ME677c4 LyapunovTheorems

Lyapunov Stability Lecture Notes by B.
Yao
Lyapunov Stability
Stability of Equilibrium Points

1. Stability of Equilibrium Points - Definitions
System:
( , ), ( )
o o
x f t x x t x = =
`
(L.1)
or
( 1) ( , ( )), ( )
o o
x k f k x k x k x + = = , (L.2)
where x e R
n
, t e R
+
, k e Z
+
and x
o
is the initial state

Definition [Equilibrium State]
An equilibrium state x
e
is such that
- f(t, x
e
) = 0, for all t, for CT systems.
- f(k, x
e
) = x
e
, for all k, for DT systems.
Without loss of generality, we will assume that 0 is an equilibrium state. Notice
that non-linear systems (and some linear systems) may have more than one
equilibrium state.
Lyapunov Stability Lecture Notes by B. Yao
The idea:
Linear Systems:
The closed-form solution is known ( ( ) (0)
At
x t e x = for x Ax =
`
) and is
determined by system matrix A. Therefore, stability of the equilibrium point
0 x = can be studied based on certain properties associated with the
system matrixA, namely, the eigenvalues of A or the poles of TFs.
Nonlinear Systems:
The closed-form solution is in general unknown and ( , ) f x t can have
varieties of different forms. However, at any given time and state x, ( , ) f x t
can be calculated and thus we know the direction that the solution is
evolving (since ( , ) x f x t =
`
). This information enables us to judge if an
energy-like positive definite (p.d.) scalar function will continuously decrease
or not. If the p.d. scalar function continuously decrease, then, it is obvious
that the solution ( ) x t will not become unbounded and the system is stable.
The whole idea is conceived to mimic the physical phenomena that if we
dissipate the energy of a system, (i.e., the p.d. energy function continuously
decreases), then, the system will eventually come to rest and the solution
converges to the equilibrium point. Lyapunov stability theory is a
formalization of this type of energy based stability analysis.

Definition [Ref.1]
[Stability and Uniform Stability in the sense of Lyapunov]
The equilibrium state 0 of (1) is (locally) stable in the sense of Lyapunov if for
every c > 0, there exists a ( , ) 0
o
t o c > such that, if ( )
o
x t o < then ( ) x t c < for
all t > t
o
(respectively k
o
for DT).
In addition, if o can be chosen independent of t
o
, i.e., ( ) o c , then, the origin is
(locally) uniformly stable.

stable
c = x
o = x
unstable
c = x
o = x
Asymptotically stable
c = x
o = x
stable
c = x
o = x
unstable
c = x
o = x
Asymptotically stable
c = x
o = x
Definition [Ref.1]
[Asymptotic Stability and Uniform Asymptotic Stability]
The equilibrium state 0 of (1) is (locally) asymptotically stable if
1. It is stable in the sense of Lyapunov and
2. There exists a ( )
o
t o
'
such that,
If ( ) , then, ( ) 0 as
o
x t x t t o
'
< .

The equilibrium state 0 of (1) is (locally) uniformly asymptotically stable if
1. It is uniformly stable in the sense of Lyapunov and
3. There exists a o
'
, independent of t
o
, such that, if ( )
o
x t o
'
< , then,
( ) 0as , x t t uniformly in t
o
; that is, for each c > 0, there exists
T=T(c)>0, independent of t
o
, such that

x t t t T x t ( ) , ( ), ( ) < > + < ' c c o
0 0
Note:
The definition of stability in the sense of Lyapunov is closely related to that
of continuity of solutions. An equilibrium is stable if all solutions starting at
nearby points stay nearby; otherwise, it is unstable. It is asymptotically
stable if all solutions starting at nearby points not only stay nearby, but also
tend to the equilibrium point as time approaches infinity.

Definition [Global asymptotic stability]
The equilibrium state 0 of (1) is globally asymptotically stable, if it is
asymptotically stable for any 0 o
'
> .

Definition [Exponential stability]
The equilibrium state 0 of (1) is exponentially stable, if it is stable in the sense
of Lyapunov and there exists a 0 o
'
> and constants M < and o > 0 such that

( )
( )
o
t t
o
x t e M x
o
s (L.3)
for all ( )
o
x t o
'
< . o is called the rate of exponential convergence.
Example:
Stable and Asymptotic Stable but Not Uniformly Stable

Consider the system
( )
8 sin 2 x t t t x =
`

which has the closed-form solution
( )
0
0
2 2
0 0 0 0 0
( )exp 8 sin 2
( )exp 8 cos 8sin 8 cos 8sin
t
x x t d
x t t t t t t t t t
t t t t
(
=
(

(
= + + +

}

For any
0
t , the exponential term will eventually be dominated by
2
t . Thus,
the exponential term is bounded for all
0
t t > by a constant
( )
0
c t dependent
on
0
t . Hence,
0 0
( ) ( ) ( ) x t x t c t s ,
0
t t >
0 c > ,
0
0
( , )
( )
t
c t
c
o c - = s.t.

0 0
( ) ( , ) x t t o c < , we have, ( ) x t c < ,
0
t t >

the origin is stable (in fact, asymptotically stable).

However, the system is NOT uniformly stable since there does not exit a
( ) 0 o c > such that
0
( ) ( ) x t o c < implies ( ) x t c < ,
0
t and
0
t t > . To prove
this, use contradiction argument. Suppose that the statement is true, i.e.,
0 c > , ( ) 0 o c - > s.t
0
t ,
0
( ) ( ) x t o c < , ( ) x t c <
0
t t > . Let
0
2 t nt = ,
1
2 t nt t = + . Then
( )( )
1 0
( ) ( )exp 4 1 6 2 x t x t n t t t ( = + +

Note that
( )( )
exp 4 1 6 2 n t t t ( + +

as n . Then, for
0
1
( ) ( )
2
x t o c = , n - s.t.
1
0
( )
2
( )
x t
x t
c
o
>
1 0
2
( ) ( ) x t x t
c
c
o
> =

which contradicts with the assumption (i.e,
0
t ,
0
( ) ( ) x t o c < , ( ) x t c <
0
t t > ).
Example:
Uniformly Stable and Asymptotically Stable but Not Uniformly
Asymptotically Stable

Consider the first-order system given by
1
x
x
t
=
+
`

0
0
0 0
1 1
( ) ( )exp ( )
1 1
t
t
t
x t x t d x t
t
t
t
+
| |
= =
|
+ +
\ .
}

Thus,
0
t t > ,
0
( ) ( ) x t x t s and the origin is uniformly stable and furthermore
asymptotically stable. However, it is not uniformly asymptotically stable. To
see this, o>0, let
0
1
( )
2
x t o
'
= and pick up a small c such that
1
2
0 ' c o < < .
Calculate the time that it takes for the solution to be in the c-set:
0
0
1
( ) ( )
1
t
x t x t
t
c
+
= <
+

( )
0
1 1
2
t t
o
c
'
> +

0 0 0
1 1 as
2 2
t t t t
o o
c c
' '
| |
> +
|
\ .

