Lecture notes 2011/2012

Dynamical systems, 3
Mihaela Cistelecan, UT Cluj-Napoca

These notes were prepared using:

1. Hale, J.K., Kocak, H. Dynamics and Bifurcations. Springer-Verlag,
1991.
2. Khalil, H. Nonlinear Systems. Prentice Hall. 1996.

1 The Stability of the equilibrium points. Lya-

punov stability
1.1 Motivation for studying
1) In physical systems, energy is a descriptor of the system dynam-
ical behavior
The pendulum mathematical model:
ẋ1 = x2
(1)
ẋ2 = − gl sin x1 − k
x
m 2

has two equilibrium points: (0, 0) and (π, 0). We study the stability of the
equilibrium point at the origin. The energy of the pendulum E(x) as the sum
of its potential and kinetic energies, with reference of the potential energy
chosen such that E(0) = 0, is:
Z x1
g 1 g 1
E(x) = sin y dy + x22 = (1 − cos x1 ) + x22 (2)
0 l 2 l 2
When friction is neglected (k = 0):
dE g g g
= ẋ1 sin x1 + x2 ẋ2 = x2 sin x1 − x2 sin x1 = 0 (3)
dt l l l

1
Since for k = 0 the derivative dE dt
= 0, it follows that the system is conserva-
tive, that is there is no dissipation of energy. Hence, E(x)=const. during the
motion of the system or, dE dt
= 0 along the trajectories of the system. Since
E(x) = c forms a closed contour around x = 0, for small c, we can arrive at
the conclusion that x = 0 is a stable equilibrium point.
When friction is accounted for (k > 0), the energy is as follows:

dE g k
= ẋ1 sin x1 + x2 ẋ2 = − x22 ≤ 0 (4)
dt l m
Therefore, the energy will dissipate during the motion of the system, that
is dE
dt
≤ 0 along the trajectories of the system. Due to friction, E cannot
remain constant indefinitely while the system is in motion. Hence, it keeps
decreasing until it eventually reaches zero, showing that the trajectory tends
to x = 0 as t tends to ∞.
Thus, by examining the derivative of E along the trajectories of the system,
it is possible to determine the stability of the equilibrium point. In 1892,
Lyapunov showed that certain other functions could be used instead of en-
ergy to determine the stability of an equilibrium point.

2) When liniarization does not suffice

Consider the system of nonlinear diff eq. :

ẋ1 = x2 + λx1 (x21 + x22 )

(5)
ẋ2 = −x1 + λx2 (x21 + x22 )

where λ is a scalar parameter. These equations are a nonlinear perturbation

of the harmonic oscillator but the perturbation does not affect the linear
terms. For all values of λ, the origin is an equilibrium point and the linearized
eq. at the origin is, of course, the harmonic oscillator:
· ¸
0 1
ẋ = x (6)
−1 0

The eigenvalues of this system are ±i, which have zero real parts. To analyze
the behaviour of the nonlinear system (5) we compute the derivative of the
square of the distance of a solution from the origin:
d 2
(x + x22 ) = 2λ(x21 + x22 )2 (7)
dt 1

2
If λ < 0, then kx(t)k2 approaches zero monotonically as t → +∞. Thus, the
equilibrium at the origin is asymptotically stable. However, if λ > 0, then
all solutions of (5) with initial data x(0) 6= 0 escape to ininity. Therefore, in
this case, the origin is unstable.

1.2 Stability - definitions

In this section we give considerations and theorems concerning n-dimensional
systems, not only second order systems.

Definition 1.3.1: An equilibrium point is stable if all solutions starting at

nearby points stay nearby; otherwise, it is unstable. The equilibrium point
is asymptotically stable if all solutions starting at nearby points not only stay
nearby, but also tend to the eq. point as time approaches infinity.

Definition 1.3.2: An equilibrium solution x̄ of the system ẋ = f (x) is

called stable if for any neighborhood U of x̄ there is a neighborhood V of x̄
contained in U such that for any solution ϕ, with ϕ(0) in V , ϕ(t) belongs to
U for any t ≥ 0. The equilibrium x̄ is called unstable if it is not stable. An
equilibrium solution x̄ of system ẋ = f (x) is called ”asymptotically stable”
if it is stable and if there is a neighborhood V0 contained in V such that any
solution ϕ, with ϕ(0) in V0 , tends to x̄ when t → ∞.

