LECTURE 6
Stability
This lecture includes:
- Internal Stability
- BIBO Stability
- Lyapunov Stability
1
Stability Analysis for Linear Dynamic Systems
x0
u Dynamic System y
(1) Internal Stability (I-S)
Assume : External Input u = 0 (or known fixed)
Equilibrium xe
x f (x) xe f (xe ) 0
at xe
Behavior of x(t) near xe is of concern.
“Stable” if x(t) nhd(xe), t t0
(2) BIBO Stability
If any BI guarantees BO, the I/O Description is called BIBO-S.
2
Internal Stability
For LTI systems,
(1) If the transient remains bounded in the nhd of “O” (or dies out), the system is
stable.
(2) Poles LHP
x0
x
u y
For LTV systems,
a. “Internal Stability”
x(t ) A(t ) x(t ), x(t0 ) x0
Boundedness w.r.t. x0 & t0 .
How do you know the stability ?
b. Lyapunov Theory
x (t ) f ( x (t ), t )
x A (t ) x (t ) B (t )u (t )
3
Given x(t ) A(t ) x(t ) : Unforced System
The response due to x(t0 ) x0
x (t ) x ( t ; t 0 , x 0 ) (t , t 0 ) x 0
Definition 1
State xe is an equilibrium state at t0 iff
x(t ) x(t; t0 , xe ) t t0
xe
Definition 2
An equilibrium state xe is stable in the sense of Lyapunov
at t0 iff for every 0 (t0 , ) 0
( depending on t0 and 0 ) s.t. if x0 xe ,
( x(t; t 0 , x0 ) xe ) , t t0
4
Definition 3
xe is uniformly stable i.s.L. iff for 0, ( ) 0 s.t.
x0 xe x(t ; t0 , x0 ) xe , t t0
Remarks
(1) 0 is always an equilibrium for x A(t ) x
(2) Let xe 0.
Then, for any 0, ( ) s.t. x0 x(t , t0 , x0 ) t
if and only if r 0 s.t.
x(t , t0 , x0 ) r x0
Hence, for any , if we choose / r , then x0 implies that
x(t ) , t t0 .
5
Theorem
Every equilibrium state of x A(t ) x is stable in the sense of Lyapunov at t0
iff there exists some constant r which depends on t0 such that
(t , t0 ) r , t t0 .
If r is independent of t0 , it is uniformly stable i.s.L.
Pf)
(Sufficiency)( )
Let xe be an equilibrium state of x A(t ) x; that is
xe (t , t0 ) xe , t t0 .
Then we have
x(t ) xe (t , t0 ) x0 xe (t , t0 )( x0 xe ), t t0 .
x(t ) xe (t , t0 ) x0 xe r x0 xe t t0 .
Hence, for any , if we choose / r , then x0 xe implies that
x(t ) xe , t t0 .
(Necessity)( ) We can prove this by contradiction. (Details are omitted.)
6
Theorem
The zero state of x A(t ) x is asymptotically stable at t0
iff (t , t0 ) r (t0 ) & (t , t0 ) 0 as t .
The zero state is asymptotically stable over [0, )
iff there exist positive numbers k1 and k2 such that for any t0 0
(t , t0 ) k1e k2 (t t0 ) , t t0
Pf ) Omitted.
Note)
If x A(t ) x is asymptotically stable, the zero state is the only equilibrium state.
( If there were another equilibrium state different from the zero state, then by
choosing that equilibrium state as an initial state, the response would not
approach the zero state.)
7
A. Stable i.s.L.
x2
x1
t0
t
x
r x0
x0
t0 t
8
Lyapunov Stability Criteria
Lyapunov Methods for Stability
- 1st Method, 2nd Method
- Lyapunov divides nonlinear systems into 2 groups
1st Class : Systems whose solutions are available.
The solution is examined and stability or
instability is noted.