Thus, there does not exist a finite time ( ) T c , which is independent of
0
t ,
such that
0
( ) t t T c > + ( ) x t c < . By definition, the origin is not uniformly
asymptotically stable.

2. Lyapunov Stability Theorems for Autonomous Systems
When f in (1) does not depend on time t explicitly, i.e.,
( ), ( )
o o
x f x x t x = =
`
(L.4)
for continuous time or
( 1) ( ( )), ( )
o o
x k f x k x k x + = = (L.5)
for discrete time, then, the system becomes an autonomous system. The
behavior of an autonomous system is invariant to shifts in the time origin. Thus,
the solution x(t) depends on x
0
and t-t
0
only, and is independent of t
0
. This
leads to the following fact:

For autonomous system, uniform (asymptotic) stability is the same as
(asymptotic) stability.

Definition [Positive Definite (Semi-Definite) Function (PDF)]
A continuously differentiable function V: R
n
R is called positive definite in a
region UcR
n
containing the origin if
a. V(0)=0
b. V(x)>0, xeU and x=0
A function is called positive semi-definite if Condition b is replaced by V(x)>0.
Note:
i. For a p.d. function ( ) V x in
n
R , the set
{ }
: ( )
n
c
S x R V x c = e s for any
0 c > is a closed set containing the origin. Let
c
O be the largest connected
set in
c
S that contains the origin. Then,
c
O is a closed set and
1 2
c c
O _ O
for any
2 1
0 c c > > . Furthermore, there exists a 0
M
c > such that
M
c
O is
bounded and has a ( ) V x value of
M
c on its boundary, i.e., ( )
M
V x c = ,
M
c
x ecO . Consequently,
2
1
o o
M
c c
c
O c O _ O ,
1 2 M
c c c < s .

ii. 0 r > , let min ( )
rl
x r
c V x
=
= and max ( )
ru
x r
c V x
=
= as
{ }
:
n
r
B x R x r c = e = is
a compact set and ( ) V x is continuous. Then, 0
rl
c c < < , as
r
B c
separates
n
R into two disjoint spaces and
r c
B S c = C,
o
r
c
B O c .
iii. When ( ) V x is radially unbounded,
M
c could be any positive value.
Example 1: Example 2:
Theorem L.1 [Theorem 4.1 of Ref1] [Lyapunov Theorem]
For autonomous systems, let DcR
n
be a domain containing the equilibrium
point of origin. If there exists a continuously differentiable positive definite
function V: DR such that
( ) ( )
V dx V
V f x W x
x dt x
c c
= = =
c c
`
(L.6)
is negative semi-definite in D, then, the equilibrium point 0 is stable. Moreover,
if W(x) is positive definite, then, the equilibrium is asymptotically stable.
In addition, if D=R
n
and V is radially unbounded, i.e.,
( ) x V x (L.7)
then, the origin is globally asymptotically stable.
A
Proof of Theorem L.1:
As ( ) V x is p.d., there exists a 0
M
c > such that
M
c
O is bounded,
M
c
D O c ,
and ( )
M
V x c = ,
M
c
x ecO , where
M
c
O represents the largest connected set
containing the origin with a ( ) V x value less or equal to
M
c .

Claim 1: 0
M
c c < < , any solution starting from the set
c
O will remain in
c
O ,
i.e.,
0 c
x eO , we have
( )
0
; (0) , 0
c
x t x x t = eO > .

Proof: For any trajectory starting within
c
O ,
0
(0)
c
x x = eO or
0
( (0)) V V x c = s .
As 0 V s
`
in
M
c
D O c and the fact that ( ( )) ( ( )) ( ( ))
o
t
o
t
V x d V x t V x t t t =
}
`
,
0
( ( )) ( (0)) ( ( )) ( (0)) , 0
t
V x t V x V x d V x c t t t = + s s >
}
`

as long as the solution ( ) , 0
M
c
x t t t eO s s , which can be proved by noting
that (0)
M
c
c
x eO c O
1
. Thus, 0, ( )
c
t x t > eO . #

1
Suppose not, then, there exists
1 1
such that ( ) .
M
c
t x t t s eO As
(0)
M
c
c
x eO c O
, there exists
2 1
t t s such that
2 2
( ) and ( ) , 0
M M
c c
x x t t t t ecO eO s s . Thus,
2
( ( )) V x c t s
. Noting that
2
( ) x t
is also connected to
(0)
c
x eO
, we have
2
( )
M
c
c
x t eO c O
, a contradiction with the

assumption that
2
( )
M
c
x t ecO .
0 c > , choose 0 r > such that
{ }
:
M
n
r c
B x R x r B D
c
= e s c O , where
{ }
:
n
B x R x
c
c = e s . The following is to show that there is a 0 o > such
that any trajectory starting from the ball
{ }
:
n
B x R x
o
o = e s will remain in
the ball
r
B , and thus in B
c
, to prove the local stability of the origin.

As the surface of
r
B , i.e.,
{ }
:
n
r
B x R x r c = e = , is a compact set and ( ) V x
is continuous, ( ) V x will reach its maximum and minimum value on the set.
So let
rl
c be the minimum value, i.e., min ( )
rl
x r
c V x
=
= . Then, 0
rl
c > as
r
B c
does not contain the origin and ( ) V x is a positive definite function, and
rl M
c c s as
M
r c
B c O . It is thus clear from Claim 1 that for any 0 c > and
rl
c c < , the set
c
O is positive invariant, i.e., any trajectory starting from the set
will remain within the set
c
O . So any ball B
o
within the set
c
O will serve our
purpose as
c r
B O c . Such a B
o
does exist as max ( ) 0
u
x
c V x
o
o =
= > could be
arbitrarily small as 0 o due to the fact that ( ) V x is continuous and
(0) 0 V = . In fact, any o such that
u rl
c c c
o
s < will do as
u
c c
B
o
o
c O c O . So
every trajectory starting from B
o
will remain in
u
c
o
O , and thus in
r
B and B
c
.
By definition, 0 is locally stable in the sense of Lyapunov.
For asymptotic stability, we want to prove that
, ( ) 0, s.t. ( ) ,
a a a o
T x t t T t c c c c < - > s > +
Proof:
0 ( ( )) V V x t <
`
decreases all the time ( ( )) V x exists as
( ) 0. V x > Say ( ( )) 0, i.e., ( ( )) ( ( )) . V x c V x t V x c = > = We want to
prove that 0 c = to complete the proof. For this purpose, let us use the
contradiction argument. Namely, assume that 0 c > and show that this will
lead to a contradiction:
( ( )) as V x t c t in decreasing fashion 0, 0 T
|
c > - > s.t.
( ( )) .
o
V x t c t T t
|
c < > + ( ( )) c V x t c
|
c s < +
Consider the compact set
{ }
: ( )
n
H x R c V x c
|
c = e s s +
Since ( ) V x
`
is continuous w.r.t. x,
min ( ( )) 0 ( ) ,
x H
V x V x x H
e
= > s e
` `

