## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

**EMAT33100: Nonlinear Dynamics & Chaos
**

Lecture Notes

by

Bernd Krauskopf & Hinke Osinga

1 What is a dynamical system?

A dynamical system is a system whose behaviour can be described by an evolution operator

Φ

t

: X →X

deﬁned on a space X for all t ∈ T. The space X is called the state space or phase space.

The space T, or time, can be R, in which case we speak of a continuous time system, or

T = Z, which corresponds to a discrete time system. The evolution operator must be such

that the following two conditions hold for any initial condition x

0

∈ X and any t, s ∈ T:

1. Φ

0

(x

0

) = x

0

(“no time, no evolution”),

2. Φ

t+s

(x

0

) = Φ

t

(Φ

s

(x

0

)) (“determinism”).

If T = R, then the dynamical system is a continuous time system, which is typically given

as a vector ﬁeld

˙ x = f(x), (1)

where f is a function deﬁned on X. The evolution operator describes the ﬂow of the vector

ﬁeld. Special solutions of a vector ﬁeld are:

• Equilibria or rest points: Points x

∗

∈ X with f(x

∗

) = 0.

• Periodic orbits: For a point x

∗

∈ X there exists τ ∈ R with τ > 0 such that

Φ

τ

(x

∗

) = x

∗

. The periodic orbit is deﬁned as the closed curve

{Φ

t

(x

∗

) | 0 ≤ t < τ

∗

},

where τ

∗

> 0 is the smallest number τ such that Φ

τ

(x

∗

) = x

∗

, and it is called the

period of the periodic orbit.

2 Bernd Krauskopf & Hinke Osinga

If T = Z, then the dynamical system is a discrete system and deﬁned by a map

x →g(x), (2)

where g is a function deﬁned on X. The evolution operator is the map g itself and Φ

t

(x) =

Φ

n

(x) = g

n

(x). Special solutions of a discrete system are:

• Equilibria or ﬁxed points: Points x

∗

∈ X with g(x

∗

) = x

∗

.

• Periodic orbits: For a point x

∗

∈ X there exists n ∈ Z with n > 0 such that

g

n

(x

∗

) = x

∗

, while g

i

(x

∗

) = x

∗

for all 0 < i < n. The periodic orbit is deﬁned as the

set of n points

{g

i

(x

∗

) | 0 ≤ i < n},

and n is called the period of the periodic orbit.

Given the evolution operator, we can ﬁnd out what happens to x

0

as time evolves. With

time increasing, one would follow x

0

under the ﬂow of (1) or compute iterations of x

0

under

the function g of (2). If f in (1) is suﬃciently smooth, then we can also go back in time

and follow the ﬂow backward to see where x

0

came from. The entire future and history of

a point x

0

,

{Φ

t

(x

0

) | t ∈ T},

is called the orbit of x

0

. In discrete time, g must be invertible in order to look at the entire

orbit. However, we can always consider the forward orbit {Φ

t

(x

0

) | t ∈ T, t ≥ 0}, only.

Remark 1 Throughout the lectures we will assume that the function f associated with

a vector ﬁeld, as in (1), is suﬃciently smooth so that both past and future are uniquely

determined by the initial condition x

0

. In other words, the diﬀerential equations in this

course always have unique solutions, provided a suﬃcient number of initial or boundary

conditions is given.

Example

The logistic map

p →rp(1 −p)

is a discrete dynamical system for any r ∈ R. The state space is R and orbits are given by

the iterates of the map. Since this map is not invertible, we can only look at forward orbits.

[Note that a one-dimensional discrete system can only be chaotic if it is not invertible.]

Example

The Lorenz system

˙ x = σ(y −x),

˙ y = ̺x −y −xz,

˙ z = −βz + xy

(3)

is a three-dimensional continuous dynamical system. Even though we cannot explicitly

solve the system of ODEs, the solutions are uniquely determined in both forward and

backward time by the initial condition (x

0

, y

0

, z

0

)

T

∈ R

3

. [Note that a vector ﬁeld can only

be chaotic if its phase space is at least three-dimensional.]

