Lecture 2: Mathematical Preliminaries

SCE-517 Control of Nonlinear Systems

M. F. Emzir
2025-09-21
KFUPM, Dhahran, K.S.A.
 Outline

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3. Basic Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4. Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5. Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6. Differentiability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
7. Lipschitz Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
8. Solution of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
9. References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

KFUPM, Dhahran, K.S.A. 1 / 59

1. Introduction
 1. Introduction

● This will be based on Chapter 2 of (Marquez 2003), and Chapter 3 of (Khalil 2002).
● For you that have not studied these topics before, I strongly recommend you to read
through these chapters in details.
● Even better, if you spend some time to study (Rudin 1976). You will find many concepts
in this lecture explained in more details.

KFUPM, Dhahran, K.S.A. 3 / 59

2. Vector Spaces
 2.1 Metric Spaces

Metric spaces generalize the concept of distance from real and complex analysis.
Standard notation is defined for sets of real numbers (ℝ), complex numbers (ℂ), integers
(ℤ), and matrices (ℝ𝑚×𝑛 ).

Definition 1 A metric space is a pair (𝑋, 𝑑) of a non-empty set 𝑋 and a metric or

distance function 𝑑 : 𝑋 × 𝑋 → ℝ such that, for all 𝑥, 𝑦, 𝑧 ∈ 𝑋 the following
conditions hold:
1. 𝑑(𝑥, 𝑦) = 0 if and only if 𝑥 = 𝑦.
2. 𝑑(𝑥, 𝑦) = 𝑑(𝑦, 𝑥).
3. 𝑑(𝑥, 𝑧) ≤ 𝑑(𝑥, 𝑦) + 𝑑(𝑦, 𝑧).

Property (3) is the triangle inequality. The non-negativity of the metric, 𝑑(𝑥, 𝑦) ≥ 0, can be
derived from the other properties (How?)

KFUPM, Dhahran, K.S.A. 5 / 59

 2.2 Vector Spaces

A vector space (or linear space) adds algebraic structure to a set by defining addition of
elements and multiplication by scalars from a field 𝔽 (typically ℝ or ℂ).

Definition 2 A vector space over 𝔽 is a non empty set 𝑋 with a function +𝑣 : 𝑋 ×

𝑋 → 𝑋, and a function ⋅𝑠 : 𝔽 × 𝑋 → 𝑋 such that, for all 𝛼, 𝜇 ∈ 𝔽 and 𝑥, 𝑦, 𝑧 ∈ 𝑋 we
have
1. 𝑥 +𝑣 𝑦 = 𝑦 +𝑣 𝑥 (addition is commutative).
2. 𝑥 +𝑣 (𝑦 +𝑣 𝑧) = (𝑥 +𝑣 𝑦) +𝑣 𝑧 (addition is associative).
3. ∃0 ∈ 𝑋 : 𝑥 +𝑣 0 = 𝑥 (0 is the neutral element in the operation of addition).
4. ∃−𝑥 ∈ 𝑋 : 𝑥 +𝑣 (−𝑥) = 0 (Every 𝑥 ∈ 𝑋 has a negative −𝑥 ∈ 𝑋 such that their
sum is the neutral element defined in (3)).
5. ∃1 ∈ 𝔽 : 1 ⋅𝑠 𝑥 = 𝑥 (1 is the neutral element in the operation of scalar
multiplication).

KFUPM, Dhahran, K.S.A. 6 / 59

 2.2 Vector Spaces

6. 𝛼 ⋅𝑠 (𝑥 +𝑣 𝑦) = 𝛼 ⋅𝑠 𝑥 +𝑣 𝛼 ⋅𝑠 𝑦 (first distributive property).

7. (𝛼 + 𝜇) ⋅𝑠 𝑥 = 𝛼 ⋅𝑠 𝑥 +𝑣 𝜇 ⋅𝑠 𝑥 (second distributive property).
8. 𝛼(𝜇 ⋅𝑠 𝑥) = (𝛼𝜇) ⋅𝑠 𝑥 (scalar multiplication is associative).

The essential feature of the definition is that the set 𝑋 is closed under +𝑣 , ⋅𝑠 .
Example 2.2.1: Example of real vector spaces:
1. ℝ𝑛 ,
2. 𝑃𝑛 be the set of all univariate polynomials of degree at most 𝑛 ∈ ℕ with real
coefficients.
3. 𝒞([0, 1]): the space of continuous function on interval [0, 1]

KFUPM, Dhahran, K.S.A. 7 / 59

 2.3 Linear Independence and Basis

Definition 3 A finite set of vectors {𝑥𝑖 } of a vector space 𝑋 is said to be linearly

dependent if there exists a corresponding set of scalars {𝑎𝑖 }, not all zero, such that

∑ 𝑎𝑖 𝑥𝑖 = 0
𝑖

On the other hand, if ∑𝑖 𝑎𝑖 𝑥𝑖 = 0 implies that 𝑎𝑖 = 0 for each 𝑖, the set {𝑥𝑖 } is said
to be linearly independent.

Definition 4 A basis in a vector space 𝑋, is a set of linearly independent vectors 𝐵

such that every vector in 𝑋 is a linear combination of elements in 𝐵.

Any vector in a finite-dimensional vector space has a unique representation as a linear

combination of basis vectors.
KFUPM, Dhahran, K.S.A. 8 / 59
 2.3 Linear Independence and Basis

Definition 5 The dimension of a vector space 𝑋 is the number of elements in any

of its basis.

Theorem 6 Every set of 𝑛 + 1 vectors in an 𝑛-dimensional vector space 𝑋 is linearly

dependent. A set of 𝑛 vectors in 𝑋 is a basis if and only if it is linearly independent.

KFUPM, Dhahran, K.S.A. 9 / 59

 2.4 Subspaces

Definition 7 A non-empty subset 𝑀 of a vector space 𝑋 is a subspace if for any

pair of scalars 𝛼 and 𝜇, 𝛼𝑥 + 𝜇𝑦 ∈ 𝑀 whenever 𝑥, 𝑦 ∈ 𝑀 .

According to this definition, a subspace 𝑀 in a vector space 𝑋 is itself a vector space.

● In any vector space 𝑋, 𝑋 is a subspace of itself.
● In ℝ3 , any plane or line passing through the origin is a subspace. A subspace must
contain the origin, since for any vector 𝑥 in the subspace, 0 = 1𝑥 − 1𝑥 must also be in
it.

Definition 8 Given a set of vectors 𝑆, in a vector space 𝑋, the intersection of all

subspaces containing 𝑆 is called the subspace generated or spanned by S, or simply
the span of S.

KFUPM, Dhahran, K.S.A. 10 / 59

 2.4 Subspaces

Theorem 9 Let 𝑆 be a set of vectors in a vector space 𝑋. The subspace 𝑀 spanned

by 𝑆 is the set of all linear combinations of the members of 𝑆.

