Notes on Dynamical Systems

Preliminary Lecture Notes

c Draft date: April 12, 2019

Adolfo J. Rumbos

April 12, 2019

2
Contents

1 Introduction 5
1.1 Integral Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Fundamental Existence Theory 9

2.1 Setup for Proof of Local Results . . . . . . . . . . . . . . . . . . 10
2.2 Proof of Local Existence and Uniqueness . . . . . . . . . . . . . . 13
2.3 Extension of Solutions . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Continuous Dependence on Initial Conditions . . . . . . . . . . . 27

3 Flows 33
3.1 Flow Domains and Flow Maps . . . . . . . . . . . . . . . . . . . 34
3.2 Properties of Flow Maps . . . . . . . . . . . . . . . . . . . . . . . 35

4 Continuous Dynamical Systems 47

4.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2 Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3 Infinitesimal Generators . . . . . . . . . . . . . . . . . . . . . . . 51
4.4 Fixed Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.5 Limit Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.6 Properties of Limit Sets . . . . . . . . . . . . . . . . . . . . . . . 56
4.7 Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.8 Limit Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.9 Liapunov Stability . . . . . . . . . . . . . . . . . . . . . . . . . . 76

5 Two–Dimensional Systems 85
5.1 Analysis of the Lienard System: Part I . . . . . . . . . . . . . . . 86
5.2 Analysis of the Lienard System: Part II . . . . . . . . . . . . . . 88

A Definitions and Facts from Real Analysis 89

A.1 Topology of Euclidean Space . . . . . . . . . . . . . . . . . . . . 89
A.2 Sequences in Euclidean Space . . . . . . . . . . . . . . . . . . . . 92
A.3 Properties of Continuous Functions . . . . . . . . . . . . . . . . . 92

3
4 CONTENTS
Chapter 1

Introduction

These notes provide an introduction to the theory of dynamical systems. We

will begin by proving the fundamental existence and uniqueness theorem for
initial value problem for a system of first–order, ordinary differential equations.
We will then proceed to establish results dealing with continuous dependence
on initial conditions and parameters, and the question of extending solutions.
The main goal on this part of the course will be the definition of a flow of a
vector field in RN . We will then study properties of flows, which are also known
as dynamical systems.

1.1 Integral Curves

We will presently formulate one of the fundamental question in the study of
dynamical systems that we will be studying in this course.
Let U be an open subset in RN and I be an open interval in R. Consider a
vector valued function, F : I × U → RN , defined on I × U , given by
 
f1 (t, x1 , x2 , . . . , xN )
 f2 (t, x1 , x2 , . . . , xN ) 
F (t, x1 , x2 , . . . , xN ) =  , (1.1)
 
..
 . 
fN (t, x1 , x2 , . . . , xN )

where f1 , f2 , . . . , fN are real valued functions defined on I × U . F is called a

(time dependent) vector field defined on U ; at each point (x1 , x2 , . . . , xN ) in the
open set U , and at “time” t ∈ I, F (t, x1 , x2 , . . . , xN ) gives a vector in RN .
Let J be an open interval contained in I. A vector–valued function

u: J → U

given by
u(t) = (x1 (t), x2 (t), . . . , xN (t)),

5
6 CHAPTER 1. INTRODUCTION

is said to be a C 1 curve in U if all the component functions

xi : J → R, for i = 1, 2, . . . , N,

are differentiable with continuous derivatives, x0i (t), for i = 1, 2, . . . , N . The

vector  0 
x1 (t)
 x02 (t) 
u0 (t) =  . 
 
 . .
x0N (t)
is called the tangent vector to the curve u at u(t), for t ∈ J.
The first question that we would like to answer is the the following: Given
a point po ∈ U and a time to ∈ I, is there a C 1 curve, u : J → U , defined on
some subinterval, J, of I which contains to , such that

u(to ) = po ,

and
u0 (t) = F (t, u(t)), for all t ∈ J?
In other words, given a point, po , in U , is it possible to find a C 1 curve through
po when t = to , and such that its tangent vectors are prescribed by the vector
field F ? If this is the case, we say that x(t) = u(t), for t ∈ J, solves the initial
value problem (IVP)
dx


 = F (t, x);
dt (1.2)


x(to ) = po .
Here, x = x(t) denotes the vector valued function
 
x1 (t)
 x2 (t) 
x(t) =  .  ,
 
 .. 
xN (t)
dx
and is the tangent vector at t,
dt
x01 (t)
 
0
dx   x2 (t) 

=  . .
dt  .. 
x0N (t)

If a C 1 curve, u : J → U , satisfying the IVP (1.2) exists, we will call it an

integral curve of the vector field F going through the point po when t = to . We
will see in subsequent chapters that for an integral curve to exist, it is sufficient
that the vector field, F , be continuous in a neighborhood of the point (to , po ).
1.2. FLOWS 7

Example 1.1.1. In this one–dimensional example, we let U = R, to = 0,

po = 0, and
F (t, x) = tx1/3 , for all (t, x) ∈ R2 .
We find a solution to the IVP
dx


 = F (t, x);
dt (1.3)


x(0) = 0,

by first separating variables in the ODE

dx
= tx1/3
dt
to get
3x2/3 = t2 + c, (1.4)
for some arbitrary constant, c. Using the initial condition in the IVP (1.3), we
obtain from (1.4) that c = 0, so that

t3
x(t) = √ , for t ∈ R
3 3
solves the IVP (1.3).
Note that the function defined by

x(t) = 0, for all t ∈ R,

also solves the IVP (1.3). Thus, uniqueness of a solution to the IVP (1.2) is not,
in general, guaranteed for a continuous vector field, F .

1.2 Flows
One of our goals in this course is to define the flow of a vector field, F . Heuris-
tically, the flow of a vector field, F : I × U → U , where I is an open interval
containing 0, is a map, θ, from a domain D ⊆ I × U to U , which assigns to each
point (t, p) ∈ D, the value of up (t), where up : Jp → U is an integral curve of F
that solves the IVP
dx


 = F (t, x);
dt (1.5)


x(0) = p.
(The interval of existence, Jp , for up is assumed here to contain both 0 and t).
Thus,
θ(t, p) = up (t), for (t, p) ∈ D.
Observe that for the map, θ, to be well defined, through each point p ∈ U there
must exist a most one integral curve, up , of F . As was pointed out in Example
8 CHAPTER 1. INTRODUCTION

1.1.1, the IVP (1.2) is not always guaranteed to have a unique solution. We
will see in the next chapter that uniqueness of the integral curve going through
a point (to , po ) in I × U is guaranteed for the case in which, in addition to the
vector field, F , being continuous, the field F (t, x) is assumed to satisfy a local
Lipschitz condition in the second variable; that is, there exists a ball of radius
ro > 0 around po , denoted by Bro (po ), such that B ro (po ) ⊂ U (here, B ro (po )
denotes the closure of the open ball Bro (po ) = {x ∈ Rn | kx − po k < ro }; that
is, B ro (po ) = {x ∈ Rn | kx − po k 6 ro }); a positive number δo > 0 such that
[to − δo , to + δo ] ⊂ I; and a constant Ko , depending on po , for which

kF (t, x) − F (t, y)k 6 Ko kx − yk, for x, y ∈ B ro (po ), and |t − to | 6 δo .

The symbol kxk denotes the Euclidean norm of a vector x ∈ RN ; in other words,
q
kxk = x21 + x22 + · · · + x2N .

An example of a vector field for which a local Lipschitz condition holds is a C 1

vector field, F ; that is, a vector field as given in (1.1) in which the component
functions have continuous partial derivatives
∂fi ∂fi
, , for i, j = 1, 2, . . . , N.
∂t ∂xj

We will also need to give a more precise definition of the domain, D, for
the flow map, θ. This will entail knowing that each integral curve, up , through
p ∈ U is defined on a maximal interval of existence Jp containing to ; that is,
Jp contains any open interval containing to on which a integral curve satisfying
the IVP (1.5) is defined. This will also be proved in the next chapter.
Finally, we also want the flow map,

θ: D → U

to be continuous. This will require proving results regarding continuous depen-

dence on initial conditions. We will prove these continuity results for the case
in which F is Lipschitz continuous in the next chapter.
Chapter 2

Fundamental Existence
Theory for Ordinary
Differential Equations

Let I denote an open interval and U be an open subset of RN . Assume that

F: I ×U →U

is a continuous vector field, which satisfies a local Lipschitz condition at (to , po ) ∈

I × U ; that is, there exist positive numbers ro and δo such that B ro (po ) ⊂ U
and [to − δo , to + δo ] ⊂ I, and a constant Ko for which

kF (t, x) − F (t, y)k 6 Ko kx − yk, for x, y ∈ B ro (po ), and |t − to | 6 δo . (2.1)

We begin this chapter by proving that there there exists a positive constant, δ,
such that δ < δo and a C 1 curve

u : (to − δ, to + δ) → U

such that
u(to ) = po ,
and
u0 (t) = F (t, u(t)), for all t ∈ (to − δ, to + δ).
We shall refer to this as the Local Existence and Uniqueness Theorem for Lip-
schitz continuous vector fields. We will state it here for future reference.
Theorem 2.0.1 (Local Existence and Uniqueness Theorem). Let I ⊆ R be an
open interval, and U ⊆ RN be an open set. Let (to , po ) ∈ I × U and suppose
that the vector field F : I ×U → U satisfies a local Lipschitz condition at (to , po )
given in (2.1). Then, there exists δ > 0 and a C 1 function

u : (to − δ, to + δ) → U

9
10 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

that solves the IVP

dx


 = F (t, x);
dt (2.2)


x(to ) = po .
over the interval (to − δ, to + δ).
The main goal of this chapter is to provide a proof of Theorem 2.0.1 and its
extensions. In the next section we set up the framework that will be needed in
the proof.

2.1 Setup for the Proof of Theorem 2.0.1

We first look for a local solution of IVP (2.2) among a class of functions defined
on some bounded interval [a, b] contained in (to − δo , to + δo ) and which contains
to . We begin by considering the class of continuous functions,
u : [a, b] → RN ,
defined on [a, b]. We will denote the class by C[a, b]. This is a vector space
which can be endowed with the norm
kuk = max ku(t)k, for all u ∈ C[a, b]. (2.3)
t∈[a,b]

Observe that the symbol k · k in (2.3) is used to mean two things: (1) to the
right of the equal sign in (2.3), ku(t)k denotes the Euclidean norm of the vector
u(t) in RN ; (2) to the left of the equal sign in (2.3), kuk denotes the norm of the
function u : [a, b] → RN being defined by the expression in (2.3). The context
will make clear which sense of the symbol k · k is being used.
The norm, k · k, defined in (2.3) gives rise to metric, or distance function,
d(·, ·), in the space C[a, b]:
d(u, v) = ku − vk, for all u, v ∈ C[a, b]. (2.4)
The space C[a, b] together with the metric, d(·, ·) is a complete metric space;
meaning that any Cauchy sequence of functions, (um ), in C[a, b], converges to
a function in C[a, b].
Definition 2.1.1 (Cauchy Sequence in C(a, b)). A sequence of functions, (um ),
in C[a, b] is said to be a Cauchy sequence if, for any ε > 0, there exists M ∈ N
such that
m, n > M ⇒ d(um , un ) < ε,
where d(·, ·) is the metric in C[a, b] defined in (2.4).
Proposition 2.1.2 (Completeness of C(a, b)). Every Cauchy sequence of func-
tions, (um ), in C[a, b] converges to a continuous function, u : [a, b] → RN , in the
sense that
lim kum − uk = 0,
m→∞
where k · k is the norm in C[a, b] defined by the expression in (2.3).
2.1. SETUP FOR PROOF OF LOCAL RESULTS 11

A proof of Proposition 2.1.2 may be found in various texts on Real Analy-

sis. For instance, in [Bar76], it is stated as the Cauchy Criterion for Uniform
Convergence on page 121.
Definition 2.1.3 (Banach Space). A vector space with norm, k · k, that is also
complete with respect to the metric generated by that norm, is called a Banach
space.
Thus, Proposition 2.1.2 states that the space C[a, b] with the norm, k · k,
defined in (2.3) is a Banach space.
Let I denote an open interval and U be an open subset in RN . Suppose that
the vector field
F : I × U → RN
is continuous. Let (to , po ) ∈ I × U and let [a, b] be a closed interval containing
to in its interior with [a, b] ⊂ I. Given a function u ∈ C[a, b] with the property
that u(t) ∈ U for all t ∈ [a, b], we can define a new function

T (u) : [a, b] → RN

as follows:
Z t
T (u)(t) = po + F (τ, u(τ )) dτ, for all t ∈ [a, b], (2.5)
to

where the integral on the right–hand side of the equal sign in (2.5) is to be
understood as a vector in RN with components given by
Z t
fi (τ, u(τ )) dτ, for i = 1, 2, . . . , N.
to

Put
v(t) = T (u)(t), for all t ∈ [a, b], (2.6)
N
where T (u) : [a, b] → R is as defined in (2.5). It then follows from (2.6), (2.5)
and the Fundamental Theorem of Calculus that v is differentiable over (a, b),
with
v 0 (t) = F (t, u(t)), for all t ∈ (a, b). (2.7)
Furthermore,
v(to ) = po . (2.8)
The expressions in (2.7) and (2.8) suggest one way to prove the existence of
a solution to the IVP in (2.2) defined over some interval (a, b) containing to .
Suppose that we can find a function, u : [a, b] → RN , that is a solution of the
fixed–point equation

T (u)(t) = u(t), for t ∈ [a, b]. (2.9)

Then, that function will satisfy

u(to ) = po (2.10)
12 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

and
u0 (t) = F (t, u(t)), for all t ∈ (a, b), (2.11)
by virtue of (2.7) and (2.8). In other words, u solves the IVP (2.2) over the
interval (a, b).

Observe from the fixed–point equation in (2.9) that any solution of (2.9)
must lie in the image of T as well as the domain of T . The domain of T consists
of functions u in C[a, b] such that u(t) ∈ U for all t ∈ [a, b]. From the definition
of the map T in (2.5) we see that it is not necessarily the case that, if u(t) ∈ U for
all t ∈ [a, b], then T (u)(t) ∈ U for all t ∈ [a, b]. Thus, in order to find a solution
of the fixed–point equation in (2.9) we will need to restrict the domain of T so
that T (u)(t) ∈ U for all t ∈ [a, b] for all functions, u, in the restricted domain.
We can do this by using the assumption that U is open. Let B ≡ Br (po ), where
0 < r < ro , be an open ball around po such that B ⊂ U . We can then define a
subset of C[a, b], which we will denote by C([a, b], B), as follows:

C([a, b], B) = {u ∈ C[a, b] | u(t) ∈ B for all t ∈ [a, b]}.

Thus, the map T is defined on C([a, b], B) and it remains to show that the
interval [a, b] can be chosen so that T maps C([a, b], B) to itself. We can do this
by taking advantage of the assumption that the vector field, F , is continuous;
so that, it is bounded on the compact set [to − δo , to + δo ] × B ro (po ), where δo
and ro are as given in the statement of the Lipschitz condition in (2.1). In fact,
let
Mo = max kF (t, x)k; (2.12)
|t−to |6δo
x∈B ro (po )

then, for any u ∈ C([a, b], B), using (2.5), we obtain that
Z t
kT (u)(t) − po k 6 kF (τ, u(τ ))k dτ
to (2.13)
6 Mo |t − to |,

for all t ∈ [a, b], where we have also used (2.12). Thus, if we choose [a, b] a priori
to be [to − δ, to + δ], for some δ > 0 with δ < δo , we obtain from (2.13) that

kT (u)(t) − po k 6 Mo δ, for all t ∈ [a, b]. (2.14)

Thus, by choosing δ so that

r
δ6 , (2.15)
Mo
we get from (2.14) and (2.15) that

kT (u)(t) − po k 6 r, for all t ∈ [a, b], (2.16)

where
[a, b] = [to − δ, to + δ]. (2.17)
2.2. PROOF OF LOCAL EXISTENCE AND UNIQUENESS 13

It then follows from (2.16) that, if [a, b] is given as in (2.17), where δ is chosen
so that δ < δo and (2.15) holds, then

T : C([a, b], B) → C([a, b], B); (2.18)

in other words, T maps the metric space C([a, b], B) to itself.

In order to prove that the map T defined in (2.5) has a fixed point, we may
apply one of various fixed point theorems to appropriately chosen [a, b] and B.
We will apply the Contraction Mapping Principle of Banach.

Definition 2.1.4 (Contraction). Let X denote a metric space with metric

d(·, ·). A function T : X → X is said to be a contraction if there exists k > 0
such that k < 1 and

d(T (x), T (y)) 6 k d(x, y), for all x, y ∈ X. (2.19)

Remark 2.1.5. If the metric in X is generated by a norm, k · k, then (2.19) in

the definition of a contraction reads

kT (x) − T (y)k 6 k kx − yk, for all x, y ∈ X. (2.20)

Theorem 2.1.6 (Contraction Mapping Principle). Let X be a complete metric

space and T : X → X be a contraction. Then, T has a unique fixed point in X.
That is, there exists a unique u ∈ X such that

T (u) = u.

Theorem 2.1.6 is proved in various Real Analysis books. In one of the recom-
mended texts for this course, [Hal09], it is presented as the Contraction Mapping
Principle of Banach-Cacciopoli [Hal09, Theorem 3.1] and proved on page 5.
The proof of the local existence and uniqueness theorem for locally Lipschitz
continuous vector fields, which we will present in the next section, boils down
to proving that there is an appropriately chosen interval, [a, b], for which the
map
T : C([a, b], B) → C([a, b], B),
defined in (2.5), is a contraction. The proof of Theorem 2.0.1 will then follow
from the Contraction Mapping Principle (Theorem 2.1.6).

2.2 Proof of the Local Existence and Uniqueness

Theorem
In the previous section we saw that, if [a, b] = [to − δ, to + δ], where δ > 0
satisfies δ < δo and (2.15), then the map T defined in (2.5) maps the metric
space C([a, b], B) to itself. We will now show that, by restricting δ further, T
will be a contraction.
14 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

Let u, v ∈ C([a, b], B) and consider

Z t
T (u)(t) − T (v)(t) = (F (τ, u(τ )) − F (τ, v(τ ))) dτ, for all t ∈ [a, b], (2.21)
to

where we have used the definition of the map T in (2.5). Since B ⊂ Bro (po ),
we can apply the Lipschitz condition in (2.1) to obtain from (2.21) that
Z t
kT (u)(t) − T (v)(t)k 6 Ko ku(τ ) − v(τ )k dτ , for all t ∈ [a, b]. (2.22)
to

Thus, using the definition of the norm in C([a, b], B) in (2.3), we get from (2.22)
that
kT (u)(t) − T (v)(t)k 6 Ko |t − to |ku − vk, for all t ∈ [a, b],
which implies that
kT (u) − T (v)k 6 Ko δ ku − vk, (2.23)
by virtue of the definition of the norm in C([a, b], B). Consequently, setting

k = Ko δ, (2.24)

we obtain from (2.23) that

kT (u) − T (v)k 6 k ku − vk. (2.25)

Thus, if we restrict δ further so that

1
δ< , (2.26)
Ko
we see from (2.24) and (2.25) that T is a contraction.
We are now in a position to prove the Fundamental Theorem of Ordinary
Differential Equations.
Proof of Theorem 2.0.1: Choose δ > 0 so that
 
r 1
δ < min δo , , , (2.27)
Mo Ko
where r > 0 is chosen so that r < ro and

B = Br (po ) ⊂ U ;

the constants δo , ro and Ko are given by the Lipschitz condition assumption on

the vector field
F : I × U → U,
which was stated in (2.1); and Mo is given in (2.12); namely,

Mo = max kF (t, x)k.

|t−to |6δo
x∈B ro (po )
2.2. PROOF OF LOCAL EXISTENCE AND UNIQUENESS 15

Put [a, b] = [to − δ, to + δ]. It then follows from (2.27), the calculations in the
previous set-up section, (2.24) and (2.25) that the map

T : C([a, b], B) → C([a, b], B),

defined in (2.5), is a contraction. Consequently, by the Contraction Mapping

Principle (Theorem 2.1.6), there exists a unique function u ∈ C([a, b], B) satis-
fying
u(t) = T (u)(t), for all t ∈ [to − δ, to + δ],
or Z t
u(t) = po + F (τ, u(τ )) dτ, for all t ∈ [to − δ, to + δ]. (2.28)
to

It then follows from (2.28) and (2.9)–(2.11) in Section 2.1, that u is the unique
solution to the IVP (2.2) over the interval (to − δ, to + δ). This completes the
proof of the Fundamental Theorem for Ordinary Differential Equations. 