Let
1
:
o
t T t = +
1
1
1
1 1 1
( ( )) ( ( )) ( ( ))
( ( )) ( ) ( ( )) ( )
t
t
t
t
V x t V x t V x d
V x t d V x t t t
t t
t
= +
s + =
}
}
`

which indicates that as , ( ( )) , t V x t a contradiction!
Example: Pendulum Motion
State-Equation

1 2
2 1 2
sin
x x
x g x bx
=
`
`

2
sin ml mgl b u u u =
`` `

where
1 2
2
, , ,
g b
x x g b
l ml
u u = = = =
`

Equilibrium Points:

2
2 2
1 1
1 2
0
0 0
sin 0 0, , 1,2,
0 sin
x
x x
x x m m
g x bx
t
=
= =

`
= = =
=

)

In reality, there are only two equilibrium points: up position
( )
0, 0
T

and down position
( )
, 0
T
t
Stability of the Equilibrium Point
( )
, 0 t :
Define
1 1
2 2
x x
x x
t =
=

Equilibrium point
( )
, 0
T
t corresponds to
( )
0, 0
T
in
( )
1 2
,
T
x x plane.

New State-Equation:
1 2
2 1 2
sin
x x
x g x bx
=
`
`

Energy-function:
( )
1
2 2
1 1 2 1 2
0
1 1
( ) sin 1 cos
2 2
x
V x g x dx x g x x = + = +
}

Note that
(0) 0 V = , and ( ) 0 V x > ,
( )
{ }
1
: , x D x x t t e = e
( ) V x is locally positive definite.

( )
1 2 1 2 2 1 2
1 2
2
2
( )
( ) ( ) sin sin
-
W x
V V V
V x f x x x g x x x g x bx
x x x
bx
c c c
= = + = +
c c c
=
` ` `

Since ( ) W x is positive semi-definite (why?), the equilibrium point
( )
0, 0
T
x = is locally stable by Lyapunov Theorem L1.

Q: Is the equilibrium point locally asymptotically stable?

Cannot judge asymptotic stability by simply applying Lyapunov Theorem
L1 in the above analysis. The reason is that ( ) W x is only positive semi-
definite, and Lyapunov Theorem L1 needs ( ) W x to be positive definite to
conclude asymptotic stability. To obtain asymptotic stability, we have two
options:

Option 1:
Try to find another Lyapunov function ( ) V x such that ( ) V x
`
is negative
definite as follows. Define a new function as
( )
( )
1
2 2 2
1 1 2 2 1
1
( ) 1 cos
2
1 1 1
1 cos
4 2 2
T
s
V x x Px g x
b x bx x x g x
= +
= + + +
,
2
1 1
2 2
1
1
2
b b
P
b
(
(
=
(
(
(

Note that
(0) 0
s
V = , and ( ) 0
s
V x > , x D e since 0 P >
( )
s
V x is positive definite.

( )
2 2 2
1 2 2 1 1 2 2
2
2 1 1
1 1 1
( ) sin
2 2 2
1 1
sin
2 2
s
V x b x x bx bx g x bx bx
bx bgx x
= + +
=
`

which is negative definite over
( )
{ }
1
: ,
s
D x x D t t = e c (why?). By
Lyapunov Theorem L1, x =0 is asymptotically stable.

Conservative Estimate of Region of Attraction:

{ }
: ( )
M
c s s M
x D V x c O = e s
where
{ }
{ }
1
2 2
inf ( )
inf ( )
3
2
8
s
M x D s
s x
c V x
V x
b g
t
t
ec
=
<
=
= +

Option 2:
Use LaSalles Invariance Principle Theorem as introduced below.
For autonomous systems, when ( ) W x in the above theorem is only positive
semi-definite, asymptotic stability may still be obtained by applying the
LaSalles Theorem introduced below. For this purpose, we need to introduce a
few definitions. Let ( ) x t be a solution of (L.4). A point p is said to be a positive
limit point of ( ) x t if there is a sequence
{ }
n
t , with
n
t as n , such that
( )
n
x t p as n . The set of all positive limit points of ( ) x t is called the
positive limit set of ( ) x t . A set Mis said to be an invariant set with respect to
(L.4) if a solution belongs to Mat some time instant, then it belongs to M for
all future and past time, i.e.,
(0) ( ) , x M x t M t e e
A set Mis said to be a positively invariant set if (0) ( ) , 0 x M x t M t e e > .

Theorem L.2 [Ref1] [LaSalles Invariance Principle Theorem]
For autonomous systems, let DcR
n
be a domain containing the equilibrium
point of origin and D Oc be a compact set that is positively invariant with
respect to (L.4).Let V: DR be a continuously differentiable function(not
necessarily positive definite) such that
( ) ( ) 0
V
V f x W x
x
c
= = s
c
`
(L.8)
in O. Let

{ }
| ( ) 0 S x V x = eO =
`
(L.9)
and Mbe the largest invariant set in S. Then, every solution starting in O
approaches M as t A

A direct application of LaSalles Invariance Principle indicates that, in Theorem
L.2, if S contains no solution other than the origin, then, the origin is
asymptotically stable with a region of attraction containing O.

Example: Pendulum Motion

Lemma L.1 [Lemma 4.1 of Ref1]:
If a solution of (L.4) is bounded and belongs to D for 0 t > , then, its positive
limit set L
+
is a non-empty, compact, invariant set. Moreover, ( ) x t
approaches L
+
as t .

Proof:
Since ( ) x t is bounded, by the Bolzano-Weierstrass theorem, it has a
converging subsequence { ( ), 1,2, }
n
x t n = that ( )
n
x t y as n . By
definition, y L
+
e . This shows that L
+
is non-empty.

To prove L
+
is compact, we need to show that L
+
is bounded and closed.
y L
+
e , by definition, { } s.t. ( )
n n
x t x t y - as n . Since
{ }
( ) ,
M
x t x t s , then,
M
y x L
+
s is bounded.