Linear systems
Theorem 1: If all the eigenvalues of the coefficient matrix A in the linear
system ẋ = Ax have negative real parts, then its equilibrium point x̄ = 0 is
asymptotically stable. Moreover, there are positive constants K and α such
that:

kx(t)k = keAt x(0)k ≤ Ke−αt kx(0)k (8)

for all t ≥ 0, x(0) ∈ R2 .
If one of the eigenvalues of the coefficient matrix A has positive real part,
then the equilibrium point x̄ = 0 is unstable.

Theorem 2: Let f be a continuous function. If all the eigenvalues of the

Jacobian matrix Df (x̄) have negative real parts, then the equilibrium point
x̄ of the differential equation ẋ = f (x) is asymptotically stable.

3
1.3 Lyapunov stability
When linearization fails we may apply the Lyapunov’s Method.This is a
method that helps us decide whether an equilibrium is stable, asymptotically
stable or unstable. It was given by Lyapunov in 1892. The idea is to find
a function whose properties determine the nature of equilibrium. Unfortu-
nately, there is no algorithm for finding the desired function. Lyapunov’s
method only ensures that if such a function exists, we can draw the right
conclusion. The right function should be guessed.

Consider the autonomous system ẋ = f (x) where f : D → Rn , D ⊂ Rn .

Let V : D → R be a continuously differentiable function defined in a domain
D ⊂ Rn that contains the origin. The derivative of V along the trajectories
ẋ = f (x), denoted by V̇ (x), is given by:
 
Xn Xn
£ ∂V ¤ f 1 (x)
∂V ∂V ∂V  ...  = ∂V f (x)
V̇ (x) = ẋi = fi (x) = ∂x ... ∂x
∂xi ∂xi 1 1
∂x
i=1 i=1 fn (x)
(9)
The derivative of V along the trajectories of a system is dependent on the
systems equations. Hence V̇ (x) will be different for different systems. If
φ(t, x) is the solution of ẋ = f (x) that starts at initial state x at time t = 0,
then:
d
V̇ (x) = V (φ(t, x))|t=0 (10)
dt
Therefore, if V̇ (x) is negative, V will decrease along the solution of
ẋ = f (x).

3.1. Lyapunov’s stability theorem: Let x = 0 be an equilibrium

point for ẋ = f (x) and D ⊂ Rn be domain containing x = 0. Let V : D → R
be a continuously differentiable function, such that:

V (0) = 0 and V (x) > 0 in D − {0}

(11)
V̇ (x) ≤ 0 in D

Then, x = 0 is stable. Moreover, if V̇ (x) < 0 in D − {0} then x = 0 is

asymptotically stable. The function V is called the Lyapunov function.
Definition 1.3.3: The surface V (x) = c, for some c > 0, is called a
Lyapunov surface or a level surface .

4
 The condition V̇ ≤ 0 implies that when a trajectory crosses a Lyapunov
surface V (x) = c, it moves inside the set Ωc = {x ∈ Rn |V (x) ≤ c} and can
never come out again. When V̇ < 0, the trajectory moves from one Lyapunov
surface to an inner Lyapunov surface with a smaller c. As c decreases, the
Lyapunov surface V (x) = c shrinks to the origin, showing that the trajectory
approaches the origin as time progresses.

Positive / Negative definite / semidefinite function

Definition 1.3.4: A function V (x) satisfying the conditions

• V (0) = 0
• V (x) > 0 for x 6= 0
is said to be positive definite . If it satisfies the weaker condition V (x) ≥ 0
for x 6= 0) it is said to be positive semidefinite .

Definition 1.3.5: A function V (x) is said to be negative definite or

negative semidefinite if −V (x) is positive definite or positive semidefinite,
respectively. If V (x) does not have a definite sign, it is said to be indefinite.

With this terminology, we can rephrase Lyapunov’s theorem , to say

that: ”the origin is stable if there is a continuously differentiable
positive definite function V (x) so that V̇ (x) is negative semidefinite,
and it is asymptotically stable if V̇ (x) is negative definite”.

A class of scalar functions V (x) for which sign definiteness

can be easily checked is the class of functions of the quadratic
form: n X
n
X
V (x) = xT P x = pij xi xj (12)
i=1 j=1

where P is a real symmetric matrix.