EMAT33100 - Nonlinear Dynamics & Chaos 3

2 What is chaos?

Chaos theory is so young that mathematicians still have not decided on the deﬁnition of

chaos!! Based on the classical notions, the evolution of an initial condition x

0

under an

operator Φ

t

does not behave chaotically if:

• Φ

t

(x

0

) goes to an equilibrium when t →∞.

• Φ

t

(x

0

) goes to a periodic orbit when t →∞.

• Φ

t

(x

0

) escapes to ∞ as t →∞.

This is not chaotic behaviour, because it is predictable.

A system is chaotic if (a subset of the) orbits are conﬁned to a bounded region, but still

behave unpredictably. A good example is the Lorenz system (3) with σ = 10, ̺ = 28 and

β = 8/3. Arbitrary orbits seem to accumulate on an object called the butterﬂy attractor.

Two initial conditions that are very close to each other behave similarly, since they both

trace the butterﬂy attractor. On the other hand, one would expect that two nearby initial

conditions trace a similar path, at least for a very long time; the number of turns in

one of the butterﬂy’s wings followed by a number of turns in the other wing, etcetera,

should be the same, at least initially. Chaotic behaviour means that this is not true, in

the sense that fairly quickly the two orbits start to behave very diﬀerently and there is

no trace or “memory” of the fact that they were once so close. This property is called

sensitive dependence on initial conditions and can be roughly translated as “all decimal

places matter.”

3 Phase portraits and topological equivalence

Example

Consider the one-dimensional linear continuous time dynamical system

˙ x = αx, (4)

4 Bernd Krauskopf & Hinke Osinga

with x ∈ R and a parameter α ∈ R. We can immediately write down the solution of (4),

namely

x(t) = e

αt

x(0)

is uniquely determined by the choice of initial condition x(0) ∈ R. In dynamical systems

theory, one is interested in getting a rough idea what solutions look like — just a sketch

will do — and one is particularly keen on knowing what happens to solutions in the long

run, as time t →∞.

Since we know the solution, we can easily draw a sketch depending on the choice for

the parameter α. Qualitatively, we have the following diﬀerent possibilities:

→t

x

↑

x(0)

u

α = 0

α < 0

α > 0

If α < 0, x(t) →0 as t →∞,

If α = 0, x(t) = constant,

If α > 0, x(t) →±∞ as t →∞.

In this picture, we choice for x(0) is some positive number. If we choose x(0) < 0, then

the solution either drops to −∞, if α > 0, stays constant, if α = 0, or it increases to 0,

if α < 0; this is the exact mirror image of the above picture, where the mirror is at the

t-axis. The case α = 0 is rather unusual, and we will forget about it in this example.

An extremely simple, and much favoured way of sketching the qualitative behaviour is

as follows:

u ,

0

α < 0

→x

u ,

0

α > 0

→x

The pictures plot the phase space which is R in our case, so we draw a line. The dynamical

system has one equilibrium, namely the origin, which is also indicated in the pictures. If

α < 0, then solutions tend to 0, which is indicated by the arrows. The arrows point in

the opposite direction for the case α > 0. In particular, the pictures immediately reveal

that the equilibrium, the origin, is stable for α < 0 and unstable for α > 0. We call such

pictures phase portraits.

One-dimensional dynamical systems have another, extremely handy, way of visualising

their behaviour. We can namely compare the velocity, and in particular the direction of

the velocity, ˙ x with the position x by simply plotting the graph of ˙ x, that is, the graph of

the RHS of (4). Again, we have the cases α < 0 and α > 0 as follows:

EMAT33100 - Nonlinear Dynamics & Chaos 5

α < 0

u ,

0 →x

`

`

`

`

`

`

` `

`

`

`

`

`

`

``

˙ x

↑

˙ x = αx

α > 0

u ,

0 →x

˙ x

↑

˙ x = αx

The graph ˙ x = αx directly relates to the direction of the arrows on the x-axis. If the graph

lies above the x-axis, then the arrow points to the right — x is increasing, because ˙ x > 0;

where it lies below the x-axis, thearrow points to the left — x is decreasing, because ˙ x < 0.