KFUPM, Dhahran, K.S.A. 11 / 59

 2.5 Normed Vector Spaces

Vector spaces provide algebraic structure but lack a notion of distance. A norm is
introduced to a vector space to define this distance, creating a normed vector space.

Definition 10 A normed vector space (or simply a normed space) is a pair

consisting of a vector space 𝑋 and a norm ‖ ⋅ ‖ : 𝑋 → ℝ such that for all 𝑥, 𝑦 ∈ 𝑋
and 𝛼 ∈ ℝ:
1. ‖𝑥‖ = 0 if and only if 𝑥 = 0.
2. ‖𝛼𝑥‖ = |𝛼| ‖𝑥‖.
3. ‖𝑥 + 𝑦‖ ≤ ‖𝑥‖ + ‖𝑦‖.

Notice that the non-negativity of the norm, ‖𝑥‖ ≥ 0, can be derived from the other
properties. This shows that every normed vector space is also a metric space with the
distance defined by 𝑑(𝑥, 𝑦) = ‖𝑥 − 𝑦‖.

KFUPM, Dhahran, K.S.A. 12 / 59

 2.5 Normed Vector Spaces

𝑝-norm: For 𝑥 ∈ ℝ𝑛 and 1 ≤ 𝑝 ≤ ∞, the 𝑝-norm is defined as:

1
𝑛 𝑝

‖𝑥‖𝑝 = (∑|𝑥𝑖 |𝑝 )
𝑖=1

This makes ℝ𝑛 a normed vector space. Important special cases are:

1. 1-norm: ‖𝑥‖1 = ∑𝑛𝑖=1 |𝑥𝑖 |
2. 2-norm (Euclidean): ‖𝑥‖2 = √∑𝑛𝑖=1 |𝑥𝑖 |2
3. ∞-norm (max norm): ‖𝑥‖∞ = max1≤𝑖≤𝑛 |𝑥𝑖 |
● The 2-norm is the most frequently used. Since many properties are independent of the
specific norm, the subscript is often dropped.
● All 𝑝-norms on ℝ𝑛 are equivalent, 𝑘1 , 𝑘2 > 0 such that:
𝑘1 ‖𝑥‖𝑎 ≤ ‖𝑥‖𝑏 ≤ 𝑘2 ‖𝑥‖𝑎 , ∀𝑥 ∈ ℝ𝑛
KFUPM, Dhahran, K.S.A. 13 / 59
 2.5 Normed Vector Spaces

Two frequently used inequalities involving 𝑝-norms in ℝ𝑛 ¹ are the following:

Theorem 11 (Hölder's Inequality) Let 𝑝 > 1 and let 𝑞 be such that 1

𝑝 + 1
𝑞 = 1.
Then for all 𝑥, 𝑦 ∈ ℝ𝑛 :

|𝑥⊤ 𝑦| ≤ ‖𝑥‖𝑝 ‖𝑦‖𝑞 (1)

Theorem 12 (Minkowski's Inequality) Let 𝑝 ≥ 1. Then for all 𝑥, 𝑦 ∈ ℝ𝑛 :

‖𝑥 + 𝑦‖𝑝 ≤ ‖𝑥‖𝑝 + ‖𝑦‖𝑝 (2)

Minkowski’s inequality is simply the triangle inequality for the 𝑝-norm.

¹These two inequalities are valid in other normed vector spaces

KFUPM, Dhahran, K.S.A. 14 / 59
3. Basic Topology
 3. Basic Topology

A few elements of basic topology will be needed. Let 𝑋 be a metric space. We say that:
1. An open ball with center 𝑝 ∈ 𝑋 and radius 𝑟 > 0 is the set 𝐵𝑟 (𝑝) = {𝑞 ∈ 𝑋 : 𝑑(𝑝, 𝑞) <
𝑟}.
2. A point 𝑝 is an interior point of a set 𝐴 ⊂ 𝑋 if there exists 𝐵𝑟 (𝑝) such that 𝐵𝑟 (𝑝) ⊂ 𝐴.
3. An open set is a set where all points are interior points.
4. A neighborhood of a point 𝑝 ∈ 𝑋 is an open set 𝑁𝑟 (𝑝) ⊂ 𝑋 containing 𝑝.
5. Let 𝐴 ⊂ 𝑋 and consider a point 𝑝 ∈ 𝑋. Then 𝑝 is said to be a limit point of 𝐴 if every
neighborhood of 𝑝 contains a point 𝑞 ≠ 𝑝 such that 𝑞 ∈ 𝐴. Note that 𝑝 itself need not
be in the set 𝐴.
6. The complement of 𝐴 ⊂ 𝑋 is the set 𝐴𝑐 = {𝑝 ∈ 𝑋 : 𝑝 ∉ 𝐴}.
7. A set 𝐴 ⊂ 𝑋 is said to be closed if it its complement 𝐴𝑐 is open. Equivalently, 𝐴 is
closed if and only if itcontains all of its limit points.
8. A set 𝐴 ⊂ 𝑋 is bounded if there exist a real number 𝑏 and a point 𝑞 ∈ 𝐴 such that
𝑑(𝑝, 𝑞) < 𝑏 for all 𝑝 ∈ 𝐴.
KFUPM, Dhahran, K.S.A. 16 / 59
 3.1 Basic Topology in ℝ𝑛

The previous concepts can be specialized to the Euclidean space ℝ𝑛 .

● An open ball with center 𝑝 ∈ 𝐴 ⊂ ℝ𝑛 and radius 𝑟 is the set 𝐵𝑟 (𝑝) defined as:
𝐵𝑟 (𝑝) = {𝑥 ∈ ℝ𝑛 : ‖𝑥 − 𝑝‖ < 𝑟}

This is often called .

● Open Set: A set 𝐴 ⊂ ℝ𝑛 is open if for every 𝑝 ∈ 𝐴, one can find an open ball 𝐵𝑟 (𝑝) ⊂
𝐴.
● Bounded Set: A set 𝐴 ⊂ ℝ𝑛 is bounded if there exists a real number 𝑀 > 0 such that:
‖𝑥‖ < 𝑀 , ∀𝑥 ∈ 𝐴
● Compact Set: A set 𝐴 ⊂ ℝ𝑛 is compact if it is closed and bounded.
● Convex Set: A set 𝐴 ⊂ ℝ𝑛 is convex if, for any 𝑥, 𝑦 ∈ 𝐴, the point 𝜃𝑥 + (1 − 𝜃)𝑦 also
belongs to 𝐴 for all 𝜃 such that 0 < 𝜃 < 1.

