Nonlinear Systems and Control Lecture # 9 Lyapunov Stability
p. 1/1
Quadratic Forms
n n
V (x) = xT P x =
i=1 j=1 2
pij xi xj ,
P = PT
2
min (P ) x
xT P x max (P ) x
P 0 (Positive semidenite) if and only if i (P ) 0 i P > 0 (Positive denite) if and only if i (P ) > 0 i V (x) is positive denite if and only if P is positive denite V (x) is positive semidenite if and only if P is positive semidenite P > 0 if and only if all the leading principal minors of P are positive
p. 2/1
Linear Systems
x = Ax V (x) = xT P x, P = PT > 0
(x) = xT P x + xT P x = xT (P A + AT P )x def xT Qx V =
If Q > 0, then A is Hurwitz Or choose Q > 0 and solve the Lyapunov equation
P A + AT P = Q
If P > 0, then A is Hurwitz Matlab: P = lyap(A , Q)
p. 3/1
Theorem A matrix A is Hurwitz if and only if for any Q = QT > 0 there is P = P T > 0 that satises the Lyapunov equation
P A + AT P = Q
Moreover, if A is Hurwitz, then P is the unique solution Idea of the proof: Sufciency follows from Lyapunovs theorem. Necessity is shown by verifying that
P =
0
exp(AT t)Q exp(At) dt
is positive denite and satises the Lyapunov equation
p. 4/1
Linearization
x = f (x) = [A + G(x)]x G(x) 0 as x 0
Suppose A is Hurwitz. Choose Q = QT > 0 and solve the Lyapunov equation P A + AT P = Q for P . Use V (x) = xT P x as a Lyapunov function candidate for x = f (x)
V (x) = xT P f (x) + f T (x)P x = xT P [A + G(x)]x + xT [AT + GT (x)]P x = xT (P A + AT P )x + 2xT P G(x)x = xT Qx + 2xT P G(x)x
p. 5/1
V (x) xT Qx + 2 P
G(x)
For any > 0, there exists r > 0 such that
G(x) < , x < r xT Qx min (Q) x
2
xT Qx min (Q) x
2
V (x) < [min (Q) 2 P ] x
, x <r
Choose
< min (Q) 2 P
V (x) = xT P x is a Lyapunov function for x = f (x)
p. 6/1
We can use V (x) = xT P x to estimate the region of attraction Suppose Take
V (x) < 0,
x =r
0< x <r
c = min xT P x = min (P )r 2 {xT P x < c} { x < r}
All trajectories starting in the set {xT P x < c} approach the origin as t tends to . Hence, the set {xT P x < c} is a subset of the region of attraction (an estimate of the region of attraction)
p. 7/1
Example
x1 = x2 x2 = f x
x=0
x1 + (x2 1)x2 1 = 0 1 1 1
A=
has eigenvalues (1 j 3)/2. Hence the origin is asymptotically stable Take Q = I, P A + A P = I P =
min (P ) = 0.691
T
1.5 0.5 0.5 1
p. 8/1
V (x) = xT P x = 1.5x2 x1 x2 + x2 1 2 V (x) = (3x1 x2 )(x2 ) + (x1 + 2x2 )[x1 + (x2 1)x2 ] 1 = (x2 + x2 ) (x3 x2 2x2 x2 ) 1 2 1 1 2 5 (x) x 2 +|x1 | |x1 x2 | |x1 2x2 | x 2 + V x 4 2 1 2 where |x1 | x , |x1 x2 | 2 x , |x1 2x2 | 5 x V (x) < 0 for 0 < x
2 2
2 def 2 < = r 5
2 Take c = min (P )r = 0.691 = 0.618 5 {V (x) < c} is an estimate of the region of attraction
p. 9/1
Example:
x = g(x) g(0) = 0; xg(x) > 0, x = 0 and x (a, a)
x
V (x) =
0
g(y) dy
x The origin is asymptotically stable
V (x) =
[g(x)] = g 2 (x) < 0, x (a, a), x = 0
If xg(x) > 0 for all x = 0, use
1 V (x) = 2 x2 + x
g(y) dy
0
p. 10/1
1 V (x) = 2 x2 +
g(y) dy
0
is positive denite for all x and radially unbounded since V (x) 1 x2 2
V (x) = xg(x) g 2 (x) < 0, x=0
The origin is globally asymptotically stable
p. 11/1
Example: Pendulum equation without friction
x1 = x2 x2 = a sin x1 V (x) = a(1 cos x1 ) + 1 x2 2 2 V (0) = 0 and V (x) is positive denite over the domain 2 < x1 < 2 V (x) = ax1 sin x1 + x2 x2 = ax2 sin x1 ax2 sin x1 = 0
The origin is stable
Since V (x) 0, the origin is not asymptotically stable
p. 12/1
Example: Pendulum equation with friction
x1 = x2 x2 = a sin x1 bx2 V (x) = a(1 cos x1 ) + 1 2 x2 2
V (x) = ax1 sin x1 + x2 x2 = bx2 2
The origin is stable
V (x) is not negative denite because V (x) = 0 for x2 = 0 irrespective of the value of x1
p. 13/1
The conditions of Lyapunovs theorem are only sufcient. Failure of a Lyapunov function candidate to satisfy the conditions for stability or asymptotic stability does not mean that the equilibrium point is not stable or asymptotically stable. It only means that such stability property cannot be established by using this Lyapunov function candidate Try
V (x) = =
1 T 2x P x 1 2 [x1
+ a(1 cos x1 ) p11 p12 p12 p22
x2 ]
x1 x2
+ a(1 cos x1 )
p11 > 0,
p11 p22 p2 > 0 12
p. 14/1
V (x) = (p11 x1 + p12 x2 + a sin x1 ) x2 + (p12 x1 + p22 x2 ) (a sin x1 bx2 ) = a(1 p22 )x2 sin x1 ap12 x1 sin x1 p22 = 1, p11 = bp12 0 < p12 < b, D = {x R2 | |x1 | < }
+ (p11 p12 b) x1 x2 + (p12 p22 b) x2 2
Take p12 = b/2
1 V (x) = 2 abx1 sin x1 1 bx2 2 2
V (x) is positive denite and V (x) is negative denite over D The origin is asymptotically stable
Read about the variable gradient method in the textbook
p. 15/1
