se
9 Another example and chaotic billiards
9.1 Example 2: Driven damped pendulum
(Strogatz 8.5)
Add a constant torque τ to the damped pendulum from Lecture 1:
γ g τ
θ̈ = − θ̇ − sin θ +
m l I0
with I0 the moment of inertia.
This equation can be written as a dimensionless dynamical system
by a suitable rescaling of t and θ̇ (Problem set 2) . In dimensionless
units we have:
dθ
=y
dt0
dy τl (1)
= −αy − sin θ +
dt0 I0g
|{z}
I
where α = dimensionless damping, I = dimensionless torque.
This system is identical to that of the phase difference in a Josephson
junction (superconducting device): Strogatz 4.6. It can also be used
as a lowest-order approximative model for a large number of problems.
We can view the system Eq. (1) as a flow on a cylinder with θ pe-
riodic −π < θ ≤ π and keep aperiodic angular velocity y.
9.1.1 Analysis for small α and I
Simple pendulum (Strogatz 6.7) When α = I = 0 Eq. (1)
simplifies to a simple (undamped, undriven) pendulum:
dθ
=y
dt0
dy
= − sin θ .
dt0
1
Dynamical systems 2019 [email protected]
This is a Hamiltonian system with V 0(θ) = sin θ ⇒ potential energy
V (θ) = − cos θ. On the cylinder there are two distinct fixed points
given by the potential minimum at (θ∗, y ∗) = (0, 0) (nonlinear center)
and the potential maximum at (θ∗, y ∗) = (π, 0) (saddle).
On the cylinder periodic orbits comes in two types: librations and
rotations
Note that due to index theory librations always encircle fixed points
whose indices sum to +1, while rotations instead encircle the cylinder.
For the simple pendulum two homoclinic orbits [homoclinic because
(θ∗, y ∗) = (−π, 0) and (θ∗, y ∗) = (π, 0) are the same point on the
cylinder] separate regions with closed orbits in the forms of librations
and rotations:
2
Dynamical systems 2019 [email protected]
Which orbit is obtained depends on the energy E of the initial condi-
tion.
Along any trajectory we have the dimensionless energy E = y 2/2 −
cos θ = const (conservation of energy in Hamiltonian systems). At
the saddle point (θ∗, y ∗) = (−π, 0) we have E = 1, the energy for the
homoclinic orbits. Librations have E < 1 and rotations E > 1. Due
to energy conservation, E = y 2/2 − cos √ θ = 1, the two homoclinic
orbits can be parameterized as yh± = ± 2 + 2 cos θ.
Weakly damped pendulum When 0 < α 1 and I = 0 we
have a weakly damped pendulum: the homoclinic orbits are bro-
∗ ∗
ken and the fixed
p point (θ , y ) = (0, 0) becomes a stable spiral
(λ1,2 = −α/2 ± i 1 − α2/4).
Letting α > 0 destroys the homoclinic orbit due to the dissipative
damping term. Is it possible to recover the homoclinic orbit by input
of energy from the driving term, I > 0? Yes! by Melnikov’s method.
Weakly damped and weakly driven pendulum For general val-
ues of α and I the system (1) has fixed points where y ∗ = 0 and
sin θ∗ = I. Depending on the value of I we have 0 (I > 1), 1 (I = 1),
or 2 (0 ≤ I < 1) fixed points:
3
Dynamical systems 2019 [email protected]
When the driving is weak (0 < I 1) the position of the saddle point
is shifted to (θ∗, y ∗) = (π − asin(I), 0) (the constant driving shifts the
unstable equilibrium from up-side down to an angle).
Evaluate the energy function from the unperturbed system, H =
2
y /2 − cos θ, along a trajectory originating from the saddle. For
most values of α and I we end up at a point different from the sad-
dle after one lap on the cylinder. Such trajectories have non-zero
∆H = Hstart − Hend (H is assumed to vary smoothly with the small
parameters α and I):
Aim: Find critical small parameters I and α such that ∆H = 0
and consequently ∆y = 0 for the perturbed trajectory (’perturbed’ =
governed by Eq. (1) with non-zero α and I).
The change in H when going from θ = −π − asin(I) to θ = π −
asin(I) (the sought upper homoclinic orbit) along perturbed trajectory
is given by (Melnikov integral, upper integration bound tf is the time
at which θ = π − asin(I) is reached (assuming I ≥ Ic so that θ =
4
Dynamical systems 2019 [email protected]
π − asin(I) is reached at all):
Z tf Z tf
0
∆H = dt Ḣ = dt0[ ẏ y + |{z}
θ̇ sin θ]
−∞ −∞
|{z}
−αy−sin θ+I y
tf Z tf
dθ
Z
= dt0y(I − αy) = dt0 0 (I − αy)
−∞ −∞ dt
Zπ−asin(I) √
= dθ(I − αy) = [Use y = yh+ = 2 + 2 cos θ + O(I, α)]
−π−asin(I)
Zπ−asin(I) √
= dθ(I − α 2 + 2 cos θ) + O(α2, αI)
−π−asin(I)
= 2πI − 8α + O(α2, αI)
Thus ∆H = 0 is a simple zero for Ic = 4α/π + O(α2). We have found
an orbit that has the same value of H after one lap on the cylinder,
and which is isolated (∆H is simple zero, any modification of Ic gives
∆H 6= 0) ⇒ the found orbit is the sought homoclinic orbit.
The same evaluation for the lower homoclinic orbit (going from
θ = π − asin(I) to θ = −π − asin(I)) gives ∆H = −2πI − 8α, i.e.
