Stability Theory of Lyapunov

N.
Rouche
P. Habets
M. Laloy
Stability Theory by
Liapunov's Direct Method
Applied Mathematical Sciences 22
Springer-Verlag New York Heidelberg Berlin
N. Rouche
P. Habets
M. Laloy
U.C.L.
Institut de Mathematique Pure et Appliquee

Chemin du Cyclotron 2
B-1348
Louvain-la-Neuve
Belgium
AMS Subject Classifications:
34D20, 93D05 (Primary), 34Dxx,

34H05
Library of Congress Cataloging in Publication Data

Rouche, Nicolas.
Stability theory by Liapunov's direct method.
(Applied mathematical sciences ; v. 22)
Bibliography: p.
Includes indexes.
1. Differential equations. 2. Stability.
3. Liapunov functions. I. Habets, P., 1943joint
author. II. Lal oy, M. , 1946j oint author.
III. Title. IV. Series.
QAl.A647 vol. 22
[QA372]
5l0'.8s [515'.352] 77-7285
All rights reserved.

No part of this book may be translated or reproduced in any form
without written permission from Springer-Verlag.
~1977 by Springer-Verlag, New York Inc.
9 8 7 6 543 2 1
ISBN 978-0-387-90258-6
DOl 10.1007/978-1-4684-9362-7
ISBN 978-1-4684-9362-7 (eBook)
PREFACE
This monograph is a collective work.
The names appear-
ing on the front cover are those of the people who worked on
every chapter.
But the contributions of others were also
very important:
C. Risito for Chapters I, II and IV,
K. Peiffer for III, IV, VI, IX
R. J. Ballieu for I and IX,
Dang Chau Phien for VI and IX,
J. L. Corne for VII and VIII.
The idea of writing this book originated in a seminar

held at the University of Louvain during the academic year
1971-72.
Two years later, a first draft was completed.
However, it was unsatisfactory mainly because it was exce~sively
abstract and lacked examples.
It was then decided
to write it again, taking advantage of -some remarks of the

students to whom it had been partly addressed.
The actual
text is this second version.

The subject matter is stability theory in the general
setting of ordinary differential equations using what is
known as Liapunov's direct or second method.
We concentrate
our efforts on this method, not because we underrate those

which appear more powerful in some circumstances, but because
it is important enough, along with its modern developments,
to justify the writing of an up-to-date monograph.
Also
excellent books exist concerning the other methods, as for

example R. Bellman [1953] and W. A. Coppel [1965].
Liapunov's second method has the undeserved reputation
of being mainly of theoretical interest, because auxiliary
iv
PREFACE
functions appear to be so difficult to construct.
We feel
this is the opinion of those people who have not really tried.
Indeed, many mathematicians have tackled only theoretical
problems.
On the other hand, too many of those involved in
applications are unaware of the useful theorems or are

victims of the myth of the elusive Liapunov function.
aim, in writing this book, has been twofold:
Our
to describe the
present state of the most useful Parts of the theory, and to

appeal to the practical man with a wealth of applications
taken from many varied fields.
Chapters I and II constitute an elementary self-contained
treatment of stability theory.
first.
They should normally be read
Almost every other chapter can be studied without
further prerequisite, except that some definitions or

propositions of Chapter VI are needed in Chapters VII, VIII,
and IX.
The whole of Chapter VI is used in Section IX.6.
We are also grateful to H. Everard, S. Spinacci and

Kate MacDougall for their particularly expert typing of
successive versions of the manuscript.
Finally, it is a
pleasure to acknowledge the financial support of this work by

"Fonds National de 1a Recherche Scientifique".
Louvain-1a-Neuve, October, 1975
SOME NOTATIONS AND DEFINITIONS
This books requires a familiarity with some basic

concepts from the theory of ordinary differential equations.
As a general rule we have used symbols which are common place
in mathematics.
Let us however point out the following
notations:
!JR, the set of real numbers,

!JR, the extended real number system,
a ::: 0, a is a positive real number,
a
> 0,
a is a strictly positive real number,
[a,b], closed interval,

]a,b[, open interval,
(alb) or aTb, according to context, scalar product in
!JRn,
Ilxll, norm of point x in ~,

d (x, M) = inf I Ix-y II, distance from x E!Jf n to M C!Jf n ,
yEM
= {x
Be:
E!Jfn , II x II < d, open ball with center at the origin
and radius
B(a,e:)
e:
{x E ~,
>
0,
II x-a II
< d, open ball with center
a E!JR n
a.nd radius e: > 0,

B (M, e:) = {x E!JR n , d (x,M) < d, e: - neighborhood of the set
M C!JRn ,
to
E, unit
n,
B(M,e:) n
M
e:
dx
= dt
~!
e: - neighborhood of
with respect
n C!JR n ,
n
matrix,
' time derivative of the function
' jacobian matrix of the function
J+, see p. 7,
~,
M E!JR n
see definition p. 12.
f: !Jfn ~ !Jfm, x ~ f(x),
vi
'ttx, universal quantifier; read "for all
x" or "given
3x, existential quantifier; read "for some

"there exist
x",
x" or
x".
For g.eneral concepts on differential equations which

are not defined in this text we refer to Ph. Hartman [1964],
E. Coddington and N. Levinson [1965] or N. Rouche and
J. Mawhin [1973].
Let
C 9f
The following definitions might be useful.

and
f: A
is said
9f, x
f (x)
be a real
valued function.
The function
increasing i f
i.e.~
'ttx E A,
for all
be:
~o
'tty E A, x < y
and
strictly inc;J;'easing if
in
implies
A, x < y
'ttx E A, 'tty E A, x < y
f (x)
f (y) ;
implies f(x)
f(y).
implies
f (x) < f (y) ,
decreasing if
'ttx E A, 'tty EA, x < y
strictly decreasing if
f (x)
implies
'ttx E A, 'tty E A, x < y
f(x)
implies
> f (y) ,
monotonic if it is increasing on
A or decreasing on
A,
A.
Let
a E
the extended closure of
superior (upper limit) of

lim sup f(x)
x+a
at
lim inf f(x)

x+a
inf{sup{f(x): x EB(a,o), x
0>0
Then the limit
t- a}}E 9f.
f
at
is
sup{inf{f(x): x E B(a,o), x ;, an E !:if.

0>0
a EA, the function
continuous at a if
A.
is
Similarly the limit inferior (lower limit) of
If
fey),
lim inf f(x)

Y+rv
is said to be lower semi~
f(a).
If
vii
lim sup f(x) < f(a), the function f is said to be upper

x->-a
semi-continuous at a. It is easy to verify that a function
f
is continuous at
semi-continuous at
A function
if and only if it is lower and upper
a.
V: ~l+n ->- ~, (t,x) ->- V(t,x)
be positive definite (with respect to

function
a E
(i)
-V
V(t,x) > a(lIxll).
is positive definite, the function
negative definite (with respect to

V(t,x)
x) if there exists a
Je such that
V(t,O)
(ii)
If
is said to
the function
definite (with respect to

(t,x,y) ->- V(t,x,y)
respect to
V(t,O)
and
x).
A function
V: ~l+n+m ->- ~,
is said to be positive definite with

a E
Je
=0
V (t, x, y) > a
(ii)
If
is said to be
is said to be positive semi-
if for some function
V(t,O,O)
(1)
x).
(I Ix I I )
An important class of positive definite functions are the

positive quadratic forms
where
is a symmetric positive definite matrix (T
transpose)
denotes
TABLE OF CONTENTS
ELEMENTS OF STABILITY THEORY
1.
A First Glance at Stability Concepts
2.
Various Definitions of Stability

and Attracti vi ty . .
3.
Auxiliary Functions
11
4.
Stability and Partial Stability
13
5.
Instability
19
6.
Asymptotic Stability
25
7.
Converse Theorems
44
8.
Bibliographical Note
47
CHAPTER I.
CHAPTER II.
SIMPLE TOPICS IN STABILITY THEORY
49
1.
Theorems of E.A. Barbashin and N.N. Krasovski

for Autonomous and Periodic Systems
50
2.
A Theorem of V.M. Matrosov on

60
3.
Introduction to the Comparison Method
73
4.
Total Stability
80
5.
The Frequency Method for Stability

of Control Systems . .
84
6.
Non-Differentiable Liapunov Functions
89
7.
95
CHAPTER III.
STABILITY OF A MECHANICAL
EQUILIBRIUM
97
1.
Introduction
97
2.
The Lagrange-Dirichlet Theorem

and Its Variants
98
3.
Inversion of the Lagrange-Dirichlet Theorem

Using Auxiliary Functions
105
4.
Inversion of the Lagrange-Dirichlet Theorem

Using the First Approximation
108
5.
6.
7.
........................
Mechanical Equilibrium in the Presence
of Gyroscopic Forces .........................
Bibliographical Note .........................
of Dissipative Forces
CHAPTER IV.
113
115
126
STABILITY IN THE PRESENCE OF

FIRST INTEGRALS ......
128
.. ...
128
1.
Introduction
2.
General Hypotheses
129
3.
How to Construct Liapunov Functions
130
4.
Eliminating Part of the Variables
134
5.
Stability of Stationary Motions
139
6.
Stability of a Betatron
145
7.
Construction of Positive Definite Functions
151
8.
165
CHAPTER V.
INSTABILITY
168
1.
Introduction
168
2.
Definitions and General Hypotheses
170
3.
Fundamental Proposition
172
4.
Sectors
173
5.
Expellers
180
6.
Example of an Equation of Nth Order
184
7.
Instability of the Betatron
188
8.
Example of an Equation of Third Order
191
9.
Exercises
195
10.
.......... ........................ .
.;
Bibliographical Notes
CHAPTER VI.
A SURVEY OF QUALITATIVE CONCEPTS
198
201
1.
Introduction
201
2.
A View of Stability and Attractivity

Concepts . . . . . . . . .
204
xi
3.
Qualitative Concepts in General
207
4.
Equivalence Theorems for Qualitative

Concepts . . . .
215
5.
A Tentative Classification of
Concepts .
222
6.
Weak Attractivity, Boundedness,

Ultimate Boundedness
227
7.
235
8.
238
CHAPTER VII.
ATTRACTIVITY FOR AUTONOMOUS

EQUATIONS
241
1.
Introduction
241
2.
General Hypotheses
242
3.
The Invariance Principle
242
4.
An Attractivity and a Weak Attractivity

Theorem . . . .
246
5.
Attraction of a Particle by a
Fixed Center . . . . .
249
6.
A Class of Nonlinear Electrical

Networks .
254
7.
The Ecological Problem of Interacting

Populations .........
260
8.
269
CHAPTER VIII. ATTRACTIVITY FOR NON AUTONOMOUS

EQUATIONS . . .
270
1.
Introduction, General Hypotheses
270
2.
The Families of Auxiliary Functions
271
3.
Another Asymptotic Stability Theorem
280
4.
Extensions of the Invariance Principle

and Related Questions .
285
5.
The Invariance Principle for Asymptotically

Autonomous and Related Equations ..
294
6.
Dissipative Periodic Systems
305
7.
310
xii
CHAPTER IX.
THE COMPARISON METHOD
313
1.
Introduction
313
2.
Differential Inequalities
314
3.
A Vectorial Comparison Equation in

Stability Theory ...........................
320
4.
Stability of Composite Systems
327
5.
An Example from Economics
332
6.
A General Comparison Principle
336
7.
342
APPENDIX I.
DINI DERIVATIVES AND MONOTONIC

FUNCTIONS . .
345
1.
The Dini Derivatives
345
2.
Continuous Monotonic Functions
347
3.
The Derivative of a Monotonic

Function . . . .
350
4.
Dini Derivative of a Function along the

Solutions of a Differential Equation ..
352
APPENDIX II.
THE EQUATIONS OF MECHANICAL

SYSTEMS . . . . .
355
APPENDIX III. LIMIT SETS
363
LIST OF EXAMPLES
368
BIBLIOGRAPHY
370
AUTHOR INDEX
386
SUBJECT INDEX
393
CHAPTER I
The first two chapters are of an introductory character.

Of the matters they exhibit, some
~ve
been known for a long
time, others belong to the last fifteen years.
Almost all
will be considered over again in subsequent chapters, where

the results will be extended or deepened.
However, the next
few pages are meant to give a fair idea of what stability and
Liapunov's direct method are.
Further, they should prove
helpful to those concerned with simple practical applications.

Of course, the rest of the book has been written to cope with
less simple applications and, unfortunately or not, everyday
practice proves how numerous they are
1.
1.1.
A First Glance at Stability Concepts
The English adjective "stable" originates from the Latin
"stabilis", deriving itself from "stare", to stand.
Its first
acceptation is "standing firmly", "firmly established".
I.
natural extension is "durable", not to mention the moral

meaning "steady in purpose, constant".
As it is, this con-
cept of stability seems to be clear and of good use in everyday

life.
The layman might well wonder what reasons can be in-
voked to refine or complicate it.
There are many, as we
shall see.
Very early, the stability concept was specialized in
mechanics to describe some type of equilibrium of a material
particle or system.
Consider for instance a particle subject
to some forces and possessing an equilibrium point
qO'
The
equilibrium is called stable if, after any sufficiently small

perturbations of its position and velocity, the particle
remains forever arbitrarily near
velocity.
qo' with arbitrarily small
We shall not dwell on the well known example of a
simple pendulum, whose lowest position, associated with zero

velocity, is a stable equilibrium, whereas the highest one,
also with zero velocity, is an unstable one.
Formulated in precise mathematical terms, this mechanical
definition of stability was found useful in many situations,
but inadequate in many others.
This is why, with passing
years, a host of other concepts have been introduced, each of

them more or less related to the first definition and to the
common sense meaning of stability.
They were created either
for definite technical or physical purposes, or for reasons

of symmetry or completeness of the theory, or else to suit the
fancy of their inventors.
Later in this book (Chapter VI),
we shall try, with much care, to separate the wheat from the
chaff.
1.
Stability concepts
1.2.
As contrasted with mechanical stability, the other con-
cept known as Liapunov's stability has the following features:

first, it pertains no more to a material particle (or the
equations thereof), but to a general differential equation 1
second, it applies to a solution, i.e. not only to an
equilibrium or critical pOint.
Let
x = f(t,x),
where
and
variable), f
assume
(1.1)
are real n-vectors, t
is defined on !If x!lf n
is the time (a real
and
x = dx/dt.
We
smooth enough to ensure existence, uniqueness and
continuous dependence of the solutions of the initial value

problem associated with (1.1) over ~ x ~n.
For simplicity,
we assume further that all solutions to be mentioned below

exist for every
Let
E~.
II . II
Jesignate any norm
on
A solution
x(t)
of (1.1) is called stable at
more precisely, stable at
= t
in the sense of
A.M. Liapunov [1892) if, for every

such that if
x(t)
e > 0, there is a
8 > 0
is any other solution with
Ilx(t o ) - X(t O) I I < 8, then

t ~ to'
to' or,
Ilx(t) - x(t) I I < e
Otherwise, of course, x(t)
for all
is called unstable at
to'
Thus, it turns out that stability at
to
but continuous dependence of the solutions on

uniform with respect to
1.3.
Exercise.
is nothing
xo
= x(t O)'
t E [to'oo [.
Prove that stability at
to
implies stability
at any other initial time (usually with different values for
8).
I.
Hint:
use the fact that, if
passing through
xo
at
x(t;to'x O)
is the solution
to' then the mapping
is a homeomorphism; i.e., it and its inverse are one to one

and continuous.
1.4.
We may gain some geometrical insight into this stability
concept by considering again a pendulum, whose equation is
+ w2 sin x
= 0,
with
and
w E~.
This second order
equation is equivalent to the first order system

y
.
-w 2 s~n
x.
As is well known, the origin of the (x,y)-plane is a center,

i.e. all the solutions starting near the origin form a family
of non-intersecting closed orbits encircling the origin.
Given
disk
e > 0, consider an orbit entirely contained in the

Be
of radius
choose any other disk

orbit.
with center at the origin.

Bo
of radius
0, contained in this
Clearly, every solution starting in
initial time remains in
Be'
Further,
Bo
at any
This demonstrates stability of
the origin for any initial time.

On the other hand however, any other solution corresponding to one of the closed orbits is unstable.
In fact,
the period of the solution varies with the orbit and two
points of the (x,y)-plane, very close to each other at
t = to'
but belonging to different orbits, will appear in opposition

after some time.
This happens however small the difference
between periods.
But it remains that, in some sense, the
orbits are closed to each other.
Similar examples led to a
1.
Stability concepts
new concept called orbital stability, to be discussed later

in this book, in connection with the stability of sets of
points.
1.5.
To say a little more about possible variations on the
theme of stability, notice that in the case of the pendulum,

the equilibrium
=y =0
solution approaches it when
is such that no neighbouring

t
00,
appropriate friction were present.
as it would do if some
In many practical situa-
tions, it is useful to require, bes'ides mere Liapunov

stability of a solution
tions
x(t)
tend to
x(t), that all neighbouring solu-
x(t)
when
00.
This leads to the
notion of asymptotic stability.

1.6.
Many other examples can illustrate the necessity of
creating new specific concepts.
The last one to be mentioned
here will be borrowed from celestial mechanics.
Following
common sense, the solar system is called stable if it is

"durable" (cf. 1.1), Le. if none of its constituent bodies
escapes to infinity, and further if no two such bodies meet
each other.
But the velocities are unbounded if and only if
two bodies approach each other.
Therefore, stability in
this sense (it is called Lagrange stability), simply means

that the coordinates and velocities of the bodies are bounded.
Boundedness of solutions thus appears as a legitimate and
natural type of stability.
In the next section, we introduce a small number of
definitions, in fact the most widely used and studied.
I.
2.
Various Definitions of Stability and Attractivity
2.1.
We presented above the concept of a stable solution
x(t)
for equation (1.1).
variable
=x
z
where
by a new
- x(t), then (1.1) becomes
g(t,z)
g(t,O)
If we replace
=0
f(t,z + x(t - f(t,x(t
==
for every
E~.
(2.1)
The origin is a
critical point of (2.1) and stability of the solution
of
(1.1) is equivalent to stability of this critical point for

(2.1).
Naturally, passing from (1.1) to (2.1) is not always
possible, for
has to be explicitly known; nor is it
always rewarding, for it often happens that (2.1) is more

complicated than (1.1):
autonomous,
for instance, when (1.1) is
(2.1) generally is not.
Nevertheless, we shall,
in this chapter and the next, concentrate on stability of

critical points.
2.2.
Equation considered, general hypotheses.
Let us consider
a continuous function
f: I x
where
I = ] T , 00 [
~n,
for some
(t,x)
T
f(t,x)
or
domain (i.e. an open connected set) of

origin.
We assume that
f(t,O)
=0
= -00,
and
n is a
~n, containing the
for every
t E l , so
that for the differential equation
x=
f(t,x)
(2.2)
the origin is an equilibrium or critical point.

f
Further, let
be smooth enough in order that, through every
(to'x O) E I x
(2.2).
n,
there passes one and only one solution of
We represent this solution by
x(t;to'x O)' thus
2.
Stability and attractivity
displaying its dependence on initial conditions.

definition
= xO.
For the right maximal interval

where x(.:to'x O) is defined, we write J + (to'x O) or simply
or
Of course J + C [to'oo[, but we do not assume
w=
x(tO:to'x O)
By
Bp = {x E ~n:
Let us also recall that we write
00.
/lxll < pl.

In all definitions below, we use the logical quantifiers
3
and
V in a systematic manner.
This somewhat rigid way
of presenting things is meant to avoid looseness in expression

and ambiguities, a non-negligible danger in the manipulation
of such delicate concepts.
2.3.
Stability.
The solution
stable (A.M. Liapunov [1892]) if

(Vx o E Bo) ("It E
of (2.2) is called
("IE> 0) ("Ito E I) (30 > 0)
Ilx(t:to'x o ) II < e: Le., given
J+)
and
to E I, there is a
and
t E
J+
=0
0 > 0
such that for all
I Ix(t:to'x o ) I I
one has
<
unstable if it is not stable, i.e. if
E > 0
Xo E Bo
E.
(3E > 0) (3t O E I)
("10 > 0) (3X O E Bo) (3t E J+)

Ilx(t:tl)'x o ) II ~ e: Le., for
some E > 0 and to E I and each 0 > 0 there is an
xOE Bo
and a t E
such that
J+
Ilx(t:to'x o ) II ~ E.
uniformly stable (K.P. Persidski [1933]) i f
("IE> 0) (30 > OJ
("Ito E I) (V'x O EBo) Nt E
<
E > 0
there is a
for all
2.4.
J+
to E I, all
Remark.
[to'ro[.
Ilx(t:to'x o ) II.
J+)
= O(E)
Ilxoll
such that
< 0..
E: Le., given
I Ix(t:to'x o ) I I < E
and all
t ~ to.
As already noticed, we did not presuppose

It follows that, in principle, any solution
mentioned in the above definitions may cease to exist after

some finite time.
However, if
BE C Q, the solutions
I.
mentioned in the definitions of stability and uniform

stability may be continued up to
approach the boundary of
g.
in fact, they cannot
+~:
This is of course not the
case for the solutions mentioned in the definition of

instability.
Attractivity.
2.5.
attractiveif
The solution
x = 0
of (2.2) is called
(Vt o E I) (3n > 0) (Ve > 0) (Vx o E Bn)
(30' > 0, to + a E J+) (Vt ~ to + a, t E J+)

i.e., for each
e > 0
and each
such that
t
to E I
there is an
I IXol I < n
to + a EJ+
and
to + a E J+
there is a
and for each
= o(to,e,x O)
I Ix(t;to'x o ) I I < e
> 0
for all
e > 0
(vt o E I) (3n > 0) (Ve > 0) (30 > 0)

[(Vt ~ to + a., t E J+)
and
i.e., for each
t 'E I
0
for all
uniformly attractive if
(Vto E I) to + a E J+
and
Remark.
and all
and for each

and
and all
t > to + 0'.
[(Vt ~ to + a, t E J+) I Ix (t;to'xO) II < e];
to + a E J+ and
to E I
E Bn)
(3n > 0) (Ve > 0) (3a > 0) (Vx O E Bn)
Le., for some n > 0 and each

such that
to + a E J+
Ilxoll < n
~Vxo
Ilx(t;to'x o ) II < e];
n = n(t O)
there is an
such that
0' = o(to,e)
Ilx(t;to'x o ) II < e
2.6
= n(t O)'
to + 0'.
equi-attractive if
all
Ilx(t;to'x o ) II < e;
e > 0 there is a
Ilx(t;to'xo) II < e
= a (e)
for all
> 0
Ilxoll < n,
t > to + o.
As in Remark 2.4, if
Be C g, the solutions
mentioned in all three definitions above exist over
[to'~[.
Thus, in the definition of attractivity, all solutions starting

from
approach-the origin as
equi-attractivitYi they tend to
...
~.
In the case of
uniformly with respect
2.
Stability and attractivity
to
Xo E Bn'
whereas in the case of uniform attractivity,
they tend to
to
Xo E Bn
uniformly with respect to
and
E I.
2.7.
Attractivity does not imply stability!
pertaining to an autonomous system in
For an example
~2, see R.E. Vinograd
[1957], also presented in W. Hahn [1967].
The orbits are
not the same, but similar to those of Figure 1.1, where

is a curve separating bounded and unbounded orbits.
The
origin is unstable, in spite of the fact that every solution

tends to it as
+ ~.
Figure 1.1.
2.8.
x=
origin i,s unstable and attractive
Exercise (J.L. Massera [1949]).

f{t,x) (x,f E
attractivity.
2.9.
Th~
Exercise.
~),
Hint:
For a scalar equation
attractivity of the origin implies equiuse the uniqueness of the solutions.
For Equation (2.2), equi-attractivity of the
origin implies its stability.
10
I.
2.10.
ELID-1ENTS OF STABILITY THEORY
The region of attraction of the origin at time
to
is the set
If
A(t O)
does not depend on
attraction is uniform.
every
to' we say that the region of
Moreover, if
~n
= A(t O)
for
to' the origin is said to be globally attractive.
Some kind of uniformity in

global attractivity.
and
Xo E Bn' x(t;to'x O)
and
can also be added to
The origin is said to be uniformly
globally attractive if, for any

+
as
n > 0, any
+
to E I
and any
00, uniformly with respect to
This concept is obtained from uniform
attractivity (Section 2.5) by replacing

(Vn
(3n > 0)
by
> 0).
2.11.
ASymptotic stability.
The solution
o of (2.2)
is called
asymptotically stable (A.M. Liapunov [1892]) if it is stable
and attractive;
equi-asymptotically stable (J.L. Massera [1949]) if it is
stable and equi-attractive; (Cf. Exercise 2.9).
uniformly asymptotically stable (I.G. Malkin [1954]), if it
is uniformly stable and uniformly attractive;
globally asymptotically stable (E.A. Barbashin and
N.N. Krasovski [1952]) if it is stable and globally attractive;
uniformly globally asymptotically stable (E.A. Barbashin and
N.N. Krasovski [1952]), if it is uniformly stable and
uniformly globally attractive.
3.
Auxiliary functions
2.12.
Exercise.
11
For the equation
it = -x, with
x E 9/,
the origin is uniformly globally asymptotically stable.

2.13.
Exercise.
x E9/ and
Consider, for
t > 0, the
differential equation defined by
x - xt
for
~- 2
for
1 < xt,
~+ 2
for
xt < -l.
.~
t2
-1 < xt < 1,
Prove that the origin is equi-asymptotically stable, but not

uniformly nor globally asymptotically stable.
For more examples, see for instance J.L. Massera (1949]
or W. Hahn (1967].
The following theorem is important.
2.14.
Theorem.
of
or periodic in
Let
f(t,x)
t.
in Equation (2.2) be independent
Then, stability of the origin
implies uniform stability, and asymptotic stability implies

uniform asymptotic stability.
For the proof, see T. Yoshizawa (1966], pp. 30-31.
Notice, by the way, that there is no such theorem for
attractivity or uniform attractivity:
indeed, uniform
attractivity implies stability (see 2.9) and therefore

attractivity would imply stability, which is not the case,
even for autonomous systems (see 2.7).
3.
3.1.
Auxiliary Functions
What is known as Liapunov's direct or second* method
for the study of stability, makes an .es.s.ent.i.al use .of.
*The
first method rests on the consideration of some explicit

representation of the solutions, in particular by infinite
:>eries.
12
I.
auxiliary functions, also called

functions.
or Liapunovlike
Those of the simplest type are

V: I
where
~iapunov
derivative
functions
n .... 91, (t,x) .... V(t,x)
n are as in Section 2.2.
and
If
cl
x(t)
is some solution of Equation (2.2.>, the
v*(t)
of the time function
V*(t)
= V(t,x(t
exists and is given by
"'*(t) = (~~ (t,x(t
f(t,x(t) + ~~ (t,x(t.
(3.1)
Then, if one introduces the function

V: I x n .... !JI, (t,x) .... V(t,x)
defined by
(~~ (t,x) I f(t,x+ ~~ (t,x), it follows
V(t,x)
that
V*(t)
= V(t,x(t.
Hence, one finds out that computing

time t
often write
Vet)
for
at
function of
instead of
and
x.
t.
x(t),
For simplicity, we shall
v*{t), accepting the ambiguity of
using the same symbol for
V as a function of
and a
In the same way, we shall use
V(t)
V*{t).
Another useful type of function is the so-called
function of class
a function
with
at some given
does not require a knowledge of the solution
but only the value of
3.2.
v*(t)
alO}
a E 5t'.
in the sense of W. Hahn [1967], i.e.,
a: !JI+ .... !JI+, continuous, strictly increasing,
= O.
For such a function, we shall usually write
4.
Stability and partial stability
3.3.
Consider now a function
such that
every
V(t,O)
and,
13
V(t,x)
defined as above, but
(t,x) E I x n: V(t,x)
a(1 Ixl I).
>
said to be positive definite on
n.
definite on
3.4.
n'
and
Such a function is
Stating that
positive definite, i.e., without mentioning

some open neighborhood
a E
for some function
V(t,x)
is
n, means that for
of the origin, V
is positive
n'.
A well known necessary and sufficient condition for a
function
V(t,x)
to be positive definite on
there exists a continuous function
V*:
= 0, v*(x)
V(t,O) = and
V*(O)
>
for
(~(t,x)
x C
IT, x
IT
is that
such that
+ ~
and further
E I x n)V(t,x)
>
that
V*(x).
We assume that the reader knows how to recognize a

positive definite quadratic form in
Sylvester's criterion.
x, for instance by using
For all these notions, we refer to
N. Rouche and J. Mawhin [1973], vol. II.
4.
4.1.
Stability and Partial Stability
Theorems 4.2 and 4.3 to follow pertain to Equation (2.2)
and to the accompanying general hypotheses.

4.2.
Theorem (A.M. Liapunov [1892]).
function
every
V: I x n
(t,x) E I x
(i)
(ii)
fie
If there exists a
such that, for some
n:
V(t,x) > a(llxll); V(t,O)

V(t,x) < 0;
then, the origin is stable.
0;
a E
and
~l
14
I. ELEMENTS OF STABILITY THEORY
Let
Proof.
tinuous and
place of
Bo
X(titO'XO)
and every
for every
4.3.
xo
x(t)
Writing
in
~ V(to'x O) < a(E).
.Ye, one obtains that
n:
such
> 0
t E J+:
Ilx(t) II < E.
Theorem (K.P. Persidski [1933]).
E I
is con-
= O(tO,E)
E Bo.
Q.E.D.
I f one adds to the
hypotheses of Theorem 4.2 that for some

(t,x)
As
and using (ii), one gets for every
aellx(t) II) ~ Vet,x(t

Since
be given.
E > 0
0, there is a
V(to'O)
V(to'x o ) < a ed
that
Xo E
and
to E I
b E
5e and every
Vet,x) : bel Ixl I), then the origin is uniformly
stable.
Indeed, in this case, 0
4.4.
can be chosen to be independent
The concept of partial stability.
Let
n > 0
and
m > 0
be two integers, and consider two continuous functions
f: I x
n x
~m ~ ~n, g: I x
defined as above and

origin.
for every
~m ~ ~m, where
n is a domain of
We assume that
t E I
n x
f(t,O,O)
~n, containing the
=0
and
get,O,O)
and
and further that
is
enough in order that, through every point of
=0
are smooth
I x
~m,
there passes one and only one solution of the differential

system
x
f(t,x,y),
g(t,x,y).
To shorten our notation, we shall write

(x,y) E ~n+m
and also
for the vector
Z(tito'ZO) = (X(tito'ZO)' y(tito'ZO
for the solution of (4.1) starting from

following definition of
(4.1)
part~al
Zo
at
to.
stability was given by
The
4.
v.v.
Rumiantsev [1957]:
the solution
stable with respect to
i.e., given
that
> 0
and
Ilx(t;to'zO) II <
15
of (4.1)
is
if
to E I
there exists 0 > 0
for all
uniform stability with respect to
Zo EBo
x
and all
such
t E J+.
is defined in the same
way, following the example of Section 2.3 above.

Remark 2.4 is no longer true if partial stability
replaces stability, because the domain defined by
I Ixl I <
is unbounded in the (x,y)-space and a solution, even if it

remains in this domain, can escape to infinity in a finite
time:
consider for example, the trivial system
= y2
4.5.
.lf1
for
x, y E
=.
/V}
Theorem (V.V. Rumiantsev [1957]).

function
V: I x n x ~m
(H)
such that, for some
V (t, x, y) > a (I Ix I I ), V (t, 0, 0)
0;
V(t,x,y) < 0;
then the origin
=0
is stable with respect to
Moreover, if for some
... ~
If there exists a
(t,x,y) E I x n x ~m:
a E 5e and every
(i)
x = -x,
b E 5e and every
x.
(t,x,y) E I
x ~m:
(iii)
V(t,x,y) ~ b(llxll + Ilyll);
then the origin is uniformly stable in
x.
The proof is left as an exercise.

4.6.
A.
first example.
differential equation
Consider, for
x E ~n, the linear
16
x=
where
and
functions of
symmetric.
are
(D(t) + A(tx
n x n
on
],,00[, D
Choosing
< 0
Dare
V(t,x)
diagonal and
(x,x)
(where
V(t,x) = 2(x,D(t)x).
scalar product), we get

ments of
matrices, being continuous
for every
tE I, then
skew-
(x,x)
is a
If the eleV ~ 0
and,
following Theorem 4.3, the origin is uniformly stable.

4.7.
Stability of some steady rotations of a rigid body
(cf. N.G. Chetaev [1955]).

fixed point
Consider a rigid body with a
in some inertial frame of reference, and no
exterior forces applied whatsoever.
Let
A,B
and
be the
principal moments of inertia.of the body with respect to
0,
w its (vectorial) angular velocity in the inertial
and
frame.
The so-called Euler equations for the components
and
of
p,q
w in the principal axes of inertia .read (cf.,
e.g., H. Cabannes [1966])

Ap
(B-C)qr,
(C-A)rp,
ct
(4.2)
(A-B)pq.
The steady rotations around the first axis correspond to the

p
defined by
critical point
x,y,z
PO' q
=P
0, r
- PO' Y
O.
Using neW variables
q, z = r, we shift the
critical point to the origin:
If
B-C yz,
It
C-A
"""B (po+x) z,
-C
A-B
(4.3)
(PO+x)y.
A < B < C, the steady rotation is around the largest axis
4.
17
of the ellipsoid of inertia.
auxiliary function suiting
An
Theorem 4.2 is
which is a first integral, therefore such that
V=
O.
It
follows that the origin is stable for (4.3), and even uniformly stable since the system is autonomous.
On the other hand, if A > B ~ C, one obtains a
similar result using the first integral
Therefore, the steady rotations of the body around the

largest and the smallest axes of its ellipsoid of inertia are
stable with respect to
p,q,r.
The auxiliary functions
(4.4) and (4.5) are combinations of the integrals of energy

and of moment of momenta.
A general method for constructing
such combinations will be presented in Chapter IV.
4.8.
Stability of a glider.
Consider a glider which might
as well be a hovering bird, and suppose its plane of symmetry

coincides at any moment with a vertical plane in an inertial
frame of reference.
of inertia
Let
and
horizontal line.
be the velocity of its center
the angle between this velocity and a
The longitudinal "axis" of the glider is
assumed to make a constant angle

mass of the glider and
with
v.
m be the
the acceleration of gravity.
notations which are usual in aerodynamics, let

cL(a)
Let
CDCa)
Using
and
be the coefficients of drag and lift respectively.
The eguations for

mt
and
-mg sin
mvG = -mg cos
read:
- cD(a)v ,
e+
CL(a)v
18
I.
and
Putting
we transform the equations into
-'x
~
dT
= -sin e -
~~
(-cos
ay2,
(4.6)
+ y2)/y.
Figure 1.2.
The glider
We have introduced a non-vanishing drag only for future

reference (see Section 6.18).
For more complete equations,
including an equation of momentum, cf. N.G. Chetaev [1955].

If
points
Yo
= 0,
= 1,
the above equations have the critical

60
= 2kn
for
an integer, all of them
corresponding to one and the same horizontal flight with

constant velocity.
6 0 = O.
Let us therefore concentrate on
YO
1,
It is easily verified that the function

v (y, 6)
= !- - y
3
cos 6 + 3
is a first integral for the equations (4.6) with

Further, in some neighborhood of
(1,0), V(y,6)
a
>
O.
except
5.
19
Instability
that
V(l,O)
= O.
Therefore, if the critical point
(1,0)
is shifted to the origin by a suitable translation of the

axes, V expressed in the new variables satisfies the
hypotheses of Theorem 4.3 (remember Section 3.4) and uniform
stability is thus proved.
For more details on this problem,
see N.E. Zhukovski [1891], F.W. Lanchester [1908], B. Etkin

[1959], and Section 6.18 below.
Instability
5.
We go back to the general setting of Section 2.2 and

prove now some sufficient conditions for instability.
Theorem (N.G. Chetaev [1934])
5.l.
to E I,
~l
> 0
function
(i)
(il)
(with
BE en), an open set
V: [to'oo[ ?<
o < V(t,x)
I f there exist
'I' EB
+qf such that, on
k < 00, for some
and a
[to'oo [
x '1':
k~
V(t,x) :: a (V(t,x, for some
a E
5t'~
i f further
(iii)
the origin of the x-space belongs to
a'l'~
then, the origin is unstable.

Proof.
an
now
In fact, due to (iii), there exists for every
Xo E'I' n Be
x(t~to'xO)'
such that
V(to'x o ) > 0
abbreviated as
x(t).
(see (1.
As long as
one gets, owing to (i) and (ii), that for every

k >
V(t,x(t
V{to'x O) + It V(T,x(TdT
to
> V(to'X O) + a{V{to'X O (t-t o )
> 0
Consider
x{t) E '1',
t E J+:
20
I.
Therefore, x(t)
must leave
after some time.
cause of (iv), it cannot leave
through
a~
But be-
E BE.
Therefore,
Q.E.D.
Geometrically speaking, we have proved there exist a

sequence
{xOi: xOi E ~, xOi .... O}
Yi
connecting an initial point
in
aBg'Yi
being included in
and a family of paths*

xOi
to some terminal point
except for its terminal
point.
5.2.
The two following corollaries of 5.1 were in fact
established long before 5.1 was known.

5.3.
Corollary (Liapunov's first theorem on instability
(1892]).
If there exist
open set
C Bg
V: [to'oo[ x Bg
and
.... !}t
5&'1
to E I, g > 0
(with
such that, on
[to'oo[ x
an
~:
b EX;
a E .5e;
V(t,x) ~a(llxll), for some
(H)
en),
function
o < V(t,x) ~b(lIxll), for some
(i)
Bg
i f further:
(iii)
the origin of the x-space belongs to
a~;
then, the origin is unstable.

5.4.
Corollary (Liapunov's second theorem on instability
[1892]).
(ii)
*If
!}tn,
If, in Corollary 5.3, (ii) is replaced by

V(t,x) = cV(t,x) + W(t,x)
on
[to'oo[ x ~,
is a continuous function on an interval

by definition f([a,b]) is a path.
[a,b]
into
5.
where
~
21
Instability
c > 0
and
W: [to'ro[ x
~ ~ ~
is continuous and
0, then the origin is unstable.
5.5.
If the differential equation is autonomous and if
depends on
only, then (i) and (ii) in Chetaev's Theorem
can be simplified to:

(1)
V(x) > 0
on
'1
(ii)
~(x) > 0
on
~.
5.6.
Instability of some steady rotations of a rigid body
(cf. N.G. Chetaev [1955]).
Let us consider the steady
rotations of a rigid body around the intermediate axis of

its ellipsoid of inertia (cf. 4.7).
(4.3).
If
The equations are still
V(x,y,z) = yz, one computes
A-B y 2}
(PO + x) (C-A
-S- z 2 + -C-
The orientation of a principal axis of inertia is at our

disposal.
Therefore, we may choose
two at least of the quantities

decide that either
{ ex, y, z) : x
with
e;
C > A >
2
+ y
+ z
A,B,C
<
o.
Further, i f
are unequal, we may
C > A > B.
or
Po >
e;
small enough in order that
Then we choose
y > 0, z > o}
Po + x
>
on
~.
All
the hypotheses of Theorem 5.1 are verified and the origin is

unstable:
the steady rotations around the intermediate axis
(or as the case may be, the equatorial axis) are unstable with
respect to
5.7.
(p,q,r).
Instability proved by using the first approximation.
The main use of Corollary 5.4 is to help proving instability

by consideration of the linear approximation.
This is a useful
22
I.
Suppose Equation (2.2)
way of looking at many applications.

is particularized as
x = Ax
wher~
is an
n x n
+ g(t,x)
(5.1)
Ax + g(t,x)
real matrix and
all the properties required from
has
in Section 2.2.
f(t,x)
Then, the following theorem holds true.

5.8.
Theore~
value of
(A.M. Liapunov [1892]).
has strictly positive real part and if

Ilg{t,x)11 + 0
Ilxll
uniformly for
Proof.
If at least one eigen-
as
(5.2)
t E l , then the origin is unstable for (5.1).
The following lemma will be used and is stated here
without proof (see, e.g. N. Rouche and J. Mawhin [1973]).

Lemma.
A
If at least one eigenvalue of the
real matrix
has strictly positive real part, then to every positive
definite quadratic form

a quadratic form
Vex)
U(x)
Vex)
>
for
and a constant
(~~I
and
n x n
for some
Ax)
E ~n, there corresponds

c > 0, such that
cV + U
(5.3)
x.
Now using (5.3), one computes
aVI g(t,x.
cV + U + (ax
(5.2) shows that

W
=U
+ e~~1 get,x
is positive definite in some neighborhood of the origin.
5.
23
Instability
Therefore, by Corollary 5.4, the origin is unstable for

(5.1)
5.9.
Q.E.D.
More about the' steady' rotations of a rigid body.
we restrict ourselves to the case where
If
C > A > B,
Theorem 5.8 yields another proof of the result established

in Section 5.6.
The linear part of Equations (4.3) admits
of a strictly positive real root equal,_to
r. 2
~o
(C-A) (A-B>] 1/2

BC
and.the origin is un$table.

contr~sted
with Theorem 5.1, yields no instability result
when either
all three
5.10.
Notice that Theorem 5.8, as
C > A
r~levant
=B
or
=A
> B, for in such cases,
eigenvalues vanish.
Watt's governor.
As Watt's governor is a well-known
device, it will suffice to present it as in Figure 1.3 to

define our principal symbols.
Figure 1.3.
Watt's governor
x. ELEMENTS OF STABILITY THEORY
24
The angles
freedom.
and
correspond to the two degrees of
If friction is disregarded, the equations (see
e.g. T. Levi-Civita and U. Amaldi [1922-27], Vol. 2, Part 1)

read
~t c/S)
-k (~-~o) ,
1 2
6
~t c/,~)
h
were
and
=C
e,l,m,g
2 . 2
+ 2 m1 s~n~,
and
"2
/' =
'0/
au
ar
2 m12, U
2 mgl cos
are positive constants.
The steady motion
~o'
22
0, 6
60
gil cos ~O
is unstable, as can be established using Theorem 5.8.
In
fact, the eigenvalue equation for the linear part of the

equations is
(2
~2 + S.in2~0]A3 + e~ Sin2~O[l + 3 cos2~0 + 2 :21A

k
+ - - 2 eO
2ml
80
When
< 0, then
~(O)
SID
2 ~O
(5.4)
= O.
~(A) +
< 0, whereas
there is a strictly positive eigenvalue.
00
But
We conclude therefore that the steady
motion being considered is unstable.
For more details on
Watt's governor and the use of friction to stabilize its

steady motions, we refer to L.S. Pontryagin [1961].
A wealth of further illustrations of Chetaev's
Theorem S.l will appear in Chapter III.
00:
Al + A2 +
and thus at least one eigenvalue has a strictly
positive real part.
5.11.
Analogously, when
eO > 0, there is a strictly negative eigenvalue.
A3
when
6.
Asymptotic stability
25
6.
In this section, we still consider equation (2.2) and
6.1.
the corresponding hypotheses.
Uniform asymptotic stability
has been studied long before (simple) asymptotic stability.

The first theorem which was proved corresponds to thesis (b)
of the following statement and is due to A.M. Liapunov [1892].
Thesis (a) gives an interesting estimate of the region of
A(t O)'
attraction
Theorem.
6.2.
~l
Suppose there exists a
function
V: I x n ... .%' such that, for some functions

(t,x) E I x n:
and every
til
a (11xll l
V(t,x) < bcllxll>;
V(t,x) < -c Cli x II)
(iil
Choosing
a, b, c E.5e
~ > 0
such that
B~
n, let us put for every
t E I
v~;~
{x En: v(t,x)
<
a(a)}.
Then
for any
(al
uniformly in
(b)
Proof.
and any
to E I
to'xo
-1
Xo E Vto,a
when t -+ 00;
x(t;to'x o ) .... 0
the origin is uniformly asymptotically stable.
(a)
We choose an
> 0
deduce from (i) that for every
such that
Ba C nand
t E I
(6.1)
For any
to E I
and
Xo E v~~,a' it follows from (ii) that
x (t) E V- l
t,a
I.
26
for any
t E
J+,
and therefore from
approach the boundary of
Now
rl.
and also choose a
n > 0
cannot
[to'oo[.
such that
larger than
cannot be larger than
Ilx(t) II
x(t)
J+ (to'x O)
Hence
cr
that
(6.1)
e > 0, let us choose
For any
b (n) < a (e)
b(a)!c(n).
for every
t E [to,to+crJ, for if this were the case, one should obtqin

for
= to
V(t,x(t
+ cr:
::. V(to'xO) -
which contradicts (i).

tl E [to,tO+cr]
since
c(IIx(s) II)ds .: b(a) - c(q)cr < 0,

to
Therefore, there exists a
such that
b(llx(t l ) II) ~ ben) < a(e), and
is decreasing, one obtains for
Therefore, for
to + cr: I Ix(t)1 I < e
to + cr
and part (a) of
the thesis is proved.

(b)
Uniform stability of the origin follows from
Theorem 4.3.
Further, for any
0 > 0
such that
This shows that the attractivity is uniform.
b(o)
<
a(a):
Q.E.D.
Several remarks on this theorem can prove helpful.

6.3.
It might be important in practical cases to obtain an
upper estimate of the time needed by the solutions to reach

a given
Xo E
e.
It follows from the above proof that for
-1
Vto,a' one can choose any number larger than
cr = b(a)!c(n), n
being known as soon as
is.
6.
27
Asymptotic s tabili ty
6.4.
Assumptions (i) and (ii) are equivalent to (i) and
(ii)'
6.5.
V(t,x) < -c' (V(t,x
The existence of a
for some
(i)
(ii)
and
c E
for some
Je
function
and every
V(t,x) > a(lIxll); V(t,O)
c' E Ye.
VCt,x)
Ct,x) E I x
such that,
n:
0;
V(t,x)::. -cCllxll);
does not imply uniform asymptotic stability, nor even

asymptotic stability!
This is shown by the following counter-
example, borrowed from J.L. Massera [1949].

Let
e- t
g: [0,00 [ .... ~ be a
5fl
function coinciding with
except at some peaks where it reaches the value
Figure 1.4. is a diagram of

each integer value of
to abcissa
t.
g (t).
There is one peak for
The width of the peak corresponding
is supposed to be smaller than
The function
g2(t)
Figure 1.4.
1.
(1/2)n.
28
I. ELm1ENTS OF STABILITY THEORY
Consider now the following differential equation

g(t)
g (t) x.
The form of its general solution, namely
x (t)
shows obviously that the origin is not asymptotically stable
for this equation.
However, if we choose the auxiliary
function
2
-x- [3 g2 (t)
V(t,x)
we see first that
ftog 2 (T)dT],
V(t,x) : x , since
2,
and secondly that
6.6.
V(t,x)
2
-x , as is easily computed.
Thus, we have established that hypotheses (i) and (ii)
of Section 6.5 do not imply asymptotic stability.
Are there
nevertheless any conclusions which can be drawn out of them?

To answer this question, let us introduce a new definition:
the solution
if
=0
of (2.2) is called weakly attractive
(Vt o E I) ( 3 n > 0) ( vxO E Bn) ( 3 {til C J+, ti
i .... 00)
6.7.
x(t i ) .... 0
Exercise.
as
....
as
i .... 00.
The hypotheses of Section 6.5, imply that
the origin is weakly attractive.

6.8.
Exercise.
By the way, the following property will be
useful in the sequel:
weak attractivity along with uniform
stability imply attractivity.
6.
29
6.9.
Corollary to Theorem 6.2'.
N.N. Krasovski [1952]).
(E.A. Barbashin and
The origin is uniformly globally
asymptotically stable if the assumptions of Theorem 6.2 are

satisfied for
n=
~n
and
a(r)
6.10.
Exercise.
as
(6.2)
~.
Corollary 6.9 remains true i f (6.2) is
replaced by
= lim
lim a(r)
b(r).
r~~
6.11.
An
RLC circuit with parametric excitation.
To
illustrate Theorem 6.2, let us consider the scalar equation
x+ ax
where
a > 0, b
= b o (l
+ b(t)x = 0
+ ef(t
a bounded function from
~....
with
~.
(6.3)
bO
0, and
f(t)
is
This equation can be
interpreted as representing an RLC circuit with time-varying

capacitance, or a mechanical oscillator with viscous friction
and a time-varying spring parameter.
Equation (6.3) is
equivalent to
y,
-ay - b(t)x.
The auxiliary function

V(x,y)
is positive definite.
According to Theorem 6.2, the origin
will be uniformly asymptotically stable if the time derivative

V(t,x,y)
- 2 Y - (b-bO)XY -
ab
30
I. ELEMENTS OF STABILITY
is negative definite.
THEO~Y
This happens, following Sylvester's
criterion, if, for some
a
(6.4)
This condition is satisfied for any small enough E, a result

which can be interpreted as follows in the language of
electrical engineers.
work:
There are two opposing forces at
a parametric excitation proportional to
the damping force
ax.
and a load,
Satisfying (6.4) amounts to choosing
the resistance, designated by
a, large enough for the
load to absorb all the energy provided by the excitation.

In this Case, the origin is asymptotically stable.
In the
opposite case, i.e. if the load is not large enough, one may
expect the energy balance of the system to increase and the
origin to become unstable.
Of course, this is but a heuristic
view of the problem.

6.12.
A damped pendulum.
Consider the damped pendulum
described by the scalar equation
x +
+ sin x
0,
for which, as everyone knows, the origin is asymptotically

stable.
To check this by Liapunov's method, one might think
it natural, at first sight, to choose the total energy

VI =
as an auxiliary function.
for the time derivative
+ (1 - cos x)
This is not a good choice however,
2
Vl = -x
is not negative definite:
it proves stability, not asymptotic stability.

"natural" choice may not always fit Theorem 6.2.
Therefore, a
Finding a
6.
31
suitable auxiliary function is often a matter of habit and

feeling.
For instance
v2 = x2
+ (x+x)2 + 4(1 - cos x)
is such that
and can be used to prove uniform asymptotic stability of the

origin.
It could hardly be said however, that
V2
has any
physical interpretation, not that it is "natural" in any

sense!
As it is often very difficult to exhibit an auxiliary
function whose time derivative is negative definite, an
alternative way of proving asymptotic stability will be to
work out some more elaborate theorems, allowing one to use
functions like
whose derivative is only
VI
course along with some more information.
<
0, but of
Much effort will be
made in this direction in subsequent sections.

6.13.
Asymptotic stability proved by using the first
approximation.
Theorem 6.2 can be used to prove asymptotic
stability by consideration of the linear approximation.

Suppose Equation (2.2) is particularized as
where
is an
real matrix and
the properties required from
f(t,x}
x = Ax
+ g(t,x),
Ax + g(t,x)
in Section 2.2.
has all
Then
the following theorem holds true.

6.14.
of
have strictly negative real parts and if
I Ir(t,x) I I ~ 0
Ixll
as
If all eigenvalues
x ~ 0,
32
I.
t E I, then the origin is uniformly
uniformly for
asymptotically stable.
The proof consists in exhibiting a quadratic form in
x, which, considered as an auxiliary function, satisfies
all the hypotheses of Theorem 6.2.
This quadratic form is
built up using the following lemma.

6.15.
the
n x n
(A.M. Liapunov [1892]).
real matrix
If all eigenvalues of
A have strictly negative real parts,
then to every negative definite quadratic form
U(x)
for
x E ~n, there corresponds one and only one quadratic form

vex)
which is positive definite and such that
(~~ ,Ax)
= u.
(See e.g. N. Rouche and J. Mawhin [1973]).

Theorem 6.14 has no immediate extension to the case
where the matrix
is a continuous function of
t.
There
exist counter-examples proving that the origin can be unstable

for a linear equation
x=
A(t)x
even though, for any
the real part of every eigenvalue of

negative.
A(t)
t,
is strictly
Cf. Exercise V.9.l.
6.16.
Exercise.
Prove Theorem 6.14.
6.17.
Exercise.
Let
tinuous function of
A(t)
t E I
be an
n x n
real matrix, con-
and periodic of period
T > O.
Suppose the second member of the differential equation
x= A(t)x + g(t,x)
from
f(t,x)
ponents of
has, as above, all the properties required
in Section 2.2.
x= A(t)x
If all the characteristic ex-
have strictly negative real parts, and if
6.
33
\\g(t,x)\\ .... 0
Ilxll
uniformly for
as
x .... 0
t E I, then the origin is uniformly
The latter condition, along with the
existence of a characteristic exponent with strictly positive

real part, implies instability of the origin.
Hint:
use
the existence of a differentiable, regular, periodic matrix

set)
such that
S(t)-l(A(t)S(t) - Set~
is constant.
Concerning linear periodic differential equations and

characteristic exponents, see e.g. N. Rouche and J. Mawhin
[1973]
6.18.
Asymptotic stability of a glider.
Let us come back
to the problem of the glider, already presented in Section 4.8.

We consider here the case of a motion with non-vanishing
drag, i.e. described by the Equations (4.6).
These equations
admit of the critical point
corresponding to a rectilinear down motion at constant

velocity.
Without loss of generality, we assume that
o > 6 0 > -nI2.
Transferring the origin to the critical point
by the change of variables
= Yo
+ Yl' 6
computing the terms of first order in
6 0 + 61
and
Yl,6 1 , we obtain for
the linear variational equation
It is readily verified that the eigenvalues of the second
I.
34
member have strictly negative real parts.
Therefore,
according to 6.14, the critical point is asymptotically

stable.
6.19.
Theorems like 6.14 and 6.17 are very useful.
They
have a drawback, however, as compared to other theorems

using an auxiliary function:
they do not yield any estimate
of the region of attraction.

6.20.
Exercise (J.L. Massera [1949], see also H.A. Antosiewicz

If we replace in Theorem 6.2, Hypotheses (ii) by
[1958]).
(ii)
there exist a function
U: I x n
!if and a
c E 5t' such that
function
U(t,x) > c(llxll),

and for any
with
V(t,x) + U(t,x)
uniformly on
0,
U(t,O)
as
00.
P1 ~ I Ixl I ~ P2' then the origin is equi-
6.21.
Exercise (H.A. Antosiewicz [1958]).
uniformly stable and if there exists a

V: I x
every
(t,x)
(i)
(ii)
5f1
!if such that, for some functions
If the origin is
function
a, c E 5t' and
E I x n:
V(t,x) > a(IIxll): V(t,O)
0:
V(t,x) < -c(llxll):
then the origin is equi-asymptotica11y stable.

6.22.
Let us now introduce for the first time in this book
(but it will occur very often in the sequel!) a theorem
6.
35
which makes use of two auxiliary functions.

6.23.
two
Theorem (L. Salvadori [1972]).
~l
functions
V: I x fl
that, for some functions

I
-+-
Suppose there exist
and
W: I x fl
-+-
a, b, c EJe and every
~ such
(t,x) E
x 1'2:
(i)
(ii)
W(t,x)
(iii)
bJllxll); W(t,O) = 0;
V(t,x) < -c(W(t,x;
(iv)
Choosing
0;
W(t,x)
a >
is bounded from below or from above.

B C 1'2, we put for any
such that
t E I
l
Vt,a. = {x E 1'2: V(t,x) < a(a)}.
Then
-1
(al
the region of attraction
(b)
the origin is asymptotically stable.
Proof.
(al
We choose an
a >
A(t O) :) Vt,a.;
such that
Ba. efland
deduce, as in Theorem 6.2, that for any to E I and

-1
+
x o EVto'a.: J (to'x o ) = [to'oo[. Let us now prove that
W(t,x(t
-+-
as
-+-
If this were not the case, two
00.
mutually exclusive behaviors would be possible for
W, and
we shall rule them out one after the other.

First there might exist a
that, for every
to +
0:
V(t,x(t
-+-
_00
as
W(t,x(t
would follow from (iii) that

that
>
V(t,x(t
-+-
00,
and a
k > 0.
~ -c(k)
k > Osuch
But then it
and therefore,
which contradicts (i).
Secondly, there might exist two increasing sequences

{ti},{tl}
such that
ti
-+-
00
as
-+-
00
and for every
36
I.
= 1,2, :
some
ti < tl < t i + l , and further such that, for
k:> 0:
(6. 5)
k/2 < W(t,x(t
< k
for every
E ]ti,ti [.
Of course, one might as well have written

(6.6)
According to whether
W is bounded from above or from
below, we use (6.5) or (6.6) in much the same way for both
cases.
M > 0
So let us suppose there is an
W(t,x) < M.
It
is clear from (6.5) that
such that
ti - ti ~ k/2 H.
using (iii), we get
< V(to,x(t O
l<i<n
t!
~V(S,x(s) )ds
ti
The last member becomes negative for
large enough, and
this again contradicts (i).
The reasoning would be similar
for
Thus, W(t,x(t
W bounded from below.
and (ii) shows that

(b)
x(t)
as
as
00
00.
Part (b) of the thesis is immediate:
stability
derives from Theorem 4.2; attractivity results from the fact

that, for every
origin.
6.24.
-1
to: Vt
O,C(
is a neighborhood of the
Q.E.D.
The function
in several ways.
identifying
W in Theorem 6.23 can be particularized
For instance, we obtain Corollary 6.25 by
W(t,x)
with
(xix)
and Corollary 6.26 by
6.
37
putting
= V(t,x).
W(t,x)
By the way, these two corollaries
were known long before Theorem 6.23.

6.25.
Corollary (M. Marachkov [1940]).
exists a
:ifl
function
some functions
(i)
(ii)
such that, for
(t,x) E I x
n:
V(t,x) > a(I Ixl I); V(t,O) = 0;

Vet,x) < -cxlx.
for every
attraction
(b)
n .... 9R
a, c EYf: and every
If, moreover, f(t,x)

(a)
V: I x
Suppose there
is bounded on
such that
at
I x
n,
c n,
Bat
then
the region of
-1
A(t O) :J Vto,at;
the origin is asymptotically stable.
There is no restriction of generality to suppose that
n is a bounded set.
Then the bound on
to prove that the time derivative of
f(t,x)
(xix)
enables one
along the solutions
is bounded.
6.26.
a
Corollary (J.L. Massera [1956]).
:ifJ.
function
V: I x
a, c E Yf: and every

(i)
(ii)
Suppose there exists
n .... 9R such that, for some functions
(t,x) E I x n:
0;
V(t,x) ~ -c(V(t,x;
then
(a)
for every
a. >
of attraction
(b)
such that
-1 .
A (to) :J V .
to'a.'
Bat
c n,
the region
the origin i-s asymptotically stable.
One may even prove that the origin is equi-asymptotica11y

stable.
I.
38
6.27.
A pendulum with time-varying friction.
Let us
generalize Example 6.12 and consider the pendulum with

variable friction described by the equation
x+
where
~l
is a
h(t)x + sin x
function from
0,
(6.7)
+~.
We are looking
for some hypotheses, as mild as possible, concerning
h(t)
and which will entail asymptotic stability of the origin of

the phase plane.
V(t,x,x)
Let us try, as an auxiliary function

+
(x
. x) 2 +
a 2 S1n
bet) (1 - cos x),
which is the sum of a quadratic function
(x
+ ax)
+ bet) x
2
and of terms of order at least

b (t) = 1 + ah(t) - a
in
x,x.
If we put
we obtain
-(het) - a cos x)x 2 - a sin 2 x + ah' (t) (1 - cos x)

a 2 (1 - cos x)x sin x
-(h(t) - a)x 2 - a(2-h'(t
tj
where
(1 - cos x) +
contains only terms of order at least
x,x, all independent of
&3
in
t.
Now, if there exist two constants
a > a
and
< 2
such that
(i)
(ii)
then
h(t).:: a > a >

h'(t) < S <
and
-v
O~
2~
are positive definite, as can be readily
6.
established.
39
Further, let us define
x + (1 - cos x)
2"
2
the derivative of which, namely
is bounded from above.
Asymptotic stability of the origin
follows then from Theorem 6.23.

Notice that Corollary 6.25 could not be used to get
this result, because of the requirement that
f(t,x)
be
bounded and therefore that a bound should have been assumed

for
h(t).
Corollary 6.26 could not be used either, at
least using the functionV(t,x,x)
chosen above, because
satisfying hypothesis (ii) of this corollary would also

require further hypotheses on
point of
while
where
h' (tl
x =0
h(t).
and a
h(t)
is bounded from below:
In fact, consider a
approaching
00
hypothesis (ii) of
Corollary 6.26 would be violated.

Notice further that, although condition (ii) of this
section may appear somewhat odd, for instance from a
physicist's point of view, it remains that (i) alone would
not entail asymptotic stability, as we shall prove presently
by way of a counter-example.
Consider first (after
J.K. Hale [1969]) the similar problem

(6.8)
which, for any
a, admits the solution
= a(l+e -t ).
The
orbit of such a solution is contained in the line of equation

y +
= a.
Of course, the origin of the phase plane is not
I.
40
X or y
8
."
Figure 1. 5.
The phase plane for equations (6.7) and (6.8)
Let us now corne back to Equation (6.7) with

and let us, for some
a, draw the line
phase plane, as shown in Figure 1.5.
x + x = a
=2
+ et
in its
The phase velocity,
i.e., the velocity of the point representing a solution in

the (x,x)-plane, is
(x,x)
or, using (6.7)
(x, -(2+et )x - sin x).
(6.9)
The analogous velocity for (6.8) would be

(6.10)
Consider now a triangle such as

the x axis.
side
AB
aAB, with
AB
parallel to
It is readily seen that at any point of the
and for any
t, the phase velocity (6.9) enters
6.
41
x<
this is because
the triangle:
and any point of the side
o.
Similarly, for any
aB, this velocity enters the
triangle if
B has been chosen at the left of the point of
abcissa
Consider now two solutions, one
1T.
x(t) of (6.7),
the other
yet)
time
Comparing (6.9) and (6.10) shows that the solu-
to.
of (6.8), both starting from
tion of (6.7) enters the triangle at
A.
A at some
But this solution
cannot come out of the triangle, neither by crossing

by crossing
aA
aBo
Suppose now that having left
fo:t' the first time at time
ing point of
reaches
P.
aA.
rx
P.
..
X(T)
to
AP
Let
T.
be' the correspond-
yet)
since
T' < T,
>
AB, nor
A, it meets
be the time at which
frx itdx fry dy
is the orbit of
is the segment
A to
T'
Observe that
T -
where
Let
v __
~
T'
x(t)
from
to
which is also the orbit of
while
yet)
from
Therefore, we may write
= -(2+e,. )X(T)
- sin X(T) > -(2+e
We conclude that at point
P, x(t)
T'
)y(T') - y(T')
.
= y(T').
would enter the triangle
with a nonvanishing velocity, which is absurd.

We sum up our results: for
satisfying condition (i, for any
x + it
= a,
h(t) = 2 + e t
a
(a function
and any point on
we have found a solution of (6.7) remaining forever
in the triangle
aBA.
Thus, condition (i) alone cannot ensure
asymptotic stability of the origin for (6.7).

Exercise (L. Salvadori [1972]).
6.28.
two
~l
functions
that, for some

(t,x) E:
x n:
a E
V:
x n ... 91
and
91, some functions
Suppose there exist

W:
xn ... 91
such
b, c E? 5t' and every
42
I.
(i)
V(t,x} > ai
W(t,x) > b(llxll); W(t,O) == 0;
(til
(Hi)
V(t,x) S. -c (W(t,x ;
(iv)
W(t,xl < 0;
then the origin is equi-asymptotica11y stable for

equation (2.2)
The two following exercises yield improvements of
Corollaries 6.25 and 6.26.
In the first one (of Marachkov),
the hypotheses can be weakened while proving the same thesis.

In the second one (of Massera), the thesis can be made
stronger without modification of the hypotheses.
6.29.
Exercise.
Use Exercise 6.28 to prove that the
hypotheses of Corollary 6.26 imply equi-asymptotic stability

of the origin.
6.30.
Exercise (J.R. Haddock [1972], A.S. Oziraner [1972]).
Show that in Corollary 6.25, Hypothesis (i) can be replaced

by:
for every
5f1
k: I
function
ge
V(t,O)
(i)
eHl
(iH)
V: I x n ~ ~
and
a E Jt'
V(t,x) > k(t)a(llxll); V(t,O}

Vet,x} < 0;
~
00
0.
and a continuous function
n:
k (t) > 0; k (t)
V(t,x)
Suppose there exist
such that for some function
(t,x) E I x
then
n,
Exercise (N.G. Chetaev [1955]).
6.31.
a
(t,x) E I x
as
00;
0;
and every
6.
Asymptotic
stability
43
-1
(a)
(b)
the origin is equi-asymptotically stable.
6.32.
A(t O)
Partial asymptotic stability.
:::>
Vt
O,a.
Let us get back to the
Equations (4.1) along with the accompanying hypotheses as

presented in Section 4.4.
Here we suppose further that all
solutions of (4.1) exist on
[to'~[.
The solution
z = 0
of (4.1) is said to be uniformly aSymptotically stable with

x, if it is uniformly stable with respect to
respect to
and i f
(3 T) > 0) (
I I Z0 II <
> 0) ( 3 a > 0)( V z 0 :
'If E
( 'Ito E I) ('It:: to +
Ilx(t~to'ZO) II < E.
0)
stability with respect to
T) )
Asymptotic
is defined similarly
(V.V. Rumiantsev [1957).

6.33.
V: I
Theorem.
~m ... ~
x rI x
b, c E.5t' and every

(i)
(ii)
.lfl
function
such that, for some functions

(t,z) E
x rI
><
a,
~m:
a (I Ix I P ~ V Ct , Z) ~ b CI Ix I I ) ~
V(t,z) ~ -cCllxll).
Then
(a)
for any
x
and any
~m), x(t~to'zO) ... 0
when
Cb)
a. > 0
t ...
(to'zO) E
uniformly in
x [(Ba.
n r1)
to'zO
~~
the origin is uniformly asymptotically stable

with respect to
x.
The proof is along the same lines as for Theorem 6.2.

The theorem corresponding to thesis (b) appears in N. Rouche
and K. Peiffer [1967).
[1971].
See also V.V. Rumiantsev [1957) and
From this last author [1970], we borrow the
44
following statement.
Exercise.
6.34.
Define
/11Jn+k, 0
E ~
containing all components of

5:f1
m, as a vector
components of
V: I x n x~m
function
y.
-+
a,b,c E..5e and every
such that, for some functions

(t,z)
and
I x n x ~m:
(i)
(ii)
aCllxll) ~ V(t,z) ~ b{llull)1
V(t, z)
~ -c (j Iu
II ) 1
then the origin is uniformly asymptotically stable with

respect to
6.35.
a
x.
Exercise (A. Halanay [1963]).
5:f1
function
functions
(i)
(ii)
V: I x
~m
a, c E..5e and every
-+
9l such that, for some
(t,z) E I x n x
V(t,z) > aCllxll) 1 V(t,O)

V(t,z)
<;
~m.
01
-c(V(t,z 1
then the origin is asymptotically stable with respect to
7.
7.1.
x.
Converse Theorems
In most theorems given so far, the existence of a
Liapunov-1ike function is assumed.

such a function actually exists:
The question arises whether

given some stability or
attractivity property of the origin, can we build up an

auxiliary function suiting the corresponding theorem?
answer is contained in the so-called converse theorems.
Three comments are in order here:
The
7.
Converse theorems
(I)
45
First, most converse theorems are proved by
actually constructing a suitable auxiliary function.
Un-
fortunately, this construction assumes almost always a

knowledge of the solutions of the differential equation.
This is why converse theorems give usually no clue to the
practical search for Liapunov-like functions.
In this
respect, all they show is that this search, however

difficult, is not hopeless, and this is a little more than
nothing.
(2)
When the existence of such and such a function
is necessary and sufficient for some stability property,

one knows of course that any other sufficient condition will
imply the first one.
But this observation should not pre-
vent the mathematician from looking for other sufficient

conditions which might happen to be more practical, easier
to apply.
Many theorems in the rest of this book, and in
particular those utilizing several auxiliary functions,

present this type of interest.
(3)
Sometimes a stability property of a system can
be investigated by considering first a simplified system.