Note that the point where the graph intersects the x-axis is the equilibrium point.

Example

To illustrate the power of this geometric technique, we consider the following nonlinear

problem

˙ x = sin x. (5)

Again, this is a one-dimensional continuous time dynamical system and x ∈ R. Suppose

we would like to know the behaviour of the solution starting at x(0) =

π

4

.

The ODE (5) is separable, and we can, in fact, solve it explicitly. For initial condition

x(0) = x

0

, we ﬁnd that the solution must be such that

t = ln

(1 −cos x) sin x

0

(1 −cos x

0

) sin x

= ln

csc x −cot x

csc x

0

−cot x

0

.

So, what happens to the solution with x

0

=

π

4

?

By being just a little less precise, we can ﬁnd an answer remarkably quickly by looking

at the graph of ˙ x = sin x.

u

0

u

π

u

2π

u

−π

u

−2π

˙ x

↑

→x

, ,

6 Bernd Krauskopf & Hinke Osinga

The dynamical system (5) has equilibria at x = k π, for any k ∈ Z. Considering where

the graph of sin x is above or below the x-axis, we immediately deduce that −π, π, 3π,

etcetera, are attracting equilibria; the equilibria in between, −2π, 0, 2π, etcetera, are

repelling equilibria. Starting with an initial condition x(0) =

π

4

, this is in the region where

the arrows point to the right, so the solution will increase and x(t) = π in the limit t →∞.

Based on the graph of sin x, we can be more speciﬁc: the speed at which the solution

increases to π is increasing as well until x(t) =

π

2

, after which the velocity slows down to

0 (note that x(t) itself is monotonically increasing; ˙ x is positive at all time). However, we

cannot say at what time x(t) =

π

2

or something as detailed as that. Nevertheless, with the

picture of the graph of sin x we can give a qualitative behaviour of solutions starting at

any arbitrary initial condition.

Topological equivalence

Let us have a more closer look at the graph of sin x near x = π. This local picture matches

the phase portrait of (4) for α < 0 VERY well.

˙ x = sin x near x = π

u ,

π →x

˙ x

↑

˙ x = αx with α < 0

u ,

0 →x

`

`

`

`

`

`

` `

`

`

`

`

`

`

``

˙ x

↑

We call these two systems topologically equivalent in these respective neighbourhoods. The

oﬃcial deﬁnition is

Deﬁnition

Suppose we have two vector ﬁelds

(1) ˙ x = f(x) and (2) ˙ y = g(y).

Then (1) on a domain U is topologically equivalent to (2) on a domain V , if we can ﬁnd a

continuous and invertible map h : U →V that maps orbits of (1) to orbits of (2), respecting

the direction of time.

Hence, in the respective domains U and V the two vector ﬁelds have the same number of

equilibria (one in our example) and the phase portraits are identical. Note that we do not

need to have a time-wise precise mapping of the orbits.

3.1 Phase portraits for two-dimensional systems

Unfortunately, with two-dimensional systems it is already impossible to draw the graph of

the RHS (this would be a surface in four-dimensional space). Hence, for phase portraits we

EMAT33100 - Nonlinear Dynamics & Chaos 7

restrict ourselves to sketches in the phase space only, that is, on the plane. However, the

whole idea is the same: we indicate with arrows in which direction the vector ﬁeld points,

which depends on the sign of the two components of the RHS.

Example

Consider, for example, the harmonic oscillator with a friction term.

Newton’s law gives:

¨ x + c ˙ x + ω

2

x = 0

M

spring

0 →x, the displacement

THIS IS A TWO-DIMENSIONAL DYNAMICAL SYSTEM !!!

The system is not determined by the displacement x alone! We also need to know the

velocity ˙ x. Hence, the phase space is R

2

, the space of displacements and velocities. Another

way to see this, is that the equation is a second-order ODE. Dynamical systems must be

written as a system of ﬁrst-order ODEs. By deﬁning y = ˙ x we obtain the system of

equations

˙ x = y,

˙ y = −cy −ω

2

x.