KFUPM, Dhahran, K.S.A. 17 / 59

4. Sequences
 4. Sequences

Definition 13 A sequence of vectors {𝑥0 , 𝑥1 , 𝑥2 , …} in a metric space (𝑋, 𝑑),

denoted {𝑥𝑛 }, is said to converge if there is a point 𝑥∗ ∈ 𝑋 with the property that for
every real number 𝜀 > 0 there is an integer 𝑁 such that 𝑛 > 𝑁 implies that
𝑑(𝑥𝑛 , 𝑥∗ ) < 𝜀. We then write 𝑥∗ = lim 𝑥𝑛 , or 𝑥𝑛 → 𝑥∗ and call 𝑥∗ the limit of the
sequence {𝑥𝑛 }.

The limit of a convergent sequence must be an element of the space. If a sequence {𝑥𝑛 }
has a limit 𝑥′ such that 𝑥′ ∉ 𝑋, then {𝑥𝑛 } is not convergent in (𝑋, 𝑑).
Example 4.1: Consider the sequence of rational numbers {𝑥𝑛 } = {1, 1.4, 1.41, …} that
√
approaches 2.
√
1. In the space of real numbers ℝ, the sequence converges because its limit, 2, is in ℝ.
2. In the space of rational numbers ℚ, the same sequence does not converge because its
√
limit, 2, is not in ℚ.
KFUPM, Dhahran, K.S.A. 19 / 59
 4. Sequences

Definition 14 A sequence {𝑥𝑛 } in a metric space (𝑋, 𝑑) is said to be a Cauchy

sequence if for every real 𝜀 > 0 there is an integer 𝑁 such that 𝑑(𝑥𝑛 , 𝑥𝑚 ) < 𝜀,
provided that 𝑛, 𝑚 > 𝑁 .

Every convergent sequence is a Cauchy sequence. The converse is not always true, as a
Cauchy sequence may not converge to a point within the space.
Example 4.2: In the space 𝑋 = (0, 1] with the metric 𝑑(𝑥, 𝑦) = |𝑥 − 𝑦|, the sequence
{𝑥𝑛 } = { 𝑛1 } is a Cauchy sequence. However, it is not convergent in 𝑋 because its limit,
lim𝑛→∞ 𝑛1 = 0, is not in 𝑋.
An important class of metric spaces are the so-called complete metric spaces.

KFUPM, Dhahran, K.S.A. 20 / 59

 4. Sequences

Definition 15 A metric space (𝑋, 𝑑) is called complete if and only if every Cauchy
sequence converges (to a point of 𝑋). In other words, (𝑋, 𝑑) is a complete metric
space if for every sequence {𝑥𝑛 } satisfying 𝑑(𝑥𝑛 , 𝑥𝑚 ) < 𝜀 for 𝑛, 𝑚 > 𝑁 there exists
𝑥 ∈ 𝑋 such that 𝑑(𝑥𝑛 , 𝑥) → 0 as 𝑛 → ∞.

Note

Incomplete spaces contain Cauchy sequences that do not converge to a point within the
space. In a complete space, by contrast, every Cauchy sequence converges. This
property is powerful because it allows one to prove the convergence of a sequence by
verifying it is a Cauchy sequence, without needing to know its limit in advance. The set
of real numbers ℝ with the metric 𝑑(𝑥, 𝑦) = |𝑥 − 𝑦| is a fundamental example of a
complete metric space.
KFUPM, Dhahran, K.S.A. 21 / 59
5. Functions
 5. Functions

Definition 16 Let 𝐴 and 𝐵 be abstract sets. A function from 𝐴 to 𝐵 is a set 𝑓 of

ordered pairs in the Cartesian product 𝐴 × 𝐵 with the property that if (𝑎, 𝑏) and
(𝑎, 𝑐) are elements of 𝑓, then 𝑏 = 𝑐.

Functions are also called maps, operators, or transformations.

● The domain of a function 𝑓 is the set of its inputs. The range is the set of its outputs.
● A function 𝑓 is injective (one-to-one) if 𝑓(𝑥1 ) = 𝑓(𝑥2 ) implies 𝑥1 = 𝑥2 .
● It is surjective (onto) if its range equals its codomain 𝐵.
● It is bijective if it is both injective and surjective.
The restriction of a function 𝑓 to a subset 𝐷1 of its domain is a new function defined only
on 𝐷1 .

KFUPM, Dhahran, K.S.A. 23 / 59

 5. Functions

Definition 17 Let 𝐷1 ⊂ ℝ𝑛 , 𝐷2 ⊂ ℝ𝑘 , and consider functions 𝑓1 and 𝑓2 of the form

𝑓1 : 𝐷1 → ℝ𝑘 and 𝑓2 : 𝐷2 → ℝ𝑚 . If 𝑓1 (𝐷1 ) ⊂ 𝐷2 , then the composition of 𝑓2 and 𝑓1
is the function 𝑓2 ∘ 𝑓1 , defined by
(𝑓2 ∘ 𝑓1 )(𝑥) = 𝑓2 (𝑓1 (𝑥)), 𝑥 ∈ 𝐷1

Definition 18 Let (𝑋, 𝑑1 ), (𝑌 , 𝑑2 ) be metric spaces and consider a function 𝑓 :

𝑋 → 𝑌 . We say that 𝑓 is continuous at 𝑥0 if for every real 𝜀 > 0 there exists a real
𝛿 = 𝛿(𝜀, 𝑥0 ) such that 𝑑1 (𝑥, 𝑥0 ) < 𝛿 implies that 𝑑2 (𝑓(𝑥), 𝑓(𝑥0 )) < 𝜀.

Equivalently, a function is continuous if it preserves limits of sequences: if 𝑥𝑛 → 𝑥0 , then

𝑓(𝑥𝑛 ) → 𝑓(𝑥0 ). A function is continuous on a set if it is continuous at every point of that
set.
KFUPM, Dhahran, K.S.A. 24 / 59
 5. Functions

Definition 19 Let (𝑋, 𝑑1 ) and (𝑌 , 𝑑2 ) be metric spaces. Then 𝑓 : 𝑋 → 𝑌 is called

uniformly continuous on 𝑋 if for every 𝜀 > 0 there exists 𝛿 = 𝛿(𝜀) > 0, such that
𝑑1 (𝑥, 𝑦) < 𝛿 implies that 𝑑2 (𝑓(𝑥), 𝑓(𝑦)) < 𝜀, where 𝑥, 𝑦 ∈ 𝑋.

Remark

The key difference is that for continuity, 𝛿 depends on both 𝜀 and the point 𝑥0 , whereas
for uniform continuity, 𝛿 depends only on 𝜀. Uniform continuity is therefore a stronger
condition that implies continuity. The converse is not generally true; for example,
𝑓(𝑥) = 𝑥1 is continuous on (0, ∞) but not uniformly continuous. However, on a compact
set, a function is uniformly continuous if and only if it is continuous.