Let us suppose that the stability of the latter can be
easily established.
Then we can deduce from a converse
theorem the existence of a suitable auxiliary function.

Under appropriate assumptions, it might also be a good
auxiliary function for the original system and prove the
desired stability property.
This way of thinking is typical
of total stability (see Section II.4) and of the proofs of

stability using the first approximation, but is used in
several other circumstances (cf. F.C. Hoppensteadt [1966]).
Hereafter, we state without proof the converses of
46
I.
ELRMENTS OF STABILITY THEORY
three important theorems given above.

is again as in Section 2.2.
The general setting
For proofs and further results,
the reader is referred to I.G. Malkin [1952],

N.N. Krasovski [1959], T. Yoshizawa [1966] and W. Hahn [1967].
7.2.
The converses of Theorems 4.2 and 4.3 have been ob-
tained by K.P. Persidski [1933] and J. Kurzweil [1955]

respectively.
Theorem.
If the function
~l
is
stable, there exist a neighborhood
~l
and every
7.3.
V: I x ~
function
(t,x) E I x
C n
of the origin and
such that, for some
E~
~:
(a)
V(t,x) ~ a( Ilxll), V(t,O)
(b)
V(t,x)
Theorem.
and if the origin is
O~
< 0.
I f the function
is
~l
and if the
origin is uniformly stable, there exist a neighborhood
~ en
of the origin and a
such that, for some
~l
function
a E ~ and
V: I x ~
b E ~, and every
(t,x) E I x ~:
7.4.
(a)
b(llxll) ~ V(t,x) > a(llxll), V(t,O)
(b)
V(t,x)
O~
< 0.
J.L. Massera,
[1949] and [1956], was the first to ob-
tain a converse of Liapunov's theorem on uniform asymptotic

stability (cf. 6.2).
Before stating it, let us recall that
the function
defined in Section 2.2 is said to be
f(t,x)
locally liEschitzian in
if, for every point of
this point, N C
t EI
n,
n,
uniforml~
with resEect to
there exist a neighborhood
and a constant
and any pair of points
t
N
such that, for any
xl' x 2 EN, one gets:
of
8.
Bibliographical note
47
Ilf(t,x l ) - f(t,x 2 ) I I ~ kl IXI - x21 I.
The same function is
said to be lipschitz ian in
if there exists a constant
such that, for any
and any pair of points
xl' x 2 E
n,
one gets:
t E I
Ilf(t,x l ) -
fCt,x 2 ) II ~ kllxl - x211.

Theorem.
If the function
lipschitz ian in
is, on
I x
n,
locally
and if
the origin is uniformly asymptotically stable, there exist

a neighborhood
V: I x
~ + ~
C n of the origin and a function
possessing partial derivatives in
of arbitrary order, such that for some functions

and every
Ct,x) E I x
aCllxll) < V(t,x) < b(j Ixll);
Cb)
vet,x) < -cCllxll).

If
is, on
formly with respect to
I x
t, V
n,
x ~,
n,
independent of
8.
uni-
can be chosen such that all

V
are bounded on
with the same bound for all of them.

independent of
a, b, cEJt
lipschitzian in
partial derivatives of any order of

I
~:
(a)
Corollary.
and
or periodic in
or periodic in
If
t, V
is, on
can be chosen
respectively.
Several references have been given in the text, where

we have made an effort to mention the author of each theorem.
Although it would be interesting to write in detail the early
story of stability theory, for lack of space we refrain from
doing so.
Let us content ourselves here with the mention of
a few introductory texts, the contents of which might usefully

precede or complement the present one.
I.
48
I.G. Malkin [1952] contains a very explicit and careful treatment of the basic theory (without use of vector
notations, which renders the text a bit lengthy).
N.G. Chetaev [1955], although somewhat old-fashioned now,
is still interesting, in particular because it contains many
mechanical applications, a too rare feature indeed!
W. Hahn
[1959] was amongst the first western authors to write on

Liapunov's second method.
The first four chapters of his
book constitute a very good and rigorous introduction to the

subject.
They are accompanied by an extensive bibliography.
In a second book [1967], he gathered more, more varied, and

up to date material.
H.A. Antosiewicz [1958] is a survey of
the theory at the end of the fifties, the instability theorems

being excluded.
It is written in a beautifully concise style
and should be highly recommended.
J.P. LaSalle and
S. Lefschetz [1961] is interesting for the beginner:
the
exposition is definitely elementary, being in particular

limited to autonomous equations.
Finally, many textbooks
on differential equations devote some chapter to stability

theory, for instance A. Halanay [1963], C. Corduneanu [1971],
N. Rouche and J. Mawhin [1973], Vol. II, H.W. Knobloch and
F. Kappel [1974] and several others.
CHAPTER II
This chapter is varied in character.
It is meant to
complete the elementary view of stability theory which we

began to present in Chapter I.
Sections I and 2 examine in
different ways what conclusions can be drawn from the knowledge of a positive definite auxiliary function
V{t,x)
possessing a derivative which is only smaller than or equal

to zero.
Section 3 is an elementary presentation of the so-
called comparison method.
All these matters will be
investigated further in later chapters.
Section 4 deals
briefly with total stability, i.e. stability with respect to

perturbations not only of the initial conditions, but also
of the second member of the differential equation.
The main
theorem proved here will be needed in Chapter IV.
Section 5
is but a glance at the "frequency method", introduced here

because of its intrinsic importance and for the sake of
completeness, but considered later as outside the scope of
the rest of the book.
The last section is an example of
II.
50
circuit theory showing that continuously differentiable auxiliary functions are not always satisfactory, and thus
justifying the fact that, in subsequent chapters, the required
regularity conditions will be somewhat weakened.
1.
1.1.
Theorems of E.A. Barbashin and N.N. Krasovski

for Autonomous and Periodic Systems
As already observed in Section I.6.l2, it is sometimes
difficult and time consuming to find an auxiliary function

with negative definite time derivative.
In this section, we
endeavor to prove asymptotic stability, global as well as

local, and instability, by using a
V(t,x)
whose time
derivative is only smaller than or equal to zero, which is a

much more common situation.
Of course, we shall have to
compensate by introducing a new hypothesis:

set
M where
V(t,x)
namely that the
contains no complete trajectory.
Inasmuch as this condition is often realized and easily

recognized, the theorems to come are amongst the most useful
ones.
In particular, as we shall see in several instances,
they allow using the total energy as an auxiliary function

in mechanical problems with dissipation.
They have a draw-
back however, in that they pertain to autonomous and periodic

equations only.
1.2.
Our general hypotheses in this section will be those
of Section I.2.2, but we assume further that
f(t,x)
periodic with respect to
1.3.
a ~1
t, for some period
Theorem (N.N. Krasovski [1959]).

function
V: I x n
>
is
o.
!Jf, periodic in
with period
T,
1.
Theorems of Barbashin and Krasovski
such that for some function
a E .5t' and every
V(t,x) > a(jlx/l); V(t,O)
(1)
51
(t,x) E I
x rl:
0;
(H)
V(t,x) < 0; we put M = {(t,x) E I x rl: V(t,x) = oJ;
(iii)
except for the origin, M cQntains no complete

positive semi-trajectory.
Choosing
a > 0
such that
Sa C rl, let us put for every
tEl:
V-I
t,a
{x E Q: V(t,x) < a(a)}.
Then
-1
(a)
(b)
A(t O) ::>Vt
O,a
If (i) is replaced by
then the origin is unstable.

Proof.
To prove the first part of this statement, observe
first that the origin is stable, by Theorem 1.4.2 and uniformly

stable by 1.2.14.
and
to E I.
Let us choose
a > 0
such that
Sa C rl
We deduce as in 1.6.2 that every solution
starting at some
-1
is defined over
Xo E Vto,a
prove now that such a solution
as
-+- "'.
Due to uniform stability (cf. 1.6.8), it suffices to prove

that
(V~ > 0) (3t ~ to)
Ilx(t;to'x o ) II
not true, there would be an
-1
Xo E Vto,a
<~.
and a
If this were
~
> 0
such
that
(1.1)
52
II.
But then, the sequence

has a cluster point
written
Xo
*
k
*
{xk}
o )' Xo + Xo
V(t,X(tito'XO
for
= 1,2,
and, for some subsequence (also

as
k +
00.
As a function of
t,
is decreasing and bounded from below:
therefore, it tends to a limit as

continuous and periodic in
lim V(t,X(tito'XO
t +
00.
But, V
being
t, the limit is obtained thus:

lim V(tO+kT,x(tO+kTito'XO
t+oo
k+oo
(1. 2)
X(tito'X~), we deduce from (iii) the
Considering now
existence of a
> to
such that
. * ,x(t * ito'XO
*
Vet
<
and therefore that

(1. 3)
The periodicity of
Due to (1.2),
f(t,x)
yields
(1.3) and the periodicity of
V(t,x), we get
finally the following contradiction

V(to'x *
O)
.
= l~m
k+oo
- O
Vet * +kT,x(t *+kT;to'X
lim vet * ,x(t * ;to'xko
k+oo
= Vet * ,x(t * ;to'x *o
~ V(to'x o ).
v- l
is a
to,a
neighborhood of the origin, this point is asymptotically
Part (a) of the thesis is proved and, because
stable.
Then it is uniformly asymptotically stable by
Theorem 1.2.14.
1.
53
The part of the thesis concerning instability is

proved ab absurdo.
a > 0
If the origin were stable, then for any
Ba C 0, one should have
with
But (i)' shows that for some

5 > 0
In that case, for some

I Ix(tJto'xO) I I ~ 5
V(t,O) = O.
to E I
because
and Xo E BT}: V(to'xo) < O.
and every
to:
V is continuous, periodic and
But then (1.1) is verified and the expected
contradiction is obtained as above.

is unstable.
1.4.
The"refore, the origin
Q.E.D.
Remarks.
(1) Theorem 1.3 cannot be extended to
general nonautonomous systems, as is shown by the equation
x=
-p(t)x
for
J:PCt)dt < "'.

whereas
x E
ge, with pet)
>
for
t E [0,"'[
and
The origin is not asymptotically stable,
V = x2
satisfies Hypotheses (i) to (iii) of
Theorem 1.3 (cf. V.M. Matrosov [1962]1).

(2) To prove instability, Hypothesis (iii) can be
replaced by the following one:
(iii)'
The set
M II {(t,x): V(t,x) < O}
contains no
complete positive semi-trajectory.

In the Corollary to follow, the general setting is
as in Theorem 1.3, with the exception that
1.5.
Corollary.
0 =
gen.
In the hypotheses of Theorem 1.3, if one
assumes further that
= gen
and that
a(r)
the origin is globally asymptotically stable.
'"
as
+ "',
54
II.
Proof.
a(r)
Indeed, due to the fact that
Xo E !Jfn
there is an
a. > 0
....
for every
00,
-1
such that
Xo E Vt O,a. ' and

Q.E.D.
thus
x (t;to'x O) .... 0
1.6.
Several examples illustrate hereafter Theorems 1.3
and Corollary 1.5.
t ....
as
00.
Another example, pertaining to the
asymptotic stability of a mechanical equilibrium appears in

Section III. 6.
1.7.
Exercise.
A simple pendulum with a constant torque
applied to it and some viscous friction admits of the equation, for

and
x E!Jf:
x + ax
+ (lsin x
are real quantities.
L, where
a > 0, w
One assumes that
<
w2
What about the stability behavior of the equilibrium positions?

(see E.A. Barbashin [1967]).
1.8.
A transistor oscillator.
One of the simplest types
of transistor oscillators is the based tuned circuit shown

in Figure 2.1.
It is studied in all elementary books on
electronic engineering and its equation has been investigated
I: n
Figure 2.1.
A transistor oscillator
1.
55
by many mathematicians since Van der Pol.
Of course, in
the days of Van der Pol, it was the equation of a vacuum tube
oscillator!
genera~e
Although the purpose of this circuit is to
electrical oscillations, we shall limit ourselves
here to exhibit the stability properties of its rest state:

in case of instability, it generates oscillations, in case
of asymptotic stability it does not.
It consists of a linear RLC circuit connected as
shown between emitter and collector, but with an ideal transformer in the base circuit.
We refer to specialized books
for the derivation of the equation, which reads

(1. 4)
where
is the current through the inductor
a
is a ~l
= RC
I
> 0
'
w2
1
Lc
> 0 and
L,
g: 91 .... 91
function, depending on the characteristics of the
transistor.
In fact
is strictly increasing and
There is a unique equilibrium at
= x = O.
g(O)
O.
The
derivative of the auxiliary function
V(x,x)
= 12
2 x
21 w2 x 2
is computed easily and reads

xg(x) - ax 2
where
g'(O)
h(x)/x .... 0
>
a, V
respect to
.
x
as
x .... O.
[g'(O) - a]x 2 + hex)
Accordinglyas
g'(O) < a
or
is negative or positive definite, but with

only:
vanishes if and only if
as Equation (1.4) shows and except at the origin, x
O.
0
But,
56
II.
implies
O.
Therefore, there is no complete positive
semi-trajectory in the set
M of the Theorem 1.3.
We con-
clude that the origin is unstable or asymptotically stable

g' CO) > a
according to whether
g' CO) < a.
or
The
functions to consider in applying Theorem 1.3 are

asymptotic stability and
1.9.
-V
for
for instability.
Example of an equation of third order (E.A. Barbashin
and N.N. Krasovski [1952], and E.A. Barbashin [1967]).

f
and
with
cp
be two functions from
0, cp
f (0)
9R .... 9R, f
of class
0, and let
continuous with
CP(O)
Let
)fl
a > O.
The equation
CP(x) +
'i'+ ax +
f(x) = 0
(1. 5)
arises in several problems of control theory and in the

dynamics of systems containing a gyroscopic pendulum, like
the monorail (cf. H. Leipho1z [1968]).
z
=y+
ay
Putting
=x
and
transforms (1.5) into
y,
z - ay,
-CP(y) - f(x).
(1. 6)
Let us look for some conditions which would entail global

asymptotic stability of the origin
= y = z = O.
The time derivative of the function

V(x,y,z)
= aF(x)
1 z2
+ f(x)y + w(y) + 2
where
F(x)
W(y)
1.
57
reads
V(x,y)
[f' (x)y - a<jl(y)]y.
Let us put
= aF(x)
W(x,y)
+ f(x)y +
~(y)
and assume the following:

(i)
(ii)
(iii)
f (x) x > 0
for
x ., 0;
[a<jl (y) - yf'(x)]y > 0

W(x,y)
....
co
as
x2 + y
Because of (iii), to prove that
for any
2
....
00.
V(x,y,z)
is positive
W(x,y,z) > 0
definite, we only have to prove that

(x,y) ., (0,0).
y ., 0;
x, i f
for any
The way to prove that, consists in consider-
ing first the integral
J:
[a<P(n) - nf' (x) ] dn
which, because of (ii), is strictly positive for any

y ., O.
to
Multiplying by
f(x)
x, if
and integrating with respect
x, one gets
= aF (x) ~ (y) -
(x)
(1. 7)
a function which, due to (i) and the way it was built up, is
strictly positive for every
(x,y)., (0,0).
Observing then
that
W(x,y)
and because
W(x,y) > 0
(2~(y) + yf(x2
4~(y)
~(y)
> 0
for any
for
4aF(x)~(y) - y2f2(x)
4~(y)
y., 0, we see that effectively
(x,y)" (0,0).
That
~(y)
> 0
results
II.
58
from the second member of (1.7) being positive.

As far as
is concerned, it is obviously
vanishes if and only if

the
O.
But the plane
< 0
and
in
(x,y,z)-space contains, besides the origin, no complete
orbit.
In fact, for an orbit in this plane, one would obtain
identically
=y =
The second equation (1.6) yields
.
z = z =
then identically
(1.6) gives
O.
f{x)
O.
Finally, the third equation
0, which is equivalent to
x = O.
All
the hypotheses of Corollary 1.5 being satisfied, the origin

is globally asymptotically stable.
1.10.
Exercise (J.P. LaSalle and S. Lefschetz [1961]).
Consider the equation

x+ f(x)x + ax + bx
where
a > 0, b > 0
=y = z =
= y,
in order that the
= z,
-f{y)z - ay - bx.
the following auxiliary function may be used:

V(x,y,z)
1.11.
is a continuous function from
be globally asymptotically stable for
the equivalent system
Hint:
Find some conditions on
~ +~.
origin
and
= 0,
= 2a
1 (bx+ay) 2
z 2 + byz + b JYof{n)ndn + 2
An example from chemical kinetics in biology.
Genetic
control mechanisms in bacteria can be described (cf. R. Rosen

[1970] ). using a primary gene product
protein
X, an enzymatic
Y determined by this gene and a metabolite
produced by the metabolic activity of the protein.

and
m be the concentrations of these products.
M
Let
A first
x,y
1.
59
model of the genetic mechanism corresponds to the following

equations
x
y
.
m
where
K,A,C,a,13
and
K
m
Ax
Cy
- a,
-
13,
ym,
yare positive constants.
equation expresses the fact that the metabolite

inhibitor for the primary gene product.
is that
is an
A common assumption
m varies slowly, in such a way that
approximated by zero.
The first
m can be
In this case, the equations become
(B. Goodwin [1963]):

x
Ky
Cy - a,
Ax -
13,
and the stability of the equilibrium

x
a/A, y
Ky/Ca
follows, by Theorem 1.4.2, from the existence of the first

integral
V(x,y)
which is such that
Ky
A 2
- -C lny + ay + -2 x
V(x,y)
- V (a/A,Ky/Ca)
- ax
is positive
definite in some neighborhood of the equilibrium (prove'

this !).
A criticism to Goodwin's equations is that, the influence of
being considered apart, y
disappears at a
constant rate, whereas it is more plausible to assume that

it disappears at a rate proportional to its concentration.
II.
60
In that case, the equations become

x
Ky
Cy - a,
Ax - Sy,
and the equilibrium
SKy
_ Ky
ACa ' y - Ca
is asymptotically stable as follows, by Theorem 1.3, from

the existence of the auxiliary function
V(x,y) = - ~ lny + ay + ~A (AX-Sy)2
whose derivative reads
V(x,y)
2.
2.1.
A Theorem of V.M. Matrosov on Asymptotic Stability

Section I was entirely devoted to proving asymptotic
stability or instability through the use of an auxiliary

function
V(t,x)
whose time derivative
V(t,x)
is not
negative definite, but only smaller than or equal to zero.

The corresponding theorems were restricted to the autonomous
or periodic case.
In this Section, we present a theorem of
V.M. Matrosov dealing with the general nonautonomous case.

The crucial condition that the set
M where
V(t,x)
contains no complete positive semi-trajectory can no longer

be used as such.
The difficulty is overcome by introducing
a second auxiliary function defined in some appropriate

neighborhood of
M and possessing adequate properties in
order that the solutions do not spend too much time near
M.
2.
61
Matrosov theorem on asymptotic stability
2.2.
We consider again Equation I(2.2), and the general
hypotheses associated with it.
As the proof of the theorem
itself is somewhat intricate, it will be rewarding to prove

first two lemmas.
2.3.
(lower bound for the transit time of a
If
solution).
x(t)
solution
and
tl
are in
t2
is defined on
[tl' t 2 1
(t,x) E S'l: Ilf(t,x)11 < A where

Ilx(t l ) - x(t 2 ) II > r > 0
I, with
tl < t 2 , if a
and if, for every
is a constant, then
implies that
t2 - tl 2: riA.
In fact
~.
Q.E.D.
2.4.
Lemma (upper bound for the escape time of a solution).
If there exist a
c. EYe, a set
.1(1
E C S'l
function
W: I x
and a constant
n +!JI, a function
L > 0
such that, for
(t,x) E I x S'l:
every
(i)
(ii)
IW(t,x)
< L~
max(d(x,E),IW(t,x)I) 2: c(llxll)~
then, for any
n >
O.~
a solution
x (t)
cannot remain in
the set
u = {x E S'l: c(1 Ixl I)

on a period of duration
Proof.
interval
Suppose
> n,d(x,E) < n}
2L/n.
x(t) E U for every
[tl,t21.
in some closed
It follows then from (ii) that
62
II.
Using (i), one gets successively
which yields the expected result.
We are able to write that
the absolute value of the integral is. equal to the integral

of the absolute value because
fore does not change sign on
2.5.
jfl
W is continuous and there[t l ,t 2 ].
Theorem (V.M. Matrosov [1962]1).
V*: n .... 9f, three functions

A > 0
constants
(t,x) E I x
L > 0
and
a (11x II) :; V(t,x)
(ii)
V(t,x) < V*(x) <
(iv)
(v)
choosing
a,b, c E .5:' and two
such that, for every
IW(t,x) I <
O~
put
{x
-1
A~
such that
{x
V*(x)
IW(t,x) I) > c("x")~
Ilf(t,x)11 <
Vt,a
En:
L~
max(d(x,E) ,
a > 0
< b("x")~
Ben,
a
let us put for every
tEl
Then
jfO
n:
(i)
(iii)
Let there exist two
V: I x n .... 9f, W: I x n .... 9f, a
functions
function
Q.E.D.
En:
V(t,x) ~ ala)}.
2.
for any
Cal
to E I
(b)
proof.
an
and any
to'x o ' when
uniformly in
63
00;
The reasoning begins as in Theorem I.6.2.
every
and deduce from (i) that for
B C fI
a
such that
a > 0
We choose
tEl
l
v-t,a
For any
to E I
(2.1)
C fl.
-1
Xo E Vto,a' it follows from (E) that
and
x(t) E
l
v-t,a
E J+, and therefore from (2.1) that

+
approach the boundary of fl. Hence J (to'x O)
for any
n > 0
e > 0, choose any
For any
x(t)
[to'oo[.
such that b(c-l(n)) < ale).

-1
Further, let us consider any
(to'x o ) E I x Vto '
suppose ab absurdo that the corresponding solution

remains for all
u
U*
t > to
{x E
Ba
cannot
and
x(t)
in the union of the sets
(I Ix I I)
>
n, d (x, E) < n},
{x EB: c(IIxll) > n,d(x,E) > n/2L

a
Then, on any subinterval
[T l ,T 2 ] C [to'oo[, with
T2 - Tl
2L/n, one would have two possibilities:

(cd
write
Vet)
x(t) E U*
either
for
V(t,x(t
T2
V(T 2 ) - V(T l ) =
Tl
V(T)dT
for
and
t
vet)
E [T l ,T 2 ] ; then if we
for
T2
Tl
V*(x(TdT
V(t,x(t, we get
~ -2~L/n
II.
64
where
9.= inf { I V* (x) I: x E U*};
(S)
or there is a t E [T l ,T 2 ]
x(t) E u\u*.
such that
But it follows from Lemma 2.4 that there is a t E [T l ,T 2 )

for which
instants
Since
T2
x(t)
U.
At the first mentioned of these
d(x,E) < n/2, while at the second
x(t)
is continuous, there exist two instants
E [T l ,T 2 ]
Ti'
such that
d(x(Ti),E) = n/2,
d(x(T 2 ),E) = n,
and, if we suppose for definiteness that

d(x(t),E) > n/2
for
We deduce from Lemma 2.3 that

vet)
d(x,E) > n.
is monotonic decreasing
2,
Ti < T
E [Ti,T 2 ].
IT2 - Til> n/2A.
Thus, as
T'
2V(T)dT < -9.n/2A.

Ti
In either case
In order to juxtapose several intervals like

k > 1,
let us put, for some integer

ti = to + 2iL/n
1 < i
< k.
Then
VetO) +
k-l
L (V(t+
l )
1.
i=O
- vetil)
and further,
V(t k )
a(a) - k min[29.L/n,9.n/2A).
[T l ,T 2 ),
2.
on asymptotic stability
M~t~osov theo~em
Of course, if
is chosen large enough, this inequality
contradicts (i).
Noticing that
is independent of
exists a
a > 0
tl E [to,tO+a]
65
k, determined in this way,
(to'x O)' we obtain the following:

-1
(to'x O) E I x Vto,a' a
and, for every
I Ix(t l ) I I ~ c
such that
there
-1
The rest of
(n).
the proof is a trivial transposition from Theorem I.6.2.

2.6.
The pendulum with time varying friction.
Q.E.D.
Let us con-
sider again a pendulum with time varying friction described

by the equation
x + h(t)x + sin x
O.
using the total energy

V
.2
=~
2
+ (l-co,s x)
as auxiliary function, and in view of obtaining the largest

possible region
Q = {
V- l
it seems appropriate to choose
to,a'
2 .2 1/2
(x,x): (x +x )
small.
<
'IT
5t' of Theorem 2.5 can be chosen as
existence of the function
V= -h(t)x 2
and
k2
2r
a(r)
2
/'IT
The derivative of
such that
W = x.
kl
The
V reads
h(t) > k2 > 0, we may choose

E
{(x,x) E Q: x
This function is bounded on
there exists a function
= a}.
Let
and further,
c E 5t' such that
max(dx,X),E),I*I) = max(lxl,lxl)
=
of
If we suppose that there exist two constants
= -k 2x.2 , in such a way that
us use
arbitrarily
> 0
is obvious and it would be of
no use to specify it further.
V*
E:
One can then verify that the function
class
kl
for some
max(lxl,lh(t)x + sin xl) ~ c[(x2+x2)1/2).
II.
66
This can be proved as follows.
First
Ih(t)x + sin xl ~ h(t) Ixl - Isin xl > k2lxl - ISin xl

and
Ih(t)x + sin xl > Isin xl - h(t) Ixl > Isin xl - k1lxl.
Hence
max(lxl,lh(t)x+sinxl) >
The function
F(x,x)
is continuous.
= max(lxl,k2Ixl
Further it is
(x,x) t O.
Let us write
c * (r)
= inf{F(x,x),
> 0
- Isin xl,lsin xl - k1lxl)

for any
r < (x 2+x.2 ) 1/2
(x, x) E 0,
~ ~
- E}.
This function is monotonic increasing, and vanishes only for

r
J.
= O.
As is well known (see e.g., N. Rouche and
Mawhin [1973]), there exists a function
for every
r: c(r) < c*(r).
c E !It such that
Hence, the desired result.
All the hypotheses of Theorem 2.5 being verified, tt

origin is uniformly asymptotically stable.
any
CI. <
Further, for
~,
.2
V~~'CI. = {(x,x): ~
+ (l-cos x) <
~~
2
} C A(t O)'
This result extends in some way those of Examples I.6.12 and

I.6.27.
h{t)
However, we know from I.6.27 that the condition

k1
is not necessary for asymptotic stability, a fact
which illustrates one of the limitations of Theorem 2.5.
2.
M~tro~ov
2.7.
Asymptotic stability of a gyroscope (V.M. Matrosov
[1962]1).
point
OZ
67
theorem on asymptotic stability
Consider a symmetric rigid body with a fixed

in some inertial frame of reference OXYZ, where
is a vertical axis oriented upward.
Let
Oxyz
be a system
of axes along the principal directions of inertia of the body

at
and fixed in the body.
With suitable hypotheses
about friction forces and the driving torque along
Oz, the
equations of motion read

Ai>
(A-C)qr + mgz oY2 - aR/ ap,
(2.2)
Aq
(C-A)pr - mgz oYl - aR/ aq,
(2.3)
cr
M(t,r) ,
(2.4)
Yl
rY2 - qy3 ,
(2.5)
Y2
PY 3 - rYl'
(2.6)
Y3
qY l - PY2'
(2.7)
2
2
2
Yl + Y2 + Y3
In these equations
A,A,C
inertia of the body at

Oxyz
(2.8)
1.
are the principal moments of
01 p,q,r
are the components in
of the instantaneous rotation of the body in the
inertial frame; R
= R(p,q)
geneous form of degree
k > 2, with coefficients continuous
and bounded functions of

torque around
Oz
is a positive definite homo-
t, and
M(t,r)
of the friction and driving forces.
Further, m is the mass of the body, g

gravity,
Zo
the acceleration of
the z-coordinate of the center of inertia of
the body and finally

inertial
is the resultant
fr~me
Yl'Y 2 'Y3
are the components in the
of the unit vector along
suppose hereafter that
Zo
<
Oz.
We shall
o.
We assume that equation (2.4) possesses a solution
68
II.
ret)
which is a bounded function of
t.
Replacing
ret)
in the other equations and replacing also
Y3
by
by its
value taken from (2.8), we get the following four equations

Ap
(A-C)qr(t) + mgz OY2 - OR/Op,
Aq
(C-A)pr(t) - mgz OY1 - aR/aq,
Y1
2
2
r(t)Y2 - qll - Y1 - Y2 ,
Y2
P/1 -
l1
2
- Y2 - r(t)y 1
We choose the plus sign for the square root in order for the
equilibrium
correspond to
ing upwards.
= q = Y1 = Y2 = 0 of
Y3 = 1, and therefore
Remembering that
these equations to
to the
Oz
axis point-
zo < 0, we see that the
center of inertia at equilibrium is below
O.
As auxiliary functions suiting Theorem 2.5, we choose
One computes easily that
v = _p
and, as
one has
ClR _ q ClR
Clp
aq
= -kR
is negative definite with respect to
= {(p,q,y 1 ,y 2 ):
=q =
O}
n n,
where
appropriate neighborhood of the equilibrium.
and
is an
One also computes
2 2
ClR
aR
W = mgz O (Y 1 +Y2) - Cr(t) (qy 2 +PY 1 ) + Y1 Clq - Y2 ap +
A(p2+q2)1l - yi Due to the boundedness properties of
Y~.
r(t)
q,
and of the
2.
coefficients of
69
R, one recognizes that hypothesis (iv) of
Theorem 2.5 is satisfied.
All the other hypotheses being
verified also, the equilibrium is uniformly asymptotically

stable.
Notice that V.M. Matrosov [1962]1 proves further
the instability of the equilibrium when
zo > O.
2.8.
Exercise (A Lienard type equation).
g(x)
be continuous functions from !Jl ... !Jl
!Jl 3 ... !Jl.
be a continuous function from
x+
Let
fey)
and
and
h(t,x,y)
For the equation
h(t,x,x)x + f(x)g(x)
0,
which is equivalent to the system
y,
-h(t,x,y)y - f(y)g(x),
the origin is uniformly asymptotically stable if

(i)
there exist continuous functions

k 2 (x,y)
kl (x,y)
(ii)
f (y) > 0
(iii)
g (x)x > 0
from
~
!Jl2 ... !Jl
kl(x,y)
and
y ~ 0
such that for
h(t,x,y) ::: k 2 (x,y) > 0;

for every
for every
Y1
g(O) = 0;
and
further, the origin is uniformly globally asymptotically

stable i f
(iv)
Hint:
lim f:g(OdE; =
Ixl"'oo
I~t:oo
and
r
0
d
i(~)
00.
.the auxiliary functions

V(x,y)
and
00
y
= f0
i(~) +
fX
og(E;)dE;
(2.9)
II.
70
dT]
fYo f(T])
can be used successfully in this problem.

is interesting to observe how
has been constructed,
using a separation of variables.

V(x,y)
= G(x)
By the way, it
One tries the form
~(y).
Its time derivative reads

V(x,y) = G'(x)y - ~'(y)[h(t,x,y)y + f(y)g(x)].
In view of simplifying
and
in order that
identically.
V, one tries then to determine

G'(x)y -
vanishes
~'(y)f(y)g(x)
By separation of variables, one gets

G' (x) _
~'(y)
g(x) -
f (y)
and, of course, both members have to be equal to one and the

same constant.
Choosing the constant equal to
1, one obtains
(2. 9)
2.9.
Example of a chemical reactor.
Consider a continuously
agitated tank reactor where a first order irreversible

reaction
A'" B
takes place (cf. R.B. Warden, R. Aris and N.R. Amundson

[1964]).
Let
be the volume of the vessel into which a
stream of reactant
constant concentration
flows at a constant rate

Co
and temperature
is stirred so that the concentration

of the reactant
TO'
q, with
The vessel
and temperature
are constant throughout the volume.
mass balance for reactant
yields the equation
T
A
2.
71
dC
V dt = q(CO-C) - VF(C,T)
where
F(C,T)
is the reaction rate.
(2.10 )
A heat balance gives

(2.11)
where
U* (T)
is the heat removed to the coolant,
heat evolved per mole of consumed

of the reactant and
A,C p
-lIH
the
the specific heat
its density.
Let us introduce the normalized variables

tq/V,
and let us put further
G
= VF/qc O
and
vu* /qc o (-lIH)
But generally,
F = kC exp(- ~)
RT
where
k,E
and
are appropriate constants.
Hence, we
may put
G = ~K (n)
where
K(n) > O.
Equations (2.10) and (2.11) become

~
1 -
nO - n +
~(l+K(n
~K(n)
- U(n)
where the dot represents from now on, the derivation with
respect to
'"(.
Let us study the stability of a constant regime

~
~s'
ns
through the use of the auxiliary function

V(~,n)
with
72
II.
are left undefined at this stage.
P (n)
~he
functions
and
One computes that
V(s,n)
(s-ss) [-1 - K(n) + p(n)K(n)]
+ (s-ss) [ss(K(n s ) - K(n
- pen) (l+K(n
+ p(n){n s - n + ss(K(n) - K(n s
+ U(n s ) - U(n) + P(n)K(n)}
+ q(n)K(n)]
+ P (n) [ss (K (n s ) - K(n + p (n){n s - n + SS (K (n) - K(n s
+ U(n s ) - U(n)}]
+ q(n) [n s - n + ss(K(n) - K(n s
Let us now choose
on
and
+ U(n s ) - U{n)].
such that
does not depend
any more, by putting
p (n)
1 + K(n) > 0
K(n)
,
q (n)
- K{n)
[ss{K{n s ) - K(n
+ p(n){n s - n + ss(K(n) - K(n s
+U{n s ) -U(n)}1.
Then
-q(n)K(n)H(n)
where
It follows from Theorem I.4.2 that the constant regime

is stable if, in some neighborhood of
(i)
is positive definite, i.e.
(ss,n s )'
q' ens) > 0;
3.
Introduction to comparison method
(ii)
73
H' (n S ) > O.
is negative, i.e.
It can be shown that these conditions are nothing but RouthHurwitz conditions for the linearized system.
under Hypothesis (ii), ~
(~,n)-space,
Observe that
is negative semi-definite in the
because it is independent of
~.
Theorem 2.5 shows that, under the same hypothesis,

the stability is asymptotic since
if
=0
if and only
n = n s ' whereas, on the set
the auxiliary function
3.
3.1.
V(~,n)
W(~,n)
=n
- ns
is such that
Introduction to the Comparison Method
In this section, we still consider Equation I(2.2),
with no change of the general hypotheses.
For convenience,
let us rewrite this equation here:
f(t,x).
(3.1 )
Suppose that there exists a scalar differential equation

u = w(t,u)
with a critical point at the origin
stability properties.
(3.2)
u
0, and some known
The so-called comparison method studies
the relationship which should exist between (3.1) and (3.2)

in order that the stability properties of (3.2) entail the
corresponding properties for (3.1).
An
appropriate re-
1ationship is obtained by using a differential inequality
II.
74
due to T. Wazewski [1950], involving an auxiliary function

which, although positive definite, will not be required to
have a total time derivative smaller than or equal to zero.
This is the kind of generalization of the classical stability
and asymptotic stability theorems which we now set out to
study.
3.2.
Preliminary remark.
It is customary to have the second
member of a differential equation defined on some open subset

of the (t,x)-space.
This is to avoid needless intricacies
in the computation of derivatives
x(t).
In this section
however, it will prove helpful to define the second member

00
(t,u)
I x ~+, where
of (3.2) on
~+
[o,~[.
Of course,
the reader may always imagine any continuous extension of

in the region
u < O.
that (3.2) admits of
00
Nevertheless, when it will be supposed

u = 0
as a critical point and that
this point is stable, or asymptotically stable, positive

perturbations only will have to be considered.
The following lemma is due to T. Wazewski [1950].
uniqueness of the solutions of the comparison equation is
assumed here for convenience.
necessary:
But this condition is not
cf. Chapter IX of this book, where the comparison
technique is dealt with over again in a more general setting.

3.3.
Lemma.
Let
00:
I x ~+
~+, (t,u)
w(t,u)
tinuous and such that (3.2) has a unique solution

through any
(to'u O> E I x ~+.
future interval where

v: [to' a [
+ ~
u(t;to'u O)
be such that
Let
be conu(t;to'u o )
[to,a[ be the maximal
is defined.
Let
3.
Introduction to comparison m.ethod
(ii)
then
vet)
vet) ~ w{t,v{t))
u(t)
on
on
75
[to,a[;
[to,a[.
For the proof, see Chapter IX.
3.4.
Theorem (C. Corduneanu [1960]).
a function
and a .sfl
as described in Lemma 3.3, with
function
(i)
(ii)
w(t,O) = 0,
V: I x n ... 9R, such that for some
a E!Ie and every
function
n:
(t,x) E I x
V(t,x) > a(lIxll), V(t,O) = 0,

V(t,x) < w(t,V(t,x));
then
(a)
stability of
of
(b)
u =
x =0;
asymptotic stability of
u =
i f moreover, for some function
(iii)
implies stability
implies equi-
= 0;
b E.Ye and every
(t,x) E I x n:
V(t,x) ~b(IJxll);
then
(c)
stability of
(d)
Proof.
implies uniform
x = 0;
uniform asymptotic stability of
u =
x = 0.
implies
Because of (i), (ii) and the lemma, we have for any
(to'x O) E I x
and
u =
uniform stability 6f
x(t;to'x O)
and any
t > to
where
u(t;to,V(to'x O))
are defined, that

(3.3)
76
II.
(a)
to
E I
u = 0
The origin
u o < 0*
for arbitrary
I IXol I < 0
implies
to' and thus

(b)
=0
Uo
0*.
<
But then
for any
is stable.
Since (3.2) is a scalar equation, the assumed
stability Ccf. I.2.8).
there exists an
implies equi-asymptotic
Therefore, for any given
n*(t o ) > 0
and for any
a a(t o ') > 0, such that

that
= V(to'x O)
o(t o ')
I Ix(t;to'x O) I I < E
one deduces from (3.3) that

~
is defined over
u(t;to'u O)
For continuity reasons, there exists a
such that
such that
t > to' u(t;to'u O) < a ()
and
Notice, in passing, that
o * (to' ) > 0
> 0, there exists a
and
[to'oo[.
being stable, for any
and
E I,
E > 0,
t
>
to + a
imply
But now, again for continuity reasons,
u(t;to'u o ) < aCE).
there exists an
to
n(t O) > 0
such that
I IXol I < n
implies
But then one gets from C3.3) that
that
Xo E Bn
and
t ~ to + a
imply that
x(t;to'x O) E B, and
therefore the origin is equi-attractive.

(c)
To prove thesis (c) , one has only to observe
that in the proof a), 0*

of
and
to
(d)
(b) , n*
and
One merely has to observe that in the proof of
can be chosen independent of
further that the same is true for

3.5.
can be chosen independent
Particular cases.
For
Q.E.D.
n.
w(t,u)
to' and
= 0,
Theorem 3.4(a)
is reduced to Liapunov's Theorem I.4.2 on stability, whereas

3.4(c) yields Persidski's Theorem I.4.3 on uniform stability.
3.
77
Choosing w(t,u) = -c(u) for some function c E~,

we observe that for the equation
= -c(u), the origin
u = 0
is uniformly asymptotically stable.
Because we have
assumed uniqueness for the solutions of (3.2), c

chosen such that for
o
u
>
has to be
0,
du
Jo c(u) =
Hypothesis (ii) reads here
(3.4)
00.
V(t,x) ~ -c(V(t,x.
Therefore,
3.4(b) is reduced to Massera's Theorem 1.6.26, and 3.4(d)

to Liapunov's Theorem 1.6.2.
observe that
To establish the latter point,
V(t,x) ~ -c[a(1 Ixl I)]
and that
is a
Je.
function of class
Because uniqueness has been assumed throughout for

the solutions of differential equations in hand, condition
(3.4) is no restriction whatsoever.
Indeed, if in the above
mentioned theorems of Liapunov and Massera, the function
was chosen such as to let the integral in (3.4) converge,

uniquepess would be violated for the solutions of
x = f(t,x)
(prove this!).
Various ways of choosing the function
w(t,u)
are
illustrated in the following exercises.

3.6.
Exercise.
If
c E Je and if
tinuous, prove that the origin

for the equation
-<P(t)c(u).
r<P(t)dt =
<p: I ....
=0
iii'+
is con-
is uniformly stable
If, moreover
00,
the origin is equi-asymptotically stable.

3.7.
Exercise.
equation
If
A: I ....
is a continuous function, the
U = A(t)U has a critical point at the origin,
II.
78
which is stable, uniformly stable or equi-asymptotically

stable according to
(Vt o E I) (3A > 0) (vt
or
(3A > 0) (Vt o E I) (Vt > to)
A (s)ds < A,
to
or
as
t+ "".
The following example illustrates the second of these three

possibilities.
3.8.
Example.
Consider the differential equation
is a unit
(E sin t + A(t,xx
n x n
(3.5)
where
identity matrix and
A(t,x)
n x n
matrix which is skew symmetric and satisfies
an
appropriate regularity conditions on some domain I x n. The

2
2
2
auxiliary function Vex) = Xl + x 2 + + xn has a time
derivative
V(t,x)
But the solution
. = 2u sin t
stability for
= 2V(x) sin t which is not sign constant.

u = 0 of the scalar comparison equation
is uniformly stable.
x = 0
This entails uniform
and Equation (3.5).
Compare this
result with 1.4.6.

The following exercise makes use of the comparison
method to generalize Chetaev's Theorem 1.6.31 on equi-asymptot:
3.
79
stability, while Exercise 3.10 is an extension of Theorem 3.4

to partial stability.
3.9.
Exercise (N.P. Bhatia and V. Lakshmikantham [1965]).
~l
Suppose there exist two

k: I
that
of
!II
w(t,O)
= 0,
u = w(t,u),
and every
(i)
(ii)
(iii)
functions
V: I
w: I
x Q +
!II+
!II ,
!II such
that uniqueness holds for the solutions
and at last, that for some function
a E
st
(t,x) E I x Q:
V(t,x) ~ a(llxll), V(t,O) = 0;
k(t)V(t,x) + k(t)V(t,x) < w(t,k(t)V(t,x;
k(t) > 0;
i f further
(iv)
k(t)
+ ~
then stability of
of
as
u =
+ ~;
implies equi-asymptotic stability
= 0.
The hypothesis that
no means essential.
k(t)
be a
~l
function is by
But to dispense with it, one should
resort to a Dini derivative, a device to appear in subsequent

chapters only.
If
k (t)
is continuous and
kV E ~l, the
statement above can be reduced to Theorem 1.6.31 of Chetaev,

by putting
3.10.
w(t,u)
= and
replacing
Exercise (C. Corduneanu [1964]).
V(t,x)
by
k(t)V(t,x).
The following statement
pertains to the system of Equation 1(4.1), described in

detail in Section 1.4.4.
Suppose there exist a function
Lemma 3.3, with
V: I
x Q x
!11m
w(t,O) = 0, and a
)(1
w as described in
function
!II, such that for some function
a E!Ie
80
II.
and every
(t,x,y) E I x
(i)
x ~m:
V{t,x,y) > a{llxll), V{t,O,O)
.
V(t,x,y)
(H)
0;
< w(t,V(t,x,y));
then
Ca)
spect to
(b)
stability of
x
of
=0
implies stability with re-
= Y = 0;
u = 0
asymptotic stability with respect to
implies equi-
of
=Y=
provided the solutions of I(4.1) do not approach
0,
in a
finite time;
if moreover, for some function
I
E~
and every
(t,x,y)
x ~m:
(iii)
V(t,x,y) < b( II xII + IIyll);
then
(c)
uniform stability of

(d)
of
u = 0
implies uniform
x = Y = 0;
= 0,
along
with the same assumptions as in (b) for .the existence of

solutions, implies uniform asymptotic stability with respect
to
of
x = Y
= O.
4.
4.1.
Total Stability
All stability-like concepts considered up to this
section pertain to variations of the initial conditions

only.
We introduce here a new type of stability, where
variations of the second member of the equation are also

taken into account.
This is relevant to most practical
4.
Total stability
81
problems, where significant perturbations occur not only at

the initial time, but during the motion.
4.2.
Our general hypotheses remain those of Section 1.2.2.
In particular, we still assume that the differential

equation
x
is sudh that
f(t,O)
= f{t,x)
for all
(4.1)
t E I.
Another differ-
ential equation will have to be considered along with (4.1),

namely
(4.2)
y = f(t,y) + g{t,y)
g: I x 0 + ~n
where
ditions -as
f6r all
satisfies the same regularity con-
f, thus ensuring global existence and uniqueness
solution~
of (4.2).
This function
will play the
role of a perturbation term added to the second member of

(4.1).
Not much will be supposed about it besides these
regularity conditions, and of course some kind of bound.

Peculiarly, it will not be assumed that
g(t,O)
0, and
therefore the origin will not be, in general, a solution of

(4.2).
4.3.
The solution
of (4.1) is called totally stable
(or stable under persistent disturbances) whenever
such that
Ilg(t,x) II
then
< 02'
82
II.
4.4.
Theorem (I.G. Malkin [1944]).
ge, three functions
function
V: I x n ....
constant
M such that, for every
(i)
(ii)
5(1
a, b, c E.5e and a
(t,x) E I
n:
a(llxll) ~ V(t,x) ~ b(llxll)~

V(t,x) ~ -c(1 Ixl I), V
tions of (4 .1)
(iii)
If there exist a
II
computed along the solu-
~~ (t,x) II ~ M~
then the origin is totally stable for (4.1).

Proof.
Let us write
Vp(t,y)
for the time derivative of
along the solutions of the perturbed equation (4.2), i.e.
Vp(t,y)
av
= (ax
(t,y)
f(t,y) + g(t,y
av
at
(t,y).
Choose
1' 0 < 1 < e, such that b (1 ) < ate), and choose

02 < kc(ol)/M for some k, o < k < 1. Observe that
V(t,x) : bC ( 1 ) < a (el

for
a (el
(t, x) E I
(t,x) E I
for
x
aBe
But i f
Vp(t,y) < -c(llyll) +

~
4.5.
-c(ol) + M
1 :
V(t,x) >
and
(to'YO) E I
In fact, for
I x
Bo
Ilg(t,y)11 < 2' there can exist no
if
tl > to
Bo
and
1
such that
Ilyll : e, one gets
(~~ (t,y) I g(t,y

kc(ol)
M
<
o.
Q.E.D.
Theorem (I.G. Malkin [1944] and S. Gorsin [1948]).
is lipschitz ian in
If
on
and if the origin is uniformly asymptotically stable,
then the origin is totally stable.
4.
Total stability
proof.
83
One has only to observe, using Theorem 1.7.4 and its
corollary, that the hypotheses above imply those of

Theorem 4.4.
4.6.
Q.E.D.
Obviously, the hypotheses of Theorem 4.4 do not imply
that any solution

fact, g(t,y)
t
00.
y(t;to'yo)
tends to
as
00:
in
does not vanish, nor does it fade down as
However, some kind of asymptotic property can be
proved, as is apparent from the following theorem.

Theorem (cf. I.G. Malkin [1952]).
In the hypotheses of
Theorem 4.4, (VE > 0) (30 1 > 0) Nn > 0) (302 > 0) (Vt o E I)
if
and i f
!!g(t,y)!! < 02 '
then there is aT> 0
Proof.
For any
such that
E > 0, let
01
(Vt
and
the definition of total stability.

Bo
starting from
remains in
in the appropriate region.
let
01
and
02
<
02
total stability, with

02 < kc(01)/M
o{
< / /y/ / ~ E
T) y(t;to'YO) E Bn.
be chosen as in
02
Thus every solution
BE' providing
Consider an
! !g(t,y) / /
n, 0 < n < 01
<
02
and
be chosen as in the definition of

E
replaced by
for some
k, 0 < k < 1.
and if
/ /g(t,y) / / < 02
n.
Further let
Then, for
on
[to'oo[ x BE' one
gets
av (t,y)
Vp(t,y) < -c(//y//) + (ax
g(t,y
84
II.
Thus, for
Bo'
enters at least once inside
YO E BO ' y(t1t O'YO)

1
But it remains thereafter in
stability.
4.7.
Exercise.
01
4.8.
n'
because of total
Show that in the definition of total

and
02
may be replaced by a single
02
Exercise CW. Hahn [1967]).
V(t,x) ~ b(llxll)
4.9.
Q.E.D.
stability, 01
Prove that, in Theorem 4.4,
can be replaced by:
Exercise (J.L. Massera
[19~6
V(t,O)
and 1958]).
O.
If the origin
is totally stable for a linear differential equation
x = A(t)x,
(A
a continuous matrix of order
n), then it is
uniformly asymptotically stable.

4.10.
Exercise (J.L. Massera [1956 and 1958]).
it is not true that, for an equation

f(O)
= 0,
stability.
Hint:
and such that
5.1.
f (x), f
E~l,
total stability implies uniform asymptotic

consider a function
at a sequence of points
5.
Prove that
xf(x) < 0
xi
for
f: 9f ... 9f, vanishing
approaching zero monotonically,

x
0, x
xi.
The Frequency Method for Stability of Control Systems

In this section, we shall illustrate, using the example
of a nuclear reactor, one of the very few practical methods

for constructing a Liapunov function, namely the so-called
frequency method, due to V.M. Popov [1962].
5.2.
The example of a nuclear reactor.
Standard parameters
to describe the state of a nuclear reactor are the fast
5.
Frequency method for stability of control
v, v > 0, and the temperatures
neutron's density
85
u E ~n
of its various constituents (cf. H.B. Smets [1961]).
The
neutron's density satisfies an equation
v
where the reactivity
= kv
(5.1)
is a linear function of the state
k = ko - rTu - ~v, r E ~n, rT
the transpose of
r, ~ E~.
Using Newton's law for heat transfer, one gets as equation

for the temperatures
u = Au
with
- bv
A a real matrix of order
(5.2)
nand
b E ~n.
The system
of equations (5.1), (5.2) has the equilibrium
koA-lb/(~+rTA-lb),
Uo
Vo
Remembering that
= ko/(~+rTA-lb).
v > 0
and using the transformation of
Vo
= - -ko
v
T -1
(In -- + r A x),
Vo
variables
one obtains, after some computation, the system
x = Ax - b~(a),
e = ~(a),
a = c Tx where
~(a)
= Vo(ea-l),
by the way that
a~(a)
cT
> 0
y~,
= _rTA- l
when
(5.3)
and
~
O.
A system like (5.3) is equivalent to what is known as

an indirect control system.
S. Lefschetz [1965].
On this notion, see e.g.,
We shall not dwell any further here on
II.
86
the technical description of a nuclear reactor, nor on the

derivation of its equation, but rather we shall prove some
sufficient conditions for the global asymptotic stability of
the critical point at the origin for the system (5.3) considered as defining the unknown functions
and
a.
The
following lemma, which we quote without proof, will be used

to build an appropriate Liapunov function.
5.3.
Let
Lemma (V.A. Yacubovich [1962], R.E. Kalman [1963]).

A
be a matrix of order
n, whose eigenvalues have
strictly negative real parts, D a symmetric, positive

definite matrix of order
and let
and
n, let
b E !JIn, b t- 0, k E!JI n
be real scalars,
0,
> O.
Then a
necessary and sufficient condition for the existence of a

symmetric matrix
definite) and a
is that
of order
q E!JI n
(a)
T
A B + BA
(b)
Bb - k = IT q,
(necessarily positive
such that
= -qqT
- ED;
be small enough and that the inequality
be satisfied for all real

5.4.
w.
The following theorem pertains to Equations (5.3).
Remember that
~(a)a >
whenever
Theorem (R.E. Kalman [1963]).
a t- O.
Suppose all eigenvalues of
have strictly negative real parts, y
>
If further, there exist real constants

(i)
(ii)
a.: 0, S .: 0, a + S > 0;
for all real
w:
and
a
and
cTb + y
>
A
O.
S such that
5.
Frequency method for stability of control
87
(5.4)
then the origin
0, a
is globally asymptotically
stable for the system (5.3).

~.
Consider, for some matrix
to be determined later,
the function
V(x,a)
= xTBX
+ a (a-cTx) 2 + S
Ia~(~)d~,
o
whose derivative along the solutions of (5.3) reads
V(x,a)
XT(ATB+BA)X - 2xT(Bb-k)~(a)
- T~2(a) - 2ay~(a)a,
where
This quantity
21
T
SA c + ayc,
should not be mistaken for the
of
Section 5.2.
Now
with
Vex,a)
I Ixl I +
lal.
is positive definite and approaches

Using Lemma 5.3, we compute
from which it follows readily that

definite provided that, for
V(x,a)
is negative
w E ~,
But this inequality is equivalent to (5.4) since
88
II.
Re[(2~y+iwS) (cT(iwE-A)-lb +
+-)
~w
Sy + Re[2~ycT (iwE-A) -~ + ScT (iwE-A+A) (iwE-A) -~)

S (cTb+y) + Re [ (2~ycT+SC?A) (iwE-A) -~]
T
-1_
+ 2Re [k (iwE-A) 0).
The thesis now follows from Corollary 1.6.9.

5.5.
Remark.
Q.E.D.
Of course, Theorem I.6.2 and its Corollary I.6.9
have been established for an equation in normal form
x = f(t,x),
which is not the form of (5.3) considered as
defining the unknown functions
and
cr.
But it can be
proved (e.g., S. Lefschetz [1965]) that (5.3) can be reduced

to the normal form, and therefore, the difficulty mentioned
is not essential.
5.6.
Theorem 5.4 is typical of a class of propositions based
on a frequency response diagram of the linear part of the

control system.
The results obtained by this method are, in
a way, stronger and more effective than those considered in

the preceding sections.
Indeed, they apply to a whole class
of systems, namely those corresponding to any function

with
~(cr)cr
>
when
function explicitly.
cr
0, and they yield the auxiliary
Further, the so-called frequency
criterion is the best possible in the sense that the proposed

Liapunov function proves asymptotic stability if and only if
this criterion is satisfied.
However, the scope of the method
is unquestionably narrower than the one of Liapunov's direct

method as such.
6.
Non-differentiable Liapunov functions
6.
6.1.
89
Non-Differentiable Liapunov Functions
All auxiliary functions introduced up to here were
functions.
~l
But, as the example of a transistorized circuit
will show, it may happen that the "natural" Liapunov function

is not that regular.
Hence, the interest of generalizing the
theorems of Liapunov's second method to encompass the case of

less smooth functions
V.
Using the results of Appendix I,
it is possible to prove anew most theorems of Chapters I and

II, while imposing on the auxiliary functions nothing more
than a local Lipschitz condition with respect to
continuity in
(t,x).
and
These considerations motivate the level
of generality adopted in most succeeding chapters.
As for the
preceding theorems, it is a mere exercise to rewrite them in

this new setting.
For example, Theorems I.4.2 and I.6.2, and
Corollary I.6.9 would read as follows, and their proofs use

mainly Theorems 2.1, 3.3 and 4.3 of Appendix I.
6.2.
Theorem.
V: I x n
+ ~
that, for some

(i)
(ii)
If there exists a continuous function

which is locally lipschitz ian in
a E Yt' and every
V(t,x)
D+V(t,x)
and such
(t,x) E I x n:
a(llxll); V(t,O) = 0;
~
0;
then, the origin is stable.

6.3.
Theorem.
V: I x
+ ~
Suppose there exists a continuous function

that, for some functions

(i)
(ii)
a,b,c EYt' and every
a(llxll) < V(t,x) < b(llxll);

D+V(t,x) < -c (II xii).
and such
(t,x) E I x n:
90
II.
Choosing
t
a > 0
SIMPLE TOPICS IN
S~ABILITY
THEORY
Ba E n, let us put for every
such that
E I
{x E
n:
V(t,x) < a(a)}.
Then
(a)
for any
uniformly in
(b)
6.4.
to'x o
to EI
when
and any
t
l
Xo E v-to,a
x(t;to'x o )
->
+~OO;
Corollary.
The origin is uniformly globally asymptotically
stable if the assumptions of Theorem 6.3 are satisfied for
n=
6.5.
~n
and
a(r)
->
00
as
->
00.
Example of a transistorized network.
In this section
we assume that the reader has some familiarity with the

terminology and symbolism of electrical engineering.
the network shown in Figure 2.2, where
Tl, ,T n
Consider
transistors
are plugged into a linear resistive n-port with
Figure 2.2.
Transistor network
6.
Non-differentiable Liapunov functions
sources.
91
Each transistor will be described by the circuit
diagram of Figure 2.3, which embodies the models of
iI
E o----+--'
VV\r-+--~c
C,{v,)
B
Figure 2.3.
Model of the transistor
H.K. Gummel [1968], generalizing the classical Ebers and

Moll's model.
In Figure 2.3, the subscripts pertain to
The circuit consists of nonlinear resistors

q l f 2 (V 2 ), voltage controlled current sources
rlfl(v l )
and nonlinear junction capacitors
No bulk resistors for the base

(C)
(B), emitter
Plfl(v l )
Tl
and
r l f 2 (v 2 ),
Cl (v l ),C 2 (v 2 )
(E)
or collector
have been considered, since these can be thought of as
embedded in the linear resistive n-port.

equations are as follows:
The corresponding
92
II.
For our purpose, we suppose the following:
(i)
the constants
(E)
the functions
continuous
(iii)
the
Ci :
> o~
f i : (jf ->!if are strictly increasing
functions~
(jf -> (jf
that for some

Let
are
are continuous functions, such
E > 0, Ci(v)
E > 0
for any
v.
i = (i l ,i 2 , ,i 2n ), v = (v l ,v 2 , ,v 2n ), C(v) =
diag[Cl(vl),C2(v2)''C2n(v2n)]'
F(V) = [fl(vl),f2(v2), ,f2n(v2n)].
The transistors equations
then read
i = C(v)v + TF(v)
(6.1)
In a similar way, we obtain the following admittance equation

for the resistive n-port:
i = -Gv + b,
where
is a constant matrix of order
constant 2n-vector.
(6.2)
2n
and
is a
From (6.1) and (6.2), we get the network
equation
C(v)v
-TF(v) - Gv + b.
(6.3)
To investigate the stability of a critical point
(V 10 , ,v 2n 0)' i.e., a point such that
(6.4)
we shall use the auxiliary functions
6.
Non-differentiable Liapunov fUnctions
v.
If ~ c.
V
If
vI
iO
(v. ) dv.
~
I,
93
1 < i < 2n.
v lO ' we get using (6.4)
Observing that at
vI
= v lO '
the left derivative of
VI
equals
minus the right derivative, we obtain
Therefore,
One obtains a similar inequality for

D+Vi(v i ), i
for any
and further
odd or even.
Consider next the auxiliary function

V(v)
where
di
>
O.
We obtain right away, for any

V(v) > e:
and
L d.V. (v.)
1;:i;:2n ~ ~ ~
L d.lv.-vol
1<i<2n ~ ~ ~
v,
II.
94
l~i~2n
(d.G .. I dIGI) Iv.l. - vl.ol.

l. l.l.
l<j<2n J Jl.
J~I
In order to state a condition for
D+V(V)
to be
negative definite, the following definition will be useful:

a matrix
(a ij )
of order
will be said to be column sum
dominant if
a ii -
l<j<n
la .. I > 0
Jl.
1 < i
<
n.
J~I
Then, of course, D+V(V)
is negative definite if
D(T+G), where
D = diag(d i ), is column sum dominant, and using Corollary 6.4,

we obtain the following result.
6.6.
v
Theorem (cf. I.W. Sandberg [1969]).
= Vo
The rest solution
of the network equation (6.3) is globally asymptotically
stable if there exists a positive diagonal matrix

which makes
6.7.
D(T+G)
Exercise.
column sum dominant.
Consider a network without sources
and with transistors such that

Suppose further that for any
that the solution
if
v = 0
f 2i _ l (v)
i: Pi > r i
= f 2i (v),
and
(b = 0)
f 2i (0) = O.
qi > rio
Prove
is globally asymptotically stable
is a positive definite matrix.
function
diag(d i )
Hint:
use as Liapunov
7.
7.
95
Theorem 1.3 and Corollary 1.5 prove asymptotic stability

or similar properties by using the fact that the origin is
the only complete positive semi-trajectory contained in the
set of points where
V(t,x)
O.
This idea, applied to
autonomous systems, appears in E.A. Barbashin and N.N. Krasovski

[1952] for stability and in N.N. Krasovski [1956] for instability.
The extension to periodic systems comes from
N.N. Krasovski [1959].
More bibliography can be found in this
last reference.
Theorem 2.5 is substantially due to V.M. Matrosov [1962],
who gave also several variants of the same proposition, along
with some instability results.
type, but where
Another theorem of the same
W is a vector function (N. Rouche [196B])
will be introduced in Chapter VII, followed by several more

recent results.
The comparison method, a detailed account of which will
be the object of Chapter IX, has been used by R. Conti [1956]
to prove a continuability result.
C. Corduneanu applied it
to stability problems [1960], as well as to partial stability

[1964].
As for total stability, if the second member of the
differential equation is viewed as belonging to some appropriate
function space, the perturbations
are estimated in the
~opology
g(t,x)
in Definition 4.3
of uniform convergence.
Various
extensions of the concept of total stability have been

proposed and studied.
They correspond to variations of the
topology of the space of second members.
For instance,
N.N. Krasovski [1959] studied stability under persistent
96
II.
disturbances that are bounded in the mean, and I. Vrko~ [1959],

cited by W. Hahn [1967] introduced integral stability.
Concerning the frequency method, a subject matter which
will not be considered any further here, the relevant
references have been given in the text.
For more details,
see e.g. V.M. Popov [1966], or A. Halanay [1963].

Theorem 6.6 on the transistorized network is a slight
generalization of the one given by I.W. Sandberg [1969], for
whom the sufficient condition of global asymptotic stability
was the existence of a positive diagonal matrix
making
diagldi)G
dominant.
and
(diag di)T
diag(d i )
simultaneously column sum
On an analogous problem of circuit theory, see
D. Mitra and C. So. Hing [1972].
CHAPTER III
STABILITY OF A MECHANICAL EQUILIBRIUM
1.
Introduction
This chapter is devoted to stability questions concerning mechanical equilibria, or in other words to stability
of critical points for differential equations having the
Lagrangian or Hamiltonian form, with or without friction
forces.
The central theorem in this context was stated by
J.L. Lagrange [1788]:
roughly speaking, it asserts that a
mechanical equilibrium of a conservative system is stable at

each point where the potential function is strictly minimum.
Lagrange himself (as well as S.D. Poisson [1838]) failed to
give a proof for a potential function more general than a
quadratic form.
G. Lejeune-Dirichlet [1846] gave an elegant
general proof which yielded the model for the entire direct
method of Liapunov.
In this chapter, we first study Lagrange-Dirich1et's
theorem along with some of its variants:
it concerns stability
98
III.
as such or some kind of partial stability.
Next, we consider
the inversion of this theorem, a classical incompletely solved

problem of theoretical mechanics.
Suppose that, at some
equilibrium position, the potential function has no strict

minimum.
One knows of some cases where stability then still
prevails (cf. Theorem 2.6).
The question is:
conditions would enforce instability?
what additional
Several answers will be
proposed, first using an auxiliary function and then by considering the first approximation.
But to find the mildest
possible conditions remains an open problem, and will be left

to the reader as a nice pastime for winter evenings.
The last
section of the chapter is devoted to asymptotic stability or

instability of a mechanical equilibrium in the presence of
complete dissipation.
2.
2.1.
The Lagrange-Dirichlet Theorem and Its variants
We consider first a conservative Lagrangian system as
described in Section 13 of Appendix II, but without assuming

T
to be positive definite with respect to
q.
In this case,
the Lagrangian equations of motion cannot be solved for the
q,
generalized acceleration vector

reducible to the normal form
and thus they are not
f(t,x).
That is why,
rigorously speaking, no theorem of Chapters I or II can be

applied to study the stability of the equilibrium
=q
= O.
However, it is possible to prove, by a direct line of reasoning from the integral of energy, that if the potential function
IT(q)
has a strict minimum at
= 0,
one gets what might be
called partial stability of the equilibrium with respect to q.

This is the content of Dirichlet's original contribution.
2.
Lagrange-Dirichlet t'heorem and its variants
99
Modern versions of this theorem actually assume a bit more,

namely positive definiteness of
T, and prove a bit more,
namely stability with respect to
and
q.
In this last
context it is substantially equivalent and more convenient

to consider the Hamiltonian equations, which assume de facto
The equilibrium is then considered at the
the normal form.
origin of phase space; i.e., the space of

cerning stability with respect to
and
p.
Con-
as well as some further
historical discussion, we refer the reader to N. Rouche and

K. Peiffer [1967], where one will also find an extension of
the Lagrange-Dirichlet theorem to mechanical systems with
time-dependent constraints.
2.2.
We now consider the
Section 14 of Appendix II.
system described in
H~iltonian
Let us, for convenience, transcribe
the equations of motion:

q
ilH
ilp (q,p),
ilH
P = - ilq (q,p)
(2.1)
where
H(q,p) = T(q,p) + IT(q).
The general hypotheses considered in the Appendix'will remain
in force up to Section 5 included.
2.3.
where
General Hypotheses.
II(O)
2.4.
=~
Sfl , and
The potential energy
IT (q)
and
Theorem.
is stable.
The kinetic energy
is symmetric and
B (q)
definite.
We recall them briefly.
.ill
ilq
If
B (0)
is
1 p TB(q)p
T(q,p) = 2
is positive
5(1, and further
(0) = O.
II(q) > 0
for
0, the origin
q = p
100
III.
This is an immediate consequence of Theorem I.4.2,

when observing that
is an integral of the equations of
motion and is positive definite on some neighborhood of the

origin.
2.5.
As observed first by P. pain1eve [1904] and then by
A. Wintner [1941], Theorem 2.4 is not the best possible:
the
hypothesis can be weakened, as is shown by the following

theorem.
In a way, the improvement is academic, but it will
throw some light on the inversion problem, to be dealt with

in further sections.
2.6.
Theorem.
Suppose that for any

~ C ~
there exists an open set

for every
q E
>
such that
IT(q) > 0: then the origin
o~:
B C 11,
with
C B
n
is
q = p = 0
stable.
Proof.
As was shown in Appendix II, if
appropriately, there exists a function

T(q,p) ~ a(1 Ipl I).
and choose
>
IT(O)
a E
has been chosen

~
such that
Let then
e > 0
such that
B C 11, and for every

n
q E Bn: IIT(q) I < a(e)/2.

tinuous and
11
= o.
be given arbitrarily
This is possible since
IT
Further, choose an open set
is con~
according to the hypothesis of the theorem, and select
C B
n
>
such that
~ <
inf{IT (q): q E
a~}
and, therefore that

~
Finally, choose
~ > 0
<
a(e)/2.
such that for every
qo' I Iqol
<
2.
101
Lagrange-Dirichlet theorem and its variants
Then, (qo'PO)
firstly that
I Iq(t) I I < n < e
IT(q(t
a(llp(t)ll)
H(q(t),p(t
and secondly that
2.7.
being the initial conditions, one gets
I Ip(t) I I < e
for every
to' because
for every
t ~ to' because
- IT(q(t < aCe). Q.E.D.
In the hypotheses of Theorem 2.6 and with
Exercise.
II q (t) II < e
chosen as in the proof, show that
IS
and
tE9/ (replacing "for every
for every
II p (t) II < e
= H(qO'PO) < ~,
= H(qO'PO)
T(q(t),p(t
t > to")
Stability is thus proved for past as well as future.

