Lecture notes 2012/2013

Dynamical systems, 6
Mihaela Cistelecan, UT Cluj-Napoca

These notes were prepared entirely from

Khalil, H. Nonlinear Systems. Prentice Hall. 1996.

1 Fundamental properties
For the state equation ẋ = f (t, x) to be a useful mathematical model for a
physical system, its solution should have the following properties:

• existence

• uniqueness

• continuous dependence on initial conditions

• continuous dependence on parameters

The meaning of these properties is as follows. In experimenting with a

physical system, we expect that starting the experiment from a given initial
state at time t0 , the system will move and its state will be defined in the
future time t > t0 . Moreover, with a deterministic system, we expect that
if we could repeat the experiment exactly, we would get exactly the same
motion and the same state at t > t0 . For the mathematical model to predict
the future state of the system from its current state at t0 , the initial value
problem
ẋ = f (t, x) with x(t0 ) = x0
must have a unique solution. This is a question of existence and uniqueness
that can be ensured by imposing some constraints on the right-hand side
function f (t, x). The key constraint is the Lipschitz condition - see below.

1
In addition, an essential factor in the validity of any mathematical model
is the continuous dependence of its solutions on the data of the problem.
This refer to the fact that arbitrary small errors in the data will not result
in large errors in the solution obtained by the model. If f is differentiable
with respect to its parameters, then the solution will be differentiable with
respect to these parameters. In this context, we may study the effect of small
parameter variations on the system, using the sensitivity equations.

1.1 Existence and uniqueness

Note that solution x(t) of :

ẋ = f (t, x) with x(t0 ) = x0 (1)

is: Z t
x(t) = x0 + f (s, x(s))ds (2)
t0

Thus, the study of existence and uniqueness of the solution of the differential
eq. (1) is equivalent to the study of existence and uniqueness of the solution
of the integral equation (2).
However, a differential equation with a given initial condition might have
several solutions. For example, the scalar equation

ẋ = x1/3 with x(0) = 0 (3)

has a solution x(t) = (2t/3)3/2 . This solution is not unique, since x(t) ≡ 0
is another solution. Noting that the right-hand side of (3) is continuous in
x, it is clear that continuity of f (t, x) in its arguments is not sufficient to
ensure uniqueness of the solution. Extra conditions must be imposed on the
function f .

Theorem - Local existence and uniqueness Let f (t, x) be piecewise

continuous in t and satisfy the Lipschitz condition

kf (t, x) − f (t, y)k ≤ Lkx − yk (4)

∀x, y ∈ B = {x ∈ Rn |kx − x0 k ≤ r}, ∀t ∈ [t0 , t1 ]. Then, there exists some
δ > 0 such that the state equation:

ẋ = f (t, x) with x(t0 ) = x0 (5)

2
has a unique solution over [t0 , t0 + δ].
A function satisfying (4) is said to be Lispchitz in x, and the positive
constant L is called a Lipschitz constant. We also use the words locally Lip-
schitz and globally Lipschitz to indicate the domain over which the Lipschitz
condition holds.
We explain locally / globally Lipschitz only for a function f (x) depending
only on the x variable.
A function f (x) is said to be locally Lipschitz on a domain D ⊂ Rn if
each point of D has a neighborhood D0 such that f satisfies the Lipschitz
condition for all points in D0 with some Lipschitz constant L. A locally
Lipschitz function on a domain D is not necessarily Lipschitz on D, since
the Lipschitz condition may not hold uniformly (with the same Lipschitz
constant L) for all points in D.
A function f (x) is said to be globally Lipschitz if it is Lipschitz on Rn .
When f : R → R the Lipschitz condition can be written as:

|f (y) − f (x)|
≤L (6)
|y − x|
which implies that on a plot of f (x) versus x, a straight line joining any
two points of f (x) cannot have a slope whose absolute value is greater than
L. Therefore, any function f (x) that has infinite slope at some point is not
locally Lipschitz at that point. For ex., any discontinuous function is not
locally Lipschitz at that point.