To prove closeness of L
+
, let us assume that if y is a limit of
{ }
, 1,2,...,
k
y L k
+
e = , i.e., as .
k
y y k Since
k
y y by assumption,
( )
1 1
0, 0, s.t. , N k N c c > - > >
2
k
y y
c
<

11 12 1 1
( ) ( ) ( )
n
x t x t x t y
21 22 2 2
( ) ( ) ( )
n
x t x t x t y
31 32 3 3
( ) ( ) ( )
n
x t x t x t y
. . . .
1 2
( ) ( ) ( )
n n nn n
x t x t x t y
+
y
Let
1 11
, t t = and pick
2
t to be one of
2
{ }
k
t s.t.
2 12 2 2
1
and ( ) .
2
t x y t t > <
Similarly, pick
3
t to be one of
3
{ , }
k
t k I e s.t.
3 13 3 3
1
and ( ) .
3
t x y t t > <
Let us now consider { ( ), 1,2,...}
n
x n t = . This sequence has the following
properties. First, when n ,
n
t . Secondly,
1
( )
n n
x y
n
t < . Thus,
2
0, ( ) 0, N c c > - > s.t.,
2
, k N > ( )
2
k k
x y
c
t < . Therefore, for any n s.t.
1 2
( ) max{ ( ), ( )}, n N N N c c c > =
( ) ( )
2 2
n n n n
x y x y y y
c c
t t c s + < + =

which shows that ( )
n
x y t as n . By definition, y L
+
e . This shows
that L
+
is closed.

To prove L
+
is invariant, let us assume that y L
+
e and look at the solution
( ; ) x t y with an I. C. of (0) x y = . We need to prove that , ( ; ) t x t y L
+
e .

Let ( ; )
o
x t x be the original solution when defining y L
+
e . By definition, { }
i
t - ,
s.t. as , i & ( ; ) .
i i o
t x t x y Consider the time sequence { }
i
t t +
where t is any fixed value.
( )
( ; ) ; ( ; )
i o i o
x t x x x t x t t + =
Since ( ; )
i o
x t x y , by the continuous dependence of solutions on the initial
conditions
( )
lim ( ; ) lim ; ( ; ) ( ; )
i o i o
i i
x t x x x t x x y t t t

+ = =
which indicates that ( ; ) x y L t
+
e . This shows that L
+
is invariant w.r.t. (L.4).

0 V V s
`
is non-increasing.
V is differentiable on a compact set O V is lower bounded.
( ( )) as V x t c t .
Consider L
+
for ( ; ). ,
o
x t x p L
+
e by definition,
n
t - s.t. as , n
& ( ; )
n n o
t x t x p . Thus,
( ( ; )) & ( ( ; )) ( )
n o n o
V x t x c V x t x V p .
So ( ) V p c = . Now consider ( ; ) x t p . , ( , ) t x t p L
+
e . Thus, , ( ( ; )) t V x t p c = .
0 ( ; ) V x t p S L M
+
= e c
`

By Lemma L.1, ( ; )
o
x t x L M as t
+
c .

2.2 Linear Time Invariant System

Theorem L.3
The following conditions are equivalent:

(a) The equilibrium 0 of the nth order system
x Ax =
`
(L.10)
is globally asymptotically stable (exponentially stable).

(b) All eigenvalues of A have negative real parts.

(c) For any positive definite symmetric matrix Q, there exists a unique
positive definite symmetric matrix P which is the solution of the
following Lyapunov equation
2

T
PA A P Q + = . (L.11)

Note: (c) indicates that the PDF function V (x) = x
T
Px is a Lyapunov function for
the system.

2
The MATLAB command for solving Lyapunov equation is lyap in continuous time and
dlyap in discrete time.
Proof: We will demonstrate that (c) is a necessary and sufficient condition for
(a) and (b).
Sufficiency:
Assume that given a positive definite symmetric matrix Q there exists a
positive definite symmetric matrix P which satisfies (L.11). Define the PDF
function V(x)=x
T
Px. Taking the time derivative of V along the trajectories of
(L.10), we obtain

{ }
( )
T T T
V x x A P PA x x Qx = + =
`
(L.12)
Thus, V and V
`
are both PDF, and the system is globally asymptotically
stable. To prove exponential stability, we notice that
min max
( ) , ( ) ,
T T T T
x Qx Q x x x Px P x x > s
where
min max
( ) and ( ) Q P are respectively the minimum eigenvalue of Q
and the maximum eigenvalue of P, both of which are positive. Thus, defining
min max
( )/ ( ) 0 Q P o = > , and using the notation ( ) ( ( )) V t V x t = , we obtain
min
max
( ) ( )
( ) ( )
T
T
V x Q x x
V x P x x
> =
`

Thus,
( ( )) ( ( )) V x t V x t o s
`
. (L.13)
Integrating (L.13), we obtain
( ( )) ( (0))
t
V x t e V x
o
s (L.14)
which implies that V 0 exponentially. Since
2
min min
2
( ) ( ) ( )
T
V x P x x P x > = ,
where
min
( ) 0 P > is the minimum eigenvalue of P, x must also converge to
zero exponentially.

Necessity:
We first define the unit ball:
{ }
1
: 1
n T
B v R v v = e = ,
and the vector induce
2
norm of a matrix M
n n
R

e :
{ }
1 1
max max
2 2
max max ( ) ( )
T T T
v B v B
M Mv v M Mv M M M o
e e
= = = =
Assume that the system given by (L.10) is asymptotically stable. Thus, all
eigenvalues of A have negative real parts and, as a consequence, A is
nonsingular and so is the solution matrix , for 0
At
e t s < . Since Q is positive
definite, there exists a nonsingular matrix,
1/ 2
Q , such that
1/ 2 1/ 2
( ) ( )
T
Q Q Q = .
Thus, the matrix
T
A t At
e Qe is positive definite for 0 s t < . Since all
eigenvalues of A have negative real parts, the matrix

0
T
A t At
P e Qe dt
=
}
(L.15)
exists, is unique and
2
P < . (Remember that
1
,
t m t
e t e L L

e for any
e C such that Re() < 0 and any 0 < m < ). Thus, P, as defined by (L.15),
is positive definite. We now prove that P, as defined by (L.15), satisfies the
Lyapunov equation (L.11).

{ }
0
T T
T T A t At A t At
A P PA A e Qe e Qe A dt
+ = +
}

{ } { }
0
/ lim
T T
A t At A t At
t
d dt e Qe dt e Qe Q Q
= = =
}
,
since
2
lim 0
At
t
e
= . Q.E.D.

2.3 LTI Discrete Time Systems
Definition [Change of V(k,x) relative to a state trajectory]
Consider the system (L.2). The change of AV(k+1,x) relative to (L.2) is given
by
( 1, ) ( 1, ( 1)) ( , ( )) V k x V k x k V k x k A + = + + (L.16)
( 1, ( , ( ))) ( , ( )). V k f k x k V k x k = +

Theorem L.4 The following conditions are equivalent:
(a) The equilibrium 0 of the nth order system
x(k+1) = Ax(k) (L.17)
is globally asymptotically stable (exponentially stable).
(b) All eigenvalues of A have magnitudes less than 1.
(c) For any positive definite symmetric matrix Q, there exist a unique
positive definite symmetric matrix P which is the solution of the following
Discrete Time Lyapunov equation

T
A PA P Q = . (L.18)

Proof: The proof is very similar to the continuous time case and its left as an
exercise. Q.E.D.
2.4 Lyapunovs Indirect Method

