TTK4150 Nonlinear Control Systems

Department of Engineering Cybernetics

Norwegian University of Science and Technology
Fall 2022 - Solution to Assignment 3

1. From V (x) = 12 (x21 + x22 ) and

ẋ1 = x2
ẋ2 = −(x1 + x2 ) − x21 x2

we have

V̇ (x) = x1 x2 − (x1 + x2 )x2 − x21 x22

= x1 x2 − x1 x2 − x22 − x21 x22
= −(1 + x21 )x22 ≤ 0 ∀x ∈ R2 \{0}

Thus V̇ is negative semidefinite since V̇ = 0 on the line x2 = 0 (where x1 ∈ R). When V̇ =

0 we must have x2 = 0 and also ẋ2 = 0. From ẋ2 = 0 we must have −(x1 + x2 ) − x21 x2 = 0
or equivalently −x1 = 0 since x2 = 0. For x1 = x2 = 0 we also have ẋ1 = 0, such
that (0, 0) is an equilibrium point of the system, and it follows that V̇ = 0 identically
on x = (0, 0). Hence, by Corollary 4.1, the origin is asymptotically stable. Furthermore
since V is radially unbounded, by Corollary 4.2 we conclude that the origin is globally
asymptotically stable.

2. (a) Consider the following Lyapunov function candidate

1 2
x̃1 + mx̃22

V (x̃) =
2
Differentiating V along the trajectories of the system yields

V̇ (x̃) = x̃1 x̃˙ 1 + mx̃2 x̃˙ 2

= x̃1 x̃2 + x̃2 −f3 (x̃1 + x1d )3 − x31d − f1 x̃1 − dx̃2 + ũ
 

Requiring V̇ (x̃) = −(d + k2 )x̃22 , we must thus solve

−(d + k2 )x̃22 = x̃1 x̃2 + x̃2 −f3 (x̃1 + x1d )3 − x31d − f1 x̃1 − dx̃2 + ũ
 

for the input ũ. This yields

ũ = f3 (x̃1 + x1d )3 − x31d + f1 x̃1 − x̃1 − k2 x̃2 .

 

(b) The closed-loop system is found by inserting ũ into the system equations

x̃˙ 1 = x̃2
˙ (1)
mx̃2 = −x̃1 − (d + k2 )x̃2
↓
x̃˙ 1
    
0 1 x̃1
= (2)
x̃˙ 2 1 1
− m − m (d + k2 ) x̃2

1
 1. V̇ from the previous question is negative semidefinite (that is, V̇ = 0 on the
line x̃2 = 0). V is a continuously differentiable, positive definite, radially un-
2
bounded
n o that V̇ (x̃) ≤ 0 for all x̃ ∈ R and V (0) = 0. Let
function such
S = x̃ ∈ R2 |V̇ = 0 , i.e. S = {x̃ ∈ R2 |x̃2 = 0}. Inserting x̃2 = 0 gives

x̃2 ≡ 0 ⇒ x̃˙ 2 ≡ 0 ⇒ x̃1 ≡ 0

i.e. the only solution that can stay in S is the trivial solution x̃ = 0. LaSalle’s
theorem states that the origin is a globally asymptotically stable equilibrium
point of the closed-loop system.
2. Since the closed-loop system is linear, we can analyze its stability properties by
considering the eigenvalues. The eigenvalues of the closed loop system can be
calculated from
λ −1
1 1 =0
m
λ + m
(d + k2 )
which gives
1 1
λ2 + λ
(d + k2 ) + =0
 m m 
  s 2
1 d + k2 d + k2 4
λ = − ± −  (3)
2 m m m

Since d, k2 , m > 0 s 2
d + k2 d + k2 4
> −
m m m
and the eigenvalues will always lie in the left half plane which means that A is
Hurwitz. Any linear system on the form