1.2 Locally Lipschitz property

Lemma: Let f : [a, b] × D → Rm be continuous for some domain D ⊂ Rn .
Suppose [∂f /∂x] exists and is continuous on [a, b]×D. If, for a convex subset
W ⊂ D, there is a constant L ≥ 0 such that
° °
° ∂f °
° °
°
° ∂x
(t, x)°≤L
°
(7)

on [a, b] × W , then

kf (t, x) − f (t, y)k ≤ Lkx − yk (8)

for all t ∈ [a, b], x ∈ W and y ∈ W .

3
 Remark: Due to equivalence of norms, the choice of a norm on Rn does
not affect the Lipschitz property of a function. It only affects the value of
the Lipschitz constant.
In the following we show by two examples how to use the above theorem.
Example: The function:
" #
−x1 + x1 x2
f (x) = (9)
x 2 − x1 x2
is continuously differentiable on R2 . On any compact subset of R2 , f is
Lipschitz. Suppose we are interested in calculation a Lipschitz constant over
the convex set:
W = {x ∈ R2 | |x1 | ≤ a1 , |x2 | ≤ a2 }
The Jacobian matrix [∂f /∂x] is given by
" #
∂f −1 + x2 x1
[ ]=
∂x −x2 1 − x1
Using k k∞ for vectors in R2 and the induced matrix norm for matrices,
we have:
° °
° ∂f °
° °
° °
° ∂x °
= max{| − 1 + x2 | + |x1 |, |x2 | + |1 − x1 |}
∞
All points satisfy
| − 1 + x2 | + |x1 | ≤ 1 + a2 + a1
|x2 | + |1 − x1 | ≤ a2 + 1 + a1
Hence ° °
° ∂f °
° °
° ° ≤ 1 + a1 + a2
° ∂x °
∞
and a Lipschitz constant can be taken as L = 1 + a1 + a2 .
Example: Consider the function
" #
x2
f (x) = (10)
−sat(x1 + x2 )
where the saturation function sat(·) is defined by:

 −1 f or
 y < −1
sat(y) = y f or |y| ≤ 1 (11)


1 f or y > 1

4
 The function f is not continuously differentiable on R2 . Let us check its
Lipschitz property by examining f (x) − f (y). Using k · k2 for vectors in R2
and the fact that the saturation function sat(·) satisfies

|sat(η) − sat(ξ)| ≤ |η − ξ| (12)

we obtain

kf (x) − f (y)k22 ≤ (x2 − y2 )2 + (x1 + x2 − y1 − y2 )2

(13)
≤ (x1 − y1 )2 + 2(x1 − y1 )(x2 − y2 ) + 2(x2 − y2 )2
Using the inequality

" #" # Ã" #!

1 1 a 1 1
a2 + 2ab + 2b2 = [a b] ≤ λmax k[a b]k22 (14)
1 2 b 1 2

we conclude that:

kf (x) − f (y)k2 ≤ 1.618kx − yk2 ∀x, y ∈ R2 (15)

Here we have used xT P x ≤ λmax (P )xT x for all x ∈ Rn where P is
symmetric and positive definite and λmax (·) is the maximum eigenvalue of
the matrix.
The local Lipschitz property of a function is basically a smoothness re-
quirement. It is implied by continuous differentiability. Except for discon-
tinuous nonlinearities, which are idealizations of physical phenomena, it is
reasonable to expect models of physical systems to have locally Lipschitz
right-hand side functions.