Remark 2.2.1 (Another Proof of Uniqueness). Uniqueness of the solution of

the IVP (2.2) over the interval of existence [to − δ, to + δ], where

δKo < 1, (2.29)

for the case in which the vector field, F , satisfies a local Lipschitz condition
as described in (2.1), follows from the uniqueness assertion of the Contraction
Mapping Principle. However, it can also be proved directly as a consequence of
the local Lipschitz condition and the integral representation (2.28) for a solution
to the IVP (2.2). We present the independent proof of uniqueness in this re-
mark because it uses calculations and results that will be needed in subsequent
sections.
Put J = [to − δ, to + δ], where δ satisfies (2.29), and suppose that u : J → U
and v : J → V both solve the differential equation (DE)

dx
= F (t, x) (2.30)
dt
over the open interval, J, contained in I. Furthermore, assume that J is small
enough so that J ⊂ (to − δo , to + δo ) and u(t), v(t) ∈ Bro (po ), for all t ∈ J.
Put
w(t) = u(t) − v(t), for all t ∈ J. (2.31)
Observe that if both v and u solve the IVP (2.2) over J, then

w(to ) = 0, (2.32)

by virtue of (2.31). Put

g(t) = kw(t)k, for all t ∈ J. (2.33)

16 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

Then, g is continuous over J and it therefore takes on a maximum value, M ,

over the compact interval J; that is,
M = max g(t) (2.34)
t∈J

is attained at some point t1 ∈ J; so that

M = g(t1 ). (2.35)
Observe that Z t1
w(t1 ) − w(to ) = w0 (τ ) dτ. (2.36)
to

Thus, using (2.32), (2.31) and (2.30), we obtain from (2.36) that
Z t1
w(t1 ) = (F (τ, u(τ )) − F (τ, v(τ ))) dτ. (2.37)
to

Thus, applying the Lipschitz condition in (2.1) to the expression in (2.37), we

get Z t1
kw(t1 )k 6 Ko ku(τ ) − v(τ )k dτ ,
to
from which we get Z t1
g(t1 ) 6 Ko g(τ ) dτ , (2.38)
to

by virtue of (2.31) and (2.33). Consequently, using (2.34) and (2.35), we obtain
from (2.38),
M 6 Ko δM. (2.39)
We claim that
M = 0. (2.40)
Otherwise, we would obtain from (2.39) that
1 6 Ko δ,
which is in direct contradiction with (2.29). Thus, (2.40) must be true, and so,
by virtue of (2.34), (2.33) and (2.31),
u(t) = v(t), for all t ∈ J.

2.3 Extension of Solutions

In the previous section, we obtained existence of a unique solution of the IVP
(2.2) in some interval (to − δ, to + δ) around to for the case in which the vector
field, F (t, x), satisfies a local Lipschiptz condition in the second variable as
described in (2.1). In this section we prove that a unique solution of the IVP
(2.2) can be obtained in a maximal interval of existence, Jpo , where Jpo is an
open interval containing to such that Jpo ⊆ I.
2.3. EXTENSION OF SOLUTIONS 17

Example 2.3.1. Let U = R2 and let F : U → R2 denote the vector field given
by  2
x
F (x, y) = ;
0
to = 0 and po = (1, 0). In this case the IVP (2.2) leads to the two–dimensional
system
dx


 = x2 ;
 dt

(2.41)
dy


= 0,


dt
subject to the initial conditions

x(0) = 1, y(0) = 0. (2.42)

Solving the system in (2.41), we are lead to the solution

 
1
u(t) = , 0 , for t < 1,
1−t
satisfying the initial condition in (2.42). Thus, in this case, the maximal interval
of existence is J(1,0) = (−∞, 1). Note that, as t approaches 1 from the left,
1
ku(t)k = → +∞.
1−t
Remark 2.3.2. Observe that the vector field, F , in Example 2.3.1 does not
depend explicitly on t. Thus, the system in (2.41) that it generates does not
depend explicitly on t either. Systems that are explicitly independent of the
“time” variable, t, are said to be autonomous. We will spend a large portion
of this course studying autonomous systems.
Remark 2.3.3. Note that the domain, I × U , of the vector field generating
the autonomous system in Example 2.3.1 can be thought to be R × U ; that is,
I = R for autonomous systems. The result of Example 2.3.1 shows that even in
the case of an autonomous systems, existence of a solution defined for all t ∈ R
is not guaranteed. Later in this section, we will see which conditions on the
vector field will imply existence of solutions which are defined for all times, t.
Such solutions are known as global solutions.
To prove the main extension result of this section, we begin with the following
lemma.
Lemma 2.3.4 (Unique Continuation Lemma). Let I ⊆ R be an open interval,
and U ⊆ RN be an open set. Suppose that the vector field F : I×U → U satisfies
a local Lipschitz condition in the second variable at every (t, p) ∈ I × U , and
let u and v denote two C 1 solutions of the differential equation
dx
= F (t, x), (2.43)
dt
18 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

which are defined in an open interval J ⊆ I. If u and v agree at a point to ∈ J;

that is,
u(to ) = v(to ), (2.44)
then
u(t) = v(t), for all t ∈ J. (2.45)
Proof: Define a subset, Jo , of J as follows:

Jo = {t ∈ J | u(t) = v(t)}.

Then, to ∈ Jo , by (2.44); so that Jo 6= ∅. Furthermore, Jo is closed; since Jo is

the pre-image under u − v of the closed set {0}, where 0 denotes the zero vector
in RN . We will show that Jo is also open. In fact, let t1 be any point in Jo and
put p1 = v(t1 ) = v(t1 ). Then, by the local existence and uniqueness theorem
for ODEs, Theorem 2.0.1, the IVP
dx

 dt = F (t, x);





x(t1 ) = p1 ,


has a unique solution defined in an open interval J1 = (t1 − δ, t1 + δ). We may

assume that δ > 0 is small enough so that J1 ⊂ J. By uniqueness, we get that
u(t) = v(t) for all t ∈ J1 , since u(t1 ) = v(t1 ). It then follows that J1 ⊂ Jo .
Hence, Jo is also open. Thus, Jo is a non–empty, closed subset of J which is
also open. Hence, by the connectedness of J, Jo = J. Thus, u and v agree on
all of J, which we wanted to show. 
Given a point (to , po ) ∈ I × U , let Jpo denote the collection of all open
intervals, J, that contain to and on which there is defined a C 1 solution, u, of
the differential equation in (2.43) with u(to ) = po . Define
[
J po = J po ; (2.46)

that is, Jpo is the union of all intervals in Jpo . Next, define a function,

upo : Jpo → U,

by
upo (t) = u(t), for t ∈ Jpo , (2.47)
where u is a solution of the differential equation in (2.43) defined on J ∈ Jpo with
t ∈ J and u(to ) = po . By the result of Lemma 2.3.4 (the Unique Continuation
Lemma), if v : Je → U is another solution of the differential equation in (2.43)
defined in Je ∈ Jpo with t ∈ Je and v(to ) = po , then v(t) = u(t), since v(to ) =
u(to ), and so v and u agree on any open interval that contains to . Thus, the
function, upo , defined in (2.47) is well defined. Observe that upo solves IVP
(2.2) over the entire interval Jpo .
2.3. EXTENSION OF SOLUTIONS 19

The open interval Jpo defined in (2.46) is called the maximal interval of
existence for the IVP (2.2). The interval Jpo is maximal in the sense that Jpo
cannot be a proper subset of an open interval on which there is defined a solution
of the IVP (2.2). This is evident from the definition of Jpo in (2.46).
We summarize what we have just proved in the following theorem.
Theorem 2.3.5 (Existence and Uniqueness on a Maximal Interval). Let I ⊆ R
be an open interval, and U ⊆ RN be an open set. Suppose that the vector field
F : I × U → U satisfies a local Lipschitz condition in the second variable at
every (t, p) ∈ I × U . For each (to , po ) ∈ I × U , there exists a unique C 1 function

upo : Jpo → U

which solves the IVP (2.2) over a maximal interval of existence, Jpo .
Definition 2.3.6. Let u : Jo → U be a solution of the IVP (2.2) defined in some
open interval, Jo , that contains to . Let J be an open interval which contains to
and such that Jo ⊂ J. We say that a C 1 curve

v: J → U

is an extension of u if v solves the IVP (2.2) over the interval J, and

v(t) = u(t), for all t ∈ Jo .

Lemma 2.3.7 (Extensibility Lemma). Let I denote an open interval and U an

open subset of RN . Suppose that F : I × U → RN is continuous and satisfies
a local Lipschitz condition in the second variable at every (t, p) ∈ I × U . Let
(to , po ) ∈ I × U and a, b ∈ R be such that a < to < b and [a, b] ⊂ I. Suppose
that u : (a, b) → U is a solution of the IVP (2.2) defined on the open interval,
(a, b). Then, there exists and extension of u to an interval (a, b + δ), for some
δ > 0, if and only if there exists p ∈ U and a sequence (tm ) in (a, b) such that

lim (tm , u(tm )) = (b, p). (2.48)

m→∞

Proof: If u has an extension to (a, b + δ), for some δ > 0, the (2.48) holds for
any sequence (tm ) that converges to b by the continuity of the extension at b.
To prove the converse, assume that (2.48) holds true for a sequence (tm ) in
(a, b). Note that it follows from (2.48) that

lim tm = b, (2.49)
m→∞

and
lim u(tm ) = p. (2.50)
m→∞

We may assume that the sequence (tm ) increases to b, by passing to a subse-

quence, if necessary.
We first show that
lim− u(t) = p. (2.51)
t→b
20 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

Proceeding by contradiction, if (2.51) is not true, there exists ε > 0 and a

sequence (sm ) in (a, b) such that (sm ) increases as m → ∞ with

lim sm = b, (2.52)
m→∞

and
ku(sm ) − pk > ε, for all m ∈ N. (2.53)
Passing to a subsequence, if necessary, we may assume that

tm < sm , for all m ∈ N.

Now, let ro > 0 be such that

ro < ε, (2.54)
where ε > 0 is as given in (2.53), and B ro (p) ⊂ U , and let δo > 0 be such that
[b − δo , b] ⊂ I. Put
Mo = max kF (t, x)k. (2.55)
t∈[b−δo ,b]
x∈B ro (p)

Consider the function

g(t) = ku(t) − pk, for t ∈ (a, b).

It follows from (2.50) that, there exits a natural number, N1 , such that

g(tm ) < ro , for all m > N1 . (2.56)

In view of (2.56), (2.54) and (2.53), it follows from the Intermediate Value
Theorem that, for each m > N1 , there exists τm such that

tm < τm < sm , for m > N1 , (2.57)

and
g(τm ) = ro , for m > N1 ; (2.58)
furthermore, we may assume that

g(t) 6 ro , for t ∈ [tm , τm ], and m > N1 . (2.59)

It follows from (2.58) and (2.59) that

ku(τm ) − pk = ro , for m > N1 , (2.60)

and
u(t) ∈ B ro (p), for t ∈ [tm , τm ], for m > N1 . (2.61)
From the assumption that u solves the IVP (2.2) in (a, b) we obtain that
Z t
u(t) = po + F (τ, u(τ )) dτ, for all t ∈ (a, b). (2.62)
to
2.3. EXTENSION OF SOLUTIONS 21

It follows from (2.62) that

Z tm
u(tm ) = po + F (τ, u(τ )) dτ, for all m ∈ N, (2.63)
to

and Z τm
u(τm ) = po + F (τ, u(τ )) dτ, for all m ∈ N. (2.64)
to

Subtracting (2.63) from (2.64) we then obtain that

Z τm
u(τm ) − u(tm ) = F (τ, u(τ )) dτ, for all m ∈ N. (2.65)
tm

Next, use (2.55) and (2.61) to obtain from (2.65) that

ku(τm ) − u(tm )k 6 Mo |τm − tm |, for all m > N1 . (2.66)

Now, use (2.50), (2.52) and (2.57) to derive from (2.66) that

lim ku(τm ) − u(tm )k = 0. (2.67)

m→∞

Applying the triangle inequality, we obtain

ku(τm ) − pk 6 ku(τm ) − u(tm )k + ku(tm ) − pk, for all m. (2.68)

Thus, combining (2.50), (2.67) and (2.68) yields

lim ku(τm ) − pk = 0, (2.69)

m→∞

which, in conjunction with (2.60), implies that ro = 0; but this contradicts

ro > 0. Hence, (2.51) must be true.
Next, consider the IVP
dx

 dt = F (t, x);


(2.70)


x(b) = p.


By the Local Existence and Uniqueness Theorem (Theorem 2.0.1), there exists
δ > 0 and a unique continuous function

v : [b − δ, b + δ] → U,

which solves the IVP (2.70) in the interval (b − δ, b + δ). We may choose δ so
that
δK < 1, (2.71)
where K is a Lipschitz constant for F over the set

[b − δo , b + δo ] × B ro (p);
22 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

for some δo > 0. We may assume that δ < δo .

b : (a, b + δ) → U as follows
Next, define a function u
(
u(t) if t ∈ (a, b);
u
b(t) =
v(t) if t ∈ (b − δ, b + δ).

We claim that u
b is an extension of u. To prove this claim, we need to show that

u(t) = v(t), for all t ∈ (b − δ, b). (2.72)

In order to prove (2.72), define w : [b − δ, b] → U as follows

(
u(t) − v(t) if b − δ 6 t < b;
w(t) = (2.73)
0 if t = b.

It follows from (2.51), which was proved in the first part of this proof, that w
is continuous on [b − δ, b]. By continuity and (2.51), δ can be chosen so that

u(t) ∈ Bro (p) and v(t) ∈ Bro (p), for all t ∈ [b − δ, b]. (2.74)

By continuity, there exists t1 ∈ [b − δ, b] such that

kw(t1 )k = max kw(t)k ≡ M1 . (2.75)

t∈[b−δ,b]

Observe that Z t1
w(t1 ) − w(b) = w0 (τ ) dτ. (2.76)
b
Thus, using (2.73) and the assumption that u and v solve the differential equa-
tion in IVP (2.2), we obtain from (2.76) that
Z t1
w(t1 ) = (F (τ, u(τ )) − F (τ, v(τ ))) dτ. (2.77)
b

Thus, applying the Lipschitz condition to the expression in (2.77), we get

Z t1
kw(t1 )k 6 K ku(τ ) − v(τ )k dτ ,
b

or Z t1
kw(t1 )k 6 K kw(τ )k dτ ,
b
from which we get
M1 6 KδM1 . (2.78)
in view of (2.75). We then see that, by virtue of (2.71), (2.78) leads to a
contradiction, unless M1 = 0. Hence, we obtain from (2.75) that w(t) = 0 for
all t ∈ [b − δ, b], which implies (2.72), and the proof of the extensibility lemma
is now complete. 
2.3. EXTENSION OF SOLUTIONS 23

Remark 2.3.8. We can use a similar argument to the one used in the proof of
Lemma 2.3.7 to prove that, under the same assumptions of the Lemma, u has
an extension to (a − δ, b), for some δ > 0, if and only if there exists q ∈ U and
a sequence (sm ) in (a, b) such that

lim (sm , u(sm )) = (a, q).

m→∞

Lemma 2.3.7 will help us answer the following question: Suppose that Jpo =
(a, b) is the maximal interval of existence for IVP (2.2), for some (to , po ) ∈ I ×U ,
and let u : Jpo → U denote the corresponding integral curve. If b < ∞, what
happens to u(t) as t tends to b? We will see, as a consequence of the following
proposition, that either u(t) tends to some point on the boundary, ∂U , of U , or
ku(t)k → ∞ as t → b− . Example 2.3.1 on page 17 of these notes provides an
instance of this general result.

Proposition 2.3.9 (Escape in Finite Time Theorem). Let I denote an open

interval and U an open subset of RN . Suppose that F : I×U → RN is continuous
and satisfies a local Lipschitz condition in the second variable at every point
(t, p) ∈ I × U . Let Jpo = (a, b) denote the maximal interval of existence for
IVP (2.2), for some (to , po ) ∈ I × U , and let u : Jpo → U be the corresponding
integral curve. If b < ∞, then, for any compact set, C ⊂ U , there exists ε > 0
such that
t ∈ (b − ε, b) ⇒ u(t) 6∈ C.

Proof: Suppose, to the contrary, that there exists a compact set, C ⊂ U , and a
sequence, (tm ), in (a, b) such that tm tends to b as m → ∞ and u(tm ) ∈ C for all
m ∈ N. Then, since [to , b] × C is compact, passing to a subsequence if necessary,
we may assume that (tm , u(tm )) converges (b, p), for some p ∈ C; that is,

lim (tm , u(tm )) = (b, p).

m→∞

Hence, the hypotheses of Lemma 2.3.7 are satisfied. Consequently, there exists
an extension of u to (a, b + δ), for some δ > 0. However, this contradicts the
maximality of Jpo . Thus, the proposition follows. 

Remark 2.3.10. Under the same assumptions of Proposition 2.3.9, we may

prove that if a > −∞, then, for any compact set, C ⊂ U , there exists ε > 0,
such that t ∈ (a, a + ε) ⇒ u(t) 6∈ C.

Remark 2.3.11. Proposition 2.3.9 is very useful in verifying that a solution

of IVP (2.2) exists for all t > to for the case in which F (t, x) is known to be
continuous over I × RN , where I is an open interval containing [to , ∞), and
we can obtain an a priori estimate on the norm, ku(t)k, for the solution u on
IVP (2.2) over the maximal interval of existence Jpo = (a, b). More specifically,
arguing by contradiction, suppose that b < ∞, and that we can, a priori, obtain
the estimate
ku(t)k 6 R, for all t ∈ [to , b), (2.79)
24 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

and some R > 0. We can then consider the compact set

C = {x ∈ RN | kxk 6 R}.
It follows from (2.79) that u(t) ∈ C for all t ∈ [to , b), which is in direct contra-
diction with the result of Proposition 2.3.9. In the remainder of this section, we
will see a few applications of this idea.

For the remainder of this section, assume that I = R and U = RN , so that

the vector field,
F : R × RN → RN ,
is defined for all times, t ∈ R. We will ask the question: Under which conditions
on the field, F , is Jpo = R? That is, under which conditions on F is the
solution to the IVP (2.2) defined for all times t ∈ R? We will see that the
following growth condition on the vector field F (t, x) is sufficient:
Definition 2.3.12 (At Most Linear Growth). Let F : R × RN → RN be a con-
tinuous vector field. We say that F (t, x) has at most linear growth in the second
variable if and only if, for every T > 0, there exist non–negative constants, cT
and dT , such that
kF (t, x)k 6 cT + dT kxk, for all t ∈ [−T, T ], and all x ∈ RN . (2.80)
Proposition 2.3.13 (Global Existence Theorem I). Let F : R×RN → RN be a
locally Lipschitz continuous vector field. Suppose that F (t, x) has at most linear
growth in the second variable. Then, for any (to , po ) ∈ R × RN , the solution
to the IVP (2.2) exists for all t ∈ R; in other words, the maximal interval of
existence, Jpo , is the entire real line.
Proof: Assume, by way of contradiction, that Jpo = (a, b) is such that b < ∞
(the case in which a > −∞ can be treated in an analogous manner). By the
growth assumption in (2.80), there exist non–negative constants, cb and db , such
that
kF (t, x)k 6 cb + cd kxk, for all t ∈ [to , b] and all x ∈ Rn . (2.81)
Let u : (a, b) → RN be the solution to the IVP (2.2). We can then write
Z t
u(t) = po + F (τ, u(τ )) dτ, for all t ∈ (a, b). (2.82)
to

Using the estimate in (2.82), we get from (2.82) that

Z t
ku(t)k 6 kpo k + (cb + db ku(τ )k) dτ, for all t ∈ [to , b),
to

from which we get

Z t
ku(t)k 6 kpo k + cb (b − to ) + db ku(τ )k dτ, for all t ∈ [to , b). (2.83)
to
2.3. EXTENSION OF SOLUTIONS 25

Putting Co = kpo k + cb (b − to ) in the inequality in (2.83) yields

Z t
ku(t)k 6 Co + db ku(τ )k dτ, for all t ∈ [to , b). (2.84)
to

Define g : [to , b) → R by
Z t
g(t) = Co + db ku(τ )k dτ, for t ∈ [to , b). (2.85)
to

It follows from the Fundamental Theorem of Calculus and (2.85) that g is dif-
ferentiable on (to , b) with

g 0 (t) = db ku(t)k, for t ∈ (to , b). (2.86)

Using the estimate in (2.84), we obtain from (2.86)

g 0 (t) 6 db g(t), for t ∈ (to , b), (2.87)

where we have also used (2.85). Rewrite the differential inequality in (2.87) as

g 0 (t) − db g(t) 6 0, for t ∈ (to , b), (2.88)

and multiply on both sides of (2.88) by e−db t to obtain

d  −db t 
e g(t) 6 0, for t ∈ (to , b). (2.89)
dt
Integrating on both sides of the inequality in (2.89) from to to t leads to

g(t) 6 g(to )edb (t−to ) , for t ∈ (to , b). (2.90)

Consequently, using (2.84) and (2.85), we get from (2.90) that

ku(t)k 6 Co edb (b−to ) , for t ∈ [to , b). (2.91)

Setting Ro = Co edb (b−to ) , we see from (2.91) that the values u(t) lie in the com-
pact set B Ro (0). This is in direct contradiction with the result of Proposition
2.3.9. Hence, b = ∞. 
Example 2.3.14 (Linear Systems). Let f : R → RN be a continuous function,
and let
aij : R → R, for i, j = 1, 2, . . . , N,
be continuous real valued functions. Define the matrix valued function

A(t) = [aij (t)], for t ∈ R.