2.8.
Exercise (A. Wintner [1941]).
one degree of freedom
Consider a system with
(q E 9/), such that

q
and
2.9.
ncO)
= O.
Then the origin
t-
=p =0
is stable.
Let us now show that for a mechanical system with one
degree of freedom
(n
= 1),
the sufficient stability condition
given by Theorem 2.6 is also necessary.
Observe first that if
q E 9/, negating the hypothesis of Theorem 2.6 amounts to

asserting that, for some
n > 0, with
n(q) < 0
q E [O,n]
either for any
Both cases can be treated alike.
[-n,n] C fl, one has
or for any
q E [-n,O].
Therefore, it will suffice
to prove the following theorem.

2.10.
Theorem.
(i)
(ii)
(iii)
Suppose that
q E 9/;
B(O) > 0;
(3n > 0,
[O,n] C fl) (vq E [O,n])
n (q) < 0;
102
STABILITY OF A r-1ECHANICAL EQUILIBRIUM
III.
then the origin
Proof.
Suppose
for some constant
is unstable.
has been chosen small enough in order that,

a > 0
and every
q E
choose initial conditions

arbitrarily small.
q(O)
O.
0 < q(t) < n:

p(t)
and further, since
2.11.
to
Po >
O~PO
(2.2)
q(t)
for
But one deduces from (2.2) and (iii)
that as long as
and
B(O)pO > 0, and therefore
q(t)
Then
+ TI(q(t
some time after
Therefore, q(t)
0, qo
a < B(q).
The integral of energy reads
B(q(tP(~)
Observe that
to
n:
q{t)
B(O)
> B(q(t
PO'
= B(q(tp(t),
that
q(t) ~ apO'
becomes never smaller than zero, and
aPot, which proves instability.
Q.E.D.
Unfortunately, this line of reasoning cannot be extended
~n
n > 1, except if one assumes the very strong
for
condition that
TI(q) < 0
its end points at
on some line segment having one of

along with some further hypotheses.
A particular case of this situation is proposed in the following exercise.

2.12.
and
Exercise.
TI(q)
T(q,p)
For
(ql,q2)
if and only if
q2
E~, suppose
= O.
TI(q) > 0
Find conditions on
to ensure the instability of the origin (cf. G. Hamel
[1903], L. Silla [1908]).

Hint:
try a solution
q(t) = (ql(t),O).
2.13.
Let us now show, by way of a counterexample (cf. M. Laloy
[1975]), that the hypothesis of Theorem 2.6 is not a necessary
2.
Lagrange"-Oirichlet theorem and i.ts variants
condition of stability, as soon as
the equations
n > 2.
103
Indeed, consider
ql
PI'
PI
an (q) ,
- aql
(2.3)
q2
P2'
P2
an (q) ,
- aq2
(2.4)
where
n (q)
for
0, and
where
= q2
ql
ncO)
O.
= O.
One has
n(q) < 0
at every point
However, the origin is stable, as is
seen by applying Theorem 2.6 separately to (2.3) and to (2.4).

Observe that
2.14.
is a
~oo
function.
So we know now that negating the hypothesis of
Theorem 2.6 does not entail instability.
It will be found
interesting to discuss how far we can go, in these conditions,

i.e., by negating the hypothesis in question, on the way
towards proving instability.
The present sUbsection utilizes
some notions to be introduced in Chapter V only:
thus, it
will be wise not to include it in a first reading of the book.

The opposite of the hypothesis of Theorem 2.6 reads:
(3n > O,Bn
The set
em ('9'
open set 'l',0 E'I' C Bn) (3q E a'l')
A = {q E !Jfn: n(q) ::: 0
and contains the origin.

of
that
Let
This
Let
or
AO
n(q) < O.
Ilqll ~ n}
(2.5)
is closed
be the connected component
A which contains the origin, and suppose ab absurdo

AO C Bn'
e: > 0
'I'
Then
AO
be such that
is compact and disjoint from

'l'
= {q:
is an open subset of
A\AO'
d(AO,q) < e:} C C(A\A O) C B

n
and
for
every
q
E
a'l':
Bn'
n (q) > 0, which contradicts (2.5)
III.
104
Consider now the

G = {(t,q,p) E!if x
~et
x~: t E!if, Ilqll +
IIpil
<
o}.
T(q,p) +IT(q)::
n,
(2.6)
Due to the existence of the energy integral, any solution

starting in
G cannot come out of it except by violating
the condition
I Iql I + I Ipi
clear that the origin
If (2.5) is verified, it is
n.
<
is, for any
(0,0)
(q,p)
t, a cluster
point for
G(t)
Therefore, G
V.2.3.
{(q,p): (t,q,p) E G}.
is an absolute sector in the sense defined at
We have proved even more:
this sector possesses a
connected component extending to the subset of !if x n x !ifn

where
I Iq I I + I I p I I = n.
Owing to the existence of the counterexample 2.13, the
absolute sector
G need not to be an expeller.
This is
illustrated by the following exercise.
2.15.
Exercise (M. Laloy [1975]).
q E!if and
Show that, for
p E!if, if the equations of motion are
h
were
= -q 8
TIC q )
p,
or
2 1
s~n
is unstable, whereas the set

expeller.
2.16.
Notice that
Conjecture.
necessary if
2.17.
TI{q)
TI
(lIT
- (lq (q),
q
and
TI(O)
= 0,
the origin
G defined in (2.6) is not an
is a
)(2
function
The hypothesis of Theorem 2.6 becomes

is analytic?
So, except for
= 1,
we have no condition which
would be necessary and sufficient for stability.
In the next
3.
Inversion of the Lagrange-Dirichlet theorem
105
two sections, we shall try to find some sufficient conditions

for instability.
This question is improperly known as the
problem of "inversion of the Lagrange-Dirichlet Theorem".

Before proceeding to this part of our study, let us propose
another exercise which yields a useful complement to
Theorem 2.10.
2.18.
Exercise.
qi
Consider the system
Pi'
Pi
where
(i)
~2
1I (0)
< n
function such that,

0
Ben
n
(ii)
-i
1 <
aqi (q)
is a
1I
all
and
-0
lI(q) <
and a1l
for some
with
> 0
q E B
n'
II (q) = -U(r), where
r = 2"
l~i~n
qi'
Then the origin is unstable.
3.
3.1.
Inversion of the Lagrange-Dirichlet Theorem Using

Auxiliary Functions
We come back now to a Hamiltonian system with
of freedom.
degrees
The following theorem is typical of a class of
inverse Lagrange-Dirichlet theorems to be illustrated in the

present section.
3.2.
(with
Theorem (N.G. Chetaev [1952]).
Ben)
g
(i)
(ii)
(iii)
If there exists
such that
=
0 E
{q EB g : lI(q) < O}
q,~
a6~
alII q)
(aq
< 0
for every
q E
6~
> 0
106
III.
then, the origin
Proof.
STABILITY OF A HECHANICAL EQUILIBRIUTVI
is unstable.
We first define the set
'I! = {(q,p): q E 9,
Clearly
Ilpll <
a'l!.
(0,0) E
that, for every
H(q,p) < 0,
The function
(q,p)
(a)
0 < V(q,p):
(b)
(qlp) > OJ.
-(qlp)H
is such
'I!
aH + (ap
aHI p)]H.
-[-(q I aq)
Using Euler's theorem on homogeneous functions, we obtain that

V
= -[2T
(aTI
aq q) -
(lKl
aq q)]H
As it changes nothing to lower the value of
, we choose
small enough to get
= 2T
S(q,p)
Such an
to
and
V(q,p) > 0
(aTI q) > 0
aq
exists since
S(O,p)
S(q,p)
= pTB(O)p
for every
when
(q,p) E
I Iql
and
w.
p i
O.
is quadratic with respect

But then
Instability is thus proved
through Theorem I.S.1 and Remark I.S.S.

3.3.
<
Exercise (N.G. Chetaev [1952]).
Q.E.D.
Prove the following
generalization of Theorem 3.2.

Let assumptions (i) and (ii) of Theorem 3.2 be satisfied.
Suppose further there exists a
on
into
(i)
(ii)
~n
~1
f(q)
such that:
f(O) = 0;
the matrix
is
positive definite:
(iii)
function
(11I1
aq
f) < 0
for every
q E 9:
3.
Inversion of the Lagrange-Dirichlet theorem
then the origin

Hint:
3.4.
q = p
use the function
107
is unstable.
= -(f,p)H.
Theorems 3.2 and 3.3 are only partial inverses of the
Lagrange-Dirichlet theorem.
Indeed, they do not cover the
cases of unstable systems with
n(q) > 0:
compare with
Theorem 2.10.
3.5.
Exercise (L.N. Avdonin [1971]).
Prove that assumption
(iii) of Theorem 3.2 can be replaced by
(~nl
aq q) ~ 0
(iii-a)
Hint:
3.6.
for every
E e.
(iii-a) implies (iii).

Corollary (N.G. Chetaev [1952]).
If, in Theorem 3.2,
assumption (iii) is replaced by

(iii-b)
is analytic, n(q)
2<i<oo
n. (q)
~
where
n~ (q)
~
is a homogeneous form of degree i, and for some
k > 2

Proof.
~n
The proof follows from Theorem 3.2, since
(.-.-1
aq
q)
=
+
an.
(-~ 1 q)
2<i<oo aq
k+1::i<oo
kn -
2<i<k-1
(i-k)n.
and it is apparent that this quantity is

3.7.
(k-i ) n.
~
Corollary (A.M. Liapunov [1897]).
assumption (iii) can be replaced by:
< 0
for
q E
In Theorem 3.2,
e.
Q.E.D.
108
III.
(iii-c)
TI(q)
ST~BILITY
= ITm(q)
+ R(q)
OF A HECHANICAL EQUILIBRIUM
where
TIm(q)
is a negative
definite, homogeneous form of degree

R(q)
and
is such that
11~~II/llqllm-1
->-
as
q ... 0.

3.8.
Exercise.
a plane.
Let
Oxy
be orthogonal coordinate axes in
Suppose a material particle of mass
to each of the four points

d = (0,-1)
(1,0), b
is connected
(0,1), c
(-1,0),
by means of four identical springs of constant
and whose natura1.1ength (i.e., their length in the unstretched state) is equal to
is an equilibrium.
ing to whether
1 < 2
or
1
1 > 2.
In other words, stability
is smaller than the diagonal of the
abed, and instability when
4.
Obviously, the origin
Show that it is stable or unstable accora-
prevails as long as
square
1 > 0.
is larger.
Inversion of the Lagrange-Dirichlet Theorem Using

the First Approximation
4.1.
The Hamiltonian system studied here is the same as in
the preceding sections.
Remember, in particular, that
B(O)
IT 2 (q)
4.2.
+ R(q), where
TI 2 {q)
I Iql I
->-
as
->-
0.
Suppose
is a quadratic form and

Then, if
TI 2 {q)
IT{q)
I laR/aql I /
takes strictly negative
values, the origin is unstable.

Proof.
Let
IT 2 (q)
= 21
q Cq
with
Hamilton's equations (2.1) read
symmetric.
Then
4.
Another inversion of the Lagrange-Dirichlet theorem
B(O)p + gl(q,P), p
109
-Cq + g2(q,P),
where
(B(q) - B(O))p
gl (q,p)
and
g2(q,P)
- aq [R(q) +
21
p B(q)p]
are such that
II
(gl (q,p) ,g2 (q,p))

IICq,p) II
II
... 0
as
II (q,p) II ... o.
From Theorem 1.5.8, the origin is unstable whenever the matrix
has at least one eigenvalue with strictly positive real part.

The corresponding characteristic equation is
det [
-;I.E
B (0)
-C
o.
-;I.E
For non-zero eigenvalues, it reads successively
det
r"
-e 1 [ AE
-;I.E
B (0)
det
[-A 2E- eB(O)

0
det(A 2E + CB(O)).
B(O)
A-:.j
-A-1e]
-E
(4.1 )
III.
110
Since
B(O)
is symmetric positive definite, there exists a
regular matrix
Q such that
= QTQ.
B(O)
But then the equa-
tion may be transformed thus:
= det(A 2E+CB(0
Putting
= QT-q,
one gets that
takes negative values, QCQT
II2
= '21
-T
T-
As
q QCQ q.
has a strictly negative eigenvalue.
Therefore, (4.1) has a strictly positive solution.

4.3.
This theorem includes Corollary 3.7 when
case
TI2
Q.E.D.
m = 2; in this
is negative definite.
stable systems such that
But it does not include un1 T

II 2 (q) = '2 q Cq > 0. In order to
cope with such systems, we introduce the following lemma, that

describes
4.4.
IT(q)
Lemma.
when
Suppose
has a single zero eigenvalue.
II (q)
is analytic and
!....!!.
aq2
(0)
is a
positive semi-definite matrix with a single zero eigenvalue.

Then, using appropriate coordinates, TI can be written as
(4.2)
where
q = (q2, ,qn)' B(ql,q)
and
S(ql)
are analytic,
B(O,O)
is positive definite and
m > 3.
Proof.
It follows easily from the assumptions that, using
appropriate coordinates
II (q)
where
BO
= 2"1
-T
q BOq + f (ql ,q)
is a positive definite constant matrix and

3
O(llqll ).
For each
ql' IT
is minimum for
such that
(4.3)
4.
Another inversion of the Lagrange-Dirichlet theorem
o.
III
(4.4)
Since
the implicit function theorem applies and there exists, in some

neighborhood of the origin
q = y(ql)
0, a unique analytic solution
of (4.4), such that
yeO)
y' (0)
O.
(O, 0)
Further
so that
(4.5)
Next, consider the change of variables
The potential function
IT
can be written
0<6<1.
Using (4.4) with

IT
y(r l ), we obtain
= 2'I-T
r {BO
a f
+ --2
- (l).,y(r l ) + 6r)) r
aq
which is the desired form of
IT
if we set
112
III.
Using (4.3) and (4.5), one gets that

Q.E.D.
4.5.
Theorem (W.T. Koiter [1965]).
If there exists
E > 0
such that
(i)
= {q E BE: n(q) < O}
(H)
E as;
(iii)
is analytic and
l;
.L1!.
(0)
aq2
is a positive semi-
definite matrix with a single zero eigenvalue;

then the origin
Proof.
is unstable.
From Theorem 3.2 it is sufficient to prove that
(E1!.1
(lq q) <
for every
q E
e.
On the other hand, from Lemma 4.4, we can choose

(4.2)
Since
origin and
B (ql'Q)
an I q)
o
as in
takes negative values arbitrarily near the
large enough to get

(~q
n(q)
is positive definite, we can choose

S (0) t
o.
- +.!.
= -T
q [B(ql,q)
2
2n
+ m-2
-2-
Further
~
L
l<i<n
aB
~q
0
"(
m + .!.
.., ql ) ql
2 -'l'q [
-(ql,q)q;]q
l<i<n
(lB-a
qi
(ql,q
-) qi ]q
5.
Mechanical equilibrium with dissipation
But for
_1_
m- 2
q E
-'I'q (
and
aB
L
-"l~i~n aqi
113
small enough
(ql ' -)
)q qi q
<
and
Hence
(~~ I q)
5.
5.1.
< 2II < 0
for
q E
e.
Q.E.D.

of Dissipative Forces
Up to this point, we have considered conservative
mechanical systems only.

forces are present?
What happens when some dissipative
This question is examined below.
Roughly
speaking, complete dissipation ensures asymptotic stability of

the origin of the (q,q)-space if
instability if
II
II
has no minimum.
is minimum there, and

However, the proof requires
that the minimum be an isolated equilibrium.
This is expressed
in the following theorem, generalizing and making rigorous

the considerations of Kelvin (see W. Thomson and P.G. Tait
[1912]) on the subject.
The mechanical system referred to
here is the Lagrangian system described in Section 13 of

Appendix II, but with some dissipative forces added to it:
vector.
5.2.
(q,q), defined and
~l
on
Theorem (L. Salvadori [1966]).

(i)
II
has a minimum at
n
If
q = 0;
!JIn
114
III.
(ii)
the equilibrium at
(iii)
=0
is
isolated~
the dissipation is complete, i.e. for some
then the equilibrium
(Q I q)
a E Je:
function
q
-a ( II q II)
is asymptotically stable.
If Hypothesis (i) is replaced by

(i-a)
TI
has no minimum at

Proof.
=q=
q = 0;
is unstable.
We consider the total energy as the auxiliary function.
It reads
= T(q,q)
V(q,q)
+ TI(q).
and due to (iii), V
(Qlq)
One knows that
if and only if
V(q,q)
o.
With
a view to applying Theorem 11.1.3, one should first show that

the set
= {(q,q) E
equilibrium
Bp: q
= O}
contains, besides the
0, no other positive semi-trajectory.
But, by (ii), it contains no other static solution whereas any

positive semi-trajectory contained in
Let us now show that

One knows that
V(q,q)
is a static solution
is positive definite with respect to
q.
Hypotheses (i) and (ii) together imply that the minimum of

is a strict minimum.
q
=0
where
Indeed, there is a neighborhood of
TI(q) > O.
If the minimum were not strict,
there would exist, in every neighborhood of

q
where
a minimum of
TI
TI
would equal zero.
0, a point
But such a point would be
IT, i.e. a static solution, and (ii) would be
violated.
If, on the other hand, IT
has no minimum at
there exists, in every neighborhood of

with
IT (qO) < O.
0, a point
But then of course, V(qo'O) < O.
= 0,
qo
Therefore,
6.
Mechanical equilibrium under gyroscopic forces
115
all the hypotheses of Theorem II.l.3 are verified for

and Theorem 5.2 is proved.
Bp
Q.E.D.
On an extension of this theorem to some non-autonomous

mechanical systems, cf. VIII 3.5.
5.3.
Remark.
Notice that we can state here a necessary and
sufficient condition:
isolated equilibrium
has a minimum at
if there is complete dissipation, the

q
=q=
O.
is stable if and only if
IT
Further, the stability is always
asymptotic in this case.

5.4.
Exercise.
Show that the condition
IT(q)
along
with complete dissipation do not entail stability.

Hint:
5.5.
consider, for
Exercise.
ge, the equation q + q3
Hypothesis (ii) of
can be weakened as follows:

there is no equilibrium in
stability still prevails.
Theore~
if, for some

Be;
e; >
0,
Be;
c n,
{q: IT(q) < a}, then in-
This is a significant extension of

2
qE9t':
3 3
= qlq2'
6.
6.1.
O.
5.2 for instability
the theorem, as is shown by the simple example for

IT (ql,q2)

of Gyroscopic Forces
Statement of the problem.
Consider a conservative system
as described in Section 13 of Appendix II. The kinetic energy

l'T
assumes the form T = 2

q A(q)q.
We suppose that A(q) is
defined and
5(2
on some open neighborhood of the origin of
q-space, and that

function
IT(q)
A(O)
is defined and also
5(2
The potential
on the same set
116
III.
and is such that
an
o.
aq (0)
n(O)
The equations of motion
read
(6.1)
One knows that, if
at
has no minimum at
may be unstable.
q = 0, the equilibrium
The present section is devoted
to studying the influence on this equilibrium of the adjunction

of appropriate forces in the second member of (6.1).
Except
at the beginning of our study, these forces will be gyroscopic,

or at least non-energic forces.
The corresponding problem,
which is practically important, is known under the name of

"gyroscopic stabilization".
6.2.
Suppose the forces
Q(q,q)
to be added in the second
member of (6.1) are of the form

Q{q,q)
where
(6.2)
is a constant matrix (not necessarily antisymmetric
for the time being) and

P{q,q)
II
(q,g)
II
Suppose, moreover, that
I I (q,q) I I
as
n(q)
O.
is a function of class
5f3
and therefore assumes the form
21 q TCq + R(q}
n(q)
where
matrix.
ij/qiT ~
~
as
~
~
0 , and
is a constant symmetric
This last hypothesis will remain in force until
Section 6.8.
The equations of motion, which read
6.
117
can be reduced to the following form
(6.3)
for some appropriate function
F(q,S!>
II(q,q)11
-+
as
such that
II (q,q) II
-+
o.
Equation (6.3) is the normal form for the differential equation of second order
A(O)q + Gq + Cq + F(q,q)
= O.
The eigenvalues corresponding to its linear part are the roots

of the polynomial
~(A)
= det(A 2 A(0)+AG+C).
(6.4)
A first instability result is immediate.

6.3.
Theorem.
In the general hypotheses above, if
has
no vanishing eigenvalue and has an odd number of strictly

negative eigenvalues, the origin is unstable for
Proof.
to
Since
as
det C < 0, one has

tends to
00.
strictly positive real root.

to Theorem I.S.S.
~(O)
Therefore,
< O.
~(A)
But
(6.3).
~(A)
tends
has at least one
Hence, the instability, owing
Q.E.D.
This result, already mentioned by Kelvin for the case
III.
118
G, appears also in N.G. Chetaev
of an antisymmetric matrix
[1955]
6.4.
Exercise.
Let
A(O)
=E
and
be diagonal with an
even number 2s of vanishing diagonal elements and an odd

number of strictly negative diagonal elements.
Suppose, with-
out loss of generality, that the first 2s diagonal elements

are equal to zero.
Then, if the upper left principal diagonal
minor of order 2s of
(6.3).
is
> 0, the origin is unstable for
For an antisymmetric
G, this result is due to
Van Chzhao-Lin [1963].

6.5.
The degree of instability.
The characteristic poly-
nomial of the linear part of (6.1) (i.e., the equations of

motion considered before the adjunction of gyroscopic terms)
reads
det(;>..2A(0) + C).
But a classical result shows (see
e.g. R. Bellman [1970]) that there exists a regular matrix

such that
pTA(O)P = E, while
pTCP
is diagonal.
Of course,
Therefore, the eigenvalues of the linear part of (6.1) are

the roots of
det(;>..2 E+ p TCP )
O.
Now, according to Sylvester's
theorem, the number of strictly negative diagonal elements of

pTcp
equals the number of strictly negative eigenvalues of
C.
The latter number equals, therefore, the number of eigenvalues

with strictly positive real part for the linear part of
(6.1).
This number is called the degree of instabilitv
of the system.
C
Theorem 6.3 may be reformulated thus:
has no vanishing eigenvalue and if the degree of in-
stability is odd, the equilibrium
=q=
cannot be
if
6.
119
stabilized by the adjunction of forces of type (6.2).

6.6.
Let us now prove an auxiliary result which will determine
to a large extent the rest of our discussion, namely that if

G is antisymmetric, 6(A) = 6(-A).
Indeed, one gets
successively
= det(A 2A(O)
= det(A 2A{O)
6{A)
+ AG + C)
det[(A 2A{O) + AG + C)T)
- AG + C)
6 (-A)
It follows from this property that
a(A)
has no root with
strictly positive real part if and only if all the roots of

6(A)
are pure imaginary.

The following theorem will give us a key to gyroscopic
stabilization.
6.7.
Theorem.
2s > 0
Assume the matrix
possesses an even number
of strictly negative eigenvalues, all the other
eigenvalues being strictly positive.

antisyrnrnetric matrix
Then there exists an
G such that every root of
6(A)
is
pure imaginary and non-vanishing.

Proof.
Let
be the matrix introduced in Section 6.S and
which is such that

Putting
pTA(O)P =
and
C*
= 'pTCp
G* = pT GP , we observe that the roots of
is diagonal.
6 (A)
are
also those of
Naturally, if, G is antisyrnrnetric, so is

diagonal elements of
c*
G*.
Further, the
are the eigenvalues of
C and
there is no loss of generality to assume that the first

are the negative ones.
Let us now construct
G*
in the
2s
120
III.
*
g2i,2i-l
1 < i < s, and all the other elements of
following way:
choose
for
G*
equal to zero.
Then
f:,. * (X)
2 *2
+ X g2k,2k-l]
One verifies readily that if

for every
g;k,2k-l
k, 1 ~ k ~ s, then
imaginary roots.
IT
2s+l<i<n
f:,.*(X)
2 *
(X +C.).
1.
is chosen large enough

has nothing but pure,
Q.E.D.
Suppose now the equations (6.1) are linear and the
6.B.
matrix
satisfies the hypotheses of Theorem 6.7.
It fol-
lows from this theorem that the eigenvalues of the system

can be transformed into pure imaginary eigenvalues by the
adjunction of appropriate gyroscopic forces.
Now a linear
system of differential equations with constant coefficients,

whose matrix can be diagonalized and whose eigenvalues have
real parts smaller than or equal to zero has a
equilibrium at the origin.
st~ble
(It may be an interesting exercise
to prove this, and to prove that the matrix of (6.1) can be

diagonalized.)
Therefore, gyroscopic stabilization is possible
for a system of this form.
But it is a significant fact that
all we can obtain in this way is stability, not asymptotic

stability.
The appearance of nonlinear terms in (6.1) can
destroy the stabilizing effect of the gyroscopic forces.
The
example to follow is an illustration of this observation.

6.9.
A glance at the restricted problem of three bodies.
When the nonlinear terms
in (6.3) depend on
only, it
6.
121
may be very difficult to determine whether the equilibrium

at the origin is or is not stable.
equilibrium point known as
L4
This is the case for the
in the restricted problem of
three bodies (and also, by reason of symmetry, for the

point
LS )'
The relevant equations, as given e.g. by
H. Leipholz [1968], read
. [20 -20]q' + Cq + F(q)
q +
is the hessian matrix, computed
where
at
(6.5)
0, of the potential function
IT(q), whereas
!R
- Cq contains but terms of order at least 2.
Clq
IT(q), through the change of variables
y
= q2
t 13,
where
= ql
F(q)
The function
+ 2 - p,
is some strictly positive parameter,
becomes
IT (x,y)
The matrix
C can be written explicitly as
313
- 4'
- -2-
313 1
- -2- (2 - p)
(2 -
p)
4'
It is readily verified that if
<
p(l-p)
<
(6.6)
27 '
the origin is stable for the system (6.5) where
has been
122
III.
replaced by zero.
V.I. Arnold [1961] and A.M. Leontovich
[1962] have shown that stability still prevails for (6.5) and
p
belonging to the interval (6.6), with the possible
exception of a denumerable number of values.
Later A. Deprit
and A. Deprit-Bartholome [1967] have shown that the critical

values were only three in number, and they worked them out.
And now, A.P. Markeev [1969] has proved that stability still
prevails for one of these values, and one gets instability for
the other two.
6.10.
let
This example demonstrates the following proposition:

A
and
Theorem 6.7.
be as above, C
Let
verifying the hypotheses of
be an antisymmetric matrix chosen such
that the origin be stable for the equation
A(O)q + Gq + Cq =
It may be possible to choose a nonlinear function
jj/qjr
~~
as
R(q), with
q"" 0,
such that the origin becomes unstable for

H
BR
Aq + Gq + Cq - Bq
= O.
(6.7)
Another problem, more important from the point of view

of gyroscopic stabilization, is the following one:
and
be given, C
let
A,C
verifying the hypotheses of Theorem 6.7.
Is it always possible to find an antisymmetric matrix

that the origin be stable for (6.7)?
such
Apparently, this problem
remains open.
Notice finally that it is easy to exhibit some examples
where a nonlinear function
depending on
stability of a gyroscopically stabilized system.
destroys the
o.
6.
6.11.
123
In this and the next few sections, we shall resort to
Liapunov's direct method to dispose of the hypotheses according to which

depend on
q.
det C t 0
and the gyroscopic forces do not
Let us consider the differential equations

d aT
aT
dt "'7 - aq
!lq
=-
alI
aq - G(q)q
(6.8)
verifies the hypotheses of 6.1, II is a !if 2 funcalI (0)

tion such that II (0) = 0, aq
= 0 and G is !ifl and
antisymmetric.
where
Clearly, if
origin
=q =0
has a strict minimum at
II
is stable, since the function
verifies the hypotheses of Theorem I.4.2.

even if
lI(q)
admits at
q = 0
0, the
H
=T
+ IT
On the other hand,
a strict maximum with a
negative definite hessian matrix, it may happen as follows

from Theorem 6.7, that the equilibrium be stable, due to the
The situation is, therefore,
presence of gyroscopic forces.
more intricate than it was in Sections 2 to 5.

6.12.
Let us assume that the gyroscopic forces are Lagrangian,
Le. that for some
function
!if2
5t"(q,q), one has 5t"
G(q)q
By the way, this is always the case for a constant
suffices to choose
uP
=- 2
G, for it
It is easy to show that 5t"
q Gq.
is of the form
La.
l<i<n
where the
ai
are
!if 2
(q)
q. ,
functions.
defined up to a linear form in
At last, since
5t" is
with constant coefficients,
124
III.
one is free to assume that
for
ai(O)
i, in such
every
a way that
~(q,q) = T +
is statiOnary at
= q = 0.
Y- IT
The following theorem, yielding
a sufficient instability condition of the origin for (6.8), is

akin to Theorem 3.2.
6.13.
e: >
Theorem (V.V. Rumiantsev [1966]).
(with
(i)
(ii)
Ben)
e:
such that
{q E B , IT(q) < O}
e:
$;
the set
'I' = {(q,q): q E
is
(iii)
If there exists an
$, and
the function
e,
q E B , H <
e:
and
(q!
iL~
O}
3q
(0,0) E 3'1';
2T+Y+ (q!
il~)
is strictly
3q
positive on'!';
Proof.
Clearly
(0,0) E 3'1'.
is such that, for every

(a)
(b)
The function
_(q!
il~)H
ilq
(q,q)E'I':
< V(q,q)
il~
-H(2T + y+ (q ! aq-
> 0.
Instability then follows from Theorem 1.5.1 and

Remark 1.5.5.
6.14.
Corollary.
Q.E.D.
In Theorem 6.13, Hypothesis (iii) can be
replaced by
(iv)
~(2)
(q,q)
>
on'!',
6.
9P(2)
where
125
is the quadratic form associated with the
hessian matrix of
at
g=
O.
Indeed, V = -H( ~(2) + o( // (q,g) //2.
Proof.
6.15.
Exercise.
.
q.
depend on
Suppose that
in equation (6.3) does not
Suppose further that the hypotheses of
Theorem 6.7 are satisfied.

an anti symmetric
Then, whatever the form of
F,
G can always be chosen such that
Hypothesis (iH) of Theorem 6.13 is not verified.

Hint:
by a suitable change of variables, bring the linear
Er + G*r + C*r = 0, with
part of the equation to the form

G*
and
6.16.
C*
as in the proof of 6.7.
Remark.
There exist mechanical systems whose equation
assumes the form (6.B) without the kinetic energy verifying

the hypotheses on
potential function
with
and
T2
d
'
= 2l q' T
A(q)q,
E~.
T = T2 , IT
= IT
in 6.11.
Indeed, consider a system with
and a kinetic energy

Tl
= bT
(q)q
and
T = T2
+ Tl + TO'
TO = d(q), b E ~n
The equation of motion is again (6.B) with

- d
and
G(q)
T
rab
~q
- (abT)TJ
aq- _.
by P. Hagedorn [1972], if the differential forms

are exact, then
As observed
bT(q)dq
= O.
rn much the same way, the equations of Routh with a

fixed
(cf. Appendix 11.14) assume the form (6.B) with
T2 ,T1 ,T O replaced respectively by
R2 ,R1
and
RO'
That is
why the instability theorems proved in this section are often

known as "inverses of Routh's stability Theorem" (cf. IV.S
and IV.B).
126
III.
7.
STABILITY OF A MECHANICAL .EQUILIBRIUM
Good historical surveys of the inversions of LagrangeDirichlet's theorem will be found in L. Salvadori [1968] and
P. Hagedorn [1971].
Both papers deal also with dissipative
systems and stationary motions, and devote a subsection to

systems with gyroscopic forces.
We make an effort below to
complete the bibliographies of Salvadori and Hagedorn, in

such a way that, in regard to the inversion of the LagrangeDirichlet theorem, our list combined with theirs should not
be too far from exhaustive.
Several authors claimed that there is always instability in the absence of a strict minimum at the equilibrium
point (cf. e.g. P. Appel [1932] and B. Lanczos [1962]; see
however P. Appel [1953]), which is not true, as we have shown.
Besides those already cited, many partial inversions have been
published, associated with the names of J. Hadamard [18.97],
P. painlev~ [1897], A. Kneser [1895-1897], G. Hamel [1903],
L. Silla [1908].
P. Hagedorn, in the paper cited above [1971] gives an
interesting partial inversion:
in case
functions, the equilibrium is unstable if

maximum.
and
IT
IT
are
~2
has a strict
We did not deal with this theorem here, because the
type of proof, i.e. showing by variational methods the existence

of a motion with appropriate properties, is too remote from
the subject of this book.
By the same method, P. Hagedorn
[1975] proved recently an interesting sufficient condition of

instability for the systems mentioned in Remark 6.16:
function
q
!2
bTA-lb + (IT-d)
if the
has a relative strict maximum at

q
=q=
is unstable.
7.
127
As for dissipative systems, and here without any claim

at completeness, let us mention, besides those already cited,
the early contribution of P. Duhem [1902], as well as
V.M. Matrosov [1962]1 and W.T. Koiter [1965].
siders time-dependent systems.
Matrosov con-
As we shall see later
(N. Rouche [1968]), the result embodied in Theorem 6.2 can

be obtained straight-forwardly by using two Liapunov-1ike
functions with a nice physical interpretation:
first the
total energy and second the vector of conjugate momenta:

cf. Section VIII.3.
CHAPTER IV
STABILITY IN THE PRESENCE OF FIRST INTEGRALS
1.
Introduction
The principal drawback of Liapunov's direct method is

that no general procedure is known to construct auxiliary
functions suiting specific theorems.
That is why, in stability
problems, one should a priori neglect no available information

concerning the solutions.
In particular, the first integrals
will often be helpful, either to facilitate the search for

auxiliary functions or to eliminate part of the variables and
thus decrease the number of equations to examine.
Both points
of view will be developed later, in Sections 3 and 4

respectively.
Section 5 deals with an important case where
first integrals are known, namely the stationary motions of

mechanical systems with ignorable coordinates.
studies a particular motion of this type:
particles in the betatron.
Section 6
the orbiting
And the last section gathers
practical criteria concerning the various methods of
2.
General hypotheses
129
constructing positive definite functions.

2.
2.1.
General Hypotheses
Starting with this chapter, our general hypotheses will
be somewhat weakened.
cost in the proofs.
This will be done at practically no

Further, as we have seen in Section II.6
and as we shall see again in Theorem 3.5, it may be quite

helpful, if not necessary, to consider auxiliary functions which
are not differentiable.
Let us therefore describe the general
setting of our future work.

For
and
an integer
> 2, I
= ]T,OO[
n an open connected subset of
for some
TE
~, consider a con-
tinuous function
f: I
-+
~n,
(t,x)
-+
f(t,x),
and the corresponding Cauchy problem

f(t,x)
(2.1)
(2.2)
for some point
(to'x O) E I
n.
There passes through
at least one non-continuable solution.
Such solutions will
be written
x:
J -+ ~n,
-+
x(t)
is, of course, an open interval.
where
write
J = ]Il,W[
and
J+ =
We shall often
Without mention to the
[to'w[.
contrary, uniqueness of the solutions will not be required.

Further, we assume that the origin
and is a critical point of (2.1):
of
~n
belongs to
i.e. for any t E I: f(t,O)
n
O.
130
IV.
2.2.
A function
W: I
lipschitz ian with respect to
~,
(t,x)
and continuous, is called
a first integral of Equation (2.1) if

every
(t,x) E I
W(t,x), locally
n+W(t,x)
for
This condition implies (see Appendix I)
x Q.
that for any solution .x(t)
of (2.1),
(2.2), W(t,x(t
simultaneously increasing and decreasing:
is
therefore it is
constant.
3.
3.1.
How to Construct Liapunov Functions
A trivial case is when one knows a 'first integral which
is also positive definite.
Then, of course, stability of the
origin follows immediately from Theorem I.4.2.

that there are
first integrals
and let us write
W(t,x)
Wi(t,x), 1
for the m-vector
m < n,
(Wl(t,x), .. ,Wm(t,x.
W(t,O)
=0
A natural approach is to try combining the
Wi
Without loss of generality, we assume that

tEl.
Suppose now
a single positive definite function.
for
into
The following theorem
proves helpful in this direction.

3.2.
Theorem (G.K. Pozharitskii [1958]).
continuous function
<p:
~m
+ gf
positive definite* if and only if

definite, where
Proof.
111
such that
<p
I IW(t,x) I I
is any norm in
= 0,
Moreover, <P(W(t,x
<p
(W(t,x
is
is positive
gfm.
The condition is obviously sufficient.
necessity, notice first that since
<P(O)
There exists a
To prove its
is continuous and
E ~ with
<P(W) ~ b(1
Iwl I).
being positive definite, there is a
*A 5fl positive definite function was defined in I.3.3. What

is used here is a trivial extension of this definition to
continuous functions.
3.
How to construct Liapunov functions
function
a E .5t'such that
a(llxll) ~

Q.E.D.
Practically, one will choose the norm, I Iw(t,x) I I
~i
Wi lt,X)2, in such a way that if, as has been assumed, the
Wi
are locally lipschitz ian with respect to
will be true of
IIW(t,x) II.
x, the same
And, therefore,
is
IIW(t,x) II
a first integral and an auxiliary function suiting Theorem I.4.2.

3.4.
If this test of Pozharitskii fails, one may try to find
a supplementary function
tion
<P{V(t,x),W(t,x
V(t,x)
and some continuous combina-
such that this last function be
positive definite, and decreasing along the solutions.

reasonable choice for
with
<p
would be
max(V(t,x),W(W(t,x),
!JRm
W a continuous function on
!JR, or equivalently
into
(see Exercise 3.6), max(V(t,x), I Iw(t,x) I I).
Hence, the fol-
lowing theorem concerning two real valued functions.

3.5.
Theorem.
Let
V(t,x),W(t,x)
into
!JR, both locally lipschitz ian in
a first integral, and let

(i)
(ii)
(iii)
I x n
and continuous, W
a E.5t' be such that
o for
V(t,O)'" 0, W(t,O)
max(V(t,x),W(t,x
D+V(t,x) < 0
be two functions on
>
t E I1
a(llxll)
for every
for
(t,x) E I x n1
(t,x) E I x n
such that
V(t,x) :::. W(t,X)1

then the origin is scable.
tion
If there exists further a func-
b E.5t' such that

(iv)
max(V(t,x),W(t,x
the stability is uniform.
b(llxll)
for
(t,x) E I x
n,
IV.
132
Proof.
As appears from Theorem 2.7 of Appendix
D+[max(V(t,x) ,W(t,x 1 < 0

Theorem I.4.2 applies.
3.6.
Exercise.
functions on
1
Let
I x rl
n, with
I,
(t,x) E I x rl.
for
Therefore,
Q.E.D.
V(t,x)
~
into
V(t,O)
and
W(t,x)
~m
and
and
W(t,O)
be two continuous
respectively,
O.
There exists
~: ~m
max(V(t,xl,~(W(t,x)
is positive definite, if and only if
max(V(t,x),1 IW(t,x) I I)
Exercise.
V and
and
is positive definite, where
111 I
In the notations of Exercise 3.6, suppose
W do not depend on
W(x).
~ such that
~m.
is any norm in
3.7.
t.
We write their values
V(x)
Then the following statements are equivalent
(and this equivalence has some practical value) :

(a)
there exists an open neighborhood

origin, N C rl, such that for
max (V (x) , I IW(x) I I) > 0:
(b)
the origin
is for
V(x)
Remark.
W(t,x) : W*(x) : 0
G
= {x E
W(x)
= O.
The situation becomes simpler if one
knows of a continuous function
put
x E N\{O}:
It may happen that hypothesis (ii) of Theorem 3.5
is difficult to check.
that
of the
a strict minimum
constrained by the equation

3.8.
rl: W*(x)
neighborhood of
G\{O}.
W* (x)
for every
O}
and let
on
rl
into
(t,x) E I x rl.
N C rl\{O}
!JR such
Let us
be an open
Hypothesis (ii) will be verified in
each of the two following circumstances:

(a)

V (t, x) ~ a * (I I x II )
on
a* E Je such that
I x N:
3.
How to construct Liapunov functions
133
there exists a continuous function
(b)
into
and
such that
v*(x) > 0
on
V(t,x)
>
Bp
for some
p >
a* E Je such that, for every
One may choose
on
on
I x N
G\{O}.
n can be restricted
In the former case, noticing that

to a ball
v*(x)
V*(x)
0, there exists a function
(t,x) E I x (Bp\N)
a ( II x II) = min (a * ( II x II ) , a* ( II x II
In the latter case, it is clear that, for some open

neighborhood
of
G\{O}, with
M\{O}CN, one has
Vx E M\{O}: V*(x) > O.

Then there exists a function
every
a* E Je such that, for
(t,x) E I x M,
V(t,x) ~ inf{V*(y): !lxll ~ Ilyll ~ p, y EM} 2: a*Cllxll),

and condition 3.8 a) is satisfied with
The hypothesis on
n+V(t,x)
M sUbstituted to
N.
in Theorem 3.5 can be
weakened at the expense of an extra requirement on
V(t,x).
This is the object of the following theorem, where
W*
and
G are defined as above.

3.9.
into
Theorem.
~,
Let
V(t,x),W(t,x)
both locally lipschitz ian in
a first integral with

W*
is continuous.
(i)
be two functions on
W(t,O)
= 0,
and continuous, W
W(t,x) > W*(x) > 0, where
If
b(llxll) ~ V(t,x) ~ a(llxll)
for every
(t,x) E I x N and some functions

b E 5e, and
of
I x n
G\{Oh
N C n \ {O}
a E Je and
is an open nel.ghborhood
134
IV.
(ii)
n+V(t,x} < 0
for every
(t,x) E I x N
such that
V(t,x} > W(t,x);

then the origin is stable.
tion
If there exists further a func-
c E.5e such that

(iii)
c(llxll);:W(t,x)
forevery
(t,x)EIxn;

3.10.
Exercise (A.M. Liapunov [la93]).
potheses of this section, suppose
~m
~
is a first integral.
In the general hy-
W(t,x)
Suppose
on
V(t,x)
I x n
on
I x n
into
into
is a "Liapunov function" subject to the constraint
W(t,x)
0, Le., for some
such that
.5e and every
= 0:
W(t,x)
V(t,x) > a(llxll), V(t,O)
(i)
Vet,x)
(ii)
(t,x) E I x n
0;
~ O.
Prove that the origin is stable with respect to the perturbations satisfying the equation
('Ito E I) ('Ie: > O) (30 > 0)
Ilx(t;to'x o ) II < e:
3.11.
with
Exercise.
and
(\fXO
W(t,x}
=0
i.e.
E Bo: W(to'x O) = 0) ('It E J + )
(conditional stability).
Use the following differential equation
in
it
sin (ty) - x,
~,
to prove that the hypotheses in
Exercise 3.10 do not entail stability (unconditional stability).

Hence, the interest of Theorem 3.5.
4.
4.1.
Eliminating Part of the Variables
Let us now see what kind of results can be obtained by
using the first integrals to eliminate part of the variables
4.
Eliminating part of the variables
and thus decrease the number of

Suppose we know
m, 1
135
equat~ons
to deal with.
m < n, first integrals, or in other
words a vectorial first integral
W(t,x) = (WI (t,x) , 'Wm(t,x.
Let us separate Equation (2.1) into two subsystems as

y
g(t,y,z),
.
z
where
x
y, g E gen-m
(y,z).
where
(4.1)
h(t,y,z),
z, h E
and
gem.
We shall write
Suppose at last that the equation
S E
gem
W(t,y,z)
can be uniquely solved in some neighborhood
of the origin of x-space and yields the continuous solution

z
= z(t,y,S),
such that
= O.
z(t,O,O)
The Equations (4.1)
can be replaced by the system
where
For
gO(t,y) + r(t,y,S),
0,
g(t,y,z(t,y,O
goCt,y)
gO(t,y).
and
(4.2)
r(t,y,S) = g(t,y,z(t,y,S -
0, the first of these equations reduces to
y=
(4.3)
gO(t,y).
Total stability of the origin for this reduced equation is a

key hypothesis in the following
theorem~
Sufficient conditions
insuring total stability have been given in Section II.4.

4.2.
Theorem.
Let
n = Bp
for some
a E.5t'
p > 0
such that for every
(t,y,S) E I x B p
(i)
(ii)
and suppose
IIz(t,y,S)II < aClI(Y,S)II),

Ilr(t,y,S)1I < a(IISII),
136
IV.
(iii)
the origin is totally stable for Equation (4.3);
then the origin

Proof.
= 0,
=0
is stable for Equation (4.1).
Hypotheses (i), (iii) and (ii) imply respectively
that:
('Ie: > 0) (311 > 0) [llyll
<
('111
0) (302
>
0) ('Ito E I) (301
>
11
and
>
IIBII
But
W(t,x)
<
03
<
11) II BII
is continuous and
n1 => [llzll
<
e:1;
0) [I IYol I < 01' I Ir(t,y,B) I I < 02
and
('102 > 0) (303' 0
<
<
t E J+1 => [lly(t) II
<
n1;
31 => Ilr(t,y,B) II
<
21.
W(t,O)
= O.
Therefore,
We conclude that
('Ito EI)(Ve: > 0)(30 > 0) [IIxo " < and
Indeed, it will be sufficient to choose
4.3.
tEJ+1 => IIx(t)II < e:.
= min(01,04).
Q.E.D.
In the next theorem, Theorem 3.5 is applied to the re-
duced Equation (4.3).

of both methods:
In a way, it combines the advantages
eliminating part of the variables and using
the first integrals to construct an appropriate auxiliary

function.
4.4.
V(t,y)
Theorem.
on
Suppose there exist a continuous function

x
Il'
into
91 for some neighborhood
the origin of y-space, a function

two functions
E I x Il')
a, c E !It' such that
pet)
on
into
Il'
91
of
and
(V(t,y) E I x Il') (V(t,y')
4.
Eliminating part of the variables
(i)
(ii)
(iii)
137
Iv(t,y) - V(t,y') I ~ pet) Ily - y'II,

V(t,y)
a<llyll); V(t,O) = 0,
+
D(4.3)V(t,y)
= -c(V(t,y,
where the subscript
(4.3) indicates a derivative computed along the

solutions of Equation (4.3).
Suppose further, that there exists a function
into
!ii, a function
at each point where

(iv)
(v)
cP
E 5t' and a constant
such that,
is defined,
p(t)q(t)
(y,S)
L.
=0
is stable for Equation (4.2).
Moreover, if there exists a function

(t,y) E I
(vi)
on
Ilr(t,y,S)11 ~ q(t)CP("S"),
Then the origin
every
get)
b E 5t' such that, for
n'
V(t,y) < b("yll),

Proof.
Consider System (4.2) and its first integral
c-l(Lcp(1
lsi I.
W(S)
This integral is not locally 1ipschitzian,
but it does not matter!
Because of (ii), V and
W verify
the hypotheses stated in Remark 3.8, and therefore Hypothesis

(ii) of Theorem 3.5 is satisfied.
theorem is trivially satisfied.
notice that the derivative of
Hypothesis (i) of the same

As for Hypothesis (iii),
V(t,y)
along the solutions of
(4.2), namely
V(t,y,s) = lJ.m+ sup

h+O
Vet + h,y + hgo(t,y) + hr(t,y,S

h
may be estimated thus, using (i) above
- V(t,y)
138
IV.
+
+
D V(t,y,S) ::. D(4.3)V(t,y) + pet) Ilr(t,y,S) II.
But using (iii), (iv) and (v):
D+V(t,y,S) ::. -c(V(t,y
+ LCP<llsll).
The second member of this inequality is negative if

W(S).
thus verified.
4.5.
V(t,y) >
Hypothesis (iii) of Theorem 3.5 is
Q.E.D.
As is well known (cf. e.g. L. Cesari [1959]), stability
may be destroyed by a change of coordinates, even if this

change is continuous.
(y,z)
of
(y,S)
Therefore, stability of the origin
for (4.1) is not necessarily equivalent to stability
=0
for (4.2).
The following exercise gives a clue
to settle this question.

4.6.
Exercise.
Suppose the origin of the x-space, x E
is stable for some differential equation, and let
~n,
CP(t,y)
be a continuously differentiable change of coordinates such

that
CP(t,O)
~~ (t,y)
=0
for every
be everywhere regular.
and the jacobian matrix

Show that the origin of
y-space is stable for the corresponding differential equation

in
y, i f
cp -1
wi th respect to
5.
5.1.
is continuous in
at
x = 0, uniformly
tEl.
Stability of Stationary Motions
The above theorems were largely inspired by the classical
stability problem of stationary motions of Lagrangian systems.

For such motions, E.J. Routh [1877] (see also [1975]) gave
simple conditions insuring stability, but only with respect
5.
Stability of stationary motions
139
to perturbations leaving unchanged the values of the conjugate

momenta.
Such a drastic restriction is seldom if ever
realized.
Hence, the interest of the following generalizations.
We consider the Routh's equations (Appendix II{11.4

and assume that the stationary motion to be studied corresponds to the critical point
these equations.
That
=q
= x,
0, c
has been chosen to be
reduce the generality of our study.

variables
0, q
of
does not
Introducing the new
q = y, S = c - c, we bring the Routh's
equations to the general form
y,
.
S
Y{t,x,y,S) ,
The critical point is now
(S.l)
0.
(x,y,S)
(O,O,O).
It will be
convenient to write Routh's function (Appendix II{11.3
in
terms of the new variables in the obvious form

~(t,x,y,S)
where
in
~2
y.
= ~2{t,x,y)
and
~l
+ ~l{t,x,y,S) + ~O(t,x,S) - n{t,x)
are respectively quadratic and linear
The stability of the critical point will be studied
using the Hamiltonian function

H (t,x,y, S)
where
on
y.
a~1
(
- y)
Cly
W{t,x, S) = n -
~O
~ {=
+ Yt'(t, 0, 0, 0)
Notice by the way that

The time derivative of
H(t,x,y,S)
= - a
at
~ ( t,
,.;;
(=
-
W(t,O,O)
,,
~2 + W
does not depend
= for
all
t.
H along the motions reads

Yt'(t,O,O,O
(Q
Iy)
IV.
140
and does not depend on
if
is independent of
t.
The
following theorem is a consequence of Theorem 3.9, if one

identifies
5.2.
V with
Theorem.
Hand
If there exists an open neighborhood
N C n\{o}
of
for every
(t,x,y,S) E
b' E
I I~I I.
W with
E n:
{{x,y,~)
I
= 0,
(x,y)
such that
O}
N and some functions
a,b,a',
Ye:
(i)
a{llyll) ~ Ye2 (t,x,y) ~ b(llyll);
(H)
a' (1Ixll) ~ W(t,x,S) ~ b' (lIxll +
Ciii)
)"
- a
at/ v(=
- =/HJ( t,O,O,O
(Q y)
II sll);
wherever
< 0
then
(a)
the equilibrium
(x,y,S) = (0,0,0)
of (5.1) is
uniformly stable and therefore the corresponding generalized

steady motion is stable with respect to
(b)
if the integral of momentum
continuous in
(q,g,i)
(q,q);
aT
or
(t,q,q,r)
is
t, this
partial stability is uniform;
(q,q,S)
(c)
if
r(t,q,g,c+S)
(0,0,0)
is continuous in
q,g,S
at
t, stability
(or, in case b), uniform stability) obtains with respect to

all variables
5.3.
q,q,r.
We consider next the case of a system with time-
independent constraints and let the forces

(Qly)
o.
We allow
IT
and
be dissipative:
Q to be time-dependent.
The
constraints being time-independent, one gets, using obvious

notations, that
.9P(t,x,y, S) = .9P2 (x,y) + geO(x, S) - IT (t,x)
5.
Stability of stationary motions
Further
that
is positive definite in
~2
does not depend on
S.
141
and one observes
The following theorem is
then a simple consequence of Theorem 3.5 and Remark 3.8 b)

5.4.
Theorem.
If there exists a continuous function
and an open neighborhood
N C n\{O}
{(x,y, S) E n: S = 0, (x,y) 'I O}

I
N and some
of the set
such that for
(t,x,y, S) E
E Y.r:
(i)
W(t,x,S} '::,W*(x,S);
(ii)
W*(x,O).:: a(llxll);
and for
W*(x,S)
et,x,y,S) E I x n
(iii)
oat
[net,x) - n(t,o) ~ 0; (Qly) ~ 0;
then, the steady motion is stable with respect to

moreover, for some
(iv)
q,q,r.
If
b E .5t':
W(t,x,S} ~ b(llxll + Ilsll),

5.5.
Example:
regular precessions of a symmetrical top
(v.v.
Rumiantsev [1971).
Consider a rigid body of mass
with an axis of symmetry and a fixed point
on this axis.
We assume
to be the origin of some inertial frame of
reference
OXYZ
with
OZ
vertical in the field of gravity.
Oxyz
will be a system of orthogonal axes, fixed in the body,
with
Oz
along the axis of symmetry.
In this system, the
coordinates of the center of mass will be written

zo
>
0.
(O,O,zO)'
The Eulerian angles specifying the position of the
body with respect to
OXYZ
will be written
$,8,$:
we use
for these angles the definitions of H. Goldstein [1950].

A
If
is the common value of the two equal principal moments of
142
IV.
inertia and
is the value of the third one, the Lagrangian
function reads
L
=T -
1 2
2 2
1
V = 2 (AS + Aq> sin a) + 2 C(lj! + ~
The variables
and
cos a)
- rrgzo cos a.
are ignorable and yield the first
lj!
integrals
C(~ + ~ cos a)
cl '
A~ sin 2 a + C cos a(~ + ~ cos a)

The Routh's function reads
(c 2 - c l cos a)2
--~--~~~2~----
2 A sin a
- mgz o cos
and the Routh's equations are

cl
AS +1\
(c 2 - c l cos a)
sin a
(c 2 - c l cos a)2cos a
A sin 3a
- mgz o sin a
0,
o.
They admit the static solution
a
'IT
aOElO, 2[,a
0,
-2
c l = c l ' c l ::: 4,Amg Z o cos ao'
corresponding to the stationary motions
5.
Stability of the betatron
cl
0, cP
d,
0/
143
I cf - 4Amgz O cos
2A cos
c l cos
e0
+ A~0/
eO
eo
(5.2)
. 2e 0
s~n
Noticing that
possesses a relative strict minimum at

c2
(Exercise:
for
prove this!), we deduce from
Theorem 5.4 that the stationary motion (5.2) is stable in
a,
and
~.
6.
6.1.
e,
Stability of the Betatron
Description of the system.
The betatron is an axially
symmetric accelerator where a particle with some electric

charge
describes a circular trajectory in a transverse
time-varying magnetic field.
The hatched parts in Figure 4.1
Figure 4.1.
The betatron: cross-section and trajectory.
144
IV.
represent a section through the pole pieces of the electromagnet.
We choose the axis of symmetry of the device as
axis in a system of cylindrical coordinates

write
(hr,he,h z )
For symmetry reasons, it may
....
be assumed that the magnetic induction

e
(r,e,z), and we
for the three unit-vectors of the
associated orthonormal base.
on
Oz
B does not depend
and has a vanishing component along
he'
Further,
it seems to be a fair approximation to assume that in the

plane of symmetry
z = 0, one has
....
rot B = O.
Indeed, let
us start from Maxwell's equations
E= -
rot
as
C at
(6.1)
....
rot B
where
is the electric field,!
the velocity of light.
that the current density
the current density and
first approximation will be to consider
....
i
is negligible.
Strictly speaking,
it does not vanish, because of the presence of the particles

being accelerated.
But it is very small anyhow.
Therefore,
the second equation becomes

..... 1 aE
rot B = c'at .
(6.2)
Using the symmetry hypotheses, one deduces from (6.1) and

(6.2) that
aBr
azwhere
r
aB
1 -1
(t,r,O) - ar (t,r,O) = - ~
c'" r
fr
But, for the values of
has been written for
Bz '
a2B (t,s,O)ds
s ---2
0
at
to be considered,
1
a2B
frs ---2
0
at
(t, s, 0) ds
(6.3)
6.
145
2
llc
is practically negligible as compared to
It is easily
verified that, equating the second member of (6.3) to

+
amounts to obtain
rot B
=0
for
= 0,
as was announced
above.
Let us now assume that the dependence of
on
and the spatial coordinates can be made explicit as in the

following formula:
B(t,r,z)
where
Br(r,O)
(i)
every
= 0,
and
get)
is proportional, for
t, to the current intensity through the coils of the
electro-magnet.
The following realistic hypotheses will be used:
(ii)
for some quantities
(iii)
6.2.
<
<
get)
s;
<
is a :2
>
>
and every
0 ~ get), get) 1 0;
function, and
Equations of motion.
for every
a, S
As
B(r,z) >
B = 0,
div
o.
one obtains that,
0,
-I:s ~:
rBr(r,z)
(s,z)ds
(6.4)
and, using (i), that
aB
az
(r,O) = O.
(6.5)
The relativistic Lagrangian function for a particle with mass

at rest
and charge
(positive or negative) reads
IV.
146
where
is the velocity of light, whereas the mass of the
"R [1
m.
,,",ving particle is
~r1/2
For the derivation of this function
v 2 ,2 + r2!2 + ,2.
with
through the use of a
vector potential, we refer to H. Goldstein [1950] and

M. La10y [1973]3.
The equations of motion read

dr
dt
=.r,
dz
dt
= z,
mre 2 + ~ erg(t)B(r,z),
c
mr 2 e +
w(t,r,z,e)
d
dt (mz)
where
w(t,r,z,8)
~
c
~
c
get) Irs B(s,z)ds

0
8g{t) Irs aaB (s,z)ds,

0
is a first integral and
depending on the initial conditions.

reference that if
=~
k,
a constant
We observe, for future
m{r 2 + r 2 e 2 + z2), one gets
d
dt (mH)e
= -m-c
9g(t)
Ir s
0
(6.6)
B(s,z)ds
the derivative being computed along the motions.

6.3.
Existence of a circular orbit.
Let us now investigate
the possibility of generating a circular orbit of the type

ret)
= r o'
ret) = 0,
z{t)
= 0,
z{t)
= 0,
= So{t).
(6.7)
6.
147
Stability of the b,etatron
It corresponds to a generalized stationary motion (see

Appendix II.12), since
eO
is not necessarily constant.
Sub-
stituting this solution in the equations of motion and setting

m
= m [1 _
r~8~(t)l-1/2
c2
'
we get
e
c
(6.8)
Due to (6.5), the latter condition is always verified.

8 0 (t) ~ 0, and since
get)
Assuming
% 0, one deduces from the two former
ones the equality
r~B(ro'O) =
rO
Io
(6.9)
s B(s,O)ds,
which is necessary and sufficient for the existence of a

stationary motion of type (6.7).
pole-pieces shape.
o < r < r O' one has
for instance if
This is a condition on the
It implies that for some value of
B(r,O) > B(rO'O), which can be realized
B(r,O)
is constant, for some
o < r < r l < r O' and strictly decreasing on

index of the magnetic field is the number
r l , on
[rl,r O].
n
(6.10 )
B(rO'O).
rO
B(r,O)
If (6.9) is satisfied,
80
= B(rO'O) [:oJn
The
defined by
One often chooses
r,
for
is given by (6.8):
near
rOo
IV.
148
(6.11)
6.4.
The stability question.
Let us now investigate the
stability of the stationary motion (6.7).
The transformation
of variables
= r,
r = get)
m
= z,
e = 2l!L
t, z = 2l!L
i
m
m
~,
does not affect the stability to be studied (cf. 4.6):

remember the existence, assumed in Section 6.1, of the bounds
on
get).
The new equations of motion admit the first integral

where
g(t)w(r,z,~)
= r 2~
w(r,z,~)
+ -e
r
s B(s,z)ds.
0
(6.12)
But (6.9) and (6.11) imply that
and therefore
e60 < O.
(6.13)
60 is a constant which, by (iii), is such that
By the transformation of variables, the stationary
motion has been changed into
= r O'
= 0, r = 0,
..
= 0, 6 = 60
(6.14)
Its stability will be studied using the auxiliary function

defined by
Hl
=H
where
ew
and
with
H =
~2
1 2~2
21 (-r2 + r 22
~ + z ) - 2 r 090
Hl
6.
149
6.5.
Lemma.
H1 <
sphere of radius
proof.
g (t)w
the motions of the system,

. along
leol and center at point (6.14)
Taking into account (6.6) , (6.11) and the fact that

is a constant, we get
_ get) ':"2
':"2
[r + z + 6 w],
get)
_ g(t) ':"2
':"2
[r + z + w(6
260 )]
H1 ==
get)
if
ew < 0:
if
ew > 0:
H1
cited above.
(6 - 260 )60 <
If
<
<
1, H1
has an isolated minimum at
point (6.14) under the constraint
w == 0.
Using (6.12) and (6.13), we get for

1
2
where
~(r,z)
in the sphere
Q.E.D.
~.
Proof.
The lemma is proved, because
6.6.
in a
(r2
!
':"2
+ z ) +
a~
az
a2~
(r 0,0)
ar2
0:
2..!J!. (r 0' 0)
ar
2
2
2
e 2 [~ (r,z) - ~ (ro'O)]
Irs B(s,z)ds.
(r 0' 0)
By (6.5) and (6.9) we obtain
iL
ar<lz
0,
(ro'o)
aB
1
ar (r 0' 0) + -r O B(rO'O).
The latter equation together with (6.10) shows that

2$
aar2
(rO'O) >
if and only if
n < 1.
On the other hand,
given (6.4)
n
az
Remember now that
2 (rO'O)
rot
fore, as was shown above:
in the plane
z =
and
there-
150
IV.
aB r
az-
(r,O) -
ClB
ar
(r,O)
= O.
Using (6.10), we obtain
i12
(lz
Thus, if
- B(rO'O).
rO
n > 0, both members of this equality are strictly
greater than
6.7.
(ro'O)
Theorem.
and the lemma is proved.

If the magnetic index
Q.E.D.
is such that
o < n < 1, the generalized stationary motion (6.7) is stable.

Proof.
Indeed, referring to Theorem 3.5 and Remark 3.8(b),
Ig(t)wl
H1
plays the role of the first integral
the role of the function
V(t,x).
W(t,x)
and
Thus (6.14) is stable
for the system in the new variables, and therefore (6.7) is

also stable for the original equation of motion.
6.8.
Remark.
Q.E.D.
The model adopted here for the betatron includes
the following features, which are of course conditions of

validity for our conclusion:
the pole pieces admit the
axis as an axis of symmetry and the plane
z = 0
Oz
as a plane
of symmetry; the electromagnetic field due to the particle

itself, or to other particles in the beam (space charge effect)
can be neglected.
7.
7.1.
Construction of Positive Definite Functions:

Practical Criteria
It was proved in Theorem 3.2 that, W(t,x)
being an
m-dimensiona1 first integral for Equation (2.1), there is a

continuous function
definite if and only if
such that
I IW(t,x) I I
~(W(t,x
is positive
7.
Construction of positive definite functions
Observe that the character of

is immaterial here:
continuous and
W(t,x)
151
being a first integral
the only requirement is that
= 0.
W(t,O)
W(t,x)
be
Theorem 3.2 is typical of a
class of useful propositions to be studied in the present

section.
7.2.
First of all, several helpful comments can be made on
the positive definiteness of
W = W(x)
I Iwi I, in case
is
time independent.
(a)
To start with, I Iw(x) I I
and only if equation
W(x)
is positive definite if
has no root besides in
some neighborhood of the origin.

(b)
if
W is
Therefore, by the implicit function theorem and

)fl, a necessary condition for
I Iw(x) I I
positive definite is that the jacobian matrix

rank strictly smaller than
(c)
Ilw(x) I I
of
aw
ax
to be
(0)
be of
m.
W, one sees that
Coming back now to a continuous
is positive definite whenever one of the components
W is different from zero on the set of points
x 1
where the other components vanish simultaneously.

(d)
z E ~-p
Wl(x)
Of course, if .x = (y,z)
for some
= = Wp(X)
with
= Y(z)
then
< P < m, and if the equations

= can be solved in some neighborhood
y
is positive definite if and only if
+ W!(Y(Z),Z)
7.3.
and
p,
of the origin, yielding a unique solution

IIW(x) I I
y E g/P
Exercise.
Wp+l(Y(z),z) +
is positive definite, of course in z-space.

Consider condition 7.2(e) above, restated in
the quasi-equivalent form:
152
IV.
Wl
(i)
x
is different from zero on the set of points
W2 ""'Wm vanish simultaneously.
where
One natural way to satisfy this condition is to choose
W1
such that:
Wl
(ii)
equations
possesses a strict minimum constrained by the
W2 (x) =
= Wm(x) =
O.
Exhibit an example where (i) is satisfied, but not (ii), and

check, on the example, the positive-definiteness of
7.4.
It would be practically useful, in case
Iw(x) 1 I.
W is twice
continuously differentiable, to get criteria for the

existence of a function
such that
(W(x
be positive
definite and the positive-definiteness be recognizable at the

matrix of its second derivatives.
This means that this
matrix, which by the way is usually called the hessian matrix,

should be positive definite at the origin.
the present section, where
dependent of
The remainder of
W is always assumed to be in-
CP'2, is devoted to this prob 1 em.

and
this respect, the function
should be discarded at once.
one gets
In fact, by putting
In
7.
where
as
equations
= 0,
vi(x)
x .,.
o.
But, as
= l, ,m,
of dimension higher than
O.
153
m < n, the
define a subspace of
~n
The following theorem is the
basis of the so-called Chetaev's method to construct positive

definite functions.
7.5.
Theorem.
with
CP(W(x
There exists a
5f2
function
cp: ~m .,. ~,
positive definite and the hessian matrix of
CP(W(x, computed at the origin, also positive definite, if

and only if there exist parameters
(i)
(ii)
l<i<m
Al, ,A m
such that
AiV i (x) :: 0;
the hessian matrix of
l~i~m
[A.W. (x) + W~(x)]

~ ~
(7.1)
is positive definite at the origin.

Proof.
The conditions are clearly sufficient since (i)
implies that the function (7.1) contains no terms linear in

x.
Let us prove that they are necessary.
and
%'i (x)
Then
Considering the
154
IV.
o<llxI12)
IIxl12
where
L OIi
for some
%'i (x) +
and
k > 0
L [..-!
l::i::m K
cp (W (x
As
Yfn
L ~kv.
(x)vk(x)
~
~
2
= L xi'
i
Therefore,
O.
is positive
> kllxl12
i,k
IIxl12
K>
for some
vf)
O.
-+
Further, in
- O.
as
-+
L OI i v i (x)
definite,
K(I
L B,kv'V
k
~
~
But
i,k
k
%'i (x) + v~~ (x)] > K IIxl12
Ai
and this proves the thesis for
= OI i /K.
Finding appropriate parameters
Ai
<
(7.2)
Q.E.D.
is made easier by
the following corollary.

