4IE fy cont'd forlater.

no
https://join.iclicker.com/MMAW
II Multivariable Control SHU
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

Summary so Far
1. We can determine local stability properties for hyperbolic fixed
points of nonlinear systems by examining the eigenvalues of the
linear approximation.
1.1 The fixed point is asymptotically stable for the nonlinear system if it
is asymptotically stable for the linear systems (all eigenvalues have
negative real part)
1.2 The fixed point is unstable for the nonlinear system if it is unstable
for the linear system. (one or more eigenvalues with positive real
part)
1.3 If any eigenvalue of the linear approximation has zero real part
(critical eigenvalues) then the fixed point of the nonlinear system
may be stable or unstable, i.e. the linearization is inconclusive.
2. Lyapunov functions can be used to determine stability in the case of
critical eigenvalues
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

The Invariance Principle

We next give an extension to Lyapunov’s Theorem due to LaSalle that
allows additional conclusions about the asymptotic behavior of the
system. A weak form of LaSalle’s theorem can be stated as
Theorem
Suppose V : Rn ! R is a positive definite scalar function and suppose
V̇  0 along trajectories of (1). If V̇ does not vanish identically along
trajectories of (1) other than the equilibrium solution, then x̄ = 0 is
asymptotically stable.
xdot = f(x) (1)
Example
Back to the simple pendulum with friction

ẋ1 = x2
g
ẋ2 = sin(x1 ) Bx2
`
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

We know that with

g 1 2
V = (1 cos(x1)) + x2
` 2
because
Notneg
def
satisfies TV
0
j x a
V̇ = -Bx22  0
c
Suppose V̇ ⌘ 0. Then x2 ⌘ 0 =) ẋ2 ⌘ 0 ng
=) sin(x1 ) ⌘ 0 =) x1 = n⇡. *we consider local stability with -pi < x1 < pi
Therefore, V̇ vanishes identically only at the equilibrium point and so
x̄ = 0 is asymptotically stable.
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

LaSalle’s Theorem

We next present a stronger version of LaSalle’s Theorem.

Theorem
Suppose V is a Lyapunov function satisfying V̇  0 along trajectories of
ẋ = f (x). Let E be the set of points where V̇ = 0 and let M be the
largest invariant set contained in E. Then all trajectories of the system
approach M as t ! 1.
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

LaSalle’s Theorem

We next present a stronger version of LaSalle’s Theorem.

Theorem
Suppose V is a Lyapunov function satisfying V̇  0 along trajectories of
ẋ = f (x). Let E be the set of points where V̇ = 0 and let M be the
largest invariant set contained in E. Then all trajectories of the system
approach M as t ! 1.

Example
Consider the nonlinear system

ẋ1 = x1 (1 + x2 )
ẋ2 = x21
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

Example
Let
1 2 1 2
V = x1 + x2
2 2
Then
defiance
V̇ = x1ẋ 1 + x2ẋ 2 2
on
= x21(1 + x2) + x2x21 = x21  0

Therefore E = {x1 = 0}, which is also an invariant set. Therefore, all

trajectories approach E as t ! 1.

ii
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

Stability of Linear Systems

We have been using functions of the form

V = x21 + x22 + · · · + x2n

as potential Lyapunov functions in several examples. The ‘sum of

squares’ or ‘norm’ function is called a Quadratic Form. A more general
quadratic form can be written as
2 3
p11 p12 . . . p1n
6 p21 p22 . . . p2n 7
6 7
V = xT P x where x 2 Rn and P = 6 . . . . 7
4 .. .
. . . .
. 5
pn1 pn2 ... pnn

P is an n ⇥ n constant matrix. To qualify as a Lyapunov function, the

matrix P should satisfy xT P x > 0 for x 6= 0.
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

Positive Definite Matrices

Definition
An n ⇥ n matrix P is Positive Definite if and only if

xT P x > 0 for x 6= 0

The matrix P is
• Positive Semi-definite if xT P x 0
• Negative Definite if xT P x < 0
• Negative Semi-Definite if xT P x  0
• P is Indefinite if it is otherwise not one of the above.
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

Positive Definite Matrices

Remark
• The matrix P is positive definite if and only if all eigenvalues of P
are positive.
• P is positive semi-definite if all eigenvalues of P are greater than or
possibly equal to zero.

