Outline Preliminary Definitions Existence and Uniqueness

Nonlinear Control
Lecture 3: Fundamental Properties

Farzaneh Abdollahi

Department of Electrical Engineering

Amirkabir University of Technology

Fall 2010

Farzaneh Abdollahi Nonlinear Control Lecture 3 1/29

 Outline Preliminary Definitions Existence and Uniqueness

Preliminary Definitions
Norm
Set
Continuous Function

Existence and Uniqueness

Existence
Existence and Uniqueness

Farzaneh Abdollahi Nonlinear Control Lecture 3 2/29

 Outline Preliminary Definitions Existence and Uniqueness

I The norm kxk of a vector x is a real-valued function s.t.

1. kxk ≥ 0 ∀x ∈ R n , kxk = 0 iff x = 0
2. kx + y k ≤ kxk + ky k, ∀x, y ∈ R n
3. kaxk = |a|kxk ∀a ∈ R, ∀x ∈ R n
I The class p − norm, p ∈ [1, ∞) are defined by

kxkp = (|x1 |p + ... + |xn |p )1/p

I kxk∞ = max |xi |
i
I The three most common norms are:
1/2
kxk1 , kxk∞ , and the Euclidean norm kxk2 = x T x
I All p-norms are equivalent in the sense that ∃c1 & c2 s.t.:
c1 kxkα ≤ kxkβ ≤ c2 kxkα ∀x ∈ R n
√
e.g.: kxk2 ≤ kxk1 ≤ nkxk2
√
kxk∞ ≤ kxk2 ≤ nkxk∞
kxk∞ ≤ kxk1 ≤ nkxk∞
Farzaneh Abdollahi Nonlinear Control Lecture 3 3/29
 Outline Preliminary Definitions Existence and Uniqueness

I An m × n matrix A defines a linear mapping y = Ax from R n into R m .

The induced p − norm of A is defined by:

kAxkp
kAkp = sup = sup kAxkp = sup kAxkp
x6=0 kxkp kxk≤1 kxk=1

I for p = 1, 2, ∞,
P we have
kAk1 = maxj m i=1 |aij |
 T
1/2
kAk2 = σmax (A)Pn= λmax (A A)
kAk∞ = maxi j=1 |aij |
I we have p
kAk2 ≤ kAk1 kAk∞
1 √
n kAk∞ ≤ kAk2 ≤ √ mkAk∞
1
m kAk1 ≤ kAk2 ≤ nkAk1
kABkp ≤ kAkp kBkp
I Hölder inequality:
1 1
|x T y | ≤ kxkp ky kq , p + q = 1, x, y ∈ R n
Farzaneh Abdollahi Nonlinear Control Lecture 3 4/29
 Outline Preliminary Definitions Existence and Uniqueness

Set

I A set S is closed iff every convergent sequence {xd } with

elements in S converges to a point in S.
I A set S is bounded if there is r > 0 s.t. kxk ≤ r for all x ∈ S.
I A set S is compact if it is closed and bounded.
I A set S is convex: if for every x, y ∈ S and every real number
θ, 0 < θ < 1, the point θx + (1 − θ)y ∈ S.
I In Euclidean space, an object is convex if for every pair of
points within the object, every point on the straight line
segment that joins them is also within the object.

Farzaneh Abdollahi Nonlinear Control Lecture 3 5/29

 Outline Preliminary Definitions Existence and Uniqueness

Set

I A set S is closed iff every convergent sequence {xd } with

elements in S converges to a point in S.
I Convergence: A sequence {xd } ∈ S, a normed linear space,
converges to x, if kxd − xk → 0 as d → ∞
I A set S is bounded if there is r > 0 s.t. kxk ≤ r for all x ∈ S.
I A set S is compact if it is closed and bounded.
I A set S is convex: if for every x, y ∈ S and every real number
θ, 0 < θ < 1, the point θx + (1 − θ)y ∈ S.
I In Euclidean space, an object is convex if for every pair of
points within the object, every point on the straight line
segment that joins them is also within the object.