The vector field, F : R × RN → RN , defined by

F (t, x) = f (t) + A(t)x, for all t ∈ R and all x ∈ RN , (2.92)

26 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

satisfies the growth condition in (2.80). In fact, we have the estimate

kF (t, x)k 6 kf (t)k + kA(t)kkxk, for all t ∈ R and all x ∈ RN , (2.93)

where v
uN N
uX X
kA(t)k = t [aij (t)]2 , for all t ∈ R.
i=1 j=1

The growth estimate in (2.80) then follows from (2.93) and the assumption that
f and aij , for i, j = 1, 2, . . . , N , are continuous functions.
The local Lipschitz continuity of F in (2.92) follows from the estimate

kF (t, x) − F (t, y)k 6 kA(t)kkx − yk, for all t ∈ R and all x, y ∈ RN ,

and the continuity of aij , for i, j = 1, 2, . . . , N .

We can then conclude, as a consequence of Proposition 2.3.13 that the linear
IVP
dx

 dt = f (t) + A(t)x;


(2.94)


x(to ) = po .


has a unique solution for each (to , po ) ∈ R × RN , which exists for all t ∈ R.
Later in these notes we will see how to compute the solution of the linear IVP
(2.94).
Corollary 2.3.15 (Global Existence Theorem II). Let F : RN → RN satisfy
the global Lipschitz condition:

kF (x) − F (y)k 6 Kkx − yk, for x, y ∈ RN . (2.95)

Then, for any (to , po ) ∈ R × RN , the solution to the IVP

dx

 dt = F (x);


(2.96)


x(to ) = po .


exists for all t ∈ R.

Proof: Observe that, for any x ∈ RN , it follows from (2.95) that

kF (x) − F (po )k 6 Kkx − po k,

from which we get that

kF (x)k 6 kF (po )k + kpo k + Kkxk.

Hence, F satisfies a growth condition of the type in (2.80) needed for Propo-
sition 2.3.13. The corollary then follows from the Global Existence I theorem
(Proposition 2.3.13). 
2.4. CONTINUOUS DEPENDENCE ON INITIAL CONDITIONS 27

Corollary 2.3.16 (Global Existence Theorem III). Let F : R × RN → RN

be continuous and bounded. Suppose that F (t, x) satisfies a local Lipschitz
condition in the second variable at every point (t, p) ∈ R × RN . Then, for any
(to , po ) ∈ R × RN , the solution to the IVP

dx


 = F (t, x);
 dt
(2.97)


x(to ) = po .


exists for all t ∈ R.

2.4 Continuous Dependence on Initial Condi-

tions
Let U ⊆ RN be an open set and I denote an open interval. Assume that
F : I×U → RN is continuous and that F (t, x) satisfies a local Lipschitz condition
in the second variable at every point (t, p) ∈ I × U . In the previous section we
saw that, for to ∈ I and p ∈ U , the IVP

dx

 dt = F (t, x);


(2.98)


x(to ) = p,


has a unique solution,

up : Jp → U,
defined on a maximal interval of existence, Jp , containing to . Similarly, for
q ∈ U , we get a unique solution,

uq : Jq → U,

of the IVP
dx


 = F (t, x);
 dt
(2.99)


x(to ) = q,


defined on a maximal interval of existence, Jq , containing to . We would like to

answer the following question in this section: Suppose that p and q are close
to each other; that is, kp − qk is small. How close will the values of, up (t) and
uq (t), of up and uq , respectively, be to each other over an interval [to , t1 ], for
some t1 > to with t1 ∈ Jp ∩ Jq ? The idea here is that if up (to ) and uq (to ) start
out being very close to each other, then the remaining values, up (t) and uq (t)
for t ∈ [to , t1 ], will remain close to one another.
28 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

Let B denote an open ball containing p and q be such that B ⊂ U and

kF (t, x) − F (t, y)k 6 Kkx − yk, for x, y ∈ B, and |t − to | 6 δo , (2.100)

for positive constants K and δo . We assume that t1 is sufficiently close to to so

that
[to , t1 ] ⊂ (to − δo , to + δo ), (2.101)
and
up (t), uq (t) ∈ B. for all t ∈ [to , t1 ]. (2.102)
Using the integral representation for the solutions of the IVPs in (2.98) and
(2.99), we have that
Z t
up (t) = p + F (τ, up (τ )) dτ, for all t ∈ Jp , (2.103)
to

and Z t
uq (t) = q + F (τ, uq (τ )) dτ, for all t ∈ Jq . (2.104)
to

Subtracting (2.104) from (2.103) we obtain that

Z t
up (t) − uq (t) = p − q + (F (τ, up (τ )) − F (τ, uq (τ ))) dτ, (2.105)
to

for all t ∈ [to , t1 ]. Taking the Euclidean norm on both sides of (2.105) we obtain
the estimate
Z t
kup (t) − uq (t)k 6 kp − qk + kF (τ, up (τ )) − F (τ, uq (τ ))k dτ, (2.106)
to

for all t ∈ [to , t1 ].

Next, use (2.102), (2.101) and the Lipschitz condition in (2.100) to obtain
from (2.106) that
Z t
kup (t) − uq (t)k 6 kp − qk + K kup (τ ) − uq (τ )k dτ, (2.107)
to

for all t ∈ [to , t1 ]. Next, set

Z t
g(t) = kp − qk + K kup (τ ) − uq (τ )k dτ, for t ∈ [to , t1 ]. (2.108)
to

It follows from (2.107) and (2.108) that

kup (t) − uq (t)k 6 g(t), for t ∈ [to , t1 ]. (2.109)

Applying the Fundamental Theorem of Calculus to the definition of g in (2.108)

we see that g is differentiable on (to , t1 ), and

g 0 (t) = Kkup (t) − uq (t)k, for t ∈ (to , t1 ). (2.110)

2.4. CONTINUOUS DEPENDENCE ON INITIAL CONDITIONS 29

Combining (2.110) with the estimate in (2.109), we obtain that g satisfies the
differential inequality

g 0 (t) 6 Kg(t), for t ∈ (to , t1 ),

which can be rewritten as

g 0 (t) − Kg(t) 6 0, for t ∈ (to , t1 ). (2.111)

Multiplying the inequality in (2.111) by e−Kt leads to

d  −Kt 
e g(t) 6 0, for t ∈ (to , t1 ). (2.112)
dt
Integrating on both sides of (2.112) from to to t ∈ (to , t1 ) then leads to

e−Kt g(t) 6 g(to )e−Kto , for t ∈ (to , t1 ), (2.113)

where
g(to ) = kp − qk, (2.114)
by virtue of (2.108). We then obtain from (2.113) and (2.114) that

g(t) 6 kp − qkeK(t−to ) , for t ∈ (to , t1 ). (2.115)

Combining (2.109) and (2.115) then yields

kup (t) − uq (t)k 6 kp − qkeK(t−to ) , for t ∈ [to , t1 ]. (2.116)

It follows from (2.116) that

max kup (t) − uq (t)k 6 C1 kp − qk, (2.117)

t∈[to ,t1 ]

where C1 = eK(t1 −to ) .

Thus, we get from (2.117) that, given ε > 0, there exists r > 0 such that

kp − qk < r ⇒ d(up , uq ) < ε, (2.118)

where
d(up , uq ) = max kup (t) − uq (t)k. (2.119)
t∈[to ,t1 ]

In fact, (2.118) follows from (2.117) and (2.119) by choosing r = ε/C1 . Another
way of expressing what we have just proved is

lim d(up , uq ) = 0,
p→q

which says that the solutions, up , of IVP (2.98) depends continuously on the
initial conditions, p, for a sufficiently small interval [to , t1 ].
We would like to extend the local continuity result we just proved to larger
closed and bounded intervals on which a solution exists. More precisely,
30 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

Theorem 2.4.1 (Continuous Dependence on Initial Conditions). Let I denote

an open interval and U an open subset of RN . Suppose that F : I × U → RN
is continuous and satisfies a local Lipschitz condition in the second variable at
every point (t, p) ∈ I × U . Let (to , po ) ∈ I × U and let Jpo denote the maximal
interval of existence for the IVP
dx


 = F (t, x);
 dt
(2.120)


x(to ) = po .


For every real number, T , such that T > to and [to , T ] ⊂ Jpo , there exists
r(T ) > 0, such that, if
kp − po k 6 r(T ),
then the solution, up , of the IVP

dx

 dt = F (t, x);


(2.121)


x(to ) = p,


exists on [to , T ]. Furthermore,

lim max kup (t) − upo (t)k = 0, (2.122)

p→po t∈[to ,T ]

where upo : Jpo → U is the solution to the IVP (2.120).

Proof: Let T > to be such that [to , T ] ⊂ Jpo . Since upo : Jpo → U is continuous,
the set
C = {upo (t) | to 6 t 6 T }
is a compact subset of U . Thus, there exists an open set, V , such that C ⊂ V ⊂
V ⊂ U and V is compact. We first show that there exists a constant, K > 0,
such that

kF (t, x) − F (t, y)k 6 Kkx − yk, for x, y ∈ V , and t ∈ [to , T ]. (2.123)

To prove (2.123), assume, by way of contradiction, that there exist sequences

(tm ) in [to , T ], and (xm ) and (ym ) in V , such that

kF (tm , xm ) − F (tm , ym )k > mkxm − ym k, for m ∈ N. (2.124)

Using the fact that [to , T ] × V is compact, we may assume, passing to subse-
quences if necessary, that there exist t ∈ [to , T ] and x, y ∈ V such that

lim tm = t, (2.125)
m→∞

lim xm = x, (2.126)
m→∞
2.4. CONTINUOUS DEPENDENCE ON INITIAL CONDITIONS 31

and
lim ym = y. (2.127)
m→∞

Let
M= max kF (t, x)k. (2.128)
(t,x)∈[to ,T ]×V

It follows from (2.124) and (2.128) that

2M
kxm − ym k 6 , (2.129)
m
where we have used the triangle inequality.
Letting m → ∞ in (2.129), we obtain from (2.126) and (2.127) that

kx − yk = 0.

Hence, x = y. Put q = x = y, and let r1 > 0, δ1 > 0 and K1 be such that

[t − δ1 , t + δ1 ] ⊂ I,

B r1 (q) ⊂ U,
and

kF (t, x) − F (t, y)k 6 K1 kx − yk, for x, y ∈ B r1 (q), and |t − t| 6 δ1 . (2.130)

By virtue of (2.125), (2.126) and (2.127), there exists N1 > K1 such that

m > N1 ⇒ xm , ym ∈ Br1 (q), and tm ∈ (t − δ1 , t + δ1 ).

It then follows from (2.130) that

kF (tm , xm ) − F (tm , ym )k 6 mkxm − ym k, for all m > N1 . (2.131)

However, (2.131) is in direct contradiction with (2.124). We therefore conclude

that (2.123) must be true.
Next, let
1
ε = dist(C, ∂V ), (2.132)
2
so that ε > 0, and define
r(T ) = εe−K(T −to ) , (2.133)
where K is the Lipschitz constant given in (2.123). Put

Cε = {x ∈ V | dist(x, C) 6 ε}. (2.134)

Then, Cε is a compact subset of V which contains C.

We claim that if kp − po k < r(T ), then the solution,

up : Jp → U,
32 CHAPTER 2. FUNDAMENTAL EXISTENCE THEORY

of IVP (2.121) is defined on [to , T ]. If not, writing Jp = (a, b), we would have that
b < T . It then follows from the Escape in Finite Time Theorem (Proposition
2.3.9) that there exists t1 ∈ [to , b) such that up (t1 ) 6∈ Cε , where Cε is defined in
(2.134). We may also assume that

up (t) ∈ V for all t ∈ [to , t1 ]; (2.135)

we also have that

kup (t1 ) − upo (t1 )k > ε. (2.136)
Now, the calculations leading to (2.116) yield that

kup (t) − upo (t)k 6 kp − po keK(t−to ) , for t ∈ [to , t1 ],

in view of (2.135) and the Lipschitz condition in (2.123); so that,

kup (t) − upo (t)k 6 kp − po keK(T −to ) , for t ∈ [to , t1 ]. (2.137)

Thus, if kp − po k < r(T ), where r(T ) is given in (2.133), we have from the
inequality in (2.137) that

kup (t) − upo (t)k 6 ε, for t ∈ [to , t1 ],

where ε > 0 is given in (2.132), which is in direct contradiction with (2.136).

Hence, kp − po k < r(T ) implies that up is defined on [to , T ]. Furthermore, we
get the estimate

kup (t) − upo (t)k 6 kp − po keK(T −to ) , for t ∈ [to , T ]. (2.138)

The statement in (2.122) follows from (2.138), and the proof of the theorem is
now complete. 
Chapter 3

Flows of C 1 Vector Fields

In this chapter we consider the autonomous system

dx
= F (x), (3.1)
dt
where F : U → RN is a C 1 vector field defined on some open set U ⊆ RN . By
the fundamental theory presented in the previous chapter, we know that for
each p ∈ U , there is a unique integral curve,

up : Jp → U,

defined on a maximal interval of existence, Jp , and which solves the IVP

dx


 = F (x);
dt (3.2)


x(0) = p.

We can therefore define a map

θ : D → U,

on a subset, D, of R × U given by

D = {(t, p) ∈ R × U | t ∈ Jp }, (3.3)

as follows
θ(t, p) = up (t), for all (t, p) ∈ D. (3.4)
One of the goals of this chapter is to give a precise definition of the map θ given
by (3.4). In particular, we will be showing that the domain of θ, D, given in
(3.3) is an open subset of R × U . We will also be showing that θ is continuous.
We will then derive some fundamental properties of θ. The set D is called the
flow domain of the vector field, F , and θ is the flow of F .

33
34 CHAPTER 3. FLOWS

3.1 Flow Domains and Flow Maps

Throughout this section, F : U → RN will denote a C 1 vector field defined in
an open subset, U , of RN . For each p ∈ U , up : Jp → U will denote the unique
solution to the IVP (3.2) defined on a maximal interval of existence, Jp , which
contains 0. Define
D = {(t, p) ∈ R × U | t ∈ Jp }. (3.5)

Proposition 3.1.1. The set D defined in (3.5) is an open subset of R × U .

Proof: Let (t1 , p1 ) ∈ D. We show that there exists δ1 > 0 and an open ball
B = Br1 (p1 ) satisfying

(t1 − δ1 , t1 + δ1 ) × B ⊂ D. (3.6)

Since (t1 , p1 ) ∈ D, there exists

up1 : Jp1 → U

which solves the IVP

dx


 = F (x);
dt (3.7)


x(0) = p1 ;
furthermore, t1 ∈ Jp1 .
Assume first that t1 > 0. Then up1 is defined on [0, t1 ]. Consequently, by
the Extensibility Lemma (Lemma 2.3.7), up1 is also defined on [0, t1 + δ1 ], for
some δ1 > 0. By the Continuous Dependence on Initial Conditions Theorem
(Theorem 2.4.1 on page 30 in this notes), there exists r1 > 0 such that kp−p1 k <
r1 implies that the solution of
dx


 = F (x);
dt (3.8)


x(0) = p,

is defined on [0, t1 + δ1 ]. Thus, setting B = Br1 (p1 ) we see that if (t, p) ∈

(t1 − δ1 , t1 + δ1 ) × B, then up : Jp → U is defined on [0, t1 + δ1 ], so that t ∈ Jp ,
and so (t, p) ∈ D. The statement in (3.6) then follows for the case in which
t1 > 0. The case in which t1 < 0 is analogous. We can therefore conclude that
D is open. 

We may define a map, θ : D → U , on the flow domain D as follows:

Definition 3.1.2 (Flow Map of a Vector Field). For each (t, p) ∈ D, let
up : Jp → U denote the unique solution to the IVP (3.7) defined on a maxi-
mal interval of existence, Jp . Put

θ(t, p) = up (t), for (t, p) ∈ D. (3.9)

3.2. PROPERTIES OF FLOW MAPS 35

The map θ : D → U is called the flow map of the vector field F . Equivalently,
θ is called the flow map of the differential equation
dx
= F (x). (3.10)
dt

3.2 Properties of Flow Maps

The following proposition shows that the flow map, θ : D → U , given in Defini-
tion 3.1.2 is continuous on the flow domain, D.
Proposition 3.2.1. The flow map, θ : D → U , is continuous on D.
Proof: Let (t1 , p1 ) ∈ D. We prove that

lim kθ(t, p) − θ(t1 , p1 )k = 0. (3.11)

(t,p)→(t1 ,p1 )

Let up1 : Jp1 → U be the unique solution of the IVP

dx


 = F (x);
dt (3.12)


x(0) = p1 ;

defined over a maximal interval of existence, Jp1 . Since (t1 , p1 ) ∈ D, we know

that t1 ∈ Jp1 . Assume first that t1 > 0; the case t1 < 0 is analogous. Then
up1 is defined on [0, t1 ]. Consequently, by the Extensibility Lemma (Lemma
2.3.7), up1 is also defined on [0, t1 + δ1 ], for some δ1 > 0. By the Continuous
Dependence on Initial Conditions Theorem (Theorem 2.4.1 on page 30 in these
notes), there exists r1 > 0 such that kp − p1 k < r1 implies that the solution of
dx


 = F (x);
dt (3.13)


x(0) = p,

is defined on [0, t1 + δ1 ]. The calculations leading to (2.137) in the proof of

Theorem 2.4.1 yield that

kup (t) − up1 (t)k 6 kp − p1 keKt , for all t ∈ [0, t1 + δ1 ], (3.14)

and for some constant K, provided that kp − p1 k < r1 . Thus, if kp − p1 k < r1 ,

we obtain from (3.14) that

kup (t) − up1 (t)k 6 kp − p1 keK(t1 +δ1 ) , for all t ∈ [0, t1 + δ1 ]. (3.15)

Next, let |t − t1 | < δ1 and kp − p1 k < r1 and apply the triangle inequality in

kθ(t, p) − θ(t1 , p1 )k = kθ(t, p) − θ(t, p1 ) + θ(t, p1 ) − θ(t1 , p1 )k

36 CHAPTER 3. FLOWS

to obtain

kθ(t, p) − θ(t1 , p1 )k 6 kup (t) − up1 (t)k + kup1 (t) − up1 (t1 )k, (3.16)

for |t − t1 | < δ1 and kp − p1 k < r1 , where we have also used the definition of the
flow map in (3.9). It then follows from (3.16) and (3.15) that

kθ(t, p) − θ(t1 , p1 )k 6 C1 kp − p1 k + kup1 (t) − up1 (t1 )k, (3.17)

for |t − t1 | < δ1 and kp − p1 k < r1 , where C1 = eK(t1 +δ1 ) . The statement in

(3.11) now follows from (3.17) and the fact that the function up1 is continuous
on Jp1 . The proof of Proposition 3.2.1 is now complete. 

For the proof of Proposition 3.2.1 all that we needed to assume is that F be
locally Lipschitz continuous. For a C 1 vector field, F , we’ll be able to prove more
than was proved in Proposition 3.2.1. We will show next that, for C 1 vector
fields, the flow map, θ : D → U , is a C 1 map; that is, the partial derivatives
∂θ ∂θ
, , for i = 1, 2, . . . , n,
∂t ∂xi
where x1 , x2 , . . . , xn are the coordinates of the initial points, p, in the IVP (3.13),
are continuous on D.
Proposition 3.2.2. Assume that the vector field, F : U → RN , is C 1 . Then,
the flow map, θ : D → U , is C 1 on D.
Before presenting a proof of Proposition 3.2.2, we will first illustrate its result
in one dimension. We will also introduce some notation that will be useful later
in these notes.
Example 3.2.3 (One–dimensional Flow Map). Let U denote an open subset
in R and f : U → R be a real-valued C 1 map. Consider the first order ODE
dx
= f (x). (3.18)
dt
Let D ⊂ R × U denote the flow domain for f and θ(t, p), for (t, p) ∈ D, be the
corresponding flow map. For fixed t ∈ R, we consider the map

p 7→ θ(t, p).

We will denote this map by θt : U → U ; thus,

θt (p) = θ(t, p), for all p ∈ U with t ∈ Jp .

