Physics 235 Chapter 4

Chapter 4
Non-Linear Oscillations and Chaos

Non-Linear Differential Equations

Up to now we have considered differential equations with terms that are proportional to the
acceleration, the velocity, and the position:

x + f ( x ) + g ( x ) = h ( t )
m

Such an equation is called a linear differential equation. Although many important

applications can be described in terms of a linear differential equation, many other applications
require the inclusion of terms that are higher than linear. In that case, the system is called non-
linear. Differential equations can become non-linear if for example:
• The restoring force is not proportional to the displacement x.
• The damping force is not proportional to the velocity v.
An example of the first case is the pendulum. The pendulum will exhibit simple-harmonic
motion if the angular displacement is small. In that case, the restoring force, which is
proportional of sinθ is approximately proportional to θ. However, if the angular displacement is
large, sinθ is no longer proportional to θ and the system will no longer exhibit simple harmonic
motion.
Systems in which the force is non-linear can be divided in to two groups:
1. Systems for which the force is symmetric around the equilibrium position (see Figure 1).
The non-linear behavior, which is added to the linear behavior, must be proportional to x3
since terms proportional to x2 would reduce the magnitude of the restoring force on one side,
while increases its magnitude on the opposite side:

F ( x ) = −kx + ε x 3

If ε > 0, the magnitude of the force is decreased compared to the linear situation. Such a
system is called a soft system. If ε < 0, the magnitude of the force is increased compared to
the linear situation. Such a system is called a hard system.
2. Systems for which the force is asymmetric around the equilibrium position (see Figure
2). The non-linear behavior, which is added to the linear behavior, must be proportional to x2
since terms proportional to x3 would reduce the magnitude of the restoring force on both
sides:

F ( x ) = −kx + λ x 2

- 1 -
Physics 235 Chapter 4

If λ > 0, the force is increased compared to the linear situation. If λ < 0, the force is
decreased compared to the linear situation. In both cases, the system is hard on one side of
the equilibrium position, and soft on the opposing side.

Figure 1. Examples of forces that are symmetric around the equilibrium position.

Figure 2. An example of a force that is asymmetric around the equilibrium position.

Phase Diagrams for Non-Linear Systems

If the potential energy of the linear system is known, it is easy to construct the phase
diagram. The procedure used to construct the phase diagram relies on conservation of energy:

1 2
E = U ( x) + mx
2

or

- 2 -
Physics 235 Chapter 4

2
x = E − U ( x)
m

Examples of phase diagram for stable and unstable equilibrium positions are shown in Figure 3.
The curves of constant energy are closed when we are dealing with a stable equilibrium, while
they are not closed when we are dealing with unstable equilibrium.

Figure 3. Phase diagrams for stable and unstable equilibrium positions.

The van der Pol Equation

An important type of non-linear differential equation was used to describe the non-linear
oscillations in electronic circuits containing vacuum tubes. The van der Pol differential equation
has the following form:

( )
x + µ x 2 − a 2 x + ω 0 2 x = 0


This equation is non-linear due to the presence of the x2 term. One can interpret this equation as
the equation of a system that has variable damping. In addition, the “damping” coefficient can
be positive and negative, depending on the value of a. Based on what we have seen for damping
in Chapter 3, we expect the following behavior for the solution of this differential equation:

- 3 -
Physics 235 Chapter 4

• If | x | >> | a |, the “damping” coefficient is positive and the amplitude of the motion will
decrease.
• If | x | << | a |, the “damping” coefficient is negative and the amplitude of the motion will
increase.
After a period of time we can imagine that the “damping” coefficient approaches zero and simple
harmonic motion, with constant amplitude, will result. The curve mapped out in phase space in
this case is called the limit cycle, and eventually no matter what the starting point, the system
will follow the limit cycle.
To study the time evolution of the van der Pol equation, we need to use numerical methods to
solve the differential equation. I will illustrate the use of Mathematica to solve this equation.
The following code will allow you to play with the solutions of the van der Pol equation (the
code can be found in the file vanDerPolSolution in the Mathematica folder under Computing
Tools on our website):

(* Set the values of the various parameters *)

mu = 0.05;
a = 1;
w0 = 1;

(* Solve the differential equations with the given set of

initial conditions. \
*)
sol = NDSolve[
{x'[t] == v[t],
v'[t] == -mu*(x[t]*x[t] - a*a)*v[t] - w0*w0*x[t],
x[0] == 4,
v[0] == 0},
{x, v}, {t, 0, 50*2*Pi}, MaxSteps -> 20000];

(* Plot the solution of the differential equations *)

ParametricPlot[Evaluate[{x[t], v[t]} /. sol], {t, 0, 50*2*Pi},

PlotRange -> All,
AxesLabel -> {"x (rad)", "v (rad/s)"}];

The results of this type of calculations, for different starting parameters and damping terms, are
shown in Figures 4 and 5.

