Lecture 07 Analysis (III) --

Stability

7.1 Bounded-Input Bounded-Output (BIBO) Stability

7.2 Asymptotic Stability
7.3 Lyapunov Stability
7.4 Linear Approximation of a Nonlinear System

Modern Control Systems 1

Bounded-Input Bounded-Output (BIBO) stablility

Definition: For any constant N, M >0

Any bounded input yields bounded output, i.e.

u(t ) N y (t ) M

For linear systems: p( s )

T ( s) C ( sI A) 1 B
q( s )
BIBO Stability All the poles of the transfer function lie in the LHP.

q( s ) 0 Solve for poles of the transfer function T(s)

Characteristic Equation

Modern Control Systems 2

Asymptotic stablility
When u(t ) 0, i. e. the system x Ax
x (t ) 0 as t

For linear systems: x Ax Bu

y Cx

Asymptotically stable All the eigenvalues of the A matrix

have negative real parts
(i.e. in the LHP)

p( s ) C adj[ sI A]B
T ( s) C ( sI A) 1 B
q( s ) sI A

sI A 0 Solve for the eigenvalues for A matrix

Note: Asy. Stability is indepedent of B and C Matrix

Modern Control Systems 3
Asy. Stability from Model Decomposition
n n
Suppose that all the eigenvalues of A are distinct. A R
Let vi the eigenvector of matrix A with respect to eigenvalue i
i.e. i , satisfying Avi i vi , i 1, ,n
Coordinate Matrix T [v1, v2 , ,vn ]
T 1 AT A T 1 AT
B T 1 B
1 1 0 0 0 1
0
2 0 0 2 C CT

2

0 z T 1 ATz T 1 Bu

n 0 0 n n y CTz Du

x (t ) T (t ) v1e1t 1 (0) v2e2t 2 (0) vn ent n (0), (0) T x (0)

1

Hence, system Asy. Stable all the eigenvales of A at lie in the LHP

Modern Control Systems 4

Asymptotic Stablility versus BIBO Stability
In the absence of pole-zero cancellations, transfer function poles are
identical to the system eigenvalues. Hence BIBO stability is
equivalent to asymptotical stability.
Conclusion: If the system is both controllable and observable, then
BIBO Stability Asymptotical Stability
Methods for Testing Stability

Asymptotically stable
All the eigenvalues of A lie in the LHP
BIBO stable
Routh-Hurwitz criterion
Root locus method
Nyquist criterion
....etc. Modern Control Systems 5
Lyapunov Stablility

A state xe of an autonomous system is called an equilibrium state,

if starting at that state the system will not move
from it in the absence of the forcing input.
In other words, consider the system x f ( x (t ), u(t ))
equilibrium state xe must satisfy f ( xe ,0) 0, t t0
Example:
x2
Equilibrium point
0 1 1
x x u (t )
2 3 1 x1
Set u (t ) 0 ,
0 1 x1e x1e 0
we get 0
2 3 x2 e x2 e 0
Modern Control Systems 6
Definition: An equilibrium state xe of an autonomous system is
stable in the sense of Lyapunov if for every 0 , exist a ( ) 0
x0 xe x (t , x0 ) xe t t0
such that for

x2

x (t )
x1

xe
x0

Modern Control Systems 7

Definition: An equilibrium state xe of an autonomous system is
asymptotically stable if
(i) it is stable
(ii) there exist a e 0 such that
x0 xe e x (t ) xe 0, as t

x2

e
x1
xe
x0

x (t )
Modern Control Systems 8
Lyapunov Theorem
Consider the system x f (x ) (6.1)

Eq. State : xe 0 f (0) 0

A function V(x) is called a Lapunov fuction V(x) if
(1) V ( x ) 0, x 0
(2) V (0) 0 for x0
dV ( x ) dV ( x )
(3) f ( x) 0
dt dx
Then eq. state of the system (6.1) is stable.
Moreover, if the Lyapunov function satisfies
dV ( x) dV ( x )
0, x 0 and 0 x0
dt dt
Then eq. state of the system (6.1) is asy. stable.
Modern Control Systems 9
Explanation of the Lyapunov Stability Theorem

1. The derivative of the Lyapunov function along the trajectory is negative.

2. The Lyapunov function may be consider as an energy function of the system.

x2

V ( x (t ))

x1
0 x (t )
Modern Control Systems 10
Lyapunovs method for Linear system: x Ax where A 0

The eq. state x 0 is asymptotically stable.

For any p.d. matrix Q , there exists a p.d. solution of the
Lyapunov equation T
A P PA Q

Proof: Choose V ( x ) x T Px

V ( x ) x Px x Px
T T

x T AT Px x T PAx
x T ( AT P PA ) x AT P PA Q
x T Qx 0, for x 0
Hence, the eq. state x=0 is asy. stable by Lapunov theorem.
Modern Control Systems 11
Asymptotically stable in the large
( globally asymptotically stable)
(1) The system is asymptotically stable for all the initial states x(t0 ) .
(2) The system has only one equilibrium state.
(3) For an LTI system, asymptotically stable and globally
asymptotically stable are equivalent.
Lyapunov Theorem (Asy. Stability in the large)
If the Lyapunov function V(x) further satisfies

(i) x , V ( x)
(ii) x ,V ( x)

Then, the (asy.) stability is global.

