Lyapunov

Lecture 34
Stability Analysis of Nonlinear Systems
Using Lyapunov Theory II

Dr. Radhakant Padhi
Asst. Professor
Dept. of Aerospace Engineering
Indian Institute of Science - Bangalore
Outline
z
Construction of Lyapunov Functions
Definitions
Invariant Sets
Limit Sets
z
LaSalles Theorem
ADVANCED CONTROL SYSTEM DESIGN
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Construction of Lyapunov Functions
Dr. Radhakant Padhi

Asst. Professor
Variable Gradient Method:

V
= g ( X ) that contains some adjustable parameters
X
T
V
Note:
Then dV ( X ) =
dX
X
To recover a unique V ,
Select a V =
V
dV ( X ) = X dX
X =0
X =0
X
V ( X ) V ( 0) =
g ( X ) dX
V = g ( X ) must satisfy
the "Curl Condition":
i.e.
gi g j
=
x j xi
X =0
However, note that the intergal value depends on the initial and final
states (not on the path followed). Hence, integration can be conveniently
done along each of the co-ordinate axes in turn; i.e.

x1
V ( X ) = g1 ( x1 , 0,......, 0)dx1
0
x2
g ( x , x , 0,......, 0)dx
2
xn
( x1 ,......, xn1 , xn )dxn
Note: The free parameter of g ( X ) are constrained to

satisfy the symmetric condition, which is satisfied
by all gradients of a scalar functions.

Theorem: A function g ( X ) is the gradient of a scalar
g ( X )
function V ( X ) if and only if the matrix

X
is symmetric; where
g ( X )
g1
x1
g
n
x
1
g1
xn
g n
xn
Proof: Please see Marquez book

(Appendix)


Proof : (Necessity)
V
Assume: g ( X ) =
X
g ( X ) 2V
=
X
X 2
2V
x 2
1
=
2V
xnx1
2V
x1x2
2V
xnx2

2V
x1xn
2
V
2
xn
7

V
V
=
x1x j x j xi
2
gi g j
=
x j
xi
g ( X )
should be symmetric.
Hence,the matrix
X

Sufficiency: Assume
gi g j
=
x j
xi
To show
= g i ( X ) i


We have:
x
V (X )=
g ( x) dx
0
x1
= g1 ( x1 , 0,......, 0)dx1
0
x2
+ g 2 ( x1 , x2 , 0,......, 0)dx2
0
xn
+ g n ( x1 , x2 , ......, xn1 , xn )dxn

0
10

V
x1
= g1 ( x1 , 0,......, 0)
x2
0
xn
g 2
( x1 , x2 , 0,......, 0)dx2
x1
g n
( x1 , x2 , ......, xn1 , xn )dxn
x1
x2
= g1 ( x1 , 0,......, 0) +
0
xn
+
0
g1
( x1 , x2 , 0,......, 0)dx2 + ...
x2
g1
( x1 , x2 , ......, xn1 , xn )dxn
xn

11

x2
= g1 ( x1 , 0,......, 0) + g1 ( x1 , x2 , 0,......, 0 ) x =0
2
+ g1 ( x1 , x2 , ......, xn1 , xn ) xn =0
n
= g1 ( x1 , 0,......, 0) + [ g1 ( x1 , x2 , 0,......, 0) g1 ( x1 , 0,......, 0) ]

+
+ [ g1 ( x1 , x2 , ......, xn ) g n ( x1 , x2 , ......, xn , 0) ]
= g1 ( x1 , x2 , ......, xn )
i.e
Similarly
V
x1
= g1 ( X )
V
= gi ( X ) , i = 1,
xi
,n

12

Example
Problem: Analyze the stability behaviour of the following system
x1 = ax1
x2 = bx2 + x1 x22
Solution: X = 0 is an equilibrium point
V
= g(X )=
Assume
X
k1 k x1

k k2 x2
A symmetric matrix
Note: g1 = g 2 = k
x2
x1

13

Example
Further, let us assume
V g1 ( X ) k1 x1
=
=
X g 2 ( X ) k2 x2
x1
x2
V ( X ) = g1 ( x1 , 0) dx1 + g 2 ( x1 , x2 ) dx2
x1
x2
= k1 x1dx1 + k2 x2 dx2
=
k1 x12 + k2 x2 2 )
(
2

14

k1 , k2 > 0
Choose
Then V ( X ) > 0
X 0 and V (0) = 0
V ( X ) is a Lyapunov function candidate.