2nd Class : Systems about which stability can be
determined directly by V-function*
( a Lyapunov function ),
9
- “The total energy of an unforced, dissipative mechanical
systems decreases as the state of the system evolves
in time”
Generalized concept V function
x V
x x
x
10
(1) Consider
x(t ) A(t ) x(t ), x(t0 ) x0
Under what conditions do the solutions x(t) behave such
that x(t ) 2 0 (monotonically decreases) as t ?
(observation)
2
Let x(t ) xT (t ) x(t ) .
d
2 T
Then x(t ) x (t ) x(t ) xT (t ) x(t )
dt
xT (t ) AT (t ) A(t ) x(t )
If AT (t ) A(t ) 0 at each t , d x(t ) 2 0 and
dt
2
hence x(t ) decreases for t t0
(not increasing)
11
x(t ) A(t ) x(t ), x(t0 ) x0 (1)
Theorem LS-II ( Uniform Exponential Stability )
Eq. (1) is uniformly exponentially stable
if an n n Q(t) such that
I Q(t ) I (†)
T
A (t )Q(t ) Q(t ) A(t ) Q(t ) I (††)
where , , are some finite positive numbers.
Brief Pf )
Let V (t ) xT (t )Q(t ) x(t ), where Q(t ) 0
then V (t ) 0 and
V (t ) xT (t ) AT (t )Q(t ) Q(t ) A(t ) Q(t ) x(t ).
So if A (t )Q(t ) Q(t ) A(t ) Q(t ) 0, then V (t ) 0.
T
It means that the system is asymptotically stable.
12
( uniformly exponentially stable.)
Lyapunov Theory for Constant System
Q is Hermitian iff Q = Q* (complex conjugate transpose).
(1) Let Q be Hermitian. Then
2 2
min x x*Qx max x
(2) Hermitian Q is called positive (semi) definite iff
x*Qx 0 x0
( x*Qx 0 x 0)
13
(3) Hermitian Q is positive (semi) definite
1. All eigenvalues are positive
(nonnegative)
2. nonsingular N s.t.
Q = N* N.
(4) Lyapunov Theorem.
x Ax is asymptotically stable iff , for any M > 0,
the equation
A*Q QA M (+)
has a unique solution Q > 0 with Q being Hermitian.
Note : (+) is called the Lyapunov equation.
14
( Proof )
() Let M I , where 0.
Existence of Q s.t., with Q 0 ,
AT Q QA M
implies that all the hypothesis of Theorem LS-II are satisfied.
Therefore x Ax, x(0) x0 is exponentially stable.
(or asymptotically stable).
() Necessity can be proved by the following:
15
Theorem
If A has negative real part eigenvalues,
then, for each symmetric M, ! solution of
AT Q QA M
of the form
AT t
Q e 0
Me At dt (++)
Further, if M > 0, then Q > 0.
16
( Proof ) If all eigenvalues of A have negative real parts,
Q in (++) is well defined. Now,
At At
A Q QA 0 A e Me dt 0 e Me Adt
T T AT t AT t
d
0
dt
e A t Me At dt T
T
e Me
A t At
0
M
To prove uniqueness, suppose Qa is also a solution. Then
AT Qa Q Qa Q A 0
e A t AT Qa Q e At e A t Qa Q Ae At 0
T T
d AT t
e (Qa Q)e At 0
dt
T
0 e A t (Qa Q)e At 0
(Qa Q)
Qa Q
17
Finally, suppose that M > 0.
Q e Me At dt
AT t
0
Q is symmetric.
Now, for a x 0 , e At x 0 , and as
T
x Qx
T
x T e A t Me At xdt 0
0
since ( e At x ) T M ( e At x ) 0 t
For x 0, xT Qx 0 Q 0.
18
Theorem
Every equilibrium state of x Ax is stable i.s.L.
if all the eigenvalues of A have nonpositive real parts and
those with zero real parts are distinct roots of the minimal
polynomial of A.