We will show that θt is differentiable at any p in U ; that is, we will show that

θt (p + h) − θt (p)
lim exists. (3.19)
h→0 h
3.2. PROPERTIES OF FLOW MAPS 37

Suppose for the moment that the statement in (3.19) is true and denote the
limit in (3.19) by v(t, p). Assume also that v is continuous. Now, since the map
t 7→ θ(t, p) solves the differential equation in (3.18) it follows that
∂θ
(t, p) = f (θ(t, p)) (3.20)
∂t
for all (t, p) ∈ D. Differentiating the expression in (3.20) with respect to p, and
applying the Chain Rule on the right–hand side of (3.20), we obtain that
 
∂ ∂θ ∂θ
(t, p) = f 0 (θ(t, p)) (t, p), (3.21)
∂t ∂p ∂p
where we have interchanged the order of differentiation on the left–hand side of
(3.21). It follows from (3.21) that, for fixed p,
∂θ
vp (t) ≡ v(t, p) = (t, p) (3.22)
∂p
solves the linear differential equation
dy
= ap (t)y, (3.23)
dt
where
ap (t) = f 0 (θ(t, p)), for t ∈ Jp . (3.24)
From (3.22) and the fact that

θ(0, p) = p, for all p ∈ U,

we obtain that
vp (0) = 1. (3.25)
Consequently, vp solve the linear IVP
dy


 = ap (t)y;
dt (3.26)


y(0) = 1.

Motivated by the previous observations, we now prove that (3.19) holds true;
in fact,
θt (p + h) − θt (p)
lim = v(t, p), (3.27)
h→0 h
where vp (t) = v(t, p) is the solution to the IVP (3.26), where ap (t) is given by
(3.24).
Assume first that t > 0 and let T > t be such that [0, T ] ⊂ Jp . By the
Continuous Dependence on Initial Conditions Theorem (Theorem 2.4.1 on page
30 of these notes), there exists δ = δ(T ) > 0 such that

|h| < δ ⇒ p + δ ∈ U,
38 CHAPTER 3. FLOWS

and θ(t, p + h) is defined on [0, T ]. Define

g(t, h) = θ(t, p + h) − θ(t, p) − v(t, p)h, for t ∈ [0, T ], and |h| < δ. (3.28)

We show that, for each t ∈ [0, T ],

g(t, h) = o(h), as |h| → 0; (3.29)

that is,
|g(t, h)|
lim = 0.
|h|→0 |h|
This will prove (3.27).
Write Z t
θ(t, p) = p + f (θ(τ, p)) dτ, (3.30)
0
Z t
θ(t, p + h) = p + h + f (θ(τ, p + h)) dτ, (3.31)
0

and Z t
v(t, p) = 1 + f 0 (θ(τ, p))v(τ, p) dτ, (3.32)
0

for t ∈ [0, T ] and |h| < δ. Substituting (3.30)–(3.32) into (3.28) then yields
Z t
g(t, h) = [f (θ(τ, p + h)) − f (θ(τ, p)) − f 0 (θ(τ, p))v(τ, p)h] dτ, (3.33)
0

for t ∈ [0, T ] and |h| < δ.

Now, using the assumption that f is differentiable on U , we can write

f (x + ∆x) = f (x) + f 0 (x)∆x + R(x, ∆x), (3.34)

where
R(x, ∆x) = o(∆x), as |∆x| → 0. (3.35)
Put
∆θ = θ(t, p + h) − θ(t, p), for t ∈ [0, T ], and |h| < δ. (3.36)
It follows from (3.36) and the Continuous Dependence on Initial Conditions
Theorem (Theorem 2.4.1) that

|h| 6 δ ⇒ |∆θ| 6 |h|eKT , for all t ∈ [0, T ], (3.37)

for some constant K.

Using (3.34) we can write

f (θ(t, p + h)) = f (θ(t, p) + ∆θ)

= f (θ(t, p)) + f 0 (θ(t, p))∆θ + R(θ(t, p), ∆θ),

3.2. PROPERTIES OF FLOW MAPS 39

so that

f (θ(t, p + h)) − f (θ(t, p)) = f 0 (θ(t, p))∆θ + R(θ(t, p), ∆θ), (3.38)

Substituting (3.38) into (3.33) yields

Z t Z t
g(t, h) = f 0 (θ(τ, p))[∆θ − v(τ, p)h] dτ + R(θ(τ, p), ∆θ) dτ,
0 0

or Z t Z t
g(t, h) = f 0 (θ(τ, p))g(τ, p) dτ + R(θ(τ, p), ∆θ) dτ, (3.39)
0 0

for t ∈ [0, T ] and |h| < δ, where we have used the definition of g in (3.28) and
the definition of ∆θ in (3.36).
Next, use the assumption that f is C 1 to obtain

M = max |f 0 (θ(t, p))|. (3.40)

t∈[0,T ]

Take the absolute value on both sides of (3.39), apply the triangle inequality,
and use (3.40) to obtain the estimate
Z t Z t
|g(t, h)| 6 M |g(τ, p)| dτ + |R(θ(τ, p), ∆θ)| dτ, (3.41)
0 0

Let ε > 0 be arbitrary. By (3.35), there exists η > 0 such that

ε
|∆θ| < η ⇒ |R(θ(t, p), ∆θ)| < |∆θ|, (3.42)
T e(M +K)T
for all t ∈ [0, T ]. Note that, in order to obtain the uniform estimate in (3.42),
we have also used the fact that the set

C = {θ(t, p) | t ∈ [0, T ]}

is a compact subset of U .
Next, by making δ smaller, if necessary, so that

δ < ηe−KT , (3.43)

it follows from (3.43), (3.37) and (3.42) that

ε
|h| < δ ⇒ |R(θ(t, p), ∆θ)| < |h|e−M T , for all t ∈ [0, T ]. (3.44)
T
Thus, combining the estimates in (3.41) and (3.44), we obtain that, if |h| < δ,
where δ satisfies (3.43),
Z t
|g(t, h)| 6 ε|h|e−M T + M |g(τ, p)| dτ, for all t ∈ [0, T ]. (3.45)
0
40 CHAPTER 3. FLOWS

Applying the result of Problem 5 in Assignment 1 (Gronwall’s Inequality) to

the estimate in (3.45) we see that

|h| < δ ⇒ |g(t, h)| 6 ε|h|e−M T eM t , for all t ∈ [0, T ],

from which we get that

|h| < δ ⇒ |g(t, h)| 6 ε|h|, for all t ∈ [0, T ].

This proves (3.29). Hence,

|θ(t, p + h) − θ(t, p) − v(t, p)h|

lim = 0,
h→0 |h|

which implies (3.27). Thus, the partial derivative of θ(t, p) with respect to p
exists and
∂θ
(t, p) = v(t, p),
∂p
where vp (t) = v(t, p) for all t ∈ Jp solves the linear IVP in (3.26). The IVP in
(3.26) can be solved to yield
Z t 
0
v(t, p) = exp f (θ(τ, p)) dτ . (3.46)
0

∂θ
Since f 0 is continuous on U , it follows from (3.46) that is continuous on D.
∂p
Proof of Proposition 3.2.2: First observe that

∂θ
(t, p) = u0p (t)
∂t
for all t ∈ Jp , where up : Jp → U is the C 1 solution to the IVP in (3.13). Thus,

∂θ
(t, p) = F (θ(t, p),
∂t
∂θ
and the continuity of on D follows from the continuity of the flow map, θ,
∂t
and the vector field, F .
Next, let (t, p) ∈ D and assume that t > 0. According to Theorem 2.4.1,
for T > t, there exists r = r(T ) > 0 be such that θ(t, p + h) is defined for all
t ∈ [0, T ] and all h ∈ RN such that khk 6 r. Fix t ∈ (0, T ) and define

θt : Br (p) → U

by
θt (q) = θ(t, q), for all q ∈ Br (p).
3.2. PROPERTIES OF FLOW MAPS 41

We first prove the map θt is differentiable at p; in other words, there exists a

linear transformation
Dθt (p) : RN → RN
such that

θ(t, p + h) = θ(t, p) + Dθt (p)h + R(p, h), for h ∈ RN with khk 6 r, (3.47)

where
R(p, h) = o(khk), as khk → 0. (3.48)
Following the outline of the argument given in Example 3.2.3 for the one–
dimensional case, we will show that

Dθt (p) = V (t, p),

where V (t, p) is an N × N matrix which solves the IVP

dY


 = A(t, p)Y ;
dt (3.49)


Y (0) = I,

where A(t, p) is the matrix given by

A(t, p) = DF (θ(t, p)), for t ∈ Jp , p ∈ U ; (3.50)

I denotes the N × N identity matrix, and

Y (t) = [yij (t)]

denotes a matrix–valued function of the real parameter t.

The fundamental theory developed in Chapter 2 applies to the IVP in (3.49).
In fact, it follows from the definition of A(t, p) in (3.50), the fact that the flow
map, θ, is continuous, and the assumption that F is C 1 that A is continuous.
Thus, since the equation in (3.50) is linear, we obtain that the IVP in (3.50)
has a unique solution, V (t, p), defined on for t ∈ Jp . We show that

Dθt (p) = V (t, p);

in other words, according to (3.47) and (3.48), we show that

kθ(t, p + h) − θ(t, p) − V (t, p)hk
lim = 0. (3.51)
khk→0 khk
As in the one–dimensional situation discussed in Example 3.2.3, we define

g(t, h) = θ(t, p + h) − θ(t, p) − V (t, p)h, for t ∈ [0, T ], and khk < r. (3.52)

Next, as in (3.30)–(3.32) in Example 3.2.3, write

Z t
θ(t, p) = p + F (θ(τ, p)) dτ, (3.53)
0
42 CHAPTER 3. FLOWS
Z t
θ(t, p + h) = p + h + F (θ(τ, p + h)) dτ, (3.54)
0
and Z t
V (t, p) = I + DF (θ(τ, p))V (τ, p) dτ, (3.55)
0

for t ∈ [0, T ] and khk < r, where the integral in (3.55) is understood as a
matrix integral (i.e., the integral is a matrix whose components are integrals).
Substituting (3.53)–(3.55) into (3.52) then yields
Z t
g(t, h) = [F (θ(τ, p + h)) − F (θ(τ, p)) − DF (θ(τ, p))V (τ, p)h] dτ, (3.56)
0

for t ∈ [0, T ] and khk < r. Now, using the assumption that F is differentiable
on U , we can write

F (x + ∆x) = F (x) + DF (x)∆x + R(x, ∆x), (3.57)

where
R(x, ∆x) = o(k∆xk), as k∆xk → 0. (3.58)
Put
∆θ = θ(t, p + h) − θ(t, p), for t ∈ [0, T ], and khk < r. (3.59)
It follows from (3.59) and the Continuous Dependence on Initial Conditions
Theorem (Theorem 2.4.1) that

khk 6 r ⇒ |∆θ| 6 khkeKT , for all t ∈ [0, T ], (3.60)

and some constant K.

Using (3.57) we can write

F (θ(t, p + h)) = F (θ(t, p) + ∆θ)

= F (θ(t, p)) + DF (θ(t, p))∆θ + R(θ(t, p), ∆θ),

so that

F (θ(t, p + h)) − F (θ(t, p)) = DF (θ(t, p))∆θ + R(θ(t, p), ∆θ), (3.61)

Substituting (3.61) into (3.57) yields

Z t Z t
g(t, h) = DF (θ(τ, p))[∆θ − V (τ, p)h] dτ + R(θ(τ, p), ∆θ) dτ,
0 0

or Z t Z t
g(t, h) = DF (θ(τ, p))g(τ, p) dτ + R(θ(τ, p), ∆θ) dτ, (3.62)
0 0

for t ∈ [0, T ] and khk < r, where we have used the definition of g in (3.52) and
the definition of ∆θ in (3.59).
3.2. PROPERTIES OF FLOW MAPS 43

Next, use the assumption that F is C 1 to obtain

M = max kDF (θ(t, p))k. (3.63)

t∈[0,T ]

Take the Euclidean norm on both sides of (3.62), apply the triangle inequality,
and use (3.63) to obtain the estimate
Z t Z t
kg(t, h)k 6 M kg(τ, p)k dτ + kR(θ(τ, p), ∆θ)k dτ, (3.64)
0 0

Let ε > 0 be arbitrary. By (3.58), there exists η > 0 such that

ε
k∆θk < η ⇒ kR(θ(t, p), ∆θ)k < k∆θk, (3.65)
T e(M +K)T
for all t ∈ [0, T ]. Note that, in order to obtain the uniform estimate in (3.65),
we have also used the fact that the set

C = {θ(t, p) | t ∈ [0, T ]}

is a compact subset of U .
Next, by making r smaller, if necessary, so that

r < ηe−KT , (3.66)

it follows from (3.66), (3.65) and (3.60) that

ε
khk < r ⇒ kR(θ(t, p), ∆θ)k < khke−M T , for all t ∈ [0, T ]. (3.67)
T
Thus, combining the estimates in (3.64) and (3.67), we obtain that, if khk < r,
where r satisfies (3.66),
Z t
−M T
kg(t, h)k 6 εkhke +M kg(τ, p)k dτ, for all t ∈ [0, T ]. (3.68)
0

Applying Gronwall’s Inequality to the estimate in (3.68) we see that

khk < r ⇒ kg(t, h)k 6 εkhke−M T eM t , for all t ∈ [0, T ],

from which we get that

khk < r ⇒ kg(t, h)k 6 εkhk, for all t ∈ [0, T ].

This proves (3.51) by virtue of (3.52); that is,

kθ(t, p + h) − θ(t, p) − V (t, p)hk
lim = 0,
khk→0 khk
which shows that θt is differentiable at p with

Dθt (p) = V (t, p),

44 CHAPTER 3. FLOWS

where V (·, p) is the solution to the IVP (3.49) on Jp .

To complete the proof of Proposition 3.2.2, we first show that V (t, p) depends
continuously on the parameter p.
Let T and r be as in the first part of this proof. By virtue of the continuity
of the map t 7→ V (t, p), for fixed p ∈ U , and the assumption that F is C 1 , we
can find constants M1 and M2 with

M1 = max kV (t, p)k, (3.69)

t∈[0,T ]

and
M2 = max kDF (θ(t, p + h)k. (3.70)
t∈[0,T ]
khk6r

Using the integral representation (3.55) for the solution to the IVP (3.49) we
obtain, for khk 6 r, that
Z t
V (t, p + h) − V (t, p) = [DF (θ(τ, p + h))V (τ, p + h) − DF (θ(τ, p))V (τ, p)] dτ,
0

which may be written as

Z t
V (t, p + h) − V (t, p) = DF (θ(τ, p + h))[V (τ, p + h) − V (τ, p)] dτ
0
Z t
+ [DF (θ(τ, p + h)) − DF (θ(τ, p))]V (τ, p) dτ.
0

Taking Euclidean norms on both sides of the pervious equation, applying the
triangle inequality, and using (3.69) and (3.70), we obtain that
Z t
kV (t, p + h) − V (t, p)k 6 M2 kV (τ, p + h) − V (τ, p)k dτ
0
Z t
+M1 kDF (θ(τ, p + h)) − DF (θ(τ, p))k dτ.
0
(3.71)
Let ε > 0 be given. Using the continuity of DF and the of the flow map θ, we
obtain δ > 0 such that δ < r and
ε
khk < δ ⇒ kkDF (θ(t, p + h)) − DF (θ(t, p))k < , (3.72)
2T M1 eM2 T

for all t ∈ [0, T ]. It then follows from (3.71) and (3.72) that khk < δ implies
that
Z t
ε
kV (t, p + h) − V (t, p)k < M2 kV (τ, p + h) − V (τ, p)k dτ + M2 T , (3.73)
0 2e
3.2. PROPERTIES OF FLOW MAPS 45

for all t ∈ [0, T . Applying Gronwall’s inequality to (3.73) then yields that
ε
kV (t, p + h) − V (t, p)k < eM2 t , for all t ∈ [0, T ], (3.74)
2eM2 T
provided that khk < δ. It follows from (3.74) that
ε
khk < δ ⇒ kV (t, p + h) − V (t, p)k < , for all t ∈ [0, T ]. (3.75)
2
It remains to show that the map

(t, p) 7→ V (t, p)

is continuous for t ∈ [0, T ] and p ∈ U . Let ε > 0 be given and let s ∈ R be such
that t + s ∈ [0, T ] . Choose δ > 0 as in the previous part of this proof so that
(3.75) holds true. By the continuity of the map

t 7→ V (t, p), for t ∈ Jp ,

way may assume, by making δ smaller if necessary that

ε
|s| < δ ⇒ t + s ∈ [0, T ] and kV (t + s, p) − V (t, p)k < . (3.76)
2
It then follows from (3.75) and (3.76) that |s| < δ and khk < δ implies that

kV (t + s, p + h) − V (t, p)k = k(V (t + s, p + h) − V (t + s, p)) + (V (t + s, p) − V (t, p))k

6 kV (t + s, p + h) − V (t + s, p)k + kV (t + s, p) − V (t, p)||

< ε,

and the proof of Proposition 3.2.2 is now complete. 

46 CHAPTER 3. FLOWS
Chapter 4

Continuous Dynamical
Systems

We saw in the previous chapter that for any C 1 vector field, F : U → RN , defined
in an open set U ⊆ RN , there exists a corresponding flow map, θ : D → U ,
defined on a flow domain D ⊆ R × U . In other words, for each p ∈ U , the
function up : Jp → U defined by

up (t) = θ(t, p), for all t ∈ Jp ,

is the unique solution to the IVP

dx


 = F (x);
dt (4.1)


x(0) = p,

defined on a maximal interval of existence, Jp . We also saw in Chapter 3 that

the flow map, θ : D → U , is a C 1 map. In this chapter we begin by noting that,
in addition to being C 1 , the flow map, θ, satisfies

θ(0, p) = p, for all p ∈ U, (4.2)

and
θ(t + s, p) = θ(t, θ(s, p)), (4.3)
for all p ∈ U , and all t, s ∈ R with t + s, s ∈ Jp and t ∈ Jθ(s,p) . The identity in
(4.2) follows from the fact that the map t 7→ θ(t, p), solves the IVP (4.1). The
identity in (4.3) was proved in Problem 1 of Assignment #2.
For the case in which the identity in (4.3) holds true for all t, s ∈ R, the flow
of F defines an action of the group R, under addition, on the set U . In general
if G denotes a group, and S a set, an action of G on S is a map

ϕ: G × S → S

47
48 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

satisfying
ϕ(e, s) = s, for all s ∈ S, (4.4)
where e is the group identity, and

ϕ(gh, s) = ϕ(g, ϕ(h, s)) for all s ∈ S, and all g, h ∈ G. (4.5)

In the case in which θ(t, p) is defined for all t ∈ R and p ∈ U , we see that
ϕ(t, p) = θ(t, p) for all (t, p) ∈ R × U satisfies the group action axioms in (4.4)
and (4.5) for G = R, with addition as the group operation, and S = U , in view of
(4.2) and (4.3), respectively. For the particular case in which the action of R on
U is defined by the flow map of a C 1 vector field on U , the map (t, p) 7→ θ(t, p)
is C 1 ; this will be the definition of a continuous dynamical system that we will
use in these notes.

4.1 Definition of Continuous Dynamical Systems

A continuous dynamical system on an open set U ⊆ RN is a C 1 map

θ: R × U → U

which satisfies the group action axioms:

θ(0, p) = p, for all p ∈ U, (4.6)

and
θ(t + s, p) = θ(t, θ(s, p)), for all p ∈ U, and all t, s ∈ R. (4.7)
Thus, a continuous dynamical system can be thought of as a C 1 action of the
group (R, +) on the open set U ⊆ RN .
Example 4.1.1. Let F : U → RN be a C 1 vector field. Then, the flow map, θ,
of F defined in the previous chapter is a continuous dynamical system if, for all
p ∈ U , the IVP
dx


 = F (x);
dt


x(0) = p,
has a solution that exists for all t ∈ R.
Example 4.1.2. Define θ : R × R2 → R2 by
     
p cos t − sin t p
θ t, = , for t, p, q ∈ R.
q sin t cos t q

To see that θ defines a dynamical system, first observe that

    
p p
θ 0, =I ,
q q
4.2. ORBITS 49

where I denotes the 2 × 2 identity matrix. Consequently,

    
p p
θ 0, = .
q q

Finally, use the trigonometric identities

cos(t + s) = cos t cos s − sin t sin s

sin(t + s) = cos t sin s + sin t cos s

to verify that
      
p cos t − sin t cos s − sin s p
θ t + s, = ,
q sin t cos t sin s cos s q

from which we obtain that

      
p p
θ t + s, = θ t, θ s, .
q q

For each t ∈ R, the dynamical system, θ : R × U → U , induces a map on U ,

denoted by θt : U → U , and given by

θt (p) = θ(t, p), for all p ∈ U. (4.8)

The map θt : U → U defined by (4.8) is C 1 . Furthermore, it follows from (4.6)

and (4.7) that
θt ◦ θ−t = id,
where id denotes the identity map in U , and

θ−t ◦ θt = id.