(6)

The general form of a vector ﬁeld is:

˙ x = f(x, λ),

where x ∈ X = R

n

and λ ∈ R

m

is a vector of parameters. The function

f : R

n

×R

m

→R

n

with

f(x, λ) =

¸

¸

f

1

(x, λ)

.

.

.

f

2

(x, λ)

**assigns a vector to every point in phase space ⇒ a force ﬁeld.
**

You have probably seen the example of the harmonic oscillator before. If no friction is

present in the system, that is, c = 0, then the motion is perfectly sinusoidal. If c > 0,

then the amplitude of the oscillations dies out as t → ∞. In the physically unlikely

(but mathematically completely logical) case that c < 0, the amplitude of the oscillations

increases. This can be visualised as follows by plotting the x-coordinate versus time.

8 Bernd Krauskopf & Hinke Osinga

c > 0

10 0

1.5

-1.5

x

t

c = 0

10 0

1.5

-1.5

x

t

c < 0

10 0

3

-3

x

t

Forgetting about time, the phase portrait for these three cases plot the orbits in the (x, y)-

plane. With arrows we indicate the direction of the ﬂow, but we are sloppy in providing

information about how fast the motion is. For example, the pictures give information about

the amplitude, but not the frequency of the oscillations. In particular, the frequency is the

same for all periodic orbits in the phase portrait for c = 0, which cannot be deduced from

the picture.

c > 0

1.5 -1.5

5

-5

←

y

x

c = 0

1.5 -1.5

5

-5

→

→

→

y

x

c < 0

3 -3

10

-10

←

y

x

By simply considering the direction of the arrows in the phase plane, one can often deduce

the stability properties of the equilibria. If there is uncertainty about this, then we can

also do a bit more algebra to make sure. This is the topic of the next section, but we

ﬁrst show another example where the phase portrait does give all relevant information by

considering the direction of the arrows alone.

Example

Consider the following model in population dynamics of two species that compete, say for

the same limited food source, such that they inhibit each other’s growth. A simple model

for this type of interaction is

˙ x = x(1 −x −αy) =: f(x, y) and

˙ y = λy(1 −y −βx) =: g(x, y).

(7)

The two species are represented by x and y, which stand for fractions of an expected

population size (so x = 1 would be considered as a healthy situation). In the absence of

species y, species x satisﬁes what is known as logistic growth: the species will increase

until it reaches its expected value x = 1 (or decrease if it started at a value larger than

1). Both x = 0 and x = 1 are equilibria, where x = 0 is repelling and x = 1 is attracting.

EMAT33100 - Nonlinear Dynamics & Chaos 9

Similarly y grows logistically in the absence of x. The two parameters α and β are positive

and indicate the inhibition eﬀect.

The nullclines of (7) are:

˙ x = 0 → x = 0 or y =

1

α

(1 −x),

˙ y = 0 → y = 0 or y = 1 −βx.

Note that there is no reason to decide whether α < 1 or α > 1 and the same holds for β.

Hence, there are four possible conﬁguration for the nullclines if we ignore the possibility

α = 1 and/or β = 1 (this is far less likely than α = 1 and β = 1). Note that we are only

interested in the positive quadrant, because we are dealing with species.

α, β < 1

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

y

x 0

1

1

α

1

1

β

P

∗

s

s

s

s

·

`

,

·

`

,

α, β > 1

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

y

x 0

1

1

α

1

1

β

P

∗

s

s

s

s

·

`

,

·

`

,

β < 1 < α

`

`

`

`

`

`

`

`

`

`

`

`

y

x 0

1

1

α

1

1

β

s

s

s

·

`

,

,

,

,

`

`

α < 1 < β

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

`

y

x 0

1

1

α

1

1

β

s

s

s

·

`

,

·

·

·

**Using that we know the direction of the arrows along the coordinate axes (absence of the
**

other species), we can “ﬁll out” the rest of the phase portrait for each of the four cases.

Note that the cases α, β < 1 and α, β > 1 lead to the existence of a steady state P

∗

in the

positive quadrant oﬀ the coordinate axes. This steady state does not exist if β < 1 < α or

α < 1 < β.

