TTK4150 Nonlinear Control Systems

Department of Engineering Cybernetics

Norwegian University of Science and Technology
Fall 2020 - Solution to Assignment 2

1. (a) By using the fact that the derivatives are zero in an equilibrium point, the following
equations must be true

0 = x∗2
f3 f1 ∗ d
0 = − x∗3 − x1 − x∗2 − g
m 1 m m
f3 f1 ∗
= − x∗3 − x −g
m 1 m 1
Inserting (0, 0) into the above equations leads to an illegal expression since g is not
zero but 9.81m/s2 . Therefore (0, 0) is not an equilibrium point.
With u = u0 we have the equations

0 = x∗2 (1)
f3 f1 ∗ u0
0 = − x∗3 1 − x1 − g + (2)
m m m
Inserting (x∗1 , x∗2 ) = (x1d , 0) into 2 gives

0 = −f3 x31d − f1 x1d − mg + u0

thus
u0 = f3 x31d + f1 x1d + mg (3)
(b) In the original system equations we insert x1 = x̃1 + x1d , x̃2 = x2 and u = u0 + ũ.
We have

x̃˙ 1 = x2 = x̃2
f3 f1 d u
x̃˙ 2 = − x31 − x1 − x2 − g +
m m m m
f3 3 f1 d (f3 x31d + f1 x1d + mg + ũ)
= − (x̃1 + x1d ) − (x̃1 + x1d ) − x̃2 − g +
m m m m
The resulting system equations are

x̃˙ 1 = x̃2 (4)

mx̃˙ 2 = −f3 (x̃1 + x1d )3 − x31d − f1 x̃1 − dx̃2 + ũ
 
(5)

In the equilibrium point for ũ = 0 we have

0 = x̃∗2
0 = −f3 (x̃∗1 + x1d )3 − x31d − f1 x̃∗1 − dx̃∗2
 

= −f3 (x̃∗1 + x1d )3 − x31d − f1 x̃∗1

 

The equilibrium point is now in the origin.

1
 (c) The Jacobian is calculated for (4)–(5).
" # " #
∂f1 ∂f1  0 1
A = ∂∂fx̃21 ∂∂fx̃22  = (6)

3f3 x21d +f1
∂ x̃1 ∂ x̃2
 − m
− md
x=(0,0)

(Note that you may instead calculate the Jacobian for the original system, as long
as you use the correct equilibrium point for this system.)
To find out whether A is Hurwitz or not, the eigenvalues of the matrix must be
calculated
 " # " #
−1

λ 0 0 1 λ
λI − A = − 3f x21d +f1 = 3f3 x21d +f1 (7)
0 λ − 3m − md m
λ + md

d 3f3 x21d + f1
|λI − A| = λ(λ + )+ (8)
m m
2
d 3f 3 1d + f1
x
= λ2 + λ + (9)
m m
The eigenvalues of A are thus given as
 s  
2 2
1 d d 4 (3f3 x1d + f1 ) 
λ = − ± − (10)
2 m m m

Since f1 , f3 , x21d , m > 0

s 2
d d 4 (3f3 x21d + f1 )
> − (11)
m m m

and the eigenvalues will always lie in the left half plane which means that A is
Hurwitz. This means that (0, 0) of (4)–(5) is locally asymptotically stable.

2. Remark: the answer to this exercise may vary a lot depending on the chosen parameters
of Lyapunov function and the domain D. One possible solution set is presented below.

(a) The scalar system is given by ẋ = −x3 with equilibrium point at the origin. Suppose
V (x) = px2 where p > 0. Then V is positive definite. Taking the derivative along
the trajectory we have V̇ = 2pxẋ = 2px (−x3 ) = −2px4 which is negative definite.
Hence the origin is asymptotically stable. Further V is radially unbounded which
implies that the origin is globally asymptotically stable.
(b) The system is given by

ẋ1 = −x1 − x2
ẋ2 = x1 − x32

where it can be seen that the equilibrium point is given by

(x∗1 , x∗2 ) = (0, 0)

2
 A general quadratic Lyapunov function candidate is given by
1
V (x) = xT P x, P = PT
2
which is positive definite if and only if all the leading principal minors of P are
positive, that is

p11 > 0
p11 p22 − p212 > 0

(and it follows that p22 > 0). The derivative of the Lyapunov function candidate
along the trajectories of the system is given by

