we divide the
polynomial F (x) 2/3 = 0 by (x + 1) then it leaves us with a quadratic polynomial which will
have all of the roots other than x = 1. Choosing the former method, we find that x = 2 is
also a solution to F (x) 2/3 = 0, and so xB = 2.
7.5.3 Estimate the period of the limit cycle of x
+ k(x2 4)x + x = 0 for k >> 1.
We shall solve this by defining z = x 1, which gives us a form in some ways more similar to
what is given in the book, but in others is a bit different. Doing this, we find that
z + k((z + 1)2 4)z + z = 0.
Then following the process of example 7.5.1, we notice that since
1
z + k z((z

+ 1)2 4) = d/dt(z + k[ (z + 1)3 4z]),

3
we can set
F (z) =

1
(z + 1)3 4z
3

so that if we let w = z + kF (z), then

w = z + k z((z

+ 1)2 4) = z.
(Note that the difference between the way we did it here and the way we could have done had
we left things in terms of x is that if we had left things, we would have w = 1 z instead).
Now we let y = w/k so that z = k[y F (z)] and y = z/k.

Now, the period is given by

tB

T =2

dt,
tA

since the time of the period is approximately twice the time it takes to get through just one
of the slow branches. Now since
dy
dz
dz
F 0 (z)
= ((z + 1)2 4) .
dt
dt
dt
Now since dy/dt = z/k we get dz/dt = z/(k((z + 1)2 4), from which it follows that
dt

k((z + 1)2 4)
dz.
z

Now we must find the limits of integration. We do this just as we did for the last excercise.
Since F 0 (z) = z 2 + 2z 3 has zeroes at z = 1 and z = 3. We want the positive one of these
for our zA value, and for zB we see that F (3) = 9 + 1/3. The value of zB will be the other
value of z for which F (z) = 9 + 1/3. Solving this equation (using mathematica or polynomial
division) for z we find that if z 6= 3 then z = 3. Thus zB = 3 so we must integrate from 3 to
1 (because of the direction of the flow along this part of the nullcline we want the upper limit
to be the smaller one). Thus we find using mathematica to integrate for us that
Z

(z + 1)2 4
dz
z
3
2k(8 ln 27).
2k

7.5.6 (Biased van der Pol) Suppose the van der Pol oscillator is biased by a constant force: x
+
(x2 1)x + x = a, where a is some real parameter and > 0 as usual.
a) Find and classify all the fixed points.
We first write out our system in Lienard form. First we define F (x) so that
x
+ (x2 1)x =

d
(x + F (x)),
dx

which works out if F (x) = x3 /3x, just as it did in the non-biased case. Then everything
works out the same as it did in the non biased case (see Example 7.5.1 from Strogatz),
only that in our case, working everything through as is the afore mentioned example, we
find that y = (x a)/. Thus, our system becomes

x
y

= [y F (x)]
xa
=

Consequently, our x-nullcline becomes y = F (x) and our y-nullcline becomes x = a.

Thus our intersection and only fixed point is at x = (a, F (a)) = (a, a3 /3 a). Next we
compute the Jacobian matrix and evaluate at x
5