ẋ = Ax

is exponentially stable as long as A is Hurwtitz, and the system (2) is hence

globally exponentially stable.
(c) From (1) it is easy to see that increasing k2 will introduce more damping into the
system. Therefore the overall system dynamics get slower as the controller gain k2
increases.
(d) The closed-loop system has two poles, but we only have one control parameter k2 .
Trying to place the two poles arbitrarily by choosing only one parameter would give
two equations with only one unknown, which is an overdetermined system and has
(in general) no solution. If k2 is found such that the first eigenvalue given by (3)
has a desired value, the second eigenvalue is also given by (3) and can no longer be
chosen arbitrarily.
(e) In Assignment 2 the input u = u0 + ũ was used to shift the equilibrium point to a
desired equilibrium point. Since the disturbance w is not counteracted by the term
u0 , the transformed form of the system is now

x̃˙ 1 = x̃2
f3
f1 d w ũ
x̃˙ 2 = − (x̃1 + x1d )3 − x31d − x̃1 − x̃2 + +
m m m m m

2
 Inserting for ũ we get

x̃˙ 1 = x̃2
1 d + k2 w
x̃˙ 2 = − x̃1 − x̃2 +
m m m
a system which is not in equilibrium at the origin. Since the origin is not an equi-
librium point of the system (in error coordinates), it is also not stable. Hence our
previous control input u is unable to make the desired setpoint stable in the presence
of an unknown, constant disturbance.

3. The system is given by

ÿ + h(y)ẏ + g(y) = 0

(a) Using x1 = y and x2 = ẏ, the state equation becomes

ẋ1 = x2
ẋ2 = −h(x1 )x2 − g(x1 )

To be an equilibrium point of the system, the origin must satisfy ẋ = 0. We see that
ẋ1 = 0 implies that x2 = 0. Inserting this in ẋ2 = 0 yields

0 = −g(x1 )

Thus, for x = 0 to be an isolated equilibrium point, we must have g(x1 ) = 0 if and

only if x1 = 0. Note that this also makes the origin the unique equilibrium point of
the system.
Rx
(b) We first need to find the conditions ensuring that V = 0 1 g(y)dy + 12 x22 is positive
definite. We see that this is the case if Rthe function g(x1 ) satisfies the condition
x
x1 g(x1 ) > 0 ∀ x1 ̸= 0 since then the term 0 1 g(y)dy is positive definite.
The time derivative of V along the trajectories of the system is

V̇ (x) = x2 ẋ2 + ẋ1 g(x1 )

= −g(x1 )x2 − h(x1 )x22 + g(x1 )x2
= −h(x1 )x22

V̇ is negative semidefinite in a neighbourhood of the origin D if the function h(x1 )

satisfies the condition h(x1 ) > 0 ∀ x1 ∈ D\{0}. Let S = {x ∈ R2 | V̇ = 0} = {x ∈
R2 | x2 = 0}. To stay identically in S, a solution must satisfy

ẋ2 = 0 ⇒ −g(x1 ) = 0

Hence, for the origin to be the only solution, we must have g(x1 ) = 0 if and only if
x1 = 0 (equivalent to what we found in a). By Corollary 4.1, the origin is asymp-
totically stable.
(c) To show global asymptotic stability, the function V R(x) needs to be radially un-
z
bounded. This is the case if the function g satisfies 0 g (y) dy → ∞ as |z| → ∞.
The condition h(x1 ) > 0 must also hold for all x1 ̸= 0, i.e. D = R2 . Then, by
the arguments provided in b and Corollary 4.2, the origin is globally asymptotically
stable.

3
4. (a) To be an equilibrium point of the system

ẋ1 = −x1 + x2 + x21 + x22

ẋ2 = −x1 − x2 + x21 + x22

a solution must satisfy ẋ = 0, that is

ẋ1 = 0 =⇒ x21 + x22 = x1 − x2

ẋ2 = 0 =⇒ x21 + x22 = x1 + x2

From this, we see that we need

x1 − x2 = x1 + x2 =⇒ x2 = 0, x1 ∈ R

Let x2 = 0. To ensure that ẋ = 0, we must also have

x21 − x1 = 0 =⇒ x1 = 0 ∨ x1 = 1.

Hence, the equilibrium points of the system are (0, 0) and (1, 0).
(b) Consider the general Lyapunov function candidate V (x) = 21 xT P x, where P =
[ pp11 p12
12 p22 ] is to be chosen as a positive definite, symmetric matrix. The derivative of

V along the trajectories of the system is given by

V̇ (x) = xT P ẋ
= −x1 + x2 + x21 + x22 (x1 p11 + x2 p12 ) + −x1 − x2 + x21 + x22





(x1 p12 + x2 p22 )
−x21 (p11 x22 (p12 x21 x22


= + p12 ) + x1 x2 (p11 − 2p12 − p22 ) + − p22 ) + + (x1 (p12 + p11 )
+ x2 (p22 + p12 ))

Choosing p12 = 0, p11 = p22 and p22 = 1, we have

V̇ (x) = −(x21 + x22 )(1 − x1 − x2 )

which entails that all the principal minors of P are positive, i.e. p11 > 0 and
p11 p22 − p212 > 0. This ensures that V is positive definite. Moreover, we see that

V̇ (x) < 0, ∀x ∈ D\{0}

where D = {x ∈ R2 | x1 + x2 < 1} (choosing the elements of P differently may yield

a different region D). By Theorem 4.1 the origin is asymptotically stable.

Notice that Theorem 4.1 states only that x is asymptotically stable, but nothing
specifically about from where the solution will converge to the origin. So long as there
exists some region D containing the origin where V is a strict Lyapunov function,
then Theorem 4.1 states that there is a (small) δ such that if ∥x(0)∥ ≤ δ, then x(t)
will converge to the origin as t → ∞ (Definition 4.1 of asymptotic stability).
Therefore any choices of D are equally good for only showing local asymptotic sta-
bility. However, the choice of D will influence the estimate of the region of attraction
obtained in the next task.

4
 2

1.5

0.5

-0.5

-1

-1.5

-2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Figure 1: Sketch of the estimated region of attraction, and D is the region left of the dashed line.

(c) Since V is radially unbounded, the level sets Ωc = {x ∈ R2 |V (x) ≤ c}, where c is
chosen such that V̇ (x) ≤ 0 ∀x ∈ Ωc , are positively invariant. We can use this to
find an estimate of the region of attraction. To get the biggest possible Ωc , choose c
as the smallest value V along the boundary of D = {x ∈ R2 | x1 + x2 < 1} found in
b. That is,

c = min V (x)
x1 =1−x2
1 2
x1 + x22


= min
x1 =1−x2 2
1
= min − x2 + x22
2
The minimum of V (x) is found by
 
∂ 1 2
− x2 + x2 = −1 + 2x2 = 0
∂x2 2

which implies that x = 21 , 21 gives the minimum V 12 , 12 = 41 .

n o
Let E = x ∈ Ω 1 V̇ (x) = 0 = (0, 0) = M , since the system cannot stay in any of
4

the other the points where V̇ = 0 (you can verify this by inserting into the system
dynamics). Note that, although V̇ (1, 0) = 0, the equilibrium point (1, 0) is not in
Ω 1 and thus not included in E. This means that (0, 0) is the largest invariant set in
4
E, and with respect to Theorem 4.4, Ω 1 can be taken as an estimate of the region
4
of attraction, shown in Figure 1.

5. (a) Inserting x1 = x2 = 0 into the system dynamics gives ẋ1 = 0 and ẋ2 = 0, confirming
that the origin is an equilibrium point.
While we do not know enough about f1 and f2 to solve for all possible equilibrium

5
 points, it is possible to check if there are equilibrium points on the set x21 +2x22 −4 = 0.
Inserting into the system dynamics gives

ẋ1 = 4x21 x2
ẋ2 = −2x31

where we see that x1 = 0 gives ẋ1 = 0 and ẋ2 = 0.√ Since we restricted the dynamics
to the set x21 + 2x22 − 4 = 0, x1 = 0√gives x2 = ± 2. Hence the system has at least
two more equilibrium points, (0, ± 2).
(b) A set C is invariant with respect to the system if, for every trajectory x,

x(t0 ) ∈ C =⇒ x(t) ∈ C ∀t ∈ R

To show that C = {x ∈ R2 | x21 + 2x22 − 4 = 0} is an invariant set of the system,

define the new variable z = x21 + 2x22 − 4. We need to show that ż = 0 when x ∈ C,
i.e. when z = 0, and ẋ satisfies the system equations. The time-derivative of z is
found as

ż = 2x1 ẋ1 + 4x2 ẋ2

= 2x1 4x21 x2 − f1 (x1 ) x21 + 2x22 − 4



+4x2 −2x31 − f2 (x2 ) x21 + 2x22 − 4




= −2x1 f1 (x1 ) x21 + 2x22 − 4 − 4x2 f2 (x2 ) x21 + 2x22 − 4

= − (2x1 f1 (x1 ) + 4x2 f2 (x2 )) x21 + 2x22 − 4

= −2 (x1 f1 (x1 ) + 2x2 f2 (x2 )) z

where it can be seen that z = 0 yields ż = 0. This means that the relation-
ship between x1 and x2 as given by z stays constant if z = 0 initially. Hence,
{x ∈ R2 |x21 + 2x22 − 4 = 0} is an invariant set for the system.
(c) The function

2
V (x) = x21 + 2x22 − 4
is radially unbounded. The derivative of V with respect to time is

V̇ (x) = 2 x21 + 2x22 − 4 (2x1 ẋ1 + 4x2 ẋ2 )

2
= −4 (x1 f1 (x1 ) + 2x2 f2 (x2 )) x21 + 2x22 − 4
n o
which is negative semidefinite. Let D = R2 . The set Ωc = x ∈ R2 |V (x) ≤ c, V̇ (x) ≤ 0 =
{x ∈ R2 |V (x) ≤ c} is a compact positively invariant set for any finite c > 0 due to
the radially unboundedness of V . Let Ω = Ωc , the set E is given by
n o
E = x ∈ Ω|V̇ (x) = 0
= x ∈ Ω|x21 + 2x22 − 4 = 0 ∪ {x ∈ Ω| x1 f1 (x1 ) + 2x2 f2 (x2 ) = 0}


= x ∈ Ω|x21 + 2x22 − 4 = 0 ∪ (0, 0)



Since both {x ∈ Ω|x21 + 2x22 − 4 = 0 } and (0, 0) are invariant sets (recall that the
origin is an equilibrium point), the largest invariant set in E is given by M = E, and
by Theorem 4.4 it can be concluded that every solution starting in Ω approaches
{x ∈ Ω|x21 + 2x22 − 4 = 0} or the origin as t → ∞.

6
 (d) Since the set C contains equilibrium points as shown in a, it cannot be a limit cycle
(once a trajectory x(t) reaches an equilibrium point, it will stay there for all future
time).

6. (a) The Jacobian of the system

ẋ1 = (x1 + 2x2 ) (−x1 − 2)

ẋ2 = −5x2 (2 + 2x1 + x2 )

is given by
 
∂f (x) − (x1 + 2) − (x1 + 2x2 ) −2 (x1 + 2)
=
∂x −10x2 −5 (2 + 2x1 + x2 ) − 5x2

Evaluation [∂f /∂x] at the origin we have

 
∂f (0) −2 −4
A= =
∂x 0 −10

and the eigenvalues are λ1,2 = −2, −10. Thus the origin is asymptotically stable.
(b) The derivative of the Lyapunov function along the trajectory of the system is

V̇ (x) = x1 (x1 + 2x2 ) (−x1 − 2) − 5x22 (2 + 2x1 + x2 )