- 4 -
Physics 235 Chapter 4

Figure 4. Solution of the van der Pol equation with µ = 0.05, a = 1 rad, ω0 = 1 rad/s, x0 = 4
rad, v0 = 0 rad/s (left) and µ = 0.05, a = 1 rad, ω0 = 1 rad/s, x0 = 1 rad, v0 = 0 rad/s (right).

Figure 5. Solution of the van der Pol equation with µ = 0.05, a = 1 rad, ω0 = 1 rad/s, x0 = 2
rad, v0 = 0 rad/s (left) and µ = 0.5, a = 1 rad, ω0 = 1 rad/s, x0 = 2 rad, v0 = 0 rad/s (right).

Figure 6. Solution of the van der Pol equation with µ = 0.05, a = 1 rad, ω0 = 1 rad/s, x0 = 4
rad, v0 = 0 rad/s (left) and µ = 0.5, a = 1 rad, ω0 = 1 rad/s, x0 = 4 rad, v0 = 0 rad/s (right).

- 5 -
Physics 235 Chapter 4

Looking at Figure 5 we observe that the shape of the limiting curve is a function of the damping
parameter. Figure 6 shows that an increase of the damping parameter decreases the time
required to approach the limit cycle.

The Plane Pendulum

A good example of a system with non-linear behavior is the plane pendulum. In many
introductory physics courses, the simple pendulum is used to introduce harmonic motion.
However, the simple pendulum will only exhibit simple harmonic motion if the angular
displacement is small (small enough to ensure that sinθ is approximately θ). If we can not make
the assumption that the angular displacements are small, the simple pendulum will show non-
linear behavior. The differential equation describing the system is given by

θ + ω 0 2 sin θ = 0

where

g
ω 02 =
l

where l is the length of the pendulum.

Before we look in detail at the phase diagrams for this pendulum, we should first consider
what we expect to see. Let’s assume that at time t = 0 s, the pendulum is located at its
equilibrium position, and it has a linear velocity v0 at that time.
• If the velocity at t = 0 is small, the system will oscillate back and forth between a minimum
and maximum value of the angular displacements, and the phase diagram is a closed curve
(see Figure 7).
• If the velocity at t = 0 is sufficiently large, the system will continue to rotate in one direction,
and the angular velocity will show an oscillatory behavior (but without a sign reversal).
Examples of this response are shown in Figure 8.
• The limiting case will be the case where the velocity at t = 0 is such that the pendulum makes
it just to its highest possible position (y = 2l). Assuming the mass of the pendulum is
connected via a weightless rod to the pivot point, the total energy of the mass will have to be
2mgl (kinetic energy will be equal to 0 at the top position). The kinetic energy at the
equilibrium must therefore be 2mgl, and the linear velocity at that postion will be equal to
2√gl. The angular velocity at the equilibrium position will thus be equal to (2√gl)/l = 2√(g/l)
= 2ω0. For the phase diagrams shown in Figures 7 and 8, we have assumed ω0 = 1 rad/s, and
the “critical” angular velocity is thus equal to 2 rad/s.

- 6 -
Physics 235 Chapter 4

Figure 7. Solution of the differential equation associated with the plane pendulum ω0 = 1
rad/s, x0 = 0 rad, v0 = 1 rad/s (left) and ω0 = 1 rad/s, x0 = 0 rad, v0 = 1.5 rad/s (right).

Figure 8. Solution of the differential equation associated with the plane pendulum ω0 = 1
rad/s, x0 = 0 rad, v0 = 2 rad/s (left) and ω0 = 1 rad/s, x0 = 0 rad, v0 = 2.5 rad/s (right).

