MASSACHUSETTS INSTITUTE OF TECHNOLOGY

Department of Electrical Engineering and Computer Science

6.241: Dynamic Systems|Fall 2003

Recitation 6

Lyapunov Methods

In this recitation we review concepts of stability and Lyapunov's direct and indirect methods for
analysing the stability of a system around an equilibrium point. We then present a series of
examples.

De nitions of Stability
Consider an unforced (time-invariant) nonlinear system described by x_ (t) = f (x(t)). An equilibrium
point of the system, x, is one for which f (
x) = 0. Since any system with x not located at the ori-
gin can be translated to obtain an equivalent system with the equilibrium point at the origin (e.g.,
let z = x x), we assume in our de nitions that the equilibrium point of our system is at the origin.

The system is called stable in the sense of Lyapunov (i.s.L.) around the equilibrium point at the
origin if given any  > 0, there exists a Æ > 0 such that if kx(t0 )k < Æ then kx(t)k <  for all t > t0 :
A system is called asymptotically stable around the equilibrium point at the origin if it is stable in
the sense of Lyapunov and there exists a > 0 such that if kx(t0 )k < then x(t) ! 0 as t ! 1: If
limt!1 x(t) = 0 no matter where x(t0 ) lies in the state space, then the system is said to be globally
asymptotically stable.

Lyapunov's Direct Method

In general, proving that a nonlinear system of the form x_ (t) = f (x(t)) is asymptotically stable
around the origin is a diÆcult task, as it is diÆcult to write down a closed form solution of x(t)
in terms of x(t0 ). For linear time-invariant systems (x_ (t) = Ax(t) + Bu(t)), we have a closed form
solution, i.e., Z t
x(t) = eA(t t0 )
x(t0 ) + eA(t )
Bu( )d: (1)
t0

For any A (regardless of whether it is diagonalizable or not), the linear system x_ = Ax is asymp-
totically stable at the origin if and only if all the eigenvalues of A lie in the open left-half complex
plane1 . This is due to the decaying exponential terms in x(t). On the other hand, if the system
has at least one eigenvalue in the open right-half plane, then the linear system is unstable (i.e. not
stable i.s.L.). Stability of the system (i.s.L.) if it has eigenvalues on the imaginary axis depends on
the algebraic and geometric multiplicity of these eigenvalues (See lecture notes page 138).
1
We assume that the origin is the only equilibrium point, i.e., A has full rank

1
To analyze the stability of a general nonlinear systems, we can carefully choose a scalar-valued
function, V (x), of the state variables and see how V (x) evolves as the states evolve.

Lyapunov Function. Let V be a continuous map from IRn to IR, then V (x) is called a locally
positive de nite (lpd) function around x = 0 if
1. V (0) = 0
2. V (x) > 0 on 0 < kxk < r1 for some r1
V is called a Lyapunov function if V is di erentiable, locally positive de nite, and V_ , the derivative
of V along trajectories of the system, is locally negative semi-de nite (V_ (x)  0 on 0 < kxk < r2
for some r2 ).

Lyapunov Theorem for Local Stability: If there exists a Lyapunov function of the system
x_ = f (x(t)), then x = 0 is a stable equilibrium point i.s.L. If in addition, V_ (x) < 0, for 0 < kxk < r
for some r, then x = 0 is asymptotically stable (locally in the region 0 < kxk < r).

Global asymptotic stability (G.A.S.) of the origin may be concluded if there exists a V (x) that is
positive de nite and V_ (x) that is negative de nite in the entire state space. In addition, V must
be radially unbounded, that is, V (x) ! 1 as kxk ! 1. For a proof, see [1], page 109-111.

Lyapunov's Indirect Method

Instead of looking for a Lyapunov function to be applied directly to the nonlinear system, one can
linearize the system around the origin and attempt to conclude \local" stability of the origin using
quadratic Lyapunov functions for the linearized system. More speci cally, if the linearized system's
A matrix has eigenvalues in the open LHP, the nonlinear system is then locally asymptotically stable
(L.A.S.). In addition, a Lyapunov function for the nonlinear system is quadratic, i.e., V (x) = x0 P x
where P > 0 solves the following Lyapunov equation: AT P + P A = Q; where Q > 0 is an
arbitrary but positive de nite matrix. See lecture 14 for details.

Examples

Example 1: In this example, we will prove global stability of the equilibrium point using the
direct method.
x_ 1 = x1 x2
x_ 2 = x1 x32
First, note that the origin is the only equilibrium point of the system. Furthermore, linearizing
about the origin, we obtain the dynamics of the linearized system,
    
Æx_ 1 =
1 1 Æx1 :
Æx_ 2 1 0 Æx2
p
The eigenvalues of the linearization are given by 0:5  j 0:5 3, so we may conclude, via Lyapunov's
indirect method, that the origin is locally asymptotically stable. However, the indirect method does

2
not provide any information about how close to the origin we have to be to guarantee stability.
Thus, we turn to the Lyapunov's direct method.