Theorem L.5 [Theorem 4.7 of Ref1]
Consider the autonomous system (L.4) with the origin as an equilibrium
point. If f: DR
n
is continuously differentiable and D is a neighborhood of
the origin. Let

0
( )
x
f x
A
x
=
c
=
c
(L.19)
Then,
(a) The origin is locally asymptotically stable if A is asymptotically stable
or all eigenvalues of A have negative real parts.
(b) The origin is unstable if one or more of the eigenvalues of A has
positive real part. A

Note:
(1). Both the Lyapunovs indirect method (Theorem L.5) and the direct
method (Theorem L.1) can be used to judge the local stability of an
equilibrium point when the linearized system matrix A is either
asymptotically stable or unstable. However, the indirect method does
not tell anything about the region of attraction
3
(or domain of attraction)
while the direct method gives at least some conservative estimate of the
domain of attraction. For example, if conditions for asymptotic stability in
Theorem 1 are satisfied and O
c
={xeR
n
| V(x) s c} is bounded and
contained in D, then, every trajectory starting in O
c
remains in O
c
and
approaches the origin as t. Thus O
c
is an estimate of the region of
attraction.

(2). When some of the eigenvalues of A have zero real parts and all the rest
eigenvalues have negative real parts, local stability of the origin cannot
be concluded from the above theorem. In such a case, local stability of
the origin depends on higher-order nonlinear terms also. Advanced
stability theorems such as Center Manifold Theorem may be used to
judge local stability of the origin.

3
The region in which all trajectories converge to the equilibrium point as t approaches .
(a).
( )
0
( )
(0) 0: ( ) . . . & lim 0
g x
x
g x
f f x Ax H OT
x
= = + =
,

By assumption, Re{ ( )} 0,
i
A i < . Thus, 0, 0 s.t. Q P > - >
T
PA A P Q + = .
Consider ( ) ( ) is p.d.
T
V x x Px V x = . and
( ) [ ( )] [( ) ]
{ } 2 ( ) 2 ( )
T T T T T
T T T T T
V x x Px x Px x P Ax g x Ax g Px
x PA A P x x P g x x Qx x P g x
= + = + + +
= + + = +
`
` `

As
0
( )
lim 0
x
g x
x
= 0 > , 0, s.t. { : },
r
r x B x x r - > e = <
( ) g x x s . Thus, :
r
x B e
2
min max
2
min max
( ) 2 ( ) ( ) 2 ( )
[ ( ) 2 ( )]
T
K
V x x Qx x P g x Q x x P x
Q P x

s + s +
=
`

Let
min
max
( )
2 ( )
Q
P
< . Then, 0 K > and is n.d. V

`
. By Lyapunov Theorem, 0 is
asymptotically stable.
Stability of Equilibrium Trajectories or Time-Varying Solutions
Suppose ( ) y t is the solution to the system
( ) ( , ) y t g t y =
`

i.e.
( ) ( , ( )) y t g t y t =
`

We would like to study stability of the solution ( ) y t . The problem arises when
we are interested in either the stability of a limit cycle of a system
(correspondingly, ( ) y t represents the limit cycle) or tracking a time-varying
trajectory in a control problem (correspondingly, ( ) y t represents the desired
state trajectories). Define a new state-variable x as

( ) ( ) ( ) x t y t y t =
Then
( ) 0 x t = ( ) ( ) y t y t =
and
( ) ( ) ( , ) ( , ( ))
( , ( )) ( , ( ))
( , )
x y t y t g t y g t y t
g t x y t g t y t
f t x
= =
= +
A
`
` `

Note that ( ,0) 0 f t = , and the origin 0 x = is an equilibrium point for the
transformed system. Thus, the stability of the time-varying solution ( ) y t is
equivalent to the stability of the origin of the transformed system.

Note:
Even though the original system may be autonomous, i.e., ( ) ( ) y t g y =
`
, the
transformed system is non-autonomous since

( , ) ( ( )) ( ( )) f t x g x y t g y t = +

is in general an explicit function of t due to the appearance of ( ) y t .
Lyapunov Stability Theorems for Non-autonomous (or Time-Varying)
Systems
Consider the non-autonomous system (L.1) where f: [0,)DR
n
is
piecewise continuous in t and locally Lipschitz in x on [0,)D, and DcR
n
is
a domain that contains the equilibrium point of origin x=0. Note that an
equilibrium at origin of a non-autonomous system could be a translation of
a non-zero time-varying solution of an autonomous system (e.g., trajectory
tracking of an autonomous system). The solution of a non-autonomous
system may depend on both t-t
0
and t
0
, and the Lyapunov function V(x,t) in
general depends on t also. To characterize the positive definiteness of a
time function, we introduce the following addition definitions.

Definition [Ref.1] [Class-K Function]
A continuous function o: [0,a) R
+
is said to be a
class-K function if,
(a) o (0) = 0.
(b) o is strictly increasing.
It is said to belong to class K
if a= and
o(r) as r. +

r 0
o
K
K

Definition [Ref.1] [Class-KL Function]
A continuous function |: [0,a)[0,) R
+
is
said to belong to class-KL if, for each fixed t,
the mapping |(r,t) belongs to class K with
respect to r and, for each fixed r, the mapping
|(r,t) is decreasing with respect to t and |(r,t)
0 as t .
+

Definition [Locally Positive Definite Function (LPDF)]
A continuous function V: R
+
D R
+
is said to be a Locally Positive
Definite Function (LPDF) if there exists a class K function o such that
(a) V (t, x) ( ) x o > for all t > 0 and for all
x s r, for some r > 0.
(b) V (t, 0) = 0 +

r 0
|
t
|
|
2
t
1
t
1
r
( )
1
, r t |
( )
,0 r |
( )
2
, r t |
( )
1
, r t |
0
( )
x o
x
( )
, V t x
( )
1
, V t x
Definition [Positive Definite Function (PDF)]
A continuous function V: R
+
R
n
R
+
is said to
be a Positive Definite Function (PDF) if there
exists a class K
function o such that

(a) V (t, x) ( ) x o > for all t > 0 and for all
xeR
n
.
(b) V (t, 0) = 0 +

Definition [Decrescent Function]
A continuous V: R
+
D R
+
is said to be locally decrescent if there exists
a class K function o such that
V (t, x) ( ) x o s for all t > 0 and for all x s r, for some r > 0.
It is decrescent if o is a class K
function and the above inequality is valid

for all x in R
n
. +

0
( )
x o
x
( )
, V t x
( )
1
, V t x
Examples
1) V (x
1
, x
2
) =
2 2
1 2
x x + is a PDF and decrescent.
2) V (t, x
1
, x
2
) = (t + 1) (
2 2
1 2
x x + ) is a PDF but not decrescent.
3) V (t, x
1
, x
2
) =
2 2
1 2
( )
t
e x x
+ is not a PDF.
4) V (x
1
, x
2
) =
2 2
1 2
sin ( ) x x + is a LPDF and decrescent (but not PDF).
Lemma 4.2
Let
1 2
& o o be class K functions on
| |
0, a ,
3 4
& o o be class K
functions,
and | be a class KL function. Denote the inverse of
1
i i
by o o
. Then
(a)
1
1
o
is defined on
1
[0, ( )] a o and belongs to class K
(b)
1
3
o
is defined and belongs to class K

