Math 1280

Notes 3

1 The pendulum equation

We now come to a particularly important example, the equation for an oscillating
pendulum. This topic rst appears in the text in Chapter 1, section 1.3.
In elementary physics it is usually said that the equation of a undamped pendulum
is something like
00
+ ! 2 = 0:
Here represents the angle made by the swinging pendulum with the vertical and !
is a nonzero constant. But this is a considerable simpli cation { we will see where
it comes from later. The correct equation can be derived by looking at the geometry
of the forces involved. Three derivations are given in the problems in section 1.3.
On page 21, the equation is given as
00 g
+ sin = 0:
L
Here g is the gravitational force, and L the length of the pendulum. Note that the
mass of the pendulum does not appear.

I will assume that Lg = 1, which is unlikely but will simplify the equations. We
switch to a system, setting = x; 0 = y. This gives
x0 = y
(1)
y 0 = sin x
The equilibrium points are easily seen to be
(0; 0) ; ( ; 0) ; (2 ; 0) ; :::; ( ; 0) ; ( 2 ; 0) ; :::
So there are in nitely many of them. But we will see that there are only two types
of behavior. The linearized system at (x0 ; y0 ) is
u0 = v
v 0 = cos (x0 ) u
At (0; 0) we get
u0 = v
v0 = u

1
which give circles, and (0; 0) is a center. At ( ; 0) we get

u0 = v
v0 = u

and this gives a saddle. As we move along the x-axis, we alternate between centers
and saddles for the linearized system.

Before guring out what these tell us, look at the linearized system around (0; 0).
If we use the usual method of coming up with an equivalent 2nd order equation, we
get
u00 = u;
or
u00 + u = 0:
This tells us where that simpli ed equation
00
+ !2 = 0

came from. It is just the linearized equation, where ! 2 = Lg has been set equal to 1.
This means that this equation is valid near x = 0; y = 0; which is the same as saying
and 0 are close to zero, but not necessarily valid for large oscillations. In fact, it
is not valid for large oscillations.

Now we have to consider whether the nonlinear system (1) has the same local
picture near the equilibrium points as the linearized system { the u; v system. As
discussed in class, and in section 9.3, and summarized in Table 9.3.1, the saddle points
will de nitely remain saddle points, but we can't be sure about centers. They might
become stable or unstable spirals.

To learn more, we look for what we will call an \energy function". There are
various ways to derive such a functions. In the previous notes we had another system,
and we consider x2 + y 2 as the energy function. Here something a bit di erent is
needed.

There are a number of ways to derive the needed function, each involving some
sort of \trick", which you probably would not think of yourself. One way is to think
of y as a function of x, and write (1) as a rst order equation, valid as long as x0 6= 0.

2
From the equations
dx
= f (x; y)
dt
dy
= g (x; y) ;
dt
we expect that the derivative of y with respect to x should be given by dy = dx . This
dt dt
seems dubious mathematically, but if we try it on (1), we get a separable equation
which we can solve.
dy sin x
=
dx y
Z Z
ydy = sin x dx
1 2
y = cos x + c
2
for some scalar c. Therefore, we choose the energy function H (x; y) = 21 y 2 cos x,
and the phase curves are the graphs of the equation H (x; y) = c. We will see how
to do this a bit later.

The derivation was suspicious, so we should check the answer. Does the equation
H (x; y) = c give us correct trajectories? We can calculate dtd H (x (t) ; y (t)) when
(x (t) ; y (t)) solves (1):

d 1 2
y cos x = 2yy 0 + (sin x) x0
dt 2
= 2y ( sin x) + (sin x) y = 0: (2)

Thus, H (x (t) ; y (t)) is constant. We say that H is \constant along trajectories."

Before seeing what this tells us about the phase plane, here is another way to
derive the function H: Our equations are

x0 = y
y 0 = sin x

Write this instead as

x00 + sin x = 0:

3
Multiply this by x0 :
x0 x00 + x0 sin x = 0: (3)
Then integrate each side. To do this you have to nd antiderivatives for x0 x00 and
x0 sin x. You should check, using the chain rule, that
d1 0 2
x (t) = x0 (t) x00 (t)
dt 2
d
cos x (t) = x0 (t) sin x (t)
dt
Hence, using (3), dtd 12 x0 (t)2 cos x (t) = x0 (t) (x00 (t) + sin x (t)) = 0: Since x0 = y;
this gives the same function H.