7.6.
Corollary.
(ii)
In Theorem 7.5, (ii) may be replaced by
l::i::m
Ai %'i (x) > 0
vi(x) = 0
Proof.
for
for every
such that
i = l, ,m.
The new condition is necessary, as is shown by (7.2)
and using the choice above:
Ai
= OI i /K.
easily proved for the new theorem.

such that
Bo C n
and let
= {x
There exists an open neighborhood
L Ai %'i(x)
a > 0
xt-O
and
> 0
over
B > 0
N.
Sufficiency is also
Indeed, let
En: vi(x)
N of
be
> 0
= 0,
aBo
such that
l, ,m}.
Further, there exist constants
such that
A. %'. (x) > -a

~
and
7.
155
over the compact set
~
ex
Ii
Therefore, one has, for
(lBo.\N.
A.
%'.
(x)
Ii
x E aBo
vf (x) > 0,
and, since this function is a quadratic form, the inequality
is true for all
~(W(x
O.
=
Therefore, the function
~ I AiWi(x)
i
Wf(X)
is positive definite with a hessian matrix positive definite

at the origin.
Q.E.D.
Corollary 7.6 can be exploited as follows:

the equations for some variables (say
into
Ii
A. %'.(x)
l.
l.
solving
p < m), one substitutes
and then uses any available criterion
(e.g. sylvester's criterion) to check the positive definiteness

of the quadratic form thus obtained, which possesses
n - p
variables only.
7.7.
is positive
As observed in Exercise 7.3, / /W(x)/ /
definite if one of the components of
N, say
strict minimum constrained by the equations

= Wm(x) = O.
WI' admits a
W2 (x) =
The next two theorems result in a criterion to
recognize this fact, using derivatives of
W of order not
higher than the second.

7.8.
Theorem.
Suppose there exist real numbers
such that
vl(x) +
2<i<m
A.V. (x) - O.
~ l.
Then
(a)
there exists a
l.I
E:!7R such that
A2 , ,A m
IV.
156
cj>(W(x
= Wl(x) +
2<i<m
[A.W.(X) + ].lWf(X)]
~ ~
be positive definite, with the hessian matrix of
~(W(x,
computed at the origin, also positive definite, if and only

if
(b)
~l(x) +
2<i<m
such that
Proof.
Ai ~i(x) > 0
vi(x)
0, i
x 1 0
for every
= 2, ,m.
The condition is necessary because
CP(W(x
%:1 (x) +
L [A.~.(X) +].lV~(x)]
2:;:i:;:m
+o(llxI1 2 )
whereas, by hypothesis
~l (x) +
L [A.~.
2<i<m
(xl + ].lV~(x)] > 0
x 1
for
o.
It is also sufficient, for it can be proved as in Corollary 7.6,

that i f
~l(x) +
A.~. (xl> 0
[A. ~. (x) + ].lV~ (x)] > 0 for x 1
2<i<m
for x 1 0, V4 (x)
= 0,
= 2, ,m,
then
~l (xl +
7.9.
2:::i~
Corollary.
(a)'
W1 (xl
Q.E.D.
Thesis (al in Theorem 7.8 implies

admits at
=0
strained by the equations

7.10.
o.
a strict minimum conW2 (x)
= =
Wm(x)
Let us introduce a last theorem concerning the case,
already touched upon in Section 7.2(d), where some of the

equations
Wi (x)
their arguments.
=0
can be solved with respect to some of
More precisely, suppose
[awl
Clx 0
is of
o.
7.
Construction of
rank
~ositive
p, 1
for some
P < m, and
a{w1 , ,Wp )
a(x 1 ,,x)
Then, if we write
(x p +1 , ,x n ), the equations
(7.3)
0
is regular.
157
definite functions
= ... = Wp(x)
W1 {x)
=
=0
(x 1 , ... ,x p )' Z
can
be solved in some appropriate neighborhood of the origin,

under the form
7.11.
Theorem.
(ClW/Clx)0
Suppose the above conditions concerning
and the jacobian matrix (7.3) are satisfied.
there exists a
1/1(0)
= Y(z).
)f2
and
definite in
1/1: ~m-p....
function
Then
~, such that
1/I(Wp +1 (Y(z),z), ,Wm(Y(z),z
is positive
z, with a hessian matrix, computed at the origin,
also positive definite, if and only if there exist real
Ap+1,,A m such that the function
numbers
ptl~k~
AkWk(Y(z),z)
be positive definite with a hessian matrix,
computed at the origin, also positive definite.

Proof.
Let us show first that the linear terms are lacking
in the expansion of the function

k
p + 1, ,m
every
such that
Wk(x)
where
x .... O.
around
qk(x)
= qk(x)
O.
Indeed, if
This happens because, for
0, i
Wi(x)
2
+ oellxll )
is quadratic in
vi(X)
where
Wi(x)
= 0,
= -%'i(x)
0i(1 Ixl 12)/1 Ixl 12 .... 0
concerning (7.3) implies that
= Wk(Y(z),z),
Wk(z)
= 1, ,p,
k
and
one has
p + 1, . ,m,
(7.4)
0(ll x I1 2 ) .... 0
IIxl1 2
as
+ i(ll x I1 2 )
as
one has
x .... O.
(7.5)
The hypothesis
IV.
158
k=p+1, ,m
for some appropriate quantities
Yki.
Therefore,
and the expected result (7.4) is obtained through replacing

vi (x)
by its value (7.5).

The condition in the statement of the theorem is
obviously sufficient.
Suppose
Let us show that it is also necessary.
~(Wp+1(Z), ,Wm(z
is positive definite with a
positive definite hessian matrix, where

a
function.
Let
~(~P+1'
'~m)
is
<Xk = (a~/a~k) 0' k = P + 1, ,m.
Remembering (7.4), one gets that
~ (Wp +1 (z), 'Wm(z
where
0 (
II z I 12) / I I z 112
quadratic part (in z) of
for some
h > O.
p+1~k~m
-+-
<Xk %-k (z ) +
as
0, and
-+-
qk(Y(z),z).
(I I z I I 2 )
~k (z)
is the
By hypothesis,
Hence, the function
is positive definite with a positive definite hessian

matrix.
7.12.
Q.E.D.
Exercise.
Find two quadratic forms
W2 (x 1 ,x 2 ), xl' x 2 E 9f, such that
W1 (x 1 ,x 2 ),
2
W1 (x 1 ,x 2 ) + W2 (x 1 ,x 2 )
is positive definite, whereas no linear combination
7.
Hint:
159
observe, in connection with Theorem 7.11, that the
hessian matrix of
WI
and
W2 , computed at the origin,
cannot be positive definite.

7.13.
vertical rotations of a top.
The example of the top
illustrates several possibilities and drawbacks of the

various theorems presented in this section.
the problem as in N.G. Chetaev [1955].
O.
body with a fixed point
Let
Let us formulate
We consider a rigid
Oxyz
be a system of ortho-
gonal axes chosen along the principal axes of inertia of

the body at
0, and let
A, B = A and
inertia with regard to these axes.
Let
be the moments of
=y =0
and
z > 0
be the coordinates of the center of inertia of the
body.
Let
p,q,r
be the components, in
Oxyz, of the
instantaneous angular velocity of the body with respect to

some inertial reference frame.
At last, let
y, y', y"
be
the directional cosines of the upward vertical with respect

to the moving axes
Oxyz.
We are interested in the stability of the vertical

rotations
p
0, q
0, r
= r O'
0, y'
0, y"
= 1,
this stability being considered with regard to the variables

p,q,r,y,y',y".
In fact, there will be at our disposal a
sufficient number of first integrals to enable us to solve

the stability question without resorting to any other equations
of motion.
The variations of the variables with respect to
the critical point will be designated by

p =
~,
n,
a, y'
~,n,s,a,S,y,
= a,
y"
=1
with
+ 5.
160
IV.
Well-known first integrals for this problem read

W1
A(1;2 + n 2 ) + C (1,;2 + 2r OI,;) + 2mgzo,
W2
A (I;a + nS) + ceol,; + rOo + 1,;) ,
W3
W4
+ S2 + 0 2 + 20.
1;.
When trying to apply Pozharitskii's Theorem 3.2 by first

computing
Iwi 12 =
L w~,
one observes that in this ex-
1~i~4 ~
pression, the terms of second degree do not constitute a

positive definite quadratic form, and therefore deciding of
the positive-definiteness of
Iwl 1
will not be easy.
Theorem 7.5 probably yields a better clue, for here only the
quadratic terms would have to be examined.
But finding the
Ai and computing the expression (7.1) remains cumbersome, all

four squares
w~
appearing in (7.1).
The method used by
N.G. Chetaev [1955] is somewhat simpler.

of real quantities
A1 , ,A 4
be positive definite.
to Chetaev's book.
and
He looks for a set
such that
For details of computation, we refer
It appears, in fact, that the easiest
procedure is suggested by 7.2(d) and Theorem 7.11.

jacobian matrix of the
equations
W3
=0
and
Wi
is of rank 2 at the origin.
W4
can be solved for
yielding, up to terms of order 2,
0, 0
Substituting in
W1
and
The
W2 ' we get
I;
and
The
0,
7.
W
l
A(~2 + n 2 )
mgz(ex
W
2
A(~ex
+ nS)
- crO ex
+ S2) +
+ S2
2
161
At last, looking for a linear combination
W
l +
AW2 =
A~
AA~ex
CrO 2
(mgz + A -2-) ex
cro
+ An 2 + AAnS - (mgz + A -2-)S
we see that both quadratic forms, in
+
and
(~,ex)
respectively, are positive definite if and only if

4Arngz.
(n,S)
c2r~
>
This becomes apparent by applying Sylvester's
criterion
AA
AA
2"
2"
croA
-(mgz + -2-)
> 0
or
A
2
cro
A + --2- A + mgz < o.
This inequality will be verified for some
if and only if
which is a sufficient condition for stability.

7.14.
Regular precessions of a satellite.
Consider a rigid
satellite attracted by newtonian forces towards the earth

center
Let
be the center of mass OD the satellite and
and let
Oxyz
be an inertial reference frame.

G~n~
a system of orthogonal axes chosen along the central axes of

inertia.
Let
A, B
=A
and
be the central moments of
162
IV.
inertia.
The satellite will be assumed to be sufficiently
remote from the earth to justify the hypothesis of the

classical restricted problem:
around
i.e., the motion of the body
has no effect upon the motion of
of course obeys Kepler's laws.

located in the plane
velocity
G, which then
We suppose the orbit to be
0, with
~O' anticlockwise around
radi~s
and angular
The problem is thus
Oz.
reduced to studying the motion of the satellite around
under the action of the gravity torque, which reads
Here
are the unit vectors along the axes
a,B,Y
are the directional cosines of
OG
G~ll1;,
and
with respect to
(see e.g. E. Leimanis [1965] or V.V. Beletskii [1966]).
G~ll1;
Let us now introduce the orbital system of axes

Gx'y'z', as shown in Figure 4.2, with
Gy'
tangent to the orbit and
-+
-+
w = pel + qe 2 + re 3
0")-
Gz'
Gx'
directed as
parallel to
Oz.
Let
OG,
be the instantaneous angular velocity
of the satellite with respect to
Gx'y'z'.
Following
y
Figure 4.2.
Orbiting satellite
163
7.
F.L. Chernous'ko [19641, there is a class of motions

called regular precessions, characterized by a fixed position
of the axis
in the orbital system.
G~
Every motion of this
type is described by specifying the directional cosines
y,
y',y"
q = 0,
of
G~
the value of
with respect to
Gx'y'z'
r, which is a constant.
and, since
There are three sub-
families of motions:
C-A = -wo C"

y"',
0,
= -4w o
r = r
/l
C-A y", Y =
A
y" =
with
-00
/l - y,,2
y" with -1
y"
r
y'
- y,,2,
y'
y" with -1
< r <
00,
= y'
<
y"
(I)
< 1,
= 0,
< :V" <
(II)
1,
0, y" =
1,
(III)
whose stability will be studied using the following first

integrals of the equations of the relative motion:
r + woy",
WI
(7.6)
2
2
2
2
W2 = A(p2+l) + cr2 - wO(C-A)y" + 3w O(C-A)y ,
(7.7)
y 2 + y,2 + y,,2
(7.8)
W3
l.
For the detailed calculation, we refer to Chernous'ko's

paper.
y,
Le t us
y" -
cons~'d
y"
er th e
'
var~a t '~ons
p, q,
t"
~
=r
-c-
+ Wo C-A -y"
of the variables appearing in (7.6) and (7.7)
with respect to the unperturbed motion (I).
The integral
Wl
written in the new variables and adjusted in such a way that

Wl(O)
reads
WI
= f;
+ wOo.
When solving the equation
'
IV.
164
WI
for
and substituting in
W2 , one gets
a function which is positive definite for

entails stability with respect to
C > A.
p,q,r,y
This
and
y", whereas
is a direct consequence of
y'
Equation (7.8).
For the unperturbed motion (II), the new variables
will be
y
=r
+ 4wO C-A
c:-
y ,
h - -,,2
Solving
WI
y"
'
- y,
=y
and the first integrals
WI
- y",
y"
and
W3
where
become
+ woz,
WI
W3
2yo + 2y"z + 0 2 + z2 + y,2
= W3 =
for
and
and substituting in
W2 ,
one gets, up to terms of second order,
and this function is positive definite if
A > C.
We reason alike for the unperturbed motion (III), using

as new variables
W2
r - rand
After computation,
becomes, up to terms of second order,

W2
= A(p 2 +q 2 )
222
2
2
2
2
) + wO(C-A) (y +y' ) + 3w O (C-A)y
crwO(y +y'
This function is positive definite, if
-r
+ 1.
y"
> -w O
C-A
c:-
or
A > C, and i f
accordingly as
r > +4w O
y"
A-C
c:-
-I, when
C > A
or
y"
= +1,
accordingly as
r <
A > C.
C-A
Wo c:-
or
when
C > A
or
r < 4wO C-A

c:-
8.
165
The general conclusion is that the stationary motion

(I) is stable with respect to
p,q,r,y,y'
and
y", if the
ellipsoid of inertia of the satellite is of the flat type,

the motion (II) if it is of the oblong type, whereas the
motion (III) can be stable in both cases, but only when the
angular velocity
-r
7.15.
What kind of difficulty would one encounter
Exercise.
satisfies a suitable inequality.
if, to study the stability of motion (I) hereabove, one tries,

as for (II) and (III), first to solve
Wl
W3
0, and then
W2 ?
to substitute in
8.
The earliest result concerning stability in the

presence of first integrals concerns stationary motions of
autonomous lagrangian systems and is due to E.J. Routh [1877J.
With the notations of Section 5 adapted, in some obvious
way, to the autonomous case, let
.9i!'0 (x, 13), IT
13
= S,
IT (x)
and
IT (x) - .9i!'O(x,l3)
Q =
O.
.9i!'2
.9'2 (x, y), .9i!'0
Routh observed that if, Jor
admits a strict minimum at
then the stationary motion defined by
=y =0
x
and
= 0,
S
is stable for all perturbations satisfying the equation

13
= S.
This proposition of Routh was extended by
A.M. Liapunov [1893J to non-autonomous equations

possessing a first integral
W(t,x):
x = f(t,x)
see Exercise 3.10.
The awkward restriction concerning the perturbations

was partially removed by A.M. Liapunov:
also V.V. Rumiantsev [1968J.
on this point see
It was completely removed, in
the setting of autonomous mechanical systems, by L. Salvadori

[1953].
Indeed, using a suitable Liapunov function in
IV.
166
connection with Theorem 1.4.2, he proved stability of the

stationary motion for arbitrary perturbations.
On this ex-
tension, see also G.K. Pozharitskii [1958] and.

V.V. Rumiantsev [1968].
On the other hand, L. Salvadori [1966] extended his
result of 1953 to a class of dissipative systems, again by
constructing an appropriate auxiliary function and using
Theorem 1.4.2.
Of course, the dissipation is limited to
non-ignorable coordinates.
This result is still generalized
in L. Salvadori [1969] to the case of a function

~o(x,~)
IT (x) -
considered along with a more general dissipation,
reduced however to non-ignorable coordinates.

proof is here completely different:
The type of
it makes use of
families of Liapunov functions with one parameter, an ingenious technical trick to be dealt with at length in
Chapter VIII.
In the mean time, C. Risito [1967] proved
Theorem 4.2 and used it to get a similar extension to

dissipative systems, with the restriction, however, that the
dissipation, although again limited to non-ignorable coordinates, has to be complete in this setting.
As becomes apparent from these historical considerations,
the methods used to tackle this problem of extending Routh's
initial observation have been varied ones.
At the end, it
beoame clear that the utmost generalization, namely the one

presented in P. Habets and C. Risito [1973] and in Section 5
above, which by the way includes non-autonomous systems, is
but a simple consequence of Theorem 3.5.
This theorem
appears in the same paper of P. Habets and C. Risito.
Its
forerunners oan be found in P. Habets and K. Peiffer [1973]

and M. Laloy [1973]3.
8.
167
A generalized version of Theorem 4.2 appears in
c.
Risito [1974] where the bound on
can be time-dependent
and where, moreover, one may find conditions for uniform

stability.
c.
c.
A first version of Theorem 4.4 appears in
Risito [1971].
It has been generalized in P. Habets and
Risito [1973] and further in
c.
Risito [1974].
A survey of stability results for stationary motions

was given by S. Pluchino [1971], along with an extension of
Salvadori's result to non-holonomic dissipative systems.
The study of the betatron is due to M. Laloy [1973]4.
Theorem 3.2 has already been attributed to
G.K. Pozharitskii [1958].
The method of constructing
weighted combinations of first integrals and of their squares

goes back to N.G. Chetaev [1961].
The theorems of Section 7
emphasize the usefulness of those criteria where positivedefiniteness can be recognized at the derivatives of second
order, thus enabling one to apply Sylvester's criterion.
They come essentially from C. Risito [1975].
Interesting
comparisons between the various methods of constructing

Liapunov functions can also be found in V.V. Rubanovskii and
S. Ia. Stepanov [1969].
CHAPTER V
INSTABILITY
1.
1.1.
Introduction
Inasmuch as stability is a desired property in many
circumstances, it is important to have at one's disposal some

effective means of recognizing instability.
ject of the present chapter.
This is the ob-
However, before studying instab-
ility as such, we shall deal at some length with new concepts such as sectors, expellers, etc., and this deserves
some preliminary comments.
Let us go back to Chetaev's Theorem I.s.l, where the
role played by the function
V(t,x)
is twofold:
it is used
to prove first that no solution starting from inside the open

set
crosses
remain in
as
a~
and then that no such solution can
approaches infinity.
Roughly speaking,
the first of these properties will be referred to hereafter

as
peller.
being a sector, and the second as
being an ex-
When combined, they imply instability.
One should
1.
Introduction
169
here emphasize the following fact which, as we shall see, has

important theoretical and practical consequences:
it is pos-
sible to prove separately, i.e. by using two distinct auxiliary functions, that
ler.
is a sector and that it is an expel-
The following example makes it clear.

In the general hypotheses used for Chetaev's Theorem
1.5.1, suppose for simplicity that the differential equation

on hand is autonomous.
Suppose further that for some
with
Be C Q, there exists an open set
tion
V:
on
Be
91 and a function
e > 0,
func-
b E Jte
':
(i)
(ii)
V (x) > 0;
V(x)
~ b (V (x) ) ;
assume further that there exists a neighborhood

and a second auxiliary function
(iii)
(iv)
W:
+91
of
a~
such that
W(x}
W(x} > 0
if, at last, 0 E
and
W(x) > 0
on
'
N n Be;
a', then the origin is unstable, as can be
shown by reasoning almost as in Chetaev's Theorem, the function

of
W being such that no solution issued from some point

~
can approach
a'.
1.2.
Exercise.
Give a detailed proof of the statement above.
1.3.
This Chapter will show how the ideas of sector and ex-
peller can be used to decompose the concept of instability

into two simpler ones, each of them becoming the object of a
separate study.
This systematic approach appears as a nat-
ural continuation of Chetaev's results and, even more directly, of K. P. Persidski's theory of sectors.
It is worth
V.
170
INSTABILITY
mentioning finally that our treatment takes closed as well as

open sectors into account, and that the topological principle
of Wazewski is used to get a useful generalization of the
basic theory.
From a practical point of view, closed sectors
will often coincide with hypersurfaces defined by first integrals.
2.
2.1.
Definitions and General Hypotheses
Our general hypotheses remain here those of Section
IV.2.1, to which we add however the requirement that, for any

(to'x O) E I x
n,
the Cauchy problem
:it
f(t,x)
(2.1)
= xo
(2.2)
x(t O)
has a unique sOlution.
This supplementary hypothesis is es-
sential in the present context, because we shall often resort

to the argument of continuity of the solutions with respect
to the initial conditions.
will be written
The solution of (2.1) and (2.2)
x(t;to'x O)'
Instability is of course the contrary of stability.

It will prove helpful, for reference purposes, to recall here
the explicit definition of instability:
the critical point
at the origin is said to be unstable if

(3e > 0) (:lt o E I) ('115 > 0) (3x O E Be) (3t E
J+)
x(t;to'x o ) fl Be'
All the following theorems have the form of sufficient conditions and they should, in a complete version, begin as
Chetaev's Theorem, by:
"If there exist an
e > 0
with
2.
Definitions and general hypotheses
BC
n
e:
and a
sume that
By reason of simplicity, we as-
e:
Let
171
and
Ce:
to
=I
are chosen here once and for all.
x Be:'
G C Ce:
For any
and
t E l , we de-
fine
G(t)
We call
= ix:
G*
{(t,x) E G: x 'I- O},
{(t,x) E aG
the side-boundary of
esis is that all sets

~n
the origin of
2.2.
(t,x)E G},
A set
n c e: :
G.
x 'I- O}.
Another general hypoth-
to be mentioned below are such that
~s a cluster point for
G(t o )'
is called a sector if, for every
0 > 0, one
at least of the two following conditions is satisfied:

(i)
(ii)
2.3.
(h O
E G* (to)
(3x O E
A set
Eo)
n Eo)
(Vt E J+)
(t,x (t1to'xO
X(t1 t o 'X O)
(3t E J+)
e:
E G1
Be:'
is called an absolute sector if, for every
Xo E G* (to)' one at least of the two following conditions is

satisfied:
(i)
(Vt E J+)
(t,X(t1t o 'X O E G1
(ii)
(3t E J+)
x(t1to'xO)
Clearly
Ce:
sector is a sector.
e:
Be:'
is an absolute sector, and every absolute

It may be somewhat surprising to observe
that, in the definition of a sector

tioned which do not belong to
G.
G, some points are men-
This peculiarity will be
justified a posteriori by the role it
will play in several
proofs below.
2.4.
A set
is called an expeller if
172
V.
2.5.
INSTABILITY
It will be called an absolute expeller if
3.
Fundamental Proposition
The followin9: statements are eg:uiva1ent:

(a)
the origin is
(b)
there exists a sector which is an absolute
(c)
there exists an absolute sector which is an eXEe11er;
(d)
there exists an absolute sector which is an absolute
unstab1e~
eXEe11er~
eXEe11er.
Proof.
Obviously (b)
and (c).
(a)
(a), (c)
(d).
(b)
But if the origin is unstable, there exists a

xOi E BE
ti > to
x(ti~tO,xOi) ~
such that
= 1,2, .
CE
(a) and (d)
Therefore, it will be sufficient to prove that
sequence of points
The set
and a sequence of time-values

BE and
xOi
for
G defined as the intersection with
of the trajectories of all solutions
absolute sector and an absolute expeller.
x(t;to'x Oi )
is an
Q.E.D.
Conditions ensuring instability will therefore be obtained by combining, in various obvious ways, sufficient conditions for the existence of a sector, an absolute sector, an
expeller or an absolute expeller.
By the way, the set
mentioned in Chetaev's Theorem 1.5.1 is an absolute sector

and an absolute expeller.
The next two sections will be de-
voted to a fairly detailed study of sectors and expellers

respectively.
4.
Sectors
173
4.
4.1.
Sectors
Sectors and absolute sectors will often be characterized
by the way the solutions of the differential equation cross

their side boundary.
Roughly speaking, they can cross it all
from outside to inside {see Theorem 4.2, (i) to (iii
and
this type of behavior leads to absolute sectors, or all from

inside to outside (see Thorem 4.5 and 4.6) and this yields
sectors, or in a more complicated way, some from outside to
inside and the others from inside to outside (see Theorem
4.7), and this yields again a sector.
A few more definitions
are needed to describe precisely such behaviors of the solutions.

A point
(s,a) E L
an ingress point of
('it E ]s,s + T])
an egress point of
(\it E ] s, s + T])
(3T> 0, s + T E J+(s,a
o
(t,x{t;s,a E G;
G
if
if
{3T > 0,
(t,x{t;s,a
a consequent point of
(\it E [s - T,s[)
will be called
+ T E J + (s,a
G;
if
(3T > 0, s - T E J{s,a

o
(t,x{t;s,a E G.
To recognize the fact that a point of the side-boundary

is, or is not, an ingress, an egress or a consequent point is
often an easy matter and can, in many instances, be deduced
from a careful examination of the second member of the differential equation.
One may also resort to some simple crit-
eria making use of auxiliary functions.

be given below:
Several of them will
see e.g. 4.3 and 4.4 and some lemmas in Sec-
v.
174
tion 6.
INSTABILITY
For the time being, assume that we can in general
recognize such points and let us prove sufficient conditions

for a set
to be a sector or an absolute sector.
We deal
first with absolute sectors, because they are simpler.

4.2.
Theorem.
Each of the following conditions is suffici-
ent for a set

(i)
to be an absolute sector:
G is closed in
Ce
and no point of
is an
egress point;
(ii)
is open and no point of
is a consequent
point;
(iii)
is open,
oG
oG
and no point of
is an
egress point;
(iv)
Proof.
Ce
for some
H positively invariant.
(i) and (ii) are obvious; (iii) follows from (ii) and
4.3
from Lemma
4.3.
=Hn
Lemma.
below; (iv) is also obvious.
If
is open,
oG
an egress point, then no point of

Proof.
Suppose a point
oG
L
Q.E.D.
and no point of
is
is a consequent point.
(s,a) E L
is a consequent
point, Le. (3T> 0, s - T EJ(s,a)) (Vt E [s - T,s[)

(t,x(t;s,a)) E G.
cause
oG
oG,
Let
N be a neighborhood of
contains points outside
G.
which, be-
The continuity
with respect to the initial conditions implies that there

exist solutions starting from some neighborhood
N'
of
(s - T, x(s - T; s,a)), N' C G and reaching these pOints outside
G.
Thus there would exist an egress point, which is

Q.E.D.
excluded.
4.4.
Exercise.
If
is open and no point of
is a
4.
Sectors
175
consequent point, then no point of
is an egress point.
As appears from the theorems to follow, sufficient conditions to get a sector are more involved.
4.5.
Theorem.
Assume
and for every

(i)
(ii)
(iii)
(iv)
then
6(t)
that, for some set
G closed in
Ce:
to:
is connected;
the origin is a cluster point of
G(t);
0
contains at least one cluster point of
aBe:
no point of
G(t);
is an ingress point;
G is a sector.
Proof.
Assume, on the contrary, that
Then, for some
G is not a sector.
6, 0 < 6 < e:, we may write the opposites of
propositions (i) and (ii) in the definition of a sector.

(a)
The opposite of (ii) reads:

(4.1)
Let us put
This set is defined for every
borhood of the origin of
Threfore, due to (ii) here-
above, G(t)
that
H(t)
H (t) C Be:
where
t
'I~.
/..;&n.
to' and is an open neigh-
On the other hand,
and then, because of (iii):
(means "the complementary of".
to'
G(t)
o
for, otherwise, G(t)
aH(t) 'I
(4.1)
implies
aCt) ~ (H(t) 'I ~,
Thus, for every
would be the union of the two open
(4.2)
176
V.
disjoint and non empty sets

and
G(t}
a(t) ()
H (t)
and
INSTABILITY
&(t) ()C'H(t),
would not be connected.
(b)
From the opposite of (i) in the definition of a
sector, one deduces that
Xo E G(t O) () aBo' there exist a
Therefore, for every

and a
for
t' > ~
such that
t E [t,t'].
For every
t(x O}'
mum of these
(t,x(t
ELand
T(X O}
x o ' let
(t,x(t}) ~ G
be the infini-
T(X O)
Let us show that
t(x o }
is upper semi-
continuous (on this notion, see e.g. E. J. McShane [1944]).

If it were not so, there would exist in G(t O) () aBo a point
o
i
i
0
i
Xo and an infinite sequence {x O}, Xo ~ x o ' such that Xo
approaches
Xo
as
for every small enough
00
lim T(X O}
and
o
T(X )
< T -
=T
> T(X O)'
and every
Then,
large
enough, one would get
and
But this is impossible because
to the initial conditions and
is continuous with respect
is closed in
Therefore, there exists a

Xo E G(t O) () aBo: T(X O) < T.
But due to (4.1), none of the
solutions on hand comes out of

exist) and further, no point of
Thus, on the one hand
C.
e:
Be:
L
(and therefore ceases to

is an ingress point.
4.
177
Sectors
and on the other, due to (iv),

x(t;to'x O)
One concludes that
(Vt ~ T) aCt)
aH(t)
= ~,
G(t).
but this as-
sertion contradicts (4.2).
Q.E.D.
This theorem leads naturally to the following one, concerning the case of a set
G, no more necessarily closed.
Notice however that if
is closed, Theorem 4.6 is weaker
than Theorem 4.5.

4.6.
Theorem.
Assume that, for some open set
and for
t > to:
every
(i)
(ii )
(iii)
aCt)
is connected;
the origin is a cluster pOint of

aB
0
contains
e:
G (t) ;
at least one cluster point of
G(t);
(iv)
then
Proof.
every point of
is an egress pOint;
js a sector.
All the hypotheses of Theorem 4.5 are satisfied if
one substitutes to
given
the set
= G nee:'
Then for any
0, either proposition (ii) in the definition of a sec-
tor is satisfied, or it is not.
But as every point of

no value of
such that
In the latter case
is an egress point, there will be

(t,x(t;to'x O)) E L, hence the
theorem.
In Theorems 4.5 and 4.6, the points of
Q.E.D.
are con-
178
V.
strained to some kind of uniform behavior:
INSTABILITY
in one case, no
one of them can be an ingress point, whereas in the other,

they all have to be egress points.
Such a restriction is
somewhat relaxed in the following theorem, which is adapted

from Wazewski's topological principle (see e.g. P. Hartman
[1964]).
needed.
Before we can state it, two more definitions are

If
X is a topological space and if
is called a retract of
ping on
B into
if there exists a continuous map-
A, which is the identity on
of mapping is called a retraction (of

4.7.
Theorem.
(i)
For some open set
A.
into
This kind
A).
G, assume that
all consequent points of

and let
A C B C X, A
G are egress points,
be the set of these consequent points.
Suppose there exists for every
0 E ]O,E[
a set
zoC
[G*(t o ) U S(t o )] n Bo such that

(ii)
Zo n G*(t o ) ~ ., Zo n S(t O) ~ .~
(iii)
writing
z6
retract of
then
every
but not of
z'0'.
G is a sector and even, which is a little more, for
oE
such that
]O,E[, there exists an
(t,x (t~to'xO
EG
for every
trajectory startinq from

time at a point of
Proof.
tE J+, or the positive semi-
(to'x O)
meets
dG
for the first
de
E
If the thesis is wrong
(30
>
0) (VxO E Zo) (3T
>
to)
and
(4.3)
The
4.
Sectors
179
points of
being egress points, one shows as in Theorem
4.5, that
T(X O)
is upper semi-continuous.
Owing to (4.3),
one may show alike that it is lower semi-continuous.

the function
then
7r
7rl
, Z' ....
"1'
is continuous.
Thus,
. ()
If
7r
is a retraction mapping
is a retraction of
Z6
on
Z6
ns,
on
and this
cannot exist.
4.8.
Z6 n s,
Q.E.D.
Some geometrical insight is necessary to realize what
happens in Theorem 4.7.

serves this purpose.
The picture presented in Fig. 5.1
It has been drawn for
=3
and the
t-axis
Fig. 5.1.
The geometrical situation of Theorem 4.7.
is not represented.
One assumes that
The set
G(t)
= G(t O)
G(t O)
for
is the "pyramid" OABCD.

t
to
and, for sim-
plicity, that the differential equation is autonomous.

of course
S(t)
the faces
OAD
S(t o)'
and
OBC
Assume that
S(t O)
Then
is the union of
(remember that the origin is ex-
V.
18U
c1uded).
For some
G(t O)
Zo
represent the
n aBo
with the
Oxy
plane.
with the arc
~, then
Zo n
S(t O) = {a} U {b}.
intersection of
identifies
0 < E, let the arc
INSTABILITY
If one
4.9.
Remark.
It is interesting to compare Theorems 4.6 and
4.7.
Suppose
to (iii) of 4.6.
is some open set satisfying Hypotheses (i)

It can be shown that if every point of
is an egress point (last hypothesis of 4.6), then every point

of
is a consequent point, and therefore, in the notations
of 4.7:
= S.
In this setting, 4.6 is not a particular
case of 4.7, for (iii) of 4.7 is an additional assumption.

However, it is easy to prove that if
t
to
and if
tersection with
G(t O)
BE
G(t) = G(t O)
is a cone (i.e. if
G(t O)
for
is the in-
of a set which is a union of rays issu-
ing from the origin), then, again in the setting described

above, the hypotheses of 4.6 imply those of 4.7.
4.10.
Remark.
As it is usually more difficult to prove the
existence of a retraction than to prove its non existence,

the following generalization of Theorem 4.7 will prove helpful:
S
Hypothesis (iii) of Theorem 4.7 may be replaced by:
is not a retract of
Z6
s.
This statement, as well as
those contained in Remark 4.9 may be considered as exercises.

The proofs as well as some further discussion of the relations between Theorems 4.6 and 4.7 appear in M. Laloy [1974]2'
5.
5.1.
Expellers
There is a wide variety of theorems giving sufficient
conditions for a set to be an expeller or an absolute expeller.
Some of them are given below, selected from the most
5.
Expellers
181
simple and effective ones.
The general idea behind them is
that a function
should be bounded from above in
V(t,x(t
some way, and simultaneously increasing at a sufficient pace

for the bound to be attained after some finite time if the
solution
on
x(t)
were to remain in the set.
can of course vary with
5.2.
Theorem.
into
~,
t, as in the following theorem.
Let there exist a function
locally lipschitzian in
continuous functions
function
b E K
(i)
(ii)
The upper bound
aCt), c(t)
V(t,x)
defined on
(Vx O EG*(t o
V(to'x O) > 0:
c (t) > 0
and continuous, two real
c(s)ds
....
[to'oo[
and a
G C c:
such that, for some set
and
on
00
as
t ....
00.
I
to
(iii)
(V(t,x) E G*
such that
V(t,x) .: aCt),
(5.1)
o+V(t,x) ~ 0,
(5.2)
o+V(t,x) ~ c(t)b(V(t,x
then
Proof.
every
(5.3)
+ o+a(t):
is an absolute expeller.
If this were wrong, there would exist an
such that
J+
V(t,x) > 0)
(t,x(t:to'x O
E G for every
[to'oo[, and, due to (5.2), V(t,x(t
t >t
0
t E J+.
Therefore
~ V(to'x O) > 0
It follows then from (5.3) that for every
and finally, by (5.1),
for
182
V.
an inequality becoming wrong for large enough

5.3.
Exercise.
INSTABILITY
t.
Q.E.D.
Prove in detail the following proposition
which has been used implicitly in the proof of Theorem 5.2.

Let
h(t)
be a continuous function on
~, and assume that
for every
every
5.4.
t
~
to
h(t O) > 0
D+h(t) > 0
h(t) > 0
Then
for
If Hypothesis (i) in Theorem 5.2 is replaced

~n
by (i)' the origin of
into
to'
Corollary.
G(t o )
and further that

h(t) > O.
such that
[to'oo[
{x: V(to'x)
is a cluster point of the set
> O}, then
is an expeller.
Theorem 5.2 leads to a particularly simple corollary

when neither
nor
depend on
(which happens of course
mainly in the case of autonomous equations).

5.5.
Suppose
corollar~.
G(t) = G(t O)
Suppose further that for some

exists a
5ffl
function
does not vary with
then
with
into
e'
Be' C
~,
t ~ to'
n,
there
such that
and that:
(Vx E G* (to
Vex) >
(ii)
(Vx E~*(tO)'
vex) > 0)
o~
Vex) > O~
Let
n'
= max
nE1O,n'], the set

and compact.
A (n)
on
(i)
Proof.
vex)
e' > e:
for every
If
{V(x): x E G(t o )}'
h: x E G(t ), Vex)
A (n)
0
is the infimum of
'is an increasing function of
For every
> n}
is non empty
Vex)
on this set,
A (n) > o.
n, with
One
knows then (see e.g. N. Rouche and J. Mawhin [1973] ) that

Vex)
b(V(x.
bE.5t'
By choosing
with
b (n) < A (n)
aCt) = n'
and
Of course
c(t) = 1, one
5.
Expellers
183
verifies all the hypotheses of Theorem 5.2.
Q.E.D.
In the proof of this corollary, the inequality

vex) > 0
has been used in an essential way.
weakened however and be replaced by
It can be
Vex) ~ 0, at the expense
of another hypothesis making sure that the solution will not

vex) = O.
spend too much time in the region where
In this
sense, the following theorem is a natural extension of Corollary 5.5.

Theorem.
5.6.
for some
'I'
[to'oo[ x
Let Equation (2.1) be autonomous and
real function
Vex)
in
'I'
Be:.
defined and locally lipschitzian on
Be:.
e:' > e:, Be' en, and such that
for some
(i)
(Vx E '1'*)
(H)
vex) > 0;
(Vx E iji: vex) >
(iii)
(Vn >
0)
n,D+V(x) = O}
0;
{x: x E if, Vex) =

n
contains no compact, invariant,
the set
0)
D+V
F
non-empty subset;
then
Proof.
If the thesis is wrong, there is an
that for every

limit set(*)
A+
to: x(t;x O) E '1'.
x E '1'*
such
But then the positive
of this solution is non-empty and
A+
C V.
Due to (i), V(X O) > 0, and therefore, by (ii), V(x(t;x O

increasing.
V(x(t;x o
Vo
on
A+.
But
~
Vo
Vex)
as
But as
t
A+
is bounded on
iji
00, for some
Vo'
and thus
Of course
is invariant, D+V(X) = 0
(iii) is violated.
(*)on this notion, cf. Appendix III.
is
on
vex)
A+, and
Q.E.D.
V.
184
INSTABILITY
As is apparent, this theorem is akin to LaSalle's invariance principle VII.3 as well as to Theorem II.l.3 of N. N.
Krasovski.
Some other technical means can be used to ensure
that the solution leaves the set
=0
o+V(x)
and, in particu-
lar, one can resort to a second auxiliary function possessing

suitable properties on an appropriate neighborhood of
set.
this
This kind of idea, which led to Matrosov's Theorem
II.2, will not be illustrated here, for the sake of conciseness.
On the other hand, and going back now to Theorem 5.2,
let us demonstrate another way of weakening the main assumption on
o+v, this time by introducing a hypothesis on the
second derivative of
5.7.
Theorem.
defined on
V.
Let there exist a real

with
V(t,x)
(Vx O E G*(t O
(ii)
(V(t,x) E G*)
and
then
function
locally liEschitzian in
V(t,x)
x
and
G C C:
continuous, such that for some set

(i)
5fl
V(to'x O) > 01
if
V(t,x) > 0, then
V(t,x) < 0
o+V(t,x) ~ 01
The proof is similar to that of Theorem 5.2 and is left

to the reader.
5.8.
Remark.
A corollary similar to 5.4 can be appended to
5.5, 5.6 and 5.7, thus yielding three sufficient conditions

for a set
6
6.1.
Let
to be an expeller.
Examp 1 e
0f
an
Equat~on
0f
Nth Or d er
be a real continuous function defined on some
6.
Example of an equation of nth order
interval
Ja,a[, where
Suppose further that
185
a < 0 < S, and such that

f
f(O)
= o.
is regular enough to ensure unique-
ness of the solutions of the nth order equation
or of the equivalent system
xl = x 2 '
x2
= x3 '
(6.1)
It will be shown under these conditions that if

if
is an isolated root of
n > 3
and
fez), the origin is unstable
for (6.1).
Of course, if
open interval
is an isolated root, there exists an
Ja',S' [, with
does not vanish on
Ja',S'[
a' < 0 < a', such that

except at the origin.
fez)
Without
Ja',S'[ = Ja,S[.
loss of generality, we assume hereafter that
The instability will be proved successively for the various

following cases:
a)
zf(z) > 0
on
]a,o[
or on
b)
zf(z) < 0
on
]a,O[
and on
bl )
b2)
n = 2m
with
m even;
b3)
n = 2m
with
m odd.
]O,S[;
]O,S[,
and
is odd;
The following simple lemma on consequent points will be found

useful.
6.2.
Lemma.
For any open set
G C C
if there exists an open neiqhborhood
and any point

N of
PEL,
and a real
186
V.
function
x
W(t,x)
defined on
N nG, and
W(t,x)
Exercise.
is not a conse-
The following variant of Lemma 6.2 will be
used in Section 8:
on
P, then
at
G.
quent point of
6.3.
=0
G, locally 1ipschitzian in
W(t,x) > 0, n+W(t,x) ~ 0
and continuous and such that
on
the lemma remains valid if the conditions
Ware replaced by the requirement that
and
W(t,x)
6.4.
INSTABILITY
_00
when
(t,x)
o+W(t,x) > 0
P.
Let us now prove step by step the expected property.

a)
zf(z) > 0
We consider only the case
The other one would be treated alike.
Choose an arbitrary
E, 0 < E < min (lal,8), and consider the open set
E ~, x E ~n, 0 < x. < E
{(t,x): t
point
for
ishes.
of the side boundary
of
W= x
is not a consequent point.
is an absolute sector.
is such that
1, . ,n, and at least one of the

Using the function
i = 1, ... ,n}.
for
]0,8[.
on
Any
xi
x., van-
xi' say
and Lemma 6.2, we see that
Then, by Theorem 4.2 (ii), G
But it is also an absolute expeller,
as is readily seen by using Theorem 5.7 and the function
v = xl
- E.
b1 )
By the way, this proof holds also for

Let
E < min
(I ex I ,8)
and
= !if x,!"
= 2.
n
'I'
{x E BE' Vl(x) > O}, where
-xlx n + x 2 x n _ l - x 3 x n _ 2
+ .. + (-l)mx x
+ (_l)m+l
m m+2
calculates that
for
an expression which is larger than
m = (n - 1)/2.
for
xl
O.
Then
is an absolute sector by Lemma 6.2 and Theorem 4.2 (ii).
One
G
But
6.
Example of an equation of n
th
order
it is also an absolute expeller, for

potheses of Theorem 5.6.
only i f
xl (t)
VI
In particular,
xl = 0, and the hyperplane
variant set except for the origin.
0, and thus
Let
e:
x 2 (t)
187
satisfies the hy-
VI (x)
x1 = 0
Indeed
= 0
if and
contains no inimplies
xl (t) - 0
0, etc.
be as above and
= t}l x
'P, 'P
= {x
E Be:'
2
V2 (x) = xm+l - xlf(x l )
an expression which is larger than
for
xl
O.
Then, as
above, G is an absolute sector by Lemma 6.2 and Theorem 4.2

(ii).
But it is also an absolute expeller by Theorem 5.6,
because
"2
only if
= 0,
xl
and the hyperplane
xl
=0
contains no invariant set, except for the origin.

b3)
One verifies easily that

1 2
i
(-1) x i x n - i +2 + -x
2 m+l
2<i<m
W=
is a first integral for (6.1).

'I'
= {x
= a},
E B , W(x)
e:
Theorem 4.2 (iv), G
it is an expeller.
there are pOints
( )
V3 x
=1
O.
e:
[to'co[
l f(z)dz
x '1',
is defined as above.
with
By
Let us show that
In every neighborhood of the origin,

x
where
W(x)
=0
and
V3 (x) > 0, with
(_ 2i - I
,
)
2
x2'x
2'1+ 1 + 1x2'1+1 x n-2i
1
n-2-
,'m-l
~1~
Choose e.g. x 2
to
and
Let
-r
0, x n _ l
and all the other
One calculates that for

2m - 1 x2
m+l
x E 'P
21 IXI
0
f(z)dz > o.
x,'s equal
1
188
V.
But the hyperplane
xm+l
=0
no nonempty invariant subset.
INSTABILITY
contains, except for the origin,

Therefore, the hypotheses of
Theorem 5.6 as modified in Remark 5.8 are satisfied.
7.
7.1.
Instability of the Betatron
The stability of the circular stationary motion in the
betatron has been studied in Section IV.6, to which we constantly refer hereafter.
The result to be proved now is that
this motion, as described by equations IV. (6.7) is unstable if

the index
1.
of the magnetic field is strictly larger than
The proof will be facilitated by first establishing the
following two lemmas.

7.2.
Lemma.
If
n > I, the function
w(u)
Ju+ro
o
Proof.
defined on
sB(s,O)ds
u + r
admits a strict maximum at
W(u)
O.
It results from IV. (6.9) that
~(O)
du
o.
Further,
and using IV. (6.9) again,
And by the definition IV. (6.10) of the index of the magnetic

field,
hence the result.

7.3.
as
Lemma.
If
Q.E.D.
~(u)
is a real function defined on
]-rO'oo[
7.
Instability of the betatron
there exists an
7.4.
z
 0
[1jJ
(u)
1jJ (0)],
such that
u d<p < 0
on
du
Let us now introduce the new variables,
r, r,
and
by the equations
m
g (t)
e = gm(t) e - eO
z
where
mo
60
remaining unchanged.
.. .
(r,r,e,z,z)
space
r,
e
The origin in the
Considering, in the space of
(r,r, e,z,z), some compact neighborhood
K of the orbit of
the solution IV. (6.7), we see that the image of

space
Let
E'
let
(7.1)
0'
of the new coordinates corresponds to the
stationary motion on hand.
.
(r,r,6,z,z)
(t)
m
gm- Z,
z
the variable
r O'
K in the
contains a ball with center at the origin.
be its radius, and write

be the supremum of
mover
= min
(E' ,EO)'
Further,
K.
It is readily verified that the jacobian of the transformation (7.1) vanishes nowhere.
Let
(8)
be the system of
differential equations deduced from the original equations of

motion (which we shall call
of coordinates, and let
of the mass
g
(cf.
(s)
here) by the transformation
m(t,r;r,e,z)
m(r,r,e,z).
be the new expression
It follows from the hypotheses on
(H) in Section IV.6.l) that the instabiUty of the
origin for
(8)
is equivalent to the instability of the
stationary motion for
(S).
Indeed,one.has
190
.
6
V.
.z
and similar inequalities for
e-
and
eO on the other.
.z
and
on the one hand, and
It will therefore be enough
to prove that the origin is unstable for

Of course, (5)
INSTABILITY
(5).
admits the first integral
eJr+ro
W(t,r,6,z) = get) [(r + rO) (6 + 6 0 ) + C 0
sB(s,z)ds],
and, using IV. (6.11) and IV. (6.9), one verifies that
if(t,O,O,O)
O.
From IV. (6.5) and the last equation of mo-
tion, which reads

d
~
Jr s
dt (mz) = c e get)
and from the fact that
=Hn
CE
(s,z)ds,
W = 0, one deduces that the set
H = {(t,r,~,~,z,i): t E l ,
is positively invariant.
aB
az
W=
0, z = i
O}
Hence, by Theorem 4.2. (iv),
Let us now prove, using the function
V = g(t)r r
connection with Corollary 5.4, that it is an expeller.

every neighborhood
N
nG(t O)
where
in
In
N of the origin, there are points of
V> O.
Further, it requires but a few cal-
culations to verify, in particular by using the third equation of motion, i.e.

d
dt mr = mre 2 + ec 6rg(t)B(r,z),
that on
. 6ll [_
V=
e2 -
M. (-)
.!-2]
r + r
c 2 r du
From Lemma 7.3 and from Hypothesis (ii) of IV.6.l, it follows
8.
Example of an equation of third order
that on
G:
V >
-k
Finally,
x
for
e2
c2
[-
r = V/g(t)r
..
.
(r, r, "!1 , z , i) ,
d4> (F)
du
for
as
191
.:.2
+ r ].
V;' 0, one obtains, writing
that for every
(t,x) E G*
such that
V > 0:
V >
-~
ke:
All the hypotheses of Corollary 5.4 are therefore verified.
8.
8.1.
Let
Sl
ExamEle of an Eguation of Third Order

be some open neighborhood of the origin of ~3
and consider a continuous function
91, where
for some
]T,oo[
g(t,x)
E fJI.
on
x Sl
Assume that
into
9
is
regular enough to ensure uniqueness of the solutions of the

differential equation
~ d 2y
d 3y
--3 = g(t,y, ,
2)'
dt
dt dt
which is equivalent to the system
(8.1)
Assume further that
g(t,O) = 0
for every
at last that for some real quantities

for any
tEl.
Suppose
a, a, 0 < a < a, and
xl < 0, one has

2
g(t,xl,axl,a xl) < a xl

and
2
3
g(t,xl,axl,a xl) > a xl:
Under these conditions, the origin of fJl3
is, as we shall
192
V.
see, unstable for (8.1).
INSTABILITY
The following two lemmas will be
used in proving this proposition.
For their own interest,
they are stated with some more generality than is strictly

needed.
8.2.
Lemma.
For any open set
G C C
if there exists an open neighborhood

tion
W(t,x)
on
and any point
e:
of
PEL,
and a func-
into !Jf, locally lipschitz ian in
and continuous, such that
then
(i)
(V(t,x) E N
G)
W(t,x) > 0
and
D+W(t,x) < 0;
(ii)
("(t,x) E N
L)
W(t,x)
and
D+W(t,x) 'i' 0;
is an egress point and a consequent point of

Let
point.
Then, in every interval
such that
Proof.
and suppose
(t 1 ,x(t 1 ;s,a
there is no
(t,x(t
(s,a)
G.
(t2,x(t21s,a
G.
t1
Because of (i) and (ii),

t E ]s, s + T[:
t2 E [s,t 1 [
Therefore, there is a
G,
is not an egress
]s,s + T[, there is a
G.
such that
and the continuity of the solutions with
respect to the initial conditions implies the existence of an

ingress point.
Suppose on the other hand that
point.
is not a consequent
By arguing as above, one shows that in this case also,
there is an ingress point.
The conclusion is that, in both
cases, the following obvious proposition would be violated:

for any open set
G C Ce:
and any point
exists an open neighborhood

on
into
of
PEL, if there
and a function
!Jf, locally 1ipschitzian in
W(t,x)
and continuous,
such that:
(i)
W(t,x) > 0
and
D W(t,x) < 01
8.
Example of an equation of third order

(ii)
W(t,x) = 0
at
P;
is not an ingress point of
then
8.3.
Lemma.
193
For any open set
GC
G.
Q.E.D.
and any point
P = (s,a) E L, if there exists an open neighborhood

an even integer
m> 1
J(m
and a
function
N of
W(t,x)
on
P,
N
into !JI such that

(i)
(Y(t,x) EN
wei) (s,a)
(ii)
w(m)
(iii)
W(t,x)
n~)
for
o < i
0;
m;
(s,a) < 0;
is an egress pOint and is not a consequent point.
then
8.4.
Let us now prove the instability of the origin for
(8.1)
whatever
First of all, Theorem 4.7 will be used to prove that,

to E I
and
> 0
with
n" C 0, the set
GI = {(t,x): t E l , aX l < x 2 < aXI} () C

is a sector.
Notice that
xl < 0
for any point of
To
Gl
start with, let us show that every consequent point of
GI
is an egress point.
of
For every point
(t,x l ,x2 ,X 3 )
L,
one has
either
(8.2)
or
(8.3)
or
(8.4)
Every point
verifying (8.2) and
is an
egress point and a consequent point, as is shown by Lemma

Wl = aX I - x 2 In
verifying (8.3) and x 3 -
8.2, along with the auxiliary function

the same way, every point
S2xI < 0
8.2 and
is a consequent and egress point:

W2
= x2
- axl.
use again Lemma
194
V.
Moreover, no other point of

Indeed, if
P EL
INSTABILITY
is a consequent point.
verifies (8.4) with
x3
>
0, then
W3
for which
veri-
fies the hypotheses of Lemma 6.2 as modified in Exercise 6.3.

If
verifies (8.4) and
x3 < 0, the same conclusion is ob-
tained using W4 = 1/ (Sx l - x 2 ) Let now P verify (8.2):

i f Ci. 2xl - x3 > 0, then P is not a consequent point by Lemma
6.2 considered along with WI; i f Ci. 2xl - x3 = 0, P is not a
consequent point either, for
Lemma 8.3 are satisfied.

satisfies (8.3) and
=W
1 (P) =
WI (P)
whereas
P
The reasoning is the same if
Let us now check the other hypotheses of Theorem 4.7.

0 E ]O,e:[, let us choose two consequent points
For any
P1
to
belonging
2 = (to,x12,x22,x32)
and verifying (8.2) and (8.3) respectively. Consider
Co
the line segment

AX i2 , 0 < A < 1, i
1T(P)
P1
if
Z' = { (t,x) : t
0
consequent points of
by
and
(to,x11,x21,x31)
to' xi
= (1
A)X i l +
1, 2, 3}, and let us call
1T: S ...
z'0
G1
The mapping
verifies (8.2) and
verifies (8.3) is a retraction.
{P 2 }
is not a retract of
1T(P)
the set of
nS
= P2
defined
if
On the other hand, {P1 } U
All the hypotheses of Theorem
Z8'
4.7 are satisfied, and therefore
G1
is a sector.
The only thing which remains to be proved is that

along with the function
p1ete.
G1
But this follows from 5.5 considered

V =
xl'
The proof is therefore comQ.E.D.
9.
Exercises
195
9.
9.1.
Exercises
Consider the non autonomous linear system:

(-b + a cos 2 bt)X l + (b - a sin bt cos bt)x 2 ,
Xl
(9.1)
(-b - a sin bt cos bt)x l
+ (-b + a sin 2 bt)X 2 ,

where
and
are two real constants with
b < a < 2b.
Show that the matrix of the system (9.1) has two eigenvalues
with the same time-independent strictly negative real part,
and, however, the origin is unstable.
Hint:
One computes easily that the common real part of

the eigenvalues is: a - 2b
2
The set
> O}
{(t,x,y): t
xl cos bt
is an absolute sector and an absolute expeller.
To prove it, use the function

one has
E~, x 2 sin bt -
(a - b)V.
= x2
sin bt - xl cos btl
Then, apply Lemma 6.2 and Theorems
4.2 and 5.2.

A particular case of (9.1) was dealt with by L. Markus
and H. Yamabe [1960] who proved the instability by exhibiting
a class of unbounded solutions.
The proof suggested above is
due to M. Laloy [1974]2'

9.2.
For the system:

2
2x l x 2 - x 2 '
4x x - x 2
122
the origin is unstable (K. P. Persidski [1947]).
196
V.
Hint:
G = ~ x 1/1, where
Consider the set
{(X l 'X 2 ): 0 < x 2 < 3x l }.
INSTABILITY
1/1 =
Using Lemma 8.2 and the function
3x l - x 2 ' one can see that all the hypotheses of Theorem 4.6 are satisfied.
the function
9.3.
j!5 V 3
Then, by means of
= Xl
one proves that
V ->
+ (Xl - X2 ) , and applying Theorem 5.2,

In fact, on
Consider the system:
+ b l xl x 2 +
(9.2)
+ b 2 xl x 2 +
suppose the trinomials
have real roots
and
G,
gl(A)
(then, as
g.~ (A)
Ail' Ai2'
2 (i
a.~ + b.A
+ C.A
~
~
Suppose moreover that
1,2)
a2 < 0
has a root
g2(0) < 0, g2(A)
has a root
A2l > 0).
Then, each
of the following conditions is sufficient for the instability

of the origin
(i)
cl > 0
and
gl (A)
(ii)
cl > 0
and
< A
11;
A2l < A12
(iii)
cl < 0
and
A12 < O.
has a second root
G =!if x 1/1, where
Hint: . Consider the set

2
{(x l ,x 2 ) E!if : 0 < x 2 < AllX l }.
WI
= AllX l
and
- x2
4.6, prove that
one gets, for
and, for
x2
G
Xl
W2
= x2 '
1/1 =
By means of the functions

and using Lemma 8.2 and Theorem
is a sector.
A12 ~ All;
In fact, for
x 2 = Allx l ,
0:
WI
-x l g 2 (A ll ) < 0,
W
2
2
a 2x l
0,
(9.3)
9.
197
Exercises
Moreover, G
Indeed,
a)
Vl
in the cases (i) and (iii), use the function

2
2
alx l + b l x l x 2 + c l x 2 For x E W, one has
xl' with V
-> 0,
Vl
and
if and only if
= AIIx1 .
x 2 = Allx l ,
x2
in view of (9.3), the set defined by

does not contain any invariant set.
Moreover,
xl
0,
So, all the hypotheses
of Theorem 5.6 are verified.

b)
Vex)
x~ ~
in case (ii), use Corollary 5.5 with the function
defined as follows:
A2l x l , Vex)
vex)
= x2
if
x 2 > A21xl' and, if
A2l x l
More complete stability results for Equation (9.2)

(and, more generally, for equations with homogeneous right
member) can be found, e.g., in W. Hahn [1967]).
9.4.
Consider the third order scalar equation
-a d 2 y
dt 2
where
and
_ b
- k (y) ,
are real constants and
an open interval containing
O.
(9.4)
dt
key)
We suppose
is defined on
k(O)
=0
and
continuous and regular enough to ensure uniqueness of solutions.
We write
heAl
for
A(A 2 + aA + b).
Equation (9.4)
is equivalent to the system:
Suppose
xl
x2'
x2
x3 '
x3
-ax 3 - bX 2 - k(x l )
a < 0, a
4b
and, for
y < 0:
v.
198
< key) < -Yh(t lal -
~2
INSTABILITY
(9.5)
- 3b).
Then, the origin is unstable (M. Laloy [1974]2; this completes

some results of A. Huaux [1964]; see also R. Reissig, G.
Sansone and R. Conti [1969]).
Hint:
maximum for
Notice that the function hOd has a relative

1
"3( lal 1a2_ 3b) and has a root A2 > AI
Al
Then, taking
and
Al
= A2 ,
[3
the hypotheses relative
to system (8.1) are verified if (9.5) is satisfied.

9.5.
Prove the following generalization of Theorem III.3.2.
(cf. M. Laloy [1975]):
Ben)
E
if there exists an
>
(with
such that:
e = {q E BE' Il(q) < O} ~ $;

OE ae
( all I q) + Il(q) < 0, Vq E e,
aq
(i)
(ii)
(iii)
then the origin

Hint:
is unstable.
Use the absolute sector
{(q,p): 0 < I Iql

function
I Ipl I
< E, H(q,p)
= ~x W,
= a},
with
W=
and the auxiliary
(q I p); a trivial modification of Corollary
5.5 implies that
G is an expeller.
10.
As already noticed in Chapter I, the classical instability theorem of N. G. Chetaev [1934] generalized two
previous results of A. M. Liapunov [1892].
It already used,
at least implicitly, the concept of a sector.
But the method
of sectors as such was developed first by K. P. Persidski

[1946], [1947] and later by S. K. Persidski [1961], [1968],
In.
[1970].
199
The main contribution was due to the former of these
authors, when he introduced a sector for which all points of

the side-boundary are egress points.
Some difficulties arose
concerning the definition of a sector and the proof of sufficient conditions for a set to be a sector.
They have been
discussed by A. D. Myshkis (1947), using arguments of topological algebra, as well as by J. L. Massera [1956] and M. Laloy
[1973]1'
The definitions of a sector and an absolute sector
as given in the present text differ slightly from the original

ones and are those of M. Laloy [1973]1' [1973]2'
As compared to those of N. G. Chetaev, K. P. and S. K.
Persidski, the sectors considered here are more general in
two respects:
first they are not necessarily open sets
(see J. A. Yorke [1968] and M. Laloy [1973]1)' and further

they admit in their side-boundary, the coexistence of ingress
and egress points (M. Laloy [1973]1)' a result which has been
obtained by using the so-called topological principle of
Wazewski [1947].
Notice also that the sectors considered
here are subsets not of ~n, but of ~x~n.

been possible to think of a sector as a set
variable with
difficulty:
for
G(t).
t.
T.
It would have
G(t)
of ~n,
This point of view leads to a specific
the necessity to define some kind of continuity

On this point, cf. Dang Chau Phien and N. Rouche
[1970].
On the other hand, the conditions for a set to be an
expeller, as they appear in the original theorems, have been
improved in various ways.
In particular, the requirements
put on the auxiliary function (or functions) have been

weakened, for example by Kh. I. Ibrachev [1947], N. N.
200
V.
INSTABILITY
Krasevski [1959], V. M. Matresev [1962]1' N. Reuche [1968],

L. Salvaderi [1971], J. A. Yerke [1968].
It is shewn in N.
Reuche [1969] that the upper beund en the

be chesen as a functien .of
functiens can
t.
Case a) .of the preblem .of an equatien .of nth .order

dealt with in Sectien 6 was already treated by D. W. Muller
[1965] with mere stringent regularity hypetheses.
auther censiders alse cases b l ) and b 2 ).
The same
These results have
been generalized, in M. Laley [1973]3' te the case .of a nen

autonemeus beunded secend member.
n
= 2m
fer
In this paper, the case
m .odd is alse censidered.
The necessity .of a
beunded secend member was dispensed with in B. D'Onefrie,

R. Sarne and M. Laley [1974].
At last in M. Laley [1974]2'
the hypethesis .of an iselated reet fer
.one censiders secend members depending en
is weakened and
t
and en all
ceerdinates.
The instability .of the betatren is established in M.
Laley [1974]2'
The example .of Sectien 8 has been inspired by
J. D. Schuur [1967].
Many mere references, including the
centributiens .of the Alma-Ata scheel, appear in V. M.

Matresev [1965].
CHAPTER VI
1.
1.1.
Introduction
Up to this point, we have studied a small number of con-
cepts such as stability, attractivity, asymptotic stability,

etc.
They all pertain to the origin of
~n
for a differ-
ential equation
x=
f(t,x)
(1.1)
with a continuous second member defined on

values in
~n
(the symbols
ing here as in Section IV.2).
and
n with
have the same mean-
An exception was the orbital
stability, as described in Section I.l.4:

the stability of a set.
I x
this is properly
It appears that the attractivity or
asymptotic stability of a set are natural concepts fitting

many practical applications:
in Chapter VII.
several examples will be seen
But as soon as a set, and especially an un-
bounded set, is substituted to a point, namely the origin,

the number of potentially useful concepts increases rapidly:
VI.
2U2
in fact several types of "uniformities" with respect to spatial

coordinates have to be considered in addition to the familiar
uniformity with respect to
to
(the one which makes the dis-
tinction between stability and uniform stability).
The num-
ber of concepts increases to such an extent that one may well

wonder which ones should be studied, and which
ones shouldn't.
The situation appears even worse when observing that some

other qualitative concepts such as boundedness, dissipativity,
. possess also many variants corresponding to different
types of uniformities.
The purpose of this chapter is to
throw some light, and, if possible, to produce some order in

this forest of concepts.
The interest of such an undertaking
is partly theoretical and, as will be
seen, it can be carried
through only at the expense of introducing some new extremely

concise symbolism.
This is why, from Section 4 below to the
end of the chapter, the familiar aspect of our statements and

formulas suddenly changes.
But except for Section IX.6, the
rest of the book can be read without knowledge of this new

language.
If he is interested mainly in the applications,
the reader can skip Sections 3 to 6 and only cast a glance at

Section 7.
The expression qualitative concept designates hereafter any concept such as stability, attractivity, boundedness, etc.
"Qualitative" is meant as an acceptable substi-
tute for "stabilitylike":
both adjectives have a rather
loose signification, but as they are, they will be found useful.

The qualitative concepts studied below pertain to
bounded or unbounded closed sets.
An important feature is
1.
203
Introduction
n,
that these sets are considered as subsets not of
Q:
but of
in other words, they can overlap the frontier of
n.
Several examples in Chapter VII will show why and how this is
relevant to many a practical application.
For the sake of
brevity, we do not consider time-varying sets:
on this point
see T. Yoshizawa [1966] and P. Habets and K. Peiffer [1971].

For the same reason, we decided to leave outside the scope of
this chapter any qualitative property of a set "with respect
to another set".
Thus partial stability, and similar "part-
ial" properties, will be absent because, as is known, they

are irreducible to any kind of stability of a set (see K.
Peiffer and N. Rouche [1969]).
The terminology adopted is, as far as possible, in
agreement with that of H. A. Antosiewicz [1958], J. L.
Massera [1956-1958], T. Yoshizawa [1966] and N. P. Bhatia
and G. P. Szego [1967].
1.2.
Our general hypotheses in this chapter are those of
Section IV.2, with the exceptions that uniqueness is assumed

throughout and that the origin is no more necessarily a point
of
n,
nor a critical point for Equation (1.1).

x(t;to'x O)
the expression
As usual,
will often be abbreviated as
x(t).
The following notations will remain in force throughout the chapter:
any
M is a closed non-empty subset of
0 > 0,
B (M, 0)
Remember that
{x E iJln:
Mo
{x E n:
Mo \ M.
d(x,M)
d(x,M) < o},

d(x,M) < o} ,
is the distance from
to
M.
Q.
For
204
VI.
2.
2.1.
A View of Stability and Attractivity Concepts
Table 6.1 exhibits the two stability definitions we
are interested in.
Definition
~xt-sl ("tEJ+) x (t)EM
("tOEI) ("e:>0) (30)0)
Sl
M is
("e:>0) (30)0) ("tOEI) ("xt-M ("tEJ+) x (t)EMe:
S2
Table 6.1.
In this table, S
pendent of
a)
to'
means stable, US
uniformly stable.
Definitions
If
The
can be chosen inde-
Two remarks are appropriate here:

Sl
and
S2
ent to those obtained by writing

b)
U S
Stability concepts.
adverb uniformly obviously means that
are respectively equival-
Mo
instead of ~.
M is uniformly stable, M
sely, if Equation (1.1) is autonomous and

M
e:
is uniformly stable.
later, mutatis mutandis,
is stable.
M
Conver-
is stable, then
This statement will be extended

(cf. Section 4.10) to periodic sec-
ond members.
2.2.
Six types of attractivities will be considered, as
listed in Table 6.2.