(c)
1 2
o o belongs to class K where
( ) ( ) ( )
1 2 1 2
x x o o o o =
(d)
3 4
K o o

e
(e)
( )
( )
1 2
( , ) ( ), r s r s o o | o = belongs to class KL
Lemma 4.3
Let ( ) V x be a continous positive definition function defined on a domain
n
D R c that contains origin. Let
r
B D c for same 0 r > . Then, there
exists class K functions
1 2
& o o , defined on [0, ] r , such that
( ) ( )
1 2
( ) x V x x o o s s
.
r
x B e If ,
n
D R = then
1 2
& o o will be defined in [0, ] and the above
inequality holds for all
n
x R e . Furthermore, if ( ) V x is radially unbounded,
then,
1 2
& o o can be chosen to belong to class K
.
0
( )
1
x o
x
( )
V x
( )
2
x o
3.1 Continuous Time Systems

Definition [Derivative of V(t, x) relative to a state trajectory]
Consider the system (L.1). The Derivative of V(t, x) relative to (L.1) is given by

( , ) ( , )
( , ) ( , ) ,
V t x V t x
V t x f t x
t x
c c
= +
c c
`
(L.20)
where
1 2
( , ) ( , ) ( , ) ( , )
, , ,
n
V t x V t x V t x V t x
x x x x
(
c c c c
=
(
c c c c

Theorem L.6 [Ref2]
The equilibrium point 0 of (L.1) is locally stable in the sense of Lyapunov if
there exists a LPDF V(t,x) such that
( , ) 0 V t x s
`

for all t > t
o
and all x such that x r < for some r > 0. A

0 ( , ) ( , )
o o o
V V t x V t x V s s =
`

As V is LPDF, K o - e s.t.
( )
( , )
o
x V t x V o s s .
1
( )
o
x V o
s
As V is continuous w.r.t. x & ( ,0) 0 V t = , we have, 0, c > ( , ) 0,
o
t o c - > s.t.
{ : },
o
x B x x
o
o e = s
( , ) ( )
o o o
V V t x o c = s ,
1
( ( )) x o o c c
s =
Thus, 0 is stable in the sense of Lyapunov.

0
( )
x o
x
( )
0
, V t x
c
( )
0
, t o c
0max
V
Theorem L.7 [Ref2]
The equilibrium point 0 of (L.1) is locally uniformly stable in the sense of
Lyapunov if there exists a locally decrescent LPDF V(t,x) such that
( , ) 0 V t x s
`

for all t >0 and all x such that x r < for some r > 0. A

As V is decresent,
2
K o - e s.t.
( )
2
( , ) V t x x o s . Then, pick up ( ) o c
s.t.
2
( ) ( ) o o o c s . Then,
o
x B
o
e
2 2
( , ) ( ) ( ) ( )
o o o o
V V t x x o o o o c = s s s
( )
( , ) ( )
o
x V t x V o o c s s s
, x t c s
0
( )
1
x o
x
( )
0
, V t x
c
( )
o c
0max
V
( )
2
x o
Theorem L.8 [Ref2]
The equilibrium point 0 of (L.1) is locally uniformly asymptotically stable if
there exists a locally decrescent LPDF V(t,x) such that ( , ) V t x
`
is a LPDF. A

Theorem L.9 [Ref2]
The equilibrium point 0 of (L.1) is globally uniformly asymptotically stable if
there exists a decrescent PDF V(t,x) such that ( , ) V t x
`
is a PDF. A

Theorem L.9b [Theorem 4.10 of Ref1] - for Exponential Stability
If there is a decrescent PDF
( )
, :[0, ) V t x D R such that
1 2
( , )
a a
k x V t x k x s s
3
( , ) or , 0,
a V V
f t x k x V V t x D
t x

c c
+ s s > e
c c
`

for some positive constants
1 2 3
, , and k k k , then, the equilibrium point 0 of
(L.1) is exponentially stable. If the assumptions hold globally, then, the
equilibrium point 0 is globally exponentially sable. A
Proof: Using comparison lemma to show that
0
( )
0 0
( , ( )) ( , ( ))
t t
V t x t V t x t e

s
In adaptive control problems, it is often the case that ( , ) V t x
`
is only negative
semi-definite, i.e., ( , ) 0 V t x s
`
. If the system is autonomous, then, LaSalles
Invariance Principle Theorem L.2 may be applied to obtain asymptotic tracking.
For non-autonomous system, LaSalles Invariance Theorem L.2 cannot be
applied. Instead, the following Barbalats lemma should be used.

Lemma L.1 [Ref1] Barbalats Lemma
Let |(t) be a uniformly continuous real function of t defined for t>0.
Suppose that
0
lim ( )
t
t
d | t t
}
exists and is finite. Then,
( ) 0 as t t | A
Using Barbalats lemma, the following Lyapunov-like lemma can be obtained.

Lemma L.2 [Ref2] Lyapunov-Like Lemma
If a scalar function V(t,x) satisfies the following conditions
- V(t,x) is lower bounded
- ( , ) V t x
`
is negative semi-definite.
- ( , ) V t x
`
is uniformly continuous in time (A sufficient condition is that
( , ) V t x
``
is bounded)
then, ( , ) 0 as V t x t
`
A
Proof of Lemma L.1:
Prove by contradiction, i.e., let us assume that ( ) 0 t |
/
as t first and
try to prove that this will lead to a contradiction.
As ( ) t | is assumed to be uniformly continuous, then, 0, ( ) 0
| | |
c o c > - >
s.t.
1 2
, , t t as long as
1 2
t t
|
o s , we have
1 2
( ) ( ) t t
|
| | c <
By assumption, if ( ) 0 t |
/
, then, 0
t
c - > , s.t.
3
0, ( ) , T t T T > - > s.t.
3
( )
t
t | c > .So let us choose
1
.
2
t |
c c = Then, for this
|
c , ( ) 0
t |
o c - > s.t.
3 3
[ , ) t t t
|
o e + or
3
t t
|
o s , we have,
3
1
( ) ( )
2
t
t t
|
| | c c < = . Thus
( ) ( )
3 3
3
1 1
( ) ( ) ( ) ( ) &
2 2
sign ( ) sign ( )
t t t
t t t t
t t
| | | | c c c
| |
> > =
=

3 3 3
3 3 3
1 1
( ) ( ) ( )
2 2
t t t
t t t
t t t
d d d
| | |
o o o
|
| t t | t t c t c o c
+ + +
= > =
} } }

which contradicts the assumption that ( )
o
t
d | t t
}
exists, as if lim ( )
o
t
t t
d | t t
}

exists, then, 0 c > , 0, s.t. , 0
t
T t T
c c
o - > > > , ( )
t
t
t
d
o
| t t c
+
<
}
. This
shows that our assumption that ( ) 0 t | is not true ( ) 0as t t | .

