E1 242: Nonlinear Systems and Control (2023j)

HW #3 Solutions

1. HK 4.14
Solution: As g(y) ≥ 1 for all y ∈ R,
Z x1 Z x1
1
yg(y)dy ≥ ydy = x21 .
0 0 2

Thus,
Z x1
1 1
V (x) := yg(y)dy + x1 x2 + x22 ≥ x21 + x1 x2 + x22 ≥ xT Qx,
0 2 2
 
1 1
where Q := .
1 2

We can verify that Q is a positive definite matrix. Also, V (0) = 0. Hence, V (x) is a
positive definite function. Moreover, as xT Qx grows unbounded as x grows unbounded in
any direction, V (x) is also radially unbounded.
Along the trajectories of the system, we have

V̇ (x) = x1 g(x1 )ẋ1 + x2 ẋ1 + x1 ẋ2 + 2x2 ẋ2 = x1 g(x1 )x2 + x22 − (x1 + 2x2 )g(x1 )(x1 + x2 )
= −g(x1 )(x21 + 2x1 x2 + 2x22 ) + x22
≤ −(x21 + 2x1 x2 + 2x22 ) + x22 = −(x1 + x2 )2 ≤ 0, ∀x ∈ R2 .

Since V is a radially unbounded positive definite function and V̇ (x) ≤ 0 for all x ∈ R2 , we
can say that the origin is stable and any arbitrarily large sub-level set of V (x) is positively
invariant.
So, now we apply one of the corollaries of LaSalle’s invariance principle. Notice that

V̇ (x) = −g(x1 )(x1 + x2 )2 − (g(x1 ) − 1)x22 .

Since we only know that g(x1 ) ≥ 1, it is not possible to write the set

E := {x ∈ R2 | V̇ (x) = 0}.

exactly. However, notice that

V̇ (x) ≤ −(x1 + x2 )2 ≤ 0,

which means

V̇ (x) = 0 =⇒ x1 + x2 = 0.

That is,
E ⊆ Ē := {x ∈ R2 | x1 + x2 = 0}.
LaSalle’s guarantees convergence to the largest invariant set in E. But since, we don’t
know what E is, let us compute the largest invariant set in Ē.

x1 + x2 ≡ 0 =⇒ ẋ1 + ẋ2 ≡ 0 and ẋ2 ≡ 0 =⇒ ẋ1 ≡ 0 =⇒ x2 ≡ 0 =⇒ x1 ≡ 0.

Thus, the largest invariant set in Ē is the origin itself, which also belongs to E. Hence,
{(0, 0)} is also the largest invariant set in E and as result, we can say that the origin is
G.A.S. by one of the corollaries of the LaSalle’s invariance principle.

1
2. HK 4.19 Given the dynamics

M (q)q̈ + C(q, q̇)q̇ + Dq̇ + g(q) = u, (Ṁ − 2C) is skew-symmetric, D ⪰ 0 and M ≻ 0.

Assume P (q) is positive definite function of q and g(q) = 0 has an isolated root at q = 0
∂P (q)
(a) Use u = 0 and V (q, q̇) = 21 q̇ T M (q)q̇ + P (q) where, g(q) = ∂q as a Lyapunov function
candidate to show that the origin is stable
(b) With u = −Kd q̇ where Kd is a positive diagonal matrix, show that the origin is
asymptotically stable.
(c) With u = g(q) − Kp (q − q ∗ ) − Kd q̇ where Kd and Kp are positive diagonal matrices and
q ∗ is the desired position. Show that (q ∗ , 0) is an asymptotically stable equilibrium point
Solution: (a) Clearly, V (.) is positive definite. Now,
1
V̇ = q̇ T M (q)q̈ + q̇ T Ṁ (q)q̇ + g(q)q̇.
2
Substituting for M (q)q̈ from the dynamics,
1
V̇ = q̇ T (Ṁ − 2C − 2D)q̇ = −q̇ T Dq̇ ≤ 0, ∀[q q̇]T ,
2
since xT (Ṁ −2C)x = 0 as (Ṁ −2C) is skew-symmetric matrix and D ≻ 0. From Lyapunov
stability theorem, the origin is stable.
(b) Substituting u = −Kd q̇ in the dynamics and using the same Lyapunov function and
substituting for M (q)q̈ in V̇ ,
1
V̇ = q̇ T (−2D − 2Kd )q̇ ≤ 0, ∀[q q̇]T .
2