Farzaneh Abdollahi Nonlinear Control Lecture 3 5/29

 Outline Preliminary Definitions Existence and Uniqueness

Continuous Function
I A function f mapping a set S1 into a set S2 is denoted by f : S1 → S2 .
I f is continuous at x if, given  > 0, there is δ > 0 s.t

kx − y k < δ ⇒ kf (x) − f (y )k <  (1)

I A function f is continuous on set S if it is continuous at every point of S

I A function f is uniformly continuous on S if given  > 0 there is δ > 0
(dependent only on ,not the point in the domain) s.t. (1) holds for all
x, y ∈ S.
I For uniform continuity, the same constant δ works for all points in the set.
I f is uniformly continuous on a set S⇒ it is continuous on S. But the
opposite is not true in general.
I If S is a compact set, then continuity ≡ uniform continuity.

Farzaneh Abdollahi Nonlinear Control Lecture 3 6/29

 Outline Preliminary Definitions Existence and Uniqueness

Continuous Differentiable Function

I A function f : R → R is differentiable at x if
f (x + h) − f (x)
f´(x) = lim
h→0 h
I A function f : R n → R m is continuously differentiable at a point x0
∂fi
if ∂xj
exist and are continuous at x0 for 1 ≤ i ≤ m, 1 ≤ j ≤ n.
I A function f is continuously differentiable on a set S
if it is continuously differentiable at every point of S.

Farzaneh Abdollahi Nonlinear Control Lecture 3 7/29

 Outline Preliminary Definitions Existence and Uniqueness

Existence
I This section provides sufficient condition for uniqueness and existence
solution of the initial value problem

ẋ = f (t, x), x(t0 ) = x0 (2)

I Existence of solution is provided by continuity

I A solution of (2) over an interval [t0 , t1 ]:
x : [t0 , t1 ] −→ R n s.t. ẋ(t) is defined, ẋ(t) = f (t, x(t)) ∀t ∈ [t0 , t1 ]
I If f is continuous in t and x the solution x(t) is continuously
differentiable.
I Assume f is continuous in x but only piecewise continuous in t x(t) is
only piecewise continuously differentiable.
I This allows time-varying input with step changes in time.

Farzaneh Abdollahi Nonlinear Control Lecture 3 8/29

 Outline Preliminary Definitions Existence and Uniqueness

Existence

I A differential equation might have many solutions, e.g.

ẋ = x 1/3 , x(0) = 0 (3)

I x(t) = (2t/3)3/2 and x(t) = 0 the solution is not unique.

I ∴f is continuous continuity is not sufficient to ensure uniqueness.
I Continuity of f guarantees at least one solution.

Farzaneh Abdollahi Nonlinear Control Lecture 3 9/29

 Outline Preliminary Definitions Existence and Uniqueness

Existence and Uniqueness

I Theorem 3.1 (Lipschitz condition: 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 − y k ∀x, y ∈ B = {x ∈ R n |kx − x0 k ≤ r },

∀t ∈ [t0 , t1 ]

Then, there exists δ > 0 such that the state equation ẋ = f (t, x) with
x(t0 ) = x0 has a unique solution over [t0 , t0 + δ].
I The function f satisfying Lipschitz condition is called Lipschitz in x
I The constant L is called the Lipschitz constant.
I A function can be locally or globally Lipschitz.

Farzaneh Abdollahi Nonlinear Control Lecture 3 10/29

 Outline Preliminary Definitions Existence and Uniqueness

Existence and Uniqueness

I A function f (x) is said to be locally Lipschitz on a domain D ⊂ R n

(open and connected set) if each point of D has a neighborhood D0 such
that f (x) satisfies the Lipschitz condition for all points on D0 with some
Lipschitz constant L0 .
I A function f (x) is Lipschitz on a set W if it satisfies Lipschitz condition
for all points with the same Lipschitz constant.
I ∴ A locally Lipschitz function on D is not necessarily Lipschitz on D since
the Lipschitz condition may not hold uniformly (with the same Lipschitz
constant) for all points in D.
I A function f (x) is said to be globally Lipschitz if it is Lipschitz on R n .
I The same terminology holds for f (t, x) if the Lipschitz condition is hold
uniformly in t for all t in a certain interval.

