Chapter 3: Lyapunov Stability I: Autonomous

Systems
1 Denitions
Consider the autonomous system
x = f(x) f : D R
n
(1)
Denition 1 : x
e
is an equilibrium point of (1) if
f(x
e
) = 0.
We want to know whether or not the trajecories near an equilibrium point
are well behaved.
Denition 2 : x
e
is said to be stable if for each > 0,
= () > 0
x(0) x
e
< x(t) x
e
< t t
0
otherwise, the equilibrium point is said to be unstable.
IMPORTANT: This notion applies to the equilibrium, not the system. A
dynamical system can have several equilibrium points.
1
-
6
&%
'$
s
x
e
r
x
0
@
@I

H
H
H
H
HHj

Figure 1: Stable equilibrium point.

-
6
&%
'$
s
x
e
r
x
0

H
H
H
H
HHj

Figure 2: Asymptotically stable equilibrium point.

Denition 3 : x
e
of the system (1) is said to be convergent if there exists
1
> 0 :
x(0) x
e
<
1
lim
t
x(t) = x
e
.
Equivalently, x
e
is convergent if for any given
1
> 0, T such that
x(0) x
e
<
1
x(t) x
e
<
1
t t
0
+T.
Denition 4 : x
e
is said to be asymptotically stable if it is both stable and con-
vergent.
2
Denition 5 : x
e
is said to be (locally) exponentially stable if there exist two
real constants , > 0 such that
x(t) x
e
x(0) x
e
e
t
t > 0 (2)
whenever x(0) x
e
< . It is said to be globally exponentially stable if (2)
holds for any x R
n
.
Note: Clearly, exponential stability implies asymptotic stability. The con-
verse is, however, not true.
Remarks: We can always assume that x
e
= 0. Given any other equilibrium
point we can make a change of variables and dene a new system with an
equilibrium point at x = 0. Dene:
y = x x
e
y = x = f(x)
f(x) = f(y + x
e
) g(y).
Thus, the equilibrium point y
e
of the new systems y = g(y) is y
e
= 0, since
g(0) = f(0 + x
e
) = f(x
e
) = 0.
Thus, x
e
is stable for the system x = f(x) if and only if y = 0 is stable for
the system y = g(y).
3
2 Positive Denite Functions
Denition 6 : V : D R is positive semi denite in D if
(i) 0 D and V (0) = 0.
(ii) V (x) 0, x in D {0}.
V : D R is positive denite in D if
(ii) V (x) > 0 in D {0}.
V : D R is negative denite (semi denite)in D if V is positive
denite (semi denite).
4
Example 1 : (Quadratic form)
V (x) : R
n
R = x
T
Qx , Q R
nn
, Q = Q
T
.
Since Q = Q
T
, its eigenvalues
i
, i = 1, n, are all real. Thus
V () positive denite
i
> 0, i = 1, , n
V () positive semidenite
i
0, i = 1, , n
V () negative denite
i
< 0, i = 1, , n
V () negative semidenite
i
0, i = 1, , n
2
Notation: Given a dynamical system and a function V we will denote

V (x) =
dV
dt
=
V
x
dx
dt
= V f(x)
=
_
V
x
1
,
V
x
2
, ,
V
x
n
_
_
_
f
1
(x)
.
.
.
f
n
(x)
_
_
.
Example 2 : Let
x =
_
ax
1
bx
2
+ cos x
1
_
and dene V = x
2
1
+ x
2
2
. Thus, we have

V (x) =
V
x
f(x) = [2x
1
, 2x
2
]
_
ax
1
bx
2
+ cos x
1
_
= 2ax
2
1
+ 2bx
2
2
+ 2x
2
cos x
1
.
2
Notice that the