This means that every sub-level set of V (.) (which is compact) is positively invariant to
the dynamics. Hence in every sub-level set of V (.), V̇ = 0 if and only if q̇ = 0. Since
q̇ ≡ 0 ⇒ q̈ ≡ 0 =⇒ g(q) ≡ 0. Since g(q) = 0 has an isolated root at q = 0, we can choose
a small enough sub-level set of V in which the g(q) ≡ 0 =⇒ q ≡ 0. Thus, [0 0]T is the
largest invariant set in set of points such that V̇ = 0, in a sufficiently small sub-level set
of V . Hence, by LaSalle’s theorem, [0 0]T is an asymptotically stable equilibrium point.
(c) Consider a candidate Lyapunov function V (x) = 21 q̇ T M (q)q̇ + 21 (q − q ∗ )T Kp (q − q ∗ ).
Substituting u = g(q) − Kp (q − q ∗ ) − Kd q̇ in the dynamics and computing V̇ ,
1
V̇ = q̇ T (Ṁ (q) − 2C(q, q̇))q̇ − q̇ T (D + Kd )q̇
2
⇒ V̇ = −q̇ T (D + Kd )q̇, since (Ṁ − 2C) is skew-symmetric.
⇒ V̇ ≤ 0.

This means that every sub-level set of V (.) (which is compact) is positively invariant to
the dynamics. Hence in every sub-level set of V (.), V̇ = 0 if and only if q̇ = 0. Since
q̇ ≡ 0 ⇒ q̈ ≡ 0 =⇒ Kp (q − q ∗ ) ≡ 0 =⇒ (q − q ∗ ) ≡ 0 as Kp is a positive diagonal
matrix. Thus, the largest invariant set in set of points such that V̇ = 0 is [q ∗ 0]T . Hence,
by LaSalle’s theorem, [q ∗ 0]T is an asymptotically stable equilibrium point.

3. HK 4.20 Suppose the set M in LaSalle’s theorem consists of a finite number of isolated
points. Show that lim x(t) exists and equals one of these points.
t→∞
Solution: By LaSalle’s theorm, we have, x(t) → M as t → ∞, i.e., ∀ϵ > 0, ∃T > 0 such
that inf ||x(t) − y|| < ϵ, ∀t > T .
y∈M

2
 min ||y −y ||
j i
Let us choose ϵ < 2 , ∀yj , yi ∈ M , yi ̸= yj . Then, as x(t) is a continuous function
of time t, we see that ∀t > T , x(t) must be confined to the ϵ-ball around one of the yi ’s.
Otherwise there would be a t > T such that inf y∈M ||x(t) − y|| > ϵ. Further, such an ϵ-ball
around each yi contains no other yj . Hence, as x(t) → M , solutions x(t) → yj for some
yj ∈ M .

4. HK 4.21
Solution: (a) The time derivative of V along the trajectories of the given system is

∂V ∂V T
 
∂V
V̇ (x) = − ≤ 0, V̇ (x) = 0 ⇐⇒ = 0.
∂x ∂x ∂x

(b) The sub-level set Ωc is compact and also positively invariant because of (a). Further,
∂V
since V is twice continuously differentiable, is locally Lipschitz everywhere and Lipscitz
∂x
on compact sets. Thus, we have existence and uniqueness of solutions for all future time
for solutions starting in Ωc . Since this true for any c > 0, we can say that the claim in the
problem is true.
(c) By a corollary of LaSalle’s invariance principle, we can say that all solutions converge
to the largest invariant set in the set where V̇ (x) = 0 ⇐⇒ ∂V ∂V
∂x = 0. Since ∂x = 0, only
at a finitely many points and each one is an equilibrium point, by (a), we can now use HK
4.20 to prove the result.

5. HK 4.22
Solution: Suppose ∃P = P T > 0 that satisfies the given Lyapunov equation. Then, the
time derivative of V (x) = xT P x along the trajectories of ẋ = Ax is

V̇ (x) = xT [P A + AT P ]x = −xT C T Cx.