Consider using a candidate Lyapunov function V (x) of the form ax21 + bx22 where a and b are some
positive parameters to be determined. Clearly V (x) is positive de nite over the entire state space
and V (x) is radially unbounded. Now, we need to check that V_ (x) is negative de nite:
 

V_ (x) = 2ax1 2bx2
 x1 x2 = 2ax21 2bx42 + 2x1 x2 (b a):
x1 x32
If we choose a = b = 1=2, then V_ (x) = x21 x42 < 0 for all x 6= 0. Thus, this system is G.A.S.

Example 2: In this example, we nd that the origin is locally asymptotically stable. We cannot
conclude global asymptotic stability of the origin due to the presence of more than one equilibrium
point; but, using Lyapunov's direct method, we can nd the set of initial conditions for which the
system trajectories will eventually end up at the origin (this ball is called the region of attraction).

Consider the autonomous (time-invariant) system described by the di erential equations:

x_ 1 = (x1 x2 )(x21 + x22 1)
x_ 2 = (x1 + x2 )(x21 + x22 1)
This system has an in nite number of equilibrium points: one at the origin and the rest on the
unit circle. Since there are multiple equilibria, none of the equilibria can be globally asymptotically
stable; furthermore, since the points on the unit circle are not isolated, none can be locally asymp-
totically stable. We wish to examine the equilibrium point at the origin. Linearizing the system
about the origin, we nd that that the linearized dynamics have eigenvalues at 1  j , so we can
conclude, via the indirect method, local asymptotic stability (L.A.S.) of the origin.

Now, using the direct method, we can nd the set of initial condtions for which the system trajec-
tories converge to 0. Consider V (x) = ax21 + bx22 , with a; b > 0, which is positive de nite for all x1
and x2 (in IR2 ). Now,

 
  (x1 x2 )(x21 + x22 1)
V_ (x) = 2ax1 2bx2 (x1 + x2 )(x21 + x22 1)
= (2ax21 + 2bx22 + 2x1 x2 (b a))(x21 + x22 1):

Note that x21 +x22 1 < 0 when kxk2 < 1, x 6= 0. If b = a = 1=2, then V_ (x) = (x21 +x22 )(x21 +x22 1) < 0
for all nonzero x in the open unit circle. Thus the origin is L.A.S. within the region kxk2 < 1.

Example 3: In this example, we see that sometimes it is hard to nd a Lyapunov function. Still,
we cannot always rush to the conclusion that the system is not stable, even if it is impossible to
nd a Lyapunov function for that system. For more on \converse" theorems (i.e. those that detail
conditions for which a Lyapunov function is guaranteed to exist), see [1].

Analyze the stability of the origin for the system described by:
x_ 1 = x1 + x22
x_ 2 = x2

3
This system has only one equilibrium point at the origin. Linearizing the system at the origin
reveals that the system has two eigenvalues at 1, and the origin is therefore at least locally
asymptotically stable. To make further conclusions about the stability of the origin (namely its
region of attraction) we attempt to use a Lyapunov function of the form V (x) = ax21 + bx22 , only
to nd that it is inconclusive. More speci cally, in this case, V_ (x) = 2ax21 + 2ax1 x22 2bx22 ; this
function is not negative de nite for any nonzero values of a and b (note that we need a; b > 0 to
keep V (x) positive de nite).

We might try to search for another Lyapunov function, but this is not an easy task. Alternatively,
in this case, it is possible to nd a closed form solution since x2 evolves independently of x1 .
Speci cally,

x2 (t) = e t x2 (0)
so,
x_ 1 (t) = x1 (t) + e 2t x22 (0):
The e 2t term can be seen as some input u(t), and the closed form solution of the linear di erential
equation in x1 can be obtained from equation 1 above,
Z t
x1 (t) = e t x1 (0) + e (t
x2 (0)e
) 2 2
d:
0
So,
x1 (t) = x1 (0)e t + e t (1 e t )x22 (0)
x2 (t) = x2 (0)e t :
So, as t ! 1, x1 (t) ! 0, and x2 (t) ! 0 for any initial conditions x2 (0) and x1 (0). Thus, this
system is G.A.S.

Example 4: Finally, we have basically dealt with CT in this recitation, but Lyapunov theory
extends to the DT case, x(k + 1) = f (x(k)) with minor changes. Speci cally, in DT, an equilibrium
point, x is one for which f (x) = x; and we de ne V_ as follows, V_ (x) = V (f (x)) V (x):

Consider the system (Example 13.7 of lecture notes):

x2 (k)
x1 (k + 1) =
1 + x22 (k)
x1 (k)
x2 (k + 1) = :
1 + x22 (k)
This system has an equilibrium point at the origin. Now let V (x) = x21 + x22 , which is positive
de nite, and evaluate V_ (x(k)),
 
1
_V (x(k)) = V (x(k)) 1 0
[1 + x22 (k)]2
Thus, we can conclude that the equilibrium point is stable i.s.L.

References
[1] Khalil, H.K. \Nonlinear Systems," Prentice Hall: 1996.