V (x) depends on the systems equation f(x) and thus it
will be dierent for dierent systems.
5
Note:
V = x
T
Qx, where Q = Q
T
if Q = Q
sym
+Q
asym
Q
sym
=
Q+Q
T
2
Q
asym
=
QQ
T
2
Then, V = x
T
Q
sym
x +x
T
Q
asym
x
where, Q
asym
=
_
_
0 a b
a
.
.
.
.
.
.
b 0
_
_
i.e. skew symmetric
and x
T
Q
asym
x = 0
V = x
T
Q
sym
x Q is always symmetric
Theorem: Rayleigh-Ritz theorem

min
{Q} x
2
x
T
Qx
max
{Q} x
2
6
3 Stability Theorems
Theorem 1 : (Lyapunov Stability Theorem) Let x = 0 be an equilibrium point
of x = f(x), and let V : D R be a continuously dierentiable function
such that
(i) V (0) = 0,
(ii) V (x) > 0 in D {0},
(iii)

V (x) 0 in D {0},
thus x = 0 is stable.
Theorem 2 : (Asymptotic Stability Theorem) Under the conditions of Theorem
1, if V () is such that
(i) V (0) = 0,
(ii) V (x) > 0 in D {0},
(iii)

V (x) < 0 in D {0},
thus x = 0 is asymptotically stable.
7
Proof of theorem 1: Choose r > 0 and dene
B
r
= {x R
n
: x r} D
is contained in D. B
r
, so dened is a closed and bounded (compact) set (a
sphere).
We now construct a Lyapunov surface inside B
r
and show that all trajec-
tories starting near x = 0 remain inside the surface. Let
= min
x=r
V (x) (thus > 0)
Now choose (0, ) and denote

= {x B
r
: V (x) }.
Thus, by construction,

B
r
. Assume now that x(0)

V (x) 0 V (x) V (x(0)) t 0.

Trajectories starting in

at t = 0 stays inside

for all t 0. By the

continuity of V (x), > 0:
x < V (x) < (B

B
r
).
It then follows that
x(0) < x(t)

B
r
t > 0
and then
x(0) < x(t) < r t 0
which means that the equilibrium x = 0 is stable. 2
Proof of theorem 2: Similar, only that

V (x) < 0 in D implies that

is shrinking until eventually it becomes the single point x = 0. 2

8

p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
p
B
B
B
B
B
B
B
B
B
m
?
mg

l
Figure 3: Pendulum without friction.
4 Examples
Example 3 : (Pendulum Without Friction)
Using Newtons second law of motion we have,
ma = mg sin
a = l = l

where l is the length of the pendulum, and is the angular acceleration. Thus
ml

+mg sin = 0
or

+
g
l
sin = 0
choosing state variables
_
x
1
=
x
2
=

we have
_
x
1
= x
2
= f
1
(x)
x
2
=
g
l
sin x
1
= f
2
(x)
To study the stability of the equilibrium at the origin, we need to propose a
Lyapunov function candidate V (x). This is dicult! In this case we try
the total energy, which is a positive function. We have
E = K +P (kinetic plus potential energy)
=
1
2
m(l)
2
+mgh
9
where
=

= x
2
h = l(1 cos ) = l(1 cos x
1
).
Thus
E =
1
2
ml
2
x
2
2
+ mgl(1 cos x
1
).
We now dene V (x) = E. We see that because of the periodicity of cos(x
1
),
we have that V (x) = 0 whenever x = (x
1
, x
2
)
T
= (2k, 0)
T
, k = 1, 2, .
Thus, V () is not positive denite. However, restricting the domain of x
1
to
the interval (2, 2); i.e., we take V : D R, with D = ((2, 2), R)
T
.
We have that V : D R > 0 is indeed positive denite. Also

V (x) = V f(x)
=
_
V
x
1
,
V
x
2
_
[f
1
(x), f
2
(x)]
T
= [mgl sin x
1
, ml
2
x
2
][x
2
,
g
l
sin x
1
]
T
= mglx
2
sinx
1
mglx
2
sin x
1
= 0.
Thus

V (x) = 0 and the origin is stable by Theorem 1. 2
10
Example 4 : (Pendulum with Friction) We now modify the previous example
by adding the friction force kl

ml

= mg sin kl

dening the same state variables as in example 3 we have

_
x
1
= x
2
x
2
=
g
l
sin x
1

k
m
x
2
.
Again x = 0 is an equilibrium point. The energy is the same as in Example
3. Thus
V (x) =
1
2
ml
2
x
2
2
+mgl(1 cos x
1
) > 0 in D {0}