Farzaneh Abdollahi Nonlinear Control Lecture 3 11/29

 Outline Preliminary Definitions Existence and Uniqueness

Existence and Uniqueness

I A function f (t, x) is said to be locally Lipschitz on [a, b] × D ⊂ R × R n if
each point of x ∈ D has a neighborhood D0 such that f (t, x) satisfies the
Lipschitz condition for same Lipschitz constant L0 on [a, b] × D0 .
I If it is true for ∀[a, b] ⊂ [t0 , ∞]=⇒ f is locally Lipschitz on [t0 , ∞] × D.
I If f is scalar, f : R −→ R, the Lipschitz condition can be expressed as:

|f (y ) − f (x)|
≤L
|y − x|
I The line connecting every two points of f , cannot have a slope > L.
I ∴ If a function has infinite slope at some points, the function cannot be
locally Lipschitz at those points.
I Discontinuous functions cannot be locally Lipschitz at the points of
discontinuity.

Farzaneh Abdollahi Nonlinear Control Lecture 3 12/29

 Outline Preliminary Definitions Existence and Uniqueness

Existence and Uniqueness

I Example: f (x) = x 1/3 is not locally Lip. at x = 0 since

f 0 (x) = (1/3)x −2/3 −→ ∞ as x −→ 0.
I If f 0 (x) in some region is bounded by k, then f is lip on that region with
Lip. constant L = k.
I This fact is also true for vector valued functions
I Lemma 3.1: Let f : [a, b] × D −→ R m be continuous for some domain
D ∈ R n . If for a convex subset W ⊂ D there is a constant L ≥ 0 s.t.
∂f
k (t, x)k ≤ L on [a, b] × W ,
∂x
then kf (t, x) − f (t, y )k ≤ Lkx − y k for all t ∈ [a, b], x ∈ W , and y ∈ W .
I ∴ a Lipschitz constant can be calculated using ∂f /∂x

Farzaneh Abdollahi Nonlinear Control Lecture 3 13/29

 Outline Preliminary Definitions Existence and Uniqueness

Proof of Lemma 3.1

1 1
I Let k.kp be any norm p ∈ [1, ∞] and determine q s.t. p + q = 1. Fix t
on [a, b] and assume x ∈ W , y ∈ W .
I Define γ(s) = (1 − s)x + sy , s ∈ R, γ(s) ∈ D,
I W ⊂ D is convex γ(s) ∈ W for 0 ≤ s ≤ 1.
I Take z ∈ Rm s.t.

kzkq = 1, z T [f (t, y ) − f (t, x)] = kf (t, y ) − f (t, x)kp

I set g (s) = z T f (t, γ(s)). Since, g (s) is a continuously differentiable

real-valued function over the open interval which includes [0, 1], from
mean-value theorem , ∃s1 ∈ (0, 1) s.t.

g (1) − g (0) = g 0 (s1 )

Farzaneh Abdollahi Nonlinear Control Lecture 3 14/29

 Outline Preliminary Definitions Existence and Uniqueness

Proof of Lemma 3.1 Cont’d

I Evaluating g at s = 0 and s = 1:
∂f
z T [f (t, y ) − f (t, x)] = z T (t, γ(s1 ))(y − x)
∂x
I and using chain rule in calculating g 0 (s) and Hölder inequality,
|z T w | ≤ kzkq kw kp :

∂f
kf (t, y ) − f (t, x)kp ≤ kzkq (t, γ(s1 ))

ky − xkp ≤ L ky − xkp
∂x p

Farzaneh Abdollahi Nonlinear Control Lecture 3 15/29

 Outline Preliminary Definitions Existence and Uniqueness

Existence and Uniqueness

I If f is Lip. on W , ⇒ it is uniformly continuous on W , (prove it)
but the converse is not true
I The function f (x) = x 1/3 is continuous on R, but it’s not locally lip on
x = 0.
I Lip. condition is weaker than continuous differentiability condition :