A:
toUA:
Some common denominations are:
is an attractor,
M
is a
to-uniform attractor,
EA:
M is an equi-attractor.
UA:
is a uniform attractor.
Thus, to-uniformity corresponds to
and
(J
independent of
to' the prefix equi to uniformity with respect to
x o ' and
A6
AS
A4
A3
A2
Al
"
"
"
"
"
(35)0) ("tOEl) ("e:>O) (30)0) ("XOEN)
(35 >0)
Table 6.2.
Attractivity concepts.
("e:>0) (30'>0) ("XOE~) ("toEl)
"
"
"
"
("e:>0) (30)0) ("XOE~)
("tOEl) (35)0)
"
"
"
(35)0)
("e:>0) ('tIxOE.~) (30'>0) ('ttOEl)
"
"
(to+OEJ) and ("t~tO+O,tEJ) X(t)EMe:
"
("e:>0) ("IXOEN) (30)0)
(35)0) ("tOEl) ("e:>0) <"XOEN) (30)0)
(VtOEl) (35)0)
Definition
is
UA
EA
to UA
V1
'<
rt
.....
<:
.....
rt
()
rt
"i
III
rt
III
P-
III
::s
'<
J-"
rt
f-'
J-"
cr
III
CJl
rt
206
VI.
uniform without any further qualification to the strongest
o and
concept:
cr
both independent of
to
and
We
notice the following:

c)
If
Al
to
A6
Mo
to .91.
d)
is
nn is positively invariant, definitions
are equivalent to those obtained by substituting
The diagram of implications presented in Fig. 6.1
obvious.
Fig. 6.1.
t6
is
A3
.A2 _ Al
t4
Implications between attractivity concepts.
If Equation (1.1) is autonomous, every horizontal arrow may be

reversed and the number of distinct concepts reduces to
2.
The extension of this property to periodic equations is considered in Sections 4.11 to 4.14:
e)
it is no trivial matter.
To each attractivity concept
ponds a global attractivity concept

stituting
2.3.
(tIo > 0)
to
Ai' there
corres-
(GA.), obtained by sub~
(3 0 > 0).
The following observations pertain to the concepts de-
fined in both Tables 6.1 and 6.2.

f)
Men
If
is compact and
is chosen such that
B(M,) en, every solution mentioned in definitions

and
Al
to
g)
A6
If
exist for all
Men
Sl' S2
to.
is a closed set with a compact boundary
n = ~) and
(for instance the complement of an open ball in

is negatively invariant, and if
tion mentioned in definitions
B(M,} C
n,
then every soluto
exists
3.
Qualitative concepts in general
for all
2.4.
x E
Al
to
0
Along with I.2.12 and I.2.13, the following exercises
show that
even if
> t
2.5.
and
not.
are meaningful and distinct concepts,
For the equation
-x/(l + t), with
t > 0, the origin verifies
Exercise.
x E!if
A6
M is compact.
Exercise.
~
207
and for
x = (1
For the equation

M
{-U
AS' but not
{+U, A3
A3
+ x)2(1 - x)
is verified but
with
A4
is
Show in addition that the same conclusion holds for
Example I.2.7 if
2.6.
Exercise.
with
x E!if
2.7.
Exercise.
and
M is the origin.
For the equation
t > 0, the set
Let
equation where, for
x = f(t,x)
t > 0
is defined as in I.2.l3 for
(2)
f(t,x)
(th-lx) (1 - x2)/t for
(3)
f(t,x)
1 - x
3.
3.1.
verifies
be a scalar differential
M = {a} lJ {1}
(1 + x)2(1 - x)/(l + t)
M = {-I} U {+I}
(1)
Then
for
<
0,
0 < x < 1,
1 < x.
verifies
Al
but not
A2
nor
A4 .
Qualitative Concepts in General
Tables 6.1 and 6.2 exhibit some kind of logical struc-
ture common to all concepts of stability and attractivity.

This structure might be roughly depicted as follows:
given concept is applicable to a set
SIll C
M if, given a set
described by a sequence of quantified variables (exis-
tential or universal) and a set
56 = ME' the solutions start-
VI.
208
ing from .s remain in
!j

in the future, or enter
!j
after
a given lapse of time and remain in it thereafter.

Now we want to play the following game:
specified the sets .Qf! and
!j
after having
in some reasonable way, to
generate a maximum number of concepts by interchanging the

quantified variables in every possible manner.
The result
will be a huge number of concepts, and it will make sense to

try some classification, to recognize the most familiar
amongst them and to wonder why some aroused interest and many
others did not.
The game will be considerably eased by reducing our symbolism to as few letters and signs ,as possible:
D. Bushaw [1969].
we follow
By doing this, we deviate from the usual
mathematical symbolism, but it will be realized that the conciseness thus obtained more than compensates for the disadvantage.
3.2.
Definition of a general qualitative concept.
A set
will be said to verify a qualitative concept with respect to

Equation (1.1) if the following proposition
(3.1)
is true.
The rest of this section is mainly devoted to defin-
ing the symbols in (3.1).

(1)
W is an abbreviation for a word whose letters
are borrowed from the following alphabet

S,R,TO'=O}'
{d,e,s,r,TO'SO,D,E,
Each letter represents a quantified variable, as
shown in Table 6.3; lower-case letters correspond to the
3.
209
existential quantifiers, and upper-case letters to the

universal ones.
As compared with Tables 6.1 and 6.2, Table
6.3 contains a new variable, namely

sample interval.
p, the upper bound of
It will be needed below to cover the cases
of weak attractivity and boundedness.
Name of the variable

Estimate of the ini-
tial perturbation
Estimate of the perturbation at time
Delay of sample
Associated
letters
Variable
interval
Upper bound of
sample interval
Initial time
to
Initial position
Xo
I
I
Table 6.3.
30 > 0
Yo > 0
3E > 0
"IE > 0
3a > 0
Ya
3p > a
yp > a
'0
3t o E I
TO
"Ito E I
~O
3xOE N
=0
YXOEN
~= Mo \
M.
W is subject to the following obvious con-
straints:
(i)
->
Variables for a qualitative concept.

I = 1<,00[,
The word
Meaning of
these letters
no letter is repeated in
w;
any letter may
210
VI.
appear as lower- or upper-case, not both,

(ii)
a variable used in the definition of the domain

of another variable must appear, in the word,
before this one:
for instance, s
appear before
or
or
has to
R.
A word verifying these two conditions will be called wellformed.
An example of a well formed word is
('fie> 0) (30 > 0) ('fIx O E Mo \
M) ('flo> 0).
It
will be written
TOEd3 0 S.
Let us now proceed with the explanation of the symbols
appearing in (3.1).
As will be seen, the meaning of
A will be different according as
and
ent or not in
W, and according as
(existential) or
(universal).
and
y,
pare pres-
appears under
Further, B will admit two
values also, but this time independently of what precedes in

the proposition.
(2)
and for
if
or
The asterisk
() J)
and of course
T, where
T
stands for
for
('f;ft
(3t E [to + a, to + p]
E [to + a, to + p] ()
does not appear in
parentheses above should be replaced by

is absent from
II
II
S, and for the
A can be freely chosen as represent,
Notice however that if
W,
in every other case.
"&"
The letter
appears in
stands for the implication
W contains the upper-case letter
ing
(X) [
if
if it does not.
logical conjunction
(4)
"to + a E J"
stands for
"to E J"
(3)
"~"
In detail, we assume the following
O.
W, cr
J).
in the
Similarly, if
W, "to + p]" should be replaced above by
3.
(5)
means
The letter
The symbol
S+
stands for
= Me:"
"x (t) E ~
211
S-
and
S-, where
or
means
"x (t) E.9J
corresponds to stability and
=n \
Me:".
to instability.
The meanings of the various symbols have to be kept in

mind to understand what follows.
They are summarized in
Tables 6.4 and 6.5.

3.3.
Remark.
To every concept
there corresponds another concept

wri ting N'
with
AV(M)
constructed as above,
C'
derived from
instead of ...Q(, where J4f'
an open neighborhood of
M.
M)
(.A"(M) \
by
n,
(J
Such concepts are
studied in detail by N. P. Bhatia and G. P. Szego [1967].

They are associated with the prefix "semi".
is existential and
is verified, so is
shown that, if furthermore
Clearly, if
C'.
It can be
is closed with a compact boun-
dary and if
Xo
is quantified universally, C
equivalent.
For lack of space, C'
and
C'
are
concepts will be con-
sidered outside the scope of our study.

3.4.
Remark.
Consider the following sequence of parenthe-
ses:
("to E I) (38 > 0) ("x o EN) (:Ie: > 0) ...
Normally, 8
varies with
to
and
e:
with
to
and
is conceivable however to imagine a concept where

depend on
x o ' but not on
the scope of this study.
to'
xo.
e:
It
would
Such a feature is also beyond
It has been considered by C.
Avramescu [1973].
3.5.
Some helpful familiarity with the symbolism introduced
above may be gained by transcribing a few well-known
212
VI.
Symbols
Particular
symbols
Conditions
Meaning
to E J
&
't
i f r or R appears in
otherwise
T
3t E [to'<X>[ nJ
i f r or R appears inW 'ft E [to,to+pj nJ

otherwise
x(t)E!j= Me:
Table 6.4.
x(t)E!j= rl \ Me:
Meaning of
* (A,B) if neither s nor S appears

in
Symbols
'ft E [to'<X>[ nJ
B+
Particular
symbols
W.
Conditions
Meaning
to + a EJ
*
A
if s appears in W
&
if S appears in W
i f r or R appears inW
't
otherwise
T
ifr or R appears in
otherwise
B+
Meaning of
3t E [to+a ,to+pj nJ
3tE[t O+a,00[ (\ J
'ft E [to+a ,to+pj n J

'ft E [to +a ,00 [ n J
x (t) E !j = M
e:
x (t) E ~ = rl \ M
e:
B
Table 6.5.
3t E [to,tO+pj n J
* (A,B) if s or S appears in W.
3.
definitions into the new system.
213
This is done in Table 6.6.
In order to make the comparison easier, the formulas in the

second column have been given in a "complete" form, i.e. using
every variable.
The simplified formulas of the third column
are obtained by using one or more of the equivalence theorems

of the next section.
3.6.
Exercise.
If a concept
C1
derives from a concept
C2
by moving a parenthesis with a universal quantifier to the

right, or with an existential one to the left, or by substituting a universal quantifier to an existential one, then
214
VI.
Terminology
Formula
Simplified formula
1. M is uniformly
stable
2. Solutions uniformly
bounded with respect
DeSRToBO,y*(T,S+)
to M
locally bounded with edSRToBo,Y*(T,S+)
respect to M
4. M is a uniform
attractor
5. M is a uniform global attractor
ultimately bounded
with respect to M
7. M is a uniform weak
attractor
8. M is a uniform global weak attractor
~.
M is unstable
Table 6.6.
Some examples of concepts.
4.
Equivalence theorems
4.
4.1.
215
Equivalence Theorems for Qualitative Concepts
Various examples of two different, but essentially
equivalent formulations of a concept are given in Table 6.6.

Such equivalences, proved by purely logical considerations,
will be established in the present section.
But it happens
also that two concepts are equivalent if some further assumption is made concerning the differential equation (for
example that it is autonomous, or periodic, or that its solu-
tions are continuable up to
00)
or the set
that it is compact, or has a compact boundary).
(for example
The present
section deals also with such conditional equivalences.

The only two propositions of the first type which we
shall give are the following ones, and they are obvious.
4.2.
Theorem.
If
WI
and
W2
are two well-formed words
obtained from each other by permutation of two adjacent lowercase (upper-case) letters, the corresponding concepts are
equivalent, Le.
[WI' y
4.3.
Theorem.
(a)
If
If
(A,a))
[W 2 , Y
(A,a)).
is a well-formed word, then
WI s W2
WI s W2' y
WI SR W2
(T, a)
0=>
wl w2 T a.
is a well-formed word and if
W2
contains only
universal quantifiers or is empty (contains no letter), then

(b)
(c)
WI SR W2 'Y
WI Sr W2 'Y
(d)
WI SR
(e)
WI sR
*
*
w2 'Y *
w2 ,y *
(T, a)
WI W2T13;
(T,13)
WI W2 T13;
,13)
W1W2T13;
(T,13)
WI s
(T
w2 ,y *
(T, 13)
VI.
216
4.4.
Exercise.
where
W2
Prove a dual proposition to 4.3 for the case
contains only existential quantifiers or is empty.
Let us now proceed to a few theorems of the second type,

considering successively solutions which can be continued up
to
+ 00, autonomous equations, periodic equations and, fin-
ally periodic equations associated with a set

compact boundary.
M having a
Most proofs are simple and will be either
omitted or merely outlined.
However, in order to illustrate
clearly the type of proofs which are relevant in this context, Theorems 4.10 and 4.11 are demonstrated in detail.
4.5.
Theorem.
If every solution of the differential equa-
+ 00
tion can be continued to
and if
W is a well-formed
word, then
[W, y
4.6.
Theorem.
If
(A, S)
~
<==*>
[W, A, S J
[0,00[, M is a compact set, every
solution of the differential equation can be continued to

+ 00
is a well-formed word, then

(for
to
0)
(4. 1)
i f and only i f
WI (~to ~ 0)DW2(~xO E Mo )W 3 ,y * (A,S+).
Proof.
It is sufficient to prove the "only if".
(4. 2)
From the
continuity of the solutions, the continuability hypothesis

and the fact that
M is compact, one deduces that

(4. 3)
(4.1) can be rewritten as
4.
217
(4.4)
Combining (4.3) and (4.4), one gets
(4.5)
Indeed, let us in
WI
choose the universal variables arbit-
rarily and the existential ones as for (4.4).

to
and
value
arbitrarily.
Then we adopt for
coming from (4.3).
Choose then
n
in (4.4) a
Further, we choose in
W2
the
universal variables arbitrarily and the existential ones as

for (4.4).
Then, for every
x(O:to'x o ) E Mn'
Xo E
M~,
But then by (4.4):
(4.3) yields
W3 ,y
Xo =
(A,x(t;O,xO)E
ME)' which completes the proof of (4.5), since
x(t:O,x O) =
Q.E.D.
Remark.
The same theorem holds for
TO
located anywhere in
the word in (4.2).

4.7.
Theorem.
If the differential equation is autonomous
and if the word
where
WI
or
WlTOW2
W2
is well-formed, then
can be empty.
The proof, which is left as an exercise, is based on

the fact that, if (1.1) is autonomous, one has for every
and
(1)
t' E I
t' E J(to'x O)
J(to'x O) ;
that
is equivalent to:
t' + t
o-
t' E
0
VI.
218
Corollary.
4.8.
and i f
If the differential equation is autonomous

is well-formed, then
If (1.1) is
w # 0, obtaining re-
w-periodic for some
sults similar to 4.7 and 4.8 requires some further hypotheses

on
M, W1
or
W2 .
The following lemma is essential in de-
riving such results.
By way of exception, it mentions an-
other domain than
for the variable
4.9.
Lemma.
with
w > 0, if
exists a
Y
to
(A,S)
to'
If the differential equation is

WITOW2
w-periodic
is a well-formed word and if there
such that the concept
is verified, so is
Wl <Vto E [to,tO+wJlW2'
WI T OW2 ,y
(A,S).
This can be proved by observing that, for all

and any integer
(i)
m, if
t' E J(to'x O)
to E I
to = to + mw E I, then:
is equivalent to
t' + t
- t' E
0
J(to'x O);
The following results concerning periodic differential

equations pertain to the concepts
Sl' S2
and
Al
to
A6
of Section 2, i.e. to the various kindsof stability and attractivity.

4.10.
Theorem.
If
Men
is closed with a compact boundary
and is positively invariant, if the differential equation is

w-periodic and if each solution
is defined at least on
x(t;to'x O)
with
Xo E aM
[to,t o + w], then stability (S1) is
4.
219
equivalent to uniform stability (8 2 ),

Proof.
It is SUfficient to show that
Choose any
to E I
a value of
E > O.
and an arbitrary
suiting
The general solution
81
where
to
x(t;to'x o )
Let then
is replaced by
<5
to + w.
01 > 0,
to E [to,t o + wJ, every
xo E B(M,ol) \ M and every
t E [to,t o + wJ: x(t;to'x O) E
B(M,OO)'
81
But then, because
for every
to + w.
is verified, x(t;to'x o ) E ME
Lemma 4.9 then shows that
82
verified.
4.11.
If the hypotheses of Theorem 4.10 are satisAl . . . A2
and
Consider an arbitrary
Proof.
is also
Q.E.D.
Theorem.
fied, one has
A4
to E I
value of
suiting A4 for to
in the proof of 4.10, a 1 with
AS
A6
a
0 be
One may find, as
and let
to + w.
B (M, (51)
en
such that
nfxo E B(M,ol) \ M) ("to E (to,t o+ wJ)

x(t o + w;to'x o ) EB(M,OO)'
We then choose an arbitrary
E > 0
and determine a value of
(J+ w).
A
(in A6 , we shall adopt the value
4
To the possible values (4.1) of to and x O' there corres(J
suiting
ponds initial values

A6
is verified with
[to,t o + w))
4.12.
be
being continuous with res-
pect to the initial conditions, there exists a

B(M,Ol} en
8 2,
implies
81
x(t o + w;to'x O)
("to E I)
suiting
replaced by
and the thesis follows from 4.9.
Exercise.
A4
Hence
nftO E
Q.E.D.
Prove the other implications appearing in
the statement of Theorem 4.11.
VI.
220
4.13.
The fo11o~....ing example shows that, even in the hypothe-
ses of Theorem 4.11, A2

autonomous system in
does not imply
A3
Consider an
~2, the orbits of which are represented
in Fig. 6.2.
Fig. 6.2.
Phase plane for Equations (4.6).
We are interested in the equations of

written in cartesian coordinates.
this system
Let us however consider
these equations written as follows in polar coordinates:

r
R(t,8)
(4.6)
1/I(r,8)
with
and
1/1 2n-periodic in
8.
It will not be necessary
to specify these two functions any further.

verifies
A1 .
(4.6) into a
The transformation
= r, a =
2n-periodic differential system.
ponds to a rotation of the plane around

city equal to
The origin
1.
For the new equations, A1
8 + t, changes
It corres-
with angular ve1oand
A2
are
4.
221
clearly satisfied whereas

E >
and
is not.
Indeed, given any
0 > 0, one has for each point
that sup {aCtO): to E [0,2n]}
ro ~ 0
Exercise.
4.14.
A3
(ro,60 )
E Bo
with
00.
Suppose we try to use a reasoning similar
to that of 4.11 to prove that, under the hypotheses of 4.10,

A2
4.15.
Where shall we fail?
A3
In Theorem 4.5 and in several other theorems to be
proved in the rest of this chapter, there appears the hypothesis that every solution of (l.l) may be continued up to
00.
Sufficient conditions to get this are given by the following

lemma.
Lemma.
4.16.
(i)
(ii)
(iii)
Consider a concept such that

does not appear in
W;
is a universal variable, or
WlS;
13+ ;
suppose further that
M, either is compact, or is closed with
compact boundary and negatively invariant; if moreover

chosen such that
B{M,E)
00.
Hypotheses (i) to (iii) imply that
* (A,B)
has to
be written in one of the three following explicit forms:

(l )
where
(2)
(t 0
+ a E
J)
&
[('It E
[t 0 +
a, 00 [
fl J)
x (t)
E ME]
can vanish, or
(to + a E J)
=;>
[('tt E [to + a,oo[ flJ) x{t) EM E ]
or
(3)
(to + a E
J) ~
is
en, then every solution mentioned
in this concept can be continued up to

Proof.
[(3t E [t o+ a,oo[ fl
J)
x{t) EM E ].
222
VI.
In the first two cases, x(t}
in
if
B(M,E)
t > to + a, t E J,
remains, for
M is compact or in
B(M,E) \ M otherwise.
Both sets are bounded and do not touch
an.
In the third case, suppose a solution
x(t)
the right maximal interval
implies that
B(M,E) \ M)
x(t)
and never comes back.
x(t) ~ ME'
is defined on
for some
ultimately leaves
to + cr E J
such that
[to,t*[
Hence the thesis.
t* E~.
B(M,E)
(or
So there exists a
and for every
This
>
to + a, t E J:
This contradicts (3).
Q.E.D.
A Tentative Classification of Concepts
5.
Let us attempt to classify the concepts whose word
is made up of letters borrowed from Table 6.3.

ing the definitions of the sets N
and
~,
While retain-
we generate, by
mere sUbstitutions and germutations, 46.080 formally different concepts, all well-formed!
out all uninteresting items.
Hence the necessity to rule

We propose the following rea-
sonab1e restrictions:
(1)
to
fier:
and
Xo
will always be preceded by the same quanti-
one is generally interested in the behavior either of
all solutions starting near

(2)
and
M, or of only one;
will appear before
a:
the reason is that we
normally want to fix an estimate of the initial perturbation

and then choose a sample delay to satisfy the final perturbation.
Respecting both conditions reduces to 5.376 the number
of formally different concepts.
and
Indeed, for
0, E, x O' a
P, the only possibilities are those listed in Table 6.7.
5.
A tentative classification of concepts
II
III
X
0
IV
Xo
Xo
Xo
Xo
Xo
Xo
223
Table 6.7. Admissible orders for the variables

As
to
may be inserted at any place, one gets
possibilities.
= 42
With all possible choices of quantifiers for
(to'x O)' 0, E, a, p
8+
0, E, xo,a,p.
and
t, and the binary choice between
and 8-, one gets the announced 5.376 well-formed concepts.

They can first be grouped as suggested in Table 6.8
in stability-like
ties concerning all
M.
(8+)
or instability-like
(T O,3 0 )
or one
(LO'~O)
(8-)
proper-
solutions near
Four families are obtained, each of 1.344 concepts.
VI.
224
1. Stability-like concepts
TO~OS+
2. Instability-like concepts
TO/;OS
3. Complete instability-like concepts
TO!:!OS
4. Incomplete stability-like concepts
TO/;OS
Table 6.S.
Fundamental families of concepts.
Each family can be divided into natural equivalence classes,

two concepts being considered "equivalent" if they differ
only by the positions of
and
in
W:
in this case,
they do not differ by more than some spatial or time uniformity conditions.
This divides the 1.344 concepts into 32
classes of 24 concepts, plus 32 classes of IS concepts

(Exercise:
why 24 and why IS?).
At last, in any listing of
concepts, advantage can be taken of the partial order induced

between them by the implication relation:
a concept will be
said to be stronger than another one if the former implies the

latter.
Let us now examine the important family of stabilitylike concepts.
In Table 6.9, each class of the family is
represented by its maximal element in the sense of partial

order:
TO
and
=0
occupy the utmost right position.
The
table begins by the strongest concepts, so strong indeed as

to be meaningless, and it ends with the weakest ones, so
weak as to be trivial.
It should contain 64 rows, but it has
only 42 because the following simplifications have been introduced:

(I) owing to Theorem 4.3, SRT, SrT and SRT
are equivalent:
5.
A tentative classification of concepts
Formula
1.
2.
3.
4.
5.
6.
7.
e D
d E
D e
E d
d e
e d
8.
9.
10.
11.
12.
13.
14.
e D
d E
D e
E d
d e
e d
15.
16.
17.
18.
19.
20.
21.
e D
d E
D e
E d
d e
e d
22.
23.
24.
25.
26.
27.
28.
e D
d E
D e
E d
d e
e d
29.
30.
31.
32.
33.
34.
35.
e D
d E
D e
E d
d e
e d
36.
37.
38.
39.
40.
41.
42.
e D
d E
D e
E d
d e
e d
E D S R
"
"
"
"
"
"
E D s R
."
"..
E D s r
"
"
"
.."
E D s R
."
..
.
.
"
"
"
"
.
.
.
."
."
"
"
"
"
TO ='0' Y
."
"
"
.
..
.
"
..
.
Table 6.9.
solutions around M are boundec

M is stable
(locally bounded)
(locally bounded)
M is a global attractor
(ultimately bounded)
M is an attractor
"
"
"
"
"
"
."
meaningless
."
.
"
"
.
."
"
"
"
.
.
"
"
"
(T, S+)
.
.
.
..
E D s r
"
"
.
.
E D S r
Terminology
TO ='0' Y
"
."
225
* (T,
.. s+)
"
"
".
"
"
"
"
"
"
T
."
.
.
"
"
M is a global weak attractor

(weakly ultimately bounded)
M is a weak attractor
trivial
."
"
Family of stability-like concepts.
226
(2)
VI.
and
and
e
D can be interchanged.
because
and
TO' or
This is not true for
EO' or both, may appear between
d.
Table 6.9 also gives the common denominations of some concepts.
A point of divergence with previously known defini-
tions is weak ultimate boundedness which, for N. Pavel [1972],

corresponds to number 10 and for us to 37.
For lack of space,
the adverb "uniformly" has been omitted throughout in Table

6.9.
Similar tables can be set up for each of the four families of Table 6.B.
For example, Table 6.10 presents, for
instability-like concepts, the ten definitions corresponding

to the first ten of Table 6.9.
Here, each class is repre-
sented by its minimal element in sense of partial order, so

that each concept of Table 6.10 is the negation of the corresponding one of Table 6.9.
Formula
Terminology
-)
l.
s r
2.
"
"
"
"
3.
"
"
"
"
4.
"
"
"
(unbounded)
5.
"
"
"
M is unstable
6.
"
"
"
7.
"
"
"
B.
S r
"
"
9.
"
"
"
10.
"
"
"
Table 6.10.
'0 ~O' y
(" S
trivial
Some classes of instability-like concepts.
6.
Attractivity, boundedness
6.
6.1.
227
Weak Attractivity, Boundedness, Ultimate Boundedness

The purpose of this section is to examine in some de-
tail those classes of Table 6.9 which aroused most interest

amongst scientists.
To avoid useless repetitions, we use of
course the equivalence theorems of Section 4.
Stability and
attractivity has been studied already and were presented in

Tables 6.1 and 6.2.
The variable
is absent there in vir-
tue of Theorem 4.3 (b) and (e) respectively.

ty is presented in Table 6.11, where
Weak attractivi-
does not appear
owing to Theorem 4.3 (a).

Formula
d
WA 2
WA3
WAS
WA6
WA l
WA 4
TO
TO
TO
TO
=0
<,8+
=0
<,8+
=0
TO
=0
<,8+
=0
<,8+
=0
Table 6.11.
a weak attractor
<,8+
TO
is
<,8+
an equi-weak attractor
a uniform weak
attractor
Weak attractivity.
The implications between these concepts are those of Fig. 6.3,

where the horizontal arrows can be reversed if Equation (1.1)
is autonomous.
Fig. 6.3.
Implications between weak attractivity concepts.
VI.
228
6.2.
Theorem.
If
MU
n is a neighborhood of
M and if
is closed with a compact boundary, then

(a)
weak attractivity
tivity
(b)
Proof.
WA 2
(WA1 )
equi-weak attrac-
(WA 4 ) :
WAS'
We prove the first equivalence only:
the reasoning
would be substantially the same for the second one.
Only the
direct implication has to be established, namely
Suppose
WA l
is satisfied and
WA 4
is not, i.e.
(6.1)
Choose a
to
suiting (6.1) and then a
Mo C n U
moreover that
in (6.1) and an
e*
M.
Choose some
x Oi E Mo * \ Me *.
sequence
hOi}
Let us use this
+
J (to'x O)
Pi +
such that
0* < 0
in
WA1
00
Mo * \ Me *
Since
has a cluster point

Xo
as for
WA 1 , such
to be used
suiting (6.1) in these conditions.
by (6.1), there exists a sequence

points
and a sequence of
is bounded, the
e* C Mo
Then there exists a tE
x (t) E Me *.
Then
Xo E Mo * \
\ M.
By continuity of the solu-
tions with respect to the initial conditions, one gets, for

i
large enough, that
and since, for every

x(t;to'x O)
i, x(t;tO'x Oi )
Me *, then also
Me *, which is a contradiction.
Q.E.D.
6.
229
6.3,
Exercise.
If
is a neighborhood of
M un
.....
is w-periodic, WA 2
M and
WA 3
If
M has a compact boundary,
is w-periodic, WA2 ~ WA 6 I f Men has a compact boundary and is positively invariant and if, for Xo E ClM,
and if
x(t;to'x o )
WA1
WA2
If
[t o ,t O+w1, then
does not depend on
and if
M satis-
fies the hypotheses of Theorem 6.2, all concepts of weak

attractivity are equivalent.
6.4.
Remark.
An alternative way to introduce a class of
concepts also likely to be grouped under the heading of "weak

attractivity" is to consider the variations of
The following relation between both kinds of concepts is

proved in P. Habets and K. Peiffer [19731:
tive1y invariant and if
WEr
and
WES
if
M is nega-
are well-formed, then
The following theorem links attractivity and weak attracti vi ty,

6.5.
Proof.
Theorem.
Each
Ai
A.
J.
.::::::;.
WAi ,
1 < i < 6.
has the form
w1 sW2 , [to + o E J]
and
[Vt E [to + o,co[
n J ,13 + ],
W1 and W2 are well-formed and do not contain p.

But this implies that w1 sW 2 , (3t = to + 0 EJ)I3+ and, taking
+
p = 0, that w1 rw 2 , (3t E [to,t o + p] nJ)13 , which is the
where
corresponding weak attractivity concept.
Q.E.D.
VI.
230
6.6.
Boundedness, ultimate boundedness and weak ultimate
boundedness concepts are presented in Tables 6.12, 6.13 and

6.14.
They are studied here for their own interest and further
because they lead naturally to the important notion of a dissipative system (cf. 6.18 and Chapter VIII).
The solutions are

with respect to M
Formula
Bl
TO
B2
B3
B4
TO
D ::0
D ::0
T ,13+
bounded
T,13+
bounded uniformly in to
::0
T,13+
equi-bounded
::0
T,13+
uniformly bounded
TO
TO
Table 6.12.
Boundedness.
...
The solutions are

with respect to M
Formula
-0
(T,13+)
,y
*
*
TO'y
(T, 13 +)
s ::0
,y
(T,13+)
(T,13+) uniformly ultimately

bounded
,y
UB 2
e D TO ::0 s
PB3
e D
::0 s
UBI TO e D
UBS
e D TO
s ::0
,y
*
*
UB 6
e D
s ::0 TO'y
PB 4 TO e D
...
Table 6.13.
(T,13+) ultimately bounded
(T,13+) equi-ultimately bounded
Ultimate boundedness.
6.
231
Formula
WUB l
WUB 2
WUB 3
WUBS
WUB 6
TO
WUB 4
TO
Table 6.14.
TO
~O
T,a+
~O
T,a+
~O
TO
TO
~O
EO
=0
T, 13+
T,a+
TO
T,a+
+
T,a
Weak ultimate boundedness.
OWing to Theorem 4.3 (b), Table 6.12 contains neither

p.
nor
In the same way, and due to Theorem 4.3 (e) and (a) res-
pectively, Table 6.13 contains no
and Table 6.14 no
Some immediate implications are given in Fig. 6.4.
a.
If (1.1)
is autonomous, the horizontal arrows can be reversed.
or
j"
UB 3 - -.....~ UB 2
Fig. 6.4.
OJ"
--_a UBI
Implications between boundedness and

ultimate boundedness concepts.
6.7.
Exercise.
and
For
M = {O}, show, by using
the following system of equations:

x
y2(1 - xy),
0,
that even in the autonomous case, B2
does not imply
B3
232
VI.
~2, M =
{a}
one does not necessarily have that
Bl
6.8.
Exercise.
Hint:
For
w-periodic,
B2
Use the following change of variables in the
differential system of Exercise 6.7:

6.9.
and
Exercise.
= x,
=Y
Show that for an unbounded set
pact boundary, all concepts
Bi , i
= 1,
+ cos t.
M with com-
2, 3, 4, are trivially
verified.
6.10.
and
Exercise (T. Yoshizawa [1959]).

M
= {a} en =
If
is
w-periodic
~n, equi-boundedness (B 3 ) is equivalent
to uniform boundedness (B 4 ).
The following theorem exhibits some relations between
boundedness and ultimate boundedness concepts.
Proposition
6.11 (b) is due to T. Yoshizawa [1959].

6.11.
Theorem.
If
UB 4
(b)
ultimate boundedness
Proof.
(a)
(B 4 )
Let
to
E [to,t o + sl]}' sl
(1'2)'
in
(b)
= 2
UB
Let
be chosen in
where
(UB 2 )
plus uniform bounded-
I.
0 > 0, B3
If
UB 4
sup {llx(t;to'x o )
is verified
is satisfied with
being the value of
II:
Xo
EN,
corresponding
This proves (a)
correspond to
UB 2 , for every
that
(B 3 );
imply equi-ultimate boundedness (UB 5 ).
then for each
l'
= max
0
equi-boundedness
to
= {a} e n = ~n:
(a)
ness
for
(to'x O)
x(t o + sl;tO'x O)
be chosen as for
0
= l
E
E M.
1
in
B4
UB 2
Then, in virtue of
, there exists an
Let
Nl
and let
sl > 0
such
be a neighborhood
6.
of
X(t o + sl;tO'x O)
neighborhood
N*
such that
of
Xo E N*, and, by
233
Then, for some
Nl C Me:
1
x o ' x(t o + sl;tO'x O) E Nl
for every
B4 , one gets that

(6.2)
Let us write
B(M,S)
say
sl = sl(N*)
and choose
arbitrary.
Then
is covered by a finite number of such neighborhoods,

(1 0) ("10 > 0) ("Ito E I) (30"
which is nothing but
M = {OJ
(b)
implies uniform
(UB 6 )
(B 4 ) ;
to prove that
sin x
with
B4 does not imply
n =!
UB 1
WUB. + uniform boundedness

~
Suppose
e: 2 , where
0
Use the equation
< 6.
Then, due to
to
w-periodic,
(UB 2 )
Theorem.
Proof.
is
M = {oJ
1 < i
implies uniform
Exercise.
6.14.
If
(UB 2 )
boundedness
e:
Q.E.D.
en = ~n, then
(a)
6.13.
EB(M,O
UB S '
Exercise (N. Pavel [1972]2)'
6.12.
and
s~O)("1xo
e: l
WUB.
B4 , UB.
~
e: 2
is satisfied for
e: = e: l
will be satisfied for
is, in
B4 , the value of
and
0" = P1
e:
p = P
1
and
corresponding
Q.E.D.
VI.
234
6.15.
and
Theorem (V. A. Pliss [1964]).

M
{o}
c n
n=
is
w-periodic
~n
(WUB 2 )
(UB 2 ).
Theorem (N. Pavel [1972]).
M = {o} C
~n, then weak ultimate boundedness
implies ultimate boundedness

6.16.
If
If
is
w-periodic,
and every solution can be continued to
+ 00, all concepts of weak ultimate boundedness and ultimate

boundedness are equivalent.
Proof.
It will be enough to prove that
WUB l
implies
UB 6 .
But this implication will be established if we show that

WUB l
WUB 2 , for then, using 6.15 and 6.12 (b), one
implies
gets the chain

then
to
is a
to
WUB l
be fixed in
to + w
01 > 0
in
WUB 2
===#>
WUB 1 .
and let
e:
UB 2
= e: l
=*'
Let
UB 6
correspond to
By continuity, for every
0 > 0, there
such that
Then there exists a t E [to'oo[

is verified with
WUB 2
==9
replaced by
with
x(t) E Me:
- [to,t o +
w].
and
Therefore,
is verified, owing to Lemma 4.9.
6.17.
Exercise.
and every
If
is
Q.E.D.
w-periodic, M
solution can be continued to
WUB 2
{o} C n
~n
+ 00, then all con-
cepts of class 30 in Table 6.9 are equivalent to all concepts

of ultimate boundedness and weak ultimate boundedness.
6.18.
for
An equation like (1.1) will be called dissipative if,

M
{o}, all concepts
of weak ultimate boundedness, of
ultimate boundedness and of class 30 (in Table 6.9) are verified for the equation.
In N. Levinson [1944],
(1.1) was
7.
235
called dissipative if it verified
UB 2
Theorem 6.16 and Exer-
cise 6.17 yield sufficient conditions for an equation to be

dissipative.
The following exercise gives particularly simple
conditions to get the same result.

6.19.
I
Exercise.
[O,OO[
If
is
w-periodic, M
{oJ
en = ~n,
and every solution can be continued to
+ 00, the
differential equation is dissipative if and only if

(3e: > 0) ("x O E ~) (3t > 0)
Hint:
Using Theorem 4.6, one can replace in all con-
cepts of weak ultimate boundedness, of ultimate boundedness

and of class 30,
("to
7.
7.1.
A set
by
E~)
(for
to = 0).
M is said to verify a concept of asymptotic
stability if it possesses some kind of attractivity along

with some kind of stability.
It results from the number of
concepts listed in Tables 6.1 and 6.2 that there are, formally,
12 types of asymptotic stability.
Some of them are equival-
ent however.
A set
(i)
M is said to be
asymptotically stable
and attractive
(ii)
(51)
if it is stable
(EAS)
and equi-attractive
uniformly aSymptotically stable

uniformly stable
(A 6 )
(51)
(Al );
equi-asymptotically stable
stable
(iii)
(AS)
(52)
if it is
(A 4 );
(UAS)
if it is
and uniformly attractive
VI.
236
Global asymptotic stability corresponds to asymptotic

stability like global attractivity to attractivity.
For in-
stance, M is said to be
(iv)
globally aSymptotically stable

stable
(v)
(51)
(GA5)
and globally attractive
if it is
(GAl):
uniformly globally aSymptotically stable if it is

uniformly stable
attractive
(52)
and uniformly globally
(GA6 ).
Let us now prove a frequently used partial converse of

Theorem 6.5.
7.2.
Theorem.
If
then
52 + WA.
J.
Proof.
A., (1
J.
to the sixth one.

WA6
choice
suiting
'.
6).
It is possible, with some care, to prove all six im-
plications at a time.
and
Let us, for simplicity, limit ourselves
More precisely, we shall show that if
are satisfied, then
= 01
in
A6
as in
A6
is satisfied with the same
WA6
We first choose a
WA6
and then select, to prove
Let
02 < 01
in
correspond to
WA6 , we get that for
'
in
01'
52'
= 02
5electing
and some
(7.1)
x(t ) E Mo .
a
2
ta
can be chosen such that
= 01
A6 , an arbitrary
(Vx o E MOl \ M) (Vt o E I) (:Ita E [to,t o + p'])
Clearly
52
x(t a )
M and so
7.
x (t ) E Mo \
a
2
= ',
01
237
But then combining (7.1) and
M.
0 = O2 , to = t a , Xo
and = '
= x(t a ,
52
(with
one gets that for
\ M) ('It E [t ,""[)
a
Finally, since
and
cr = p
ta
to + p, one obtains for
that
(lito E I) (lIx o E Mo
7.3.
0 = 01 , = '
Exercise.
M) ('It E [t
Prove that, for
i = 1
+ cr,""[)
x(t) E M.
Q.E.D.
or
2, the unique
(continuabi1ity) hypothesis of Theorem 7.2 can be dispensed

with.
OWing to Lemma 4.16, if
MC
either is compact, or
is closed with compact boundary and negatively invariant,

then the (continuabi1ity) hypothesis of Theorem 7.2 can be
dispensed with.
We have already observed that attractivity does not imply stability (cf. I.2.7).
However, we can prove the fo11ow-
ing.
7.4.
Theorem.
attractivity
If
(A4 )
MC
is positively invariant, equi-
implies stability
(51).
The proof is left as an exercise (see Exercise I.2.9).

7.5.
Exercise.
Suppose
is
w-periodic and
MC
either
is compact and positively invariant, or is closed with a compact boundary and invariant.
tion
x(t7to'xO)
with
Suppose further that each solu-
Xo E aM
238
VI.
[to,t o + w).
Then asymptotic stability is equivalent to uni-
form asymptotic stability.

Hint:
Applying Theorem 4.10, one has
Then show that, even if

Al
imply
A6 .
A4"
is not periodic, 52
....
Exercise.
= {o}
52'
along with
Using Theorem 4.11, one gets that
(The case
7.6.
51
A4
has been treated in Theorem 1.2.14.)
If the hypotheses of Exercise 7.5 are satis-
fied, global asymptotic stability is equivalent to global uniform asymptotic stability.

Hint:
One may reason as in 7.5, while changing the
implication "52
ply
and
Al
GA 4 ". The case
8.
imply
= {o}
A4" into "52
and
GAl
im-
appears in T. Yoshizawa [1966].
In his paper of 1892, A. M. Liapunov proposed a fairly

general definition for the stability of a solution of a differential equation.
tion
x*(t)
if the norm
Roughly speaking, he said that a solu-
is stable with respect to some function

I Ih(x(t
I Ix(t o ) - x*(t o) I I
- h(x*(t
II
remains small whenever
is chosen small enough.
the simplest case only:
hex)
=x
hex),
and
x*(t)
But he studied
= o.
Later, K. P. Persidski [1933] introduced the uniformity

in
to' and I. G. Malkin [1954], for asymptotic stability,
the uniformity in
[1957]
xO'
The example due to R. E. Vinograd
(cf. 1.2.7) of an unstable attractor led naturally to
splitting asymptotic stability into two simpler concepts,

namely stability and attractivity.
Later, the notion of weak
8.
239
attractivity appeared as a natural one, when one is trying

to prove some kind of asymptotic stability by using a theorem
like 7.2 (see also I.6.8).
The definition of weak attracti-
vity as proposed here is slightly different from the one of

N. P. Bhatia [1966].
Boundedness was studied in detail by
T. Yoshizawa [1966].
On weak ultimate boundedness, see P.
Habets and K. Peiffer [1973].

The extension of these definitions to sets is due to
several authors:
V. I. Zubov [1957], T. Yoshizawa [1966],
N. P. Bhatia and G. P. Szego [1967].
Time-varying sets were
considered by T. Yoshizawa, and later, in an effort to encompass partial stability, P. Habets and K. Peiffer [1973] mentioned the stability of a set with respect to another set.
An early work in the same direction is A.
N. Michel [1969].
A number of other concepts have been studied, as for example,

exponential stability by many authors, stability in tube-like
domains, stability of conditionnally invariant sets, by V.
Lakshmikantham and S. Leela, etc.
Systematic presentations of the most classical concepts with an effort to unify the terminology appear in J. L.
Massera [1956] and H. A. Antosiewicz [1958].
Rather exten-
sive sets of definitions can be found in the books of V. I.

Zubov [1957], L. Cesari [1959], W. Hahn [1959], [1967],
T. Yoshizawa [1966], N. P. Bhatia and G. P. Szego [1967],
V. Lakshmikantham and S. Leela [1969] and A. Halanay [1966].
A classification of concepts by means of various types
of bounds on the solutions has been presented by W. Hahn
[1967].
A different approach was adopted by D. Bushaw [1969]
24U
VI.
who defines a large number of concepts by combining in

various ways a sequence of quantified variables.
The same
type of symbolism was used later by P. Habets and K. Peiffer

[1976], whose general presentation has been adopted here.
Other contributions of the same type are those of M. Dana
[1972], V. M. Matrosov [1973] and C. Avramescu [1973].
CHAPTER VII
ATTRACTIVITY FOR AUTONOMOUS EQUATIONS
1.
Introduction
Attractivity properties are studied in this chapter

and the next one.
As contrasted with Chapters I to V, the
origin will play no particular role here:
it will not be as-
sumed that it is a critical point of the differential equation

on hand, nor even that it belongs to
Q.
The reason will be-
come clear as soon as we tackle some applications:
the at-
tractivity of a set is frequently encountered as a natural

question.
The autonomous case is treated separately in this chapter for several reasons.
First, strong results can be ob-
tained rather simply in this setting.
Further, an important
use can be made of the invariance property of limit sets:

in the non autonomous case, this property can be only partially
preserved, at the expense of some supplementary hypothesis.
Finally, there exist very good and important examples to
VII.
242
illustrate the autonomous case, whereas this is less true for

the non autonomous one.
Many significant theoretical results
concerning the latter are more recent indeed.
2.
Let
General Hypotheses
n +~,
be some open set of ~, f:
a continuous function and
Xo
n.
a point of
time will always be chosen equal to
O.
x + f(x)
The initial
Thus, our general
Cauchy problem will be:

(2.1)
x=f(x)
Xo
x(O) =
(2.2)
By the way, the uniqueness of solutions is not assumed.

non-continuable solution will be written
we shall often write J = ]a,oo[ and J +
tive limit set of a solution
x: ]a,oo[ +~n, and
[0,00[.
The posi-
will be designated by
A+(x).
The properties of limit sets are studied in Appendix III.
3.
3.1.
The Invariance Principle
The following theorem is important from a theoretical
point of view:
it shows in fact which conclusions remain, in
Liapunov's direct method, when a single function is considered

along with hypotheses as weak as possible, not much more indeed than a negative derivative.
Every inessential considera-
tion being discarded in this way, there remains something like

the essence of the method.
Further, this theorem, which is
known under the name of invariance principle, is amongst the

most useful ones in the applications.
3.
The invariance principle
3.2.
243
Theorem (J. P. LaSalle [196B]).
of (2.1), (2.2) and

tion such that
where
V: n ... ge
D+V(x)
Proof.
x(J+).
If
Ilx (t) II
...
""
{x
En:
as
thesis is trivially satisfied.

A+
rt.
an
an.
D+V(x)
...
Then A+ () n C M
V(t.) ... c
~
x(t i ) ... x *
as
Otherwise, one has either
as
i ... "".
i'" "", and, as
x* E A+()
A+ ()
n.
But
such that
{V(t i )}
is bounded
c E ge,
V(t) ... c
as
t ... w.
is the same for every
In other words: Vex) = c
on
A+ ()
is serni-invariant, D+V(x) = 0
on
A+ ()
the concept of serni-invariance, cf. Appendix III).

A+ () n CE
n.
is continuous, V(x*) = c.
is decreasing, and therefore

x* E A+ ()
{til C J+
Therefore, for some
Therefore the limiting value of

But as
and the
In the latter case, consider some point
from below and decreasing.
point
A+ = q,
and the thesis is again trivially satisfied, or
ti ... wand
= oJ.
"", then
There exists an increasing sequence
But
be a solution
M is the union of all non-continuable orbits, each of
which is a subset of
A+ C
a locally lipschitzian func-
0 on
<
Let
n.
n
(on
Therefore
and the thesis is obtained by invoking again the
semi-invariance of
A+ ()
n.
Q.E.D.
More precise and practically important conclusions concerning the asymptotic behavior of
x(t)
are given in the
following corollary.
3.3.
Corollary.
t ... w, then
If
(J+)
If
x(t)
x(t) ... M U
an
does not approach infinity when

on every compact set when
is bounded, x (t) ... M U
an
when
In this statement, the expression
t ... w.
t ... "".
"x(t)
tends to
244
VII.
M U an
set
FOR AUTONOMOUS EQUATIONS
on every compact set" means that for every compact
K, i f
JK
x (t) .... M U an
3.4.
ATTRACTIVITY
is defined as
{t E J: x(t) E K}, then
for values of
contained in
J K
In connection with Theorem 3.2, it is important to be
able to identify the set
M.
It is often easier to recognize
E \ M, i.e. the largest subset of
no point of which be-
longs to a non-continuable orbit entirely contained in
E.
This is usually possible by considering the vector field

x .... f(x).
But auxiliary functions can be used also as is
shown by the following two propositions.
The first one, which
is rather crude, is obvious; the second one is more refined.

3.5.
proEosition.
neighborhood of
W(x) ':J 0
on
P.
Let
pen
and
W: P'
Then, i f
be compact, p' C n
.... 91 a .sfl
x(t)
an open
function such that
is a solution such that
x (0) E P: x(J+) rt. P.

Unfortunately, E \ M is not often, if ever, compact.
This is because, in many applications, M is compact:
of the simplest case, where
M is a critical point.
think
Of
course, Proposition 3.5 is no more true if the hypothesis of

compacity of
is removed.
In a way, it's a pity, because
one would have liked to draw some useful conclusions from the
very simple assumption that
o.
W(x) ':J
(See however Section
4.4. )
3.6.
n, let
Proposition.
{Pi}
For some sets
nand
{Q,}
J
and
(u P,) . U (U Q,)
~
with
PCP * C
be two sequences of open
sets such that

i
P*
::) P.
3.
Let us write
P*
in
Q!
closure of
= Pj n
p~
Q! = Q. n P. Assume that the

J
J
~n) is disjoint from an. Assume that,
(in
245
and
considered as a metric space, the
Q'j n ap = cp.
disjoint and such that
.lfl
quences of
are pairwise
J
If there exist two se-
functions
Wi: Pi
+ ~
Q~
W~:
Q.
Iw~ (x) I >
ct.
and
+.9R
such that
(i)
on
p' Wi (x) =
(ii)
on
Q'
ct.
> 0;
0, Wi (x) 'I 0;
..
W~(x)
is bounded,
then, for every solution
x(t)
such that
x(O)
for some
E P:
x (J+) rt P.
Proof.
(a)
such that
xeD) E Pi
If
E Pi
x(t)
for every
would get, for all these

fore
W. (x(D
i, there is no
for some
t
t, that
E [O,T[.
Otherwise, one
Wi(x(t
0, and there-
= 0, which is excluded by (i).
open with respect to
T > 0
p'
is
i
leaves P
But
P: as a consequence, x(t)
immediately.
for every
for every
xeD) E
If
(b)
comes out of
E [O,OO[
Q!.
x(t)
is bounded:
E [O,w[, then either
and
x(t)
of
Q~
upon which
touches one of the
Q!
J
P~
remains in
remains in
Q~
x(t)
remains in some
ct.,
whereas
J
J
In the latter case,
Iw~(x) I >
this is a contradiction.
comes out of
x(t)
w = "", !.
and it follows that
In the former case, x(t)
connected component
W.*
for some
Q~
while remaining in
P.
Therefore it
and, as proved above, comes out of
P, which is another contradiction.

The importance of
p*
Q.E.D.
will be better understood if
one knows that in practical situations, the interior of
P*
VII.
246
will often be empty.

W*
that
Notice also, that if one assumes further
is defined and continuous on some compact set con-
taining
.*
w. (x)
Qj' then (ii) can be replaced by:
on
this set.
4.
An Attractivity and a Weak Attractivity Theorem
Corollary 3.3 implies that if
4.1.
bounded away from

other hand, if
x(J+)
this corollary that

theorem
then x(t)
a~,
x(J+)
M
as
is bounded, and
t
is unbounded, there is no indication in

x(t)
might approach some set.
function
fix).
Theorem.
M
= as n
zian functiort
pose that
Let
S C
be closed with respect to
able orbit.
and
Suppose there exists a locally Lipschit-
a~.
V: ~
~ such that
D+V(X) ~ 0
= O}
{x E~: D+V(X)
By the way, the
was introduced in this theorem to avoid impos-
~(x)
ing a bound on
put
The
to follow yields sufficient conditions for a set,
possibly an unbounded set, to be attractive.
4.2.
On the
00.
on
~.
Sup-
contains no non-continu-
Suppose at last that there exist a continuous,
strictly positive function

four numbers
A, B, C, D
>
~:
and, for every
~ +~
p > 0,
xES'
B (M,P):
(i)
(ii)
(iii)
~(x) Ilf(x) II < A;
(11xll > B)
V(x)
Proof.
[~(x)D
Vex)
-C];
-D;
then, every solution

M as
===;.
x(t)
such that
x(J+) C S
tends to
w.
If the thesis is wrong, there exists a sequence
4.
Attractivity and weak attractivity
{til C J+
p >
and a
d (x (t i ) ,M) > 2p
i.
as
x(t i )
But
A+
there would exist a point of

eluded by Theorem 3.2.
t .... w
such that
for every
247
... co
outside
Ilx(t i + l ) - x(t i ) I I > P.
every
Ti E ]ti,ti + l [
t E [ti,T i [
fore, for
V(X(T 1.
At last, since
t ...
4.3.
co,
<
V(x(t
Further, for
such that
d(x(t),M)
and
Ti
p.
(a)
There-
IT.
1
ds
- t o ~(x(s
D V(x(sds < -C
ti
is decreasing, V(x(t
...
-co
which is excluded by (iii).

Remarks.
and
for otherwise,
Ilx(T i ) - x(t i ) I I
large enough, one has
- V(x(t.
'"
For an appropriate subsequence, again
{til, one has
for every
...
M, which is ex-
noted
there is a
as
Q.E.D.
In the applications, any suitable cri-
terion can be used to recognize that the set

non-continuable orbit.
contains no
In particular Proposition 3.6 can
prove helpful.
(b)
One might well wonder why the attractive set of
Theorem 4.2 has been chosen as a subset not of
an.
IT but of
The reason was one of convenience, and it will be no
limitation in any case, for it is always possible to decide

that
fIx)
is not defined on the attractive set.
(c) It has been proved that
w may be finite.
x(t) ... M as
t ...
00,
and
The reader who is familiar with Chapters I
and II knows that in a proof ab absurdo like this one, one

always shows that a function
V(x(t
... -
co,
whereas on some
248
VII.
other ground, it is bounded from below.

that V(x(t
V(x(t
+ - "", one uses the fact that
= ""
and that
is bounded from above by a strictly negative quantity
multiplied by
t.
This reasoning was impossible here.
is why one has used, to "push"

fact that
4.4.
But usually, to prove
V(x(t
towards
This
- <0, the
x(t) + "".
A theorem like VI.7.2 has shown that weak attractivity
is a useful concept.
The following theorem leads obviously
to some sufficient conditions for a compact set to be weakly

attractive.
It gives us an opportunity to draw some interes-
ting conclusion from the main hypothesis of Proposition 3.5
(namely that
w(x)
0), which, as will be remembered, led to
a rather disappointing result.

4.5.
weakly to a set
{til C ]a,w[
4.6.
M as
Let
Let us put
ti + wand
SeQ
(i)
(E)
M = as
i + ""
an.
Suppose there exist a
V: n
+~
and a 5(1
function
,""x E Q) D+V(x) < 0;

('1fx E E) W(x) ~ 0
tends weakly towards
where
x(t)
M as
= ix
such that
E S: D+V(X) = O};
x(J+) C s, x(t)
t + "".
If the thesis is wrong, there exists a solution
such that
nand
such that:
then, for every solution
Proof.
x(t i ) + M as
be closed with respect to
locally lipschitz ian function

+~
tends
t + W, if there exists a sequence
such that
Theorem.
bounded.
W: n
x: ]a,w[ +~n
We shall say that a solution
x(J+) C S
Then there exists an
and which does not tend weakly to

E >
and a T E J+
such that
x(t)
M.
5.
Attraction of a particle
x(t) E S \ B(M,E)
00.
249
for every
E [t,w[, and therefore
But one knows, by Theorem 3.2, that

x(t) ~ A+
is compact, that
A+
since
other hand, there exists a sequence

and
~ 0
W(x(t.
l.
00:
t ~
when
{x E S \ B(M,E): W(x)
t ~
as
and further,
On the
00.
such that
{to }
l.
indeed, W(x)
S \ B(M,E), IW(x)1
on the compact set

away from
i ~
as
A+ C E
t.
l.
00
being bounded
cannot be bounded
00.
But this is absurd, for
O}
are two disjoint compact sets.
A+
and
Q.E. D.
5.
5.1.
Attraction of a Particle by a Fixed Center
Let
be the origin of some inertial frame of ref-
erence and assume a particle of mass equal to

to two forces:
an attraction towards
friction.
Let
and
vectors.
and
dP>O
dr
along
f or every
~
of phase space, we
o will derive from a !if1
P(r), defined for

r > 0
r > 0
po-
and such that
The friction force will be directed
with opposite direction and an amplitude given by
a function
k(v) > 0
(,~)
(r 2 + v 2 )1/2.
The attraction towards

tentia1 function
For simplicity, we
for the euclidean norms of the two
Further, for the vector
shall use the norm
and some kind of
be respectively the position vec-
tor and the velocity of the particle.

shall write
is subject
k(v)
defined for
0, with
k(O)
0,
and a convergent integral
I: ifvr
dv.
The last hypothesis is verified for instance if

portional to
va
for some
k(v)
is pro-
E [0,2[, which might be accept-
VII.
250.
able for a physicist in many circumstances.

5.2.
The equations of motion read
~,
(5.1)
dP ~ _ k (v) v
dr r
v
We choose as Liapunov function the total energy
V(r,v)
where
T(v)
V 2 /2
T(v) + per),
is the kinetic energy.
One computes
readily
= -k(v)v
where the equality obtains if and only if

5.3.
O.
Let us now prove that for every solution of (5.1),
is bounded in the future.
(~o'~o) , then, for every
ret)
More precisely, if a motion
t
~
E [to'w[
IVo
ro +
v
dv,
k(v)
the second member being finite by hypothesis.

cp(r,v)
and assume there exists a
cp(ro'v o )'
~(t)
[ Vo
tl > to
Let us write
k~V) dv,
such that
r(t l )
rl >
Then, one would get

(5.2)
which is impossible, as we shall prove now.

Observe that if
is positive,
5.
</J{r{t),v{t
251
-v +
is negative, because the last term is negative and

Further, if there exist two values
to
t'
t" < tl
V{r{t'),v{t'
with
t'
and
= ret"),
ret')
vet')
</J(r(t"),v{t".
to
fore we get that
z(t), to
</J{rl,v l )
to
future.
vet")
and
and
tl
to
to
y,
b.
There-
</J(rO'v O)' which contradicts
We have thus proved that
(5.2).
Every hypothesis of
Lemma 5.6 below is verified if we identify

-</J(r(t),v(t
such that
one gets successively
V{r{t"),v{t", whence
at last </J(r(t'),v(t'
til
v.
<
ret)
is bounded in the
Next we consider successively two types of potential
functions.
5.4.
Assume first that
00
as
o.
0, and we may suppose, without loss of generality, that
o.
Per)
does not tend to
P(O)
Since
per)
is increasing, it tends to a limit when
With a view
to apply Theorem 4.2, let us identify
n with the phase space 916

and let
(,~)
\ {O}
of the non-zero vectors
be precisely the origin.
Every solution
of (5.1) is bounded in the future.
This has just been proved
for
is decreasing and
ret).
As for
vet), since
bounded from below, T(v)

the future.
The set
the phase space where
E
v
and therefore
is
are bounded in
=0
and
O.
At every point of
= 1,
2 or 3)
Let us choose, for every such point
(r,O),
an open neighborhood
N(r,O)
fine the auxiliary function

(5.1),
is constituted of those points of
E, one component at least of

is non vanishing.
, say
where
W(r,O)
r.
(for
r.
]
(,~)
0, and let us de-
= vj
But, owing to
252
VII.
a quantity which doesn I t vanish on

small enough.
N (r, 0) Ii E
N (r, 0)
for
It follows then from Proposition 3.6 that
contains no non-continuable orbit.
Therefore, all the hypoth-
eses of Theorem 4.2 are satisfied for any motion, if one

chooses
~(x)
identically equal to
and a set
enough to encompass the orbit of the motion:
large
the origin of
the phase space is a global attractor.

5.5.
Removing now the restriction that
to
00
when
0, we identify the set
with the hyperplane

and
P(r)
W be as above.
=a
and
with
does not tend
M of Theorem 4.2
~6 \ M.
Let
At last, let
1
cp (!.,~)
(1 + v) (1 + k (v
One computes easily that
We know that for every solution

an
rl > 0
B[M,r l ]
such that
(!.(t)
(!.(t),~(t,
,~(t
n n will play the role of
E B[M,rll.
S
in 4.2.
there exists
The set
Then all the
hypotheses of 4.2 are satisfied obviously, except perhaps

B > r l , we see that
_ r2)1/2 > O. If we
II(!.,~)II "> B implies that v > (B 2
1
limit our study to those functions k(v) which are, for v
(ii) which we verify now.
Choosing
large, greater than some strictly positive constant, we can

prove that
-k(v)v
(1 + k (v (1 + v)
5.
253
is smaller than some strictly negative constant in the appropriate region of phase space.
Theorem 4.2 applies, and as
it applies to every solution, M is a global attractor.

5.6.
Lemma.
Let
yet)
and
z(t)
be two real)fl
defined on some non-empty compact interval
[a,b]
functions
As-
C~O'
sume that
yea) <
(i)
(ii)
for every
z'
zeal
t', til E [a,b]
y (t') = y (t "): z (t ')
I*
= {t *
This set is closed.

approaching some
y' (t) > 0:
such that
z (t II )
t' < til
E [a,b]:
*
*
(Vt E [a,t])
yet) < y(t)}.
{t~} C I*
Indeed, let
and let
3.
be a sequence
be fixed in
[a,t*[.
For
*
large enough, yet) < y(t.)
Therefore
and,
such that
z(b).
Let
every
E [a,b]
(t) ~ 0 ~
and
Proof.
for every pair
(iii)
then
y(b)~
E I
*.
and thus yet) .: yet * ).

*
*
*
Further, for any t E I : y' (t ) ~ 0
-
3.
owing to (ii),
f*
z'
The set
I**
[a,b]
\ I*
('r)dT
(5.3)
O.
is open and is even a denumerable
union of open pairwise disjoint intervals:

I**
where
may equal
1 < i < P
Next
~.
]ai,b i [
y(a i )
To prove this, suppose first that

a!
3.
inf {t E [a. ,b.]: yet)

3.
3.
= y(b i )
for every
y(a.) < y(b.)

3.
and let
3.
y(a i ) + y(b.)
3.
}.
i.
VII.
254
One shows easily that

which is absurd.
then
bi
a~
J.
and also that
E ]a. ,b. [
J.
J.
Neither can one have
wouldn't belong to
We conclude that for every
1*, and this is absurd again.

i
J.
f1**
y(a i ) > y(b i ), for
z (b .) -
and
a' E I *
z (a.)
J.
> 0,
z' (,)d, > O.

-
(5.4)
It results from (5.3) and (5.4) that
J[a,b]
6.
6.1.
z' (,) d, = z (b) -
z (a)
> O.
Q.E.D.
A Class of Nonlinear Electrical Networks
In electrical engineering, there are many examples of
networks whose l'lormal operation consists in some kind of

switching from one equilibrium to another, and which should be
prevented from oscillating in any manner:
think for instance
of the bistable devices used in computer technology.
In this
section, we establish sufficient conditions for a certain

class of networks to be such that, starting from any initial
conditions, they approach an equilibrium as
6.2.
00.
The networks are those whose equations, following R. K.
Brayton and J. Moser [1964] can be derived from the knowledge

of a single state function called the mixed potential (or by
some others the "hybrid dissipation function").
More pre-
cisely, the variables describing the state of the network

will be supposed to be, for some positive integers
s, a current vector
i E ~r
and a voltage vector
rand
v E ~s.
6.
Nonlinear electrical networks
The components of
components of
255
are voltages across capacitors, and the
are currents through inductors.
potential is a real function
The mixed
of the form
P(v,i) = _vTxi - G(v) + F(i),

where
is a real
s x r
matrix (whose elements, by the
way, are equal either to
or to
-1, but
immaterial for what follows) and where

real
5(1
~s
functions, defined on
this fact will be
and
~r
and
are two
respectively.
The equations of the network read

C
where
order
sand
and
~t =
U~
di
:lP
(6.1)
L dt
:Iv'
are two square matrices, respectively of
r, usually functions of
and
i.
A descrip-
tion of the type of networks admitting equations of the form

(6.1) and the actual derivation of these equations appear in
the above mentioned paper by K. K. Brayton and J. Moser.
The
following theorem is an interesting example of construction

of a Liapunov function.
6.3.
Theorem.
(i)
.Assume that
is a function of
definite for
(ii)
E ~s~
is a function of
definite for
(iii)
v, symmetric and positive
for some
i, symmetric and positive
E ~r~
E ~r
and some constant, symmetric
and positive definite matrix

F (i)
(iv)
!.2
iTRi + aTi'
G(v) + IlxTvl1
'
-+-
00
as
-+ 00;
of order
r:
VII.
256
(v)
x E qeS, any
for any
(i,v) c!ifr x!ifs
and some
E ]0,1[:
where
(vi)
the system possesses a finite number of equilibriums)
then every solution of (6.1) tends to an equilibrium as

t ....
00.
Proof.
Since
(6.2)
the equations of the network read

C
dv
dt
=-
Xi _
3G
3v '
To get simpler notations, let us define the functions

g
and
by the following equations:

f
= 3p
3v = - Xi -
g =
3p
- ar=
xTv
3v
- Ri - a.
(6.3)
Our auxiliary function will be

V
= g TR-1 g
- P,
(6.4)
with a derivative, computed along the solutions of (6.1),

given by
6.
257
But
~ =
dt
I
XTC-I f - RL- g,
and therefore
This derivative vanishes at every equilibrium point, i.e. at

every point where
f = g = O.
Let us show next that it is
strictly negative everywhere else.
Since
C and
are
symmetric and positive definite, the matrices
CI / 2
LI/2
Therefore we
exist, are real, regular and symmetric.
and
may write
or
with
From this result and Hypothesis (v), one deduces that
_ ~~ =
(y _ Kx)T(y _ Kx) + xTx _ (Kx)TKx > I Iy _ Kxl 12
+
But Hypothesis (i) and (ii) imply that
only when
ishes.
vanishes, and
Therefore
dV < 0
dt
ellxll 2
vanishes when and
when and only when
van-
everywhere, except at the equi-
librium points.
Let us show at last that
00
00
when
In order to achieve this, we first write
Ilill + Ilvll
P
as a function
VII.
258
of
and
g.
We sUbstitute to
in (6.2) its expression
One obtains after some easy calculation, that
where
U(v)
T
T -1 T
[v X - a lR [X v - al + G(v).
Therefore, by
(6.4) ,
v
But
1 T-1
'2
g R g +
U (v)
-1
R being positive definite, the same is true of
Then Hypothesis (iv) implies that

II v II ~
But
00.
II g II ~
II i II ~
if
00
V ~
is evident from (6.3) and the fact that
v f
and
00
I Igl I +
when
00
as
00,
is regular.
Getting back now to Theorem 3.2 and Corollary 3.3, we

observe first that
Further, the fact that

with
II i II + II v II
= M = the
V
set of equilibrium points.
is decreasing and tends to infinity
implies that every solution is bounded.
Therefore every solution tends to the set of equilibrium

points.
But since there is a finite number of such points,
every solution tends to some equilibrium point
as
00.
Q.E.D.
6.4.
As an elementary illustration, consider the circuit of
Fig. 7.1, where the rectangle represents a nonlinear resistor, and where the capacitance
resistance
are constant.
the mixed potential that here

equal to
1.
One obtains
C, the inductance
and the
It appears from the theory of

X
is of order
and is
6.
f( V)
Fig. 7.1.
_ vi -
259
JV
RLC circuit.
f(v)dv + '2
1 R,2
1 ,
whence the equations of the circuit,

C dv
dt
L
di
dt
- i - f(v),
v -
Rio
We assume that the characteristic function
is such that
there is a finite number of equilibriums and that

+
Ivl
tends to
00
with
Ivl.
IV
o
f(v)dv
Then the circuit will always
approach an equilibrium if
1/2 -1 -1/2
L
R C
< 1,
or in other words, if
2
L < R C.
More involved illustrations of Theorem 6.3 will be

found for instance in R. K. Brayton and J. Moser [1964] or in
T. E. Stern [1965].
VII.
260
7.
7.1.
The Ecological Problem of Interacting Populations

The problem dealt with in this section yields a nice il-
lustration of Theorem 4.2 but it touches also several other

questions of stability theory, for example partial asymptotic
stability and the use of first integrals.
sidered by A. J. Lotka [1920]
It was first con-
(cited by N. S. Goel et altr.
[1971]) as a problem of chemical reactions or competing

species and then by V. Volterra [1931] when he tried to explain some cyclic variations of fish catches in the Adriatic.
Let us quote N. S. Goel et altr.:
"It was apparently ob-
served that the populations of two species of fish
commonly
found in these catches varied with the same period, but somewhat out of phase.
One of these was a species of small fish
, and the other was a species of a larger fish
It
seemed as though the large fish ate the small ones, grew, and
multiplied until the population of small ones diminished to
such a level that there were insufficient numbers for the survival of the large ones.
As the population of the large one
declined, that of the small species prospered to the degree

that a larger number of large fish could be supported, etc.".
7.2.
Let us call
Nl
the number of fish preyed upon and
the number of predators.
N2
It will appear natural to assume
that in the absence of predators, the birth rate
(dNl/dt)/N l
amongst the first population is a constant, which we write

al
Assume similarly that in the absence of the first popu-
lation, the death rate amongst predators is a negative constant, which we write
-a 2
In the presence of predators,
the birth rate of fish preyed upon has to be corrected by a
7.
Interacting populations
261
negative number, which we suppose to be proportional to the

number of predators.
It may be considered that the probabil-
ity of encounter between fish of the first population and predators varies like the product
Nl N2
Introducing a symmetric
correction for the death rate of predators, one gets the following pair of equations:
Nl(ll l - Al N2 ) ,
(7.1)
N2 (-a 2 + A2Nl )
where
for
Il
1l 2 ,
l'
Al
or
Nl > 0
and
N2 > 0
are
A2
>
o.
They make sense only
since they are concentrations of
animals or chemical products.
From a mathematical point of
view, they can of course be studied in the entire

plane.
ant.
One verifies that both
Nl
and
N2
axes are invari-
There are two critical points, namely the
('1"2)
(0,0), and the point
(,, "2)
(N l ,N 2 )-
or~g~n
~~, ~~
(n, ,n2 )
The former is unstable, as is shown by the existence of an

exponentially increasing
zero.
Nl , for
N2
identically equal to
The latter is stable, because one verifies readily
that the function
is a first integral for the Equations (7.1) and that is is

positive definite around
(n l ,n 2 ).
The last property is
proved by noticing that the Taylor series of

(n l ,n 2 )
begins by the terms

1 A2
112 2
1 Al Al
ill 2
(N - - ) .+ - (N 2 - - )
2 il2
1
A2
2 A2 III
Al
around
262
VII.
Following Volterra, we now generalize the problem to
7.3.
the case of
dN i
dt
Here
ki
species, for which the equations read
= Ni(k i + S-l
I
i 1 < j < n aijN j },
1 0, it means that the
ith species increases at the expense of the

when it is
one.
Si
< 0, the ith one diminishes on behalf of the jth
The
are
jth one, whereas
form an anti symmetric matrix.