V (x) = V f(x)
=
_
V
x
1
,
V
x
2
_
[f
1
(x), f
2
(x)]
T
= [mgl sinx
1
, ml
2
x
2
][x
2
,
g
l
sinx
1

k
m
x
2
]
T
= kl
2
x
2
2
.
Thus

V (x) is negative semi-denite. It is not negative denite since

V (x) = 0
for x
2
= 0, regardless of the value of x
1
(thus

V (x) = 0 along the x
1
axis).
We conclude that the origin is stable by Theorem 1, but cannot conclude
asymptotic stability. 2
The result is disappointing. We know that a pendulum with friction cin-
verges to x = 0. This example emphasizes the fact that all of the Lyapunov
theorems provide sucient but not necessary conditions for stability.
11
Example 5 : Consider the following system:
x
1
= x
1
(x
2
1
+ x
2
2

2
) + x
2
x
2
= x
1
+ x
2
(x
2
1
+ x
2
2

2
).
To study the equilibrium point at the origin, we dene V (x) = 1/2(x
2
1
+ x
2
2
).
We have

V (x) = V f(x)
= [x
1
, x
2
][x
1
(x
2
1
+ x
2
2

2
) +x
2
2
, x
1
x
2
(x
2
1
+x
2
2

2
)]
T
= x
2
1
(x
2
1
+ x
2
2

2
) +x
2
2
(x
2
1
+x
2
2

2
)
= (x
2
1
+ x
2
2
)(x
2
1
+x
2
2

2
).
Thus, V (x) > 0 and

V (x) 0, provided that (x
2
1
+ x
2
2
) <
2
, and it follows
that the origin is an asymptotically stable equilibrium point. 2
12
5 Asymptotic Stability in the Large
Denition 7 : The equilibrium state x
e
is globally asymptotically stable (or A.S.
in the large), if it is stable and every motion converges to the equilibrium as
t .
Question Can we infer that if the conditions of Theorem 2 hold in the whole
space R
n
, then the asymptotic stability of the equilibrium is global?
Answer: No! and the next example illustrates this.
13
-
6
s
x
0
V =
1
V =
2
V =
3

3
>
2
>
1
Figure 4: The curves V (x) = .
Example 6 : Consider the following positive denite function:
V (x) =
x
2
1
1 +x
2
1
+ x
2
2
.
The region V (x) is closed for values of < 1. However, if > 1, the
surface is open. Figure 4 shows that an initial state can diverge from the
equilibrium state at the origin while moving towards lower energy curves. 2
Denition 8 : Let V : D R be a continuously dierentiable function. Then
V (x) is said to be radially unbounded if
V (x) as x .
Theorem 3 : (Global Asymptotic Stability) Under the conditions of Theorem 2,
if V () is radially unbounded then x = 0 is globally asymptotically stable.
14
6 Positive Denite Functions Revisited
Denition 9 : A continuous function : [0, a) R
+
is said to be in the class
K if
(i) (0) = 0.
(ii) It is strictly increasing.
is said to be in the class K

if in addition : R
+
R
+
and (r)
as r .
In the sequel, B
r
represents the ball
B
r
= {x R
n
: x r}.
Lemma 4 : V : D R is positive denite if and only if there exists class K
functions
1
and
2
such that

1
(x) V (x)
2
(x) x B
r
D.
Moreover, if D = R
n
and V () is radially unbounded then
1
and
2
can be
chosen in the class K

.
Example 7 : Let V (x) = x
T
Px, where P is a constant positive denite sym-
metric matrix. Denote
min
(P) and
max
(P) the minimum and maximum
eigenvalues of P, respectively. We have:

