2 Lyapunov Stability

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

1

2 Lyapunov Stability

Whereas I/O stability is concerned with the effect of inputs on outputs, Lya-
punov stability deals with unforced systems:
ẋ = f (x, t) (1)
where x ∈ Rn , t ∈ R+ , and f : Rn × R+ → Rn . The system in (1) is
autonomous (time invariant) if f = f (x). Otherwise, it is nonautonomous
(time varying).
If f (x0 , t) = 0, then x0 is an equilibrium. Then, the unique solution of Eq.
(1) with x(0) = x0 is x(t) = x0 .
Definition. The equilibrium x0 is stable (or L-stable or stable in the sense
of Lyapunov) if for  > 0 there exists δ > 0 such that
||x(0) − x0 || < δ ⇒ ||x(t) − x0 || < 
x(t)
6

6

δ 6
x0 ? ?
6 6
δ
? 

-
t

The equilibrium x0 is unstable if it is not stable.


The equilibrium x0 is asymptotically stable if it is stable and there exists
δ > 0 such that
||x(0) − x0 || < δ ⇒ x(t) → x0 as t → ∞
The equilibrium x0 is globally asymptotically stable if it is stable and
x(t) → x0 as t → ∞
2

Consider the linear time-invariant system


ẋ = Ax, A ∈ Rn×n (2)
The eigenvalues of A are λi , i = 1, · · · , n. The solution of (2) is
x(t) = exp(At)x(0) (3)
where ∞ 1
(At)k = 1 + At + 12 A2 t2 + · · ·
X
exp(At) =
k=0 k!
which converges for t ∈ [0, ∞]. If A has distinct eigenvalues λi , and eigen-
vectors ei , then we have the following eigendecomposition:
A = EΛE−1
E = row{ei }
Λ = diag{λi }
In this case, it is not hard to show that
Ak = EΛk E−1
Λk = diag{λki }
exp(At) = E exp(Λt)E−1
exp(Λt) = diag{exp(λi t)}
We have the following stability results for the system in Eq. (2) and its
equilibrium x0 = 0.
(a) The system is stable if Re{λi } ≤ 0, i = 1 · · · n, and there are no repeated
eigenvalues on the imaginary axis.
(b) The system is unstable if there is at least one λi with Re{λi } > 0.
(c) The linear system is (globally) asymptotically stable if Re{λi } < 0, i =
1 · · · n.

2.1 Lyapunov’s Linearization (or First) Method

Consider the system ẋ = f (x) with f (x0 ) = 0. Let x(t) = x0 + δx(t). Then,
ẋ = δ ẋ = f (x0 + δx)
∂f

= f (x0 ) + T
δx + O(||δx||2 )
∂x x=x0

3

Neglecting higher order terms, we have the linearization


∂f

δ ẋ = Aδx, A = (4)
∂xT x=x0

If the equilibrium δx = 0 of (4) is:


(a) aysmptotically stable then x0 is an asymptotically stable equilibrium of
the nonlinear system;
(b) unstable then x0 is an unstable equilibrium of the nonlinear system;
(c) stable then no conclusion can be drawn about the nonlinear system.

2.2 Lyapunov’s Direct or Second Method

Consider the system in Eq. (1) with x = x0 an equilibrium. We assume that


x0 = 0 which can be accomplished with a change of coordinates.
A function V (x, t) is C 1 if it is continuously differentiable. It is a locally
positive-definite function [lpdf] if V (0, t) ≡ 0 and there exists r > 0 such
that V (x, t) > 0, ∀x 6= 0 s.t. ||x|| < r. It is a positive-definite function [pdf]
if V (0, t) ≡ 0 and V (x, t) > 0, ∀x ∈ Rn , x 6= 0.
It is radially unbounded if
V (x, t) → ∞ as ||x|| → ∞

It is locally negative definite if −V is a lpdf.