Farzaneh Abdollahi Nonlinear Control Lecture 3 16/29

 Outline Preliminary Definitions Existence and Uniqueness

Existence and Uniqueness

∂f
I Lemma 3.2 If f (t, x) and [ ∂x ](t, x) are continuous on [a, b] × D for
n
some domain D ⊂ R , then f is locally Lip. in x on [a, b] × D.
I Proof:
I For x0 ∈ D, let r be so small that the ball D0 = {x ∈ R n |kx − x0 k ≤ r } is
contained in D
I The set D0 is convex and compact
∂f
I By continuity, ∂x is bounded on [a, b] × D0 .
∂f
I Let L0 is a bound for ∂x on [a, b] × D0
I By Lemma 3.1, f (t, x) is Lip. on [a, b] × D0 with Lip. constant L0 .
∂f
I Lemma 3.3: If f (t, x) and [ ∂x ](t, x) are continuous on [a, b] × R n , then
∂f
f is globally Lip. in x on [a, b] × R n iff [ ∂x ] is uniformly bounded on
n
[a, b] × R .
I x(t) is uniformly bounded if ∃c > 0, independent of t0 > 0, and for every
a ∈ (0, c), there is β = β(a) > 0, independent of t0 , s.t.
kx(t0 )k ≤ a⇒kx(t)k ≤ β, ∀t ≥ t0
Farzaneh Abdollahi Nonlinear Control Lecture 3 17/29
 Outline Preliminary Definitions Existence and Uniqueness

Example 1
 
−x1 + x1 x2
f (x) =
x2 − x1 x2
I f is continuously differentiable on R 2 =⇒ f is locally Lip. on R 2 .
∂f
I f is not globally Lip. since ∂x is not uniformly bounded on R 2 .
I However, it is Lip. on any compact set on R 2 .
I Find the Lip. constant on set W = {x ∈ R 2 ||x1 | ≤ a1 , |x2 | ≤ a2 }.
 
−1 + x2 ∂f x1
I fist find jacobian matrix =
−x2 1 − x1
∂x
I Use ∞ norm for vectors and induced norm for matrices:
∂f
k∞ =
k max{| − 1 + x2 | + |x1 |, |x2 | + |1 − x1 |}
∂x
| − 1 + x2 | + |x1 | ≤ 1 + a2 + a1 , |x2 | + |1 − x1 | ≤ a2 + 1 + a1
∂f
k k∞ ≤ 1 + a1 + a2 L0 = 1 + a1 + a2
∂x
Farzaneh Abdollahi Nonlinear Control Lecture 3 18/29
 Outline Preliminary Definitions Existence and Uniqueness

Example 2  
x2
f (x) =
−sat(x1 + x2 )
I f is not continuously differentiable on R 2 .
I Lip. condition is evaluated by definition.
I Use k.k2 and also note that

|sat(η) − sat(ζ)| ≤ |η − ζ|
∴ kf (x) − f (y )k2 ≤ (x2 − y2 )2 + (x1 + x2 − y1 − y2 )2
= (x1 − y1 )2 + 2(x1 − y1 )(x2 − y2 ) + 2(x2 − y2 )2

I We have T         2
a 1 1 a 1 1 a
a2+ 2ab + = 2b 2 ≤ λmax
b 1 2 b 1 2 b
2
√
I ∴kf (x) − f (y )k2 ≤ 2.618kx − y k2 , ∀x, y ∈ R 2
Farzaneh Abdollahi Nonlinear Control Lecture 3 19/29
 Outline Preliminary Definitions Existence and Uniqueness

I If we use the more conservative inequality

a2 + 2ab + 2b 2 ≤ 2a2 + 3b 2 ≤ 3(a2 + b 2 )

√
I The Lip constant 3 is obtained.
I Therefore
I Type of norm does not affect the Lip. property, but it does affect the Lip.
constant
I If the Lip. condition is satisfied for some L0 , it is also hold for all L > L0 .
I Lip. constant is not unique
I Theorem 3.1 is a local theorem
I It guarantees the existence and uniqueness for the interval [t0 , t0 + δ].
I Existence and uniqueness for the interval [t0 , t1 ] is not clear.