In each of the four cases, the origin is a source or repelling point. The diagram shows

that for α, β < 1 the steady states (1, 0) and (0, 1) are saddles, while the positive steady

state P

∗

is stable. Hence, the populations reach a balanced situation where both x and y

10 Bernd Krauskopf & Hinke Osinga

are positive. Unfortunately, this case seems quite rare in nature. Typically at least one of

α and β is larger than 1.

For α, β > 1 the two steady states on the axes (1, 0) and (0, 1) are both stable and P

∗

is a saddle. Hence, there are two attractors in this system and it depends on the initial

conditions whether the x population or the y population goes extinct. This means that

there must exist a curve through P

∗

that separates the positive quadrant into two regions.

On one side of this separatrix initial conditions converge to (1, 0) and on the other side they

converge to (0, 1). The separatrix separates the two basins of attraction. Since the origin

is a source, we can conclude that the separatrix must also pass through (0, 0). However, it

is not possible, in general, to ﬁnd an explicit equation for the separatrix. Nevertheless, it

is clear such a curve must exist. Initial conditions chosen exactly on the separatrix neither

go to (1, 0) nor to (0, 1), but to P

∗

instead. The separatrix is, therefore, also called the

stable manifold of P

∗

.

If β < 1 < α then (1, 0) is a saddle and (0, 1) is stable. Hence, the x-population dies

out. If on the other hand α < 1 < β then (0, 1) is a saddle and (1, 0) is stable, so that the

y-population dies out. Intuitively, this is clear from the equations (7). If β < 1 < α then y

has a stronger inhibitive eﬀect on x than the x on y. So, x will lose. Similarly, y will lose

if α < 1 < β.

We see that in three of the four cases one of the populations goes extinct. This illustrates

the principle of competitive exclusion that is common in nature. Typically, in nature, one

of the two species will move away, or choose another food source or so to survive. Hence,

naturally, a species attempts to adjust the inhibition the other has on it. Well, at least the

most inhibited species tries.

Let us assume that we are in the situation β < 1 < α. This means that y has a stronger

inhibitive eﬀect on x than the other way around. Hence, x will try to adapt, so that it

may survive. This means mathematically that x will try to change the parameter α by

looking for another food source, for example. If α decreases only slightly, the situation stays

topologically the same, and x will still go extinct (it only takes a little longer). However,

when α decreases dramatically, so that α < 1, then the situation becomes topologically

diﬀerent and x manages to survive. The moment of change in the two topologically diﬀerent

cases comes about when α = 1. We call such a transition between two topologically diﬀerent

phase portraits a bifurcation.

Eigenvalues and the stability of equilibria

Recall that a point x

0

of the vector ﬁeld ˙ x = f(x) is an equilibrium or steady state if

f(x

0

) = 0, that is, if there is no force at x

0

. The Jacobian Df(x

0

) at the equilibrium x

0

is

deﬁned as

Df(x

0

) =

∂f

1

∂x

1

. . .

∂f

1

∂x

n

.

.

.

.

.

.

∂f

n

∂x

1

. . .

∂f

n

∂x

n

¸

¸

¸

¸

¸

¸

, where f(x) =

f

1

(x)

.

.

.

f

n

(x)

¸

¸

¸

,

EMAT33100 - Nonlinear Dynamics & Chaos 11

and is an n × n matrix. The eigenvalues of this matrix (counted with multiplicity) are

important for determining the stability properties of the equilibrium x

0

. We deﬁne the

following three numbers:

n

0

= # of eigenvalues of Df(x

0

) with zero real part,

n

+

= # of eigenvalues of Df(x

0

) with positive real part,

n

−

= # of eigenvalues of Df(x

0

) with negative real part.

An equilibrium x

0

is called hyperbolic if n

0

= 0. In this case, we call x

0

an attractor if

n

−

= n and n

+

= 0, a repellor if n

+

= n and n

−

= 0, and a saddle point both n

+

> 0 and

n

−

> 0. In general, we cannot be certain about the stability properties of x

0

if n

0

= 0.

