Example 1. Simple strong Lyapunov function.

Example 2. Stability by Linearization

For the following system of equations …nd all equilibrium points and inves-
tigate
( their stability and their type by linearization.
x0 = ln(2 y 2 )
y 0 = exp(x) exp(y)

1. Solution. There are two equilibrium points: x1 = (1; 1)and x2 = ( 1; 1).

" #
0 2 yy2 +2
The Jacobian of the right hand side is: . Its values in
ex ey
" # " #
0 2 0 2
x1 and x2 are A1 = , and A2 = . The eigen-
e e 1=e 1=e
p p
values to A1 are 12 e 1
2 e
2 8e;and 12 e2 8e 1
2 e that are conjugate
complex numbers with negative real parts. Therefore we observe stable spi-
ral around the equilibrium point x1 . The eigenvalues to A2 are , eigenvalues:
1 1
p 1 1 1
p 1
e 2 8e + 1 2 ; e 2 8e + 1 2 , one postive and one negative. There-
fore we x2 is a saddle point and is unstable.
Example 3. (
x0 = 2y
Consider the following system of ODEs: :
y0 = x (1 x2 )y
Show the asymptotic stability of the equilibrium point in the origin and …nd
it’s domain of attraction.
Solution.
We try the test function V (x; y) = x2 + 2y 2 that leads to cancellation of
mixed terms in the directional derivative Vf along trajectories:

1
 Vf (x; y) = 4xy 4xy 4y 2 (1 x2 ) = 4y 2 (1 x2 ) that is not positive
for jxj 1. Therefore the origin is a stable stationary point. Checking the
behavior of the system on the set of zeroes to Vf (x; y) inside the stripe jxj < 1
1
we consider (Vf ) (0) = f(x; y) : y = 0; jxj < 1g. On this set y 0 = x and the
1
only invariant set in (Vf ) (0) is the origin. The LaSalles invariance principle
implies that the origin is asymptotically stable and the domain of attraction is
the largest set bounded by a level set of V (x; y) = x2 + 2y 2 inside the stripe
jxj 1. The largest such set will be the interior of the ellipse x2 + 2y 2 = C
such that is touches the lines x = 1. Taking points ( 1; 0) we conclude that
1 = C. and the boundary of the domain of attraction is the ellipse x2 + 2y 2 = 1
p
with halfs of axes 1 and 0:5 :
y
0.625
0.5
0.375
0.25
0.125
0
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
-0.125
x
-0.25
-0.375
-0.5
-0.625

2
 How to …nd a Lyapunov function?

If the right hand side of the equation is a higher degree polynomial, then
it is often convenient to …nd to …nd Lyapunov’s function in a systematical way
in the form of polynomial with unknown coe¢ cients and unknown even degrees
like 2m.
Consider the system

x0 = 3x3 y
y0 = x5 2y 3

Try a test function V (x; y) = ax2m + by 2n , a; b > 0.

Vf (x; y) = rV f (x; y) =
= a2m(x)2m 1
3x3 y + b2n(y)2n 1
x5 2y 3
= 6amx2m+2 2ma(x)2m 1
y + 2nby 2n 1 5
x 4nby 2n+1
| {z } | {z } | {z } | {z }
good<0 bad indef inite bad indef inite good<0

We choose …rst powers m and n so that inde…nit terms would have same
powers of x and y.

2m 1 = 5; =) m = 3
2n 1 = 1; =) n = 1

Then Vf (x; y) = 18ax8 6x5 y + 2bx5 y 4nby 4 :We choose a = 1 and b = 3

to cancel inde…nite terms. Then

V (x; y) = x6 + 3y 2
Vf (x; y) = 18x8 12y 4 < 0; (x; y) 6= (0; 0)

Therefore V is a strong Lyapunov’s function in the whole plane and the equi-
librium is a globally asymptotically stable equilibrium point, because V (x; y) =
x6 + 3y 2 ! 1 as k(x; y)k ! 1.
Example 4. Investigate stability of the equilibrium point in the origin.

3
 x0 = y x3
y0 = x5

We try our simplest choice of the Lyapunov function: V (x; y) = x2 + y 2 and

arrive to

Vf (x; y) = 2xy 2x4 + 2yx5

It does not work because the expression Vf (x; y) includes two inde…nite terms:
2xy and 2yx5 that change sign around the origin. We try a more ‡exible expres-
sion by looking on particular expressions in the right hand side of the equation:
V (x; y) = x6 + y 2 where @V =@x = 6x5 with the same power of x as in the
equation, and the parameter that can be adjusted later. V is a positive def-
inite function: V (0) = 0 and V (z) > 0 for z 6= 0.The level sets to V look as
‡attened in y - direction ellipses. The curve x6 + 3y 2 = 0:5 is depicted:

y 0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
-0.8 -0.6 -0.4 -0.2 -0.05 0 0.2 0.4 0.6 0.8
-0.1
-0.15 x
-0.2
-0.25
-0.3
-0.35
-0.4