min
(P)x
2
x
T
Px
max
(P)x
2

min
(P)x
2
V (x)
max
(P)x
2
.
Thus,
1
,
2
: [0, ) R
+
, and are dened by

1
(x) =
min
(P)x
2

2
(x) =
max
(P)x
2
.
2
15
Lemma 5 : x = 0 is stable if and only if there exists a class K function ()
and a constant such that
x(0) < x(t) (x(0)) t 0. (3)
A stronger class of functions is needed in the denition of asymptotic stability.
Denition 10 : A continuous function : [0, a) R
+
R
+
is said to be in the
class KL if
(i) For xed s, (r, s) is in the class K with respect to r.
(ii) For xed r, (r, s) is decreasing with respect to s.
(iii) (r, s) 0 as s .
Lemma 6 : The equilibrium x = 0 of the system (1) is asymptotically stable if
and only if there exists a class KL function (, ) and a constant such that
x(0) < x(t) (x(0), t) t 0. (4)
16
6.1 Exponential Stability
Theorem 7 : Suppose that all the conditions of Theorem 2 are satised, and in
addition assume that there exist positive constants K
1
, K
2
, K
3
and p such
that
K
1
x
p
V (x) K
2
x
p

V (x) K
3
x
p
.
Then the origin is exponentially stable. Moreover, if the conditions hold glob-
ally, the x = 0 is globally exponentially stable.
Proof: According to the assumptions of Theorem 7, the function V (x) sat-
ises Lemma 4 with
1
() and
2
(), satisfying somewhat strong conditions.
Indeed, by assumption
K
1
x
p
V (x) K
2
x
p

V (x) K
3
x
p

K
3
K
2
V (x)
i.e.

V (x)
K
3
K
2
V (x)
V (x) V (x
0
)e
(K
3
/K
2
)t
x [
V (x)
K
1
]
1/p
[
V (x
0
)e
(K
3
/K
2
)t
K
1
]
1/p
or
x(t) x
0
[
K
2
K
1
]
1/p
e
(K
3
/K
2
)t
.
2
17
7 The Invariance Principle
Asymptotic stability is always more desirable that stability. Lyapunov func-
tions often fail to identify asymptotic stability. We now study an improve-
ment over the Lyapunov theorems studied earlier.
Denition 11 : A set M is said to be an invariant set with respect to the dy-
namical system x = f(x) if:
x(0) M x(t) M t R
+
.
Example 8 : Any equilibrium point is an invariant set, since if at t = 0 we have
x(0) = x
e
, then x(t) = x
e
t 0. 2
Example 9 : For autonomous systems, any trajectory is an invariant set. 2
Example 10 : A limit cycle is an invariant set (a special case of Example 9).
2
Example 11 : If V (x) is continuously dierentiable (not necessarily positive
denite) and satises

V (x) < 0 along the solutions of x = f(x), then the set

l
dened by

l
= {x R
n
: V (x) l}
is an invariant set. 2
18
Theorem 8 : The equilibrium point x = 0 of the autonomous system (1) is
asymptotically stable if there exists a function V (x) satisfying
(i) V (x) positive denite x D, where we assume that 0 D.
(ii)

V (x) is negative semi denite in a bounded region R D.
(iii)

V (x) does not vanish identically along any trajectory in R, other than
the null solution x = 0.
Example 12 : Consider again the pendulum with friction of Example 4:
x
1
= x
2
(5)
x
2
=
g
l
sinx
1

k
m
x
2
. (6)
Again
V (x) > 0 x (, ) R,

V (x) = kl
2
x
2
2
(7)
which is negative semi denite since

V (x) = 0 for all x = [x
1
, 0]
T
(so x = 0
stable but cannot conclude AS). We now apply Theorem 8. Conditions (i)
and (ii) of Theorem 8 are satised in the region
R =
_
x
1
x
2
_
with < x
1
< , and a < x
2
< a, for any a R
+
. We now check
condition (iii), that is, we check whether

V can vanish identically along the
trajectories trapped in R, other than the null solution.
By (7) we have

V (x) = 0 0 = kl
2
x
2
2
x
2
= 0
thus x
2
= 0 t x
2
= 0
and by (6), we obtain
0 =
g
l
sinx
1

k
m
x
2
andthus x
2
= 0 sinx
1
= 0
restricting x
1
to x
1
(, ) we have that the last condition is satised if
and only if x
1
= 0. Thus,