In this case, V (x) is positive definite (positive semidef.) if and only if all the
eigenvalues of P are positive (nonnegative), which is true if and only if all
the leading principal minors of P are positive (all principal minors of P are
nonnegative).
If V (x) = xT P x is positive definite (positive semidef.) we say that the
matrix P is positive def. (positive semidef.) and write P > 0 (P ≥ 0).

5
 Example: Linear system stabilityFor a linear autonomous system ẋ = Ax
we can choose the Lyapunov function as V (x) = xT P x where P = P T . Then,
V̇ (x) = xT (AT P + P A)x = −xT Qx. If we can find P > 0 and Q ≥ 0 so that
AT P + P A = −Q, then the system ẋ = Ax is stable. If, Q > 0 then the
system is asymptotically stable.
Example: Control law designFor a linear control system ẋ = Ax + Bu
we can use the Lyapunov function to design the control law u = F x (state
feedback control). We can choose the Lyapunov function as V (x) = xT P x
where P = P T . Then, V̇ (x) = xT (AT P +P A+F T B T P +P BF )x = −xT Qx.
If we can find the matrices F , P > 0 and Q ≥ 0 so that AT P +P A+F T B T P +
P BF = −Q, then the system ẋ = (A + BF )x is stable. If, Q > 0 then the
system is asymptotically stable.
Conclusion: Important
• Lyapunov’s theorem can be applied without solving the diff. eq. ẋ =
f (x)
• There is no systematic method for finding Lyapunov functions. In
some cases, there are natural Lyapunov functions like energy functions
in electrical or mechanical systems. In other cases, it is a matter of
trial and error.
• The Lyapunov’s Theorem conditions are only suffficient. Failure of a
Lyapunov function to satisfy the conditions for stability or asymptotic
stability does not mean that the equilibrium is not stable or asymp-
totically stable. It only means that such stability property cannot be
established by using this Lyapunov function... further investigation is
necessary.
The variable gradient method

Example: Pendulum,revisited From (4) we see that for the Lyapunov

function chosen as V (x) = E(x) the origin is stable but not asymptotically
stable. However, it is known that for the pendulum with k 6= 0 the origin
is asymptotically stable. We show in the following that by choosing the
Lyapunov function as follows the term x22 /2 is replaced by the more general
quadratic form xT P x/2, P > 0:
· ¸· ¸
1 p11 p12 x1 g
V (x) = [ x1 x2 ] + (1 − cos x1 ) (13)
2 p12 pp22 x2 l

6
 For the quadratic form to be positive definite, the elements of the matrix
P must satisfy
p11 > 0, p22 > 0, p11 p22 − p212 > 0 (14)
The derivative V̇ (x) is as follows:
g g k k
V̇ (x) = (1−p22 )x2 sin x1 − p12 x1 sin x1 +(p11 −p12 )x1 x2 +(p12 −p22 )x22
l l m m
(15)
We can choose p11 , p12 and p22 such that V̇ (x) < 0. Since the terms
x2 sinx1 and x1 x2 are sign indefinite, we will cancel them by taking p22 = 1
and p11 (k/m)p12 . With these choices, from (14), p12 must satisfy 0 < p12 <
(k/m). Let us take p12 = 0.5(k/m). Then, V (x) is given by:
1g k 1k 2
V̇ (x) = − x1 sinx1 − x (16)
2lm 2m 2
The term x1 sinx1 > 0 for all 0 < |x1 | < π. Taking D = {x ∈ R2 | |x| < π}
we see that V (x) is positive definite and V̇ (x) is negative definite over D.
Thus, we conclude that the origin is asymptotically stable.
The procedure that exploits the idea given in the above example (we
investigated the expression for V (x) and went back to choose the parameters
of V (x) so as to make V̇ (x) negative) is known as the variable gradient method
. We briefly describe this procedure in the following.
Let’s denote g(x) = ∇V = (∂V /∂x)T . The derivative V̇ (x) along the
trajectories of ẋ = f (x) is given by:
∂V
V̇ (x) = f (x) = g T (x)f (x) (17)
∂x
The idea now is to try to choose g(x) such that it would be the gradient
of a positive definite function V (x), and at the same time, V̇ (x) would be
negative definite.
See the literature for further information...
3.4. Domain / region of attraction
The domain / region of attraction (or basin, or region of asymptotic sta-
bility) is defined as follows.