It then follows that θt is invertible with inverse θ−t . Hence, θt is C 1 with an

inverse which is also C 1 . We say that θt is a diffeomorphism of U . Hence, a
dynamical system θt induces a family of diffeomorphisms, {θt }t∈R , of the set U
into itself.

Example 4.1.3. For the dynamical system defined in Example 4.1.2,

θt : R2 → R2

is a rotation of the plane through an angle of t (in radians) of the plane R2 .

4.2 Orbits
In the study of group actions, it is of interest to look at orbits of points in the
set on which the group acts.
50 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

Definition 4.2.1 (Orbits). Let θ(t, p) denote a dynamical system on an open

set U ⊆ RN . Given p ∈ U , the orbit of the flow, θt , through p is the set, γp ,
defined by
γp = {x ∈ U | x = θ(t, p) for some t ∈ R};
in other words, γp is the image of the map θp : R → U defined by

θp (t) = θ(t, p), for all t ∈ R.

Example 4.2.2. For the dynamical p system θ : R × R2 → R2 given in Example

4.1.2, γ(p,q) is a circle of radius p + q 2 around the origin for the case (p, q) 6=
2

(0, 0); and γ(0,0) = {(0, 0)}. Figure 4.2.1 shows γ(1,0) , γ(0,0) and another typical

JJ γ
]
'$(1,0)
]
r
(1, 0) p
&%

Figure 4.2.1: Phase Portrait of θ in Example 4.1.2

orbit of the dynamical system θ defined in Example 4.1.2. The arrows on the
the two circular orbits portrayed in the figure indicate the direction on the orbit
induced by the parametrization θ(p,q) : R → R2 , defined by
  
p
θ(p,q) (t) = θ t, , for all t ∈ R, (4.9)
q

at t increases.

Definition 4.2.3 (Phase Portrait). A depiction of all possible kinds of orbits

that a dynamical system can have is known as a phase portrait of the system.

Example 4.2.4. Figure 4.2.1 shows the phase portrait of the dynamical system,
θ : R × R2 → R2 , given in Example 4.1.2. Observe that, for (p, q) 6= (0, 0) and
4.3. INFINITESIMAL GENERATORS 51

θ(p,q) as defined in (4.9),

kθ(p,q) (t)k2 = (p cos t − q sin t)2 + (p sin t + q cos t)2

= p2 cos2 t − 2pq sin t cos t + q 2 sin2 t

+p2 sin2 t + 2pq sin t cos t + q 2 cos2 t

= p2 + q 2 , for all t ∈ R,
p
which shows that γ(p,q) lies in the circle of radius r = p2 + q 2 around the
origin in R2 . On the other hand, if (x, y) ∈ Sr ((0, 0)), the circle of radius r
around the origin in R2 , where r > 0, by letting
y  
q
t = arctan − arctan ,
x p
we can show that  
x
θ(p,q) (t) = .
y
In other words, (x, y) ∈ γ(p,q) . Consequently, γ(p,q) = Sr ((0, 0)), for r =
p
p2 + q 2 , as claimed in Example 4.2.2. On the other hand, if (p, q) = (0, 0),
then γ(p,q) = {(0, 0)}. Thus, the singleton {(0, 0)} and concentric circles around
the origin are the only kinds of orbits that the dynamical system, θ(t, p), defined
in Example 4.1.2 can have.

4.3 Infinitesimal Generator of a Dynamical Sys-

tem
Given a dynamical system, θ : R × U → U , on an open set U ⊆ RN , we can
define a vector field, F : U → RN , as follows
∂
F (x) = θ(t, x) , for all x ∈ U ; (4.10)
∂t t=0

in other words,
θ(h, x) − θ(0, x)
F (x) = lim , for all x ∈ U. (4.11)
h→0 h
Since we are assuming that the dynamical system, θ : R×U → U is a C 1 map, it
follows that F : U → RN defined in (4.10) is a C 1 vector field defined in U . We
show next that θ : R × U → U is the flow map for the vector field F . The vector
field, F , defined in (4.11) is called the infinitesimal generator of the dynamical
system θt , for t ∈ R.
Thus, we need to show that the map, θp : R → U , given by
θp (t) = θ(t, p), for all t ∈ R (4.12)
52 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

is the unique solution to the IVP

dx


 = F (x);
dt (4.13)


x(0) = p,
Using the group action axiom for θ(t, p) in (4.7), we see that θp defined in
(4.12) satisfies
θp (t + h) = θ(t + h, p) = θ(h, θ(t, p)), for t, h ∈ R.
We then have that
θp (t + h) = θ(h, θp (t)), for t, h ∈ R. (4.14)
Thus, for h 6= 0 we obtain from (4.14) that
θp (t + h) − θp (t) θ(h, θp (t)) − θ(0, θp (t))
= , (4.15)
h h
where we have also used the group action axiom for θ(t, p) in (4.6).
Next, letting h → 0 in (4.15) and using the definition of the field, F , in
(4.11), we obtain that
θp0 (t) = F (θp (t)),
which shows that θp solves the differential equation in the IVP (4.13). Finally,
since θp (0) = p, by the group action axiom for θ(t, p) in (4.6), we see that
θp : R → U solves the IVP in (4.13), which was to be shown.
Remark 4.3.1. We have seen that every continuous dynamical system, θ : R ×
U → U , has an associated C 1 vector field, F : U → RN , given by (4.10); namely,
the infinitesimal generator of θ. However, it is not the case that every C 1 vector
field, F , has a dynamical system, θ : R × U → U , in the sense defined in Section
4.1. The issue at hand is that solutions to the IVP (4.13), for given p ∈ U ,
might not be defined for all t ∈ R.
Example 4.3.2. Let θ : R×R2 → R2 be the dynamical system given in Example
4.1.2. To find the infinitesimal generator of θ, we first compute
       
∂θ x − sin t − cos t x x
t, = , for all ∈ R2 ;
∂t y cos t − sin t y y
so that     
x ∂θ x
F = 0,
y ∂t y
  
0 −1 x
=
1 0 y
 
−y
= ,
x
4.4. FIXED POINTS 53
 
x
for all ∈ R2 . It then follows that the infinitesimal generator of θ is the
y
vector field, F : R2 → R2 , given by
     
x −y x
F = , for all ∈ R2 .
y x y
In other words, the dynamical system, θ : R × R2 , given by
     
p cos t − sin t p
θ t, = , for t, p, q ∈ R,
q sin t cos t q
is the flow of the linear system of differential equations

dx
 dt = −y;





 dy

 = x.
dt

4.4 Fixed Points and Equilibrium Solutions

Let θ : R × U → U be a C 1 dynamical system in U with infinitesimal generator
F : U → RN . A fixed point of the flow θt in U is a point p∗ ∈ U such that
θ(t, p∗ ) = p∗ , for all t ∈ R,
or
θp∗ (t) = p∗ , for all t ∈ R. (4.16)
Taking the derivative with respect to t on both sided of (4.16) we obtain that
θp0 ∗ (t) = 0, for all t ∈ R,
so that, by the definition of the infinitesimal generator of θt ,
F (θp∗ (t)) = 0, for all t ∈ R. (4.17)
Combining (4.17) with (4.16) we obtain
F (p∗ ) = 0. (4.18)
Thus, a fixed point of the dynamical system with infinitesimal generator F : U →
RN is a solution to the equation
F (p) = 0. (4.19)
Solutions to (4.19) are also known as equilibrium points, or singular points. A
point p ∈ U which is not a fixed point of the system generated by F is called a
regular point of F . A solution to the system
dx
= F (x),
dt
defined by (4.16) where p∗ ∈ U satisfies (4.18) is called an equilibrium solution.
54 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

Example 4.4.1. Consider the two–dimensional system


dx
 dt = λx + y;



(4.20)

 dy

 = λy,
dt
where λ 6= 0.
In this case the field, F : R2 → R2 , is given by
     
x λx + y x
F = , for all ∈ R2 .
y λy y

The equilibrium points are therefore solutions to the system


λx + y = 0;
(4.21)
λy = 0.

Since λ 6= 0, the only solution to the system in (4.21) is the origin, (0, 0), in R2 .
Thus, (0, 0) is the only equilibrium point of the system in (4.20).

4.5 Limit Sets

The dynamical system corresponding to the two–dimensional system (4.20) in
Example 4.4.1 is given by
       
p 1 t p p
θ t, = eλt for t ∈ R, ∈ R2 . (4.22)
q 0 1 q q
   
x(t) p
Thus, if ∈ γ p , the orbit of
  , it follows from (4.22) that
y(t) q q
   
x(t) p p
6 eλt 2 + t2 , for all t ∈ R. (4.23)
y(t) q

Hence, if λ < 0, we obtain from (4.23) that

 
x(t)
lim = 0,
t→∞ y(t)

from which we conclude that, if λ < 0, then

    
p 0
lim θ t, = .
t→∞ q 0
 
0
We then say that is an ω–limit point (“omega limit point”) of the orbit
0
γ p  .
q
4.5. LIMIT SETS 55

Definition 4.5.1 (Limit Sets). Let θ : R × U → U denote a dynamical system

on an open set U ⊆ RN , and let γp , for p ∈ U , be an the orbit of the point p
under the flow θt .

• (ω–limit point) A point q ∈ U is said to be an ω–limit point of γp if there

exists a sequence of real values, (tm ), such that tm → +∞ as m → ∞ and

lim θ(tm , p) = q.
m→∞

• (ω–limit set) The set of all ω–limit sets of the orbit γp is called the ω–limit
set of γp and is denoted by ω(γp ).

• (α–limit point) A point q ∈ U is said to be an α–limit point of γp if there

exists a sequence of real values, (tm ), such that tm → −∞ as m → ∞ and

lim θ(tm , p) = q.
m→∞

• (α–limit set) The set of all α–limit sets of the orbit γp is called the α–limit
set of γp and is denoted by α(γp ).

Example 4.5.2. Let θ : R × R2 → R2 be the dynamical system given in (4.22)

corresponding to the two–dimensional system in (4.20). If λ < 0, then
 
0
ω(γ p ) =
  ,
q 0
 
p
for all ∈ R2 .
q
   
p 0
On the other hand, if 6= , and λ < 0, then
q 0

α(γp ) = ∅.
q

To see why the last claim is true, note that, if q 6= 0,

  
p
θ t, > |q|eλt → ∞, as t → −∞.
q

Consequently,   
p
θ tm , → ∞, as m → ∞,
q
for any sequence of real numbers, (tm ), such that tm → −∞ as m → ∞;
therefore,   
p
lim θ tm ,
m→∞ q
does not exist for the case q 6= 0 and λ < 0.
56 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

Next, if q = 0 and p 6= 0, while λ < 0, note that

  
p
θ t, > |p|eλt → ∞, as t → −∞,
q

and the same result is obtained.

Finally, observe that, regardless of the sign of λ,
 
0
α(γ0 ) = ω(γ0 ) = ,
0 0 0
 
0
since is the equilibrium point of the system.
0

4.6 Properties of Limit Sets

Definition 4.6.1 (Invariant Sets). Let θ : R × U → U be a dynamical system
on U , an open subset of RN . A subset A of U is said to be invariant under the
flow θ if for every q ∈ A, θ(t, q) ∈ A for all t ∈ R.
Example 4.6.2. For p ∈ U , the orbit γp is invariant under the flow. In fact, if
q ∈ γp , then q = θ(to , p), for some to ∈ R. We then have that, for any t ∈ R,

θ(t, q) = θ(t, θ(to , p)) = θ(t + to , p) ∈ γp .

In the next proposition we will see that the limit sets of an orbit is also
invariant.
Proposition 4.6.3. Let θ : R × U → U be a dynamical system on U , an open
subset of RN . For any p ∈ U , the ω–limit set and α–limit set of γp are closed,
invariant subsets of U .
Proof: We prove the assertions in the proposition for ω(γp ); the arguments for
α(γp ) are analogous.
We first show that ω(γp ) is invariant. Let q ∈ ω(γp ); then there exists a
sequence of real numbers, (tm ), such that tm → ∞ as m → ∞ and

q = lim θ(tm , p). (4.24)

m→∞

Now, by the continuity of the map θt : U → U for each t ∈ R, we obtain from

(4.24) that
θ(t, q) = lim θ(t, θ(tm , p)) (4.25)
m→∞

for any t ∈ R. On the other hand,

θ(t, θ(tm , p)) = θ(t + tm , p) = θ(tm , θ(t, p)). (4.26)

Thus, combining (4.25) and (4.26), we see that

θ(t, q) = lim θ(tm , θ(t, p)),

m→∞
4.6. PROPERTIES OF LIMIT SETS 57

which shows that

θ(t, q) ∈ ω(γθ(t,p) ), for any t ∈ R. (4.27)

However, γθ(t,p) = γp , for any t ∈ R, since θ(t, p) ∈ γp , for any t ∈ R (see, for
instance, Problem 3 in Assignment #4). Thus, (4.27) can be written as

θ(t, q) ∈ ω(γp ), for any t ∈ R,

which shows that ω(γp ) is invariant under the flow.

Next, we show that ω(γp ) is closed. This is equivalent to showing that the
complement, ω(γp )c , is open.
Arguing by contradiction, suppose that there exist q ∈ ω(γp )c and a sequence
of points, (qm ), in ω(γp ) such that

lim kqm − qk = 0. (4.28)

m→∞

Now, since each qm is in ω(γp ), we can construct a sequence of real numbers,

(tm ) such that tm → ∞ as m → ∞, and

1
kθ(tm , p) − qm k < , for all m ∈ N. (4.29)
m
Applying the triangle inequality we obtain that

1
kθ(tm , p) − qk < + kqm − qk, for all m ∈ N, (4.30)
m
where we have also used the estimate in (4.29). It follows from (4.30), (4.29)
and the Squeeze Lemma that

lim kθ(tm , p) − qk = 0,
m→∞

which shows that q ∈ ω(γp ), contradicting the assumption that q ∈ ω(γp )c . This
contradiction shows that ω(γp )c is open, and therefore ω(γp ) is closed. 

It is possible for the ω–limit set of an orbit to be empty. For instance, let
θ : R × R2 → R2 be given by
       
p 1 t p p
θ t, = eλt for t ∈ R, ∈ R2 , (4.31)
q 0 1 q q
 
x(t)
where λ > 0, and consider the orbit γ  .
1 Points, , in γ1 are of the
0 y(t) 0

form    
x(t) 1
= eλt , for all t ∈ R.
y(t) 0
58 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

We then have that

 
x(t)
= eλt → ∞, as t → ∞,
y(t)

since λ > 0. It then follows that ω(γ1 ) = ∅.

  0
1
The reason the limit of θ t, as t → ∞ fails to exist is that the orbit
  0
1
of is unbounded for positive values of t.
0
Definition 4.6.4 (Semi–Orbits). Let θ : R×U → U denote a dynamical system
on an open set U ⊆ RN . Given p ∈ U , the forward semi–orbit of the flow, θt ,
through p is the set, γp+ , defined by

γp+ = {x ∈ U | x = θ(t, p) for some t > 0}.

Similarly, the backward semi–orbit through p, denoted by γp− , is the set

γp− = {x ∈ U | x = θ(t, p) for some t < 0}.

Observe that γp = γp− ∪ γp+ .

Example 4.6.5. Let θ : R × R2 → R2 be as given in (4.31), where λ > 0; then

γ+1 = {(x, y) ∈ R2 | x > 1, y = 0}

0

and
γ−1 = {(x, y) ∈ R2 | 0 < x < 1, y = 0}.
0

In the next proposition we will see that, if γp+ is bounded, then ω(γp ) is
nonempty. Similarly, if γp− is bounded, then α(γp ) is nonempty. In the previous
example, note that γ−1 is bounded and α(γ1 ) = {(0, 0)}; so that, α(γ1 ) 6= ∅.
0 0 0

Proposition 4.6.6. Let θ : R × U → U be a dynamical system on U , an open

subset of RN , and p ∈ U . If γp+ lies in a compact subset of U , then ω(γp ) is
nonempty, compact and connected. Similarly, if γp− lies in a compact subset of
U , then α(γp ) is nonempty, compact and connected.
Proof: We prove the result for the ω–limit set; the arguments for the α–limit
set are analogous.
Assume that γp+ ⊂ K, where K is a compact subset of U . For any sequence,
(tm ), of real numbers such that tm → ∞ as m → ∞, the set {θ(tm , p)} is a
subset of K. Hence, since K is compact, there exists q ∈ K and a subsequence,
(tmk ) such that tmk → ∞ as k → ∞, and

q = lim θ(tmk , p);

k→∞
4.6. PROPERTIES OF LIMIT SETS 59

that is, q ∈ ω(γp ), and therefore ω(γp ) 6= ∅.

Next, observe that, if q ∈ ω(γp ), then

q = lim θ(tm , p),

m→∞

for a sequence of positive numbers, tm , tending to infinity as m → ∞. We then

have that θ(tm , p) ∈ K for all m. Consequently, since K is closed, q ∈ K. Thus,

ω(γp ) ⊆ K.

since ω(γp ) is closed, by Proposition 4.6.6, and K is compact, it follows that

ω(γp ) is compact.
It remains to show that ω(γp ) is connected. Assume to the contrary that
there exist nonempty open subsets, U1 and U2 , of U such that

U1 ∩ U2 = ∅, (4.32)

U1 ∩ ω(γp ) 6= ∅ and U2 ∩ ω(γp ) 6= ∅, (4.33)

and
ω(γp ) ⊆ U1 ∪ U2 . (4.34)
Define
C1 = ω(γp ) ∩ U1 and C1 = ω(γp ) ∩ U2 . (4.35)
It follows from (4.32) and (4.34) that

C1 ∩ C2 = ∅, (4.36)

and
ω(γp ) = C1 ∪ C2 . (4.37)
Next, we see that C1 and C2 are compact. This follows from the fact that
C1 and C2 are both closed subsets of ω(γp ), which was shown to be compact
previously in this proof. To see that C1 is closed, let (qm ) be a sequence of
points in C1 such that
lim kqm − qk = 0, (4.38)
m→∞

for some q ∈ ω(γp ). We show that q ∈ C1 . If this is not the case, it follows
from (4.37) that q ∈ C2 . Then, by the definition of C2 in (4.35), q ∈ U2 . It then
follows from (4.38) that there exists M ∈ N such that

m > M ⇒ qm ∈ U2 .

However, this is impossible in view of (4.32) since qm ∈ U1 for all m ∈ N. We

therefore conclude that q ∈ C1 , and therefore C1 is closed. A similar argument
shows that C2 is closed.
From (4.36) and the fact that C1 and C2 are compact, it follows that

δ = dist(C1 , C2 ) > 0. (4.39)

60 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

By virtue of (4.33) and (4.33), we can find q1 ∈ C1 and q2 ∈ C2 . Then, there

exist sequences of real numbers, (tm ) and (sm ), such that

tm < sm , for all m ∈ N, (4.40)

and
tm → ∞ and sm → ∞, as m → ∞, (4.41)
with
δ δ
kθ(tm , p) − q1 k < and kθ(sm , p) − q2 k < , for all m ∈ N. (4.42)
4 4
Hence, for each m ∈ N,
δ δ
inf kθ(tm , p) − qk < and inf kθ(sm , p) − qk < . (4.43)
q∈C1 4 q∈C2 4
Define the functions f1 : R → R and f2 : R → R by

f1 (t) = dist(θ(t, p), C1 ) = inf kθ(t, p) − qk (4.44)

q∈C1

and
f2 (t) = dist(θ(t, p), C2 ) = inf kθ(t, p) − qk (4.45)
q∈C2

for all t ∈ R. By the continuity of the map t 7→ θ(t, p) and the continuity of the
distance fucntion, it follows that f1 and f2 are continuous functions.
It follows from (4.43) and the definition of f1 and f2 in (4.44) and (4.45),
respectively, that
δ δ
f1 (tm ) < and f2 (sm ) < . (4.46)
4 4
Next, let q 2 ∈ C2 be such that f2 (tm ) = kθ(tm , p) − q 2 k. It then follows from
(4.39) and the triangle inequality that

δ 6 kq1 − q 2 k 6 kq1 − θ(tm , p)k + kθ(tm , p) − q 2 k,

so that
δ
δ<
+ f2 (tm ),
4
where we have used the first inequality in (4.42). We therefore obtain that
3δ
f2 (tm ) > . (4.47)
4
Combining (4.47) and the first inequality in (4.46) that
δ
f1 (tm ) − f2 (tm ) < − < 0. (4.48)
2
Similar calculations show that
δ
f1 (sm ) − f2 (sm ) > > 0. (4.49)
2
4.6. PROPERTIES OF LIMIT SETS 61

It follows from (4.40), (4.48), (4.49), the continuity of f1 and f2 , and the in-
termediate value theorem that there exists a sequence, (τm ), of real numbers
with
tm < τm < sm , for all m, (4.50)
and
f1 (τm ) − f2 (τm ) = 0, for all m ∈ N,
or
dist(θ(τm , p), C1 ) = dist(θ(τm , p), C2 ) for all m ∈ N. (4.51)
In view of (4.41) and (4.50) we also see that

τm → ∞, as m → ∞. (4.52)

Since θ(τm , p) ∈ K, for all m, by assumption, there exists a subsequence, (τmk ),

such that
lim θ(τmk , p) = q, (4.53)
k→∞

for some q ∈ K.
It follows from (4.52) and (4.53) that

q ∈ ω(γp ). (4.54)

Also, from (4.53), (4.51) and the continuity of the distance function we obtain
that
dist(q, C1 ) = dist(q, C2 ). (4.55)
From (4.54) and (4.37) we obtain that either q ∈ C1 or q ∈ C2 . If q ∈ C1 , then

dist(q, C1 ) = 0, (4.56)

and, by virtue of (4.39),

dist(q, C2 ) > δ > 0. (4.57)
Observe that (4.56) and (4.57) are in contradiction with (4.55). Thus, we must
have that q ∈ C2 , which leads to

dist(q, C1 ) > δ > 0 and dist(q, C2 ) = 0,

which, again, contradict (4.55). This contradiction establishes that ω(γp ) is

connected, and therefore the proof of Proposition 4.6.6 is now complete. 