V (x) = 0 does not vanish identically along any
trajectory other than x = 0. Thus x = 0 is asymptotically stable by Theorem
8. 2
19
Theorem 9 : The null solution x = 0 of the autonomous system (1) is asymp-
totically stable in the large if the assumptions of theorem 8 hold in the entire
state space and V () is radially unbounded.
Example 13 :
x
1
= x
2
x
2
= x
2
x
1
(x
1
+ x
2
)
2
x
2
.
where is a positive scalar.
To study the stability of x = 0 we dene V (x) = x
2
1
+ x
2
2
(Radially
unbounded). Thus dierentiating V (x)

V (x) =
V
x
f(x)
= 2x
2
2
[1 + (x
1
+ x
2
)
2
]
and V (x) > 0 and

V (x) 0 since

V (x) = 0 for x = (x
1
, 0). Assume now
that

V = 0:

V = 0 x
2
= 0 , x
2
= 0 t x
2
= 0
x
2
= 0 x
2
x
1
(x
1
+x
2
)
2
x
2
= 0
and considering the fact that x
2
= 0, the last equation implies that x
1
= 0.
It follows that

V (x) does not vanish identically along any solution other
than x = [0, 0]
T
. Thus, x = 0 is globally asymptotically stable. 2
Theorem 10 : (LaSalles theorem) Let V : D R be a continuously dieren-
tiable function and assume that
(i) M D is a compact set, invariant with respect to the solutions of (1).
(ii)

V 0 in M.
(iii) E : {x : x M, and

V = 0}; that is, E is the set of all points of M such
that

V = 0.
(iv) N: is the largest invariant set in E.
Then every solution starting in M approaches N as t .
20
8 Region of Attraction
Example 14 : Consider the system dened by
_
x
1
= 3x
2
x
2
= 5x
1
+ x
3
1
2x
2
.
We are interested in the stability of x = 0. Consider
V (x) = 12x
2
1
x
4
1
+ 6x
1
x
2
+ 6x
2
2
= 3(x
1
+ 2x
2
)
2
+ 9x
2
1
+ 3x
2
2
x
4
1
(8)

V (x) = 6x
2
2
30x
2
1
+ 6x
4
1
. (9)
According to Theorem 2, if V (x) > 0 and

V < 0 in D {0}, then x = 0 is
locally asymptotically stable.
Studying V and

V we conclude that dening D by
D = {x R
2
: 1.6 < x
1
< 1.6} (10)
we have that V (x) > 0 and

V < 0, x D {0}.
Question: Can we conclude that any solution starting in D converges to the
origin? No!
Plotting the trajectories as shown in we see that, for example, the trajectory
initiating at the point x
1
= 0, x
2
= 4 is quickly divergent from x = 0 even
though the point (0, 4) D.
The problem is this: D is not an invariant set and there are no guarantees
that trajectories starting in D will remain within D. Thus, once a trajec-
tory crosses the border |x
1
| =

5 there are no guarantees that

V (x) will be
negative. 2
We now study how to estimate the region of attraction.
Denition 12 : Let (x, t) be the trajectories of the systems (1) with initial
condition x at t = 0. The region of attraction to the equilibrium point x
e
,
denoted R
A
, is dened by
R
A
= {x D : (x, t) x
e
, as t }.
We nos estimate this region based on LaSalles Theorem.
21
Theorem 11 : Let V : D R be a continuous dierentiable function and
assume that x
e
is an equilibrium point and
(i) M D is a compact set containing x
e
, invariant with respect to the
solutions of (1).
(ii)

V is such that

V < 0 x =x
e
M.

V = 0 if x = x
e
.
Under these conditions we have that
M R
A
.
Example 15 :
x = kx +x
3
V =
1
2
x
2

V = x x
= kx
2
+ x
4
=
_
K x
2
_
x
2

V x
2
for k > x
2

V x
2
for k > 2V (t) {V (0) > V (t)}

V x
2
for k > 2V (0)

V x
2
for k > x
2
(0)

V 2V (t) for k > x

2
(0)
V (t) V (0) exp(2t) for K > x
2
(t)
1
2
x
2
(t)
1
2
x
2
(0) exp(2t)
|x(t)| |x(0)| exp(t) for K > |x(0)|
2
22
9 Analysis of Linear Time-Invariant Systems
Consider the autonomous linear time-invariant system given by
x = Ax, A R
nn
(11)
and let V () be dened as follows
V (x) = x
T
Px (12)
where P R
nn
is (i) symmetric and (ii) positive denite (iii) P is a constant.
Thus V () is positive denite. Also