V̇ (x) = ẋT P x
 T   
−x1 − x2 p11 p12 x1
=
x1 − x32 p12 p22 x2
= −p11 x1 x2 − p12 x1 x2 + p22 x1 x2 − p11 x21 + p12 x21 − p12 x22 − p22 x42 − p12 x1 x32
= − (p11 − p12 ) x21 − (p11 + p12 − p22 ) x1 x2 − p12 x22 − p22 x42 − p12 x1 x32

In order to eliminate the undesirable terms, pi is chosen according to

p11 + p12 − p22 = 0

p12 = 0
⇒ p11 = p22

which fulfill the requirements imposed in order to guarantee V (x) to be positive

definite. The derivative of V (x) with respect to time is

V̇ (x) = −p11 x21 − p11 x42

< 0 ∀x ∈ R2 − {0}

Since V (x) is radially unbounded it follows that the origin is globally asymptotically
stable.
(c) The system is given by

ẋ1 = −x1 + x22

ẋ2 = −x2

where it can be seen that the equilibrium point is (x∗1 , x∗2 ) = (0, 0). A general
quadratic Lyapunov function candidate is given by
1
V (x) = xT P x, P = PT
2
which is positive definite if and only if all the leading principal minors of P are
positive

p11 > 0
p11 p22 − p212 > 0

3
 (and it follows that p22 > 0). The derivative of the Lyapunov function candidate
along the trajectories of the system is given by

V̇ (x) = ẋT P x
 T   
−x1 + x22 p11 p12 x1
=
−x2 p12 p22 x2
  T  
−x1 + x22 p11 x1 + p12 x2
=
−x2 p12 x1 + p22 x2
2


= −x1 + x2 (p11 x1 + p12 x2 ) − x2 (p12 x1 + p22 x2 )
= p12 x32 − p11 x21 − p22 x22 − 2p12 x1 x2 + p11 x1 x22

By choosing p12 = 0, the term x32 and x1 x2 vanishes and the derivative is rewritten
as

V̇ (x) = −p11 x21 − p22 x22 + p11 x1 x22

= −p11 x21 − (p22 − p11 x1 ) x22
 
2 p22 p22
= −p11 x1 − p11 − x1 x22 < 0, ∀ > x1
p11 p11
n o
n p22
By taking D = x ∈ R |x1 < p11 , where pp11 22
may be chosen arbitrarily large, it
follows that the equilibrium point is asymptotically stable.
Intuitively, we can argue that it is likely that the equilibrium is globally asymp-
totically stable by looking at the original system. The second state x2 will approach
0 independently of x1 . Eventually the term −x1 will dominate in ẋ1 , and x1 will also
approach 0.
Showing this formally will be revisited in a future assignment.
(d) The system is given by

ẋ1 = (x1 − x2 ) x21 + x22 − 1

ẋ2 = (x1 + x2 ) x21 + x22 − 1

where it can be seen that the equilibrium points are given by

(x∗1 , x∗2 ) = (0, 0)

and the set

x∗2 ∗2
1 + x2 = 1

This implies that the origin can not be globally asymptotically stable, since any
initial condition in {x ∈ R2 |x21 + x22 = 1} implies that the solution stays in that set.
A general quadratic Lyapunov function candidate is given by
1
V (x) = xT P x, P = PT
2
which is positive definite if and only if all leading principal minors of P are positive,
that is

p11 > 0
p11 p22 − p212 > 0

4
 (and it follows that p22 > 0). The derivative of the Lyapunov function candidate
along the trajectories of the system is given by
V̇ (x) = ẋT P x
 T   
(x1 − x2 ) (x21 + x22 − 1) p11 p12 x1
=
(x1 + x2 ) (x21 + x22 − 1) p12 p22 x2
2x1 x2 p12 − x1 x2 p11 + x1 x2 p22 + x21 p11 + x21 p12 − x22 p12 + x22 p22 x21 + x22 − 1



=
p12 − 12 p11 + 12 p22
 
T p11 + p12
x x21 + x22 − 1


= x 1 1
p12 − 2 p11 + 2 p22 p22 − p12
= xT Qx x21 + x22 − 1

By choosing Q such that xT Qx > 0 ∀x 6= 0 and taking D = {x ∈ R2 |x21 + x22 < 1},
it can be seen that
V̇ (x) < 0 ∀x ∈ D
Choosing p12 = 0, the matrix P is positive definite if and only if
p11 > 0
p22 > 0
and the matrix Q is positive definite if and only if
p11 > 0
p22 > 0
Thus by choosing p11 , p22 > 0 and p12 = 0 we have shown that the origin of the
system is asymptotically stable.

