Nonlinear Systems and Control

Lecture # 21

L2 Gain

&

The Small-Gain theorem

p. 1/1
Theorem 5.4: Consider the linear time-invariant system

x = Ax + Bu, y = Cx + Du

where A is Hurwitz. Let G(s) = C(sI A)1 B + D .

Then, the L2 gain of the system is supR kG(j)k

p. 2/1
Lemma: Consider the time-invariant system

x = f (x, u), y = h(x, u)

where f is locally Lipschitz and h is continuous for all

x Rn and u Rm . Let V (x) be a positive semidefinite
function such that
V
V = f (x, u) a( 2 kuk2 kyk2 ), a, > 0
x
Then, for each x(0) Rn , the system is finite-gain L2
stable and its L2 gain is less than or equal to . In particular
s
V (x(0))
ky kL2 ku kL2 +
a
p. 3/1
Proof
Z Z
V (x( ))V (x(0)) a 2 ku(t)k2 dt a ky(t)k2 dt
0 0

p. 4/1
Proof
Z Z
V (x( ))V (x(0)) a 2 ku(t)k2 dt a ky(t)k2 dt
0 0

V (x) 0
Z Z
V (x(0))
2 2 2
ky(t)k dt ku(t)k dt +
0 0 a

p. 4/1
Proof
Z Z
V (x( ))V (x(0)) a 2 ku(t)k2 dt a ky(t)k2 dt
0 0

V (x) 0
Z Z
V (x(0))
2 2 2
ky(t)k dt ku(t)k dt +
0 0 a
s
V (x(0))
ky kL2 ku kL2 +
a

p. 4/1
Lemma 6.5: If the system

x = f (x, u), y = h(x, u)

is output strictly passive with

uT y V + y T y, >0

then it is finite-gain L2 stable and its L2 gain is less than or

equal to 1/

p. 5/1
Lemma 6.5: If the system

x = f (x, u), y = h(x, u)

is output strictly passive with

uT y V + y T y, >0

then it is finite-gain L2 stable and its L2 gain is less than or

equal to 1/
Proof

V uT y y T y
1
= 2 (u y)T (u y) + 1 T
2 u u 2 y T y
1 T T


2 2 u u y y

p. 5/1
Example

x1 = x2 , x2 = ax31 kx2 + u, y = x2 , a, k > 0

p. 6/1
Example

x1 = x2 , x2 = ax31 kx2 + u, y = x2 , a, k > 0

a 4
V (x) = 4 x1 + 12 x22

p. 6/1
Example

x1 = x2 , x2 = ax31 kx2 + u, y = x2 , a, k > 0

a 4
V (x) = 4 x1 + 12 x22

V = ax31 x2 + x2 (ax31 kx2 + u)

= kx22 + x2 u = ky 2 + yu

p. 6/1
Example

x1 = x2 , x2 = ax31 kx2 + u, y = x2 , a, k > 0

a 4
V (x) = 4 x1 + 12 x22

V = ax31 x2 + x2 (ax31 kx2 + u)

= kx22 + x2 u = ky 2 + yu

The system is finite-gain L2 stable and its L2 gain is less

than or equal to 1/k

p. 6/1
Theorem 5.5: Consider the time-invariant system

x = f (x) + G(x)u, y = h(x)

f (0) = 0, h(0) = 0
where f and G are locally Lipschitz and h is continuous
over Rn . Suppose > 0 and a continuously
differentiable, positive semidefinite function V (x) that
satisfies the HamiltonJacobi inequality
T
V 1 V V 1

T
f (x)+ G(x)G (x) + hT (x)h(x) 0
x 2 2 x x 2

x Rn . Then, for each x(0) Rn , the system is

finite-gain L2 stable and its L2 gain
p. 7/1
Proof
V V
f (x) + G(x)u =
x x
2
T
1 2 1 V V
 
T
u 2 G (x) + f (x)

2 x x
T
1 V V 1 2

T 2
+ G(x)G (x) + kuk
2 2 x x 2

p. 8/1
Proof
V V
f (x) + G(x)u =
x x
2
T
1 2 1 V V
 
T
u 2 G (x) + f (x)