Given an n-dimensional linear system

ẋ = Ax

let us try to find a quadratic Lyapunov function of the form

V = xT P x
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

Stability of Linear Systems

Computing the derivative of V = xT P x along trajectories of ẋ = Ax yields

V̇ = ẋT P x + xT P ẋ
= xT AT P x + xT P Ax
= xT (AT P + P A)x

Let’s define Q = AT P + P A. Then we have

V = xT P x
V̇ = xT Qx
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

Stability of Linear Systems

Computing the derivative of V = xT P x along trajectories of ẋ = A yields

V̇ = ẋT P x + xT P ẋ
= xT AT P x + xT P Ax
= xT (AT P + P A)x

Let’s define Q = AT P + P A. Then we have

V = xT P x
V̇ = xT Qx
https join.iclicker.comMMAW
Suppose both P and Q are positive definite. What can you conclude?

a System is unstable a asymptotically stable

in a exponentiallystable d in stable
b
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

Stability of Linear Systems

Computing the derivative of V = xT P x along trajectories of ẋ = A yields

V̇ = ẋT P x + xT P ẋ
= xT AT P x + xT P Ax
= xT (AT P + P A)x

Let’s define Q = AT P + P A. Then we have

V = xT P x
V̇ = xT Qx

Suppose both P and Q are positive definite. What can you conclude?
The conclusion is that the linear system is (exponentially) stable.
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

The Lyapunov Equation

The problem is how to select an appropriate positive-definite matrix P .

Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

The Lyapunov Equation

The problem is how to select an appropriate positive-definite matrix P .

Suppose we turn the problem around. Choose Q to be any

positive-definite matrix and consider the equation

ATP + P A + Q = 0

where P is the unknown and A and Q are given. Suppose that we solve
Equation (1) for P and find that P is positive-definite. What conclusions
can you draw?
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

The Lyapunov Equation

The problem is how to select an appropriate positive-definite matrix P .

Suppose we turn the problem around. Choose Q to be any

positive-definite matrix and consider the equation

ATP + P A + Q = 0

where P is the unknown and A and Q are given. Suppose that we solve
Equation (1) for P and find that P is positive-definite. What conclusions
can you draw?
The equation (1) is called a Lyapunov Equation.

Px 0
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

Stability of Linear Systems

The leads to the following result to determine stability of linear systems

Theorem (globally) asymptotically stable
------- (i.e. all eigenvalues of A have negative real
The n ⇥ n matrix A is stable
part) if and only if for any symmetric positive definite n ⇥ n matrix Q
there is a unique symmetric positive definite matrix P satisfying the
Lyapunov Equation
ATP + P A + Q = 0

Remark
This gives a way to determine stability of linear systems by solving a
system of linear algebraic equations; in other words, without computing
matrix eigenvalues.
Nonlinear Systems Linearization Stability Lyapunov Theory Basin Global Stability The Invariance Principle Stability of Linear Systems

Example
Remark
The Matlab Control Systems Toolbox has the built-in function lyap,
which solves the Lyapunov equation.
Matlab Script

>> A = [0 1; 2 3];
>> Q = [1 0; 0 1];
>> P = lyap(A, Q)
>> P =
1.0000 0.5000
0.5000 0.5000
>> eig(P )
ans =
0.1910
1.3090
I Multivariable Control

A PD control Proportional Derivative Control

G g content ation
i M q c
qg q
Edinclude motor inertias by adding diagonal matrix
to the prior definition

First assume G
g 0 gravity free

Define independendent joint PD control law affiant

where
a
Kpg Kaq q q q if q
Thiagonal
matrices
Kp Kd 70

Note and x̅ 0 is
eq.pt
Q Does this controller make x̅ O a stable eq.pt
is this constable

Try Lyapunov function

x
q qTM g if qtKpq 0 x O

Effectors
replaced bysprings
virtuality.fr
i qtM q.tn q j qtkp
q.q7g qᵗmlq qtkpq
qt u

ily jul 25
it Kaq q.TKa q 0 X

Étq
depend

This prove weak form of stability in the sense of

Lyapunov

Try Lasalle's thin i.e invariance principle

suppose O
q if 0

c no
Kpq Hq O 8 0

I e V vanishes only at eq.pt q q 0,0

Asymptotically stable A 5