We use H to plot the phase plane. By Theorem 9.3, the saddle points remain
saddle points.
Now let's evaluate H at the equilibrium points. H (0; 0) = 1; H ( ; 0) = 1;
H ( ; 0) = 1;and so H = 1 at 2k and +1 at (2k + 1) , for each integer
k. Notice that 1 is the minimum possible value of H; but there is no maximum
possible value, because of the term 21 y 2 .
To get the complete phase plane we must consider the graph of the equation
H (x; y) = c for every allowable value of the constant. c:
First suppose c = 0. Then 21 y 2 cos x = 0. Thus,
p
y= 2 cos x:
p
This gives y = 0 at x = 2 and x = 2 , and y = 2 when x = 0. For 0 < x < 2 ;
dy sin x
dx
= y
< 0; and so y decreases. But H is symmetric around the x and y axes
(why?), and so we can ll in a complete loop around (0; 0). Also, H has period
2 in x; and this allows us to reproduce this loop around (2 ; 0) ; ( 2 ; 0) ; and so
forth.

plot of 12 y 2 cos x = 0
Notice next that if c < 1 then there are no points, because H (x; y) can't be
any smaller than 1: And if c = 1 then we must have y = 0; cos x = 1; so
x = 0; 2 ; 2 ; etc.

4
 For any value of c, we see that we always have the symmetry around the x and
y axes. But a spiral cannot have that symmetry. Choosing values of c between 1
and 0 gives a sequence of ellipse-like curves by the same argument as when c = 0:
So we must have centers, not spiral points.
Since ( ; 0) and ( ; 0) are saddle points, a trajectory with H (x; y) = 1 connects
these two points. This is then a heteroclinic orbit, as was seen in a previous example.
Again, this repeats below the y axis, and every 2 in x .

plot of 12 y 2 cos x = 1
Now we combine these on one plot:

plot of 12 y 2 cos x = c for c = 0; 1

Finally, what happens if c > 1? Then, we see that y is never zero, since 12 y 2 =
c + cos x and jcos xj 1: Here is a computer plot of 21 y 2 cos x = 2; added to the
previous picture.

5
 plot of 12 y 2 cos x = 0; 1; 2

The diagrams above have no arrows. But these are easy to supply. For y > 0
the direction is to the right, while for y < 0 it is to the left.
Next, we add damping. This time we will add a linear damping term, giving the
equation
x00 + kx0 + sin x = 0;
where k > 0: As usual, write as a system:

x0 = y
y 0 = sin x ky:

The equilibrium points are the same as before. First we linearize around (0; 0) . I
will denote the Jacobian matrix,

fx fy
gx gy

by DF: Then we have

0 1
DF(x;y) = :
cos x k
0 1
so at (0; 0) our matrix A = DF(0;0) is : The eigenvalues turn out to be
1 k
1
p p
2
k + 12 k 2 4 and 21 k 12 k 2 4: We will assume that the damping is small,
so that jkj < 2: In this case we get two complex eigenvalues with negative real part.
The equilibrium point (0; 0) is asymptotically stable, with spiraling in orbits, at least
locally.

To determine the global picture we again let H (x; y) = 12 y 2 cos x: This time
we nd that
H_ (x; y) = y ( sin x ky) + sin x (y) = ky 2 :
This is similar to the previous example, x00 + x03 + x3 = 0:1 The function q (t) =
1
2
y (t)2 cos (x (t)) is decreasing and bounded below and so q (t) tends to a limit as
t tends to in nity. However this does not mean that (x; y) necessarily approaches
(0; 0) : Every other equilibrium point along the x-axis { all those of the form (k ; 0)
1
So, for consistent notation, I should have called this function E: But if we don't know ahead
of time how it will come out (H_ = 0 or E_ 0), then we don't know what to call it.

6
where k is an even integer, also is a stable spiral, so nearby solutions wind onto it.
Notice that the energy function H approaches its minimum value of 1.
To check this we can use the computer to draw some solutions. Pplane works
ne, though I used another program below.

In this example also, we could imagine that possibly the inwardly spiraling curves
in this picture don't spiral all the way to zero, but instead spiral onto a curve, maybe
a circle or ellipse, or something close to these. Again one can use the fact that
H_ (x; y) < 0 if y 6= 0 to show that this does not happen.

A more complete picture of the phase plane for the damped pendulum equation
appears at the end of section 9.3.

2 Homework{due January 27 at the beginning of

class.

section 9.3, #19 in 9th edition, # 17 in 8th edition. Part (b) refers to the linearized
system. You might prefer to use u; v variables here to di erentiate from the nonlinear
system. You can nd the exact solution and thus the parametric curves for the
linearized system, but not for the original system.

Section 9.3, # 24, 25 in 9th edition, # 22, 23 in 8th edition. In these you are
asked, among other things, to graph x vs t: Pplane will do this for you. Under
"graph" at the top you can choose to graph either x or y vs t .