Vf (x; y) = 6x5 ( y x3 ) + 2 yx5 = 6x5 y + 2 x5 y 6x8

We get again two inde…nite terms, but they are proportional and the choice

4
 = 3 cancels them:
Vf (x; y) = 6x8 0

Therefore the origin is a stable equilibrium point. Vf (x; y) = 0 on the whole

y axis that in our "general" theory is denoted by Vf 1 (0).We check invariant
sets of the system on the set Vf 1 (0): We observe that x0 = x3 (only this fact is
important) and y 0 = 0 (it does not matter for Vf 1 (0) that is y axis). Therefore
f0g is the only invariant set on the y - axis. Trajectories starting on the y -
axis go across it in all points except f0g. The LaSalle’s invariance principle
implies that all trajectories approach f0g as t tends to in…nity and the origin is
asymptotically stable.
The test function V (z) ! 1 as kzk ! 1. It implies that the whole plain is
a region or domain of attraction for the equilibrium point in the origin.

How to …nd a strong Lyapunov’s function?

Example 4.
It is theoretically possible to …nd a strong Lyapunov function for the same
system as in the Example 3.
Looking on the previous week Lyapunovs function x6 + 3y 2 we see that it’s
"weekness" followed from the fact that both level sets of V and velocities of the
system were orthogonal to the y - axis. It implied that Vf (z) = 0 on the y -
axis. To go around this problem a strong Lyapunov function must have level
sets that deviate slightly from the normal to the y - axis. Adding a relatively
small inde…nite term xy 3 to the function x6 + 3y 2 we get this e¤ect. A level set
corresponding x6 + xy 3 + 3y 2 = 0:7 of this new Lyapunovs function looks as a
slightly rotated version of level sets for the previous (weak) Lyapunovs function.
Why like that ? Take a simpler example with an ellipse curve x2 + 2y 2 = 1
and another that is x2 + xy + 2y 2 = 1 " #
1 0:5
This quadratic form is positive de…nite: the matrix is :A quadratic
0:5 2
form xT Ax = Q(x) is positive de…nite if and only if det A > 0 and all subma-
trices Ai from the upper left corner have positive determinants: det Ai > 0:
Level sets of the positive de…nite quadratic form with mixed tems like x2 +
xy + 2y 2 are ellipses with symmetry axes (that are orthogonal eigenvectors to
A) and are rotated with respect to coordinate axes:

5
 y 0.75

0.625

0.5

0.375

0.25

0.125
0
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
-0.125
x
-0.25

-0.375

-0.5

-0.625

-0.75

We try to introduce the test function V (x; y) = x6 + xy 3 + 3y 2 with an

inde…nite mixed term xy 3 added, that would similarly with the ellipses, give
slightly rotated level sets so that trajectories would cross them strictly inside
on the y - axis:

y 0.5

0.4

0.3

0.2

0.1
0
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
-0.1
x
-0.2

-0.3

-0.4

-0.5

We claim that the test function V (x; y) = x6 + xy 3 + 3y 2 is positive de…nite

and is a strong Lyapunovs function namely that Vf (x; y) < 0 for (x; y) 6= (0; 0).
Because of the geometry of the vector …eld f of our equation z 0 = f (z)
velocities on the y axis cross such level sets strictly towards inside, implying the
desired strict inequality Vf (z) < 0; z 6= 0 on the y axis. We need to check that

6
V (x; y) = x6 + xy 3 + 3y 2 is positive de…nite (it is not trivial) and to show that
Vf (z) < 0; z 6= 0 for all z 2 R2 (it requires some non-trivial analysis).
A very useful inequality in analysis is Young’s inequality
Lemma. If a; b 0, then

ap bq
ab +
p q
for every pair of numbers p; q 2 (1; 1) satisfying the conjugacy relation.

1 1
+ =1
p q

The simplest example of Young’s inequality:

1 2
ab a + y2
2
We show that the test function V (x; y) = x6 + xy 3 + 3y 2 is positive de…nite
in a domain around the origin.

We calculate Vf = V for the system from the Example 3:

x0 = y x3
y0 = x5

7
 Example 5.

Consider the Lienard equation: x00 + x0 + g(x) = 0; and investigate stability

of the equilibrium in the origin. The second order equation can be rewritten as
a system z 0 = f (z):

x0 = y
y0 = g(x) y

where g satis…es the following hypothesis: g is continuously di¤erentialble for

jxj < k for some k > 0, xg(x) > 0; x 6= 0:
Physically this equation is a Newton equation for a non-linear spring. For
example if g(x) = sin(x) it describes a pendulum with friction where air resis-
tance is proportional to velocity.

