LaSalle in Variance Principle
LaSalle in Variance Principle
LaSalle in Variance Principle
Pantelis Sopasakis
Problem Formulation: The following state space model (nonlinear, time invariant) describes the
dynamics of a pendulum with linear friction term:
(
=
(
1 2
2
2
1
sin x kx
x
x
x
dt
d
(1)
Defined on the domain | | x D 9 = t t, . The total energy of the system is chosen as a candidate
Lyapunov function for the system:
( ) 0 1 cos
2
1
1
2
2
+ = x x x V (2)
Then ( ) 0
2
2
kx x V = and in fact it is ( ) 0 = x V
if and only if 0
2
= x . Although it is possible to
calculate analytically the solution of (1) and corroborate that the origin is an asymptotically stable
point, Lyapunovs theorem fails to prove asymptotic stability (by means of a completely natural
Lyapunov function). In what follows we state and prove a complementary result to Lyapunovs
stability theorem, known as La Salles (invariance) Principle. This is used in a combination with
standard Lyapunov-like theorems to derive asymptotic stability properties of nonlinear systems when
the Lie derivative candidate Lyapunov function (along the trajectory of the system) is merely negative
semidefinite (and not negative definite). Other interesting results accrue from the introduction of a new
notion: the -limit set of a system.
Definition 1. Let ( ) t x x = be the solution of the dynamical system ( ) x f x = . A point
n
p 9 e will
be called a positive limit of the trajectory ( ) t x if there is a sequence ( )
n n
t , such that =
n n
t lim and
( ) =
n n
t x lim . The set of positive limits of ( ) ( )
t
t x , is called positive limit set (also known as the -
limit set) and henceforth will be denoted as
+
L .
Definition 2 (Positive invariant set). A set is called a positive invariant set of the system ( ) x f x =
if:
( ) ( ) 0 , 0 > e e t M t x M x (3)
Definition 3 (Negative invariant set). A set is called a negative invariant set of the system
( ) x f x = if:
( ) ( ) 0 , 0 s e e t M t x M x (4)
Definition 4 (Functional invariant set). A set M is called functional invariant if it is both positive and
negative invariant.
Definition 5 (Asymptotic Stability with respect to a set). A trajectory ( ) ( )
t
t x is said to approach the set
asymptotically as t , if for every 0 > c , there is a 0 > T such that:
( ) ( ) T t M t x d > < , , c (5)
Where d is a metric.
The following lemma is necessary in order to prove La Salles invariance principle.
Assumptions : Hereinafter :
n
f D9 is always assumed to be a Lipschitz-continuous function
over the domain
n
D_9 which is compact and contains the origin. The function | ) : 0, V D is
always assumed to be continuously differentiable.
Lemma 1 (Topological Description of Positive Limit Sets). Let ( ) t x x = be a solution of the
dynamical system ( ) x f x = . Assume that this solution is bounded over D for all 0 > t . Then the
positive limit set L
+
is:
1. Not empty
2. Norm compact
3. Positive Invariant
4. Is approached asymptotically be the trajectory ( ) ( )
t
t x (with respect to the standard metric).
Proof
1. We show that L
+
is not empty: Since ( ) ( )
t
t x is a bounded trajectory, according to the
Bolzano-Weierstrass theorem there is a sequence of times ( )
n n
t such that
( ) ( ) ( ) ( )
t n n
t x x c t and let ( ) ( )
k
n
k
x t be a convergent sub-sequence. Then L
+
is not empty
since it contains the limit of this convergent subsequence.
2. We show that the set L
+
is norm-compact. We need to show that L
+
is norm-closed and norm-
bounded. This calls for the following two steps:
i. We show that it is bounded: We have already proved that L
+
is not empty. Then for
every
+
eL y there is a sequence ( )
n n
t satisfying =
n n
t lim such that
( ) y t x
n n
= lim (As it follows from the definition of positive limit sets). But the
sequence ( ) ( ) ( ) ( )
t n n
t x t x c is bounded, hence y are bounded and L
+
is a bounded set.
ii. We now prove that it is closed: It suffices to show that every convergent sequence with
elements from L
+
converges to a limit inside L
+
. For each
` +
eL y we can find a
sequence ( )
+
c L y
i i
with elements in L
+
such that y y
i
as i . For each i there
is a sequence ( )
j
j i
t
,
satisfying
j i
t
,
as j , such that ( )
i j i
y t x
,
as j .