Proof of Lemma L.2:
0
0 ( ) (0) decreasesas
( ) is lower bounded
t
V V t V V d t
V t
t

s = +
)
}
` `

( ) converges as or lim exists.
t
o t
V t t V dt
}
`

Since V
`
is uniformly continuous, by Barbalats lemma, 0 V
`
as t .

Lemma [Ref1] Characterizing Lyapunov Stability via Class K Functions

The equilibrium point 0 is

- uniformly stable iff there exists a class Kfunction o and a positive
constant c, independent of
0
t , such that
( )
0 0 0
( ) ( ) , 0, ( ) x t x t t t x t c o s > > <

- uniformly asymptotically stable iff there exist a class KLfunction | and
a positive constant c, independent of
0
t , such that
( )
0 0 0 0
( ) ( ) , , 0, ( ) x t x t t t t t x t c | s > > <

- globally uniformly asymptotically stable iff the above inequality is
satisfied for all initial state
0
( ) x t . A

3.2 Discrete Time Systems
Definition [Change of V (k, x) relative to a state trajectory]
Consider the system (L.2). The change of AV(k+1,x) relative to (L.2) is given
by
( 1, ) ( 1, ( 1)) ( , ( )) V k x V k x k V k x k A + = + + (L.21)
( 1, ( , ( ))) ( , ( )). V k f k x k V k x k = +

Theorem L.10
The equilibrium point 0 of (L.2) is locally stable in the sense of Lyapunov if
there exists a LPDF V(k,x) such that
( 1, ) 0 V k x A + s
for all k > k
o
and all x such that x r < for some r > 0. A

Theorem L.11
The equilibrium point 0 of (L.2) is globally uniformly asymptotically if there
exists a decrescent PDF V(k,x) such that ( 1, ) V k x A + is negative definite.

3.3 Linear Time-Varying Systems
The solution of the linear time-varying system described by
( ) x A t x =
`
(L.22)
is given by
0 0
( ) ( , ) ( ) x t t t x t = u
where
0
( , ) t t u is the state transition matrix.

The equilibrium point 0 of (L.22) is (globally) uniformly asymptotically stable
if and only if the state transition matrix satisfies the inequality

0
( )
0 0
( , ) , 0
t t
t t ke t t

u s > > (L.23)
for some positive constant k and .

Theorem L.12 shows that, for linear systems, uniform asymptotic stability of the
origin is equivalent to exponential stability. Note that, for linear time-varying
system, in general, uniform asymptotic stability cannot be characterized by the
location of the eigenvalues of the matrix A
Example:
Stability of Linear-Time-Varying System cannot be Judged Based on
Eigenvalues of ( ) A t

Consider the LTV system given by

2
2
1 1.5cos 1 1.5sin cos
( ) , ( )
1 1.5sin cos 1 1.5sin
t t t
x A t x A t
t t t
(
+
= =
(
+

`

Verify by yourself that

( )
( ) 0.25 0.25 7, A t i t =

0.5
0.5
cos sin
( ,0) , . ., ( ,0) ( ) ( ,0)
sin cos
t t
t t
e t e t d
t i e t A t t
dt
e t e t
(
| |
u = u = u
( |
\ .

Thus, all eigenvalues of ( ) A t are in LHP and, furthermore, are independent
of t. Yet the system is unstable since some solutions become unbounded.

Suppose that the equilibrium point 0 of (L.22) is uniformly asymptotically
stable, and A(t) is continuous and bounded. Let Q(t) be a continuous,
bounded, symmetric positive definite matrix. Then, there is a continuously
differentiable, bounded, symmetric positive definite matrix P(t) such that
( ) ( ) ( ) ( ) ( ) ( )
T
P t P t A t A t P t Q t = + +
`
(L.24)
Hence, V(t,x)=x
T
P(t)x is a Lyapunov function for the system that satisfies
the conditions of Theorem L.9 . A

Proof:
Let
( ) ( , ) ( ) ( , )
T
t
P t t Q t d t t t t
= u u
}
(L.25)
It can be verified that P(t) given above satisfies (L.24). Details are omitted.

3.4 Linearization (Lyapunovs Indirect Method)

Consider the non-autonomous system (L.1) with the origin as an equilibrium
point. Suppose that f: R
+
DR
n
is continuously differentiable, and the
Jacobian matrix [cf/cx] is bounded and Lipschitz on D, uniformly in t. Let
0
( , )
( )
x
f t x
A t
x
=
c
=
c
(L.26)
Then, the origin is an exponentially stable equilibrium point for the nonlinear
system (L.1) if and only if it is an exponentially stable equilibrium point for
the linearized linear system (L.22). A

Q: Is there a function that satisfies conditions of Lyapunov Theorems?

A: Yes as seen from the following converse Lyapunov Theorems, which
explicitly construct the needed Lyapunov functions. However, the
construction assumes exact knowledge of the solutions, which is not
helpful in practical search for an Lyapunov function as we do not know
the solution in reality. However, knowing existence of such functions
may be very useful in the analysis of nonlinear systems.

4. Converse Lyapunov Theorems

Theorem L.15 [Theorem 4.14 of Ref1] - Converse for Exponential Stability
Let 0 x = be an equilibrium point for the nonlinear system ( , ) x f t x =
`
, where
:[0, )
n
f D R is continuously differentiable on
{ }
:
n
D x R x r = e < and
the J acobian matrix
f
x
c
c
is bounded on D, uniformly in t . Let K, and
o
r be
positive constants with
o
r
r
K
< . Let
{ }
:
o
n
o r o
D B x R x r = = e < . Assume that
the trajectories of the system satisfy
( )
( ) , ( )
o
t t
o o o
x K x t e x t D

s e
Then, there is a function :[0, )
o
V D R that satisfies the inequalities:
2 2
1 2
( , ) c x V t x c x s s
2
3
( , )
V V
f t x c x
t x
c c
+ s
c c

4
V
c x
x
c
s
c

for some positive constants
1 2 3 4
, , and . c c c c Moreover, if r = , and the origin
is globally exponentially stable, then ( , ) V t x is defined and satisfies the
aforementioned inequalities on
n
R . Furthermore, if the system is
autonomous, V can be chosen independent of t .

Proof: Show that
( , ) ( ; , ) ( ; , )
t
T
t
V t x t x t x d
o
| t | t t
+
=
}

satisfies all the requirements, where ( ; , ) t x | t is the solution of the system
that starts at ( , ) t x , and o is a positive constant that is large enough.