Let φ(t; x) be the solution of ẋ = f (x) that starts at initial state x at

time t = 0. Then, the region of attraction is defined as the set of all points
x such that limt→∞ φ(t; x) = 0.

7
Finding the exact region of attraction analytically might be difficult or even
impossible. Still, a conservative approximation may be found using the Lya-
punov function, that is: if there is a Lyapunov function that satisfies the con-
ditions of asymptotic stability over a domain D, and if Ωc = {x ∈ Rn |V (x) ≤
c} is bounded and contained in D, then every trajectory starting in Ωc re-
mains in Ωc and approaches the origin as t → ∞. Thus, Ωc is an estimate
of the region of attraction. This estimate may be conservative - it may be
much smaller than the actual region of attraction.
Example: Consider the second-order dif. eq. :

z̈ + 2aż + z + z 3 = 0 (18)
where a is a constant satisfying 0 < a < 1, which is equivalent to the
system:

ẋ1 = x2
(19)
ẋ2 = −x1 − 2ax2 − x31
The origin is the only equilibrium point and
√ the eigenvalues of the liniariza-
tion at the origin are −a ± iβ, where β = 1 − a2 . Consequently, it follows
that the origin is asymptotically stable. Let us now try to estimate the basin
of attraction of the origin by using a Lyapunov function. With the intent of
determining a quadratic Lyapunov function, we first put the linear part of
the vector field into Real Jordan Normal Form. If we use the new variables
y defined by:
· ¸ · ¸
1 0 −1 1 β 0
y = P x, P = P =β (20)
−a β a 1
then the system becomes:

ẏ1 = −ay1 + βy2

(21)
ẏ2 = −βy1 − ay2 − β1 y13
We now take the Lyapunov function as the quadratic function:
1 2
V (y1 , y2 ) = (y + y22 ) (22)
2a 1
then a simple computation yields that:
1 3
V̇ (y1 , y2 ) = −(y12 + y22 ) − y y2 (23)
aβ 1

8
 Now, the main task is to determine the largest subset of R2 containing
the origin where −V̇ is positive definite. This is a difficult task, so let us
settle for the largest such disk. Observe that the level sets of the function
1 3
y12 + y22 are circles about the origin and the level sets of the function aβ y1 y2
are similar to hyperbolas. From the symmetry of −V̇ , it is evident that the
radius r0 of the largest circle inside which −V̇ is positive definite must satisfy:
1 4
r02 − r =0 (24)
aβ 0
√
that is, r0 = aβ.
Thus, every solution of (19) with initial value y(0) satisfying ky(0)k < r0
approaches the origin as t → +∞.

Brute approximation of the region of attraction

You may want to transform the nonlinear system into an equivalent form
ẋ = Ax + g(x) where ẋ = Ax is the system linearization around the chosen
equilibrium point and g(x) is the nonlinear part of the system. Note that on
the domain of attraction:

V (x) = xT P x ≤ c (25)

the following is true:

V̇ (x) = −xT Qx + 2xT P g(x) (26)

where the first term is negative when Q > 0 and the second term is in general
indefinite.
Since for a matrix M the following inequalities are true:

λmin (M )kxk2 ≤ xT M x ≤ λmax (M )kxk2

(27)
λmin (M )kxkkyk ≤ xT M y ≤ λmax (M )kxkkyk

where
kxk2 = xT x
,
and the function g(x) satisfies:

kg(x)k
→ 0 as kxk → 0 (28)
kxk

9
we may reason as follows:

∀γ > 0, ∃r > 0 : kg(x)k < γkxk ∀kxk < r

V̇ (x) < −xT Qx + 2γλmax (P )kxk2 , ∀kxk < r (29)

Since xT Qx ≥ λmin (Q)kxk2 we get:

V̇ (x) < −(λmin (Q) − 2γλmax (P ))kxk2 , ∀kxk < r (30)

Choosing γ < λmin (Q)/2λmax (P ) ensures that V̇ (x) is negative definite.