> 0
The parameters
and take into account the fact that, for instan-
ce, to generate one predator, usually more than one prey has
to disappear.
Ni
Practically, Equations (7.2) make sense for
0, 1 < i < n.
the whole of ~n.

sets of ~n
Mathematically, they can be studied in

Moreover, one observes that any of the sub-
characterized by an arbitrary number of the
Ni
equal to zero, is invariant.

7.4.
The equilibriums are solutions of the equations

Ni(Siki +
~n
The origin of
1 < j
< n
aijN j } = O.
is one of them, and it is unstable as
soon as one of the
ki
is
> O.
Let
Ni = n i
be any equi-
librium, and suppose the subscripts have been arranged _in

such a way that
n.1. > 0
for
i = k + l, ,n, where
, or
n.
and
is some number equal to
Then of course
S.k. +
1. 1.
i = l, ,k
1 < j < k
n.1.
aiJn J. = 0
1 < i < k,
k + 1 < i < n.
= 0
for
1, 2,
7.
263
Consider now the auxiliary function
7.5.
=I
<
L<
N~
~.
B.n. n~ - In n
i
~ ~
i
k+l
BiNi'
which should be considered as defined on the set
Its time derivative along the solutions of (7.2) is computed

as follows:
I 
tegral.
V for every equilibrium

i, one gets
for every
has been used to de-
(nl, ,nn'.
positive definite.
There is a
(nl, ,nn'.
0, and
Further, let us subtract from
to its value at
aiJnJ.)N i
a .. n . ) N .
rive the final expression of the derivative.

function
Now if
is a first in-
V a constant equal
The function thus obtained is
Therefore, the corresponding equilibrium
is stable.
Consider next an equilibrium for which
belonging to the frontier of
W.
<
n, i.e.
One verifies that, when cor-
rected by a constant term chosen as above, the function

again positive definite around the equilibrium.
V is
Of course,
VII.
264
the positive-definiteness referred to here is relative to a

neighborhood (of the equilibrium) in
other hand, V
1 < j < k
~n.
On the
if the following conditions are satisfied:
k i 8i +
The function
~, not in
a .. n. < 0,
1J J -
(7.3)
k+l < i < n.
is no more a first integral in this case.
A simple generalization of Theorems 1.4.2 and 1.6.33 to the

case of a critical point located, as above, in
d~,
yields
parts (b) and (c) of the following proposition, which sums up

the stability informations derivable from the properties of
the function
7.6.
V.
Proposition.
which
Anyequilibrium
(nl, . ,nn)
of
n i > 0, 1 0, 1 ~ i
(n l , ,nn)
k, n.1 = 0, k+l
of (7.2) for
i ~ n, k < n,
(7.3)
is a sufficient stability condition;

(c)
in the same conditions, if the inequalities (7.3)
are strict for some values of
i, the stability is asymptotic
with respect to the variables having these

7.7.
Exercise.
its as subscripts.
State and prove the extensions of Theorems
1.4.2 and 1.6.33 referred to in Section 7.5.

7.S.
As an interesting particular case, let us consider a
ternary system where two species live on a third one.

equations read
Nl
(k l + 8 -1
1 a 13 N3 )N l ,
N2
(k
N3
-1
(k 3 - 8 3 (a i3 Nl + a 23 N2 ) )N 3 ,
-1
+ 8 2 a 23 N3 )N 2 ,
The
7.
where
i
= 1,
265
kl < 0, k2 < 0, k3 > 0, a 13 > 0, a 23 > 0

2, 3.
In general, one has
k l Sl /a 1 3
and
Si > 0,
k 2 S2 /a 23 .
If
we disregard this case, we find three and only three critical

points, namely
= N3 = 0;
(PI)
Nl
N2
(P 2 )
Nl
0, N2
(p 3)
NI
We know that
= ak3l3 3
(PI)
k3 B3
23
- > 0, N3
a
> 0, NZ
is unstable.
= 0,
N
3
As for
k 26 2
a Z3
- - > 0;
klBl
-a 13- >
O.
(P 2 ), the criterion
(7.3) yields
(7. 4)
with
a < sign because the equality has been discarded by
hypothesis.
The stability is asymptotic with respect to
Nl .
On the other hand, if we have, instead. of (7.4), the inequal-
ity
(P 2 )
is unstable.
invariant, the set

tor for any
> O.
Indeed, since the hyperplane

{(Nl ,N 2 ,N 3 ): Nl > O}
Nl
n B(P Z')
But on such a set, and if
=0
is
is a secis chosen
small enough, the auxiliary function V(N l ,N 2 ,N 3 ) = Nl is

-1
strictly positive and N = (k l + Sl a l3 N3 )
Nl > O. Simi1
larly, the stability condition for (P 3) reads
The inequalities (7.4) and (7.5) are mutually exclusive.

we continue to disregard the exceptional case where
= k262/a23'
If
k l Sl /a 13
we therefore obtain two unstable and one stable
266
VII.
equilibriums.
ATTRACTIVITY
FOR AUTONOMOUS EQUATIONS
Further, if the system starts in some neigh-
borhood of the stable one, one of the species disappears.

7.9.
Exercise.
Prove that this conclusion can be extended
to any initial point.
More explicitly, prove that if the sys-
tem starts from any point in the set
{(N 1 ,N 2 ,N 3 ): Ni > 0,
1 < i < 3}, then one of the species disappears asymptotically.

This is a precise statement of what is known as "the ecological principle of exclusion of Volterra-Lotka".
7.10.
One of the most conspicuous drawbacks of Volterra's
model is that the birth rate of the species preyed upon left
alone is a strictly positive constant, in such a way that the
corresponding population increases beyond any bound.
Of
course, in any real situation, there will occur a saturation

effect due to the limited resources of the ecological environment.
This observation was made by P. Verhulst [1845] when
discussing the Malthus theory of exponential population

growth.
This effect is taken into account in the following
= k(6
~~ ~
expression for the birth rate:

is some positive real quantity.
- N)/6, where
The Equations (7.2) are now
replaced by the following ones:

dN i
---d
t
ki
1
(6, - - (1 + sign k,)N,) +
= N,~ [--6i
-1
S~
1 < j < n
The equilibriums are obtained by equating the second

to zero.
Let
Ni
= ni ,
and assume as above that

n,
=0
for
<
n, > 0
= k+l, ,n.
ai),N),].
members
n, be such an equilibrium,
for
i = 1, ,k, and
Then of course
7.
signk.)n.
6.
267
1 < j < k
a .. n. = 0,
~J J
1, ,k
i = k+l, ,n.
The derivative of the function
V
1 <
L
i <
8i ki
+
defined as above reads
(1 + sign k i )
(N i
26 i
k+l <
?<
~
(8 i k i +
ni ) 2
L
1 < j < n
a .. n. )N .
~J
A first observation is that if the equilibrium is such that

n. > 0
~
for
i = l, ,n, the associated function
more a first integral of the differential equations.
is no
Further,
the same type of stability conclusions hold as for the simplest model, the one without saturation effect, and the stability conditions remain unchanged.
One observes however
that in the present case, the stability is necessarily asymptotic with respect to those variables for which
i < k
and
k i > O.
7.11.
Let us now try to get sufficient conditions of asymp-
totic stability with respect to all variables, for the case

where
n.~ > 0
for a1l
(Le.k=n).
Applying Theorem
II.l.3, the only thing to prove is that the set

{(nl, ,nn)}
E \
contains no non-continuable orbit, where
Assume that we renumber the
Ni
in such a way that
k. > 0
if
1, . ,1,
k. < 0
if
1+1, ,n.
VII.
268
Then of course every point of

Nl+l, .. ,N n )
and, owing to Proposition 3.6, it will be enough
to prove that the vector function

non vanishing derivative on
that, on
(nl, .. ,n l ,
is of the form
W = (SlNl"",SlN l )
E \ {(nl, ,nn)}'
has a
One computes
E,
6.l. -
1+1 ~ j < n
a .. ni(N. - n.).
l.]
(i=l, ,l)
Therefore, the stability of the equilibrium is aSymptotic

whenever the rank of the matrix
equals
if
n - 1.
Of course, this condition is never satisfied
1 < n/2.
7.12.
As an example, consider the quaternary system where
the fourth species subsists on the third and the second,

while the third subsists on the first.
The corresponding
differential equation are
e2 -
N2
-1
- S2 a 42 N4N2 ,
N2
k2N2
N3
k3 N3 + S3 (a 3lNl - a 43 N4 )N 3 ,
62
-1
8.
a ij > O.
librium
269
Under suitable conditions, there exists an equiNi = n i > 0
for
= 1, ,4.
On the set
the derivative of the function
doesn't vanish.
Indeed
Hence the equilibrium mentioned above is asymptotically stable.
8.
The invariance principle for bounded solutions of autonomous equations is due to J. P. LaSalle [1960].
Antecedent
to the theory are the theorems of E. A. Barbashin and N. N.

Krasovski appearing in Section II.l of this book and which
dealt already with periodic equations.
ciple for unbounded
. [1968].
The invariance prin-
solutions comes from J. P. LaSalle
A particular case of Proposition 3.6 on non invariant
sets was proved in N. Rouche [1968] and subsequently generalized by M. Laloy (1974) and N. Rouche [1974].
The attracti-
vity Theorem 4.2 comes from J. L. Corne and N. Rouche (1973),

where it is applied to the fixed center problem treated in
Section 5.
The problem of electrical network of Section 6
will be found, as already mentioned, in R. K. Brayton and J.

K. Moser (1964), whereas a good review of the ecological problems of interacting populations appears in N. S. Goel et altr.
[1971].
CHAPTER VIII
ATTRACTIVITY FOR NON AUTONOMOUS EQUATIONS
1.
Introduction, General Hypotheses
Proving attractivity or asymptotic stability is more

difficult in the non autonomous case than in the autonomous
one, because in the former, one cannot rely in general on any
invariance
property of the limit sets.
The situation is
more complex and the asymptotic properties which can be

proved with substantially equivalent hypotheses are weaker.
In Section 2, we introduce the one-parameter families of
Liapunov functions of L. Salvadori:
they do not appear in
the statements of the theorems, but are powerful tools of

demonstration.
We use them to prove a significant extension
of Matrosov's theorem II.2.5, yielding a new and interesting

characterization of the uniform asymptotic stability of the
origin.
By the way, in order to grade the difficulties, the
origin, instead of a set, is studied in this Section 2.
Sec-
tion 3 gives another useful extension of Matrosov's theorem.
2.
271
The families of auxiliary functions
With Section 4, we corne back to the attractivity of sets and

examine what kind of generalizations of LaSalle's Theorem
VII.3.2 can be proved when starting from the most natural and
simple hypotheses in Liapunov's direct method:
lower bound on an auxiliary function
of upper bound on its derivative
some kind of
V(t,x), and some kind
V(t,x).
Next we extend the
attractivity Theorem VII.4.2 to the non autonomous case, and

it proves helpful, in this setting, to generalize also Proposition VII.3.6 on the expulsion of the solutions from a
given set.
Section 5 is devoted to those particular types of
non autonomous equations for which the limit sets admit some
kind of invariance property:
for such equations, which are
called asymptotically autonomous, asymptotically almost

periodic, etc., there exist interesting particular extensions
of LaSalle's theorem.
Finally, Section 6 is an introduction
to dissipative systems.
Our general hypotheses are again here those of Chapter
IV.
For reference purposes, let us recall the Cauchy prob-
lem on hand:
x = f(t,x),
x(t O) = xO'
2.
2.1.
(1.1)
(1.2)
The Families of Auxiliary Functions
Let us show how a family of auxiliary functions in the
sense of L. Salvadori [1969] [1971] can be used to prove a

theorem of weak attractivity.
erable:
These families are non denum-
there corresponds a function to each possible choice
of the quantity
in the definition of stability, or of a
272
VIII.
similar quantity in another definition.
A lemma and two de-
finitions are needed before stating the theorem.

2.2.
g
Lemma (L. Salvadori [19691).
two functions on
constants
into !Jf.
Sl' S2' S3 > 0
be a set and
[f (z) > -S21
===i>
f,
Assume there exist three
fez) ~ 0,
z E S
g(z) ~ Sl'
[g (z) < -S3 1 ;
then, there exist two constants

every
Let
~,
1 > 0
such that, for
z E S:
fez) +
Proof.
~g(z)
One only has to choose
<-1.
~
such that
0 <
< S2/Sl
Q.E.D.
2.3.
If
and if
is a compact neighborhood of the origin of ~n
cn,
E CN
function
pect to
we say with V. M. Matrosov [196211 that a
V: I x N~!Jf
E
on
is non-vanishing definite with res-
if
(V8 > 0) (:In> 0) (:Is> 0) (Vt E I)(Vx E N \ B )

8
[d(x,E) < n1 ~
2.4.
Exercise.
with respect to
neighborhood
Show that
E
on
JY of
E \ {oJ
V(t,x)
is non-vanishing definite
if and only if there exists a

and a function
(Vt E I) (Vx E .5n N)
a E.5t'
such that
IV(t,x) I ~ a(llxll);
also if and only i f there exists a function

('fit E I) (Vx EN)
[\V(t,x) I > Sl
a EYe
such that
max (d(x,E) ,IV(t,x) I) ~ a(llxll).
2.
2.5.
273
The following definition is akin to the one of weak at-
tractivity (cf. VI.6).
x: ]a,w[ ~ ~n
A solution
is said to tend weakly towards the origin as
~~.
w, if there
{t.}
exists a sequence
as
of (1.1)
Our first theorem refers to this concept and can
be compared to Theorem 11.2.5.
As it is rather difficult, we
shall avoid some inessential difficulties by assuming that

both auxiliary functions
2.6.
Theorem.
b > 0, two )(1
and
Assume there exist three constants
M, a,
functions
I x Q
V(t,x)
into 91, a continuous function

compact neighborhood
every
and
V* (x)
W(t,x)
on
fl
on
into 91
and a
of the origin, N C fl, such that, for
(t,x) E I x N:
(i)
(ii)
(iii)
Ilf(t,x) II ~ M:
-a~V(t,x):
V(t,x) ~ v*(x) < 0: we put
(iv)
IW(t,x) I < b:
. (v)
if moreover
respect to
then every solution

the origin when
Proof.
of
Ware )(1.
= {x EN: V*(x) = oJ:
W is non-vanishing definite with

on
N,
such that
x(J+) C N
tends weakly to
~ roo
The solution
on hand cannot approach the frontier
fl, and therefore for
x: J+
[to'~[
To prove
the
theorem, one only has to show that there exists, for every
E
> 0, a T > 0
such that
I Ix(t)1
a time interval of duration

n
and
T.
such that for every

x E G(E)
cannot remain> E
on
OWing to (v), there exist

t E I
B(E,n)
and every
(N \
BE)'
274
VIII.
one gets
IW(t,x)1 >~.
Defining the compact set
H(E)
H(E)
by
(N \ BE) \ B(E,n/2),
we observe that, by (iii),

(i6 > 0) (Y(t,x) E I x H (E
Let us now use
and
V(t,x) < - 13.
(2.1)
W to construct a function
v E: J x (N \ BE) -..!JI with a strictly negative derivative.

Notice first that
G(E)
can be split into two disjoint parts,
namely
G(l) (E) = {x c G(E):
G(2) (E)
{x E G(E):
Let
be a
(Yt E I)
W(t,x) < -0,
(Yt E I)
W(t,x) > O.
function such that
"E)
= 1
for
E [0,n/2],
"E(T)
=0
for
<
Its derivative is obviously bounded.
n.
For
1, 2, let us
define
a (i) (x) = " (d (x,E
for
x E G(i) (E),
a(i) (x)
for
x ~ G(i) (E).
These functions will not usually be ir1 , because

not irl.
is
However, they have bounded Dini derivatives, for
the derivative of
for
d(x,E)
v E is bounded, and one gets further that
(t,x) E I x (N \ BE)' as well as similar inequalities
for the other Dini derivatives.
Define next
h:
E
2.
I x (N \ B) ~~
275
by the equation
h~ (t,x) = (a (1) (x) _ a (2) (x

""
W(t,x).
And now let us prove that if one chooses

the functions
V(t,x)
D+h(t,x)
and
pectively with the functions
=I
(N , B),
can be identified res-
and
of Lemma 2.2.
In-
deed (see Appendix I.l.2 e):

V(t,x) ~ o~
(1)
D+h (t,x) = (a (1) (x) - a (2) (x W(t,x) + (Da (1) (x)
W(t,x), where D means D+ or D+ according to
(2)
- Da~2) (x
the sign of
W.
OWing to the construction mode of the

< O.
the first term is
The second one is bounded, for it is
constituted of three bounded

(3)
every
functions~
At last, D+h(t,x)
x It: H(),
for
a~i),
Da(l)
< - ~ for every
and
Da (2)
n/2, and, on the other hand, V(t,x) ~
t E I
vanish when
-e
for
and
d(x,E) <
(t,x) E I x
H().
Since the hypotheses of Lemma 2.2 are verified, there
exist two quantities
for
V >
(t,x) E I x (N \ B).
from below by
T(t O)
-a - Vb.
= sup
1 > 0
such that, if
But on this set, v
is bounded
Let us put
{v(to'x o ) + a + Vb
1
No solution can remain in

of length
and
N \ B
during an interval of time
T, for if this were the case, one would get for
this solution that
276
VIII.
which is absurd.
2.7.
Q.E.D.
Another proof of the same theorem.
With a view to com-
pare the one-parameter families of auxiliary functions to the

more classical types of proof, let us outline now another
demonstration of the same theorem, this time inspired by II.
2.5.
E, G(E)
We retain the definitions of
start anew from (2.1).

remain in
2b/~.
G(E)
and
H(E)
We first show that no solution can
for a period equal to or longer than
Indeed, IW(t,x) I
and
~ ~
being
on
G(E), and since
is continuous, one would get in the opposite case

2b > IW(t + T) - W(t)1
t+T
= ft
li(a) Ida > ~T
2b
which is absurd.
Let us put
= min
{BT,Bn/2M} > 0, k
yi ~ a + sup [V(to'x O): Xo EN \ BE]}
= min
T(t O)
and
sume, ab absurdo, that the solution may remain in

during the time interval
val in
[to,tO+T].
sub-intervals of length
(1)
to (2.1),
kT.
As-
N \ BE
Let us cut this interT:
i
Then, for each
{i EN:
l, . ,k.
i:
either
x(t) E H(E)
for every
t E I, and owing
2.
(2)
or
x(t) E G(e)
t E Ii
therefore two values
t.
l.
l.
and
n/2 < d(x(t) ,E) < n
It~l. - t.l. I
the case,
>
for which
t.* E I.
and
out loss of generality, that

d(x(t.),E)
t' E Ii.
for some
there exists a value
277
l.
x(t)
In this case,
G(e), and
(and we assume, with-
l.
ti < t i*) , such that
= n/2,
d(x(t~),E)
l.
= n,
for every
n/2M.
I f this is
Indeed,
< Mit ~ -
Since
l.
tl.'
is negative everywhere and smaller than
I.
-6
on
H{d,
and therefore
at last
2.8.
veto + T) - veto) < -ky
~-V{to)
- a, whence
veto + T) < -a, which is absurd.
Q.E.D.
The differences between these two proofs are interest-
ing to notice.
In the former, W is changed into
he
to
get a strictly negative derivative, then combined with

yield
v e ' and using
pelled from
N \ Be.
to
v , the solutions are shown to be exe
In the latter, W is used to prove that
the solution is expelled from an appropriate neighborhood of

E(e); after which
is shown to decrease at a sufficient
pace for the solutions to be expelled from

the use of Hypothesis (i), i.e. the bound on
Be.
Observe
f(t,x), in
both demonstrations.
In the former, it yields an upper bound
on the derivative of
h.
e
It is proved below that a bound on
278
Wf
VIII.
would be enough.
But this new hypothesis would not suf-
fice in the latter proof, where the bound on
ficient transit time of the solution through
H() n G(),
and therefore a sufficient decrease of

potheses on
V.
gives a suf-
At last, the hy-
W might be weakened for the second proof, where
they are used to show that the solutions are expelled from
G(), a result which can be obtained in various ways.
2.9.
Corollary.
The thesis of Theorem 2.6 remains true if
Hypothesis (i) is replaced by

(i')
Proof.
(V(t,x) E I
N)
IIW(t,x)f (t,x) II .::. M.
One only has to observe that, for
IDa(i) (x)1 =
2,
II~I
Dd(x(t),E)I'::'cllx(t)11
T
d(x(t),E)
=
where
= 1,
is the bound on the derivative of
cllf(t,x(t II
In these
V.
conditions
D+h(t,x) < 2cllf(t,x)W(t,x) II.
2.10.
Corollary.
In Theorem 2.6 as well as in its version
modified by Corollary 2.9, the quantity

independent of
to
Q.E.D.
provided that
may be determined
is bounded (and no more
bounded from below only).

The thesis of the theorem will then be expressed saying that the solutions starting from some point in
weakly to the origin, uniformly with respect to
t
00.
Notice that
tend
to' when
was, from the beginning, independent
2.
279
We know that uniform stability plus weak attractivity

entails attractivity.
Therefore, if we add to the hypotheses
of Theorem 2.6 any conditions of uniform stability, we get a

theorem of equi-asymptotic stability.
Further, if Corollary
2.10 applies, one obtains uniform asymptotic stability.
This
leads to the following theorem, generalizing Theorem II.2.5.

2.11.
Theorem.
c > 0, two ~l
Assume there exist two constants

functions
V(t,x)
into !ii, a continuous function

functions
and
bEJe
and
V* (x)
W(t,x)
(H)
(iii)
(iv)
(v)
I x 0
into 5it, two
on
and at last a compact neighborhood
N of the origin, NCO, such that, for every

(i)
on
M and
(t,x) E I x 0:
II f (t,x)W(t,x) II ~ M~
a(llxll) ~ V(t,x) ~b(llxll)~
~(t,x) ~ V*(x) ~ O~ we put
= {x EO: V*(x)
O}~
Iw(t,x)1 ~ c~
if moreover
respect to
is non-vanishing definite with

on
N,
then the origin is uniformly asymptotically stable.

Assume for a moment that we include in our general
hypotheses a local Lipschitz condition for
tailing uniqueness of the solutions.
on
I x 0, en-
One knows then (cf.
I.7.4) that the classical Liapunov conditions for uniform

asymptotic stability (i.e. (ii) hereabove along with a negative definite
V) are also necessary.
Clearly, any other
sufficient condition ensuring this property can be but stronger

than this one, or at least equivalent.
In the case of Theorem 2.11, we really have an equivalence.
Indeed, the Liapunov hypotheses imply those of 2.11
280
VIII.
with
reduced to the origin and an identically vanishing
function
W.
Notice that if we had kept, instead of (i) the
hypothesis (i') ! !f(t,x)!! < M, the Liapunov hypotheses would

not have implied those of 2.11!
The method of one-parameter
families of auxiliary functions led us therefore to an origina1 characterization of uniform asymptotic stability of the
origin.
3.
3.1.
Another Asymptotic Stability Theorem
Theorem 2.11 in the last section was an extension of
Matrosov's Theorem II.2.S, to the case of possibly unbounded

second members
f(t,x).
Another extension of the same theorem
is considered here, which consists in replacing the second

auxiliary function, the one written
W, by some more elabor-
ate means of proving that the solution is repelled by a given

compact set.
This is the object of the following lemma, to
be compared with Proposition VII.3.6.
In addition to its use
in the asymptotic stability theorem in question, this lemma

can serve other purposes, for instance to prove that a set
is an absolute expeller in the sense of Chapter V (see M.
La10y [1974]2).
3.2.
Lemma.
{Mi : 1 < i
subsets of
Let
K be a compact subset of
n 1 }, {N j : 1 ~ j ~ n 2 }
such that
M =
be two sequences of open
fJM.) U (U
n and let
N.)
J
K.
Assume the
N.
are pairwise disjoint and there exist two sequences of
lf1
functions
such that
3.
Anotner asymptotic stability theorem
(i)
(Vi: 1
x
uniformly in
as
Mi ~
K, x E
n l ) Ui(t,x)
281
(:Jl3 l > 0) (Vj: 1 < j < n 2 )(V(t,x) E I x N.)lw.(t,x)I<131~

J
J
(ii)
(:J~
(iii)
>
O){(V~: 1
~nl) (V(t,x)

~~
(Yj: 1 ~ j ~n2)(V(t,x) EI x N.) Iw. (t,x) I > ~~

J
J
assume finally that

(iv)
(:JA > 0) (V(t,x) E: I x Q)
then there exist two numbers

x(t)
of (1.1) cannot remain in
time-interval of duration
Proof.
of an
and
such that a solution
D = {x: d(x,K) < a}
on a
T.
One shows easily (compare with VII.3.6) the existence

r > 0
such that, if we put

M.1.*
N.*
= M.1.
\ {x: d(x,8M i ) ~ r},
N.
J
\ {x: d(x,8N.) ~ r},
M.*
the whole family of the

open covering of
K.
M*
Choose now
13 2
a > 0
and
l.
N*
j
still constitutes an
Let us write
* U (U
(U M.)
i
l.
N~).
J
such that
o
and
II f (t,x) II < A~
such that
< 13 2 <
~r
~
a < d(K,8M*)
(3.1)
and further that, owing
to (i),
(V(t,x) EI x (M i n D
where
D = {x: d(x,K)
a}.
IUi(t,x)1 ~ 13 2 ,
1 < i ~ n l , (3.2)
Let us put further
282
VIII.
P.
Qj
Let us
= M.1 n
= N.J n
P*
i
Qj*
0,
0,
Mi*
n
Nj* n
now show that a solution
main in any of the

greater than
P.1
or
= 2a/f;,
be a sOlution such that

for one of the
Qj's).
Qj
0,
1 < i
0,
~ n2
x(t)
of (1.1) cannot re-
for a period of duration
where
= max
x(t O) E Pi
If
~ j
nl ,
x(t)
{a l ,a 2 l.
Let then
(one would reason alike
remains in
Pi
from
to
tl = to + ~, one gets, owing to (iii) and (3.2) that
up to
2a ~ 2a 2 ~ IUi(tl,x(t l
- Ui (to,x(to1 > f;(t l - to) = 2a,
which is absurd.
Assume now that
x(t)
leaves
"barrier"
without leaving
P.
P.1 \ Pi*
least equal to
stants where
r.
x (t)
for some
be in
If
i.
0, i t has to cross the
and therefore to travel a distance at

If
and
tl
are respectively the in-
t2
enters and leaves
the barrier, one gets
that
which contradicts (3.1).

for some
be in
leaving
j.
If the solution leaves
0, it has to cross the barrier
without
and there-
fore again to travel a distance at least equal to

and
t2
are here also the instants where
But then either
x(t)
remains in
t2
and one gets the inequalities (3.3),
of
Pi' but then it crosses
again.
x(t)
x(t l )
leaves the barrier, one knows that
*
Pi.
x(t O)
Assume on the contrary that
r.
If
tl
enters and
belongs to some
Pi
~
between
x(t)
tl
and
comes out
P. \ P~, and one obtains (3.3)

1
Q.E.D.
3.
Another asymptotic stability theorem
3.3.
Exercise.
Let
for some integer

a 5fl
function
283
n and let,
K be a compact subset of
and a neighborhood
k > 0
W: N
~k
of
K, N C
n,
be such that its derivative along
the solutions of (1.1), written
W(x), doesn't depend on
and that
(i)
('Ix E
(ii)
t-
W(x)
K)
O~
k - I
at least
components
identically on
Wi
of
W vanish
K~
then, except for (iv), all the hypotheses of Lemma 3.2 are
verified.
Hint:
use Hypothesis (i) and the compactness of
construct two sequences

tively
k - m
nents of
{M i
and
W vanish on
K and
doesn't vanish identically on
m= 0
m= I
to
containing respec-
{N. }
J
m elements, where
and
if all compo-
if one component of
K.
Lemma 3.2 enables one to render Theorem 11.2.5 on

asymptotic stability more versatile by merely changing the
hypotheses concerning
W.
The generalized theorem reads ex-
plicitly as follows.
3.4.
a Sf0
Theorem.
Let there exist a Sfl
function
a constant
(i)
(ii)
(iii)
function
V*: n +~, two functions

A > 0
and
< v < E
n +~,
bE.5t'
(t,x) E I x
and
n:
a(lIxll) < V(t,x) ~ b<llxll)~
.
V(t,x)
V* (x)
II f (t,x) II
2.
~ O~
put
{x
with
BE C
n,
V * (x)
E n:
A~
suppose further that for every pair of numbers
V: I
the set
and
E,
{x E E: v 0
such that
Ba C
n,
let us put for
t E I,
-1
V
{x En: V(t,x)
t,a
< a(a)}.
Then
(a)
for any
x(t;to'x O) .... 0
(b)
-1
Xo E Vt
:
O,a
( to' x 0)' when t .... 00;
and any
to E I
uniformly in

3.5.
Example of a general Lagrangian system.
Consider a
mechanical system such as described at Section 13 of Appendix

II, i.e. with a kinetic energy
domain !Jtn x
n,
T(q,q) which is
a potential function
there exist Lagrangian forces

terms of
t, p, q
as
on some
also lf1
n (q)
stationary at the origin, with nCO) = O.
~1
n,
on
Assume further that
Q(t,q,q), also expressible in
Q(t,p,q).
The equations of motion for
such a system are Hamiltonian equations modified by the adjunction of generalized forces.
Assume that
They read
aH
aq (p,q) + Q(t,p,q),
n(q)
admits an isolated minimum at
aH
ap (p,q).
and
that
T
aH
H(t,p,q) =Q (t,p,q) (p,q) ~-a(llpll)
ap
for some function

proaches
set
a EYe.
Assume further that
when
Q(t,p,q)
p"" O.
apThe
of Theorem II.2.5 as modified in Theorem 3.4 above is
{(p,q): p =
a}.
The
(with appropriate sets

component of
Pi
will play the role of the

M.) in Lemma 3.2.
~
vanishes on
E.
As for
Ui
Of course every
p,
observe first
4.
Extensions of the invariance principle
285
that, owing to Proposition 7 of Appendix II and the fact

that
every
and
q
and
vanish simultaneously, Q(t,O,q) = 0

t.
Therefore, on
p=
aB
aq
E,
an
(O,q) =
aq'
a vector which, by hypothesis, vanishes at

Then the
Ui
Pi
for
q = 0
only.
possess all the properties required from the
in 3.4 and 3.2.
The full power of Lemma 3.2 will not be
used here, since there is in our example, no function like

the
We conclude that the origin
=q =0
phase space is uniformly asymptotically stable.
of the
This extends
to some types of non autonomous mechanical systems the result

embodied in Theorem III.5.2.
It is pleasant to observe that
the auxiliary functions used in connection with this mechanical problem are the total energy and the vector of conjugate
momenta, two functions which are obviously meaningful to the
physicist.
4.
Extensions of the Invariance Principle

and Related Questions
4.1.
We now come back to the attractivity of sets and remove
from our general hypotheses the requirement for the origin of

~n
4.2.
to belong to
and to be a critical pOint of (1.1).
In the autonomous case, LaSalle's theorem proved that
A+ n n was a subset of
invariant set of
M, where
E = {x E
n:
M is the largest semi-
D+V(X)
= OJ.
In the absence of
a semi-invariance property, all that can be proved in the non

autonomous case is that
A+ n n
is a subset of
E.
But how
VIII.
286
shall we define
tion of
in the present case where
but also of
t?
V* (x)
is a func-
In his paper of 1968, LaSalle
supposed the existence of a function

n +V(t,x)
n+v
and defined
V*(x)
as
such that
Here, we shall find interesting to assume that

bounded above by a function of
and
will be identified sometimes with
but rather that some function of
x(t)
t
n+V(t,x)
is
x, which by the way
n+V(t,x)
we shall not immediately show that
= OJ.
{x: v*(x)
and
itself.
Further,
approaches some set,

x(t)
tends to zero.
Later, this function will be identified with the distance

from
x(t)
to some set.
Most results in this section are
not very far from being variants of one another:
they form
altogether a set of tools enabling one to cope with rather

varied situations.
A final remark:
a majority of the
theorems below are consequences of a unique fundamental lemma,

which by fue way formalizes a type of reasoning already encountered in this book (see Theorem 1.6.23).
4.3.
~:
Fundamental Lemma.
la,co[ ...
(i)
(ii)
(iv)
then
~(t)
Proof.
If
 0) (:la E.5it') (:lA > 0) (\ft:
~(t)
?. d
n+<P (t) < -a (~(t) ) ;

n+1jJ(t) < A
... 0
~(t)
as
(or
n+~(t) ~ -A);
t ... co.
does not approach
and an increasing sequence

tn ...
<p: la,co[ ... Slf,
be two continuous functions such that
assume further that

(iii)
a ESlf, let
For some
and for every
{t}
n
0, there exist an
E > 0
of time-values such that
n: tn+l - tn ~ E/A
and
~(tn)
>
2E.
4.
But then by (iv) , l//(t) > e:
on
[t ,t +
n n
287
Xl or on [t n
+ X) - 4>(t ) <
n
x,tnl. Then, as the case may be, 4> (t n

e:
e:
or 4> (t ) - 4>(t - =-) < - A
- 1\ a(e:)
In both cases,
a(e:).
n
A n
4>(t) .... _00 as t .... 00, which contradicts (i)
Q.E.D .
4.4.
The properties which can be proved are of a different
character according as, for the solution on hand, w
in
00,
which case there is no need for the second member to be

bounded, or
w is arbitrary, and then some kind of bounded-
ness has to be assumed for
case where
00
f(t,x).
Let us first study the
and no bound is assumed for the second
member.
4.5.
Theorem (arbitrary solutions).
and let
and
every
I x n
and continuous.
and a function
into
~,
If there exist
a EYe
such that, for
V(t,x) > -A;
w(t,x)
(ii)
(iii)
D+W(t,X) > -B
W(t,x(t
> 0;
D+V(t,X) < -a(w(t,x;
(iv)
which
be a subset of
(t,x) E I x S:
(i)
then
A, B > 0
numbers
W be two functions on
locally lipschitz ian in

two
Let
00
.... 0
and
as
(or
t ....
00
D+W(t,x) ~ B);
for every solution
x(t)
for
x(J+) C S.
This is an immediate consequence of Lemma 4.3.

requirement of a lower bound on
The
may be relaxed, but only
for those solutions which don't tend to
00
as
t ....
00.
This
is shown in the next theorem.

4.6.
Theorem (solutions which don't tend to
00).
Let
be
288
VIII.
n and let
a subset of
I
and
be two functions on
~, locally lipschitz ian in
into
B > 0
If there exists a number

that, for every
(ii)
(iii)
and continuous.
a E
and a function
Je
such
E I x S:
(t,x)
~(t,x)
0:
D+V(t,x} < -a(w(t,x}}:

D+W(t,x)
(iv)
-B
D+ ~(t,x)
(or
B):
if moreover
(i') for every compact set
such that
then
W(t,x(t)}
V(t,x} > -A
as
which does not tend to
00
Proof.
00,
I x (C
tends to
00
x(t)
and for which
For every solution of the mentioned type, there exists

C
{t n }
and a sequence
and for every
n: x(t n )
doesn't tend to
00
E C.
as
decreasing, it tends to a limit as

it admits a lower bound.
such that
Exercise.
tn
as
Then, owing to (i'),

+
00
But since
00,
vet}
is
and in particular,
Therefore Lemma 4.3 again entails
the announced result.

4.7.
n n):
for every solution
00
as
00
on
A > 0
x(J+} C S.
and
a compact set
n
C, there exists an
Q.E.D.
The following statement is interesting in
that, like Theorem VII.3.2, it draws some conclusion of

asymptotic behavior from the consideration of a single auxiliary function.
Let
I x
into
~ such that
schitzian in
A
and
be a subset of
n and
V(t,x)
a function on
exists,:be locally lip-
and continuous: if there exist two numbers
(t,x)
E I x S:
4.
(i)
V(t,x) < 0:
(ii)
V(t,x) > -A;

O+V(t,x) > -B
(iii)
then
V(t,x(t
w =
which
4.8.
289
For
00
as
for every solution
00
x(t)
for
x(J+) C S.
and
independent of
lji
O+V(t,x) ~ B):
(or
t, Theorem 4.6 admits the
following corollary, which is akin to Theorem 1 (b) of J. P.

LaSalle [1968].
4.9.
Corollary.
n; let
pect to
1ipschitzian in
Let
n,
be a subset of
closed with res-
V: I x n + 9f be a function which is locally

x
and continuous, and let
lji:
9f be a
locally 1ipschitzian function: if there exists a number
B > 0
(t,x) E I x S:
O+V(t,x) < -lji (x) :
(ii)
(iii)
lji (x) .:: 0:
O+lji(t,x) ~ -B
(iv)
(or
O+lji(t,x) ~ B);
if moreover
(i')
for every compact set

such that
nne E
then
A+
x(t)
such that
Proof.
C, there exists an
V(t,x) > -A
on
I x
(C
A > 0
n) :
{x E S: 1)J (x) = O}, for every solution
00
and
x(J+) C S.
If the solution in question tends to
00
as
00,
its limit set is empty and the thesis is trivially verified.

Otherwise, the solution is such, owing to Theorem 4.6, that
1)J(x(t
closed in
n \
fore
as
n.
is closed in
If there existed a point
E, its distance to
lji(x(t
Since
00.
x*
of
n,
is
in
would be strictly positive.
couldn't approach
as
00.
ThereQ.E.D.
290
VIII.
4.10.
A+ C E U (as II
One deduces from this corollary that
an).
sen,
In particular, if
then
situation described in Theorem 1
(~)
A+ C E, which is the
of LaSalle [1968].
Further, in this paper, the regularity conditions imposed on

ware somewhat different from those adopted here.
The the-
sis of Corollary 4.9 will appear insufficiently precise in

many practical circumstances.
The following corollary gives
conclusions which are stronger in some sense.

4.11.
Corollary.
If one adds to Corollary 4.9 the hypothe-
sis that for every

tion
xES: W(x)
a ESt'; then
x (t)
which does not tend to

4.12.
as
as
00
a (d (x,E))
00.
00
for some func-
for every solution
Let us next examine the case of a solution which can-
not be continued up to infinity, or in other words for which
w is finite.
As will become apparent from the theorems to
follow, we shall be obliged, in compensation, to impose

kind of bound on
f(t,x)
and further, to content ourselves
with weaker conclusions.

4.13.
Theorem.
pect to
~,
n:
and
be a subset of
(H)
(Hi)
(iv)
closed with res-
W be two functions on
x
I x
into
and continuous: if, for every

A, Band
a E $I(, such that for every
C:
(i)
n,
C C S, there exist three numbers
and a function
D > 0
x
locally lipschitzian in
compact set
let
Let
Ilf(t,x) II :. A:
w(t,x) > 0:
V(t,x) > -B:
D+V(t,x) ~ -a(W(t,x)):
(t ,x) E
4.
(v)
o+~(t,x) > -0
291
o+~(t,x) < 0);
(or
then
~
as
Proof.
* (t)
min {~(t,x(t, d(x(t),an),
w, for every solution

If
x(t)
} +
1 + Ilx(t) II
there is nothing left to prove.

1::,
.... w
< I:: < 1,
1)
B(an,d]
as
w,
{t}
Vet)
is bounded from be-
It remains to verify Hypotheses (iii) and (iv) of the
fundamental Lemma 4.3.

to some given
as above.
tion
x(J+) C S.
ns.
One shows as in Theorem 4.6 that

low.
} + 0
such that
n
belongs to the compact set
x (t )
[B(O,~ -
Ilx (t) II
Otherwise, there exists an
such that, for some sequence
as
such that
min {d(x(t),an),
1 +
1::,
Now
only if
* (t)
x(t)
is greater than or equal
belongs to some compact set
One knows that there exists for this set a fun.c-
a E Je such that, for every
(t,x)
in the set:
o+V(t,x) ~ -a(~(t,x, and a fortiori: o+V(t,x) ~ -a(~*(t,x,
~*
is obvious.
At last, again on the
same compact set, f(t x)
is bounded.
The same is true for
D+(d(x(t).,"))
1
+ Ilx (t) II
where the meaning of
=d
~+[,
is also bounded, either above or below.
o+~*(t)
Therefore
The fundamental
Q.E.D.
lemma can thus be applied and the proof is complete.

4.14.
The following corollary parallels Corollary 4.9, in
that it introduces a function
independent of
and
thereby leads to a situation considered by J. P. LaSalle

[1968].
292
VIII.
4.15.
Coro11ar;y:.
n:
respect to
let
closed with
I x
and continuous; let
into
1jI
91, lo-
be a contin-
into iJI; if, for every compact set
C C S, there exist two numbers

every
n,
be a subset of
V be a function on
cally 1ipschitzian in
uous function on
Let
A and
B > 0
such that, for
(t,x) E I x C:
(i)
Ilf(t,x)11
(ii)
V(t,x)
nne
A+
such that
Proof.
-B;
D+V(t,x) ~ -1jI(x) ~ 0;
(iii)
then
~A;
E = {x E S: 1jI(x) =
oJ,
for every solution
x(J+) C S.
For every compact set
C C S, there exists obviously
a function
a E.5t' such that, for every
a(d(x,E.
If
1jI(t,x)
d(x,E)
x E C: 1jI (x)
is identified with the function
of Theorem 4.13, Hypothesis (iv) of this theorem is
satisfied.
On the other hand, since
is bounded on
I x C,
Hypothesis (v) of this theorem is verified for the same

choice of
1jI (t,x).
Putting
as nan
min [d(x(t),E), d(x(t),M),

as
t .... w, for every solution
follows that
sequence
A+
= M, one concludes that

1
] .... 0
1 + Ilx (t) II
such that
x(J+) C S.
It
C E, for otherwise there would exist a
{tn } C J+, tn .... w, and a point
YES \ E
such
that
x(t) .... y, and the lim inf of each of the following

n
sequences:
{d(x(t ),M)}, { 1
} ,
n
1 + IIx(tn )II
would be strictly positive, which is absurd.
Q.E.D.
4.
293
As in Section 4.10, it follows from the inclusion
4.16.
It + II l6 C E, that
It + C E U M, and if
C l6, that
It + C E.
This is the situation described in Theorem 1 (a) of J. P.

LaSalle [1968].
More precise forms of asymptotic behavior can
be proved at the expense of reinforcing the boundedness hypothesis on
f(t,x).
This is the object of the following
theorem, which is presented as an exercise, because its

proof is not very different from that of Theorem 4.13.
Exercise.
4.17.
Let
be two functions on
1/1
in
be any subset of
I
and continuous; if, for every
three numbers
for every
A, B, D
(t,x) E I
(i)
(ii)
(iii)
(iv)
(v)
>
and
into 9f, locally lipschitzian
l6
l6; let
p > 0, there exists
and a fUnction
a EJt such that,
[S \ B(M,P)]:
Ilf(t,x) II ~ A;
1/1
(t,x) > 0;
V(t,x) > -B;

D+V(t,x) < -a(1/I(t,x;
D+1/I(t,x) > -D
D+1/I(t,x) < D);
(or
thEm
1/I*(t)
as
min {1/I(t,x(t,d(x(t),M)}
w, for every solution
x(t)
such that
x(J+) C
s.
The next two statements, again left as exercises, yield

sufficient conditions for a solution to approach some set as
t
w.
4.18.
Exercise.
If one adds to the hypotheses of Theorem
4.17 that for every

1/1
(t,x)
>
(t,x) E I
d(x,E*), then
x (t)
l6
and some set
E* U M
as
t + w.
E* C l6:
294
VIII.
4.19.
M
Exercise.
= as n an;
Let
let
lipschitzian in
V: I x
x
(H)
(Hi)
then
that
V(t,x)
n .... !Jf be a function which is locally

A
and
(t,x)
E I
B >
x
and a
p > 0,
a EYe
function
[S \ B(M,p)]:
-B;
D+V(t,x) .::. -a(d(x,M));
x(J ) C
5.
let us put
Ilf(t,x) II .::. A;
x(t) .... M as
n;
and continuous; if, for every
there exist two nwnbers
(i)
be any subset of
t .... w for every solution
x(t)
such
s.
The Invarianceprinciple for Asymptotically

Autonomous and Related Equations
5.1.
As has been shown in Chapter VII, a solution
of an
autonomous differential equation approaches the largest invariant set contained in the set
V(x)
where the derivative
of the auxiliary function vanishes.
No such property
exists for non autonomous equations, because in this case,

the limit sets are not invariant.
But there are special
classes of non autonomous equations, for instance periodic,

asymptotically autonomous, almost periodic, etc. for which the
limit sets possess some easily recognizable property which we
shall term here, for convenience, "pseudo-invariance".
In
these cases, a conclusion of the following type is obtained:

any solution
set of
approaches the largest pseudo-invariant sub-
E.
A good way to establish these pseudo-invariance re-
suIts is to prove first a regularity theorem for the solutions
5.
295
of the differential equation.

ing in itself.
This theorem is also interest-
We state and prove it hereafter for differen-
tial equations of the Caratheodory type, a kind of equation

frequently encountered in control theory.
On the fundamental
theory of such equations, cf. E. A. Coddington and N.

Levinson [1955].
5.2.
General Hypotheses.
Let
and consider the space Y
be a domain of iJI x iJln
IjI
of functions
f (t,x)
on
IjI
into
~n, with the following prOperties:
(i)
is Lebesgue measurable in
(ii)
is continuous in
(iii)
for any compact subset

a
= inf
{t: (t ,x) E K}
for fixed
for fixed
K of
and
1jI,
[a,b]
such that
mK(t)
t;
i f we put
= sup
EK} , there exists a real function
{t: (t,x)
~(t)
[a,b], or,
for some
p E ]1,"'[, mK(t)p
ble over
[a,b], and further, for every

~
on
is bounded almost every-
where and Lebesgue measurable over
K: Ilf(t,x) II
x;
is Lebesgue integra(t,x) E
mK(t).
As is customary, we shall not distinguish between two

functions of
Ywhen they are equivalent, i.e. when, for
every fixed
x, they differ on a subset of measure zero of the
appropriate set of
ogy on
.?
Suppose
values.
f
is a compact subset of
ous functions
z(t)
and
1jI.
g
If
Let us now introduce a topolare functions of Y

ZK
is the family of continu-
z CiJI
K, we define
on some interval
such that their graph is in
and
into iJln
and
296
VIII.
dK(f,g)
[f
sup
E ZK
As is apparent, d K
Ilf(T,z(T - g(T,z(TIIPdT]l/P.
JZ
yields the desired topology~
associated with it.
Let
Obviously, if
quence, stating that
"f i .... g
as
for "whatever the compact subset

-+
and
This topology is clearly Haus-
dorff and therefore metrisable.
is a family of semi-distances on
d(f,g)
{f i }
be a distance
CY
is some se-
i .... 00" is but a short cut

K
1/J, dK(fi,g) .... 0
of
as
COli
We shall consider, for some point

some function
(to'x O)
and
~ the Cauchy problem
x=
f(t,x),
(5.1 )
Further, there will be a sequence of such problems, i.e.
x.
(5.2)
= f. (t,x.)
~
(to'x O)
Conditions (i) to (iii) guarantee that, through

(resp.
(tOi'x Oi
'
there passes at least one Caratheodory

(resp.
solution of the corresponding problem (5.1)
(5.2.
All solutions mentioned below will be understood in the sense

of Caratheodory.
In the lemma below, all trajectories will be confined
to some compact cylindrical subset
T
=J
x B, where
"x - Xo II < r}
the only function
of
- 1, to + 1] and
0
for some quantities 1 > 0
J
[t
mK (t)
~,
defined as
and
{x E.~n:
r > O.
of Hypothesis (iii) to be mentioned
explicitly will be associated with
and the function
there will be no possible misunderstanding if we write it

simply
met).
As
f,
5.
5.3.
Lemma.
297
In these general hypotheses, if

f.
as
00
and if
(tOi'x Oi )
{x.: J
~
B}
is a se-
quence of solutions of the corresponding problems (5.2), then

(a)
there exists a subsequence
and a function
k
00,
x: J
uniformly for
xi(k) (t)
x(t)
as
t E J:
(b)
x(t)
is a solution of problem (5.1):
(c)
if there exists no other solution of problem
(5.1), x. (t)
Proof.
The
(a)
x(t)
tinuous.
If
as
xi
trajectories are in
t E J.
uniformly for
00,
are uniformly bounded, since their

T.
t I , t2
i, p > 1
every
such that
{xi(k): k = 1, 2, }
and
Let us show that they are equicon-
are any two points of
J, one gets for
q = p/(p-l), and using Holder's in-
equality
II xi (t 2 ) - Xi(t l )II ~ I
Jtl2
+1(211 f (T,X i (TlliidTI
IIfi(T,xi(T - f(T,x i (TII dT I
[dT(fi,f) + [fJm(T)PdT]l/P]lt2-tlll/q.
If
P = 1
and
M is a bound on
met), one obtains as easily
that
The equicontinuity of the

fi
f.
p = 1.
xi
follows in both cases, since
This result may not be immediately evident for

But observe that the equicontinuity obtains for the
subfamily of the
xi
corresponding to every
some sufficiently large
n.
greater than
On the other hand, the equicon-
tinuity is obvious for the finite family

(b)
We write henceforth
{xi}
{xl' ,xn }.
instead of
{xi(k)}
VIII.
298
for the uniformly convergent subsequence which we know to

exist owing to the classical theorem of Arzela-Ascoli.
One
gets easily that

xi(t) = x iO + It f(T,x(TdT + It [f(T,xi(T-f(T,x(T]dT
tOi
tOi
+ It
t
But, for
[f. (T ,x. (T
- f (T ,x. (T ]dT.
(5.3)
1 1 1
Oi
p > 1,
and, by the dominated convergence theorem, the second member

approaches zero
as
+~.
Further
l i t [fi(T,xi(T - f(T,xi(TdTII
tOi
~dT(fi'f)lt-tOill/q,
and again, the right member approaches zero.

limit for
Passing to the
in both members of (5.3) proves that
is a solution of the Cauchy problem (5.1).
x(t)
p = 1
The case
can be taken care of even more simply.

(c)
This part of the thesis is obvious.
Q.E.D.
With a view to extending the conclusions of this

theorem in some way to non continuable solutions, we first
prove the following lemma, for which the general hypotheses
remain unchanged.
5.4.
Lemma.
with
~l C '1'2 C ~2 C '1'.
and
r > 0
Let
'1'1' '1'2
be two bounded open sets of

There exist two quantities
(to'x o ) E'I'l'
'1',
1 > 0
5.
(a)
center
21, radius
the compact cylinder of length
if
f. ... f
with
{{toi'x Oi )} C 0/ 1
and
i ... 00, for every sequence
as
0/ 2;
is contained in
(to'x O)
(b)
for
299
...
(tOi'x Oi )
as
(to'x O)
00, and
sufficiently large, all solutions of the Cauchy prob-
[to - 1,t O + 1]
lem (S.l) and (S.2) exist on the interval
and have their trajectories contained in the cylinder mentioned under (a).
Proof.
norm:
Suppose we use on
I I (t,x)1 I
It I +
the distance induced by the
0/
I Ixl
than half the distance from

l' > 0
One chooses
r > 0
smaller
0/ 1
to the frontier of
0/2' and
such that
f
where
I.
m{t)
+1'
0
m{T)dT <
to-l'
is associated here with
compact cylinder of radius

center
(to'x O) E 0/ 1
sis (a) is proved.
is contained in
(tOi'x Oi )
 1, con-
E.
N,
fi (t,xi),x i (to)
i ~ N, a
x Oi ' doesn't
[to - 1,t O + 1].
But whenever
0/ 2
[fi{T'X i (T-f(T,xi(T]dT + ftt

Oi
(to'XO)E
4 (2l)1/q
exists and is contained in

x Oi + ftt
and
and therefore the-
Assume that for some
exist on the whole interval

xi (t)
= min
Any
~2
approaching some fixed

N
solution
and
On the other hand, thesis
0/ , and select an integer
and
f{t,x)
r, length 1
for the solutions of problem (5.1).

sider a sequence
f(T,Xi(TdT.
Oi
300
VIII.
One gets that

Ilx. (t) - x011 < IlxO' - x011 + d- (f. ,f) (2l)1/q + 4 .
~
~
'1'2 ~
So, xi(t)
doesn't approach the frontier of
'1'2
and hence
[to - 1,t O + 1], which is absurd.
cannot cease to exist on
That it remains within the afore-said cylinder follows from

the above inequalities.
The proof for
=1
is left to
the reader.
5.5.
as
Q.E.D.
Theorem.
i
x.: ]()(.
~
,w. [
{(tOi'x Oi )} C '1'
let
00;
(toi'x Oi )
In the general hypotheses above, let
(to'x O) E'1'
!fin
fi (t,x), x(t Oi )
solution
()(i(k)
[t l
tl
<
i + "".
For
i = 1, 2,
... ,
x
let
Then there exist a non continuable

of problem (5.1) and an increasing
= 1,
{i(k): k
with
<
= xOi'
be some sequence such that
be a non continuable solution of
x: ]a.,w[ +!fIn
subsequence
t l , t2
as
f.~
()(
<
tl
t2
<
wiCk)
<
t2
2, }
<
w, one gets for
and
xi(k) (t)
x(t)
large, that
uniformly on
,t 2 ]
Proof.
a less
A similar theorem is proved in P. Hartman (1964), in

gene~al
setting.
The proof, which is almost the same
in both cases (the only significant differences are in the

lemmas), is outlined here for completeness.
tence of
We show the exis-
on a right maximal interval only, but the reason-
ing runs alike for left and right.
Let
'1'1' '1'2""
sequence of bounded open sets such that
Wi C '1'i+l
every
be a
for
and
'1'
U
1 < i <
00
Suppose, without loss of generality, that
(to'x O) E '1'1
and
5.
let
211
with
301
be the length of the compact cylinders associated

in Lemma 1.
~l
(to'x O).
Let
(tOi'x Oi )
in
tend to
Lemma 1 shows that there exists a subsequence
xi(k) (t), k
= 1,2, . ,
of solutions of the corresponding
problems (5.2), such that

uniformly on
belongs to
xi(k) (t)
[to,t o + 11]'
~l'
tends to some
(to + 11,x(t O + 1 1
Either
or it does not.
x(t)
If it does, we start from
this point as a new initial point to prove, by the same argument, the existence of a new subsequence, again written
xi{k) (t). with the same convergence property, but this time
on
[to,t o + 21 1 ].
Repeating this process proves either the
existence of a subsequence
over
[to'~[
xi{k) (t)
for some
done in
~1'
x(t)
uniformly on every finite interval, or allows
one to reach a point outside

~
converging to
~1'
But this point will be in
r > 1, and we can repeat in
what we have
choosing of course a new length
1r
The rest
of the proof is obvious.

5.6.
Q.E.D.
Invariance properties of limit sets.
The differential
equation to be considered from now on, namely
x = f{t,x)
will have its second member
where
= ]T,~[
for some
TE
(5.4)
defined on a set
~
or
=-
~,
and
open subset of ~n, while its range is still in ~n.

reason for particularizing in this way the set
=I
n is an
The
of the pre-
vious section will become apparent when we introduce below

the translates of
f.
Further, f
fy Hypotheses (i) to (iii) above.
is still supposed to veri-
302
VIII.
x: ]a,w[ +~n
Let us recall that if
is a non conx*
tinuable solution of Equation (5.4), a point

said to be a positive limit point of
sequence
+
of
i ....
as
t. +
J.
A+(X).
and
and is repre-
Amongst the properties of the limit sets,
only their invariance will be studied here.

we shall need the following proposition:
not empty, then
is
The set of all positive limit points
00.
is called the positive limit set of
sented by
if there exists a
of time-values such that
x*
of
w=
In this context,
if
A+(x)
is
~.
The translate of the function
by a given amount
is the function defined thus
a > 0
The following two hypotheses concerning
will be used suc-
cessively below.
f* E ~
(A)
There exists an
(B)
For every sequence
there is a subsequence
f* E ~ such that
such that
+ f*
S(f,B)
obtainable in this way.

led a limit function and
a
t.
J.
+ f*
+
{ti(k): k = 1, 2, }
ti(k)
Let us designate by
such that
as
a +
~.
as
00,
and a function
k + ~.
the set of all functions
Any such function

x.
as
= f * (t,x)
f*
f*
will be cal-
a limit equation.
Several simple remarks are appropriate here:

(1)
(A) implies (B).
(2)
~being
a vector space for the usual addition of func-
tions and product of a function by a scalar, the subset of

functions of
of
$T.
possessing property
(A)
is a linear subspace
5.
303
(3)
The same is true for (B) .
(4)
If
a' > 0, f
possesses property (A) , for any
f*
as
f*
is actually a constant with respect to
a ...
but also
co,
equation is autonomous.
f a + a , ... f*a'
as
...
a+a'
Therefore
co
...
and the limit
In this sense, an equation satisfy-
ing property (A) is asymptotically autonomous.
(5)
A rephrasing of (B) is as follows:

such that
lates of
(6)
If
t i ...
co
i ...
co,
the family
is relatively compact in
{f t
} of transi
possesses property (B) and is continuous, and if
one uses for
the topology of uniform compact convergence,
then it is known that

g(t,x)
as
for every sequence
and
h(t,x)
f(t,x)
is the sum of two functions
such that, for fixed
periodic in the sense of Bohr and

One may then say that
f(t,x)
x, g
h(t,x) ... 0
is almost
as
co.
is asymptotically almost per-
iodic, since the limit function is almost periodic.

the continuity assumption on
t ...
without
f, the class of functions sat-
isfying (B) becomes larger of course, and its extent might

well deserve some further exploration.
A set Fen
is said to be semi-invariant with res-
pect to equation (5.4) whose second member is supposed to

(to'x O) E I x F, there
possess property (A), if, for every
is at least one non continuable solution

the limit equation
x * (tl E F
for every
A set
Fen
f* (t,x)
t
with
x*: lu,w[ ... ~n
x * (to)
= xo '
of
such that
E )a,w[.
is said to be quasi-invariant with res-
pect to equation (5.4) whose second member is supposed to

possess property (B) if, for every
(to'x O) E I x F, there
VIII.
304
exists a function
E S(f,B)
f*
and at least one non continu-
x*
able solution
x (t) E F
such that
x = f * (t,x)
of
for every
with
E )Il,w[.
As is well known (see e.g. P. Hartman [1964), for any

solution of an autonomous differential equation with continu-
n~
ous right member, A+(x)
is semi-invariant.
If uniqueness
of the solutions is assumed, this set is even invariant.

Let us now prove two simple but important theorems, first of
semi-invariance for equations whose second member satisfies
Hypothesis (A), and then of quasi-invariance for the case of
property (B).
5.7.
Theorem.
For every solution
of Equation (5.4)
n ~
whose second member possesses property (A), A+(x)
is
semi-invariant.
Let
Proof.
{t. }
*
Xo E A+ (x)
be such that
x (t i )
xOi
If
E I
t.
to
ti - to
to belong to
deft
the
i -
= Xo*
x(t i )
as
is the limit function of

i
00
Let
Put
00
f, then
large enough for
I, one may write that:

t
,f * )
as
00
is a solution of
tends to some solution
of the equation
x = f * (t,x).
this subsequence belongs of course to

thesis.
*
Xo
x=
ti
_ t
(t,x),
By Theorem 5.5, an appropriate subsequence of
x(t + ti - to)
x * (to)
and starting with
Further, x(t + ti - to)

x(t o ) = x oi
and
00
*
f (t ,x)
for any
As observed above,
(J ~.
x * (t), with
But the limit of
A+(x) n~.
Hence the
Q.E.D.
6.
305
Dissipative periodic systems
5.8.
Theorem.
For every solution
of Equation (5.4) whose
second member possesses property (B), A+(x) n ~
is quasi-
invariant.
The proof is a kind of obvious paraphrase of the preceding one.
5.9.
Extensions of the invariance principle.
Suppose now
for simplicity that we come back to the general hypotheses

described in Sections I and 4.1, which are a particular case
of the Caratheodory conditions of the present section.
As-
sume further that the hypotheses of Corollary 4.9 are satisfied.

then
As has been observed already (see 4.10), if

A+ C E.
fore that
If one knows that
x(t)
A+
as
A+
is compact, and there-
~, and that
C~,
f(t,x)
satisfies
property (A) (or (B, it follows immediately from Theorem 5.7

(or Theorem 5.8) that
x(t)
approaches the largest semi-
invariant (or quasi-invariant) subset of
E.
A similar con-
clusion holds of course in correspondence with Corollary 4.15.
6.
6.1.
Dissipative Periodic Systems
Consider the Cauchy problem
x=
(6.1)
f(t,x),
x(O) = x o '
where
f(t,x)
(6.2)
is supposed to be defined and continuous on
~x ~n, sufficiently regular to ensure uniqueness of the
solutions, and
w-periodic, i.e. f(t +w,x)
w> 0
(t,x) E .91 x ~n.
6.2.
and any
= f(t,x)
Suppose that all solutions can be continued to
We define the translation operator

m
',n
for some
+~.
306
VIII.
and construct what is known as a semi-flow (see, for instance,

N. P. Bhatia and
o.
Hajek [1969]), a function

~n ...
on
<7.. ./Y.
.A x
../
(k , xo ) ... Tk xo '
...".n,
!:K
where ~+ = {a, 1, 2, }, verifying the following properties:

(i)
(ii)
(iii)
TOXo = xO:
Tk(TlxO) = Tk+lxO
TkxO
k
for any
and
in J.V+:
is continuous with respect to
x o ' for any
in ~+.
Further, we define the solution of the semi-flow

through
Xo E n
as
k
x: /11'+ ... !}In,

its positive
k ... T x o'
x},
and the corresponding positive limit set
...
6.3.
every
and
A set
Xo E S
S c!}ln
and
ki
T xo'" x
as
i ... oo}.
will be said to be invariant if, for
and every
x(n'w:O,x O) E S
negative integers).
00
n' E
(~
tJ:
x (n 'w:O,x O)
is defined
is the set of all positive and
Paraphrasing what has been done for auto-
nomous ordinary differential equations, one proves easily the

following proposition.
Proposition.
If
y-+ (xl
- n n
empty and compact, -+
A (xl
is bounded, -+
A (xl
is invariant and
is non-
6.
6.4.
Similarly, one can transpose LaSalle's Theorem VII.3.2.

Theorem.
Let
a solution through
S
307
be a compact subset of ~n
Xo
such that
y+(x) C S.
and
Suppose
is a continuous function such that, for every
+ ~
V(~xO) ~ V(Tk-lx O).
x
V:
k > 0:
M is the largest invariant subset
If
of
{x E S: Tx E Sand
E
then
x(k)
M as
V(Tx) = V(x)},
~.
The proof is a mere repetition of the proof of Theorem

VII.3.2.
6.5.
We are
now ready to obtain a dissipativity condition
for Equation (6.1).

Theorem.
Suppose all solutions of (6.1) can be con-
tinued to infinity and there exists a positive constant

(i)
V:
~ \
V(x) ~ Hllxll), with
+5/1
such that:
$: [R,~[ +9/
ous increasing function such that
(ii)
for any
such that
e BR
and
a continu-
$(r)
Tx
+ ~
as
BR:
V(Tx) < V(x);

then the system is dissipative.
Proof.
Due to the observation stated as Exercise VI.6.19,
the only thing to prove is that, for each

exists a
t > 0
such that
x (t; O,x O) E BR
wrong, there would exist an

every
t > 0: x(t;O,x ) E
-
~n
x E ~n, there
If this were
Xo E ~n \ BR such that for

, B
R
Then, from assumptions
VIII.
308
(i) and (ii), the solution
of the flow
Xo
through
jr
is such that
y+(x) C S = {x E ~n \ BR: vex) < vex )}
0
and
is compact.
E=
is empty.
Further, (ii) implies that the set
{x E S: Tx E S
and
V(Tx)
= Vex)}
Hence a contradiction, by Theorem 6.4.
Q.E.D.
Notice that V. A. pliss [1964] proves that the conditions of Theorem 6.5 are also necessary.
A condition such
as (ii) is easy to verify using Dini derivatives.

6.6.
Exercise.
(ii-I) for any

(ii-2)
V(x)~
that either
6.7.
V:
such that
for any
Tkx E BR
or
Theorem.
function
Replace, in Theorem 6.5, assumption (ii) by
BR
and
Tx
k > 0
BR, there exists a

V (Tkx) < vex).
Suppose that, for some
BR: V(Tx) <

such
R> 0, there exists a
!JI x (gil \ BR), locally lipschitz ian in
and
continuous, such that:

(i)
(ii)
Vet +
IIl,X)
= V(t,x)
V(t,x) ~ <p(llxll)
with
p: [R,oo[
ous increasing function such that
!JI a continup(r)
00
as
r . .,. 00;
(iii)
for each
(t,x) in !JI x (~n \ BR), D+V(t,X) < o~
then the system is dissipative.