KFUPM, Dhahran, K.S.A. 25 / 59

 5. Functions

Example 5.1:
1. Uniformly Continuous: The function 𝑓(𝑥) = 2𝑥 is uniformly continuous on ℝ. For any
𝜀 > 0, we can choose 𝛿 = 2𝜀 . Then, if |𝑥 − 𝑦| < 𝛿, we have |𝑓(𝑥) − 𝑓(𝑦)| = |2𝑥 −
2𝑦| = 2|𝑥 − 𝑦| < 2𝛿 = 𝜀. Since 𝛿 depends only on 𝜀, the function is uniformly
continuous.
2. Continuous but Not Uniformly Continuous: The function 𝑓(𝑥) = 𝑥2 is continuous on ℝ
but not uniformly continuous. Let 𝑦 = 𝑥 + 2𝛿 . Then |𝑥 − 𝑦| = 2𝛿 < 𝛿. The condition on
2
the function values becomes |𝑥2 − (𝑥 + 2𝛿 ) | = | − 𝑥𝛿 − 𝛿4 | = 𝛿 |𝑥 + 4𝛿 |. Hence it
2

depends on both 𝛿 AND 𝑥. Say if we choose an 𝜀 = 1, we can always choose a large

enough 𝑥 such that any given 𝛿 > 0, 𝛿|𝑥 + 4𝛿 | > 1. Since we can always find such 𝑥, 𝑦
for any 𝛿, the function is not uniformly continuous.

KFUPM, Dhahran, K.S.A. 26 / 59

 5.1 Bounded Linear Operators and Matrix Norms

Now consider a function 𝐿 mapping a vector space 𝑋 into a vector space 𝑌 .

Definition 20 A function 𝐿 : 𝑋 → 𝑌 is said to be a linear operator (or a linear

map, or a linear transformation) if and only if for any 𝑥1 , 𝑥2 ∈ 𝑋 and any 𝛼, 𝜇 ∈ ℝ:
𝐿(𝛼𝑥1 + 𝜇𝑥2 ) = 𝛼𝐿(𝑥1 ) + 𝜇𝐿(𝑥2 )

The function 𝐿 is said to be a bounded linear operator if there exists a constant 𝑀

such that:
‖𝐿(𝑥)‖ ≤ 𝑀 ‖𝑥‖, ∀𝑥 ∈ 𝑋

The smallest such constant 𝑀 is called the operator norm of 𝐿.

KFUPM, Dhahran, K.S.A. 27 / 59

 5.1 Bounded Linear Operators and Matrix Norms
● For linear operators between finite-dimensional spaces, 𝐿 : ℝ𝑛 → ℝ𝑚 , the operator can
be represented by an 𝑚 × 𝑛 matrix 𝐴 such that 𝐿(𝑥) = 𝐴𝑥. The operator norm
becomes the induced matrix norm:
‖𝐴𝑥‖𝑝
‖𝐴‖𝑝 = sup = sup ‖𝐴𝑥‖𝑝
𝑥≠0 ‖𝑥‖𝑝 ‖𝑥‖𝑝 =1

● This norm is induced by the vector 𝑝-norm. Common cases for 𝑝 = 1, 2, ∞ are:
1. 1-norm (max absolute column sum):
𝑚
‖𝐴‖1 = max ∑|𝑎𝑖𝑗 |
1≤𝑗≤𝑛
𝑖=1
2.
2-norm (spectral norm):

‖𝐴‖2 = √𝜆max (𝐴⊤ 𝐴)

KFUPM, Dhahran, K.S.A. 28 / 59
 5.1 Bounded Linear Operators and Matrix Norms

where 𝜆max (𝐴⊤ 𝐴) is the largest eigenvalue of 𝐴⊤ 𝐴.

3. ∞-norm (max absolute row sum):
𝑛
‖𝐴‖∞ = max ∑|𝑎𝑖𝑗 |
1≤𝑖≤𝑚
𝑗=1

KFUPM, Dhahran, K.S.A. 29 / 59

6. Differentiability
 6. Differentiability

Definition 21 A differentiable mapping 𝑓 of an open set 𝐷 ⊂ ℝ𝑛 into ℝ𝑚 is said to

be continuously differentiable in 𝐷 (written as 𝒞1 (𝐷)) if its derivative ∇𝑓 is a
continuous function on 𝐷.

Proposition 22 (Chain Rule). Let 𝑓1 : 𝐷1 ⊂ ℝ𝑛 → ℝ𝑘 , and 𝑓2 : 𝐷2 ⊂ ℝ𝑘 → ℝ𝑚 . If 𝑓1 is

differentiable at 𝑎 ∈ 𝐷1 and 𝑓2 is differentiable at 𝑏 ∈ 𝐷2 such that 𝑓2 (𝑏) = 𝑎, then 𝑓1 ∘ 𝑓2
is differentiable at 𝑏, and ∇(𝑓1 ∘ 𝑓2 )(𝑏) = ∇𝑓1 (𝑓2 (𝑏))∇𝑓2 (𝑏)

Theorem 23 (Mean-Value Theorem) Let 𝑓 : [𝑎, 𝑏] → ℝ be continuous on the

closed interval [𝑎, 𝑏] and differentiable in the open interval (𝑎, 𝑏). Then there exists a
point 𝑐 ∈ (𝑎, 𝑏) such that
𝑓(𝑏) − 𝑓(𝑎) = ∇𝑓(𝑐)(𝑏 − 𝑎)

KFUPM, Dhahran, K.S.A. 31 / 59

 6. Differentiability

Remark

● The Mean-Value Theorem can be generalized for a differentiable function 𝑓 : 𝐺 ⊂

ℝ𝑛 → ℝ. For any two points 𝑥, 𝑦 in an open set 𝐺 where the line segment between
them is also in 𝐺, there exists a point 𝑧 on that segment such that:
𝑓(𝑦) − 𝑓(𝑥) = ∇𝑓(𝑧) ⋅ (𝑦 − 𝑥)

where ∇𝑓 is the gradient of 𝑓. This is derived by applying the single-variable

theorem to a parametrized function along the line segment connecting 𝑥 and 𝑦.
● A useful extension of this result to functions 𝑓 : ℝ𝑛 → ℝ𝑚 is given below. In the
following theorem Ω represents an open subset of ℝ𝑛 .