Hence, the stability properties of an equilibrium x

0

for a nonlinear dynamical systems

are determined by the linearisation about x

0

. This is stated in the following important

theorem:

Theorem 2 (Hartman & Grobmann) If the vector ﬁeld

(1) ˙ x = f(x)

has a hyperbolic equilibrium x

0

, then there exists a neighbourhood U of x

0

such that (1)

on U is topologically equivalent to the linearised system

(2) ˙ x = Df(x

0

) x

on an (arbitrary) neighbourhood V of the origin.

For example, the system ˙ x = sin x near x = π is topologically equivalent to the system

˙ x =

d

dx

sin x

x=π

x = (cos π) x = −x,

as we have seen before.

The Hartman-Grobmann Theorem is very powerful: it says that the dynamics of any

nonlinear dynamical system is similar to the dynamics of a linear system locally near its

hyperbolic equilibria. So, what is the possible dynamics of a linear system?

Let us, for simplicity, consider all linear vector ﬁelds on R

2

. The case for R

n

is similar.

Hence, we consider an arbitrary 2 ×2 matrix A and the linear system

¸

˙ x

˙ y

= A

¸

x

y

. (8)

The origin is the only equilibrium. It is not so hard to ﬁnd explicit solutions of (8) for

speciﬁc choices of A and deduce the behaviour, so here we only list the possibilities. The

eigenvalues of A are determined by the zeroes of its characteristic polynomial

char

A

(λ) = det(λI −A) =

¸

λ 0

0 λ

−A

= λ

2

−trace(A) λ + det(A).

Depending on the trace and determinant of A we can ﬁgure out the values for n

0

(A),

n

−

(A), and n

+

(A). In particular, n

0

(A) = 0 only if det(A) = 0 (which is not very common

for arbitrary 2 ×2 matrices). We get the following picture in the (trace(A), det(A))-plane:

12 Bernd Krauskopf & Hinke Osinga

trace(A)

det(A)

2

2a

1

1a

3 3

Eigenvalues in

1a

and

2a

are complex conjugate.

On the trace(A)-axis we have n

0

= 0.

In

1

and

1a

: n

−

= 2 ⇒ origin is an attractor

In

2

and

2a

: n

+

= 2 ⇒ origin is a repellor

In

3

: n

+

= n

−

= 1 ⇒ origin is a saddle point

For each of the regions, we have the following qualitative phase portraits, each given with

an example taken either from the harmonic oscillator or from the competing species model.

Eigenvalues Phase portrait Example

1

C

u u

x

y

s ,

·

·

`

`

»

´

Competing species, α, β > 1

linearisation about (1, 0)

A =

¸

−1 −α

0 −λ(β −1)

1a

C

u

u

x

y

1.5 -1.5

5

-5

←

Harmonic oscillator

c = 0.7, ω = 3

A =

¸

0 1

−9 −c

2

C

u u

x

y

s ,

`

`

·

·

´

»

Competing species

linearisation about (0, 0)

A =

¸

1 0

0 λ

2a

C

uu

x

y

3 -3

10

-10

←

Harmonic oscillator

c = −0.3, ω = 3

A =

¸

0 1

−9 −c

3

C

u u

x

y

s ,

`

·

º >>

> >.

Competing species, α, β < 1

linearisation about (1, 0)

A =

¸

−1 −α

0 λ(1 −β)

**EMAT33100 - Nonlinear Dynamics & Chaos 13
**

Theorem 3 Consider the two linear vector ﬁelds

(1) ˙ x = Ax and (2) ˙ x = Bx,

where A and B are n × n matrices. If n

0

(A) = n

0

(B) = 0, that is, if the origin is a

hyperbolic equilibrium for both systems, then (1) and (2) are topologically equivalent if and

only if n

+

(A) = n

+

(B) and, consequently, also n

−

(A) = n

−

(B).

In particular, the above theorem says that

1

and

1a

, as well as

2

and

2a

are topologically

equivalent! For one-dimensional systems the phase portraits of two topologically equivalent

systems are identical. For two-dimensional systems, we need to be a little more ﬂexible.

Imagine that the phase portraits are drawn on a piece of rubber. If with a bit of pulling

and stretching two phase portraits are identical, then the associated dynamical systems

are topologically equivalent.