Example 4.6.7. In Example 4.2.4 we showed that the orbits of the system

dx
 dt = −y;



(4.58)

 dy

 = x,
dt
62 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

consist of the singleton {(0, 0)} and concentric circles around the origin. Some
of this orbits are shown in Figure 4.2.1 on page 50. For instance,

γ1 = {(x, y) ∈ R2 | x2 + y 2 = 1}.

0

We show that
ω(γ1 ) = γ1 . (4.59)
0 0

Indeed, if (x, y) ∈ γ1 , there exists to ∈ [0, 2π) such that

0

(x, y) = θ(to , 1, 0) = (cos(to ), sin(to )).

Setting tm = to + 2πm, we see that

tm → ∞, as m → ∞,

and
(x, y) = θ(tm , 1, 0), for all m ∈ N,

by the fact that sin and cos are periodic functions of period 2π. Consequently,

(x, y) = lim θ(tm , 1, 0),

m→∞

and therefore (x, y) ∈ ω(γ1 ). Thus,

0

γ1 ⊆ ω(γ1 ). (4.60)

0 0

To see why the reverse inclusion holds, let (p, q) ∈ ω(γ1 ). Then, there exists
0

a sequence of real numbers, (tm ), such that

tm → ∞, as m → ∞,

and
lim θ(tm , 1, 0) = (p, q).
m→∞

Note that kθ(tm , 1, 0)k = 1 for all m, so that, by continuity of the norm,
k(p, q)k = 1, which shows that (p, q) ∈ γ1 . Consequently,
0

ω(γ1 ) ⊆ γ1 . (4.61)

0 0

Combining (4.60) and (4.61) fields (4.59).

4.6. PROPERTIES OF LIMIT SETS 63

Example 4.6.8 (Negative Gradient Flows). Let U be an open subset of RN

which contains the origin, 0. Let V : U → R be a C 2 function satisfying V (x) > 0
for all x ∈ U \{0} and V (0) = 0. Put F = −∇V and, for p ∈ U , consider the
IVP
dx


 = F (x);
dt (4.62)


x(0) = p.
In this example we show that 0 is an equilibrium point of F and, if F has no
equilibrium points in some neighborhood, Br (0), of 0, other than zero, then
ω(γp ) = {0} for all p ∈ Br (0) which are sufficiently close to 0. We will prove
these facts in stages.

(a) The vector 0 is an equilibrium point of the system in (4.62).

b be any unit vector in RN and observe that, for any t 6= 0,

Proof: Let w

b − V (0) > 0.
V (tw) (4.63)

Dividing the expression in (4.63) by t > 0 and letting t → 0+ yields

∇V (0) · w
b > 0. (4.64)

Similarly, dividing the expression in (4.63) by t < 0 and letting t → 0−

yields
∇V (0) · w
b 6 0. (4.65)
Combining (4.64) and (4.65) we obtain that

∇V (0) · w
b = 0, b ∈ Rn with kwk
for all w b = 1,

which implies that ∇V (0) = 0; hence F (0) = 0 and therefore 0 is an equi-

librium point of the differential equation in (4.62). 

(b) Let p ∈ U and suppose that p is not an equilibrium point of F . Then, the
function V is strictly decreasing along γp ; that is, the function V (up (t))
decreases with increasing t, where up : Jp → U is the unique solution to the
IVP in (4.62) defined on a maximal interval of existence Jp .

Proof: Let up : Jp → U be the solution to IVP (4.62) defined on a maximal

interval of existence Jp , where p is not an equilibrium point of F . We then
have, by uniqueness, that

F (up (t)) 6= 0, for all t ∈ Jp (4.66)

Observe that, by the Chain Rule,

d
[V (up (t))] = ∇V (up (t)) · u0p (t),
dt
64 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

or
d
[V (up (t))] = ∇V (up (t)) · F (up (t)), (4.67)
dt
since up solves the differential equation in (4.62). Consequently, using the
assumption that F = −∇V , we obtain from (4.67) that

d
[V (up (t))] = −k∇V (up (t))k2 , for all t ∈ Jp . (4.68)
dt
Thus, if p is not an equilibrium point of F , it follows from (4.68) and (4.66)
that V is strictly decreasing along the orbit γp as t increases. 

(c) For every r > 0 with Br (0) ⊂ U , such that Br (0)\{0} contains no equilib-
rium points of F , there exists δ > 0 such that, Bδ (0) ⊂ U and, for every
p ∈ Bδ (0), up (t) ∈ Br (0) for all t ∈ Jp ∩ [0, ∞).

Proof: Let r > 0 be such that Br (0) ⊂ U and Br (0) contains no equilibrium
points of F other than 0. Put

ε = inf V (x). (4.69)

kxk=r

By the assumption that V > 0 on U \{0}, the compactness of ∂Br (0), and
the continuity of V we have that ε > 0; thus, since V (0) = 0 and V is
continuous, there exists δ > 0 such that

kxk < δ ⇒ V (x) < ε. (4.70)

It follows from the result of part (a) that

V (up (t)) 6 V (p), for all t ∈ Jp ∩ [0, ∞). (4.71)

Consequently, if kpk < δ, it follows from (4.71) and (4.70) that

V (up (t)) < ε, for all t ∈ Jp ∩ [0, ∞). (4.72)

We claim that (4.72) and (4.69) imply that, if kpk < δ, then

up (t) ∈ Br (0), for all t ∈ Jp ∩ [0, ∞). (4.73)

Suppose that (4.73) does not hold. Then there exists t1 > 0 such that
t1 ∈ Jp and kup (t1 )k > r. By the intermediate value theorem, there exists
t ∈ (0, t1 ) such that kup (t)k = r; thus, by (4.69),

V (up (t) > ε,

which is in direct contradiction with (4.72). Hence, (4.73) must hold true
for every p ∈ Bδ (0). 
4.6. PROPERTIES OF LIMIT SETS 65

(d) Let δ > 0 be as obtained in the previous part. Then, for every p ∈ Bδ (0),
the forward semi–orbit, γp+ , is bounded. Deduce therefore that ω(γp ) 6= ∅.

Proof: From the result of the previous part, it follows that if p ∈ Bδ (0),
then
up (t) ∈ Br (0), for all t ∈ Jp ∩ [0, ∞). (4.74)
It then follows from the global existence result proved in the lecture notes
that Jp ∩[0, ∞) = [0, ∞); in other words, up (t) is defined for all t > 0. Thus,
the positive semi–orbit, γp+ , is defined an

γp+ ⊆ Br (0), (4.75)

where we have used (4.74); that is, γp+ is bounded.

It follows from (4.75) that γp+ lies in a compact subset of U ; hence, ω(γp ) 6=
∅; furthermore,
ω(γp ) ⊆ Br (0). (4.76)


(e) Let r > 0 be as given in part (c) and δ > 0 as obtained in part (c). Prove
that, for any p ∈ Bδ (0), ω(γp ) = {0}.

Proof: First observe that, if p = 0, then γp = {0}, since 0 is an equilibrium

point of the system. Thus, the statement is true in the case p = 0. So,
assume for the rest of this argument that p ∈ Bδ (0)\{0}.
Observe that, for p ∈ Bδ (0)\{0},

up (t) 6= 0, for all t > 0, (4.77)

by uniqueness of the solution to the IVP in (4.62).

We next observe that, by (4.68) in part (b) and (4.77), V decreases along
γp+ as t increases. We then have that lim V (up (t)) exists. We prove that
t→∞

lim V (up (t)) = 0. (4.78)

t→∞

Arguing by contradiction, if (4.78) does not hold true, then there exists
εo > 0 such that
lim V (up (t)) = εo . (4.79)
t→∞

Thus, since V (up (t)) decreases with t, it follows that

V (θ(t, p)) > εo , for all t > 0, (4.80)

where we have written θ(t, p) for up (t). Combining (4.80) with (4.72) in the
proof of part (c), we obtain that

εo 6 V (θ(t, p)) 6 ε, for all t > 0, (4.81)

66 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

where ε is as given in (4.69).

Define
C = {θ(t, p) | t > 0 and εo 6 V (θ(t, p)) 6 ε}. (4.82)
It follows from the continuity of the functions V and θ and (4.74) that C
is a closed subset of a compact set; hence, C is compact. It is also the case
that θ(t, p) 6= 0 for all t > 0, since V (θ(t, p)) > εo for all θ(t, p) ∈ C, and
V (x) = 0 only when x = 0. The compactness of C defined in (4.82) and the
continuity of ∇V then implies that there exists a positive constant ν such
that
d
[V (θ(t, p))] 6 −ν, for all t > 0, (4.83)
dt
where we have used (4.68). Integrating on both sides of the inequality in
(4.83) from 0 to t, we obtain
V (θ(t, p)) 6 V (p) − νt,
which shows that
1
V (θ(t, p)) < 0, for t > V (p),
ν
which is impossible since V (x) > 0 for all x ∈ U . Hence, (4.79) is not
possible, and therefore (4.78) must be true.
Now, given any x ∈ ω(γp ), there exists a sequence of positive real numbers,
(tm ), such that tm → ∞ as m → ∞ such that
lim kθ(tm , p) − xk = 0. (4.84)
m→∞

It then follows from the continuity of V and (4.84) that

lim V (θ(tm , p)) = V (x). (4.85)
m→∞

Hence, using (4.78), it follows from (4.85) that

V (x) = 0, for all x ∈ ω(γp ). (4.86)
Therefore, since V (x) > 0 for all x ∈ U \{0} and V (0) = 0, we obtain from
(4.86) that
x ∈ ω(γp ) ⇒ x = 0;
in other words, ω(γp ) = {0}. We therefore conclude that ω(γp ) = {0} for
all p ∈ Bδ (0), which was to be shown. 

4.7 Cycles and Periodic Solutions

In Example 4.6.7 we saw that the two–dimensional system in (4.58) has an orbit,
γ(1,0) , which is a simple closed curve in R2 ,

γ(1,0) = {(x, y) ∈ R2 | x2 + y 2 = 1}.

4.7. CYCLES 67

Definition 4.7.1 (Cycles). An orbit of a dynamical system is a called a cycle

when it is a simple closed curve.
Let θ : R × U → U be a dynamical system in an open set U ⊆ RN . Suppose
that for p ∈ U , the orbit of p, γp , is a cycle. Then, there exists a positive number
T such that θp (T, p) = p and θp : [0, T ] → U is a parametrization of γp . In other
words, θp : [0, T ) → U is one–to–one, and γp = θp ([0, T ]). The function θp is
said to be periodic with period T .
Definition 4.7.2 (Periodic Solutions). Let U ⊆ RN be open and F : U → RN
be a C 1 vector field. A solution u : R → U of the differential equation
dx
= F (x), (4.87)
dt
which is not an equilibrium solution, is said to be periodic if there exists a
positive number, τ , such that

u(t + τ ) = u(t), for all t ∈ R. (4.88)

The smallest positive number, τ , for which (4.88) holds true is called the period
of u.
Example 4.7.3. Suppose that the differential equation in (4.87) has a flow,
θ : R × U → U , and that the orbit, γp , for p ∈ U , is a cycle. Let T > 0 be such
that
θ(T, p) = p, (4.89)
θp ([0, T ]) = γp ,
and
θp : [0, T ) → U is one–to–one. (4.90)
We show that the function
θp : R → U
is periodic with period T . To see why this claim is true, observe that, for any
t ∈ R,
θ(t + T, p) = θ(t, θ(T, p)) = θ(t, p),
where we have used (4.89). We therefore have that

θp (t + T ) = θp (t), for all t ∈ R,

which shows that θp is periodic. To see that T is the period of θp , suppose that
0 < τ < T , and
θp (t + τ ) = θp (t), for all t ∈ R. (4.91)
Letting t = 0 in (4.91) we obtain that

θp (τ ) = θp (0),

which contradicts (4.90) since τ ∈ (0, T ). Hence, cycles in a dynamical system

correspond to period solutions.
68 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

4.8 Limit Cycles

Example 4.8.1. Consider the two–dimensional system

 dx = −y + µx(1 − x2 − y 2 );

 dt

(4.92)

 dy 2 2

 = x + µy(1 − x − y ),
dt
where µ is a real parameter.
We would like to understand the dynamics of the system in (4.92). To do so,
we first look for equilibrium points. We will then look at orbits and compute
limit sets of those orbits.
To find the equilibrium points of the system in (4.92), we solve the algebraic
system
−y + µx(1 − x2 − y 2 ) = 0;

(4.93)
x + µy(1 − x2 − y 2 ) = 0.
Note that (0, 0) is a solution of the system in (4.93), so that (0, 0) is an equilib-
rium point. We next see if there are other equilibrium points. So, suppose that
(x, y) solves the system in (4.93) and x2 + y 2 6= 0. We claim that xy 6= 0. For
suppose that x = 0 and x2 + y 2 6= 0. It then follows from the first equation in
(4.93) that y = 0, which is impossible. Similarly, if y = 0 and x2 + y 2 6= 0, we
obtain from the second equation in (4.93) that x = 0, which is impossible. We
therefore get that, if (x, y) solves the system in (4.93) and x2 + y 2 6= 0, then
x 6= 0 and y 6= 0. Thus, let (x, y) be a solution of (4.93) with x2 + y 2 6= 0.
Multiplying the first equation in (4.93) by x and the second equation by y then
leads to the equivalent system

−xy + µx2 (1 − x2 − y 2 ) = 0;

(4.94)
xy + µy 2 (1 − x2 − y 2 ) = 0.

Adding the equation in (4.94) then leads to

µ(x2 + y 2 )(1 − x2 − y 2 ) = 0,

which is equivalent to
µ(1 − x2 − y 2 ) = 0, (4.95)
since we are assuming that x2 + y 2 6= 0. It follows from (4.95) that either µ = 0
or x2 + y 2 = 1. If µ = 0, then we obtain from (4.93) that (x, y) = (0, 0), which
is impossible. Similarly, if x2 + y 2 = 1, we get from (4.95) that (x, y) = (0, 0), a
contradiction. We have therefore shown that (0, 0) is the only equilibrium point
of the system in (4.92).
We next show that, for µ < 0, if (p, q) ∈ B1 (0, 0), then ω(γ(p,q) ) = {(0, 0)}.
We will proceed as in Example 4.6.8 by considering the function V : R2 → R
given by
V (x, y) = x2 + y 2 , for all (x, y) ∈ R2 ,
4.8. LIMIT CYCLES 69

and first showing that if (p, q) ∈ B1 (0, 0), then V (u(p,q) (t)) decreases with in-
creasing t; in other words, V decreases along the orbit, γ(p,q) . Here, the function
u(p,q) : J(p,q) → R2 denotes the solution to the system in (4.95) subject to the
initial condition:
up (0) = (p, q), (4.96)
where J(p,q) is the maximal interval of existence. In order to show that V (u(p,q) (t))
decreases with increasing t, first compute
d 
V (u(p,q) (t)) = ∇V (u(p,q) (t)) · u0(p,q) (t)

dt

= 2x(−y + µx(1 − x2 − y 2 )) + 2y(x + µy(1 − x2 − y 2 ))

= 2µ(x2 + y 2 )(1 − (x2 + y 2 )).

We therefore have that
d  
V (u(p,q) (t)) = 2µV (u(p,q) (t))[1 − V (u(p,q) (t))]. (4.97)
dt
Setting v(t) = V (u(p,q) (t)), for t ∈ J(p,q) , we can re-write the differential equa-
tion in (4.97) as
dv
= 2µv(1 − v). (4.98)
dt
dv
Note from (4.98) that < 0, whenever 0 < v < 1, for the case µ < 0. Thus, if
dt
(p, q) ∈ B1 (0, 0), it follows from (4.96) that
ku(p,q) (t)k2 6 k(p, q)k2 < 1, for all t ∈ J(p,q) ∩ [0, ∞),
from which we get that
u(p,q) (t) ∈ B1 (0, 0), for all t ∈ J(p,q) ∩ [0, ∞), and (p, q) ∈ B1 (0, 0). (4.99)
It follows from the Proposition 2.3.9, on page 23 in these notes, that u(p,q) (t) is
defined for all t > 0. Furthermore, we obtain from (4.99) that
+
γ(p,q) ⊂ B1 (0, 0).

It then follows from Proposition 4.6.6 that, if (p, q) ∈ B1 (0, 0), then ω(γ(p,q) ) 6= ∅
and
ω(γ(p,q) ) ⊆ B1 (0, 0).
We next show that
ω(γ(p,q) ) = {(0, 0)}, for all (p, q) ∈ B1 (0, 0). (4.100)
In order to prove (4.100), we solve (4.98) by separation of variables and partial
fractions to obtain
vo e2µt
v(t) = , for all t ∈ R, (4.101)
1 − vo + vo e2µt
70 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

where vo = p2 + q 2 < 1. We then see from (4.101) that, if (p, q) ∈ B1 (0, 0) and
µ < 0, then
lim v(t) = 0,
t→∞
or
lim V (u(p,q) (t)) = 0,
t→∞

from which we get that

lim kθ(t, p, q)k = 0, whenever (p, q) ∈ B1 (0, 0), (4.102)

t→∞

where we have written θ(t, p, q) for u(p,q) (t), for all t > 0. Finally, to show (4.100)
holds true, assume that (p, q) ∈ B1 (0, 0) and let (x, y) ∈ ω(γ(p,q) ). Then, there
exists a sequence of positive numbers, (tm 0, such that tm → ∞ and

lim θ(tm , p, q) = (x, y). (4.103)

m→∞

It then follows from (4.102), (4.103), and the continuity of the flows and of the
norm that
k(x, y)k = 0,
from which get that (x, y) = (0, 0) and (4.100) follows.
Example 4.8.2 (Continuation of Example 4.8.1). In this example we show that
if µ < 0 in system (4.92), and (p, q) ∈ B1 (0, 0)\{(0, 0)}, then α(γ(p,q) ) = S 1 ,
where S 1 = {(x, y) ∈ R2 | x2 + y 2 = 1}.
It follows from (4.101) that, if 0 < vo = p1 + q 2 < 1, then

V (u(p,q) (t)) < 1, for all t ∈ J(p,q) .

Thus, u(p,q) (t) ∈ B1 (0, 0) for all t ∈ J(p,q) . The global existence results proved
in in these notes then imply that u(p,q) (t) is defined for all t ∈ R. Furthermore,
−
γ(p,q) ⊂ B1 (0, 0);

it then follows from Proposition 4.6.6 that α(γ(p,q) ) 6= ∅ and

α(γ(p,q) ) ⊂ B1 (0, 0).