KFUPM, Dhahran, K.S.A. 32 / 59

 6. Differentiability

Theorem 24 Consider the function 𝑓 : Ω ⊂ ℝ𝑛 → ℝ𝑚 and suppose that the open

set Ω contains the points 𝑎 and 𝑏 and the line segment 𝑆 joining these points, and
assume that 𝑓 is differentiable at every point of 𝑆. Then there exists a point 𝑐 on 𝑆
such that
‖𝑓(𝑏) − 𝑓(𝑎)‖ ≤ ‖∇𝑓(𝑐)(𝑏 − 𝑎)‖

Example. Consider the function 𝑓(𝑥) = 𝑥2 on ℝ. By the Mean-Value Theorem, for any
𝑥, 𝑦 ∈ ℝ, there exists a point 𝑐 between 𝑥 and 𝑦 such that
𝑓(𝑥) − 𝑓(𝑦) = 𝑓 ′ (𝑐)(𝑥 − 𝑦) = 2𝑐(𝑥 − 𝑦)

To show that 𝑓 is not uniformly continuous, we must show that for some 𝜀 > 0, no single
𝛿 > 0 works for all 𝑥, 𝑦. Let 𝜀 = 1. For any 𝛿 > 0, choose 𝑥 = 1𝛿 and 𝑦 = 𝑥 + 2𝛿 . Then |𝑥 −
𝑦| = 2𝛿 < 𝛿. The point 𝑐 is in (𝑥, 𝑦), so 𝑐 > 𝑥 = 1𝛿 . Then,
KFUPM, Dhahran, K.S.A. 33 / 59
 6. Differentiability

1 𝛿
|𝑓(𝑥) − 𝑓(𝑦)| = |(2𝑐)(𝑥 − 𝑦)| = |2𝑐| |𝑥 − 𝑦| > |2𝑥| |𝑥 − 𝑦| = 2( )( ) = 1 = 𝜀
𝛿 2

Since for any 𝛿 > 0 we can find a pair (𝑥, 𝑦) such that |𝑥 − 𝑦| < 𝛿 but |𝑓(𝑥) − 𝑓(𝑦)| ≥ 𝜀,
the function is not uniformly continuous on ℝ.

Theorem 25 (Inverse Function Theorem) Let 𝑓 : ℝ𝑛 → ℝ𝑛 be 𝒞1 (𝐷) with 𝐷

open, containing the point 𝑥0 ∈ ℝ𝑛 , and let ∇𝑓(𝑥0 ) be invertible. Then there exist an
open set 𝑈0 containing 𝑥0 and an open set 𝑊0 containing 𝑓(𝑥0 ) such that 𝑓 : 𝑈0 →
𝑊0 has a continuous inverse 𝑓 −1 : 𝑊0 → 𝑈0 that is differentiable and for all 𝑦 =
𝑓(𝑥) ∈ 𝑊0 satisfies

∇𝑓 −1 (𝑦) = [∇𝑓(𝑥)]−1

KFUPM, Dhahran, K.S.A. 34 / 59

7. Lipschitz Continuity
 7. Lipschitz Continuity

1. Lipschitz continuity is a stronger form of

continuity that is crucial for studying
differential equations.
2. For a Lipschitz continuous function,
there exists a double cone (white) whose
origin can be moved along the graph so
that the whole graph always stays
outside the double cone
Figure 1: Illustration of a Lipschitz function
on bounded interval

KFUPM, Dhahran, K.S.A. 36 / 59

 7. Lipschitz Continuity

Definition 26 A function 𝑓 : ℝ𝑛 → ℝ𝑚 is said to be locally Lipschitz on an open set

𝐷 ⊂ ℝ𝑛 if every point of 𝐷 has a neighborhood 𝐷0 ⊂ 𝐷 over which the restriction of
𝑓 satisfies
‖𝑓(𝑥1 ) − 𝑓(𝑥2 )‖ ≤ 𝐿 ‖𝑥1 − 𝑥2 ‖, ∀𝑥1 , 𝑥2 ∈ 𝐷0 (3)

for some constant 𝐿 > 0.

It is said to be Lipschitz on 𝐷 if it satisfies (3) for all 𝑥1 , 𝑥2 ∈ 𝐷 with the same
Lipschitz constant 𝐿. Finally, 𝑓 is said to be globally Lipschitz if it satisfies (3) with
𝐷 = ℝ𝑛 .

KFUPM, Dhahran, K.S.A. 37 / 59

 7. Lipschitz Continuity

Note

If 𝑓 : ℝ𝑛 → ℝ𝑚 is Lipschitz on 𝐷 ⊂ ℝ𝑛 , then, given 𝜀 > 0, we can define 𝛿 = 𝐿,

𝜀
and
we have that
𝜀
‖𝑥1 − 𝑥2 ‖ < 𝛿 ⇒ ‖𝑓(𝑥1 ) − 𝑓(𝑥2 )‖ ≤ 𝐿 ‖𝑥1 − 𝑥2 ‖ < 𝐿( ) = 𝜀, 𝑥1 , 𝑥2 ∈ 𝐷
𝐿
which implies that 𝑓 is uniformly continuous. However, the converse is not true; that is,
not every uniformly continuous function is Lipschitz. The next theorem gives an
important sufficient condition for Lipschitz continuity.

KFUPM, Dhahran, K.S.A. 38 / 59

 7. Lipschitz Continuity

Theorem 27 If a function 𝑓 : ℝ𝑛 → ℝ𝑚 is continuously differentiable on an open set

𝐷 ⊂ ℝ𝑛 , then it is locally Lipschitz on 𝐷.

Proof. Consider 𝑥0 ∈ 𝐷 and let 𝑟 > 0 be small enough to ensure that the closed ball
𝐵̄ 𝑟(𝑥0 ) = {𝑥 ∈ 𝐷 : ‖𝑥 − 𝑥0 ‖ ≤ 𝑟} is contained in 𝐷. Consider two arbitrary points 𝑥1 , 𝑥2 ∈
𝐵̄ 𝑟(𝑥0 ) . Since the closed ball is a convex set, the line segment 𝛾(𝜃) = 𝜃𝑥1 + (1 − 𝜃)𝑥2 , for
𝜃 ∈ [0, 1], is contained in 𝐵̄ 𝑟(𝑥0 ) .
Now define the function 𝜑 : [0, 1] → ℝ𝑚 as 𝜑(𝜃) = 𝑓(𝛾(𝜃)). By the Mean-Value Theorem,
there exists a point 𝑐 on the line segment joining 𝑥1 and 𝑥2 such that
‖𝑓(𝑥1 ) − 𝑓(𝑥2 )‖ ≤ ‖∇𝑓(𝑐)(𝑥1 − 𝑥2 )‖

