Anda di halaman 1dari 10

98

5.3 Poincar Surface-of-Section Plots

How do we handle higher dimensional systems that are autonomous, that is if the forces or the
potentials are not time-dependent? Nonlinear dynamics scientists use a variety of tools to study
such systems, including laboratory experiments, numerical simulations, and perturbation and
asymptotic analytic methods. Here, we will consider the simplest approach: numerical
simulation used to produce a different type of Poincar section: the Poincar surface-of-section,
or SOS for brevity.

First, it is worth reminding ourselves that one degree-of-freedom systems (i.e. with a 2-D phase
space) are relatively simple: weve already studied methods to understand them, including
finding integrals of motion, calculating the phase portrait, and applying fixed point analysis. If
you think back, however, youll realize that every example we considered entailed only regular
motion - in fact, it can be shown that motion with one degree of freedom cannot be chaotic.

On the other hand, we did find chaotic motion in the driven pendulum and in the cos 2 potential
example above. Both of these have an effective phase space dimension of 3 (2 space + 1
extended proportional to time). A 3-D phase space is the minimum dimension that allows for
chaotic motion. For a time-independent force in Newtons second law, the phase space can only
have an even dimension, so we begin with 4-D phase spaces, i.e. two degrees of freedom.

You might be thinking, well, since the 4 dynamical variables depend on time, we can just use the
extended phase space and visualize orbits and make a strobe plot. But a strobe plot for a 4-D
phase space would be 4-dimensional! Not so easy to visualize. The Poincar surface-of-section
does not use the extended phase space, but even it works best with a 3-D phase space. How do
you then visualize a 4-D or a 6-D phase space? Remember integrals of motion?

Even though chaotic systems dont have all the integrals of motion needed for a complete
solution, you can nonetheless often find some first integrals. And if you have an N-dimensional
phase space you need only find N3 integrals to reduce your problem to 3-D. Let's assume we
have already reduced the dynamical system to 3-D and see how the method works.

5.3.1. SOS Basics

To apply Poincares surface-of-section technique, you will need:


1. A 3D phase space.
2. A convenient surface embedded in phase space through which orbits pass repeatedly.
3. A numerical solution for particle orbits.

Example: Conservative force system with 2 degrees of freedom (4D phase space):
1a. Therell be two coordinates, call them x and y, and 2 velocities, v x and v y , yielding a
4D phase space, ( x, y,v x ,v y ) .

1b. For a conservative system there is a potential energy function U, and the mechanical
energy E is conserved

PHY 380.03 Spring 2013 2013 R. Martin


99

E = 12 m(v x2 + v y2 ) + U ( x, y ) = constant

1c. Use this energy integral (or another integral if there is one) to eliminate one of the
variables and reduce the phase space to a 3D subspace of the original 4D space. For
example, eliminate v y using conservation of energy

2
v y2 =
m
[ E U ( x, y )] v x2
This equation relates v y to the other three variables. If we substitute the right-hand
side in to the equations of motion for each occurrence of v y , we'll reduce the system to
3 variables: x, y, and v x . We've now met condition (1) for applying the SOS technique:
we have our 3-D phase space, ( x, y,v x ) .

2. Choose an SOS surface: It's best to familiarize yourself with some orbits first so you
become familiar with the types of motion possible. Then choose a convenient surface
through which the trajectories pass repeatedly. Suppose youve done that and you
know that orbits regularly pass through the surface y = y * = constant . Then that would
be a good choice for the SOS surface.

3. Create the SOS plot: Youll need to follow many orbits numerically, and determine the
values of the remaining variables, ( x,v x ) , each time the orbit passes thru the SOS plane
y = y * = constant with positive v y . The positive v y restriction is needed so structures
on the plot dont overlap (it can be relaxed for systems with appropriate symmetry).
The surface-of-section, or SOS, plot is the locus of all points determined in this way,
i.e. you plot points in the xv x -plane at the locations where the orbits hit the plane. Like
stroboscopic plots, youll usually need lots and lots of points to obtain a useful SOS,
which requires integrating many orbits for long periods of time.