V is positive definite radially unbounded function. So, we can use the corollary to LaSalle’s
invariance principle.
E := {x | V̇ (x) = 0} = {xT C T Cx = 0}.
The largest invariant set in E is
T
xT C T Cx ≡ 0 =⇒ xT0 eA t C T CeAt x0 ≡ 0 =⇒ x0 = 0 =⇒ x ≡ 0,

where the second last implication is from the definition of observability in the hint given in
the problem. Thus, from the corollary to LaSalle’s invariance principle, we can conclude
that x = 0 is G.A.S. and hence A is hurwitz.
Now, suppose that A is Hurwitz. Let
Z ∞
T
P = eA t C T CeAt dt.
0

It is clear that P is symmetric and positive semi-definite. Now, note that

xT P x = 0 =⇒ CeAt x ≡ 0 =⇒ x = 0,

where again we have used the definition of observability. Hence P is positive definite.
Now,
Z ∞ Z ∞
T AT t T At T
PA + A P = e C Ce Adt + AT eA t C T CeAt dt
Z0 ∞ 0
∞
d AT t T At AT t T At
= e C Ce dt = e C Ce = −C T C,
0 dt 0

3
 since A Hurwitz means eAt → 0 as t → ∞. We can verify uniqueness of the solution for P
by contradiction by assuming that there are at least two distinct solutions P1 and P2 that
satisfy
P1 A + AT P1 = −C T C = P2 A + AT P2 =⇒ (P1 − P2 )A + AT (P1 − P2 ) = 0.
Tt
Pre-multiplying by eA and post-multiplying by eAt gives
T d h AT t i
eA t [(P1 − P2 )A + AT (P1 − P2 )]eAt = 0 =⇒ e (P1 − P2 )eAt = 0, ∀t ≥ 0.
dt
T
In other words, eA t (P1 − P2 )eAt is a constant for all t ≥ 0 and hence
T T T
eA t (P1 − P2 )eAt = eA 0 (P1 − P2 )eA0 = (P1 − P2 ) = lim eA t (P1 − P2 )eAt = 0,
t→∞

where the last equality is true because A is Hurwitz. This means P1 = P2 , which is a
contradiction. Thus, there is a unique solution to the Lyapunov equation.
6. HK 4.35
Solution: Notice that r1 + r2 ≤ 2 max{r1 , r2 }. So,
α(r1 + r2 ) ≤ α(2 max{r1 , r2 }) ≤ α(2 max{r1 , r2 }) + α(2 min{r1 , r2 }) ≤ α(2r1 ) + α(2r2 ).
Since the domain of α is [0, a), the above inequality is well defined for r1 , r2 ∈ [0, a/2).
7. HK 4.39
Solution: (a) Notice that
1
V (t, x) ≥ (a sin(x1 ) + x2 )2 + (1 − cos(x1 )),
2
as g(t) > a > 0. Thus, V (t, x) is a positive definite function in the set where x1 ∈
(−2π, 2π). Next,
1
V (t, x) ≤ (a sin(x1 ) + x2 )2 + [1 + a(β − a)](1 − cos(x1 )),
2
as g(t) ≤ β. The RHS is again a positive definite function in the set where x1 ∈ (−2π, 2π)
as β > a. Thus, V (t, x) is also decrescent.
(b)
V̇ = aġ(t)(1 − cos(x1 )) + [(a sin(x1 ) + x2 )a cos(x1 ) + (1 + ag(t) − a2 ) sin(x1 )]ẋ1
+ (a sin(x1 ) + x2 )ẋ2
≤ aγ(1 − cos(x1 )) + [(a sin(x1 ) + x2 )a cos(x1 ) + (1 + ag(t) − a2 ) sin(x1 )]x2 +
+ (a sin(x1 ) + x2 )[− sin(x1 ) − g(t)x2 ]
= aγ(1 − cos(x1 )) − (g − a cos(x1 ))x22 − a sin2 (x1 ) − a2 sin(x1 )(1 − cos(x1 ))x2
≤ −(α − a)x22 − a(2 − γ)(1 − cos(x1 )) + [2a − a2 sin(x1 )x2 ](1 − cos(x1 )) − a sin2 (x1 ).
Now, note that
V̇ ≤ −(α − a)x22 − a(2 − γ)(1 − cos(x1 )) + [2a + a2 ||x||2 ]||x|| + a||x||2
≤ −(α − a)x22 − a(2 − γ)(1 − cos(x1 )) + k||x||3
for some k > 0 in a small enough ball around the origin. Here ||.|| is the Euclidean norm.
(c) From the inequality in (b), we see that V̇ is negative definite in a small enough ball
around the origin. We can also show that there exists a negative of class K function of
norm of x that bounds V̇ . So, we can use the Lyapunov stability theorem to guarantee
that the origin is a.s.