Next, we show that

α(γ(p,q) ) ⊆ ∂B1 (0, 0) = S 1 , for 0 < p2 + q 2 < 1. (4.104)

Let (x, y) ∈ α(γ(p,q) ); then, there exists a sequence of negative numbers, (tm ),
such that
tm → −∞ as m → ∞, (4.105)
and
lim θ(tm , p, q) = (x, y). (4.106)
m→∞
4.8. LIMIT CYCLES 71

It follows from (4.101) and (4.105) that, if µ < 0 and 0 < vo < 1, then
vo
lim v(tm ) = lim = 1,
m→∞ m→∞ (1 − vo )e−2µtm + vo

from which we get that

lim kθ(tm , p, q)k = 1. (4.107)
m→∞

It follows from (4.106), (4.107), and the continuity of the flow and the norm
that
k(x, y)k = 1,
which proves (4.104). In order to prove the reverse inclusion,

S 1 ⊆ α(γ(p,q) ), for 0 < p2 + q 2 < 1, (4.108)

let (x, y) ∈ S 1 then x2 + y 2 = 1, and so there exists a unique ϕ ∈ [0, 2π) such
that
x = cos ϕ and y = sin ϕ. (4.109)
For (p, q) ∈ B1 (0, 0) with p2 + q 2 6= 0, denote θ(t, p, q) by (x(t), y(t)) and put
p
r(t) = [x(t)]2 + [y(t)]2 , for t ∈ R,

so that p p
r(t) = V (θ(t, p, q)) = v(t), for all t ∈ R.
It then follows from
1/2
vo
r(t) = p , for all t ∈ R. (4.110)
(1 − vo )e−2µt + vo

Next, put  
y(t)
ϕ(t) = arctan , for t ∈ R,
x(t)
so that
y(t)
tan(ϕ(t)) = , for t ∈ R. (4.111)
x(t)
Differentiation on both sides of the equation in (4.111) with respect to t then
yields
dϕ y dx 1 dy
sec2 ϕ =− 2 + ,
dt x dt x dt
or
y 2 dϕ
 
y dx 1 dy
1+ 2 =− 2 + ,
x dt x dt x dt
from which we get that
dϕ y dx x dy
=− 2 + 2 . (4.112)
dt x + y 2 dt x + y 2 dt
72 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

Since (x(t), y(t)) solves the system in (4.8.1), we can substitute the expressions
dx dy
for and in (4.8.1) into (4.112) to obtain
dt dt
dϕ −y 2 + µxy(1 − x2 − y 2 ) x2 + µxy(1 − x2 − y 2 )
=− + ,
dt x2 + y 2 x2 + y 2
which leads to
dϕ
= 1. (4.113)
dt
We can solve (4.113) to obtain

ϕ(t) = t + ϕo , for all t ∈ R,

and some ϕo ∈ [0, 2π). We then have that

θ(t, p, q) = r(t)(cos(t + ϕo ), sin(t + ϕo )), for all t ∈ R, (4.114)

where r(t) is given in (4.110). Define the sequence of real numbers

tm = ϕ − ϕo − 2mπ, for all m ∈ N. (4.115)

Then
tm → −∞ as m → ∞. (4.116)
Substituting t in (4.114) for tm given in (4.115) we obtain from (4.114) that

θ(tm , p, q) = r(tm )(cos ϕ, sin ϕ), for all m ∈ N, (4.117)

where we have used the 2π–periodicity of sin and cos. Thus, using the definition
of r(t) in (4.110) and (4.116), we obtain from (4.117) that

lim θ(tm , p, q) = (cos ϕ, sin ϕ) = (x, y),

m→∞

for 0 < p2 +q 2 < 1, where we have used (4.109); thus, (x, y) ∈ ω(γ(p,q) ), provided
that 0 < p2 + q 2 < 1. We have therefore established the inclusion in (4.108).
Combining (4.108) with (4.104) yields that, if µ < 0 and γ(p,q) is an orbit of the
system in (4.92) with 0 < p2 + q 2 < 1, then

α(γ(p,q) ) = S 1 .

Example 4.8.3 (Continuation of Example 4.8.2). Assume (p, q) ∈ S 1 = ∂B1 (0, 0)

and that u(p,q) : J(p,q) → R2 is the unique solution to the system in (4.92) subject
to the initial condition:

x(0) = p and y(0) = q, (4.118)

where
p2 + q 2 = 1.
4.8. LIMIT CYCLES 73

Then, there exists ϕo ∈ [0, 2π) such that

p = cos ϕo and q = sin ϕo . (4.119)

We show that

u(p,q) (t) = (cos(t + ϕo ), sin(t + ϕo )), for all t ∈ R. (4.120)

Put
v(t) = (cos(t + ϕo ), sin(t + ϕo )), for all t ∈ R. (4.121)
First note that, in view of (4.119), the function v = v(t) given in (4.122) satisfies
the initial condition in (4.118). Writing x(t) = cos(t + ϕo ) and y(t) = sin(t + ϕo )
for all t ∈ R, we compute that

x0 (t) = − sin(t + ϕo ) = −y(t), for all t ∈ R,

and
y 0 (t) = cos(t + ϕo ) = x(t), for all t ∈ R,
so that, since
[x(t)]2 + [y(t)]2 = 1, for all t ∈ R,
we see that x = x(t) and y = y(t) solve the system in (4.92). By uniqueness of
the solution to the IVP in (4.92) and (4.118), we have that

v(t) = u(p,q) (t), for all t ∈ R,

so that the assertion in (4.120) follows.

We then have that if (p, q) ∈ S 1 , α(γ(p,q) ) = ω(γ(p,q) ) = S 1 (see Example
4.6.7 on page 61 in these notes).

Example 4.8.4 (Continuation of Example 4.8.3). Assume that p2 + q 2 > 1.

Let u(p,q) : J(p,q) → R2 be the solution to the IVP in (4.92), for µ < 0, subject
to the initial condition
u(0) = (p, q). (4.122)
As in Example 4.8.1, let V (x, y) = x2 + y 2 for all (x, y) ∈ R2 , and put

v(t) = V (u(p,q) (t)), for all t ∈ J(p,q) .

Then, v satisfies the first order differential equation in (4.98), which can be
solved to yield
vo
v(t) = , for t < t1 , (4.123)
vo − (vo − 1)e−2µt
for the case v > 1, where
vo = p 2 + q 2 ,
and  
1 vo − 1
t1 = ln > 0, (4.124)
2µ vo
74 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

since µ < 0. Observe that for t < t1 ,

−2µt < −2µt1 , (4.125)

since µ < 0. Exponentiating on both sides of (4.125) yields

vo
e−2µt < , (4.126)
vo − 1

where we have used the definition of t1 in (4.124). Consequently, since vo > 1,

we obtain from (4.126) that

vo − (vo − 1)e−2µt > 0 for all t < t1 ,

where t1 is given in (4.124); hence, v(t) in (4.124) is indeed defined for all t < t1 .
Thus, J(p,q) = (−∞, t1 ), and so the ω–limit set of γ(p,q) is not defined for the
case µ < 0 and p2 + q 2 > 1. On the other hand, since by virtue of (4.123),
vo
lim v(t) = lim = 1, (4.127)
t→−∞ t→−∞ vo − (vo − 1)e−2µt

for µ < 0, we can prove that

α(γ(p,q) ) = S 1 , for p2 + q 2 > 1. (4.128)

In order to prove (4.128), first note that, by virtue of (4.98), if µ < 0 and
vo = p2 + q 2 > 1, then v(t) increases with increasing t; so that,

v(t) < vo , for all t < 0,

from which we get that

√
ku(p,q) (t)k < vo , for all t < 0.
−
Thus, γ(p,q) is bounded and therefore, by Proposition 4.6.6, α(γ(p,q) ) 6= ∅ for
p2 + q 2 > 1, provided that µ < 0.
Next, let (x, y) ∈ α(γ(p,q) ). Then there exists a sequence, (tm ), of negative
numbers such that
tm → −∞ as m → ∞, (4.129)
and
lim u(p,q) (tm ) = (x, y). (4.130)
m→∞

Thus, by the continuity of the function V we obtain from (4.130) that

lim V (u(p,q) (tm )) = V (x, y),

m→∞

or
lim v(tm ) = V (x, y). (4.131)
m→∞
4.8. LIMIT CYCLES 75

Thus, using (4.127) and (4.129), we obtain from (4.131) that

V (x, y) = 1,

which implies that (x, y) ∈ S 1 . We have therefore shown that

α(γ(p,q) ) ⊆ S 1 , for p2 + q 2 > 1, and µ < 0. (4.132)

Next, let (x, y) ∈ S 1 and ϕ ∈ [0, 2π) be such that

x = cos ϕ and y = sin ϕ. (4.133)

Similarly, let ϕo ∈ [0, 2π) be such that

p q
√ = cos ϕo and √ = sin ϕo . (4.134)
vo vo

We then have that

u(p,q) (t) = r(t)(cos(t + ϕo ), sin(t + ϕo )), for all t < t1 , (4.135)

where p
r(t) = v(t), for all t < t1 , (4.136)
where v(t) is given in (4.123).
Put
tm = −2mπ + ϕ − ϕo , for m = 1, 2, 3, . . . ; (4.137)
then,
tm → −∞ as m → ∞. (4.138)
Substituting tm in (4.137) for t in the expression defining u(p,q) (t) in (4.135)
yields

u(p,q) (tm ) = r(tm )(cos(ϕ), sin(ϕ)), for m = 1, 2, 3, . . . , (4.139)

where we have used the 2π–periodicity of sin and cos. Thus, using (4.138),
(4.127) and (4.133), we obtain from (4.139) that

lim u(p,q) (tm ) = (x, y),

m→∞

which shows that (x, y) ∈ α(γ(p,q) ); so that

S 1 ⊆ α(γ(p,q) ), for p2 + q 2 > 1, and µ < 0. (4.140)

Combining (4.132) with (4.140) yields (4.128).

In Examples 4.8.2 through 4.8.4 in this section, we have shown that S 1 is a

cycle of the system in (4.92) with the property that if either 0 < p2 + q 2 < 1 or
p2 + q 2 > 1, then α(γ(p,q) ) = S 1 , for µ < 0. We then say that, for µ < 0, the
orbit S 1 is a limit cycle of the system in (4.92).
76 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

Definition 4.8.5 (Limit Cycles). Let U be an open subset of RN and F : U →

RN be a C 1 vector field. A closed orbit, γ, of the system

dx
= F (x)
dt
is said to be a limit cycle of the system if there exists a neighborhood, Vγ ⊂ U ,
of the orbit γ such that for all p ∈ Vγ either ω(γp ) = γ or α(γp ) = γ.

In Examples 4.8.2 and 4.8.4 we saw that if V = RN \{0}, then α(γ(p,q) ) = S 1

if (p, q) ∈ V \S 1 . In Chapter 5 of these notes we will study criteria that can be
used to determine whether a given, general, two–dimensional system has limit
cycle.

4.9 Liapunov Stability

Let U denote an open subset of RN and F : U → RN be a C 1 vector field.
Suppose that x is an equilibrium point of the system

dx
= F (x). (4.141)
dt

Assume also that there exists r > 0 such that Br (x) ⊂ U , and Br (x)\{x}
contains no equilibrium points of F ; in other words, x is an isolated equilibrium
point of F in U . In this section we are interested in conditions that will guarantee
that if a solution of (4.141) begins near the equilibrium point, x, then it will
remain near x for all t > 0. This is the concept of stability which we make
precise in the following definition.

Definition 4.9.1 (Liapunov Stability). Let x be an isolated equilibrium point

of the system in (4.141) and let r > 0 be such that Br (x) ⊂ U , and Br (x)\{x}
contains no equilibrium points of F . We say that x is stable if, for every r ∈
(0, r), there exists δ > 0 such that, if kp−xk < δ, then the solution, up : Jp → U ,
to the IVP
dx


 = F (x);
dt (4.142)


x(0) = p,
exists for all t > 0, and there exists t1 > 0 such that

up (t) ∈ Br (x), for t > t1 .

Example 4.9.2. In part (c) of Example 4.6.8 on page 63, we saw that for
F = −∇V , where V : U → R is a C 2 function satisfying V (x) > 0 for all
x ∈ U \{0} and V (0) = 0, and if 0 is an isolated equilibrium point, then, for
every r > 0 with Br (0) ⊂ U , such that Br (0)\{0} contains no equilibrium
points of F , there exists δ > 0 such that, Bδ (0) ⊂ U and, for every p ∈ Bδ (0),
4.9. LIAPUNOV STABILITY 77

up (t) ∈ Br (0) for all t ∈ Jp ∩ [0, ∞), where up : Jp → U is the unique solution
to the IVP in (4.142). This implies that up (t) is defined for all t > 0 and

up (t) ∈ Br (0), for all t > 0.

Hence, x = 0 is stable according to Definition 4.9.1.

Definition 4.9.3 (Asymptotic Stability). An isolated equilibrium point, x, of
the system in (4.141) is said to be asymptotically stable if x is stable and there
exists δ > 0 such that, if kp − xk < δ, then ω(γp ) = {x}.
Example 4.9.4. Let F = −∇V , where V is as given in Example 4.6.8 on page
63. In parts (d) and (e) of that example we showed that there exists δ > 0 such
that, for any p ∈ Bδ (0), ω(γp ) = {0}. In other words, x = 0 is asymptotically
stable.
Definition 4.9.5 (Unstable Equilibrium Points). An isolated equilibrium point,
x, of the system in (4.141) which is not stable is said to be unstable.
Example 4.9.6. Consider the system

dx 3
 dt = y + µx ;



(4.143)
 dy
−x + µy 3 ,


 =
dt
where µ > 0. We show that (0, 0) is an unstable equilibrium point of the system
in (4.143).

Solution: First note that (0, 0) is the only equilibrium point of the
system in (4.143). Indeed, suppose that (x, y) is an equilibrium point
of the system in (4.143) with (x, y) 6= (0, 0). We then have that

 y + µx3 = 0;


(4.144)
−x + µy 3 = 0.


We see that x 6= 0 and y 6= 0. For, suppose that x = 0, the from the

second equation in (4.144), we get that y = 0 since µ > 0, which is
impossible since we are assuming that (x, y) 6= (0, 0). Similarly, y
cannot be 0. Next, multiply the first equation in (4.144) by x and
the second equation by y to get that

 xy + µx4 = 0;


(4.145)
−xy + µy 4 = 0.


Adding the two equations in (4.145) we get that

µ(x4 + y 4 ) = 0,
78 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

which yields that

x4 + y 4 = 0, (4.146)
since µ > 0. Note that (4.146) is impossible for (x, y) 6= (0, 0). We
have therefore shown that (0, 0) is an isolated equilibrium point of
the system in (4.143).

Next, we show that µ > 0 implies that (0, 0) is unstable.

We will prove that, for every r > 0 and every δ ∈ (0, r), there exists
(p, q) ∈ Bδ (0, 0), such that (p, q) 6= (0, 0) and ku(p,q) (t1 )k > r for
some t1 > 0. In fact, let (p, q) 6= (0, 0) be such that p2 + q 2 < δ and
let u(p,q) : J(p,q) → R2 denote the solution to the system in (4.143)
subject to the initial condition

(x(0), y(0)) = (p, q). (4.147)

Suppose by way of contradiction that

u(p,q) (t) ∈ Br (0, 0), for all t ∈ J(p,q) ∩ [0, +∞). (4.148)

It follows from (4.148) and Proposition 2.3.9 on page 23 in this notes

that u(p,q) is defined for all t > 0 and

u(p,q) (t) ∈ Br (0, 0), for all t > 0. (4.149)

Define V (x, y) = x2 + y 2 for all (x, y) ∈ R2 and put

u(p,q) (t) = (x(t), y(t)), for all t > 0.

Applying the Chain Rule we obtain that

d 
V (u(p,q) (t)) = 2µ([x(t)]4 + [y(t)]4 ),

for all t > 0. (4.150)
dt
Note also that, since (p, q) 6= (0, 0), it follows from uniqueness that

u(p,q) (t) 6= (0, 0), for all t > 0. (4.151)

In view of (4.151) and (4.150), we obtain from (4.150) that

d  
V (u(p,q) (t)) > 0, for all t > 0,
dt
since µ > 0. Consequently, V (u(p,q) (t)) is increasing in t. Thus,

V (u(p,q) (t)) > V (p, q), for all t > 0. (4.152)

Using the definition of V and (4.149), we obtain from (4.152) that

ro 6 ku(p,q) (t)k 6 r, for all t > 0, (4.153)

4.9. LIAPUNOV STABILITY 79

where ro = k(p, q)k. Since we are assuming that (p, q) 6= (0, 0), we
have that ro > 0. Put

ν= min 2µ(x4 + y 4 ). (4.154)

ro 6k(x,y)k6r

Then, ν > 0 since ro > 0 and x4 +y 4 = 0 if and only if (x, y) = (0, 0).
It follows from (4.153), (4.150) and (4.154) that

d  
V (u(p,q) (t)) > ν, for all t > 0. (4.155)
dt
Integrating the inequality in (4.155) from 0 to t then yields

V (u(p,q) (t)) > V (p, q) + vt, for all t > 0,

which implies that

ku(p,q) (t)k → ∞ as t → ∞,

which contradicts (4.149). This contradiction shows that (0, 0) is an

unstable equilibrium point for the system in (4.143) for µ > 0. 

The arguments used to prove stability in Example 4.6.8 on page 63 and in

Example 4.9.6 to prove instability are instances of a more general procedure
developed by Liapunov. This procedure is described in Chapter X of [Hal09]
and presented as Liapunov’s direct methods. We will describe the Liapunov
technique in the remainder of this section and apply to some two–dimensional
systems in the next chapter.
The main object in Liapunov’s approach to stability is a C 1 function, V : U →
R, which is positive definite in a neighborhood, Ω ⊂ Ω ⊂ U of the origin which
decreases along orbits as t increases. We will make these ideas precise in what
follows.

Definition 4.9.7 (Lie Derivative). Let F : U → RN be a C 1 vector field on a

open subset, U , of RN . Given a C 1 function, V : U → R, we define the derivative
of V along orbits of the system

dx
= F (x) (4.156)
dt

to be the map V̇ : U → R given by

V̇ (x) = ∇V · F (x), for all x ∈ U. (4.157)

Remark 4.9.8. Observe that if u : J → U is a solution to the differential

equation in (4.156) defined on a maximal interval of existence, J, then V̇ (u(t))
gives the rate of change of the function V along the orbit determined by u.
80 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

Definition 4.9.9 (Liapunov Function). Let U be an open subset of RN and Ω

be a open subset of U with Ω ⊂ U . A C 1 function, V : U → R, is said to be a
Liapunov function of the system in (4.156) in on the set Ω if and only if

V̇ (x) 6 0, for all x ∈ Ω. (4.158)

Example 4.9.10. If V : U → R is a C 2 function, and F = −∇V , then V is a

Liapunov function for the system in (4.156) since, in this case,

V̇ (x) = ∇V (x) · F (x) = −k∇f (x)k2 6 0, for all x ∈ U.

Definition 4.9.11 (Positive Definite Functions). Let Ω ⊂ Ω ⊂ U be a neigh-

borhood of 0. Let V : U → R be a continuous function satisfying V (0) = 0. We
say that V is positive semi–definite in Ω if V (x) > 0 for all x ∈ Ω. We say that
V is positive definite in Ω if V (x) > 0 for all x ∈ Ω\{0} and V (0) = 0.
Definition 4.9.12 (Negative Definite Functions). Let Ω ⊂ Ω ⊂ U be a neigh-
borhood of 0. Let V : U → R be a continuous function satisfying V (0) = 0. We
say that V is negative semi–definite in Ω if −V is positive semi–definite We say
that V is negative definite in Ω if −V positive definite.

We are now ready to state the first Liapunov stability theorem of this section.
Theorem 4.9.13 (Liapunov Stability Theorem). Let x = {0} be an isolated
equilibrium point of the system in (4.156). Suppose that the system in (4.156)
has a Liapunov function, V , in a neighborhood Ω ⊂ Ω ⊂ U of 0. Assume also
that V is positive definite in Ω. Then x = 0 is a stable equilibrium point of the
system in (4.156). In addition, if V̇ is negative definite in Ω, then ω(γp ) = {0}
for all p ∈ Ω.

The argument to prove Theorem 4.9.13 has been outlined in Example 4.6.8
for the case of negative gradient flows.
We also state here an instability criterion.
Theorem 4.9.14 (Instability Criterion). Let x = {0} be an isolated equilibrium
point of the system in (4.156). Let Ω be a neighborhood of 0, and suppose that
V : U → R is a C 1 function such that V̇ is positive definite in Ω. Suppose that
for every δ > 0, there exists p ∈ Bδ (0)\{0} such that V (p) > 0. Then, x = 0 is
unstable.

The proof of Theorem 4.9.14 is outlined in Example 4.143.

We end this section with the following extension of the Liapunov stability
theorem, which is presented as Theorem 1.3 in [Hal09, pg. 316].
Theorem 4.9.15 (The “S&M Theorem”). Suppose that the system in (4.156)
has a Liapunov function, V , in an open subset, Ω, of U with Ω ⊂ U . Let

S = {x ∈ Ω | V̇ (x) = 0}, (4.159)

4.9. LIAPUNOV STABILITY 81

and denote by M the largest invariant set of the system in (4.156) which is
contained in S. Let p ∈ Ω and suppose that γp+ is bounded and contained in Ω.
Then, ω(γp ) ⊆ M .