Figures 7 and 8 were made using Mathematica and can be reproduced using the following
code, which is also contained in Mathematica notebook planePendulum that is available from
our website:

(* Set the values of the various parameters *)

w0 = 1;

(* Solve the differential equations with the given set of

initial conditions. \
*)
sol = NDSolve[
{x'[t] == v[t],
v'[t] == -w0*w0*Sin[x[t]],
x[0] == 0,

- 7 -
Physics 235 Chapter 4

v[0] == 2},
{x, v}, {t, 0, 50*2*Pi}, MaxSteps -> 20000];

(* Plot the solution of the differential equations *)

ParametricPlot[Evaluate[{x[t], v[t]} /. sol], {t, 0, 5*2*Pi},

PlotRange -> All,
AxesLabel -> {"x (rad)", "v (rad/s)"}];

Another way to look at the phase diagram is to use contour plots. Rather than solving the
differential equation for a specific set of initial parameters, and thus a specific energy, the
contour plot provides an overview of the solution for a range of initial conditions. Consider a
point in the phase diagram corresponding to an angular position θ and an angular velocity ω .
The kinetic and potential energies at this location are:

1 2 1 2 2
T= I ω = ml θ
2 2

and

U = mgl (1 − cos θ )

The total energy E is thus equal to

1 2 2
E = T +U = ml θ + mgl (1 − cos θ )
2

A contour plot, showing the total energy E as function of θ and ω , shows the nature of the
motion that will result when we give the pendulum specific initial conditions. An example of a
contour plot, obtained for m = 1 kg, l = 1 m is shown in Figure 9. This contour plot was
generated using the following Mathematica code, which is also contained in Mathematica
notebook planePendulumContours that is available from our website:

l = 1;
g = 9.8;
ContourPlot[(1/2)*l^2*w^2 + g*l*(1 - Cos[theta]),
{theta, -4Pi, 4Pi}, {w, -15, 15}];

- 8 -
Physics 235 Chapter 4

Figure 9. Contour plot of the phase diagram of the plane pendulum (l = 1 m).

Chaos
Important principles of chaos can be studied using a simple driven pendulum. Consider the
pendulum shown in Figure 10, which is driven about its pivot point. We assume that the
pendulum is damped with a damping coefficient b. The total torque around the pivot point is
equal to

N = Iθ = ml 2θ

The total torque is equal to the sum of the torques due to the gravitational force, the damping
force, and the driving force:

N = −bθ − mgl sin θ + N d cos (ω t )

The differential equation describing the motion of this driven pendulum is given by

b g N
θ = − 2 θ − sin θ + d2 cos (ω t )
ml l ml

- 9 -
Physics 235 Chapter 4

This differential equation is a non-linear equation due to the sinθ term. This equation has the
following general form:

x = −cx − sin x + F cos (ω t )



Figure 10. A damped pendulum, driven about its pivot point.

The solution to this equation can be studied using numerical methods, and we will focus on
the results obtained with Mathematica. The code required to study the solutions of this equation
can be found in the file chaosPendulum in the Mathematica folder under Computing Tools on
our website:

(* Set the values of the various parameters *)

c = 0.2;
w = 0.694;
F = 0.52;
pi = N[Pi];
cycles = 50;
steps = 30;

(* Solve the differential equations with the given set of

initial conditions. \

- 10 -
Physics 235 Chapter 4

*)
sol = NDSolve[
{x'[t] == v[t],
v'[t] == -c*v[t] - Sin[x[t]] + F*Cos[w*t],
x[0] == 0.8,
v[0] == 0.8},
{x, v}, {t, 0, (cycles*(2Pi)/w)}, MaxSteps -> 20000];

(* Create a function "reduce" that translate all angles back to

the region \
between - Pi and + Pi *)
reduce[x_] := Mod[x, 2pi] /; Mod[x, 2pi] <= pi;
reduce[x_] := (Mod[x, 2pi] - 2pi) /; Mod[x, 2pi] > pi;

(* Create a table with the angles mapped to the - Pi to Pi

region. *)
points = Flatten[
Table[
{x[t], v[t]} /. sol,
{t, 0, (cycles*(2Pi)/w), (1/steps)(2Pi/w) // N}
], 1
];
xposition = Table[{i, 1}, {i, Length[points]}];
newpoints = MapAt[reduce, points, xposition];

(* Plot the solution of the differential equations : v[t] vs t.