V = x
T
Px + x
T
P x
= x
T
(A
T
P + PA)x
or

V = x
T
Qx (13)
PA + A
T
P = Q. (14)
Here the matrix Q is symmetric, since
Q
T
= (PA+ A
T
P)
T
= (A
T
P +AP) = Q
If Q is positive denite, then

V () is negative denite and the origin is (glob-
ally) asymptotically stable. To analyze the positive deniteness of the pair
of matrices (P, Q) we need two steps:
(i) Choose an arbitrary symmetric, positive denite matrix Q.
(ii) Find P that satises equation (14) and verify that it is positive denite.
Equation (14) appears very frequently in the literature and is called Algebraic
Lyapunov equation.
The procedure described above for the stability analysis based on the pair
(P, Q) depends on the existence of a unique solution of the Lyapunov equation
for a given matrix A. The following theorem guarantees the existence of such
a solution.
23
Theorem 12 : The eigenvalues
i
of a matrix A R
nn
satisfy e(
i
) < 0 if
and only if for any given symmetric positive denite matrix Q there exists a
unique positive denite symmetric matrix P satisfying the Lyapunov equation
(14)
Proof: Assume rst that given Q > 0, P > 0 satisfying (14). Thus
V = x
T
Px > 0 and

V = x
T
Qx < 0 and asymptotic stability follows from
Theorem 2.
For the converse assume that e(
i
) < 0 and given Q, dene P as follows:
P =
_

0
e
A
T
t
Qe
At
dt
which is symmetric. We claim that it is also positive denite. To see this,
assume the opposite: i.e. that x =0 such that x
T
Px = 0. But then
x
T
Px = 0
_

0
x
T
e
A
T
t
Qe
At
x dt = 0

_

0
y
T
Qy dt = 0 with y = e
At
x
y = e
At
x = 0 t 0
x = 0
since e
At
is nonsingular t. This contradicts the assumption. Thus P > 0.
We now show that P satises the Lyapunov equation
PA +A
T
P =
_

0
e
A
T
t
Qe
At
A dt +
_

0
A
T
e
A
T
t
Qe
At
dt
=
_

0
d
dt
(e
A
T
t
Qe
At
) dt
= e
A
T
t
Qe
At
|

0
= Q
To complete the proof, there remains to show that this P is unique. Suppose
that there is another solution

P =P. Then
(P

P)A+ A
T
(P

P) = 0
e
A
T
t
_
(P

P)A+ A
T
(P

P)
_
e
At
= 0

d
dt
_
e
A
T
t
(P

P)e
At
_
= 0
which implies that e
A
T
t
(P

P)e
At
is constant t. This can be the case if and
only if P

P = 0, or equivalently, P =

P. 2
24
10 Instability
Theorem 13 : (Chetaev) Consider the autonomous dynamical systems (1) and
assume that x = 0 is an equilibrium point. Let V : D R have the following
properties:
(i) V (0) = 0
(ii) x
0
R
n
, arbitrarily close to x = 0, such that V (x
0
) > 0
(iii)

V > 0 x U, where the set U is dened as follows:
U = {x D : x , and V (x) > 0}.
Under these conditions, x = 0 is unstable.
Example 16 : Consider again the system of Example 3.20 (textbook)
x
1
= x
2
+x
1
(
2
x
2
1
x
2
2
)
x
2
= x
1
+x
2
(
2
x
2
1
x
2
2
)
The origin of this system is an unstable equilibrium point. We now verify
this result using Chetaevs result. Let V (x) = 1/2(x
2
1
+ x
2
2
). Thus we have
that V (0) = 0, and moreover V (x) > 0 x R
2
=0, .e., V () is positive
denite. Also

V = (x
1
, x
2
)f(x)
= (x
2
1
+x
2
2
)(
2
x
2
1
x
2
2
).
Dening the set U by
U = {x R
2
: x , 0 < < }
we have that V (x) > 0 x U, x =0, and

V > 0 x U, x =0. Thus the
origin is unstable, by Chetaevs result.
2
25