= −x21 (2 + x1 + 2x2 ) − 4x1 x2 − 5x22 (2 + 2x1 + x2 )

In the domain D, we have

2 + x1 + 2x2 ≥ 1
2 + 2x1 + x2 ≥ 1

Hence,

V̇ (x) ≤ −x21 − 4x1 x2 − 5x22 , ∀x ∈ D

T
= −x Qx

where  
1 2
Q=
2 5
2
is positive definite (since q11 > 0 and q11 q22 − q12 > 0). Thus V̇ is negative definite
which implies that the origin is asymptotically stable.

Alternatively, use a simpler completion of squares:

V̇ (x) ≤ −x21 − 4x1 x2 − 5x22 , ∀ x ∈ D

= −(x1 + 2x2 )2 − x22 < 0

which yields the same result.

7
(c) Since V is radially unbounded, the level sets Ωc = {x ∈ R2 | V (x) ≤ c}, where V̇ (x) ≤
0 ∀ x ∈ Ωc are positively invariant sets and can thus be used as estimates of the region
of attraction. To get the largest estimate, we choose c as the smallest value of V
along the boundary of D. We start by considering the boundary 1 + x1 + 2x2 = 0:
1
1 + 4x2 + 5x22


min V (x) = min
x1 =−1−2x2 x2 2
To find the minimum, we solve
∂
1 + 4x2 + 5x22 = 0


∂x2
4 + 10x2 = 0

which gives x2 = −0.4, x1 = −0.2. Next, we consider the boundary 1 + 2x1 + x2 = 0:

1
1 + 2x2 + 5x22


min V (x) = min
1 1
x1 =− 2 − 2 x2 x2 8

The minimum is given by

∂
1 + 2x2 + 5x22 = 0


∂x2
2 + 10x2 = 0

yielding x2 = −0.2, x1 = −0.4. Since

V (−0.4, −0.2) = V (−0.2, −0.4) = 0.1

an estimate of the region of attraction can thus be taken as Ω0.1 = {x ∈ R2 | V (x) ≤ 0.1}.
(d) Since Ω0.1 is bounded and contained in D, it is positively invariant. Thus, any tra-
jectory starting in the set cannot leave it. Moreover, by Theorem 4.4, all trajectories
starting in the set will converge to the origin, which we can see is the case from
Figure 2. On the other hand, the set Ω2.5 is not bounded and contained in D. This
implies that it is possible for the trajectories not to converge to the origin in this
set. Even though V (x) decreases while x is in D, D itself is not positively invariant,
meaning that it might be possible for the trajectories to leave D (as is the case here).
Hence, D cannot be taken as an estimate of the region of attraction. Note also that
since Ω0.1 is only an estimate of the region of attraction, the true region is bigger
and therefore, some trajectories in Ω2.5 \Ω0.1 do in fact converge to the origin.

8
Figure 2: Phase portrait for the system in Exercise 6. The boundaries 1 + x1 + 2x2 = 0 and 1 +
2x1 + x2 = 0 are given by the red, dashed lines. The blue line shows a trajectory that
escapes Ω2.5 and D.

7. The system is given by

ẋ1 = x2
ẋ2 = −x1 − 2x2 + x2 cos x2
and the Lyapunov function candidate is given by
1
V (x) = xT P x, P = [ 41 13 ]
2
which is positive definite and radially unbounded. The derivative of V along the trajec-
tories of the system is given by:
V̇ (x) = ẋT P x
 T   
x2 4 1 x1
=
−x1 − 2x2 + x2 cos x2 1 3 x2
= 4x2 x1 + x22 − x21 − 3x1 x2 − 2x2 x1 − 6x22 + x1 x2 cos x2 + 3x22 cos x2
≤ −2x22 − x21 − x1 x2 + x1 x2 cos x2
where we use that |cos x2 | ≤ 1 in the last line. To deal with the cross-term containing
cos x2 , we can use Young’s inequality:
1
xy ≤ ϵx2 + y 2 , ∀ϵ > 0 ∀x, y ∈ R
4ϵ
with x = x1 and y = x2 cos x2 . This gives
1
V̇ (x) ≤ − 2x22 − x21 − x1 x2 + ϵx21 + x22 cos2 x2
  4ϵ
1
≤ − x22 2 − − x1 x2 − x21 (1 − ϵ)
4ϵ