Using the property of induced matrix norms, we get

KFUPM, Dhahran, K.S.A. 39 / 59

 7. Lipschitz Continuity
‖𝑓(𝑥1 ) − 𝑓(𝑥2 )‖ ≤ ‖∇𝑓(𝑐)‖ ‖𝑥1 − 𝑥2 ‖

Since 𝑓 is continuously differentiable on the compact set 𝐵̄ 𝑟(𝑥0 ) , its derivative ∇𝑓 is

bounded on this set. Let 𝐿 = sup𝑥∈𝐵̄ 𝑟(𝑥 ) ‖∇𝑓(𝑥)‖. Then we have
0

‖𝑓(𝑥1 ) − 𝑓(𝑥2 )‖ ≤ 𝐿 ‖𝑥1 − 𝑥2 ‖

This shows that 𝑓 is locally Lipschitz on 𝐷. ∎

Definition 28 A function 𝑓(𝑥, 𝑡) : ℝ𝑛 × ℝ → ℝ𝑛 is said to be locally Lipschitz in 𝑥

on an open set 𝐷 × [𝑡0 , 𝑇 ] ⊂ ℝ𝑛 × ℝ if every point of 𝐷 has a neighborhood 𝐷1 ⊂
𝐷 over which the restriction of 𝑓 with domain 𝐷1 × [𝑡0 , 𝑇 ] satisfies (3). It is said to
be Lipschitz in 𝑥 on 𝐷 × [𝑡0 , 𝑇 ] if it satisfies (3) for all 𝑥1 , 𝑥2 ∈ 𝐷 and all 𝑡 ∈ [𝑡0 , 𝑇 ].

KFUPM, Dhahran, K.S.A. 40 / 59

 7. Lipschitz Continuity

Theorem 29 Let 𝑓 : ℝ𝑛 × ℝ → ℝ𝑛 be continuously differentiable with respect to 𝑥

on a convex set 𝐷 × [𝑡0 , 𝑇 ] and assume that the Jacobian matrix of 𝑓 with respect to 𝑥
satisfies
‖ ∇𝑓(𝑥, 𝑡) ‖ ≤ 𝐿

on 𝐷 × [𝑡0 , 𝑇 ]. Then 𝑓 is Lipschitz continuous in 𝑥 on 𝐷 with constant 𝐿:

‖𝑓(𝑥, 𝑡) − 𝑓(𝑦, 𝑡)‖ ≤ 𝐿 ‖𝑥 − 𝑦‖, ∀𝑥, 𝑦 ∈ 𝐷, ∀𝑡 ∈ [𝑡0 , 𝑇 ]

Example 7.1 (Example 3.1 K): The function

−𝑥1 + 𝑥1 𝑥2
𝑓(𝑥) = [ ].
𝑥2 − 𝑥1 𝑥2

KFUPM, Dhahran, K.S.A. 41 / 59

 7. Lipschitz Continuity

The function 𝑓 ∈ 𝒞1 (ℝ2 ). Hence, it is locally Lipschitz on ℝ2 . It is not globally Lipschitz

since ∇𝑓 is not uniformly bounded on ℝ2 . On any compact subset of ℝ2 , 𝑓 is Lipschitz.
Suppose that we are interested in calculating a Lipschitz constant over the convex set
𝑊 = {𝑥 ∈ ℝ2 | |𝑥1 | ≤ 𝑎1 , |𝑥2 | ≤ 𝑎2 }. The Jacobian matrix is given by

−1 + 𝑥2 𝑥1
∇𝑓 = [ ]
−𝑥2 1 − 𝑥1

Using ‖ ⋅ ‖∞ for vectors in ℝ2 and the induced matrix norm for matrices, we have
‖ ∇𝑓 ‖∞ = max{| − 1 + 𝑥2 | + |𝑥1 |, |𝑥2 | + |1 − 𝑥1 |}

All points in 𝑊 satisfy

| − 1 + 𝑥2 | + |𝑥1 | ≤ 1 + |𝑥2 | + |𝑥1 | ≤ 1 + 𝑎2 + 𝑎1

and
KFUPM, Dhahran, K.S.A. 42 / 59
 7. Lipschitz Continuity

|𝑥2 | + |1 − 𝑥1 | ≤ |𝑥2 | + 1 + |𝑥1 | ≤ 𝑎2 + 1 + 𝑎1

Hence,
‖ ∇𝑓 ‖∞ ≤ 1 + 𝑎1 + 𝑎2

and a Lipschitz constant can be taken as 𝐿 = 1 + 𝑎1 + 𝑎2 .

Example 7.2 (Example 3.2 K): The function

𝑥2
𝑓(𝑥) = [ ]
−sat(𝑥1 + 𝑥2 )

where the saturation function is defined as

𝑢, if |𝑢| ≤ 1
sat(𝑢) = {
sgn(𝑢), if |𝑢| > 1

KFUPM, Dhahran, K.S.A. 43 / 59

 7. Lipschitz Continuity

is not continuously differentiable on ℝ2 (why?). Is it Lipscitz: Using ‖ ⋅ ‖2 for vectors in ℝ2

and the fact that the saturation function sat(⋅) satisfies
|sat(𝜂) − sat(𝜉)| ≤ |𝜂 − 𝜉|

we obtain

‖𝑓(𝑥) − 𝑓(𝑦)‖22 = (𝑥2 − 𝑦2 )2 + (−sat(𝑥1 + 𝑥2 ) + sat(𝑦1 + 𝑦2 ))2

≤ (𝑥2 − 𝑦2 )2 + ((𝑥1 + 𝑥2 ) − (𝑦1 + 𝑦2 ))2
= (𝑥2 − 𝑦2 )2 + ((𝑥1 − 𝑦1 ) + (𝑥2 − 𝑦2 ))2

Let 𝑎 = 𝑥1 − 𝑦1 and 𝑏 = 𝑥2 − 𝑦2 . Then the expression becomes

𝑏2 + (𝑎 + 𝑏)2 = 𝑎2 + 2𝑎𝑏 + 2𝑏2

Using the inequality

KFUPM, Dhahran, K.S.A. 44 / 59
 7. Lipschitz Continuity

⊤ 2
𝑎 1 1 𝑎 1 1 𝑎
𝑎2 + 2𝑎𝑏 + 2𝑏2 = [ ] [ ][ ] ≤ 𝜆max ([ ])‖[ ]‖
𝑏 1 2 𝑏 1 2 𝑏
2
√
1 1
where 𝜆max ([1 2]) = 2≈ 2.618. We conclude that
3+ 5

√
‖𝑓(𝑥) − 𝑓(𝑦)‖2 ≤ 2.618 ‖𝑥 − 𝑦‖2 , ∀𝑥, 𝑦 ∈ ℝ2

Here we have used a property of positive semidefinite symmetric matrices; that is,
𝑧 ⊤ 𝑃 𝑧 ≤ 𝜆max (𝑃 )𝑧 ⊤ 𝑧, for all 𝑧 ∈ ℝ𝑛 , where 𝜆max (⋅) is the maximum eigenvalue of the
matrix. A more conservative (larger) Lipschitz constant will be obtained if we use the more
conservative inequality

𝑎2 + 2𝑎𝑏 + 2𝑏2 ≤ 2𝑎2 + 3𝑏2 ≤ 3(𝑎2 + 𝑏2 )