Global asymptotical stability

The stability type of an equilibrium point is a local property. However,

in this section we deal with the following question: under what conditions
will the region of attraction be the whole space Rn . This will be the case
if we can show that for any initial state x, the trajectory φ(t; x) approaches
the origin as t → ∞, no matter how large kxk is. If an asymptotically stable
equilibrium point at the origin has this property, it is said to be globally
asymptotically stable .
Barbashin-Krasovskii Theorem Let x = 0 be an equilibrium point
for ẋ = f (x). Let V : Rn → R be a continuously differentiable function such
that:

V (0) = 0 and V (x) > 0 ∀x 6= 0

kxk → ∞ ⇒ V (x) → ∞ (V is radially unbounded) (31)
V̇ (x) < 0, ∀x 6= 0
then x = 0 is globally asymptotically stable.
Example: Let’s suppose we are given the system:

ẋ1 = x2
(32)
ẋ2 = −h(x1 ) − ax2

with yh(y) > 0 for all y 6= 0. The Lyapunov function:

· ¸ Z x1
δ T ka2 ka
V (x) = x x+δ h(y)dy (33)
2 ka 1 0

is positive definite for all x ∈ R2 and radially unbounded. The derivative:

10
 V̇ (x) = −aδ(1 − k)x22 − aδkx1 h(x1 ) (34)
is negative definite for all x ∈ R2 since 0 < k < 1. Therefore, the origin
is globally asymptotically stable.
Remark: If the origin x = 0 is a GLOBALLY asymptotically stable equi-
librium point of a system, then it must be the unique equilibrium point of the
system. For if there was another equilibrium point x̄, the trajectory starting
at x̄ would remain at x̄ for all t ≥ 0; hence, it would not approach the origin
which contradicts the claim that the origin is globally asymptotically stable.
Therefore, global asymptotic stability is not studied for multiple equilibria
systems like the pendulum equation.

2 Home-works
1. For each of the following systems, use a quadratic Lyapunov function
candidate to show that the origin is asymptotically stable:

ẋ1 = −x1 + x22 ẋ1 = (x1 − x2 )(x21 + x22 − 1)

ẋ2 = −x2 ẋ2 = (x1 + x2 )(x21 + x22 − 1)
(35)
ẋ1 = −x1 + x21 x2 ẋ1 = −x1 − x2
ẋ2 = −x2 + x1 ẋ2 = x1 − x32
2. Using V (x) = x21 + x22 , study the stability of the origin of the system:

ẋ1 = x1 (k 2 − x21 − x22 ) + x2 (x21 + x22 + k 2 )

(36)
ẋ2 = −x1 (k 2 + x21 + x22 ) + x2 (k 2 − x21 − x22 )

when: (a) k = 0 and (b) k 6= 0.

3. Consider the second-order linear control system ẋ = Ax + Bu, A ∈
R2×2 , B ∈ R2×1 where the coefficient matrix A has (a) real positive eigenval-
ues and (b) complex eigenvalues with positive real part. Investigate, using
the Lyapunov function, if there exists a linear control law u = Kx ∈ R able
to stabilize the control system.
4. Discuss the stability of the following systems, given in polar coordi-

11
nates:
(ṙ = r(r − 4); θ̇ = 1)
(ṙ = ar2 + br3 ; θ̇ = 1)
(ṙ = r(λ − r); θ̇ = 1) (37)
(ṙ = ar2 ; θ̇ = 1)
(ṙ = ar(r − 1)(r − λ); θ̇ = 1)
where a, b are constants and λ(t) is time-variant.
5. Consider the following system of diff. eq.

ẋ1 = ax21 + bx1 x2 + cx22

(38)
ẋ2 = −x2 + lx21 + mx1 x2 + nx22

where a, b, c, l, m and n are constants. Show that:

1. if a 6= 0 then the origin is unstable;
2. if a = 0, bl < 0, then the origin is asymptotically stable;
3. if a = 0, bl > 0, then the origin is unstable;
4. if a = b = 0, cl2 6= 0, then the origin is unstable;
5. if a = b = c = 0, then the origin is stable but not asymptotically;
6. if a = l = 0, then the origin is stable but not asymptotically.
Show that the origin of the second-order system ÿ + y 3 = 0 is a stable
equilibrium point.
6. Show that the equilibrium point at the origin of the system:

ẋ1 = x1 x2 + ax31 + bx1 x2

(39)
ẋ2 = −x2 + cx21 + dx21 x2
1. is asymptotically stable if either a+c < 0, or a+c = 0 and cd+bc2 < 0;
2. is unstable if either a + c > 0, or a + c = 0 and cd + bc2 > 0.
7. Consider the second order system:

ẋ1 = x2
(40)
ẋ2 = −x1 − x32
What is the stability type of the equilibrium point at the origin ?
8. Discuss the stability properties of the origin for the system:

ẋ1 = x2 + x1 x2 + ax1 x22

(41)
ẋ2 = −x1 − x21 + x22
for various values of a.