Farzaneh Abdollahi Nonlinear Control Lecture 3 20/29

 Outline Preliminary Definitions Existence and Uniqueness

I In general, we cannot extend δ s.t. t + δ = t1

I ∴ there is a maximum interval [t0 , T ] that the unique solution which starts
from (t0 , x0 ) exists.
I T might be smaller than t1 , in this case when t −→ T , the solution leaves
the set on which f is locally Lip.
I Example 3.3 ẋ = −x 2 , x(0) = −1
I f is locally Lip. for all x ∈ R.
I It is locally Lip. on all compact subset of R
1
x(t) = Unique solution on [0, 1]
t −1
I As t −→ 1 x(t) leaves the set.
I Finite escape time indicates that the trajectories go to infinity in finite
time.
I ∴ The trajectory has finite escape time at t = 1

Farzaneh Abdollahi Nonlinear Control Lecture 3 21/29

 Outline Preliminary Definitions Existence and Uniqueness

When the solution interval can be extended indefinitely?

I One way to keep the solution x(t) always in the set: f (t, x) be glob. Lip.
I Theorem 3.2 (Global Existence and Uniqueness) Suppose that
f (t, x) is piecewise continuous in t and satisfies

kf (t, x) − f (t, y )k ≤ Lkx − y k ∀x, y ∈ R n , ∀t ∈ [t0 , t1 ]

Then, ẋ = f (t, x), x(t0 ) = x0 has a unique solution on [t0 , t1 ].

I Example 3.4: ẋ = A(t)x + g (t) = f (t, x)
I where A(t) and g (t) are piecewise continuous functions in t.
I Over any finite interval, elements of A(t) and g (t) are bounded

kA(t)k ≤ a using any induced norm

I All conditions of Theorem 3.2 is satisfied since ∀x, y ∈ R n and t ∈ [t0 , t1 ]:

kf (t, x) − f (t, y )k = kA(t)(x − y )k ≤ kA(t)kkx − y k ≤ akx − y k

Farzaneh Abdollahi Nonlinear Control Lecture 3 22/29
 Outline Preliminary Definitions Existence and Uniqueness

I Example 3.4. Contd.

I Linear System has a unique solution over [t0 , t1 ].
I t1 can be arbitrarily large if A(t) and g (t) are piecewise continuous
functions, system has a unique solution for t ≥ t0 and cannot have ”finite
escape time”.
I The global Lip. condition is reasonable for linear systems.
I In general, it is rarely satisfied for nonlinear systems
I Local Lip. condition is essentially related to smoothness of f
I It is automatically satisfied if f is continuously differentiable
I Except for hard nonlinearities which are idealization of nonlinear
phenomena, physical system models satisfy Lip. condition
I Continuous functions which are not locally Lip. are rare in practice.
I However, the global Lip. condition cannot be satisfied by many physical
systems.

Farzaneh Abdollahi Nonlinear Control Lecture 3 23/29

 Outline Preliminary Definitions Existence and Uniqueness

I Theorem 3.2 provides conservative condition on unique solution of

nonlinear systems
I Example 3.5: ẋ = −x 3 = f (x)
∂f
I f (x) is not globally Lip. since Jacobian ∂x is not bounded in R.
I However, for x(t0 ) = x0 , the unique solution is given by
s
x02
x(t) = sign(x0 )
1 + 2x02 (t − t0 )

I By having some knowledge about the solution x(t), one can prove less
conservative condition for uniqueness using local Lip. condition on f