This example shows that any two-degree-of-freedom conservative system can be studied with the
SOS technique. In fact, you dont necessarily have to actually do the math described in step
(1c): its sufficient to know that it can be done. Practically speaking, all you need to do is steps
(2) and (3), making sure that every orbit you calculate has the same value of the conserved
quantity, usually the energy E. Thus its often useful to rewrite your initial conditions in terms
of E.

Its also worth noting that the SOS method works for higher dimensional Newtonian systems, as
long as you can find a sufficient number of integrals of motion to reduce the phase space to 3D.
So a three degree-of-freedom system (6D phase space) requires knowledge of 3 integrals of
motion to be usefully studied by the SOS technique.

PHY 380.03 Spring 2013 2013 R. Martin


100

5.3.2. Interpreting SOS plots

SOS plots are interpreted in a similar way to stroboscopic Poincar section plots, but its a little
easier to draw the possible orbits and their SOS for this case, so Ill try. Let's look at a system
with a 3D phase space (, choosing the SOS plane to be the y = 0 plane.

1. Periodic orbits are represented by a discrete set of points. Here are two examples::
Simple periodic orbit: S.O.S. plot is a single point

Doubly periodic orbit: S.O.S. plot has two points that are visited each time around:

PHY 380.03 Spring 2013 2013 R. Martin


101

Multiply periodic orbits: S.O.S. plot has discrete set of points that is repeated forever,
usually forming a well-defined pattern - Ill let you imagine this one -- its too hard to
draw.

In practice, periodic orbits are very rare on SOS plots since periodic orbits usually
represent a set of measure zero in phase space and are notoriously difficult to locate
numerically.

3. Quasi-periodic orbit : SOS plot looks like a closed curve.

The typical example of quasiperiodic motion is motion spiraling around the surface of a
torus (a donut), where the SOS plane slices the donut in two. This is what Ive attempted

PHY 380.03 Spring 2013 2013 R. Martin


102

to draw above - the part of the torus (and the orbit) below the y = 0 plane is not shown for
clarity - you are to imagine the orbit taking a helical path along the torus surface and
plotting a point each time it hits the y = 0 plane going upward. Ive labeled the odd-
numbered SOS points to show that the orbit is not periodic - point 9 is a little beyond the
first point 1, and if the orbit were to keep circling the torus it would eventually fill in a
closed curve. Even though the orbit is not strictly periodic, it is still well-behaved and
regular. These are the quasi-periodic orbits we saw in the strobe plots.

If you dont like my artwork, we can try to visualize a quasiperiodic orbit SOS with the
help of Mathematica. Heres an orbit in a 3D phase space (1 , 2 , 2 ) that has a distorted
torus shape (this one actually comes from a double pendulum simulation):

If we choose the plane y = 0 as our SOS plane, we intersect this orbit with that plane:

PHY 380.03 Spring 2013 2013 R. Martin


103

If you're used to these kinds of plots you can guess that the SOS plot will have two "D"-
shaped curves where this orbit intersects the plane. A quick way to look right at the SOS
plane is by plotting the orbit but with a very restricted range on the 1-variable; in
Mathematica: PlotRange->{-0.001, 0.001}, then changing the viewpoint so you
are looking straight down on the 22-plane: ViewPoint->{0, 0, 1}. Then you get
the following plot:

You can see that the points on the SOS plane look like they are beginning to fill in two
curves (if I had integrated the equations of motion for a much longer time it would have
filled in better - but this is just a sample to show how it works). Thus, the complex 3D
orbit can be visualized with this much simpler view of the surface-of-section plot - as long
as we keep in mind what this 'slice' of the orbit represents, we can go on and put other
orbits with the same energy on this plot. The fact that orbits can't cross in phase space will

PHY 380.03 Spring 2013 2013 R. Martin


104

assure that the slice of the orbits on the SOS plot will also not cross, except at some fixed
points (like hyperbolic points) that represent limiting case behavior.

4. Chaotic orbits again appear as a set of points that appear to fill a 2D region of the plot.
Usually, the dots cover the region randomly, although there can be some structure which
is indicative of underlying structure in the dynamics. Its too hard to draw, so Ill give
another short-time example from Mathematica. For a higher initial energy, a chaotic
double pendulum orbit looks like this

Now intersect his with the y = 0 plane to get an SOS plot:

The points now look pretty well scattered (that short line segment is probably an artifact of
the approximate way of plotting the SOS). Again, I havent integrated the orbit out long
enough to see that it fills a 2D region, but I hope you get the point: you can tell the

PHY 380.03 Spring 2013 2013 R. Martin


105

difference between quasiperiodic non-chaotic orbits and chaotic orbits by a look at a


completed SOS plot.

