Sunteți pe pagina 1din 16

Outline

Lyapunov Stability Theory


M. S. Fadali
Professor of EE

Equilibrium of nonlinear system

Stability definitions.

Lyapunovs (second) direct method.

Class

functions.

Equilibrium Point

Lyapunov Stability

Consider autonomous systems.

x2

x1

The equilibrium

= open connected subset of


locally Lipschitz
Equilibrium point

is stable if

0 such that

Otherwise,

is unstable.

Can stay arbitrarily close to the equilibrium.


4

Convergence
The equilibrium

Convergence and Stability


Independent properties.
Convergence does not imply stability.
Example: Trajectories always go to a circle
of radius r before converging. Unstable.

is convergent if

s.t.
_|

x2

x1

Asymptotic Stability
The equilibrium
if it is both

Stable (i.s. Lyapunov)

2.

Convergent.

Equilibrium at the Origin

is asymptotically stable

1.

W.l.og., assume

If equilibrium is at
axes:

of
, translate the

Definition does not include rate of


convergence.

Example

Exponential Stability
The equilibrium
s.t.

is exponentially stable if

The equilibrium
is globally exponentially
stable if the condition holds
exponentially stable
asymptotically stable stable

Equilibrium:

10

Lyapunovs Direct Method

Directional Derivative

Examine stability of nonlinear system


directly.
Generalize concept of energy function.
Gives sufficient stability or instability
conditions (in general).
Possible difficulty, choice of suitable
Lyapunov (generalized energy) function.

Consider a continuous function f with


continuous partial derivatives.
Directional derivative of at in the
direction

_|

11

Some authors assume

12

Lie Derivative

The Lie derivative of

Definite Functions
along

= open connected subset of

Example

Positive Definite:
Negative Definite:
Positive Semidefinite:
Negative Semidefinite:
13

14

Example: Quadratic Forms

Lyapunov Function

Positive definite:

Decreasing (or non-increasing) along the


trajectories of the system.

Derivative negative definite (or semidefinite)

Positive Definite
Negative Definite
Positive Semidefinite
Negative Semidefinite

15

16

Lyapunov Surface
10

Boundary of a region in state


space where
Surface
encloses all
if
surfaces
Shape of surface depends on
: may or may not be the
sphere
1

Given a C1 positive definite function


in
If the derivative of
along the
trajectories of the system is

Lyapunov Stability Theorem

0.5

-0.5

-1

10

-2

-1.5
-3

a) negative semi-definite then the equilibrium is


stable in the sense of Lyapunov.
b) negative definite then the equilibrium is
asymptotically stable.
c) positive definite then the equilibrium is
unstable.

-1

17

18

Asymptotic Stability

Lyapunov Stability

inside

If
has a negative derivative along the
trajectories, the Lyapunov surface shrinks.

decreases inside the Lyapunov surface.


The surface continues to shrink along the
trajectories until it converges to its
minimum zero value.
The minimum value of the function is
(at the equilibrium point).

: a trajectory starting inside


never leaves it.
can choose s. t. the largest ball
is

x2

x1

19

20

Instability

Remarks

: Lyapunov surfaces grow along


grows continuously.
any trajectory and
diverges along any trajectory.
Note: condition is very severe because an
unstable system may have some stable
trajectories (e.g. saddle point).

The theorem provides sufficient


conditions for stability and sufficient
conditions for instability.
If the test fails, there is no conclusion.
It is often difficult to find a suitable
Lyapunov function for nonlinear systems.
For linear systems, the theorem can
provide a necessary and sufficient
condition.

21

22

Laypunov Stability Analysis

Scalar System (Slotine & Li)

Select a positive definite function


for
use in stability analysis: Lyapunov function
candidate.
2. Evaluate the derivative
along the
system trajectories.
3. Use the stability theorem: success means
= Lyapunov function
that
If the test fails, no conclusion (sufficient)
Can try another function
(it may be
difficult to find one that works)
1.

Show that the system is asymptotically


stable.
Lyapunov function

,
23

24

Example: Frictionless Pendulum

Stability Analysis

Equilibrium:
Physically motivated choice: total energy

Stable equilibrium.

25

Example: Pendulum + Friction

26

Stability Analysis

Equilibrium:
Physically motivated choice: total energy
Stable equilibrium (we know it is
asymptotically stable).
27

28

Using a Different

Global Stability (in the Large)


Requires
radially unbounded.
Radially unbounded
continuously differentiable and
as
uniformly in .

Example:
if either

locally

or

or

both.

Asymptotically stable equilibrium


29

Why Radially Unbounded?


1
0.8
0.6

Thm. Global Asymptotic Stability

1.6

1.6

1.2

1.6

30

1.2

8
0.

0.
8

1.2

4
0.

Given a C1 positive definite, radially


unbounded function

If the derivative of
along the
trajectories of the system is negative
then
is globally
definite
asymptotically stable.

0.4

0.4

0.
8

0.2
0
8
0.

0.4

-0.4
-0.6
-0.8

0.
4

Example:

-0.2

1.2

1.6

1.2

1.2

-1
2

-2.5

8
0.

0.
8

1.6

-2

-1.5

-1

1.6

-0.5

0.5

1.5

closed for
For
, trajectories can diverge while moving
towards a Lyapunov contour associated with a
lower value.