Proof.
Again the only thing to prove is the existence of an
R' > 0
such that for every
such that
x(t~O,xO)
E BR,.
E ~n
there is a
> 0
It is clear from assumptions (ii)
6.
309
and (iii) that all solutions can be continued to

choose
R'
Let us
such that
E [O,w], Ilxll = RL
(R') > sup {V(t,x): t
It follows then from assumptions (ii) and (iii) that for

every
and
E BR and
x
Tx
BR"
then
~ OJ
x(tjO,x O)
x(tjO,x) E BR
Using (iii), we get that
V(O,Tx)
E BR,.
Hence if
for every
= V(w,Tx)
6.8.
w(x)
/Z'J
W(x)
= 0,
into
/Ul
~j
Q.E.D.
in
such that
for each
(iii-2)
Replace assumption (iii) in Theorem 6.7 by:
for each
in ~ x (gpn \ BR), D+V(t,x) < 0

tl E [O,w] with D+V(tl,x) < O.
(t,x)
and there exists a

6.10.
as
D+W(t,x) ~ O.
Exercise.
(iii')
V(O,x)
All
W is a locally lipschitzian function from
0, where
BR
6.9.
(t ,x)
for each
n , -
BR ,
Replace assumption (iii) in Theorem 6.7 by:
Exercise.
(iii-I)
This proves the theorem.
E [O,w].
< V(O,x).
assumptions of Theorem 6.5 are satisfied with

auxiliary function.
Various examples of second and third order dissipative
equations appear in V. A.Pliss [1964].
As a simple illus-
tration, let us consider the class of electrical networks

introduced in Section VII.6 and described by the equations
dv
C dt
= f,
L di
= g,
dt
with
and
defined in Section VII.6.3.
again the notations of the same section, that
Suppose, using
310
(a)
VIII.
R, Land
are constant, symmetric, positive definite
matrices:
(8)
= aCt)
is
Theorem.
w-periodic for some

If
v T dG > 0
dV
w > o.
for large enough
I Ivl I,
the system is dissipative.

Proof.
Xo E ~n, there exists a t > 0
every
BR
The only thing to prove is that for some

such that
R > 0
and
x(t;O,xo)E
Consider the auxiliary function
whose derivative
is negative for
I I (v,i) I I
large enough.
Theorem 6.7 ap-
plies.
Q.E.D.
7.
The one-parameter families of auxiliary functions have

been introduced by L. Salvadori [1969] and then used in several subsequent papers, for instance L. Salvadori [1971], P.
Fergola and V. Moauro [1970], L. Gambardella and L. Salvadori
[1971], L. Gambardella and C. Tenneriello [1971], A. D'Anna
[1973].
Concerning this method, see also W. Hahn [1971].
Theorem 2.5 on weak attractivity which has been chosen to illustrate this method is adapted from a theorem of V. M.
Matrosov [1962]1 (cf. Section 11.2 of the present book) on
asymptotic stability.
In his paper of [1971], L. Salvadori
showed how the bound on the second member of the differential
7.
311
equation could be dispensed with:

in Corollary 2.9.
this result appears here
The comparison of the two methods of proof
of Theorem 2.6 is borrowed from J. L. Corne [1973].
The
latest and most general paper on one-parameter families of

auxiliary functions is L. Salvadori [1974].
The use of a vectorial auxiliary function meant to compensate for the fact that
V(t,x)
appears in N. Rouche [1968].
is not negative definite
In connection with Section 3,
one will remember that such a function is used to expel the

solutions from some compact set.
This result has been ex-
tended successively by N. Rouche [1971], M. Laloy (1974)2

and N. Rouche [1974) and inspired Lemma 3.2.
Section 4 is adapted from J. L. Corne and N. Rouche
[1973].
It has already been mentioned in the text that the
important Corollaries 4.9 and 4.15 are due to J. P. LaSalle

[1968], who however limits his study to solutions contained
in some closed subset of
n.
As for Section 5, two papers opened the way towards

rather general regularity theorems and the problem of asymptotically autonomous equations:
and Z. Opial [1960).
they are L. Markus [1956)
The generalizations of the invariance
principle for periodic equations appears in J. P. LaSalle

[1962], for asymptotically autonomous equations in T.
Yoshizawa [1963) (see also T. Yoshizawa [1966)
and for
asymptotically autonomous equations in R. K. Miller [1965).

A regularity theorem concerning equations with continuous
second members and the topology of uniform convergence will be
found in P. Hartman [1964).
For a topology akin to the one
312
VIII.
adopted in the present text, but in connection with Volterra

integral equations, and therefore encompassing ordinary differential equations, see R. K. Miller and G. R. Sell [1968],
[1970].
Similar regularity theorems along with pseudo-invari-
ance properties studied in the setting of dynamical systems

associated with non autonomous differential equations appear
in R. K. Miller and G. R. Sell [1970] and G. R. Sell [1971].
Theorems 5.7 and 5.8 as they appear here come from N. Rouche
[1976].
Concerning the dissipative systems dealt with in Section 6, cf. V. A. Pliss [1964], which contains an extensive
bibliography, as well as C. Corduneanu [1957], V. M. Gerstein
[1969], V. V. Nemytskii [1965] and, at a more abstract level,
J. E. Billotti and J. P. LaSalle [1971] and J. K. Hale,
J. P. LaSalle and M. Slemrod [1972].
The comparison method
has been applied to dissipative equations by V. M. Matrosov

[1969], a paper which also gives an extensive bibliography,
and N. Pavel [1971].
Various applications appear in A. de
Castro [1953], B. Manfredi [1956] as well as in the book of

J. M. Skowronski [1969] and in several papers of the same
author.
Cf. also S. Ziemba [1961].
CHAPTER IX
1.
Introduction
The role of the comparison method, as already studied

in Section II.3, might be roughly characterized as follows:
the solutions of the comparison equation push on before themselves those of the original equation.
Following V. M.
Matrosov [1973], this can be viewed as an interesting extension of the notion of a mathematical model.
In the usual
sense, a mathematical model operates approximately like the

system it represents, whereas the solutions of the comparison
equation fit approximately those of the original one, but
while remaining "on the same side" at any time.
speak of a one-sided model.
One might
The comparison method is suf-
ficiently important in itself to justify the presence here

of a complete chapter devoted to this subject.
Further, it
introduces to some typical applications.

In Section II.3, we considered a scalar comparison
314
IX.
equation.
After some preliminaries on differential inequal-
ities, we generalize here this theory to the case of vectorial equations and apply it thereafter to two practical
problems:
the stability of composite, inter-connected sys-
terns, and a problem of market stability in a Walrasian economy.
A last section shows how the comparison method can be
applied, besides stability and attractivity, to a wide variety of qualitative concepts.
2.
2.1.
Let
Differential Inequalities
r~m
be an open subset of
for some integer
m,
and consider the differential equation

u
F(t,u)
associated with a function
F: I
~m
We shall need on
(2.1)
x
~ +~m,
some kind of partial order de-
fined in the following natural way:

we write
u > v
larly, u > v
function
if
u. > v.
~
i
v.
2.2.
for u
for every
u. > v.
~
and
1, ... ,m.
for
i = 1, ..
in .c:;f'm,
Simi-
,m.
The
will be said to be quasi-monotone increasing if,
= l, ,m,
and
will mean that
for every pair of points

every
(t,u) + F(t,u).
(t,u)
one gets
and
Fi (t,u)
(t,v)
~
in
I x
Fi(t,V)
and
whenever
u > v.
Most of this section will be devoted to explaining and
proving the following important property:
if
is continu-
ous and quasi-monotone increasing, all solutions of (2.1) issuing from an initial point
by two particular solutions:
(t ,u ) E I x ~ are bracketed
o 0
a maximum and a minimum solu-
2.
Differential inequalities
tion.
solution
u+: [to'w[
315
~m
with initial conditions
(to'u O)
is said to be a (right) maximum solution if every

/1I)m
other solution u: [to'W[
+ ~ , passing through
(to'u o )' is
u+(t) ~ u(t)
such that
Similarly, a solution
tial conditions
u (t)
(to'u O)
defined on
and defined on some interval

t E [to'W [
[to'w[, with ini-
is called a (right) minimum solu-
Hon if, for any other solution
for every
t E [to'W[ n [to'W[.
for every
u(t)
issuing from
u-(t)
[to'W[, one has
n [to'W[.
[to ,u O[
~
u(t)
To prove the existence of
maximum and minimum solutions, we shall need the following

lemmas.
Lemma.
2.3.
G: I x ~
Let
gem
monotone increasing function and
be a continuous quasiv: [to'w[
v=
w finite or infinite, be a solution of

w: [to'w[
(i)
w(t O)
vetO)
(iii)
D-w(t) < G(t,w(t
wet) < vet)

a)
for
t E 1to,to +
t E 1t o 'w[,
D+wi(t O)
<
[.
small enough,
This is obvious, by rea-
w(t o ) < vetO).
i, w. (t ) = v. (to).
~
If
t E 1to'w[.
for every
son of continuity, if
some
for
Let us prove first that, for
wet) < vet)
G(t,v).
D+W(t O) < G(to,w(to~
Proof.
~m, for some
is a continuous,function such that
ii)
then
Suppose then that for
One has
Gi(tO,w(t O
Gi (to,v(t o
dv.
-2:.(t o)
dt
Hence
D+ (w.
v.)(t O) < 0
~
which implies that, for
small enough and
t E 1t o ,t o + [,
316
IX.
wi(t) < vi (t).

b)
E,W[
t.
w(t)
such that
these
Assume now that there exist values of
gets
and let
be the infimum of
Notice by the way that, for a partial order,
is not equivalent to
for
v(t)
t E ] to +
E ]to,T[
and
>!
W(T)
wi(T) = vi(T).
Then
V(T).
T E [to + E,W[, w(t) < v(t)

For one
at least, one
Therefore, as above
n > 0
and there exists an
such that for some
E [T - n,T[:
wi(t) > vi(t), which contradicts the definition of
2.4.
Lemma.
If we replace assumptions (i),
T.
Q.E.D.
(ii) and (iii)
of Lemma 2.3 by
(i)
(ii)
(iii)
then
w(t O)
v(t O);
D+W(t O) > G(to,w(t O;

D_w(t) > G(t,w(t
w(t) > v(t)
for
for
E ]to'w[;
E ]to'W[.
The proof is similar to the one of Lemma 2.3.
We are
now ready to prove an existence theorem for maximum and minimum solutions.
2.5.
Theorem.
If
is a continuous quasi-monotone increas-
ing function, there exist through every point

~
(to'u O) E I x
a unique non continuable maximum solution and a unique non
continuable minimum solution.

Proof.
tion.
Let us prove first the existence of the maximum soluConsider a cylinder
2.
where
and
(i)
(ii)
are chosen such that

x '.,
T C I
b
= Ma
I t is known that
317
M>
where
M
sup
IIF(t,u) II.
(t,u) ET
can be chosen large enough in order
v > 1, the Cauchy problem
that, for any integer
1, ... ,m
has at least one solution

Further, any solution
fined on
and
[to,t o + al
(v-I)
v~
for
of (2.1) through
1
v-I
= F. (t,v (v-I) )
v
and
(to'u O)
+ ----- > G~(t,v
[to,t o + a)
[to,t o + a) + B[UO,b).
v(V):
>
2.
v-
1)
is de-
It follows from Lemmas
2.3 and 2.4 that
and the sequence of functions

to a function
u+(t).
the functions
v(v)
val
u
[to,t o + a)
v(V) (t)
converges pointwise
But this convergence is uniform since

are equicontinuous on the compact inter-
(cf. J. Dieudonne [1960, p. 136J).
Further,
is a solution of (2.1) as results from passing to the
limit in the equation

v(V) (t)
= Uo
t
t
[F(cr,V(V) (cr
+~)dcr.
It is a maximum solution, as follows from (2.2).
By a stand-
ard argument, this solution can be continued to the right,
318
IX.
until it approaches the boundary of
(see e.g. N. Rouche
and J. Mawhin [1973, Vol. I, p. 82]).
At last, the unique-
ness of the maximum solution is obvious.
The existence of
the minimum solution is proved in the same way.
Q.E.D.
The following result is basic for the comparison m

method.
2.6.
Comparison Lemma.
.
tone increas~ng
Let
.
funct~on and
solution through some point
w~
[to'w[ ~ !iRm,
(t, v (t
uo ;
(ii)
Dv(t)
2.
F(t,v(t
t
a)
for
is any derivative of
in
Assume
v:
and
Dv(t)
~.
the maximum
w, is a continuous function such that
vetO)
for any
t E ]to'W[,
vet).
Then
vet)
<
u+(t)
[to'w[.
Consider the function

vet)
vet) - ft F(T,v(TdT.
to
It follows from (ii) that

]to'w[
(to' u O) E I x
(i)
where
Proof.
be a continuous, quasi-mono-
+ [to'w [/1}Jm
u:
~ ~
is a decreasing function on
and, by continuity, on
Therefore, for any integer
[to'W[.
Hence
1, one may write:

1, ... ,m;
As in Theorem 2.5, let
v(V): [to,t o + a] ~!iRm
be a solution
2.
319
of
It was proved in Theorem 2.5 that for a small enough, the

functions
u+.
v(V)
converge uniformly to the maximum solution
On the other hand, it follows from Lemma 2.3 that

vet) < v(V)(t)
f or
By passing to the limit as
v +
t E]t 0' t 0 + a[.

~,
one gets that

(2.3)
b)
It remains to prove that (2.3) holds for any
t E [to'w[.
all
If it were not true, let
tE]to+a,w[
VeT) .::., U+(T).
such that
be the infimum of
v(t>.::.,u+(t).
The reasoning under a) with
proves that, for
a'
small enough and
Bycontinuity
replacing
to
t E ]T,T + a' [:
vet) ~ u+(t), which contradicts the definition of
T.
Q.E.D.
The same type of proof yields the following result.

2.7.
Lemma.
Let
ing function and
F
u
through some point
w~
be a continuous, quasi-monotone increas[to'w[
gfm
(to'u o ) EI x V.
the minimum solution

Assume
w, is a continuous function such that
~m
v: [to'w[
(t,v(t
+ ~
E I x V
and
(i)
(ii)
v(t O):::'
uO~
Dv(t):::. F(t,v(t
where
Dv(t)
u-(t)
for any
for
t E ltO'w[,
is any Dini derivative of

t E [to'w[.
vet).
Then
vet) >
320
IX.
3.
A Vectorial Comparison Equation in Stability Theory
3.1.
The general idea of this section is to generalize the
comparison method as introduced in 11.3 to the use of a vector comparison equation like (2.1).
More precisely, consider
the equation
x=
(3.1)
f(t,x)
along with the general hypotheses of Section 1.2.2 and a vector comparison equation
(3.2)
F(t,u)
with a critical point at the origin
F: I x ~
that
creasing.
~m
u+ : K
[to'w[
solution of (3.2) through

+
Let
u-(t)
Assume further
is continuous and quasi-monotone in-
Given some initial conditions
shall write
= O.
(to'u O) E I x
~m, t + u+(t)
be the vector
we
for the maximum

m
Similarly, u : L +~,
(to'u o )
represents the minimum solution through
e E ~m
~,
e = (1, 1, , 1)
(to'u o ).
and suppose
I :) {u: 0 < u < eL
3.2.
As already noticed in 11.3.2, the only solutions
(3.2) to be considered here are such that
u(t)
O.
of
There-
fore the relevant stability concepts for (3.2) are the following straightforward modifications of our previous ones.
origin
u = 0
The
will be called stable if
('Ie> 0) ('Ito E I) (:10 > 0) ('lu 0 : 0 <
Uo
< oe) ('It ~ to,t E K)
u+(t)< e e,
and uniformly stable if
('Ie> 0) (:10 > 0) ('Ito E I) (Vu O: 0 <
Uo
< oe) ('It ~ to ,t E K)
u+(t) < ee.
3.
A vectorial comparison equation
Provided
321
1, it is clear that for the maximum solu-
tions mentioned in these definitions, K

3.3.
[to'oo[.
Similarly, one defines uniform attractivity as
(:10 > 0) ('f/;
and
> 0) (:10 > 0) (~to E I) (~uo: 0 < U o < ee) to +
(~t ~ to + o,t
K) u+(t)
<
(J
E K
Ee.
In Chapter VI, we studied six variants of this definition.

Here, due to the quasi-monotonicity of
F, only three of them
are distinct, as will be shown now.

Proposition.
Wl
In the language of Chapter VI, let
be a word obtained by combining some of the expressions

(:Ie> 0),
(~E
> 0)
and
E I), and let
(~to
void or identical with
(~to
E I).
be either
W2
Then for Equation (3.2),
the concepts
p: Wl (:10 > 0) (~uo: 0 ~
(~t ~ to + o,t E K)
o ~ ee) W2 (to + 0 E K)
and
u+(t) < Ee,
are equivalent.
Proof.
Obviously
q.
Suppose now
is no loss of generality to assume that

which case
K(to,ee)
responding to
and for any
o = ee
u O: 0
[to'oo[.
in
Uo <
q.
is true.
0 < 1,
Let us choose
There
1, in
in
cor-
From the Comparison Lemma 2.6
e, one has
322
IX.
which implies that

position
K(tO'u O)
[to'oo[
and pro-
p.
Q.E.D.
Among the three distinct attractivity concepts for

(3.2), equi-attractivity turns out to be particularly useful.
Its definition reads:
(Vt o E I)
>0) (VE> 0)
(~o
(~cr
> 0) (Vu O: 0
~ Uo <
oe) to + cr E K
and
(Vt
to + cr,t E K) u + (t) < Ee.
The proposition above implies that it is equivalent to attractivity for Equation (3.2).
The instability concept relevant for Equation (3.2) is:
3.4.
(~E
> 0) (~to E I) (Vo > 0) (~uO: 0 ~ Uo ::. oe) (~t ~ to,t E L)

u-
3.5.
Theorem (V. M. Matrosov [1962]2)'
a function
V: I
~m
(ii)
fEe.
as described in 3.1 and a continuous function

a E
that, for some function

(i)
(t)
and every
and such
(t,x) E I
n:
max Vi (t,x) ~ a(llxll); V(t,O) = 0;

D+ V(t,x) .::. F(t,V(t,x;
then
stability of
(b)
equi-attractivity of
tivity of
(a)
implies stability of
u
b EJe
n
(iii)
=0
0;
implies equi-attrac-
x = 0;
if moreover, for some function

x
max Vi(t,x) .s.b<llxll),

i
and every
(t,x) E I
3.
323
then
(c)
bility of
(d)
implies uniform sta-
x = 0:
uniform attractivity of
u = 0
implies uniform
x = O.
attractivity of
Proof.
u = 0
uniform stability of
Because of (i) and (ii) and the Comparison Lemma 2.6,
we have, for any
(to'x o ) E I x Q
+
u (t:to,V(to'x O
and any
to' t E
is defined, that
(3.3)
The origin being stable for (3.2), there exists a
o* (to,e)
t
such that, for arbitrary
> 0
to: u+ (t:to'u O) < a(e)e.
u+ (t:to'u O)
V(to'x o )
such that
~
0 *e.
uo
0 *e
[to'oo[.
By continuity, there
I IXOI I < 0
implies
Uo
But it follows then from (3.3) that
Ilx(t:to'x O) II < e
Thus
for any
and Thesis (a) is proved.
x = 0
is stable,
This proof was a kind of repeti-
tion of the corresponding proof in Theorem II.3.4.
The same
is true for (b), (c) and (d).

3.6.
Example.
x
Q.E.D.
Let us try, for the
= e-tx
system of equations,
+ y sin t - (x 3 + xy2)sin 2 t,
y = x sin t + e-ty -
(x 2y + y3)sin 2 t,
to use the auxiliary function

V = Ax
and
Notice in passing that
is defined over
exists a
u O: 0
2
+ 2Bxy + Cy ,
in connection with Theorem I.4.2.
(A > 0, B2 - AC < 0)
One computes
324
IX.
an expression which is positive for any

ger}.
E {krr,
Hence Theorem I.4.2 cannot be applied.
an inte-
One can how-
ever use the quadratic vector function

V
x + y)
, (x -
y)
2 T
)
which is such that
V < 2
-t
+ sin t
-t
- sin t
Jv.
whereas, for each of the equations
= 2(e-t
sin t)u,
the origin is uniformly stable, as can be shown by direct

integration:
lui
luOI exp {2[e
-t
0 - e- t cos to
< IU o Ie
3.7.
Exercise
continuous function
(E)
(iii)
= o.
as described in 3.1, a
x, and a continuous function

a E.5f'
and every
k: I ~~
(t,x) E I x
V(t,O)
mrx Vi (t,x) ~a(llxll),
such
n:
0:
...
D+(k(t)V(t,x)) < F (t,k (t)V(t,x)):

k(t) > 0, k(t)
then stability of
of
).
n ~ ~m which is locally lip-
V: I x
that, for some function

(i)
-to
t]}
(N. P. Bhatia and V. Lakshmikantham [1965]).
schitzian in
exp (2e
+cos
=0
'"
as
-)0.
00;
implies equi-asymptotic stability
3.
325
Notice that assumptions (i) and (ii) can be replaced

by
(i ')
m~x Vi(t,x) ~ k(t)a(1 Ixl I), V(t,O)
0;
o+V(t,x) ::. F(t,V(t,x.
(ii ')
These results should be compared with Theorem I.6.3l

and Exercise II.3.9.
3.8.
Exercise.
With assumptions (i), (ii) and (iii) of
Theorem 3.5 and if

tivity of
'l':)
{u: u
O}, uniform global attrac-
u = 0, i.e.
('fI5 > 0) (Ve: > 0) (30 > 0) (Vt o E I) CVu o : 0 ::. u o < 5e) to+O E K
and
('fit ~ to + o,t E
K)
u+(t) < e:e
implies uniform global attractivity of

3.9.
Theorem.
is locally lipschitzian in
a E.5t'
(i)
(ii)
as des-
V: I x Q +!7tm which
and such that, for some func-
(t,x) EI x Q:
and every
max Vi(t,x) < a(llxll);

i
for any 5 > 0 and to E I, there exists an
veto ,xo ) > 0;
such that
xo E B(0,5 0 )
(iii)
= O.
cribed in 3.1 and a continuous function
tion
o+V(t,x) ~ F(t,V(t,x;
then instability of
u = 0
implies instability of
x = o.
Proof.
The origin being unstable for (3.2), there exist
e: * > 0
and
to E I
such that
(Elt> t , tEL) u-(t)
e:*e.
(V5 * > 0) (Elu O: 0 ::. u ::. 5 * e)

Let
e:
be such that
a(e:) < e: *
326
IX.
0* > 0
0 > 0, let us choose
For a given
This is possible, because of (ii).

o < u < o*e
0
and
such that
Let us now fix

such that
u-(t)
u O:
E*e.
is chosen according to (3.4) and
If
theorem is proved, since a solution cannot

without leaving
B(O,E).
If
cease to exist
E J(to'x o ), we can apply
Lemma 2.7 and obtain
Hence
I Ix(t) I I > E.
Q.E.D.
There exist several variants of this theorem.
Let us
give two of them.

3.10.
Theorem
exist a function
function
V: I x
(V. M. Matrosov [1962]2).

F
as described in 3.1 and a continuous
n +~m,
which is locally lipschitzian in
and such that, for some constant
(i)
(ii)
Suppose there
M and every
(t,x)
VI (t,x) < M;
D+V(t,X) ~ F(t,V(t,x;
then, the following "instability" concept

(:It E I) (~o > 0) (:Ix E B(O,o (:It> to,t E L)
0 0 u-(t;t ,V(to'x
100
implies instability of
= O.
> M
E I
x
x
4.
Stability of composite systems
3.11.
327
Exercise (V. M. Matrosov [1962]2)'
exists a 5C'2
V: I x n .... !iR
function
Suppose there
(t,x) E I x n:
(i)
(ii)
(iii)
V(t,x)
is bounded from above;
V > aV + 2bV, where

for any
a ~ 0, and
b > 0
if
a = 0;
t E I, the function
takes positive values in any neighborhood of x

then the critical point
x = 0
is unstable.
A wealth of related results appear in V. M. Matrosov

[1962]2'
4.
4.1.
Stability of Composite Systems
The difficulty of studying the stability properties of
a system increases usually with its dimension.
One way to
get over the difficulty of building a Liapunov function for a

large system might be described as follows:
one considers
the system as constituted by the interconnection into a complex whole of several smaller subsystems, each of which is
simpler to study.
The complete system will be called com-
posite.
4.2.
More precisely, suppose the basic building block of our
system is a so-called transfer system, described by the equations
x.
fi (t,x i ) + DiU i ,
y. = H.x.,
~
l, ...
,m,
0;
328
IX.
n.
x. E~ ~
where
p.
is the state vector, u i E ~ ~
is the input
~ni
and Yi E ~qi the output. We assume that the f.~ : I x
-+ ~ni
satisfy the general hypotheses I. 2. 2 and that the ma-
trices
Di
and
Hi' being rectangular with the appropriate
number of rows and columns, are constant.
An example of such
a transfer system is the tansistor model given in II.6.S,

where
and
Xi
The
u.
m transfer systems are interrelated by the fol-
lowing equations:
u.
1 < j
< m
B . y.,
~J
1, 1m,
are appropriate constant matrices.
where the
Then the
system as a whole is described by the equations

f~ (t,xi) +
4
4.3.
i
1 ~ j <
~ ~J J
Theorem (F. N. Bailey [1965]).
= l, ,m
in System (4.1).
1
there exist ~ -functions
Suppose
Suppose for
n.
Vi: I x ~ ~
and
constants
(4.1)
D.B .. H.x J., i = 1, . ,m.
-+
Bii = 0
for
= l, ,m,
and positive
such that:
(i)
(H)
(Hi)
av.
av.
-2:. + ~ (t,x. )f(t,xi:) < -c 3 .llx.1I 1

at
oXi
~
'~
~
avo
II.,x~ (t,x i ) I I < c 4 llx.111

a ~
~
~
assume at last that the eigenvalues of the matrix

elements are defined by
A whose
4.
Stabilitv of composite systems
= c 24;
a ..
~J
I ID.BkHkl I
1 < k < m
have strictly negative real

then the critical point
329
i ~ j,
/ 2c 3i c IJ"
parts~
x = 0
of System (4.1) is uni-
formly, globally, asymptotically stable.

Proof.
Considering the comparison equation
u = Au
(4.2)
the only thing one has to verify, owing to Theorem 3.5 and
Exercise 3.8 is that
<
But one computes easily that
v.
aVo
~
::;--t
0
< -c 3 .
~
(lv.~
avo~
+-~-(t,x.)f.(t,x.)
oX i
1/ x.~ 1/ 2
c 3i
+ c4 . (
~ 1
< - -llx.11
2
~
oXi
L<
< j
+-~-(t,x.)
1 < j
I I D. B" H. II
~
~J
< m
"I
I I x . I I ) I I X.
IID.B.
~
~k
112
1 < J <
j
< m
a"V .
~J
~J
J J
/I
<
1 < j
c 24'
2
+.,...2:.-(
'i'
IID.B .. H.llllx.ll)
,,:c3i 1 ~ j < m
~ ~J J
J
i
<
D.B"H.x.
"I
I/x.1I 2
J
Q.E.D.
IX.
330
4.4.
Exercise.
In Theorem 4.3, one may also define
4.5.
A simple example of a composite system is a closed
loop, as shown on Fig. 9.1.
The input of every subsystem is
the output of the preceding one.

the matrix
A by
In that case, p
built up from the submatrices
Bij
= qi-l
and
is
Y2
1 - - - .-----
Ym
~ .. -------.
Sm
Fig. 9.1.
0
0
5j
Uj
Closed loop system.
Ip
Yl
r'
is a Pi x Pi unit matrix. To investigate the

Pi
stability. of such a system, we shall need the following lemma.
where
4.6.
Lenuna.
The solution
of the differential
4.
331
Stability of composite systems
system
- a 1 u 1 + blum'
i
- aiu i + b i u i-1'
with
a. > 0
~
and
(4.3)
2, . , m,
is globally asymptotically stable
b. > 0
~
if
ai
b.
(4.4)
II - > 1.
~
Proof.
The characteristic equation for (4.3) is

II
1 

-
II
1 < i < m
Re A < O.
Hence
m functions
V.
defined as in
and
and
ai'
Q.E.D.
Consider System (4.1) with
tive constants
IA + ail
4m
posi-
satisfying assump-
tions (i), (ii) and (iii) of Theorem 4.3.
If the gain esti-
mates
n~
= cc 4i
3i
[ c2i
c 1i
]1/2
110.11
~
IIHII
~
of the subsystems are such that

IT
1 < i < m
n.2
< 1,
then the zero solution of the composite system (4.1) is uniform1y globally asymptotically stable.
332
IX.
Proof.
The comparison equation (4.2) is of the form (4.3),
with
a.
c3 . / 2c2 . ,
~
But owing to Lemma 4.6, u = 0
1, ... , m,
is globally asymptotically
stable for (4.2) since

a.
b,'"~
TI
1 
TI
- r >1,
2
<
<
m
1
i
m c 4i c 2 i 1 IDiHi_ll 12
ni
has to be interpreted as another symbol for
H .
m
Q.E.D.
5.
An Example from Economics
We show in this section how the comparison method can

be used to prove that, under some suitable conditions, a market tends to some given evolution independent of initial conditions.
In the Walrasian approach to price evolution, the
price of any commodity, be it services or goods, is supposed

to increase when demand exceeds supply and to decrease otherwise.
On the other hand, the supply is an increasing func-
tion of the price, whereas the opposite occurs for the demand (see e.g. J. R. Hicks [1939) or P. A. Samuelson [1947).
The model discussed below comes, after simplification, from
D. D. Siljak [1973).
Suppose we divide the market into
modi ties, the
ith group being composed of
groups of comk.
items.
The
5.
An example from economics
subscripts
333
j, running from
and
the groups, while
running from
the commodities in one group.
to
to
n, will designate
k.
l.
will label
For instance
will be the
price of the
Ith item of the ith group, p.

a column vector
l.
formed by the prices in ith group, and p a column vector
formed from the
and
Pi's.
On the other hand, the capitals
with proper subscripts will designate demand and
supply respectively, whereas
D - S
is called the excess
demand.
The equations for the prices in the Walrasian approach
are
Pil = hil [t, (DU (t,p) - Su (t,p)) 1
where
hil(t,O)
and
Clh il
- - (t,x) > 0,
Clx
ClD'l._
l < 0,
__
ClPil -
All the functions introduced up to here are defined on some

appropriate domain (for instance most of them are positive)
which we shall not need to specify any further.
Anyhow, the
equations of the problem may be written under the general form

with
Pil (t,p) = gil (t,p)
(5.1)
It is natural to ask under what conditions all solutions of (5.1) approach some particular solution
ing
=P
Writ-
- PO' we get equations of the form

P
with
PO'
f(t,O)
g(t,p) - g(t,PO(t))
0,
Clf il
ClP il
(t,P)
= f(t,P)
Clg il
~Pl.'l
(5.2)
(t,po + P) < O.
These conditions are consistent with the following further
IX.
334
hypotheses.
that, for
= 1,
i, j
~ ail (t,Pi)P il
(v)
~ Pilb il (t,P)
IIp,ll =
~
ai
and
Sij
such
... , n,
= ail (t,P i )
fil (t,P)
(iv)
where
Assume there exist constants

every
(iii)
+ bil (t,P) ;
T
Pia i <
T
Pib i <
T
a ,P ,p, ;
~
s .. llp,11
~J
1 < j < n
IIpjll
~l'
l~'
Equation (iii) can be viewed as a way to consider separately the evolution of prices in one group and the interactions between groups, including readjustments in one group
due to variations in the others; (iv) refers to the adjustments of prices in one group and (v) is a bound on the reactions between groups.
Notice that (v) is implied by the fol-
lowing simple condition
Indeed
L P'lb'l
~
~
Ip'll !b'l!
~
<
I IPil I Ib'll.
~
Condition (iv) is sufficient for exponential stability of the

origin for the "decoupled" system
as is shown by using the Liapunov function
Vi
This suggests using the vector Liapunov function
I !P i ! I.
5.
An
example from economics
335
, ... ,
(P T
P )
1.1.
1/2 ,
(pTp )1/2)
nn
and the comparison equation

u. = - ~1..u1.' + L S.. u. = F. (u)
1.
j 1.J J
1.
(5.3)
to study the stability properties of the origin for Equation

(5.2) under assumptions (i) to (v).
But since (5.3) is
linear with constant coefficients, the origin will be globally

asymptotically (exponentially) stable provided all eigenvalues
of the matrix
C, whose elements
Cij
are given by
(i
"I j),
have negative real parts; (observe that by hypothesis,

C .. > 0 for i "I j). L. Metzler [1945] has
1.J proved that this is the case if and only if the so-called
and
Hicks conditions, familiar to economists, are satisfied:

principal determinants of
C must alternate in sign:
> O, ,(-l)j
(vi)
the
> 0,
1 .::. j .::. n.
An
extensive list of equivalent criteria appears in the re-
view article by M. Fiedler and V. ptak [1960].

Let us now show how (i) to (vi) imply global asymptotic
stability for the origin in Equation (5.2) provided all solutions of this equation can be continued to infinity.
putting
Vi
lip1. II
h7P.,
1. 1.
we compute, i f
V.1. "I 0,
Indeed,
336
IX.
-1 T
V. P [a.(t,P.) + b.(t,P)]
-1 T
V. P.P.
~
Ij
Vi (t)
= 0,
SijV j
then for every
h > 0
Vi (t + h) - Vi (tl < h
E [~~i + h]
t-
V. (Tl
~
E [t,t + h]
Vi (Tl
t-
-1
s~J
.. IIp.II
]
V.
lip II
i
(T)
sup
< h
Since the
aiv i +
< -
If
[-a.V.
~
(T)
s~],vJ' (T)].
j.
V's are continuous functions, dividing by
taking the limit
0+, one gets
D+V. (t) < - a.V. (t) +

~
hand
Ij s~J
.. v.J (tl.
As is readily verified, all the hypotheses of Theorem 3.5 are

satisfied, and this proves the announced property.
6.
6.1.
A General Comparison Principle
The comparison method, which was applied in Section 3
to the concepts of stability and attractivity, can of course

be also used successfully in connection with most concepts
introduced in Chapter VI.
It shows how a qualitative pro-
perty relating to some supposedly simple differential equation entails a similar property for another, usually more complex, equation.
In this chapter, we analyse from a theoreti-
cal point of view, this kind of transfer of a property from

one equation to another:
of the property itself, we shall
6.
A general comparison principle
337
retain just enough for the transfer to take place.
In other
words, a comparison principle will be derived, which will

apply to a whole class of concepts.
We consider again Equations (3.1) and (3.2) with the
general hypotheses 3.1, and, in order to make things easier,
we assume that all solutions under consideration are unique
and continuable to
+~.
using the symbolism of Chapter VI and given any quali-
6.2.
tative concept
c:
W,y * (A,B)
relating to Equation (3.1), we define a similar concept
for equation (3.2) by substituting:

(1)
(2)
E.9I' to
0
K(tO'U O)
to
Xo E .91' ~
J(to'xo)~ notice that, owing to our con-
tinuability hypothesis, both intervals equal

(3)
proposition
o _
= [u(t~to'uo) E 91]
to
[to'~[
B.
The other quantities appearing in
W,y
changed.
wO
thus obtained is ad-
We assume that the word
missible.
The new concept
son concept with respect to

V(t,x)
if the sets .91'0
or
cO
(A,B)
remain un-
will be said to be a compari-
C and some auxiliary function
~O
are such that:
(i)
if
is universal and defined after
(H)
if
is universal and defined before
338
IX.
(iii)
if
(iv)
if
is existential and defined after
Xo
is existential and defined before
(Vu o E JafO) (i!x O E ~) (Vt o E I)

(v)
V(to'X O)':: u O;
~O) (Vx E n: V(t,x) .::. u)
(Vt E I) (Vu E
~O
Notice that these requirements don't define

completely.
to
~.
x E
~O
nor
This definition of a comparison concept will per-
haps appear puzzling at first sight.

clear in the sequel.
Its meaning becomes
Anyhow, it is fairly easy to build such
a concept, as is shown by the following two lemmas.

6.3.
I
Lemma.
n .... ~m
(t,x) E I
a concept
Suppose there exists a continuous function
a E 5If:
such that, for some function
n:
C
max Vi(t,x) ~ a(1 Ixl I), V(t,O)

i
with
= B(O,o) n n
Xo
= O.
universal and defined after
and
B(O,e:)
V:
and every
Consider
to'
n n.
Then a comparison
sup
V~
concept is defined by the sets
~O
{u: u > 0,
U.
<
~-xEJaf
(to'x)},
~o = {u: max u. < a(e:)}.

i
Proof.
As
choose
Uo
x E
Xo
is universal and defined after
= V(to'x O)'
such that
Further, for any
V(t,x) .::. u,
to' we can
tEl, u E ~
and
6.
339
Q.E.D.
x E B(O,E)
Hence
fl.
Similarly, if
Xo
is defined before
to' one can prove
the following lemma.

Lemma.
6.4.
Suppose there exists a function

a E.5te
and that for some functions

(t,x) E I
V: I x fl
b E.5te
and
~m
and every
aCllxll) ~ m~x Vi(t,x) ~ b(llxll).
x fl:
Consider a
concept
.s= B(O,O)
with
universal and defined before

and
fl
Sj = B(O,e:)
Then a comparison
fl.
concept is defined by the sets

~O
{u: u > 0, max u. < b (0) },

~
i
SjO
{u: max u. < aCE)}.

~
i
~O
In most cases, the definitions of the sets

SjO
and
can be simplified to yield a new concept, which we shall
call auxiliary concept and which implies the comparison concept.

6.5.
Lemma.
Let
cO
be a comparison concept as defined by
Lemma 6.3 or 6.4 and suppose the corresponding assumptions

are satisfied.
(i)
if
Assume further that

0
is defined before
b ' E.5te
tion
to' there exists a func-
such that for any
(t,x) E I
x fl
max V. (t,x) #0
/,;10
~
and
Proof.
Let
0*,
follows, Wl ' wO2
' 1 'les t h e comparlS0n

'
, lmp
concept
E
be the values of
are parts of
and
W3
a way that the appearances of

If
0, E
in
c*.
In what
a word chosen in such
are left explicit.
is universal, one has the following implica-
tions:
~ u ~ o*e})w~,yO * (AO,SO)
(:io * > O:.s:t' {u: O)W~(VUo E {u:

~
Wl (Vo > 0)
~W~(Vo
O)W~(Vuo
>
,,>#O)W~,yO * (AO,SO) ::: CO.
Consider next an existential
Given
{u:
<
0 * > 0, there exist
0 >
such that
u < o*e}, in which case
is satisfied.
One proves in the same way that
E
implies
and the concept
..J 0
J
C ~
.-J*
6.
341
Indeed, given anyone of the quantities
E, E*, one can always
choose the other one in such a way that
E* < ace).
At last, combining the above arguments proves that

Q.E.D.
6.6.
General Comparison Theorem.
function
function
V: I x
Assume there exists a
as described in 3.1 and 6.1, and a continuous
n+
~m
which is locally lipschitzian in
and such that, for every
F(t,V(t,x.
EI
(t,x)
Then a qualitative concept
n:
C
D+V(t,X) <
will be satis-
fied whenever the corresponding comparison concept
cO
is
satisfied.
Proof.
Let
ation and
= W,
* (A,B)
* (AO,B O)
cO = WO, yO
parison concept.
be the concept under considerbe the corresponding com-
The problem is, knowing that
cO
is true,
to prove that the existential variables can be chosen in

such that
* (A,B)
is true.
following recurrence.
Wand
is
The first
variables being fixed in
I f i t is universal, we fix i t in
W, and, except
WOo, i f this variable

x O' we choose the same value in
0 such that V(to'x ) ~ u
O
x o ' let us choose u o E Jaf
O
Notice that, if
to
such that for every

2)
is not fixed yet, we can choose

to
E I: V(to'x O)
u O
wO
such that
cO
is satisfied.
we adopt the same value for the variable in

0'
such that vet ,x O) < u
x o ' we choose
Except for
For
uo
If the next variable is existential, consider the
corresponding variable in
W.
Let us choose them by the
wO, let us look at the next one.
1)
for
342
IX.
Let us now show that

and
t
yO
in
are equivalent since

AO or
(A,S)
=K =
is true.
[to'oo[.
First
The variable
A is chosen according to the rules 1) or 2)
used for the variables in
W.
From Lemma 2.6 we deduce that
and from the definition of the comparison concept (cf. Conditions (v

~
7.
x(t) E
~.
Q.E.D.
The comparison method has already been used, in substantially the same form as here, by R. Conti [1956] in a
theorem on the continuability of solutions using a scalar
comparison equation.
It was applied by C. Corduneanu [1960]
to the study of stability and asymptotic stability.
V. M.
Matrosov [1962]2 and R. Bellman [1962] both extended the results of Corduneanu to the case of vectorial comparison equations.
These extensions are based on the work of T. Wazewski
[1950] on differential inequalities, an account of which appears in several books; for instance W. Walter [1964], J.
Szarski [1967] or V. Lakshmikantham and S. Leela [1969].
The
last one contains many original contributions of the authors

in several domains, as for instance conditional stability.
A fairly complete survey of the results obtainable by the
comparison method, not only in stability theory, but in the
study of existence and uniqueness of solutions, continuability,
convergence of approximations, etc. is given in a series of
papers by V. M. Matrosov [1968-1969].
7.
343
Most results of Section 3 come from V. M. Matrosov

[1962]2'
Theorem 3.5 generalizes the corresponding result
of Matrosov in the sense that negative
V.
J.
functions can be
used here, whereas in the original paper, all these functions

had to be positive and their sum had to be positive definite.
Exercise 3.7 comes from N. P. Bhatia and V. Lakshmikantham
[1965].
Theorem 3.9 is a variant of a result of V. M.
Matrosov, where one used a comparison concept independent of

the equation in
x.
Theorem 3.10 is an example of Matrosov's
results on instability.
Section 4 is based on a paper by F. N. Bailey [1965].
An
application to a control system describing the longitudi-
nal motion of a plane appears in A. A. Piontkovskii and L. D.

Rutkovskaya [1967].
Economic problems akin to the one treated
in Section 5 appear in D. D. Siljak [1973].

ted systems,
On interconnec-
see also A. N. Michel [1974].
The comparison principle presented in Section 6 comes

from P. Habets and K. Peiffer [1973].
Some generalizations
to comparison equations depending not only of the auxiliary

variable
u, but also on
x, are due to V. M. Matrosov [1963]
and P. Habets and K. Peiffer [1975].
On the other hand, with
a view to generalize the monotonicity hypothesis, V.

Lakshmikantham, A. R. Mitchell and R. W. Mitchell [1975] have
established some comparison results for differential equations on cones.
Amongst the questions directly related to the subject
matter of this chapter, but which have not been treated here
for lack of space, let us cite the combined use of the com-
344
IX.
parison method and the theory of sectors by V. M. Matrosov

[1965], partial stability in C. Corduneanu [1964], K. Peiffer
and N. Rouche [1969], and C. Risito [1972], and at last several results of Dang Chau Phien [1968] on stability and
asymptotic stability, where the hypothesis on
max V. (t,x)
i
is substantially weakened with respect to the theorems of

the present chapter.
APPENDIX I
DINI DERIVATIVES AND MONOTONIC FUNCTIONS
1.
1.1.
Let
a, b, a
function
f: la,b[
The Dini Derivatives

b, be two real numbers and consider a
<
+ ~,
f(t)
and a point
to E la,b[.
We assume that the reader is familiar with the notions of

lim sup f (t)
t .... to
The same concepts of
mean
lim inf f (t) .

t + to
and
lim sup
and
inf, but for
to +
simply that one considers, in the limiting processes,
only the values of
>
to'
A similar meaning is attached to
Remember that, without regularity assumptions on

f, any lim sup or inf exists if we accept the possible values
+
00
or
00.
after by !Ji.
The extended real line will be designated hereThus~
=!Ji' U
{_oo} U {+oo}.
The four Dini deriv~tives of
at
to
fined by the following equations:

+
D f(t O)
D+f(t O)
lim sup
t + t o+
lim inf
t + t o+
lim sup
+ t
o-
f (t) - f(t O)
t - to
f (t) - f(t O)
t - to
f (t) - f (to)
t _ t
0
are now de-
346
APPENDIX I:
f(t) - f(t O)
D_f (to) = lim inf _ _ _ _--101t + t ot - to
They are called respectively the upper right, lower right,
upper left and lower left derivatives of
the function
D+f(t)
upper right derivative of

similarly for
1.2.
D+, D-
Remarks.
and
on
at
into ~
]a,b[
on the interval
to'
Further,
is called the
]a,b[, and
D.
a) It is clear that, in the absence of regu-
larity assumptions on
f, any Dini derivative may equal
or
b)
However, if there is a Lipschitz condition for
00
00.
some neighborhood of
on
to' then all four derivatives are fin-
ite.
c)
The four Dini derivatives of
]a,b[
are equal if and only if
f
f
at some point
toE
has a derivative at
to'
This derivative is then of course equal to the common value

of the Dini derivatives.
d)
The well known properties of lim sup and lim inf yield
the elementary rules of calculus applicable to the Dini derivatives.
For example, if
tions defined on
fl
and
f2
are two real func-
] a,b [, one gets for any
t E] a,b [
that
and
as long as the additions are possible
[(+
(0)
+ (-
(0)
is an
example of an.addition which is not possible].

e)
and
Another important property is that if

g
is !ifl , then
is continuous
Appendix I
347
t: get) > 0, one has
i f for some
(1)
D+(fg) (t)
f(t)g' (t)
+ g(t)D+f(t)
(2)
i f for some
t: get) < 0, one has
D+ (fg) (t)
f (t)g' (t)
+ g(t)D+f(t),
where
g' (t)
is the ordinary derivative.
The proof is as follows:

D+ (fg) (t) = lim sup
(fg) (t + h) -
h ~ 0+
lim sup[f(t + h)
g (t + h) -
h ~ 0+
g (t)
get + h) - get)
lim f (t + h)
h ~ 0+
(fg) (t)
Hence the expected result.
f(t + h) - f(t)
+ get) - - - - = - h - - - - ]
f(t + h) - f(t)
+ lim sup g (t)--....,h,...----h ~ 0+
On the properties of the lim sup
and lim inf which enable one to write the above equalities,
as well as on other rules of calculus for Dini derivatives,
we refer to E. J. McShane [1944].
2.
2.1.
Theorem.
Continuous Monotonic Functions

Suppose
is increasing on
every
la,b[
is continuous on
if and only if
D+f(t) ~ 0
Then
for
E la,b[.
Remember that, in this book, f

on
la,b[.
la,!;>[
if, for any
is called increasing
t l , t2 E ]a,b[, tl < t 2 , one has
f(t l ) .::. f(t 2 )

Proof.
The condition is obviously necessary_
that it is sufficient.
Let us prove
348
a)
APPENDIX I:
D+f(t) > 0
Assume first that
two points
[a,B]
points.
B,
a, BE ]a,b[, a <
there exist a
t t
V with
Of course,
]a,b[.
with
Let
If there exist
f(a) > feB), then
f(a) > V > feB)

f{t) > V.
such that
on
and some points

be the sup of these
[a,B], and,
is an interior point of
due t9 the continuity of
f:
f{~)
V.
Therefore, for every
tEH,B[ :
f{t) - f{~) < 0
t
D+f{~) ~ 0, which is absurd.
and
b)
Assume now, as in the statement of the theorem, that
D+f{t) ~ 0
on
]a,b[.
For any
E > 0, one gets
D+ [f (t) + Et]
Hence
f{t) + Et
true for any
2.2.
is increasing on
E, f{t)
Remarks.
a)
And since this is
is also increasing on
]a,b[.
Q.E.D.
This theorem remains true if one replaces
D+f(t) ~ 0
the inequality
]a,b[.
by
D+f{t) ~ 0, because the lat-
ter implies the former.

b)
One proves similarly that
placed by
D+f(t) > 0
D-f(t) ~ 0: it suffices to substitute the inf of
the points
where
f{t) < V
to the sup of the points
f{t) > V.
where
c)
In the new theorem thus obtained, D
by
D.
2.3.
can also be re-
may be replaced
As a consequence, we get the following statement.
Theorem.
Suppose
is increasing on
derivatives of
]a,b[
f
is
is continuous on
]a,b[.
Then
if and only if any of the four Dini

>
on
]a,b[.
Appendix I
2.4.
349
Corollary.
function
If any Dini derivative of the continuous
is
> 0
on
]a,b[, the same is true of the
other three.
2.5.
Remark.
Analogous monotonicity properties can be es-
tablished using less than the continuity of
(cf. E. J.
McShane [1944]).
Functions with a bounded Dini derivative.
2.6.
The follow-
ing theorem is used to estimate the average rate of decrease

of a function possessing a Dini derivative bounded from below.
It is a straightforward consequence of Theorem 2.3.

D* f
Hereafter, the symbol

derivatives of
represents any of the four Dini
f.
Theorem.
Let
such that for any
f: [a,b] ...
t E l a,b [
be a continuous function
and some
A > 0:
D*f(t) > -A.
(2.1)
Then
f(a) - f(b) < A.
b - a
Proof.
One deduces from (2.1) that
therefore, using Theorem 2.3, that

on
la,b[.
2.7.
Therefore
f(b) + Ab
D*(f(t) + At) ~ 0, and

f(t) + At
is increasing
f(a) + Aa.
Q.E.D.
Dini derivative of the maximum of two functions.
cerning three functions
f, g, h
such that
Con-
h(t) = max (f(t),
g(t)), the following theorem gives an estimation of a Dini

derivative of
of
and
h(t)
in terms of the corresponding derivatives
g.
Theorem.
Let
f, g, and
be three continuous func-
350
APPENDIX I:
tions on
If
[a,b)
D+f (t) < 0
D+h(t) < 0
DIN I DERIVATIVES AND MONOTONIC FUNCTIONS
!it, such that
into
D+get)
and
for
h (t) = max (f (t) ,g (t.

t E ) a, b [, then
t E )a,b[.
for
Proof.
Otherwise, one would have, by Theorem 2.1, for two
points
a', b'
a < a' < b'
with
b, that
max (f(b'),
g(b'
> max (f(a'),g(a', and therefore either
f(a')
or
feb') >
g(b') > g(a'). But, using Theorem 2.1 again, this

+
contradicts either D f < 0 or D+g < O.
Q.E.D.
3.
3.1.
The Derivative of a Monotonic Function
The theorem stated (without proof) in this section is a
key theorem for Liapunov's direct method.

Lebesgue integral of a function
which will be written
t
a
It mentions the
on an interval
[a,b),
f(T)dT
On this concept, we refer to E. J. McShane [1944] or to A. N.

Kolmogorov and S. V. Fomin (1961).
Only one of its elemen-
tary properties will be recalled below, in order to clear the

statement of the theorem.
3.2.
A subset
of the real line
!JR is said to have
measure zero if there exists, for every

countable collection
Ul i
~E
and
E~Ii
1 1 , 1 2 ,
< ,
where
> 0, a finite or
of open intervals such that
~Ii
is the length of
Ii.
When a property is verified at each point of some interval

[a,b)
E!JR, except at the points of a set of measure zero,
one says that the property is true almost everywhere on

[a,b]
or for almost all
E [a,b).
Appendix I
351
If a function
over
to
almost everywhere on
ble on
is Lebesgue integrable
g: [a,b] ... 91
[a,b], then any function

f
f.
f: [a,b] ... 91
[a,b]
which is equal
is also Lebesgue integra-
[a,b], and the integral of
equals the integral of
Therefore, it makes sense to speak of the integral over
[a,b]
on
of a function which is defined only almost everywhere
[a,b]:
it can be extended to the whole of
[a,b]
by
choosing arbitrary values at the points where it was originally undefined.

3.3.
f
Theorem.
If
f: [a,b] ... 91
has a finite derivative
is an increaSing function,
f' (t)
this derivative is Lebesgue integrable and one has,
[a,b]~
for any
t E [a,b],
f(t) = It f' (T)dT + h(t)
a
where
is an increasing function
and
h' (t)
vanishes
[a,b].
For a proof, see E. J. McShane [1944] or H. L. Royden

[1963]
3. 4.
Corollary.
.In the hypothese s

feb) - f(a)
3.5.
Remarks.
a)
f Theorem 3.3,
(3.1)
f' (T)dT.
This inequality becomes an equality if
one adds the hypothesis that

[a,b].
fa
is absolutely continuous on
On this point, cf. the reference books already men-
tioned.
It does exist an example of a function
into
which is increasing, uniformly
~,
on
[a,b]
continuous, whose
derivative vanishes almost everywhere, and such that
352
feb)
APPENDIX I:
>
f(a).
Of course, for this function, which is not ab-
solutely continuous
feb) - f(a) >
f'(T)dT
O.
Cf. K. Kuratowski [1961], p. 187.

b)
Since the derivative of
f, when it exists, equals
all four Dini derivatives, the inequality (3.1) can also be

written under the form
fb D+f(T)dT
a
feb) - f(a),
or similarly while replacing
D+
ber that it is valid when
is increasing.
4.
by
D+, D
or
D_
Remem-
Dini Derivative of a Function along the Solutions

of a Differential Equation
4.1.
For some
T,
_00
< T <
00
n C~,
and some open subset
consider a continuous function

f: ]T,OO[
n -+-~n, (t,x) -+- f(t,x)
and the associated differential equation

let
V: ]T,oo[
x Q-+-!if
Further,
be a continuous function, satisfying
a local Lipschitz condition for

to
x = f(t,x).
x, uniformly with respect
t.
4.2.
One has often to verify that a function like
V(t,x)
is, so t9 say, decreasing along the solutions of the differential equation.

J
This means that for any solution
x: J -+- ~n,
x = f(t,x),
the function
an open interval, of the equation
V: J -+- !if n , t -+- vet)
= V(t,x(t
is decreasing.
The follow-
Appendix I
353
ing theorem is crucial,
for it enables one to check this pro-
perty without any knowledge of the solutions.

4.3.
Theorem (T. Yoshizawa [1966]).
eses, let
ting
x: J
~n
In these general hypoth-
be any solution and let
*E
J.
Put-
x(t*) = x*, one gets
+_ *
V(t* + h,x* + hf(t*,x*
D Vet ) = lim sup
h
h + 0+
Proof.
One has, for
h.> 0
V(t* + h,x(t* + h
- V(t*,x*)
(4.1)
small,
- V(t*,x(t*
V[t * + h,x * + hf(t * ,x * ) + he(t * ,x * ,h)] - Vet * ,x * )
~ V(t* + h,x* + hf(t*,x*

where
with
some neighborhood of
D+vet * )
hand
x*
+ khl le(t*,x*,h) I I - V(t*,x*),
is a Lipschitz constant on
Therefore
vet * + h,x(t * + h
lim sup
h
h + 0+
< lim sup

h + 0+
V(t* + h,x* + hf(t*,x*
- V(t*,x*)
One obtains similarly for

V(t* + h,x(t* + h
- Vet * ,x(t *
h > 0
- V(t*,x(t*
V (t * + h, x * + hf (t *, x*
small, that
>
- kh I I e (t *, x* , h) I I-v (t *, x* ),
whence
*
* *
Vet * + h,x * + hf(t * ,x
V(t,x)
D+V(t*) > lim sup -----------------------------------h + 0+
h
Q.E.D.
4.4.
Remarks.
a)
We shall admit the symbol
D+vet * ,x * )
to
represent the second member of (4.1), and this quantity will

be called occasionally the upper right Dini derivative of
354
APPENDIX I:
V(t,x)
(along the solutions of the differential equation).
b)
It is a simple consequence of Theorems 2.1 and 4.3

D+V(t,x)
that i f
on
]T,OO[ X
n,
then
v(t,x)
is increas-
ing along the solutions of the differential equation.

analogous statement for
c)
decreasing is obvious.
There is a theorem similar to 4.3 for any other
Dini derivative
d)
V(t,x)
The
D+, D-
and
D.
It is noticeable that no uniqueness property has
been assumed for the solutions of the differential equation.

e)
x
n,
gn
If,
one has
for some
e >
and every
(t ,x) E]T,OO[
D+V(t,X) > -e, then for any solution
and any points
a, b E J, a < b,
Veal
b
V(b) < e.
a
This is a consequence of Theorems 2.6 and 4.3.
x: J
->-
APPENDIX II
THE EQUATIONS OF MECHANICAL SYSTEMS
Assuming some knowledge of analytical mechanics, we
gather here a few precise definitions concerning Lagrangian
and Hamiltonian systems and recall their most fundamental
properties.
1.
To a mechanical holonomic system with
degrees
of freedom with generalized (or Lagrangian) coordinates

q E ~n
q E ~n, there will cor-
and generalized velocities
respond a kinetic energy of the form

T: I x n x YR n ... :if, (t,q,q) ... T{t,q,q),
where
I
]T,OO[
is some domain (connected open set) of

for some
With respect to
E~.
polynomial of the second degree.
~n
q, T
and
is a
It will often be written
thus:
T{t,q,q)
where
d
is an
21 q.TA{t,q)q
n x n
T.
+ b{t,q) q + d{t,q)
is a scalar, all defined on
will be supposed to be ~l
2.
is an
matrix, b
I
n.
n x 1
matrix and
Further, A, band
functions.
The potential energy will be designated by
IT.
In this book, we are prevented from choosing the usual letter
356
APPENDIX II:
EQUATIONS OF MECHANICAL SYSTEMS
to represent the potential function, because in the domain
of Liapunov's direct method, V

iliary functions.
IT
3.
IT
is the ritual symbol for aux-
will be supposed to be a
x rl ....
5fl
function
<JP, (t,q) .... IT (t,q).
We shall also consider Lagrangian forces which
do not derive from a potential function.
They will form on
n-vector depending on velocities, coordinates and time:

I
We assume
4.
x rl x
gj'n .... !Jf,
(t,q,q) .... Q(t,q,q).
to be continuous.
The Lagrangian equations of motion are, in the
usual notations,
(4.1 )
where
=T
- IT.
The question arises immediately whether
Equation (4.1) can be solved with respect to

be brought to normal form
x=
around every point
where
5.
A(t,q)
This can be done

is regular.
A situation frequently encountered is when the
matrices
dent of
(t,q)
f(t,x).
q, and therefore
Band
t.
vanish identically and
independent constraints.
det A(q) i
is indepen-
In this case we write the latter A (q).
compact form is obtained for
respect to
O.
and therefore
tive definite.
for instance in case of time
Then Equation (4.1) can be solved with
in the neighborhood of some point

But
A(q)
det A(q)
Such a
(t,q,q)
if
is everywhere positive semi-definite
if and only if
A(q)
is posi-
The following useful property of the kinetic
Appendix II
357
energy is easy to prove.

Proposition.
N
of
qo' N C Q
For
qo E Q, there exists a neighborhood
and a function
a E
(q,q) EN x ~n, qTA(q)q ~ a(1 Iql I), if and only if
A(qO)
(Hint:
choose
compact, remember
is continuous
and can be diagona1ized by an orthogonal change of coordinates)

6.
The dissipative forces are an important family of
generalized forces:
Q is called dissipative if, for every

Q T.q
Of course
the power supplied by the forces
Q.
The case of complete
dissipation is when there exists a function

that, for every
is
a E Jt
such
(t,q,q) E I x Q x ~n: QT(t,q,q)q ~ -a(1 Iql I).
Dissipative forces which are linear with respect to

called viscous frict"ion forces.
are
As observed by A. I. Lur I e
[1968J, they are not the only ones to be derivable from a

Rayleigh dissipation function (on this notion, see for instance H. Goldstein [1950J).
energic forces (in the term-
inology of G. D. Birkhoff [1927J) are particular dissipative

T
forces, namely those for which
Q (t,q,q)q = 0
identically.
Gyroscopic forces are non energic forces which are linear

with respect to
7.
Q(t,q,O) =
Proof.
and some
q.
Proposition.
for every
If
Q(t,q,q)
is dissipative,
(t,q) E I x Q.
Suppose on the contrary that for some
.,.
.,.
i, 1 
(the case
By continuity, one would get
q= O. Let us,
*
(O,O, ,q., 0, ,0),
1
borhood of
.*
< 0
would be treated alike).
*.
Qi (t ,q ,q) > 0
in some neigh-
in this neighborhood, choose

with
.*
q. > O.
1
One would obtain
*.* > 0, which is excluded.

QT (t * ,q * ,~)q
ment:
Q.E.D.
8.
This proposition introduces the following state-
if
is quadratic in
and
is dissipative,
the Lagrange equations (1) have an equilibrium at every point

where
an/aq = O.
This covers obviously the case where
vanishes identically.
When
9.
write
is independent of
T(q,q)), i.e. when
and when further
= 0,
A, band
t
d
(in this case we

are time-independent,
the Lagrange equations admit of a
first integral called the Painleve integral.

E(q,q)
If
Q" 0, E
easily that
tive of
It reads
.T aL
1.T.
q -- - L = -2 q A(q)q - d(q) + IT(q).
aq
is no more a first integral, and one computes
T.
E(t,q,q) = Q(t,q,q) q, where E is the deriva-
along the solutions of (4.1).
quadratic with respect to

identically, then
q, i.e. if
If further
Band
E(q,q) = T(q,q) + IT(q)
is
C vanish
is the total en-
ergy and, as a first integral, is called the energy integral.

As already mentioned, this situation arises mainly when the
constraints are time independent.
10.
Other first integrals are the so-called integrals
of conjugate momenta.
ordinate
qk
identically,
When
does not depend on some co-
and the corresponding force
Qk
vanishes
Appendix II
359
aL
aqk
(t,q,q)
is a first integral of the motion.

11.
Ignorable coordinates and equations of Routh.
Consider the
mechanical system described in Sections 1 to 4.
Suppose the number

some
of degrees of freedom is
m, 1 < m < n
aT
aqi
n-m+l, ... ,n.
Henceforth, we shall write

n - m
last
> 2, and for
(ql, . ,qn-m)
Lagrangian coordinates and

m.
for the first
(rl, ,rm)
for the
Observe therefore that, by doing this, we change
the assignment of the symbol
q!
The coordinates
are said to be ignorable if one has further that
(1)
write it
the potential does not depend on
the only non vanishing generalized forces corres-
pond to the first

pend on
rand
we shall
IT(t,q);
(2)
with
r:
n - m
r:
degrees of freedom and do not de-
we shall write them
(Ql, .. ,Qn-m)
= Q(t,q,q).
The kinetic energy may be written under the form
*.
**.
T(t,q,q,r) = T2 (t,q,q) + U(t,q,q,r) + T2 (t,q,r)
**.
+ Tl (t,q,q) + Tl (t,q,r) + TO(t,q)

where
T2
is quadratic in
is quadratic in
in
r.
r, Tl*
q,
is bilinear in
is linear in
and
**
(lI,t), T2
**
Tl
The regularity assumptions for all functions
is linear
T, IT
360
APPENDIX II:
and
remain those of Sections 1 to 3.
The Lagrange equa-
tions of motion read

d
dT
dt
dCt
dq
dT
dr
where
(11. 2)
this case, r
Assuming that for
r,
is positive definite with respect to
Equation (11.2) can be solved for

neighborhood of
(11.1)
= c
is a constant of integration.
-.
q, T2** (t,q,r)
some
(T - II) + Q
in some appropriate
q, yielding a function
= r(t,q,q,c).
In
can be eliminated from the equations of motion.
An elegant procedure to achieve this is by using Routh's function, obtained by substituting

T - cTr.
in terms of
t,q,q,c
in
One gets readily for this function
- II(t,q),
+ T1* + T01..
r=r(t,q,q,c)
R(t,q,q,C)
the detailed computation appearing in A. I. Lur'e [19681.

For our theoretical purposes, we only need to observe here
that, with respect to
q, R
is a polynomial of the second de-
gree, and in this polynomial, the terms of the second degree

do not depend on
c.
This enables us to write
R(t,q,q,C)
-II (t,q) ,
where
q.
and
R1
By the way
R2
R2
USing
(11.3)
are respectively quadratic and linear in

T*
2
R, one gets the equations of motion under the
convenient form, called Routh's equations,
Appendix II
361
d
aR
--dt aq
c=
aR
aq
= Q,
(11.4)
o.
In case of time-independent constraints, the formulas

above become simpler:
the argument
disappears everywhere
and further, all functions with subscript

identically.
Rl
and
12.
or
vanish
As is well known, there exist time-dependent
constraints such that

tions
RO
disappears everywhere, but the func-
do not vanish.
Stationary motions.
q = q, q = 0, c =
Consider an equilibrium
of Equations (11.4).
A generalized sta-
tionary motion (C. Risito [1972]) is a motion of the original

system corresponding to such an equilibrium, with
course given as a function of
of
by Equation (11.2).
equations of motion are autonomous,
If the
is a constant and the
corresponding motion is called stationary.
Another adjective
sometimes used instead of stationary is merostatic.

13.
For the sake of references, let us now describe a
type of system most frequently encountered in the applications.
~n
The kinetic energy

and is, for every
form with respect to

defined and ~l
q.
n.
on
T(q,q)
is defined and
~l
on
q, a positive definite quadratic

The potential function
n(q)
is
When one has to study an equilibrium
of such a system, it is convenient to locate i t at the point

q = 0, i.e. to assume
a way that
n (0) = o.
gin be a point of
n.
~(o) = 0, and to adjust

aq
in such
This of course requires that the or i-
APPENDIX II:
362
14. For the system just described, with

lT
2 q A(q)q,
the equation
= -dT = A(q)q
dq
can be uniquely solved for

as a function of
and
T(p,q)
where
B (q) = A
-1
(q).
q.
q
The kinetic energy expressed
reads
1 T
- p B(q)p
2
Of course
taneously positive definite.