√
resulting in a Lipschitz constant 𝐿 = 3.
KFUPM, Dhahran, K.S.A. 45 / 59
8. Solution of Differential Equations
 8. Solution of Differential Equations

● Linear ODEs, like

𝑥̇ = 𝐴𝑥 + 𝐵𝑢,

have closed-form solutions, e.g.,

𝑡
𝑥(𝑡) = 𝑒𝐴𝑡 𝑥0 + ∫ 𝑒𝐴(𝑡−𝜏) 𝐵𝑢(𝜏 )𝑑𝜏
0

.
● In contrast, nonlinear differential equations generally lack closed-form solutions.
● For nonlinear systems, two primary concerns are the existence and uniqueness of
[Link] section focuses on deriving sufficient conditions for the existence and
uniqueness of the solution to the initial value problem:
𝑥̇ = 𝑓(𝑥, 𝑡), 𝑥(𝑡0 ) = 𝑥0 (4)
KFUPM, Dhahran, K.S.A. 47 / 59
 8.1 Contraction Mapping

Definition 30 Let (𝑋, 𝑑) be a metric space, and let 𝑆 ⊂ 𝑋. A mapping 𝑓 : 𝑆 → 𝑆

is said to be a contraction on 𝑆 if there exists a number 𝛾 < 1 such that
𝑑(𝑓(𝑥), 𝑓(𝑦)) ≤ 𝛾𝑑(𝑥, 𝑦) (5)

for any two points 𝑥, 𝑦 ∈ 𝑆.

It is a straightforward exercise that every contraction is continuous (in fact, uniformly

⊤
continuous) on 𝑋. Example of a contraction on ℝ2 is 𝑓(𝑥) = [ 12 𝑥1 , − 12 𝑥2 ] .

Theorem 31 (Contraction Mapping Principle) Let 𝑆 be a closed subset of a

complete metric space (𝑋, 𝑑). Every contraction mapping 𝑓 : 𝑆 → 𝑆 has one and
only one point 𝑥 ∈ 𝑆 such that 𝑓(𝑥) = 𝑥.

KFUPM, Dhahran, K.S.A. 48 / 59

 8.2 Local Existence and Uniqueness of Solutions

Theorem 32 (Local Existence and Uniqueness) Consider the nonlinear

differential equation (4) and assume that 𝑓(𝑥, 𝑡) is piecewise continuous in 𝑡 and
satisfies the Lipschitz condition
‖𝑓(𝑥1 , 𝑡) − 𝑓(𝑥2 , 𝑡)‖ ≤ 𝐿 ‖𝑥1 − 𝑥2 ‖

for all 𝑥1 , 𝑥2 ∈ 𝐵 = {𝑥 ∈ ℝ𝑛 : ‖𝑥 − 𝑥0 ‖ ≤ 𝑟} and for all 𝑡 ∈ [𝑡0 , 𝑡1 ]. Then, there exists

some 𝛿 > 0 such that (4) has a unique solution in [𝑡0 , 𝑡0 + 𝛿].

Proof.
● The proof uses Theorem 31 to show the existence and uniqueness of a solution.
● Step 1: Reformulate the problem.
1.
The ODE 𝑥̇ = 𝑓(𝑥, 𝑡) with 𝑥(𝑡0 ) = 𝑥0 is equivalent to the integral equation:

KFUPM, Dhahran, K.S.A. 49 / 59

 8.2 Local Existence and Uniqueness of Solutions
𝑡
𝑥(𝑡) = 𝑥0 + ∫ 𝑓(𝑥(𝜏 ), 𝜏 )𝑑𝜏
𝑡0

2. Let 𝑋 = 𝒞([𝑡0 , 𝑡0 + 𝛿]) be the space of continuous functions on [𝑡0 , 𝑡0 + 𝛿] with the
sup norm ‖𝑥‖𝑐 = sup𝑡∈[𝑡0 ,𝑡0 +𝛿] ‖𝑥(𝑡)‖. This is a complete metric space. Now define
the operator 𝐹 : 𝑋 → 𝑋, 𝑦 ↦ 𝐹 (𝑦) as
𝑡
(𝐹 (𝑦))(𝑡) = 𝑥0 + ∫ 𝑓(𝑦(𝜏 ), 𝜏 )𝑑𝜏
𝑡0

3. It turns out that 𝑥 is the fixed point of 𝐹 ; i.e., 𝑥 = 𝐹 (𝑥).

● Step 2: Define the space and the operator.
1. Define a closed subset 𝑆 = {𝑧 ∈ 𝑋 : ‖𝑧 − 𝑥0 𝟙‖𝑐 ≤ 𝑟}. 𝑆 is also a complete metric
space. Here, 𝟙 is the continuous function that is identically equal to 1 on [𝑡0 , 𝑡0 + 𝛿].
● Step 3: Show 𝐹 maps 𝑆 to itself (𝐹 : 𝑆 → 𝑆).
1. For any 𝑧 ∈ 𝑆, we need to show 𝐹 (𝑧) ∈ 𝑆, i.e., ‖𝐹 (𝑧) − 𝑥0 𝟙‖𝑐 ≤ 𝑟.
KFUPM, Dhahran, K.S.A. 50 / 59
 8.2 Local Existence and Uniqueness of Solutions
2. 𝑡
‖𝐹 (𝑧) − 𝑥0 𝟙‖𝑐 = sup ‖𝐹 (𝑧)(𝑡) − 𝑥0 ‖ ≤ sup ∫ ‖𝑓(𝑧(𝜏 ), 𝜏 )‖ 𝑑𝜏
{𝑡∈[𝑡0 ,𝑡0 +𝛿]} {𝑡∈[𝑡0 ,𝑡0 +𝛿]} 𝑡0

3. Using the triangle inequality and Lipschitz condition: ‖𝑓(𝑧(𝑡), 𝑡)‖ ≤ ‖𝑓(𝑧(𝑡), 𝑡) −
𝑓(𝑥0 , 𝑡)‖ + ‖𝑓(𝑥0 , 𝑡)‖ ≤ 𝐿‖𝑧(𝑡) − 𝑥0 ‖ + 𝑀 where 𝑀 is a bound on ‖𝑓(𝑥0 , 𝑡)‖.
4. Hence:
𝑡
‖𝐹 (𝑧) − 𝑥0 𝟙‖𝑐 ≤ sup ∫ (𝐿‖𝑧(𝜏 ) − 𝑥0 ‖ + 𝑀 )𝑑𝜏
{𝑡∈[𝑡0 ,𝑡0 +𝛿]} 𝑡0

𝑡
≤ sup ∫ (𝐿𝑟 + 𝑀 )𝑑𝜏
{𝑡∈[𝑡0 ,𝑡0 +𝛿]} 𝑡0

= 𝛿(𝐿𝑟 + 𝑀 )
5. To ensure ‖𝐹 (𝑧) − 𝑥0 𝟙‖𝑐 ≤ 𝑟, we need to choose 𝛿 such that 𝛿 ≤ 𝐿𝑟+𝑀 .
𝑟

● Step 4: Show 𝐹 is a contraction on 𝑆.