Proof: similar to earlier proofs and radially


unbounded prevents pathological cases.
31

32

Example

Chetaevs Theorem
The equilibrium
of
,
continuously differentiable in
is unstable if a function
satisfying

Equilibrium
pos. def. and radially

unbounded.

I.

arbitrarily close to

II.

Equilibrium
stable.

s.t.

III.

is globally asymptotically
33

Proof: Main idea

34

Theorem (Apostol, p.83)

Show that a trajectory starting at


arbitrarily close to in the set will cross
the sphere
,
arbitrary.
Unstable: cannot find
s.t.

If
subset

x2

is continuous on a compact
, metric space, then
s. t.

>0

>0

x1

Needed in the proof.

35

36

Proof

Proof Cont.
is nonempty
arbitrarily close to s.t.

since

bounded

Starting at
arbitrarily close to , the
trajectory leaves as
and crosses

For
bounded (compact) set

closed &
have bounds in

Unstable: cannot find

s.t.

37

Example (Khalil p.126)

38

Functions of Class
Class
with
I.

locally Lipschitz, satisfy


in
(

: continuous function

II.

Class
with

Equilibrium

strictly increasing.
: continuous function

I.
II.

is unstable

III.
39

strictly increasing.
as
40

Lemma 3.1

Example

is positive definite if and


only if class functions
and
s. t.
Choose

If
then
class

and
and

is radially unbounded,
can be chosen in
is positive definite and radially
unbounded.
41

Lemma 3.2
The equilibrium of
if and only if a class
and a constant s.t.

42

Class
is stable
function

Continuous function
s.t.
I. For fixed ,
is in class w.r.t. .
II. For fixed ,
is strictly decreasing
w.r.t. .
III.
as

Class function
: Upper bound
for stable trajectory.
43

44

Exponential Stability

Proof

Given a C1 positive definite function


. If
I.
is negative definite.
II.
s.t.

in

/
/
/

Then
is exponentially stable.
If the conditions hold globally,
is
globally exponentially stable.

/
45

Example

46

Invariance Principle

Show that the system is globally


exponentially stable (obvious).

Lyapunov function

In some cases, we can prove asymptotic


stability even if
.
Exploit the relationship between the state
variables of the system.
(Positively) Invariant Set :
is an
invariant set w.r.t. a dynamical system
if
Starting in , the system never leaves .

47

48

Examples: Invariant Sets

Limit Set

Equilibrium point
Limit cycle.
Any system trajectory.

= trajectory of

is a (positive) limit set of


a time sequence
s.t.
as
, i.e.

if

_|

= set of points to which

tends as

49

50

Example

Lemmas

Asymptotically stable equilibrium point:


limit set of any trajectory in its domain of
attraction.
Stable limit cycle
limit set of any sufficiently close trajectory.

If

is the (positive) limit set of

Then

i
ii
iii
51

= bounded solution of

is bounded, closed and nonempty.


approaches

as

is an invariant set w.r.t.


52

Theorem 3.7

Theorem 3.6

1.
2.
3.

The equilibrium point


asymptotically stable if
satisfying

of
a function

is

is positive definite
is negative semidefinite in a bounded
region
.
is only zero for the trivial trajectory
(nonzero on all nontrivial trajectories).

The equilibrium point


of
is
globally asymptotically stable if a function
satisfying

1.

is positive definite

2.

is negative semidefinite

3.

is only zero for the trivial trajectory


(nonzero on all nontrivial trajectories)

4.

is radially unbounded.

53

54

Example: Pendulum with Friction

Theorem 3.7
The equilibrium
of of
globally asymptotically stable if
satisfying
1.
is positive definite
2.

Equilibrium:

asymptotically stable.
55

is
a function

is negative semidefinite in a bounded


.
region

3.

is only zero for the trivial trajectory


(nonzero on all nontrivial trajectories)

4.

is radially unbounded.
56

Example

La Salles Theorem
continuously differentiable

unbounded

compact set invariant w.r.t. the solutions

, radially

of

= largest invariant set in

Then every solution starting in

is globally asymptotically stable.

approaches

as

57

58

Corollaries

Remarks
Unlike Lyapunov theory, La Salles Thm.
1. Requires
to be continuously
differentiable but not necessarily positive
definite.
2. Is applicable to any attractor and not just
an equilibrium point i.e. it can be used to
determine the stability of a limit cycle.
3. Largest invariant set =union of all
invariant sets.

continuously differentiable, positive


definite in
.
is negative semidefinite in

in

no solution remains
other than the trivial trajectory.

is an asymptotically stable
Then
equilibrium of

59

If
then

and
is radially unbounded
is globally asymptotically stable
60

Example

Stability of Limit Cycle


pos. semidef. In

Equilibrium:
Invariant Set:

iff (a)
For

1st Quadrant

(b)

, or
on circle of radius

61

62

Apply La Salles Theorem

Refine Answer

Given
define
compact (closed & bounded) by construction
and any trajectory starting in
remains in , i.e. is an invariant set

measure of distance to the limit cycle

Choose

excludes but includes the limit cycle


Limit cycle is stable (attractive) while is
unstable because arbitrarily close points to
converge to the limit cycle.

and
invariant sets
= largest invariant set in ,
La Salles Thm.: every motion in converges to
or
(limit cycle)
63

64