H(p,q)
T(q,q)
A(q)
and
The Hamiltonian
B(q)
are simul-
function
T(p,q) + n(q)
is equal to the total energy expressed as a function of

and
q, and the Hamilton's equations of motion are

p
dH
dP
APPENDIX III
LIMIT SETS
1.
I =
Let
set of ~n.
for some
]T,~[
n an open
and
E~
We consider the differential equation

x = f{t,x)
f: I x n
where
= ]a,w[
x: J
equation.
of
that
~n
is some continuous fUnction.
y Eli
is called a positive limit point
i f there exists a sequence
W and
x (t n )
tive limit set of the solution

limit points.
It is designated by
{t
+
n
y
of time-values, such
n
as
-+
00.
A+ (x)
recall that the positive semi-orbit of

to E J
2.
is the set
Theorem.
The posi-
is the set of its positive

Ne9:ative limit
points and negative limit sets can be defined alike.
some
Let
be a non-continuable solution of this
A point
{tn } C J, t
(I)
Let us
corresponding to
y+{X,t o ) = {x{t): t E [to'W[}.
For every
to E J: Y+ (x,t o ) = y+ (x,t O)
UA+{x).
Proof.
Since
A+{x) C y+{X,t o ), it is clear that
y (x,t o ) :J y (x,t o ) U A (x).
Suppose now that some point
y EY+{X,t o ).
There exists a
APPENDIX III:
364
{til C J, ti ~ to
sequence
t.
w as
y E A+(x).
00, then
an infinite subsequence
t' < w.
by some
i
-+
The
such that
x(t i )
If
Otherwise, there exists
are such that
l.
y.
which is bounded from above
{t~}
t~
LIMIT SETS
x (t! )
as
But since they are infinite in number and bounded,
00.
{t'.'}
there exists a subsequence
til < W.
approaching some
=y
Hence, by continuity, x(ti) + x(t")
as
i + 00, and there-
y E Y+ (x,t O)'
fore
3.
Proof.
when
Theorem.
Let
i
Q.E.D.
A+(X)
is a closed set.
{Yi} C A+(x)
be some sequence such that
00, and let us prove that
there exists a sequence
cular sequences
{t ij }.
given.
Let us choose
< 1/2.
Choose further
-+
00
and
j.
i,
and
We shall now construct some parti-
and such that
t22 > t12

t33 > t13
i
d(x(tii),y i ) < l/i
d(x(t 33 ),
Of course
for every
be
d(x(t 22 )'Y2)
and such that
4, 5,
{t lj }
i-
tii .... w
On the
other hand, we impose no restriction whatsoever to the

for
For every
Suppose first of all that
Y3 ) < 1/3, and so on for

when
E A+(x).
such that
j + 00.
when
Yi
t ij
From the two propositions
(Ve: > 0) (:IN> 0) (Vi
N)
d(x(t ii ) ,Y i ) < e:/2,
(Ve: > 0) (:IN' > 0) (Vi .::,N')

we deduce easily that
x(t ii ) .... Y
as
d(Yi'Y) < e:/2,

i .... 00, and therefore
Q.E.D.
4.
Theorem.
If
empty, compact and connected.
is bounded, A+(x)
is non
Appendix III
365
y+ (x,t o)
Proof.
Since
+
Further, A (x)
t
compac t se.
.
1S
There would exist a sequence

such that
d(x(ti),A 2 ) ... 0
as
d(A I ,A 2 ) = 0
to < tl < t'I

0 and
ti ... W, d(x(ti),A I )
...
Therefore, for every large enough
i"'~.
i, there would exist a
such that
But the
cluster point, since
t 1 < t~ < t~
1
x(t i* )
and
would admit a
is bounded.
This point would
A+, but also it would be apart from
least the distance
A+
by at
0/4, which is a contradiction.
Theorem.
In
would be the union of two compact
< t2 < ti <
is connected.
A2 , such therefore that
and
Al
0 > O.
5.
"'+(x)
Le t us now prove t h at
disjoint sets
belong to
is not empty.
is compact, since it is a closed subset of a
the opposite case, A+(x)
for some
compac t , "'+(x)
If
A+(x)
Q.E.D.
is bounded, x(t) ... A+(x)
as
... W.
~.
If it were not the case, there would exist an
and a sequence of time-values

when.
i'" ~
and
d(x(t.) ,A+(x
a subsequence of the
point would belong to
larger than
6.
E,
to' t l ,
x(t i )
A+(x)
> E.
and be at a distance of
exists a positive limit point in
n.
for every
8.
Exercise.
to
that
Theorem.
If
But there would exist

This
A+(x)
Q.E.D.
which is impossible.
Exercise.
t i ...
approaching a limit point.
Prove that
7.
such that
E >
=~
as soon as there
is a periodic solution, one has
A+(x) = y+(x,t o ) = {x(t): t E J}.

If the differential equation is autono-
mous and if the solutions possess the uniqueness property, then
APPENDIX III:
366
A+(x)
Proof.
If
LIMIT SETS
y+(x,t O).
to E J
is periodic, the theorem is a direct conse-
quence of Exercise 7.
If it is not, let us first deduce from
Theorem 2 that
n
to E
y+ (x,t o ) = (
n
to E
y+ (x,t o ))
U A+ (x).
The theorem follows from the fact that the intersection in

the second member is empty.
9.
Q.E.D.
For autonomous equations, the most important pro-
perty of the limit sets is that they are semi-invariant.
To
define this notion, let
(2)
f(x}
be the differential equation, with
defined on some open subset
solutions is not required.
interval
and such that
for every
x(t}
E J.
ant if, for every
Uniqueness of the
First of all, a set

Xo
E F
Fen
is
and for all non-
of Equation (2), defined on some

x(t o}
Further, Fen
Xo
of ~n.
said to be invariant if, for every

continuable solutions
= xo'
one has
x(t) E F
is said to be semi-invari-
E F, there is one such non-continuable
solution with the same property.
If uniqueness of the solu-
tions is assumed, semi-invariance is of course equivalent to

invariance.
10.
The following regularity theorem will be used to
prove the semi-invariance of the limit sets.

Theorem.
such that
Let
{x Oi } en
as
be a sequence of points
i ... 00, for some
Let
x.:
~
Appendix III
lai,w i [
~n
367
be a non-continuable solution of the Cauchy
x = f(x),
problem
uable solution
= x Oi '
x(O)
x: ]a,w[
= Xo
1,2, }
such that for every
Xi (k) (t)
and an increasing subsequence
x(t)
of the Cauchy problem
~n
f (x) , x (0)
one gets, for
Then there exist a non-contin-
large, that
uniformly on
t l , t2
with
{Uk) : k
a < t
<
t2 < w,
ai(k) < t 1 < t 2 < W i (k)

[t l ,t 2 ] as k + 00
and
For the proof, cf. P. Hartman [1964].

11.
Theorem.
mous, A+(x)
(see Exerx * E A+(x) n fl. We know that w =

0
Let ft.1 } be such that ti + 00 and x(t i ) + Xo*
cise 6).
i
is semi-invariant.
Let
~.
as
nn
If the differential equation is autono-
00.
00
Put
x(t. )
l.
of the Cauchy problem
= XOi
Then
f (x) , x (0)
x(t + t i )
= XOi
is a solution
It follows
from Theorem 10 that there exists an appropriate subsequence

of the
x (t + t.)
Cauchy problem
approaching some soluti,on
x * (t)
of the
f(x), x(O) = xo'

* uniformly on every com-
pact subinterval of the interval

Therefore, x*(t) ~ A+(x)
J*
for every
of definition of
t E J*.
x*.
Q.E.D.
LIST OF EXAMPLES
Page
Attraction of a particle by a fixed center (VII.5) .
249
Betatron
stability of (IV.6)
145
instability of (V.7)
188
Chemical kinetics (11.1.11)

Chemical reactor (11.2.9)
58
70
A class of nonlinear electrical networks (VII.6)
254
30
Ecological problem of interacting populations (VII.7)
260
332
17
33
Damped pendulum (1.6.12)
An example from economics (IX.5)

Glider
stability of a (1.4.8)
asymptotic stability of a (1.6.18)
(V.6)
184
84
38
Nth order equation (instability for a)

Nuclear reactor (11.5.2)
Pendulum
with time varying friction (1.6.27)
(11.2.6)
30
damped pendulum (1.6.12)

Regular precessions of a satellite (IV.7.14)
Regular precessions of a symmetric top (IV.5.5)
161
141
Restricted problem of three bodies (111.6.9)

Steady rotations of a rigid body
stability of (1.4.7)
16
21
instability of (1.5.6)
Third order equation
global asymptotic stability for a (11.1.9)

instability for a (V.8)
56
191
List of Examples
Transistorized network (II.6.S)

Transistor oscillator (II.l.B)
369
, ..................... .
Vertical rotations of a top (IV.7.l3)
90
54
159
BIBLIOGRAPHY
Amaldi, U., see Levi-Civita, T.
Amundson, N. R., see Warden, R. B.
Antosiewicz, H. A. [1958}, A survey of Liapunov's second
method, Contributions to the theory of non linear oscillations, vol. IV, ed. by S. Lefschetz, Princeton Univ. Press,
141-166.
Appell, P. [1932}, Traite de mecanique rationnelle, tome IV,
fascicule I, Gauthier-Villars, Paris, 235.
Appell, P. [1953}, Traite de mecanique rationnelle, tome II,
2d edition, Gauthier-Villars, Paris, 339.
Aris, R., see Warden, R. B.
Arnold, V. I. [1961}, The stability of the equilibrium position of a Hamiltonian system of ordinary differential
equations in the general elliptic case, Dokl. Akad. Nauk.
SSSR, 137, 255-257, Soviet Math. Dokl., ~, 247-249.
Avdonin, L. N. [1971}, A theorem on the instability of equilibrium, Prikl. Mat. Meh., 35, 1089-1090, J. Appl. Math.
Mech., ~, 1036-1038.
-Avramescu, C. [1973}, Quelques remarques sur les concepts
qualitatifs attaches a une equation differentielle, Proceedings of the conference on differential equations and
their applications, Iasi, Romania, October, 24-27.
Bailey, F. N. [1965}, The application of Liapunov's second
method to interconnected systems, SIAM J. Control, sere A,
l, 443-462.
Barbashin, E. A. [1967}, Introduction to the theory of stability, Wolters-Noordhoff, Groningen, The Netherlands,
1970~ translation of the Russian edition, Moskow, 1967.
Barbashin, E. A., and Krasovski, N. N. [1952}, On the stability of motion in the large (Russian), Dokl. Akad. Nauk.
SSSR, ~, 453-456.
Beletskii, V. V. [1966}, Motion of an artificial satellite
about its center of mass, Israel program for scientific
translations, Jerusalem.
Bellman, R. [1953}, Stability theory of differential equations,
McGraw-Hill, New York.
Bellman, R. [1962}, Vector Liapunov Functions, SIAM J. Control,
sere A, l, 32-34.
Bellman, R. [1970}, Introduction to matrix analysis, McGrawHill, New York.
Bibliography
371
Bhatia, N. P. [1966], Weak attractors in dynamical systems,

Bol. Soc. Mat. Mexicana, 11, 56-64.
Bhatia, N. P., and Hajek, o. [1969], Local semi-dynamical systems, Lecture Notes in Mathematics, 90, Springer Verlag,
Berlin.- Heidelberg - New York.
-Bhatia, N. P., and Lakshmikantham, V. [1965], An extension of
Liapunov's direct method, Mich. Math. J., 12, 183-191.
Bhatia, N. P., and Szego, G. P. [1967], Dynamical systems:
stability theory and applications, Lecture Notes in Mathematics, 35, Springer Verlag, Berlin - Heidelberg - New
York.
-Billotti, J. E., and LaSalle, J.P. [1971], Dissipative
periodic processes, Bull. Amer. Math. Soc., 77, 10821088.
Birkhoff, G. D. [1927], Dynamical systems, American Mathematical Society Colloquium Publications, vol. IX, Providence,
Rhode Island.
Brayton, R. K., and Moser, J. K.[1964], A theory of nonlinear
network, Quart. Appl. Math., ~, 1-33, 81-104.
Bushaw, D. [1969], Stabilities of Liapunov and Poisson types,
SIAM Rev., 11, 214-225.
Cabannes, H. [1966], Cours de mecanique generale, Dunod, Paris.
Cesari, L. [1959], Asymptotic behavior and stability problems
in ordinary differential equations, Springer Verlag,
Berlin - Heidelberg - New York.
Chernous'ko, F. L. [1964], On the stability of regular precession of a satellite, Prikl. Mat. Meh., 28, 155-157,
J. Appl. Math. Mech., 28, 181-184.
-Chetaev, N. G. [1934], A theorem on instability (Russian),
Dokl. Akad. Nauk. SSSR, 1, 529-531.
Chetaev, N. G. [1952], On the instability of equilibrium in
some cases where the force function is not maximum (RuSsian), Prikl. Mat. Meh., !i, 89-93.
Chetaev, N. G. [1955], The stability of motion, Pergamon Press,
New York, 1961; translation of the Russian edition,
Moskow, 1955.
Coddington, E. A., and Levinson, N. [1955], Theory of ordinary differential equations, McGraw-Hill, New York.
Conti, R. [1956], Sulla prolungabilita delle soluzioni di un
sistema di equazioni differenziali ordinarie, Boll. Un.
Mat. Ital., 11 (3), 510-514.
372
BIBLIOGRAPHY
Conti, R., see Reissig, R.

Coppel, W. A. [1965], Stability and asymptotic behavior of
differential equations, D. C. Heath and Company, Boston.
Corduneanu, C. [1957], Systemes differentiels admettant des
solutions bornees, C. R. Acad. Sci. Paris, ser. A-B, 245,
21-24.
Corduneanu, C. [1960], Applications of differential ine~uali
ties to stability theory (Russian), An. Sti. Univ. 'AI I.
Cuza" , Iasi Sect. I a Mat., ~, 47-58.
Corduneanu, C. [1964], Sur la stabilite partielle, Revue
Roumaine de Math. Pures et Appl., ~, 229-236.
Corduneanu, C. [1971], Principles of differential and integral
equations, Allyn and Bacon Inc., Massachusetts.
Corne, J. L. [1973], L'attractivite mais c'est tres simple,
These de doctorat, Louvain-la-Neuve.
Corne, J. L., and Rouche, N. [1973], Attractivity of closed
sets proved by using a family of Liapunov functions, J.
Differential Equations, 13, 231-246.
Dana, M. [1972], Conditions for Liapunov stability, J. Differential Equations, 12, 590-609.
Dang Chau Phien [1973], Stabilite d'ensembles et fonctions
vectorielles de Liapunov, Seminaires de mathematique
appliquee et mecanique, 58, Louvain University.
Dang Chau Phien, and Rouche, N. [1970], Stabilite d'ensembles
pour des equations differentielles ordinaires, Riv. Mat.
Univ. Parma, 11, (2),1-15.
D'Anna, A. [1973], Asymptotic stability proved by using vector
Liapunov functions, Ann. Soc. Sci. Bruxelles, ser. I, 87
(2), 119-139.
De Castro, A. [1953], Sulle oscillazioni non-lineari dei
sistemi in uno 0 piu gradi di liberta, Rend. Sem. Mat.
Univ. Padova, 22, 294-304.
Deprit, A., and Deprit-Bartholome, A. [1967], Stability of
the triangular lagrangian points, Astronom. J., 72, 173179.
Dieudonne, J. [1960], Foundations of modern analysis, Academic press, New York - London.
D'Onofrio, B., Sarno, R., Lal~~, M. [1974], Unboundedness properties of solutions of n
order differential equations,
Boll. Un. Mat. Ital., (4), 10, 451-459.
Bibliography
373
Duhem, P. [1902], .sur les conditions necessaires pour la stabilite de l'equilibre d'un systeme visqueux, C. R. Acad.
Sci. Paris, 135, 939-941.
Etkin, B. [1959], Dynamics of flight:
'I'Jiley, New York.
Stability and control,
Fergola, P., and Moauro, V. [1970'], On partial stability,

Ricerche Mat., 19, 185-207.
Fiedler, M., and Ptak, V. [1960], On matrices with non ~osi
tive off diagonal elements and positive principal m1nors,
Czechoslovak Math. J., 12 (87), 382-400.
Fomin, S. V., see Kolmogorov, A. N.
Gambardella, L., and Salvadori, L. [1971], On the asymptotic
stability of sets, Ricerche Mat., ~, 143-154.
Gambardella, L., and Tenneriello, C. [1971], On a theorem of
N. Rouche, Rend. Accad. Sci. Fis. Mat. Napoli, (4), ~,
145-150.
Gerstein, V. M. [1969], The dissipativity of a certain twodimensional system, Differencial'nye Uravnenija, ~, 14381444.
Goel, N. S., Maitra, S. C., and Montroll, E. W. [1971], Nonlinear models of interacting populations, Academic Press,
New York - London.
Goldstein, H. [1950], Classical mechanics, Addison-Wesley,
Reading.
Goodwin, B. [1963], Temporal organization of cells, Academic
Press, New York.
Gorsin, S. [1948], On stability of motion under constantly
acting disturbances, Izv. Akad. Nauk. Kazah. SSR, 56,
Ser. Mat. Meh., , 46-73.
Gummel, H. K. [1968], A charge-control transistor model for
network analysis programs, Proc. IEEE, 56, 751.
Habets, P., and Peiffer, K. [1973], Classification of stability-like concepts and their study using vector Lyapunov
functions, J. Math. Anal. Appl., 43, 537-570.
Habets, P., and Peiffer, K. [1975], Attractivity concepts and
vector Lyapunov functions, Nonlinear Vibration Problems,
l6! 35-52.
Habets, P., and Risito, C. [1973], Stability criteria for systems with first inte~rals, generalizing theorems of Routh
and Salvadori, Equat10ns differentielles et fonctionnelles
non lineaires, ed. P. Janssens, J. Mawhin et N. Rouche,
Hermann, Paris, 570-580.
374
BIBLIOGRAPHY
Hadamard, J. [1897], Sur certaines proprietes des trajectoires

en dynamique, J. Math. Pures et Appl., ser. V, ~, 331387.
Haddock, J. R. [1972], A remark on a stability theorem of M.
Marachkoff, Proc. Am. Math. Soc., 31, 209-212.
Hagedorn, P. [1971], Die Umkehrung der Stabilitatssatze von
Lagrange-Dirichlet und Routh, Arch. Rational Mech. Anal.,
g, 281-316.
Hagedorn, P. [1972], Eine zusatzliche Bemerkung zu meiner
Arbeit: "Die Umkehrung der Stabilitatssatze von LagrangeDirichlet und Routh", Arch. Rational Mech. Anal., 47, 395.
Hagedorn, P. [1975], Uber die Instabilitat konservativer
Systeme mit gyroskopischen Kraften, Arch. Rational Mech.
Anal., ~, 1-9.
Hahn, W. [1959], Theory and application of Liapunov's direct
method, Prentice-Hall, Englewood Cliffs, 1963; translation
of the German edition, Springer Verlag, Berlin, 1959.
Hahn, W.
[1967], Stability of motion, Springer Verlag, Berlin.
Hahn, W. [1971], On Salvadori's one-parametric families of

Liapunov functions, Ricerche Mat., 20, 193-197.
Hajek, 0., see Bhatia, N. P.
Halanay, A. [1963], Differential equations: stability, oscillations, time lags, Academic Press, New York, 1966; translation of the Rumanian edition, Bucharest, 1963.
Hale, J. K. [1969], Ordinary differential equations, WileyInterscience, New York.
Hale, J. K., LaSalle, J. P., and Slemrod, M. [1972), Theory of
a general class of dissipative processes, J. Math. Anal.
Appl., 39, 177-191.
Hamel, G. [1903], Uber die Instabilitat der Gleichgewichtslage
eines Systems von zwei Freiheitsgraden, Math. Ann., ~,
541-553.
Hartman, P. [1964], Ordinary differential equations, John
Wiley, New York.
Hicks, J. R. [1939], Value and capital, Oxford University
Press, London, Clarendon Press, Oxford, 1945 (2d edition).
Hing, C. So., see Mitra, D.
Hoppensteadt, F. C. [1966), Singular perturbations on the infinite interval, Trans. Amer. Math. Soc., 123, 521-535.
Huaux, A. [1964], Sur la methode directe de Ljapunov, Universite Libre de Bruxelles, Faculte Poly technique de Mons.
Bibliography
375
Ibrachev, Kh. I. [1947], On Liapunov's second method, Izv.

Akad. Nauk. Kazah. SSR, Ser. Fiz. Mat., ~, 101-110.
Kalman, R. E. [1963], Liapunov functions for the problem of
Lurie in automatic controls, Proc. Nat. Acad. Sci. USA,
i2,., 201-205.
Kappel, F., see Knobloch, H. W.
Kneser, A. [1895-1897], Studien uber die Bewegungsvorgange in
der umgebung instabiler Gleichgewichtslagen, J. Reine
Angew. Math., 115, 1895, 308-327 and 118, 1897, 186-223.
Knobloch, H. W., and Kappel, F. [1974], Gewohnliche Differentialgleichungen, B. G. Teubner, Stuttgart.
Koiter, W. T. [1965], On the instability of equilibrium in
the absence of a minimum of the potential energy, Nederl.
Akad. Wetensch. Proc. ser. B, ~, 107-113.
Kolmogorov, A. N., and Fomin, S. V. [1960], Measure, Lebesgue
integrals and Hilbert space, Academic Press, New York London, 1961; translation of the Russian edition, Moskow,
1960.
Krasovski, N. N. [1956], The inversion of the theorems of
Liapunov's second method and the question of stability of
motion using the first approximation, Prikl. Mat. Meh.,
~, 255-265.
Krasovski, N. N. [1959], Problems of the theory of stability
of motion, Stanford Univ. Press, Stanford, California,
1963; translation of the Russian edition, Moskow, 1959.
Krasovski, N. N., see Barbashin, E. A.
Kuratowski, K. [1961], Introduction to set theory and topology,
Pergamon Press, Oxford.
Kurzweil, J. [1955], Reversibility of Liapunov's first theorem
on stability of motion (Russian), Czechoslovak Math. J.,
~, 382-398.
Lagrange, J. L. [1788], Mecanique analytique, Paris, reed.
Mallet - Bachelier, Paris, 1853-1855.
Lakshmikantham, V., and Leela, S. [1969], Differential and
integral inequalities, theory and applications, Academic
Press, New York - London.
Lakshmikantham, V., Mitchell, A. R., and Mitchell, R. W.
[1975], Maximal and minimal solutions and comparison results for differential equations in abstract cones, University of Texas at Arlington, TR 27.
Lakshmikantham, V., see Bhatia, N. P.
376
BIBLIOGRAPHY
Laloy, M. [1973]1' Une extension de la methode des secteurs

pour l'etude de l'instabilite
Bruxelles, ser. I, ~, 17-49.
la Liapunov, Ann. Soc. Sci.
Laloy, M. [1973]2' Prolongements de la methode des secteurs de

Persidskii, Ann. Soc. Sci. Bruxelles, ser. I,
~,
141-164.
Laloy, M. [1973]3' utilisation des integrales premieres dans

l'etude de la stabilite, Sem. Math. Appl. Meca., Louvain, 62.
Laloy, M. [1973]4' Stabilite
1a Liapunov du betatron, Equa-
tions differentie11es et fonctionnelles non lineaires, ed.

P. Janssens, J. Mawhin, N. Rouche, Hermann, Paris, 582595.
La10y, M. [1974]1' La
m~thode
des secteurs appliquee
l'equa-
tion sca1aire du n e ordre dnx/dt n = f(t,x), An. Sti. Univ.

"Al. I. Cuza", Iasi Sect. I a Mat., ~, 39-51.
Laloy, M. [1974]2' Problemes d'instabilite pour des equations
differentie1les ordinaires et fonctionne1les, These de
doctorat, Louvain-la-Neuve.
La10y, M. [1975], On equilibrium instability for conservative
and partially dissipative mechanical systems, Sem. Math.
App1. Meca., Louvain, 82.
Laloy, M., see D'Onofrio, B.
Lanchester, F. W. [1908], Aerial flight II: aerodonetics, Constable, London.
Lanczos, B. [1962], Variational principles in mechanics, Handbook of Engineering Mechanics, W. Flugge, ed., McGrawHill, New York.
LaSalle, J. P. [1960], Some extensions of Liapunov's second
method, IRE Trans. Circuit Theory CT-I, 520-527.
LaSalle, J. P. [1962], Asymptotic stability criterion, Proc.
Sympos. Appl. Math., vol. 13, 299-307, Amer. Math. Soc.,
Providence, R. I.
LaSalle, J. P. [1968], Stability theory for ordinary differential equations, J. Differential Equations, !, 57-65.
LaSalle, J. P., and Lefschetz, S. [1961], Stability by
Liapunov's direct method with applications, Academic press,
New York.
LaSalle, J. P., see Bil1iotti, J. E.; see Hale, J. K.
Bibliography
377
Leela, S., see Lakshmikantham, V.

Lefschetz, S. [1965}, Stability of nonlinear control systems,
Academic Press, New York.
Lefschetz, S., see LaSalle, J. P.
Leimanis, E. [1965}, The general problem of the motion of
coupled rigid bodies about a fixed point, Springer Verlag,
Berlin - Heidelberg - New York, p. 300.
Leipholz, H. [1968}, Stability theory, Academic Press, New
York, 1970; translation from the German edition, Stuttgart,
1968.
Lejeune-Dirichlet, G. [1846], Uber die Stabilitat des Gleichgewichts, G. Lejeune- Dirichlet's Werke, Vol. II, Georg
Reimer, Berlin, 1897; first published in Abh. Preussische
Akad. Wiss., 1846.
Leontovich, A. M. [1962], On the stability of Lagrange's
periodic solutions of the restricted three-body problem,
Dokl. Akad. Nauk. SSSR, 143, 525-528, Soviet Math. Dokl.,
l, 425-430.
Levi-Civita, T., and Amaldi, U. [1922-1927], Lezioni di meccanica razionale, Zanichelli, Bologna, vol. I, 1922, vol.
Ill' 1926, vol. 1I 2 , 1927.
Levinson, N. [1944}, Transformation theory of non-linear
differential equations of second order, Ann. of Math.,
45, 723-737.
Levinson, N., see Coddington, E. A.
Liapunov, A. M. [1892}, prob1eme general de 1a stabi1ite du
mouvement, Photo-reproduction in Annals of Mathematics,
Studies nO 17, princeton Univ. Press, Princeton, 1949, of
the 1907 French translation of the fundamental Russian
paper of Liapunov published in Comm. Soc. Math., Kharkow,
1892.
Liapunov, A. M. L1897}, Sur l'instabilite de l'equi1ibre dans
certains cas ou 1a fonction de forces n'est pas un maximum, J. Math. Pures App1., ser. V, l' 81-94.
Lotka, A. J. [1920}, Analytical note on certain rhythmic relations in organic systems, Proc. Nat. Acad. Sci. USA, ~,
410-415.
Lur'e, A. I. [1968}, Mecanique ana1ytique, Librairie Universitaire, Louvain.
McShane, E. J. [1944}, Integration, Princeton University Press,
Princeton.
Maitra, S. C., see Goe1, N. S.
378
BIBLIOGRAPHY
Malkin, I. G. [1944}, Stability in the case of constantly acting disturbances, Prikl. Mat. Meh., ~, 241-245.
Malkin, I. G. [1952}, Theorie der Stabilitat einer Bewegung,
translated by W. Hahn and R. Reissig from the Russian edition, 1952, R. Oldenburg, Munich, 1959.
Malkin, I. G. [1954}, On the reversibility of Liapunov's
theorem on asymptotic stability (Russian), Prik1. Mat.
Meh., 18, 129-138.
Manfredi, B. [1956}, Su11a stabilita del moto di sistemi a p~u
gradi di liberta in condizioni non lineari, Boll. Un. Mat.
Ital., sere III, 64-71.
Marachkov, M. [1940}, On a theorem on stability (Russian),
Bull. Soc. Phys.-Math., Kazan, 12, 171-174.
Markeev, A. P. [1969}, On the stability of the triangular
libration points in the circular bounded three-body problem, Prikl. Mat. Meh., 33, 112-116, J. Appl. Math. Mech.,
~, 105-110.
-Markus, L. [1956}, Asymptotically autonomous differential systems, Contr. Theor. Nonlinear Oscillations, ed. by S.
Lefschetz, vol. 3, Princeton Univ. Press, 17-29.
Markus, L., and Yamabe, H. [1960}, Global stability criteria
for differential equations, Osaka Math. J., 12, 305-317.
Massera, J. L. [1949}, On Liapounoff's conditions of stability,
Ann. of Math., 50, 705-721.
Massera, J. L. [1956-1958}, Contributions to stability theory,
Ann. of Math., 64, 1956, 182-206, Erratum Ann. of Math.,
~, 1958, 202. -Matrosov, V. M. [1962]1' On the stability of motion, Prik1.
Mat. Meh., 26, 885-895, J. Appl. Math. Mech., 26, 13371353.
-Matrosov, V. M. [1962]2' On the theory of stability of motion,
Prikl. Mat. Mech. 26, 992-1002, J. Appl. Math. Mech.,
26, 1506-1522.
Matrosov, V. M. [1963], On the stability of motion II (Russian), Trudy Kazan. Aviacion. Inst., 80, 22-33.
Matrosov, V. M. [1965], Development of the method of Liapunov
functions in stability theory, Proceedings second AllUnion Conference in theoretical and applied mechanics,
L. "I. Sedov ed., 1965; English trans1. Israel program for
scientific translation, Jerusalem, 1968.
Matrosov, V. M. [1968-1969], Comparison principle with vector
Liapunov functions I-IV, Differencia1'nye Uravnenija, i,
1968, 1374-1386, 1739-1752, 5, 1969, 1171-1185, 21292145.
-
Bibliography
379
Matrosov, V. M. [1973], Comparison method in system's dynamics,

Equations differentielles et fonctionnelles non lineaires,
ed. P. Janssens - J. Mawhin - N. Rouche, Hermann, Paris,
407-445.
Mawhin, J., see Rouche, N.
Metzler, L. [1945], Stability of multiple markets:
conditions, Econometrica, 13, 277-299.
the Hicks
Michel, A. N. [1969], On the bounds of the trajectories of

differential systems, Int. J. Control, 10, 593-600.
Michel, A. N. [1974], Stability analysis of interconnected
systems, SIAM J. Control, 12, 554-579.
Miller, R. K. [1965], Asymptotic behavior of solutions of nonlinear differential equations, Trans. Amer. Math. Soc.,
115, 400-416.
Miller, R. K., and Sell, G. R. [1968], Existence, uniqueness
and continuity of solutions of integral equations, Ann.
Mat. Pur a Appl., (4), 80, 135-152.
Miller, R. K., and Sell, G. R. [1970], Volterra integral equations and topological dynamics, Mem. Amer. Math. Soc.,
102.
Mitchell, A. R., see Lakshmikantham, V.
Mitchell, R. W., see Lakshmikantham, V.
Mitra, D., and Hing C. So. [1972], Existence conditions for Ll
Lyapunov functions for a class of nonautonomous systems,
IEEE Trans. Circuit Theory, CT-19, 594-598.
Moauro, V., see Fergola, P.
Montroll, E. W., see Goel, N. S.
Moser, J. K., see Brayton, R. K.
Muller, 1-1. D. [1965], Zum Nachweis der Instabilitat der
Rubelagen bei gewissen Differentialgleichungen, Z. Angew.
Math. Mech., 45, 359-360.
Myshkis, A. D. [1947], Sur un lemma geometrique qui s'applique
dans la theorie de stabilite au sens de Liapunov, Dokl.
Akad. Nauk. SSSR, 55, 299-302.
Nemytskii, V. V. [1965], Some modern problems in the qualitative theory of ordinary differential equations, Russian
Math. Surveys, 20 (IV), 1-34.
Opial, Z. [1960], Sur la dependance des solutions d'un systeme d'equations differentielles de leurs seconds membres,
application aux systemes presque autonomes, Ann. Polon.
Math., ~, 75-89.
380
BIBLIOGRAPHY
Oziraner, A. S. [1972], On certain theorems of Liapunov's

second method, Prik1. Mat. Meh., 36, 396-404, J. App1.
Math. Mech., 36, 373-381.
-Pain1eve, P. [1897], Sur 1es positions d'equi1ibre instable,
C. R. Acad. Sci. Paris, sere A-B, 25 (2), 1021-1024.
Pain1eve, P. [1904], Sur 1a stabi1ite de l'equi1ibre, C. R.
Acad. Sci. Paris, sere A-B, 138, 1555-1557.
Pavel, N. [1971], On dissipative systems, Boll. Un. Mat. Ital.,
!, 701-707.
Pavel, N. [1972]1' A generalization of ultimately bounded systems, An. St1. Univ. "A1. I. Cuza" Iasi Sect. I a Mat. ,
18, 81-86.
Pavel, N. [1972J 2 , On the boundedness of solutions of a system
of differential equations, Tohoku Math. J., 24, 21-32.
Peiffer, K., and Rouche, N. [1969], Liapunov's second method
applied to partial stability, J. Mecanique, !, 323-334.
Peiffer, K., see Habets, P., see Rouche, N.
Persidski, K. P. [1933], On the stability of motion in first
approximation, Mat. Sb., 40, 284-293.
Persidski, K. P. [1946], On the theory of stability of solutions of differential equations (Russian), Doctoral dissertation, Moskov. Gos. Univ., a summary was published
in Uspehi Mat. Nauk., !, 250-255.
Persidski, K. P. [1947], On Liapunov's second method, Izv.
Akad. Nauk. Kazah, SSR, sere Fiz.-Mat., 42, 48-55.
Persidski, S. K. [1961], On the second method of Liapounov,
Prik1. Mat. Meh., 25, 17-23, J. App1. Math. Mech., 25,
20-28.
Persidski, S. K. [1968], Investigation of stability of solutions of some nonlinear systems of differential equations,
prik1. Mat. Meh., 32, 1122-1125, J. App1. Math. Mech., 32,
1141-1144.
-Persidski, S. K. [1970], Investigating the stability of solutions of systems of differential equations, Prik1. Mat.
Meh., ~, 219-226, J. App1. Math. Mech., ~, 209-215.
Piontkovskii, A. A., and Rutkovskaya, L. D. [1967], Investigation of certain stability theory problems by the vector
Lyapunov function method, Avtomat. i Te1emeh., 10, 2331, Automat. Remote Control, ~, 1422-1429.
-P1iss, V. A. [19641, Non1oca1 problems of the theory of oscillations, Academic Press, New York - London, 1966, translated from the Russian edition, Moscow - Leningrad, 1964.
Bibliography
381
Pluchino, S. [1971], Osservazioni sulla stabilita dei moti

merostatici di un sistema olonomo od anolonomo, Boll. Un.
Mat. Ital., !, 213-219.
Poisson, S. D. [1838], Traite de mecanique, 3e ed., Societe
beIge de Librairie, Bruxelles.
Pontryagin, L. S. [1961], Ordinary differential equations,
Addison-Wesley, Reading, Massachusetts, 1962; translation
of the Russian edition, Moskow, 1961.
popov, v. M. [1962], Absolute stability of nonlinear systems
of automatic control, Avt. Telemeh., 22, 961-979, Aut.
Rem. Control, ~, 857-875.
-Popov, V. M. [1966], Hyperstability of control systems,
Springer Verlag, Berlin - Heidelberg - New York, 1973;
revised translation of the Romanian edition, Bucarest,
1966.
Pozharitskii, G. K. [1958], On the construction of Liapunov
functions from the integrals of the equations of the perturbed motion (Russian), Prikl. Mat. Meh., 22, 145-154.
ptak, v., see Fiedler, M.
Reissig, R., Sansone, G., and Conti, R. [1969], Nichtlineare
Differentialgleichungen hoherer Ordnung, Edizioni Cremonese, Roma.
Risito, C. [1967], On the Liapunov stability of a system with
known first integrals, Meccanica, ~, 197-200.
Risito, C. [1970], Sulla stabilita asintotica parziale, Ann.
Mat. Pur a Appl., (4), ~, 279-292.
Risito, C. [1971], Some theorems on the stability and the partial asymptotic stability of systems with known first integrals, Comptes rendus des journees nationales du C.B.R.M.,
Mons, 24-26 mai, 53-56.
Risito, C. [1972], The comparison method applied to the stability of systems with known first integrals, Proc. 6th
Int. Conf. on Nonlinear Oscillations, Poznan, 1972, Nonlinear vibration problems, 15, 1974, 25-45.
Risito, C. [1974], Metodi per 10 studio della stabilita di
sistemi con integrali primi noti, Seminaires de mathematique appliquee et mecanique, 74, Louvain University; to
appear in Ann. Mat. Pura App~
Risito, C. [1975], On the Chetayev method for the construction
of a positive definite first integral, Ann. Soc. Sci.
Bruxelles, ser, I, ~, 3-10.
Risito, C., see Habets, P.
382
BIBLIOGRAPHY
Rosen, R. [1970], Dynamical system theory in biology, Wi1eyInterscience, New York.

Rouche, N. [1968], On the stability of motion, Int. J. NonLinear Mechanics, l, 295-306.
Rouche, N. [19691, Quelques criteres d'instabi1ite a 1a
Liapunov, Ann. Soc. Sci. Bruxe11es, ser. I, ~, 5-17.
Rouche, N. [1971], Attractivity of certain sets proved by
using several Liapunov functions, Symposia Mathematica,
~, 331-343.
Rouche, N. [1974], Theorie de 1a stabi1ite dans 1es equations
differentie11es ordinaires, Stability Problems, 1st ClME
session, 1974, Edizioni Cremonese, Rome, 111-194.
Rouche, N. [1975], The invariance principle applied to noncompact limit set, Boll. Un. Mat. Ita1. 11, 306-315.
Rouche, N., and Mawhin J. [1973], Equations differentie11es
ordinaires, Masson, Paris.
Rouche, N., and Peiffer, K. [19671, Le theoreme de LagrangeDirichlet et 1a deuxieme methode de Liapounoff, Ann. Soc.
Sci. Bruxe11es, Ser. 1, ~, 19-33.
Rouche, N., see Corne, J. L.l see Dang Chau Phien.
Routh, E. J. [1905], The advanced part of a treatise on the
dynamics of a system of ri~~d bodies, Republished by Dover,
New York, 1955, from the 6
edition, 1905.
Routh, E. J. [1975], Stability of motion, Repub1. of selected
papers, Taylor and Francis Ltd., London.
Royden, H. L. [1963], Real analysis, McMillan, New York.
Rubanovskii, V. V., and Stepanov, S. Ia. [1969], On the Routh
theorem and the Chetaev method for constructing the
Liapunov function from the integrals of the equations
of motion, Prik1. Mat. Meh., 33, 904-912, J. App1. Math.
Mech., ll, 882-890.
Rumiantsev, V. V. [19571, On the stability of a motion in a
part of variables (Russian), Vestnik Moskov. Univ. Ser. I
Mat. Meh., !, 9-16.
Rumiantsev, V. V. [19661, On the stability of steady motions,
Prik1. Mat. Meh., 30, 922-933, J. App1. Math. Mech., ~,
1090-1103.
-Rumiantsev, V. V. [19681, On the stability of steadystate motions, Prik1. Mat. Meh., ~, 504-508, J. App1. Math.
Mech., ~, 517-521.
Bibliography
383
Rumiantsev, V. v. [1970], On the optimal stabilization of

controlled systems, Prikl. Mat. Meh., 34, 440-456, J. Appl.
Math. Mech., l!, 415-430.
-Rumiantsev,
part of
lineare
di Alta
V. V. [1971], On the stability with respect to a

the variables, Symposia Math., 6, Meccanica non
e stabilita, 23-26 Febbraio, 1970, Roma, Ist. Naz.
Matematica, Academic Press, New York, 1971.
Rutkovskaya, L. D., see Piontkovskii, A. A.

Salvadori, L. [1953], Un' osservazione su di un criterio di
stabilita del Routh, Rend. Accad. Sci. Fis.-Mat. Napoli,
20, 269-272.
Salvadori, L. [1966]; Sull' estensione ai sistemi dissipativi
del criterio di stabilita del Routh, Ricerche Mat., 15,
162-167.
Salvadori, L. [1968], Sulla stabilita dell' equilibrio nella
meccanica dei sistemi olonomi, Boll. Un. Mat. Ital., i,
333-344.
Salvadori, L. [1969], Sulla stabilita del movimento,
che, 24, 218-239.
Matemati~
Salvadori, L. [1971], Famiglie ad un parametro di funzioni di

Liapunov nello studio della stabilita, Symposia Math., ~,
309-330.
Salvadori, L. [1972], Sul problema della stabilita asintotica,
Rendiconti dell' Accad. Naz. Lincei, (8), 53, 35-38.
Salvadori, L. [1974], Some contributions to asymptotic stability theory, Ann. Soc. Sci. Bruxelles, ser, I, ~, 183194.
Salvadori, L., see Gambardella, L.
Samuelson, P. A. [1947], Foundations of economic analysis,
Harvard University Press, Cambridge, Massachusetts.
Sandberg, I. W. [1969], Some theorems on the dynamic response
of nonlinear transistor networks, Bell Syst. Tech. J.,
i!, 35-54.
Sansone, G., see Reissig, R.
Sarno, R., see D'Onofrio, B.
Schuur, J. D. [1967], The asymptotic behavior of a solution
of the third order linear differential equation, Proc.
Amer. Math. Soc., 18, 391-393.
Sell, G. R. [19711, Topological dynamics and ordinary differential equations, Van Nostrand Reinhold Company, London.
Sell, G. R., see Miller, R. K.
384
BIBLIOGRAPHY
Si1jak, D. D. [1973], Competitive economic systems: stability,

decomposition and aggregation, Proceeding of the 1973
IEEE Conference on Decision and Control, San Diego, California, December 5-7, 265-275.
Sil1a, L. [1908], Su11a instabi1ita dell'equilibrio di un sistema materia1e in posizioni non isolate, R. Accad. Lincei,
.!I, 347-355.
Skowronski, J. M. [1969], Multiple nonlinear lumped systems,
Polish scientific publishers, Warsaw.
Slemrod, M., see Hale, J. K.
Smets, H. B. [1961], Stability in the large of heterogeneous
power reactors, Acad. Roy. Be1g. Bull. C1. Sci., 47, 382405.
Stepanov,
s.
Ia., see Rubanovskii, V. V.
Stern, T. E. [1965], Theory of nonlinear networks and systems,

Addison-Wesley, Reading, Massachusetts.
Szarski, J. [1967], Differential inequalities, Polish Scientific Publishers, Warsaw.
Szego, G. P., see Bhatia, N. P.
Tait, P. G., see Thomson, W.
(Lord Kelvin).
Tennerie110, C., see Gambardella, L.

Thomson, W. (Lord Kelvin), and Tait, P. G. [1912], Principles
of mechanics and dynamics, Cambridge U. P. (last ed.),
1912, reprinted by Dover Publications, Inc. 2 vol., 1962.
Van Chzhao-Lin [1963], On the converse of Routh's theorem,
Prikl. Mat. Meh., 27, 890-893, J. Appl. Math. Mech., 27,
1354-1360.
-Verhulst, P. [1845], Recherches mathematiques sur 1a loi
d'accroissement de 1a population, Hemoires Acad. Roy.
Bruxe11es, ~, 3-39.
Verhulst, P. [1847], Deuxieme memoire sur 1a loi d'accroissement de 1a population, Memoires Acad. Roy. Bruxe11es, ~,
3-32.
Vinograd, R. E. [1957], The inadequacy of the method of characteristic exponents for the study of nonlinear differential equations (Russian), Mat. Sbornik, 41 (83), 431-438.
Volterra, V. [1931], Le~on sur 1a theorie mathematique de 1a
1utte pour 1a vie, ~authier-Vi11ars, Paris.
I. [1959], Integral stability, Czechoslovak Math. J.
9 (84), 71-129.
Vrko~,
Bibliography
385
Walter, W. [1964], Differential and integral inequalities,

Springer Verlag, New York - Heidelberg - Berlin, 1970;
translation of the German edition, 1964.
Warden, R. B., Aris, R., and Amundson, N. R. [1964], An analysis of chemical reactor stability and control VIII, Chern.
Eng. Sc., 19, 149-172.
Wazewski, T. [1947], Sur un principe topo10gique de l'examen
de l'allure asymptotique des integrales des equations differentielles ordinaires, Ann. Polon. Math., 20, 279-313.
Wazewski, T. [1950], Systemes des equations et des inegalites
differentielles ordinaires aux deuxiemes membres monotones
et leurs applications, Ann. Soc. Polonaise Math., 23, 112166.
Wintner, A. [1941], The analytical foundation of celestial
mechanics, University Press, Princeton.
Yacubovich, V. A. [1962], Solution of certain matrix inequalities occuring in the theory of automatic controls, Dokl.
Akad. Nauk. SSSR, 143, 1304-1307.
Yamabe, H., see Markus, L.
Yorke, J. A. [1968], An extension of Chetaev's instability
theorem using invariant sets, and an example, Seminar on
differential equations and dynamical systems, ed. by G. S.
Jones, Springer Verlag, Berlin, 100-106.
Yoshizawa, T. [1959], Liapunov's function and boundedness of
solutions, Funkcial. Ekvac., ~, 71-103.
Yoshizawa, T. [1963], Asymptotic behavior of solutions of a
system of differential equations, Contributions Differential Equations, ~, 371-387.
Yoshizawa, T. [1966], Stability theory by Liapunov's second
method, The Math. Soc. of Japan, Tokyo.
Zhukovski, N. E. [1891], On soaring of birds (Russian), Trudy
Otdel. Fiz. Nauk. Obshch. Lyubit. Yestestvz., 4 (2), 29,
1891, Reprinted in Collected Works, 4, Gostekhizdat, 1949,
5.
Ziemba, S. [1961], Some problems of the Warsaw group on the

theory of nonlinear oscillations in the last five year
programme, Proc. Internat. Symp. Nonlinear Oscillations,
T II, Kiev, 161-171.
Zubov, V. I. [1957], The methods of Liapunov and their applications, Noordhoff, Groningen, 1964; translated from the
Russian edition, Leningrad, 1957.
AUTHOR INDEX
M1ALDI, Uo1 I.5.10, p. 24

AMUNDSON,
II.2.9, p. 70
N.R.~
ANTOSIEWICZ, H.A.; I.6.20, p. 34~ I.6.21, p.

VI.1.1, p. 203~ VI.B, p. 239.
APPEL,
loB, p.
4B~
III.7, p. 126
P.~
ARIS, R.
ARNOLD,
V.I.~
AVDONIN,
34~
II. 2.9, p. 70
III.6.9, p. 122
L.N.~
AVRAMESCU,
C.~
III.3.5, p. 107
VI.3.4,
p. 211; VI.B, p. 240
BAILEY, F.N.; IX.4.3, p.
32B~
IX.7, p. 343
BARBASHIN, E.A.~ I.2.11, p. 10; I.6.9, p. 29~ II.1.7, p. 79;

II.1.9, p. 56~ II.7, p. 95~ VII.B, p. 269
BELETSKII,
BELLMAN,
V.V.~
R.~
IV.7.14, p. 162
III.6.5, p.
11B~
IX.7, p. 342
BHATIA, N.P.~ II.3.9, p. 79~ VI.1.1, p. 203; VI.3.3, p. 211;

VI.B, p. 239~ VIII.6.2, p. 306; IX.3.7, p. 324~
IX.7, p. 343
BILLOTTI, J .Ed VIU 7, p. 312
BIRKHOFF, G.D.; AII.6, p. 357
BRAYTON, R.K.; VII.6.2, pp. 254,
VII.B, p. 269
BUSHAW,
D.~
VI.3.1, p.
20B~
255~
VII.6.4, p.
259~
VI.B, p. 239
CABANNES, H.; I.4.7, p. 16

CESARI,
L.~
CHERNOUS'KO,
IV.4.5, p.
F.L.~
13B~
VI.B, p. 239
IV.7.14, p. 163
CHAETAEV, N.G.~ I.4.7, p. 16~ I.4.B, p. IB~ I.5.1, p. 14~

I.5.6, p. 21~ I.6.31, p. 42~ I.B, p. 4B; III.3.2, p. 105~
III.3.3, p. 106~ III.3.6, p. 107~ III.6.3, p. IlB;
IV.7.4, p. 153~ IV.7.13, pp. 159, 160~ IV.B, p. 167~
V.I0, p. 19B
CODDINGTON,
E.A.~
VIII.5.1, p. 295
Author Index
3B7
CONTI, R.; II.7, p. 95; V.9.4, p. 19B; IX.7, p. 342

COPPEL, W.A.;
(i)
CORDUNEANU, C.; I.B, p. 48; II.e.4, p. 75; II.3.10, p. 79;

II.7, p. 95; VIII.7, p. 312; IX.7, pp. 342, 344
CORNE, J.L.; VII.8, p. 269; VIII.7, p. 311
DANA, M.; VI.B, p. 240
DANG CHAU PHIEN; V.10, p. 199; IX.7, p. 344
D'ANNA, A.; VIII.7, p. 310
DE CASTRO, A.; VIII.7, p. 312
DEPRIT, A.; III.6.9, p. 122
DEPRIT-BARTHOLOME, A.; III.6.9, p. 122
DIEUDONNE, J.; IX.2.5, p. 317
D'ONOFRIO, B.; V.I0, p. 200
DUHEM, P.; II I. 7, p 127
ETKIN, B.; I.4.8, p. 19
FERGOLA, P.; VIII.7, p. 310
FIEDLER, M.; IX.5, p. 335
FOMIN, S.V.; AI.3.1, p. 350
GAMBARDELLA, L.; VIII.7, p. 310
GERSTEIN, V.M.; VIII.7, p. 312
GOEL, N.S.; VII.7.1, p. 260; VII.B, p. 269
GOLDSTEIN, H.; IV.5.5, p. 141; IV.6.2, p. 146; AII.6, p. 357
GOODWIN, B.; II.1.11, p. 59
GORSIN, S.; II.4.5, p. B2
GUMMEL, H.K.; II.6.5, p. 91
HABETS, P.; IV.B, pp. 166, 167; VI.1.1, p. 203; VI.6.4, p. 229;
VI.8, pp. 239, 240; IX.7, p. 343
HADAMARD, J.;
III.7,~.
126
388
AUTHOR INDEX
HADDOCK, J.R.; I.6.30, p. 42

HAGEDORN, P.; III.6.16, p. 125; III.7, p. 126
HAHN, W.; I.2.7, p. 9; I.2.13, p. 11; I.3.2, p. 12;
I.7.1, p. 46; I.8, p. 48; II.4.8, p. 84; II.7, p. 96;
V.9.3, p. 197; VI.8, p. 239; VIII.7, p. 310
HAJEK, 0.; VIII.6.2, p. 306
HALANAY, A.; I.6.35, p. 44; I.8, p. 48; II.7, p. 96;
VI.8, p. 239
HALE, J.K.; I.6.27, p. 39; VIII.7, p. 312
HAMEL, G.; III.2.12, p. 102; III.7, p. 126
HARTMAN, P.; V.4.6, p. 178; VIII.5.5, p. 300;
VIII.5.6, p. 304; VIII.7, p. 311; AIII.10, p. 367
HICKS, J.R.; IX.5, p. 332
HING, C.SO.; II.7, p. 96
HOPPENSTEADT, F.C.; I.7.1, p. 45
HUAUX, A.; V.9.4, p. 198
IBRACHEV, KH.I.; V.10, p. 199
KALMAN, R.E.; II.5.3, p. 86; II.5.4, p. 86
KAPPEL, F.; I.8, p. 48
KNESER, A.; II.7, p. 126
KNOBLOCH, H.W.; I.8, p. 48
KOlTER, W.T.; III.4.5, p. 112; III.7, p. 127
KOLMOGOROV, A.N.; AI.3.1, p. 350
KRASOVSKI, N.N.; I.2.11, p. 10; I.6.9, p. 29; I.7.1, p. 46;
II.1.3, p. 50; II.1.9, p. 56; II.7, p. 95; V.10, p. 200;
VII.8, p. 269
KURATOWSKI, K.; AI.3.5, p. 352
KURZWEIL, J.; I.7.2, p. 46
LAGRANGE, J.L.; I.1.6, p. 5; III.1, p. 97
LAKSHMlKANTHAM, V.; II.3.9, p. 79; VI.8, p. 239; IX.3.7,
p. 324; IX.7, pp. 342, 343
Author Index
389
LALOY, M.: II.2.13, p. 102, III.2.15, p. 104, IV.6.2, p. 146,

IV.8, pp. 166, 167, V.4.10, p. 180, V.9.1, p. 195,
V. 9.4, p. 198, V.9.5, p. 198, V.I0, pp. 199, 200,
VII.8, p. 269, VIII.3.1, p. 280, VIII.7, p. 311
LANCHESTER, F.W., I.4.8, p. 19
LANCZOS, B., III.7, p. 12'6
LASALLE, J.P., I.8, p. 48, 11.1.10, p. 58, VII.3.2, p. 243,
VII.8, p. 269, VIII.4.2, p. 286: VIII.4.8, p. 289,
VIII.4.10, p. 290, VIII.4.14, p. 291, VIII.4.16, p. 293,
VIII.7, pp. 311, 312
LEELA, 5., VI.8, p. 239, IX.7, p. 342
LEFSCHETZ, S., 1.8, p. 48, 11.1.10, p. 58, 11.5.2, p. 84,
II.5.5, p. 88
LElMANIS, E., IV.7.14, p. 162
LEIPHOLZ, H., 11.1.9, p. 56, 111.6.9, p. 121
LEJEUNE-DIRICHLET, G., 111.1, p. 97
LEONTOVICH, A.M., 111.6.9, p. 122
LEVI-CIVITA, T., I.5.10, p.
~4
LEVINSON, N., VI.6.18, p. 234, VIII.5.1, p. 295

LIAPUNOV, A.M., 1.1.2, p. 3, I.2.3, p. 7, 1.2.11, p. 10,
I.3.1, p. II, 1.4.2, p. 13, 1.5.3, p. 20, 1.5.4, p. 20;
1.5.8, p. 22; I.6.1, p. 25; I.6.14, p. 31, I.6.15, p. 32;
111.3.7, p. 107; 111.4.2, p. 108; IV.3.10, p. 134;
IV.8, p. 165; V.10, p. 198; VI.8, p. 238
LOTKA, A.J.; VII.7.1, p. 260: VII.7.9, p. 266
LUR'E, A.I.; AII.6, p. 357; AII.11, p. 360
McSHANE, E.J.; V.4.5, p. 176; AI.1, p. 347: AI.2.5, p. 349;
AI.3.1, p. 350: AI.3.3, p. 351
MALKIN, I.G.: 1.2.11, p. 10: I.7.1, p. 46; I.8, p. 48;
11.4, p. 82: 11.4.5, p. 82: 11.4.6, p. 83: VI.8, p. 238
MANFREDI, B.; VIII.7, p. 312
MARACHKOV ,M., 1.6.25, p. 37
MARKEEV, A.P.; 111.6.9, p. 122
MARKUS, L.: V.9.1, p. 195; VIII.7, p. 311
MASSERA, J.L.; 1.2.8, p. 9; 1.2.11, p. 10; 1.2.13, p. 11;
1.6.5, p. 27; 1.6.20, p. 34: 1.6.26, p. 37; 1.7.4, p. 46;
11.4.9, p. 84; II.4.10, p. 84; V.I0, p. 199; VI.l.1, p. 203;
VI.8, p. 239
390
AUTHOR INDEX
MATROSOV, V.M.; 11.1.4, p. 53; 11.2.5, p. 62; 11.2.7, PP. 67, 69;
11.7, p. 95; 111.7, p. 127; V.10, p. 200; VI.8, p. 240;
VIII.2.3, p. 272; VIII.7, pp. 310, 312; IX.1, p. 313;
1X.3.5, p. 322; IX.3.10, p. 326; IX.3.11, p. 327;
IX.7, pp. 342, 343, 344
MAWHIN, J.; 1.3.4, p. 13; 1.5.8, p. 22: 1.6.15, p. 32: 1.6.17,
p. 33; I. 8, p. 48; I I. 2 6, p. 66: V. 5. 5, p. 182: IX. 2 . 5 ,
p. 318
METZLER, L.: IX.5, p. 335
MICHEL, A.N.: VI.8, p. 239; IX.7, p. 343
MILLER, R.K.; VIII.7, pp. 311, 312
MITCHELL, A.R.: IX.7, p. 343
MITCHELL, R.W.: IX.7, p. 343
MITRA, D.; 11.7, p. 96
MOAURO, V.; VIII.7, p. 310
MOSER, J.: VII.6.2, pp. 254, 255; VII.6.4, p. 259: VII.8, p. 269:
MULLER, D.W.: V.10, p. 200
MYSHKIS, A.D.; V.10, p. 199
NEMYTSKII, V.V.; VIII.7, p. 312
OPIAL, Z.; VIII.7, p. 311
OZIRANER, A.S.: 1.6.30, p. 42
PAINLEVE, P.: 111.2.5, p. 100; 111.7, p. 126
PAVEL, N.: VI.5, p. 226: VI.6.12, p. 233: VI.6.16, p. 234:
VIII. 7, p. 312
PEIFFER, K.: 1.6.33, p. 43; 111.2.1, p. 99: IV.8, p. 166:
VI.1.1, p. 203; VI.6.4, p. 229; VI.8, pp. 239, 240:
IX 7, pp 344
PERSIDSKI, K.P.; 1.2.3, p. 7; 1.4.3, p. 14: 1.7.2, p. 46;
V.9.2, p. 195: V.10, p. 198: VI.8, p. 238
PERSIDSKI, S.K.; V.10, pp. 198, 199
PIONTKOVSKII, A.A.: IX.7, p. 343
PLISS, V.A.; VI.6.15, p. 234: VIII.6.5, p. 308; VIII.6.10,
p. 309; VIII.7, p. 312
PLUCHINO, S.; IV.8, p. 167
Author Index
POISSON,
III.1, p. 97
S.D.~
PONTRYAGIN,
POPOV,
391
I.s.10, p. 24
L.S.~
II.s.1, p.
V.M.~
POZHARITSKII,
G.K.~
II.7, p. 96
84~
IV.3.2, p. 1301 IV.8, pp. 166, 167
PTAK, V.1 IX.s, p. 335

REISSIG, R.1 V.9.4, p. 198
RISITO,
IV.8, PP. 166, 1671 IX.7, p. 3441 AII.12, p. 361
C.~
ROSEN, R.1 II.1.11, p. 58

ROUCHE, N.~ I.3.4, p. 131 I.s.8, p. 221 I.6.1s, p. 321
I.6.17, p. 32~ I.6.33, p. 431 I.8, p. 481 II.2.6, p. 66~
II.7, p. 9s~ III.2.1, p. 991 III.7, p. 127~ V.S.S, p. 182~
V.10, pp. 199, 200~ VI.1.1, p. 2031 VII.8, p. 2691
VIII.7, pp.-311, 312~ IX.2.s, p. 318~ IX.7, p. 344
ROUTH, E.J.1 IV.s.1, p. 1381 IV.8, p. 165
ROYDEN,
H.L.~
AI.3.3, p. 351
RUBANOVSKII, V.V.1 IV.8, p. 167

RUMIANTSEV, V.V.1 I.4.4, p. ls~ I.4.s, p. ls~ I.6.32, p. 43;
I.6.33, p. 43~ III.6.13, p. 124~ IV.s.s, p. 141~ IV.8,
pp. 165, 166
RUTKOVSKAYA,
L.D.~
IX.7, p. 343
SALVADORI, L.~ I.6.23, p. 3s~I.6.28, p. 41~ III.s.2, p.

III.7, p. 126~ IV.8, pp. 165, 166~ V.10, p. 200~
VIII.1, p. 270~ VIII.2.1, p. 271~ VIII.2.2, p. 272~
VIII.7, pp.-310, 311
SAMUELSON,
SILJAK,
SILLA,
D.D.~
SANSONE,
SARNO,
SELL,
J.W.~
IX.7, p. 343
II.6.6, p.
94~
II.7, p. 96
V.10, p. 200
J.D.~
G.R.~
SKOWRONSKI,
SLEMROD,
332~
V.9.4, p. 198
G.~
R.~
SCHUUR,
IX.s, p.
III.2.12, p. 1021 III.7, p. 126
L.~
SANDBERG,
IX.s, p. 332
P.A.~
M.~
V.IO, p. 200
VIII.7, p. 312
J.M.~
VIII.7, p. 312
VIII.7, p. 312
113~
392
AUTHOR INDEX
SMETS, H.Bd II.S.2, p. 85

STEPANOV, S. IA.; IV.8, p. 167
STERN, T.E.; VII.6.4, p. 269
SZARSKI, J.; IX.7, p. 342
SZEG5, G.P.; VI.1.1, p. 203; VI.3.3, p. 211; VI.8, p. 239
TAIT, P.G.; II1.S.1, p. 113
TENNERIELLO, C.; VIII.7, p. 310
THOMSON, W.
(Lord Kelvin); 111.5.1, p. 113
VAN CHZHAO-LIN; 111.6.4, p. 118

VERHULST, P.; VII.7.10, p. 266
VINOGRAD, R.E.; 1.2.7, p. 9; VI.8, p. 238
VOLTERRA, V.; VII.7.1, p. 260; VII.7.3, p. 262; VII.7.9, p. 266
VRKOC, 1.; II. 7, p. 96
WALTER, W.; IX.7, p. 342

WARDEN, R.B.; 11.2.9, p. 70
WAZEWSKI, T.; II.3.1, p. 74; 11.3.2, p. 74; V.10, p. 199;
IX.7, p. 342
WINTNER, A.; 111.2.5, p. 100; III.2.8, p. 101
YACUBOVICH, V.A.; 11.5.3, p. 86
YAMABE, H.; V.9.1, p. 195
YORKE, J. A.; V. 10, pp . 199, 200
YOSHIZAWA, T.; 1.2.14, p. 11; 1.7.1, p. 46; V1.lo1, p. 203;
VI.6.10, p. 232; VI.7.6, p. 238; VI.8, p. 239, VIII.7,
p. 311; AI.4.3, p. 353
ZHUKOVSKI, N.E.; 1.4.8, p. 19
ZIEMBA, S.; VIII.7, p. 312
ZUBOV, V.I.; VI.8, p. 239
SUBJECT INDEX
asymptotically almost
periodic, 303
uniformly globally
attractive, 10, 206
asymptotically autonomous
equation, 303
weakly attractive, 28, 227
asymptotically stable,
asymptotic stability, 10,
235
asymptotically stable
with respect to x,
43
equi-asymptotically
stable, 10, 235
globally asymptotically
stable, 10, 236
for comparison equations,

321
Attraction, region of
attraction, 10
uniform region of
attraction, 10
attractor, 204, 227
equi-attractor, 204
equi-weak attractor, 227
uniformly globally
asymptotically stable,
10, 236
uniform attractor, 204
partially asymptotically
stable, 43
uniform weak attractor, 227
uniformly asymptotically
stable, 10, 235
uniformly asymptotically
stable with respect
to x, 43
asymptotic stability by the
first approximation, 31
to-uniform attractor, 204
auxiliary function, 12
family of auxiliary
functions, 271
boundedness, 230
equi-boundedness, 220
equi-ultimate boundedness,
230
asymptotic stability of a
glider, 33
ultimate boundedness, 230
asymptotic stability of a
gyroscope, 67
uniform boundedness, 230
attractive, attractivity, 8,
204
equi-attractive, 8, 204
globally attractive, 10,
206
uniformly attractive, 8,
204
uniform ultimate boundedness,

230
weak ultimate boundedness,
231
class
.5e, 12
column sum dominant, 94

composite system, 327
394
SUBJECT INDEX
concept, comparison
concept, 337
qualitative concept,
202, 208
consequent point, 173
first integral, 130

forces
dissipative forces, 357
gyroscopic forces, 357
Lagrangian forces, 356
decreasing, strictly
decreasing, viii
nonenergic forces, 357
definite function, ix
viscous friction forces, 357
positive definite
function, 13, ix
negative definite
function, ix
definite semi-positive,
ix
construction of
positive definite
function, 150
non-vanishing definite
function, 272
frequency method, 84
gyroscopic stabilization, 116
Hamilton equations, 362
Hamiltonian function, 362
ignorable coordinates, 359
increasing, strictly increasing,
viii
index of a magnetic field, 147
degree of instability 118
ingress point, 173
differential inequalities,
74, 314
instability, 7, 172, 226
Dini derivative
upper right (left)
derivative D+(D-), 346
instability by the first

approximation, 21
lower right (left)

derivative D+(D_), 346
dissipation, complete
dissipation, 357
instability of the betatron, 188
instability of a rigid body, 21

integrals of conjugate momenta,
358
invariant set, invariance, 366
dissipative equation,
dissipative system, 234
invariant set for

semi-flow, 306
domain, 6
pseudo-invariance, 294
egress point, 173
quasi-invariance, 303
energy integral, 358
semi-invariance, 303, 366
~cess
demand, 353
expeller, 171
absolute expeller, 172
Lagrange equations, 356

Lagrange-Dirichlet theorem
(inversion of -), 105
395
SUBJECT INDEX
Liapunov function, 12
semi-flow, 306
limit superior (upper limit),

viii
side-boundary, 171
limit inferior (lower limit),

viii
limit point, 363
solution, maximum solution,

minimum solution, 315
stable, stability, 7, 204
Lagrange stability, 5
negative limit point,

363
orbital stability, 5, 201
positive limit point,

363
partial stability, stable

with respect to x, 15
limit set, 363

negative limit set,
363
positive limit set,

363 .
Lipschitzian, 47
stable at to' 3
stable under persistent
disturbances, total
stability, 81
stable with respect to some
function, 238
merostatic motion, 361
stability for comparison

equation, 320
mixed potential, 254
uniform stability, 7, 204
monotonic function, viii
uniform stability with

respect to x, 14
orbit, positive (semi-)

orbit, 363
stability of the betatron, 143
Painleve integral, 358
stability of composite systems,

327
path, 20
positive equadratic forms,

ix
stability of a glider, 17
stability of a rigid body, 16
quasi-monotonic function,
314
stability of stationary motions,

138
retraction, retract, 178
stationary motions, 361
Routh's equations, 360
generalized stationary
motion, 361
Routh's theorem
(inversion of -), 125
transfer system, 327
sector, 171
translation operator, 305
absolute sector, 171

semi-continuous, viii
unstable, 7, 226
unstable at to' 3
396
SUBJECT INDEX
weak limit, to tend weakly, 248

word, 208
well formed word, 210

Stability Theory of Lyapunov

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Stability Theory of Lyapunov

Încărcat de

Drepturi de autor:

Formate disponibile

N.

Applied Mathematical Sciences 22

Springer-Verlag New York Heidelberg Berlin

Institut de Mathematique Pure et Appliquee

AMS Subject Classifications:

34D20, 93D05 (Primary), 34Dxx,

Library of Congress Cataloging in Publication Data

All rights reserved.

ISBN 978-1-4684-9362-7 (eBook)

The names appear-

But the contributions of others were also

The idea of writing this book originated in a seminar

Two years later, a first draft was completed.

However, it was unsatisfactory mainly because it was exce~sively

abstract and lacked examples.

It was then decided

to write it again, taking advantage of -some remarks of the

text is this second version.

our efforts on this method, not because we underrate those

excellent books exist concerning the other methods, as for

functions appear to be so difficult to construct.

On the other hand, too many of those involved in

applications are unaware of the useful theorems or are

present state of the most useful Parts of the theory, and to

They should normally be read

Almost every other chapter can be studied without

further prerequisite, except that some definitions or

The whole of Chapter VI is used in Section IX.6.

We are also grateful to H. Everard, S. Spinacci and

pleasure to acknowledge the financial support of this work by

Louvain-1a-Neuve, October, 1975

SOME NOTATIONS AND DEFINITIONS

This books requires a familiarity with some basic

Let us however point out the following

!JR, the set of real numbers,

a is a strictly positive real number,

[a,b], closed interval,

Ilxll, norm of point x in ~,

E!Jfn , II x II < d, open ball with center at the origin

< d, open ball with center

a.nd radius e: > 0,

' time derivative of the function

' jacobian matrix of the function

see definition p. 12.

f: !Jfn ~ !Jfm, x ~ f(x),

SOME NOTATIONS AND DEFINITIONS

'ttx, universal quantifier; read "for all

3x, existential quantifier; read "for some

For g.eneral concepts on differential equations which

The following definitions might be useful.

'ttx E A, 'tty E A, x < y

f (x) < f (y) ,

'ttx E A, 'tty EA, x < y

'ttx E A, 'tty E A, x < y

the extended closure of

superior (upper limit) of

lim inf f(x)

Then the limit

sup{inf{f(x): x E B(a,o), x ;, an E !:if.

a EA, the function

Similarly the limit inferior (lower limit) of

lim inf f(x)

is said to be lower semi~

SOME NOTATIONS AND DEFINITIONS