V (X )= g
( X ) f ( X ) = [ k1 x1
ax1
k2 x2 ]
2
bx2 + x1 x2
= k1ax12 + k2 (b + x1 x2 ) x22
Let us choose k1 = k2 = 1. Then
V ( X ) = ax12 + (b + x1 x2 ) x22
15

Unless we know about a, b at this point nothing can be
said about V ( X ). Let us assume
V (X )=
2
ax1
a > 0, b < 0. Then
( b x1 x2 ) x
2
2
> 0 (for small x1 x2 )
V ( X ) < 0 in some domain D R 2 and 0 D

i.e V ( X ) is negative definite in D
The system is locally asymptotically stable!
16
Krasovskiis Method
Let us consider the system X = f ( X )
f
: Jacobian matrix
Let A ( X )
X
Theorem :
If the matrix F ( X )
A ( X ) + AT ( X ) is ndf for all X D
(0 D ) ,
then the equilibrium point is locally asymptotically stable and a

Lyapunov function for the system is
V (X )= f T (X ) f (X )
Note: If D = R n and V ( X ) is radially unbounded,
then the equilibrium point is globally asymptotically stable.
17
Krasovskiis Method
Claim-1: Since F ( X ) is ndf, A( X )is invertible.
Proof (by contradiction):
Let A( X ) be singular
Then Y0 0 : A( X )Y0 = 0
But
Y0T F Y0 = Y0T ( A+AT )Y0
= Y0T ( AY0 ) + (Y0T AT )Y0 = 0

i.e.
F is not ndf.
Hence, A( X ) is non-singular (i.e., it is invertible).

18
Krasovskiis Method
Claim-2:
The invertibility (and continuity) of A ( X ) guarantees

that the function f ( X ) can be uniquely inverted.
Justification:
This is perphaps straight forward from unifrom convergence
property of Taylor series expansion.
This leads to the conclusion that the dynamic system has only one
equilibrum point in D. i.e. f ( X ) 0,
X 0, X D.
V ( X ) = f T ( X ) f ( X ) is pdf.
19
Krasovskiis Method
V (X )= f T f + f T f
T f
T f
X+X
f
= f
X
X
T
= f T ( AT + A) f
= f TF f
Hence, if F ( X ) is negative definite, V ( X ) is ndf.
So, by Lyapunov's theorem, X = 0 is asymptotically stable.
20
Krasovskiis Method
Note: The global asymptotic stability of the system is guaranteed
by the Global version of Lyapunov's direct method.
Comment: While the usage of this result is fairly
straight forward, its applicability is limited in
practice since F ( X ) for many systems
do not satisfy the negative definite property.

21
Generalized Krasovskiis Theorem

Theorem :
Let
f ( X )
A( X )
X
A sufficent condition for the origin to be asymptotically stable is that

two pdf matrices P and Q: X 0, the matrix
F ( X ) = AT P + PA + Q
is negative semi-definite in some neighbourhood D of the origin.
In addition, if D = R n and V ( X )
f T ( X ) P f ( X ) is radially unbounded,
then the system is globally asymptotically stable.

22
Generalized Krasovskiis
Theorem
Proof : V ( X ) = f T ( X ) Pf ( X )
V ( X ) = f T P f + f T P f
= f P
X
= f T PAT f +
f T
X Pf
X +
X
f T AP f
T
= f T ( PAT + AP + Q Q) f
= f T ( PAT + AP + Q) f f T Qf
nsdf
< 0 (ndf)
ndf
Hence, the result.

23
Example
Problem:
Analyze the stability behaviour of the following system

x1 = 6 x1 + 2 x2
x2 = 2 x1 6 x2 2 x23
Solution:
2
f
=
A =
2
2 6 6 x2
X
12
4
T
F = A+ A =
2
4 12 12 x2

24
Example
Eigen values of F :
+ 12
+ 12 + 12 x2
=0
( + 12) + ( + 12)12 x2 2 16 = 0
2
2 + 24 + 144 + 12 x2 2 + 144 x2 2 16 = 0
2 + ( 24 + 12 x2 2 ) + (128 + 144 x2 2 ) = 0
1,2
1
= (24 + 12 x2 2 )
2
(24 + 12 x2 )
2 2
4 (128 + 144 x2 )

25
Example
= (12 + 6 x2 )
2
(12 + 6 x2 ) (128 + 144 x22 )

2 2
0 < (*) < (12+6 x22 )
<0
x2 R
A is ndf in R 2
Morever, V ( X ) = f T ( X ) f ( X )
= (6 x1 + 2 x2 ) + (2 x1 6 x2 2 x2
2
as
3 2
X = 0 is globally asymptotically stable.