Proof: Since γp+ is bounded, it follows from Proposition 4.6.6 on page 58 in

these notes that ω(γp ) is nonempty, compact and connected. It is also the case
that the solution, up : Jp → U , to the equation in (4.156) subject to the initial
condition
x(0) = p (4.160)
exists for all t > 0 and

up (t) ∈ Ω, for all t > 0. (4.161)

Next, use the assumption that V is a Liapunov function in Ω to see that

d
[V (up (t))] = V̇ (up (t)) 6 0, for all x ∈ Ω, (4.162)
dt
where we have used (4.158) and (4.161). It follows from (4.162) that V (up (t))
is monotone non–increasing with t.
We now see that V (up (t)) is also bounded below for t > 0. Suppose, by way
of contradiction, that there exists a sequence of positive numbers, (tm ), such
that
tm → ∞ as m → ∞
and
lim V (up (tm )) = −∞. (4.163)
m→∞

The hypotheses that γp+ is bounded and γp+ ⊂ Ω imply that there exists a
subsequence, (tmk ), of (tm ) such that

lim up (tmk ) = y, (4.164)

k→∞

for some y ∈ Ω. It follows from (4.164) and the continuity of V that

lim V (up (tmk )) = V (y), (4.165)

k→∞

where V (y) ∈ R, since y ∈ Ω ⊂ U . On the other hand, by virtue of (4.163),

lim V (up (tmk )) = −∞,

k→∞

which contradicts (4.165). Hence, V (up (t)) must be bounded below.

We therefore have that V (up (t)) is monotone non–increasing as t increases,
and bounded from below for t > 0. Consequently, lim V (up (t)) exists. Set
t→∞

lim V (up (t)) = c. (4.166)

t→∞
82 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

We show that
V (y) = c, for all y ∈ ω(γp ). (4.167)
Let y ∈ ω(γp ); then there exists a sequence of positive numbers, (tm ), such that

tm → ∞ as m→∞ (4.168)

and
lim up (tm ) = y. (4.169)
m→∞

Then, using the continuity of V , we obtain from (4.169) that

lim V (up (tm )) = V (y). (4.170)

m→∞

Hence, in view of (4.168), we see that (4.167) follows from (4.166) and (4.170).
Since, ω(γp ) is invariant under the flow of F by virtue of Proposition 4.6.3 on
page 4.6.3 of these notes, we see from (4.167) that

V (uy (t)) = c, for all t ∈ Jy . (4.171)

Thus, differentiating on both sides of (4.171) with respect to t

V̇ (uy (t)) = 0, for all t ∈ Jy ,

so that, in particular,
V̇ (y) = 0,
which shows that y ∈ S, where S is as defined in (4.159). We have therefore
shown that
ω(γp ) ⊆ S;
consequently, since ω(γp ) is invariant, by Proposition 4.6.3, and M is the largest
invariant set contained in S, it follows that

ω(γp ) ⊆ M,

which was to be shown. 

Example 4.9.16 (Negative Gradient Flow Revisited). Let U be an open subset

of RN and let V : U → R be a C 2 function. Put F (x) = −∇V (x) for all x ∈ U .
Assume that V has a (strict) local minimum at x ∈ U ; that is, there exists r > 0
such that Br (x) ⊂ U and

V (x) < V (y), for all y ∈ Br (x)\{x}. (4.172)

Assume also that Br (x)\{x} contains no equilibrium points of F .

We show that there exists an open neighborhood, Ω, of x, such that Ω ⊂ U ,
with the property that
p ∈ Ω ⇒ ω(γp ) = {x}. (4.173)
4.9. LIAPUNOV STABILITY 83

In order to prove this claim, we let

ε= min (V (x) − V (x)). (4.174)

x∈∂BBr (x)

It follows from (4.172), the continuity of V , and the compactness of ∂BBr (x)
that ε > 0.
Put
Ω = {x ∈ U | V (x) < V (x) + ε}, (4.175)
where ε is as defined in (4.174). Then, Ω is an open neighborhood of x.
Observe that V is a Liapunov function in U . In fact, using the assumption
that F = −∇V , we get that

V̇ (x) = −k∇V (x)k2 , for all x ∈ U, (4.176)

where we have used (4.157) and the assumption that F = −∇F , so that

V̇ (x) 6 0, for all x ∈ U. (4.177)

Let p ∈ Ω, where Ω is as defined in (4.175). It then follows that

V (p) < V (x) + ε. (4.178)

Let up : Jp → U denote the unique solution to the IVP

dx


 = F (x);
dt (4.179)


x(0) = p,

where Jp is the maximal interval of existence.

Next, apply the Chain Rule and use the definition of V̇ in (4.157) to obtain
that
d
[V (up (t))] = V̇ (up (t)), for all t ∈ Jp ,
dt
so that
d
[V (up (t))] = −k∇V (up (t))k, for all t ∈ Jp , (4.180)
dt
by virtue of (4.176). It follows from (4.180) that
d
[V (up (t))] 6 0, for all t ∈ Jp , (4.181)
dt
from which we get that

V (up (t)) 6 V (p), for all t ∈ Jp ∩ [0, ∞). (4.182)

Combining (4.178) and (4.182) yields that

V (up (t)) < V (x) + ε, for all t ∈ Jp ∩ [0, ∞),

84 CHAPTER 4. CONTINUOUS DYNAMICAL SYSTEMS

which shows that

γp+ ⊂ Ω. (4.183)
Next, we show that

up (t) ∈ Br (x), for all t ∈ Jp ∩ [0, ∞). (4.184)

If not, by continuity and the intermediate value theorem, there exists t1 > 0
such that
kup (t1 )k = r.
It then follows from (4.174) that

V (up (t1 )) > V (x) + ε. (4.185)

On the other hand, it follows from (4.181) and (4.178) that

V (up (t1 )) 6 V (p) < V (x) + ε,

which is in direct contradiction with (4.185). We have therefore established

(4.184).
Finally, the assumption that Br (x)\{x} contains no equilibrium points of
F = −∇V implies that

S = {x ∈ Ω | V̇ (x) = 0} = {x};

consequently,
M = {x}. (4.186)
In view of (4.177), (4.183) and (4.184), we can now apply Theorem 4.9.15 to
obtain (4.173) from (4.186), which we wanted to show.
Chapter 5

Two–Dimensional Systems

The mail goal of this chapter is to give an analysis of the system

  3 
dx x
 dt = y − µ 3 − x ;



(5.1)
 dy


= −x,

dt
where µ is a real parameter. The system in (5.1) is known as the Lienard equa-
tions and comes about as a way to solve the second order, nonlinear differential
equation
ẍ + µ(x2 − 1)ẋ + x = 0, (5.2)
where
dx d2 x
ẋ = and ẍ = 2 .
dt dt
The equation in (5.2) is known as the Van der Pol equation in electric circuits
theory. Notice that if (x(t), y(t)) is a solution of the system in (5.1) then x =
x(t) is a solution to the second order differential equation in (5.2). In fact,
differentiating the first equation in (5.1) with respect to t we obtain,

ẍ = ẏ − µ(x2 − 1)ẋ. (5.3)

Substituting the second equation in (5.1) into (5.3) leads to (5.2).

In the next section we will consider the case µ < 0 and use the method of
Liapunov discussed in the previous section to show that (0, 0) is asymptotically
stable. More precisely, there exists a neighborhood, Ω, of the origin, (0, 0), such
that, if (p, q) ∈ Ω, the ω(γ(p,q) ) = {(0, 0)}. In a subsequent section, we consider
the question of what the α–limit set of γ(p,q) would be. This is equivalent to
looking u(p,q) (−t) = (x(−t), y(−t)) as t → ∞. Set z(t) = x(−t) for all t ∈ R
such that −t ∈ J(p,q) . Then,

ż(t) = −ẋ(−t) and z̈(t) = ẍ(−t),

85
86 CHAPTER 5. TWO–DIMENSIONAL SYSTEMS

for all t ∈ R such that −t ∈ J(p,q) . It then follows from (5.2) that z solves the
second order equation
z̈ − µ(x2 − 1)ż + z = 0. (5.4)
The Lienard system corresponding to (5.4) is then
  3 
dx x


 dt = y − ν − x ;
3

(5.5)
 dy


= −x,

dt
where ν = −µ > 0 for the case µ < 0. In Section 5.2 we will show that that the
system in (5.5) has a unique asymptotically stable limit cycle for ν > 0. Thus,
the Van del Pol equation in (5.2) has a non–trivial periodic solution.

5.1 Analysis of the Lienard System: Part I

Consider the Lienard system in (5.1),
  3 
dx x


 dt = y − µ − x ;
3

(5.6)
 dy


−x,

=
dt
where µ < 0. Observe that (0, 0) is the only equilibrium point of the system. We
will apply Theorem 4.9.15 on page 80 to prove that there exists a neighborhood,
Ω, of (0, 0) in R2 with the property that

(p, q) ∈ Ω ⇒ ω(γ(p,q) ) = {(0, 0)}. (5.7)

In order to prove (5.8), define V : R2 → R2 by

V (x, y) = x2 + y 2 , for all (x, y) ∈ R2 , (5.8)

Next, compute
  3 
x
V̇ (x, y) = 2x y − µ −x − 2xy
3
(5.9)
x2
 
2
= −2µx −1 ,
3
so that
V̇ (x, y) 6 0, provided that x2 6 3, (5.10)
since we are assuming that µ < 0. Define

Ω = {(x, y) ∈ R2 | V (x, y) < 3}. (5.11)

5.1. ANALYSIS OF THE LIENARD SYSTEM: PART I 87

It then follows from (5.9) that V is a Liapunov function in Ω.

Next, we show that

ω(γ(p,q) ) = {(0, 0)}, for all (p, q) ∈ Ω. (5.12)

In order to prove (5.12), we first verify the hypotheses of the S&M Theorem
+
(Theorem 4.9.15 on page 80) for the semi–orbit γ(p,q) and the set Ω defined in
(5.11). Since Ω is bounded, it suffices to show that
+
γ(p,q) ⊂ Ω, for all (p, q) ∈ Ω. (5.13)

Assume that (p, q) ∈ Ω. Then,

V (p, q) < 3. (5.14)

We show that
V (u(p,q) (t)) < 3, for all t > 0. (5.15)
If (5.15) not the case, there exists t1 > 0 such that

V (u(p,q) (t1 )) = 3, (5.16)

and
V (u(p,q) (t)) < 3, for 0 6 t < t1 . (5.17)
It follows from (5.17) and (5.10) that

V̇ (u(p,q) (t)) 6 0, for 0 < t < t1 ,

from which we get that

V (u(p,q) (t)) 6 V (p, q) < 3, for 0 6 t < t1 , (5.18)

where we have also used (5.14). I follows from (5.19) and continuity that

V (u(p,q) (t1 )) < 3, (5.19)

which is in direct contradiction with (5.16). This contradiction establishes

(5.15), which shows that (5.13) holds true. We can therefore apply the S&M
Theorem to conclude that

ω(γ(p,q) ) ⊆ M, for all (p, q) ∈ Ω, (5.20)

where M is the largest invariant subset of

S = {(x, y) ∈ Ω | V̇ (x, y) = 0}. (5.21)

It follows from (5.8), (5.9) and (5.21) that

√ √
S = {(0, y) ∈ Ω | − 3 < y < 3}. (5.22)
88 CHAPTER 5. TWO–DIMENSIONAL SYSTEMS

Next, let (0, yo ) ∈ S be such that yo > 0 and write u(0,yo ) (t) = (x(t), y(t))
for all t ∈ J(0,yo ) . It follows from the first equation in (5.6) that
x0 (o) = yo > 0.
Consequently, by the continuity of the derivative of u(0,yo ) , there exists δ > 0
such that [0, δ) ∈ J(0,yo ) and
yo
x0 (t) > , for all t ∈ [0, δ). (5.23)
2
It then follows from (5.23) that
Z t
yo
x(t) − x(0) = x0 (τ ) dτ > t, for all t ∈ (0, δ),
0 2
from which we get that
x(t) > 0, for all t ∈ (0, δ),
and so the orbit γ(0,yo ) , for yo > 0 must leave the set S given in (5.22). Similarly,
the orbit γ(0,yo ) , for yo < 0 must leave the set S. Hence, the largest invariant
set, M , contained in S must be {(0, 0)}; in other words, M = {(0, 0)}. Hence,
it follows from (5.20) that
ω(γ(p,q) ) = {(0, 0)}, for all (p, q) ∈ Ω,
which was to be shown.

5.2 Analysis of the Lienard System: Part II

In this section we show that the ω–limit set an orbit, γ(p,q) , of the Lienard
system in (5.5),   3 
dx x
 dt = y − ν 3 − x ;



(5.24)
 dy


 = −x,
dt
where ν > 0, is a limit cycle for any (p, q) 6= (0, 0) in a neighborhood of the
origin in R2 .
The analysis given in the previous section, shows that, for the system in
(5.24) with ν > 0, the equilibrium point (0, 0) is unstable; furthermore, α(γ(p,q) ) =
{(0, 0)} for every
(p, q) ∈ Ω = {(x, y) ∈ R2 | x2 + y 2 < 3}.
In other words, the origin is a source for the system in (5.24) with ν > 0. In
this section, we will try to understand the structure of the ω–limit set of the
orbit, γ(p,q) , for (p, q) ∈ Ω\{(0, 0)}. In order to do this, we present the analysis
given by Hirsch and Smale in [HS74, pp. 217–225] for a system similar to the
one in (5.24).
Appendix A

Definitions and Facts from

Real Analysis

In this Appendix, we list all the concepts and results from analysis that are used
in these notes. They may be found in undergraduate texts on real analysis (see,
for example, Bartle’s The Elements of Real Analysis, [Bar76]). Another good
reference is the first three chapters of Spivak’s Calculus on Manifolds, [Spi65].

A.1 Topology of Euclidean Space

The open ball of radius r about p in N –dimensional Euclidean space, RN , is
the set
Br (p) = {x ∈ RN | kx − pk < r},
where k · k denotes the Euclidean norm in RN ; in other words,
q
kxk = x21 + x22 + · · · + x2N , for all x ∈ RN .

The closure of Br (p), denoted by B r (p), is defined by

B r (p) = {x ∈ RN | kx − pk 6 r}.
Remark A.1.1. In the one–dimensional case, N = 1, the open ball, Br (p), for
p ∈ R, will simply be the open interval (p − r, p + r).
Definition A.1.2 (Open Sets). A subset, U , of RN is said to be open if, for
every p ∈ U , there exists r > 0 such that
Br (p) ⊂ U.
Definition A.1.3 (Closed Sets). A subset, K, of RN is said to be closed if its
complement,
K c = {x ∈ RN | x 6∈ K}
is an open subset of RN .

89
90 APPENDIX A. DEFINITIONS AND FACTS FROM REAL ANALYSIS

Proposition A.1.4 (Facts About Open Sets). Let U , V , Uα , for α in some

indexing set Λ, denote open subsets of Rn . Then,
1. the finite intersection, U ∩ V , is open;
[
2. the (possibly infinite) union Uα is open;
α∈Λ

3. U c is closed.
Proposition A.1.5 (Facts About Closed Sets). Let K, L, Kα , for α in some
indexing set Λ, denote closed subsets of Rn . Then,
1. the finite union, K ∪ L, is closed;
\
2. the (possibly infinite) intersection Kα is closed;
α∈Λ

3. K c is open.
Definition A.1.6 (Relative Topology). Let X be subset of RN . A subset, Y ,
of X, is set to be relatively open in X if there exists an open set U ⊆ RN such
that Y = U ∩ X. Similarly, Y is relatively closed in X if there exists a closed
set K ⊆ RN , such that Y = K ∩ X. Observe that the empty set, ∅, and X are
relatively open and closed in X.
Definition A.1.7 (Connectedness). A subset X of RN is said to be connected
if the only subsets of X which are both relatively open and relatively closed in
X are ∅ and X.
Example A.1.8 (Intervals in R). Intervals of real numbers are connected in R.
Definition A.1.9 (Open Cover). Let X be subset of RN . A collection of open
sets, {Uα | α ∈ Λ}, for some indexing set Λ, is said to be an open cover for X
if and only if [
X⊆ Uα
α∈Λ

Definition A.1.10 (Compactness). A subset, K, of RN is said to be compact

if and only if every open cover for K has a finite subcover; that is, for every
collection of open set, {Uα | α ∈ Λ}, such that
[
X⊆ Uα ,
α∈Λ

there exist finitely many indices, α1 , α2 , . . . , αn , such that

X ⊆ Uα1 ∪ Uα2 ∪ · · · ∪ Uαn .

Example A.1.11 (Examples of Compact Sets). Let a, b and R denote real

numbers with a < b and R > 0.
A.1. TOPOLOGY OF EUCLIDEAN SPACE 91

1. The closed and bounded interval, [a, b], is compact.

2. Closed and bounded subsets of RN are compact; where a set X ⊂ RN is
said to be bounded if there exists R > 0 such that X ⊂ BR (0).
3. Finite sets of points are compact.
Proposition A.1.12. Let U be an open subset of RN and K a compact subset,
K, of U . There exists and open subset, V , of U such that
K ⊂ V ⊂ V ⊂ U,
where V is compact.
Proof: Let U be an open subset of RN and and K ⊂ U be compact. Then, for
each x ∈ K there exists a positive real number rx such that
B2rx (x) ⊂ U.
Consequently,
Brx (x) ⊂ U, for all x ∈ K.
Since K is compact, the open cover, {Brx (x)}x∈K , for K has a finite subcover,
Br1 (x1 ), Br2 (x2 ), . . . , Brn (xn ).
We then have that
n
[
K⊂ Br1 (x1 ).
i=1
Put
n
[
V = Bri (xi ).
i=1
Then V is open with K ⊂ V . Furthermore,
n
[ n
[
V ⊂ Bri (xi ) ⊂ B2ri (xi ) ⊂ U.
i=1 i=1

Observe that the set V is bounded. In fact, from

n
[
V ⊂ B2ri (xi ),
i=1

we obtain that
V ⊂ BR (0),
where
R = max kxi k + 2 max ri .
16i6n 16i6n

It then follows that V is compact. 

Proposition A.1.13. Let K denote a compact subset of RN , and F a closed
subset of RN . If K ∩ F = ∅, then dist(K, F ) > 0; where
dist(K, F ) = inf kx − yk.
x∈K, y∈F
92 APPENDIX A. DEFINITIONS AND FACTS FROM REAL ANALYSIS

A.2 Sequences in Euclidean Space

Definition A.2.1 (Convergence). A sequence of points, (xm ), in RN is said to
converge to x ∈ RN if
lim kxm − xk = 0.
m→∞

Definition A.2.2 (Cauchy Sequence). A sequence of points, (xm ), in RN is

said to be a Cauchy sequence if, for every ε > 0, there exists M ∈ N such that

m, n > M ⇒ kxm − xn k < ε.

Proposition A.2.3 (Completeness). Evert Cauchy sequence in RN converges;

that is, if (xm ) is a Cauchy sequence in RN , then there exists x ∈ RN such that

lim kxm − xk = 0.
m→∞

Definition A.2.4 (Subsequences). Let (xm ) be a sequence of points in RN

and (mk ) an infinite sequence of natural numbers satisfying mk < mk+1 for all
k ∈ N. Then, the sequence (xmk ) is called a subsequence of (xm ).

Proposition A.2.5 (Bolzano–Weierstrass Property of Compact Sets). A sub-

set, K, of RN is compact if and only if every infinite subset of K has a subse-
quence which converges to some point in K.

A.3 Properties of Continuous Functions

Definition A.3.1 (Continuity at a Point). Let U be an open subset of RN . A
function
F : U → Rm
is said to be continuous at x ∈ U if and only if

lim kF (y) − F (x)k = 0.

ky−xk→0

Definition A.3.2 (Continuity on a Set). Let X be a subset of RN . A function

F : X → Rm

is said to be continuous on X if and only if the pre-image of every open subset,

V , of Rm ,
F −1 (V ) = {x ∈ X | F (x) ∈ V },
is relatively open in X.

Theorem A.3.3 (Intermediate Value Theorem). Let X be a subset of RN and

I and open interval of real numbers,
A.3. PROPERTIES OF CONTINUOUS FUNCTIONS 93

1. If F : X → Rm is continuous and X is connected, then image of X under

F,
F (X) = {y ∈ Rm | y = f (x), for some x ∈ X},
is a connected subset of Rm .
2. Suppose that f : I → R is continuous, and let a, b ∈ I be such that a < b.
If
f (a) < y < f (b)
or
f (a) > y > f (b),
for some y ∈ R, then there exists x ∈ (a, b) such that

f (x) = y.

Theorem A.3.4 (Extremal Value Theorem). Let K be a compact subset of

RN .
1. If F : K → Rm is continuous, then the image of K under F , F (X), is
compact.

2. Suppose that K 6= ∅. Let f : K → R be continuous. Then, there exist x1

and x2 in K such that
F (x1 ) = max f (x)
x∈K

and
F (x2 ) = min f (x)
x∈K
94 APPENDIX A. DEFINITIONS AND FACTS FROM REAL ANALYSIS
Bibliography

[Bar76] R. G. Bartle. The Elements of Real Analysis. John Wiley & Sons, 1976.
[Hal09] J. K. Hale. Ordinary Differential Equations. Dover, 2009.

[HS74] M. Hirsch and S. Smale. Differential Equations, Dynamical Systems

and Linear Algebra. Academic Press, 1974.
[Spi65] Michael Spivak. Calculus on Manifolds. Addison–Wesley Publishing
Company, 1965.

95