For LTI:

( )
( ; , )
A t
x Ax t x e x
t
| t

= =
`

Then,
( ) ( ) ( ) ( )
( )
( , )
T T
t t
T A t A t T A t A t
t t
P t
V t x x e e x d x e e d x
o o
t t t t
t t
+ +

(
= =
(

} }

( ) as
T
A s As
o
P t e e ds P
o
o
=
}
.
which is the solution of Lyapunov equation for Q I = .

Remark:
One application of the above converse Theorem is to show that the
linearization Theorem L.14 also has a only if part, which is stated by
Theorem 4.15 of textbook.
- Converse Theorem for Uniformly Asymptotic Stability
Let 0 be an equilibrium point for the nonlinear system
( , ) x f t x =
`

where :[0, )
n
f D R is continuously differentiable on
r
D B = , and the
J acobian matrix
f
x
c
c
is bounded on D, uniformly in t . Let | be a class KL
function and
o
r be a positive constant that ( , 0)
o
r r | < . Assume that the
trajectory of the system satisfies
( )
( ) ( ) , , ( )
o
o o o o r
x t x t t t x t D B | s e =
(an alternative way to state the uniformly asymptotic stability). Then, there is
a continuously differentiable function :[0, )
o
V D R such that
( ) ( )
1 2
( , ) x V t x x o o s s
( )
3
( , )
V V
f t x x
t x
o
c c
+ s
c c

( )
4
V
x
x
o
c
s
c

where
1 2 3 4
, , , and o o o o are class K functions. If the system is autonomous,
V can be chosen independent of t .

Proof:
( )
( , ) ( ; , )
t
V t x G t x d | t t
=
}

where G is a class K function selected by using Masseras Lemma
(Lemma C.1).
5. Boundedness and Ultimate Boundedness

Theorem L.17 [Theorem 4.18 of Ref1]:
Let
n
D R c be a domain that contains the origin and :[0, ) V D R be a
continuously differentiable function such that
( ) ( )
1 2
( , ) x V t x x o o s s
3
( , ) ( ), 0, 0, and
V V
f t x W x x t x D
t x

c c
+ s > > > e
c c

where
1 2
and o o are class K functions and
3
( ) W x is a continuous positive
definite function. Take 0 r > such that
r
B D c and assume that
( )
1
2 1
( ) r o o
<
Then, there exists a class KL function | and for every initial state ( )
o
x t
satisfying
( )
1
2 1
( ) ( )
o
x t r o o
s , there is a
( )
( ), 0
o
T x t > ) such that the solution
satisfies
( )
( ) ( ) , ,
o o o o
x t x t t t t t t T | s < < +
( )
1
1 2
( ) ( ) ,
o
x t t t T o o
s > +

Intuitive Explanation of Theorem L.17 for Autonomous Systems

{ : ( ) } x V x
c
c O = s
{ : ( ) }
c
x V x c O = s
max ( ) min ( )
x x
V x V x
c

c
= =
= =
( ) ( )
1 2
( ) x V x x o o s s

If the assumptions of Theorem L.17 are true:
3
( ) ( ),
V
V f x W x x
x

c
= s >
c
`

which is also true for any and
c
x
c
eO O ( ) ( ) ( )
o o
V k V t V t k t t s s
`
.
Then the solution will eventually enter
c
O after
( )
o
V t
T
k
c
= , which also
indicates that after
o
x t t T
c
s > .
( )
( )
( )
2 2
1 2
1
1 2
1 1
max ( ) max ( )
( ) ( ) or
( )
min ( ) min ( )
x x
x x
V x x
V x x
c c

c
c
c

c o o
o o
o o
c o o
= =
= =
= s =
s
s
= > =

D
( ) V x c =
B
c
B

6. Input-to-State Stability

6.1 Linear Time-Invariant Systems:
( )
( ) ( )
t
At A t
o
o
x Ax Bu x t e x e Bu d
t
t t
= + = +
}
`

For stable system,
Re[ ( )] 0
At t
i
A e ke

< s for some & 0 k >
where
| | { }
0 min Re ( )
i
i
A < s . Then,
( )
0
( )
0
( )
0
( ) ( )
( )

[1 ]
t
At A t
o
t
At A t
o
t
t t
o
t t
o
x t e x e Bu d
e x e B u d
ke x ke d B u
k
ke x e B u
t
t
t

t t
t t
t
s +
s +
s +
= +
}
}
}

which means that bounded inputs will lead to bounded states.

6.2 Nonlinear Systems
Definition: [Input-to-State Stable (ISS)]
System ( , , ) x f t x u =
`
is said to be input-to-state stable if there exists a
class KL function | and a class K function such that for any initial
state ( )
o
x t and any bounded input ( ) u t , the solution exists for all
o
t t >
and satisfies
( )
( ) ( ) , sup ( )
o
o o
t t
x t x t t t u
t
| t
s s
| |
s +
|
\ .

Theorem L.18 [Theorem 4.19 of Ref1]:
Let :[0, )
n
V R R be a continuously differentiable function such that
( , , ) [0, ) :
n m
t x u R R e
( ) ( )
1 2
( , ) x V t x x o o s s
( )
3
( , , ) ( ), 0
V V
f t x u W x x u
t x

c c
+ s > >
c c

where
1 2
, o o are class K
functions, is a class K function, and

3
( ) W x
is a continuous positive definite function on
n
R . Then, the system
( , , ) x f t x u =
`
is ISS with
1
1 2
o o
= .

For any bounded input, i.e., ( ) u t M s , for some M and t . Consider:
( ),
( )
0
u t
v
t
t t
t
t
s
=

>

Then, ( ; , , )
o o
x t x v t is the same as ( ; , , )
o o
x t x u t . Let
{ }
sup ( )
o
t t
v
t
t
s s
=
From Theorem L.17, the above theorem is true. #

Note: [Lemma 4.6 of Ref1]
An immediate consequence of the converse Lyapunov theorem and the
above theorem is that, if the unforced system, i.e., ( , ,0) x f t x =
`
, has a
globally exponentially stable equilibrium point at the origin and ( , , ) f t x u is
continuously differentiable and globally Lipschitz in ( , ) x u , uniformly in t ,
then, the system ( , , ) x f t x u =
`
is input-to-state stable.

References:
[Ref1]: Khalil, H. K. (2002), Nonlinear Systems, Third edition, Prentice-Hall.
[Ref2]: Slotine, J .J .E. and Li, Weiping (1991), Applied Nonlinear Control,
Prentice-Hall.

ME677c4 LyapunovTheorems

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

ME677c4 LyapunovTheorems

Hochgeladen von

Copyright:

Verfügbare Formate

Lyapunov Stability Lecture Notes by B.

, a contradiction with the

< . Then, 0 K > and is n.d. V

Lyapunov Stability Lecture Notes by B. Yao

function o such that

function and the above inequality is valid

is defined and belongs to class K

Lyapunov Stability Lecture Notes by B. Yao

functions, is a class K function, and

Das könnte Ihnen auch gefallen