Modern Control Systems 12

Sylvesters criterion
A symmetric n n matrix Q is p.d. if and only if all its n
leading principle minors are positive.

Definition
The i-th leading principle minor Qi i 1,2,3,, n of an n n
matrix Q is the determinant of the i i matrix extracted from
the upper left-hand corner of Q.

Example 6.1: q11 q12 q13

Q q21 q22 q23 Q1 q11
q31 q32 q33
q11 q21
Q2 Q3 Q
q21 q22
Modern Control Systems 13
Remark:
(1) Q1 , Q2 , Qn are all negative Q is n.d.
(2) All leading principle minors of Q are positive Q is n.d.

Example:
V ( x) 2 x12 4 x1 x3 3 x12 6 x2 x3 x32
2 0 4 x1
Q1 2 0
x1 x2 x3 0 3 6 x
2 Q2 6 0
0 0 1 x3 Q3 24 0
2 0 2 x1
x1 x2 x3 0 3 3 x
2 Q is not p.d.
2 3 1 x3
Modern Control Systems 14
Example: 0 1
x x
1 1
p11 p12
Let Q I , Assume P
p12 p22
Solve for AT P PA I
0 1 p11 p12 p11 p12 0 1 1 0
1 1 p 1 1 0 1
12 p 22 12p p 22
p11 p12 1 3 1
P 1 2
12p p22 2
p11 3 0 P 50 P is p.d.
System is asymptotically stable

The Lyapunov function is: V ( x) xT Px 1 (3 x12 2 x1 x2 2 x22 )

2
V ( x) ( x12 x22 )
Modern Control Systems 15
Linear approximation of a function around an operating point xe
Let f (x) be a differentiable function.
Expanding the nonlinear equation into a Taylor series about the
operation point xe , we have
df ( x ) ( x xe ) d 2 f ( x ) ( x xe ) 2
f ( x ) f ( xe )
dx x xe 1! dx 2 x xe 2!

Neglecting all the high order terms, to yield

df ( x ) ( x xe )
f ( x ) f ( xe ) f ( x e ) m ( x xe )
dx x xe 1!
f (x )
f ( x ) f ( xe ) m ( x x e )
f (x )

where
f ( xe )
df ( x )
m
dx x x
x
e
Modern Control Systems
xe x
Modern Control Systems 16
Multi-dimensional Case:

Let x be a n-dimensional vector, i.e. x R

n

f ( x1 ,, xn )
f f f
f ( x1e ,, xne ) x xe ( x1 x1e ) x x e ( x2 x2 e ) x xe ( xn xne )
x1 x2 xn
f f f f
f ( x1e ,, xne ) ( x xe ), where , ,
x x x xe x1 x x xn x x
x xe
e e

Let f be a m-dimensional vector function, f ( x) : Rn Rm

i.e.
f1 ( x1 ,, xn )
f ( x ,, x )
f ( x1 ,, xn ) 2 1 n

f (
m 1 x , , x )
n

Modern Control Systems 17

Linear approximation of a function around an operating point xe

Special Case: n=m=2

f
f ( x ) f ( xe ) ( x xe ) A( x xe )
x x xe
f1 f1

f x1 x xe
x2 x xe A
where x [ x1 , x2 ] T
and
f
x x xe f 2
2
x1 x xe
x2 x xe

Modern Control Systems 18

Linear approximation of an autonomous nonlinear systems x (t ) f ( x (t ))

Let xe be an equilibrium state, from

x f ( x (t )) A( x - xe )
f1 f1

where A
f
x1 x xe
x2 x xe
x x xe
f f 2
2
x1 x xe
x2 x xe

The linearization of x (t ) f ( x (t )) around the equilibrium state xe is

z Az where z x - xe and z x - x e x

Modern Control Systems 19

Example : Pendulum oscillator model

From Newtons Law we have

d 2
J MgLsin 0
dt 2

where J is the inertia.

Define x1 , x2
(Reproduced from [1])

x 1
x2
MgL
x 2 - J sinx1

We can show that xe 0 is an equilibrium state.

Modern Control Systems 20

Example (cont.):
Method 1: f1 ( x2 ) x2 f1 ( x2 ) - f (0) ( x2 - 0) z2

f 2 ( x1 ) sinx1
d (sinx1 )
f 2 ( x1 ) - f 2 (0) sinx1 - sin0 ( x1 - 0) z1
dx1 x 0
1

The linearization around the equilibrium state xe 0 is

z1
z2
z - MgL z
2 J
1

where z x and z x

Modern Control Systems 21

Example (cont.):
MgL
Method 2: f1 ( x ) x2 , f 2 ( x ) - sinx1
J
f1 f1

f
x1 x xe
x2 x xe
x x xe
f f 2
2
x1 x xe
x2 x xe

0 1
- MgL A
J 0

The linearization around the equilibrium state xe 0 is

z1 0 1 z z2
z Az - MgL
1
MgL
2 0 z2 - z1
J J

where z xModern z Systems

andControl x 22