9
 where we once again use that |cos x2 | ≤ 1, which we now can do since the term has a
definite sign. Moreover, V̇ is now on the form V̇ (x) = −xT Qx, with
1
 
1−ϵ 2
Q= 1 .
2
2 − 4ϵ1

We want to choose ϵ > 0 such that Q is positive definite (as this will make V̇ negative
definite). We know that Q > 0 if and only if all the principal minors are positive, that is

1−ϵ>0
 
1 1
(1 − ϵ) 2 − − >0
4ϵ 4

Solving the last inequality, we find that ϵ must satisfy

√ √
2− 2 2+ 2
<ϵ< , (4)
4 4
which also fulfills the first inequality. Hence, we have shown that the origin of the system
is globally asymptotically stable.

8. The system is given by q

ẋ1 = − x21 + x22 − 11x2
ẋ2 = 2x1 − 5x2
 1 −1 
and the Lyapunov function candidate is given by V (x) = 12 xT P x with P = −1 3 , which
is positive definite as the leading principal minors are positive, and radially unbounded.
The derivative of V along the trajectories of the system is given by:

V̇ = xT P ẋ
 T   p 
x1 1 −1 − x21 + x22 − 11x2
=
x2 −1 3 2x1 − 5x2
 q 
2 2
= (x1 − x2 ) − x1 + x2 − 11x2 + (−x1 + 3x2 ) (2x1 − 5x2 )
q
= − (x1 − x2 ) x21 + x22 − 11x1 x2 + 11x22 − 2x21 + 5x1 x2 + 6x1 x2 − 15x22
= − (x1 − x2 ) ∥x∥2 − 2x21 − 4x22
≤ − (x1 − x2 ) ∥x∥2 − 2∥x∥22
≤ |x1 − x2 |·∥x∥2 − 2∥x∥22

We can rewrite the term x1 − x2 as the inner product between the vector [1 − 1]T and
the vector x, that is |x1 − x2 | = |[1 − 1]x|. The Cauchy-Schwarz inequality then gives:
√
V̇ ≤ 1 + 1∥x∥2 ·∥x∥2 − 2∥x∥22
 √ 
= − 2 − 2 ∥x∥22 < 0 ∀x ∈ R2 \{0}

Thus, since V̇ is negative definite (on R2 ) and V is positive definite and radially un-
bounded, by Theorem 4.2 the origin is globally asymptotically stable

10
 9. (Khalil 4.35) Since α is a class K function, it must be strictly increasing in its argument
and positive for positive arguments. Therefore, for any r′ ≥ r > 0, we have α(r′ ) ≥
α(r) > 0. Now, if r1 ≥ r2 , we must consequently have that r1 + r2 ≤ 2r1 . Using the
properties of α, we thus have

α(r1 + r2 ) ≤ α(2r1 ) ≤ α(2r1 ) + α(2r2 ) ∀ r1 , r2 ∈ [0, α/2)

| {z }
≥0

The same holds if r2 ≥ r1 , that is

α(r1 + r2 ) ≤ α(2r2 ) ≤ α(2r1 ) + α(2r2 ) ∀ r1 , r2 ∈ [0, α/2)

The proof can be shortened by covering both cases at once:

α(r1 + r2 ) ≤ α(2 max(r1 , r2 )) ≤ α(2r1 ) + α(2r2 ).