Farzaneh Abdollahi Nonlinear Control Lecture 3 24/29

 Outline Preliminary Definitions Existence and Uniqueness

I Theorem 3.3: Let f (t, x) is piecewise continuous in t and is locally Lip.

in x for all t ≥ t0 and all x ∈ D ⊂ R n . Let W be a compact subset of D,
x0 ∈ W and every solution of ẋ = f (t, x), x(t0 ) = x0 lies entirely in W .
Then, there is a unique solution that is defined for all t ≥ t0 .
I Proof:
I The proof is based on the fact that if the solution remains in the set W , it
cannot have ”finite escape time”.
I By Theorem 3.1, the unique solution exist in the interval [t0 , t0 + δ]. From
the previous discussion we know that if T is finite, the solution must leave
D, however, since the solution never leaves W , we conclude that T = ∞.
I The problem in applying this theorem is to show that the solution never
leaves the set W .
I We desire to check the assumption that every solution lies in a compact
set without actually solving the differential equation.
I Lyapunov’s stability theorem is an important tool for this purpose.

Farzaneh Abdollahi Nonlinear Control Lecture 3 25/29

 Outline Preliminary Definitions Existence and Uniqueness

Example 3.6:

ẋ = −x 3 = f (x)

I f (x) is locally Lip. on R


x(t) > 0 =⇒ ẋ < 0
I
x(t) < 0 =⇒ ẋ > 0
I Let x(0) = a, and compact set W = {x ∈ R||x| ≤ a}
I It is clear that no solution can leave the set W .
I There is a unique solution for t ≥ 0.

Farzaneh Abdollahi Nonlinear Control Lecture 3 26/29

 Outline Preliminary Definitions Existence and Uniqueness

Summery

I Solution exitance for ẋ = f (x, t) is achieved by continuity or at least

piecewise continuity of function f in t.
I Lipschitz condition can provide sufficient condition for unique solution
I Theorem 3.1: Let f (t, x) be piecewise continuous in t and satisfy the
Lipschitz condition:

kf (t, x) − f (t, y )k ≤ Lkx − y k ∀x, y ∈ B = {x ∈ R n |kx − x0 k ≤ r },

∀t ∈ [t0 , t1 ]

Then, there exists δ > 0 such that the state equation ẋ = f (t, x) with
x(t0 ) = x0 has a unique solution over [t0 , t0 + δ].

Farzaneh Abdollahi Nonlinear Control Lecture 3 27/29

 Outline Preliminary Definitions Existence and Uniqueness

Summery

I Locally Lipschitz
I The condition is satisfied on a subset D ⊂ R n
I It guarantees unique solution over [t0 , t0 + δ]
I A function f (x) is Lipschitz on a set W if it satisfies Lipschitz condition for
all points with the same Lipschitz constant.
I To check the Lipschitz conation a convex subset W ⊂ D, it is sufficient to
∂f
satisfy: k ∂x (t, x)k ≤ L on [a, b] × W .
I To find Lip. constant, L, type of norm does not affect the Lip. property,
but it does affect the Lip. constant.
I Lip. constant is not unique.
I Continuously differentiability of f (t, x) on [a, b] × D guarantees f to be
locally Lip.

Farzaneh Abdollahi Nonlinear Control Lecture 3 28/29

 Outline Preliminary Definitions Existence and Uniqueness

Summery

I Globally Lipschitz
I The condition is satisfied on R n
I It guarantees unique solution over [t0 , t1 ], (no matter how large t1 is)
∂f
I Continuously differentiability of f (t, x)+ uniformly boundedness of ∂x on
n
[a, b] × R guarantees f to be globally Lip.
∂f
I uniformly boundedness of ∂x is a killer condition and difficult to be
satisfied for nonlinear systems in practice.
I By finding a compact subset W in which every solution of ẋ lies entirely,
locally Lip. also guarantees a unique solution for all t ≥ t0 .

Farzaneh Abdollahi Nonlinear Control Lecture 3 29/29