26
Invariant Sets & La Salles Theorem
Dr. Radhakant Padhi

Asst. Professor
Invariant Set
A set M is said to be an "invariant set" with respect to the
system X = f ( X ) if:
X ( 0) M
X (t ) M , t > 0
Examples:
(i) An equilibrium point (M = X e )
(ii) Any trajectary of an autonomous system ( M = X (t ))
28
Invariant Set
(iii) A limit cycle
(iv) M = R n
n
X
:
V
X
l
R
(
)
(v) l =
where, V ( X ) is a continuously differentiable function
0
X
=
f
X
such
that
along
the
solution
of
V
X
(
)
(
)
Note: (1) V ( X ) need not be pdf.
(2) The condition implies that once the trajectory crosses

the surface V ( X ) = c, it can never come out again.
29
Limit Set
Definition:
Let X (t ) be a trajectory of the dynamical
system X = f ( X ) .Then the set N is called
the limit set (or positive limit set) of X (t ) if
for any p N , a sequence of times {tn } [0, ]
such that
X (tn ) p as tn .
Note: Roughly, the limit set N of X (t ) is

whatever X (t ) tends to in the limit.
30
Limit Set
Example:
(i) An asymptotically stable equilibrium point is the
limit set of any solution starting from a close
neighbourhood of the equilibrium point.
(ii) A stable limit cycle is the limit set for any
solution starting sufficiently close to it
31
Some Useful Results

Lemma-1:
If the solution X (t , t0 , X 0 ) of the system X = f ( X )
is bounded for t > t0 ,then its limit set N is:
(ii) closed
(i.e. it is a non empty "compact set")
(iii) Non-empty
(i) bounded
Moreover, as t , the solution approaches N .

Lemma-2: The limit set N of a solution X (t , t0 , X 0 ) of the autonomous
system X = f ( X ) is invariant with respect to the same system.
32
A Useful Theorem
(Subset of LaSalles Theorem)
Theorem : The equilibrium point X = 0 of the autonomous system X = f ( X )
is asymptotically stable if:
(i) V ( X ) > 0 (pdf) X D
[0 D]
(ii) V ( X ) 0 (nsdf) in a bounded region R D

(iii) V ( X ) does not vanish along any trajectory in R
other than the null solution X = 0
Morever,
If the above conditions hold good for R =
and V ( X ) is radially unbounded,
then X = 0 is globally asymptotically stable.

33
Proof of the Theorem

V 0
The system is stable
i.e. for each > 0 , >0 :
X0
<
X (t ) <
or, Any solution starting inside the closed ball B

will remain within the closed ball B
The solution (starting within B ) is bounded.
34

Hence, X (t ) tends to its limit set N B
and B is compact.
(By Lemma - 1)
Moreover, V ( X ) is continous on the compact set B

and V ( X ) 0 ,
i.e V ( X ) = L
V ( X ) L 0 as t
X N
(N : the limit set)
Note that N is invariant set with respect to the

system X = f ( X )
(By Lemma - 2)

35

Any solution that starts in N will remain within it
for all furture time.

However, along that solution V ( X ) = 0 , as V ( X ) = L
But, by the assumption of the theorem, V ( X ) does not
vanish along any trajectory other than the null solution X = 0
Hence, Any solution starting in R B converges to
X = 0 as t

36
Example 1:
Pendulum with Friction
Example: (Pendulum with friction)
x1 = x2
k
g
x2 = sin x1 x2
m
l
1 2 2
V ( X ) = ml x 2 + mgl (1 cos x1 )
2
> 0 X D = (, ) R
V ( X ) = kl 2 x 2 2 : nsdf
[Note: 0 D]

37
Example 1:
Pendulum with Friction
Now let us examine the condition
V ( X ) = 0 t
kl 2 x 2 2 = 0
x2 = 0 t
x2 = 0. Hence
g
k
sin x1 + x2 = 0
l
m
sin x1 = 0 ( x2 = 0) x1 = 0
as x1 (, )
Hence, V ( X ) happens only for X = 0.

Hence, X = 0 is locally asymptotically stable !
38
Example 2
Example:
x1 = x2
x2 = x2 x1 ( x1 + x2 ) x2
2
Solution:
Let V ( X ) = x 21 + x2 2 , > 0
V
V ( X ) =
f X
X ( )
T
x2
= [ 2 x1 2x2 ]
2
x
x
x
x
(
)
2
1
1
2
2
= 2 x1 x2 2
x22
2 x1 x2 2 ( x1 + x2 ) x22

39
Example 2
2
V ( X ) = 2 x2 2 1 + ( x1 + x2 )
0 (nsdf)
Now V ( X ) = 0
x2 (t ) = 0
x2 = 0
x2 x1 ( x1 + x2 ) x2 = 0 (However, x2 = 0)
2
x1 = 0
x1 0
i.e. X = =
x2 0


40
Example 2
Here we have :
(i) V ( X ) does not vanish along any trajectory
other than X = 0
(ii) V 0 in R n
(iii) V ( X ) is radially unbounded,
Hence, the origin is Globally asymptotically stable.