3. (a) The phase portrait can be seen in Figure 1.

Please purchase Metafile/EMF/WMF/RTF To PDF Converter on http://www.verypdf.com to remove this watermark.

x1 ' = x12 - x22

x2 ' = 2 x1 x2

0
x2

-1

-2

-3

-4

-4 -3 -2 -1 0 1 2 3 4
x1

Figure 1: Phase portrait

Please purchase Metafile/EMF/WMF/RTF To PDF Converter on http://www.verypdf.com to remove this watermark.

The origin is not stable in the sense of Lyapunov. Given any ε > 0, no matter
how small a δ we choose for the region of initial condition there always some initial
conditions close to the x1 -axis which will exit the ε-region before converging to the
origin.

5
 (b) We need to show two conditions for asymptotic stability. First we need to show that
for any given ε > 0 we could always find a δ > 0 such that kx (0)k < δ =⇒ kx (t)k <
ε, ∀t ≥ 0. Furthermore we need to show that when the initial condition is on some
domain every trajectory converges to the origin.
The solution is given by x (t) = eαt x (0). We then have |x (t)| ≤ |x (0)| for all t ≥ 0
since α < 0. Given any ε > 0, choose δ = ε to show that for all |x (0)| < δ = ε it
follows that |x (t)| < ε, ∀t ≥ 0. Thus the origin is stable. From the solution it is
easy to see that for any δ we have
|x (0)| < δ =⇒ lim x (t) = lim eαt x (0) = 0.
t→∞ t→∞

as α < 0.

4. We first introduce a change of variables

z1 = x1 − x∗1
z2 = x2 − x∗2
where (x∗1 , x∗2 ) = (−1, 1). We could write our differential equations as
ż1 = ẋ1 = −x21 x2 − 2x1 x2 + x21 + 2x1
= − (z1 − 1)2 (z2 + 1) − 2 (z1 − 1) (z2 + 1) + (z1 − 1)2 + 2 (z1 − 1)
= z2 1 − z12

ż2 = ẋ2 = x31 + 2x21 + x21 x2 + 2x1 x2

= (z1 − 1)3 + 2 (z1 − 1)2 + (z1 − 1)2 (z2 + 1) + 2 (z1 − 1) (z2 + 1)
− z1 + z2 − z13 − z12 z2


=
− (z1 + z2 ) 1 − z12


=
and the new system becomes
 
z2
1 − z12


ż = (12)
− (z1 + z2 )
 
Let our Lyapunov function candidate be in the form V (z) = z T P z where z T = z1 z2
and  
p11 p12
P =
p12 p22
is a positive definite symmetric matrix. Each component on the matrix P will be assigned
later to a specific value such that our function candidate V becomes a Lyapunov function
(in this case P does NOT have to be unique). Expanding the Lyapunov function candidate
we have V (z) = p11 z12 + 2p12 z1 z2 + p22 z22 . Taking the derivative of the function along the
trajectory we have
V̇ = 2p11 z1 ż1 + 2p12 ż1 z2 + 2p12 z1 ż2 + 2p22 z2 ż2
= 2p11 z1 z2 1 − z12 + 2p12 z22 1 − z12 − 2p12 z12 + z1 z2 1 − z12 − 2p22 z1 z2 + z22 1 − z12

= 2p11 z1 z2 − 2p11 z13 z2 + 2p12 z22 − 2p12 z22 z12 − 2p12 z12 + 2p12 z14 − 2p12 z1 z2
+2p12 z13 z2 − 2p22 z1 z2 + 2p22 z13 z2 − 2p22 z22 + 2p22 z22 z12
= −2p12 z12 + 2 (p11 − p12 − p22 ) z1 z2 − 2 (p22 − p12 ) z22 + H.O.T.

6
where the higher order term is given by

H.O.T. = −2p11 z13 z2 − 2p12 z22 z12 + 2p12 z14 + 2p12 z13 z2 + 2p22 z13 z2 + 2p22 z22 z12

We could rewrite the derivative as

V̇ = −z T Qz + H.O.T.

where  
2p12 −p11 + p12 + p22
Q=
−p11 + p12 + p22 2 (p22 − p12 )
To prove asymptotic stability it is enough to require our function candidate V to be a
Lyapunov function on a neighborhood of the origin (does not have to be in the global
region since we are not talking about global asymptotic stability). This means that it
is enough to show V̇ to be negative definite on a neighborhood of the origin. From this
point of view we could see that near the origin, the quadratic term z T Qz dominates the
higher order term H.O.T. Hence, V̇ will be negative definite in the neighborhood of the
origin if the term −z T Qz is negative definite. In this case it is enough to show that Q is
a positive definite matrix. Thus a sufficient condition for asymptotic stability is to have
P and Q to be positive definite. By checking all the leading priciple minors of P and Q
to be positive we might end up having nonunique solution of P and Q. One choice is by
setting
p11 = 3, p12 = 1, p22 = 2
where we have    
3 1 2 0
P = , Q=
1 2 0 2
Since P and Q are positive definite matrices then (z1 , z2 ) = (0, 0) is asymptotically stable,
or equivalently (x1 , x2 ) = (−1, 1) is asymptotically stable.

Another approach is to utilize that V̇ can be written:

 T   
z2 p11 p12 z1
z12


V̇ = 1−
− (z1 + z2 ) p12 p22 z2
= 1 − z12 z1 z2 p11 − z1 (z1 + z2 ) p12 + z22 p12 − z2 (z1 + z2 ) p22

= − 1 − z12 z12 (p12 ) + z1 z2 (−p11 + p12 + p22 ) + z22 (p12 + p22 )

= − 1 − z12 z T Qz

1
 
p12 2
(−p11 + p12 + p22 )
Q= 1
2
(−p11 + p12 + p22 ) p12 + p22
We can thus see that V̇ is negative definite on a domain D = {z ∈ R2 |1 − z12 > 0} under
the condition that z T Qz > 0∀z 6= 0, i.e. that Q is positive definite. Furthermore, P still
needs to be positive definite for V to be a Lyapunov function candidate.

P > 0 : p11 > 0 p11 p22 > p212 ⇒ p22 > 0

1
Q > 0 : p12 > 0 p12 (p12 + p22 ) − (−p11 + p12 + p22 )2 > 0
4
1
p12 (p12 + p22 ) > (p12 + p22 − p11 )2
4

7
 Choosing p12 = 1 > 0 ⇒ p11 p22 > 1 and 1 (1 + p22 ) > 14 (1 + p22 − p11 )2 .
Choosing p22 = 2 > 0 ⇒ p11 > 12 and 1 (1 + 2) > 14 (1 + 2 − p11 )2 ⇒ 12 > (3 − p211 ).
Choosing p11 = 3 > 0 ⇒ 3 > 21 and 12 > 0.  
T 3 1
To conclude: By using the Lyapunov function V (z) = z P z, P = we have shown
1 2
that the equilibrium is asymptotically stable.

5. The pendulum system is given by

ẋ1 = x2
g k
ẋ2 = − sin x1 − x2
l m
A general quadratic Lyapunov function candidate is given by
1 T g
V (x) = x P x + (1 − cos x1 )
2 l   
1  p11 p12 x1 g
= x1 x2 + (1 − cos x1 )
2 p12 p22 x2 l

The quadratic form 12 xT P x is positive definite if and only if all the leading principal
minors of P are positive

p11 > 0
p11 p22 − p212 > 0 (13)

The derivative of the Lyapunov function candidate along the trajectories of the system
is given by
g
V̇ (x) = xT P ẋ + ẋ1 sin x1
l   
  p11 p12 ẋ1 g
= x1 x2 + ẋ1 sin x1
p12 p22 ẋ2 l
g
= (x1 p11 + x2 p12 )ẋ1 + (x1 p12 + x2 p22 )ẋ2 + sin x1 · ẋ1
l  
h g i g k
= p11 x1 + p12 x2 + sin x1 x2 + (p12 x1 + p22 x2 ) − sin x1 − x2
l l m
   
g g k k
= (1 − p22 )x2 sin x1 − p12 x1 sin x1 + p11 − p12 x1 x2 + p12 − p22 x2
l l m m 2

The elements p11 , p12 , and p22 should be selected such that V̇ (x) becomes negative definite.
The signs of x2 sin x1 and x1 x2 change based on the quadrant of x1 and x2 and therefore,
these two elements should be eliminated. This happens by choosing p22 = 1 and p11 =
(k/m)p12 . Then, from (13), p12 should satisfy 0 < p12 < (k/m) to have a positive definite
V . By choosing p12 = 21 (k/m)

1g k 1k 2
V̇ (x) = − x1 sin x1 − x
2lm 2m 2

8
 Obviously, x1 sinx1 > 0 for all −π < x1 < π. Therefore, by selecting

Br = {x ∈ R2 | |x1 | < π},

it concludes that V (x) is positive definite and V̇ (x) is negative definite in Br . Thus, we
can conclude that the origin is locally asymptotically stable.

6. The function V (x) = 0.5 (x21 + x22 ) is positive definite at all points which are not in the
origin. Then

V̇ = x1 ẋ1 + x2 ẋ2
= x1 x2 + αx1 β 2 − x21 − x22 + x2 −x1 + αx2 β 2 − x21 − x22

= α x21 + x22 β 2 − x21 − x22

Defining
U , x ∈ R2 | kxk2 ≤ r, 0 < r < β


which is nonempty, it follows that V and V̇ are positive definite in U . By the Chetaev’s
theorem the origin is unstable.

7. (Khalil 4.9) The function is given by

(x1 + x2 )2 2
V (x) = 2 + (x1 − x2 )
1 + (x1 + x2 )

(a) Let x1 = 0, then V (x) is given by

x22
V (x) = + x22
1 + x22
x2
and it can be seen that V (x) = 1+x2 2 + x22 → ∞ as |x2 | → ∞.
2
Let x2 = 0, then V (x) is given by

x21
V (x) = + x21
1 + x21
x21
and it can be seen that V (x) = 1+x21
+ x21 → ∞ as |x1 | → ∞.
(b) On the set x1 = x2 the function is given by

4x21
V (x) =
1 + 4x21
4x21
and it can be seen that V (x) = 1+4x21
→ 1 as |x1 | → ∞, and V (x) is therefore not
radially unbounded.

9
8. Since P is positive definite we should have p11 > 0 and p11 p22 − p212 > 0 (i.e. all leading
principal minors are positive, see Appendix). The eigenvalues λ of P are found from the
equation

det (λI − P ) = 0
 
λ − p11 −p12
⇐⇒ det =0
−p12 λ − p22
⇐⇒ (λ − p11 ) (λ − p22 ) − p212 = 0
⇐⇒ λ2 − (p11 + p22 ) λ + p11 p22 − p212 = 0

Then we have
 q 
2 2
λmin = 0.5 p11 + p22 − (p11 + p22 ) − 4 (p11 p22 − p12 )
 q 
2 2
λmax = 0.5 p11 + p22 + (p11 + p22 ) − 4 (p11 p22 − p12 ) .

where
(p11 + p22 )2 − 4 p11 p22 − p212 = (p11 − p22 )2 + 4p212


(14)
To show λmin I ≤ P we need to show that

Λmin = P − λmin I
 
−λmin + p11 p12
=
p12 −λmin + p22
is positive semidefinite. The leading principal minors of Λmin are

µmin,1 = −λmin + p11

µmin,2 = det (Λmin )
 q 
2 2
and we have µmin,1 = 0.5 p11 − p22 + (p11 − p22 ) + 4p12 > 0 (since |p11 − p22 | <
q
(p11 − p22 )2 + 4p212 ) and µmin,2 = det (P − λmin I) = 0. Thus Λmin is positive semidefi-
nite. This implies that

xT Λmin x ≥ 0
⇐⇒ xT P x − λmin xT x ≥ 0
⇐⇒ xT P x − λmin kxk22 ≥ 0
⇐⇒ xT P x ≥ λmin kxk22

for all x. To show P ≤ λmax I we need to show that

Λmax = λmax I − P
 
λmax − p11 −p12
=
−p12 λmax − p22
is positive semidefinite. The leading principal minors of Λmax are

µmax,1 = λmax − p11

µmax,2 = det (Λmax )

10
  q 
2 2
and we have µmax,1 = 0.5 p22 − p11 + (p11 − p22 ) + 4p12 > 0 and µmin,2 = det (λmax I − P ) =
0. Thus Λmax is positive semidefinite. This implies that

xT Λmax x ≥ 0
⇐⇒ λmax xT x − xT P x ≥ 0
⇐⇒ λmax kxk22 − xT P x ≥ 0
⇐⇒ xT P x ≤ λmax kxk22

for all x.