8
 A Lyapunov function is naturally to choose as a total energy of the system:
2 Z x
(y)
V (x; y) = + g(s)ds
2 0

Indeed it is positive de…nite in the region = f(x; y) : jxj < k g because

g(s)s > 0 in according to given conditions. The directional derivative of V
along f is
2
Vf (x; y) = y ( g(x) y) + g(x)y = (y)

V is a Lyapunov’s function, but not strong because Vf (x; y) is negative

de…nite in . Vf 1 (0) is the whole x - axis. Checking values of f on Vf 1 (0) we
observe that trajectories of the system are orthogonal to Vf 1 (0) in all points
on Vf 1 (0) except the origin. It implies that f0g is the only invariant set on
Vf 1 (0) that attracts all trajectorie starting in a small neighborhood of the
origin. Therefore the origin is asymptotically stable.
Our next problem is to …nd a possibly large domain or region of attraction
for the equilibrium point.If we …nd a closed level set for V in , it will be a
boundary for a domain of attraction. It will might not be the largest possible
and depends on a clever choice of Lyapunov’s function V .
2
We cannot solve this problem for a general expression V (x; y) = (y) 2 +
Rx
0
g(s)ds.
Example 6.
Choose a particular g(x) = x + x2 in the previous example.

x0 = y
0
y = (x + x2 ) y

Observe that the system has two equilibrium points: ( 1; 0) and (0; 0)
" #
0 1
Linearization gives Jacoby matrix A(x; y) = ; A( 1; 0) =
1 2x 1
" # " #
0 1 0 1
Observe that det = 0 1 = 1 < 0 it implies by Grob-
1 1 1 1
man - Hartman " that ( 1; #0) is a "saddle point.
# " #
0 1 0 1 0 1
A(0; 0) = , det = 1 > 0, trace =
1 1 1 1 1 1
1 < 0;

9
 2
(traceA(0; 0)) =4 = 1=4 < 1 = det A(0; 0). It imples that the origin is stable
spirals.
y 6

0
-1.25 0 1.25

g(x) = x + x2

We can …nd an explicit expression for the Lyapunov’s function V (x; y) =

(y)2 Rx
2 + 0 g(s)ds.

2 3 2
(x) (x) (y)
V (x; y) = + +
2 3 2
n o
2 2 2 3
This function is positive de…nite on the set = (y) > (x) 3 (x)
The level set 12 y 2 + 12 x2 + 13 x3 = 16 is depicted by the red line.The level set
1 2 1 2 1 3
2 y + 2 x + 3 x = 0 is depicted by the blue line. We will investigate them
analytically a bit later.

10
 2 2 2 2
Vf (x; y) = rV (x; y) f = xy + (x) y (y) xy (x) y = (y) 0 valif
2
in the whole plane R :
We check which invariant sets are contained in Vf 1 (0) on that is a part
of x - axis f(x; 0) : x > 3=2g that is a thick black line on the picture above.
Notice that Vf 1 (0) on contains two equilibrium points ( 1; 0) and (0; 0)
and they both are invariant sets. We like to …nd a largest domain 1
bounded by a part of a level set of V such that 1 does not include the point
( 1; 0). Then 1 contains only one invariant set that is the origin (0; 0). This
set 1 is the domain of attraction for the asymptotically stable equilibrium in
(0; 0).
Such largest level set of V must go through the second equilibrium point
( 1; 0) and it’s value there is V (x; y) = V ( 1; 0) = 1=6. The domain of at-
2
traction is the egg - shaped domain bounded by the closed curve (y) =
2 3
1=3 (x) + 32 (x) or as a union of explicit two branches:

s
2 2 3
y= 1=3 (x) + (x)
3

11
It is a part of the red level set on the picture. To see that this curve is closed e
consider derivative of the function
d 2 3
dx 1=3 (x) + 23 (x) = 2x 2x2 = ( 2) x (x + 1). It implies that
the functions has a maximum in x = 0, and minimum at x = 1. V (x) has zero
in x = 1 and another zero in x = 1=2:
2 3 2 3
1=3 (x) + 32 (x) = 1=3 (1=2) + 23 (1=2) = 1=3 (1=4) + 13 (1=4) =
x=1=2
1=3 1=3 = 0;

One can try to …nd an even larger region of attraction for the equilibrium
point in the origin. It cannot include the equilibrium in ( 1; 0) because it is
p
unstable (a saddle point). We can extend 1 to a rectangle [1; 0] 0; 3=3 in
the second quadrant by checking signs of x0 and y 0 on it’s left and upper sides.
Actual region of attraction is even a bit larger as one can see on the phase
portrait

12
 Example 7. Exercise 5.13 from L.R.
Investigate stability of the equilibrium point in the origin and …nd a possible
domain of attraction for the following system.

x01 = x2 (1 + x1 x2 )
x02 = 2x1

We try choose the Lyapunov function V as

V (x1 ; z2 ) = 2x21 + x22

We could try …rst a function V (x1 ; x2 ) = ax21 + x22 , check Vf and then decide
which value a suites best.