12
 9. Discuss the stability and instability of the equilibrium points of the
system:
ẋ1 = −x2 + x32
(42)
ẋ2 = −x1 + x31

3 Lab-works
1. Draw the Lyapunov function for the systems given in the lecture notes, in
3-D as the function V (x1 , x2 ) and in 2-D as the function V (t).
2. Draw the phase portraits and the Lyapunov functions for the problems
given in the above section (Home-works).

4 Matlab
Study the Matlab functions lyap, contour, mesh.
Example: Given the nonlinear system
½
ẋ1 = x2
(43)
ẋ2 = −x1 + (x21 − 1)x2
compute the domain of attraction of its equilibrium point (0, 0).
Method 1: We choose the Lyapunov function as the quadratic function
V (x1 (t), x2 (t)) = x(t)T P x(t), P > 0, P = PT > 0
and compute its time-derivative as follows:
dV
V̇ = = 2(−xT Qx + p12 x31 x2 + p22 x21 x22 ) (44)
dt
where: · ¸
p12 0.5(p22 + p12 − p11 )
Q= (45)
0.5(p22 + p12 − p11 ) p22 − p12
Note that we should require p22 > 0 (P > 0). On the other hand, if
p12 = 0 it turns out that det(Q) = −0.52 (p22 − p11 )2 ≤ 0 which forbids
Q > 0.
We may compute the worst case scenario for V̇ < 0 as follows:
x21 x22 < (x21 + x22 )(x21 + X22 ) = kxk2 kxk2 = r4
p p (46)
|x31 x2 | < x21 |x1 ||x2 | < (x21 + x22 ) (x21 + x22 ) (x21 + x22 ) 4
= kxk = r 4

13
where r2 = kxk2 = x21 + x22 . Thus, we should find a conservative (small)
approximation of the domain of attraction from:

−xT Qx + (p12 + p22 )r4 < 0 (47)

or
−λmin (Q) + (p12 + p22 )r2 < 0 (48)
Therefore:
λmin (Q)
r2 < (49)
p12 + p22
is an approximation of the radius of the region of attraction.
It might be possible to find combinations of matrices P and Q which
shows a bigger domain of attractions than others. In order to find the best
matrices P and Q we must reason as follows:
By denoting the eigenvalues of the matrix Q: λ1 = λ and λ2 = λ + α,
λ, α > 0, from (45) we get:

2λ + α = p22
λ(λ + α) = p12 (p22 − p12 ) − 0.52 (p22 + p12 − p11 )2

or, equivalently:

λ2 − p22 λ + p12 (p22 − p12 ) − 0.52 (p22 + p12 − p11 )2 = 0 (50)

Since:
p22 1 √
λ1,2 = ± ∆
2 2
∆ = p222 − 4p12 (p22 − p12 ) + (p22 + p12 − p11 )2
we require a positive great minimum eigenvalue by ∆ = 0. In addition, from
the positivity of P and Q we should require: p22 > 0, p11 > 0, p11 p22 > p212 ,
p12 > 0, p22 > p12 ,

p12 (p22 − p12 ) − 0.52 (p22 + p12 − p11 )2 > 0

We may choose:
· ¸ · ¸
1 0 3 1
Q= ; P = (51)
0 1 1 2

14
 2

25 20
24

10
5

10
1.5 22
30

15
1 20
25

5
20 18
0.5

10
15 16

5
0

10
10 14

5 −0.5
12
0

5
−1 10
2

10
2 8

5
−1.5

20 25
15
10
0
0 6
−2
−2 −2 −2 −1 0 1 2

Figure 1: The Lyapunov function and its level sets

Using the following Matlab code:

————————
[x, y] = meshgrid(−2 : 0.1 : 2, −2 : 0.1 : 2);
z = 3 ∗ x. ∗ x + 2 ∗ y. ∗ y + 2 ∗ x. ∗ y;
subplot(1, 2, 1), mesh(x, y, z)
subplot(1, 2, 2), [c, h] = contour(x, y, z); clabel(c, h), colorbar
—————————-
we show the Lyapunov function and its level sets.
Remark: The Matlab function contour shows the level sets of the Lya-
punov function.
With Q and P chosen as in (51) we get:

r2 < 1/3 → r < 0.5774

However, if we use the following Matlab code:

——————————————————————-
[x, y] = meshgrid(−.9 : 0.1 : .9, −.9 : 0.1 : .9)
z = (x. ∗ x + y. ∗ y) − 2 ∗ x. ∗ x. ∗ y. ∗ y − 1 ∗ x. ∗ x. ∗ x. ∗ y;
subplot(2, 2, 1), mesh(x, y, z)

15
 0.6 0.
0.4

0
2 2 0.8
0.5

1
0.6 0.8
0.2 0.6

0.4
0

0.8 0.6
0 0.4

0.4
0.2
−2
−0.5 0 0.2
1 .2 0
1 0.4

1
0 0 0.6

0
−1 −1 −0.5 0 0.5

0.6
0.5

0.70.8
0.4

0.3
1 0.3 0.7
0.5 0.2
0.6
0.1

0.06.5 .4
0
0.4 0.05.6
0.5 0.5
0
0.4

0.3
0.1
0 0.3

0.3
1 −0.5 0.2

0..87
1 0.2

0.3
0.4

0
0 0 0.5
0.6
−1 −1 −0.5 0 0.5

Figure 2: -V̇ for x1 , x2 ∈ [−0.9, 0.9] and x1 , x2 ∈ [−0.8, 0.8]

subplot(2, 2, 2), [c, h] = contour(x, y, z); clabel(c, h), colorbar

[x, y] = meshgrid(−.8 : 0.1 : .8, −.8 : 0.1 : .8)
z = (x. ∗ x + y. ∗ y) − 2 ∗ x. ∗ x. ∗ y. ∗ y − 1 ∗ x. ∗ x. ∗ x. ∗ y;
subplot(2, 2, 3), mesh(x, y, z)
subplot(2, 2, 4), [c, h] = contour(x, y, z); clabel(c, h), colorbar
—————————————————————–
we get Fig. 2. Note that for x1 , x2 ∈ [−0.8, 0.8] V̇ < 0 (since −V̇ > 0)
but for x1 , x2 ∈ [−0.9,√0.9] this is not true anymore. √ Thus, the stability is
guaranteed for r < 0.8 2 = 1.1314 but not for r < 0.9 2 = 1.2728.

Method 2: The system (43) may be also written as follows:

· ¸ · ¸
0 1 0
ẋ = Ax + g(x) = x+ (52)
−1 −1 x21 x2

where
∂f
A= |(0,0)
∂x

16
 8

−2

−4

−6

−8
−4 −3 −2 −1 0 1 2 3 4

Figure 3: Phase portrait

is the Jacobian of the system in the equilibrium point (0, 0). Note that:
q q
kg(x1 , x2 )k = 0 + (x1 x2 ) < (x1 + x2 ) x21 + x22 = kxk3
2 2 2 2

Using the Matlab lyap function for Q chosen as the identity matrix:
>> P = lyap([0, 1; −1, −1]0 , [1, 0; 0, 1])
we get:
P =[ 1.5000 , 0.5000 ; 0.5000 , 1.0000 ]
>> eig(P )
ans =
1.8090, 0.6910
Thus, using (30):

λmin (Q) 1
γ = r2 < =
2 ∗ λmax (P ) 2 ∗ 1.8090
r < 0.5257
We show a phase portrait in Fig 3.

17
 35

30

25

20
V(t)

15

10

0
0 2 4 6 8 10
t

Figure 4: The time history of the energy for different initial conditions

We also show the time history of V (t) for P = [3 1; 1 2] in Figure 4 for

different initial conditions inside and outside the attraction domain. Note
that when the initial condition is inside the domain of attraction the energy
decreases monotonically.
In Figure 5 we show the difference between the time history of V (t) for
P = [3 1; 1 2] and P = [1 0; 0 1] for an initial condition contained in the
domain of attraction.
The Matlab code used was based on the sequence:
[t,x]=ode23(’vpol11’,[0,10], [1,1]), P=[3 1; 1 2]; V=[];
for i=1:length(t)
V=[V x(i,:) * P * x(i,:)’];
end plot(t,V,’b’), hold on

18
 4.5

3.5

2.5
V(t)

1.5

0.5

0
0 2 4 6 8 10
t

Figure 5: The time history of the energy versus the time history of the state
vector norm for the same initial condition

19