2 x x
T
1 V V 1 2

T 2
+ G(x)G (x) + kuk
2 2 x x 2

1 2 2 1
V kuk kyk2
2 2

p. 8/1
Example
x = Ax + Bu, y = Cx

p. 9/1
Example
x = Ax + Bu, y = Cx
Suppose there is P = P T 0 that satisfies the Riccati
equation

T 1 T T
PA + A P + 2
P BB P + C C=0

for some > 0.

p. 9/1
Example
x = Ax + Bu, y = Cx
Suppose there is P = P T 0 that satisfies the Riccati
equation

T 1 T T
PA + A P + 2
P BB P + C C=0

for some > 0. Verify that V (x) = 21 xT P x satisfies the

Hamilton-Jacobi equation

p. 9/1
Example
x = Ax + Bu, y = Cx
Suppose there is P = P T 0 that satisfies the Riccati
equation

T 1 T T
PA + A P + 2
P BB P + C C=0

for some > 0. Verify that V (x) = 21 xT P x satisfies the

Hamilton-Jacobi equation

The system is finite-gain L2 stable and its L2 gain is less

than or equal to

p. 9/1
 The Small-Gain Theorem

u1-e1- y1

H1 -
+
6

y2 e + u
?

 H2  2 + 2


p. 10/1
 The Small-Gain Theorem

u1-e1- y1

H1 -
+
6

y2 e + u
?

 H2  2 + 2


ky1 kL 1 ke1 kL + 1 , e1 Lm
e , [0, )

p. 10/1
 The Small-Gain Theorem

u1-e1- y1

H1 -
+
6

y2 e + u
?

 H2  2 + 2


ky1 kL 1 ke1 kL + 1 , e1 Lm
e , [0, )

ky2 kL 2 ke2 kL + 2 , e2 Lqe , [0, )

p. 10/1
 " # " # " #
u1 y1 e1
u= , y= , e=
u2 y2 e2

p. 11/1
 " # " # " #
u1 y1 e1
u= , y= , e=
u2 y2 e2

Theorem: The feedback connection is finite-gain L stable if

1 2 < 1

p. 11/1
 " # " # " #
u1 y1 e1
u= , y= , e=
u2 y2 e2

Theorem: The feedback connection is finite-gain L stable if

1 2 < 1

Proof

e1 = u1 (H2 e2 ) , e2 = u2 + (H1 e1 )

p. 11/1
 " # " # " #
u1 y1 e1
u= , y= , e=
u2 y2 e2

Theorem: The feedback connection is finite-gain L stable if

1 2 < 1

Proof

e1 = u1 (H2 e2 ) , e2 = u2 + (H1 e1 )

ke1 kL ku1 kL + k(H2 e2 ) kL

ku1 kL + 2 ke2 kL + 2

p. 11/1
ke1 kL ku1 kL + 2 (ku2 kL + 1 ke1 kL + 1 ) + 2
= 1 2 ke1 kL
+ (ku1 kL + 2 ku2 kL + 2 + 2 1 )

p. 12/1
ke1 kL ku1 kL + 2 (ku2 kL + 1 ke1 kL + 1 ) + 2
= 1 2 ke1 kL
+ (ku1 kL + 2 ku2 kL + 2 + 2 1 )

1
ke1 kL (ku1 kL + 2 ku2 kL + 2 + 2 1 )
1 1 2

p. 12/1
ke1 kL ku1 kL + 2 (ku2 kL + 1 ke1 kL + 1 ) + 2
= 1 2 ke1 kL
+ (ku1 kL + 2 ku2 kL + 2 + 2 1 )

1
ke1 kL (ku1 kL + 2 ku2 kL + 2 + 2 1 )
1 1 2
1
ke2 kL (ku2 kL + 1 ku1 kL + 1 + 1 2 )
1 1 2

p. 12/1
ke1 kL ku1 kL + 2 (ku2 kL + 1 ke1 kL + 1 ) + 2
= 1 2 ke1 kL
+ (ku1 kL + 2 ku2 kL + 2 + 2 1 )

1
ke1 kL (ku1 kL + 2 ku2 kL + 2 + 2 1 )
1 1 2
1
ke2 kL (ku2 kL + 1 ku1 kL + 1 + 1 2 )
1 1 2
ke kL ke1 kL + ke2 kL

p. 12/1