13
 Vf (x1 ; x2 ) = 2ax1 x2 (1 + x1 x2 ) + 2x2 2x1
= 4x1 x2 2ax1 x2 2ax21 x22 = 2ax21 x22 0
f or a = 2

We conclude that the equilibriom 0 is stable. Vf (x1 ; x2 ) = 2ax21 x22 = 0 on

both coordinate axes. We chech which invariant sets are contained in Vf 1 (0).
If x1 = 0, then x01 = x2 , x02 = 0. Therefore only f0g is an invariant set on
the x2 axis.
If x2 = 0, then x01 = 0, x02 = 2x1 . Therefore only f0g is an invariant set on
the x1 axis.
Trajectories '(t; ) starting inside ellipses V (x1 ; z2 ) = 2x21 + x22 = C >0 are
contained inside these ellipses. It implies that their positive orbits O+ ( ) are
bounded and have compact closure in R2 :
It implies according to the LaSalle’s theorem that all these solutions '(t; )
approach the maximal invariant set in Vf 1 (0) that in our particular case con-
sists of one point (0; 0). Therefore the equilibrium point in the origin is asymp-
totically stable. It is also globally stable because the Lyapunov function V (x)
tends to in…nity as kxk ! 1 , making that arbitrary large elliptic discs from
the family 2x21 + x22 < C are regions of attraction.

Example 8.
(
x0 = x 2y + xy 2
Consider the following system of ODE: .
y 0 = 3x 3y + y 3

1. Show asymptotic stability of the equilibrium point in the origin and …nd
the region of attraction for that.
Hint: applying Lyapunovs theorem, use the elementary Young’s inequal-
ity 2xy x2 + y 2 to estimate inde…nite terms with xy: (4p)
1
Solution. Choose a test function V (x; y) = 2 x2 + y 2
Vf = x( x 2y + xy 2 ) + y 3x 3y + y 3 = xy x2 3y 2 + y 4 + x2 y 2
= x2 1 y2 y2 3 y 2 + xy 0 ?????

14
 We apply the inequality 2xy x2 + y 2 to the last term and collecting
terms with x2 and y 2 arrive to the estimate
Vf x2 0:5 y2 y 2 2:5 y2
p
It implies that Vf < 0 for (x; y) 6= (0; 0) and jyj < 1= 2.Therefore the
origin is asymptotically stable.
The attracting region is bounded by the largest levle set of V - a circle
p
having the center in the origin that …ts to the domain jyj < 1= 2, namely
x2 + y 2 < 1=2.
Another more clever choice of a test function is V (x; y) = 3x2 + 2y 2 :
Vf = 6x( x 2y+xy 2 )+4y(3x 3y+y 3 ) = 4y 4 12y 2 6x2 +6x2 y 2 = 4y 2
3 y2 6x2 1 y 2 < 0
for jyj < 1, therefore the ellipse 3x2 + 2y 2 < 2 is a domain of attraction
for the asymptotically stable equilibrium in the origin.
One can also observe the asymptotic stability of the origin by linearization
with variational matrix
" #
1 2
A= , with characteristic polynomial: 2 + 4 + 9 = 0, and
3 3
p p
calculating eigenvalues: i 5 2; i 5 2 with Re < 0. But linearization
gives no information about the set of attraction.

Example 9 on instability
Consider the following system of ODEs. Prove the instability of the equilib-
rium( point in the origin, of the following system
x0 = x5 + y 3
(4p)
y 0 = x3 y 5
using the test function V (x; y) = x4 y 4 and Lyapunov’s instability theorem.
Solution. " #
x5 + y 3
Denoting f (x; y) = , consider how V (x; y) = x4 y 4 changes
x3 y 5
" # " #
x5 + y 3 4x3
along trajectories of the system. f (x; y) rV (x; y) = =
x3 y 5 4y 3
x5 4x3 + y 3 4x3 x3 4y 3 + y 5 4y 3 = x5 4x3 + y 5 4y 3 = 4(x8 + y 8 ) > 0:
Point out that the function V (x; y) = x4 y 4 is positive along the line
y = x=2, x > 0 arbitrarily close to the origin. It implies according to the
instability theorem, that the origin is an unstable equilibrium.

15