Dynamical systems 2019 [Link]@[Link].

se

5 Closed orbits and limit cycles

In Lecture 4 we connected the local dynamics close to fixed points
in non-linear flows to the dynamics observed in linear flows. In this
lecture we consider non-local properties of the flow, such as closed
orbits and limit cycles.
As shown in the last lecture, non-linear terms may destroy the
structurally unstable closed orbits around centers. But in systems
with symmetry (conservative, inversion symmetry, time-reversal sym-
metry, volume conserving) closed orbits around centers often become
structurally stable (insensitive to small perturbations of the flow).

5.1 Conservative systems

5.1.1 Hamiltonian systems
Newton’s law
F = ma = mẍ
If the force F is conservative (no friction), it can be written as the
negative gradient of a scalar potential V (x):
∂V
F =− .
∂x
Then Newton’s law can be written as the following dynamical system
(introduce momentum p ≡ mẋ):
p
ẋ =
m (1)
∂V
ṗ = − .
∂x
A Hamiltonian dynamical system takes the form
∂H
ẋ =
∂p
∂H
ṗ = −
∂x

1
Dynamical systems 2019 [Link]@[Link]

for some function H. In the case of Newton’s law above, choose H to

be the Hamiltonian (energy function)
p2
H≡ + V (x)
2m
to get
∂H p
ẋ = =
∂p m
∂H ∂V
ṗ = − =− .
∂x ∂x
In conclusion, Newton’s law with a conservative force, Eq. (1), is one
example of a Hamiltonian dynamical system.
In Hamiltonian systems, the energy E = H(x, p) is conserved since
∂H ∂H
Ḣ = ẋ + ṗ = 0 .
∂x |{z}
∂H
∂p |{z}
∂H
∂p − ∂x

Systems with at least one conserved quantity (integral of motion)

are called conservative systems. Remark: An integral of motion is a
combination of phase space coordinates that is constant along a tra-
jectory. A ‘constant of motion’ may in addition depend explicitly on
time. In classical mechanics continuous symmetries imply conserva-
tion laws (Noether’s theorem):
Symmetry Integral of motion
Time independence Energy
Space independence Momentum
Rotational symmetry Angular momentum

5.1.2 General conservative systems

More generally: Given a dynamical system ẋ = f (x) a function E(x)
is a conserved quantity if
E(x(t)) = constant ⇔ Ė = 0

2
Dynamical systems 2019 [Link]@[Link]

along trajectories and E(x(t)) is non-constant on every open set in

phase space. The latter condition implies that E(x(t)) cannot be
a trivial constant. For example E(x(t)) = x21 + x22 is constant when
evaluated along a circular trajectory and it is thus a conserved quantity
if the system has circular trajectories, while E(x(t)) = 7 (or any other
constant) is constant everywhere in phase space and is therefore not
a conserved quantity.

5.1.3 Fixed points in conservative systems

A conservative system cannot have any attracting fixed points:
If x∗ was an attractive fixed point, then all points in the basin of
attraction move towards the fixed point and must therefore have the
same value of E (the value at the fixed point). This implies that E is
constant on an open set in phase space and the flow is therefore not
conservative by definition.
Find which fixed points can occur in classical frictionless mechanics:

ẋ = ∂H
∂p
p2
with H≡ + V (x) .
ṗ = − ∂H
∂x 2m
Linearize
∂2H ∂2H
!  1
∂x∂p ∂p2 0 m
J= 2 ∂2H = 2 .
− ∂∂xH2 − ∂p∂x − ∂∂xV2 0

At any fixed point we have

√ τ ≡ trJ = 0 and ∆ ≡ det J(x∗) =
V 00(x∗)/m, i.e. λ1,2 = ± − det J:
q
00 ∗
If V (x ) < 0 then λ1,2 = ± |V 00(x∗)|/m (saddle)
q
00 ∗
If V (x ) > 0 then λ1,2 = ±i |V 00(x∗)|/m (center)

Fixed points are located at ẋ = p = 0 and ṗ = −V 0(x) = 0 ⇒

Centers at potential minima and saddles at potential maxima.