5. Transient (Unbounded) orbits: For orbits that interact with a system for a finite time
and then leave, the SOS is a set of discrete dots in no particular order. This one is a little
tricky, since a discrete set of points for a single orbit might look, at first glace, like a
periodic orbit. The difference will be obvious by plotting the orbit itself: a transient orbit
leaves the system and doesnt repeat itself. Plotting many transient orbits on the same
SOS plot could also masquerade as a chaotic region: many transients will tend to cross
the SOS plane in a finite number of regions. Again, you have to look at some sample
orbits to tell the true nature of the orbits.

5.3.3. Numerics of SOS plots: Interpolation Method


Determining the points to be plotted on a surface-of-section plot is a bit more difficult than
finding strobe plot points. In the strobe case, you just chose regular times and integrated the
orbits to those times. For an SOS plot you need to find the dynamical variables on a surface in
space, regardless of what the time is when they hit that surface. The simplest approach is a
simple linear interpolation between the two positions just before hitting the surface and just after.
Here's a summary of the steps:

1. Integrate the equations of motion with a numerical integration routine.


2. At each time step use IF logic to determine whether the SOS surface (usually a plane, to
make life simple) has been crossed with positive velocity. If a crossing occurs between
the n th timestep and the (n + 1) st timestep, simply use linear interpolation to estimate the
actual crossing location.
3. For example, for our conservative force example above, the SOS plane is the plane y = y* .
Once you've determined when the orbit crosses this plane, refine the variable to the SOS
plane as follows: set
y x v x
= =
Ly L x Lv x

where y = y n +1 y * is the distance between, Ly = y n +1 y n , and similarly for x and v x .


Thus, you can solve for x * and vx* on the SOS plane, and thus calculate the point to plot
for this piercing of the plane.

4. Repeat steps (2) and (3) until you stop the orbit. You'll now have a set of points to plot
that are the SOS representation of that orbit. To complete the plot, run other orbits to
sample all regions of phase space.

Note that linear interpolation may not give good results if you are taking large time steps; i.e. the
orbit may not be approximately linear between t n and t n +1. You can decrease the time step to
obtain better results, at the expense of longer run times to integrate the orbit.

PHY 380.03 Spring 2013 2013 R. Martin


106

5.3.4. Numerics of SOS plots: Hnon's Method (Trick)

The problem with locating the SOS plane is that you know where it is in space (at y = y * , e.g.),
but your numerical integrator is following the orbit in time. Hnon's trick is to change the
independent variable in the numerical integration from time to space: thus for SOS plane at
y = y * , choose independent variable y and integrate from y n to y * . Here's how:
1. Use the chain rule to change time derivatives to y derivatives:
d dy d d
= = vy
dt dt dy dy
so, e.g.
dx dx 1
x = v y or = x
dy dy v y

2. Using the chain rule, then, we convert the equations of motion to be dependent on y. If
we start with
x = v x
y = v y
v x = Fx
v y = Fy

youll now be integrating


dx v x
=
dy v y
dy
=1
dy
dv x Fx
=
dy vy
dv y Fy
=
dy vy

(or, alternatively, you can compute vy from energy conservation). Of course, the y-
equation is now trivial and neednt be integrated. If however, you need to know the
exact time that the orbit hits the SOS plane, you could get that from integrating
dt dy = 1 / vy .

PHY 380.03 Spring 2013 2013 R. Martin


107

4. Now integrate the primed equations from y n to y * to obtain the SOS values of x * and
vx* . This method is much more precise than interpolation and, although it can add a bit
of extra execution time, the tradeoff is generally worth it: the SOS plane must be
located precisely if you want to be able to discern quasiperiodic orbits (closed curves)
from chaotic orbits which are nearby (which can imitate closed curves for long time
periods).

PHY 380.03 Spring 2013 2013 R. Martin

Anda mungkin juga menyukai