9. The system is given by

ẋ1 = −x1 + 2x2 − x2 x3

ẋ2 = −x2
ẋ3 = −x3 + x1 x2 + x22

By setting ẋ2 = 0, we see that x∗2 = 0. Inserting this into ẋ1 and ẋ3 gives (x∗1 , x∗2 , x∗3 ) =
(0, 0, 0) as the only possible equilibrium of the system.

A general quadratic Lyapunov function candidate is given by

1
V (x) = xT P x, P = PT
2
For a positive definite P , we have that

λmin kxk2 ≤ xT P x ≤ λmax kxk2 ∀x

with λmin , λmax > 0 being the smallest and largest eigenvalue of P , respectively.
As given in the hint, let  
p11 p12 0
P = p12 p22 0 
0 0 p33
For P to be positive definite, the coefficients will have to be chosen such that the principal
minors of P are positive, that is

p11 > 0 (15)

p11 p22 − p212 > 0 (16)
p11 p22 − p212 > 0


p33 ⇒ p33 > 0 (17)

The derivative of V along the trajectories of the system is given by

V̇ (x) = ẋT P x
= p11 x1 ẋ1 + p12 x1 ẋ2 + p12 x2 ẋ1 + p22 x2 ẋ2 + p33 x3 ẋ3
= − p11 x21 + 2p11 x1 x2 − p11 x1 x2 x3 − p12 x1 x2 − p12 x1 x2 + 2p12 x22 − p12 x22 x3
− p22 x22 − p33 x23 + p33 x1 x2 x3 + p33 x3 x22
= − p11 x21 − (p22 − 2p12 )x22 − p33 x23 + 2(p11 − p12 )x1 x2 + (p33 − p11 )x1 x2 x3

11
 To cancel out the terms with indeterminate sign, choose p12 = p11 and p33 = p11 . In order
for V̇ to be negative and for there to be a constant k3 > 0 such that V̇ ≤ −k3 kxk22 , choose
p22 such that p22 > 2p12 . Since p12 = p11 > 0, the condition (16) holds for this choice of
p22 , such that P is positive definite. The derivative of V is then

V̇ = −p11 x21 − (p22 − 2p11 )x22 − p11 x23 ≤ − min(p11 , p22 − 2p11 )kxk22 ∀x ∈ R3

The conditions of Theorem 4.10 are fulfilled with a = 2, k1 = λmin , k2 = λmax and
k3 = min(p11 , p22 − 2p11 ), all positive. Since they hold on all of R3 , the origin is globally
exponentially stable.

10. The system is given as

ẋ1 = −4x1 + 3x2
(18)
ẋ2 = x1 − 2x2 − x32

The Jacobian matrix of the considered system is

   
∂f −4 3
A= =
∂x 1 −2 − 3x22

For simplicity, choose P = I. This gives

 
T −8 4
F (x) = A (x)P + P A(x) =
4 −4 − 6x22

The matrix F (x) is negative definite iff all its Eigenvalues are negative. The eigenvalues
are given by
 
λ + 8 −4 
 −4 λ + 4 + 6x22  = 0
 

(λ + 8)(λ + 4 + 6x22 ) − 16 = 0
λ2 + 4λ + 6x22 λ + 8λ + 32 + 48x22 − 16 = 0
λ2 + (12 + 6x22 )λ + (16 + 48x22 ) = 0

 
1
q
2 2 2 2
λ1,2 = −(12 + 6x2 ) ± (12 + 6x2 ) − 4(16 + 48x2 )
2
q
2
= −(6 + 3x2 ) ± (6 + 3x22 )2 − (16 + 48x22 )

where
q
0 < (6 + 3x22 )2 − (16 + 48x22 ) < (6 + 3x22 )

which means that

q
λ1,2 = −(6 + 3x22 ) ± (6 + 3x22 )2 − (16 + 48x22 ) < 0 ∀ x2 ∈ R

12
 and F (x) is negative definite on R2 . Moreover,

V (x) = f T (x)f (x)

= (−4x1 + 3x2 )2 + (x1 − 2x2 − x32 )2 → ∞ as ||x|| → ∞

(if kxk were to approach ∞ along the line 4x1 = 3x2 , the second term of V (x) would still
approach ∞, and vice versa). It can therefore be concluded that by Krasovskii’s theorem,
the origin of the system (18) is globally asymptotically stable.