4 Stable and unstable manifolds

Linearisation about hyperbolic equilibria and topological equivalence are local results. We

know what a vector ﬁeld looks like near a hyperbolic equilibrium, but how do we paste

this information together to obtain the entire phase portrait? An important global object

for this purpose is the following.

Deﬁnition

Let x

0

be a saddle point of the vector ﬁeld ˙ x = f(x). The set of all points “ending up at

x

0

” under the ﬂow Φ

t

of the vector ﬁeld

W

s

(x

0

) = {x ∈ R

n

| Φ

t

(x) →x

0

as t →∞}

is called the stable manifold of x

0

. Similarly, all points “coming from x

0

”

W

u

(x

0

) = {x ∈ R

n

| Φ

t

(x) →x

0

as t →−∞}

is called the unstable manifold of x

0

.

The stable and unstable manifolds of saddle points in two-dimensional vector ﬁelds are

one-dimensional curves. In general, the dimension of the unstable manifold is equal to n

+

and the dimension of the stable manifold is n

−

. It is not easy to ﬁnd stable and unstable

manifolds; it is not possible, in general, to compute them analytically. However, we do

know the following:

Theorem 4 Let x

0

be a saddle point of ˙ x = f(x). Then x

0

is also a saddle point of the

linearised system

s ,

´

´

´

´

´

´

´

´

´

´´

´-

´

´´

.

`

E

u

(x

0

)

E

s

(x

0

)

W

u

loc

(x

0

)

W

s

loc

(x

0

)

˙ x = Df(x

0

) (x −x

0

).

Let E

s

(x

0

) and E

u

(x

0

) denote the stable

and unstable eigenspaces of the linearised

system. In a small enough neighbourhood

U of x

0

, there exists a local piece of W

s

(x

0

),

called W

s

loc

(x

0

), that is a smooth manifold and that can be considered as the graph of some

function h : E

s

(x

0

) → E

u

(x

0

). Furthermore, W

s

loc

(x

0

) is tangent to E

s

(x

0

) at x

0

. The

same is true for the local unstable manifold W

u

loc

(x

0

) which is deﬁned in the same way.

14 Bernd Krauskopf & Hinke Osinga

Example

The planar pendulum is a good example to illustrate the global (un)stable manifolds.

`

`

`

`

`

`

`

`

`

θ

}

M

>

>

>

>

>

>.

·

l

M g

Mg sin θ

˙

θ

The planar pendulum can be considered as a two-dimensional

dynamical system, very similar to the harmonic oscillator. The

equations of motion are

M l

¨

θ + c

˙

θ + M g sin θ = 0,

Where M is the mass, l is the length of the pendulum, g the

gravitational constant, and c the dampings coeﬃcient. By res-

caling time t →

l

g

t, and deﬁning x = θ, y =

l

g

˙

θ (or simply

assuming l = g = 1), we obtain equations in a simple form.

Deﬁne D =

c

m

√

g l

, then

˙ x = y,

˙ y = −Dy −sin x.

(9)

Since x = θ is the angular displacement from the vertical down position, this variable is

periodic with period 2π, which means that the phase space is a cylinder. Alternatively, we

can assume X = R

2

and we will see the periodicity in x from the phase portrait.

To analyse the dynamics, we compute the equilibria and determine their stability. The

equilibria are given by

y = 0 and sin x = 0 ⇔ x = k π, k ∈ Z.

The Jacobian matrix is

Df

¸

x

y

=

¸

0 1

−cos x −D

.

Hence, for the equilibria (2k π, 0) the matrix becomes

Df

¸

2k π

0

=

¸

0 1

−1 −D

,

with eigenvalues

λ

1,2

=

−D

2

±

1

2

√

D

2

−4

that are ±2i if D = 0 and complex conjugated for small D. For the equilibria (2k π +π, 0)

the matrix becomes

Df

¸

2k π + π

0

=

¸

0 1

1 −D

,

with eigenvalues

λ

1,2

=

−D

2

±

1

2

√

D

2

+ 4,

EMAT33100 - Nonlinear Dynamics & Chaos 15

so that always one is positive and one negative.