41
LaSalles Theorem
Let V : D R be a continuously differentiable (not necessarily pdf) function
and (i) M D be a compact set, which is
invariant with respect to the solution of X = f ( X )
(ii) V 0 in M
(iii) E =
{X :
X M and V ( X ) = 0}
i.e. E is the set of all points of M : V = 0

(iv) N is the largest invariant set in E
Then Every solution starting in M approaches N as t .

42
Lasalles Theorem
Remarks:
(i) V ( X ) is required only to be continuously differentiable
It need not be positive definite.
(ii) LaSalle's Theorem applies not only to equilibrium
points, but also to more general dynamic
behaviours such as limit cycles.
(iii) The earlier theorems (on asymptotic stability) can
be derived as a corollary of this theorem.
43
References
z
H. J. Marquez: Nonlinear Control

Systems Analysis and Design, Wiley,
2003.
J-J. E. Slotine and W. Li: Applied

Nonlinear Control, Prentice Hall, 1991.
H. K. Khalil: Nonlinear Systems, Prentice

Hall, 1996.
44

45

Lyapunov

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Lyapunov

Загружено:

Авторское право:

Доступные форматы

Lecture 34

Stability Analysis of Nonlinear Systems

Using Lyapunov Theory II

Construction of Lyapunov Functions

Construction of Lyapunov Functions

Dr. Radhakant Padhi

Variable Gradient Method:

Variable Gradient Method:

( x1 ,......, xn1 , xn )dxn

Note: The free parameter of g ( X ) are constrained to

Variable Gradient Method:

function V ( X ) if and only if the matrix

Proof: Please see Marquez book

ADVANCED CONTROL SYSTEM DESIGN

Variable Gradient Method:

ADVANCED CONTROL SYSTEM DESIGN

Variable Gradient Method:

ADVANCED CONTROL SYSTEM DESIGN

Variable Gradient Method:

ADVANCED CONTROL SYSTEM DESIGN

Variable Gradient Method:

+ g n ( x1 , x2 , ......, xn1 , xn )dxn

Variable Gradient Method:

ADVANCED CONTROL SYSTEM DESIGN

Variable Gradient Method:

= g1 ( x1 , 0,......, 0) + [ g1 ( x1 , x2 , 0,......, 0) g1 ( x1 , 0,......, 0) ]

ADVANCED CONTROL SYSTEM DESIGN

Variable Gradient Method:

ADVANCED CONTROL SYSTEM DESIGN

Variable Gradient Method:

ADVANCED CONTROL SYSTEM DESIGN

Variable Gradient Method:

V ( X ) is a Lyapunov function candidate.

Variable Gradient Method:

a > 0, b < 0. Then

> 0 (for small x1 x2 )

V ( X ) < 0 in some domain D R 2 and 0 D

A ( X ) + AT ( X ) is ndf for all X D

then the equilibrium point is locally asymptotically stable and a

Y0T F Y0 = Y0T ( A+AT )Y0

= Y0T ( AY0 ) + (Y0T AT )Y0 = 0

Hence, A( X ) is non-singular (i.e., it is invertible).

The invertibility (and continuity) of A ( X ) guarantees

ADVANCED CONTROL SYSTEM DESIGN

Generalized Krasovskiis Theorem

A sufficent condition for the origin to be asymptotically stable is that

then the system is globally asymptotically stable.

Hence, the result.

ADVANCED CONTROL SYSTEM DESIGN

Analyze the stability behaviour of the following system

ADVANCED CONTROL SYSTEM DESIGN

ADVANCED CONTROL SYSTEM DESIGN

(12 + 6 x2 ) (128 + 144 x22 )

0 < (*) < (12+6 x22 )

X = 0 is globally asymptotically stable.

Invariant Sets & La Salles Theorem

Dr. Radhakant Padhi

Note: (1) V ( X ) need not be pdf.

(2) The condition implies that once the trajectory crosses

Note: Roughly, the limit set N of X (t ) is

Some Useful Results

Moreover, as t , the solution approaches N .

(ii) V ( X ) 0 (nsdf) in a bounded region R D