3
Dynamical systems 2019 [Link]@[Link]

Example: Double-well potential

x2 x4
V (x) = − +
2 4
p2
The paths in the phase plane are contours of constant H(x) = 2
−
x2 x4
2
+ 4
= E:

A and C are centers (potential minima) and B is a saddle (potential

maximum). Isolated centers in Hamiltonian systems are robust to
non-linear perturbations (non-linear centers). They are surrounded
by a band of closed orbits despite the non-linear terms. The closed
orbits around the centers correspond to small oscillations around the
potential minima. The stable and unstable manifold of the saddle
point connect at the fixed point and form a homoclinic orbit (this
trajectory is not strictly periodic because it takes an infinite amount
of time to reach the fixed point). The homoclinic orbit serves as a
separatrix: outside of it, the system has high enough energy to make
large-amplitude oscillations over both potential minima.

Closed orbits in conservative systems More generally, the fol-

lowing holds for two-dimensional conservative systems ẋ = f (x) with
smooth f : If x∗ is an isolated fixed point and E(x) is a local mini-
mum of E, then the paths in the vicinity of x∗ are closed (Strogatz
Theorem 6.5.1).

4
Dynamical systems 2019 [Link]@[Link]

5.1.4 How to find conserved quantities

There is no general method to find conserved quantities in a given
system. As seen above, a system that can be written as ẍ = −V 0(x)
has a conserved quantity E = y 2/2 + V (x) with y = ẋ.
Another method is to search for an explicit relation between y and x
by dividing ẏ by ẋ. For example
ẋ = y
ẏ = y 2 − x
gives
ẏ dy y 2 − x
= =
ẋ dx y
Let ỹ = e−xy to get
dỹ dy x x
= −e−xy + e−x = −e−x = −e−2x
dx dx y ỹ
Solve by separation of variables

−2x ỹ 2 1 −2x
ỹdỹ = −e xdx ⇒ = e (1 + 2x) + E
2 4
with an integration constant E, which forms an integral of motion
2
−2x y 1
E(x, y) = e − e−2x(1 + 2x) .
2 4

5.2 Reversible systems (Strogatz 6.6)

Another mechanism to form closed orbits is invariance under the si-
multaneous transformation t → −t and y → −y (reversible system).
This transformation mirrors trajectories in the upper half-plane on
the lower half-plane.
Special case: time-reversible system if reversible and ẋ = y.

5
 Dynamical systems 2019 [Link]@[Link]

Example Consider the system

ẋ = y − y 3
ẏ = −x − y 2
The system has three fixed points: (x∗1, y1∗) = (−1, −1), (x∗2, y2∗) =
(0, 0), and (x∗3, y3∗) = (−1, 1).
Linear stability analysis
∂f ∂f
!
2
 
∂x ∂y 0 1 − 3y
J= ∂g ∂g =
∂x ∂y
−1 −2y
Classification of fixed points:
Fixed point (x∗1, √
y1∗) = (−1, −1) (x∗2, y2∗) = (0, 0) (x∗3, y3∗√
) = (−1, 1)
Eigenvalues 1± 3 ±i −1 ± 3
Type Saddle
√
Center Saddle√
(1+ 3,1) (−1+ 3,1)
Stable direction √ √ - √ √
5+2
√ 3 5−2
√ 3
(1− 3,1) (−1− 3,1)
Unstable direction √ √ - √ √
5−2 3 5+2 3
Linear stability analysis shows that the origin is a center, but the non-
linear terms could destabilise, there is no conservation law as for the
conservative systems. However, the system is reversible: invariant
under the simultaneous transform t → −t and y → −y.
Close to the origin, the center causes swirl for positive values of y.
Time reversibility mirrors the solution in the line y = 0 and trajecto-
ries must therefore close (this rules out spirals):

6
Dynamical systems 2019 [Link]@[Link]

To sketch the full phase portrait, start with the nullclines and lin-
earized dynamics around the saddle points (need to evaluate the di-
rection of the stable and unstable manifolds of the saddle points.)

The unstable trajectory going up from the lower saddle point must
continue upwards (ẏ > 0) until y = 0 is reached with vertical slope
(ẋ = 0 at y = 0). Due to the reversibility there must be a mirrored
trajectory coming from the upper saddle. The combined heteroclinic
trajectory joins two fixed points.
Likewise, the unstable trajectory down from the upper saddle point
must move down and meet the y = 0 line at some x > 0. At the
crossing point it joins with its mirror from the lower saddle point,
forming a second heteroclinic trajectory. The heteroclinic trajectories
form a heteroclinic cycle surrounding a region of closed orbits.