From the theory, we can conclude that equilibria (2k π + π, 0) are saddle points, with

eigenvectors

¸

1

1

and

¸

−1

1

**if D = 0. For small D > 0 the equilibria (2k π, 0) are (spiral)
**

attractors. The theory does not allow us to say something about the equilibriua (2k π, 0)

for D = 0, because these are not hyperbolic. However, we know that a pendulum without

friction will have periodic motion. Hence, near the equilibria, the dynamics becomes:

−π 0 π 2π −π 0 π 2π

No friction: D = 0 Friction: D > 0

If D = 0, then the periodic motion where the pendulum swings back and forth is separated

from the periodic motion where the pendulum goes over the top by the stable and unstable

manifolds. If the pendulum is (almost) exactly up (unstable situation) it will fall down

and again end up (almost) exactly up. This means that the stable and unstable manifolds

connect up as shown below.

6.3 -6.3

3

-3

−π 0 π

ր

ւ

ց

տ

ր

ւ

ց

տ

→

→

→

←

←

←

0

1

2

3

1

pendulum goes over the

top counter-clockwise.

2

pendulum does not go

over the top.

3

pendulum goes over the

top clockwise.

In the presence of friction, practically all initial conditions end up at the vertical down

position. The exception is formed by the stable manifolds, consisting of points that end

up in the vertical up position (highly exceptional!)

6.3 -6.3

3

-3

−π 0 π

ր

ւ

ց

տ

ր

ւ

ց

տ

→

→

← ←

0

All typical initial

conditions end up

in the rest position.

- New York university mathematical modeling lecture notes
- 485
- sepope
- no idea
- RMP_v066_p1389
- Uzawa-LucasModel
- Reference 1 9
- Timothy J. Lewis and Michael R. Guevara- Chaotic Dynamics in an Ionic Model of the Propagated Cardiac Action Potential
- 1+1=3.pdf
- Fractales y Dow Jones
- Dynamical Systems Theory
- Primer Adapted Dynamic Analysis in vibration problems
- -Victor-Schauberger-and-the-Turbulence-of-Water-♦-Franz-Pichler-♦-excerpted-paper-♦-Olafur-Eliasson-Surroundings-Surrounded-♦-2001
- General System Theory
- 00260881.pdf
- 04484213
- CBasquerotto_CILAMCE
- Duffing Equation
- 8._HamiltonianSystems.ppt
- 2 Chaotical Generators
- Chaos Theory a Brief Introduction IMHO
- ADL Dottorato2011 Part1 Slides
- mamatAMS37-40-2012
- fragile-watermarking
- PCA1 (1)
- Lecture 13
- AQA A maths grids
- Spark Med
- What is Complexity.pdf
- Carlos Castro- On Nonextensive Statistics, Chaos and Fractal Strings

Skip carousel

- tmp9457.tmp
- tmpD747
- Davis - Dead Cities and Other Tales (2002) - Synopsis
- Design a New Image Encryption using Fuzzy Integral Permutation with Coupled Chaotic Maps
- tmp63EF.tmp
- tmp209A.tmp
- tmpF6B3.tmp
- tmp8195.tmp
- tmpE0A5.tmp
- UT Dallas Syllabus for nats2v10.007.09s taught by (rch072000)
- tmp4ED8.tmp
- Strategic Theory for the 21st Century
- tmpF88A.tmp
- tmp6B2D.tmp
- tmp1B69
- tmpDD93.tmp
- Tmp 7904
- Classical Number Theory based Vs Chaos based Cryptography for Wireless Sensor Networks
- tmp1D19
- Napoleoni - Rogue Economics (2008) - Synopsis
- Tmp 2924
- Tmp 5608
- tmp47B
- Data Hiding Technique using Adaptive Pixel Pair Matching
- tmp15D3.tmp

Sign up to vote on this title

UsefulNot usefulRead Free for 30 Days

Cancel anytime.

Close Dialog## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Close Dialog## This title now requires a credit

Use one of your book credits to continue reading from where you left off, or restart the preview.

Loading