10. (a) For class KL functions we know that, for each fixed r, the mapping β(r, s) is de-
creasing with respect to s. Therefore,

β (r, t) ≤ β (r, 0)

Since
∥x(t)∥ ≤ β (∥x(0)∥ , t) ,
we must consequently have

∥x(t)∥ ≤ β (∥x(0)∥ , 0) ,

To show that x = 0 is stable, we want δ > 0 such that

∥x(0)∥ ≤ δ ⇒ ∥x(t)∥ ≤ ϵ ∀t ≥ 0

From the above, we see that if we choose ϵ > β (δ, 0), then we will have ∥x(t)∥ ≤ ϵ for
any ∥x(0)∥ < δ, which is possible because lim β(r, 0) = 0 and β is strictly increasing
r→0
with respect to r. Hence, x = 0 is stable.

β(δ,0)
β(δ,0)

Figure 3: Plot of β (δ, 0) vs. ϵ

(b) Since β is of class KL then lim β (∥x (0)∥ , t) = 0. Thus lim ∥x (t)∥ = 0. By
t→∞ t→∞
definition of the norm, it follows that lim x (t) = 0. We conclude that the origin of
t→∞
the system is globally asymptotically stable.

11
11. We have
ẋ1 = x2
(5)
ẋ2 = − (x1 + x2 ) − h (x1 + x2 )
We want to determine the gradient, g(x) = ∇V , of the Lyapunov function so that
∂g1 ∂g2
= (6)
∂x2 ∂x1
and

V̇ (x) = g(x)T f (x) < 0 ∀ x ̸= 0 (7)

Z x
V (x) = g T (y)dy > 0 ∀ x ̸= 0
0

Looking at the system dynamics (5), we have the function h(·) of which we know very
little. In order to satisfy (7) we should choose g2 such that −g2 (x)h(x1 + x2 ) is negative.
Since we know that zh(z) > 0 ∀z, a smart choice for g2 is

g2 (x) = β(x1 + x2 )

with β > 0 being a constant. Then the symmetry condition (6) gives
∂g1 ∂g2
= =β
∂x2 ∂x1
Integrating this with respect to x2 in order to find g1 yields
Z x2
g1 (x) = β dy = βx2 + ϕ(x1 )
0

where ϕ is a function of x1 only, to be found later. We now have

V̇ (x) = g(x)T f (x) = (ϕ(x1 ) + βx2 ) x2 + β (x1 + x2 ) (−(x1 + x2 ) − h(x1 + x2 ))

= ϕ(x1 )x2 + βx22 − β(x1 + x2 )2 − β(x1 + x2 )h(x1 + x2 )
= ϕ(x1 )x2 + βx22 − βx21 − 2βx1 x2 − βx1 x2 − β(x1 + x2 )h(x1 + x2 )
= ϕ(x1 )x2 − βx21 − 2βx1 x2 − β(x1 + x2 )h(x1 + x2 )

Now choose ϕ(x1 ) = 2βx1 to cancel the cross-term and get

V̇ (x) = −βx21 − β(x1 + x2 )h(x1 + x2 ) < 0∀x ∈ R2 \{0}

since zh(z) > 0 ∀z ̸= 0 and β > 0. The function V is constructed by integrating V̇ along
any path from 0 to x. We integrate along the axes to get
Z x1 Z x2
V (x) = g1 (y1 , 0) dy1 + g2 (x1 , y2 )dy2
0 0
Z x1 Z x2
= 2βy1 dy1 + β (x1 + y2 ) dy2
0 0
 x  x
1 2 1 x2 1 2 2
= 2β y1 + βx1 [y2 ]0 + β y2
2 0 2 0
β
= βx21 + βx1 x2 + x22 = xT P x
2
12
where
β
 
β 2
P = β β
2 2

and

β > 0
2 2
β β β2
− = > 0
2 4 4
which implies that P > 0, and so V is positive definite on R2 and radially unbounded. By
Theorem 4.2 it is concluded that the origin is globally asymptotically stable. Alternatively,
g can be found similarly to Example 4.5.

13