*)
ParametricPlot[
Evaluate[{t, v[t]} /. sol], {t, 0, (cycles*(2Pi)/w)},
PlotRange -> {{((cycles - 20)*(2Pi)/w), (cycles*(2Pi)/w)},
{-3, 3}},
AxesLabel -> {"t/t0", "v (rad/s)"}
];

(* Plot the phase diagram (using the reduced angle). *)

ListPlot[
newpoints,
PlotRange -> {{-3, 3}, {-3, 3}},
AxesLabel -> {"x (rad)", "v (rad/s)"}
];

- 11 -
Physics 235 Chapter 4

(* Make the poincare plot. *)

poincare = Table[newpoints[[n]], {n, 1 + 25steps,
Length[points], steps}];
Length[poincare];
ListPlot[
poincare,
PlotRange -> {{-3, 3}, {-3, 3}},
AxesLabel -> {"x (rad)", "v (rad/s)"},
PlotStyle -> PointSize[0.015]
];

There are many different ways of looking at the results of this calculation. Consider the case
where c = 0.2, ω = 0.694, and F = 0.52. A graph showing the velocity versus time is shown in
Figure 11.

Figure 11. Velocity versus time for the driven pendulum with c = 0.2, ω = 0.694, and F =
0.52.

The period shown in Figure 11 does not include the initial period when transient effects may
dominate. The motion is periodic, but not simple harmonic. The average velocity is negative,
indicating that on average the solution will move to more negative angles.
Another way to look at the solution is to construct the phase diagram. The phase diagram for
the solution shown in Figure 11 is shown in Figure 12. Figure 12a shows the velocity versus
angle. As expected, the system will move towards more negative angles. Figure 12b shows the
similar information to what is shown in Figure 12a, except that the angle is reduced to an angle
between –π and +π.

- 12 -
Physics 235 Chapter 4

a)

b)

Figure 12. Phase diagrams for the driven pendulum with c = 0.2, ω = 0.694, and F = 0.52.
The diagram shown in b) contains the same information as in a) except that the angles are
mapped on the range between –π and +π.

The phase diagrams shown in Figure 12 are often difficult to interpret. An approach that was
developed to emphasize the possible periodic nature of the solution is the Poincare
representation. In this representation, we plot the velocity versus position at times separated by
one period (2π/ω ). If the motion is periodic, each new point in the Poincare diagram will be
plotted at the same position as the previous point: the plot shows thus a single point. Note: the
number of points shown will provide information about the periodicity; the actual position
does not provide much information. The Poincare representation of the phase diagrams shown
in Figure 12 is shown in Figure 13. There are 3 points in the Poincare representation, suggesting
a periodic solution (but not simple harmonic).
The solution of the driven pendulum is very sensitive to the parameters that are being used in
the calculation. The transition of chaos is characterized by an increase in the number of points in
the Poincare diagrams. Consider what happens if we change one of the parameters of our
differential equation. Figure 14 shows the results of a calculation when we increase F from 0.52

- 13 -
Physics 235 Chapter 4

to 0.62. The solution has become a simple harmonic, as can be seen by the fact that the Poincare
plot contains only a single point.

Figure 13. Poincare representation of the motion of a driven pendulum with c = 0.2, ω =
0.694, and F = 0.52.

Figure 13. Solutions of the motion of a driven pendulum with c = 0.2, ω = 0.694, and F =
0.62.

- 14 -
Physics 235 Chapter 4

Figure 14. Solutions of the motion of a driven pendulum with c = 0.2, ω = 0.694, and F =
0.72.

Figure 15. Solutions of the motion of a driven pendulum with c = 0.2, ω = 0.694, and F =
0.82.

- 15 -
Physics 235 Chapter 4

Figure 15. Solutions of the motion of a driven pendulum with c = 0.2, ω = 0.694, and F =
0.92.

Figures 14, 15, and 16 show the results of increasing the force F to 0.72, 0.82, and 0.92. The
solution is chaotic when F = 0.72 and F = 0.82, but the solution is periodic when F = 0.92.

Logistic Equations
Chaos can also develop when we work with maps. Maps are used to describe the evolution
of a system, and can developed if we know how the future depends on the immediate past. For
example, consider a situation where the value of a system parameter at time tN+1, xN+1, depends on
the value of this parameter at time tN, xN:

x N +1 = α xn (1 − xn )

The function describing this relation is called the logistic equation, and it depends on both x and
on α:

f (α , x ) = α x (1 − x )

- 16 -
Physics 235 Chapter 4

For certain values of α , the system will evolve towards a stable equilibrium while for other
starting value the system may evolve towards more than one possible solution. Two examples,
for two different values of α, are shown in Figure 16.