1.3 Globally Lipschitz property

The theorem given in the previous section guarantees existence and unique-
ness only over an interval [t0 , t0 + δ] where δ may be very small. This means
that the phenomenon that the trajectory (solution) may have a finite escape
time is not taken into account.
For example, consider the scalar system:

ẋ = −x2 with x(0) = −1

5
 The function f (x) = −x2 is locally Lipschitz for all x ∈ R. Hence, it is
Lipschitz on any compact subset of R. The unique solution
1
x(t) =
t−1
exists over [0, 1). As t → 1, x(t) leaves any compact set.
A global Lipschitz condition would take care on the ”finite escape time”,
as shown by the theorem below.
Theorem: (Global existence and uniqueness) Suppose f (t, x) is piece-
wise continuous in t and satisfies :

kf (t, x) − f (t, y)k ≤ Lkx − yk

kf (t, x0 )k ≤ h
∀x, y ∈ Rn , ∀t ∈ [t0 , t1 ]. Then, the state equation

ẋ = f (t, x) with x(t0 ) = x0

has a unique solution over [t0 , t1 ].

The global Lipschitz property is restrictive. One can easily construct
smooth meaningful examples which do not have the global Lipschitz property,
but do have unique global solutions, which is an indication of the conservative
nature of this theorem.
Example: Consider the scalar system

ẋ = −x3 = f (x)

The function f (x) does not satisfy a global Lipschitz condition since the
Jacobian ∂f /∂x = −3x2 is not globally bounded. Nevertheless, for any
initial state x(t0 ) = x0 , the equation has the unique solution:
v
u
u x20
x(t) = sign(x0 )t
1 + 2x20 (t − t0 )

which is well defined for all t ≥ t0 .

However, the global Lipschitz condition may be imposed on a linear sys-
tem, as follows. Consider the linear system:

ẋ = A(t)x + g(t) = f (t, x)

6
 where A(·) and g(·) are piecewise continuous functions of t. Over any
finite interval of time [t0 , t1 ], the elements of A(t) and g(t) are bounded.
Hence kA(t)k ≤ a and kg(t)k ≤ b, where kgk can be any norm on Rn and
kAk is the induced matrix norm. The conditions of the above given theorem
are fulfilled:

1 kf (t, x)−f (t, y)k = kA(t)(x−y)k ≤ kA(t)kkx−yk ≤ akx−yk, ∀x, y ∈ Rn ,

t ∈ [t0 , t1 ]

2 kf (t, x0 )k = kA(t)x0 + g(t)k ≤ akx0 k + b ≤ h for each finite x0 , ∀t ∈ [t0 , t1 ]

1.4 Continuous dependence on initial conditions and

parameters
For the solution of (1) to be of any interest, it must depend continuously
on the initial state x0 , the initial time t0 and the right-hand side function
f (x, t).
Let y(t) be a solution of (1) that starts at y(t0 ) = y0 and is defined on
the compact time interval [t0 , t1 ]. The solution depends continuously on y0
if solutions starting at nearby points are defined on the same interval and
remain close to each other in this interval
Another important problem to study is how f = f (t, x, λ) depends on
some constant parameters λ. This problem is related to the case when the
constant parameters represent physical parameters of the system and the
study of perturbation of these parameters accounts for modeling errors or
changes in the parameter values due to aging. Let x(t, λ0 ) be a solution of
ẋ = f (t, x, λ0 ) defined on [t0 , t1 ], with x(t0 , λ0 ) = x0 . The solution is said
to depend continuously on λ if for any ² > 0, there is δ > 0 such that for
all λ in the ball {λ ∈ Rp |kλ − λ0 k < δ} , the equation ẋ = f (t, x, λ) has
a unique solution x(t, λ) defined on [t0 , t1 ] with x(t0 , λ) = x0 , and satisfies
kx(t, λ) − x(t, λ0 )k < ² for all t ∈ [t0 , t1 ].
More about this subject in the next section.