∆H = 0 cannot be satisfied ⇒ the lower homoclinic orbit is always
broken if α > 0 and I > 0.
In conclusion, for small α and I we have a homoclinic bifurcation
at Ic ≈ 4α/π:
For I < Ic the damping overtakes the driving and the motion is
damped to zero.
At I = Ic a limiting trajectory (the homoclinic orbit) is created that
allows the pendulum to make one full lap, but in infinite time.
5
Dynamical systems 2019 [email protected]
For I > Ic the system is bistable, it shows a stable limit cycle (full
lap rotations) and a stable spiral decaying the motion to zero. Which
attractor is reached depends on whether if the initial position is in
the basin of attraction of the limit cycle (green hashed area above) or
the spiral (blue hashed), separated by the stable/unstable manifolds
of the saddle point.
9.1.2 Larger α and I
When I > 1 no fixed points exists. Librations are not possible (Index
theory: closed orbits must encircle a fixed point). It is possible to
show (Strogatz 8.5) that a unique attracting limit cycle in the form
of a rotation exists (the strong torque give the pendulum a stationary
full-lap rotation).
For larger values of α it is possible to show that the limit cycle form
in an infinite-time bifurcation (Strogatz 4.6), i.e. not a homoclinic
bifurcation as for small values of α.
Using numerical simulations one obtain a phase-diagram (Strogatz
Fig. 8.5.10)
Here lines show bifurcations, black text denote system attractor(s).
6
Dynamical systems 2019 [email protected]
9.2 Coupled oscillators (Strogatz 8.6)
As we saw in the previous example the topology of the dynamics
matters: the flow on a cylinder allows for new type of periodic orbit
(rotations) not encircling a fixed point.
In a system with two periodic coordinates we have a flow on a torus.
9.2.1 Flow on a torus
The dynamics
θ̇1 = f1(θ1, θ2)
θ̇2 = f2(θ1, θ2)
represents a flow on a torus (θ1 and θ2 both 2π periodic):
Easier to visualise on a square with periodic boundaries (opposite end-
points are the same points)
7
Dynamical systems 2019 [email protected]
9.2.2 Example: Uncoupled oscillators
Simplest case f1 = ω1 = const. and f2 = ω2 = const.. Uncoupled
dynamics:
The lines have slope: dθ2/dθ1 = θ̇2/θ̇1 = ω2/ω1.
Two special cases:
• ω2/ω1 rational (can be written as a ratio p/q with p and q
integers) ⇒ closed orbits
• ω2/ω1 irrational ⇒ trajectory never closes, covers entire torus
(quasiperiodic, new long-term behaviour . Only appears on the
torus.)
9.2.3 Other examples
• Phase locking and synchronization (Strogatz 8.6).
• Integrable hamiltonian systems: Can be solved using ’action-
angle coordinates’, the solutions are simply uncoupled oscillators
on (hyper) tori. This will be discussed in a later lecture, but a
simplified version, integrable billiard systems comes next.
8
Dynamical systems 2019 [email protected]
9.3 Billiards
In billiard systems a particle has constant velocity until it hits a
boundary, where it bounces using the law of reflection (incident angle
θi=outgoing angle θo):
Billiards are important as simple models of problems in optics, statis-
tical physics (gases), electrons in metal (Lorentz gas), etc.
Billiard systems constitute examples of Hamiltonian systems:
p2
0 Inside billiard area
H(x, p) = + V (x) , with V (x) =
2m ∞ At walls
It is possible to show (later) that a Hamiltonian system is integrable
if there exists d independent integrals of motion. These systems show
periodic (or quasiperiodic) motion (regular patterns)
The type of dynamics depend on the geometry of the billiard table.
9
Dynamical systems 2019 [email protected]
Integrable billiards -
• Rectangle: p2x and p2y are integrals of motion. The dynamics
closely resembles that of uncoupled oscillators on the torus in
Section 9.2.2:
The billiard trajectory (blue) is obtained by mirroring of the
uncoupled-oscillator trajectory (red) around the centers of the
vertical and horizontal axes. Depending on the ratio vy /vx we
get a closed orbit (rational ratio) or a quasiperiodic orbit (irra-
tional ratio).
• Circle: Kinetic energy p2/(2m) and angular momentum w.r.t.
center are integrals of motion.
• Ellipse: Kinetic energy p2/(2m) and product of angular momen-
tum w.r.t. the 2 foci is conserved
10
Dynamical systems 2019 [email protected]
Integrable motion is always predictable. Two close-by initial con-
ditions stay together for a large time (they may separate, but not
exponentially fast).
Note: If the billiard table is two-dimensional, we have a four-dimensional
dynamical system (x,y,vx,vy ). We plot a two-dimensional projection
⇒ it may look like trajectories cross, but they do not in the four-
dimensional phase space.
Now consider cases where one symmetry is broken
9.4 Non-integrable billiards
Introduce circle in center of rectangular billiard (left).
Energy (p2x + p2y )/(2m) still conserved, but not the individual compo-
nents p2x and p2y ⇒ non-integrable system for most initial conditions.
Convex surface causes closeby trajectories with closeby angles to
separate ⇒ trajectories with small initial distance show exponential
separation after many bounces. Further: (most) trajectories are ape-
riodic (infinitely long!) and densely fill space.
These systems are chaotic: (almost all) trajectories depend sensi-
tively on the initial condition.
Any disturbance (in physical system) or numerical inaccuracy (in nu-
merical simulations) is exponentially amplified ⇒ long-term prediction
impossible.
11