Figure 16. Examples of the logistic equation map for two different values of α.

In the example shown in Figure 16b, there are two solutions to the logistic map. For higher
value of α there may be more solutions. The best way to view the possible modes of evolution is
to use a bifurcation diagram. In a bifurcation diagram, we plot the value of xn as function of the
parameter α . We can construct a bifurcation diagram using numerical methods, and we will
focus on the results obtained with Mathematica. The code required to study the solutions of this
equation can be found in the file bifurcation in the Mathematica folder under Computing Tools
on our website. A bifurcation diagram for the logistic equation discussed in this section is shown
in Figure 17. The two diagrams in Figure 17 show the bifurcation diagrams that are created for a
starting value x0 = 0.1 (left) and x0 = 0.9 (right). We see that for value of α below 3 there is only
one solution. For values of α between 3 and 3.45, there are two possible solutions. For values
between 3.45 and 3.55 there are four possible solutions. Above α = 3.55 the system has many
possible solutions, and the system is said to be chaotic. However, as Figure 17 shows, for very
specific values of α the system returns back to a highly-ordered state with a small number of
solutions. As can be seen, the bifurcation diagram does not change when we change the initial
conditions, except for the special cases shown in Figure 18 where x0 = 0.0 (left) and x 0 = 1.0
(right).

- 17 -
Physics 235 Chapter 4

Figure 17. Bifurcation diagrams for f(α, x) = α x(1 – x) for a starting value of x0 = 0.1 (left)
and x0 = 0.9 (right).

Figure 18. Bifurcation diagrams for f(α, x) = α x(1 – x) for a starting value of x0 = 0.0 (left)
and x0 = 1.0 (right).

Changes to the logistic equation will change the bifurcation diagram. Consider for example
the following logistic equation:

f (α , x ) = α x (1 − x )
2

The corresponding bifurcation diagram is shown in Figure 19. Many of the features are similar,
except that we see that the range the first doubling of the solutions does not occur until α
exceeds 4.
When the time evolution of a system depends sensitively on the initial conditions, the system
is said to show the characteristics of chaos. In the bifurcation diagrams shown in Figure 17, we
see that a particular system can have non-chaotic behavior for a specific range of parameters and
chaotic behavior for a different range of parameters. A good way to examine the development of
chaos is by looking at the Lyapunov exponent λ . The Lyapunov exponent is based on the
differences between solutions if we make a small change in the initial conditions. Consider two

- 18 -
Physics 235 Chapter 4

starting values of x: x 0 + ε and x 0. The difference between the solutions for these starting
conditions is

d n = f n (α , x0 + ε ) − f n (α , x0 )

It turns out that this difference grows as

d n = ε e nλ

Combining these equations we see that

1 ⎡ f (α , x0 + ε ) − f (α , x0 ) ⎤ 1 ⎡ df n (α , x ) ⎤
n n
1 nλ
( )
λ = ln e = ln ⎢
ε
⎥ = ln ⎢ ⎥
n n ⎣ ⎦ n ⎢⎣ dx x0 ⎥
⎦

Using the chain rule for derivations, we can rewrite this equation as

1 ⎡ df n (α , x ) ⎤ 1 ⎡ df (α , x ) df (α , x ) df (α , x ) df (α , x ) ⎤
λ= ln ⎢ ⎥ = ln ⎢ .... ⎥
n ⎢ dx ⎥ n ⎢
⎣ dx dx dx dx x0 ⎥
⎦
⎣ x0 ⎦ xn−1 xxn− 2 x1

If we take the limit of n to infinity, we get the definition of the Lyapunov exponent

1 n−1 df (α , xi )
λ = lim ∑ ln
x→∞ n dx
i=1

Figure 19. Bifurcation diagram for f(α, x) = α x(1 – x)2 for a starting value of x0 = 0.1.

- 19 -
Physics 235 Chapter 4

A plot of the Lyapunov exponent as function of α for the logistic function f(α, x) = α x(1 – x) is
shown in Figure 20.

Figure 20. Lyapunov exponent as function of α for the logistic equation f(α, x) = α x(1 – x).

If the Lyapunov exponent is negative, the solutions will eventually converge. In Figure 20 we
can thus easily identify the region where a stable solution is achieved. Bifurcation (doubling of
the number of solutions) will occur when the Lyapunov exponent is zero. Chaos will occur
when the Lyapunov exponent is larger than 0.

- 20 -