7
Dynamical systems 2019 [Link]@[Link]

Reversible or conservative systems tend to have trajectories connect-

ing one fixed point with itself (homoclinic orbits, c.f. the double-well
potential in Section 5.1.3) or with another fixed point (heteroclinic
trajectory).
Note that the homoclinic orbit is not a periodic orbit: it takes an
infinite amount of time to reach the fixed point.
Homoclinic orbits and heteroclinic cycles (several heteroclinic trajec-
tories forming a loop) can attract (or repel) other trajectories, or act as
divisors between regions of different kinds of dynamics (separatrices):
in the example above separating a band of closed orbits from other
trajectories.
It is also possible that isolated closed orbits, not containing fixed
points, attract or repel trajectories. Such orbits are limit cycles.

5.3 Limit cycles (Strogatz 7)

Systems with limit cycles are useful in order to model self-sustained
oscillations (oscillations without external periodic forcing), such as the
firing of a pacemaker, cycles in the body , oscillating chemical reac-
tions, unwanted or dangerous self-excitations in mechanical systems.
Simplest construction using uncoupled polar coordinates

ṙ = f (r)
θ̇ = ω

where f (r) has zeroes for r > 0. r describes a one-dimensional

system whose fixed points determine the stability of the limit cycles.
Examples of stable, unstable, and half-stable limit cycles:

8
Dynamical systems 2019 [Link]@[Link]

Note that the system above is given in terms of polar coordinates and
that the limit cycle appears after changing to Cartesian coordinates.
Limit cycles (in the Cartesian coordinates) can not be found using
linear-stability analysis ⇒ non-local effect, non-linear terms are nec-
essary. They can be stable, unstable and half-stable (it is also possible
to have a limit cycle from outside|inside, while band of closed orbits
on the inside|outside).
More non-trivial examples of limit cycles will be discussed later in the
course (van der Pol oscillators).

5.4 To prove that closed orbits exist

Poincaré-Bendixson theorem Assume a smooth flow in a bounded
domain D of the plane. Assume further that D does not contain any
fixed point and that there exists a trajectory that is confined in D
for all times. Then at least one periodic orbit exists in D. This is
a consequence of the fact that trajectories for smooth flows cannot
intersect in two dimensions.
To satisfy the condition that a confined trajectory exists, one can
construct a trapping region, i.e. choose D such that the flow points
inward everywhere. If it is possible to construct a trapping region,

9
Dynamical systems 2019 [Link]@[Link]

then the Poincaré-Bendixon theorem ensures that at least one closed

orbit exists in D.

Trapping region D
Periodic orbit

As a consequence in two dimensions: any trajectory will approach

either a fixed point, a periodic orbit, or a union of fixed points and
homo|hetero-clinic orbits. In higher dimensions: infinite wandering of
trajectories is possible (trajectories never repeats).

5.5 To rule out closed orbits or limit cycles

• If a gradient system: ẋ = −∇V (x) with potential function V .
During one revolution of a supposed periodic orbit with period
time T the potential changes by ∆V ≡ V (xT ) − V (x0)
Z T Z T Z T
∆V = dt V̇ = dt ẋ · ∇V = − dt |ẋ|2 < 0
0 0 0

But ∆V bust be 0 because x(T ) = x(0) for the periodic orbit.

Hence, our assumed periodic orbit cannot exist.
This criterion rules out limit cycles in one-dimensional systems,
(f (x) can be written as ∇V ). When d > 1 gradient systems
are atypical.
• Construct a function V (x) (Lyapunov function) that
– is positive everywhere except at a fixed point x∗ where it
is zero
– decreases along any trajectory (V̇ ≡ ẋ · ∇V < 0).

10
Dynamical systems 2019 [Link]@[Link]

If a Lyapunov function can be constructed, the fixed point x∗ is

globally attracting and no closed orbits can exist.
• Dulac’s criterion: consider a differentiable function g(x) such
that ∇ · (g ẋ) does not change sign in some domain. If such
function exists, there are no closed orbits in the domain (Green’s
theorem).

11