KFUPM, Dhahran, K.S.A. 51 / 59
 8.2 Local Existence and Uniqueness of Solutions
1. We need to show ‖𝐹 (𝑦) − 𝐹 (𝑧)‖𝑐 ≤ 𝛾 ‖𝑦 − 𝑧‖𝑐 for any 𝑦, 𝑧 ∈ 𝑆 and some 𝛾 < 1.
2. 𝑡
‖((𝐹 (𝑧))(𝑡) − (𝐹 (𝑦))(𝑡)‖ ≤ ∫ ‖𝑓(𝑦(𝜏 ), 𝜏 ) − 𝑓(𝑧(𝜏 ), 𝜏 )‖ 𝑑𝜏
𝑡0

3. Using the Lipschitz property of 𝑓:

𝑡
‖((𝐹 (𝑧))(𝑡) − (𝐹 (𝑦))(𝑡)‖ ≤ ∫ 𝐿 ‖𝑦(𝜏 ) − 𝑧(𝜏 )‖ 𝑑𝜏 ≤ 𝐿𝛿 ‖𝑦 − 𝑧‖
𝑡0

4. Taking the sup norm: ‖𝐹 (𝑦) − 𝐹 (𝑧)‖𝑐 ≤ 𝐿𝛿 ‖𝑦 − 𝑧‖𝑐 .

5. For 𝐹 to be a contraction, we need 𝐿𝛿 < 1, which means 𝛿 < 𝐿1 .
● Step 5: Conclusion.
1. By choosing a small enough 𝛿 > 0 that satisfies both 𝛿 < 𝐿1 and 𝛿 ≤ 𝐿𝑟+𝑀 𝑟
, the
operator 𝐹 becomes a contraction on the complete metric space 𝑆.
2. By the Contraction Mapping Theorem, 𝐹 has a unique fixed point in 𝑆.
3. This fixed point is the unique local solution to the ODE on the interval [𝑡0 , 𝑡0 + 𝛿].
KFUPM, Dhahran, K.S.A. 52 / 59
 8.2 Local Existence and Uniqueness of Solutions

Note

Theorem 32 provides a sufficient, not necessary, condition for the local existence and
uniqueness of the solution to (4). Theorem 33 offers a stronger, more conservative
condition for global existence and uniqueness.

KFUPM, Dhahran, K.S.A. 53 / 59

 8.3 Global Existence and Uniqueness of Solutions

Theorem 33 (Global Existence and Uniqueness) Consider again the nonlinear

differential equation (4) and assume that 𝑓(𝑥, 𝑡) is piecewise continuous in 𝑡 and
satisfies the global Lipschitz condition
‖𝑓(𝑥1 , 𝑡) − 𝑓(𝑥2 , 𝑡)‖ ≤ 𝐿 ‖𝑥1 − 𝑥2 ‖

for all 𝑥1 , 𝑥2 ∈ ℝ𝑛 and for all 𝑡 ∈ [𝑡0 , 𝑡1 ]. Then, (4) has a unique solution in [𝑡0 , 𝑡1 ].

Proof. See (Khalil 2002) Appendix C.1. ∎

KFUPM, Dhahran, K.S.A. 54 / 59

 8.3 Global Existence and Uniqueness of Solutions

Remark

This theorem is often too conservative to be of practical use. The main condition is the
existence of a global Lipschitz constant, which is very restrictive. The local Lipschitz
condition of Theorem 32, on the other hand, is not very restrictive and is satisfied by
any function 𝑓(𝑥, 𝑡) satisfying mild smoothness conditions.

Example 8.3.1 (Example 3.3 K): Consider the scalar system 𝑥̇ = −𝑥2 , with 𝑥(0) = −1.
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 𝑥(𝑡) = 𝑡−1 1
exists over [0, 1). As 𝑡 → 1, 𝑥(𝑡)
leaves any compact set.

KFUPM, Dhahran, K.S.A. 55 / 59

 8.3 Global Existence and Uniqueness of Solutions

Note

“Finite escape time” describes a solution reaching infinity in finite time, as in Example
8.3.1. To guarantee a solution can be extended indefinitely, it must remain in a set
where 𝑓(𝑥, 𝑡) is uniformly Lipschitz. Theorem 33 uses a global Lipschitz condition to
establish the existence of a unique solution over an arbitrarily large interval [𝑡0 , 𝑡1 ].

Example 8.3.2 (Example 3.4 K): Consider the linear system

𝑥̇ = 𝐴(𝑡)𝑥 + 𝑔(𝑡) = 𝑓(𝑥, 𝑡) (6)

where 𝐴(𝑡) and 𝑔(𝑡) are piecewise continuous functions of 𝑡. Over any [𝑡0 , 𝑡1 ], the
elements of 𝐴(𝑡) are bounded. Hence, ‖𝐴(𝑡)‖ ≤ 𝑎, where ‖ ⋅ ‖ is any induced matrix
norm. The conditions of Theorem 33 are satisfied since (why?)

KFUPM, Dhahran, K.S.A. 56 / 59

 8.3 Global Existence and Uniqueness of Solutions

‖𝑓(𝑥, 𝑡) − 𝑓(𝑦, 𝑡)‖ = ‖𝐴(𝑡)(𝑥 − 𝑦)‖ ≤ ‖𝐴(𝑡)‖ ‖𝑥 − 𝑦‖ ≤ 𝑎 ‖𝑥 − 𝑦‖

for all 𝑥, 𝑦 ∈ ℝ𝑛 and 𝑡 ∈ [𝑡0 , 𝑡1 ]. Therefore, Theorem 33 shows that (6) has a unique
solution over [𝑡0 , 𝑡1 ]. But 𝑡1 can be arbitrarily large, we can also conclude that if 𝐴(𝑡) and
𝑔(𝑡) are piecewise continuous for all 𝑡 ≥ 𝑡0 , then the system has a unique solution for all
𝑡 ≥ 𝑡0 . Hence, the system cannot have a finite escape time.

KFUPM, Dhahran, K.S.A. 57 / 59

9. References
 9. References

Khalil H. 2002. Nonlinear Systems. Prentice Hall

Marquez HJ. 2003. Nonlinear Control Systems: Analysis and Design. Hoboken, N.J: John
Wiley
Rudin W. 1976. Principles of Mathematical Analysis, Vol. 39. McGraw-Hill

KFUPM, Dhahran, K.S.A. 59 / 59