Figure 3: Plot of the nullclines for problem 7.5.6

J|x =

(x2 1)
1/
0

=
x

(a2 1)
1/

From this we find that our trace and determinent are going to be given by T = (a2 1)
and D = 1. Since our determinent is always positive we are never going to have any
sort of saddlish behaviour. Furthermore, we have that the system is unstable when
1 < a < 1 and unstable otherwise. Furthermore, for sufficiently high values of a the
value of T 2 4D = 2 (a2 1)2 4 will be positive, implying that the fixed point will be
a node. Clearly, this cant always be the case, since plugging in a = 1 readily checks out
as giving us a negative value for T 2 4D, impplying that we do have spiral behaviour for
certain values of a.
b) Plot the nullclines in the Lienard plane. Show that if they intersect on the middle branch
of the cubic nullcline, the corresponding fixed point is unstable.
See Figure 1 for a numerically produced plot given that a = .4 and = 5, which has been
laid on top of a vector field of that system.
To see that the fixed point in cases like these are unstable, we resort to the work we did
for part (a) with respect to finding stabilities of fixed points for various regions.
c) For >> 1, show that the system has a stable limit cycle if and only if |a| < ac , where
ac is to be determined. (Hint: Use the Lienard plane.)
For |a| < ac = 1, the y-nullcline intersects the x-nullcline on the middle branch of the
cubic nullcline. We know that when this is the case the intersection (fixed point), is
unstable, so all initial conditions move away from from that point. Furthermore, they all
move toward the cubic nullcline, and once there, they will move up along the nullcline if
it has reached the negative leg, and will move down if on the positive leg. Eventually, the
trajectory will reach either a local maximum or a local minimum (depending on which
leg you are on). I for instance, you are on the positive leg and get to the minimum, you
will move just past where the cubic nullcline is keeping you from quickly shooting past
it and into the negative x realm, but eventually it will hit the negative leg of the cubic

Figure 4: Plot of the nullclines for problem 7.5.6

nullcline, and then move slowly upwards, then hit the peak and move quickly over to the
positive leg, and then repeat. See Figure 2 for a numerical example.
If, however |a| > ac , there can be no stable limit cycle, since the x nullcline intersects
the cubic nullcline along one of the positive branches, and consequently, once a trajectory
hits that nullcline, it will move slowly along it and toward the intersection (since if it hits
below the intersection, then x < a so x is positive and similarly negative if x > a). Thus
since all trajectories end up moving slowly along nullclines and toward the fixed point,
there can be no stable limit cycles. See Figures 3 and 4 for examples.
d) Sketch the phase portrait for a slightly greater than ac . Show that the system is exciteable.
See Figure 5 for the phase portrait in the case when a = 1.1. Notice that all of the
trajectories end up getting stuck near the local minimum of the cubic nullcline. Now, if a
slight distubance moves the state of the system to just below the fixed point then it will
shoot accross to the negative leg of the cubic, and then go along the whole of the cycle
but get stuck again at the globally attracting rest state.
7.6.2 (Calibrating regular perturbation theory) Consider the initial value problem x
+ x + x = 0,
with x(0) = 1, x(0)

= 0.
a) Obtain the exact solution to the problem.
We rearange to get the equation x
= (1
+ )x, which is easily seen, given the initial
conditions, to have the solution x(t) = cos( 1 + t).
b) Using regular perturbation theory, find x0 , x1 , and x2 in the series expansion x(t, ) =
x0 (t) + x1 (t) + 2 x2 (t) + O(3 ).
We find that for the differential equation to hold for the expansion, we must have that
x
0 +
x1 + 2 x
2 = (1 + )(x0 + x1 + 2 x2 ).
From this it follows that
[
x0 + x0 ] + [
x1 + x0 + x1 ] + [
x2 + x1 + x2 ]2 = 0.
7

Figure 5: Phase portrait for a > ac

Figure 6: Phase portrait for a >> ac

Figure 7: Phase portrait for a slightly greater than ac

The key in using this is that for this sum to be equal to zero, each of the bracketed terms
has to be equal to zero; in addition, each of the bracketed terms must satisfy the initial
conditions. Solving for x0 we have the diffeq x
0 = x0 , which, given the initial conditions,
is clearly seen to have the solution x0 (t) = cos t. Our next differential equation has x0 in
it, so we write it as x
1 + cos t + x1 = 0. We know how to solve this sort of differential
equation from last quarter, but Mathematica is good for this sort of thing. It gives the
solution x1 (t) = cos(t) 12 t sin(t), (after running FullSimplify). Now we get to use this
equation in our next differential equation, which is x
= cos(t) 21 t sin(t) cos(t). The
solution, which I got from mathematica, is x2 (t) = cos(t) 18 t2 cos(t) 38 t sin(t).
c) Does the perturbation solution contain secular terms (these are terms that grow without
bound as t )? Did you expect to see any? Why?
Yes, both x1 and x2 grow without bound as t , because of the t and t2 terms in their
equations. We shouldnt expect to see any in a good aproximation to the solution, since
the actual solution is not secular.