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
1 , 1 , 3 1 , 2 1 , 1
2 , 2 , 3 2 , 2 2 , 1
3 , 3 , 3 3 , 2 3 , 1
, , 3 , 2 , 1
3 2 1
k
k
k
n k n n n
k
t x t x t x t x
t x t x t x t x
t x t x t x t x
t x t x t x t x
y y y y y
| | | |
We construct a sequence ( )
i i
t as follows:
Pick
2 , 1 2
t > t such that ( )
2
1
2 2
< y x t
Pick
3 , 1 3
t > t such that ( )
3
1
3 3
< y x t
What is the same, pick a
j j
t
, 1
> t such that ( )
j
y x
j j
1
< t
Obviously
i
t as i and it also holds that ( ) ,... 2 ,
1
= < i
i
y x
i i
t .
Given 0 > c , there are integer numbers and such that ( ) N i y x
i i
> < ,
2
c
t and
M i y y
i
> < ,
2
c
. Then according to the triangular inequality, one has that:
( ) { } N M i y x , max , = > < c t
i
(6).
Hence, ( ) y x
i
t as i . Therefore the set L
+
is closed.
From (i) and (ii) it is immediate that L
+
is norm-compact.
3. We prove that L
+
is positive invariant: Let
` +
eL y and ( ) y t; | be the solution of the
system ( ) x f x = satisfying the initial condition ( ) y x = 0 . There is a sequence ( )
i i
t with
i
t as i such that ( ) y t x
i
as i . We then have that ( ) ( )
o i i
x t t x ; | = .
Due to the uniqueness of the solution (Picard theorem) we have that:
( ) ( ) ( )
i o i
t x t x t t ; ; | | = + (7)
Then for some i:
0 > +
i
t t (8)
Due to the continuity property of with respect to t:
( )
( )
( ) ( ) ( ) y t t x t x t t
i i o i i
; ; lim ; lim
7
| | | = = + (9)
Thus:
( )
+ +
e > e L y t L y t , 0 , , | (10)
4. We show that ( )
+
L t x as t : Assume that this is not the case. Then there is
an 0 > c and a sequence of times ( )
n n
t with
n
t as n , such that:
( ) ( ) c >
+
L t x d
n
, (11)
Where d is the standard metric. According to the Bolzano Weierstrass theorem and because
the sequence ( ) ( )
n n
t x is bounded, it has a convergent subsequence. Let ( ) ( )
k
i
k
t x be a
convergent subsequence with ( )
*
x t x
k
i
as k . Since L
+
is closed, this limit is inside
L
+
thus ( ) 0 ,
*
> >
+
c L x d . Therefore
+
eL x
*
which is a contradiction. Therefore L
+
is
positive invariant.
This completes the proof.
Having proved above lemma we are now ready to state and prove La Salles invariance principle.
Theorem (La Salles Invariance Principle): Let D c O be a compact and positive invariant set with
respect to the system ( ) x f x = . Let 9 D V : be continuously differentiable with ( ) 0 s x V
over
. Let be the set of states for which ( ) 0 = x V
. Therefore:
O c c c
+
E M L (14)
This completes the proof.
We now state a very important implication which accrues directly from La Salles invariance principle
and complements Lyapunovs stability theory:
Corollary 1. Let ( ) x f x = have the origin as equilibrium point and 9 D V : be a continuously
differentiable function over the domain
n
D 9 c which contains the origin such that 0
V over D.
Let ( ) { } 0 | = e = x V D x E
. If no solution of the system can remain inside E except for ( ) 0 t x ,
then the origin is an asymptotically stable equilibrium point.
In case { } 0 = E this corollary boils down to the stability theorem of Lyapunov. If for the
system ( ) x f x = for which the origin is an equilibrium point, the following implication holds:
( ) ( ) 0 0 = = t x x V
(15)
Then, the origin is an asymptotically stable equilibrium point.
To this end we can go back to the problem we stated in the introduction. For the dynamical system
described by (1), equation (2) gives a Lyapunov function for it. We have that
( )
| |
0 0 sin 0 0 0 0
1
,
1 2 2
0
2
2
1
= = = = = =
e
>
x x x x kx x V
x
k
t t
(16)
Then, according to the last corollary, the origin is an asymptotically stable point for the system (1).
System (1) belongs to a more general class of systems of the form:
( ) ( )
(
=
(
2 2 1 1
2
2
1
x h x h
x
x
x
(17)
with | | 9 = e a a D x , where h
1
and h
2
are locally Lipschitz functions with the
properties ( ) 0 0 =
i
h and ( ) 0 > y yh
i
for every ( ) a a y , e . The origin is an asymptotically stable
equilibrium point for systems of this class. The corresponding Lyapunov function we use to prove this
is:
( ) ( )
}
+ =
1
0
2
2 1
2
1
x
x dy y h x V (18)
References
1. H.K. Khalil, Nonlinear Systems, Prentice Hall editions (3rd edition), New Jersey, 2002, ISBN:
0-13-067389-7.
2. P. Sopasakis, The Realm of Linear State Space Control Systems, Oct. 2011, Available online
at http://users.ntua.gr/chvng/en (Access date: 22 Oct. 2011)