1.5 Differentiability of solutions and sensitivity equa-

tions
Suppose f (t, x, λ) is continuous in (t, x, λ) and has continuous first partial
derivatives with respect to x and λ for all (t, x, λ) ∈ [t0 , t1 ] × Rn × Rp . Let

7
λ0 be a nominal value of λ, and suppose the nominal state equation

ẋ = f (t, x, λ0 ) with x(t0 ) = x0 (16)

has a unique solution x(t, λ0 ) over [t0 , t1 ]. A theorem proves that for all
λ sufficiently close to λ0 , that is, kλ − λ0 k sufficiently small, the state eq:

ẋ = f (t, x, λ) with x(t0 ) = x0 (17)

has a unique solution x(t, λ) over [t0 , t1 ]:
Z t
x(t, λ) = x0 + f (s, x(s, λ), λ)ds (18)
t0

that is close to the nominal solution x(t, λ0 ).

In this context the sensitivity function S(t) is defined, to provide first
order estimates of the effect of parameter variations on solutions. That is:

x(t, λ) ≈ x(t, λ0 ) + S(t)(λ − λ0 ) (19)

An approach for calculating S(t) is to solve for the nominal solution and
the sensitivity function simultaneously, by numerically solving the system:

ẋ = f (t, x, λ), x(t0 ) = x0

(20)
ẋλ = ( ∂f (t,x,λ)
∂x
)xλ ∂f (t,x,λ)
+ ( ∂λ ), xλ (t0 ) = 0
The sensitivity function S(t) can be calculated by solving (20) at the
nominal value λ = λ0 .
Example: Consider the system:

ẋ1 = x2 = f1 (x1 , x2 )
(21)
ẋ2 = −c sin x1 − (a + b cos x1 )x2 = f2 (x1 , x2 )

and suppose the parameters a, b, c have the nominal values:

a0 = 1 b0 = 0 c0 = 1
The nominal system is given by:

ẋ1 = x2
(22)
ẋ2 = − sin x1 − x2

The Jacobian matrices [∂f /∂x] and [∂f /∂λ] are given by:

8
 " #
∂f 0 1
= (23)
∂x −c cos x1 + bx2 sin x1 −(a + b cos x1 )
" #
∂f h i 0 0 0
∂f ∂f ∂f
= = (24)
∂λ ∂a ∂b ∂c −x2 −x2 cos x1 − sin x1
Evaluate these Jacobian matrices at the nominal parameters to obtain:
" #
∂f 0 1
|nominal = (25)
∂x − cos x1 −1
" #
∂f 0 0 0
|nominal = (26)
∂λ −x2 −x2 cos x1 − sin x1
To solve for S(t) numerically, we solve (20) at the nominal values of the
parameters. Let:
" # " #
∂x1 ∂x1 ∂x1
x3 x5 x7
S= = ∂a
∂x2
∂b
∂x2
∂c
∂x2 |nominal (27)
x4 x6 x8 ∂a ∂b ∂c

Then, (20) is given:

ẋ1 = x2 x1 (0) = x10

ẋ2 = − sin x1 − x2 x2 (0) = x20
ẋ3 = x4 x3 (0) = 0
ẋ4 = −x3 cos x1 − x4 − x2 x4 (0) = 0
(28)
ẋ5 = x6 x5 (0) = 0
ẋ6 = −x5 cos x1 − x6 − x2 cos x1 x6 (0) = 0
ẋ7 = x8 x7 (0) = 0
ẋ8 = −x7 cos x1 − x8 − sin x1 x8 (0) = 0

1.6 Homework
1. For each of the following functions find whether (a) f is locally Lipschitz
and (b) f is globally Lipschitz:

 
" # " # x3 sat(x1 + x2 )
x1 + sgn(x2 ) x1 + sat(x2 )
f (x) = ; f (x) = ; f (x) = 
 x22


x2 x1 + sin x2
x1

9
 2. Derive the sensitivity equations for the system:

ẋ1 = tan−1 (ax1 ) − x1 x2

ẋ2 = bx21 − cx2
as the parameters a, b, c vary from their nominal values a0 = 1, b0 = 0, c0 = 1.

10