11. (Optional exercise) (Khalil 4.10)

R1 ∂
(a) Given f (x) = 0 ∂x f (σx) xdσ
Z 1 Z 1 T
T T T ∂ ∂
x P f (x) + f (x) P x = x P f (σx) xdσ + f (σx) xdσ P x
0 ∂x 0 ∂x
Z 1 Z 1  T
T ∂ T ∂
= x P f (σx) xdσ + x f (σx) dσP x
0 ∂x 0 ∂x
Z 1 Z 1 T !
∂ ∂
= xT P f (σx) dσ + f (σx) dσP x
0 ∂x 0 ∂x
Z 1  T !
T ∂ ∂
= x P f (σx) + f (σx) P dσx
0 ∂x ∂x

∂ ∂

T
and by using P ∂x f (σx)+ ∂x f (σx) P ≤ −I the expression may be upper bounded
by
xT P f (x) + f T (x) P x ≤ xT (−I) x = −xT x = − kxk22
(b) A Lyapunov function candidate is given as V (x) = f T (x) P f (x) where P is sym-
metric and positive definite. To show that V (x) is positive definite, we need to show
that f (x) = 0 if and only if x = 0. In other words we need to show that the origin
is a unique equilibrium point. Suppose, to the contrary that there is a p 6= 0 such
that f (p) = 0. Then

pT p ≤ − pT P f (p) + f T (p) P p = 0


(19)

which is a contradiction since p 6= 0 (in order to satisfy the inequality (19), p needs
to be equal to zero). Hence the origin is a unique equilibrium point.

To see that the function is radially unbounded, we first assume that V (x) bounded
such that

V (x) = |V (x)| ≤ a, ∀x ∈ Rn

which means that

|V (x)| = |f (x)> P f (x)| ≤ λmax (P )||f (x)||22 ≤ a.

13
 We can conclude if V (x) is bounded then ||f (x)||2 needs to be bounded by
r
a
||f (x)||2 ≤ =c
λmax (P )
where c > 0. Now we see if there exist a c > 0 such that ||f (x)||2 ≤ c. Here,
x> P f (x) + f (x)> P x ≤ −x> x
is used. First we see that
x> P f (x) + f (x)> P x = 2x> P f (x) ≤ −x> x. (20)
Next, we have
2x> P f (x) ≤ 2||x||2 λmax (P )||f (x)||2 ≤ ||x||2 b, (21)
where
b = 2λmax (P )c.
It can be seen that (20) is equivalent to
2x> P f (x) ≤ −||x||22 . (22)
By summing 22 and (21) together we get
4x> P f (x) ≤ ||x||2 (b − ||x||2 )
4x> P f (x)
≤ b − ||x||2
||x||2
4x> P f (x)
lim ≤ b − lim ||x||2
x→∞ ||x||2 x→∞

0 ≤ −∞,
which is clear contradiction. Thus as ||x||2 → ∞, the magnitude of f (x) must
approach ∞, which shows that V (x) → ∞ as ||x||2 → ∞.
(c) We have shown that V (x) is positive definite and radially unbounded. The time
derivative of the function is found as
V̇ (x) = f˙T (x) P f (x) + f T (x) P f˙ (x)
 T  
∂ T ∂
= f (x) ẋ P f (x) + f (x) P f (x) ẋ
∂x ∂x
 T  
∂f (x) T ∂f (x)
= f (x) P f (x) + f (x) P f (x)
∂x ∂x
 T  
T ∂f (x) T ∂f (x)
= f (x) P f (x) + f (x) P f (x)
∂x ∂x
 T !
∂f (x) ∂f (x)
= f T (x) P + P f (x)
∂x ∂x
≤ −f T (x) f (x)
= − kf (x)k22
Since origin is a unique equilibrium point and all of the conditions hold globally, the
origin is a globally asymptotically stable equilibrium point.

14