Note)
minimal polynomial of A : the monic polynomial ( A)
of least degree s.t. ( A) 0
Theorem
The zero state is asymptotically stable iff all the
eigenvalues of A have negative real parts.
j
m
t j 1 k t
Proof. 0
e At
dt
0
k 1 j 1
Wk j
( j 1)!
e dt
t j 1
Wk j e k t dt
0 ( j 1)!
e Rei t
finite 19
BIBO - Stability
x
Input Output
System
u( t ) y (t )
Definition ( BIBO Stability )
A relaxed system (i.e., a system with zero initial
conditions) is BIBO - stable iff, for a bounded input,
the output is bounded.
20
(i.e u (t ) k1 , t (, ), k1 ,
k2 y (t ) k 2 , t (, ))
x0 0
u (0) y (0)
system
u (t ) y (t )
k1 k2
t t
21
Theorem Ⅰ
A relaxed single-variable system
t
y (t )
g ( t , ) u ( ) d
is BIBO stable iff k s.t.
t
g ( t , ) d k , t ( , ).
22
( Proof )
Sufficiency (←)
t
y (t )
g ( t , ) u ( ) d
t
g ( t , ) u ( ) d
t
k g ( t , ) d k k t .
Necessity (→)
t1
Suppose g (t1 , ) d for some t
Define u ( ) sgn g (t1 , )
Then u ( ) 1 but
y ( t1 ) ; ( Contradiction )
23
Theorem Ⅱ
A relaxed MIMO System
t
y (t )
G ( t , ) u ( ) d
is BIBO stable iff
t
g ij ( t , ) d k
for i , j , t ( , )
( Note )
G ( t , ) g i j ( t , ) i 1 , , q
j 1 , , p
Corollary to Theorem I
t
cf ) y (t ) u (t ) g ( )d
Time-invariant SISO case :
0
u (t ) g ( ) d
o
g (t) dt k
g ( ) d M
0
k M
BIBO stable. 24
Theorem III ( BIBO stability of LTI System in terms of the system poles )
A relaxed SISO system with the transfer function g ( s)
given as a proper rational function is BIBO stable if
and only if all the poles of g ( s) have negative real parts.
( Proof ) : omitted
25
Internal Stability vs BIBO Stability
Theorem IV
Consider x A(t ) x B(t )u
y C (t ) x
If B ( t ) & C ( t ) are bounded on (, ) , then the
asymptotic stability of the zero-state ( equilibrium )
implies the BIBO stability of the zero-state response.
( The converse is not true. )
( Proof )
t
t0
C ( t ) ( t , ) B ( ) d
t
t0
g (t , ) d
C ( t ) ) ( t , ) B ( ) d
k1 k 2 (t , ) d
k1 k 2 k e d
k3 (t )
26
Ex 1)
x x u
y x
Note that
1
gˆ ( s )
s 1
det(I A) ( 1)( 1)
The system is BIBO stable but the zero-state is not (internally) stable,
because a positive eigenvalue.
Definition ( Total-Stability )
A system is T-stable (totally stable ) if for any initial state and for any
bounded input, the output and the state variable are bounded.
27
Theorem V
A system is T-stable if B ( ) and C ( ) are bounded
on ( , ) and the equilibrium (O) of x A x
is asymptotically stable.
( Proof ) Exercise.
28
Time-Invariant Case
x Ax Bu
y Cx
Recall :
Transfer Function G ( s) C ( sI A ) B
The zero-state response is BIBO stable iff all the poles
have negative real parts.
Theorem VI
Every equilibrium state of x Ax is stable iff all
the eigenvalue of A have nonpositive real parts and
those with zero real parts are simple zeroes of the
minimal polynomial of A.
Note)
minimal polynomial of A : the monic polynomial ( A)
of least degree s.t. ( A) 0 29
Theorem VII
The zero state of x Ax is asymptotically stable iff
all the eigenvalues of A have negative real parts.
Theorem VIII
If an LTI dynamic equation FE is controllable and
observable, then the following are equivalent.
(1) Total stability
(2) BIBO stability
(3) Asymptotic stability
(4) All the poles have negative real parts
(5) All the eigenvalues of A have negative real parts.
30