It is negative definite if −V is a pdf.
Let Br = {x ∈ Rn | ||x|| < r}.
Examples
V (x) = 1 − cos x is a lpdf but it is not radially unbounded.
V (x) = x2 is a pdf and it is radially unbounded.
V (x) = xT Px is a pdf and radially unbounded if P is a symmetric positive-
definite matrix.
Theorem 1. The equilibrium x0 = 0 of (1) is stable if there exists a C 1 lpdf
V (x, t) and r > 0 such that
V̇ (x, t) ≤ 0, ∀t ≥ 0, ∀x ∈ Br
4

where V̇ is evaluated along the trajectories of (1), i.e.,


∂V ∂V
V̇ = + T f (x, t)
∂t ∂x
V is called a Lyapunov function.
Example. Consider the equation of the simple pendulum

θ̈ + sin θ = 0
Let x1 = θ, x2 = θ̇ so that
ẋ1 = x2
ẋ2 = − sin x1
If we adopt V (x1 , x2 ) = (1 − cos x1 ) + 21 x22 as a Lyapunov function we have

V̇ = ẋ1 sin x1 + x2 ẋ2


= x2 sin x1 − x2 sin x1
= 0
Hence x = 0 is stable.
Example. Consider Euler’s equation Iω̇ + ω × Iω = 0 where I = IT > 0.
Letting V = 12 ω T Iω, we have

V̇ = ω T Iω̇
= −ω T ω × Iω
= 0
Hence, ω = 0 is a stable equilibrium.
Theorem 2. The equilibrium x0 = 0 of Eq. (1) is asymptotically stable if
there exists a C 1 lpdf V such that −V̇ is a lpdf.
Theorem 3. The equilibrium x0 = 0 of Eq. (1) is globally asymptotically
stable if there exists a C 1 pdf V such that V is radially unbounded and −V̇
is a pdf.
Note: Theorems 1-3 give sufficient conditions for stability.
Example. Consider the system
ẋ = −ax, a > 0
5

Letting V = 12 x2 > 0, x 6= 0, we have V̇ = xẋ = −ax2 < 0, x 6= 0. Therefore


x = 0 is a globally asymtotically stable equilibrium.
Example. Consider the system

Mq̈ + Kq = u(t), u = −Dq̇


M = MT > O, K = KT > O, D = DT > O

Letting x1 = q, x2 = q̇, we have

ẋ1 = x2
ẋ2 = −M−1 Kx1 − M−1 Dx2

Let
  
K O x1
V = 21 q̇T Mq̇ + 12 qT Kq = 12 [xT1 xT2 ]    > 0 (x 6= 0)
O M x2
which is a radially unbounded pdf. Then,
  
O O x1
V̇ = q̇T (Mq̈ + Kq) = −q̇T Dq̇ = −[xT1 xT2 ]    ≤0
O D x2
Hence, x = 0 is a stable equilibrium. It is in fact globally asymptotically
stable. How do we show this when V̇ does not contain q?

2.3 Krasovskii-LaSalle Theorem

Consider the autonomous system

ẋ = f (x), f (0) = 0 (5)

Suppose there exists a C 1 pdf V (x) that is radially unbounded and

V̇ ≤ 0, ∀t ≥ 0, ∀x ∈ Rn

Define the invariant set according to

M = {x ∈ Rn | V̇ (x) = 0}

If M contains only x = 0, then x = 0 is globally asymptotically stable.


6

In the previous example,


V̇ = −q̇T Dq̇ = 0 ⇒ q̇ = 0
If q̇ ≡ 0, then q̈ = 0 and
Mq̈ + Dq̇ + Kq = 0 ⇒ Kq = 0
and hence q = 0. Therefore the invariant set is given by M = {0} and x = 0
is globally asymptotically stable.
Theorem 4. The equilibrium x = 0 is unstable if there exists a C 1 lpdf
function V such that V̇ is an lpdf.
Example. Consider the system
ẋ = ax, a > 0
Letting V = 12 x2 > 0, x 6= 0, we have V̇ = xẋ = ax2 > 0, x 6= 0. Therefore
x = 0 is an unstable equilibrium.

2.4 Stability of Linear Systems

Consider the linear time-invariant system


ẋ = Ax (6)
Let us select
V = xT Px, P = PT > O (7)
as a Lyapunov function candidate. Hence
V̇ = ẋT Px + xT Pẋ
= xT (AT P + PA)x
If,
PA + AT P = −Q, Q = QT > O (8)
then
V̇ = −xT Qx < 0 (x 6= 0)
Hence, if given Q = QT > O, the Lyapunov equation (8) has a symmetric
positive-definite solution P, then the eigenvalues of A have negative real
parts, i.e, the system (6) is asymptotically stable.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy