Stability

CLASSICS IN MATHEMATICS
Nam Parshad Bhatia Giorgio P. Szego
Stability Theory
of Dynamical
Systems
Springer
Classics in Mathematics
N.P. Bhatia G.P. Szego
Stability Theory
of Dynamical Systems
Springer
Berlin
Heidelberg
New York
Barcelona
Hong Kong
London
Milan
Paris
Tokyo
N. P. Bhatia G. P. SzegO
Stability Theory
of Dynamical Systems
Springer-Verlag Berlin Heidelberg New York 1970
Professor Nam Parshad Bhatia

University of Maryland
Division of Mathematics, Baltimore
and Institute for Fluid Dynamics
and Applied Mathematics, College Park
Maryland, U.S.A.
Professor George Philip Szego

University of Milan
Milan, Italy
Geschaftsfiihrende H erausgeber:
Professor Dr. B. Eckmann

Eidgenossische Technische Hochschule Zurich
Professor Dr. B. L. van der Waerden

Mathematisches Institut der Universitat Zurich
This work is subject to copyright. All rights are reserved, whether the whole or part of the
material is concerned, specifically those of translation, reprinting, re-use of illustrations, broadcasting, reproduction by photocopying machine or similar means, and storage in data banks.
Under 54 of the German Copyright Law where copies are made for other than private use, a fee
is payable to the publisher, the amount of the fee to be determined by agreement with the publisher.
@ by Springer-Verlag, Berlin Heidelberg 1970. Printed in Germany
Library of Congress Catalog Card Number 70-126892
Title No. 5144
To Sushiela and Emilia
Preface
This book contains a systematic exposition of the elements of the
theory of dynamical systems in metric spaces with emphasis on the stability theory and its application and extension for ordinary autonomous
differential equations.
In our opinion, the book should serve as a suitable text for courses
and seminars in the theory of dynamical systems at the advanced undergraduate and beginning graduate level, in mathematics, physics and
engineering.
It was never our intention to write a treatise containing all known
results on the subject; but we have endeavored to include most of the
important new results and developments of the past 20 years. The
extensive bibliography at the end should enhance the usefulness of the
book to those interested in the further exploration of the subject.
Students should have completed an elementary course in ordinary
differential equations and have some knowledge of metric space theory,
which is usually covered in undergraduate courses in analysis and topology.
Each author strongly feels that any mistakes left in the book are
attributable to the other author, but each would appreciate receiving any
comments from the scientific community.
We are obliged to Professor AARON STRAUSS for reading the entire
typescript and pointing out several corrections. We would also like to
thank Doctors FLORENCIO CASTILLO, LAWRENCE FRANKLIN, CESI.AW
OLECH and GIULIO TRECCANI for help in proofreading the galleys.
June 1970
N. P. BHATIA G. P. SzEGO
Contents
Notation . . .
Introduction
I. D y n am i c a 1 Sy st e m s .
1. Definition and Related Notation . .

2. Examples of Dynamical Systems . . .
Notes and References . . . .
II. Elementary Concepts
1. Invariant Sets and Trajectories. .
2. Critical Points and Periodic Points . .
3. Trajectory Closures and Limit Sets
4. The First Prolongation and the Prolongational Limit Set . .
Notes and References . . . .
. . . . . . . .
III. R e c u r s i v e C o n c e p t s . .
1. Definition of Recursiveness
. . . . . . . . . .
2. Poisson Stable and Non-wandering Points .
. . . .
3. Minimal Sets and Recurrent Points . . .
4. Lagrange Stability and Existence of Minimal Sets
Notes and References . . . . . . . . .
IV. D i s p e r s i v e C o n c e pt s
1. Unstable and Dispersive Dynamical Systems
2. Parallelizable Dynamical Systems
Notes and References . . . . . . .
V. Stability Theory.
1. Stability and Attraction for Compact Sets . . . . . . . . . .
2. Liapunov Functions: Characterization of Asymptotic Stability .
3. Topological Properties of Regions of Attraction
4. Stability and Asymptotic Stability of Closed Sets .
5. Relative Stability Properties . . . . . . . . . . . . . .
6. Stability of a Motion and Almost Periodic Motions
Notes and References . . . . . . . . .
V. Flow near a Compact Invariant Set . . .
1. Description of Flow near a Comi:act Invariant Set
2. Flow near a Compact Invariant Set (Continued)
Notes and References . . . . . . . . . . . . . . .
x
1
5
6
10
12
12
15
19
24
30
31
31
31
36
41
42
43
43
48
55
56
56
66
79
84
99
106
111
. 114
. 114
. 116
. 117
IX
Contents
VII. Higher Prolongations
. 119
1. Definition of Higher Prolongations .

2. Absolute Stability . . .
:3. Generalized Recurrence
Notes and References . . .
VIII. @1 - Li a pun o v
Equations
Functions for
. 120
. 124
. . . 129
. . 133
Ordinary
Differential
. 134
1. Introduction. . . . . . . . . . . .
.
2. Preliminary Definitions and Properties
. . .
3. Local Theorems . . . . . . . . . .
.
4. Extension Theorems . . . . . . . . .
.......
5. The Structure of Liapunov Functions .
. ....
6. Theorems Requiring Semidefinite Derivatives . . . .
.
7. On the Use of Higher Derivatives of a Liapunov Function . .
Notes and References . . . . . . . . . . . . . . . .
.
IX. Non-continuous Liapunov Functions

Differential Equations.
for
References. . .
Author Index
Subject Index . .
145
150
156
160
162
Ordinary
. . . 166
Introduction. . . . . . . . . .
A Characterization of Weak Attracto:-s
. . . .
. . .
Piecewise Differentiable Liapunov Functions
. .
Local Results . . . . . . . . . . . . . .
.
Extension Theorems . . . . . . . . . . .
. . . . .
Non-continuous Liapunov Functions on the Region of weak Attraction . . . . . .
. . . . . . . . .
. . . . .
.
Notes and References . . . . . . . . . . . . . . . . . . . . . .
1.
2.
3.
4.
5.
6.
134
136
138
166
169
172
176
177
179
183
185
. 221
. . 223
Notation
Set Theoretic Notation
Throughout the book standard set theoretic notations are used. Thus(, V, (\,
stand for set inclusion, set union and set intersections, respectively. For a given set
M, fJM, $ M, M, ((2. (M) denote the boundary, interior, closure, and complement of
the set M, respectively. For given sets A, B, the set A-Bis the set difference.
Other standardly used set theoretic notations are:

X
a metric space with metric e.
2X
family of all subsets of X.
R
set of real numbers.
Rn
real n-dimensional euclidean space.
R+
set of non-negative reals.
Rset of non-positive reals.
the empty set.
0
absolute value of a real number.
II
euclidean distance norm.
II II
the scalar product of vectors x, y in Rn.
<x,y)
for given xE X and e > 0 is the open ball of radius X > 0 centered
S (x, lX)
at x, i.e., the set {y: (} (x, y) < X}.
the set {y: (} (y, M) < X}, where MC X and X > 0 are given.
S(M, X)
the closed ball of radius X > 0 centered at x, i.e., the set {y: (} (x, y)
S [x, X]
< lX}.
S [M, X]
H (x, X)
H (1Vf, X)
{xn} or {xn}
Xn~X
((2.1
((2.2
the set {y: (} (y, M) ~ X}.

the spherical hypersurface of radius X > 0 centered at x, i.e., the set
{y: (} (x, y) = lX}.
the set {y: (} (y, M) = X}.
a sequence.
sequence {xn} converges to x.
family of continuously differentiable functions.
family of twice continuously differentiable functions.
Notation Pertaining to Dynamical Systems

(X, H, n)
n
n'
n.~
y(x)
y+(x) (y-(x))
dynamical system on a space X (I, 1.1, p. 5).

phase map of a given dynamical system (I, 1.1, p. 5).
transition corresponding to a given t E R (p. 6).
motion through x (p. 6).
trajectory through ..-r (II, 1.9, p. 14).
positive (negative) semi-trajectory through x (II, 1.9, p. 14).
Notation
A+ (x) (A-(x))
D+(x) (D-(x))
J+(x) (j-(x))
Dt (x) (D:; (x))
]~ (x) (]; (x))
Dt(M}
D+(M, U)
Aw(M)
A (M)
A,l(M)
<R,
IJ)
J.
XI
positive (negative) limit set of x (II, 3.1, p. 19).

positive (negative) prolongation of x (II, 4.1, p. 24).
positive (negative) prolongational limit set of x (II, 4.1, p. 25).
for given ordinal number cX, the cX-th positive (negative) prolongation
of x (VII, 1.12, p. 123).
for given ordinal number cX, the cx-th positive (negative) prolongational limit set of x (VII, 3.1, p. 129).
uniform positive prolongation of a set M (VII, 2.11, p. 128).
for given M, U in X, the positive prolongation of M, relative to U
(II, 4.10, p. 29).
region of weak attraction of M c X (V, 1.1, p. 56).
region of attraction of MC X (V, 1.1, p. 56).
region of uniform attraction of M ( X (V, 1.1, p. 56).
the generalized recurrent set (VII, 3.6, p. 131).
operator used in the definition of higher prolongation (VII, 1.1,
p. 120).
operator used in the definition of higher prolongation (VII, 1.1,
p. 120).
Introduction
The theory of dynamical systems may be said to have begun as a
special topic in the theory of ordinary differential equations with the
pioneering work of HENRI PoINCARE in the late 19th century. PoINCARE,
followed by IVAR BENDIXSON, studied topological properties of solutions
of autonomous ordinary differential equations in the plane. The Poincare-Bendixson theory is now a standard topic of discussion in courses
on ordinary differential equations, and is adequately covered in all its
details in the books of, say, CODDINGTON and LEVINSON [1], LEFSCHETZ
[1], HARTMAN [1], SANSONE and CONTI [lJ, and NEMYTSKII and STEPANOV [1]. Of these, HARTMAN'S book contains the most detailed and recent
exposition.
Almost simultaneously with PoINCARE, A. M. LIAPUNOV developed
his theory of stability of a motion (solution) for a system of n first order
ordinary differential equations. He defined in a precise fonn the concept
of stability, asymptotic stability, and instability; and gave a "method"
(the second or direct method of Liapunov).for the analysis of the stability
properties of a given solution of an ordinary differential equation. Both
his definition and his "n1ethod" characterize, in a strictly local setting,
the stability properties of a solution of the differential equation. As such,
the Liapunov theory is strikingly different from the Poincare theory,
in which, on the contrary, the study of the global properties of differen-\
tial equations in the plane play a 1najor role.
One of the main aspects of the Poincare theory is the introduction of
the concept of a trajectory, i.e., a curve in the x, x plane; parametrized
by the time variable t, which can be found by eliminating the variable t
from the given equations, thus reducing it to a first order differential
equation connecting x and x. In this way, Po1NCARE set up a convenient
geometric framework in which to study qualitative behavior of planar
differential equations. PoINCARE was not interested in the integration of
particular types of equations, but in classifying all possible behaviors
1
Bhatia/Szego, Stability Theory
Introduction
of the class of all second order differential equations. By introducing

this concept of trajectory, PoINCARE was able to formulate and solve, as
topological problems, problems in the theory of differential equations.
In the above fashion PoINCARE paved the way for the formulation
of the abstract notion of a dynamical system, which can be essentially
attributed to A.A.MARKOV and H. \VHITNEY. These two authors separately noticed that one could study the qualitative theory of families
of curves (trajectories) in a suitable space X, provided that these families
are somehow restricted in their possible behavior, e.g., if they are
defined, as having been generated by a general one-parameter topological transformation group acting on X.
Great impetus to the theory of dynamical systems came from the
\York of G.D.BIRKHOFF, who may truly be considered as the founder
of the theory. His celebrated 1927 monograph on Dynamical Systems
(BIRKHOFF [1]) is the basis of much of the research which came in the
1930's and 1940's and even today it is not outdated. BIRKHOFF established
the two main streams of work on the theory of dynamical systems, namely, the topological theory and the ergodic theory.
In 1947, V. V.NEMYTSKII and V. V. STEPANOV [1] completed their
"Qualitative Theory of Differential Equations" which to this day has
served as a standard reference for all the major development in the
theory of dynamical systems up to the middle 1940's. In 1949, NEMYTSKII [10] wrote a survey paper on the topological problems in the theory
of dynamical systems, which sums up almost all the research into the
topological theory to the end of 1940's.
During the 1950's a relatively large effort went into the generalization
of the concept of a dynamical system to topological transformation
groups. Thus in 1955 the book of W.H.GoTTSCHALK and G.A.HEDLUND
[1] appeared, and a large body of research has appeared since in print.
In this connection the work of R. ELLIS, H. FURSTENBERG, J. AUSLANDER,
H.CHu, F.HAHN, S. KAKUTANI, besides GOTTSCHALK and HEDLUND, is
noteworthy. On the other hand problen1s on structural stability in ordinary differential equations have led to the efforts at introducing the
concepts and methods of differential topology and the theories of
S. SMALE, D. V. ANosov, J. MosER, lVI. PEIXOTO, and L. MARKUS. In this
connection ANosov's monograph [1] is noteworthy.
More recently the basic theory has been extended by bringing in, the
stability problems a la Liapunov, characteristically absent in earlier
works on dynamical systems and topological transformation groups. In
this connection, TARO URA's work, in particular his theory o.f prolongations, and its connections with stability has clearly shown that a significant portion of stability theory is topological in nature and hence belongs
to the main stream of the theory of dynamical systems. An attempt in
Introduction
bringing in the direct method of Liapunov was made by V.I.ZuBov [l].

However, ZuBov mainly carried over to flows in metric spaces, results
and methods previously known in differential equations without attempting to develop an independent theory.
The present volume was thus conceived to present in an easy and
readable fashion the recent research in the theory of dynamical systems
on metric spaces, and especially the stability theory together with its
concrete applications to the theory of differential equations.
This book does not introduce several interesting areas of modern day
research, such as the theory of structural stability (which requires
son1e knowledge of differential topology), ergodic theory, and the general
theory of topological transformation groups. To keep the presentation
at a level easily accessible to undergraduate students in their junior
or senior years (who have had some exposure to metric spaces and
differential equations), we have not introduced local dynamical systems,
although most results are true for such systems. Local semi-dynamical
systems (N.P.BHATIA and O.HAJEK [1]), flows without uniqueness
(G. P. SzEGO and G. TRECCANI [1]), are other major areas of development
today. The present book should be helpful to all those desiring to
work in any one of the above mentioned fields of study and research,
which are not covered in this volume.
As to material covered in this volume, Chapters I-VII contain the
basic theory of dynamical systems in metric spaces and Chapters VIII
and IX contain applications and extensions of the stability theory
(Chapter V) to dynamical systems defined by ordinary differential equations. Specifically, Chapter I contains the definition of a dynamical
system and some examples to indicate various fields of application.
Chapter II contains elementary notions which remain invariant under
certain topological transformations of dynamical systems. Chapter III
deals mainly with minimal sets and their structure. Chapter IV is devoted
to the study of dispersive and parallelizable dynamical systems and concludes the part of the book devoted to the basic theory. Chapter V develops the main theme of the book, i.e., the stability and attraction theory.
The theory presented here differs rather strongly from the one developed
by ZuBov, being essentially based on the concept of weak attraction
(absent in ZuBov's work). Chapter VI is devoted to a more specific
problem: the classification of flows near a compact invariant set. Some
results are given, but many problems in this are still open. Chapter VII
contains the theory of higher prolongations originated by T. URA with
applications to absolute stability and generalized recurrence. Chapter VIII
deals with the geometrical theory of stability for ordinary autonomous
differential equations including various extensions of Liapunov's direct
method. Chapter IX is again devoted to a more specific problem of
1*
Introduction
characterizing stability and attraction concepts via non-continuous Liapunov functions; these are concepts like the weak attractor, which are
not characterizable by continuous Liapunov functions._
Regarding the formal structure of the book, each chapter is divided
into sections, followed by an un-numbered section of notes and references.
In each section, individual items (Definitions, Theorems, etc.) are
numbered consecutively. Each item may be subdivided into consecutively
numbered subitems. References to the same chapter do not mention
the chapter number. Thus, for example, reference 2.5.3 indicates, section 2, item 5, subitem 3 of the same chapter. References to other chapters contain an indication of the chapter number. Thus II, 3.17 denotes
item 17 of section 3 in Chapter II. References to the bibliography are
given by the author's name followed, if necessary, by an item number
between brackets.
Chapter I
Dynamical Systems
In this chapter we introduce the definition of a dynamical system or
what is also called a continuous flow. Several general examples are given
to motivate and prepare the reader for the study of the theory of dynamical systems. Throughout the book the symbol X denotes a metric
space with metric e and R stands for the set of real numbers.
1. Definition and Related Notation

1.1
Definition. A dynamical system on X is the triplet (X, R, n;),
where n; is a 1nap from the product space XX R into the space X satis(~dng the following axioms :
l.1.1
n (x, 0)
=x
n(n(x, t1), t2)

1.1.2
(group axiom),
1.1.3
for every x EX (identity axiom),
== n(x, t1 + t2)
for every x EX and t1 , t2 in R
n is continuous (continuity axiom).
Given a dynamical system on X, the space X and the map n are

respectively called the phase space and the phase map (of the dynamical
system). Unless otherwise stated a dynamical system on X is alw~ys
assumed given.
In the sequel we shall generally delete the symbol n;. Thus the image
n;(x, t) of a point (x, t) in XxR will be written simply as xt. The identity
and the group axioms then read
1.1.4
x0= x
for every x EX,
1.1.5
xt1 (t2) = x(t1
+ t2 )
for every x EX and t1 , t2 in R.
In line with this notation, if MC X and A C R, then MA is the set

{xt: x EM and t EA}. If either .Zl! or A is a singleton, i.e., M = {x} or
I. Dynamical Systems
A = {t}, then we simply write xA and Mt for {x} A and M {t}, respectively. For any x EX, the set xR is called the trajectory through x (see
II, 1.9).
The phase map determines two other maps when one of the variables
x or t is fixed. Thus for fixed t E R, the map nt: X ~ X defined by
nt (x) = xt is called a transition, and for a fixed x E X, the map nx:
R ~ X defined by nx(t) = xt is called a motion (through x). Note that nx
maps R onto xR.
The following theorem expresses an important property of the transitions.
1.2
itself.
Theorem. For each tE R, nt is a homeomorphism of X onto
Proof. For any t ER the transition 11/ is continuous as n is such.

To see that nt is one-to-one and onto observe that if yt = zt, then
y = z follows from y = y 0 = y(t - t) = yt(-t) = zt(-t) = z(t - t) =
z 0 = z. Again, if y E X, then nt (x) = y for x = y (-t) is easily verified.
Finally to see that nt has a continuous inverse we need only show that
the transition n-t is the inverse of nt. To see this note that for any two
transitions nt and n 8 , the composition nt o n 8 is the transition 'JT}+s,
because for any x E X,
n' o n 8 (x)
n'(n8 (x)) = nt(xs) = xs(t) = x(t
+ s) =
nt+ 8 (x).
Note also that the transition n is the identity, since for any x EX,
n (x) = x 0 = x. Since now n-t o n' = n'-t = n, the transition n-t
is the inverse of n'.
1.3
Exercises.
1.3.1
Show that the transitions nt, t ER, form a commutative group
with the group operation being the composition of transitions.
1.3.2
For any xE X and [a, b] CR, the set x[a, b] is compact and
connected.
2. Examples of Dynamical Systems

2.1
Ordinary Autonomous Differential Systems. Consider the autonomous differential system
2.1.1
~; =
x = t (x),
where /: Rn~ Rn (Rn is the real n-dimensional euclidean space) is continuous and moreover assume that for each x E Rn a unique solution
<p (t, x) exists which is defined on R and satisfies <p (0, x)
x. Then it is
well known (see for example CODDINGTON and LEVINSON [1], chapters 1
and 2) that the uniqueness of solutions implies
and considered as a function from RxRn into Rn, <pis continuous in its
arguments (section 4, chapter 2 in CODDINGTON and LEVINSON [1]). It
is clear that the map n : Rn X R --+ Rn such that n (x, t) = <p (t, x) defines
a dynamical system on Rn. We remark that the conditions on solutions
of 2.1.1, as required above, are obtained, for example, if the function f
satisfies a global Lipschitz condition, i.e., there is a positive number k
such that
2.1.3
11/(x) -
/(y) JI< k
llx
-y!I
for all x, v in Rn.
2.2
Ordinary Autonomous Differential Systems (Continued). To
illustrate that the theory of dynamical systems as defined in this chapter
is applicable to a much larger class of ordinary autonomous differential
systems we consider a system
2.2.1
~;=X=f(x),
xER"
where/: D ~Rn is a continuous function on some open set D (Rn, and

for each x ED, 2.2.1 has a unique solution <p (t, x), <p (0, x) = x defined
on a maximal interval (ax, bx), - oo < ax < 0 < bx < + oo. For each
x ED define y+ (x) = {<p (t, x): 0 :St< bx}, and y-(x) = {<p (t, x):
ax < t < O}. y+ (x) and y-(x) are respectively called the positive and
negative trajectory through the point x ED. We will show that to each
syste1n 2.2.1, there corresponds a system
2.2.2
dx
dt
= x. =
g (x ) ,
x E Rn ,
where g: D--+ Rn, such that 2.2.2 defines a dynamical system on D with
the property that for each x ED the systems 2.2.1 and 2.2.2 have the
same positive and the same negative trajectories. Thus in general it is
sufficient to consider 2. 2. 2 instead of 2. 2.1.
If D =Rn, then given 2.2.1, we set
2.2.3
dx
dt
= x
g (x)
/(x)
= 1 + flj(x)JI
where !! II is the euclidean-distance norm. If D =f= Rn, then oD =f= 0 and

is closed. In this case, given 2.2.1, we set
2.2.4
dx
dt
==
f (x)
(] (x, oD)
g (x) == 1 + 11/(x)ft 1 + (] (x, oD)
'
where e(x, oD) = inf{llx -yli: yE oD}. One can now show that 2.2.3
or 2.2.4 defines a dynamical system on Rn or D, whenever the system
2.2.1 has unique solutions. Moreover, 2.2.3 and 2.2.4 have the same positive and negative trajectories as 2.2.1.
Ordinary Autonomous Differential Systems Containing Para2.3
meters. Consider the differential system
2.3.1
x= f(x, ),
xE Rn,
where is a parameter in a metric space M, and f: Rn X M -+ Rn is

continuous. Assume that for each (x, ) E RnxM there is a unique solution <p (t, (x, )) defined on R and satisfying <p (0, (x, )) = x. Set X =
Rnxl\II, and define n: XXR-+ X by setting n((x, ), t) = (<p (t, (x, )), ).
Then it is easily verified that n defines a dynamical system on Rn X M
=--=
x.
2.4
Ordinary Non-autonomous Differential Systems. Consider the
differential system
2.4.1
x=
/(t, x),
x E Rn, t ER,
where f: RX Rn -+ Rn is continuous. Assume that for each (t0 , x0) E RX Rn,

2.4.1 possesses a unique solution rp (t, t0 , x0 ), <p (t0 , t0 , x0) = x0 , defined for
all t E R. Consider the map n: XX R -+ X, where X = RX Rn, defined
by n ((t, x), s) = (s + t, <p (s + t, t, x)). Then n defines a dynamical system
on X. This dynamical system in fact corresponds to the equivalent
autonomous differential system in Rn+t
2.4.2
y = g(y),
where y = (t, x), and g (y)
(1, I (y)).
The Bebutov Dynamical System. Consider the set X of all continuous functions /: R-+ Rn. Now define e: XxX-+ [O, + oo) as
follows: For any integer m, let Im= [-m, m], and for/, gE. X set
2.5
g) = max {II f (t) - g (t)

<Jm (f, g)
2 5 2
l?m (/' g) = 1 + <Jm (f, g) '
and finally
2. 5.1
<1m (/,
2.5.3
e(f, g) = 2
00
11 : t E Im},
2m l?m (/, g).
m=1
One may now verify that e defines a metric on X (i.e., X is a metric

space with metric e given by 2.5.3) and that with this metric the space
Xis complete. One can verify in fact that if {In} is any sequence in X,
then e(/n' /) ~ 0 as n ~ oo if and only if {/n (t)} converges to f (t) uniformly on every compact subset of R.
On the metric space X defined above we define a dynamical system

as follows: n: XX R ~ X is given by
1'& (I, t) = g
where g is given by g(s) = l(t + s) for all sE R. To verify that n is a
dynamical system, note first that the identity axiom is trivially satisfied.
To see the group axiom, let n(n(I, t), s) = h and n(/, t) = g. Then
h(-r) = g(s + -r) = l(t + s + -r) by definition. Hence h = n(f, t + s).
Finally, to see that n is continuous, let In be a sequence in X, {tn} a
sequence in R, such that In~ f and tn ~ t. To show is :rc(fn, tn) ~ n(I, t).
Let KC R be compact. Let n(ln, tn) = gn, and :rc(f, t) = g. V./e must
show that given e > 0 there is an integer N such that 11 gn (-r) - g (-r) 11 < e
for n > N and TE K. Now gn(-r) - g(-r) ~ ln(tn + -r) - l(t + -r) =
ln(tn + 'l') - in(t + T) + fn(t + T) -l(t + 'l'). Thus
lfgn(T) - g(-r) If< !lln(tn + T) - ln(t + T) I!+ llln(t + T) - l(t + T) If

Note now that fn(t) ~ /(t) uniformly on any compact subset of R so
that the sequence {In (t)} is equi-continuous. The result is now easily seen
from the last inequality. The dynamical system n defined above has
many interesting properties. This has been included to show that the
theories expounded in this book are applicable to situations which may
not necessarily arise out of ordinary differential equations. However,
we shall not study such examples in any detail.
2.6
Dynamical Systems on a Torus. Consider a differential system
in the plane
2. 6.1
dx
dt
= I (x),
x E R2 ,
where I: R 2 ~ R2 is continuous and the conditions of example 2.1 are

satisfied. Set x = (<p, 0), I= (111 / 2) and assume that the functions 11
and 12 are periodic of period 1 in each of the variables 0 and q;. Thus for
all real <p and 0
2.6.2
li(<p, 0) = li(<p
+ 1, 0) == fi(<p, 0 + 1),
i = 1, 2.
It is customary to think of the equation 2.6.1 satisfying condition 2.6.2

as representing a dynan1ical system on the torus, instead of in the plane
R 2 the immediate advantage being the fact that the torus is a compact
2-manifold. To see this clearly one takes as a convenient representation
of the torus, the square
{(<p, 0): 0 < <p < 1, 0 < 0 < 1}
in which the pairs of opposite sides <p = 0, <p = 1 and 0 = 0, 0 = 1 are
identified. This yields, in particular, that the vertices (0, 0), (0, 1),
(1, 0), (1, 1) are all identified. The plane R 2 is now projected on the torus
by identifying any point (rp, 0) in the plane with a point (p, 0) of the
10
square for which <p - p, and 0 - 0 are integers. One usually expresses
this by <p = ~ (mod 1), and 0 = 0 (mod 1). The projection of trajectories
of 2.6.1 in the plane onto the torus then yields the trajectories of a dynamical system on the torus. Thus 2.6.1 is said to define a dynamical
system on the torus when condition 2.6.2 holds.
Notes and References

Section 1. The introduction of the definition of a dynamical system cannot be
attributed to any one person. Interesting historical remarks on the development
of the basic concepts of dynamical systems can be found in the papers by V. V.NEMYTSKII [10] and by G.D.BIRKHOFF [2, vol. 2, p. 710].
The first abstract definition of a dynamical system can be found in the works
of A.A.MARKOV [1] and of H.WHITNEY [1, II]. Most concepts were introduced by
H. Po1NCARE [1] and by G.D. BIRKHOFF in the framework of the theory of dynamical systems defined by ordinary differential equations.
The theory of dynamical systems received new impetus in the recent years by
the publication of the book by V.V.NEMYTSKII and V.V.STEPANov.
Generalizations to topological transformation groups have appeared since then
in the monographs of G.T.WHYBURN, W.H.GoTTSCHALK and G.A.HEDLUND,
D.MoNTGOMERY and L.ZIPPIN, L.AusLANDER, L.W.GREEN and F.HAHN.
In this volume we devote ourselves to dynamical systems in a metric space.
We do this for simplicity of the proofs. Indeed most of the results that we present
can be proved with due care in more general topological spaces.
Various generalizations of the concepts of a dynamical system have been studied
in the literature, among them we mention the discrete dynamical systems (also
called cascades), i.e., the triple (X, I, n) where I is the set of integers (see for
instance W.H.GoTTSCHALK and G.A.HEDLUND [1]}, the local dynamical systems
(X, Jx, n), which are defined only on intervals Jx CR, OE Jx (see for instance
T.URA [4] or O.HAJEK [1]).
The definition of local dynamical systems may be considered to have been
motivated by ordinary differential equations satisfying conditions of existence and
uniqueness of solutions but not necessarily those of global extendability. Example
2.2 shows that local systems defined by ordinary differential equations can be
reparametrized to obtain global systems. It is however not known if the same is
possible for local systems not defined by ordinary differential equations (see
M.I.GRABAR [3]).
A further generalization of dynamical systems is the concept of dynamical
systems without uniqueness or semigroups of multivalued mappings: (X, R,f),
where f: Xx R ~ X is multivalued and satisfies certain axioms. These have been
investigated by E.A.BARBASHIN, B.M.BuDAK, by M.I.MINKEVITCH, and by
G.P.SzEGO and G.TRECCANI [1] among others.
Semigroups of multivalued mappings defined only on the positive (or negative)
reals (X, R+ ,f) have been investigated by I. U. BRoNSHTEIN [2] and by E. RoxIN
[3, 5, 6]. A complete advanced theory is presented in the recent monograph by
N.P.BHATIA and O.HAJEK [1] where the local semi-dynamical systems are investigated. These generalizations of dynamical systems are usually motivated by the
need of finding abstract frameworks for the investigation of systems whose behavior
does not satisfy the axioms 1.1. For instance ordinary differential equations without global extendability of solutions may define a local dynamical system, ordinary
11
differential equations without uniqueness may define a semigroup of multivalued

mappings (Rn, R,f), integral equations (R. K. MILLER and G. SELL [1]) define, in
some cases, semi-dynamical systems (X, R+, n) and functional differential equations
are used by N.P.BHATIA and O.HAJEK [1] as example of local semi-dynamical
systems.
A natural extension of the theory of dynamical systems and in particular of
semigroups of multivalued mappings is the theory of dynamical polysystems,
which deals with one-parameter families of dynamical systems (see D. BusHAW
[1, 3]).
A complete presentation of special results for dynamical systems on the plane
is due to 0. HAJEK. In his recent monograph [1] he was able to extend most results
of the Poincare-Bendixson theory for planar second order differential equations to
dynamical systems on the plane.
Lastly, we want to point out that in all the theories mentioned above the
mapping n is usually assumed to be continuous, but not continuously differentiable.
There exist works (see for instance S. SMALE [2, 3, 5, 6, 7], J. MosER and D. V. ANosov [1]) in which n is assumed to be continuously differentiable. The type of problems
solved and the techniques used in these later works are considerably different from
the ones used in this book, and close to the works on ergodic problems.
Section 2. 2.1 In order for the differential equation 2.1.1 to define a dynamical
system it must satisfy conditions which insure existence, uniqueness and extendability of its solutions. These conditions can be found in most books on differential
equations (N.P.BHATIA and G.P.SzEGo [1, section 3.1], P.HARTMAN [1, chapter
1-2], E.CoDDINGTON and N.LEVINSON [1, chapter 1-2] and V.V.NEMYTSKII
and V. V. STEPANOV [1, chapter 1]. Notice that beside the problems of when a
differential equation defines a dynamical system, also the more important inverse
problem has been discussed, i.e., what are the conditions for a flow to be represented as the system of solutions of differential equations (M. I. GRABAR [1, 2, 3]).
2.3 The differential equation 2.3.1 when is a function belonging to a certain set
defines a control system, i.e., a particular realization of a dynamical polysystem
(D.BUSHAW [1, 2]).
2.4 In addition to the one given here there exists another interesting axiomization
of the non-autonomous differential equation 2.4.1 due to G. SELL [5]. He defines
first of all a mapping n* : C x R ~ C, where C is the space of continuous functions
defined by n* (t, -r) = f(x, t + -r) and proves that n* defines a dynamical system on
C. Consider next the set F = {f(t + -r): -rE R} and the metric space X = Fx R
with a suitable metric, where f is a function which ensures uniqueness of solutions
of equation 2.4.1. Then the mapping n: XxR~X defined by n((x,f),-r) =
(<p(x,f,-r), f(t + -r)), here <p(x,f,T) denotes the solution of equation 2.4.1 with
<p (x, f, 0) = x, defines a dynamical system.
2.5 The dynamical system and the space of continuous functions was investigated
by M.V.BEBUTOV [3, 4, 5].
Other models of dynamical systems of particular type are used in the so-called
symbolic dynamics (see M.MORSE [4]), in the theory of Hamiltonian systems
(see G.BIRKHOFF [1], J.MosER, and C.CONLEY, where additional references can
be found), and in the theory of linear dynamical systems (see R. E. KALMAN [1]
and F.H.SHOLOHOVICH [2]).
Chapter II
Elementary Concepts
In this chapter we introduce certain fundamental maps Q: X ~ 2x
(2x is the set of all subsets of X) corresponding to a given dynamical
system (X, R, n). The basic topological properties such as closedness,
and connectedness of the image sets Q(x) are studied. The maps introduced are such that their images have in gener.al certain invariance
properties, so that we start with the notion of invariance of a set in
section 1. A dynamical system (X, R, n) is assumed given. R.+ and Rwill denote respectively the set of non-negative and non-positive real
numbers.
1. Invariant Sets and Trajectories

1.1
Definition. A set M C X is called invariant whenever
1.1.1
xtE M
for all xE Mand tE R.
It is called positively invariant whenever 1.1.1 holds with R replaced

by R+, and is called negatively invariant if the same holds with R replaced
by R-. Note that 1.1.1 is equivalent to MR = M.
The following theorems exhibit some properties of such sets.
1.2
Theorem. Let {Mi} be a collection of positively invariant, negatively invariant, or invariant subsets of X. Then their intersection and
their union have the same property.
Proof. To fix our ideas let the sets kl;, be positively invariant. Let
M' = V Mi, and M" = f\ M;,. For any x EM', we have x EM.;,
for some i. Thus xt E M.;, for all t E R+ since Mi is positively invariant.
Hence indeed xt E M' for all t E R+ as M' ) Mi. M' is therefore
positively invariant. Now let xE M". Then xE M;, for every i and
by positive invariance of eachM.;,, xt EM.;, for each i and each tER+. Hence
1. Invariant Sets and Trajectories
13
xt E (\ M;, = M" for each t E R:+ and M" is positively invariant. The
proof for negative invariance and for invariance are entirely analogous.
Theorem. Let M C X be positively invariant, negatively in-
1.3
variant, or invariant. Then the closure M has the same property.
Proof. Consider the case of invariance. Let x E M and t E R. Then

there is a sequence {xn} in M such that xn ~ x. By invariance of M we
have xnt E M for each n. Since xnt ~ xt we have xt E M. Thus M is
invariant. The proofs for positive and negative invariance are entirely
analogous.
Theorem. A set M C X is positively invariant if and only if
1.4
the set X - M is negatively invariant. M is invariant if and only if

X -- M is invariant.
Proof. Let M be positively invariant. If x E X - M and t E R- then

we nlust show that xt EX - M. Suppose not. Then xt EM and since
-tER+ we have xt(-t)=x(t-t)=xO=xEM by positive invariance of M. This contradiction shows that X - Mis negatively invariant. Proof of the converse is entirely similar. The proof of the second
result follows from the first.
As a final useful result on invariance we have
Theorem. A set M C X is invariant if and only if it is both po-
1.5
sitively and negatively invariant.

We have left the simple proof as an exercise.
The following important results on invariance are really corollaries
of the first three fundamental theorems on invariance established above.
1.6
Theorem. Let M C X be invariant. Then so is its boundary
oM and its interior J (M). The converse holds whenever M is open or
closed.
Proof. If Mis invariant then so is X - M by Theorem 1.4. Consequently both M and X - 111 are invariant by Theorem 1.3. Hence
oM = M f\ X - Mis invariant by Theorem 1. 2, and J (M) = X - X - M
is invariant again by Theorem 1.4. For the converse note that if
oM is invariant then so is J (ltJ). For otherwise there is an x E J (M)
and tE R with xt~ J(M). Since the set (assume for convenience t > 0)
x [O, t] is connected it follows that x [O, t] (\ oM =f: 0. Hence there is a
14
-c:, 0 < -c:
<
t} with X't' E oM. But then x
X-C: ( -T)
E ()Ji[ by :nvariance
of oM, which is a contradiction to x E J (M). We have thus proved that
if oM is invariant, then so is J (M). Consequently if M is closed then
M = oM V J (M) is invariant, and if M is open then M = J (11) is
invariant. The proof that when J (llf) is invariant then so is l\II for closed
or open Jill is even simpler and left to the reader.
1.7
Theorem. If M ( X is positively or negatively invariant then
so is its interior J (M).
Proof. Assume M is positively invariant. Then X - M and hence

X - M are negatively invariant. Consequently J (M) = X - X - 1Yf
is positively invariant.
1.8
Exercise. The sets X and 0 are positively and negatively invariant.
We now introduce the notion of a trajectory.
1..9
Definition. \Ve introduce the maps y, y+, and y- from X into

2x by defining for any x EX,
1.9.1
y (x) = {xt: t ER},
1.9.2
y+(x)
1. 9. 3
y- (x) =- {xt: t E R-}.
= {xt: tE R+},
For any x EX, the sets y (x), r+ (x)> and y-(x) are respectively called the
trajectory, the positive semi-trajectory, and the negative semi-trajectory
through x (or of x). Note that for any x EX, y (x) = xR, etc.
1.10
Exercises.
1.10.1 For any x E X the sets y (x), y+ (x), and y- (x) are, respectively,
invariant, positively invariant, and negatively invariant.
1.10.2 For the dynamical system defined by the differential equations
3.3.2 and 3.3.4, prove that the open and the closed unit disc are invariant
sets containing more than one trajectory.
For a given map Q: X ~ 2x and MC X we let Q(M) = V {Q(x):

x E .l\II}. Thus it is clear that for any llf C X, y (M) is invariant, y+ (M)
is positively invariant, and y-(M) is negatively invariant. This leads to
the following characterization of invariance.
1.11
Proposition. A set M C X is invariant, positively inva(iant, or
negatively invariant if and only if, respectively, y (M) == M, y+ (M) = M,
ory-(M) = M.
Another characterization of invariance is the following.
15
2. Critical Points and Periodic Points
1.12
Proposition. A set JJf C X is invariant, positively invariant, or
negatively invariant if and only if for each x E .ZV.l, respectively, y (x) (M,
y+ (x) C M, or y-(x) C !f.
1.13
Exercise. Give detailed proofs of Propositions 1.11 and 1.12.
The sets y (x), y+ (x), y- (x) as defined above need not in general have
elementary topological properties like closedness, compactness, or openness. In the next section we single out two important classes of points
whose trajectories are compact.
vVe close thiS. section by the following useful proposition whose proof
is immediate.
1.14
Proposition. A set l\tf C Xis invariant, positively invariant, or
negatively invariant if and only if each of its components has the same
property.

2.1
Definition. A point x E X is said to be a cri!ical point (or a rest
point, or an equilibrium point) if x = xt for all t E R.
The following theorem contains several characterizations of a critical

point.
2.2
Theorem. Let x E X. Then the following are equivalent.
2.2.1
x is critical,
2.2.2
{x}
2.2.3
= y (x),
{x} = y+ (x),
2.2.4
{x}==y-(x),
2.2.5
{x} = x [a, b] for some a< b.
2.2.6
There is a sequence {tn}, tn
>
0, tn-+ 0 with x
xtn for each n.
The proofs of the equivalence of 2.2.1 with 2.2.2-2.2.5 are almost

trivial once the equivalence of 2.2.1 and 2.2.6 is established. For this we
need the following important lemma.
2.3
Lemma. If x E X and x
for all integers n.
xt for some t E R, then x
x (nt)
Proof. If x = xt, then indeed x(-t) = xt(-t) = x(t - t) = x. Thus

we need prove the result for positive integers only. This follows from
simple induction. For if x = x (nt) and x == xt, then substituting xt for x
16
in the right hand side of the first equation we get x = xt (nt)

= x((n + 1) t).
= x (t + nt)
Proof of Theorem 2.2. We will only prove that 2.2.1 is equivalent

2.4
to 2.2.6. First assume 2.2.1. Then indeed 2.2.6 holds trivially as x = xt
for all t E R. Now assume 2.2.6 and let t E R. If t = ktn for some integers
k and n then indeed x = xt by Lemma 2.3. Otherwise for each integer n
there is an integer k,,, with kntn < t < (kn+ 1) tn and moreover an integer
m > n with
kin< kmtm < t < (km+ 1) tm < (kn
+ 1} tn.
Thus clearly the so constructed sequence {kin} has the property that
kntn--+ t. Now by the continuity axiom
x(kin)-+ xt
and since x = x(kntn) for each n by Lemma 2.3, we have x = xt. Since
t ER was arbitrary, xis critical.
We now prove some theorems about critical points. But first the
following lemma.
2.5
Lemma. If x =F xt for some x EX and t ER (i.e., xis not critical)
then there exist open neighborhoods U of x and V of xt such that V = Ut
and U (\ V = 0.
Proof. Note that if WC Xis open, then Wt = 7r} (W) is open because
nt is a homeomorphism {I, 1.2). Now let W1 and W2 be disjoint open
sets containing x and xt respectively. For example, if e(x, xt) = e (> 0),
then take W1 = S (x, e/2) and W 2 = S (xt, e/2). Then W 1t is open and
contains xt. Now set V ;:::= W 1t (\ W 2 and U = V(-t). Then V C W 2
and UC W1 , so that U and V are disjoint. Indeed x E U and xt EV so
that the open sets U and V are the required neighborhoods of x and
xt respectj vely.
We are now ready to prove some more results about critical points.
2.6
Theorem. A point x E X is critical if and only if every neighborhood of x contains a semi-trajectory.
Proof. The necessity is trivial as y (x) =+ {x} when xis critical so that
'Y (x) is contained in every neighborhood of x. For sufficiency assume that
x is not critical. Then there is a tE R+ with x =f= xt and consequently by
Lemn1a 2.5, there are disjoint open neighborhoods U of x and V of xt
with V = Ut. Since for each y E U we have yt E V, U cannot contain a
positive semi-trajectory. Similarly U cannot contain a negative semitrajectory. This contradiction shows that x = xt for every t ER and so x
is critical.
17
2.7
Theorem. The set of all critical points in Xis closed.
Proof. If the set of critical points is not closed then there is a sequence
{xn} of critical points with xn--+ x and x is not critical. Thus there is a
tE R with x =t= xt and hence, by Lemma 2.5, there are open sets U
containing x and V containing xt with Ut = V and U (\ V = 0. Since
xn -~ x we have xn E U for all sufficiently large n. Then for the t above
xntE V and in particular xnt~ U. But xn's are critical and therefore
xn = xnt E U. This contradiction proves the theorem.
As yet another application of Lemma 2.5., we have
2.8
Theorem. Let x, y EX and
t --+ - oo). Then x is critical.
f! (')'t,
x)--+ 0 as t--+
+ oo
(or as
Proof. Let U be a neighborhood of x. Since e(yt, x) --+ 0 as t--+ + oo

there exists T 2 0 such that yt E U for all t > T. Hence U contains the
positive semi-trajectory y+(yT) and therefore from Theorem 2.6 xis a
critical point.
2.9
Exercise. Prove that if x is critical and y =t= x, then x =t= yt for
every t ER. This is sometimes expressed by saying that no motion
starting at a point y E X reaches a critical point x =i= y in a finite tin1e.
(Hint: If xis critical, then {x} is invariant, and so is X - {x}.)
We now introduce the periodic points.

2.10
Definition. A pO'int x EX is said to be periodic if there is a T =i= 0
such that
2.10.1
xt = x (t
+ T)
for .all t E R.
A number TE R for which 2.10.1 holds will be called a period of x. If a

point x E X is periodic then both the motion nx and the trajectory y (x) are
said to be periodic.
Note that every x EX has the period T = 0, but it may not be

periodic. Further, if x EX is critical, then every TE R is a period of x,
and indeed x is periodic.
The following characterization of a periodic point is very useful.
2.11
with x
Proposition. x EX is periodic if and only if there is a T
=t= 0
xT.
Proof. Necessity is obvious. To see sufficiency, take any t ER and

apply the group axiom to obtain:
xt = xT(t) = x(t
+ T).
vVe now study some properties of periodic points.

2
18
2.12
Theorem. If x EX is periodic, but not critical, then there is
T > 0 such that T is the smallest positive period of x. Further if T is
any other period of x, then..,;= nT for some integer n.
Proof. Consider the set {t
>
0: t is a period of x} = P. Note that

P =f= 0 since if..,; =f= 0 is a period of x, then so is -T, as x = x..,; implies
x (-'t') = x..,; (-T) = x (..,; - ..,;) = x 0 = x. Since either ..,; or -..,; is positive, the set Pis non-empty. Now set inf P = T. \Ve claim that T > 0.
Indeed T > 0, and if T = 0, then there is a sequence {tn} in P with
tn--+ 0. Since x = xtn for each n, Theorem 2.2 shows that x is critical,
which contradicts our hypothesis. Thus T > 0. Note also that T is a
period of x. This follows from the fact that there is a sequence {tn} in P
with tn--+ T since T = inf P. Now x == xtn--+ xT by the continuity
axiom. Hence x == xT. Thus T is a period of x. By definition of T it is
also the smallest positive period. Finally, let t ER be any period of x.
If t =F nT for any integer n, then there is an integer n with nT < t <
(n + 1) T. But by Lemma 2.3, nT is a period of x. Thus we have x =
xt === x(nT). This gives xt(-nT) = x(nT) (-nT} = x(nT - nT) =
xO==x, i.e., x=xt(-nT)=x(t-nT), showing that t-nT is a
period of x. Since 0 < t - nT < T, we get a contradiction to the fact
that Twas the smallest positive period of x. This completes the proof.
The above theorem motivates the following definition.
2.13
Definition. If a point x is periodic but not critical, then the
smallest positive period of x is called its fundamental or primitive period.
In contrast to the situation for critical points (Theorem 2. 7) it is in
general not true that the set of all periodic points is closed. However
we can say the following:
2.14
Theorem. Given any x > 0, the set of all x such that xis periodic
with a positive period T ~ c:x is closed.
The proof will be made to depend on the following important lemma.
2.15
Lemma. If {xn} is a sequence of periodic points with positive
periodic Tn--+ 0, and xn--+ x, then xis critical.
Pruof. For given t ER, there are integers kn such that
knTn
< t<
knTn
+ Tn.
Since Tn--+ 0, we have knTn--+ t. Then xn = xn (knTn)--+ xt, and since

xn--+ x we have x = xt. As t E R was arbitrary, x is critical.
2.16
Proof of Theorem 2.14. Indeed if {xn} is a sequence of periodic
points with periods Tn <ex and xn ~ x, then also xn = xnTn--+ x. Since
19
0 < Tn < <X, either Tn--+ 0 in which case x is critical by Lemma 2.15
and hence periodic, or there is a subsequence Tn,,--+ ..,;, 0 <..,; < <X. But
then by the continuity axiom xnkTnk--+ x..,; and also xnkTnk = xnk--+ x.
Hence x = x..,; and x is periodic with a posit~ve period ..,; < <X. This proves
the theorem.
2.17
Exercise. If x E X is periodic, then y (x) is compact. Does the
converse hold? Prove also that if x is periodic, then y+ (x) is compact.
Does the converse of the last statement hold? (Hint: If x is periodic with
period T > 0, then y (x) = y+ (x) = x [O, T].)

We have seen above that trajectories of periodic points are compact
invariant sets. In general, however, the trajectory of a point x EX need
not even be closed (it is almost never open). Using Theorem 1.3 we see
that y (x) is invariant for any x E X, in fact it is the smallest closed invariant set containing x. We now introduce two n1aps Q: X--+ 2x such
that for each x E X, Q(x) is closed and invariant. These will be the socalled limit sets.
3.1
Definition. Define maps A+, A- from X into 2x by setting for
each xE X,
3.1.1
A+ (x)
and xtn -~ y},
= {y E X:
there is a sequence {tn} in R with tn--+
+ oo
3.1.2
A-(x) =- {y E X: there is a sequence {tn} in R with tn--+ - oo
and xtn--+ y}.
For any x EX, the set A+ (x) is called its pos1:tive (or omega) limit set,
and the set A-(x) is called its negative (or alpha) limit set.
Exercise. Show that if x E X is periodic, then A+ (x) = .A-(x)

y (x). (Partial converses of this statement hold. See III, 2.5 and 2.9.)
3.2
Before proceeding further we give some examples of limit sets.

3.3
Examples of Limi:t Sets.
3.3.1
Consider the differential system defined in R 2 by the differential
equations (polar coordinates)
3.3.2
2*
dr
dt
=r(l- r),
20
It can easily be verified that the solutions are unique and all solutions
are defined on R. Thus 3.3.2 defines a dynamical system. The trajectories
are shown in Fig. 3.3.3. These consist of: (i} a critical point, namely the
origin 0, (ii) a periodic trajectory y coinciding with the unit circle, (iii)
spiralling trajectories through each point P = (r, 0) with r =f= 0, r =f= 1.
For points P with 0 < r < 1, A+(P) is the unit circle and A-(P) is the
singleton {O}. For points P with 1' > 1, A+(P} is the unit circle and
A-(P) = 0.
Fig.3.3.3
3.3.4
Consider another differential system defined in R 2 by (polar
coordinates)
dr
dt
=r(l-r),
d(J
dt
sin2 0
+ log
- 1- 3
sin2 0
ifO<r<!,
1
'Y
log-1-
'Y
sin2 ()
sin2 0
if! < r < 1,

if r
== 1,
ifr>l.
'Y
log-r -
One may verify that this system also defines a dynamical system on R 2
The trajectories are shown in Fig. 3.3.5. These consist of: (i) three critical
points, namely the origin 0 and the points A = (1., 0) and B = (1, n) on
the unit circle, (ii) a trajectory y 1 = {(1, 0): 0 < 0 < n} on the unit
circle, (iii) a trajectory y 2 = {(1, 0): n < 0 < 2n} on the unit circle, and
(iv) spiralling trajectories through points P = (r, 0) with r =f= 0, r =f= 1.
For any P = (r, 0) with 0 < r < 1, A+(P) is the unit circle, and.A-(P)
is {O}. For any P with r > 1, A+(P) is the unit circle and A-(P) = 0.
For any P = (1, 0) with 0 < () < n, A+(P) = {B} and A-(P) ={A}.
For any P == (1, 0) with n < 0 < 2n, A+(P) ={A} and A-(P) = {B}.
21
Fig.3.3.5
3.3.6
Consider in the space R 2 the differential equations
i = f (x, y), y = g(x, y),
3.3. 7
where the functions
f and g are given by

0 if
f (x, Y) = {
3.3.8
Here
y(1 -
(1
+ y 2 ) (1 -
x2 )
p(x) q(y))
if
lxl <
1,
+1ifx>1
g(x,y) = -1 if x < -1
{
x
if jxl < 1.
p (x), q (y) are any continuously differentiable functions satisfying
< 21
if 0 < x < 1,
p(x)
if x
< 0,
p(x}
=2
if x
>
0< p(x)
3.3.9
lxl >
0
1
O<q(y)<2
q(y)
1,
if y< 0,
if y> 0.
It is easily verified that such a system describes a dynamical system as

shown in Fig. 3.3.10. This dynamical system has the straight lines
x = 1 as the positive limit set of all points in the strip -1 < x < +1,
except the origin.
22

y
+1
-1
Fig.3.3.10
We now proceed to establish some important properties of the limit

sets and their relationship with the semi-trajectory closures.
3.4
Theorem. For any x EX
3.4.1
A+(x) andA-(x) are closed invariant sets,
3.4.2
y+(x)
y+(x) V A+(x)
and
y-(x) = y-(x) V A-(x).
Proof. Consider the case of A+(x). Let {Yn} be a sequence inA+(x)

with Yn ~ y. We wish to show that y EA+ (x). Indeed for each positive
integer k, there is a sequence {t!} in R with t! ~ + oo and xt! ~ Yk
We may assume without loss of generality that e(yk, xt!) < 1/k and
t! > k for n > k. Consider now the sequence {tn} in R with tn = t~. Then
tn ~ + oo and we claim that xtn ~ y. To see this, observe that
Since 1/n and e()1, Yn) tend to zero we conclude that e(y, xtn) ~ 0. Consequently, xtn~Y and yE A+(x). We have thus proved that A+(x) is
closed. To see thatA+ (x) is invariant, let y EA+ (x) and t ER be arbitrary.
There is a sequence {tn} in R with tn ~ + oo and xtn ~ y. Then by the
continuity axiom xtn (t) ~ yt. Since xtn (t) = x (tn + t) and Tn = (tn + t)
~ + oo we have yt EA+ (x) and A+ (x) is invariant.
23
We now prove 3.4.2. For this remember that y+(x) == xR+. By the
--definition of A+(x) we have y+(x)) y+(x) V A+(x). To see that y+(x)
(y+(x) VA+(x), let yEy+(x). Then there is a sequence {Yn} iny+(x)
such that Yn--+ y. Now Yn = xtn for a tn ER+. Either the sequence {tn}
has the property that tn--+ + oo, in which case y EA+ (x), or there is a
subsequence tnk --+ t E R+ (as R+ is closed). But then xtnk--+ xt E y+ (x),
and since also xtnk--+ y we have y
V A+(x). This proves 3.4.2.
xt E y+ (x). Thus y+ (x) Cy+ (x)
3.5
Exercises. For any x E X prove that
3.5.1
A+(x)
3.5.2
A+(x)=A+(xt)
== (\ {y+(y):
y E y+(x)} = (\ {y+(xn): n is an integer}.

for
tER.
We now prove connectedness properties of limit sets. The principal

theorem is
3.6
Theorem. If the space X is locally compact, then a positive
limit set A+ (x) is connected whenever it is compact. Further, whenever a
positive limit set is not compact, then none of its components is compact.
(In the following proof and elsewhere in this book we use the notation
S (M, e), S [M, e], and H (M, e) for the sets {x: e(x, M) < e}, {x: e(x, M)
< e}, and {x: e(x, M) = e}, respectively. Here MC X and e > 0.)
Proof. Let A+ (x) b~ compact, and let it be not connected. Then

A+(x) = P V Q, where P, Q are non-empty, closed, disjoint sets. Since
A+(x) is compact, so are P and Q. Further, since Xis locally compact,
there is an e > 0 such that S [P, e], S [Q, e] are compact and disjoint.
Now let y E P and z E Q. Then there are sequences {tn}, {rn}, tn--+ + oo,
Tn--+ + oo, such that xtn--+ y, and XTn--+ z.
We may assume without loss of generality, that xtn ES (P, e),
x..,;n ES (Q, e), and Tn - tn > 0 for all n. Since the trajectory segments
x [tn, ..,;n], n = 1, 2, ... , are compact connected sets, they clearly intersect
H (P, e) and H (Q, e). Thus, in particular, there is a sequence {Tn},
tn < Tn < Tn, such that xTn EH (P, e) which is compact. We may therefore assume that xTn--+ y, and as Tn--+ + oo, we have y E A+(x). However, yft P V Q, which is a contradiction. This establishes the first part
of the theorem.
To prove the second part of the theorem we need the following topological theorem, which we give without proof.
3.7
Topological Theorem. Let S be a Hausdorff continuum (a compact connected Hausdorff space), let U be an open subset of S, and
let C be a component of U. Then U - U contains a limit point of C.
24
3.8
Proof of the Second Part of Theorem 3.6. Notice that the space
X is a locally compact Hausdorff space, and everything that has been
said above goes through in such a space. Now X possesses a one-point
compactification. So let X =XV {ro} be the one-point compactification
of X by the ideal point ro. Extend the dynamical system (X, R, n) on
x to a dynamical system (X, R, n) on X, where i is given by n(x, t) ==
n(x, t) for x EX, t ER, and ii (ro, t) = ro for all t ER. If now for x EX,
- (x) =A+ (x)
A+ (x) denotes the positive limit set of x, then clearly A+
V {ro}, whene~er xE X and A+(x) is not compact. However, A+(x) is
compact, as X is compact, and by the first part of the theorem it is
c.onnected. A"+(x) is therefore a Hausdorff continuum. Further A+ (x)
is an open set in.J+ (x). Now A+ (x) -A+(x) = {ro}, and so by Theorem 3. 7
every component of A+ (x) has ro as a limit point, and so is not compact.
This proves the theorem completely.
Exercises.
3.9
3.9.1
Show that if y+(x) is compact, then A+(x) is a non-empty compact and connected set.
3.9.2
Show that if y+(x) is compact, then

e(xt, A+(x))--+ O
as
t-+
+ oo.
3.9.3
Show that in a locally compact space X, A+(x) is
and compact if and only if y+ (x) is compact.
non~empty
3.9.4
Show that 3.9.2 may not be true if y+(x) is not compact.
3.9.5
If X is not locally compact, then Theorem 3.6 may not hold.
(Hint: Consider the restriction of the dynamical system in example 3.3.4

to the set R 2 - y 1 V y 2 .)
4. The First Prolongation and the Prolongational Limit Set

In this section we introduce yet other maps D+, D-, l+, l- from X
into 2x. These maps play an important role in the theory of dynamical
systems, particularly in the stability theory and the theory of dispersive
systems.
Definition. Define maps D+, D-, l+, l- from X into 2x by
setting for each x E X,
4.1
4.1.1
.D+(x) = {y EX: there is a sequence {xn} in X and a sequence
{tn} in R+ such that xn--+ x and xntn--+ y },
25
D-(x) ={YE X: there is a sequence {xn} in X and a sequence

{tn} in R- such that xn-+ x and x,.tn-+ y},
4.1.2
4.1, 3
J+ (x) = {y E X: there is a sequence {xn} in X and a sequence
{tn} in R+ such that xn-+ x, tn-+ + oo, and xntn -+ y},
J-(x) ={yEX: thereis a sequence {xn} inXandasequence
{tn} in R- such that xn-+ x, tn-+ - oo, and x,/,.-+ y}.
4.1.4
For any x E X, the set D+ (x) is the first positive and the set D-(x) the
first negative prolongation of x.
The sets ]+ (x) and J-(x) are called, respectively, the first positive
and the first negative prolongational limit set of x.
It is clear that for any x EX, JJ+(x) ) y+(x), D-(x) ) y-(x), J+(x) )
A+(x), and J-(x)) A-(x). That these inclusions may be proper is
shown by the following simple example.
4.2
Example. Consider the planar differential system (cartesian
coordinates)
dx1
dt = -
X1,
dx2
dt =
X2.
These equations define a dynamical system with the so-called saddle

point at the origin. The trajectories are shown in Fig. 4.2.1. For any
point P = (x1 , 0), D+ (P) = y+ (P) V {(x1 , x2) : x1 = O}, D-(P) = y- (P),
J+(P) = {(x1 , x2): x1 = O}, J-(P) = 0 if x1 =f= 0 and J-(P) ={(xv x2):
x2 = O} if x1 = 0. For points P = (x1 , x2) with x1 =f= 0, x 2 =f= 0, D+(P) =
y+(P), D-(P) = y-(P), J+(P) = J-(P) = 0. Clearly, for points P =
--(x1' 0), D+ (P) ::/= y+ (P), and J+ (P) =t= A+ (P) (what are y+ (P) and
A+(P) ?).
Xz
Fig.4.2.1
26
We now discuss some basic properties of these sets.
4.3
Theorem. For any x E X,
4.3.1
D+ (x). is closed and positively invariant,
4.3.2
J+(x) is closed and invariant,
4.3.3
D+(x) =y+(x) V J+(x).
Analogous results hold for D-(x) and J-(x).
Proof. It is sufficient to prove 4.3.2 and 4.3.3. We first prove 4.3.2.

Let {Yn} be a sequence in J+(x) with Yn--+ y. For each integer k, there
are sequences {x!} in X and {t!} in R+ with x!--+ x, t!-+ + oo, and
x!t!--+ Yk We may assume by taking subsequences if necessary that t! > k,
e(x!, x) < 1/k, and e(x!t!, Yk) < 1/k for n > k. Now consider the
sequences {x:}, {t:}. Clearly x:--+ x, t:--+ + oo, and x:t:--+ y. To see
for example that x:t:--+ y note that e(x:t:, y) < e (x:t:, Yn) + e (Yn' y)
< 1/n + e(Yn' y). Thus yE J+(x) and J+(x) is closed.
To see that J+(x) is invariant, let y E J+(x) and t ER. There is a
sequence {xn} in X and a sequence {tn} in R+ such that tn--+ + oo,
xn--+ x, and xin--+ y. Now consider the sequence {tn + t}. Clearly
tn + t -* + oo, and xn (tn + t) = Xin (t) --+ yt. Since xn--+ x, we have
yt E J+ (x). As t E R was arbitrary, J+ (x) is invariant. This completes
the proof of 4. 3. 2.
To see 4.3.3 observe that D+(x) ) y+(x) V J+(x) holds always (see
definitions). We will prove D+(x) C y+(x) V J+(x). Let y E D+(x). Then
there is a sequence {x,.} in X and a sequence {tn} in R+ with xn-+ x,
xntn--+ y. We may assume that either tn-+ tE R+ or tn--+ + oo, if necessary
by taking subsequences. In the first case xin--+ xt by the continuity
axiom and so xt = y E y+ (x) as t E R+. In the second case y E ] + (x) by
definition. Hence y E y+ (x) VJ+ (x) and 4.3.3 is proved.
We now prove connectedness properties of D+(x) and J+(x).
4.4
Theorem. Let the space X be locally compact. Then for any
x E X, D+ (x) is connected whenever it is compact. If D+ (x) is not compact, then it has no compact component.
Proof. Let D+ (x) be compact but disconnected. Then there are two
compact non-empty sets P and Q such that P V Q = D+ (x) and P (\
Q = 0. Since P and Q are compact, e(P, Q) > 0. Thus there is a ~ > 0
such that S [P, ~], S [Q, ~] are compact and disjoint. Now x E .P or
x E Q. Let x E P. Then there is a sequence {xn} in X and a sequence {tn}
in R+ such that xn--+ x, and xntn--+ y E Q. \Ve may assume xn E S [P, ~]
and xntn ES [Q, ~]. Then the trajectory segments xn [O, tn] intersect
4. The First Profongation and the Prolongational Limit Set
27
H (P, c5) for every n, and therefore there is a sequence {rn}, 0 < Tn < tn,
such that xn-rn EH (P, ") Since H (P, ") is compact we may assume that
xn-rn--+zEH(P,") Then zE.D+(x), but z~PVQ aszEH(P,"). This
contradiction shows that D+ (x) is connected.
To prove the second part of the theorem, we consider the one point
compactification X =XV {ro} of the space X, and for each xE X define
the set D* (x) by
D* (x) = {y E X: there is a sequence {xn} in X and a sequence {tn} in R+
such that xn--+ x and xin--+ y}.
The set D* (x) is always a compact subset of X. Following exactly the
same proof as given above for JJ+ (x) one can show that D* (x) is connected.
Now if D+(x) is not compact, then clearly
D* (x) = D+(x) V {ro}.

Further D+ (x) is an open subset of D* (x). By the topological Theorem 3. 7,
each component of n+ (x) has a limit point in D* (x) - D+ (x) = {ro }.
Thus no component of D+ (x) is compact. This proves the theorem.
Theorem 4.4 clearly holds for D-(x). That it also holds for J+(x) and
J- (x) will be shown next. But first the following lemma.
4.5
Lemma. Let X be locally compact. Then A+ (x) =f= 0 whenever
]+ (x) is non-empty and compact.
Proof. If possible let A+ (x) = 0. Then we claim that y+ (x) is closed
and disjoint with J+(x). That y+(x) is closed follows from y+(x) =
y+ (x) VA+ (x) = y+ (x) as A+ (x) = 0. That y+ (x) (\ ]+ (x) = 0 follows
'from the fact that if y+ (x) f\ ]+ (x) =f= 0, then by invariance of ]+ (x),
y+(x) C J+(x). Since J+(x) is compact, we will have A+(x) =f= 0 and
compact (remember that any sequence {Yn} in a compact set Q has a
convergent subsequence). This again contradicts the assumption A+(x)
= 0. Thus y+ (x) is closed and y+ (x) f\ ]+ (x) = 0. Since J+ (x) is nonempty and compact we have e(y+(x), J+(x)) > 0. Thus there is a c5 > 0
such that S [J+(x), c5] is compact and disjoint with y+(x). Now choose
any y E J+(x). There is a sequence {xn} in X and a sequence {tn} in R+
such that xn--+ x, t,.--+ + oo, and xntn--+ y. We may assume that x EJ:
S [J+(x), c5], xntn ES [J+(x), c5] for all n. Then the trajectory segments
xn [O, tn] intersect H (]+ (x), c5) and therefore there is a sequence {-rn},
0 < Tn < tn, such that xn-rn EH(]+ (x), c5). Since H (]+ (x), c5) is compact,
we may assume that xn-rn--+ z EH(]+ (x), c5). By taking subsequences we
may assume that either Tn--+ t E R+ or Tn--+ + oo. If Tn--+ t E R+, then
by the continuity axiom xn-rn--+ xt = z, i.e., z E y+ (x) which contradicts
28
y+(x) f\ S U+(x), "J = 0. If Tn ~ + CX), then zE J+(x), but this contradicts zE H(J+(x), ") as J+(x) (\ H(J+(x), c5) = 0. Thus our original
assumption A+ (x) = 0 is untenable and the lemma is proved.
As an immediate application we have
4.6
Lemma. Let X be locally compact. Then ]+ (x) is non-empty
and compact if and only if [)+ (x) is compact.
Proof. Let J+(x) be non-empty and compact. Then A+(x) is nonempty and compact. But then y+(x) is compact (prove this). Hence
D+(x) = y+(x) V J+(x) = y+(x) V J+(x) is compact. The converse is
trivial.
We now prove the connectedness properties of J+(x).
4.7
Theorem. If Xis locally compact, then for any xE XTheorem4.4
holds for J+(x).
Proof. Let J+ (x) be compact. If J+ (x) = 0 there is nothing to prove.
So let J+(x) =t= 0. If J+(x) is disconnected, then there are non-empty
compact sets P, Q such that J+(x) = P V Q and Pf\ Q = 0. Since
A+ (x) is non-empty and compact, hence connected, we have A+ (x) ( P
or A+(x) C Q. Let .A+(x) C P. Then y+(x) VP is compact and disjoint
from Q. That it is compact follows from y+(x) VP= y+(x) VP as
A+ (x) C P and y+ (x) is compact. That it is disjoint from Q follows from
the fact that if Q f\ (y+ (x) V P) =t= 0, then Q f\ r+ (x) =f= 0. But Q must
be invariant (why?). This will show that A+(x) C Q, a contradiction.
Finally, D+(x) = (y+(x) VP) V Q and since Q and y+(x) VP are disjoint compact sets we have a contradiction to the fact that D+ (x) is
connected whenever compact. This proves that J+(x) is connected whenever compact. The proof of the second part of the theorem is exactly
similar to proof of the second part of Theorem 4.4 and is left to the reader.
4.8
lixercises.
4.8.1
For any xE X, D+(x) = f\ {S(x, <X) R+:
statement for J+(x) and D-(x).
<X
>
0}. Give a similar
4.8.2
Give examples to show that Theorems 4.4 and 4. 7 and Lemmas
4.5 and 4.6 are not in general true for non-locally compact spaces X.
4.8.3
4.8.4
X, let
Give a detailed proof of the second part of Theorem 4. 7.

Given a dynamical system (X, R, n) on a locally compact space
X =XV {ro} be the one point compactification of X and define

,..,
29
,..,
ii: XxR-+X by
n (x, t) =
{n(x, t) if.f xE X
ro
and tE R,
ro and t E R.
Prove rigorously that (i, R, n) is a dynamical system on
X.
For any compact M C X, the set D+ (M) is closed and positively

invariant.
4.8.5
We conclude this chapter with a non-trivial property of the maps]+

and J- which is generally not possessed by the maps A+ and A-.
4.9
Theorem. Let x, yin X. Then y E J+ (x) if and only if x E J-(y).
Proof. y E J+ (x) if there is a sequence {xn} in X and {tn} in R such

that Xn-+ x, tn-+ oo, and Xin-+ y. Set xntn = Yn' and Tn = -tn.
Then Yn-+ y, Tn-+ - oo, and YnTn = xntn (-rn) = Xn (tn - tn) = Xn-+ x.
Consequently x E J-(y). The converse holds also.

For some applications (section V, 5) we shall need the following
concepts of relative prolongation and relative prolongational limit set.
4.10
Definition. Let M C X be compact. If U C X, the (positive)
prolongation of M, relative to U is the set
4.10.1
D+(M, U) =
V {y EX:
there is a sequence {xn} C U and a
xEM
sequence {tn}CR+ such that xn-+X and xntn-+Y},

while the prolongational limit set of M, relative to U is the set:
4.10.2
J+(M, U) =
V {y EX:
there is a sequence {xn} C U and a
xEM
sequence {tn} ( R+ such that xn-+x and tn-+ + oo, Xn tn-+ y}.
4.11
Exercises. Let M C X be compact and U C X. Prove the
following statements.
4.11.1
D+(M, U) is positively invariant.
4.11.2
]+ (M, U) is invariant.
4.11.3
If UC Xis closed, then D+ (M, U) and J+ (M, U) are closed.
4.11.4
If U C X is positively invariant, then

D+(M, U)
4.11.5
y+(M (\ U) V J+(M, U).
If U is a neighborhood of M, then D+(M, U)
== D+(M).
30

Most of the results presented in this chapter are classical.
Section 1. The trajectory y (x) is also called orbit. The word orbit designates y (x)
mostly when y(x) is compact. We do this to keep our terminology close to the
classical one. Theorems 1.2-1.7 are to be found in weaker versibns in a paper by
S. LEFSCHETZ [2].
Section 2. In 2.2.6 one may replace the null sequence {tn} by an arbitrary convergent sequence {tn}, tn ~ t, tn =F t.
Theorem 2.6 is a stronger version of a similar theorem found in the book by
V. V. NEMYTSKII and N. V. STEPANOV [1]. The problems of existence of critical
points of a dynamical system or of a differential equation on manifolds have been
discussed by many authors among which we mention L. E. EL'sGoL'c [1], F. HAAS
[5], G. S. JoNES [4], and G. S. JONES and J. A. YoRKE [1]. Some criteria due to
N. P. BHATIA, A. LAZER and G. P. SzEGO [1] and to N. P. BHATIA and G. P. SzEGo
are also presented in V,3 of this book in the context of the relations between the
properties of sets and of their regions of attraction.
Section 3. The definition of limit sets is due to G. D. BIRKHOFF [1, vol. 1, pp. 654672]. This concept had been previously used by H. PoINCARE without a formal
definition.
Alternative equivalent definitions (3.5.1) of limit set were used by S. LEFSCHETZ
[2] and by T. URA [4]. The second part of Theorem 3.6 is due to N. P. BHATIA [3].
In special phase spaces stronger results than Theorem 3.6 may be proved. For
example in R 2 the following result on the structure of limit sets is given by HAJEK
[1, p. 184]: every positive limit set consists of critical points and at most a countable number of non-critical trajectories y1i.; each compact subset of R 2 which contains no critical points intersects at most a finite number of Yn For further result
on the structure of limits sets see SIBIRSKII [1, 2]. (See also V, 6.)
Section 4. The formal definition of prolongation is due to T. URA [2]; he adopts
essentially the property 4.8.1 as a definition. This concept has been previously
used by H. PoINCARE [1, vol. 1, p. 44] in a special case (Example 4.2) without a
formal definition. The concept of prolongational limit set is due to J. AUSLANDER,
N.P.BHATIA and P.SEIBERT [1]. Theorems4.4 and 4.7 are due to N.P.BHATIA
[3]. The notion of relative prolongation is due to T. URA [5]. Most of the concepts
defined in this chapter can also be defined in more general situations for instance for
the case of local semidynamical systems (see N. P. BHATIA and 0. HAJEK [1]) or for
the case of dynamical systems without uniqueness (see G. P. SZEGO and G. TRECCANI [1]).
The structure of prolongational limit sets on special phase spaces like manifolds
or even on R2 has still not been fully investigated.
N. P. BHATIA [10] has recently introduced a somewhat different concept of
relative prolongation than the one in 4.10. This seems more useful for the study
of stability and attraction in non-locally compact spaces.
Chapter III
Recursive Concepts
1. Definition of Recursiveness
1.1
Definition. A set A C X will be said to be positively recursive
with respect to a set B C X if for each TE R there is a t > T and an
x EB such that xt EA. Negative recursiveness may be defined by using
the inequality t < T. vVe will say that a set A is self positively recursive
whenever it is positively recursive with respect to itself.
The simplest example of a self recursive set is that of a singleton {x}

where x is a periodic point (prove this I). In this chapter we study those
concepts which are connected with the concept of recursiveness defined
above. These are the concepts of Poisson stable and non-wandering
points, and the concepts of minimal sets and recurrent points. To be
sure, one may introduce other concepts by requiring that the entering
of points from a set B into the set A happens with some kind of regularity
not necessarily as stringent as in the case of a periodic point. This can
give rise to various concepts of almost periodicity, but we shall not study
this in detail (see GOTTSCHALK and HEDLUND [1] for very general
concepts of almost periodicity in topological transformation groups).
2. Poisson Stable and Non-wandering Points

2.1
Definition. A point x E X is said to be positively Poisson stable
if every neighborhood of xis positively recursive with respect to {x}.
We give some characterizations of positively Poisson stable points.
2.2
Theorem. Let x E X. Then the following are equivalent.
2.2.1
xis positively Poisson stable,
32
III. Recursive Concepts
2.2.2
t > T,
given a neighborhood U of x and a T
2.2.3
xE A+(x),
2.2.4
y+(x) =A+(x),
2.2.5
y(x) (A+(x),
2.2.6
for every e > 0 there is a t
>
>
0, xt E U for some
1 such that xt ES (x, e).
Proof. The only non-trivial statement in the above theorem is the

equivalence of 2.2.6 with either of the statements 2.2.1-2.2.5. \Ve shall
prove that 2.2.3 is equivalent to 2.2.6. Indeed 2.2.3 implies 2.2.6 trivially.
So assume 2.2.6. Then choosing a positive null sequence {en} one sees
that there is a sequence {tn}, tn > 1, such that xtn E S (x, en) consequently
xtn--+ x. Now either {tn} contains a convergent subsequence {tnk}, tnk --+
t > 1, or tn--+ + oo. In the first case xtnk--+ xt by the continuity axiom
and so x = xt and x is periodic hence x EA+ (x) (see II, 3. 2). In the
second case x E A+ (x) by definition. Hence 2. 2. 6 implies 2. 2. 3.
2.3
Exercises.
2.3.1
Give complete proof of Theorem 2.2.
2.3.2
If x is positively Poisson stable then so is xt for every t E R,
The following alternative definition of Poisson stability is customary in

the literature and is suggested by the above Theorem 2.2.
2.4
Definition. A point x E X is positively or negatively Poisson
stable whenever, respectively, x EA+ (x) or x E A-(x). It is said to be
Poisson stable if it is both positively and negatively Poisson stable. If a
point x E X is Poisson stable then both the motion nx and the trajectory
y (x) are said to be Poisson stable.
In view of Theorem 2.2 it is interesting to inquire about the consequences of the condition y+(x) = A+(x). The answer is contained in the
following theorem.
2.5
Theorem. y+ (x)
= A+ (x)
if and only if x is a periodic point.
Proof. Let y+(x) =A+(x). Indeed then xEA+(x) and as A+(x) is

invariant we see that y+(x) =A+(x) =y(x). Thus x-rEy+(x) for each
T < 0, and, therefore, there is a -r' > 0 such that x-r = x-r'. Hence by
the group axiom xt = x (t + 7:1 - -r) for all t E R, showing that x is
periodic with a period 7: 1 - T (> 0). The converse holds trivially and the
theorem is proved.
33
Remark. It is to be noted that if y+(x) = A+(x) then the point

xis indeed Poisson stable. It is, therefore, appropriate to inquire whether
there exist points which are Poisson stable but are not periodic (i.e., also
not a rest point). We give below an example of Poisson stable points which
are not periodic.
2.6
2. 7
Example.
2. 7.1
Consider a dynamical system defined on a torus by means of the

planar differential system
2.7.2
d~
dt
f(<p, 0),
dO
dt
rxf(<p, 0),
where I (<p, 0) == I ( 0 if 0
be irrational. It is easily seen that the trajectories of this system on the
torus consist of a rest point p corresponding to the point (0, 0). There is
exactly one trajectory y 1 such that A-(y1) = {p}, and exactly one trajectory y 2 such that A+(y2) = {p}. For any other trajectory y, A+(y) =
A-(y) =the torus. Further A+(y1 ) = A-(y 2) =the torus. In this
example points on the trajectory y 1 are positively Poisson stable, but not
negatively Poisson stable. Points on y 2 are negatively Poisson stable,
but not positively Poisson stable. All other points are Poisson stable.
Note that no point except the rest point pis periodic.
Fig. 2. 7. 3 describes this example.
Fig.2.7.3
2. 7.4
If in the above example we have / (<p, 0) > 0 for all <p, 0 (i.e.,
also /(0, 0) > 0), then every trajectory is dense in the torus and moreover the torus is also the positive and negative lir~1it set of each point.
This example describes a compact minimal set which is not a periodic
orbit (see section 3 for definition of minimal sets) and indeed each point
is positively and negatively Poisson stable.
The following theorem throws some light on positively Poisson stable
points x when y+(x) =t= A+(x).
3
34
2.8
Theorem. Let X be a complete metric space. Let x E X be
positively Poisson stable, and let it not be a periodic point. Then the
set A+ (x) - y (x) is dense in A+ (x), i.e., A+ (x) - y (x) = A+ (x) = y (x).
Proof. Since x is positively Poisson stable, we have A+ (x) = y (x).
To see that A+(x) -y(x) ==A+(x), it is sufficient to show that if
yE y(x) and e > 0 is arbitrary, then there is a point zEA+(x) -y(x)
such that zES(y,e). To see this notice that since yEA+(x)==A+(y),
there is a monotone increasing sequence {tn}, tn-+
oo, such that
ytn-+ y. Choose Ti > ti such that YTi E S (y, e). Then YTi ~ y [-t1 , t1 ]
(otherwise x will be periodic). Hence c5i == (! (y-r1 , y [ -t11 ti]) > 0. Set
e1 == min {e/2, e - (! (y, yri), c51 /2}. Then S (y-r11 e1) C S (y, e) and
S (y-r1 , e1) (\ y [-ti, ti] = 0. Having defined Y'l'n_ 1 and en_ 1 , choose
Tn > tn such that YTn E S (Y'l'n_ 1 , Bn_ 1) {possible because of positive Poisson
stability of x). Then define en= min {en-1/2, en_ 1 - (! (YTn_ 1 , YTn),
c5n/2}, where c5n = (! (YTn, y [-tn, tn]). Note that c5n > 0 as the motion is
not periodic. Clearly S (YTn, en) ( S (YTn_ 1 , e1, _ 1 ), and S (YTn, en)(\ Y [ -tn, tnJ
= 0. The sequence {y-rn} has the property that (! (YTn, YTn_ 1) < en_ 1 <
e/2n- 1 for n = 1, 2, .... {y-rn} is, therefore, a Cauchy sequence which
converges to a point z as the space X is complete. Since yrn E y (x), and
Tn-+ + oo, we have z EA+ (x). Further(! (y, yrn) < B, so that(! (y, z) < e.
Notice further that z~ y (x). For, otherwise, if z E y (x)
y (y), we will
have z = y-r. But there is an n such that tn > 1-r/, so that z E y [ -tn, tn].
However, z ES (yrn, en), and by construction S (YTn, en)(\ y [-tn, tn] == 0,
i.e., z ~ y [-tn, tnJ. This contradiction proves that z ~ y (x) and the theorem is proved.
==
It is now clear that

2.9
Theorem. If X is complete, then a necessary and sufficient
condition that y (x) be periodic is that y (x) == A+ (x) [== A-(x)].
2.10
Remark. Theorem 2.8 is not true if Xis not complete. This can
be shown for instance by restricting the dynamical system in example
2. 7 to a single trajectory say 'Yi Clearly for such a flow 'Yi= A+ (x)
for every x E y11 but no x E Yi is periodic.
We now introduce the notion of a non-wandering point.

2.11
Definition. A point x EX is said to be non-wandering if every
neighborhood U of x is self positively recursive.
We give some characterizations of non-wandering points.
2.12
Theorem. For any x EX, the following are equivalent.
2.12.1
x is non-wandering,
2.12.2
xE]+(x),
2.12.3
every neighborhood of xis self negatively recursive,
2.12.4
x E J-(x).
35
Proof. Assume 2.12.1. Consider a null sequence {en}, 0 < en, Bn ~ 0,

and a sequence {tn} inR with tn ~ + oo. Since eachS (x, en) is self positively
recursive, we have an xn ES (x, en) and a Tn > tn with xn-rn E S (x, en)
Since en~ 0 we have xn ~ x and xn-rn ~ x and since Tn ~ + oo we conclude x E J+ (x). Thus 2.12.2 holds. Now assume 2.12.2. Then there is a
sequence {xn} in X and a sequence {tn} in R with xn ~ x, tn ~ + oo, and
xntn ~ x. Now for any neighborhood U and T > 0 there is an N such
that tn > T for n > N and xn EU and xin EU for n > N. Thus U is
self positively recursive. Consequently x is non-wandering and 2.12.1
holds. Equivalence of 2.12.3 and 2.12.4 is proved in the same way. To
complete the proof, it is sufficient to show that 2.12.2 and 2.12.4 are
equivalent. But this is trivial. A more general result was proved in
II, 4.9.
We shall now prove a theorem which shows the connection between
Poisson stable points and non-wandering points. First the following
elementary result.
2.13
Theorem. Let x E X. Every y EA+ (x) is non-wandering.
Proof. We have to show that if y E A+(x) for some x EX, then

y E J+(y). Indeed there is a sequence tn ~ + oo with xtn ~ y. Since
tn ~ + oo, we may assume, if necessary by taking a subsequence, that
tn+l - tn > n for each n. Then setting Tn = tn+l - tn and xtn = xn
we have xn ~ y, xn (tn+l - tn) = xtn+l ~ y, and Tn = tn+I - tn ~ + oo.
Thus y E J+ (y) and y is non-wandering.
A slightly deeper result is the following.
2.14
Theorem. Let P C X be such that every x E Pis either positively
or negatively Poisson stable. Then every x E P is non-wandering.
Proof. Let {xn} in P and xn ~ x. We must prove that x E J+ (x).

Indeed for each n we have_ either xn EA+ (xn) or xn E A-(xn). Thus by
taking a subsequence, we may assume that either xn E .A+ (xn) for all n
or xn E A-(xn) for all n. Assume xn E A+(xn) for all n. For each n there is
a tn > n with e(xn, xntn) < 1/n. Then clearly e(x, xntn) < e(x, xn) +
3*
36
e(xn, xntn) < e(x, xn) + 1/n. This shows that Xin-+ x
and consequently
xE J+(x). In the second case similar considerations show that xE J-(x).
Thus by Theorem 2.12 every x E P is non-wandering.
We now give a partial converse of Theorem 2.14.
2.16
Theorem. Let X be complete. Let every x E X be non-wandering.
Then the set of Poisson stable points P is dense in X.
Proof. Let x E X an<:} 8 > 0. We will find a point y E S (x, 8) such that
y E A+(y) and y E A-(y). Let S (x, 8) = U. Since U is self positively
recursive, there is a t1 > 1 such that U f\ Ut1 =f: 0. Indeed U (\ Ut1
is open. So choose an x1 in U f\ Ut1 and positive 8 1 < 1/2 such that
S (x1 , 8 1 ) C U (\ Ut1 Set U1 = S (x1 , 8 1). Since x1 is non-wandering,
U1 is self negatively recursive, and there is a t2 < -2 such that U1 (\
U1t2 =f: 0. Choose X2 in ul (\ U1t2 and a positive 82 < 1/22 such that
S (x2 , 8 2) C U 1 (\ U it2 Set U 2 = S (x2, 8 2). Since x 2 is non-wandering,
U 2 is self positively recursive and there is a t3 > 3 such that U 2 f\ U2t3
=f: 0. Choose x3 in U2 f\ U2t3 and positive 8 3 < 1/23 such that S (x3 , 8 3) C
U 2 (\ U 2t3 Set U 3 = S (x3 , 8 3}. Proceeding in this fashion we obtain a
sequence of open sets {Un}, with the property that Un ) Un+l and
f\ {Un: n = 1, 2, 3, ... } is a singleton {y} with y E U (prove these facts
which are consequences of construction and completeness of X). We
claim that y is Poisson stable. To see this notice that by construction,
for any integer n > 2, Un (- tn) ( Un-i In particular, since yE Un for
every n, y (- tn) E Un-i Clearly, then the sequences {y (- t 2n)} and
{y (- t2n+i)} both converge toy. Since -t2n-+ oo and - t2n+i-+ - oo
we have y E A+(y) and y E A-(y).
2.16
Remark. The above theorem remains true in locally compact
spaces. The proof is the same as above except that one starts with a
relatively compact neighborhood of the point x. We also remark that
the closure of the set of Poisson stable points is called the set of central
motions.
3. Minimal Sets and Recurrent Points

In this section we study certain subsets of the set of Poisson stable
points. These sets have the property that each point in them returns
to its arbitrary neighborhood with a certain regularity. The points need
not however be periodic. We start with the definition of a minimal set.
3.1
Definition. A set M C X is caJled minimal, if it is non-empty,
closed, and invariant, and no proper subset of M has these properties.
A useful characterization of minimal sets is given by
37
3.2
Theorem. A non-empty set M C X is minimal if and only if
y(x) = M for every xE M.
Proof. Let M be minimal. Then for each x E M, r (x) ( M as M is
closed and invariant. Since r (x) is a closed invariant set we must have
-(x) = M, for otherwise (x) will be a non-empty proper subset of M,
a contradiction to minimality of M. Now assume that r (x) == M for
every x E M. If M is not minimal then there is non-empty closed invariant set PCM with P =FM. Choose xE P. Then y(x) C P, hence
y (x) =F M but x E M. This contradiction shows that M is minimal.
Minimal sets possess the following important property.

3.3
M ==
Theorem.
J'(M).
If MC X
is minimal and
J (M) =f= 0, then
Proof. Let xE J(M), and yE M. Since xE y(y), there is a tE R with

yt E $ (M). Then y E J (M) (-t). The later is however an open subset
of M, so that y is an interior point of M. Thus every point in Mis an
interior point of M. This proves the theorem.
Compact minimal sets may be chara:cterized in various ways. We

have
3.4
Theorem. Let M
following are equivalent.
C X be non-empty and compact. Then the
3.4.1
Mis minimal,
3.4.2
y(x)
3.4.3
3.4.4
y+(x) = M for every xE M,

-y-(x) == M for every x E M,
3.4.5
A+(x)
3.4.6
A-(x)
= M for every xE M,
=
=
M for every xE M,
M for every x EM.
The proof is immediate from elementary considerations and is left

to the reader.
A rest point and a periodic trajectory are examples of compact
minimal sets. By the above theorem, every point in a compact minimal
set is Poisson stable. Example 2. 7 indicated that the closure of a Poisson
stable trajectory need not be a minimal set (in that example the closure
of every Poisson stable trajectory except the rest point is the whole
38
torus, which is not minimal as it contains a rest point). G.D.BIRKHOFF

discovered an intrinsic property of motions in a compact minimal set,
which is usually called the property of recurrence. We now study this
concept.
3.5
Definition. For any x EX, the motion nx is said to be recurrent
if for each e > 0 there exists a T = T (e) > 0, such that
y(x) C S(x[t - T, t
+ T], e)
for all t ER.

3.6
Remark. It is clea~ that if a motion nx is recurrent then every
motion n'Y with y E y (x) is also recurrent. Thus we shall also speak of the
trajectory y (x) being recurrent. Moreover, a point x EX may be called
recurrent whenever the motion nx is recurrent.
3. 7
Exercises.
3. 7.1
Show that every recurrent motion is Poisson stable.
3. 7. 2
In example 2. 7 every point is positively Poisson stable, but
the only recurrent point is the rest point.
That the concept of recurrence is basic in the theory of compact
minimal sets is seen from the following theorem of Birkhoff.
Theorem. Every trajectory in a compact minimal set is re3.8
current. Thus every compact minimal set is the closure of a recurrent
trajectory.
Proof. Let M be a compact minimal set. Suppose that there is an

x E M such that the motion nx is not recurrent. Then there is an e > 0
and sequences {Tn}, {tn}, {rn}, with Tn > 0, Tn ~ + oo, and
Xin~
S(x[tn - Tn, tn
+ TnJ, e),
n = 1, 2, ....
This shows that
e(xTn, x (tn + t)) >
e whenever jtJ
< Tn, n
= 1, 2, ....
The sequences {xtn}, {xTn} are contained in the compact set Mand may
without loss of generality be assumed to be convergent. So let xtn ~ y
and Xin ~ z. Then y, z E y (x) = M. Consider now the motion n'Y. For
any t E R, the inequality
yields
e(yt, z) >: e
39
in view of the previous inequality. Thus

This contradiction proves the theorem.
z~
y (y), i.e., z~ Mas M
= y (y).
3.9
Theorem. If a trajectory y (x) is recurrent and y (x) is compact
then y (x) is also minimal.
Proof. Set y (x) = M. Let if possible M be not minimal. Then there

exists a non-empty compact invariant subset N of M, N =FM. Clearly
x ~ N (otherwise y (x) ( N which is impossible). Now let e (x, N) =
e (> 0). As nx is recurrent, there is a T > 0 such that
3.9.1
y(x) C S(x[t - T, t
+ T], e/3)
for all tE R.
Now choose any yEN. Since yEM=y(x), and y~y(x), we have

yEA+(x) or yE.A-(x). Let yEA+(x)~ Then there is a sequence {tn},
tn ~ + oo, such that xtn ~ y. By the continuity axiom, there is a
~ > 0 such that e(yt, zt) < e/3 whenever e(y, z) < ~ and jtj < T. This
shows that there is an n with
fl (yt, x (tn
+ t)) <
e/3
!tJ < T.
for
From this it follows that
e(x, x (tn + t)) > e(x, yt) -
fl (yt, x (tn
+ t)) >
for
jtj < T.
e - s/3
2e/3
This however contradicts 3.9.1. The theorem is proved.

3.10
Corollary. If the space Xis complete, then the closure y (x) of
any recurrent trajectory y (x) is a compact minimal set.
The proof follows from the observation that the conditions imply
compactness of y (x), so that the result follows from Theorem 3.9. The
details are left to the reader as an exercise.
Another way of defining a recurrent motion is provided via the
concept of a relatively dense set of numbers.
3.11
Definition. A set D of real numbers is called relatively dense if
there is a T > 0 such that
D f\ (t - T, t
+ T) =F 0
for all t ER.
3.12
Theorem. For an x E X let y (x) be compact. Then the motion
nx is recurrent if and only if for each e > 0 the set
Ke= {t: fl (x, xt)

is relatively dense.
< e}
40
Proof. Let for each e > 0 the set Ke be relatively dense. For any
e > 0 there is by definition a Te= T > 0 such that
Ke f\ (t - T, t
+ T) =f= 0
for all t ER.
As y (x) is compact, to show that the motion nx is recurrent we need

-show only that y (x) is minimal. Let r (x) be not minimal. Then there is
--minimal subset M of r (x), M =f= y (x) (see Theorem 4.4). Clearly xt! M
-(otherwise y (x) ( M which will implyy (x) = M). Set e (x, M) = 3e (> 0).
Choose any y E M. Then there is a d > 0 such that e (yt, zt) < e whenever
e(y,z) <~and !ti< T= Te. As yEM(y(x), and yt!y(x), we conclude that yE .A+(x) or y E .A-(x). Let y E .A+(x). Then there is a sequence
{tn}, t,, ~ + oo and xtn ~ y. Thus for all sufficiently large n we have
e(yt, x(tn + t)) < e for It[< T =Ts. But then for tE [tn - T, tn + T]
we have
e (x, xt) > fl (x, y (t - tn)) - e (xt, y (t - tn)) > 3e - e
2e.
This shows that

which is a contradiction. This shows that r (x) is minimal and hence the
motion nx is recurrent. The converse holds trivially. The theorem is
proved.
The above discussion was centered on compact minimal sets. Not
very much is known about the properties of non-compact minimal sets.
It can be shown that all minimal sets in R 2 consist of single trajectories
with empty limit sets (BHATIA and SZEGO [1], Theorem 1.3.6). However,
compact minimal sets contain in general more than one trajectory.
3.13
Theorem. There exist non-compact minimal sets which contain
more than one trajectory.
Proof. Consider example 2. 7 of the dynamical system on a torus T.

Consider the dynamical system obtained by restricting the given dynamical system to the complement of the rest point in this example. The
resulting space Xis not compact, but for each x EX, y (x) = X, so that
X is minimal. This proves the theorem.
3.14
Remark. In the example in the proof above the motions nx are
not recurrent. This shows that Theorem 3.8 is not necessarily true for
non-compact minimal sets.
41

Given a dynamical system (X, R, n) it is in general not known if X
contains minimal sets. The purpose of this section is to give a necessary
and sufficient condition for the existence of compact minimal sets.
4.1
Definition. For any x EX, the motion nx is said to be positively
--Lagrange stable if y+ (x) is compact. Further, if y-(x) is compact, then the
motion nx is called negatively Lagrange stable. It is said to be Lagrange
stable if y (x) is compact.
4.2
Remark. If X =Rn, then the above statements are equivalent
to the sets y+ (x), y- (x), y (x) being bounded, respectively.
4.3
Exercises.
4.3.1
If Xis locally compact, then a motion nx is positively Lagrange
stable if and only if A+ (x) is a non-empty compact set.
If a motion nx is positively Lagrange stable, then A+(x) is
4.3.2
compact and connected.
4.3.3
~
If a motion nx is positively Lagrange stable, then e (xt, A+ (x))
0 as t~
+ oo.
The notion of Lagrange stability plays an important role in the study

of minimal sets as Theorem 3.9 and 3.12 would indicate. That this notion
is also critical in the existence of compact minimal sets is demonstrated
by the following theorems.
4.4
Theorem. Every non-empty compact invariant set contains a
compact minimal set.
Proof. The proof follows by considering the family of all non-empty
closed invariant subsets of the given set. This family is partially ordered
by set inclusion and since the original set is compact, Zorn's lemma
shows that there is a non-empty compact invariant subset which contains
no proper subsets which are closed and invariant. Such a set is then
compact and minimal.
Finally we have
4.5
Theorem. The space X contains a compact minimal set if and
--only if there is an x E X such that either y+ (x) or y-(x) is compact.
Proof. Let y+ (x) be compact. Then A+ (x) is a non-empty compact
invariant subset of y+(x) and hence of X. The existence of a compact
42
minimal set now follows from Theorem 4.4. Similarly if y-(x) is compact,
then A- (x) is a non-empty compact invariant set and the existence of a
compact minimal set follows again from Theorem 4.4. The converse is
trivial, for every trajectory in a compact minimal set is Lagrange stable.
The concepts of Poisson stability, non-wandering points, minimal sets, recurrent
points, and Lagrange stability are all classical. The theorems given here are also
generally well known (see for example G.D. BIRKHOFF [1], and NEMYTSKII and
STEPANOV [1]). The use of prolongational limit sets to describe non-wandering points
is due to J. AusLANDER [3], who also introduced the notion of generalized recurrence
via higher order prolongational limit sets. AusLANDER's work is discussed in VII, 3.
G.D. BIRKHOFF did a considerable amount of work on the notion of central
motions (Remark 2.16) which he introduced. That work is of great interest in ergodic
theory and for this we refer the reader to NEMYTSKII and STEPANOV [1].
Theorems 2.5 and 2.13 do not seem to be in the classical works.
For further developments in the direction of this chapter we mention H. CHU
[1, 2], DOWKER [1], ELLIS [5, 8, 10, 11, 12], ENGLAND and KENT [1], GARCIA and
HEDLUND [1], GOTTSCHALK [11], HILMY [1-7], and SIBIRSKII [1, 2].
Sections 3, 4. The concepts of non-wandering points and recurrent points are due
to G.D. B1RKHOFF [1] who also proved the important relationship between compact minimal sets and recurrent points (Theorems 3.8, 3.9, Corollary 3.10, Theorem 4.4). The important result (Theorem 2.8) on Poisson stable points is from
NEMYTSKII and STEPANOV [1].
It is worthwhile to note that the only recurrent motions in R 2 are the periodic
ones. Consequently, all compact minimal sets in R 2 are the trajectories of periodic
points. In fact 0. HAJEK [1, p. 183] shows that all positively Poisson stable points
in R 2 are periodic. We recall also that the only non-compact minimal sets in R2
consist of a single trajectory with empty positive and negative limit sets (BHATIA
and SzEGo [1], Theorem 1.3.6). Thus Theorem 3.3 implies that all minimal sets in
R 2 have empty interiors. Theorem 3.3 is attributed by NEMYTSKII and STEPANOV
[1] to G. S. TuMARKIN. This theorem poses an important problem, viz., which
phase spaces (for example which manifolds) can be minimal.
For further study of minimal sets in topological transformation groups we refer
to W. H. GOTTSCHALK [11] as a starting point.
A special case of recurrence, namely almost periodicity, is deferred to V, 6,
because it is intimately connected with stability of motion. Note also that the
concepts of this chapter can be generalized to non-metric topological spaces, whereas
almost periodicity requires a uniformity on the space.
Chapter IV
Dispersive Concepts
This chapter is devoted to dynamical systems (X, R, n) which are
notably marked by the absence of Lagrange stable motions or Poisson
stable points or non-wandering points and in general the absence of
recursiveness. Such concepts (for the lack of a better term) may be
called dispersive concepts. The first section contains those concepts
which are definable in terms of the machinery introduced in Chapter II.
The second section studies the theory of parallelizable dynamical
systems.

1.1
Definition. Let x EX. Then
1.1.1
the motion nx is said to be positively Lagrange unstable whenever
y+ (x) is not compact. It is called negatively Lagrange unstable if y- (x)
is not compact. Finally, it is called Lagrange unstable if it is both positively
and negatively Lagrange unstable.
the point x is called positively Poisson unstable whenever
x~A+(x), negatively Poisson unstable whenever x~A-(x), and Poisson
unstable whenever it is both positively and negatively Poisson unstable.
1.1.2
1.1.3
the point xis called wandering whenever xt! J+(x).
We are now in a position to formally define some of the concepts of

dispersiveness of a dynamical system (X, R, n).
1.2
Definition. The dynamical system (X, R, n) is said to be
1.2.1
Lagrange unstable if for each x EX the motion nx is Lagrange
unstable,
1.2.2
Poisson unstable if each x EX is Poisson unstable,
1.2.3
completely unstable if every x EX is wandering,
44
IV. Dispersive Concepts
1.2.4
dispersive if for every pair of points x, y EX there exist neighborhoods Ux of x and U'Y of y such that Ux is not positively recursive
with respect to u'Y.
1.3
Remark. In 1.1.3 it is unnecessary to define concepts such as
positively wandering or negatively wandering with the obvious requirement that x~ ]+ (x) or x t! J-(x). This is so because x~ ]+ (x) if and only
if xt! J-(x). Thus we would not get any new concepts.
1.4
Exercises.
1.4.1
Call a dynamical system (X, R, n) positively Lagrange unstable
whenever for each x E X, y+ (x) is non-compact. Prove that (X, R, n) is
Lagrange unstable if and only if it is positively Lagrange unstable.
Call a dynamical system (X, R, n} positively Poisson unstable
1.4.2
whenever for each xE X, x~A+(x) holds. Show that there exists a
positively Poisson unstable dynamical system which is not Poisson
unstable.
1.4.3
Give examples, of systems for which one of the conditions 1.2.1
through 1.2.4 holds but none of the subsequent ones hold.
Show that each of the conditions 1.2.1 through 1.2.4 implies
1.4.4
the preceding one.
1.4.5
Prove that a point x EX is a wandering point if and only if there
is a neighborhood U of x and a T > 0 such that U f\ Ut = 0 for all t,
jtf > T.
To emphasize the fact that none of the concepts 1.2.1-1.2.4implies
the following one we give some examples.
1.5
Examples.
1.5.1
Consider a dynamical system in a euclidean (x1 , x2}-plane, whose
phase portait is as in Fig.1. 5.2. The unit circle contains a rest point panda
trajectory y such that for each point q E y we have A+ (q) = A-(q) = {P }.
All trajectories in the interior of the unit circle (= {P} V y) have the
same property as y. All trajectories in the exterior of the unit circle spiral
to the unit circle as t---? + oo, so that for each point q in the exterior
of the unit circle we have A+(q) = {P} Vy, and A-(q) = 0. Notice that
if we consider the dynamical system obtained from this one by deleting
the rest point p (the dynamical system is thus defined on R 2 - {P }) then
this system is Lagrange unstable and Poisson unstable, but it is not
45
completely unstable because for each q E 'Y we have J+(q) = J', 1.e.,
q E J+(q).
x,
Fig.1.5.2
1.5.3
Consider a planar dynamical system defined by the differential
equations (cartesian coordinates)
X1 =
2
Sln X 2 , X 2 =COS X 2
The dynamical system contains, in particular, trajectories 'Yk given by

'Yk = {(x1 , x2): x2 =kn
+ n/2},
k = 0, 1, 2, ....
These are lines parallel to the x1-axis. Between any two consecutive
'Yk's the trajectories are given by y = {(x1 , x2): x1 + c =sec x2}, where c
is some constant depending on the trajectory. The phase portrait between
the lines x2 = -n/2 and x 2 = +n/2 is shown in Fig.1.5.4. This system
Xz
n/2
~1
-rc/2
Fig.1.5.4
46
is completely unstable but not dispersive. This follows by noticing, for

example, that for each point p E y _ 1 , ]+ (p) == y0 , and for all other points
,pf! yk, J+(p) == 0. Thus Pf! J+(p) for every p E R 2 However, if p E y _ 1
and q E y0 , then every neighborhood of q is recursive with respect to any
neighborhood of p.
1.5.5
We now give an example of a dispersive dynamical system which
will be shown to be not parallelizable when this notion is introduced in
the next section.
Consider a dynamical system defined in R 2 by the differential equations
dt = f (X1' X2),
dx1
dx2
dt
O,
where I (x1 , x2) is continuous, and moreover f (x1 , x2) = 0 whenever the
point (x1' x2) is of the form (n, 1/n) with n a positive integer. For simplicity we assume that f (x1' x2 ) > 0 for all other points. The phase portrait
is as shown in Fig.1.5.6. Let us now consider the dynamical system
obtained from the above one by deleting the sets
In= {(x1' x2}: x1 < n, x2 == 1/n},
==
1, 2, 3, ...
from the plane R 2 This system is dispersive.
(1,1)
--------+------------~-
(2,1/z)
(J,1/3)
-----+-------------(0.Dl
Fig.1.5.6
1.6
Remark. For dynamical systems defined by differential equations in the euclidean plane R 2 the concept of Lagrange instability and
the concept of wandering are equivalent. This may easily be proved
using the Poincare-Bendixson theory of planar systems.
The notion of a dispersive dynamical system is perhaps the most

important of the notions introduced in this section. We start with several
of its characterizations.
1.7
Theorem. For a given (X, R, n) the following are equivalent.
1. 7.1
(X, R, n) is dispersive.
47
1. 7.2
For any two points x, y in X there are neighborhoods Ux of x
and UY of y and a constant T > 0 such that Ux (\ U/ = 0 for all t,
ftj > T.
1.7.3
For any two points x, yin X,
y~
J+(x).
The proof easily follows from the definitions and is left to the reader.
We now develop a criterion for dispersive flows.
1.8
Theorem. A dynamical system (X, R, n) is dispersive if and
only if for each x EX, J+(x) == 0.
Proof. Let (X, R, n) be dispersive. Let if possible x E X and J+ (x) =F 0.

Then if y E J+(x), there are sequences {xn}, {tn}, xn ~ x, tn ~ + oo,
and xntn ~ y. This shows that for any neighborhoods Ux, U'Y of x and y
respectively Uin (\ UY =f=. 0 as the element xntn = Yn is contained in this
intersection. Since tn ~ + oo, this contradicts the definition of a dispersive flow as UY is positively recursive relative to Ux. Hence ]+ (x) == 0 for
each xE X. Conversely, let J+(x) == 0 for each xE X. We claim that for
{x, y} C X there are neighborhoods Ux of x.and U'Y of y and a T > 0
such that Ui f\ UY== 0 for all t > T. For if not, then there will be sequences {xn}, {Yn}, {tn}, xn ~ X, Yn == Xin' Yn ~ y, and tn ~ + oo, so
that y E J+(x). This is absurd as J+(x) = 0. Note that UY is not positively
recursive relative to Ux. This proves the theorem.
1.9
Remark. Using the above theorem the dynamical system described in Example 1.5.5 is clearly seen to be dispersive.
vVe now give another criterion for dispersive flows, which is sometimes
more useful than the one given above.
1.10
Theorem. The dynamical system (X, R, n) is dispersive if and
only if for each x E X, D+ (x) = y+ (x) and there are no rest points or
periodic trajectories.
Proof. If (X, R, n) is dispersive, then ]+ (x) = 0 for each x E X.

Consequently D+ (x) = y+ (x) V J+ (x) == y+ (x) for each x E X, and there
are no rest points or periodic orbits. For if x is a rest point or 'Y (x) is
periodic then y (x) ==A+ (x) C ]+ (x). Conversely, if D+ (x) = y+ (x) and
there are no rest points or periodic orbits, then ]+ (x) = 0. For indeed
D+(x) == y+(x) V J+(x) = y~(x) implies that J+(x) C y+(x). J+(x) being
closed and invariant, we conclude that if J+ (x) is not empty, then y (x) (
]+ (x) Cy+ (x), i.e. y (x) == y+ (x). This shows that if r < 0 is arbitrary,
then there is a r' > 0 such that xr = xr', i.e. x == x (r' - r). Since
48
>
0, the last equality shows that the trajectory 'Y (x) is closed
and has a period 7:1 - T:. Since we assumed that there are no rest points
or periodic orbits, we have arrived at a contradiction. Thus J+ (x) = 0
for each x EX, and the dynamical system is dispersive. The theorem is
proved.
T:
7:' -
We will now introduce parallelizable dynamical systems.

2.1
Definition. A dynamical system (X, R, n) is called parallelizable
if there exists a set S C X and a homeomorphism h: X ~ S X R such
that SR== X and h(xt) = (x, t) for every x ES and t ER.
For the study of parallelizable dynamical systems we need to develop
a theory of sections. Thus we start with
2.2
Definition. A set SC X is called a section of (X, R, n) if for
each x E X there is a unique ,; (x) such that X't' (x) E S.
Not every dynamical system has a section. Indeed any (X, R, n)
has a section if and only if it has no rest points or periodic orbits.
The function T: (x) will be basic in what follows. In general 't' (x) is
not continuous, but the existence of a section S with continuous 't' (x)
implies certain properties of the dynamical system which we sum up
in the following lemma.
2.3
Lemma. If S is a section of the dynamical system (X, R, n)
with T: (x) continuous on X then
2.3.1
Sis closed in X,
2.3.2
Sis connected, arcwise coanected, simply connected if and only
if X is respectively connected, arcwise connected, simply connected.
2.3.3
If K ( S is closed in S, then Kt is closed in X for every t E R.
2.3.4
If KC Sis open in S, then KI, where I is any open interval in
R, is open in X.
Proof of 2.3.1. If {xn} in S, and xn ~ x EX, then 't'(xn) ~ 't'(x) by

continuity. Since T: (xn) = 0 for each n, we get 't' (x) = 0. Thus X't' (x) =
x 0 = x ES by definition of 't' (x) and S. That is, Sis, closed in X.
Proof of 2.3.2. We shall prove only the first part. The interested reader
can supply the proofs of the remaining parts. Let S be not connected.
Then there are disjoint closed sets 5 1 , 5 2 such that 5 1 V 5 2 = S. As
49
X =SR, we have X = S 1R V S 2R. Note that S 1R and S 2R are disjoint.

We prove that they are closed. Consider S 1R, and let {xn} in S 1R, xn-+ x.
Then r (xn) -+ 7: (x), and by the continuity axiom xnr (xn) -+ xr (x). Since
{xnr(xn)} is in S 1 and S 1 is closed we conclude that xr(x) E Sr Then
x = xr (x) (-r (x)) E xr (x) R C S 1R. Thus S 1R is closed. Similarly we
can prove that S2R is closed. Thus X being the union of two disjoint nonempty closed sets is not connected. We conclude that if X is connected,
so must be S. The converse follows similarly.
Proof of 2.3.3. The proof follows by observing that if K is closed in S,
then it is closed in X.
Proof of 2.3.4. The simple proof is left to the reader.
The following theorem now gives a criterion for parallelizable dynamical systems.
2.4
Theorem. A dynamical system (X, R, n) is parallelizable if and
only if it has a section S with r (x) continuous on X.
Proof. Sufficiency. Indeed SR= X. Define h: X-+ SxR by h(x) =
(xr(x), -r(x)). Then h is 1 - 1 and continuous by the continuity of
'l' (x) and the contin~ity axiom. The inverse h-1 : S X R -+ X is given by
h,-1 (x, t) = xt and is clearly continuous. This h is thus a homeomorphism
of X onto SxR, i.e. (X, R, n) is parallelizable. To see necessity, we note
that if the dynamical system is parallelizable, then the set Sin its definition is a section of X. Since for any x EX, x = yt for some y ES and
t E R, we set 'l' (x) = -t. Then continuity of r (x) follows from that of h.
The proof is completed.
2.5
Remark. The above theorem shows that the dynamical system
of Example 1.5.5 is not parallelizable. Notice however that the phase
space in this example is not locally compact.
The following is the most important theorem in this section.

2.6
Theorem. A dynamical system (X, R, n) on a locally compact
separable metric space Xis parallelizable if and only if it is dispersive.
The proof of this theorem depends on properties of certain sections
which we now describe.
2. 7
Definition. An open set U in X will be called a tube, if there exists
a 'l' > 0 and a subset S C U such that
2.7.1
4
SIT CU, and
50
2.7.2
for each xE U there is a unique r(x), jr(x) I< r, such that
xr (x) E S. Here IT== (-r, r).
It is clear that if 2.7.1 and 2.7.2 hold, then U =SIT. U is also called
a r-tube with section S, and S a (r - U)-section of the tube U.
If IT== R, then U is an CX)-tube, and S an (CX) - U)-section. In this
last case indeed U =SR. Note also that the function r(x) which maps U
into IT is 1 - 1 along each trajectory in U.
2.8
Lemma. Let Ube a r-tube with section S. If KC Sis compact,
then the function r(x) is continuous on Kis for any s, 0 < s < r.
Proof. To show is: if {xn} in Kls and Xn--*XEKls, then r(xn)--*T(x).

Note that the sequence {xn'l' (xn)} is in K, and we may assume that it is
convergent as K is compact. Further {r(xn)} is in Is and hence bounded
so that we may also assume that {r(xn)} is convergent. Thus let xnr(xn)
--* x* EK, and r(xn)--* r* E Is. Since Xn--* x, we have x* = xr*. Since
jr*j < s < r, we haver*= r(x) by uniqueness. The lemma is proved.
The next theorem shows that if x E X is not a rest point, then there
is a tube containing x.
2.9
Theorem. If x EX is not a rest point, then there exists a tube
containing x.
>
Proof. Since x is not a rest point, there is a T0

0. Consider the function
>
0 such that e(x, xT0 )
t+T0
1JJ (y,
t)
e (x, YT) dr.
It follows that
1P (y, t1
+ t2)
ti +t2+ T 0
t 2+ T 0
== J (] (x, yr) dr == J (] (x, y (r

ti +t2
+ t1)) d'l'
t2
t 2 +To
== J e (x, ytl (r)) dr = 1P (yt1' t2).

t2
Further the function 1P (y, t) is continuous in (y, t) and has the continuous
partial derivative
"Pt (y, t) = e (x, y (t
T0))
e (x, yt).
Since
= e(x, xT0 ) > 0,

"Pt (y, 0) > 0 for y E S (x, e).
"Pt (x, 0)
there is an e > 0 such that

Define r 0 > 0
such that x [-3r0 , 3r0 ] CS (x, e). Then, in particular, 1P (x, r 0} > 1P (x, 0)
> 1P (x, -r0}. Now choose C> 0 such that
(S [xr0 ,
CJ
VS [x (-r0 ),
CJ)
CS (x, e),
51
and such that for y E S (xr0 , C) we have 1P (y, 0) > 1P (x, 0), and for y E
S(x(-r0 ), C) we have 't/J(y, 0) < 1Jl(X, 0). Finally determine ~ > 0 such
that
S [x, ~] r 0 CS (xr0 , C),
S [x, ~] (-r0 ) CS (x (-r0 ), C),
and
S [x, ~] [ -3r0 , 3r0 ] CS (x, e).
We will show that if y E S [ x, ~], then there is exactly one r (y), Ir (y) I
<To such that 1P (y, r (y)) = 1P (x, 0). This follows from the fact that
1P (y, t) = 1P (yt, 0) is a strictly increasing function of t, and 1P (y, r 0) >
1P (x, 0) > 1P (y, -r0).
Consider now the open set U == S (x, ~) /To' and set S = {y E U:
tp(y, 0) == 1JJ(X, O)}. We claim that Sis a (2r0 - U)-section. For this we
need prove that if y E U, then there is a unique r(y), lr(y) I< 2r0 such
that yr (y) E S. Indeed for any y E U, there is a t' It' I < r 0 such that
y' == yt' E S (x, ~), and for y' E S (x, ~) there is a t", It" I < r 0 , such that
y't" ES. Thus y(t' + t") = yr(y) ES, where r(y) = t' + t", and lr(y) I
< jt'I + jt"I < 2r0 Now let if possible there be two numbers, r' (y),
r" (y), Ir' (y) [ < 2r0 , Ir" (y) J < 2r0 , such that yr' (y) E S and yr" (y) E S,
and let y' = yt' E S (x, '5), where It' I < r 0 Then 1P (y', y' (y) - t') =
1P (y, y' (y)) = 1P (yr' (y), 0), and 1P (y', r" (y) - t') = 1P (y, r" (y)) =
1P (yy" (y), 0), so that 1P (y', r' (y) - t') = 1P (y', r" (y) - t') = 1P (x, 0). Now
IT' (y) - t' [ < 3T0 , and Jr" (y) - t' J < 3r0 , and "Pt (y', t) > 0 for It J < 3r0 ,
i.e., 1P (y', t) is strictly increasing for ltl < 3r0 Hence r' (y) - t' = r" (y)
- t', or r' (y) = r" (y). The theorem is proved.
2.10
Remark. If X is locally compact, then we can restrict ~ > 0
in the above proof to ensure that S [x, ~]is compact. Thus the (2r0 - U)section S constructed in the above proof will also be locally compact.
By Lemma 2.8 we may further assume the function 'l' (x) corresponding
to the section S to be continuous on U.
In fact the following more general theorem can now be proved.
2.11
Theorem. Let x EX be not a rest point. Let r > 0 be given,
restricted only by r < Tf 4 if the motion nx is periodic with least period
T. Then there exists a tube U containing x with a (r - U)-section S.
Further, if X is locally compact, then the function r (x) corresponding
to the section Scan be assumed continuous on U.
The proof of this theorem is left to the reader.
For wandering points x E X one can prove:
4*
52
2.12
Theorem. If xE X is a wandering point, i.e., x~ ]+(x), and
moreover Xis locally compact, then there exists a tube U containing x,
with an (CX) - U)-section S, and with r(x) continuous on U.
Proof. Indeed there is a tube W containing x, with a (r - W)-section
S, and r(x) continuous on W. Since xis wandering, we claim that there
is a '5 > 0 such that S (x, '5) (\ S = S* is an (CX) - U)-section of the open
set U = S*R, which is an oo-tube containing x. To see this notice that
there is a '5 > 0, such that every trajectory 'Y (y) with y E S*, intersects
S* only at the pointy. For otherwise, there will be a sequence {Yn} in S,
Yn ~ x, and a sequence {tn} in R, tn ~ + CX) (or tn ~ - CX)), such that
Yntn ~ x, i.e., either x E J+(x), or x E J-(x), both of which are ruled out
by the assumption that xis wandering. We have thus shown that there
is a '5 > 0 such that S* = S (x, '5) (\ Sis an (CX) - U)-section of U == S* R.
Furthermore U is open, and continuity of r (x) on U follows from its
continuity on W (\ U, and continuity of the phase map n. This we leave
to the reader to verify. The theorem is proved.
For further development we need the following definition.

2.13
Definition. Given an open CX)-tube U with a section S and r (x)
continuous on U, and given sets N, K, NC K ( S, where N is open
in S and K is compact, we shall call KR the compactly based tube over
K. Then indeed T: (x) restricted to KR is continuous on KR.
2.14
Remark. A compactly based tube need not be closed in X. As an
example, one may consider a dynamical system defined in the euclidean
plane R 2, as shown in Fig. 2.14.1. The x2-axis consists entirely of rest
points, all other trajectories are parallel to the xi-axis, with each having
a rest point on the x2-axis as the only point in its positive limit set,
whereas the negative limit sets are empty. Here, for example the set
{ (x11 x2) : 0 < x2 < 1, x1 > O} is a compactly based tube, which is not
closed in X.
Xz+
~~~~~~~-!
______________________
--------~---~-'----------------------!
I
--~----...._--~-,~------------------~
I
~
~~---------~-,~-------------------
Fig.2.14.1
We can now prove the following theorem.
53
2.15
Theorem. If X is locally compact and separable, and if every
x E X is a wandering point, then there exists a countable covering
{KnR} of X, by compactly based tubes KnR each with Tn (x) continuous
on KnR.
Proof. The proof is immediate, when we notice that by using Theorem 2.11, one can find a compactly based tube containing a wandering
point of X. The rest follows by the assumption of separability of X.
We gave an example above to show that a compactly based tube
need not be closed in X. One may wonder if for a wandering dynamical
system (X, R, n), a compactly based tube is not closed. Here is a counterexample.
Example. Consider again Example 1.5.3. Any compactly based

2.16
tube containing a point p E y _ 1 is not closed, because its closure will
contain y0 which is not in such a tube. This is an example of a wandering
dynamical system which is not dispersive. In the case that (X, R, n)
is dispersive one obtains the following:
2.17
Lemma. A compactly based oo-tube U with section K of a
dispersive dynamical system (X, R, n) is closed in X.
Proof. U =KR and if {xn} is a sequence in KR, then there are sequences {Yn} inK and {rn} in R such that xn = Yn7:n. We may assume that
Yn ~ y EK as K is compact. If now xn ~ x, we claim that the sequence
{rn} is bounded, so that x E yR C KR. For otherwise if {r,.} contains an
unbounded subsequence {rnk}, say Tnk ~ + oo, then clearly x E J+ (y),
which is absurd, as J+(y) = 0 for each y EX by Theorem 1.8. The lemma
is proved.
We now prove the last lemma required to prove Theorem 2.6.
2.18
Lemma. Let U1 , U 2 be two compactly based tubes of a dispersive
dynamical system with sections K 11 K 2 and continuous functions r 1 (x)
and -r2 (x) respectively. If U 1 (\ U 2 -=!= 0 then U = U 1 V U 2 is a compactly based tube with a section K ) K 1 and a continuous function r (x).
Moreover, if the time distance between K 1 and K 2 along orbits in
U1 f\ U2 is less than r (> 0), the time distance between Kand K 2 along
orbits in U is also less than r.
Proof. U1 and U2 are invariant and closed. Therefore U1 f\ U2 is

invariant and closed. Further, K 2 (\ U1 is compact and non-empty, as
54
shown in Fig. 2.18.1. Set 5 2 = K 2 f\ U1 and 5 1 = K 1 f\ U1 Any orbit

in U1 (\ U2 intersects K 2 and hence 5 2 in exactly one point, and also
intersects K 1 and hence 5 1 in exactly one point. Thus for any x E U1 (\ U 2
we have r 1 (x) = r 2 {x) + r 1 {xr2 (x)). This is so because xr1 (x) = xr2 (x)
(r1 (xr2 (x))) = x (r2 (x) + r 1 (xr2 (x))), and there are no rest points or
periodic orbits. The function r 1 (x) is continuous on 5 2 (which is compact),
and by Tietze's theorem it can be extended to a continuous function
r (x) defined on K 2 , where 'l' (x) == T1 (x) for x E 5 2 Further if Ir 1 (x) I < r
for x E 5 2 , we can have lr(x) I< 'l' on K 2 Notice now that {xr(x): xE S 2}
= 5 11 and r (x) being continuous {X'l' (x): x E K 2} is compact as K 2 is
compact. We set now K = K 1 V {xr (x) : x E K 2}, and define r* (x) on
KR = K 1R V K 2R as follows. T* (x) = r 1 (x) for x E K 1R, and T* (x) =
r 2 (x) + r (xr2 (x)) if x E K 2R. r* (x) is continuous on KR and we need
only verify that if x E U1 (\ U2 , then T1 (x) = T2 (x) + r (xT2 (x)), which
has already been proved. The lemma is proved.
Fig. 2.18.1
Proof of Theorem 2.6. Only the sufficiency part needs proof. It

is sufficient to prove that X has a section S with r(x) continuous on X.
By Theorem 2.15 there is a countable covering {Un} of X by compactly
based tubes Un with sections Kn and continuous functions r n(x). We replace
this covering by a like covering {Un} of compactly based tubes which we
construct as follows. Set Kl= K 1 , and U1 = U 1 . Beginning with u1
and U2 we use Lemma 2.18 to enlarge K 1 to a compact set K 2 , thus
obtaining the compactly based tube U2 = U1 V U2 with r 2 (x) continuous
on U 2 This leaves K 1 unaltered. Having found un, we take it together
with U,,,+ 1 and construct similarly un+i with Kn+i) Kn, and rn+i (x)
continuous on U"'+ 1 Now set S = V Kn, then X =SR, and the function T (x) defined by T (x) = Tn (x) for x E un is continuous on X, with
the property that xr(x) ES. Moreover r(x) is unique for each x EX.
Thus X has a section S with continuous r(x) defined on X. The system
(X, R, n) is thus parallelizable and the theorem is proved.
2.19
2.20
persive.
Exercise. Show that a parallelizable dynamical system is dis-
55

The concepts in this chapter are all classical.
Section 1. The important concept of dispersiveness may be attributed to WHITNEY
[1, 2, 3], but it also appears in NEMYTSKII [10]. All definitions (except 1.2.4)
appear in NEMYTSKII and STEPANOV [1]. The classical definition of a wandering
point is contained in Exercise 1.4.5. Theorems 1.7, 1.8, 1.10wereproved by BHATIA
[3]. Theorem 1.10 was first conjectured by J. AUSLANDER [3].
Section 2. The concept of a parallelizable system is due to WHITNEY [1, 2, 3]. The
crucial Theorem 2.6 was obtained by NEMYTSKII and STEPANOV [1] in a different
form using the concept of complete instability and Nemytskii's notion of an improper saddle point (or a saddle point at infinity). The present form of the theorem is
found in NEMYTSKII [10] who showed that dispersiveness was equivalent to the
conjunction of complete instability and absence of improper saddle points. The
development in this section follows closely that of ANTOSIEWicz and DUGUNDJI [1].
Theorems 2.9 and 2.11 are due to BEBUTOV (see NEMYTSKII and STEPANOV [1]).
For further developments on the theory of sections see T. URA [6].
We would like to emphasize that in the discussion of dispersive concepts these
distinct points of view together with distinct techniques have been advanced:
(i) the absence of recursive motions, (ii) the absence of improper saddle points, and
(iii) the theory of sections leading to parallel systems. All these are intimately
connected with the theory of separatrices which is still in its infancy (see MARKUS
[11]).
Chapter V
Stability Theory
This chapter is devoted to the study of various notions of stability
and attraction and the characterization of some of them in terms of the
so-called Liapunov functions.
Sections 1, 2, 3 expose the theory for compact sets in locally compact
phase spaces. This theory seems to be fairly complete.
Section 4 is devoted to a study of stability and attraction properties
of closed sets. Here we need not to restrict to locally compact spaces.
However, it seemed to us quite unnecessary to try to document all the
stability and attraction phenomena that can occur in this case, so that
we have restricted ourselves to only those which seemed to be of most
interest.
Section 5 exposes a concept of relative stability that may be useful
in several situations. The concepts of stability studied in sections 1-5
may be termed as orbital stability concepts in contrast to the concept
of stability of motion and its effect on recurrent motions as studied
in section 6.
1. Stability and Attraction for Compact Sets

Throughout this section, the phase space X will be assumed to be
locally compact. Some of the results do not require this assumption. Such
results will be indicated in remarks or exercises. M will denote a nonempty compact subset of X.
1.1
Definitions. With a given M we associate the sets
1.1.1
Aw(M) = {xE X: A+(x)
1.1.2
A (M)
1.1.3
Au (M)
f"\
M =t= 0},
= {xE X: A+(x) =t= 0 and A+(x) CM},
{x EX:
J+ (x) =t= 0 and J+ (x) ( M}.
57
The sets Aw(M), A (M), Au(M) are respectively called the region of weak
attraction, attraction, and uniform attraction of the set M. Moreover, any
point x in Aw (M), A (M), or Au (M) may respectively be said to be weakly
attracted, attracted, or uniformly attracted to M.
The motivation for the above definition is the following proposition.
1.2
Proposition. Given M, a point x is
1.2.1
weakly attracted to M if and only if there is a sequence {tn}
in R with tn-?- + oo and (! (xtn, M) -?- 0,
1. 2. 2
attracted to M if and only if (! (xt, M) -?- 0 as t-?-
+ oo,
1.2.3
uniformly attracted to M if and only if for every neighborhood
V of 1\1 there is a neighborhood U of x and a T > 0 with Ut ( V for
t > T.
Proof. 1.2.1 and 1.2.2 easily follow from the definitions. We prove
1.2.3. First assume that for an xE X, J+(x) =F 0 and J+(x) CM. Then
A+(x) =f= 0 and A+(x) CM (II, 4.5). Assume now that there is a
neighborhood V = S (M, cX} of M (cX > 0) such that for every neighborhood U of x and T > 0 there is a t > T with Ut ct V. By 1.2.2 there
is a t0 > 0 such that xt E V for t > t0 This shows that there exist
sequences {tn}, {rn} in R, tn < Tn, tn-?- + oo and a sequence {xn} in X,
Xn-?-X, such that xntnES(M,cX), but XnTnEH(M,cX). Since H(M,cX)
may be assumed compact (the space is locally compact) we conclude that
there is a sequence {xn}, xn-?- x, and a sequence {Tn}, Tn ~ + oo, with
xnTn-?- y EH (M, cX). Then y E J+ (x) but y ~ M. This is a contradiction.
Now assume the converse requirement that for every neighborhood V
of M there is a neighborhood U of x and a T > 0 with Ut C V for
t > T. Then indeed Ut C V. This shows indeed that xt E V for t > T,
and since we may take V to be compact, we have A+(x) =f= 0. Consequently J+ (x) =f= 0. Clearly J+ (x) C U [t, + oo) for every neighborhood
U of x and every tE R+. Hence J+(x) CV for every neighborhood V
of M. Thus J+(x) C (\ {V: Vis a neighborhood of M} = M. This proves
the proposition.
The following theorem expresses an elementary property of the sets

Aw(M), A (M), and Au(M).
1.3
Theorem. For any given M,
1.3.1
Aw(M))A(M))Au(M),
1.3.2
the sets Aw (M), A (M), and Au (M)
are invariant.
58
V. Stability Theory
Proof. 1.3.1 is quite apparent from Definition 1.1, and 1.3.2 is

easily seen from the following lemma.
1.4
Lemma. Let X be any metric space (not necessarily locally
compact) and x E X. Then
1.4.1
A+ (x)
1.4.2
J+ (x)
= A+ (x) t = A+ (xt) for every t E R,

= ]+ (x) t = ]+ (xt) for every t ER.
Proof. The first equalities in 1.4.1 and 1.4.2 follow from invariance
of the sets A+(x) and J+(x). To see the second equalities, consider, for
example, the case A+ (x) t ==A+ (xt). Let z EA+ (x) t. Then there is a
y E A+(x) with z == yt and a sequence {tn}, tn-?- + oo, with xtn-?- y.
Then by the continuity axiom xtn (t) -?- yt. But xtn (t) == xt (tn) and since
tn--+ + oo, we must have yt EA+ (xt). Thus A+ (x) t CA+ (xt). The argument is clearly reversible, so that also A+ (xt) CA+ (x) t. This proves
1.4.1. The proof of the second equality in 1.4.2 is entirely analogous.
We now introduce the fundamental definitions of attraction and stability.
1.0
Definition. A given set M is said to be
1.5.1
a weak attractor if Aw (M) is a neighborhood of M,
1.5.2
an attractor if A (M) is a neighborhood of M,
1.5.3
a uniform attractor if Au (M) is a neighborhood of M,
1.5.4
stable if every neighborhood U of M contains a positively invariant neighborhood V of M,
1.5.5
asymptotically stable if it is stable and is an attractor,
1.5. 6
unstable, if it is not stable.
1.6
Remark. A weak attractor will be called a global weak attractor
whenever Aw (M) = X. Similarly for attractors, uniform attractors, or
asymptotically stable sets, the adjective global is used to indicate that
the corresponding region of attraction is the whole space.
We will presently show that the concepts introduced have distinct

meaning. For this we consider some examples.
1.7
Examples.
1.7.1
Consider Example II, 3.3.1. Take any singleton {x} with x in
the unit circle. Then {x} is a weak attractor but it is none of the other
kind of objects introduced in 1.5.
1.7.2
Consider Example II, 3.3.4. The set M= {A,B} is again a
weak attractor but none of the other kinds of sets in 1.5.
59
1. 7.3
In the above example let M be the set consisting of the points
A and B and the trajectory y 1 . Then again M is a weak attractor but
none of the rest.
1. 7.4
Consider a planar dynamical system defined by the differential
equations (in polar coordinates)
i. 7. 5
== r (1 - r),
iJ = sin2 ( J2) .
The trajectories are shown in Fig.1. 7.6. These consist of two rest points
p1 = (0, 0) and p2 = (1, 0), a trajectory y on the unit circle with {p2}
as the positive and the negative limit set of all points on the unit circle.
All points p, P =F p1 , have A+(p) = {P2}. Thus {P 2} is an attractor. But
it is neither stable nor a uniform attractor. Note that for any p =
(cX, 0), cX > 0, J+ (p) is the unit circle.
Fig.1.7.6
1. 7. 7.
In Examples 1. 7.1, 1. 7.2, and 1. 7.4, if we take M to be the unit
circle, then M is clearly asymptotically stable and also is a uniform
attractor. As an example of a uniform attractor which is not stable we
consider in the present examples any set consisting of the unit circle
and sny po int p different from the origin, but not on the unit circle.
1. 7.8
Fig.1. 7.9 describes an example of a global attractor in the plane
which is not asymptotically stable. The point 0 is the only positive limit
point of each point in the plane.
Fig.1.7.9
60
V. Stability Theory
We are now ready to prove some theorems about the concepts introduced.
1.8
Theorem. If Mis a weak attractor, attractor, or uniform attractor, then the corresponding set Aw (M), A (M), or Au (M) is open (indeed
an open neighborhood of M).
Proof. Let N denote any one of the sets Aw (M), A (M), or Au (M).
Then N is an invariant neighborhood of M. Consequently 8N is invariant
and disjoint from Mand is indeed closed. Note now that for each x EN,
A+(x) f\ M =f= 0, whereas for each x E aN, A+(x) C oN. Since aN f\ M
= 0 we conclude that N f\ aN = 0. Thus N is open.
1.9
Theorem. If M is stable then it is positively invariant. Consequently, if Mis a singleton {x}, then xis a rest point.
Proof. Clearly M is the intersection of positively invariant neighborhoods of M. Hence by II, 1.2, Mis positively invariant. Finally, if M = {x},
then {x} = y+ (x). Hence by II, 2. 2, x is a rest point.
In 1.5 we introduced asymptotic stability as the conjunction of stability and the attractor property. The next theorems are aimed at showing
that the conjunction of stability with weak attraction or uniform attraction does not lead to any new concepts. We first prove the following
important lemma.
1.10
Lemma. Let X be an arbitrary metric space. Let x E X and
roEA+(x). Then J+(x) C J+(ro).
Proof. Given ro E A+(x), and any y E J+(x), there exist sequences

{'r:}, T: ~ + oo, X't'~ ~ ro, and {t~} and {xn}, xn ~ x, t~ ~ + oo, Xi~~ y.
We can assume, if necessary by choosing subsequences, that
T: > n
for each n. Consider for each fixed k, k = 1, 2, ... , the sequence {xnT~}.
By the continuity axiom Xn't'~ ~ XT~, k = 1, 2, .... We may, therefore,
assume without loss of generality that for each fixed k, (! (xn't'~, XT~) < 1/k
for n > k. This shows that xn't': ~ OJ, because (! (ro, Xn't':) < (! (ro, XT~)
+ (!(X't'~, Xn't'~) <(!(OJ, XT:) + 1/n. Now notice that xnt: = Xn't'~(t~ - T~),
and xnt: ~ y, xn't':-+ ro, and t: - T~ > n. HenceyE J+(ro). AsyE J+(x)
was arbitrary, we have J+(x) C /+(ro), and the lemma is proved.
t: -
As an important corollary we have

1.11
Corollary. Given Mand x E Aw (M), we have
J+(x) ( J+(M) ( D+(M).
61
Proof. Indeed for x E Aco (M), A+(x) f\ M =f= 0. So take any ro E

A+ (x) (\ M and apply the above lemma.
We now give an important characterization of stability of a set M.
1.12
Theorem. A set M is stable if and only if
D+(M) = M.
Proof. Let JJ+ (M) = M, and suppose if possible that M is not stable.
Then there is an e > 0, a sequence {xn}, and a sequence {tn}, with tn > 0,
e(xn, M)-+ 0, and e(xin' M) > e. We may assume without loss of
generality that e > 0 has been chosen so small that S [M, e] and hence
H (M, e) is compact (this is possible as X is locally compact). Further,
we may assume that xn--+ x EM. We can now choose a sequence {Tn},
0 < l"n < tn, such that Xnl"n EH (M, e), n = 1, 2, .... Since H (M, e) is
compact, we may assume that Xnl"n ~ y EH (M, e). Then clearly y E
D+(x) ( D+(M), but y~ M. This contradiction shows that Mis stable.
Now assume that Mis stable. Given any neighborhood V of M there is
a positively invariant neighborhood U of M with U ( V. Since for any
-x E M, D+ (x) C WR+ for any neighborhood W of x, we get JJ+ (x) ( U
since U is positively invariant. Thus D+ (M) C V for any neighborhood
V of M. Hence .D+ (M) C f\ {V: V is a neighborhood of M} = M as M
is compact. Since M C D+ (M) holds always, we have D+ (M) = M.
This proves the theorem.
In view of Theorem 1.12 we can give the following definition.
Definition. If a given set M is unstable, the non-empty set
1.13
D+(M) - M will be called the region of instability of M. If D+(M) is
not compact, then M is said to be globally unstable.
1.14
Theorem. If M is stable and is a weak attractor, then M is an
attractor and consequently asymptotically stable.
Proof. Indeed Aco (M) is a neighborhood of M. We need show that

Aco(M) CA (M). Let xE Aw(M). Then A+(x) =f= 0 and A+(x) C J+(x)
C D+(M) by Corollary 1.11. Since Mis stable we have D+(M) = M by
Theorem 1.12, and consequently A+ (x) C M. Thus M is an attractor.
Theorem. Let M be positively invariant and a uniform attrac1.15
tor, then Mis stable. Consequently Mis asymptotically stable.
Proof. Note that MC Au (M). Thus, in particular, D+(M) =MR+

V J+(M) = M V M = M, as M =MR+ by positive invariance of M,
and J+(M) CM, as MC Au(M). Hence by Theorem 1.12, Mis stable.
62
V. Stability Theory
1.16
Theorem. If M is asymptotically stable then M is a uniform
attractor.
Proof. We need show that A(M) CAu(M). Let xEA(M). Then

A+(x) =F 0 andA+(x) CM. Choose any ro E A+(x) and apply Lemma 1.10
to get J+(x) C J+(ro) C J+(M) C D+(M). But D+(M) = M by stability
of M. Hence J+(x) CM for x EA (M). This proves the theorem.
The various implications that have been proved above are shown in
the following diagram.
1.17
Diagram.
Positively
invariant
+-----Uniform
attractor
Stable
Asymptotically stable
-----+
Weak
attractor
Attractor
1.18
Exercises.
1.18.1
If M is stable, then J-(X -M) (\ M
0.
1.18.2 If X is locally compact, then a compact set M is stable if and

only if J-(X-M) (\ M == 0.
1.18.3 In an arbitrary metric space X let M be a compact subset of X.
Define Aw (M), A (M), Au (M) as before. Show that they are invariant
sets. Further if M is a weak attractor or attractor then the respective
sets Aw (M), A (M) are open. However, if M is a uniform attractor, i.e.
Au (M) is a neighborhood of M, then Au (M) need not be open.
1.18.4 Give examples to show that Theorems 1.12 and 1.15 do not
hold in arbitrary metric spaces.
1.18.5
Show that Theorem 1.14 holds in arbitrary metric spaces.
1.18.6
Show that if Mis compact, then D+(M) is closed.
The next few theorems are devoted to the problem of determining

whether the components of a stable or asymptotically stable set inherit
the same properties. It is very simple to show that in general the properties
of weak attraction, attraction, and uniform attraction are not inherited
by the components.
1.19
Theorem. A set Mis stab1e if and only if every component of M
is stable.
63
Proof. Note that if M is compact, then every component of M is

compact. Further if M is positively invariant, so is every one of its
components. Now let M = V {Mi: i E J} where J is an index set, and Mi
are components of M. Let each Mi be stable, i.e., D+(Mi) =Mi. Then
D+(M) = V D+(Mi) = V Mi= M and M is stable. To see the converse, let D+(M) = M, i.e., Mis stable. Let Mi be a component of M.
Then D+(Mi) is a compact connected set, and D+(Mi) CM. Since
Mi is a component of M we have D+(Mi) C Mi. Clearly then D+(Mi)
= Mi as Mi C D+ (Mi) holds always. Thus Mi is stable. This proves
the theorem.
1.20
(N. P.
Remark. Theorem 1.19 holds even if Xis not locally compact

BHATIA [9]).
1.21
Theorem. Let M be asymptotically stable, and let M* be a
component of M. Then M* is asymptotically stable if and only if it is
an isolated component.
Proof. Let M* be isolated from the other components of M. Then

there is a closed neighborhood U of M* with U ( A (M) and U f\
(M - M*) = 0. Since M* is stable by Theorem 1.19, there exists a positively invariant neighborhood V of M* with V C U. Now for every
x EV, A+(x) =t= 0 as x EA (M), and A+(x) ( V ( U. But A+(x) ( M
and since U f\ (M - M*) = 0 we must have A+ (x) C M*. Thus M*
is an attractor since V C A (M*) and so A (M*) is a neighborhood of M*.
Conversely, let M* be asymptotically stable. We claim that A (M*) f\
(M - M*) = 0, so that M* is isolated. To see this, note that if M0 is a
component of M other than M*, then M 0 is positively invariant and
compact. Moreover, M* f\ M0 = 0. Thus if A (M*) f\ M 0 =t= 0, then for
an x E A (M*) f\ M0 , A+ (x) =t= 0 and A+ (x) C M* as x E A (M*) on one
hand, and A+ (x) C M0 as x E M 0 and M 0 is closed and positively invariant. This contradicts M* f\ M0 = 0. The theorem is proved.
As a corollary to the above theorem we can obtain the following very
important theorem.
1.22
Theorem. Let X be locally compact and locally connected. Let
M be asymptotically stable. Then M has a finite number of components,
each of which is asymptotically stable.
Proof. A (M) is open (and invariant). Since X is locally connected

each component of A (M) is open. The components of A (M) are therefore
an open cover for the compact set M. Hence a finite number of components of A (M) cover M. It is now easily seen that each component of
A (M) contains precisely one component of M.
64
V. Stability Theory
1.23
Exercise. Give a detailed proof of Theorem 1. 22.
1.24
Remark. Theorem 1.22 is not true without local connectedness
of X. Consider the set X consisting of the points 0, 1, 1/2, ... , 1/n, ... on
the real line. X is a compact metric space with the usual euclidean distance between real numbers. Consider any dynamical system on X.
Then one sees easily that each point of X is a rest point. The compact
set X is asymptotically stable and each point of X is a component of x.
However, the point 0 is not asymptotically stable. However, local compactness is not needed as shown by N. P. BHATIA [9].
As a final result in this section we prove an important property of
weak attractors.
Theorem. Let M be a compact weak attractor. Then D+(M)
1.25
is a compact asymptotically stable set, with A (D+(M)) == Aw(M).
Moreover, D+ (M) is the smallest asymptotically stable set containing M.
For the proof we shall need the following lemma.

1.26
Lemma. Let M be a compact weak attractor and let
Then there is T > 0 such that
D+(M) CS [M, ~] [O, T]
>
0.
== n(S [M, ~], [O, T]),
where n is the map defining the dynamical system.
Proof. Choose e, 0 < e <~,such that S [M, e] is a compact subset of

Aw (M). For x EH (M, e), define Tx = inf{t > 0: xt ES (M, e)}; since
xE Aw(M), Tx is defined. Set T = sup{Tx: xE H(M, e)}. We claim that
T < + oo. If this is not the case, there is a sequence {xn} in H(M, e)
for which Txn -?- + oo. We may assume that xn -?-XE H(M, e). Let T > 0
such that xTE S(M, s). For sufficiently large n, we then have Txn < T,
which contradicts Txn -?- + oo. Now let y E .D+ (M) - S [M, e]. Then
there are sequences {xn}, {tn} with xn-?- x EM, and tn > 0 such that
xntn ~ y. Then for all sufficiently large n there is a Tn, 0 < Tn < tn, such
that XnTn EH (M, e), and xt~ S [M, e] for Tn < t < tn. By the first part
of this proof 0 < tn - Tn < T. Then xntn = XnTn (tn - Tn) ES [M, e] [O, T].
Therefore, y ES [M, e] [O, T], since this set is closed. The lemma is
proved.
Proof of Theorem 1.25. Notice that if S [M, e] is compact, then
S [M, e] [O, T] is compact for any T > 0. Thus .D+(M), being a closed
subset of the compact set S [M, e] [O, T] (by the above lemma), is
compact. Further, as e > 0 is chosen, such that S [M, e] C Aw (M) we
have D+(M) CS [M, e] [O, T] ( Aw(M). Thus Aw(M) is an open inva1.27
65
riant set containing D+(M), and is, therefore, a neighborhood of D+(M).

By Corollary 1.11 xE Aw(M) implies A+(x) =I= 0, and A+(x) C D+(M).
Therefore, D+(M) is an attractor. Notice that Aw(M) =A (D+(M)), for
if there is an x E A (D+ {M)), then there is a t > 0 such that xt E Aw (M)
(this being a neighborhood of D+(M)), and sinceA+(x) == A+(xt), we have
x E Aw (M). To show that IJ+ (M) is stable, let xE [)+ (M). Since xE Aw (M),
we can choose an ro E A+(x) f\ M. Then J+(x) C D+(ro) C D+(M) by
Lemma 1.10. Thus
D+(x)
y+(x) V J+(x) ( D+(M) V D+(M)
D+(M),
for D+ {M) is positively invariant. This shows that D+ (D+ (M)) = D+ (M),
i.e., D+(M) is stable. We have thus proved that D+(M) is asymptotically stable. Finally, let M* be any compact set such that M ( M* (
D+(M). Then D+(M) ( D+(M*) ( D+(D+(M)) = D+(M), and so
D+(M*) = D+(M). If M* is stable, then M* = D+(M*) = D+(M).
Thus D+(M) is the smallest stable (also asymptotically stable) set containing M. The theorem is proved.
1.28
Exercises.
1.28.1 Show that if a singleton {x} is a weak attractor, then D+(x) is

invariant. Show also that for every y E D+ (x), x E A- (y) holds.
1.28.2 If Mis invariant and a weak attractor then for every x ED+ (M),
A-(x) (\ M =I= 0. Give an example to show that if M is not invariant,
then the result may not hold.
1.28.3 In Examples 1. 7 calculate D+ (M) of the weak attractors and
attractors and verify that it is an asymptotically stable compact set.
Remark. All the concepts introduced in this section were intro1.29
duced using the so-called positive concepts from Chapter II, i.e., using
positive semi-trajectories, positive limit sets, and positive prolongations.
Thus let us for the moment identify these concepts by using the adjective
"positive" before the terms introduced already, i.e., positively stable,
positive weak attractor, etc. instead of stable, or weak attractor. It is
clear that by using the dual concepts of a negative semi-trajectory, negative limit set, negative prolongation, and negative prolongational limit
set, one can define in an obvious fashion the concepts negatively stable,
negative weak attractor, negatively asymptotically stable, etc. In the
future, whenever we need such a concept we will explicitly write it with
the adjective "negative".
1.30
Exercise.
1.30.1 Let M be minimal and asymptotically stable. Then for every

--xE A (M), y+(x) is asymptotically stable with A (y+(x)) =A (M).
5
66
V. Stability Theory
1.30.2 Show that in a locally compact space X, a compact set Mis a

uniform attractor if and only if it is a weak attractor and for every
compact set U in Aw (M) and every neighborhood V of M there is a T > 0
with Ut ( V for t > T.
2. Liapunov Functions : Characterization of Asymptotic Stability

Throughout this section the space X is assumed locally compact
unless otherwise specified.
The basic feature of the stability theory a la Liapunov is that one
seeks to characterize stability and asymptotic stability of a given set
in terms of a non-negative scalar function defined on a neighborhood
of the given set and decreasing along its trajectories. It is in general not
possible to characterize stability and the various attractor properties
by means of continuous functions. However, in the case of asymptotic
stability one can give very strong theorems. These we describe in this
section.
We need the following lemma.
2.1
Lemma. Let the phase space X be arbitrary and K ( X. Let cp
be any continuous real-valued function defined on K such that cp (xt)
< cp (x) whenever x [O, t] C K, t > 0. Then if for some x, y+ (x) ( K,
we have cp (y) = cp (z) for every y, z EA+ (x).
Proof. Assume tn for each n. Then clearly
 0, and xntn [O, in - tn] c K.
Thus proceeding to the limit we have by continuity of cp, cp(y) > cp(z).
This contradicts the original assumption and the lemma is proved.
The best known result on asymptotic stability is
2.2
Theorem. A compact set M ( X is asymptotically stable if and
only if there exists a continuous real-valued function defined on a
neighborhood N of M such that
0 if x q M;
 (x)
2.2.2
 (xt) < (x) for x q M, t > 0 and x [O, t] ( N.
0 if x E M and (x)
>
2.2.1
Proof. Assume that a function as required is given. Choose ~ > 0

such that S [M, cX] C N and is compact. Let m ==min{</> (x): x EH (M, cX)}.
Then by 2.2.1 and continuity of </>, m > 0. Set K == {x ES [M, cX]:
<l>(x) < m}. ThenK is indeed compact and because of 2.2.2K is positively
2. Liapunov Functions: Characterization of Asymptotic Stability
67
invariant. This establishes that M is stable as K is a positively invariant

neighborhood of M. To see that M is an attractor, choose any compact
positively invariant neighborhood K of M with K C N. Then for any
x EK, 0 =f= A+ (x) ( K, and Lemma 2.1 shows that </J is constant on
A+(x). But this shows using 2.2.2 that A+(x) CM. Thus Mis an attractor and consequently asymptotically stable. Now let M be asymptotically stable and A (M) its region of attraction. For each x EA (M)
define
 0} .
Indeed 0 with x [T, + oo) CS (M, x). Thus
 0 for x ~ M, and 0. This is clear when we remember that M is stable and
hence positively invariant and that A (M) is invariant. Thus if <p (x) is
defined for any xE A (M), it is defined for all xt with tE R+. We further
claim that this 0) and choose B, 0 0 such that S [x, e] t CS (M, x/4) for all t > T (see 1.30.2).
Thus for y ES [x, e] we have
 O} O< t 
O}
T} - sup {e (yt, M) : O < t
< T}.
Therefore
j<p(x)- <p(y)J
< sup{le(xt,M) -e(yt,M)J: O < t < T}

< sup {e (xt, yt): O < t < T}.
The continuity axiom implies that the right hand side of the above inequality tends to zero as y ~ x, for Tis fixed for y E S [x, e]. The function <p (x) is therefore continuous in A (M). However the above function
may not be strictly decreasing along parts of trajectories in A (M) which
are not in Mand so may not satisfy 2.2.2. Such a function can be obtained
by setting
00
</J (x)
J 0. Then indeed </J (xt) < (x)
5*
68
V. Stability Theory
holds, because (xt) = f/> (x) , observe
that in this case we must have 0. Thus
in particular, letting T = 0, t, 2t, ... we get <p (x) = <p (x (nt)), n = 1, 2, 3, ....
But asymptotic stability of M implies that for x E A (M), e(xt, M) ~ 0
as t ~ oo. Thus cp (x (nt)) ~ 0 as n ~ oo, as 0, a contradiction.
We have thus proved that f/>(xt) < </J(x) for xq Mandt> 0. The theorem is proved.
2.3
Remark. Theorem 2.2 says nothing about the size of the region
of attraction of M. Thus if a function f/> (x) as in Theorem 2. 2 is known to
exist in a neighborhood N of M, we need not have either NC A (M) or
A (M) ( N. We will give an example to elucidate this point. In particular
this means that the above theorem cannot immediately be stated as a
theorem on global asymptotic stability.
2.4
Example. Consider a dynamical system defined in the real

euclidean plane by the differential equations
2.4.1
x=f(x,y),
y=g(x,y),
where
> 1
g(x, y) = -y for all (x, y), and /(x, y) = {
~
2x3y 2 - x If x 2 y2 < 1 .
x if x 2y 2
2.4.2
These equations are integrable by elementary means and the phase

portrait is as in Fig. 2.4.3.
y
Fig.2.4.3
69
The origin (0, 0) is asymptotically stable, with the set {(x, y): x2y2 < 1}
as its region of attraction. Consider now the function
2.4.4
x2
<1> (x, y)
= y2 + 1 + x2
=:
This function satisfies conditions of Theorem 2. 2 in the whole euclidean

plane. To see this one may find the derivative (x, y)
f (x, y) +
~ (x, y) < 0 for all
(x, y) =f= (0, 0), which implies that this <1> satisfies conditions of Theorem 2.2 in every neighborhood of the origin. But not every neighborhood
of the origin is contained in the region of attraction, nor does it contain
the region of attraction.
Since Theorem 2.2 guarantees asymptotic stability of M, it is clear
that A (M) (\ N is a neighborhood of M. How far the extent of this
neighborhood can be determined by a function <1> (x), as in Theorem 2.2,
is the subject of the following proposition.
2.5
Theorem. Let <1> (x) be a function defined in an open neighborhood N of Mand having properties as in Theorem 2.2. Then if K ( N
is any compact positively invariant set, we have K ( A (M). Further
for all sufficiently small cX > 0 the set K' = { x E N: <1> (x) ~ cX} has a
compact positively invariant subset P' such that P' (\ K' - P' = 0,
P' is a neighborhood of M, and P' CA (M). In particular if K'is compact, then K' CA (M).
Proof. The first part follows from Lemma 2.1. To prove the second
part, let B > 0 be chosen such that S [M, B] is a compact subset of J (N).
Set m (e) = min {<1> (x): x EH (M, e) }. Clearly m (e) > 0. Now let 0 < cX
< m(e), and set P' == K' (\ S [M, B]. We claim that P' is a set predicted
in the lemma. Clearly P' =F 0 as P') M. P'is compact, for if {xn} is a
sequence in P', then we may assume that it is convergent, because {xn}
is in the compact set S [M, B]. If now xn-+ x, then clearly x ES [M, e]
on one hand, and <1> (x) ~ cX, because <1> (xn) < cX for each n, showing that
x EK'. It follows that x E (K' (\ S [M, e]) = P'. That P' is positively
invariant may be seen by first observing that if x E P ''and xR:+' <t S (M, e),
then there is a t > 0 such that xt EH (M, e) and x [O, t] CS [M, e] ( N.
By 2. 2. 2 we get <1> (xt) ~ <1> (x) < cX, and xt E H (M, e) implies <1> (xt) >
m (e) > cX, a contradiction. :Hence xR:+' C S (M, e) ( N. It follows that
<P(xt) < </J(x) < cX for each t > 0, which shows that xR+ ( K'. Thus
xR+ C K' (\ S (M, e) C P '. That is, P ' is positively invariant. By the
first part of the theorem we have P' CA (M). ThatP'is a neighborhood
70
V. Stability Theory
of M follows from the continuity of </J and the fact that </J (x) = 0 for
x E M. Finally to see that P ()(. f\ K(J(. - P ()(. = 0, observe that K(J(. (\ H (M, e)
= 0. The last part follows by observing that if K' is compact, then it is
positively invariant.
2.6
Remark. For any given B, cX > 0 chosen as above, the set P (X
defined in the above proof is the largest, compact positively invariant
set in K'. For if SC K(J(. were a larger set, then S - P(J(. ( @(S [M, e]),
so that S - P ()(. (\ P ()(. = 0. It follows that S - P ' is compact and positively invariant. This is impossible unless S - P(J(. = 0. For, if S - P(X
were a non-empty subset of N, by the above theorem, S - P' CA (M).
Therefore, (S - P') R+ (\ P(X =F 0 as P' is a neighborhood of M. But
(S - P(J(.) R+ = S - P(J(., and (S - P') f\ P' = 0.
2. 7
Theorem. Let M C X be a compact asymptotically stable set.
Then its region of attraction A (M) contains a countable dense subset.
In particular, if A (M) = X, then Xis separable.
Proof. Take any </J (x) defined in a neighborhood N of Mand satisfying

conditions of Theorem 2.2. Choose B, cX > 0 as in the proof of Theorem 2.5
and construct P ()(. as before. As P ()(. is a compact positively invariant
neighborhood of M, and P(J(. CA (M),
= P 'R+ V P 'R- = P ' V P 'R- = P (J(.R-.

= V {P(J(. [-n, OJ: n = 1, 2, ... } == P(J(.R-.
A (M) = P 'R
Thus A (M)
Since each P ' [-n, OJ is compact and thus contains a countable dense
subset, the result follows.
The above proposition motivated the introduction of the following
definition.
2.8
Definition. A continuous scalar function </J (x) defined on a set
N C X will be said to be uniformly unbounded on N if given any cX > 0
there is a compact set KC N, K =F N, such that (]>(x) > cX for xq K.
2.9
Theorem. Let M C X be a compact asymptotically stable set
and let M be invariant. Then there exists a continuous uniformly unbounded function </J (x) defined on A (M) such that
2.9.1
2.9.2
= 0 for xE M, and </J(x) > 0 for xq M,

</J (xt) = e-t </J (x) for all x EA (M) and t ER.
</J(x)
Proof. Consider any function cp (x) defined in any neighborhood N of

Mand satisfying conditions of Theorem 2.2. Choose B, cX > 0 as in Theo-
71
rem 2.5, and consider the set P(J(.. Note that 8P' = {xE S [M, e]: <p (x) = x}.
We claim that for every point x EA (M) - M, there is a unique T(x) ER
such that XT(x) E oP(X. The uniqueness of T(x) follows from the fact that
if xr(x) E oP', then <p(x(r(x) + t)) < <p(xr(x)) = x for t > 0. As P' is
positively invariant and 0. This establishes uniqueness of
T (x) for all x for which T (x) is defined. That r (x) is defined for all x E
A (M) - P(J(. follows from the fact that P(J(. is a neighborhood of M, so
that any trajectory y (x) ( A (M) with x ~ P '' must intersect oPoc For
xE oP(X indeed T(x) is defined and T(x) = 0. For XE J(P(J(.) - M, if
there is no i (x) such that XT (x) E oP'' then y (x) c J (P()(.) - M c p oc
P(J(. being compact A-(x) =f= 0, A-(x) C P', but A-(x) (\ M = 0 (otherwise M will be unstable). Now A-(x) is compact and invariant, so that
if y E A-(x) we have A+ (y) =f= 0, and A+ (y) C A-(x). Then on one hand
we haveA+(y) (\ M = 0 and on the other handA+(y) (Mas y EA (M).
This contradicts A+ (y) =F 0. Thus y (x) (\ oP(X =F 0, and T (x) is defined
for each x EA (M) - M. Note that A (M) - M is invariant, as M and
A (M) are both invariant. For x E A (M) - M and TE R observe now
that
i (xt) == T (x) - t.
This follows from the fact that any trajectory y (x) in A (M) - M intersects oP()(. at exactly one point. Thus xt (T (xt)) = x (T (x)), i.e., by the homeomorphism axiom
2. 9. 3
x (t
+ T (xt)) =
x (r (x)) .
As y (x) can neither be periodic nor a rest point, we have t + r (xt) = T (x).
This shows further that r (xt) is a continuous function of t and T(xt) ~
oo as t ~ =f= oo. We now claim that r (x) is continuous on A (M) - M.
For any x EA (M) - M, and e > 0, the pointy== x(T(x) + e) E J (Poe)
There is therefore a neighborhood NY of y such that N'Y ( P'. Then
N+ = N'Y(-T(x) - e) is a neighborhood of x and note that for each
ro EN+, T(ro) < T(x) + e. Again z == x (r (x) - e) EA (M) - P '' the last
set being open. Thus there is a neighborhood Nz of z such that
Nz C (A(M) - P'). Then N- = Nz(-T(x) + e) is a neighborhood of x
and note that for each ro EN- we have T(ro) > r(x) - e. Thus if ro is
in the neighborhood N 8 = N+ f\ N- of x, we have
T(x) - e
< T(ro) < T(x) + e.
This proves continuity of T(x) in A (M) - M. We now show that T(x) ~

- oo as x ~ M, x EA (M) - M. If this were not true, then there will be
a T > 0 and a sequence {xn} in A (M) - M, such that xn ~ x EM and
v-T < T(xn) < 0. Since {T (xn)} is a bounded sequence it contains a conergent subsequence. We may therefore assume that T(xn) ~ T, where
72
V. Stability Theory
-T < r < 0. Then by the continuity axiom xnr(xn) ~xi. As M is

invariantxrE M, on the other hand xnr(xn) E oP(J(. which is compact. Therefore xrE oP(J(.. But oP(J(. f\ M = 0, a contradiction. We now define the
function </J (x) on A (M) as follows
</J(x)
for
xE M,
</J (x)
eT(x)
for
x EA (M) - M.
and
The above observations show that this function is continuous on A (M).
It is clearly positive for x Et M, and
2.9.4
</J (xt) = e-,;(xt) = ei-(x)-t = </J (x) e-t.
Lastly to see that this </J (x) is uniformly unbounded, recall that A (M) =
V {P(J(.[-n, OJ: n = 1, 2, 3, ... }.Each Pix[-n, O] is compact and positively invariant. Observe that if x4 Pix [-n, OJ, then r(x) > n, so that
</J (x) > en. This proves the theorem completely.
If M is not invariant, then we have:
2.10
Theorem. If M (Xis any compact asymptotically stable set,
there exists a continuous, uniformly unbounded function </J (x) on A (M)
such that
2.10.1
</J(x)
= 0 for x EM and </J(x) > 0 for xq M,
2.10.2
</J (xt)
< </J (x)
for x q M and t
>
0.
Proof. Consider any function 0 as before and
consider P ()(." For each x E A (M) - P ()(. define r (x) as before. This r (x) is
continuous and r (x) ~ 0 as x ~ Pix' x Et Pix Now define
f/> (x) == (x)
cXei-(x)
This </J (x) has the desired properties as may easily be verified.
2.11
Theorem. Let MC X be compact and let there exist a continuous
uniformly unbounded function f/> (x) defined on an open neighborhood
N of M such that
2.11.1
</J (x)
2.11. 2
</J (xt)
0 for x E M and </J (x)
< </J (x)
>
0 for x q M,
for x tt M, t > 0 and x [O, t] C N.
Then Mis asymptotically stable and N (A (M). If, in addition, any

condition guaranteeing the invariance of N holds, then N =A (M).
73
Proof. The proof follows from the observation that if </J (x) is uniformly
unbounded on N, then for any x > 0, the set K(J(. = {x: </J(x) < cX} is
compact. Then by Theorem 2.5, K(J(.C A (M). Now N = V {Kn: n =
1, 2, ... }, and since each Kn CA (M) we have NC A (M). Lastly if N
is invariant we must have N =A (M) as N is a neighborhood of M.
The remaining details of the proof will be the same as in the proof of
sufficiency of Theorem 2.2. These we leave to the reader.
For global asymptotic stability we can state the following two theorems as corollaries of the above results.
2.12
Theorem. A compact invariant set M C Xis globally ~symptoti
cally stable if and only if there exists a continuous uniformly unbounded
function </J (x) defined on X such that
2.12.1
2.12.2
= 0 for x EM, </J(x)

</J (xt) = e-t (J) (x).
</J(x)
> 0 for xq M,
Prooj. The sufficiency follows from Theorem 2.11 as Xis an invariant

neighborhood of M. The necessity follows from Theorem 2.9.
2.13
Theorem. A compact set M C Xis globally asymptotically stable
if and only if there exists a continuous uniformly unbounded function
</J (x) defined on X such that
2.13.l
</J(x)
2.13.2
</J (xt)
< </J (x)
0 for x EM, </J(x) > 0 for xq M,

for x q Mandt> 0.
Proof. The sufficiency follows from Theorem 2.11, the necessity from
Theorem 2.10.
2.14
Remark. In dynamical systems defined in locally compact metric
spaces, one may define ultimate boundedness of the dynamical system
by the property that there is a compact set K C X with A+ (x) =f= 0, and
A+ (x) ( K for each x E X, i.e., whenever there exists a compact global
attractor in X. It is shown in Theorem 1.25 that if K (Xis a compact
weak attractor, then D+(K) (the first positive prolongation of K) is a
compact positively invariant set which is asymptotically stable and has
the same region of attraction as K. One can show now that the largest
invariant set in D+ (K) is compact and asymptotically stable with the
same region of attraction as of K. These observations will allow one to
write theorems on ultimate boundedness which are similar to those on
global asymptotic stability. We leave these to the reader.
As a final result on asymptotic stability we wish to prove the following

theorem whose proof illustrates yet another technique of proving the
existence of special kinds of Liapunov functions.
74
V. Stability Theory
2.15
Theorem. If MC X is a compact asymptotically stable set
with the region of attraction A (M), then for any f3 > 0 there exists a continuous uniformly unbounded function <l>(x) on A (M) having the following properties.
2.15.1
<l>(x)
= 0
2.15.2
<1> (xt)
<
for
x EM and <l>(x) >
e-fJt <1> (x)
0 for
for x EA (M) and t
xq M,
>
0.
Proof. Consider any compact positively invariant neighborhood P

of M, PC A (M). Then there is an integer n0 such that S [M, 1/n] ( P
for n > n0 For each n > n0 we construct a function f/Jn: P ~ R+ as
follows
2.15.3
<pn (x) ==sup {e (xt, S [M, 1/n])
ef1t:
t > O}.
We claim that this function is well defined. In fact, for each n > n 0 ,
there is a Tn > 0 such that xt ES [M, 1/n] for all x E P and t > Tn. For
otherwise for some n* there will be a sequence xn in P and a sequence
tn ~ + oo, such that xinE P - S [M, 1/n*]. We may assume that xn ~
x E P, and xntn ~ y E P - S(M, 1/n*). Then y E J+(x), xE A (M), but
y q M. This contradicts uniform attraction of M (Theorem 1.16). Thus in
fact for each n > n0
2.15.4
cpn (x) = sup {e (xt, S [M, 1/n])
efJt: 0
< t < T n},
and f/Jn (x) is defined. One now easily sees that f/Jn (x) is continuous and
moreover, satisfies the inequality
2.15.5
<pn (xt) <
e-fJt CfJn (x)
for x E P, t > 0.
To see this note that
CfJn (xt) = sup {e (x (t
+ -r), S [M, 1/n]) eP-r: 0 < i'}
= sup {e (xT, S [M, 1/n]) eP<-r-t>: t < T}

= e-Pt sup {e (xT, S [ M, 1/n]) eP-r:
< e-fJt f/Jn (x), if t >
> t}
0.
Finally let max {cpn (x): x E P} ==Mn. Define cp: P ~ R+ by

2.15.6
+oo 1
( )
cp (x} = ;E 2n <p~x
n=no
This cp (x) is continuous on P, and satisfies cp (xt) < cp (x) e-fJt for x E P
and t > 0. Moreover, <p(x) = 0 if x EM, and <p(x) > 0 if x E P - M.
Now choose any cX > 0 sufficiently small such that
P'
{xE P: <p(x)
< cX}
is in the interior of P. Now define <P: A (M)

2.15. 7
rFt.
':I:'
(x)
75
R+ by
= {<p (x)
if x E Pa:,
fJ ( )
rxe
if x 'F
L P a:
where -r (x) is the unique number such that xi (x) E oPa: This <P (x) has all
the properties required in the theorem as is easily verified.
We shall give next, theorems which allow the identification of oA (M)
in the case of an asymptotically stable compact, connected set M ( X.
2.16
Theorem. Consider a dynamical system (X, R, n) on a locally
compact metric space X. Let MC X be a compact, connected invariant
set. Then a necessary condition for M to be asymptotically stable and
the open invariant set A (M) its region of attraction is the existence of
two real-valued functions 0 if xE X - M.
2.16.5
 0, there exists y 2 > 0 such that y 1
2.16. 7 For each sequence {xn} CA (M) such that either xn ~ x E oA (M)
or e(xn, M) ~ + oo, <p(xn) ~ -1.
2.16.8
(d<pd~l)
Proof. Let
2.16.9
<51
= lim <p(xt)
t=O
>
HO
0 (x) (1
<p(x) =
0 be such that S [M,
A(t) =Sup {e (xt, M): x ES (M,
<51 ]
+ ip (x)).
CA (M). Assume
M},
<51) -
for each t > 0 such A(t) exists and is positive. Since M is asymptotically
stable and S [M, <51] CA (M), A(t) < e1 for 0 ~ t < + oo, and lim A(t)
= 0. Then for each integer k one can find tk
2.16.10 A(t)
< e1/2k,
for all t with t
>
>
t-++oo
0 such that
tk.
In addition we can find an increasing subsequence {tk} such that tk --+
+oo.
Assume next
2.16.11 lk(t)
for tk
< t<
[2e1tk+i - e1 (tk
tk+l and k
> 1.
+ t)]/2k(tk+i
- tk)
76
V. Stability Theory
The function lk (t) is increasing in the interval of definition. In addition, since

2.16.12 lk (t)
> [2e1tk+ 1 - e1 (tk + tk+i)]/2k (tk+i - tk)
8,:_
2 1
> A(t),
in the interval of definition, we have

2.16.13 A(t)
<
lk (t) for tk < t < tk+t and k > 1.
For k = 1 we can assume that the function l1 (t) is defined on the interval (- oo, t 2). Since l1 (t) > e1 > A(t), the function
-{lk(t) for tk < t < tk+l' k

2.16.14 L (t) - l1(t) for - oo < t < t2
> 1
for each t > 0 satisfies the inequality

2.16.15 A(t)
<
L (t).
Such function L (t) is defined and continuous on the whole real line and
in addition it is strictly decreasing from + oo to 0. From the equation
2.16.9 and the inequality 2.16.15 it then follows that
2.16.16
e(xt, M) < L (t)
fort ER+ and x ES (M, <51).
Let L-1 (s) be the inverse function of L (t): from the properties of L (t)
it follows that L-1 (s) exists, is continuous and strictly decreasing on the
interval 0 < s < + oo. Consider the function
2.16.17
~ (s) =
s e-L-l(s), 0 < s <
+ oo,
which is positive, continuous, strictly increasing and such that ~ (s) ~ 0

as s ~ 0. We shall next assume ~ (0) = 0 and therefore define ~ (s) on
the whole interval 0 < s < + oo.
Consider the function
2.16.18 f) (x) =
~ (e (x,
M)) =
e(x, M) e-L-l(Q(x,M)).
This function satisfies all conditions of the theorem since it is defined and
continuous for all x EX, it is positive for all x EX - M, it vanishes for
x E Mand it is strictly increasing as e(x, M) increases. Consider next the
function
2.16.19 'P (x) = exp [-
J~
(e (xt, M)) dt] - 1,
we shall prove that, if for each xE A (M), <p(x) exists, we have

2.16.20 1
+ <p(xt) =
[1
+ <p(x)] exp[/ ~((!(X'i,
M)) d'i] .
From 2.16.19 it follows

2.16.21 1
+ <p(x) =exp [-i~(e(n,M))d'i] exp [-l~(e(x'r,M})d'i].
Hence, if we let
2.16.22
(1
77
= t + s, we have
+ IP (x)) exp
=exp [=exp
[j ~
(e (x-r, M))
d-r]
J~(e(x(t + s), M)) ds]
[-J~(e ((.xt) s, M)) ds] = 1 + IP (xt)
where the equality x (t + s) = (xt) s has been used.

In order for 2.16.20 to be true it is then enough that for each x EA (M),
 0, such that, for all
t > i(x), we have Xi(x) E S(M, <51). It follows that
oo
2.16.23
+oo
~ (e (xt, M)) dt + ~ (e (xt, M)) dt.
TW
J~(fl (xt, M)) dt = 0J
T(X)
If we change the integration variable in the second integral from t into

s + i(x) and notice that, as usual, x(s + i(x)) = (xi(x)) s, we have that
2.16.24
+oo
~ (e (x, t, M)) dt
+oo
~ (!_> (x (s
+ i (x)), M)) ds
T(X)
+oo
= ~
J ((] ((xi(x)) s, M)) ds.
Now
XT
(x) ES (M, <51) and hence from 2.16.16 we have
2.16.25 L (s)
> e((xi(x)) s, M)
for
s ER+.
Now, since L-1 is strictly decreasing, from the inequality 2.16.25 it

follows that
2.16.26 s = L-1 (L (s))
<
L-1 {f! ((xi (x)) s, M)},
from which we have that

2.16.27
e((xi (x)) s, M)e-s >(]((xi (x)) s, M) exp [-L-1 {e ((xi (x)) s, M) }J

=
~{e ((xi (x))
s, M)}.
Thus, since L (t) < <5 for t > 0, because of the inequalities 2.16.16 and
2.16.27, it follows that the integral 2.16.24 is less than <5 and therefore
+oo
converges. It follows that for each x EA (M),
J ~ (e (xt, M)) dt
hence for each x E A (M), <p (x) exists and the equality is proved.
exists,
78
V. Stability Theory
We want to show next that <p (x) is continuous for each x EA (M).
For that it is enough to prove that the function
2.16.28 <p* (x) ==log (1
+oo
 0 such that, for(! (y, x) < <51
>
00
2.16.29
j<p* (y) -
<p* (x) f ==
f {~ (e (yt, M)) -
~ (e (xt, M))} dt
<
e.
For (! (y, x) sufficiently small, the points x and y belong to A (M) and
therefore there exists T > 0 such that
+oo
j ~ (e (yt, M)) dt < : ,
2.16.30
+co
f ~(e(yt, M)) dt < : ;
2.16.31
these inequalities follow from the convergence of the integral 2.16.23.

Now, since the map n of the dynamical system is continuous, given
<5 2 > 0, we can find <53 > 0 such that
2.16.32
(!
(xt, yt)
<
<5 2
for
(!
(x, y)
<
<53
and 0
< t<
Since ~ (e (x, M)) = (J (x), from the continuity of

sufficiently small <52 > 0, we have
2.16.33
Thus, for
2.16.34
l~(e(xt, M)) - ~(e(yt, M))

(!
(x, y)
j<pj (x) -
<
<53 ,
I< e/2T,
T.
(J (x)
it follows that for
< t<
T.
we have
<pj (y)
I< e,
and the properties 2.16.1-2.16.5 of the theorem have therefore been

proved.
We shall now prove that also the property 2.16.6 is satisfied. For that
take y1 > 0. Let T > 0 and cX > 0 be such that (! (xt, M) > cX for 0 < t
< T and e(x, M) > y1.
Then there exists f3 > 0 such that ~ (e (xt, M)) > f3 for 0 < t < T,
hence
2.16.35 ip(x) <exp [-
fPdt]-
1 < -y 2
which concludes the proof of the property 2.16.6.

It remains to be proved that also the property 2.16. 7 is true. For that
let x EA (M) and consider a sequence {xn} CA (M) with xn ~ x. Let
<51 > 0 be such that H (M, <51) C A (M) and {rn} C R+ such that xnt ~
S (M, <51) for 0 < t <in, Xnin EH (M, <51). Clearly such a sequence {in}
79
exists and, in addition -c11' ~ + oo, if not x EA (M) (\ oA (M), which is

absurd.
Let xnr:n ~ x* E H (M, <51). From the continuity of 0, such that
~ (e (xnt, M)) > cX for 0 < t < -r:n. From the equality 2.16.20 it then follows
that
2.16.36 (1
+ qi (x"")) exp [ - J"~ (e (x"t, M)) dt] =
(1 +qi (x")).
Now the limit of the right-hand side of the 2.16.36 exists and it is equal
to zero, hence <p(xn) ~ -1. The same reasoning applies if {xn} CA (M)
and [jxn/I ~ + oo, which concludes the proof of the property 2.16.7 and
of the theorem.
2.17
Remarks.
The importance of Theorem 2.15 is only in connection with

asymptotically stable sets M which are not necessarily invariant. For
invariant setsM,4>(x) may be more easily obtained bysetting(l)(x) == efh:(x)
for all x EA (M) - M (see proof of Theorem 2.9).
2.17.1
Theorems analogous to those proved in this section can also

be derived for negatively asymptotically stable sets. A negatively asymptotically stable set is sometimes called completely unstable. But we
shall a void this term in general.
2.17.2
The definitions introduced in section 1 can be reformulated to

include closed sets having a compact boundary. Then all the theorems in
this and the previous section hold for such sets also.
2.17.3
Also Corollary 4.24 is useful in the identification of the region

of attraction of compact asymptotically stable sets.
2.17.4

In this section we shall present some additional properties of weak
attractors, their regions of attraction, and the level lines of certain
Liapunov functions for asymptotically stable sets. Again the space X
will be locally compact.
We start by recalling the result in Theorem 2.5 which may be stated
as
Theorem. Let M be a compact asymptotically stable subset of
3.1
X and 4> a function defined on a neighborhood N of M and satisfying
conditions of Theorem 2.2. Let e > 0 be such that S [M, e] is a compact
80
V. Stability Theory
subset of N. Let X, 0 <

3.1.1
< m(e), where
m(e) ==min {(J)(x): xE H(M, e)}.
Then the set

3.1.2
P(X == K(X f\ S [M, e],
where
3.1.3
K(X == {x EN: (})(x) < X},
is a compact positively invariant subset of A (M).

The following lemma expresses an important topological property of
the sets P defined above.
(X
3.2
Lemma. The set P defined in the above theorem is a retract
of A (M). Consequently P(X is a retract of every set N, P(X C N (A (M).
(X
Proof. This is so because we can define a map h: A (M) ~ P by

h (x) == x if x E P (X, and h (x) == Xi (x) if x ~ P (X, where -r (x) is defined as in
the proof of Theorem 2.9. Because of the continuity of i(x) and of the
phase map n, and the fact that i (x) == 0 for x E fJP (X, it follows that h is
a continuous map of A (M) into P(X which is an identity on P(X. Thus by
definition P(X is a retract of A (M), and hence also a retract of every subset of A (M) which contains P
(X
(X.
From the definition of stability it is clear that if a singleton {x} is

stable, then x is a critical or rest point (Theorem 1. 9). In particular, if
{x} is asymptotically stable, then x is a rest point. Our next theorem
concerns an important topological property of the region of attraction
of a rest point in a euclidean n-space Rn. The proof depends on the following lemma in the topology of n-spaces.
Lemma. Let {Un} be a monotone sequence of open n-cells in Rn,
3.3
i.e., Un ( Un+l' n = 1, 2, .... Then V {Un, n = 1, 2, ... } is an open
n-cell.
Theorem. If a rest point p E Rn is asymptotically stable, then
3.4
A (p) is homeomorphic to Rn.
Proof. Since A (p) is a neighborhood of p, there is an e > 0 such that

the closed ball S [p, e] CA (p). For each t ER (the transition nt being
a homeomorphism of Rn onto Rn), the image S (p, e) t of the open ball
S (p, e) by nt is an open n-cell. We claim now that for any given t 1 there
exists a t2 , t2 < t1 such that S (p, e) t1 CS (p, e) t2 This is so because
S (p, e) t1 being a subset of the compact set S [P, e] t1 is itself compact.
81
Further, S [p, e] t1 (A (p), and A (p) is open. Since pis uniformly attracting there exists a T > 0 such that S [p, e] (t1 + t) C S (p, e) t1 for t > T.
In particular, S [p, e] (t1 + T) CS (p, e) t1 CS [p, e] t1 Hence S [p, e] t1
C S (p, e) (t1 - T) C S [p, e] (t1 - T). Setting t 2 == t1 - T, we get
S [p, e] t1 CS (p, e) t 2 The above analysis shows that we can choose a
sequence {tn}, tn--+ - oo, such that {S (p, e) tn} is a monotone sequence of
open n-cells. By Lemma 3.3 V {S(p, e) tn; n == 1, 2, ... } is an open
n-cell. But this last union is A(p), so that A (p) is an open n-cell and hence
homeomorphic to Rn.
3.5
Corollary. If p is an asymptotically stable rest point in Rn, then
A (p) - {P} is homeomorphic to Rn - {O}, where 0 is the origin in Rn.
We can now present the following result.
3.6
Theorem. Let M C Rn be a compact invariant globally asymptotically stable set in Rn. Then Rn - M = @(M) is homeomorphic to
Rn - {0}.
For the proof of this theorem we need to establish some further results
which are important in themselves.
3.7
Lemma. Let MC X be a compact positively invariant set.
Let {rn} be a sequence of periodic trajectories with periods Tn, such that
Yn CM, and Tn--+ 0. Then M contains a rest point.
Proof. Consider any sequence of points {xn}, with xn E y n' n = 1, 2, ....

We may assume without loss of generality that xn--+ x EM, as M is
compact. We will demonstrate that x is a rest point. For suppose that
this is not the case. Then there is a i > 0, such that x =f= Xi. Let (! (x, Xi)
== cX (> 0). The spheres S (x, cX/4), and S(xi, <X/4) are disjoint. Now choose
T, 0 < T < i, such that (! (x, xt) < cx/8 for 0 < t < T. By continuity
of n there is a <5 > 0 such that (! (x, y) < <5 implies (! (xt, yt) < cX/8 for
0 < t < i. Notice in particular that if e(x, y) < <5, then
(! (x, yt)
< (! (x, xt) + (! (xt, yt) < cX/8 + cx/8 = cX/4
and e(xi, Yi)
<
if 0
< t < T,
cX/4.
Now for sufficiently large n we have T n < T and d (x, xn) < <5. Hence
e(x, xnt) < cx/4 for 0 < t < Tn < T. And as Yn is periodic of period Tn,
we have (! (x, xnt) < cX/4 for all t ER. This is impossible, because we must
have (! (x, Xni) > (! (x, Xi) - (!(xi, Xni) > cX - cx/4 = (3/4) cX. This contradiction proves that the point x EM is a rest point.
As an application of the above lemma we have
6
82
V. Stability Theory
3.8
Theorem. Let M (Rn be a compact positively invariant set,
which is homeomorphic to the closed unit ball in Rn. Then M contains a
rest point.
Proof. Consider any sequence {rn}, in > 0, in~ 0. Consider the sequence of transitions {nn}, Jtn = Jt-r:n. As Mis positively invariant each Jtn
maps M into itself. Thus by the Brouwer fixed point theorem, M contains a fixed point of each one of the transitions nn. Let xn E M be a fixed
point of Jtn. Then since xn == nn (xn) = Xnin, the trajectory 'Y (xn) = 'Y n is a
rest point or a periodic trajectory with a period in, and as Mis positively
invariant 'Yn CM. By the above lemma, M contains a rest point, and the
theorem is proved.
We are now in a position to prove the following important result.
3.9
Theorem. Let M C Rn be a compact set which is a weak attractor. Let the region of weak attraction Aw (M) of M be homeomorphic to
Rn. Then M contains a rest point. In particular, when Aw (M) =Rn (i.e.,
Mis a global weak.attractor), then M contains a rest point.
Proof. By Theorem 1.25 D+(M) is an asymptotically stable compact

set with A (D+ (M)) = Aw (M). Let <P (x) be any function for the asymptotically stable set D+ (M) as in Theorem 3.1, and consider a set P 'for <P (x).
Then P ' is compact, positively invariant, and is a retract of Aw (M). As
Aw (M) is homeomorphic to Rn we can choose a compact set B, P ' C B
( Aw(M), where Bis homeomorphic to the closed unit ball in Rn. Then
P' is a retract of B. Thus P' has the fixed point property, as B has by
the Brouwer fixed point theorem. Since P' is positively invariant, the
transition nr maps P' into P' for each i > 0. Thus for each fixed i > 0,
nT has a fixed point in P', i.e., corresponding to any i > 0 there is an
xT E P' such that nT (xT) = xT. Thus the orbit y (xT) is periodic and has a
period i, moreover y (xT) C P', because y (xT) = y+ (x-r) C P'. We have
thus shown that, corresponding to any sequence {in}, in > 0, in~ 0,
there is a sequence of closed orbits {Yn}, 'Yn = y (xTn ), with Yn having a
period in This sequence being in P', P' contains a rest point x* (Lemma 3. 7). However, M CD+ (M) C P ' C Aw (M). As M is a weak attractor
we have A+(x)f\M =f= 0 for each xEAw(M). Thus A+(x*) f\ M =f= 0.
But A+(x*) = {x*}, as x* is a rest point. Hence x* EM. The theorem is
proved.
For the following corollaries the dynamical system is assumed to be
defined on Rn.
3.10
Corollary. If the dynamical system is ultimately bounded, then
it contains a rest point.
83
This is so, because ultimate boundedness is equivalent to the existence

of a compact globally asymptotically stable set (Remark 2.14) which by
the above theorem contains a rest point.
3.11
Corollary. The region of attraction of a compact minimal weak
attractor M cannot be homeomorphic to Rn, unless Mis a rest point.
Note, however, that if a rest point p E Rn is weakly attracting, or

attracting, then A (p) need not be homeomorphic to Rn, as the analytic
example 1. 7.4 shows. As another corollary we have
3.12
Corollary. If Mis compact minimal and a global weak attractor,
then Mis a singleton. Consequently, M consists of a rest point.
Finally we prove Theorem 3.6.
Proof of Theorem 3.6. By Theorem 3.9, M contains a rest point. We

may assume without loss of generality that M contains the origin 0
and 0 is a rest point. Consider now the homeomorphism h: E - {O}
~ E - {O} defined by h(x) = x/llxll 2 , where llxll is the euclidean norm
of x. h maps the given dynamical system into a dynamical system on Rn,
with 0 becoming a negatively asymptotically stable rest point, and
Rn - Mis mapped into A-(O) - {O}, where A-(O) is now the region of
negative attraction of 0. By the Corollary 3.5, A-(O) - {O} is homeomorphic to Rn - {O}. Hence the result follows.
We shall now present one example of application of Theorem 3.9.
3.13
Example. Consider a dynamical system in Rn with only two
rest points x and y, x =f= y. Theorem 3.9 shows that y cannot be asymptotically stable with A ({y}) =Rn - {x}, since @({x} V {y}) is not homeomorphic to Rn - {0}.
3.14
Exercises.
3.14.1 Show that if Xis compact and MC X a compact global weak

attractor, then D+(M) = X. Consequently, if M =f= X, then M cannot
be stable.
3.14.2 If Mis a compact invariant asymptotically stable set in X, then
the restriction of the dynamical system (X, R, n) to the set A (M) - M
is parallelizable.
6*
84
V. Stability Theory
4. Stability and Asymptotic Stability of Closed Sets

In this section X will be an arbitrary metric space, unless otherwise
restricted. In contrast to the situation in the last three sections, several
notions of stability and asymptotic stability may be introduced. The
purpose of this section is to develop several notions of stability and attraction and to provide criteria for these in terms of Liapunov functions.
We shall first discuss stability of closed sets.
MC X will be said to be
4.1
Definition. A closed set
4.1.1
stable, if for each e > 0 and x E M, there is a
o = o(x, e) >
such that
S (x, o) R+ ( S (M, e),
equi-stable, if for each x q M, there is a
4.1.2
o = o(x) >
0 such that
xft S(M, o) R+,

uniformly stable, if for each
4.1.3
>
0, there is a
o= o(e) >
0 such
that
S (M, o) R+ ( S (M, e).
4.2
Proposition. If Xis locally compact and Mis compact, then M
is uniformly stable whenever it is either equi-stable or stable (or both).
(The converse holds always.)
Proof. (i) If Mis stable, then for a given e > 0, let ox > 0 be a number
corresponding to xE M such that S(x, ox) R+ C S(M, e). Since
V {S (x, ox): x EM} is an open cover of M, there is a finite open cover,
consisting of {S (xi, Ox.):
i = 1, 2, ... , n; xi EM}. But then there is a
i
o> 0 such that S (M, o) ( V {S (xi, ox.):
i = 1, 2, ... , n}. Notice now
i
that S (M, o) R+ ( [V {S (xi, ox.): i = 1, 2, ... , n}] R+ ( S (M, e). Thus
i
M is uniformly stable.
(ii) Let M be equi-stable. Since Mis compact and Xis locally compact,
there is a o> 0 such that S [M, o] and (hence) H (M, o) are compact.
Then for each xE H(M, o), there is a ox> 0 such that xq S(M, ox) R+.
But then x E @(Sx), where Sx := S (M, ox) R+. Since each @(Sx) is open,
and H(M, o) CV {@(Sx): xE H(M, o)}, we have an open cover of the
compact set H (M, o). Thus there are points x11 x 2 , , xn in H (M, o)
such that H (M, o) ( V {@(Sx.): i == 1, 2, ... , n}. Since V {@(Sx.):
i
i = 1, 2, ... , n} = @(f\ {Sx.:
i = 1, 2, ... , n}) we have f\ {Sx.: i =
i
i
1, 2, ... , n} S(M, e). lfnowo =min {0X1' OX:a' , oxJ, thenS(M, o) R+
Cf\ {Sx.: i = 1, 2, ... , n} C S(M, e). Thus Mis uniformly stable.
1t
1t
85
4.3
Remark. Note that part (i) of the above proof did not use the
fact that X is locally compact. Further uniform stability implies both
stability and equi-stability, but it cannot be asserted that a closed set
which is both stable and equi-stable is uniformly stable.
4.4
Proposition. If a closed set is either stable or equi-stable, then
it is positively invariant.
The proof is simple and is left as an exercise.

We now indicate the connection between various kinds of stability
and Liapunov functions.
4.5
Theorem. A closed set M is stable if and only if there exists a
function 0, there is a ~ > 0 such that ~ whenever
e (x' M) > B; also for any sequence {xn}' 0.
Proof. (i) Sufficiency. Given B > 0, set mo= inf { e}.
By 4.5.2, m 0 > 0. Then for xE M find ~ > 0 such that <p(y) < m0 for
y ES (x, b). This is also possible by 4.5.1 and 4.5.2. We claim that
S (x, ~) R+ C S (M, e). For otherwise there is y ES (x, ~) and t > 0 such
that (! (yt, M) = e. But then inf { e}, as e (yt, M) = e, i.e., m0
This contradiction proves the result.
(ii) Necessity. Let M be stable. Define
<p(x) =sup {1 !(:(x~~)
: t > o}.
This <p (x) is defined on X, and has all the properties required in the theorem. The verification is left to the reader. The theorem is proved.
4.6
Remark. Condition 4.5.2 in the above theorem is equivalent to
the requirement that if {xn} is any sequence such that xn ~ x E M, then
 0, such that lX (e (x, M)).
4.7
Theorem. A closed set M (Xis equi-stable if and only if there
is a function 
4. 7. 2
for every e > 0 there is a
0 for x~ M,
>
<~'and
4. 7. 3
 0.
0 such that 0 such that 0: x~ S(M, ~) R+},
and <p (x) = 0 for x EM. This <p (x) has all the desired properties, the
verification of which is left to the reader. Note that <p (x) < e(x, M).
Remark. Condition 4. 7.2 in the above theorem is equivalent to

4.8
the existence of a continuous strictly increasing function x (r), x (0) = 0,
such that e,
> 0 such that 
> 0 such that 0.
Proof. We leave the details to the reader, but remark that in the proof
of necessity one may choose either of the functions given in the necessity
proofs of Theorems 4.5 and 4. 7.
4.10
Remark.
4.10.1 The theorems above differ from the usual theorems on stability
in that the existence of the functions is shown in all of X rather than in a
small neighborhood of M. Indeed for sufficiency the functions need be
defined on just a neighborhood of M.
4.10.2 It should perhaps be emphasized that in contrast to the situation in section 2, the functions shown to exist in the last three theorems
need not be continuous. To further emphasize this point, we give an
example.
Example. Consider a planar dynamical system given by the

differential system in cartesian coordinates
4.11
4.11.1
The phase portrait (Fig. 4.11.3) consists of a rest point P (the origin of
coordinates) a sequence {Yn} of periodic trajectories,
4.11.2
Yn ={(xv x2}: x~
+ x~ =
1/n},
87
and between any two consecutive Yn the trajectories are spirals which
approach the outer circle as t ~ + oo. The singleton {P} is stable, but
there are no other compact stable sets. No continuous function satisfying
conditions of Theorem 4.5, 4. 7, or 4.9 exists. For if so, then such a function must necessarily be constant on each y n (prove this!) and its value
on any Yn must be less than or equal to that on Yn+i Consequently its
value at P must be greater or equal to that on any Yn
Xz
Fig. 4.11.3
We shall now discuss asymptotic stability of closed sets and its relation with the Liapunov functions.
4.12
Definition. A closed set
MC
X will be said to be
a semi-weak attractor, if for each x EM, there is a ~x > 0, and

for each y E S (x, ~x) there is a sequence {tn} in R, tn ~ + oo, such that
e(ytn, M) ~ 0,
4.12.1
a semi-attractor, if for each x EM, there is a

for each y E S(x, ~x), e(yt, M) ~ 0 as t ~ + oo,
4.12.2
~x
>
0, such that
a weak attractor, if there is a ~ > 0 and for each y E S (M,

there is a sequence {tn} in R, tn ~ + oo, such that (! (ytn, M) ~ 0,
4.12.3
4.12.4 an attractor, if there is a

e(yt, M) ~ 0 as t~ + oo,
>
0 such that for each y E S (M,
~),
~),
a uniform attractor, if there is an X. > 0, and for each e > 0

there is a T = T(e) > 0, such that x[T, + oo) ( S(M, e) for each
x ES [M, x],
4.12.5
an equi-attractor, if it is an attractor, and if there is an A > 0

such that for each e, 0 < e < il, and T > 0, there exists a ~ > 0 with
the property that x[O, TJ (\ S(M, ~) = 0 whenever B < e(x, M) <A,
4.12.6
88
V. Stability Theory
4.12. 7
semi-asymptotically stable, if it is stable and a semi-attractor,
4.12.8
tor,
asymptotically stable, if it is uniformly stable and is an attrac-
4.12.9 uniformly asymptotically stable, if it is uniformly stable and a

uniform attractor.
4.13
Definition. For any
MC X, the set
4.13.1 Aw (M) == {y EX: there is a sequence {tn} in R, tn ~ + (X), such

that (! (ytn, M) ~ {O}} is called the region of weak attraction of M,
and the set
4.13. 2 A (M) == {y E X:
of attraction of M.
4.14
(!
(yt, M)
0 as t ~ + (X)} is called the region
Proposition. If Mis an attractor then Aw (M)
== A (M).
The proof is trivial and is left as an exercise.

4.15
Proposition. If M is a weak attractor (attractor), then A (M)
(A (M)) is an open invariant set which contains S (M, c5) for some c5 > 0.
00
The proof is simple and is left as an exercise.

4.16
Theorem. If a compact set M is a semi-weak attractor (semiattractor), then it is weak attractor (attractor).
The proof is again simple and left as an exercise.
We now discuss the existence of Liapunov functions for various kinds
of asymptotic stability.
4.17
Theorem. A closed set Mis semi-asymptotically stable, if and
only if there exists a function 0 for
x($ M,
4.17.3 there is a strictly increasing function ex(), ex (0)

for > 0, such that
 a (e (x, M)),
0, defined
4.17.4 0, and for each y EM, there is
a c5,, > 0 such that if x ($ JYI, x E S (y, c5y), then 0 and
<p (xt)~ 0 as t ~ + (X).
89
Proof. (i) Sufficiency. Stability follows from Theorem 4.5. The semiattractor property follows from 4.17.3 and 4.17.4.
(ii) Necessity. Consider the function
<p(x)
: o}.
= suP{ 1 !(;(x7.~l t >
This has all the properties 4.17.1 - 4.17.4 except that it may not be
strictly decreasing along trajectories originating in any neighborhood
of points of M. Before proving this we complete our construction. We
define
00
J<p (xT) e-T dt:.
ro (x)
This w(x) has all the properties 4.17.1 - 4.17.4 except possibly 4.17.3.
The construction is now completed by setting
(P (x)
 0 such that
<p (x) is continuous in S (y, ~,,), we need prove that <p (x) is continuous in
an open set containing M. If A (M) is the region of attraction of M, then
A (M) is invariant, and indeed J (A (M)) is also invariant and open and
contains Af. We now define the set W8 = {xE J (A (M)): y+(x) CS (M, e)}.
This We js open positively invariant, and contains an open set containing
M, and has the important property that if x E J (A (M)}, then there is a
T > 0 such that xT E W 8 Now let x E J (A (M)) and let e(x, M) ==A.
There is a T > 0 such that xT E WA/ 4 Since WA/ 4 is open we can find
a neighborhood S (xT, er) C WA14 Then S (xT, er) (-T) = N is a neighborhood of x, and indeed NC J (A (M)). We can thus choose an 'Y/ > 0
such that 17 < .A/4, and S(x, 'Y/) C N. Then if yE S(x, 'Yj),
<p(x) - <p(y) =sup { 1
-- sup {-(!+
1
(xt, M)
.
e(xt, M) .
(! (xt,
M)
+ e(xt,M)
0< t<
== =
: t
> 0 - sup
(! (yt,
M)
+ e(yt, M)
r} - sup { +
1
(!
(yt, M)
.
e(yt, M) .
0<
= t
: t
> 0
<
r} ,
==
and so
{j
_ {I
- sup
l<p(x) - <p(y) I< sup
!(:(x7,~)
!(~'(;~~) I: 0 < t <
e(xt, M) - e(yt, M)
r}
I0 =< t ==< r}
+ (! (xt, M)) (1 + (! (yt, M))

<sup {le(xt, M) - e(yt, M) I: o < t <
(1
T}
< sup {e (xt, yt): O < t < T}.

By the continuity axiom the right hand side tends to zero as y ~ x,
hence <p (x) is continuous in J (A (M)). The rest of the observations on
<p (x), ro (x) are easy to verify and are left as an exercise.
90
V. Stability Theory
4.18
Theorem. Let M be a closed set. Then Mis asymptotically stable
if and only if there is a function 0,
4.18.2
<p(x) = 0 for xE M, <p(x)
>
0 for x4 M,
4.18.3 there exist strictly increasing functions <X (),

f3 (0) == 0, defined for > 0, such that
' (e (x, M)) <
 0, and there is a~ > 0 such that
if xE S(M, ~) - M then <p(xt) < <p(x) for t > 0, and <p(xt) ~ 0 as
t~ + (X).
The proof follows exactly the same lines as that of the previous theorem and is left as an exercise. We note, however, that in the proof of
necessity, since A (M) is open and invariant, and A (M) ) S (M, ~) for
some ~ > 0, the functions <p (x) and ro (x) can be taken as being defined
and continuous on A (M). In the present case (x) == <p (x) + w (x)
will have all the desired properties.
We shall now prove the following very important theorem, which in
the case of asymptotic stability of a closed invariant set M, describes the
flow in the set A (M) - M.
4.19
Theorem. If a closed invariant set M C X is asymptotically
stable, then for each x E A (M), ]+ (x) C M, and for each x E A (M) - M,
J-(x) (\A (M) = 0.
Proof. Let, if possible, x* E A (M) and y E J+ (x*), y 4 M. Set (! (y, M)
= <X (> 0). Since M is uniformly stable, there is a ~ > 0 such that
y+(S(M, ~)) C S(M, <X/2). Since x* EA (M), there is a T > 0 such that
x*T ES (M, ~). Since S (M, ~) is open, there is an 'Y/ > 0 such that
S(x*T, 'Y/) ( S(M, ~).Now N-::::::. S(x*T, 'Y/) (-T) is a neighborhood of
x* such that for each x EN, xT ES (x*T, 'Y/) and consequently x [T, + (X))
C S(M,<X/2). Now since yEJ+(x*}, there exist sequences {xn} in X and
{tn} in R+, tn~ + oo, such that xn ~ x*, xntn~Y We may assume without loss of generality, that Xn EN, and tn > T. But then xntn ES (M, <X/2}.
Thus if xin ~ y, we must have e(y, M) < <X/2. This is a contradiction,
as e (y, M) == <X. Thus J+ (x*) CM. The second statement follows from
the fact that if yE J-(x), then xE J+(y). Now let xE A (M) - M, and
assume that yEJ-(x)(\A(M). Then we have yEA(M), xE]+(y),
x4 M, which has already been ruled out.
91
4.20
Corollary. Let a closed invariant set M be asymptotically stable.
Let A (M) - M (or in particular the space X) be locally compact and
contain a countable dense subset. Then the invariant set A (M) - M is
parallelizable.
The proof follows from the above theorem, and IV, 2.6.
Remark. The considerations in section 2 show that for X locally
4.21
compact and M C X a compact invariant asymptotically stable set, if
 0 such
that 5 [M, e] C N and X sufficiently small, represent sections of the
parallelizable flow in A (M) - M. How far the same method of construction can be extended to non-compact closed sets, depends naturally on
whether the flow in A (M) - M is parallelizable.
We shall now prove that uniform asymptotic stability of a closed set

M ( X implies that the flow in A (M) - M is parallelizable, even when
the subspace A (M) - M of X is assumed to be neither locally compact
nor separable.
4.22
Proposition. Let M C X be a closed invariant uniformly asymptotically stable set with A (M) as its region of attraction. Then A (M)
- M is parallelizable.
Proof. Since M is asymptotically stable, we can find a function 0 such that
5 [M, X] (A (M), and such that for any e > 0, there is a T > 0 with
the property that y+ (xT) C 5 (M, e) for every x E 5 [M, lX]. Now let
m0 =inf { 0. Consider now any set

5 11 === {x E 5 [M, lX]: <p (x) === 17}.
We claim that if 17 < m0 , then 5 11 is a section of the flow in A (M) - M

with the property that there is a unique continuous function -r (x),
mapping A (M) - M into R such that for each x E A (M) - M,
x-r(x) E 5 17 The existence of such a section indeed shows that the flow
on A (M) - M is parallelizable (IV, 2.4). To see that 5 11 has the properties enunciated above, we consider the set P 11 =={xE5[M,X]:
<p(x) < 17}. Indeed P 11 CA (M), and Pf/) M. We note now that any
trajectory in A (M) - M can intersect 5 11 at most at one point. This is
so because if any trajectory in A (M) - M has two points x1 , x2 on 5 11 ,
then we may assume that x 2 = x 1t where t > 0. But then since
92
V. Stability Theory
5 11 (\ M == 0, <p(x2) == <p(x1t) < <p(x1), which contradicts the definition

of S,t To see that every trajectory yin A (M) - M intersects S" we note
first that if x~ P 11 , there is at> 0 such that xtE P 11 But then x[O, t]
(\ oP11 =F 0. However, 51/ == oP11 If x E P 11 , then we claim that there is a
t < 0 such that xtE S 1r For otherwise y-(x) C JP 11 In this case we can
set ~ ==inf {e (xt, M): t < O}, and ~ > 0 (otherwise M will be unstable).
If now T > 0 is such that y ES [M, X] implies e (yT, M) < ~/2 (such
T > 0 exists by uniform asymptotic stability), then x (-T) == y E P 11
CS [M, lX], but yT = x~ S (M, ~/2}. This contradiction shows that
every trajectory in A (M) - M intersects 5 11 exactly once. We now
define the function -r: A (M) - M ~ R by the requirement that x-r (x) ES 11
for xE A (M) - M. Then -r(x) is uniquely defined and is continuous.
The continuity follows in the same way as in the proof of Theorem 2.9.
\Ve have thus proved our proposition.
\Ve shall now prove the following theorem.

Theorem. A closed invariant set M is uniformly asymptotically
4.23
stable and equi-attracting with an open set N containing S (M, ~) for
some ~ > 0 as its region of attraction, if and only if there exists a continuous function 
0 for x~ M;
there exist strictly increasing continuous functions

X (0) == f3 (0) == 0 such that
4.23.2
lX (e (x,
M))
<
 O; V {En: n == 1, 2, 3, ... } == N, and this
sequence has the property that for any lX

that lX for x ~ Ene;
4.23.4
>
0 there is an integer n 0 such
<p (xt) == e-t <p (x) for x EN, t ER.
Proof. We shall not give complete details as the arguments are similar
to those used earlier. For the proof of sufficiency we remark that 4.23.3
and 4.23.4 imply that N is invariant. 4.23.1, 4.23.2, and 4.23.4 ensure
uniform stability, as well as attraction, and show that N (A (M). Since
N is an invariant neighborhood, it follows that N == A (M). Uniform
attraction and equi-attraction follow from 4.23.2 and 4.23.4. To prove
necessity, we consider the region of attraction A (M) and define a <p (x)
on A (M) as in Theorem 4.18. We then consider a section S" of the flow
in A (M) - M defined by this <p (x) (Proposition 4. 22), with the corre-
93
sponding continuous map T: A (M) - M
<p (x)
for
eT(z)
R. Lastly we define
x E A (M) - M,
and
<p (x)
for
x EM.
This p(x) is easily shown to have all the properties 4.23.1-4.23.4. Note
that to get the sequence {En}, we set E 0 = {x: <p(x) < 1}. Then define
En = E 0 [-n, OJ. These sets are closed and have the required properties.
Setting <P(x)
corollary.
= -
:<~(x)
, we obtain the following very important
4.24
Corollary. A closed invariant set M is uniformly asymptotically
stable and equi-attracting, with an open set N containing S (M, ~) for
some ~ > 0 as its region of attraction, if and only if there exists a continuous function </> (x) defined on N with the following properties:
4.24.1
-1<4>(x) < 0
4.24.2
4> (x)
xE N - M,
for
e(x, M) ~ 0,
as
4.24.3 for any e > 0 there is a ~

e(x, M) > ~,
~
4.24.4
4>(x)
4.24.5
da> (xt)
dt
-1
It=O =
as
x~
_q; (x) (1
>
0 such that </>(x)
< -e for
YE 8N,
+ q> (x)).
We shall now give a theorem along the lines of the Theorem 4.18 for
the case of uniform asymptotic stability.
Theorem. Let the space X be locally compact and separable.

Then a closed set M C X is uniformly asymptotically stable, with an
open set N containing S (M, ~) for som~ ~ > 0 as its region of attraction,
if and only if there exists a continuous function 
0 for xtl M,
there exist continuous strictly increasing functions

{J (0), such that
lX (e
4.25.3
(x, M))
<

0 there is an n0 such that lX
94
V. Stability Theory
The conditions can easily be shown to be sufficient. To prove the

necessity we need the following lemma.
4.26
Lemma. Let f (r, x) be a function from (0, 1] XX~ [O, + oo),
where X is locally compact separable metric space. Let f (r, x) be bounded
on every compact subset of (0, 1] XX. Then there exist two functions
H(r) and G(x) defined on (0, 1] and X respectively, which are bounded
on compact subsets of (0, 1] and X respectively (and may even be chosen
continuous), such that
f (r, x) < H (r) G (x).

Proof. Since Xis locally compact and separable we can find a sequence
00
of compact sets Un such that Un ( Un+l and X =
Un. We now
n=1
defineH(r) ==sup {f(r, x) + 1: xE Un where 1+1/r > n0 > 1/r},
rt(:i) :
and G (x) = sup

1 > r > 0}. The above defined H (r) and G (x)
have the required properties. Indeed H (r) is defined here as a step
function.
Proof of Necessity of Theorem 4.25. Let A (M) be the region of attraction of the set M. Since M is an attractor, there is an X > 0 such that
S (M, X) CA (M). We may choose X < 1. For each r E (0, X) define
T(r,x) =inf{r> 0: xtES(M,r) for t>i'}. We assert that T(r,x) is
bounded on any compact set KC A (M) and for fixed r. To prove this
we need to show that for each compact KC A (M), and r > 0, there
exists a T > 0 such that Kt CS (M, r) fort> T. We note first that by
stability of M, there is a <5 > 0 such that y ES (M, <5) implies y+(y)
( S(M, r). For xE A (M), choose T(x) such that xT(x) E S(M, <5). Since
S(M, <5) is open, we can choose a <J > 0 such that S(xT(x), a) is compact
and contained in S (M, <5). Then its inverse image
Nz = S(xT(x), a) (-T(x))
is a compact neighborhood of x. Nz has, moreover, the property that
NzT(x) C S(M, <5). Thus we have in fact shown that for each xE A (M),
there exists a T (x) and a (! (x) > 0 such that y E S (x, (! (x)) implies
ytES(M,r) for t>T(x). Consider now the open cover {S(x,e(x)):
x E K} of the compact set K. By the Heine-Borel theorem, there exist
a finite number of sets, say, S (x1 , (! (x1}), ... , S (xn, (! (xn)) which cover K.
We can now choose T = max (T (x1), ... , T (xn)). Then x E K implies
xtE S(M, r) fort> T.
For any given integer n > 1/X, define
<pn (x)
== sup {e (xT, S (M, 1/n)) exp (T): T >
O}.
95
We assert that <l'n (x) is continuous on A (M). To see this, note that for
e > 0 such that S(x, e) is a compact subset of A (M), there exists a
T > 0 such that
e(y-r, S (M, 1/n)) = 0
for y Es (x, e) and
>
l<l'n (x) - <f'n (y) I =
T. Therefore, if y Es (x, e) we have

J
sup {e (xT, S (M, 1/n)) exp (7:): 0
<
< T}
- sup {e(yT, S(M, 1/n)) exp (T): 0 < T < T}j
< exp (T) . sup {I e (xT, s (M, 1/n)) - e(yT, s (M, 1/n)) I: 0 <
< T}.
This implies that

l<l'n (x) - <f'n (y) I < exp (T) sup {e (xT, YT): 0 < T < T}.
Using the continuity axiom we conclude that the right hand side tends
to zero as e (x, y)-+ 0. Thus <l'n (x) is continuous on A (M). This <l'n (x)
has further the following important property
<f'n (xt)
< exp (-t) <f'n (x),
for
>
0.
To see this, note that fort> 0

<l'n (xt) =sup {e (x (t
+ T), S (M, 1/n)) exp (7:):
T > O}
= sup {e (xT, S (M, 1/n)) exp (T - t): T > t}

=exp (-t) sup {e (xT, S (M, 1/n)) exp (T): T > t}
< exp (-t) <l'n (x) as t > 0.

We now note that
<l'n (x) ==sup {e (xT, S (M, 1/n)) exp (7:): 0 <
as e(xT, S(M, 1/n)) == 0 for

<l'n (x)
>
< T (1/n, x)}
T(1/n, x). Thus
< exp (T (1/n, x)) sup {e (xT, S (M, 1/n)): -r > O}.
Since the function exp (T(lfn, x)) has the properties of the function
f(r, x) of Lemma 4.26, we can choose a function H(r), such that <l'n(x)/
H (1/n) is uniformly bounded on each compact subset KC A (M).
We now define
00
</J (x) == I <f'n (x) /H (1/n) n !, where n 0

n=n0
> 1/X.
Then </J (x) is continuous on A (M) and

</J (xt)
< e-t </J (x).
Note that </J(x) == 0 for xE M, and </J(x) > 0 for x~ M. By uniform

asymptotic 'stability, </J (x) -+ 0 if e (x, M) -+ 0; thus there is a strictly
96
V. Stability Theory
increasing function
fJ (r), fJ (0)
0 such that
(!) (x)
< fJ (e (x, M)).
Further if e(x, M) > e > 0, then for sufficiently large n, <fJn (x) > ~n > 0
for some !5n. And hence (!) (x) > !5 > 0. Thus there is a strictly increasing
function x(r), X(0) = 0 such that (l)(x) > x(e(x, M)). We now choose
k > 0 such that
k < inf {(!) (x) : e(x, M) = X}.
Consider the sets
Pk = {x EA (M): (l)(x)
<
k} f\ S [M, X],
and
Sk ={xEA(M):(l)(x) =k}f\S[M,X].
Then as shown in Theorem 4.23, Sk is the section of the flow in A (M)

consisting of all those trajectories in A (M) which are not in M. For each
xE A (M), x~ Pk, we can define T(x) by the requirement that xT(x) E Sk.
Then T (x) is continuous. Now define
and
This <p (x) has all the properties required in the theorem as may easily be
verified.
4.27
Remark. Note that if in the above theorem we assume M to be
invariant and construct <p (x) as in Theorem 4. 23, then this <p (x) need not
satisfy the property 4.25.2 in the above theorem. Indeed if that were
the case, then construction of the <fJn (x) would be superfluous and then
uniform asymptotic stability would imply equi-attraction, which is
indeed not the case.
4.28
Exercises.
4.28.1 Give proofs or provide the missing details in the proofs of

Proposition 4.4, Theorems 4.5, 4. 7, 4.9, Propositions 4.14, 4.15, Theorems 4.16, 4.17, 4.18, 4.20, 4.23, 4.24, 4.25.
4.28.2 Show that in a locally compact metric space X, a compact set
M is stable if and only if there exists a lower semi-continuous function
 0 for x EE M, (b} <p (x) is continuous on M,
and (c) <p (xt) < <p (x) if x [O, t] C N.
4.28.3 Give examples to show that the various concepts introduced
in this section are indeed distinct concepts.
97
4. 28.4 In the following figures, trajectories of certain dynamical

systems defined in the euclidean plane R 2 are shown. The x1-axis in
each case has an appropriate stability or attractor property, but no
stronger property is implied. Such a property is written just below the
figure. Construct examples of differential equations which conform to
the behavior indicated in each figure.
Fig. 4.28.5. Stable but not equistable
x,
Fig. 4.28.6. Equistable but not stable
Xzt
I
---+-l~.,_------~-1~'-----------------
I
I
I
-11I
I
I
Fig. 4.28.7. Semi-weak attractor

Xz
x,
Fig. 4.28.8. Semi-attractor
98
V. Stability Theory
Xz
Fig.4.28.9. Weak attractor
Xz
Fig. 4.28.10. Attractor
Xz
KI
Fig.4.28.11. Uniformly stable semi-attractor
Xz
Fig.4.28.12. Stable attractor
99
5. Relative Stability Properties
4.28.13 In the euclidean plane, consider the system
.
X1
= 1,
X2
= -
The solution through any point

X1=
2x1x 2
(1 + x~)
(x~,
for
X1
> 0.
xg) has the form
(x?)2
o f
>
o
+ X1,o X2==1 1+ +(t+x?)2X2
or t =-Xi,
and
The x1-axis is a uniformly stable attractor, but is not a uniform attractor.
4.29
Remark. The notions in this section are described in terms of
positive semi-trajectories and/or using sequences {xtn} with tn ~ + CX>
(on a positive semi-trajectory y+ (x)). It is clear that notions dual to
these may be obtained by using negative semi-trajectories and/or using
sequences {xtn} with tn ~ - CX> on a negative semi-trajectory y-(x). The
dual notions, whenever needed, will be described by using the adjective
negative before the appropriate term describing that notion. One should
also note that some of these negative notions have in the literature been
termed as instability notions.

In what follows the space X is assumed to be locally compact. In
this section we shall use the concept of relative prolongation D+(M, U)
and of relative prolongational limit set J+(M, U) (Definition II, 4.10)
of a set MEX, relative to the set UC X. These concepts will allow
us to characterize the properties of relative stability and attraction
of a compact set.
5.1
Definition. Given a compact set M C X and a set U ( X, the
set M is said to be
5.1.1
!5
>
5.1.2
stable, relatively to the set U, if given an s

0, such that
y+(S(M, !5) f\ U) ( S(M, s);
>
a weak attractor relative to U, if
A+(x) (\ M =F 0, for each xE U;

5.1.3
an attractor relative to U, if
A+(x) =F 0, A+(x) CM, for each xE U;

7*
0 there exists
100
5.1.4
V. Stability Theory
a uniform attractor relative to U, if
j+(x, U) =F 0, J+(x, U)
CM,
for each xE U;
5.1.5
asymptotically stable, relatively to U if Mis a uniform attractor relative to U and it is positively invariant.
5.2
Remark. If in Definition 5.1 U is neighborhood of M, then the
stability, weak attraction, attraction, uniform attraction and asymptotic
stability of M, relative to U, reduces to the stability, weak attraction,
attraction, uniform attraction and asymptotic stability of M, respectively, as defined in Definition 1.5.
5.3
Theorem. A compact set M C X is stable relatively to U ( X
if and only if M) D+(M, U).
Proof. (i) Sufficiency. Let M) D+(M, U) and let, if possible, M be

not stable relative to U. Then there is an e > 0; a sequence {xn} in
U, xn ~ x E M, and a sequence {tn}, tn > 0, such that f! (xin' M) = B.
We may assume that H(M, e) is compad. Thus the sequence {xntn} may
be assumed to converge to a point y E H (M, e). Then y E D+ (x, U)
( D+(M, U), but y~ M. This contradiction proves sufficiency.
(ii) Necessity. Let M be relatively stable with respect to U. Then
clearly D+(M, U) CS [M, e] for arbitrary B > 0. Hence D+(M, U)
( (\ {S [M, e]: B > O} = M. This proves the theorem.
5.4
Remark. In the above theorem or definitions, the set U need
not contain the point x or the set M. If, however, U) M, then in the
above theorem we may replace condition M) D+(M, U) by M =
D+(M, U). Further, in case U is a neighborhood of M, one obtains
Theorem 1.12.
We have now a theorem for the characterization of relative uniform
attraction. I ts proof follows that of the analogous proposition 1. 2 and it is
left as an exercise to the reader.
5.5
Theorem. A necessary and sufficient condition for a compact
set M to be a uniform attractor relative to U C X is that for each B > 0
and each compact set KC U we can find a real number T(K, e) > 0,
such that Kt C S (M, e) for each t > T.
We shall now prove a theorem (5.10) on relative weak attraction,
which is an extension of the weak attractor theorem (1.25). Its proof
will be based upon the two following Lemmas 5.6 and 5.8.
5.6
Lemma. Let M ( X be a compact set such that A' (M)
Let U ( A' (M) be a set with the following properties:
5.6.1
U is closed and positively invariant.
5.6.2
f\
101
=F M.
=F 0.
Then the set D+(M, U) is non-empty and compact and in addition,

givens> 0 we can find T > 0 such that
5.6.3
D+ (M, U) ( S [M, s] [O, T].
Proof. Choose s > 0 so small that the set V = H (M, s) f\ U is nonempty and compact. Let
Tx
=Inf {tE R+: xtE S(M, s)}
and
T =Sup {Tx: xE V}.

Clearly T < +ex>. If not there would exist a sequence {xn} ( V such
that rxn-+ +ex>. Since Vis compact there exists a point x E V such that
xn-+ x and XT ES (M, s), T > 0. Then Txn < T for a sufficiently large n;
which is absurd.
Let now
yE D+(M, U) - S[M, s].
Then there exist sequences {xn} C U and {tn} CR+ such that xn-+ x EM
and xntn-+ y.
Notice that from the hypothesis 5.6.1, y+ (~) ( U, xn E S (M, s) and
xntn E @{S (M, s)} for sufficiently large values of n. Hence there exists
a sequence {Tn} ( R+ such that xnrn EH (M, s) (\ U = V and xnt~ S [M,s]
for t E (Tn, tn]. Since x"tn = Xn'Z'n (tn - rn) and xnrn E V, it follows that
xntn E V [O, T] ( S [M, s] [O, T].
5.7
Remark. The hypothesis 5.6.1 is necessary for Lemma 5.6
to be true. Consider for that the planar flow shown in Fig. 5. 7.1, where P
is a rest point.
Fig.5.7.1
102
V. Stability Theory
In this case the region of weak attraction of P is not closed and if

U = A'(P), the set D+(M, U) is not compact. Consider next the planar
flow shown in Fig. 5. 7. 2.
- - - -I'-
--
Fig.5.7.2
In this flow the points P and Qare rest points, and A' (P) = R 2 - { QV y}.
Let U be the closed set above the dotted line I', this set is closed, contained in Aw (P), but it is not positively invariant. In this case we have
QE D+(P, U) and therefore Lemma 5.6 does not hold.
The hypothesis 5.6.2 on the other hand ensures that Lemma 5.6
is not trivial. In fact since U is closed, U f\ M =F 0 would imply that
D+(M, U) =t= 0.
Lemma. Let x E X and U C X be positively invariant, then

z EA+ (x) implies
J+ (x, U) ~ [)+ (z, U) .
6.8
Proof. If x~ Uthe theorem is trivially true. Assume then that xE U.

If z E A+ (x) and y E J+ (x, U) there exist then the following sequences:
+ oo, {xT:} C U, XT~ ~ z.
5.8.1
{'r:} CR+, T: ~
5.8.2
{xn} ( U, Xn ~ X.
5.8.3
{(} ( R+, t: ~
+ oo, {xnt:} (
U, xn( ~ y.
t: -
We can assume, possibly by choosing suitable subsequences that

T: > 0 for each n. Fork fixed, consider the sequence {xnT~} ~ U. We have
xnT~ ~ xr~ E U. We can then assume that, for each k,
5. 8.4
() (XnT~, XT~)
< 1/k for each n > k.
Now
5.8.5
as
n~
() (z, XnT:)
< () (z, XT~)
+ () (XT~, XnT:) < () (z, XT~) + 1/n ~ 0
+ oo.
On the other hand xnt: = XnT: (t: - r:)

{tn - T~} ( R+. Thus y E D+(z, U).
y and {XnT:} ( U, xnr: ~ z,
103
5.9
Remark. In Lemma 5.8 the set U does not have to be closed.
On the other hand the positive invariance of U is essential. This can be
shown from the following planar flow (Fig.5.9.1), in which zEA+(x)
and D+(z, U) = 0, while J+(x, U) =F 0.
Fig.5.9.1
We shall proceed next to the proof of the theorem on the prolongation of a relative weak attractor.
5.10
Theorem. Let M C X be compact and such that A 00 (M) - M
=f= 0. Let U ~ Aw (M) be a set with the following properties:
5.10.1
U is closed and positively invariant.
5.10.2
U f\ M =f= 0.
Then the set D+ (M, U) is compact and asymptotically stable relatively

to U.
Proof. From Lemma 5.6 we have that D+(M, U) is a compact set,

which is contained in UC Aw (M). We shall see that J+ [D+(M, U), U]
C D+(M, U). Let x E D+(M, U) ( U, thenA+(x) f\ M =F 0. Let z E A+(x)
(\ M. From Lemma 5.8 it follows that J+(x, U) ~ D+(z, U) ( D+(M, U).
Thus D+(M, U) is a uniform attractor relative to U. Since D+(M, U)
is also positively invariant the theorem is proved.
5.11
Remark. From what was previously seen, it follows that a
necessary and sufficient condition for a compact set M C X to be asymptotically stable relatively to a closed and positively invariant set U ( X
with M ( U, is that Mis a stable weak attractor relative to U.
5.12
Theorem. Let M C X be compact and positively invariant and
let M ( M be the largest invariant set contained in M. Then M is a
stable attractor, re]ative to M.
Proof. For any x E M, A+ (x) =f= 0 and compact, since y+ (x) C M.

Again A+(x) CM, because A+(x) is an invariant subset of M. Hence M
is an attractor, relative to M. To see that it is stable, one easily shows
104
V. Stability Theory
that J+ (M, M) is an invariant subset of M; then by definition of M,

D+ (M, M) ( It! holds, so that M is stable relative to M by 5.3.
5.13
Remark. In the case in which in Theorem 5.12 we have that
M (..IM, Mis indeed asymptotically stable.
The concepts of relative attractor and relative asymptotic stability

will be extensively used in Chapters VIII and IX. On the other hand the
concept of relative stability may be motivated by considering the example
of a limit cycle y in the plane, with the property that all trajectories
outside the disc bounded by the limit cycle y, has y as their sole positive
limit set, and all trajectories in the interior of the disc bounded by y tend
to the equilibrium point 0.
Fig.5.13.1
Notice that if U is the complement of the disc bounded by y, then y

is relatively stable with respect to U. Notice also that if y is an asymptotically stable limit cycle, then y is stable with respect to every component
of R 2 - y. These considerations lead to the following definition and theorem.
5.14
Definition. Let MC X be compact. We say that Mis component-wise stable if M is relatively stable with respect to every component
of X - M.
We have then
5.15
Theorem. Let a compact set M ( X be positively stable. Then
M is component-wise stable.
The proof is obvious and is omitted.

The converse of Theorem 5.15 is in general not true. To see this, we
consider a simple example.
5.16
Example. Let X C R 2 be given by X == {(x, y) E R 2 : y = 1/n, n
any integer, or y == O}. The space Xis a metric space with the distance
105
between any two points being the euclidean distance between the points
in R2 \Ve define a dynamical system on X by the differential equations
5.16.1
x == lyf,
y=
0.
Then the set {(O, O)} (Xis component-wise stable, but is not stable.
The question now arises, as to when the converse of Theorem 5.15
is true. For this purpose the following definition is convenient.
Definition. Let M C X be compact. We shall say that the pair

(M, X) is stability-additive if the converse of Theorem 5.15 holds for
every dynamical system defined on X which admits M as an invariant
set.
5.17
In this connection the following theorems are important.
5.18
Theorem. The pair (M, X) is stability-additive if X - M has a
finite number of components.
5.19
Theorem. The pair {M, X) is stability-additive if X - M is
locally connected.
The proof of Theorem 5.18 is immediate and is left as an exercise.
(It is indeed a corollary of Theorem 5.19.)
5.20
Proof of Theorem 5.19. Let M be a compact invariant set for a
given dynamical system on X and let M be component-wise stable. Since
Mis locally compact, there is an e > 0 such that S [M, e], and hence also
H (M, e), is compact. We claim that only a finite number of components
of X - M can intersect H(M, e). For otherwise, if an infinite number
of components of X - M intersect H(M, e), then we may choose a sequence of points {xn} in H(M, e) such that no two points of the sequence
are in the same component. Since H (M, e) is compact we may assume
that xn ~ x E H (M, e). Since Xis locally connected, there is a neighborhood of x, say N, such that N is a subset of a component of X - M.
Now there is an integer n0 such that xn EN for n > n0 and hence all xn
for n > n0 belong to the same component of X - M, which is a contradiction. Now notice that every component of X - Mis an invariant set.
If C1 , C2 , , CP are the components of X - M which intersect H(M, e),
then by component-wise stability we have positive numbers <5v <52, , <5P
such that y+ (S(M, <5i) (\ Ci) CS (M, e), i = 1, 2, ... , p. If now <5 ==
min (<51 , <5 2, ,<5p) we get y+(S(M,<5)) (S(M,e), i.e., Mis stable. To
see this last assertion, note that if x E S (M, <5), then either x E Ci for
some i = 1, 2, ... , p and hence y+ (x) CS (M, e), or xis an element of a
component of X - M which does not intersect H (M, e) and hence is
106
V. Stability Theory
contained in S (M, e}, or x EM. In either of the last two cases y+ (x)
S (M, e). The theorem is proved.
6 . Stability of a Motion and Almost Periodic Motions

In this section we shall assume throughout that the metric space X is
complete.
The concept of almost periodicity is intermediate between that of
periodicity and recurrence, and the concept of stability of motion plays
a central role in its study. We therefore first introduce the concept of
stability of a motion.
6.1
Definition. A motion nx is said to be positively (Liapunov)
stable in a subset N of X if for any e > 0, there is a ~ > 0 such that
y E N f\ S (x, ~) implies e(xt, yt) < e for t E R+.
Any motion nx is called negatively stable, or stable in both directions
in a subset N of X, if the above condition is satisfied with t ER+ replaced
by t E R- or t E R, respectively.
If in the above definition N is a neighborhood of x, then the qualifier
"in the subset N of X" will be deleted. Thus a motion nx is positively
stable if given e > 0, there is a ~ > 0 such that y E S (x, ~) implies
e(xt, yt) < e for t E R+.
6.2
Exercise. Show that a motion nx is positively stable if and only
if every motion nxt' t ER, is positively stable.
6.3
Definition. If A CBC X, then the motions through A (i.e.,
motions nx with x EA) will be called uniformly positively stable, uniformly
negatively stable, or uniformly stable in both directions in B, if given any
e > 0, there is a~> 0 such that e(xt, yt) < e for tE R+, tE R-, or tE R
respectively whenever x EA, y EB, and e (x, y) < ~6.4
Exercise. Show that if A is a compact subset of B, then the
motions through A are uniformly positively stable in B, whenever each
motion through A is positively stable in B.
We now introduce the concept of almost periodicity.
6.5
Definition. A motion nx is said to be almost periodic if for every
e > 0 there exists a relatively dense subset of numbers frn}, calle:d displacements, such that
for all t ER and each
Tn
107
It is obvious that periodic motions and rest points are special cases
of almost periodic motions. That every almost periodic motion is recurrent
follows from Theorem III, 3.12 and we leave this as an exercise. Later in
this section we shall consider examples to show that not every recurrent
motion is almost periodic, and that an almost periodic motion need
not be periodic.
The following theorems show how stability is deeply connected with
almost periodic motions. First observe the following lemma.
6.6
Lemma. If a motion nx is almost periodic, then every motion n'Y
with y E y (x) is almost periodic with the same set of displacements frn}
corresponding to a given e > 0.
Proof. Indeed for any e > 0 there is a set of displacements {rn} such that
e(xt, x(t + -rn))
<
e for tE R, and each -rn.
If y E y (x}, then there is a -r ER such that y = x-r, or x = y (--r). The

above inequality together with the group axiom then gives
e(y(t - -r), y(t - -r - -rn))

Setting t - -r = s, we see that
< e for tE R.
e(ys, y (s + -rn}) < e for s ER and each -rn,

as -r is fixed. This proves the lemma.
6. 7
Theorem. Let the motion nx be almost periodic and let y (x)
be compact. Then
6. 7.1
every motion n,, with y E y (x) is almost periodic with the same
set of displacements {-rn} for any given e > 0, but with the strict inequality< replaced by <;
6. 7.2
the motion nx is stable in both directions in y (x).
Proof of 6.7.1. For any yE y (x}, there is a sequence {xn} Cy (x) such
that xn ~ y. By Lemma 6.6, for any e > 0 there is a set of displacements {-rn}
such that e(xi, xn (t +Tm))< 8 for all t ER, xn E {xn}, and Tm E {-rn}
Now keeping t ER, and Tm E {-rn} fixed but arbitrary and proceeding to
the limit we get e(yt, y (t + Tm)) < 8 for all t ER and Tm E {-rn} This
completes the proof of 6. 7.1.
Proof of 6.7.2. Given e > 0, let {-rn} be a set of displacements
corresponding to e/3 for the almost periodic motion nx, and let T > 0
be such that {-rn}f\ [t- T, t + T] =t= 0 fort ER. Then by 6. 7.1, e(yt, y (t +
-rn)) < e/3 for all y E y (x), t ER, and each in By the continuity axiom,
for e/3 > 0 and T > 0 as above, there is a ~ > 0 such that e (y, z) < ~
108
V. Stability Theory
implies e (yt, zt) < e/3 for all ftl < T, whenever {y, z} Cy (x) as this last
set is compact. Now for any y E y (x) and e (x, y) < ~, we have for any
tE R
(! (xt, yt)
< (! (xt, x(t + Tn)) + (! (x(t + Tn), Y (t + Tn))
+ (!(y(t + Tn), yt) < e/3 + e/3 -I- e/3=e,

because for any t ER we can choose Tn such that (t
ves the theorem completely.
+ Tn I <
T. This pro-
6.8
Corollary. If Mis a compact minimal set, and if one motion in
M is almost periodic, then every motion in M is almost periodic.
6.9
Corollary. If M is a compact minimal set of almost periodic
motions, then the motions through M are uniformly stable in both directions in M.
The above corollary follows from 6. 7.2, and from Exercise 6.4.
We now investigate when a recurrent motion is almost periodic.
6.10
Theorem. If a motion nx is recurrent and stable in both directions in y (x), then it is almost periodic.
Proof. We have indeed that given e > 0 there is a ~ > 0 such that
e(xt, yt) < e for all t E R, whenever {x, y} C y (x) and e(x, y) < ~- Further, by recurrence of nx (Corollary III, 3.10 and Theorem III, 3.12),
there is a relatively dense set {rn} of displacements such that
(! (x, x-rn)
<
for each Tn.
From the above two results we conclude
e(xt, x(t
+ Tn)) <
for tE R
and each Tn. The theorem is proved.

A stronger result is the following:
6.11
Theorem. If a motion nx is recurrent and positively stable in
y (x), then it is almost periodic.
Proof.
(a) By positive stability of nx in y (x), we have given e > 0 there is a
~ > 0 such that e(x, x-r) < ~ implies e(xt, x (t + -r)) < e/2 for all t ER+.
(b} By recurrence of nx, there is a relatively dense set {-rn} sucq. that
(! (x, x-rn) < ~/2 for each Tn.
(c) By the continuity axiom, for any Tn there is a <J > 0 such that
(! (x, y) < <1 implies (! (x-rn, YTn) < ~/2. Now let t ER and i-n be arbitrary
109
but fixed. Then by recurrence of :n;x, there is a 't' < t such that e (x, xi)
<min (a, <5). Then by (c) e(x-rn, x(i- +in))< <5/2, so that
(! ( X, X
(i
+ in)) <
(!
(x, Xin)
+ (! (xin, X ('t' + in)) <
<5/2
<5/2 = <5.
Hence by (a), since t - -r > 0, we get e(x(t - -r), x(t +in))< e/2.
Further, e(xt, x(t - -r)) = e(x-r(t - -r), x(t - -r)) < e/2 by (a), as
e (x, xi) <min (a, <5) < <5, and t - 't' > 0. Thus we get
e(xt, x (t +in))< e(xt, x (t - -r))

This shows that
:n;x
+ e(x (t -
-r), x (t
+ 't'n)) <
e/2
+ e/2 == 8.
is almost periodic and the theorem is proved.
Theorem. If the motions in y (x) are uniformly positively stable

6.12
in y (x) and are negatively Lagrange stable, then they are almost peri_odic.
Proof. It is sufficient to prove that the motion :n;x is recurrent, as

the rest follows from the last theorem.
By negative Lagrange stability of :n;x we conclude that A- (x) is
compact, and indeed A-(x) C y(x). Since A-(x) is also invariant, there
is a compact minimal set M, M CA- (x). If nx is not recurrent, then
M =f: y(x), and in particular x~ M. Let e(x, M) = tx. > 0. We will show,
that every motion ny, y E A-(x}, is positively stable in y (x). To this end,
given e > 0, there is a <5 == <5 (e) > 0 (by uniform positive stability in
y (x) of motions in r (x)), such that {xn, xm} c y (x), (! (xn, xm) < <5, implies
e(xnt, xmt) < e fort> 0. Now for yEA-(x), there is a sequence {tn} in
R-, tn ~ - ex>, such that xtn ~ y. There is then an integer N such that
n > N, m > N imply e (xtn, xtm) < <5 and(! (xtn, y) < <5, and consequently e (xtn (t), xtm (t)) < e for t > 0. Keeping, in this last inequality, tE R+
and n > N arbitrary but fixed and lettingm ~ex> we get e(yt, xtn (t)) < e
for t > 0 whenever e (y, xtn) < <5. Choosing now e = tx./2 and t = -tn,
we see that e (y (-tn), x) < rx/2. Since y (-tn) EM, this contradicts the
assumption that e(x, M) = lX. The theorem is proved.
The remaining portion of this section will be devoted to finding
conditions under which a limit set A+ (x) is compact and minimal, and
further when such a set consists of almost periodic motions only. For
this, the following definition is useful.
6.13
Definition. A semi-trajectory y+ (x) is said to uniformly approximate its limit set A+(x), if given any e > 0, there is a T = T(e) > 0
such that A+(x) C S(x[t, t + T], e) for each tE R+.
6.14
Theorem. Let the motion nx be positively Lagrange stable.
Then the limit set A+(x) is minimal if and only if the semi-trajectory
y+(x) uniformly approximates A+(x).
110
V. Stability Theory
Proof. The set A+(x) is non-empty, compact and invariant. Now let
y+(x) uniformly approximate A+(x). If A+(x} is not minimal, then there
are points y, z EA+ (x) such that z ~ y (y) (otherwise y (y) =A+ (x)
for each yEA+(x), and A+(x) is minimal; Theorem III, 3.4). Let
e (z, y (y)) = e > 0. By uniform approximation there is a T > 0 such
that S (x [t, t + T], e/2} ) A+ (x) fort > 0. Further, there is a ~ > 0 such
that e(y,w)<~ implies e(yt,wt)<e/2 for ftj<T. Since yEA+(x),
there is a point x1 E y+ (x) such that e(x11 y) < ~- And because S (x [t,
t + T], e/2)) A+(x), we have in particular S(x1 [O, T], e/2)) A+(x).
Thus there is a point x2 E x1 [O, T] such that e(z, x2) < e/ 2. If then x2 = x1r,
where 0 < T < T, then e (x1 T, y-r) < e/2, so that e (z, y-r) < e (z, x2)
+ e(x2 , y-r) < e/2 + e/2 = e, as x2 = x1-r. This is a contradiction, as
e (z, y (y)) == c. Thus A+(x) is minimal. Now let A+(x) be minimal, so
that for any y EA+ (x), y (y) =A+ (x). Now assume, if possible, that
y+(x) does not uniformly approximate A+(x). Then there is an e > 0, a
sequence of intervals {(tn, Tn)}, and a sequence {Yn} C A+(x) such that
tn ~ +ex:>, (-rn - tn) ~ +ex:>, Yn ~ y (EA+ (x)), and Yn ~ s (x [tn, Tn], e).
We may also assume that e(Yn, y) < e/3 for all n. Then for arbitrary n
e(y, x [tn, Tn]) >
(!
(Yn' x [tn, Tn]) -
e(Yn' y) > 8 -
e/3 == 2e/3.
Consider now the sequence of points {xn}, where
t:
Xn = X('n ~ Tn) = xt~.
Clearly ~ + ex:>. Since y+ (x) is compact, we may assume that xt: ~

z (EA+ (x) ). Since A+ (x} is minimal, y (z) = A+ (x), so that there is a
TE R such that e(zT, y) < e/3. By the continuity axiom we can choose
a <1 > 0 such that e(zT, wT) < e/3 whenever e(z, w) < <J. Now choose
N large enough such that e(z, xN) < <1, and (-rN - tN)/2 > !Tj. Then
xNT
== x (t'tv + T) E x [tn, Tn],
and hence e(y, xNT) > 2e/3. On the other hand e(xNT, zT) < e/3, so
that
e (y, xNT) < e (xNT, zT) + e (zT, y) < e/3 + e/3 = 2e/3.
This contradiction proves the result.
The following theorem gives a sufficient condition for a positive limit
set A+ (x) to be a minimal set of almost periodic motions. No necessary
and sufficient condition is known as yet.
6.15
Theorem. Let the motion nx be positively Lagrange stable, and
let the motions in y+(x) be uniformly positively stable in y+(x). If moreover y+(x) uniformly approximates A+(x), then A+(x) is a minimal set
of almost periodic motions.
111
Proof. By Theorem:.6.14, A+(x) is a compact minimal set. In view

of Theorem 6.11 we need only prove that every motion through A+ (x)
is positively stable in y+ (x).
By uniform positive stability of motions in y+(x}, we have given
8 > 0, there is a c5 > 0 such that e (xtl, xt2) < c5 implies e (x (t1 + t)
x(t2 + t)) < e/3 for t > 0. Let {y, z} (A+(x) and e(y, z) < c5/3. Let
-r > 0 be arbitrary. We wish to estimate e(y-r, z-r). By the continuity
axiom there is a <1 > 0 such that e (y, W} < <1, e (z, u) < <1 imply e (y't', WT}
< e/3 and e(z-r, u-r) < e/3. If C=min [<1, c5/3]. There are t1 > 0 and
t2 > 0 such that e (xt11 y) < Cand e (xt2, y) < C. Thus
e(xtl, xt2) < e(xtl, y) + e(y, z) + e(z, xt2) < + c5/3 + < c5.
Consequently e(x(t1 + -r), x(t2 + -r)) < e/3, and e (y-r, x(t1 + -r)) < e/3
and also e(z-r, x (t2 + -r)) < e/3. The last three inequalities yield
(! (y-r, ZT) < (!(YT, X(t1 + -r)) + (!(X (t1 + 1}, X(t2 + -r)) + (! (x (t2, T), ZT)
I
e/3 + e/3 + e/3 = e.

This shows in fact that the motions through A+ (x) are uniformly positively
stable in A+(x). The theorem is proved.
We now give a simple example of an almost periodic motion which is
neither a rest point nor a periodic motion.
<
6.16
Example. Consider a dynamical system defined on a torus by
differential equations (see Example III, 2. 7).
dq;
dt =
d()
l,
dt = lX,
where ex is irrational. For any point P on the torus y (P) = the torus,
and since <X is irrational, no trajectory is periodic. The torus thus is a
minimal set of recurrent motions. To see that the motions are almost
periodic, we note first that if P 1 = (q;1 , 01 ), P 2 = (q;2 , 02), then e(P1 , P 2)
= (<p1 - q;2}2 + (01 - 02) 2 , where the values of <p1 - <p2 and 01 - 02
are taken as the smallest in absolute value of the differences (mod 1).
Now any motion on the torus is given by <p = <p0 + t, 0 = 00 + ext. Then
for the motions through P 1 and P 2 we have
e(Pit, P2t) = Jl (<pl -
<fJ2) 2 + (01 - 02) 2 = e(Pv P2).

Thus the motions are uniformly stable in both directions in the torus. Thus
by Theorem 6.10 the torus is a minimal set of almost periodic motions.
A definition of stability (of motion) was given in a precise form by A. M. LIA[1] (see also Notes and References of Chapter VIII). The concept of stability
of equilibrium was previously discussed by LAGRANGE and by DIRICHLET. LAGRANGE
gave also a criteria for stability of equilibrium points based upon the consideration
of the potential energy: If the potential energy has a minimum value at a given rest
PUNOV
112
V. Stability Theory
point, then that rest point is stable. It is interesting to notice that the energy is a
particular case of a Liapunov function.
The first complete development of stability theory of dynamical systems is due
to v. I. ZUBOV [1] (see also R. BASS [2] and s. LEFSCHETZ [2]). In a sense, in the
theory of stability of dynamical systems one has achieved a synthesis between the
point of view of LIAPUNOV and that of PoINCARE, in fact LIAPUNOV was interested
in local (stability) properties of n-dimensional systems and PoINCARE in global
properties of planar systems. In the theory of stability, the concept of "region of
asymptotic stability" and of "global asymptotic stability'' is of rather recent origin
and it was developed in the early fifties in the Soviet Union by M. AIZERMANN,
E. BARBASHIN and N. N. KRASOVSKII. Most of the machinery needed . for these
developments is already present in the book by NEMYTSKII and STEPANOV [1].
Results in more general framework than the one presented here can be found in the
monographs by G. P. SzEGO and G. TRECCANI (for the case of flows without uniqueness) by N. P. BHATIA and 0. HAJEK (for the case of local semiflows) and in the
papers by D. BusHAW [2], J. CLAY [2], T. URA and I. KIMURA [2], and E. RoxIN
[3, 5, 6].
Section 1. Notice that the discovery of "unstable attractors" for dynamical systems
is rather recent and it is due to P. MENDELSON [2]. On the other hand in the case of
non-autonomous differential equations unstable attractors were known and were
called quasi-asymptotically stable sets (see H. ANTOSIEWICZ [1]). Weak attractors
were introduced by N. P. BHATIA [2]. The concept of uniform attractors is used by
S. LEFSCHETZ [2] and extensively used in the paper by N. P. BHATIA, A. LAZER and
G. P. SzEGO where the characterization presented in Exercise 7.29.2 is used as a
definition. The same definition is also used in the monograph by N. P. BHATIA and
G. P. SzEGO [1] where the characterization 1.5.3 is derived. The complete diagram
1.17 is due to N. P. BHATIA [7]. Theorem 1.22 is essentially due to DEsBROW [1]
who proved it for a connected, locally compact, locally connected, metrizable space.
The weak attractor theorem (1.25) is due to N. P. BHATIA and it was previously
proved for the case of an attractor by J. AUSLANDER, N. P. BHATIA and P. SEIBERT
[1].
Section 2. This section contains results of BHATIA [5]. Some remarks are in order.
Earlier results in this direction, for example those of ZuBov [1], AUSLANDER and
SEIBERT [2] and BHATIA [1], used essentially the same methods as used for the welldeveloped theory in the case of ordinary differential equations. For results on ordinary
differential equations see, for example, A. M. LIAPUNOV, I. G. MALKIN, E. A~ BARBASHIN, N. N. KRASOVSKII, J. KURZWEIL, I. VRKOC, K. P. PERSIDSKII, s. K. PERSIDSKII, V. I. ZuBov, J. MAsSERA, H. ANTos1Ew1cz, T. YosHIZAWA, W. HAHN. The
basic feature of the results in this section is that Liapunov functions are shown to
exist on the whole region of attraction instead of on a sufficiently small neighborhood
in earlier results. The functions, in general, have sufficient properties to allow the
derivation of theorems on global asymptotic stability and ultimate boundedness
as corollaries. Indeed J. AUSLANDER and P. SEIBERT [2] established formally the
long suspected duality between stability and boundedness in locally compact
separable metric spaces.
Theorem 2.16 is essentially due to V. I. ZuBov [1]. The proof given here is
simpler than the original because here we deal only with the case of compact sets
while V. I. ZuBov considers the more general case of closed sets.
Theorems 2.2, 2.9, 2.10, 2.12, 2.15 are due to N. P. BHATIA [5].
Section 3. The results presented in this section are due to N. P. BHATIA, A. LASER
and G. P. SzEGO [1] and to N. P. BHATIA and G. P. SzEGO [2]. For the case of the
flow defined by differential equations similar results are proved by F. WILSON [2].
113
Most of the results given here could have also been proved by means of the asymptotic fixed point theorem, due to G. S. JONES [3]. For the case of ordinary differential equations on the plane some results in the same spirit (identification of the canonical region, i.e., of the regions in which the stability properties are invariant) are
given by L. MARKUS [5].
The important Corollary 3.12 can also be proved directly as follows.
Direct Proof of Corollary 3.12. By Theorem 1.25, D+ (M) is compact and globally
asymptotically stable. Let x 0 E En be arbitrary but fixed. Choose ~ > 0 sufficiently
large so that D+(M) C S(x0 , ~). Choose further e > 0 sufficiently small such that
S(D+(M), e) C S[x0 , ~]. By Theorem 1.16, D+(M) is uniformly attracting. Hence
(1.2.3) there is a T > 0 such that x tE S(D+(M), e) C S[x0 , a], whenever
xE S [x0 , a] and t 2 T. For each 't' 2 T define the map n-r:: X ~ X by n-r:(x) = n(x, 't').
Then n-r: is continuous and n-r(S[x0 , a]) C S[x0 , a]. Thus by the Brouwer fixed point
theorem S [x 0 , a] contains a fixed point of the map n-r:, i.e., there is an xE S [x 0 , a]
such that n-r:(x} = x = n(x, 't'). Hence n(x, t) = n(n(x, 't'), t) = n(x, t
't') for all
tE R, and so y(x) is a periodic trajectory. Notice that xE M, for otherwise, if x EE M,
then y (x) (\ M = 0, for M is invariant. On the other hand, since y (x) is a periodic
trajectory, we havey(x)
A+(x}, and as xE Aw(M), we must haveA+(x) (\ M =I= 0,
i.e., y(x) (\ M =I= 0. This contradiction proves that y(x) CM. Since y(x) = y(x),
we must have y (x)
M, as M is minimal. Thus M is a periodic trajectory with
period 't'. If now M is not a rest point, then it will have a least period 't'0 , and all
other periods must be the numbers m't'0 , where mis an integer. However, we have in
fact shown that all numbers 't' > Tare periods of M. This is a contradiction and so
M is a rest point, and the theorem is proved.
Section 4. Theorems for the characterization of the stability properties of closed

sets are given by N. P. BHATIA whose presentation we essentially follow. Other
characterizations were given by V. I. ZuBov [1], T. YOSHIZAWA, while sufficient
conditions were given by G. P. SzEGO and G. GEISS [1].
The notion of equistability used here was introduced in the monograph by
N. P. BHATIA and G. P. SzEGO together with Theorem 4.19. This theorem seems to
lead the way for the use of the theory of parallelizable flows in the study of problems
of asymptotic stability of closed sets and of the existence of Liapunov functions.
The exposition in this section of this topic is more general than the one of H. ANT0srnw1cz and J. DuGUNDJI.
Theorem 4.24 is an improvement of an analogous theorem due to V. I. ZuBov [1].
Section 5. This section contains some new results (5.5-5.10), but the remaining
parts are adopted from T.URA [5]. For the case of differential equations some
different results on relative stability properties were derived by U. D' AMBROSIO
and v. LAKSHMIKANTHAM [1], A. A. KAYANDE and v. LAKSHMIKANTHAM [1] and
by V. LAKSHMIKANTHAM [4].
Section 6. This section brings to a completion the classification of compact minimal
sets, viz., a rest point, a periodic trajectory, the closure of an almost periodic trajectory, and the closure of a recurrent trajectory. The relationship between almost
periodicity and stability is clarified.
The notion of an almost periodic function is due to H. BoHR and Theorem 6.7.
is due to S. BocHNER.
A. A. MARKOV [3] showed the relationship between stability of motion and almost
periodicity (Theorems 6.11, 6.12).
Definition 6.13 and the following material is due to V. NEMYTSKII. In this
connection one may also see the paper of DEYSACH and SELL [1] on the existence
of almost periodic motions.
Bhatia/Szego Stability Theory
Chapter VI
Flow near a Compact Invariant Set

This chapter is devoted to a rather deep analysis of the flow in the
vicinity of a compact invariant set of a locally compact phase space.
The main result was motivated by a desire to find if every neighborhood
of a rest point contained a semi-trajectory different than the one through
the rest point (see II, 2.6).
1. Description of Flow near a Compact Invariant Set

Our principal result is the following
1.1
Theorem. Let the phase space X be compact and M ( X a
compact invariant subset of X. Then one of the four conditions given
below holds.
1.1.1
Mis positively asymptotically stable,
1.1.2
M is negatively asymptotically stable,
1.1.3
there exist
A-(y) CM,
x~
M, y~ M such that 0 =f= A+(x) CM, and 0 =f=
1.1.4
every neighborhood U of M contains an
(Note that since M is invariant, y (x) (\ M = 0.)
x~
M with y (x) ( U.
Proof. Since X is compact, A+ (x) =f= 0, A-(x) =f= 0 for every x E X.

If Mis an open subset of X then 1.1.1 and 1.1.2 hold trivially. Let Ube
an arbitrary open neighborhood of M. Then for any x E U - M one of
the following conditions must hold.
1.1.5
A+(x) CM,
1.1.6
A+(x)
1.1. 7
A+ (x) f\
CU but 1.1.5 does not hold,
e (U) =f= 0.
1. Description of Flow near a Compact Invariant Set
115
Note now that if 1.1.6 holds, then U contains the invariant set A+(x) - M
which is non-empty. A similar analysis will show that for every x E U
one of the following conditions holds.
1.1.8
A-(x) ( M,
1.1.9
A-(x) ( U but 1.1.8 does not hold,
1.1.10
A-(x) f\
@(U)
=t= 0,
and if 1.1.9 holds then U contains the invariant set A-(x) - M which is
non-empty. To see that at least one of the conditions 1.1.1-1.1.4
must hold for M, we assume that 1.1.4 and 1.1.3 do not hold. Then we
claim that there exists an open neighborhood U of M such that @(U)
is either a weak attractor or a negative weak attractor with the region
of weak attraction (positive or negative) coinciding with @(M). To see
this note that if 1.1.4 does not hold then there is a neighborhood U of
M such that for no x E U - M, 1.1.6 or 1.1.9 holds. If now for some
x E U - M, the condition 1.1.5 holds, then 1.1.8 cannot hold for any
y~ M for then 1.1.3 will be fulfilled. Then 1.1.10 holds for every xE U
- M and consequently for every x~ M, and @(U) is a weak negative
attractor. Similarly, if 1.1.8 holds for some x E U - M, then @(U) will
be a weak attractor. To complete the proof, it is now sufficient to show
that if (Q (U) is a weak attractor with Aw (M) = (Q (M), then Mis negatively asymptotically stable, and if@ (U) is a negative weak attractor with
its region of negative weak attraction coinciding with (Q (M), then M is
asymptotically stable. Because of the dual nature of the concepts we
need only prove one of the foregoing assertions.
So let @(U) be a weak attractor with Aw(@(U)) = @(M). Then by
the weak attractor Theorem V, 1.25, D+(@ (U)) is compact and asymptotically stable with Aw(D+(@ (U))) =A (@(U)) = (Q (M). Since D+(@ (U))
C @(M), the open set @(D+(U)) == Vis an open neighborhood of M. We
claim that for each xE V, J-(x) ( M. This follows from the fact if for
some xE V, J-(x) ct M, then there is a y~ M, with yE J-(x). But then
x E J+(y) by Theorem II, 4.9. Consequently, there is ayE A( (Q (U)) = (Q (M)
with J+ (y) ct D+ (@ (U) ). Since the latter set is uniformly attracting by
Theorem V, 1.16, we have arrived at a contradiction. Hence M is a
negative uniform attractor, and since it is invariant, it is negatively
asymptotically stable. This completes the proof.
The above theorem, though proved for compact spaces X, carrie~
over to locally compact phase spaces. To see this one may use the result
in Exercise II, 4.8.4. Then the result holds for the dynamical system
on x. But since =Tl on X, the result holds in x. Thus we have
8*
116
1.2
VI. Flow near a Compact Invariant Set
Corollary. Theorem 1.1 holds for locally compact spaces X.
1.3
Exercises. Let X be locally compact and M a compact invariant
subset of X.
1.3.1
If there is a relatively compact neighborhood U of AI such that
for each x E M either A+ (x) f\ @(U) =f= 0, or there is a T > 0 with
y+ (xT) ( @. ( U), then M is negatively asymptotically stable. Show that
the converse holds also.
If M is asymptotically stable, and A (M) is relatively compact,
1.3.2
then @(A (M)) is negatively asymptoticaUy stable.
If M is asymptotically stable, then for each x($ 111, J-(x) (
1.3.3
@(A (M)) holds.
1.3.4
Let M be stable. Then for every
A+(x) CM, or A+(x) f\ M = 0.
x~
M, either A+(x) =f= 0 and
1.3.5
If Mis stable but not asymptotically stable, then every neighborhood U of M contains a compact invariant set N disjoint from M.
1. 3. 6
If M is a weak attractor, then A- (x) f\ M =f= 0 for every
xE D+(M).
1.3. 7
Give an example of a weak attractor for which condition 1.1.3
of Theorem 1.1 holds, but not the condition 1.1.4.
1.3.8
Prove that if xis a rest point, then every neighborhood U of x
contains a y =f= x with either y+ (y) or y-(y) contained in U.
1.3.9
Show that the results in this section hold for closed invariant
sets M with compact boundary oM.
2. Flow near a Compact Invariant Set (Continued)

In Theorem 1.1, the condition 1.1.3 may be fulfilled in several ways.
For example one may have points x~ M, yft M, such that 0 =f= J+ (x) ( M,
0 =f= J-(y) CM. Similarly, condition 1.1.4 of Theorem 1.1 may be fulfilled
in several ways. For example every neighborhood U of M contains a
non-empty compact invariant set disjoint from M. It is therefore desirable to try to develop more detailed descriptions of the flow near a compact invariant set than the one contained in Theorem 1.1. We shall
indicate one such generalization of Theorem 1.1 here.
2.1
Theorem. Let the phase space X be compact and M ( X a
compact invariant subset of X. Then one of the following holds. (Here
At (M) will denote the set Au (M) defined in V, 1.1.3, and A; (M) the
corresponding negative concept.)
117
2.1.1
Aj- (M) V A; (M) V M is a neighborhood of M. (Thus M is
either positively or negatively asymptotically stable, or there is a neighborhood U of M such that either 0 =F J+ (x) C Mor 0 =F J- (x) C M holds
for each x E U - M.)
2.1.2
There exist x,
0
yft M
such that
=F A+(x) ( M,
=F A-(y)
( IYI.
2.1.3
Every neighborhood U of M contains an x ft M such that
y(x) CU and
A+(x) (\ M =F 0, A-(x) (\ M =F 0,
but J+(x) V J-(x)
ct M.
2.1.4
Every neighborhood U of M contains an
y(x) ( U - M.
x~
M such that
Proof. Suppose that 2.1.1 and 2.1.2 do not happen. Then by Theorem 1.1 condition 1.1.4 of Theorem 1.1 holds. Thus every neighborhood
U of M contains an x with y (x) C U-M. Notice first that in fact we
can choose x such that y (x) C U, because for any neighborhood U of M
there is a compact neighborhood V of M with V C U and, by the above
requirement, there is an x E V with y (x) ( V - M. Now compactness of
--v ensures y (x) C V C U. Note also that in this case y (x) is compact and
therefore both A+ (x) and A-(x) are non-empty compact sets. If now
A+(x) (\ M = 0 or A-(x) (\ M = 0, then 2.1.4 holds by choosing a
y EA+ (x) in the first, or a y E A-(x) in the second case (thus ensuring
y (y) (A+ (x) C U - M for example in the first case). If therefore 2.1.4
does not hold we must have an x E U - M, with A+ (x) (\ M =F 0,
A-(y) (\ Af =f= 0 and y (~) ( U - M. However, since 2.1.2 does not hold
we cannot have bothA+(x) ( M andA-(x) ( M, and therefore 2.1.3holds.
The problem of the qualitative behavior of the flow defined in a neighborhood
of an isolated rest point of a planar. differential equation of analytic type was one
of the first dealt by H. Po1NCARE [3]. In this situation he proves that, if the differential equation has the form =Ax+ g(x}, xE R 2 , g(O) = 0 and it is such that the
power series expansion of g (x) in the neighborhood x = 0 begins with terms of
degree at least two, then, if the linear system =Ax has eigenvalues with nonzero real parts, the qualitative behavior of the nonlinear system and that of the
linear system in a sufficiently small neighborhood of the rest point x = 0 coincide.
A classification of the possible behaviors of linear systems was also developed (see
A. ANDRONOV and C. CHAIKIN [1], N. M1NORSKY [1]). Classifications for more general planar cases are given by S, LEFSCHETZ [1] and S. BAROCIO [1, 2].
118
VI. Flow near a Compact Invariant Set
The same problem was studied in the case of then-th order ordinary differential
equation by D. GROBMAN [4], A. I. PEROV [1] and by P. HARTMAN [4]. A complete
presentation of this problem can be found in the book by HARTMAN [1, chapter 9].
Recently V. V. NEMYTSKII [12] proposed a new classification for the case of a
differential equation on the plane. This classification was also used by N. P. PAPUSH. Recently M. B. KuDAEV [2] extended the Nemytskii classification to higher
order singular points and proposed a method [3], based upon the consideration of
the higher derivatives of a Liapunov function (see also Chapter VIII) for classifying
the flow.
The qualitative behavior of differential equations on manifolds or near manifolds
was in addition investigated by F. HAAS [1-4], A. DENJOY [1-5], F. B. FULLER
[1], T. SAITO [1, 2], C. L. SIEGEL [1], S. STERNBERG [1], W. KAPLAN [6], J. K.
HALE and A. STOKES and by A. SCHWARTZ [1, 2].
For the case of the second order differential equation the global problem of the
classification of the flow was studied by I. BENDixsoN. He posed and gave a solution
to the problem of the possible qualitative behaviors of the solutions in a neighborhood
of a rest point. For the case of a dynamical system defined in a locally compact
space T. URA and I. KIMURA [1] were able to propose a description. A similar description (Theorem 1.1) was with an independent proof, proposed by N. P. BHATIA [7].
The more advanced result (Theorem 2.1} was proved by N. P. BHATIA [9] in the
more general framework of dynamical system in a locally compact Hausdorff
space. Other classification theorems for dynamical systems are due to T. SAITO
[3, 4]. Classification theorems were also proved for the case oflocal semi-dynamical
systems by N. P. BHATIA [8] and by N. P. BHATIA and 0. HAJEK [1] and for the
case of flows without uniqueness by G. P. SzEGO and G. TRECCANI [1, 2].
For further research in the direction of the results of this chapter, one may find
the work of K. S. SIBIRSKII [1, 2] to be useful and interesting. SIBIRSKII introduces
the notion of VJ - ({1-) limit point, which essentially guarantees that the limit set
of a point x has precisely one minimal subset. G.D. BIRKHOFF [1, p. 204] used this
last idea to define semi-asymptotic central motions. The notion of ambits in Topological Dynamics as introduced by W. A. GOTTSCHALK [12] is also based on a similar idea.
It should be noted that in Theorem 1.1, if Mis compact but not open, then the
cases 1.1.1 and 1.1.2 cannot occur simultaneously. In fact, if 1.1.1 occurs, then
none of 1.1.2-1.1.4 occurs. On the other hand 1.1.3 and 1.1.4 may occur simultaneously. This is so, because the cases 1.1.1 and 1.1.2 characterize the flow through
all points of a neighborhood of M, whereas the cases 1.1.3 and 1.1.4 require the
existence of some points with a given behavior in each neighborhood. It will be
useful to develop description of flows containing as many cases of the nature of
1.1.1 and 1.1.2 as possible and reduce or eliminate the number of cases of the type
1.1.3 or 1.1.4. In Theorem 2.1 we have a description containing three cases of the
former (all grouped under 2.1.1) and three cases of the latter type.
The problem of the classification of the flow via Liapunov-type functions will
again be discussed in Chapters VIII and IX.
Chapter VII
Higher Prolongations
The first positive prolongation and the first positive prolongational
limit set have been shown to be useful in characterizing various concepts
in dynamical systems. Notable applications are the characterization of
stability of a compact set in a locally compact metric space (Theorem V,
1.12) and the characterization of a dispersive flow (Theorem IV, 1.8).
The first positive prolongation may be thought of as an extension of
the positive semi-trajectory. For example consider a dynamical system
(R2, R, n) which is geometrically described by Fig. 0.1.
}(_~---f's
~___,.._Y+lx)IYi
--
a-~---~----_-_-_-_- ------=-~
Fig.0.1
The first positive prolongation of the point x in the figure consists of the
semi-positive trajectory y+(x), the equilibrium points 0, P, and Q, and
the trajectories y1 , y 2 , y 3 , and y4 In a way, to get the prolongation of a
point we might find ourselves arguing that we move along the positive
semi-trajectory and approach the equilibrium point 0. So we transfer
120
VIL Higher Prolongations
to the point 0. From 0 we transfer to a trajectory which leaves 0, e.g.,

we can transfer to y 1 or y3 If we transfer to y 1 , then we approach the
equilibrium point P. So we transfer to P, and thence to a trajectory
leaving P, and so on. If indeed this procedure were laid down to define
the prolongation of a point, then notice that we would have to include
the trajectories y5 and y 6 in the prolongation of x. The definition of a
prolongation, however, excludes the trajectories y 5 and y 6 from the prolongation of x. If, however, we wished to include these in a prolongation,
then either we must change the definition of prolongation, orin a sense introduce other prolongations which will do precisely what we did with the
intuitive reasoning above. Just as the first positive prolongation is in
fact a meaningful extension of the positive trajectory, the second and
higher prolongations which will be presently introduced, will be shown to
be meaningful extensions of the first prolongation.
1. Definition of Higher Prolongations

The description of higher prolongations is facilitated by the introduction of two operations J. and fJ) on the class of maps from X into 2x.
The Operators J. and fJJ. If I': X----? 2x, we define fJJI'by
1.1
1.1.1
fJJI'(x)=f\{I'(U): UE%(x)},
where % (x) denotes the set of all neighborhoods of x. Further, J.I' is

defined by
1.1.2
J.I'(x)
where I'1(x)
= 1, 2, ... } ,
and P(x) = I'(I'(n-i) (x)), n
V {I'n (x): n
I'(x),
2, 3, ....
In the sequel the following lemma will prove useful.

1.2
Lemma. For any x E X, and I': X----? 2x, fJJI'(x) is the set of all
points y EX such that there are sequences {xn}, {Yn} in X, Yn E I'(xn),
and xn----? x, Yn----? y. Further, J.I'(x) is the set of all points y EX such
that there are points x1 , x 2 , , xk, with the property that xi+i E I'(xi),
i = 1, 2, ... , k - 1, and x1 = x, x k = y.
The proof is immediate and is left to the reader.
The following lemma gives some elementary properties of the operators fJ) and J..
1.3
Lemma.
1.3.1
fJ) 2 = fJ),
1.3.2
If M
is closed.
and J.2 = J.. Thus fJ) and J are idempotent operators.
C X is compact, then
fJJI'(M) = V {fJJI'(x):
x EM}
121
1.3.3
If v = <p (x) is a continuous real-valued function on X, such that
<p (y) < <p (x), whenever y E I' (x), then <p (y) < <p (x) whenever y E rfJI' (x)
V J.I'(x).
Proof of 1.3.1. Let I': X--?- 2x. If y E rfJI'(x), then indeed y E rfJrfJI'(x).
Thus rfJI'(x) C rfJrfJI'(x). If y E rfJrfJI'(x), then there are sequences {xn},
{Yn}, YnE rfJI'(xn), such that Xn-?-X andyn-?-Y. Foreachxk, yk, YkErfJI'(xk),
there are sequences {xk}, {Yk}, Yi! E I'(xk), such that xi!~ xk, y~--?- Yk
\Ve may assume e(xk, x;:) and e(yk, y~) < 1/k for n > k and each k.
Now consider the sequences {x:}, {Y:}. Clearly y: EI' (x:). Further x:-?- x,
and y:-?-y. Thus y E rfJI'(x). We have thus proved that rfJrfJI'(x) ( rfJI'(x).
This together with the previous observation shows that rfJrfJI' (x) == rfJI' (x).
Hence 0 2 = rfJ. The proof of J.2 = J. is even simpler and is left as an
exercise.
Proof of 1.3.2. Let {Yn} be a sequence in rfJI'(M) such that Yn -?-y.
Then there is a sequence {xn} in M such that Yn E rfJI'(xn) Since M is
compact, we may assume that xn--?- x E M. Thus y E rfJrfJI'(x) == rfJI'(x)
( rfJI'(M). This shows that rfJI'(M) is closed.
Proof of 1.3.3. If y E rfJI' (x), then there are sequences {xn}, {yn},
Yn E I'(xn), such that xn--?- x, and Yn--?- y. It is given that <p (Yn) < <p (xn).
Since <p is continuous we get by proceeding to the limit <p (y) < <p (x). If
y E J.I'(x), then there are points x == x1 , x2 , , xn == y such that xi+iE
I'(xi), i = 1, 2, ... , n - 1. Hence <p (y) = <p (xn) < <p (xn_ 1 ) < < <p (x2)
< <p (x1) = <p (x). This completes the proof of the lemma.
2x will be called transitive if J.I' = I'.
1.4
Definition. A map I': X
1.5
Exercise.
1.5.1
A map I': X--?- 2x, is transitive if and only if I'2 ==I'.
1.5.2
Given I': X--?- 2x, J.I' is transitive.
--?-
1.6
Definition. A map I': X--?- 2x will be called a cluster map if
rfJI' =I'.
1. 7
Definition. A map I': X --?- 2x will be called a c - c map if it
has the following property: For any compact set KC X and x EK, one
has either I'(x) ( K, or I'(x) f\ oK =t= 0.
1.8
Theorem. Let the space X be locally compact. Let I' be a c - c
map. Then if I'(x) is compact, it is connected.
Proof. If I'(x) is compact, but not connected, then we can write

I'(x) = M 1 V M 2 , where M1' M 2 are non-empty compact disjoint sets.
122
Since Xis locally compact, we can choose compact neighborhoods U1, U 2

of M 1 , M 2 respectively such that U1 f\. U2 === 0. Note that I'(x) f\. U1 =F 0,
but U1 ):) I'(x), as M 2 f\. U1 = 0. However, oU1 f\.I'(x)=0, contradicting the fact that I'(x) is a c - c map. Thus I'(x) is connected.
1.9
Theorem. Let the space X be locally compact. Let I' be a
c - c map. If M (Xis compact, then rfJI'(M) = M if and only if for
each neighborhood U of M, there is a neighborhood W of M such that
I'(W) ( U.
Proof. (i) Sufficiency. Note that for any c - c map I', x E I'(x). Hence
rfJI'(M) ) M always. Consider now a sequence of closed neighborhoods
{Un} of M, such that f\. Un= M. Then there is a sequence of neighborhoods {Wn} such that I'(Wn) (Un. Then rfJI'(M) ( f\. I'(Wn) ( f\. Un
=== M. Thus we have proved that rfJI'(M) = M, and sufficiency is proved.
(ii) Necessity. Indeed assume, if possible, that there is a neighborhood
U of M, such that for every neighborhood W of M, I'(W) ct U. We may
assume without loss of generality that U is compact (because Xis locally
compact). Then there is a sequence {xn}, xn---* x EM, and a sequence
{Yn}, Yn E I'(xn) such that Yn ft U, n = 1, 2, .... Indeed we may assume
that xn EU, n === 1, 2, .... But then since I'is a c - c map andI'(xn) ct U,
we must have I'(xn) f\. au =t= 0. Consequently, there is a sequence {zn},
Zn E I'(xn) (\ au, n = 1, 2, .... Since au is compact, we may assume that
Zn --7- zE au. But then zE rfJI'(x) c rfJI'(M). Butz~ M, a contradiction,
as rfJI'(M) = M. This proves necessity and the theorem is proved.
We remark now that c - c maps have also the following properties,
which help to build families of c - c maps.
1.10
Lemma.
1.10.1 Let {I'a}, ~ E A, be a family of c - c maps. Then I'= VI'' is a

c - c map.
1.10.2
If I'1 , I'2 are c - c maps, then so is the composition map
I'= I'1 I'2
1.10.3 If I' is a c - c map, then so are J.I' and rfJI'.
Proof of 1.10.1. Let K be a compact set, and x EK. If I'(x) ct K, we

must prove that I'(x) f\. aK =t= 0. Indeed if I'(x) ct K, then there is at
least one map, say I'cx, such that I'cx (x) ct K. Since x E K, and I'cx is a
c - c map, we must have I'cx (x) f\. aK =t= 0. Thus I'(x) f\. oK =t= 0, and
so I' is a c - c map.
Proof of 1.10.2. We need consider the case thatI'2 {x) C J(K), where
xE K, andK compact. This is so, becauseI'1 I'2 (x) = I'1 (I'2 (x))) I'2 (x).
123
If I'1 o I'2 (x) ct K, and I'2 (x) C J (K), then there is a y E I'2 (x) such that
I'1 (y) ct K. But then I'1 (y) f\ oK =t= 0, and since I'1 (y) ( I'1 (I'2 (x)) =
I'1 o l~(x), we have I'1 o I'2 (x) f\ oK =t= 0. Thus I'= I'1 o I'2 is a c - c
map.
Proof of 1.10.3. If I' is a c - c map, then J.I' is indeed a c - c map

by the assertions 1.10.1 and 1.10.2. To show that 0I' is a c - c map,
let x EK, where K is compact. If 0I'(x) ct K, we may assume without
loss of generality that x E .J' (K). If 0I'(x) C[ K, then there is a sequence
{xn} in K, xn----? x, and a sequence {Yn}, Yn ft K, Yn----? Y ~ K, Yn E I'(xn)
Since I'(xn) ct K, andxnE K, there is a sequence {zn}, znE I'(xn), andzn E oK.
Since oK is compact, we may assume that Zn----? z E oK. But then z E
0I'(x), so that 0I'(x) f\ oK =f= 0. This proves that 0I' is a c - c map.
We now prove the following interesting theorem.
1.11
Theorem. Let X be locally compact and M a compact subset
of X. Let I' be a c - c map which is moreover a transitive map as well
as a cluster map. Then I'(M) = M if and only if there exists a fundamental system of compact neighborhoods {Un} of M such that I'( Un)= Un.
Proof. Let {Wn} be a fundamental system of neighborhoods of M.

Since (f)I'(M) = I'(M) = M, by Theorem 1.9, we have compact neighborhoods Kn of M such that I'(Kn) ( Wn. Now notice that I'(Kn) are
closed and may be considered as compact. Setting I'(Kn) = Un, we get a
fundamental system of compact neighborhoods of M such that I'(Un)
= Un. This is so because I'(Un) = I'(I'(Kn)) = I'(Kn) == Un, as I' is
transitive. Indeed Un C I'(Un) so that equality follows.
We shall now apply the theory constructed above to a dynamical
system (X, R, n).
1.12
Definition. The Higher Prolongations. Consider the map y+:
X----? 2x which defines the positive semi-trajectory through each point
x E X. Then since y+ (x) is connected, y+ is a c - c map. Further, notice
that y+(y+(x)) = y+(x), so that y+ is a transitive map, i.e., J.y+ = y+.
We now set (f)J.y+ == (f)y+ = Dt, and call Dt (x) as the first positive
prolongation of x. This is clearly the same as defined in Chapter II, where
it is denoted by D+. Indeed Dt is a cluster map as 0 is idempotent, but
Dt is not transitive as simple examples will show. We, therefore, consider
the map (f)J.Dt and denote it by Di and call Di (x) as the second prolongation of x. Having defi:ned n:, we define n:+i = (f)J.D;i, and call
D;i (x) to be the n-th prolongation of x. This defines a prolongation of x
for any positive integer n. Using transfinite induction, we define a prolongation Dt (x) of x for any ordinal number x as follows: If x is a succes-
124
sor ordinal, then having defined-D:_ 1 , we setD! = fJ)J.D:_ 1 If xis not

a successor ordinal, then having defined D{ for every {3 < x, we set
D! = fJ) V { J.Dt: f3 < x }. This defines for each x E X, a prolongation
Dt (x) for every ordinal x.
Notice that each of the map considered above is a c - c map. Moreover, the prolongations Dt (x) for any ordinal x are closed.
vVe give below some properties of these prolongations.
1.13
Theorem. Let Q be the first uncountable ordinal number. Then
1.13.1
Dt
1.13.2
D"J is a transitive map.
= V {Dt: lX
< Q},
Proof. Recall that DJ = rfJ V { J.Dt: x < Q}, as Q is not a successor

ordinal. Let for any x EX, y E D"J (x). Then there are sequences {xn},
{Yn}, xn---* x, Yn---* y, and Yn E I'a: n (xn), where I'a:n = J.Dtn and lXn is some
ordinal number, xn < Q. Let {3 be an ordinal number such that lXn < {3
< {3 + 1 < Q (such ordinals exist). Then indeed Yn E J.DjJ (xn) for each
n, so that y E fJ)J.D{i (x) = D{+i (x). Consequently y E I'(x) where I'
stands for V {Dt: x < Q}. Indeed for any x < Q, D! (x) ( D"J (x), so
that I'(x) ( DJ (x). This proves 1.13.1. To prove that Dt is transitive,
we need show only that DJ o D"!i = D"!i. Suppose that y E DJ (x), and
zE DJ (y), so that zE D"J o DJ (x). Then there is a {3 < Q such that
y E D{i (x), and zE DjJ (y). But then zE J.DjJ (x) C rfJJ.DjJ (x) = D{+i (x)
C DJ (x). Hence J.D"!i = DJ. The theorem is proved.
1.14
Corollary. If lX > Q, then Dt = Dt. This follows by induction
as DJ is transitive, and indeed also a cluster map.
2. Absolute Stability
We shall now define a host of stability concepts with the help of the
higher prolongations introduced above.
2.1
Definition. Let X be locally compact. Let M be a compact
subset of X. The set M will be called stable of order lX, or lX-stable, if
Dt (M) = M. If M is stable of order lX, for every ordinal number lX,
then Mis said to be absolutely stable.
2.2
Remark. Stability of order 1, is the same thing as stability
defined in V, 1.5, as is evident from Theorem V, 1.12. Note also that
absolute stability is the same as stability of order Q, where Q is the
first uncountable ordinal, as is clear from Theorem 1.13 and Corollary 1.14.
One may easily prove the following theorem.
125
Theorem. A compact set MC X, where Xis locally compact,

is ~-stable if and only if for every neighborhood U of M, there exists a
neighborhood W of M, such that D! (W) C U.
2.3
We now give a few simple examples to illustrate that the various

higher prolongations introduced above are indeed different concepts.
2.4
Example. Consider a dynamical system defined on the real line.
The points of the form 1 : n , n = 0, 1, 2, ... , are equilibrium points,

and so are the points -1, and + 1. Between any two successive (isolated)
equilibrium points p, q, such that p < q, there is a single trajectory which
has q as its positive limit point, and p as its negative limit point. There
is a single trajectory with -1 as its sole positive limit point, and it has
no negative limit points, and there is a single trajectory with + 1 as its
negative limit point, and it has no positive limit points (see Fig. 2.4.1).
For any point P such that P< -1, Dt(P) ={xER: P <x< -1},
Dt (P) = Dt (P), but Dt (P) = {xE R: P < x < +1}, and Dt (P) =
{x ER: P < x}. Note also, that if Pis the point -1, then Dt (P) == P,
Dt (P) = {xE R: -1<x<1}, Dt (P) = {x: x > -1}.
----..
-1
-~---~~---..,..._-._.,.__._...._
...____
+1
Fig. 2.4.1
2.5
Example. Consider again a flow on the real line, such that we
have the equilibrium points as in the above example, and, moreover,
between any two such successive equilibrium points, say q and p, there
are two sequences of equilibrium points, say {Pn}, and {qn}, ... qn <
qn- 1 < ql < P1 < P2 < , Pn~P, andqn~q. Then direction of motion
on a trajectory between any two equilibrium points is again from left to
right, as in the previous example. In this case, if we consider the point
P == -1, then indeed Dt (P) = P, Dt (P) === P, but Dt (P} === {xE R:
-1 < x < +1}, and Dt(P) = {xE R: -1 < x}.
Proceeding in this fashion it is easy to see that we can construct
examples on the real line in which a point is stable of order n (n integer),
but is not stable of order n + 1.
2.6
Example. We now give an example of a dynamical system defined
on the real line, in which an equilibrium point is stable of every integral
order n, but not stable of order ro, where w is the first uncountable ordinal.
To obtain such an example we consider a sequence of points {Pn},
Pn ~ 0, P 1 > P 2 > P 3 > > 0. To the right of P 1 , we introduce a
sequence {P~}, such that Pf > P~ > > P 1 , and P~ ~ P 1 Between
126
P 1 and P 2 , we first introduce a sequence {P2n}, P 1 > P 21 > P 22 >

P 23 > >P2 , P2n~P2 Then for each P 2k, we introduce a sequence
{.P;k} between P 2k and P 2(k-i)' such that P~k > P~k > > P 2 k, and
P2k ~ P 2k. Between P 2 and P 3 we first introduce a sequence of points as
between P 1 and P 2 , and then between any two successive points we
introduce a monotone decreasing sequence converging to the point on
the left. We now proceed inductively. Having introduced a suitable sequence between say Pn-l and Pn, we introduce a sequence betweenPn
andPn+i similar to the one introduced betweenPn-l andPn, then between
each pair of successive points of this sequence, we introduce a monotonic
decreasing sequence converging to the point on the left. Now we are
ready to introduce the dynamical system on the real line. Each point
of the countable set of points introduced on the line is an equilibrium
point. There are no other equilibrium points, and the motion between
any two successive equilibrium points is from left to right. It is easy to
see that each point {Pn} of the first sequence introduced above has the
following property. P 1 is not stable, P 2 is stable of order 1, but not
stable of order 2, P 3 is stable of order 2 (and hence also of order 1), but
is not stable of order 3, Pn+i is stable of order n but not stable of order
n + 1. The point 0 is stable of every integral order n, but is not stable of
order ro.
If we consider Example 2.4, then it is an easy matter to show that
no continuous scalar function satisfying conditions of Theorem V, 4.9
exists for the uniformly stable equilibrium point -1. An example in
the plane, e.g., Example V, 4.11 was previously used to establish the
same thing. In fact, even the point 0 in Example 2.6 which is stable of
every integral order n is such that no continuous function satisfying
conditions of Theorem V, 4.9 can exist for this point. The question
obviously arises, as to what are the implications of the existence of a
continuous function satisfying the conditions of Theorem V, 4.9 for a
given closed set M. The answer for a compact set Min locally compact
spaces X is given by the following theorem.
2.7
Theorem. Let X be locally compact, and let M ( X be compact.
Then the following are equivalent:

2. 7.1
There is a real-valued function <p (x) satisfying conditions of
Theorem V, 4.9 which is continuous in some neighborhood of M,
2. 7.2
M possesses a fundamental system of absolutely stable compact
neighborhoods,
2. 7. 3
M is absolutely stable.
We shall need the following lemma, whose proof is immediate from
the definitions.
127
2.8
Lemma. Let 9} is a fundamental system of neighborhoods of
M, where U' = {x EX: 0, and {U ':
0 < < m0 }, where U ' == {x E U: <p (x) < X }, is a fundamental system
of compact, positively invariant neighborhoods of M. We will now show
that each U' is absolutely stable. We shall show this by using 1.3.3.
To do this, we consider the function (/) (x) defined on X, by means of
<J> (x) = (x) = m 0 for x ~ U mo This is a continuous
function which is decreasing along the trajectories. For 0 < X < {3 < m0 ,
U f3 is indeed a compact neighborhood of U rx Since <J> (x) is decreasing
along the trajectories, we get for y E DJ (x), <J> (y) < <J> (x). If D"J (U') =f= U''
then there is a {3 > such that D"JU' C( Up. Since D"J is a c - c map,
there is an x E U', and a y E DJ (x) f\ oUfJ. On one hand, therefore,
<J> (y) < <J> (x) < 0<-, and, on the other hand, <J> (y) = {3 >
This contradiction shows that DJ (U') == U', i.e., each U' is absolutely stable.
'
'
'
2. 7.2 implies 2. 7.3. This is immediate.

2.7.3 implies 2.7.1. Using Theorem 1.11 (since DJ is a c - c map
which is moreover a transitive as well as a cluster map) we first construct
a fundamental system of absolutely stable neighborhoods U 112 n, n ==
0, 1, 2, ... , such that (U112n} ( J (U 112n-1).
We now extend this system of absolutely stable compact neighborhoods to one defined over the diadic rationals, i.e., numbers of the type
' == j/2n, n = 0, 1, 2, ... ; j = 1, 2, ... , 2n, in such a way that (a) the
compact neighborhood corresponding to any diadic rational is absolutely
stable, (b) if < {3 are diadic rationals, then U' ( J (U11), (c) M ==
f\ {U':
diadic rational}. Indeed this is possible by using Theorem 1.11.
Now if x E U 1 , define 0, then <p (xt) < <p (x). This is so,
because if x E U', then since U' is positively invariant, we have xt E U',
hence <p (xt) < <p (x). Finally, to see that <p (x) is continuous on U1 , we
assume that this is not true. Then there is an x E U'' and a sequence
{xn} in U1, xn ~ x, such that <p (xn) ~ ' =f= <p (x) = 'x If X < 'x' then we
can choose diadic rationals 0<-1 , 0<-2 , such that ' < 0<-1 < 0<-2 < 'x Then for
large xn, xn E U '1 , whereas x ~ U ' 2 Since U '1 is closed, and xn ~ x, x E U'1
This is a contradiction as u'l J (U'J If again lX > lXX, then choose
diadic rationals 'v 0<-2, such that ' > 0<-1 > X2 > 'x Then xn ~ U' for
large n, whereas x E U' 2 But U'a ( J (U'), which contradicts xn ~ x.
This completes the proof of the theorem.
'
'
128
2.9
Remark. Prolongations and Stability of Closed Sets. Although
Theorem V, 1.2 gives an excellent characterization of Liapunov stability
of compact sets in locally compact spaces, a similar characterization is
not available for closed (non-compact) sets, or in general metric spaces.
Indeed we defined several concepts of stability of closed sets in section
V, 4 and it appears that if we are to reach at a characterization we must
first change the definition of prolongation for non-compact sets.
The following lemma gives an insight into what may be done.
2.10
Lemma. If the set M (Xis compact, then
Dt (M) = f\ {y+(S(M, b}): b > 0}.

The proof is elementary and is left as an exercise. \Ve only recall that
Dt (M) is by definition the set V {Dt (x): x EM}.
It is now to be noted that Dt (M) need not even be closed for closed
sets M, which are not compact. And further, in general, if for any closed
set M, we have Dt (M) = M, then the set M need not be weakly stable,.
or equistable. We now introduce the following definition.
2.11
Definition. Given any non-empty set Min X, we shall call the
set(\ {y+(S (M, b)): b > O} as the uniform (first) (positive) prolongation
of M and denote it by Dt (M).
Lemma 2.10 says that if M is compact, then D 1 (M)
D{ (M).
The uniform prolongation has further the following properties.
C X,
2.12
Lemma. For any non-empty set M
2.12.1
D1 (M) is closed and positively invariant,
2.12.2 Dt (M) =={YE X: there are sequences {xn} in X and {tn} inR+
such that e (xn, M) ~ 0 and xntn ~ y},
2.12.3
Dt (M} ) MR+.
The proofs are immediate consequences of the definition.

The uniform prolongation is useful in characterizing the equi-stability
of a closed set.
2.13
Theorem. A closed set 1\,f C X is equi-stable if and only if
Dt(M) == M.
This is an immediate consequence of the definitions and we leave the
details to the reader. vVe note that Theorem V, 1.2 falls as a corollary
of this theorem.
129
3. Generalized Recurrence
In Chapter II we introduced the first positive prolongation and the
first positive prolongational limit set, and we studied some of their
properties. We introduced the higher prolongations in section 1. We shall
now introduce also the higher prolongational limit sets and study some
of the properties. We shall then use these to characterize a notion of
generalized recurrence.
3.1
Definition. The first positive prolongational limit set Ji (x) of
any point x E X is defined by ]'{ (x) = {y E X: there are sequences
{xn} in X and {tn} in R such that xn-+ x, tn-+ + oo, andxntn-+ y}. In
Chapter II this set was simply denoted by J+ (x). Using now the operators
J. and fJ) introduced in section 1, we define for any x EX
Ji (x)
0J.Jt (x),
and if' is any ordinal number, and Jf has been defined for all p <
we set
0<.,
1: = 0(V {J.Jf: p < 0<.}).
We have immediately the following lemma as a consequence of the

definition. In the sequel we denote
simply by]{)(,.
J:
3.2
Lemma. If' > 1, then y E J{)(, (x) if and only if there are sequences {xn}, {Yn}, Yn E J~: (xn), xn-+ x, Yn-+ y, where Pn are ordinal numbers
less than <X, and kn are positive integers. (Recall that for any map I':
1 , where I'1 =I'.)
x-+ 2x, p = I'o
rn-
We leave the proof to the reader. It is also to be noted that:

3.3
Lemma. For any ordinal'> 1, y En: (x) if and only if there
(xn),
are sequences {xn}, {Yn} in x such that Xn-+ x, Yn-+ y, and Yn
where for each n, fln is an ordinal less than X and kn is a positive integer.
(In this lemma and hereafter
is simply written as D{)(, to facilitate the
use of upper indices.)
Enz:
n:
The following lemma now expresses some elementary properties of

prolongations and prolongational limit sets.
3.4
Lemma. For any x EX, and any ordinal 0<.,
3.4.1
3.4.2
J{)(, (x) is closed and invariant,

J{)(,(xt) = J{)(,(x) t = J{)(,(x), for all tE R,
3.4.3
D{)(,(x) =y+(x) V J{)(,(x),
3.4.4
D{)(, (x) is closed and positively invariant,
130
3.4.5
if the space Xis locally compact, then D' (x) and l' (x) are connected, whenever they are compact (if one is compact, then so is the
other), and if DtX (x) (l' (x)) is not compact it does not possess any compact components.
Proof of 3.4.1. 11 (x) has been proved to be closed and invariant in

Chapter II. l ' (x) is closed by construction. To prove invariance, let
lp (x) be invariant for all {J < ' Let y E l '(x), and t E R. Let xn ~ x,
Yn ~ y, Yn El~: (xn), where fJn < ' and kn is a positive integer. Then by
the induction hypothesisyntE l~:(xn). Since Ynt~ yt, we haveytEl'(x),
and the result follows.
Proof of 3.4.2. l ' (x) t = l ' (x) is a trivial consequence of invariance
of l' (x). To see l' (xt) = l' (x) t, note that 11 (xt) = 11 (x) t (Lemma V,
1.4). Now assume that lp (xt) = lp (x) t for all {J < ' Let y El' (xt).
Let Yn El~: (xnt) (wherefJn < 0<., and kn positive integers) such that xnt~ xt
and Yn ~ y. Now xn ~ X, and Yn (-t) E l;;(xnt) (-t) = l~: (xn), by the
induction hypothesis. Since Yn(-t)~y(-t), so y(-t)El'(x), andyE
]'(x) t. Hence l'(xt) ( J'(x) t. Now J'(x) t = l'(xt(-t)) t( l'(xt) (-t) t
= l ' (xt). This proves 3. 4. 2.
Proof of 3.4.3. In Chapter II we proved that D 1 (x) = y+ (x) V
11 (x). Now assume that the result is true for all {J < ' Notice that if
y' E D~ (x'), then by 3.4. 2 y' E y+ (x'), or y' E 11; (x'), where m < k. Now
if y E D'(x), let xn ~ x, Yn ~ y, Yn ED~: (xn) (where Pn < O<., kn positive
integers}. If yn E l J: (xn) (ln < kn) for infinitely many n, then y E l ' (x).
If yn E y+ (xn) for infinitely many n, then y E D 1 (x) == y+ (x) V l 1 (x). In
either case y E y+ (x) V l ' (x): Thus D' (x) ( y+ (x) V l ' (x). Since y+ (x) V
J' (x) C D' (x) is obvious, we have D' (x) = y+ (x) V l ' (x). This completes
the proof of 3. 4. 3.
Proof of 3.4.4. Positive invariance is an immediate consequence of
3.4.3, and D' (x) is closed by definition.
Proof of 3.4.5. The proof of this statement may easily be constructed
by the method adopted for the proof of a similar statement about A+ (x)
and about D 1 (x) and 11 (x) (II, 3.6, 4.4, 4. 7). This we leave to the
reader.
3.5
Exercise. Show that for any ordinal '

l ' (x)
A {D' (xt) : t E R} .
We now recall some of the notions of recurrence that have occurred

earlier, namely, a rest point, a periodic trajectory (or periodic point),
131
a positively or negatively Poisson stable motion (or point), a nonwandering point. We recall that these concepts are respectively equivalent to x == xt for all t E R, xt == x (t + T) for all t E R, x EA+ (x) or
x E A-(x), and x E It (x) which is equivalent to x E 11 (x).
Now let(() denote the class of real-valued continuous functions f on X
such that f (xt) < f (x), for all x E X and all t > 0.
3.6
Definition. Let rQ denote the set of all points x E X such that
/ (xt) == f (x), for all / E ((), and all t > 0. rQ will be called the generalized
recurrent set.
We have immediately
Lemma. rQ includes the non-wandering points in X.
3.7
Proof. Let xEJ1 (x). Lett> 0, and /E0. Then indeed xEJ1 (xt},
and there are sequences xn ~ xt, tn ~ + oo, and xntn ~ x. Then indeed
f (xin) < f(xn), and since I is continuous, we have
I (x) < I (xt) .

As/(xt) </(x) holds by hypothesis, we get f(xt) = /(x). Thus xE rQ.
Now we have
Theorem. rQ is closed and invariant.
3.8
Proof. That rQ is closed is clear. To see invariance, let first x E rQ and

-r > 0. Then for any /E (() f((x-r)t) == /(x(-r + t)) == /(x) =/(xi-). Thus
Xi E rQ. Secondly' let T < 0 and Xi($ rQ. Then there is an I E rv and a to > 0
such that /((xi-) t0 ) </(xi-). Define now gE (() by g(x) =/(xi-) for any
x E X. Then g (xt0 } = f ((xt0 } -r) = /((xi-) t0 ) 0. This remark and the above theorem yield
3.9
Lemma. Let a and b, a < b, be real numbers. Set(() a,b = {/ E (():
a < I (x) < b for all x E X}. Then x E rQ if and only if f (xt) = f (x) for all
f E (()a b and all real t.
'
From now on we shall assume that the space X is locally compact and
separable.
The following theorem shows that in the class (() of functions there is
a function which is constant along any trajectory in the recurrent set,
but is strictly decreasing along any trajectory which is not in the recurrent set.
9*
132
3.10
Theorem. There is an f E (() such that
3.10.1
if x E f'Q, then f (x)
3.10.2
if x ~ rQ and t >
=I (xt) for all real t,

0, then I (xt) < f (x).
and
Proof. Let C (X) denote the continuous real-valued functions on X,

provided with the topology of uniform convergence on compact sets.
Then C (X) contains a countable dense subset and so does (()' = (() _ 1 ,1
Let {/k}, k = 1, 2, ... , be a countable dense set in 0'. Then x E rQ if and
only 1 f k(xt) = f k (x) for k = 1, 2, ... , and real t. Set g =
00
k=1
Ii f k
2
Since Ilk (x) I < 1, it follows that g is continuous and jg(x) I < 1. Thus
gE 0'. If g(xt) = g(x), for all t > 0, then fk(xt) = fk(x} fork= 1, 2, ... ,
and so x E rQ. If x ~ rQ, there is a sequence {tn} in R+ with tn ~ + oo such
00
that g (x)
>
g (xt1} > g (xt2)
indeed f E (()', and
Define
f (x)
==
J e---t g (xt) dt.
Then
has the properties required in the theorem.
We shall now obtain a characterization of rQ by means of the prolongational limit sets. First, the following lemma.
3.11
Lemma. If f E (() and y ED{)(, (x}, then

/(y)
< /(x).
This is an immediate consequence of Lemma 1.3.3 and the definition

of D{)(,(x).
3.12
Definition. The set of all points x EX such that x E /{)(, (x) will be
denoted by rQ{)(,. And we set rQ' = V {r:z?tX: 'an ordinal number}.
The following theorem characterizes rR..

3.13
Theorem. rQ
ordinal0<-.
rQ'. That is, x E rQ if and only if x E J{)(, (x) for some
For the proof we need the following topological theorem.

3.14
Theorem. Let X be a locally compact separable metric space and
let < be a closed quasi-order on X. Let x and yin X such that x < y
does not hold. Then there is an f in C (X) such that (i) if z < z', then
f (z) < f (z'), and (ii) I (y) 0. Then if /E 0, we have /(x) < f(xt). However, we have
133
rv,
by definition of /. Thus for each IE

I (xt) = I (x) for all
therefore, xE ~- This proves f'Q' ( f'Q. To prove rQ ( f'Q',
we define a relation < on X by y
x if and only if y E D' (x) = V D ' (x).
Then
is a closed quasi-order on X. Observe that xt
x whenever
x EX, and t < 0. If x-< y but not y <( x, we write x <( <( y. Note now
that if x($ rQ', and t > 0, then xt <(
x. To see this note that xt <( x.
If x <( xt, then x E D' (xt). Thus x E D' (xt) for t > 0 and some ordinal
<X. Then xE y+(xt) V J()(,(xt). Thus either xE J'(xt) or xE y+(xt). In the
second case y (x) is periodic and so x E J 1 (xt) C J ()(, (xt). In any case then
xE J'(xt) = J'(x). Thus xE rQ' and this is a contradiction. The rest
follows from Theorem 3.14.
I (xt) 0, and,
-<
<
<
<
3.15
Remark. By a quasi-order on X one means a reflexive, transitive,
but not necessarily antisymmetric relation.

Section 1. The notion of higher prolongations is due to URA [4] who also showed
their close connection with stability and introduced the notion of stability of order lX.
The exposition here is based on AUSLANDER and SEIBERT [2]. We have followed the
enumeration of AUSLANDER and SEIBERT for prolongations. URA's enumeration
is different. For example the second prolongation of URA is Dt Dt. The prolongation Dt here is what URA labels as Dro where ro is the first countable ordinal. URA
[4, p. 195] also showed that the prolongations introduced here are the only ones
which lead to different concepts of stability. The notion of a c - c map is one of
the axioms of AUSLANDER and SEIBERT [2], for an abstract prolongation. We show
that this is the concept which leads to various properties which are needed for results
on stability. Thus articles 1.1 to 1.11 are independent of the notion of a dynamical
system. For example 1.9 contains as a particular case URA's characterization of stability given in Theorem V, 1.12. Again Theorem 1.8 shows that the property "for
given xE X, A+(x), D+(x), j+(x) are connected whenever compact" is not
dependent on the notion of a dynamical system. Finally, Theorem 1.13 is due to
AUSLANDER and SEIBERT [2].
Section 2. The notion of absolute stability is due to URA [4]. The important connection of absolute stability with the existence of continuous Liapunov functions was
established by AUSLANDER and SEIBERT [2]. Attempts on the characterization of
stability of compact sets in non-locally compact spaces (or of stability of arbitrary
sets) using URA's prolongations have not been very fruitful. For more recent work in
this direction using the notion of strong attraction see BHATIA [9].
Section 3. This section is almost exclusively a reproduction of results of J osEPH
AUSLANDER [3]. The only exception is the last statement in 3.4.5. Notice that first
part of the statement about D'(~) follows from Theorem 1.8 as D' is a c - c map.
However, the remaining part does need a separate proof. Indeed it is not too difficult to construct examples of c - c maps I' such that I' (x) is closed but not compact,
and I'(x) has a compact component.
Chapter VIII
@1-Liapunov Functions for Ordinary Differential Equations

1. Introduction
Throughout this chapter we shall assume that the ordinary differential equation
1.1
x=f(x), xERn, /: Rn-*Rn
defines a dynamical system (Rn, R, n) in the space Rn, i.e., we shall

assume that equation 1.1 satisfies conditions for the existence, uniqueness
and extendability to the whole real line of its solutions for all points
xE Rn.
This chapter is devoted to the study of sufficient conditions for stability, asymptotic stability and instability of compact invariant sets
in the flow (Rn, R, n), defined by the solutions of the ordinary differential
equation 1.1. In Chapter V (section 5.1) and in Chapter VI we have
defined various possible qualitative properties of the flow in a neighborhood of a compact invariant set. It is now our aim to analyze the qualitative properties of a given differential equation in the neighborhood
of a compact connected invariant set M ( Rn, estimate the region
I' ( Rn in which the behavior shown to hold near M does not change,
and finally to estimate oI' in the case I' =f= Rn. These conditions that we
derive are based upon the construction of suitable continuously differentiable real-valued functions called Liapunov functions.
For simplicity we have only considered the case of a compact connected set; in view of Theorems V, 1.12 and V, 1.22 this is not a restriction and with some more involved proofs practically the same results
apply to the more general case of compact sets.
The theory we present here is essentially a geometric theory whose
foundation lies on some ideas introduced by STRAUSS [5] and by LA
SALLE [4]. This theory will finally be based upon the so-called extension
theorems. These theorems allow an estimate of the region I', M ( I ' (Rn,
135
1. Introduction
in a new way, which, in contrast to the classical theorems, does not

require an analysis of the global geometrical properties of the level surfaces of Liapunov functions. Liapunov functions have already been
applied in section V, 2 for the characterization of asymptotic stability
of compact sets for the dynamical system (X, R, n). The Liapunov
functions used there were. continuous, but not necessarily continuously
differentiable. Their application for the solution of the problem was based
upon inequalities of the type V, 2.2.2 which for their evaluation required knowledge of the positive semi-trajectories y+(x) of the dynamical system (X, R, n) for all points x in a neighborhood N of the compact
set M ( X. For the case of a given differential equation 1.1, the theory
developed in section V, 2 has not much practical use for analyzing
stability properties. However, in the case of continuously differentiable
Liapunov function v = <p (x), the inequality 2.2.2 follows from the inequality
1. 2
d<p (xt)
dt
t=O
<
evaluated along the positive semi-trajectories y+ (x) such that x [O, t) ( N.

Under the same assumption of <p(x}, the function
dipd;t)
lt=o can be
evaluated without explicit knowledge of the solutions of the equation 1.1
by using the relationship
1.3
dipd;t) li=o =(grad <p(x},
X) =(grad <p(x), /(x)).
On the other hand if we let

1. 4
1P (x)
<grad <p (x), f (x)),
by integration we have
t
1.5
<p (xt) -
<p (x) =
J1P (x-r) d-r
from which the inequality V, 2.2.2 follows if 1JJ(X)

Thus from Theorem V, 2.2 we immediately have
<
0 for all
x~
M.
1.6
Theorem. Let x = f(x), x E Rn, f: Rn-+ Rn define a dynamical
system in Rn. Let M (Rn be a non-empty compact set. If there exists a
continuously differentiable real-valued function v = 
0 if
x~
M,
0 for xtf M.
136
VIII.
e -Liapunov Functions for Ordinary Differential Equations

1

The theory that we shall present is a geometric theory which is based
upon the properties of certain sets defined in the space Rn by a realvalued function and a constant. We shall proceed next with definition
of these sets and the proof of some of their properties.
Definition. Let v = <p (x) be a real-valued function defined on

Rn and assume that there exists a compact, connected set M ( X such
that
2.1
p(x) = 0 if x EM.
2.1.1
For a real number {J, consider the following sets:
< {J},
2.1.2
N' {{J)
2.1.3
K' ({J)
=
=
2.1.4
H ({J)
= {x E Rn: {J},
> {J}.
If N' ({J) ) M, let then

2.1.7
N({J)
N({J) be the component of N'({J) containing M, otherwise let

0.
Similarly we denote with K ({J), L ({J), 0 ({J) the component containing

M of the sets K' ({J), L' {{J) and O' ({J), respectively, if such exists.
Let next
2.1.8
Nc(fJ)
0 if N ({J) is not compact,
{ N ({J) if N
({3) is compact.
Similarly we shall define Kc(fl), Lc(fl) and Oc(fJ) as sets which are empty
if the sets K ({J), L ({J) and 0 ({J) are respectively not compact, and equal
to K ({J), L ({J) and 0 ({J) if the sets K ({J), L ({J) and 0 ({J) are compact,
respectively.
The sets defined above have the following properties, whose proofs
are simple and left as exercises.
2.2
Properties. Let v = <p (x) be a continuous real-valued function
defined on Rn and let M (Rn be a non-empty, compact, connected set
such that
2. 2.1
 0, then 0 ({J)
neighborhood of M.
= Oc ({J) =
Nc ({J)
0, 0 ({J) =f= 0 and 0 ({J) 1s a

0, N {{J) =f= 0 and N ({J) is a
2. 2.4
The sets N ({J) and 0 ({J) are open, while the sets K ({J), H ({J) and
L ({J) are closed.
2.2.5
N ({J) C K ({J), 0 ({J)
2. 2. 6
oN (fJ)
H (fJ),
L ({J).
ao (fJ) ~ H (fJ) .
2.3
Properties. Let v = <p (x) be a continuously differentiable realvalued function defined on Rn and let M (Rn be a non-empty, compact,
connected set, such that
 0, such that for all

,..,
{J + e, N ({J) = N c ({J).
,_
p with
The proof of these properties is based upon the Dini's implicit function
theorem and is left as exercise to the reader.
Clearly in many cases the sets Nc(fJ) and Oc(fJ) may be empty. We
shall give next a sufficient condition for Nc(fJ) to be non-empty. A similar theorem, by reversing the sign condition can also be proved for Oc {{J).
From now on we shall discuss mostly properties of Nc(fJ). The analogous
properties of Oc(fJ) can be derived and proved by the reader.
Remark. The Definitions 2.1 and the Properties 2.2 and 2.3 can
2.4
also be applied to the case in which the real valued function v = <p (x)
is defined in an open neighborhood G of the compact, connected set
M (Rn, satisfying 2.1.1. In this case definitions and properties apply to
the sets N' ({J), etc., which are contained in G.
138
VIII. @1-Liapunov Functions for Ordinary Differential Equations
2.5
Theorem. Let v = 0 if xE G - M.
real number {J > 0, such that the open set Nc ({J)
0 if xE M, p(x)
Then there exists a

is
not empty. Furthermore, for every neighborhood U ( M there is a
{J > 0 such that Ne ({J) C U.
Proof. Choose ~ > 0 such that S (M, ~) ( G. Let v =min { 0. Consider next the set
N' (v/2). The component of this set which contains M is contained in
S [M, ~/2] and therefore its closure is compact. From definition 2.1.8, it
follows that Ne (v/2) is not empty. Since this is true for all~'<~, the result
follows.
2.6
Remark. It is easy to show that Theorem 2.5 does not provide
a necessary condition for Nc(fl) to be non-empty. Consider for instance
the following continuous real-valued function defined on the x, y plane:
2.6.l
v = (x2 + y2) sin 1/(x2
+ y2).
Clearly this function is indefinite (it changes its sign) in each neighborhood of the point (0,0) but the set Nc ({J) is not empty.
In the classical stability theory much use is made of the concept of
definite functions.
2.7
Definition. Let v = <p (x) be a real-valued function defined in an
open neighborhood G of a non-empty, compact, connected set MC Rn.
We shall say that
2. 7.1

0 for all
2. 7. 2
<p (x) is negative definite if <p (x) = 0, if x E M, <p (x)
xEG-M.
<
0 for all
2. 7.3
<p (x) is definite if it is either positive or negative definite.
3. Local Theorems
We shall prove next some theorems providing sufficient conditions
for stability, asymptotic stability and instability of compact sets. The
proof of the first theorem follows from the following lemma.
3.1
Lemma. Consider a dynamical system defined by the ordinary
differential equation
139
3. Local Theorems
X == f (x), XE Rn, f: Rn-+ Rn.
3.1.1
Let v == <p (x) and w

Assume that
= 1P (x)
be real-valued functions defined 1n Rn.
@1 ,
3.1.2
v == <p (x) E
3.1.3
'f/J(x) ==(grad p(x), /(x)) < 0 for all x E Rn.
Then for each real number {J, each component of the sets N' ({J) and K' ({J)
is positively invariant.
Proof. The inequality 3.1.3 implies that <p (x) is decreasing along trajectories. Thus N' ({J) and K' ({J) are positively invariant and consequently
so are their components (see II, 1.14).
In the case in which the real-valued function v == <p (x) is defined only
on an open set G containing a continuum M (Rn, the same result as
Lemma 3.1 applies to these components of N' ({J) and K' ({J) which have
a compact closure which is contained in G. In general the following result
holds:
3.2
Lemma. Consider a dynamical system defined by the ordinary
x=f(x),
3.2.1
xERn, /: Rn-+Rn.
Let v == <p (x) and w = 1P (x) be real-valued functions defined on an open

neighborhood G of a continuum M (Rn. Assume that
3.2.2
v == 0 is the
smallest positive real number such that XT E @(N ({J)), then x-r C @(G).
From these it follows that the components of N' ({J) whose closure lies
in G are positively invariant.
We shall next prove the stability theorem.
3.3
Theorem. Consider a dynamical system represented by the
3.3.1
X==f(x),
xERn, /:Rn-+ Rn.
Let v == <p (x) and w == 1P (x) be real-valued functions defined in an open

neighborhood G of a non-empty, compact, connected set M (Rn. Assume
that
3.3.2
 0 such that 0 =f= Nc ((3) C G
and !or each neighborhood U of M there exists (3, 0 < {3 < {J such that
NC ({J) u.
Then M is (positively) stable.
Proof. From the hypothesis 3.3.5. we have that for a sufficiently small
[3 > 0, Nc ([3) C G. In addition from Lemma 3.1 we have that for all
real numbers [3 with 0 < [3 < [3, the set N.c ((3) is positively invariant. Then
1'J is stable.
3.4
Corollary. Theorem 3.3 is true if we replace condition 3.3.4 with:
3.4.1
'f/J(x)
>
0 if x E G
and condition 3.3.5 with:

3.4.2
there exists a real number [3 < 0 such that 0 =f= Oc(f3) ( G
and for each neighborhood U of M there exists [3, [3 < {3 < 0 such that
Oc(fJ) ( U.
By applying Theorem 2.5 and the analogous result on Oc(f3) to Theorem
3.3 and Corollary 3.4 we can immediately prove the following corollary,
which requires a sign condition on <p (x) which implies the property 3.3.5.
This corollary is the classical theorem of stability due to LIAPUNOV.
3.5
Corollary. Consider a dynamical system represented by the
3.5.1
x=I (x),
x E Rn,
/: Rn-+ Rn.
Let v = <p (x) and w = 1P (x) be real-valued functions defined in an open

neighborhood G of a compact, connected set ME Rn. Assume that
3.5.2
 0 or 1P (x) < 0 if x E G - M,
Then M is (positively) stable.

Remark. As shown from the example 2.6.1 condition 3.5. 3 is
3.6
more restrictive than conditions 3.3.5 and 3.4.2. Thus Theorems 3.3 and
3.4 together, cover situations which need not be covered by Corollary 3.5.
This can be directly shown by the following example.
3. Local Theorems
3.7
Example. Consider the following system on the plane (x, y):
3.7.1
{y
where r
3. 7.2
X=
141
y,
-x - yr2 (r sin 1/r - cos 1/r},
x2 + y 2, and the real-valued function

v = <p (x, y) = (x2 + y2) sin 1/(x2 + y2).
Then
3. 7.3
1jJ(X,
y) = : = -2y [sin 1/r - 1/r(cos 1/r)] 2
This real-valued function satisfies the hypotheses of both Theorem 3.3

and Corollary 3.5. The real-valued function 3. 7.2, on the other hand,
satisfies the hypotheses of Theorem 3.3, but not of Corollary 3.5.
We shall give next a theorem on asymptotic stability. For its proof
the following lemma is useful.
3.8
Lemma. Consider the dynamical system represented by the
3.8.1
X= /(x), XE Rn, /: Rn-+ Rn.
Let v = q;(x) and w = "P (x) be real-valued functions defined in an open

neighborhood G of a compact, connected set M ( Rn. Assume that
3.8.2
v = <
0.
If for an xE G, A+(x) is a non-empty, compact subset of G, then

3.8.4
1P (y) == 0 for ally EA+ (x), x E G.
Proof. Condition 3.8.3 ensures that 0, with xt E G, for t > T. Then using Lemma V, 2.1 the result
follows.
Theorem. Consider the dynamical system represented by the dif3.9
ferential equation
3.9.1
x=f(x),
xERn,
/: Rn-+Rn.
Let v == <p (x) and w = "P (x) be real-valued functions defined in an open
neighborhood G of a compact, connected, invariant set M (Rn. Assume
that
3.9.2
v = <p(x) E @1 ,
3.9.3

0, such that Nc ((3)
CG is not
0 for all xE G - M.
Then M is asymptotically stable and

3.9.6
A (M)) Nc(fl).
Proof. Choose {J > 0 in such a way that Nc ({J) C G. Then from Lemma 3.1, x E Nc ({J) implies y+ (x) C Ne ({J). Now, if M were not an attractor there would exist y E Nc(fl), such that A+(y) =F 0 and A+(y) C[ M.
On the other hand, from Lemma 3.8 it follows that 1P (x) == 0 if x EA+ (y).
From hypothesis 3.9.5 this implies that A+(y) CM. Thus Mis an attractor and 3.9.6 holds. Let us now prove that M is stable. If it were not
stable, then D+ (M) - M =F 0. Since Nc ({J) is positively invariant
(Lemma 3.1) D+(M) C Nc(fl). Consider a point xE D+(M) - M; since
Mis an attractor there exist points y, z EM, y E A-(x) (\ M, z E A+(x)
(\ M (V, 1.26.2). Consider then monotonic sequences {tn} CR+ and
{rn} ( R- with tn ~ + oo and Tn ~ - oo, xtn ~ z and X't'n ~ y.
Consider the corresponding value of the function <p. We have tX = <p (z), which contradicts the fact that <p (x) vanishes on M.
Again from Theorems 2.5 and 3.9 we can derive the classical asymptotic stability theorem.
3.10
Corollary. Consider the dynamical system 3.9.1. Let v == p(x)
and w == 1P (x) be real-valued functions defined in an open neighborhood
G of a compact connected invariant set MC Rn. Assume that
3.10.1
v = 0, x~ M,

1P(x) == (grad <p (x), /(x)) < 0, x E G -
3.10.3
x E G,
<p (x)
M.
3.11
Remark. The conditions of Theorem 3.9 imply that the realvalued function <p (x) is positive definite in Ne {{J). If this were not true
there would exist points y E (Nc ({J) - M) in which <p (x) has a minimum
and therefore grad <p (x) = 0, which contradicts condition 3. 9.5 of
Theorem 3.9. This shows that Theorem 3.9 is completely equivalent to
the classical theorem on asymptotic stability (see, for instance, BHATIA
and SzEGO [1], Theorem 3.6.15).
3. Local Theorems
143
In the proof of Theorem 3.9 the asymptotic stability of the compact

set M was deduced from the sign of 1P (x) and from the fact that M is an
attractor. The same result can be obtained under different assumptions
if we notice that if 1P (x) is sign definite for the set M and M is stable,
then M is asymptotically stable. This is done in the following theorem.
3.12
Theorem. Consider a dynamical system defined by the differential equation
3.12.1
x ==I (x),
x E Rn,
f: Rn-+ Rn.
Let v = <p (x) and w = 1P (x) be real-valued functions defined in an open

neighborhood G of a continuum MC Rn. Assume that
3.12.2
v = <p (x) E @1 ,
3.12.3
3.12.4
= 0 if x E M,
1P (x) = (grad <p (x), 1P (x)) =F 0 if x E G - M,
3.12.5
M is stable.
<p (x)
Proof. As in the proof of Theorem 3.9 we can show that for all
x E G - M, the set A+(x) is non-empty, compact and such that A+(x)
(M.
Theorem 3.9 gives also an estimate 3.9.6 of the region of attraction

A (M). This estimate can be improved in the following corollaries which
we give without proof.
3.13
Corollary. Let the conditions of Theorem 3.9 be satisfied. Let,
in addition,
3.13.1
v {Nc([3) c G: fJ >
0, Nc(fJ)
c G}.
Then
3.13.2
A (M)) N.
3.14
Corollary. Let the conditions of Theorem 3.9 be satisfied and
G:::::::. Rn. Let
3.14.1
Rn==
v {Nc(fJ), fJ >
O}.
Then the compact, connected set M (Rn is globally asymptotically stable.

By using the same technique used in the proof of Theorem 2.5 it is
possible to give a sufficient condition for a real-valued function v = 
0 if x E fJ.M.
Assume that lim <p (x) exists. Then condition 3.14.1 is satisfied.
llxll~ 00
3.16
Remark. Theorems giving sufficient conditions for negative
stability and for negative asymptotic stability (complete instability)
can be proved either by inverting the sign conditions 3.5. 7 in Corollary 3.5
and 3. 9.5 of Theorem 3.9 or by considering both in Theorem 3.3 and 3.9
the sets 0 c ({J) in place of the sets N c ({3). Corollaries 3.13 and 3.14 will then
apply to the region of negative attraction A- (M).
By means of continuously differentiable real-valued functions we can
also give sufficient conditions for the (not necessarily complete) instability of a compact, connected set M (Rn. This result applies to the case
of a continuum M ( Rn which i~ unstable, but neither completely unstable, nor attracting. This theorem gives also an estimate of the region
of instability (V, 1.13).
3.17
Theorem. Consider a dynamical system represented by the
3.17.1
X = /(x), XE Rn, /: Rn-+ Rn.
Let v = <p (x) and w = 1P (x) be real-valued functions defined on the closure B of an open set BC Rn. Assume that
3.17.2
v = p(x) E @1 , on B,
3.17.3
1P (x)
Let M
(grad 
0 a real number such that
3.17.4

x EB f\ S(M, 17),
3.17.6
there is a sequence {xn} CB, xn-+xE M, such that <p(xn)
0,
> 0.
Then M is unstable and

3.17.7
D+(M, B) f\ H(M, ij)
Proof. For each real f3

3.17.8
>
=F 0,
0 < ij < 17.
0, consider the set
OB ({J) = {x EB: fJ}.
Since B is an open set also OB ({3) is open. In addition from the hypothe8is 3.17.4, for each f3 > 0, we have:
3.17.9
oOB ({J) (\ (oB (\ s (M, 17)) = 0.
4. Extension Theorems
145
The set 0 B ({J), in general, is not connected. On the other hand from the
hypothesis 3.17.5 it follows that if xE OB ({J) f\ S (M, 'YJ), then either there
existsT > OforwhichxTEH(M,'YJ), or y+(x) C (OB({J) f\ S(M, 'YJ)). In the
latter case y+ (x) is compact, hence A+ (x) is non-empty and compact and
in addition A+(x) f\ (oB f\ S(M, 'YJ)) == 0. Since if z1 EA+(x),1J'(z1) == 0,
we have that z1 EH (M, 'YJ). Thus for each ij < 'YJ there exists a real
number T-;; > 0, such that Xirj E H(M, n). From the hypothesis 3.17.6,
given a point xn in the sequence {xn} C B, for a sufficiently small {3 > 0,
we have that xn E OB ({3) and therefore for each ij < 'YJ there exists a
real number Tn> 0 such that XninEH(M, ij), which proves the theorem.
From the proof of Theorem 3.17 the following corollaries can be
derived.
3.18
Corollary. If the hypotheses of Theorem 3.17 are satisfied for an
arbitrary real number 'YJ > 0, then M is globally unstable.
Corollary. If the hypotheses of Theorem 3.17 are satisfied, then
3.19
each compact set K ( S (M, 'YJ) (\ oB is unstable.
The following stronger theorem is given without proof.

3.20
Theorem. Let M ( Rn be a compact, connected set and let the
hypotheses of Theorem 3.17 be satisfied. Let K ( Rn be a compact,
connected set such that M C oK and (D+ (M, B) - M) ct K. Then K is
unstable.
Remark. Theorems equivalent to Theorem 3.17, Corollaries 3.18
3.21
and 3.19 and Theorem 3.20, respectively, can be proved with different
sign conditions on<p (x) and 1J' (x), namely with conditions 3.17.5 and 3.17.6,
respectively replaced by:
3.21.1
1J' (x)
<
0 if
x E Bf\ S (M, 'YJ)
and
3.21.2
there exists a sequence {xn} CB, xn--+ x EM, with <p (xn)
<
0.
The proof of those theorems and corollaries is left as exercise to the reader.
3.22
Exercise. Prove Theorem 3.9 without assuming that the continuum M is invariant.
The results derived in the previous section regarding the estimate
of the region of attraction A (M) of the compact, connected, asymptotically stable set M (Rn (Corollary 3.13 and Corollary 3.14) are decep10
Bbatia/Szego, Stability Theory
146
VIII.
e -Liapunov Functions for Ordinary Differential Equations

1
tively strong. Their application is, usually, not easy since they require
a proof of compactness of the set N (3.13.1) which is generally difficult.
On the other hand, condition 3.9.5 of Theorem 3.9 imposes a sign condition on the real-valued function 1/l(x), which, from the hypothesis made
on /(x), can immediately be translated into a sign condition on the vector
grad q;(x), from which, in turn, one should be able to derive an estimate
of the set N.
This will be done in this section.
4.1
Theorem (Local Extension Theorem). Consider a dynamical
system represented by the differential equation
Let v = <p(x) and w = 1P(x) be real-valued functions defined on Rn. Let

M ( Rn be a compact, connected set. Assume that
4.1.2
v = 
4.1.6
Mis positively asymptotically stable.
is a real number, such that for each sequence

{xn} ( N (P), with xn fr M, and 'f/J(xn) ~ 0, we have <p(xn) ~fl.
0
Then
4.1. 7
A (M)
N ((30).
If, in addition,
4.1.8
1P (x)
=* 0
if
x E 8N (P},
then
4.1.9
A (M) ) N (P)
In condition 4.1.5 the statement Xn -/+ stands for that there is a

neighborhood U of M and a subsequence {xnk} of {xn} such that Xnk ~ U.
Proof. If the theorem were false, then oA (M) r\ N ({3) =* 0. Let

E oA (M) (\ N ((3). Since the differential equation 4.1.1 defines a dynamical system, the set oA (M) is invariant (V, 1.3.2 and II, 1.6).
x0
Thus y (x0 )
4.1.10
oA (M). Let next
= inf {I 1P (y) I:
yE r (x0) (\ N (P)}.
147
Condition 4.1.5 implies that 1P (x) cannot change sign in N ({36) - M,

then the following four cases may arise:
4.1.12
== 0,
4.1.13
>
>
4.1.14
>
1/J(x) <
1J'(x) >
1J'(x) <
0, 1J'(x)
4.1.11
0,
0,
0 if xE N(/1) - M,
0 if
EN (/1) - M,
if xE N(/1) - M,
if xE N(/1) - M.
In the case 4.1.11 there exists a sequence {xn} Cy (x6 ) f\ N ((36 ) such that
1J' (xn) ~ 0 and therefore from 4.1.5, 0.
Since 1J' (x) > 0 for x E N (/1) - M, it follows that
4.1.15

<p (y) for all t with y [O, t]
C N (/1).
Since q;(x)=P for xEoN(/1}, from 4.1.15 it then follows that

A+(y) ct M, which contradicts the hypothesis 4.1.6 and proves the theorem in the case 4.1.11.
In the case 4.1.12 we have that the set N (/1) is positively invariant
(Lemma 3.1). Thus y+(x0 ) ( oA (M) f\ N(/1). Since== 0 we have that
there exists a sequence {xn} C y+(x0 ), such that <p(xn) ~ {16. This contradicts the fact that 0, such that 
<p (x0 )
in the case 4.1.11 and prove the theorem in this case.

In the case 4.1.14, since N(/1) is positively invariant (Lemma 3.1),
we have
4.1.17
<p (x0t)
<
<p (x0 )
t for all t ER+.
Hence there exists TE R+ such that <p (x0i-) < 0. Since <p (x) is continuous
there exists then a point y EA (M) f\ N ({36 ), such that <p (y) < 0, which
is absurd.
Th us in all cases
4.1.18
oA (M) f\ N ({30) == 0.
Since A (M) f\ N(f36 ) = 0, _it follows that A (M) ) N(fJ6) (4.1.7), which
completes the proof of the first part of the theorem.
Assume next that, in addition, condition 4.1. 8 holds; we shall prove that
this implies that oA (M) (\ oN ({36) = 0. For this let XoE oA (M) (\ oN ({16).
10*
148
VIII. @l-Liapunov Functions for Ordinary Differential Equations
Let next 1P (x0) > 0. Since A (M) ) N (/3), there exists y E A (M), such
that 0. On the other hand, if 1P (x0 ) > 0, we have that 1P (x) > 0
for all x EN ({J0 ) - M. Following the same arguments as in the case
4.1.11 we reach a contradiction.
Let next 1JJ(x0 ) < 0. Then 1fJ(x) < 0 for xE N(/3) - M. Let y+(x0 }
( oA (M). For t E R+, sufficiently small, we have <p (x0 t) < p0 Thus,
either x0t E N ({J0}, which is absurd, or x 0 belongs to the boundary of a
component of N' ({3), say N 1 ({J0) =F N ({3). Now for t ER+, sufficiently
small, <p (x0t) < p0 Hence x0t E N' ({3), x0t E N (p0 ), x0t E N 1 ({J0}. Thus
N ({J0) and N 1 ({3) cannot be two different components of N' ({3) and the
proof follows from that of the previous case.
This concludes the proof of the theorem.
4.2
Remark. It is possible to prove a theorem, analogous to Theorem 4.1, where instead of the sets N ({3), {J 0 > 0, we consider the sets
,..,
0 ({J0), p0 < 0. This is left as an exercise to the reader.
We shall next prove a theorem regarding the global stability properties of a compact, connected set M (Rn. The statement of this theorem
is stronger than that of Theorem 4.1.
4.3
Theorem (Global Extension Theorem). Consider the dynamical
system defined by the differential equation
4.3.1
x == f (x),
x E Rn,
/: R"' ~Rn.
Let M ( Rn be a compact connected set and v == <p (x) and w == 1P (x)

be real-valued functions on Rn. Assume that
4.3.2
v = q;(x}E le1 ,
4.3.3
q;(x)=O if xEM,,
4.3.4
1J' (x)
4.3.5
{xn} (Rn,
(grad <p (x), I (x)),
1J'(Xn}
0 imply
Xn ~ M.
Then whatever the stability properties of M are, they are global.
Proof. From Theorem 3.12 it follows that, under the hypotheses of

Theorem 4.3, positive and negative stability imply positive and negative
asymptotic stability, respectively. Hence the local stability properties
of M can be only the following: instability (in both directions), positive
asymptotic stability and negative asymptotic stability (complete instability). In the first case the theorem is true, since the conditions of
Theorem 3.17 (Remark 3.21) are satisfied for B ==Rn and any real
number 'YJ > 0. Indeed, in this case there exists a sequence {xn} C Rn,
xn ~ M along which sign <p (xn) == sign 1J' (xn): if this were not true, then
149
either the hypotheses of Theorem 3.9 or those of the theorem of negative

asymptotic stability (Remark 3.16) would be satisfied. M would then be
either positively or negatively asymptotically stable which contradicts
the hypothesis. Hence Mis globally unstable (in the sense of V, 1.13).
Consider next the case that M is asymptotically stable. Then, if the
theorem were not true, oA (M) =F 0. Consider next a point Xo E oA (M)'
since oA (M) is invariant, y+(x0 ) ( oA (M). Let next
4.3.6
==inf {!1P(Y) I: YE y+(x0 )}.
From the hypothesis 4.3.5 it follows that > 0.

Assume now that 1P (x) > 0 for all x E oA (M). Then
4.3. 7

 0. Thus M cannot
be asymptotically stable, which contradicts the hypothesis.
If, instead, we assume that 1P (x) < 0 for all x E (}, (M), then
4.3.8
<p (x0t)
<
<p (x0 )
t.
Again, fort ER+ sufficiently large we have <p (x0t) < 0, which contradicts
the hypothesis made.
Finally in the case in which M is completely unstable, we shall replace
equation 4.3.1 with the equation
4.3.9
x == -f (x).
If the hypotheses of the theorem were satisfied by equation 4.3.1 they

are still satisfied by equation 4.3.9. Now however, M is asymptotically
stable. The proof of this case is therefore reduced to the previous one.
This completes the proof of the theorem.
Remark. Theorems 4.1 and 4.3 still leave some open problems
and some unanswered questions, some of which will be clarified in the
next section. The most important of these problems is concerned with the
identification of the set N ({1), where {J0 is defined in 4.1.5, for the case in
which 1P (x) =f= 0 if x E Rn - M. This is therefore a case in which only for
some sequences {xn} ( N((3), xn ~oo, we have 1P(xn) ~ 0. Clearly only
two situations may now arise: the case in which N ((3) == V N ((3)
4.4
O~P</3
== Rn and the case in which N ({J0} is not compact, but N ((30) =F Rn. The
distinction between these two cases is an open problem. The finite version
of this problem, which has already been mentioned in Theorems 2.3 and
2.4, is the following. Assume that there exists a point x0 E Rn, such that
1P (x0) == 0, but for all diverging sequences {xn} C N ((3), xn --+ oo, 1P (xn)
fr 0. Then N ((3) may or may not have a compact closure (see Theorem
,..,
2.4) however, for all e > 0 and all fJ < {J0 - e, the set Ne ({J) is non-
150
VIII. ei-Liapunov Functions for Ordinary Differential Equations
empty. On the other hand, if there does exist a sequence {xn} C N ({J0 ),
xn ~ oo, then 1P(xn) ~ 0 may not hold. In this case, if for s~me {J > 0,
Ne ({J} =F 0, it is not trivial to find the largest real number {J such that
Nc (p) =F 0. This problem is not related to that of the estimate of A (M)
(since A (M) is estimated from 4.1.7 and 4.1.8 anyhow), but with the
problem of the geometrical properties of A (M). Clearly when the Liapunov
functions v = <p (x) are polynomials (which do not have singular point at
infinity) these problems do not arise.
We close this section by illustrating with an example the use of
Theorem 4.1.
4.o
Example. Consider the differential equation V, 2.4.1 and the

real-valued function V, 2.4.4. Let M = {(O, O)}. The (0, 0) is locally
asymptotically stable. The total time derivative of the function V, 2.2.4
with respect to the equation V, 2.3.1 has the form
4.5.1
1J' (x)
-2y2
+ (t
-2y2
:x:
2) 2
2.x2 (x2y2 - 1)
(1 + x2)2
for x2y2 > 1,

for x2y2 < 1.
We see that the only points in which 1P(x) vanishes are (0, 0) and y = 0,
lxl ~ + oo. Condition 4.1.5 holds if p0 = 1, but not if {J0 > 1, and indeed
2
A (M) 2 { (x, y) : y
+ 1 ~ xz <
1} .
5. The Structure of Liapunov Functions

The extension theorems proved in the previous section are still too
restrictive in the sense that (see next section: Theorems 6.3 and 6.4),
it is possible to characterize the property of positive or negative asymptotic stability of a continuum M ( Rn by means of Liapunov functions
whose total time derivative 1P (x) is only semidefinite. In order to prove
these theorems some additional information on the structure of the sets
N' ({J), N ({J) and Nc ({J), must be obtained. For that we shall apply the
stability theorems proved in sections 3 and 4 to the differential system
x = grad <p (x). Thus the real-valued function <p (x) is now assumed to
be twice continuously differentiable.
We shall begin our investigation by asking ourselves what properties
<p(x} must have in order that it is a Liapunov function which characterizes
the positive (or negative) asymptotic stability properties of a continuum
Mand has a sign-definite total time derivative 1P(x).
The first obvious conditions, 5.1 and 5.2, are of geometric type; their
proof is an immediate consequence of the theorems of section 3 and it is
left as an exercise to the reader.
151
o.1
Proposition. A necessary condition for a real-valued function

<p (x) to be a Liapunov function with negative definite 1P (x), suitable to
characterize the properties of asymptotic stability of a compact, connected set M C Rn, is that
5.1.1
N ({J}
Nc ({J) for all {J
<po
and
5.1.2
N ((JO)=
NC ({J),
0~{3<(3
where the sets N ({J) and Nc (/3) are defined in 2.1. 7 and 2.1.8 and
defined in 4.1. 5.
po is
In the case of global properties, we have
o.2
Proposition. A necessary condition for a real-valued function

<p (x) to be a Liapunov function with negative definite 1P (x), suitable to
characterize the properties of global asymptotic stability of a compact,
connected set M C Rn, is that
5. 2.1
N ({J)
Nc ({J) for all {J
and
5.2.2
Rn=
V Nc(fJ}
{3~0
Conditions 5.1.1, 5.1.2, 5.2.1and5.2.2 clarify the structure of Liapunov

functions used in the characterization of the positive or negative asymptotic stability properties of compact, connected sets. From the practical
point of view, however, such conditions are not easy to satisfy and they
are usually replaced by analytical conditions of the following type.
o.3
Proposition. Let v
and connected set. Let
5.3.1
<p (x) = 0
for
==
<p (x), <p: Rn~ R. Let M ( Rn be a compact
E M.
If there exist two strictly increasing functions

= {J (0) = 0 such that
5.3.2
lX
X ()
and {J () with
lX
(0)
(e (M, x)) < Jq; (x) I < {J (e (M, x)),
then such a real-valued function can be used for the characterization of

positive (or negative) asymptotic stability properties of the set M. If in
addition
5.3.3
() -+
+ ex> as -+ + oo,
then the real-valued function <p (x) can be used for the characterization of
positive (or negative) global asymptotic stability properties of the set M.
152
Notice that condition 5.3.2 is a necessary and sufficient condition

for a continuous real-valued function v = <p (x) to be definite.
5.4
Remark. Clearly (see Theorem 2.5) condition 5.3.2 implies condition 5.1.1 and together with condition 5.3.3 implies condition 5.2.2
while the converse is not true, for instance the function 2. 6.1 satisfies
condition 5.1.1, but not condition 5.3.2. On the other hand condition
5.2.2 clearly does not imply condition 5.3.3; this can be easily seen by
taking a function <p (x) which satisfies in the whole space Rn conditions
5. 3. 2 and 5. 3. 3 and therefore 5. 2. 2, then considering the function
1
r<;)(x) which satisfies 5.2. 2 but not 5.3.3.
These remarks may have somewhat clarified the relationship between

the classical conditions (5.3.2 and 5.3.3) and the geometrical conditions
5.1.1 and 5.1.2.
Still we have to discuss a problem of great practical importance, that
of the relationships between the topological properties of N' ({3) and the
existence of critical points of a continuously differentiable, real-valued
function <p (x) (the points y E Rn, such that grad <p (y) == 0).
The following theorems are an application of Theorems 4.1 and 4.3 to
a differential equation uniquely defined by the continuously differentiable
real-valued function <p (x): <p: Rn~ Rn, namely, the differential equation
x ==grad <p (x),
5.5
and others of a similar kind.
5.6
Theorem. Let v == cp(x), cp: Rn~ R. Let M (Rn be a compact,
connected set. Assume that the following conditions are satisfied:
5.6.1

>
0 such that Ne ({3)
NC (/3) =f= 0
and in addition
5.6.6
Nc ({3}
== Rn.
{J?:.O
Proof. Consider the differential equation

5.6.7
x == -
grad <p(x)
Jlgrad<p(x)ll
== g(x).
=f= 0,
153
This equation defines a dynamical system since all its solutions are
unique and defined on R (see section I, 2.1). We shall investigate its
stability properties by means of the Liapunov function v == <p (x). Consider
then the real-valued function
5.6.8
1P (x)
== (grad <p (x), g (x)) == -
II grad <p (x) 112
+ IJgrad <p(x)ll <
0.
From the condition 5.6.4 it follows that for each sequence {xn} C Rn,
1P(xn)--+ 0 implies xn ~ M. We are now in the position of applying
Theorem 4.3, and conclude that the compact set MC Rn, is globally
asymptotically stable for the differential equation 5.6. 7.
We shall next prove that for each (3 > 0 the set N' ((3) is connected.
Assume by contradiction that there exist two components of N' ((3), say
N 1 ({J) and N 2 ({J), such that N 1 ({J) f\ N 2 ({J) == 0 and let N 1 ({J) ) M. Let
now x 0 E N 2 ((3), since M is globally asymptotically stable, we have
y+(x0} f\ oN1 ({J) =F 0. Let yEy+(x0) r\ oN1 ({J), we must have q;(y) == {J.
On the other hand 0 the set N ({J) is compact. Let (3 be such that N ({J) is
not compact.
From Theorem 4.3, Mis globally asymptotically stable, hence for all
xE 8N(f3), (} (xt, M) ~ 0 as t ~
oo.
Let next
5.6.9
== sup 1P (x)
for x E Rn - Ne ((3),
where {J > 0 is defined by 5.6.3. From the hypothesis 5.6.4, we have that
> 0. Let next
5.6.10
'fi - f3
T
- -/t-
Then for all -,; > T and all x E 8N ((3) we have x-r: E Ne ({J). In particular
we have Xn7: == Yn E Nc(fJ) for a sequence {xn} oN({J), Ilxnll ~ + oo.
Let now Yn ~ y*, hence Yn (--r) ~ y* (--,;). On the other hand Yn (--r)
= xn, which contradicts the assumption that j(xnll--+ + oo, and proves
the theorem.
,1.,
In a similar way, by applying Theorem 4.1 to the differential equation 5. 6. 7, we can prove the following local result.
5.7
Theorem. Let v == <p(x), q;: Rn~ R. Let M (Rn be a compact,
connected set. Assume that conditions 5.6.1, 5.6.2 and 5.6.3 of Theorem
5.6 are satisfied. In addition assume that
5. 7.1
{J0 > {J is such that if {xn} C N ((3) and Xn fr M, then grad <p (xn)
~ 0 implies <p (xn) ~ po.
154
Then
5. 7.2
N ({J) =Ne ({J) for all {J
<
{J
and
5.7.3
N({J)
N({J).
O~/J-5-fJO
o.8
Remark. Clearly, if the differential equation
5.8.1
x=f(x),
/: Rn~R"
defines a dynamical system and

5.8.2
11' (x) = (grad <p (x), /(x)),
then the conditions of Theorems 4.1 and 4.3 on 1P(x) imply the conditions of Theorems 5. 6 and 5. 7, respectively. Thus in Theorems 4.1 and
4.3 the sets N({J) do satisfy the conditions 5.6.5, 5.6.6, 5.7.2 and 5.7.3
respectively.
One can prove a theorem similar to Theorem 5.6 also for the analytical conditions 5.3.2 and 5.3.3. Thus we have
o.9
Theorem. Let v
5.9.1
v = 0 for xE H(M, ~), ~ >
5.9.4
for {xn} (Rn, grad q;(xn)
~ 0
0,
implies xn ~ M.
Then there exist two strictly increasing

that
5.9.5
functions~()
lX(e(x, M)) < <p(x) < {J(e(x, M)), lX{O)
and {J(), such
{J(O) = O
and, furthermore,
5.9.6
lim lX ()
-+co
= + ex:>.
In addition to this, if
5. 9. 7
v = min { v for
x E @(S [M,
~]).
Proof. Consider again the dynamical system 5.6. 7 and the realvalued function 5. 6. 8.
155
The functions v = 0 such that
x0t: EH (M, "). Thus v. Since <p(x0t:) > v.
Lastly as <p(x) = 0, for xE M, <p(x) > 0 for x~ M, and <p(x) is continuous, we can define two continuous increasing functions x ( 'YJ) and
{J ('Y/) by
x (ri) =min { x('Y/) > 0 for 'YJ > 0. There exist
strictly increasing continuous functions x ('Y/) and fJ('YJ) such that
cX (11) < cX('f/) < fJ('Y/) < fJ('Y/).

With these x ('Y/) and {J ('Y/) we have
x ((! (x, M))
<
<p (x)
< {J ((!(X, M)).
It remains to be proved that there exists one function x('Y/) which satisfies condition 5.9.6. We shall prove first that for all x0 E fd (M), <p(xJ)
~ + oo as t ~ - oo. In fact, notice that for any x 0 E fd (M)
t
 'YJ > 0. Thus

1P(x0-r) <max {1P(x): x~ S (M, ")} =
-x <
where"> 0 is such that x0-r~ S(M, ")for-,; < 0. Hence fort < 0
t

<p (xo)
+ f x dt: =
<p (xo) -
xt.
Thus <p (x0t) ~ + oo as t ~ - oo.

Now the existence of our x('Y/) with the property that x('Y}) ~ + oo
as 1J ~ + oo is equivalent to the property that 0 such that the surfaces h. Consider the open set P defined by
P
= {x: 0 < k <
<p(x)
<
h}.
This open set is bounded away from the set M and is bounded by the
surfaces <p (x) = k and <p (x) = h. We claim that there is a point x0 , <p (x0 )
= k, such that x 0t E P for.t < 0. For, otherwise, for every x 0 , with <p (x0 )
= k we will have a unique t < 0 such that x.,tE {x: <p(x) == h}, so that,
156
VIII. @l-Liapunov Functions for Ordinary Differential Equations
following the argument in the proof of Theorem V, 2. 9, there will be

continuous map of the compact set as
t ~ - ex> for all x0 ~ M. The theorem is completely proved.
6. Theorems Requiring Semidefinite Derivatives

In this section we shall prove theorems which allow us to deduce
properties of attraction and asymptotic stability of a compact set M ( Rn
using conditions different from'those of Theorem 3.9.
In particular we allow "P (x) to be only semidefinite. After the
proof of the main theorem (6.1) we shall prove the corresponding extension theorems relaxing some conditions on <p (x) in Theorem 6.1. The
proofs of these extension theorems depend upon the results proved in the
previous section.
6.1
Theorem. Consider the dynamical system represented by the
X = /(x), XE Rn, j:
6.1.1
Rn~
Rn.
Let Q be a compact, positively invariant set. Let v === 
0, x E Q,
<grad <p (x), f (x)),
= 1/l(x) <
0, xE Q.
Consider the following sets :

(I) Sis the largest invariant set in Q,
(II) M
(III) P
= {xE !J: q;(x) = O},

= {xE Q: 1JJ(x) = O},
(IV) Q is the largest invariant set in P (Q ~ 5),

(V) U is the largest invariant set contained in M.
Then
(a) The sets S, P and Q are closed,
(b) M
P and Mis stable relative to Q,
( c) Q is attracting relative to Q,
157
(d) S is asymptotically stable relative to Q,
( e)
oQ (\ as =t= 0,
S = D+(Q, Q) = {y: there
is a sequence {xn} C Q and a sequence

{tn} ( R+ such that xn ~ Q and xntn ~ y} where D+(Q, !J) is the first
positive prolongation of Q relative to Q,
(f)
(g) S) M, and is the largest asymptotically stable set relative to Q,

(h) if for all xE
to Q and Q == S,
oQ,
<p (x)
= const, Qis
asymptotically stable relative
(i) Q minimal implies Q == S == U,
(j) M = Q implies Q is asymptotically stable relative to Q and

M=S,
(k) if either Q, Sor Qare homeomorphic to the unit ball, then M contains a rest point,
(1) if either QC JQ or SC JQ, then the words "relative to Q" may
be deleted" from the above sta tern ents.
Proof.
(a) This is clear, for if a set is invariant, so is its closure.
(b) For all xE M, from (ii) and (iv) we have 1/'(x) == 0 which implies
M ~ P. Stability then follows from Theorems V, 4.2, 4.5, 4.6.
( c) Since for each y ED, A+ (y)
A+ (y) C Q as A+ (y) is invariant.
C Q and "P (x) == 0 for x EA+ (y), we get
(d) Since S) Q, S is attracting relative to Q. Stability follows as S

is the largest invariant set in Q, D+(S, !J) C Q, so that D+(S, Q) == S.
(e) For if aQ f\ as== 0, then as ( Q being invariant, we have x E oS,
A+ (x) =t= 0 and A+ (x) f\ Q == 0, since Q C S, we conclude that Q is not
attracting relative to Q, contradicting (c).
(f) Follows from proof of (d).
(g) Obvious (V, 5. 12).

(h} We need only show that Q is stable relative to Q. If not, then
there is a sequence {xn} in Q, xn ~ x E oQ, and a sequence {tn}, tn > 0,
such that xntn ~ y~ Q. Indeed y ED and xn [O, tnJ C Q as Q is positively
invariant and compact. Since <p(z) = <p(x), showing that <p(y)= <p(x). This
shows, however, that y (y) C Q and <p (x) is constant on y (y). Consequently
1/'(x) = 0 on y (y), showing that y (y) C Q, a contradiction.
158
VIII. ~ 1 -Liapunov Functions for Ordinary Differential Equations
(i) If Q is minimal, then <p (x) is constant on Q and the result follows
from (h) and the fact that U C S.
(j) Follows from (h) and (c).
(k) Follows from V, 3,8.

(1) Follows from V, 5.2.
In view of Lemma 3.1 from Theorem 6.1 it immediately follows:
6.2
Corollary. Consider the dynamical system defined by the ordin ary differential equation
6.2.1
XE Rn, /:
X = f (x},
Rn~
Rn.
Let v = 0 such that the set Nc ({J) =t= 0, Theorem 6.1 holds with
6.2.2
=NC ({J).
In view of the Theorems 5.6 and 5. 7 which allow an estimate of largest

Ne ({J) from the properties of grad <p (x), we are now in the position of
proving the following theorems:
6.3
Theorem. Consider the dynamical system, defined by the ordinary differential equation
6.3.1
X=f(x),
xERn, /:
Rn~K'.
Let v = <p (x) and w = 1P (x) be real-valued functions, which satisfy the
assumptions of Theorem 6.1 in the whole space Rn. In addition assume
that
6.3.2
<p (x) E @2, x E Rn.
6.3.3
The sets S, Q, P and M defined as in Theorem 6.1 with Q =Rn
are compact.
6.3.4
There exist two real numbers {J and {J', {J
<
{J' such that
(i) The set H ({J) has a compact component which will be denoted
Hc(f3). If we denote with Nc(fJ) the component with compact closure of
N ({J') which contains He ({J), Ne ({J') =F 0.
(ii) For all sequences {xn}CRn,

Then Theorem 6.1 holds with
6.3.5
=Rn.
grad<p(xn)~Oimpliesxn~Hc(fJ).
159
Proof. From Theorem 5.6 we have that Rn=
V Nc(k).
k"2!fJ
Then since
-
the sets S, M, P, Qare compact there exists a real number {J such that
all the above mentioned sets are contained in Ne ({J) for all {J > {J.
6.4
Theorem. Consider the dynamical system defined by the ordinary differential equation
x=f(x),
6.4.1
xERn, /: Rn-+Rn.
Let v = <p (x) and w = "P (x) be real-valued functions which satisfy the
assumptions of Theorem 6.1 in the whole space Rn. In addition assume
that
v = {J' such that if {xn} C N ((JO), xn fr Hc(fl), then grad<p(xn)
-+ 0 implies <p (xn) -+ po.
Then Theorem 6.1 holds with
Q
N ({J) for all {J with {J
< {J-<
{J'.
Proof. From Theorem 5. 7 it follows that N ({J) is compact for all

{J < {J < {J'n and that Nc(fl) = N ({J).
Clearly the sets S, M, P, Q must be referred to the N ({J). For the
a pplica tion of the results the following corollary is useful.
6.o
Corollary. Consider the dynamical system defined by the ordinary differential equation
X = f (x}, XE Rn, f: Rn-+ Rn.
6.5.1
Let v
<p (x) and w = "P (x) be real-valued functions defined in Rn. Let
MC Rn be a compact, connected set. Assume that

6.5.2
v = <p (x) E f2 2,
6.5.3
q;(x) = 0 if x EM,
6.5.4
"P (x)
6.5.5
for all {xn} (Rn, grad <p"(xn)-+ 0 implies xn-+ M,
6.5.6
'f/J(x)
6. 5. 7
M is the only invariant set contained in the set E.
=
=
(grad <p (x), f (x) ),
0 if xE E (Rn, M ( E,
160
VIII. @1 -Liapunov Functions for Ordinary Differential Equations
Then if M is asymptotically stable, it is globally asymptotically stable,

if M is negatively asymptotically stable, it is globally negatively asymptotically stable and if M is unstable, it is globally unstable.
7. On the Use of Higher Derivatives of a Liapunov Function

In the previous chapter the stability properties of sets with respect
to the flow defined by the solutions of ordinary differential equations
7.1
ic = f (x), /: Rn-+ Rn,
have been characterized by the properties of a real-valued function v =

<p (x) and its total time derivative along the solutions of the differential
equation 7.1.
7.2
1Pi (x)
<grad <p (x), /(x)).
In this section we shall briefly summarize some recent results obtained

by various authors on the use of the total time derivative of order n of
the real-valued function v = <p (x) along the solutions of 7.1, which is
defined as follows
7.3
1JJ2 (x)
= <grad 'f/Ji (x), /(x)), ... , 1Pn (x)
<grad 1Pn-l (x), /(x)),
where I (x) E en-i and <p (x) E en.

Most of the results obtained are not strictly stability results, but they
lead to a more complete analysis of the qualitative behavior of the differential equation 7.1. This analysis is in accordance with the classification due to NEMYTSKII [13] of trajectories in the neighborhood of an
isolated singular point into hyperbolic, parabolic and elliptic sectors.
The first use of 1JJ2 (t) for the characterization of such qualitative properties seems to be due to N. P. PAPUSH [1]. The aim of his work is to
identify the type of the N emytskii classification of the solutions of 7.1
in a neighborhood of an equilibrium point by means of suitable sign
combinations of <p, 'f/Ji and 1jJ2
More recently M.B.KuDAEV [1] has derived additional results on the
behavior of the trajectories of the differential equation 7.1 in a neighborhood of an equilibrium point by suitable sign combinations of <p, 1JJ1, 1P~
and 'f/Ja
Most of the results by KuDAEV have been recently sharpened by
J. YORKE [2] whose results are stated next. Notice that these results by
YORKE have the extremely important and unique feature of having local
conditions.
7.4
Theorem. Let v = <p (x) be a real-valued function defined in Rn.
Let N ((3) be a bounded component of the set {x E Rn: <p (x) < [3}. Assume
7. On the Use of Higher Derivatives of a Liapunov Function
that
7.4.1
v = <p(x) E @2 ,
7 .4.2
for all z E oN ({J), "Pi (z)
7.4.3
there exists y E oN (/3) such that "Pt (y)
0 implies 1JJ2 (z)
>
161
0,
< 0,
7. 4. 4
N (/3) contains a compact invariant subset.
Then there exists a point z E fJN (/3) such that for the differential equation 7.1
7.4.5
y+ (z) ( N ({3).
7.o
Theorem. If in Theorem 7.4 conditions 7.4.1 and 7.4.2 are satisfied and instead of 7.4.3 and 7.4.4 we assume that
7.5.1
the set {yE oN (/3): 1Pt (y) > O} is non-empty and non-connected.
Then the set {zE Rn: y+(z) C N({J)} has dimension at least n - 1.
7.6
Theorem. Let M ( Rn be a compact invariant set and let
v = 
=
1Pi (x) =
0 for all x E Rn,
JJxJJ-+ oo,
0implies1/)2 (x)
>
0 for all x E Rn - M,
7.6.5
0 for some y E Rn - M.
Then either there exist points z1 and z2 in Rn - M, such that
+ oo as t-+ - oo,
f z2t f-+ + oo as t-+ + oo,
7.6.6
A+(z1} CM, Jz1tj-+
7.6. 7
A-(z2) CM,
or for all z E Rn - M
7.6.8
jzt I-+ oo as ftJ-+ oo.
Recently J. A.YORKE [7] proved the following theorem which provides a characterization of a certain type of compact invariant unstable
attractors relative to an open set U (Rn.
7.7
Theorem. Let v = <p(x) be a real-valued function defined in an
open set UC Rn. Let MC U be compact and such that U - M is
simply connected. Assume that
7. 7.1
v = <p (x) E @2 ,
7. 7.2
for each x E U - M, either 1Pi (x) =t= 0 or 1JJ2 (x) =t= 0,
where "Pi (x) and 1JJ2 (x) are defined by 7.2 and 7.3, respectively. Then
for each x E U,
7.7.3
A+(x) CM and A-(x) CM.
11 Bhatia/Szego, Stability Theory
162
Sketch of Proof. Let

7.7.4
{xE U - M: 1JJ1 (x)
O}.
Then E is closed relatively to U - M. Condition 7. 7. 2 implies that

grad 1Pi (x) =t= 0 for all x EE. It can be proved that Eis the disjoint union
of connected manifolds Ei, and that there exist at most a countable
number of such manifolds Ei. Since U - Mis simply connected, homology theory implies that each Mi separates U - M into components
one of which, Pi, is positively invariant and the other, Ni, is negatively
invariant. If there exists a point y E (U - M) (\A+ (x), then the image
point xt must eventually be in the same connected component of U - M
as y, for y ~ M. Thus 1Pi (xt) =t= 0 for large values oft. Then <p (xt) is monotonic and <p (x) must be constant on the invariant set A+ (x). Since <p (yt)
is constant 1Pi (y) = 1JJ2 (y) = 0 which contradicts 7. 7. 2. Similarly for
A-(x).
The idea of characterizing the stability properties of differential equations by
means of the sign properties of a real-valued function is due to LIAPUNOV [1]. A
similar idea in a much more geometrical context, quite near to our point of view
is to be found in the work of POINCARE [1, vol. 1, p. 73ff.]. Here PoINCAR:E developes
in R 2 a method which enables to prove the absence of limit cycles in certain domains
of the plane. This information is derived by analyzing the properties of the set
1P (x) = <grad <p (x) ,f(x)) = 0 (contact curve) where v = <p (x) is a "topographical
system", i.e., a real-valued function such that the curves <p (x) = const are nonintersecting, closed and differentiable.
Methods quite close to LIAPuNov's have also been suggested by M. HADAMARD
[1] and D. c. LEWIS [2].
For the investigation of the qualitative properties of ordinary differential equations methods other than that of LIAPUNOV have been suggested among which we
recall the topological method of T. WAzEwsKI [1] (see also F. ALBRECHT [1],
N. 0Nucmc [1], A. Pus [1, 3] and Z. SzMYDT).
In the literature various qualitative properties of differential equations which
are not considered in this book are labeled as "stability" properties. These are
essentially properties which have meaning only in the framework of differential
equations or of differentiable flows, therefore these are beyond the aim of this book.
Perhaps the most important of these "stability" properties is the so-called structural
stability, i.e., the invariance of the global qualitative structure of a differential
equation % = f (x) for "small" changes in the vector f (x). These problems are investigated in the works of L. MARKUS [7], M. PEIXOTO, S. SMALE and J. MosER,
where additional references can be found.
Another "stability" property for differential equations is that of "total stability" or stability under constantly acting perturbations which deals with the problem of the invariance of the local qualitative structure of a differential equation
x = f(x, t), when a continuously acting vector-valued function g(x, t) is added to
the vector f(x, t). Works on this problem were done by I. MALKIN [8], P. SEIBERT
[2, 3], STRAUSS and YORKE [3], and others.
A rather complete survey of the problems in the qualitative theory of differential equations is given by S. LEFSCHETZ [3].
163
\Ye want to emphasize that LIAPUNOV was originally interested only in the
stability properties of a given motion. He formulated this problem as follows. Let
y = g (y, t). Let yl = yl (t) be a solution of such equation. In order to investigate the
stability of yl (t), for all e > 0, we shall consider solutions y = y (t) such that
11 y 1 (t0 ) - y (t0) I\ < 'YJ and see if this implies that 11 y 1 (t) - y (t) 11 < e for all t > t0
This can be easily done by defining the new variable x = y - y 1 (t). Then from the
differential equation y = g (y, t) one can obtain a new differential equation % =
f (x, t) = g (x yl (t), t) - g (y1 (t), t). This equation is called equation of the perturbed
motion. Notice that the stability problem for the motion y 1 = y 1 (t) for the equation
y = g (y, t) is now reduced to the stability problem for the equilibrium point x = 0
of the equation% = f(x, t).
From the theorems shown in this chapter, it can be seen that the central point
of the analysis of the stability properties of a compact set is the construction of
continuously differentiable, real-valued function v = <p (x) such that the corresponding total time derivative (1.4) along the solutions of the differential equation 1.1
is such that 'tJl (x) = 0 if xE Mand 'tJl (x) ::f= 0 if xEf M. Only in this case the theorems
of section 3 provide a complete classification, since, if <p (x) is definite for M and
sign <p (x) =f= sign 'tJl (x), Mis asymptotically stable, if <p (x) is definite and sign <p (x) =
sign 'tJl (x), 1VI is negatively asymptotically stable, if <p (x) is indefinite then M is
(not completely) unstable; <p (x) cannot be semidefinite for M. In addition if 'tJl (x)
is definite then the extension theorems of section 4 can be applied. For the construction of such a pair of functions <p (x) and 'tJl (x), equation 1.4 can be regarded in different ways according to which are the unknowns and which the given function in the
relationship 1.4. For instance if <p (x) and/(x) are given, then clearly 'tJl (x) = d<p (xt)/dt,
on the other hand if 'tJl (x) and /(x) are given, then they give rise first to an equation of the type 'tJl (x) = <a (x), f(x)) where a (x) is an unknown vector. Then if
'tJl (x) and/(x) are such that a (x) satisfies the integrability condition 8aj/Oxi = oai/oxj,
then there exists a real-valued function <p (x), such that the relation 1.4 can be
satisfied for the given 'tJl (x) and /(x). One can also require 'tJl (x) to have a special
form. For instance we can assume that 'tjJ(x) = B(x){3(<p(x)), then find v = <p(x}
which satisfies the equation B {x) f3 (<p (x)) = <grad <p (x), f (x)) under the assumption
00
that the integral
fJ(:)
exists. The above given equation was introduced by
G. P. SzEGo [3] and it is a generalization of the one introduced by V. I. ZuBov [1],

given in V, 2.16.8.
These two approaches: that of choosing <p (x) (or at least its form) and of fixing
'tJl (x) first, represent two different ways of constructing Liapunov functions.
For the problem of constructing Liapunov functions the reader is referred to
the books by W. HAHN [1, 2] where additional references are given.
Section 2. Most of the geometric ideas used in section 2 were introduced by
A. STRAUSS [5], who developed some earlier results of J.P. LASALLE [4].
Property 2.3 can also be proved by the same technique used in the proof of
Theorem 5.6.
Section 3. All results presented in section 3 are natural extensions of the results
of LIAPUNOV to the particular case of compact sets. The theory presented here follows
in spirit the one of A. STRAUSS, even if some of the results are different. Theorem 3.9
on asymptotic stability is not formulated in the traditional way, but it is equivalent to the classical theorem. However, the proof of the classical theorem on asymptotic stability (see, for instance, BHATIA and SzEGo [1], Theorem 3.6.15) is much
more involved than that of Theorem 3.9. This is due to the fact that in the classical
theorem it was essentially shown that Mis a uniform attractor and the number T
11*
164
VIII. (21-Liapunov Functions for Ordinary Differential Equations
appearing in V, 1.29.2 was explicitly computed. Thus in that proof one was able
to prove more properties about the system than the ones essentially needed.
Notice that in the local theorems 3.9, and Corollary 3.13 estimation of the region
of attraction A (M) requires that one has to make sure that N (/3) is compact. Thus
3.13.2 essentially expresses the classical condition "fJN is the level surface of the
Liapunov function v = <p (x) which is bounded and fJN (\ (E - M) =F 0, while
N - M (\ E = 0, where E = {xE R": 'tJl (x) = O}." The condition of Theorem 3.15 on
global asymptotic stability is due to E. A. BARBASHIN and N. N. KRASOVSKII [1].
The instability theorem 3.17 is an extension of the well known theorem due to
N. CHETAEV [1, 3]. Recently J. A. YORKE [4] proved the following theorem which
is another extension of CHETAEV's result, which we state for the case of a dynamical
system defined by the differential equation %= f(x), /: U ~ R" and U ( R"
open.
Theorem. Let G C Ube closed. Let G(v) = {xE G: <p(x) = v and 'tJJ(X) = O} and
G+ = {x E G: O}. Assume that (i) G is positively invariant; (ii) {O} E G-+,
 0 for x E G; (iii) for v > 0, G (v) has no (non-empty) compact
invariant subsets. Then {O} is unstable.
A similar result was also proved by N. P. BHATIA [9].
Almost all results proved in sections 1- 3 are proved in the more general
framework of the theory of flows without uniqueness in the monograph by G. P.
SzEGO and G. TRECCANI [1].
Section 4. Preliminary ideas leading to the extension theorems can be found in the
works of D.R. INGWERSON [1] and W. LEIGHTON [1]. A complete preliminary statement was given by G. P. SzEGo [4]. Its complete proof for the case of global asymptotic stability is due to N. P. BHATIA and G. P. SzEGo [2].
The complete global extension theorem 4.3 is due to G. P. SzEGO [5].
There is a basic difference between the application of Theorem 3.9 and Corollary 3.13 on the one hand and that of Theorem 4.1 on the other. Namely in order to
use Theorem 3.9 for an estimate of A (M) one had to prove that for a certain {3,
N (/3) is compact, while in Theorem 4.1 this operation does not have to be performed
at all and in addition N ({3) may well be not compact. Clearly however, in this case,
one has to make sure (see condition 4.1.5) that in the open set N (/36) - M there do
not exist points in which 1Jl (x) = 0 nor diverging sequences along which 'tJl (x) ~ 0.
The use of the extension theorem as compared with the classical asymptotic
stability theorem has been shown in various papers (see, for instance, G. P. SzEGO
[7], G. P. SzEGO, C. OLECH and A. CELLINA [1], G. P. SzEGo, G. ARIENTI and
C. SuTTI [1]). These theorems are useful for instance when one is interested in the
stability properties of the rest point x = 0 of an ordinary differential equation
== g(x), g(O) = 0, g: R" ~ R", which admits a non-singular linear approximation
in the neighborhood of the point x = 0, i.e., such that the Jacobian matrix ] =
{ogi/ox;} evaluated at the point x = 0 (and therefore with constant elements) is
nonsingular. In this case the local stability properties of the rest point x = 0 coincide
with those of the rest point x = 0 of the linear differential equation with constant
coefficients: = ] x. The stability properties of equation 4.4.3 are easily analyzed
by algebraic criteria (ROUTH, HuRWITZ, etc., see e.g., GANTMACHER [1]). In this
situation the extension theorems allow one to solve problems which are not practically solvable within the classical theory, for instance when one is able to construct
a continuously differentiable Liapunov function vanishing at the rest poiAt x = 0,
which is locally asymptotically stable, and such that 'tJl (x) satisfies conditions 4.1.5
or 4.3.5. On the other hand <p (x) may have a structure making it very difficult to
analyze its sign properties (for instance it is a polynomial of order higher than two),
165
as required in the classical theory. The extension theorems are proved for the case
of differential equations without uniqueness in the monograph by G. P. SzEGo and
G. TRECCANI [1].
Section 5. By applying the extension theorems for the case of differential equations
without uniqueness (G. P. SzEGo and G. TRECCANI [1]) to the study of the geometrical problems presented in section 5, all those results can be proved in a stronger
form, namely under the assumption that <p (x) E ~1 , but not necessarily E ~2. Similar
results for functions of the class ~ 1 , derived from Theorem 5.9 are proved for such
functions and can be found in the monograph by N. P. BHATIA and G. P. SzEGo [1].
The technique used in the proof of Theorem 5.9 can also be applied in the
proof of relationships between the properties of the set N' ({J) and those of the
vector grad <p (x). Most of the relationships are esseniially generalization to Rn
of the well known Rolle's theorem. A first result along the line was proved
byG. P. SzEGO [8]. Additional results are proved byG. P.SzEGo and G. TRECCANI [1].
Section 6. All results of Theorem 6.1 can be found in the monograph of N. P. BHATIA and G. P. SzEGo [1] and contain as special cases previous results by E. A. BARBASHIN and N. N. KRAsovsKII [1] and by J.P. LASALLE [4, 6, 7]. A weaker version of Theorem 6.1 was proved for flows without uniqueness by G. P. SzEGO and
G. TRECCANI [1] where also theorems 6.3, 6.4 and 6.5 are proved for the case of
functions <p (x) E ~1.
Finally, we would like to point out that the problem of the existence of Liapunov functions for differential equations (converse problem) has been discussed
by many authors, notably J. J. MASSERA [5, 6], N. N. KRASOVSKII [3, 6, 7, 8, 9],
K. P. PERSIDSKII [2], I. VRKOC [1] and J. KURZWEIL [1, 2] and J. KuRZWEIL and
I. VRKOC [1].
Some results on the existence of ~ 00 -Liapunov functions for asymptotically
stable rest points were then proved by ZuBov [1].
Results on the existence of ~ -Liapunov functions for compact asymptotically
stable sets are given by F. w. WILSON [1]. J. L. MASSERA [2] and J. KURZWEIL
simultaneously proved the existence of a <tf -Liapunov function for the uniformly
asymptotically stable rest point of the differential equation x = f(x, t). The method
used in the proof is quite different from the method of ZuBov. T. YOSHIZAWA [1]
gives simpler converse theorems, but is able to prove only the existence of Lipschitz
continuous Liapunov functions.
00
00
Section 7. Most of the results of this section are due to J. A. YORKE [ 1, 7]. The
reader may consult the original papers for the complete proofs.
Chapter IX
Non-continuous Liapunov Functions for

Ordinary Differential Equations
1. Introduction
Throughout this chapter we shall assume that the ordinary differential equation
1.1
X = f (x), x E Rn,
where f: Rn-+ Rn is continuous, has unique solutions in the space Rn.

This chapter is devoted to the characterization of the (weak) attraction properties of an isolated rest point x = 0 of the differential equation 1.1.
As it must be clear from the results of the previous section, there
need not exist any continuous Liapunov v = <p (x) function with a continuous total time derivative along the solutions of the differential equation 1.1 in the case in which the rest point x = 0 of equation 1.1 is
an unstable (weak) attractor. The characterization of weak attractors
can however be done by means of lower-semicontinuous real-valued functions and their lower-right-hand-side Dini derivates along the solutions
of equation 1.1. By means of these functions we are able to give a characterization of (weak) attraction which follows exactly the theory
(extension theorems) presented in the previous section for the case of
asymptotically stable sets.
Throughout this section, for a mapping <p: Rn -+ R we shall write
1.2
lim inf <p (x) =Jim [inf {<p (x): x ES (x0 , e)}J.
X-+X 8
-+0
In this section we shall often use lower and upper-semicontinuous

real-valued functions on a set U (Rn. If UC Rn, a function q;: U-+ R is
lower-semicontinuous if and only if
1.3
lim inf 
<p (x)
for all x E U,
167
1. Introduction
and upper-semicontinuous if and only if

1.4
lim sup <p (y)

:v-+x
< <p (x)
for all x E U.
If <p: Rn-+ Risa lower-semicontinuous real-valued function and f: Rn-+ Rn

is a continuous function, we shall consider the following extended
real-valued function
1.5
D+ (x t) lt=O = lim inf i--1 [<p (xi-) - <p (x)],

T~O+
which is the lower-right-hand-side Dini derivative of the function <p (x)

along the solutions of the differential equation 1.1.
The stability results of the previous chapter were all essentially
based upon the relationships VIII, 1.3 and VIII, 1.5 which connect the
real-valued function 1/'(x) with the function <p' (x)
dqij;t) It=o
Notice
that while for the estimation of 1P (x) no knowledge about the trajectories
of the system is required, the estimation of <p' (x) requires the knowledge
of the trajectories.
Similarly for the case of lower-semicontinuous real-valued functions
one needs a connection between D+ (x t) lt=O and the real-valued function,
defined below in 1. 6.1, for the evaluation of which no knowledge of the
trajectories is required. This connection is provided by the following
theorem due to YORKE [3].
1.6
Theorem. Let v = <p (x), <p: U-+ R be a lower-semicontinuous
real-valued function defined on an open set UC Rn. Assume that there
exists a real-valued function w = ~(x), ~: U-+ Rn, which is defined and
continuous in U and such that the extended real-valued function
1. 6.1
1P (x) = <p* (x) = lim

inf i--1 [ (x)]
z~f(~)
lim inf i--1 [<p (x
T~O+
+ i-f (x) + i-y) -
<p (x)]
Jfyjj-+O
satisfies the condition

1.6.2
<p* (x)
<
~ (x)
for all x E U.
Then
t
1. 6. 3
<p (xt) - <p (x)
< J~(xi-) d-r:

0
for all x E U and t E R, such that x [O, t]

1.6.4
<p* (x)
< lim inf i--1 [ 0. Then there is an integer N such that for
n > N, the inequality
1.6.5
n-1 (
kt) t
+oJ~t (xs)ds >I~
x- k=O
n n
holds.
From 1.6.2, for each integer n > N and k with 0 < k < n - 1, there
exist h(n,k)' Y(n,k)' satisfying 0 < h(n,k) < t/n, IY(n,k) I < t/n, such that
1.6.6
kt )
~ ( x-;;:
>
-1 [
h{n,k)
kt
kt )
cp ( x-.;;:
+ h(n,k) f ( x-;-
+ h(n,k> Y (n,k) )
-ip(x:)J-e:.
Since <p is lower-semicontinuous, and f is uniformly bounded on the
compact trajectory segment x [O, t], we may further choose hcn,k) and
Y(n,k) such that the first term on the right hand side is either non-negative
(this is possible at points x (kt/n) where <pis not a maximum), or (where -
hcn,k)-;;-
holds. With such a choice, 1.6.5 yields on using 1.6.6, the inequality
t
1.6.7
e+f~(xs}ds>-=-B
0
> - n - e + <p (xt) -
<p (x)
n[(k-1
+I
cp x - - t + h(n,k-l) I (k-1
x --- t
k=l
n
n
+ h(n,k-t) Y(n,k-1))- 'P ( X :)J
In the last sum, the terms are either non-negative, or no less than
h(n,k-l) e/n. Therefore we have
t
e + ~ (xs) ds > - 3e
+ 0 was arbitrary, the inequality 1.6.3 holds. This proves the
theorem.
2. A Characterization of Weak Attractors
169

In this section we shall apply Theorem 1.6 for providing a complete
characterization of (unstable) (weak) attractors. The first theorem (2.1)
provides a sufficient condition for a non-empty compact set M C Rn to
be an attractor. Under an additional hypothesis on the systems this
condition is also necessary (Remark 2.2.3). Theorems 2.4 and 2.5 on
the other hand provide necessary and sufficient conditions for a nonempty compact set M ( Rn to be a global weak attractor and an attractor, respectively.
2.1
Theorem. Consider the dynamical system represented by the
X = f (x), XE Rn, f: Rn-+ Rn.
2.1.1
Let M ( Rn be a compact set and let v = <p (x), <p: @(M) -+ R be a lowersemicontinuous real-valued function. Assume that
2.1. 2
there exists a continuous real-valued function w = ~ (x), ~:
@ (M) -+ R- such that the extended real-valued function <p* (x) (1.6.1)
satisfies the condition <p* (x) < E(x) < 0 for all x E @(M) .
For each v ER, let K' (v) = {x: <p (x) < v}, and assume that
either K' (v) V Mis compact or is all of Rn.
Then for each x E Rn, either
2.1.4
X't'n EM for some sequence Tn-+
oo,
or
2.1.5
e(xt, M)-+ 0 as t-+ oo.
Proof. Let 2.1.4 not hold for an xE Rn; that is, for some LX0 , xtE @(M)
fort E [X 0 , + oo). Let 5 1 = K'( X0 , fort> X1 , xtE 5 1
and 5 1 V M is compact. It suffices to prove that A+ (xt) (which is
non-empty since 5 1 V Mis compact) is a subset of M. Choose yEA+(xt).
Suppose y($ M. Then there exist 'Y and " such that e(y, M) > 2y > 0
and on the set B = 5 [y, 2y ], <p* < E< -" < 0. Let = max Ill
B
Then for any t
2.1.6
(} (xt - y) < 'Y =::} x(t + i-) EB for f-r( < y/.
Choose ti-+
oo such that X 0 < t1 and ti+l >ti + y/ and (xti - y f
< y. Since <p is lower-semicontinuous and <p (xt) is monotonically
decreasing for all i,
2.1. 7
<p (y) <Ii~ inf <p (xti) < <p (xtJ
2.1.3
i-+oo
But
i tj+y/
ti+1
<p (xti+i) - <p (xt1)
<
J ~(xi-) di-< :E tjJE(xi-) d-r: <
as i-+ oo. Hence,
Ii~
inf <p (xti)
i~oo
-y "if-+ - oo
j=1
t1
= -
oo, contradicting 2.1. 7. Therefore
y E M, and A+ (xt) C M, proving 2.1.5.
170
2.2
Remarks.
2.2.1
Notice that if Mis positively invariant, the conclusion is that
Mis a global attractor.
2.2.2
In many examples of attractors, such a <p exists. For instance,
if we make the additional hypothesis that for each y ~ M the arc length
of the half-trajectory y+(y) is finite, then a function <p as described below
00
exists, letting <p (y)
JldJ: Id7:, the arc length. Then
<p is a function as
described below with

<p* (y)
<
d
dt <p (yt)
lt=O = - If (y) I = ~ (y).
2.2.3
Notice that 2.1.1 implies that K' (v) is closed relative to @(M},
i.e., the limit points of K' (v) in @(M} are in K' (v); therefore, 2.1.3 may be
guaranteed, for example, by assuming lim inf <p (y)
y_.,.M
for all x E @ (M). If <p was assumed to be defined and continuous on Rnand
constant on M, and if M was invariant, then M would be stable.
2.2.5
The assumption 2.1.2 is similar to saying <p* is negative definite.
The assumption "cp* (y) < O", however, does not imply that <p (xt) is not
constant since cp* is not necessarily a continuous function. (Examples can
be given.)
2.3
Example of an Attractor. Let dimension n = 2, and let x be in
the usual polar coordinates (e, fJ). Let M be {(O, 0), (1, O)}. Then M is
invariant for
2.3.1
JO= (2n -
O)
o + max{O, 1- e},
le= -11 - el
The phase portrait is given in Fig. 2.3.2.
(1.0)
--Fig.2.3.2
171
Define 1.
-log (1 - e)
2.3.3
 O) = { _ 0 + e
Then
2.3.4
J-1
<p*(e. O) =
l-0- 11 -
el
fore< 1,
fore> 1.
Then if we assume ~(O, e) == max{-1, -0 -11- el}, we have

<p* < ~ < 0 on @(M). Fore> 1, <p(e, 0) > e - 2n and <p((>, 0) ~ oo as
jxj = e ~ oo. Checking <p fore = 1 in particular, we see <pis lower-semicontinuous. Thus M is an attractor.
2.4
Theo:rem. Consider the differential equation 2.1.1. Let M ( Rn
be a compact set. The set M is a global weak attractor if and only if the
following conditions are satisfied.
For each c5 > 0, there exists a compact set M 6 , with MC Md
2.4.1
( S [M, c5], a lower-semicontinuous real-valued function v = </Jd (x),
q;6 : @(M6 )~R and a continuous real-valued function w == ~6 (x),
~ 6 : @(M6) ~ ( - oo, 0) such that
2.4.2
<p6 * (x)
< ~6 (x) < 0 for all x E @(M6).

Let K~(v) = {x: q;6 (x) < v}. For each real v and
2.4.3
c5K~(v) V M~
is compact or is all of Rn.
Proof. Assume 2.4.1 holds. Let M in Theorem 2.1 be M 6 in Theorem 2.4; for each y there exists a sequence i-n ~ CXJ such that y-rn ~ M 6
(whether 2.1.4 or 2.1.5 is satisfied). It follows that letting c5 = 1/2n there
exists tn > n such that e(ytn, M 6 ) < 1/2n. Hence, ytn ~ M as n ~ oo.
Therefore Mis a weak attractor.
To show the converse, assumeM is a weak attractor. Let M 6 == S [M, <5]
for all c5 > 0. For yE @(M6} let <p6 (y) =inf {t > 0: ytE Md} To see
that <p6 is lower-semicontinuous, fix y and let v E [O, <pd (y)). Then
inf e (yt, M) > c5 and for some neighborhood N of y, for all z EN,
tE[O,v]
zt ~ S [M, c5] for t E [O, v]. Hence, <p6 (z)

semicontinuous. Also note that
2.4.4
dtpi{t)
Lo=
>
v for z EN, and <pd is lower-
-1 for yE @(M.,).
So 1.6.3 is satisfied letting ~ == -1 =

1.6.2, <p* < -1 == ~ on @(M6 ).
~6
Since 1.6.3 is equivalent to
2.5
Theorem. Consider the differential equation 2.1.1. Let M ( Rn
be a compact invariant set. The set M is a global attractor if and
only if 2.4.1 holds requiring in addition that, for each c5 > 0, the sets
Md can be chosen to be invariant.
172
Proof. From Theorem 2.1, if each M 6 is invariant and 2.4.1 holds,

each M 6 is an attractor. It follows that M is an attractor. To see the
converse, let Md be the largest invariant set in S [M, c5]. Since the union
of invariant sets is invariant, M 6 exists (perhaps = M), and since the
closure of an invariant set is invariant, Md is closed. Since Mis an attractor, we may define
= sup {t: fl (yt, M) > <5} for y E @(M6).

Note that if "e(yt, M) < c5 for all t", then y(y) V Md would be an in<p6 (y)
variant set in S [M, c5], so <p6 (y) is defined for ally E @(M6).
To see <p6 is lower-semicontinuous, note that if e (yt, M) > <5 for some
t and y, then for all z in some neighborhood N of y, e (zt, M) > c5, and
<p6(z) > t, so lim inf <p6 (z) > <pd (y). As in the previous proof, 2.4.4 holds
and <p*
2.6
<
z~
-1 =
~6
on @(M6).
Remarks.
2.6.1
For all <5 > 0, we may have Md= M. If in 2.4.1 the largest
invariant subset of S [M, c5] isM, then Md = M. If Md = M for some
c50 > 0, then we may let M 6 = M 60 and <p~ = 'Pdo for all c5.
0
2.6.2
Both 2.4.1 and 2.4.3 use families of Liapunov functions; if Mis
invariant in 2.4.1 and is not a global attractor, then it can be seen from
Theorem 2.1 that Md =f: M; if Md= M for some <5, then 2.2.1 implies
xt ~ M if t ~ + oo for all x and x E R!", and M is an attractor.
3. Piecewise Differentiable Liapunov Functions

In this section we shall assume that the ordinary differential equation
3.1
x = /(x),
XE Rn,
/(0) = 0,
where/: Rn~ Rn is continuous, defines a dynamical system (Rn, R, n)

and that the rest point x = 0 is isolated.
In this section we shall use a rather restrictive class of lower-semicontinuous real-valued functions v = <p (x), <p: Rn~ Rn, namely, piecewise differentiable functions. On these functions and on their corresponding functions 'P (x) (1. 6.1) we shall impose additional conditions.
These conditions will allow us to prove the extension theorems and
to construct a special type of Liapunov function on the region of instability (V, 1.13) of an unstable attracting rest point (section 6). In addition to this application the introduction of this special kind of lowersemicontinuous function has been suggested by problems arising in the
numerical construction of Liapunov functions where piecewise differentiable Liapunov functions are simpler to construct than higher order
polynomials.
173
We shall begin this presentation by defining a few properties of the

real-valued function <p.
3.2
Property. Let v = <p (x), <p: U -+ Rn, where U C Rn is an open
set with {O} ( U, be lower-semicontinuous. We shall say that <p(x) has
the property 3.2 with respect to the dynamical system 3.1 if and only if
the function 'I' (x) (1. 6.1) satisfies 'I' (0) = 0 and for each e > 0 there
exists a real-valued function 'e (), lXe (0) = 0, lXe () > 0 for > 0,
which is non-decreasing and continuous on [e, + oo) and for each x E U
3.2.1
1P (x)
<
-tX6
(llxll).
3.3
Property. Let v = <p (x), <p: U -+ Rn, where U ( Rn is an open
set, be lower-semicontinuous. Let DC Ube the set of all points in which
 0 and a real-valued function rux 6 (y), which is lowersemicontinuous in the sphere S (x, ~),and such that for all y ES (x, ~)
3.3.1
w:,d (y) =li~!1f i--1 [wx,d (y
+ i-z)
- mx,d (y)]
<
-e, e
>
8~/(y)
and moreover
3.3.2
z ED* f\ S (x, ~), z
=f: x implies <p (z)
= <p (x).
The following example shows that there exist lower-semicontinuous

functions and differential equations for which the property 3.3 is satisfied.
3.4
Example. Consider the following planar system of ordinary
differential equations, described in polar coordinates:
.;. = - r sin 0,
3.4.1
{ ()
= r,
where r and (} are the radial and angular coordinates respectively.

Consider for each point (r0 , 0) r 0 =f: 0, the neighborhood
3.4.2
S ={fr - r0 f <a;
2n - e <
() <
2n:; 0
< (} < e}.
On each set S define the function

3.4.3
w (r, fJ)
-fJ
= { e-8
for 0 < fJ < e,

for 2n - e < (}
<
2n.
The real-valued function 3.4.3 is lower-semicontinuous and has the value

0 for fJ = 0, fr - r0 1 <a. For each point of the set S (3.4.2), in which
e > (} > 0, the function w (r, 0) is continuously differentiable with
ru = <grad w, I>= -r, while for each point of the set s in which
174
() <
2n, we have
w=
-re- 0 Consider then a point in the set S (3.4.2)
in which () = 0, i.e. the point (r, 0) ES. Then

3.4.4
w* (r, 0)
lim inf i--1 [w (r

T~O+
+ i-r, i-r + i-0)]
-r~o
Now for a given e > 0,
if'<
,..,
hence i--1 ru (r, i-r

3.4.5
+ i-0) =
w* (r, 0) = - (r
(r
r:+
2n, we have 0
< T(r + 0) <
e and
,..,
0) . Thus
+ 2n).
In summary for each point (r, 0) ES
< -e- 2nr = -k1 , k1 > 0,

while for each point of the axis 0 = 0
3.4. 7
w (r, 0) = 0.
3.4.6
w* (r, 0)
Thus each lower-semicontinuous real-valued function which is continuously differentiable on the whole plane (r, 0), except on the axis 0 = 0,
has the property 3. 3 for the system 3.4.1.
Notice that in this example the condition on ru* holds indeed only
locally (for each X < e/r + 2n) and not globally on the axis 0 = 0.
From the definitions given the following proposition can be easily
derived.
3.5
Proposition. If v = <p (x) is a lower-semicontinuous real-valued
function which has the property 3.3 in the set U (Rn, then for each
x E Uthe set y (x) (\ D* is a set of isolated points.
We shall prove next that if a real-valued function has the property
3.3, then some properties of the function which hold in the set U - D*,
hold for the whole set U.
The following theorem is a special case of Theorem 1.6. We shall
however give a complete proof, since under the stronger assumptions
made on the function <p (x) its proof is quite simple.
Theorem. Let v = <p(x), <p: U ~Rn, where U (Rn is an open
3.6
set with {O} C U, be lower-semicontinuous and have the properties 3.2
and 3.3 with respect to the dynamical system 3.1, while for each x E D*,
1P (x) (1. 6.1) is finite. Then
t
3.6.1
<p(xt) -
<p(x)
< -
JlX8 {fl~_T[f} di-
or all xE U and tE R+ such that x[O, t] ( U - S(O, e).
175
Proof. Clearly in the interval [O, t] there exists (Proposition 3.5) at

the most a finite number of points Tk such that XTk ED*. For each XTk
there exists then a real number Gk (7:k) such that
{x(Tk -
ak,
Tk) V
+ ak)} (\ D* == 0.
x(Tk, Tk
Let (} = min ak and K = max 1P (xTk). Then from the inequality 3. 2. 1

[O,t]
[0,t]
<i;{ T~dT}+ "{ TJ1~ 1X8 (11x!i)dT}

6
3.6.2
<p(xt)-<p(x)
k=i Tk-9
k=i
1:k+8
1:1-8
+ J -<Xs (!!xi DdT + J -<Xs (j !xi DdT

Tn +fJ
Now by
letting()~
o+ from
3.6.2 we obtain
t
<p (xt) -
<p (x)
< -
<X8
(l!xT! j) dT
for all x [O, t] ( U - S (0, e), which proves the theorem.

All the results above suggest the following definition:
3.7
Definition. Let U (Rn be an open set such that 0 E U. Consider
the dynamical system defined by the differential equation 3.1. Let
v = <p (x) : <p = U ~ R be a lower-semicontinuous real-valued function.
Let D ( U be the set on which <p (x) is not continuously differentiable.
Let D* = D - {0}. We shall say that <p (x) has the property 3. 7, if
and only if <p (x) has the properties 3.2 and 3.3 and in addition, 1P (x) is
finite for each x E D*.
We shall say that the extended real-valued function w = 1P(x) has the
property 3. 7 whenever it has the properties stated above. A similar
definition can be given for upper-semicontinuous real-valued functions.
3.8
Remark. It is well known that for the case of a continuous realvalued function v == <p(x), <p: U 4 R, U (Rn, OE U, the condition:
3.8.1
there exists a continuous, nondecreasing real-valued function
<X (), lX: R ~ R, <X (0) = 0 and <X () > 0 for > 0, and such that
<p(x)
< -<X(ffx//) foreachxE
U,
is equivalent to the condition:

3.8.2
<p (x)
<
0 for x E U - {O} and if {xn} ( U, <p (xn)
0 implies
xn ~ {O}.
This equivalence is not generally true for semicontinuous functions.
176
4. Local Results
We shall next apply the results of the previous section to the study
of some local properties of the flow in the neighborhood of the rest point
x == 0 of the dynamical system 3.1. Here we shall give some results on
properties which cannot be characterized by continuously differentiable
Liapunov functions.
4.1
Theorem. Let U C Rn be an open neighborhood of the rest point
x == 0, and U be compact. Let v == <p (x) have the property 3. 7. Then the
set U - {O} does not contain any recurrent trajectory.
Proof. Let 'Y (x)
CU
- {O} be a recurrent trajectory. Then 'Y (x) is a
compact minimal set. Thus e[y(x), {O}J = <1 > 0. We can choose Xa()
such that 3.2.1 holds. For Theorem 3.6 for each t > 0:
t
4.1.1
<p(xt) - <p(x)
< -
X0
(j[xrj[) dr < -
JX
(a) d-r = -Xa(a)
t.
On the other hand if 'Y (x) is recurrent then x EA+ (x), hence there exists a
sequence {tn} ( R+, tn ~ +ex:>, such that xtn--+ x. Thus from 4.1.1
4.1.2
 for n--+ - ex:> which implies that v == <p (x) is not lowersemicontinuous in the point x E U and violates the hypothesis.
4.2
Theorem. Let v == <p (x), <p: U--+ R be a lower-semicontinuous
real-valued function defined on an open set U (Rn, with 0 E U, and such
that
4.2.1
the extended real-valued function 1P (x) (1.6.1) has the property

3. 7 on the set U,
4.2.2
 O} is either compact or is U.
>
0 for x E U - {O},
Then {O} is an attractor and A ({O}) ) U.

Proof. Let yEU-{O}and assumeA+(y)==0. It must be that <p(y)
 <p (y).
It follows that {xE U: <p(x) < <p(y)} =FU and so it is compact. Then
A+(y) =F 0 and compact. We shall prove that A+(y) = {O}. By Theorem
4.1, {O} (A+ (y).
Assume z E A+(y), z =F {O}. There is then a sequence {tn} ( R+, tn--+
+ex:> such that ytn--+ z. Clearly we can construct two sequences, {-rn}
177
and {sn} such that yrn E H(z, ! !lzlj) and ysnE H(z, i llzll) and for
each n, Sn< 'l'n < tn.
We may also assume that for sn < t < 'l'n,ytE S(z, i llzll) and that
rn - sn > <J > 0 for each n. From the inequality 3.6.1 it follows that
n
<p (ytn) - <p (y)
<
~( <p (yri)
-- <p (ysi))
j=1
n ~
< - ~ f Xillzll (if lzll) dr =

j=1
< -
- Xiflzll
si
(i Ijzj j)
~ (ri - si)
i=1
nXillzll (i
Ilzl [) <J
Then for sufficiently large n, 'P (ytn) < t <p (z) and so <p (x) cannot be
lower-semicontinuous in zE U. \Vehaveproved that for each yE U,A+(y)
= {O} and so {O} is an attractor and its region of attraction contains U.
We shall prove first the global extension theorem, i.e., a theorem
which allows us to extend to the whole space the local stability properties of the rest point x == 0, which are assumed to be known.
5.1
Theorem. Let v == 
<p (x),
xn~x
5.1.2
<p(O)
== 0,
5.1. 3
the extended real-valued function 1P (x) (1.6.1) has the property
3.7 on Rn.
Then if {O} is a positive attractor, it is a positive global attractor.
Proof. Assume that the rest point x == 0 is a positive attractor and A

is its region of attraction. Let Xo E oA since oA is invariant, 'Y (xo) oA.
Now
5.1.4
inf {llYll: y E y (x0 )}
== k >
0.
As 1P (x) has the property 3. 7, we can choose
Xk
() such that
5.1.5
 0, such that x 0i- E oA and l', fr' - l' I < e, e > 0 sufficiently
12
178
small, <p (x) is continuous in x -,:' and 
<
0. Since x -,:' E oA,
0.
such that ztn---+ {O}. We have

<p (0),
which contradicts the assumption 5.1.1 and proves that oA

A =Rn.
== 0.
Thus
5.2
Remark. The most interesting case in which Theorem 5.1 can be
applied is the case of an unstable attractor. This is a case in which it is not
possible to construct a continuous Liapunov function with sign-definite
total time derivative. On the other hand it is possible to construct a
lower-semicontinuous real-valued function v == <p (x) which satisfies the
hypotheses of Theorem 5.1. It is to be noticed that such a real-valued
function must be discontinuous at the point x == 0.
A theorem analogous to Theorem 5.1 can be proved also for uppersemicontinuous real-valued function v = <p (x). I ts proof is left as an
exercise to the reader.
Theorem. Let v == <p (x), <p: Rn---+ R be a real-valued function,

5.3
which satisfies the following conditions:
5. 3.1
== <p (x)
5.3.2
<p (0)
is upper-semicontinuous,
== 0,
5. 3. 3
the extended real-valued function 1P (x) defined in 1. 6.1 has
the property 3. 7 on Rn.
Then if {O} is a negative attractor, it is a negative global attractor.
We shall proceed next with the proof of the local extension theorem.
Its proof requires the sets introduced in Definition VIII, 2.1. We therefore apply Definition VIII, 2.1 to the case of a lower-semicontinuous
real-valued function v == <p(x), and to the case in which M == {O}. In
this case, in general, the properties VIII, 2.2 do not hold, instead we
have
Properties. Let v == 
== <p (x)
is bounded in a neighborhood of {O}, there exists

0, such that N ((3) contains an open neighborhood of {O}.
If v
6. Non-continuous Liapunov Functions on the Region of Weak Attraction
5.4.2
If v = 0 such that {O} C oN(f3).
== 0,
179
then there
< f3 for x E oN ({3).
5.4.3
<p (x)
5.4.4
0 ({3) is open, while N ({3) may not be open.
We are now in the position of proving the local extension theorem for
attracting rest points.
5.5
Theorem. Consider the differential equation 3.1 which defines a
dynamical system. Let x = 0 be an isolated rest point which is an attractor. Let v = 0 such that the extended realvalued function, w = 1P (x), defined in 1. 6.1 has the property 3. 7 in
N ({3).
Then A{{O})) N({J).
Proof. Assume if possible that x0 E (oA({O}) f\ N(f30)). From the

hypotheses made, the set N ({3) f\ oA ({O}) is positively invariant, hence
y+(xO) ( [N({J0) f\ oA({O})J. Now by proceeding as in the proof of Theorem 5.1, we reach a contradiction which proves oA ({O}) f\ N ((JO) = 0
and hence N({J) C A({O}).
6. Non-continuous Liapunov Functions on the Region of Weak

Attraction
The functions used in the previous section in the case of an attractor
were constructed in a whole neighborhood of the attractor, even in the
case of an unstable attractor. In this latter case one can obtain deeper
insight in the structure of the flow by constructing a Liapunov function
on the region of instability (V, 1.13).
The proof of the main theorems rests upon the following Lemma
whose proof is similar to that of Theorem V, 2.16 and it is left as an
exercise to the reader.
6.1
Lemma. Let the rest point x = 0 of the dynamical system
defined by the ordinary differential equation 3.1 be asymptotically stable
relative to a closed, positively invariant set UC Rn with {O} C U. Then
there exist in U two continuous,. real-valued functions 0 for llxf J > 0,
6.1.3

6.1.1
0,
0,
I = (} (x) (1 + <p (x)).
dq; (xt)
dt t=O
6.2
Corollary. Let all the conditions of Lemma 6.1 be satisfied,
but instead of asymptotically stable assume that {O} is negatively asymptotically stable relative to U. Then there exist on U two continuous,
real-valued functions (x)).
The results 6.1 and 6.2 will be applied next to the construction of
a Liapunov function v = <p (x) on the region of instability I ({O}) =
JJ+ ({O}) - {O} of the rest point x == 0 of the ordinary differential equation 3.1. This rest point will be assumed to be an unstable weak
attractor. We recall that in this case the prolongation of {O}, .D+({O})
is compact.
It is desired that such a Liapunov function has the following
6.3
Properties.
6.3.1
I ({O})--+ R is lower-semicontinuous
<p(x), <p:
on the set
6.3.2
I= /({O})
6.3.3
For
each~>
<p* (x)
D+({O}) - {O}.
0 there exists e > 0 such that
lim inf [<p (x

T~O
+ -ry)
- <p (x)] r-1
<-
y~f(xY
for all x E I with [xj
<
~-
6.3.4
If D C I ({O}) is the set of all points in I ({O}) in which <p (x)
is either not continuous or does not have a continuous total time
derivative along the trajectories of the system, then the set D - {O}
is intersected by each trajectory on a set of isolated points.
The following theorem gives a sufficient condition for a Liapunov
function to have on the set I ({O}) the properties 6.3.1, 6.3.3 and 6.3.4.
6.4
Theorem. Consider a dynamical system defined by the differential equation 3.1. Let its rest point x = 0 be an unstable weak attractor.
Assume that there exist in I (6.3.2) two compact sets P and N with the
6. Non-continuous Liapunov Functions on the Region of Weak Attraction
181
following properties :
6.4.1
P is positively invariant.
6.4. 2
N is negatively invariant.
6.4.3
P V N = J({O}) V {O} = D+({O}).
6.4.4
For each x E I there exists one and only one real number t' (x) E R
such that xr(x) E P (\ N.
6.4.5
]+ (x, P) == {O} for each x E P.
6.4.6
J-(x, N) == {O} for each x EN.
Then there exists on D+({O}) a Liapunov function with the properties

6.3.1, 6.3.3 and 6.3.4.
Proof. The rest point x == 0 is positively (negatively) asymptotically

stable relatively to P (to N). By applying Lemma 6.1 and Corollary 6.2
one can construct respectively on P and on N the real-valued functions
<pp(x) and Op(x) and <pN (x) and ON (x) which satisfy the properties 6.1.16.1.4 and 6.3.1, respectively. Then the real-valued function
for x == 0,
0
6.4.7
for xE P,
<p(x) == -<pp(x)
<pN (x)
for xE N - P
satisfies the conditions 6.3.1, 6.3.3 and 6.3.4.

6.5
Theorem. Let {O} be as in Theorem 6.4 and let S be a section
of I (6. 3. 2), i.e. let S C I be such that for all x E I there exists one and
only one real number r(x) ER, such that xr(x) ES. Let next
6.5.1
P == y+ (S)
and N == y- (S).
Then the following statements are equivalent:

6.5.2
The sets P and N (6.5.1) have the properties 6.4.1--6.4.6.
6.5.3
xn--+ x E P - Sor Xn--+ x EN - Simplies sign r(xn) ==sign r(x).
6.5.4
For each e
>
0, r(x) is bounded on the set I - S(O, e).
Proof. Notice, first of all, that under the hypotheses made on the
system there always exists an infinite number of sections of I, since I
contains neither periodic orbits nor rest points (IV, 1.10, 2. 4). Let us
now proceed with the proof of the theorem.
6.5.3--+ 6.5.2. As already pointed out D+({O}) is a compact set,
which from the definition of the section Sis such that P V N = D+({O})
==IV {O}. If it were true that P (\ N =f= S V {O}, since S ( P V N,
182
there would exist a point z~ S, zE Pf\ N = y+(S) (\ y-(S). Then there

exist sequences {xn}, {Yn} CS and sequences {rn} CR+ and {an} CRsuch that Xnt'n--+ z and YnGn--+ z.
Since z E P (\ N C P V N if z =F 0 then z E /. Hence there exists
r(z) ER such that zr(z) ES. From 6.5.3 both rn and <Jn have a sign which
is equal to the sign of T (z) and since they have signs which opposite
to one another it follows that r(z) = 0, i.e., z ES. Finally since the set
J+(x, P) is invariant and J+(x, P) C P, if z =F 0 and zE J+(x, P), then
y-(z) C P. Then there exists t' (z) < 0 such that zr (z) E S and fort < t' (z),
zt E P (\ N, from which zt E S. This implies that t = t' (z) which is absurd.
The same argument can be applied to J+ (x, N). This concludes this part
of the theorem.
6.5.2--+ 6.5.3. Let x E P - S be such that there exists a sequence
{xn} ( Rn with xn--+ x, xnt' (xn) E S, sign t' (xn) =F sign t' (x). Since x E P - S,
then t' (x) < 0. Then t' (xn) --+ + ex> or else there exists a real number
r' (x) > 0 such that xr' (x) E S.
Then xnr(xn)--+ 0. Now Yn == xnr(xn) ( S ( N and since J-({O}, N)
= {O} we must have Yn ( - t' (xn))--+ 0. This is absurd if Yn (-r (xn)) ==
xn--+ x =F 0. This concludes this part of the theorem. The fact that
6.5.2 ~ 6.5.4 can be analogously proved.
Remark. Notice that if for a certain section S ( I the real6.6
valued function t' (x) can be selected to be continuous, then the properties
6.5.3 and 6.5.4 are obviously true, then from Theorem 6.5 the sets P and
N have the properties 6.4.1-6.4.6. Since the set I (Rn (6.3.2) is invariant, and locally compact, then a necessary and sufficient condition for
the dynamical system to be dispersive on I (IV, 2.4) is that there can be
selected on I a section with t' (x) continuous. On the other hand the dynamical system is dispersive on I if and only if]+(/,/) = {O}.
Thus
6. 7
Theorem. Consider a dynamical system defined by the ordinary
differential equation 3.1. Assume that the rest point x = 0 is a weak
attractor and let I (6.3.1) be its region of instability. Assume that
6. 7.1
J+ (/, /)
{O}.
Then there exists in D+({O}) a Liapunov function which satisfies the

properties 6.3.1, 6.3.3 and 6.3.4.
Proof. The theorem will be proved by showing that the hypothesis

6. 7.1 is equivalent to the conditions 6.4.5 and 6.4. 6. Clearly if both 6.4.5
and 6.4.6 are true also 6.6.1 is true. Conversely if J+ (/, /) = {O}, then
from what was pointed out in Remark 6.6 it follows that J+(x, P) = {O}
183
and J-(x, N) = {O} where P

tion of I with t' (x) continuous.
y+ (S) and N
y- (S) and S is a sec-
6.8
Remark. The function which can be constructed on the set /,
if the condition 6. 7.1 is satisfied, has a property which is stronger than
6.3.4. In this latter case, in fact, the set of discontinuity D is intersected
in one and only one point by each trajectory in /. It is easy to see that
the existence of this type of Liapunov function is a necessary and sufficient condition for 6. 7.1 to be satisfied.
6.9
Theorem. Let rest point x = 0 of the dynamical system 6.3.1
be an unstable weak attractor which is such that on the set I there exists
a Liapunov function v = <p (x) which satisfies the conditions 6.3.1, 6.3.2
and 6.3.3.
In addition condition 6.3.4 is satisfied in the strongest form, i.e., the
set D - {O} is intersected by each positive semi-trajectory in one point
at the most. Then the relationship 6. 7.1 is true.
Proof. It is clear that D is a section of I. For this it is enough to show

that for each x EI there exists t' (x) ER, such that X'l' (x) ED. If this were
not true we would have y (x) CI - D and <p (x) would be continuous relatively to I and would have a continuous total time derivative along
y(x). From 6.3.3 we would in addition haveA+(x) = A-(x) = {O}, which
is absurd. Then Dis a section of I. Let next P = y+(D) and N = y-(D),
we have J+(/, I) = J+(J, N) V J+(J, P). If J+(J, I) =F {O} we have
J+(/, P) =F {O}. Now J+(J, P) is an invariant set contained inP, hence
there would exist in Pa trajectory which would not intersect D, which is
absurd.
The idea of using non-differentiable Liapunov functions for the characterization of the stability properties of differential equations was originated by H. OKAMURA [1] and fully developed by T. YOSHIZAWA in many of his papers (see T. YosHIZAWA [1]). For the differential equation x = f(x, t) OKAMURA and YOSHIZAWA use
Lipschitz continuous functions<p (t, x) and the derivative<jJ(t, x) = lim sup'l'-1 [<p(t+ r,
T~O+
+ rf(x, t))
- <p(t, x)]. In Chapter V we have already used continuous Liapunov

functions and have been able to prove very directly the converse theorems. In this
case however we did not need to prove that <p is Lipschitz because the derivative
is taken along trajectories.
Section 1. The theorems presented in this section are due to J. A. YORKE [3], and
proved for the case of the differential equation = f(x, t). The proof we give is
different from the original. We notice that in the above mentioned paper two different cases are taken into account: the case in which the differential equation has
uniqueness of solution and the case in which it has not. In this latter case the statement of the theorem is less general since it is proved that for each point x E U there
184
exists at least one solution <p (x, t, t0) of the differential equation, with <p (x, t0 , t0) = x,
which satisfies the inequality 1.6.3. Indeed J. A. YORKE gives an example of a second order differential equation without uniqueness for which there exists one solution which does not satisfy the inequality 1.6.3 when all conditions of the theorem
are met.
Section 2. The results of this section are due to N. P. BHATIA, G. P. SzEGo and
J. A. YORKE [1]. They constitute a motivation for the introduction of non-continuous Liapunov functions and provide a Liapunov-type characterization of attractors and weak attractors.
Section 3-5. The results of these sections are due to G. P. SzEGo and G. TRECCANI
[3].
The problem which has motivated the research of these sections is that of the
numerical construction of piecewise quadratic Liapunov functions, as an alternative
to high order polynomials. Piecewise quadratic Liapunov functions have the definite
advantage on higher order polynomials to allow the computation of the hypervolume
of set bounded by a level surface in a much simpler way. The details on the construction of Liapunov functions can be found in the paper by G. P. SzEGo, G. ARIENTI
and C. SuTTI [1].
Section 6. The motivation of Theorem 6. 7 lies in the need for a finer analysis of the
(instability) properties of flows as shown in section VI, 2. The results of this section are due to G. P. SzEGo and G. TRECCANI [4].
References
AEPPLI, A.; MARKUS, L.
1. Integral equivalence of vector fields on manifolds and bifurcation of differential
systems. Amer. J. Math. 85, 633-654 (1963).
ALBRECHT, F.
1. Un theoreme de comportement asymptotique des solutions des equations des
systemes d'equations differentielles. Bull. Acad. Polon. Sci., Cl. III, 4,
737 - 739 ( 1956).
ANDREA, s. A.
1. On homeomorphisms of the plane, and their embedding in flows. Bull. Amer.
Math. Soc. 71, 381-383 (1965).
ANDRONOV, A.; CHAIKIN, c. B.
1. Theory of Oscillations (Russian). Moscow 1937. English translation, Princeton:
Princeton Univ. Press 1949.
ANOSOV, D. v.
1. Geodesic flows on closed Riemann manifolds with negative curvature (Russian).
Trudy Mat. Inst. Steklov, no. 90 (1967). English translation: Proc. Steklov
Inst. Math., no. 90, Amer. Math. Soc., Providence, 1969.
2. Roughness of geodesic flows on compact Riemannian manifolds of negative
curvature (Russian). Dokl. Akad. Nauk SSSR 145, 707-709 (1962). English
translation : Soviet Math. (Dokl.) 3, 1068-1070 (1962).
3. Ergodic properties of geodesic flows on closed Riemannian manifolds of negative
curvature (Russian). Dokl. Akad. Nauk SSSR 151, 1250-1273 (1963).
English translation: Soviet Math. {Dokl.) 4, 1153-1155 (1963).
ANTOSIEWICZ, H. A.
1. A survey of Liapunov's second method. In: Contributions to the Theory of
Nonlinear Oscillations, vol. 4 (Annals of Math. Studies, no. 41). Princeton:
Princeton Univ. Press 1958, pp. 141-166.
ANTOSIEWICZ, H. A.; DAVIS, P.
1. Some implications of Liapunov's conditions of stability. J. Rat. Mech. Anal.
3, 447-457 (1954).
ANTOSIEWICZ, H. A.; DUGUNDJI, J.
1. Parallelizable flows and Liapunov's second method. Ann. Math. 73, 543-555
(1961).
ARIENTI, G. (see SZEGO, G. P.)
ARNOL'n, V. L.; AvEz, A.
1. Problemes ergodiques de la mecanique classique. Paris: Gauthier-Villars 1967.
ARZELA, c.
1. Funzioni di linee. Atti R. Accad. Lincei Rend. 5, 342-348 (1889).
2. Sulle funzioni di linee. Mem. R. Accad. Bologna 5, 225-244 (1895).
186
References
Ascou, G.
1. Le curve limiti di una varieta data di curve. Mem. R. Accad. Lincei 18, 521-586
(1883/84).
AUSLANDER, J.
1. Mean-L-stable systems. Illinois J. Math. 3, 566-579 (1959).
2. On the proximal relation in topological dynamics. Proc. Amer. Math. Soc. 11,
890-895 (1960).
3. Generalized recurrence in dynamical systems. In : Contributions to Differential
Equations, vol. 3. New York: Wiley 1964, pp. 55-74.
4. Transformation groups without minimal sets. Math. Systems Theory 2, 93- 95
(1968).
AUSLANDER, J.; BHATIA, N.P.; SEIBERT, P.
1. Attractors in dynamical systems. Bol. Soc. Mat. Mexicana 9, 55-66 (1964).
AUSLANDER, J.; GOTTSCHALK, W. H.
1. (Editors) Topological Dynamics. Proceedings of an International Symposium
held at Colorado State University, Fort Collins, Colorado, August 1967.
New York-Amsterdam: Benjamin 1968.
AUSLANDER, J.; HAHN, F.
1. Point transitive flows, algebras of functions and the Bebutov system. Fund.
Math. 60, 117-137 (1967).
AUSLANDER, J.; SEIBERT, P.
1. Prolongations and generalized Liapunov functions. In: Nonlinear Differential
Equations and Nonlinear Mechanics, J.P. LaSalle and S. Lefschetz (Editors). New York: Academic Press 1963, pp. 456-462.
2. Prolongations and stability in dynamical systems. Ann. Inst. Fourier (Grenoble)
14, 237- 267 (1964).
AUSLANDER, L.; GREEN, L. W.
1. G-induced flows. Amer. J. Math. 88, 43-60 (1966).
AUSLANDER, L.; GREEN, L. W.; HAHN, F.
1. Flows on Homogeneous Spaces (Annals of Math. Studies, no. 53). Princeton:
Princeton Univ. Press 1963.
AUSLANDER, L.; HAHN, F.
1. Real functions coming from flows on compact spaces and concepts of almost
periodicity. Trans. Amer. Math. Soc. 106, 415-426 (1963).
AvEz, A. (see also ARNOL'D, V. L.)
1. Ergodic Theory of Dynamical Systems. Lecture Notes, University of Minnesota,
1967.
2. Anosov diffeomorphisms. In: Topological Dynamics, J. Auslander and W. H.
Gottschalk (Editors). New York-Amsterdam: Benjamin 1968, pp. 17-51.
BAmosov, V. A. (see also BARBASHIN, E.A.)
1. Invariant functions of dynamical systems (Russian). Izv. Vyssh. Uchebn.
Zaved., Mat. no. 1 (8), pp. 9-15 (1959).
2. On homeomorphisms of dynamical systems (Russian). Izv. Vyssh. Uchebn.
Zaved., Mat. no. 3 (16), pp. 21-29 (1960).
BARBASHIN, E. A.
1. Introduction to the Theory of Stability (Russian). Moscow: Izd. "Nauka" 1967.
2. Sur certaines singularites qui surviennent dans un systeme dynamique quand
l'unicite est en defaut. C. R. (Dokl.) Acad. Sci. URSS 41, 139-141 (1943).
3. Les singularites locales des points ordinaires d 'un syst eme d 'equations differentielles. C. R. (Dokl.) Acad. Sci. URSS 41, 183-186 (1943).
4. Sur la conduite des points sous les transformations homeomorphes de l'espace.
C. R. (Dokl.) Acad. Sci. URSS 51, 3-5 (1946).
References
187
5. On homeomorphisms of dynamical systems (Russian). Dokl. Akad. Nauk SSSR

61, 429-432 (1948).
6. On the theory of generalized dynamical systems (Russian) . U chen. Zap. Moskov.
Univ., no. 135, Mat. 2, 110-133 (1948).
7. On homeomorphisms of dynamical systems (Russian).Mat.Sbornik 27, 455-470
(1950).
8. Dispersive dynamical systems (Russian). Uspehi Mat. Nauk 5, 138-139 (1950).
9. On the theory of systems of multivalued transformations of a topological space
(Russian). Uchebn. Zap. Ural. Univ., no. 7, pp. 54-60 (19.50).
10. The method of sections in the theory of dynamical systems (Russian). Mat.
Sbomik 29, 233-280 (1951).
11. On homeomorphisms of dynamical systems, II (Russian). Mat. Sbornik 29,
501- 518 (1951).
12. On the behavior of points under homeomorphic transformations of a space.
(Generalization of theorems of Birkhoff) (Russian). Trudy Ural. Politehn.
Inst. 51, 4-11 (1954).
13. Conditions for the existence of recurrent trajectories in dynamical systems with
a cylindrical phase space. Differencial'nye Uravnenija 3, 1627-1633 (1967).
BARBASHIN, E.A.; BAIDOSOV, V.A.
1. On the question of topological definition of integral invariants (Russian). Izv.
Vyssh. Uchebn. Zaved., Mat. no. 3 (4), pp. 8-12 (1958).
BARBASHIN, E.A.; KRASOVSKII, N.N.
1. On stability of motion in the large. Dokl. Akad. Nauk SSSR 86, 453-456 (1952).
2. On the existence of Liapunov functions in the case of asymptotic stability in the
large. Prikl. Mat. Mekh. 18, 345-350 (1954).
BARBASHIN, E.A.; SHOLOHOVICH, F.A.
1. The mapping of a dynamical system into a time-analytic dynamical system
(Russian). Izv. Vyssh. Uchebn. Zaved., Mat. no. 1 (14), pp. 11-15 (1960).
BARBUTI, U.
1. Su alcuni teoremi di stabilita. Ann. Scuola Norm. Sup. Pisa 8, 81-91 (1954).
BAROCIO, s.
1. On certain critical points of a differential system in the plane. In : Contributions
to the Theory of Nonlinear Oscillations, vol. 3 (Annals of Math. Studies,
no. 36). Princeton: Princeton Univ. Press 1956, pp. 127 -136.
2. On trajectories in the vicinity of a three-dimensional singularity (Spanish).
Bol. Soc. Mat. Mexicana 1, 57-58 (1956).
BASS, R. W.
1. On the regular solutions at a point of singularity of a system of nonlinear
differential equations. Amer. J. Math. 77, 734- 742 (1955).
2. Zubov's stability criterion. Bol. Soc. Mat. Mexicana 4, 26-29 (1959).
BAUM, J.D.
1. An equicontinuity condition for transformation groups. Proc. Amer. Math.
Soc. 4, 656- 662 (1953).
2. Asymptoticity in topological dynamics. Trans. Amer. Math. Soc. 77, 506-519
(1954).
3. P-recurrence in topological dynamics. Proc. Amer. Math. Soc. 7, 1146-1154
(1956).
4. An equicontinuity condition in topological dynamics. Proc. Amer. Math. Soc.
12, 30-32 (1961).
5. Instability and asymptoticity in topological dynamics. Pacific J. Math. 12,
25-34 (1962).
188
References
6. Some remarks on P-limit point stability. In: Topological Dynamics, J. Auslander and W. H. Gottschalk (Editors). New York-Amsterdam: Benjamin
1968, pp. 53-66.
BEBUTOV, M. v.
1. Surles systemes dynamiques stables au sens de Liapounoff. C. R. (Dokl.) Acad.
Sci. URSS 18, 155-158 {1938).
2. Sur la representation des trajectoires d'un systeme dynamique sur un systeme
de droites paralleles. Bull. Math. Univ. Moscou, Ser. Internat. 2, fasc. 3,
pp. 1-22 (1939).
3. Surles systemes dynamiques dans l'espace des fonctions continues. Bull. Math.
Univ. Moscou, Ser. Intemat. 2, fasc. 5 (1939).
4. Surles systemes dynamiques dans l'espace des fonctions continues. C. R. {Dokl.)
Acad. Sci. URSS 27, 904-906 (1940).
5. On dynamical systems in the space of continuous functions (Russian). Bull.
Moskov. Gos. Univ., Mat. 2, no. 5, pp. 1-52 (1941).
BEBUTOV 1 M. v.; STEPANOV, v. v.
1. Sur le changement du temps dans les systemes dynamiques possedant une
mesure invariante. C.R. (Dokl.) Acad. Sci. URSS 24, 217-219 (1939).
2. Sur la mesure invariante dans les systemes dynamiques qui ne different que
par le temps. Rec. Math. (Mat. Sbornik) 7, 143-166 (1940).
BECK, A.
1. On invariant sets. Ann. Math. 67, 99-103 (1958).
2. Continuous flows with closed orbits. Bull. Amer. Math. Soc. 66, 305-307 (1960).
3. Plane flows with few stagnation points. Bull. Amer. Math. Soc. 71, 892-896
(1965).
4. Plane flows with closed orbits. Trans. Amer. Math. Soc. 114, 539-551 (1965).
BELLMAN, R.
1. Stability Theory of Differential Equations. New York-Toronto-London: McGraw-Hill 1953.
BELLMAN, R.; COOKE, K. L.
1. Differential-Difference Equations. New York: Academic Press 1963.
BENDIXSON, I.
1. Sur les courbes definies par des equations differentielles. Acta Ma th. 24, 1- 88
(1901).
BESICOVITCH, A. s.
1. Almost Periodic Functions. Cambridge : Cambridge Univ. Press 1932. Reprint,
New York: Dover 1954.
BHATIA, N. P. (see also AUSLANDER, J .)
1. Stability and Lyapunov functions in dynamical systems. In : Contributions to
Differential Equations, vol. 3. New York: Wiley 1964, pp. 175-188.
2. Weak attractors in dynamical systems. Bol. Soc. Mat. Mexicana 11, 56-64
(1966).
3. Criteria for dispersive flows. Math. Nachr. 32, 89-93 (1960).
4. Lectures on Ordinary Differential Equations: Stability Theory with Applications. Department of Mathematics, Western Reserve University, Spring,
1964.
5. On asymptotic stability in dynamical systems. Math. Systems Theory 1,
113-128 (1967).
6. Semidynamical flow near a compact invariant set. In: Topological Dynamics,
J. Auslander and W. H. Gottschalk (Editors). New York-Amsterdam: Benjamin 1968, pp. 81-96.
References
189
7. Dynamical systems. In: Mathematical Systems Theory and Economics I,

H. W. Kuhn and G. P. Szego (Editors) (Lecture Notes in Operations
Research and Mathematical Economics, vol. 11). Berlin-Heidelberg-New
York: Springer 1969, pp. 1-10.
8. Semidynamical systems. In: Mathematical Systems Theory and Economics II,
H. W. Kuhn and G. P. Szego (Editors) (Lecture Notes in Operations
Research and Mathematical Economics, vol. 12). Berlin-Heidelberg-New
York: Springer 1969, pp. 303-318.
9. Attraction and non-saddle sets in dynamical systems. J. Diff. Eqs. 8 (1970).
10. Characteristic Properties of Stable Sets and Attractors in Dynamical Systems.
Tech. Note BN-637, IFDAM, Univ. of Maryland, College Park, Maryland,
January 1970. Invited Lecture at Conference on Rational Mechanics, Rome
Feb. 23-26, 1970.
BHATIA, N.P.; HAJEK, 0.
1. Local Semi-Dynamical Systems (Lecture Notes in Mathematics, vol. 90).
Berlin-Heidelberg-New York: Springer 1969.
BHATIA, N.P.; LAZER, A.; LEIGHTON, w.
1. Applications of the Poincare-Bendixson theory. Ann. Mat. Pura Appl., Ser.
IV, 73, 27-32 (1966).
BHATIA, N.P.; LAZER, A.C.; SZEGO, G.P.
1. On global weak attractors in dynamical systems. J. Math. Anal. Appl. 16,
544-552 (1966).
BHATIA, N.P.; SzEGO, G.P.
1. Dynamical Systems: Stability Theory and Applications (Lecture Notes in
Mathematics, vol. 35). Berlin-Heidelberg-New York: Springer 1967.
2. An extension theorem for asymptotic stability. In: Differential Equations and
Dynamical Systems, J. K. Hale and J.P. LaSalle (Editors). New York:
Academic Press 1967, pp. 147-154.
3. Weak attractors in Rn. Math. Systems Theory 1, 129-134 (1967).
BHATIA, N.P.; SzEGO, G.P.; YORKE, J.A.
1. A Liapunov characterization of attractors. Boll. Un. Mat. Ital., Ser. IV, 1,
222-228 (1969).
BIRKHOFF, G.D.
1. Dynamical Systems (Amer. Math. Soc. Colloquium Publications, vol. 9). New
York 1927.
2. Collected Works, vol. 1, 2, 3. Amer. Math. Soc., New York, 1950. Reprint, New
York: Dover 1968.
3. Uber gewisse Zentralbewegungen dynamischer Systeme. Kgl. Ges. d. Wiss.
Gottingen, Nachrichten, Math.-phys. Kl., 1926, Heft 1, pp. 81-92.
4. On the periodic motions of dynamical systems. Acta Math. 50, 359-379 (1927).
5. Stability and the equations of dynamics. Amer. J. Math. 49, 1-38 (1927).
6. Proof of a recurrence theorem for strongly transitive systems. Proc. Nat. Acad.
Sci. USA 17, 650-655 (1931).
BOCHNER, S.
1. Beitrage zur Theorie der fastperiodischen Funktionen, I. Math. Ann. 96,
119-147 (1926).
BoHR,H.
1. Zur Theorie der fastperiodischen Funktionen, I, II, III. Acta Math. 45, 29-127
(1924); 46, 101-214 (1925); 47, 237-281 (1926).
BOUQUET, J.C.; BRIOT, c. A. A.
1. Recherches sur les fonctions definies par les equations differentielles. J. :Ecole
Polytech. (Paris) 21, cah. 36, pp. 133-198 (1856).
190
References
BRAUER,F.
1. Liapunov functions and comparison theorems. In: Nonlinear Differential
Equations and Nonlinear Mechanics, J. P. LaSalle and S. Lefschetz (Editors). New York: Academic Press 1963, pp. 435-441.
2. The use of comparison theorems for ordinary differential equations. In: Stability
Problems of Solutions of Differential Equations, A. Ghizzetti (Editor).
Gubbio : Oderisi 1966.
BRAUER,F.; STERNBERG, SH.
1. Local uniqueness, existence in the large, and the convergence of successive
approximations. Amer. J. Math. 80, 797 (1958).
BRIOT, C. A.A. (see BOUQUET, J.C.)
BRONSTEIN, I. U. (see also SIBIRSKII, K. S.)
1. Motions in partially ordered dynamical systems (Russian). Uchen. Zap.
Kishinev. Univ., no. 39, pp. 249-251 (1959).
2. On dynamical systems without uniqueness, as semigroups of non-single-valued
mappings of a topological space (Russian). Dokl. Akad. Nauk SSSR 144,
954-957 (1962). English translation: Soviet Math. (Dokl.) 3, 824-827
(1962).
3. Recurrence, periodicity and transitivity in dynamical systems without uniqueness (Russian). Dokl. Akad. Nauk SSSR 151, 15-18 (1963). English
translation: Soviet Math. (Dokl.) 4, 889-892 (1963).
4. On dynamical systems without uniqueness, as semigroups of non-singlevalued
mappings of a topological space (Russian). Izv. Akad. Nauk Moldav. SSR,
Ser. Estestven. Tekh. Nauk, no. 1, pp. 3-18 (1963).
5. Two examples of dynamical systems (Russian). Izv. Akad. Nauk Moldav. SSR,
Ser. Estestven. Tekh. Nauk, no. 1, pp. 73-74 (1963).
6. Recurrent points and minimal sets in dynamical systems without uniqueness
(Russian). Izv. Akad. Nauk Moldav. SSR, Ser. Estestven. Tekh. Nauk,
no. 7, pp. 14-21 (1965).
7. On homogeneous minimal sets (Russian). Papers on Algebra and Analysis,
Kishinev, 1965, pp. 115-118.
8. Recurrent points and minimal sets in dynamical systems without uniqueness.
Bui. Akad. Shtiince RSS Moldoven., 1965, no. 7, pp. 14-21.
9. A contribution to the theory of distal minimal sets and distal functions (Russian).
Dokl. Akad. Nauk SSSR 172, 255-257 (1967). English translation: Soviet
Math. (Dokl.) 8, 59-61 (1967).
BRONSTEIN, I. U.; SHCHERBAKOV, B. A.
1. Certain properties of Lagrange stable funnels of generalized dynamical systems
(Russian). Izv. Akad. Nauk Moldav. SSR, Ser. Estestven. Tekh. Nauk, no. 5,
pp. 99-102 (1962).
BROUWER, L. E. J.
1. On continuous vector distributions, I, II, III. Verh. Nederl. Akad. Wetensch.,
Afd.Natuurk., Sec. I, 11, 850-858 (1909); 12, 716-734 (1910); 13, 171-186 (1910).
BROWDER, F. E.
1. On the iteration of transformations in noncompact minimal dynamical systems.
Proc. Amer. Math. Soc .. 9, 773-780 (1958).
2. On a generalization of the Schauder fixed point theorem. Duke Math. J. 26,
291-303 (1959).
3. On the continuity of fixed points under deformations of continuous mappings.
Summa Brasil. Math. 4, 183-191 (1960).
References
191
4. On the fixed points index for continuous mappings of locally connected spaces.
Summa Brasil. Math. 4, 253- 293 (1960).
BROWN, M.
1. The monotone union of open n-cells is an open n-cell. Proc. Amer. Math. Soc.
12, 812-814 (1961).
BunAK,B.M.
1. Dispersive dynamical systems (Russian). Vestnik Moskov. Univ., no. 8,
pp. 135-137 (1947).
2. The concept of motion in a generalized dynamical system (Russian). Uchen.
Zap. Moskov. Univ., no. 155, Mat., 5, 174-194 (1952).
BUSHAW, D.
1. Dynamical polysystems and optimization. In: Contributions to Differential
Equations, vol. 2, New York: Wiley 1963.
2. A stability criterion for general systems. Math. Systems Theory 1, 79- 88
(1967).
3. Dynamical polysystems - a survey. In: Proceedings U.S.-Japan Seminar on
Differential and Functional Equations, W. A. Harris, Jr. and Y. Sibuya
(Editors). New York-Amsterdam: Benjamin 1967, pp. 13-26.
4. Stability of Liapunov and Poisson types. SIAM Rev. 11, 214-225 (1969).
CAIRNS, s. s.
1. (Editor) Differential and Combinatorial Topology. Princeton: Princeton Univ.
Press 1965.
CARATHEODORY,C.
1. Vorlesungen fiber reelle Funktionen. Leipzig: Teubner 1927. Reprint, New
York: Chelsea 1958/60.
2. Calculus of Variations and Partial Differential Equations of the First Order
(German). Leipzig-Berlin: Teubner 1935. English translation in two volumes, San Francisco: Holden-Day 1965/67.
CARTWRIGHT, M. L.
1. Some decomposition theorems for certain invariant continua and their minimal
sets. Fund. Math. 48, 229-250 (1960).
2. From non-linear oscillations to topological dynamics. J. London Math. Soc.
39, 193-201 (1964).
3. Topological problems of nonlinear mechanics. Abh. Deutsch. Akad. Wiss.
Berlin, Kl. Math., Phys., Tech. (1965) no. 1. - III. Konferenz iiber nichtlineare Schwingungen, Berlin 1964, Teil I. Berlin: Akademie-Verlag 1965,
pp. 135-142.
4. Equicontinuous mappings of plane minimal sets. Proc. London Math. Soc. 14A,
51-54 (1965).
5. Almost periodic differential equations and almost periodic flows. J. Diff. Eqs.
5, 117-135 (1969).
CAUCHY, A. L.
1. Oeuvres completes, vol. 1. Paris: Gauthier-Villars 1888.
CELLINA, A. (see SzEGO, G.P.)
CESARI, L.
1. Asymptotic Behavior and Stability Problems in Ordinary Differential Equations (Ergebnisse der Mathematik und ihrer Grenzgebiete, Neue Folge,
Heft 16). Berlin-Gottingen-Heidelberg: Springer 1959.
CHAIKIN, C. B. (see ANDRONOV, A.)
CHEN, K. T.
1. Equivalence and decomposition of vector fields about an elementary critica
point. Amer. J. Math. 85, 693-722 (1963).
192
References
CHERRY, T. M.
1. Topological properties of solutions of ordinary differential equations. Amer. J.
Math. 59, 957-982 (1937).
2. Analytic quasi-periodic curves of discontinuous type on a torus. Proc. London
Math. Soc. 44, 175-215 (1938).
3. The pathology of differential equations. J. Austral. Math. Soc. 1, 1-16 (1959).
CHETAEV, N.
1. The Stability of Motion. Moscow: G ITTL 1946; 2nd ed. 1959. English translation, London: Pergamon Press 1961.
2. On stability in the sense of Poisson (Russian). Zap. Kazansk. Mat. Obshch.,
1929.
3. Un theoreme sur l'instabilite. C.R. (Dokl.) Acad. Sci. URSS (N.S.) 2, 529-531
(1934).
4. On instability of the equilibrium in certain cases where the force function does
not have a maximum (Russian). Uchen. Zap. Kazansk. Univ., 1938.
5. On unstable equilibrium in certain cases when the force function is not maximum (Russian). Prikl. Mat. Mekh. 16, 89-93 (1952).
6. On certain questions relative to the problem of the stability of unsteady motions (Russian). Prikl. Mat. Mekh. 24, 6-9 (1960). English translation:
J. Appl. Math. Mech. 24, 5-22 (1960).
7. On the stability of "rough" systems (Russian). Prikl. Mat. Mekh. 24, 20-22
(1960). English translation: J. Appl. Math. Mech. 24, 23-26 (1960).
CHU,HSIN
1. On totally minimal sets. Proc. Amer. Math. Soc. 13, 457-458 (1962).
2. Algebraic topology criteria for minimal sets. Proc. Amer. Math. Soc. 13, 503508 (1962).
3. Fixed points in a transformation group. Pacific J. Math. 15, 1131-1135 (1965).
4. A note on compact transformation groups with a fixed end point. Proc. Amer.
Math. Soc. 16, 581-583 (1965).
CLAY, J.P.
1. Proximity relations in transformation groups. Trans. Amer. Math. Soc. 108,
88-96 (1963).
2. Invariant attractors in transformation groups. Illinois J. Math. 8, 473-479
(1964).
CODDINGTON, E. A.; LEVINSON, N_
1. Theory of Ordinary Differential Equations.
McGraw-Hill 1955.
New York-Toronto-London:
COLEMAN, C.
1. Equivalence of planar dynamical and differential systems. J. Diff. Eqs. 1,
222-233 (1965).
CoNLEY,C.C.
1. Invariant sets in a monkey saddle. In: Proceedings U.S.-Japan Seminar on
Differential and Functional Equations, W. A. Harris, Jr. and Y. Sibuya
(Editors). New York-Amsterdam: Benjamin 1967, pp. 663-668.
2. Twist mappings, linking, analyticity and periodic solutions which pass close to
an unstable periodic solution. In: Topological Dynamics, J. Auslander and
W. H. Gottschalk (Editors). New York-Amsterdam: Benjamin 1968,
pp. 129-153.
3. On the ultimate behavior of orbits with respect to an unstable critical point, I.
Oscillating, asymptotic and capture orbits. J. Diff. Eqs. 5, 136-158 (1969).
References
193
CONLEY, C.; EASTON, R.

1. Isolated invariant sets and isolating blocks (Advances in Differential and Integral
Equations). SIAM, Philadelphia, 1969.
CoNTI, R. (see also REISSIG, R. and SANSONE, G.)
1. Sull'equivalenza asintotica dei sistemi di equazioni differenziali. Ann. Mat.
Pura Appl. 41, 95-104 (1965).
2. Limitazioni in ampiezza delle soluzioni di un sistema di equazioni differenziali
ed applicazioni. Boll. Un. Mat. Ital. 11, 344-349 (1956).
3. Sulla prolungabilita delle soluzioni di un sistema di equazioni differenziali ordinarie. Boll. Un. Mat. Ital. 11, 510-514 (1956).
CooKE, K. L. (see BELLMAN, R.)
CoPPEL, W. A.
1. Stability and Asymptotic Behavior of Differential Equations. Boston: Heath
& Co. 1965.
CoRDUNEAU, C.
1. Sur la stabilite asymptotique. An. Sti. Univ. "Al. I. Cuza", Ia~i, Sect. I a Mat.
5, 37-39 (1959).
2. Sur la stabilite asymptotique, II. Rev. Roumaine Math. Pures Appl. 5, 573-576
(1960}.
3. Application des inegalites differentielles a la theorie de la stabilite (Russian).
An. Sti. Univ. "Al. I. Cuza", Iar,i, Sect. Ia Mat. 6, 46-58 (1960).
4. Sur certains systemes differentielles nonlineaires. An. Sti. Univ. "Al. I. Cuza",
Iaei, Sect. Ia Mat. 6, 257-260 (1960).
CRONIN, J.
1. Fixed Points and Topological Degree in Nonlinear Analysis (Mathematical
Surveys, no. 11). Amer. Math. Soc., Providence, 1964.
D'AMBROSIO, U.; LAKSHMIKANTHAM, V.
1. On VJ-stability and differential inequalities. In: Topological Dynamics, J. Auslander and W. H. Gottschalk (Editors). New York-Amsterdam: Benjamin
1968, pp. 155-164.
DAVIS, P. (see ANTOSIEWICZ, H. A.)
DENJOY, A.
1. Surles caracteristiques a la surface du tore. C.R. Acad. Sci. Paris 194, 830-833
(1932).
2. Surles caracteristiques du tore. C. R. A.cad. Sci. Paris 194, 2014-2016 (1932).
3. Sur les courbes definies par les equations differentielles a la surface du tore.
J. Math. Pures Appl. 11, 333-375 (1932).
4. Les trajectoires a la surface du tore. C. R. Acad. Sci. Paris 223, 5-8 (1964).
5. Surles trajectoires du tore. C. R. Acad. Sci. Paris 251, 175-177 (1960).
DESBROW, D.
1. On connexion, invariance and stability in certain flows. Proc. Cambridge Philos.
Soc. 60, 51-55 (1964).
DEYSACH, L. G.; SELL, G. R.
1. On the existence of almost periodic motions. Michigan Math. J. 12, 87-95 (1965).
DILIBERTO, s. P.
1. Qualitative behavior for classical dynamical systems (Advances in Differential
and Integral Equations). SIAM, Philadelphia, 1969.
DIRICHLET, G. L.
1. Uber die Stabilitat des Gleichgewichts. J. reine angew. Math. 32, 85-88 (1846).
DOWKER, Y. N.
1. On minimal sets in dynamical systems. Quart. J. Math., Oxford Ser., 7, 5-16
(1956).
13
194
References
DOWKER, Y. N.; FRIEDLANDER, F. G.

1. On limit sets in dynamical systems. Proc. London Math. Soc. 4, 168-176 (1954).
DUBOSHIN, G. N.
1. Foundations of the Theory of Stability of Motions. Moscow 1957.
DuGUNDJI, J. (see also ANTOSIEWICZ, H. A.)
1. Topology. Boston: Allyn & Bacon 1966.
EASTON, R. (see CONLEY, C.)
EDREI, A.
1. On iteration of mappings of a metric space onto itself. J. London Math. Soc. 26,
96-103 (1951).
ELLIS, R.
1. Continuity and homeomorphism groups. Proc. Amer. Math. Soc. 4, 969-973
(1953).
2. A note on the continuity of the inverse. Proc. Amer. Math. Soc. 8, 372-373
(1957).
3. Locally compact transformation groups. Duke Math. J. 24, 119-125 (1957).
4. Distal transformation groups. Pacific J. Math. 8, 401-405 (1958).
5. Universal minimal sets. Proc. Amer. Math. Soc. 11, 540-543 (1960).
6. A semigroup associated with a transformation group. Trans. Amer. Math. Soc.
94, 272-281 (1960).
7. Point transitive transformation groups. Trans. Amer. Math. Soc. 101, 384-395
(1961).
8. Locally coherent minimal sets. Michigan Math. J. 10, 97-104 (1963).
9. Global sections of transformation groups. Illinois J. Math. 8, 380-394 (1964).
10. The construction of minimal discrete flows. Amer. J. Math. 87, 564-574 (1965).
11. Group-like extensions of minimal sets. Trans. Amer. Math. Soc. 127, 125-135
(1967).
12. The beginning of an algebraic theory of minimal sets. In: Topological Dynamics,
ELLIS, R.; GOTTSCHALK, w. H.
1. Homeomorphisms of transformation groups. Trans. Amer. Math. Soc. 94, 258271 (1960).
EL'sGoL'c, L. E.
1. An estimate for the number of singular points of a dynamical system defined
on a manifold (Russian). Mat. Sbornik 26, 215-223 (1950). English translation: Amer. Math. Soc. Transl. no. 68 (1952).
ENGELKING, R.
1. Quelques remarques concemant les operations sur les fonctions semi-continues
dans les espaces topologiques. Bull. Acad. Polan. Sci., Ser. Sci. Math. Astronom. Phys. 11, 719-726 (1963).
ENGLAND, J. W.
1. A characterization of orbits. Proc. Amer. Math. Soc. 17, 207-209 (1966).
ENGLAND, J. W.; KENT, J. F. III
1. P-recurrence and quasi-minimal sets. In: Topological Dynamics, J. Auslander
and W. H. Gottschaik (Editors). New York-Amsterdam: Benjamin 1968,
pp. 185-204.
ERUGIN, N. P.
1. On certain questions of stability of motion and the qualitative theory of differential equations (Russian). Prikl. Mat. Mekh. 14, 459-512 (1950).
2. A qualitative investigation of integral curves of a system of differential equa~
tions (Russian). Prikl. Mat. Mekh. 14, 659-664 (1950).
References
195
3. Theorems on instability (Russian). Prikl. Mat. Mekh. 16, 355-361 (1952).

4. The methods of A. M. Liapunov and questions of stability in the large (Russian).
Prikl. Mat. Mekh. 17, 389- 400 (1953).
5. Qualitative methods in theory of stability (Russian). Prikl. Mat. Mekh. 19,
599-616 (1955).
FOLAND, N. E.; UTZ, w. R.
1. The em bedding of discrete flows in continuous flows. In: Ergodic Theory.
Proceedings of an International Symposium held at Tulane University, New
Orleans, Louisiana, October 1961. New York: Academic Press 1963, pp. 121134.
FoMIN, S. (see also KoLMOGORov, A. N.)
1. On dynamical systems in a space of functions (Russian). Ukrainsk. Mat. Zh. 2,
no. 2, pp. 25-134 (1950).
FRIEDLANDER, F. G. (see also DowKER, Y. N.)
1. On the iteration of a continuous mapping of a compact space into itself. Proc.
Cambridge Philos. Soc. 46, 46-56 (1950).
FUKUHARA,M.
1. Sur les systemes des equations differentielles ordinaires. Japan J. Math. 5
345-350 (1929).
2. Sur les systemes d'equations differentielles ordinaires, II. Japan J. Math. 6,
269-299 (1930).
3. Sur !'ensemble des courbes integrales d'un systeme d'equations differentielles
ordinaires. Proc. Imp. Acad. Japan 6, 360-362 (1930).
FULLER, F. B.
1. Notes on trajectories in a solid torus. Ann. Math. 56, 438-439 (1952).
FURSTENBERG, H.
1. The structure of distal flows. Amer. J. Math. 85, 477-515 (1963).
2. Disjointness in ergodic theory, minimal sets and diophantine approximations.
Math. Systems Theory 1, 1-50 (1967).
GANTMACHER, F. M.
1. Theory of Matrices. New York: Chelsea 1960.
GARCIA, M.; HEDLUND, G. A.
1. The structure of minimal sets. Bull. Amer. Math. Soc. 54, 954-964 (1948).
GARY, J.
1. The topological structure of trajectories. Michigan Math. J. 7, 225-227 (1960).
GEISS, G. (see SzEGO, G. P.)
GHIZZETTI, A.
1. (Editor) Stability Problems of Solutions of Differential Equations. Proceedings
of a NATO Advanced Study Institute held in Padova, Italy, August 1965.
Gubbio: Oderisi 1966.
GOTTSCHALK, W. H. (see also AusLANDER, J. and ELLIS, R.)
1. A note on pointwise nonwandering transformations. Bull. Amer. Math. Soc. 52,
488-489 (1946).
2. Almost periodicity, equi-continuity and total boundedness. Bull. Amer. Math.
Soc. 52, 633- 636 (1946).
3. Recursive properties of transformation groups, II. Bull. Amer. Math. Soc. 54,
381-383 (1948).
4. Transitivity and equicontinuity. Bull. Amer. Math. Soc. 54, 982-984 (1948).
5. Characterizations of almost periodic transformation groups. Proc. Amer. Math.
Soc. 7, 709-712 (1956).
6. Minimal sets: an introduction to topological dynamics. Bull. Amer. Math. Soc.
64, 336-351 (1958).
I
13*
196
References
7. The universal curve of Sierpinski is not a minimal set. Notices. Amer. Math.
Soc. 6, 257 (1959).
8. An irreversible minimal set. In: Ergodic Theory. Proceedings of an International
Symposium held at Tulane University, New Orleans, Louisiana, October
1961. New York: Academic Press 1963, pp. 135-150.
9. Substitution minimal sets. Trans. Amer. Math. Soc. 109, 467-491 (1963).
10. Minimal sets occur maximally. Trans. New York Acad. Sci. 26, 348-353 (1964).
11. A survey of minimal sets. Ann. Inst. Fourier (Grenoble) 14, 53-60 (1964).
12. Ambits. Invited Lecture given at the Regional Conference on Global Differentiable Dynamics held at Case-Western Reserve University, Cleveland, Ohio,
June 1969.
GOTTSCHALK, w. H.; HEDLUND, G. A.
1. Recursive properties of transformation groups. Bull. Amer. Math. Soc. 52,
637-641 (1946).
2. The dynamics of transformation groups. Trans. Amer. Math. Soc. 65, 348- 359
(1949).
3. Asymptotic relations in topological groups. Duke Math. J. 18, 481-485 (1951).
4. Topological Dynamics (Amer. Math. Soc. Colloquium Publications, vol. 36).
Providence 1955.
5. A characterization of the Morse minimal set. Proc. Amer. Math. Soc. 15, 70-74
(1964).
GRABAR, M. I.
1. The representation of dynamical systems as systems of solutions of differential
equations. Dokl. Akad. Nauk SSSR 61, 433-436 (1948).
2. Transformations of dynamical systems into systems of solutions of differential
equations (Russian). Vestnik Moskov. Univ., 1952, no. 3, pp. 3-8.
3. On change of time in dynamical systems (Russian). Dokl. Akad. Nauk SSSR
109, 250-252 (1956).
4. On a sufficient test for isomorphism of dynamical systems (Russian). Dokl.
Akad. Nauk SSSR 109, 431-433 (1956).
5. Isomorphism of dynamical systems differing only in time (Russian). Dold. Akad.
Nauk SSSR 126, 931-934 (1959).
GREEN, L. W. (see also AusLANDER, L.)
1. Transversals to a flow. In: Topological Dynamics, J. Auslander and W. H.
Gottschalk (Editors). New York-Amsterdam: Benjamin 1968, pp. 225-242.
GROBMAN, D. M.
1. Systems of differential equations analogous to linear ones. Dokl. Akad. Nauk
SSSR 86, 19-22 (1952).
2. Homeomorphisms of systems of differential equations. Dokl. Akad. Nauk
SSSR 128, 880-881 (1959).
3. Topological and asymptotic equivalence for systems of differential equations
(Russian). Dokl. Akad. Nauk SSSR 140, 746-747 (1961). English translation: Soviet Math. (Dokl.) 2, 1201-1253 (1961).
4. Topological classification of the neighborhood of a singular point in n-dimensional space. Mat. Sbornik (N.S.) 56 (98), 77-94 (1962).
5. The topological equivalence of dynamical systems (Russian). Dokl. Akad. N auk
SSSR 175, 1211-1212 (1967). English translation: Soviet Math. (Dokl.) 8,
985-986 (1967).
6. Global topological equivalence of systems of differential equations. Mat. Sbornik (N.S.) 73, 600-607 (1967).
References
197
7. The neighborhood of a singular point of a system of differential equations with

small nonlinearities (Russian). Dokl. Akad. Nauk SSSR 172, 524-526
(1967). English translation: Soviet Math. (Dokl.) 8, 124-126 (1967).
HAAS, F.
1. A theorem about characteristics of differential equations on closed manifolds.
Proc. Nat. Acad. Sci. USA 38, 1004-1047 (1952).
2. On the global behavior of differential equations on two-dimensional manifolds.
Proc. Amer. Math. Soc. 4, 630-636 (1953).
3. The global behavior of differential equations on n-dimensional manifolds. Proc.
Nat. A.cad. Sci. USA 39, 1258-1260 (1953).
4. Poincare-Bendixson type theorems for two-dimensional manifolds different
from the torus. Ann. of Math. 59, 292-299 (1954).
5. On the total number of singular points and limit cycles of a differential equation. In: Contributions to the Theory of Nonlinear Oscillations, vol. 3
(.Annals of Math. Studies, no. 36). Princeton: Princeton Univ. Press 1956,
pp. 137 -172.
HADAMARD, J.
1. Sur les trajectoires en dynamiques. J. de Math., Ser. III, 3, 331-387 (1897).
2. Surles integrales d'un system d'equations differentielles ordinaires, considerees
comme fonctions des donnees initiales. Bull. Soc. Math. France 28, 64- 66
(1900).
3. Sur !'iteration et les solutions asymptotiques des equations differentielles. Bull.
Soc. Math. France 29, 224-228 (1901).
HAHN, F. J. (see also AUSLANDER, J. and AUSLANDER, L.)
1. Recursion of set trajectories in a transformation group. Proc. Amer. Math. Soc.
11, 527-532 (1960).
2. Nets and recurrence in transformation groups. Trans. Amer. Math. Soc. 99,
193-200 (1961).
3. On affine transformations of compact Abelian groups. Amer. J. Math. 85,
428-446 (1963); errata 86, 463-464 (1964).
HAHN, F.; PARRY, w.
1. Minimal dynamical systems with quasi-discrete spectrum. J. London Math.
Soc. 40, 309-324 (1965).
2. Some characteristic properties of dynamical systems with quasi-discrete spectra.
Math. Systems Theory 2, 179-190 (1968).
HAHN, w.
1. Theorie und Anwendung der direkten Methode von Ljapunov (Ergebnisse der
Mathematik und ihrer Grenzgebiete, Neue Folge, Heft 22). Berlin-GottingenHeidelberg: Springer 1959. English translation: Theory and Application of
Ljapunov's Direct Method. Englewood Cliffs, N.J.: Prentice-Hall 1963.
2. Stability of Motion. Berlin-Heidelberg-New York: Springer 1967.
3. Uber die Anwendung der Methode von Ljapunov auf Differenzengleichungen.
Math. Ann. 136, 430-441 (1958).
HAJEK, 0. (see also BHATIA, N. P.)
1. Dynamical Systems in the Plane. New York: Academic Press 1968.
2. Critical points of abstract dynamical systems. Comm. Math. Univ. Carolinae 5,
121-124 (1964).
3. Betti numbers of regions of attraction. Comm. Math. Univ. Carolinae 5, 129-132
(1964).
4. Structure of dynamical systems. Comm. Math. Univ. Carolinae 6, 53-72 (1965);
correction 6, 211-212 (1965).
5. Flows and periodic motions. Comm. Math. Univ. Carolinae 6, 165-178 (1965).
198
References
6. Sections of dynamical systems in E 2 Czechoslovak Math. J. 15, 205-211

(1965).
7. Prolongations of sections in local dynamical systems. Czechoslovak Math. J.
16, 41-45 (1966).
8. Differentiable representation of flows. Comm. Math. Univ. Carolinae 7, 219 225 (1966).
9. Theory of processes, I. Czechoslovak Math. J. 17, 159-199 (1967).
10. Theory of processes, II. Czechoslovak Math. J. 17, 372-398 (1967).
11. Local characterization of local semi-dynamical systems. Math. Systems Theory
2, 17-26 (1968).
12. Categorical concepts in dynamical systems theory. In: Topological Dynamics,
A.
1. Differential Equations: Stability Theory, Oscillations, Time-Lags. New York:
Academic Press 1965.
HALANAY,
J.
K.
1. Integral manifolds of perturbed differential systems. Ann. of Math. 73, 496-531
(1961 ).
2. Dynamical systems and stability. J. Math. Anal. Appl. 26, 39-59 (1969).
HALE,
J. K.; LASALLE, J. P.
1. (Editors) Differential Equations and Dynamical Systems. Proceedings of an
HALE,
International Symposium held at the University of Puerto Rico, Mayaguez,

Puerto Rico, Dec. 27-30, 1965. New York: Academic Press 1967.
J. K.; STOKES, A. P.
1. Behavior of solutions near integral manifolds. Arch. Rat. Mech. Anal. 6, 133170 (1960).
HALE,
HALKIN, H.
1. Topological aspects of optimal control of dynamical polysystems. In: Contri-
butions to Differential Equations, vol. 3. New York : Wiley 1964, pp. 377 385.
2. Finitely convex sets of nonlinear differential equations. Math. Systems Theory
1, 51-54 (1967).
P. R.
1. Lectures on Ergodic Theory. Mathematical Society of Japan, Tokyo 1956.
Reprint, New York: Chelsea 1960.
HALMOS,
W. A., Jr.; SrnuvA, Y.

1. (Editors) Proceedings U.S.-Japan Seminar on Differential and Functional
Equations, held at the University of Minnesota, Minneapolis, Minn. on June
26-30, 1967. New York-Amsterdam: Benjamin 1967.
HARTMAN, P.
1. Ordinary Differential Equations. New York: Wiley 1964.
2. On local homeomorphisms of Euclidean spaces. Bol. Soc. Mat. Mexicana 5,
220-241 (1960).
3. On stability in the large for systems of ordinary differential equations. Canad.
J. Math. 13, 480-492 (1961).
4. On the local linearization of differential equations. Amer. Math. Monthly 70,
219-232 (1963).
5. The existence and stability of stationary points. Duke Math. J. 33,2, 281-290
(1966).
HARRIS,
References
199
HARTMAN, P.; OLECH, Cz.

1. On global asymptotic stability of solutions of ordinary differential equations.
Trans. Amer. Math. Soc. 104, 154-178 (1962).
HARTMAN, P.; WINTNER, A.
1. Integrability in the large and dynamical stability. Amer. J. Math. 65, 273-278
(1943).
2. On the asymptotic behavior of the solutions of a nonlinear differential equation.
Amer. J. Math. 68, 301-308 (1946).
HEDLUND, G. A. (see also GARCIA, M. and GOTTSCHALK, W. H.)
1. Sturmian minimal sets. Amer. J. Math. 66, 605-620 (1944).
2. A class of transformations of the plane. Proc. Cambridge Philos. Soc. 51, 554564 (1955).
3. Mappings on sequence spaces (Part I). Communications Res. Div. Tech. Rep.
no. 1, von Neumann Hall, Princeton, N.J ., February 1961.
4. Transformations commuting with the shift. In: Topological Dynamics, J. Auslander and W. H. Gottschalk (Editors). New York-Amsterdam: Benjamin
1968, pp. 259-290.
HILMY, H.
1. Sur la structure d'ensemble des mouvements stables au sens de Poisson. Ann.
of Math. 37, 43-45 (1936).
2. Sur les ensembles quasi-minimaux dans les systemes dynamiques. Ann. of
Math. 37, 899-907 (1936).
3. Sur les centres d'attraction minimaux des systemes dynamiques. Compositio
Math. 3, 227-238 (1936).
4. Surles mouvements des systemes dynamiques qui admettent l'incompressibilite
des domaines. Amer. J. Math. 59, 803-808 (1937).
5. Sur une propriete des ensembles minimaux. C. R. (Dokl.) Acad. Sci. URSS
14, 261-262 (1937).
6. Sur la theorie des ensembles quasi-minimaux. C. R. (Dokl.) Acad. Sci. URSS
15, 113-116 (1937).
7. Surles theoremes de recurrence dans la dynamique generale. Amer. J. Math. 61,
149-160 (1939).
HIRASAWA, Y. (see URA, T.)
HOCKING, J.; YOUNG, G.
1. Topology. Reading, Mass.: Addison-Wesley 1961.
HOPF, E.
1. Ergodentheorie (Ergebnisse der Mathematik, vol. 5, no. 8). Berlin: Springer
1937. Reprint, New York: Chelsea 1948.
HORELICK, B.
1. An algebraic approach to the study of minimal sets in topological dynamics.
Thesis, Wesleyan University, 1967.
Hu, SzE-TSEN
1. Theory of Retracts. Detroit: Wayne State Univ. Press 1955.
INGWERSON, D. R.
1. A modified Liapunov method for nonlinear stability analysis. IRE Trans.
Automatic Control 6, 199-210 (1961); discussion 7, 85-88 (1962).
jARNIK, J.; KURZWEIL, J.
1. On invariant sets and invariant manifolds of differential systems. J. Diff.
Eqs. 6, 247-263 (1969).
JONES, G. s.
1. (Editor) Seminar on Differential Equations and Dynamical Systems (Lecture
Notes in Mathematics, vol. 60). Berlin-Heidelberg-New York: Springer 1968.
200
References
2. Asymptotic fixed point theorems and periodic systems of functional differential

equations. In: Contributions to Differential Equations, vol. 2. New York:
Wiley 1963, pp. 385-405.
3. Stability and asymptotic fixed point theory. Proc. Nat. Acad. Sci. USA 53,
6, 1262-1264 (1965).
4. The existence of critical points in generalized dynamical systems. In: Seminar
on Differential Equations and Dynamical Systems, G. S. Jones (Editor)
(Lecture Notes in Mathematics, vol. 60). Berlin-Heidelberg-New York:
Springer 1968.
JONES, G. S.; YORKE, J. A.
1. The existence and nonexistence of critical points in bounded flows. J. Diff.
Eqs. 6, 238-246 (1969).
KAHN, P. J.; KNAPP, A. w.
1. Equivariant maps onto minimal flows. Math. Systems Theory 2, 319- 324
(1968).
KAKUTANI, SH.
1. A proof of Beboutov's theorem. J. Diff. Eqs. 4, 194-201 (1968).

KALMAN, R. E.
1. Algebraic aspects of the theory of dynamical systems. In: Differential Equations and Dynamical Systems. J. K. Hale and J.P. LaSalle (Editors). New
York: Academic Press 1968, pp. 133-146.
2. Introduction to the algebraic theory of linear dynamical systems. In: Mathematical Systems Theory and Economics I, H. W. Kuhn and G. P. Szego
(Editors) (Lecture Notes in Operations Research and Mathematical Economics, vol. 11). Berlin-Heidelberg-New York: Springer 1969, pp. 41- 66.
KAMKE, E.
1. Differentialgleichungen reeller Funktionen. Leipzig: Akad. Verlagsges. 1930.
Reprint, New York: Chelsea 194 7.
2. Differentialgleichungen, Losungsmethoden und Losungen, I. Gewohnliche Differentialgleichungen, 7th ed. Leipzig: Akad. Verlagsges. 1961.
3. Differentialgleichungen, Losungsmethoden und Losungen, II. Partielle Differentialgleichungen erster Ordnung fur eine gesuchte Funktion, 4th ed.
Leipzig: Akad. Verlagsges. 1959.
4. Zur Theorie der Systeme gewohnlicher Differentialgleichungen, II. Acta Math.
58, 57-85 (1932).
VAN KAMPEN, E. R.
1. The topological transformations of a simple closed curve into itself. Amer. J.
Math. 57, 142-152 (1935).
2. Remarks on systems of ordinary differential equations. Amer. J. Math. 59,
144-152 (1937)
KAPLAN,
w.
1. Regular curve-families filling the plane, I. Duke Math. J. 7, 154-185 (1940).

2. Regular curve-families filling the plane, IL Duke Math. J. 8, 11- 46 (1941).
3. Differentiability of regular curve families on the sphere. In: Lectures in Topology. Ann Arbor: Univ. of Michigan Press 1941, pp. 299-301.
4. The structure of a curve-family on a surface in the neighborhood of an isolated
singularity. Amer. J. Math. 64, 1-35 (1942).
5. Dynamical systems with indeterminacy. Amer. J. Math. 72, 573-594 (1950).
6. Analytic ordinary differential equations in the large. In: Proceedings U.S.J apan Seminar on Differential and Functional Equations, W. A. Harris, Jr.
and Y. Sibuya (Editors). New York-Amsterdam: Benjamin 1967, pp. 133 151.
References
201
KATO, J.
1. The asymptotic behaviour of the solutions of differential equations or the product space. Arch. Rat. Mech. Anal. 6, 133-170 (1960).
2. The asymptotic relation of two systems of ordinary differential equations. In:
Contributions to Differential Equations, vol. 3. New York: Wiley 1964,
pp. 141-161.
3. Asymptotic equivalences between systems of differential equations and their
perturbed systems. Funkcial. Ekvac. 8, 45-78 (1966).
4. Asymptotic equivalence. In: Seminar on Differential Equations and Dynamical
Systems, G. S. Jones (Editor) (Lecture Notes in Mathematics, vol. 60).
Berlin-Heidelberg-New York: Springer 1968, pp. 27 - 32.
5. A remark on the result of Strauss. In: Seminar on Differential Equations and
Dynamical Systems, G. S. Jones (Editor) (Lecture Notes in Mathematics,
vol. 60). Berlin-Heidelberg-New York: Springer 1968, pp. 89-98.
KAYANDE, A. A.; LAKSHMIKANTHAM, V.
1. Conditionally invariant sets and vector Liapunov functions. J. Math. Anal. Appl.
14, 285-293 (1966).
KELLEY, J. L.
1. General Topology. New York: Van Nostrand 1955.
KENT, J. F. III (see ENGLAND, J. W.)
KEYNES, H.
1. The proximal relation in a class of substitution minimal sets. Math. Systems
Theory 1, 165-176 (1967).
KIMURA, I. (see URA, T.)
KNAPP, A. W. (see KAHN, P. J.)
KNESER, A.
1. Studien iiber die Bewegungsvorgange in der Umgebung instabiler Gleichgewichtslagen, I, II. J. reine angew. Math. 115, 308-327 (1895); 118, 186223 (1897).
KNESER, H.
1. Uber die Losungen eines Systemes gewohnlicher Differentialgleichungen, das
der Lipschitzschen Bedingung nicht geniigt. Sitzungsber. Preuss. Akad.
Wiss., Phys.-Math. Kl., 1923, pp. 171-174.
2. Regulare Kurvenscharen auf den Ringflachen. Math. Ann. 91, 135-154 (1924).
KoLMoGoRov, A. N.; FoMIN, S. V.
1. Elements of the Theory of Functions and Functional Analysis. Rochester, N.Y.:
Graylock 1957.
KRAsovsKn, N. N. (see also BARBASHIN, E. A.)
1. Certain Problems of the Theory of Stability of Motion (Russian). Moscow :
Gos. Izd. Fiz.-mat. Lit. 1959. English translation, Stability of Motions.
Stanford: Stanford Univ. Press 1963.
2. On a problem of stability of motion in the large (Russian). Dokl. Akad. Nauk
SSSR 88, 401-404 (1953).
3. On stability of motion in the large for constantly acting disturbances (Russian).
Prikl. Mat. Mekh. 18, 95-102 (1954).
4. On the inversion of theorems of A. M. Liapunov and N. G. Chetaev on instability for stationary systems of differential equations (Russian). Prikl. Mat.
Mekh. 18, 513-532 (1954).
5. On stability in the large of the solutions of a nonlinear system of differential
equations (Russian). Prikl. Mat. Mekh. 18, 735-737 (1954).
6. Sufficient conditions for stability of solutions of a system of nonlinear differential equations (Russian). Dokl. Akad. Nauk SSSR 98, 901-904 (1954).
202
References
7. On the converse of K. P. Persidskii's theorem on uniform stability (Russian).

Prikl. Mat. Mekh. 19, 273-278 (1955).
8. On conditions of inversion of A. M. Liapunov's theorems on instability for
stationary systems of differential equations (Russian). Dokl. Akad. N auk
SSSR 101, 17-20 (1955).
9. Converse of theorems on Liapunov's second method and questions of stability
of motion in the first approximation (Russian). Prikl. Mat. Mekh. 20, 255265 (1956).
10. On the converse of theorems of the second method of A. M. Liapunov for investigation of stability of motion (Russian). Uspehi Mat. Nauk 9, no. 3,
pp. 159-164 (1956).
11. On the theory of the second method of A. M. Liapunov for the investigation
of stability (Russian). Mat. Sbornik 40, 57-64 (1956).
12. On stability with large initial perturbations (Russian). Prikl. Mat. Mekh. 21,
309-319 (1957).
KRECU, V. I. (see SIBIRSKII, K. S.)
KREIL, K. A.
1. Das qualitative Verhalten der Integralkurven einer gewohnlichen Differentialgleichung erster Ordnung in der Umgebung eines singularen Punktes. Jber.
Deutsch. Math. Verein 57, 111 (1955).
KREIN, M. G.
1. On some questions related to the ideas of Liapunov in the theory of stability.
Uspehi Mat. Nauk (N.S.) 3, no. 3, pp. 166-169 (1948).
KUDAEV, M. B.
1. The use of Liapunov functions for investigating the behavior of trajectories of
systems of differential equations (Russian). Dokl. Akad. Nauk SSSR 147,
1285-1287 (1962). English translation: Soviet Math. (Dokl.) pp. 1802-1804.
2. Classification of higher-dimensional systems of ordinary differential equations
by the method of Liapunov functions. Differencial'nye Uravnenija 1, 346356 (1965). English translation: Diff. Eqs. pp. 263-269.
3. Liapunov function for the region of influence of a single singular point of higher
order. Vestnik Moskov. Univ., 1965, no. 1, pp. 3-13.
KUHN, H. W.; SZEGO, G. P.
1. (Editors) Mathematical Systems Theory and Economics, I and II. Proceedings
of an International Summer School held in Varenna, Italy, June 1-12, 1967
(Lecture Notes in Operations Research and Mathematical Economics, vol. 11
and 12). Berlin-Heidelberg-New York: Springer 1969.
KURZWEIL, J. (see also jARNIK, J.)
1. On the reversibility of the first theorem of Liapunov concerning the stability
of motion (Czech). Czechoslovak Math. J. 5, 382-398 (1955).
2. The converse second Liapunov's theorem concerning the stability of motion
(Czech). Czechoslovak Math. J. 6, 217-259, 455-473 (1956). English translation: Amer. Math. Soc. Transl., Ser. II, 24, 19-77 (1963).
3. Invariant manifolds for flows. In: Differential Equations and Dynamical
Systems, J. K. Hale and J.P. LaSalle (Editors). New York: Academic Press
1967, pp. 431-468.
4. Invariant manifolds of differential systems. Z. angew. Math. Mech. 49, 11-14
(1969).
KURZWEIL, J.; VRKOC, I.
1. The converse theorems of Liapunov and Persidskii concerning the stability
of motion (Czech). Czechoslovak Math. J. 7, 254-274 (1957). English translation: Amer. Math. Soc. Transl., Ser. II, 29, 271-288 (1963).
References
203
LAGRANGE, J. L.
1. Mecanique analytique. Paris: Desaint 1788.
2. Oeuvres, vol. I (1867) and IV (1869). Paris: Gauthier-Villars.
LAKSHMIKANTHAM, V. (see also D'AMBROSIO, U. and KAYANDE, A. A.)
1. On the boundedness of solutions of nonlinear differential equations. Proc. Amer.
Math. Soc. 8, 1044-1048 (1957).
2. Upper and lower bounds of the norm of solutions of differential equations.
Proc. Amer. Math. Soc. 13, 615-616 (1962); 14, 509-513 (1963).
3. Notes on a variety of problems of differential systems. Arch. Rat. Mech. Anal.
10, 305-310 (1962).
4. Vector Liapunov functions and conditional stability. J. Math. Anal. Appl. 10,
368-377 (1965).
LAKSHMIKANTHAM, V. ; LEELA, S.
1. Differential and Integral Inequalities, I, II. New York-London: Academic
Press 1969.
LASALLE, J.P. (see also HALE, J. K.)
1. A study of synchronous asymptotic stability. Ann. of Math. 65, 571-581 (1957).
2. Asymptotic stability criteria. Proc. Symp. Appl. Math. 13, 299-307 (1962).
3. Some extensions of Lyapunov's second method. IRE Trans. Circuit Theory
7, 520-527 (1960).
4. The extent of asymptotical stability. Proc. Nat. Acad. Sci. USA 46, 363- 365
(1960).
5. Recent advances in Liapunov stability theory. SIAM Rev. 6, 1-11 (1964).
6. Liapunov's second method. In: Stability Problems of Solutions of Differential
Equations, A. Ghizzetti (Editor). Gubbio: Oderisi 1966.
7. An invariance principle in the theory of stability, differential equations and
dynamical systems. In: Differential Equations and Dynamical Systems,
J. K. Hale and J.P. LaSalle (Editors). New York: Academic Press 1967,
pp. 277 - 286.
8. Stability theory for ordinary differential equations. J. Diff. Eqs. 4, 57-65 (1968).
LASALLE, J.P.; LEFSCHETZ, s.
1. Stability by Liapunov's Direct Method with Applications. New York: Academic
Press 1961.
2. (Editors) Nonlinear Differential Equations and Nonlinear Mechanics. Proceedings of an International Symposium held at the U.S. Air Force Academy,
Colorado Springs, August 1961. New York: Academic Press 1963.
LAZER, A. C. (see BHATIA, N. P.)
LEBEDEV, A. A.
1. The problem of stability in a finite interval of time. Prikl. Mat. Mekh. 18, 75-94
(1954).
2. On stability of motion during a given interval of time. Prikl. Mat. Mekh. 18,
139-148 (1954).
LEE, E. B.; MARKUS, L.
1. Foundations of Optimal Control Theory. New York: Wiley 1967.
LEFSCHETZ, S. (see also LASALLE, J. P.)
1. Differential Equations: Geometric Theory. New York: Interscience-Wiley, 1st
ed. 1957, 2nd ed. 1962.
2. Liapunov and stability in dynamical systems. Bol. Soc. Mat. Mexicana 3, 25- 39
(1958).
3. Geometric differential equations: recent past and proximate future. In: Differential Equations and Dynamical Systems, J. K. Hale and J.P. LaSalle
(Editors). New York: Academic Press 1.967, pp. 1-14.
204
References
LEIGHTON, W. (see also BHATIA, N. P.)

1. Morse theory and Liapunov functions. Rend. Circ. Mat. Palermo, Ser. II, 13,
1-10 (1966).
LEVIN, J. J.
1. On the global asymptotic behavior of nonlinear systems of differential equations. Arch. Rat. Mech. Anal. 6, 65-74 (1960).
LEVINSON, N. (see CoDDINGTON, E. A.)
LEWIS, D. C., Jr.
1. Metric properties of differential equations. Amer. J. Math. 71, 294-312 (1949).
2. Differential equations referred to a variable metric. Amer. J. Math. 73, 48-58
(1951).
3. Reversible transformations. Pacific J. Math. 11, 1077-1087 (1961).
LIAPUNOV, A. M.
1. Probleme general de la stabilite du mouvement (Annals of Math. Studies,
no. 17). Princeton: Princeton Univ. Press 1947 [reproduction of the French
translation, Ann. de la Faculte des Sciences de Toulouse 9, 203-474 (1907);
of a Russian memoir, Obshchaya Zadacha Ustoichivosti Dvizheniya, Kharkov, 1892, and of a note, Comm. Soc. Math. Kharkov 3, 265-272 (1893)].
English translation, Stability of Motion. New York: Academic Press 1966.
LIPSCHITZ, R.
1. Sur la possibilite d'integrer completement un systeme donne d'equations diffe
rentielles. Bull. Sci. Math. Astron. 10, 149-159 (1876).
LOJASIEWICZ, s.
1. Sur !'allure asymptotique des integrales du systeme d'equations differentielles
au voisinage du point singulier. Ann. Polon. Math. 1, 34-72 (1954).
LOVINGOOD, J. A.
1. A special class of dynamical polysystems. J. Diff. Eqs. 6, 326-336 (1969).
MALKIN I. G.
1. Theorie der Stabilitat einer Bewegung. Miinchen: Oldenbourg 1959 (German
translation of the Russian original published in 1952). (A very poor English
translation of this book is: Theory of Stability of Motion. Atomic Energy
Commission, Transl. no. 3352, Dept. of Commerce, Washington, D.C., 1958.)
2. Das Existenzproblem von Ljapunovschen Funktionen. Izv. fiz.-mat. Obshch.
Kazan III, 4, 51-62 (1930); 5, 63-84 (1931).
3. Certain questions in the theory of stability of motion in the sense of Liapunov.
Amer. Math. Soc. Transl. no. 20, 1950.
4. Verallgemeinerung des Fundamentalsatzes von Liapunoff iiber die Stabilitat
der Bewegungen. C. R. (Dokl.) Acad. Sci. URSS 18, 162-164 (1938).
5. On the stability of motion in the sense of Liapounov (Russian). Mat. Sbornik
3, 47-100 (1938).
6. Sur un theoreme d'existence de Poincare-Liapounoff. C.R. (Dokl.} Acad. Sci.
URSS 27, 307-310 (1940).
7. Basic theorems of the theory of stability of motion (Russian). Prikl. Mat. Mekh.
6, 411-448 (1942).
8. Stability in the case of constantly acting disturbances (Russian). Prikl. Mat.
Mekh. 8, 241-245 (1944).
I
MARCHAUD, M. A.
1. Surles champs des demi-droites etles equations differentielles du premier ordre.
Bull. Soc. Math. France 63, 1-38 (1934).
2. Sur les champs continus de demi-c6nes convexes et leur integrales. Compositio
Math. 3, 89-127 (1936).
References
205
MARKOV, A. A.
1. Sur une propriete generale des ensembles minimaux de M. Birkhoff. C. R. Acad.
Sci. Paris 193, 823-825 (1931).
2. On a general property of minimal sets (Russian). Rusk. Astron. Zh. 1932.
3. Stabilitat im Liapunoffschen Sinne und Fastperiodizitat. Math. Z. 36, 708-738
(1933).
4. Almost periodicity and harmonizability (Russian). L. Trudy V torogo V sesoyuzn.
Mat. S'ezda 2, 227-231 (1936).
MARKUS, L. (see also AEPPLI, A. and LEE, E. B.)
1. Escape times for ordinary differential equations. Rend. Sem. Mat. Politec.
Torino 11, 271-277 (1952).
2. On completeness of invariant measures defined by differential equations. J.
Math. Pure Appl. 31, 341-353 (1952).
3. Invariant measures defined by differential equations. Proc. Ann. Math. Soc.
4, 89-91 (1953).
4. A topological theory for ordinary differential equations in the plane. Colloque
de Topol. et Geom. Diff., Strasbourg 1952.
5. Global structure of ordinary differential equations in the plane. Trans. Amer.
Math. Soc. 76, 127-148 (1954).
6. Asymptotically autonomous differential systems. In: Contributions to the
Theory of Nonlinear Oscillations, vol. 3 (Annals of Math. Studies, no. 36).
Princeton: Princeton Univ. Press 1956, pp. 17-30.
7. Structurally stable differential systems. Ann. of Math. 73, 1-19 (1961).
8. Periodic solutions and invariant sets of structurally stable differential systems.
In: Symposium Internacional de Ecuaciones Diferenciales Ordinarias,
Mexico, 1961, pp. 190-194.
9. The global theory of ordinary differential equations. Lecture Notes, University
of Minnesota, 1964/65.
10. Dynamical systems on group manifolds. In: Differential Equations and Dynamical Systems, J. K. Hale and J.P. LaSalle (Editors). New York: Academic
Press 1967.
11. Parallel dynamical systems. Topology 8, 47-57 (1969).
MARKUS, L.; Y AMABE, H.
1. Global stability criteria for differential systems. Osaka Math. J. 12, 305-317
(1960).
MARTIN, M.
1. A problem in arrangements. Bull. Amer. Math. Soc. 40, 859-864 (1934).
MASSERA, J. L.
1. On Liapunoff's condition of stability. Ann. of Math. 50, 705-721 (1949).
2. Contributions to stability theory. Ann. of Math. 64, 182-206 (1956); correction
68, 202 (1958).
3. On the existence of Liapunov functions. Puhl. Inst. Mat. Estad. Montevideo 3,
no. 4, pp. 111-124 (1960).
4. Converse theorems of Liapunov' s second method. In: Symposium Internacional
de Ecuaciones Diferenciales Ordinarias, Mexico, 1961, pp. 158- 163.
MASSERA, J. L.; SCHAFFER, J. J.
1. Linear Differential Equations and Function Spaces. New York: Academic Press
1966.
MATROSOV, v. M.
1. On the stability of motion. Prikl. Mat. Mekh. 26, 885-895 (1962). English
translation: J. Appl. Math. Mech. 26, 1337-1353 (1962).
206
References
2. On the theory of stability of motion. Prikl. Mat. Mekh. 26, 992-1002 (1962).
English translation: J. Appl. Math. Mech. 26, 1560-1522 (1962).
MENDELSON, p.
1. On Lagrange stable motions in the neighborhood of critical points. In: Contributions to the Theory of Nonlinear Oscillations, vol. 5 (Annals of Math.
Studies, no. 45). Princeton: Princeton Univ. Press 1960, pp. 219-224.
2. On unstable attractors. Bol. Soc. Mat. Mexicana 5, 270-276 (1960).
MICHAEL, E.
1. Topologies on spaces of subsets. Trans. Amer. Math. Soc. 71, 152-182 (1951).
MILLER, R. K.
1. On almost periodic differential equations. Bull. Amer. Math. Soc. 70, 792- 795
(1964).
2. Asymptotic behavior of solutions of nonlinear differential equations. Trans.
Amer. Math. Soc. 115, 400-416 (1965).
3. Almost periodic differential equations as dynamical systems with applications
to the existence of a.p. solutions. J. Diff. Eqs. 1, 337-345 (1965).
4. The topological dynamics of Volterra integral equations (Advances in Differential and Integral Equations). SIAM, Philadelphia, 1969.
MILLER, R. K.; SELL, G. R.
1. A note on Volterra integral equations and topological dynamics. Bull. Amer.
Math. Soc. 74, 804-809 (1968).
MILNOR,
J.
1. Morse Theory (Annals of Math. Studies, no. 51). Princeton: Princeton Univ.
Press 1963.
2. Sommes de varietes differentiables et structures differentiables des spheres.
3. Differential Topology (Lectures on Modern Mathematics, vol. II), T. L. Saaty
(Editor). New York: Wiley 1964.
MINKEVICH, M.
I.
1. The theory of integral funnels in generalized dynamical systems without a

hypothesis of uniqueness (Russian). Dokl. Akad. Nauk SSSR 59, 1049-1052
(1948).
2. Closed integral funnels in generalized dynamical systems without a hypothesis
of uniqueness (Russian). Dokl. Akad. Nauk SSSR 60, 341-343 (1948).
3. Theory of integral funnels in dynamical systems without uniqueness (Russian).
Uchen. Zap. Moskov. Univ., no. 135, Mat. 2, 134-151 (1948).
4. Closed integral funnels in generalized dynamical systems without a hypothesis
of uniqueness (Russian). Uchen. Zap. Moskov. Univ., no. 6, Mat. 163, 73-88
(1952).
MINORSKY,
N.
1. Nonlinear Oscillations. Princeton: Van Nostrand 1962.

MIRANDA, C.
1. Un'osservazione su un teorema di Brouwer. Boll. Un. Mat. Ital. 3, 5-7 (1940).
MoISEEv, N. D.
1. Summary of the History of Stability (Russian). Moscow 1949.
MONTEL, P.
1. Sur les suites infinies de fonctions. Ann. Sci. Ecole Norm. Sup., Ser. III, 24,
233-234 (1907).
2. Sur !'integral superieure et l'integrale inferieure d'une equation differentielle.
MONTGOMERY, D.; ZIPPIN, L.
1. Topological Transformation Groups. New York: Interscience-Wiley 1955.
References
207
M.
MORSE,
1. A one-to-one representation of geodesics on a surface of negative curvature.

Amer. J. Math. 43, 35-51 (1921).
2. Recurrent geodesics on a surface of negative curvature. Trans. Amer. Math. Soc.
22, 84-100 (1921).
3. Relations between the critical points of a real function of n independent variables. Trans. Amer. Math. Soc. 27, 345-396 (1925).
4. Symbolic Dynamics. Lecture Notes, Dept. of Mathematics, Princeton University, 1966.
MOSER,
J.
1. On a theorem of Anosov. J. Diff. Eqs. 5, 411-440 (1969).

MOSTERT, P. s.
1. One-parameter transformation groups in the plane. Proc. Amer. Math. Soc. 9,
462-463 (1958).
MOLLER,
M.
1. Uber das Fundamentaltheorem in der Theorie der gewohnlichen Differentialgleichungen. Math. Z. 26, 619-645 (1927).
2. Beweis eines Satzes des Herrn H. Kneser iiber die Gesamtheit der LOsungen, die
ein System gewohnlicher Differentialgleichungen <lurch einen Punkt schickt.
Math. Z. 28, 349-355 (1928).
3. Neuere Untersuchung iiber den Fundamentalsatz in der Theorie der gewohnlichen Differentialgleichungen. Jber. Deutsch. Math. Verein. 37, 33-48
(1928).
MYSHKIS,
A. D.
1. Generalizations of the theorem on a fixed point of a dynamical system inside of a

closed trajectory (Russian). Mat. Sbornik 34, 525- 540 ( 1954).
NAGUMO,
M.
1. Eine hinreichende Bedingung fiir die Unitat der Losung von Differentialgleichungen erster Ordnung. Japan J. Math. 3, 107-112 (1926).
2. Un theoreme relatif a !'ensemble des courbes integrales d'un systeme d'equations differentielles ordinaires. Proc. Phys. Math. Soc. Japan, III, 12, 233239 (1930).
NAGY,
J.
1. Lyapunov's direct method in abstract local semi-flows. Comm. Math. Univ.

Carolinae 8, 257-266 (1957).
2. Stability in continuous local semi-flows. Casopis Pest. Mat. 93, 8-21 (1968).
3. Stability of sets with respect to abstract processes. In: Mathematical Systems
Theory and Economics II, H. W. Kuhn and G. P. Szego (Editors) (Lecture
Notes in Operations Research and Mathematical Economics, vol. 12).
Berlin-Heidelberg-New York: Springer 1969, pp. 355-378.
V. V.
1. Sur les systemes dynamiques instables. C. R. Acad. Sci. Paris 199, 19-20
(1934).
2. Uber vollstandig unstabile dynamische Systeme. Ann. Mat. Pura Appl. 14,
275-286 (1936).
3. Surles systemes de courbes remplissant un espace metrique. C.R. (Dokl.) Acad.
Sci. URSS 21, 99-102 (1938).
4. Surles systemes de courbes remplissant un espace metrique (Generalisation des
theoremes de Birkhoff). Rec. Math. (Mat. Sbornik) 6, 283-292 (1939).
5. Systemes dynamiques sur une multiplicite integrale limite. C. R. (Dokl.) Acad.
Sci. URSS 36, 535-538 (1945).
NEMYTSKII,
208
References
6. Surles familles de courbes du type de Bendixson. C.R. (Dokl.) Acad. Sci. URSS
21, 103-105 (1938).
7. Les systemes dynamiques generaux. C.R. (Dokl.) A.cad. Sci. URSS 53, 491-494
(1946).
8. On the theory of orbits of general dynamical systems (Russian). Mat. Sbornik
23, 161-186 (1948).
9. The structure of one-dimensional limiting integral manifolds in the plane and
three-dimensional space (Russian). Vestnik Moskov. Univ., no. 10, 1948,
pp. 49-61.
10. Topological problems of the theory of dynamical systems (Russian). Uspehi
Mat. Nauk 4, 91-153 (1949). English translation: Amer. Math. Soc. Transl.
no. 103, p. 85 (1954).
11. Generalizations of the theory of dynamical systems (Russian). Uspehi Mat.
Nauk 5, 47-49 (1950).
12. Some general theorems on the distribution of integral curves in the plane.
Vestnik Moskov. Univ., Ser. I Mat. Mekh. no. 6, pp. 3-10 (1960).
13. Sur une classe importante des systemes caracteristiques sur le plan. Ann. Mat.
Pura Appl. 49, 11-24 (1960).
14. Some modern problems in the qualitative theory of ordinary differential equations. Russian Math. Survey, 1965, pp. 1-34.
15. Topological classification of singular points and generalized Liapunov functions.
Differencial'nye Uravnenija 3, 359-370 (1967).
NEMYTSKII, V. V.; STEPANOV, V. V.
1. Qualitative Theory of Differential Equations (Russian). Moscow-Leningrad,
1st ed. 1947, 2nd ed. 1949. English translation, Princeton: Princeton Univ.
Press 1960.
OKAMURA, H.
1. Condition necessaire et suffisante remplie par les equations differentielles
ordinaires sans points de Peano. Mem. Coll. Sci. Kyoto Imp. Univ., Ser. A,
24, 21-28 (1942).
OLECH, Cz. (see also HARTMAN, P. and SzEGO, G. P.)
1. On the asymptotic behavior of the solutions of a system of ordinary nonlinear
differential equations. Bull. Acad. Polon. Sci., Cl. III, 4, 555-561 (1956).
2. Remarks concerning criteria for uniqueness of solutions of ordinary differential
equations, Bull. Acad. Polon. Sci., Cl. III, 8, 661-666 (1960).
3. On the global stability of autonomous systems in the plane. In: Contributions
to Differential Equations, vol. 1. New York: Wiley 1963, pp. 389-400.
0NUCHic, N. (see also HARTMAN, P.)
1. Applications of the topological method of Wazewski to certain problems of
asymptotic behavior in ordinary differential equations. Pacific J. Math. 11,
1511-1527 (1961).
2. Relationships among the solutions of two systems of ordinary differential equations. Michigan J. Math. 10, 129_: 139 (1963).
0PIAL, Z.
1. Sur !'allure asymptotique des solutions de certaines equations differentielles de
la mecanique nonlineaire. Ann. Polon. Math. 8, 105-124 (1960).
2. Sur la dependence des solutions d'un systeme d'equations differentielles de leurs
seconds membres. Ann. Polon. Math. 8, 75-89 (1960).
OSGOOD, w.
1. Beweis der Existenz einer Losung der Differentialgleichung dy/dx = f (x, y)
ohne Hinzunahme der Cauchy-Lipschitzschen Bedingung. Monatsh. Math.
Phys. 9, 331-345 (1898).
References
209
0XTOBY, J.C.
1. Stepanoff flows on the torus. Proc. Amer. Math. Soc. 4, 982-987 (1953).
PAPUSH, N. P.
1. A study of the distribution of integral curves occupying a domain containing
one singular point. Mat. Sbornik 38, 337 (1956).
PARRY, W. (see HAHN, F.)
PEANO, G.
1. Sull'integrabilita delle equazioni differenziali di primo ordine. A tt. R. Accad.
Torino 21, 677-685 (1885/86).
2. Demonstration de l'integrabilite des equations differentielles ordinaires. Math.
Ann. 37, 182-228 (1890).
PEIXOTO, M. C.; PEIXOTO, M. M.
1. Structural stability in the plane with enlarged boundary conditions. An. Acad.
Brasil Ci. 31, 135-160 (1959).
PEIXOTO, M. M. (see also PEIXOTO, M. C.)
1. On structural stability. Math. 69, 199-222 (1959).
2. Some examples on n-dimensional structural stability. Proc. Nat. Acad. Sci.
USA 45, 633-636 (1959).
3. Structural stability on two-dimensional manifolds. In: Symposium Internacional de Ecuaciones Diferenciales Ordinarias, Mexico, 1961, pp. 188-189.
4. Structural stability on two-dimensional manifolds. Topology 1, 101-120 (1962).
5. Qualitative theory of differential equations and structure stability. In: Differential Equations and Dynamical Systems, J. K. Hale and J. P. LaSalle
(Editors). New York: Academic Press 1967, pp. 469-480.
PEROV,A.I.
1. Investigation of the neighborhood of a critical point of a multi-dimensional
differential equation in the analytic case (Russian). Dokl. Akad. N auk SSSR
166, 544-547 (1966). English translation: Soviet Math. (Dokl.) 7, 122-125
(1966).
2. On the structure of a limit set (Russian). Dokl. Akad. Nauk SSSR 176, 526-529
(1967). English translation: Soviet Math. (Dokl.) 8, 1142-1145 (1967).
PERRON, 0.
1. Uber Ein- und Mehrdeutigkeit des Integrales eines Systems von Differentialgleichungen. Math. Ann. 95, 98-101 (1926).
2. Uber Existenz und Nichtexistenz von Integralen partieller Differentialgleichungssysteme im reellen Gebiet. Math. Z. 27, 549-564 (1928).
3. Eine hinreichende Bedingung fur die Unitat der Losung von Differentialgleichungen erster Ordnung. Math. Z. 28, 216-219 (1928).
4. Uber Stabilitat und asymptotisches Verhalten der Integrale von Differentialgleichungssystemen. Math. Z. 29, 129-160 (1928).
5. Uber Stabilitat und a&ymptotisches Verhalten der Losungen eines Systems
endlicher Differenzengleichungen. J. reine angew. Math. 161, 41- 61 (1929).
6. Die Stabilitatsfrage bei Differentialgleichungen. Math. Z. 32, 703- 728 (1930).
PERSIDSKII, K. P.
1. Au sujet du probleme de stabilite. Bull. Soc. phys.-math. Kazan III, 5, no. 3,
pp. 56-62 (1931).
2. Un theoreme sur la stabilite du mouvement. Bull. Soc. phys.-math. Kazan III,
6, pp. 76-79 (1934).
3. On the stability theory of the solutions of systems of differential equations
(Russian). Bull. Soc. phys.-math. Kazan III, 8 (1936).
4. On a theorem of Liapunov (Russian). C.R. (Dokl.) Acad. Sci. URSS 14, 541544 (1937).
14
210
References
5. On the theory of stability of solutions of differential equations (Russian). Thesis,

Moscow 1946. Summary: UspehiMat.Naukl,no.1, 5-6,pp.250-255 (1946).
6. On the stability of the solutions of an infinite system of equations (Russian).
Prikl. Mat. Mekh. 12, 597-6'12 (1948).
7. On stability of solutions of a system of countably many differential equations
(Russian). Izv. Akad. Nauk Kazach. SSR 56, Ser. Mat. Mekh., no. 2, pp. 335 (1948).
8. Countable systems of differential equations and the stability of their solutions
(Russian). Uchen. Zap. Kazach. Univ. Mat. Fiz. no. 2 (1949).
9. On stability of solutions of differential equations (Russian). Izv. Akad. Nauk
Kazach. SSR 60, Ser. Mat. Mekh., no. 4, pp. 3-18 (1950).
10. On Liapunov's second method in linear normed spaces (Russian). Vestnik Akad.
Nauk Kazach. SSR, no. 7, pp. 89-97 (1958).
11. Inversion of Liapunov's second theorem on instability in linear normed spaces
(Russian). Vestnik Akad. Nauk Kazach. SSR, no. 10, pp. 31-35 (1959).
PERSIDSKII, s. K.
1. On the second method of Liapunov (Russian). Izv. Akad. Nauk Kazach. SSR,
no. 4, pp. 43-47 (1956).
2. On stability in a finite interval (Russian). Vestnik Akad. Nauk Kazach. SSR,
no. 9, pp. 75-80 (1959).
3. Some theorems on the second method of Liapunov (Russian). Vestnik Akad.
Nauk Kazach. SSR, no. 2, pp. 70-76 (1960).
4. On Liapunov's second method (Russian). Prikl. Mat. Mekh. 25, 17-23 (1961).
English translation: J. Appl. Math. Mech. 25, 20-28 (1961).
PETROVSKII, I. G.
1. Lectures on the theory of ordinary differential equations (Russian). 5th augmented ed. Moscow: Izd. "Nauka" 1964, p. 272. English translation, Englewood Cliffs, N.J.: Prentice Hall 1966.
PICARD, E.
1. Lecons sur quelques equations fonctionnelles. Paris: Gauthier-Villars 1928.
Pus, A.
1. On a topological method for studying the behavior of the integrals of ordinary
differential equations. Bull. Acad. Polon. Sci., Cl. III, 2, 415-418 (1954).
2. Characteristics of nonlinear partial differential equations. Bull. Acad. Polon.
Sci., Cl. III, 2, 419-422 (1954).
3. Sets filled by asymptotic integrals of ordinary differential equations. Bull.
Acad. Polon. Sci., Cl. III, 4, 749-752 (1956).
4. On sets filled by asymptotic solutions of differential equations. Ann. Inst. Fourier (Grenoble) 14, 191-194 (1966).
Puss, V. A.
1. Certain Problems of the Theory of Stability of Motion in the Whole. Leningrad:
Izd. Leningradsk. Univ. 1958.
PoINCARE, H.
1. Oeuvres. Paris: Gauthier-Villars 1929.
2. Les methodes nouvelles de la mecanique celeste. Paris: Gauthier-Villars 1892/99.
Reprint, New York: Dover 1960.
3. Memoire sur les courbes definies par une equation differentielle. J. de Math. 7,
375-422 (1881); 8, 251-296 (1882); 11, 187-244 (1885); 12, 151-217 (1886).
PUGH,
c. c.
1. Cross-sections of solution funnels. Bull. Amer. Math. Soc. 70, 580-583 (1964).
2. The closing lemma and structural stability. Bull. Amer. Math. Soc. 70, 584-587
(1964).
References
211
PUTNAM, C. R.
1. Unilateral stability and almost periodicity. J. Math. Mech. 9, 915-917 (1960).
REEB, G.
1. Sur certaines proprietes topologiques des trajectoires des systemes dynamiques.
Acad. Roy. Belg., Cl. Sci. Mem. 27, no. 9 (1952).
2. Sur la theorie generale des systemes dynamiques. Ann. Inst. Fourier (Grenoble)
6, .89-115 (1955/56).
3. Sur certains problemes de topologie algebrique et de topologie generale en dynamique. Bol. Soc. Mat. Mexicana 5, 199-202 (1960).
REISSIG, R.
1. Kriterien fiir die Zugehorigkeit dynamischer Systeme zur Klasse D. Math.
Nachr. 20, 67-72 (1959).
2. Stabilitatsprobleme in der qualitativen Theorie der Differentialgleichungen.
Jber. Deutsch. Math. Verein. 63, 97-116 (1960).
3. N eue Probleme und Methoden aus der qualitativen Theorie der Differentialgleichungen. Monatsber. Deutsche Akad. Wiss. Berlin 2, 1-8 (1960).
REISSIG, R.; SANSONE, G.; CONTI, R.
1. Qualitative Theorie nichtlinearer Differentialgleichungen. Pubblicazioni dell'Istituto Nazionale di Alta Matematica. Roma: Cremonese 1963.
REMAGE, R., Jr.
1. On minimal sets in the plane. Proc. Amer. Math. Soc. 13, 41-47 (1962).
ROBLIN, V. A.
1. Selected topics from the metric theory of dynamical systems (Russian). Uspehi
Mat. Nauk 9, 57-128 (1949). English translation: Amer. Math. Soc. Transl.
49, 171-240 (1965).
2. New progress in the theory of transformations with invariant measure (Russian).
Uspehi Mat. Nauk 15, 3-26 (1960). English translation: Russian Math.
Surveys 15, 1-22 (1960).
3. Lectures on the theory of entropy of transformations with invariant measure
(Russian). Uspehi Mat. Nauk 22, 3-56 (1967). English translation: Russian
Math. Surveys 22, 1-52 (1967).
Roos, B. W.
1. Note on generalized dynamical systems. SIAM Rev. 6, 269-274 (1964).
ROSEAU, M.
1. Vibrations nonlineaires et theorie de la stabilite (Springer Tracts in Nat. Phil.,
vol. 8). Berlin-Heidelberg-New York: Springer 1966.
RoxIN, E.
1. Reachable zones in autonomous differential systems. Bol. Soc. Mat. Mexicana
5' 125-135 (1960).
2. Stability in general control systems. J. Diff. Eqs. 1, 115-150 (1965).
3. On generalized dynamical systems defined by contingent equations. J. Diff.
Eqs. 1, 188-205 (1965).
4. On stability in control systems. SIAM J. Control 3, 357-372 (1966).
5. Local definition of generalized control systems. Michigan Math. J. 13, 91-96
(1966).
Roxrn, E. O.; SPINADEL, V. W.
1. Reachable zones in autonomous differential systems. In: Contributions to Differential Equations, vol. 1. New York: Wiley 1963, pp. 275-315.
SACKER, R. J.
1. A new approach to the perturbation theory of invariant surfaces. Comm. Pure
Appl. Mat. 18, 712-732 (1965).
14*
212
References
SAITO, T.
1. On the measure-preserving flow on the torus. J. Math. Soc. Japan 3, 279-286
(1951).
2. On dynamical systems in n-dimensional torus. Funkcial. Ekvac. 7, 91-120
(1965).
3. On the flow outside an isolated minimal set. In: Proceedings U.S.-Japan
Seminar on Differential and Functional Equations. New York-Amsterdam:
Benjamin 1967, pp. 301-312.
4. Isolated minimal sets. Funkcial. Ekvac. 11, 155-167 (1969).
SANDOR,ST.;V\TEXLER,D.
1. Sur la stabilite dans les systemes dynamiques. Rev. Roumaine Math. Pures
Appl. 3, 325-328 (1958).
SANSONE, G. (see also REISSIG, R.)
1. Equazioni differenziali nel campo reale. Bologna: Zanichelli 1948.
SANSONE, G.; CONTI, R.
1. Equazioni differenziali non lineari. Roma: Cremonese 1956. English translation,
Nonlinear Differential Equations. London: Pergamon Press 1965.
SARD, A.
1. The measure of the critical values of differentiable maps. Bull. Amer. Math. Soc.
48, 883- 896 (1942).
SCHAFFER, J. J. (see MASSERA, J. L.)
ScHAUDER, J.
1. Der Fixpunktsatz in Funktionalraumen. Studia Math. 2, 171-180 (1930).
ScHwARTZ, A. J.
1. A generalization of a Poincare-Bendixson ~heorem to closed two-dimensional
manifolds. Amer. J. Math. 85, 453-458 (1963); errata p. 753.
2. Flows on the solid torus asymptotic to the boundary. J. Diff. Eqs. 4, 316-326
(1968).
SCHWARTZMAN, S.
1. Asymptotic cycles. Ann. Math. 66, 270-284 (1957).
2. On the existence of strongly recurrent and periodic orbits. Bol. Soc. Mat. Mexicana 5, 181-183 (1960).
3. Global cross sections of compact dynamical systems. Proc. Nat. Acad. Sci.
USA 48, 786-791 (1962).
SCHWEIGERT, G. E.
1. A note on the limits of orbits. Bull. Amer. Math. Soc. 46, 963-969 (1940).
SEIBERT, P. (see also AUSLANDER, J .)
1. Prolongations and generalized Liapunov functions. Tech. Rep. 61-7. RIAS,
Baltimore, 1961.
2. Zurn Problem der Stabilitat unter standig wirkenden Storungen bei dynamischen Systemen. Arch. Math. (Basel) 15, 108-114 (1964).
3. A concept of stability in dynamical systems. In: Topological Dynamics,
J. Auslander and VV. H. Gottschalk (Editors). New York-Amsterdam: Benjamin 1968, pp. 423-433.
SEIFERT, G.
1. Stability conditions for the existence of almost-periodic solutions of almostperiodic systems. J. Math. Anal. Appl. 10, 409-418 (1965).
2. Recurrence and almost perodicity in ordinary differential equations (Advances
in Differential and Integral Equations). SIAM, Philadelphia, 1969.
SELL, G. R. (see also DEYSACH, L. G. and MILLER, R. K.)
1. Stability theory and Liapunov's second method. Arch. Rat. Mech. Anal. 14,
108-126 (1963).
References
213
2. A note on the fundamental theory of ordinary differential equations. Bull.

Amer. Math. Soc. 70, 529-535 (1964).
3. On the fundamental theory of ordinary differential equations. J. Diff. Eqs. 1,
370-392 (1965).
4. Periodic solutions and asymptotic stability. J. DHf. Eqs. 2, 143-157 (1966).
5. Nonautonomous differential equations and topological dynamics, I, II. Trans.
Amer. Math. Soc. 127, 241-283 (1967).
6. Invariant measures and Poisson stability. In: Topological Dynamics, J. Auslander and W. H. Gottschalk (Editors). New York-Amsterdam: Benjamin
1968, pp. 435-454.
SELL, G. R.; SIBUYA, Y.
1. Behavior of solutions near a critical point. In: Proceedings U.S.-Japan Seminar
on Differential and Functional Equations. New York-Amsterdam: Benjamin
1967, pp. 501-506.
SHCHERBAKOV, B. A. (see also BRONSTEIN, I. U.)
1. Classifications of motions stable in the sense of Poisson: pseudorecurrent
motions (Russian). Dokl. Akad. Nauk SSSR 146, 322-324 (1962). English
translation: Soviet Math. (Dokl.), pp. 1320-1322.
2. On classes of motions stable in the sense of Poisson. Pseudorecurrent motions
(Russian). Izv. Akad. NaukMoldav. SSR, Ser. Estestven. Tekh. Nauk, no. 1,
pp. 58-72 (1963).
3. Decomposition of a set of Poisson-stable motions (Russian). Dokl. Akad.
Nauk SSSR 152, 71-74 (1963).
4. Constituent classes of Poisson-stable motions (Russian). Sibirsk. Mat. Zh. 5,
1397-1417 (1964).
5. Dynamical systems: review of papers given at the Kishinev seminar on the
qualitative theory of differential equations. Differencial'nye Uravnenija 1,
260-266 (1965).
6. Minimal motions and the structure of minimal sets (Russian). Papers on Algebra
and Analysis, Kishinev, 1965, pp. 99-110.
7. On a class of motions stable in Poisson's sense (Russian). Papers on Algebra and
Analysis, Kishinev, 1965, pp. 155-160.
8. Recurrent solutions of differential equations and the general theory of dynamical systems. Differencial'nye Uravnenija 3, 1450-1460 (1967).
SHoLoHovrcH, F. A. (see also BARBASHIN, E. A.)
1. The relationship between a linear dynamical system and a certain differential
equation in Banach space (Russian). Dokl. Akad. Nauk SSSR 120, 43-46
(1958).
2. Linear dynamic systems (Russian). Izv. Vyssh. Uchebn. Zaved., Mat. 1957,
no. 1, pp. 249-257.
SHUBIN, M.A.
1. On some properties of generalized ro-limit sets in dynamical systems (Russian).
Vestnik Moskov. Univ., no. 3, 58-60 (1966).
SIBIRSKII, K. s.
1. Uniform approximation of points of dynamical limit sets and motions in them
(Russian). Dokl. Akad. Nauk SSSR 146, 307-309 (1962). English translation: Soviet Math. (Dokl.) 3, 1304-1305 (1962).
2. Uniform approximation of points and properties of motions in dynamical limiting sets (Russian). Izv. Akad. Nauk Moldav. SSR, Ser. Estestven. Tekh.
Nauk, no. 1, pp. 38-48 (1963).
214
References
SIBIRSKII, K. s.; BRONSTEIN' I. u.

1. Partially ordered group dynamical systems (Russian). Uchen. Zap. Kishinev.
Univ., no. 54, pp. 33-36 (1960).
SIBIRSKII, K. S.; KRECU, v. I.; BRONSTEIN, I. u.
1. Liapunov stability in partially ordered dynamical systems (Russian). Uchen.
Zap. Kishinev. Univ., no. 54, pp. 29-32 (1960).
SIBIRSKII, K. S.; STAKHI, A. M.
1. Limit properties of partially ordered dispersive dynamical systems (Russian).
Izv. Akad. Nauk Moldav. SSR, Ser. Estestven. Tekh. Nauk, no. 11, pp. 4249 (1963).
2. On partial ordering of groups (Russian). Papers on Algebra and Analysis, Kishinev, 1964.
SIBUYA, Y. (see also HARRIS, W. A. Jr. and SELL, G. R.)
SIEGEL, c. L.
1. Note on differential equations on the torus. Ann. Math. 46, 423-428 (1945).
SINAI, YA. G.
1. Dynamical systems with countably-multiple Lebesque spectrum, I. Izv. Akad.
Nauk SSSR, Ser. Mat. 25, 899-926 (1961). English translation: Amer. Math.
Soc. Transl. 39, 83-110 (1964).
2. Dynamical systems with countably-multiple Lebesque spectrum, II. Izv. Akad.
Nauk SSSR, Ser. Mat. 30, 15-68 (1966). English translation: Amer. Math.
Soc. Transl. 68, 34-88 (1968).
SMALE, ST.
1. Morse inequalities for a dynamical system. Bull. Amer. Math. Soc. 66,
43-49 (1960).
2. On gradient dynamical systems. Ann. of Math. 74, 199-206 (1961).
3. On dynamical systems. In: Symposium Internacional de Ecuaciones Diferenciales Ordinarias, Mexico, 1961, pp. 195-198.
4. Generalized Poincare conjecture in dimensions greater than four. Ann. of Math.
74, 361-406 (1961).
5. Stable manifolds for differential equations and diffeomorphisms. Ann. Scuola
Norm. Sup. Pisa, Ser. 3, 17, 97-117 (1963).
6. Diffeomorphisms with many critical points. In: Differential and Combinatorial
Topology, S.S. Cairns (Editor). Princeton: Princeton Univ. Press 1965,
pp. 63-80.
7. Dynamical Systems on n-dimensional manifolds. In: Differential Equations
and Dynamical Systems, J. K. Hale and J.P. LaSalle (Editors). New York:
Academic Press 1967, pp. 483-486.
SoLNCEV, Yu. K.
1. Two examples of dynamical systems determined by infinite systems of differential equations (Russian). Uchen. Zap. Moskov. Gos. Univ., Mat. 155, 156167 (1952).
SPINADEL, V. W. (see Roxrn, E. 0.)
STAKHI, A. M. (see SIBIRSKII, K. S.)
STALLINGS, J.
1. Polyhedral homotopy sphere. Bull. Amer. Math. Soc. 66, 485-488 (1960).
STEPANOV, V. V. (see also BEBUTov, M. V. and NEMYTSKII, V. V.)
1. Lehrbuch der Differentialgleichungen. Berlin: VEB Deutscher Verlag der.Wiss.
1956 (German translation of the Russian edition, Gos. Izd. Tekh. Teor. Lit.,
Moscow 1953).
STERNBERG, SH. (see also BRAUER, F.)
1. On differential equations on the torus. Amer. J. Math. 79, 397-402 (1957).
References
215
2. On Poincare's last geometrical theorem. Proc. Amer. Math. Soc. 8, 787-789

(1957).
3. On the structure of local homeomorphisms of euclidean n-space. Amer. J.
Math. 80, 623-631 (1958).
STOKES, A. P. (see HALE, J. K.)
STRAUSS, A.
1. Continuous dependence of solutions of ordinary differential equations. Amer.
Math. Monthly 71, 649-652 (1964).
2. Liapunov functions and LP solutions of differential equations. Trans. Amer.
Math. Soc. 119, 37-50 (1965).
3. Liapunov functions and global existence. Bull. Amer. Math. Soc. 71, 519-520
(1965).
4. On the stability of a perturbed nonlinear equation. Proc. Amer. Math. Soc. 17,
803-807 (1966).
5. A geometrical introduction to Liapunov's second method. In: Stability Problems of Solutions of Differential Equations, A. Ghizzetti {Editor). Gu bbio:
Oderisi 1966, pp. 19-27.
6. A note on a global existence result of R. Conti. Boll. Un. Mat. Ital. 22, 439-441
(1967).
7. Perturbing asymptotically stable differential equations (Advances in Differential and Integral Equations). SIAM, Philadelphia, 1967.
STRAUSS, A.; YORKE, J. A.
1. Perturbation theorems for ordinary differential equations. J. Diff. Eqs. 3, 1530 (1967).
2. On asymptotically autonomous differential equations. Math. Systems Theory
75-82 (1967).
3. Perturbing asymptotically stable differential equations. Bull. Amer. Math. Soc.
74, 992-996 (1968).
4. On the fundamental theory of differential equations. SIAM Rev. 11, 236-246
(1969).
5. Perturbing uniform-asymptotically stable nonlinear systems. J. Diff. Eqs. 6,
452-483 (1969).
6. Perturbing uniformly stable linear systems with and without attraction. SIAM
J. Appl. Math. 17, 725-739 (1969).
7. Identifying perturbations which preserve asymptotic stability. Proc. Amer.
Math. Soc. 22, 513-518 (1969).
8. Linear perturbation of a family of linear systems. Rep. 950, MRC, Univ. of
vVisconsin, Madison, Wisc., October 1968.
SuTTI, C. (see SzEGO, G. P.)
SZARSKI, j.
1. Remarque sur un critere d'unicite des integrales d'une equation differentielle
ordinaire. Ann. Polon. Math. 12, 203-205 (1962).
SzEGO, G. P. (see also BHATIA, N. P. and KUHN, H. W.)
1. Contributions to Liapunov's second method: nonlinear autonomous systems.
In: Nonlinear Differential Equations and Nonlinear Mechanics, J.P. LaSalle and S. Lefschetz {Editors). New York: Academic Press 1963, pp. 421430.
2. Contributions to Liapunov's second method: nonlinear autonomous systems.
Basic Eng. Trans. ASME(D) 84, 571-578 (1962).
3. On a new partial differential equation for the stability analysis of time-invariant control systems. SIAM J. Control 1, 63-75 (1962).
216
References
4. On global stability properties of nonlinear control systems. Rep. Contract

NONR-1228 (23), Sept. 1964.
5. New theorems on stability and attraction. In: Stability Problems of Solutions
of Differential Equations, A. Ghizzetti (Editor). Gubbio: Oderisi 1966.
6. Liapunov's second method. Appl. Mech. Rev. 19, 833-838 (1966).
7. An application of the extensive theorem to a control problem. Atti Accad. Naz.
Lincei Rend., Cl. Sci. Fis. Mat. Natur. 42, 766-770 (1967).
8. A theorem of Rolle's type in Rn for functions of the class C1 . Pacific J. Math. 27,
193-195 (1968).
9. Topological properties of weak attractors. In: Topological Dynamics, J. Auslander and W. H. Gottschalk (Editors). New York-Amsterdam: Benjamin
1968, pp. 455-469.
SzEGo, G. P.; ARIENTI, G.; SUTTI, C.
1. On the numerical construction of Liapunov functions. Proc. Fourth Int. Congress IFAC, Warszawa, June 16-21, 1969.
SZEGO, G. P.; GEISS, G.
1. A remark on "A new partial differential equation for the stability analysis on
time-invariant control systems". SIAM J. Control 1, 369-376 (1963).
SZEGO, G. P.; OLECH, C.; CELLINA, A.
1. On the stability properties of a third-order system. Ann. Mat. Pura Appl.,
Ser. IV, 78, 91-104 (1968).
SzEGo, G. P.; TRECCANI, G.
1. Semigruppi di trasformazioni multivoche (Lecture Notes in Mathematics, vol.
101). Berlin-Heidelberg-New York: Springer 1969.
2. Flow without uniqueness near a compact strongly invariant set. Boll. Un. Mat.
Ital., Ser. IV, 1, 113-124 (1969).
3. Noncontinuous Liapunov functions. Ann. Mat. Pura Appl., Ser. IV, 82, 1-16
(1969).
4. Liapunov functions on the region of unstable weak attraction. In: Seminar
on Differential Equations and Dynamical Systems II, J. A. Yorke {Editor)
Springer 1970.
SzMYDT, Z.
1. Sur la structure de !'ensemble engendre par les integrales tendant vers le point
singulier du systeme d'equations differentielles. Bull. Acad. Polon. Sci., Cl.
III, 1, 223 (1953).
2. Sur les systemes d'equations differentielles dont toutes les solutions sont bornees. Amer. Pol. Math. 2, 234-236 (1955).
3. Remarque sur la methode topologique de T. Wazewski. Colloq. Math. 18, 23-27
(1967).
SzMYDTOWNA, Z.
1. Sur l 'allure asymptotique des integrales des equations differentielles ordinaires.
Ann. Soc. Polan. Math. 24, 17-34 (1951).
TA, LI
1. Die Stabilitatsfrage bei Differenzengleichungen. Acta Math. 63, 99-141 (1934).
TAAM, c. T.
1. Asymptotic relations between systems of differential equations. Pacific J.
Math. 6, 373-388 (1956).
TONELLI, L.
1. Opera scelte (Selected Works), published by UMI, S. Cinquini (Editor), vol. 3.
Roma: Cremonese 1960/62.
TRECCANI, G. (see SzEGO, G. P.)
References
217
TRJITZINSKY, w. J.
1. Problemes dans la theorie des systemes dynamiques. Acta Math. 95, 191-289
(1956).
2. Aspects topologiques de la theorie des fonctions reelles et quelques consequences
dynamiques. Ann. Mat. Pura Appl. 42, 51-117 (1956).
TURAN, P.
1. On the instability of systems of differential equations. Acta Math. Acad. Sci.
Hungar. 6, 257-270 (1955).
2. On the property of the stable or conditionally stable solutions of systems of
nonlinear differential equations. Ann. Mat. Pura Appl. 48, 333-340 (1959).
TUROWICZ, A.
1. Sur les trajectoires et les quasitrajectoires des systemes de commande nonlineaires. Bull. Acad. Polon. Sci. 10, 529-531 (1962).
2. Sur les zones d'emission des trajectoires et des quasitrajectoires des systeme&
de commande nonlineaires. Bull. Acad. Polon. Sci. 11, 47-50 (1963).
URA, T.
1. Sur les courbes definies a la surface du tore par des equations admettant un
invariant integral. Ann. Sci. Ecole Norm. Sup. 69, 259-275 (1952).
2. Sur les courbes definies par les equations differentielles dans l'espace a m dimensions. Ann. Sci. Ecole Norm. Sup. 70, 287-360 (1953).
3. Sur les periodes fondamentales de solutions periodiques. Comm. Math. Univ.
St. Paul 4, 113-130 (1955).
4. Sur le courant exterieur a une region invariante; prolongements d'une caracteristique et l'ordre de stabilite. Funkcial. Ekvac. 2, 143-200 (1959).
5. On the flow outside a closed invariant set: stability, relative stability and saddle
sets. In: Contributions to Differential Equations, vol. 3. New York: Wiley
1964, pp. 249-294.
6. Local isomorphisms and local parallelizability of dynamical systems. In: Topological Dynamics, J. Auslander and W. H. Gottschalk (Editors). New YorkAmsterdam: Benjamin 1968, pp. 493-506.
URA, T.; HIRASAWA, Y.
1. Sur les points singuliers des equations differentielles admettant un invariant
integral. Proc. Japan Acad. 30, 726-730 (1954).
URA, T.; KIMURA, I.
1. Sur le courant exterieur a une region invariante: Theoreme de Bendixson.
Comm. Math. Univ. St. Paul 8, 23-39 (1960).
2. Stability in topological dynamics. Proc. Japan Acad. 40, 703-706 (1964).
URABE, M.
1. Nonlinear Autonomous Oscillations. New York: Academic Press 1967.
UTZ, W. R. (see FOLAND, N. E.)
VINOGRAD, R. E.
1. On the limiting behavior of unbounded integral curves (Russian). Dokl. Akad.
Nauk SSSR 66, 5-8 (1949).
2. On the limit behavior of an unbounded integral curve (Russian). Uchen. Zap.
Moskov. Gos. Univ. 155, Mat., no. 5, pp. 94-136 (1952).
VINOKUROV, v. R.
1. On the definition of a dynamic limit point in general dynamical systems
(Russian). Izv. Vyssh. Uchebn. Zaved., Mat., 1964, no. 3 (40), pp. 36-38.
VRKoc, I. (see also KuRZWEIL, J.)
1. On the inverse theorem of Chetaev. Czechoslovak Math. J. 5 (80), 451- 461
(1955).
15
Bhatia/Szego, StabiJity Theory
218
References
2. Integral stability (Russian). Czechoslovak Math. J. 9 (84), 71-129 (1959).

VRUBLEVSKAYA,l.N.
1. On trajectories and limiting sets of dynamical systems (Russian). Dokl. Akad.
Nauk SSSR 97, 9-12 (1954).
2. Some criteria of equivalence of trajectories and semitrajectories of dynamical
systems (Russian). Dokl. Akad. Nauk SSSR 97, 197-200 (1954).
3. On geometric equivalence of the trajectories of dynamical systems (Russian).
Mat. Sbornik 42, 361-424 (1957).
WALTER, w.
1. Bemerkung zu verschiedenen Eindeutigkeitskriterien fur gewohnliche Differentialgleichungen. Math. Z. 84, 222-227 (1964).
WAZEWSKI, T.
1. Sur un principe topologique de I' examen de !'allure asymptotique des integrales
des equations differentielles ordinaires. Ann. Soc. Polon. Math. 20, 279-313
(1947).
2. Surles integrales d'un systeme d'equations differentielles ordinaires. Ann. Soc.
Polon. Math. 21, 277-297 (1948).
WAZEWSKI, T.; ZAREMBA, s.
1. Surles ensembles de condensation des caracteristiques d'un systeme d'equations
differentielles ordinaires. Ann. Soc. Polon. Math. 15, 24-33 (1936).
WEXLER, D. (see also SANDOR, ST.)
1. Stability theorems for a system of stationary differential equations. Rev. Roumaine Math. Pures Appl. 3, 131-138 (1956).
WILSON, F. W., Jr.
1. Smoothing derivatives of functions and applications. Div. Appl. Math., Brown
Univ., Tech. Rep. 66-3, June 1966.
2. The structure of the level surfaces of a Lyapunov function. J. Diff. Eqs. 3,
323-329 (1967).
WrnTNER, A. (see also HARTMAN, P.)
1. The nonlocal existence problem of ordinary differential equations. Amer. J.
Math. 67, 277-284 (1945).
2. On the convergence of successive approximations. Amer. J. Math. 68, 13-19
(1946).
3. Asymptotic equilibria. Amer. J. Math. 68, 125-132 (1946).
4. The infinities in the nonlocal existence problem of ordinary differential equations. Amer. J. Math. 68, 173-178 (1946).
5. On the local uniqueness of the initial value problem of the differential equation
dnx/dtn = f(t, x). Boll. Un. Mat. Ital. (3) 11, 496-498 (1956).
WHITNEY, H.
1. Regular families of curves, I, II. Proc. Nat. Acad. Sci. USA 18, 275-278,
340-342 (1932).
2. Regular families of curves. Ann. of Math. 34, 244-270 (1933).
3. On regular families of curves. Bull. Amer. Math. Soc. 47, 145-147 (1941).
WHYBURN, G. T.
1. Analytic Topology (Amer. Math. Soc. Colloquium Publications, vol. 38). New
York 1942.
Wu, TA-SUN
1. Continuous flows with closed orbits. Duke Math. J. 31, 463-469 (1964).
2. Proximal relations in topological dynamics. Proc. Amer. Math. Soc. 16, 513514 (1965).
3. Left almost periodicity does not imply right almost periodicity. Bull. Amer.
Math. Soc. 72, 314-316 (1966).
References
219
YAMABE, H. (see MARKUS, L.)

YoRKE, J. A. (see also BHATIA, N. P.; JoNES, G. S. and STRAUSS, A.)
1. Asymptotic properties of solutions using the second derivative of a Liapunov
function. Ph.D. Thesis, Univ. of Maryland, College Park, Md., June 1966.
2. Invariance for ordinary differential equations. Math. Systems Theory 1, 353372 (1967).
3. Extending Liapunov's second method to non-Lipschitz Liapunov functions. In:
Seminar on Differential Equations and Dynamical Systems, G. S. Jones
(Editor)(LectureNotesinMathematics, vol. 60). Berlin-Heidelberg-New York:
Springer 1968, pp. 33-38.
4. An extension of Chetaev's instability theorem using invariant sets and an
example. In: Seminar on Differential Equations and Dynamical Systems,
G. S. Jones (Editor) (Lecture Notes in Mathematics, vol. 60). Berlin-Heidelberg-New York: Springer 1968, pp. 107-114.
5. Liapunov functions and the existence of solutions tending to zero. In: Seminar
on Differential Equations and Dynamical Systems, G. S. Jones (Editor)
Springer 1968, pp. 51- 58.
6. Space of solutions. In: Mathematical Systems Theory and Economics II,
H. W. Kuhn and G. P. Szego {Editors) (Lecture Notes in Operations Research and Mathematical Economics, vol. 12). Berlin-Heidelberg-New York:
Springer 1969, pp. 383- 603.
7. A theorem on Liapunov functions using V. Tech. Note BN-596, IFDAM, Univ.
of Maryland, College Park, Md., Feb. 1969.
YOSHIZAWA, T.
1. The Stability Theory by Liapunov's Second Method. Math. Soc. Japan, Tokyo,
1966.
2. On the stability of solutions of a system of differential equations. Mem. Coll.
Sci. Univ. Kyoto, Ser. A, 29, 27-33 (1955).
3. Note on the solutions of a system of differential equations. Mem. Coll. Sci.
Univ. Kyoto, Ser. A, 29, 249-273 (1955}.
4. On the equiasymptotic stability in the large. Mem. Coll. Sci. Univ. Kyoto, Ser.
A, 32, 171-180 (1959).
5. Liapunov's functions and boundedness of solutions. Funkcial. Ekvac. 2, 95-142
(1959).
6. Stability and boundedness of systems. Arch. Rat. Mech. Anal. 6, 409-421
(1960).
7. Asymptotic behavior of solutions of nonautonomous systems near sets.
J. Math. Kyoto Univ. 1, 303-323 (1962).
8. Stability of sets and perturbed systems. Funkcial. Ekvac. 5, 31-69 (1963).
9. Asymptotic stability of solutions of an almost periodic system of functionaldifferential equations. Rend. Circ. Mat. Palermo, Ser. II, 13, 1-13 (1964).
10. Eventual properties and quasi-asymptotic stability of a non-compact set.
Funkcial. Ekvac. 8, 79-90 (1966).
YouNG, G. (see HocKING, J .)
ZAREMBA, M. S. C. (see also WA.zEWSKI, T.)
1. Sur les equations au paratingent. Bull. Sci. Math. I, 60, 139-160 (1936).
ZEEMAN, c.
1. The generalized Poincare conjecture. Bull. Amer. Math. Soc. 67, 270 (1961).
15*
220
References
N. P.
1. Certain properties of discrete dynamical systems (Russian). Uchen. Zap. Moskov. Gos. Univ., no. 163, Mat., 6, 31-59 (1952).
ZIPPIN, L. (see also MONTGOMERY, D.)
1. Transformation groups. In: Lectures in Topology. Ann Arbor: Univ. of Michigan Press 1941, pp. 191-221.
ZUBOV, V. I.
1. The Methods of Liapunov and their Applications. Leningrad 1964.
2. Mathematical Methods for the Investigation of Systems of Automatic Control
(Russian). Leningrad: Gos. Sojus. Izd. Sudostroit. Promsyl. 1959, p. 324.
English translation, Oxford: Pergamon Press 1964.
3. Some sufficient criteria for stability of a nonlinear system of differential eq.uations (Russian). Prikl. Mat. Mekh. 17, 506-508 (1953).
4. On the theory of A. M. Liapunov's second method (Russian). Dokl. Akad. Nauk
SSSR 99, 341-344 (1954).
5. Questions of the theory of Liapunov' s second method, construction of a general
solution in the region of asymptotic stability (Russian). Prikl. Mat. Mekh.
19, 179-210 (1955).
6. On the theory of A. M. Liapunov's second method (Russian). Dokl. Akad.
Nauk SSSR 100, 857-859 (1955).
7. An investigation of the stability problem of systems of equations with homogeneous right hand members (Russian). Dokl. Akad. Nauk SSSR 114,
942-944 (1957).
8. Conditions for asymptotic stability in the case of nonstationary motions and
estimate of the rate of decrease of the general solution (Russian). Vestnik
Leningrad. Univ., Ser. Mat. Mekh. Astron. 12, 110-129 (1957).
9. On a method of investigating the stability of a null-solution in doubtful cases
(Russian). Prikl. Mat. Mekh. 22, 46-49 (1958).
10. On stability conditions in a finite time interval and on the computation of the
length of that interval (Russian). Bull. Inst. Politechn. Ia~i (N.S.) 4, 59-74
(1958).
11. Some problems in stability of motion (Russian). Mat. Sbornik 48 (90), 149-190
(1959).
12. On the theory of recurrent functions (Russian). Sibirsk. Mat. Zh. 3, 532-560
(1962).
ZHIDKOV,
Author Index
Aizermann, M. 112
Albrecht, F. 162
Andronov, .A. 117
Anosov, D. V. 7, 8
Antosiewicz, H. A. 55, 112, 113
Arienti, G. 164, 184
Auslander, J. 27, 30, 42, 55, 112, 133
Barbashin, E. A. 7, 112, 164, 165
Barocio, S. 117
Bass, R. \V. 112
Bebutov, M. V. 8, 55
Bendixson, I. 1, 118
Bhatia, N. P. 3, 7, 8, 30, 42, 55, 112, 113,
118, 133, 163, 164, 165, 183
Birkhoff, G. D. 2, 7, 8, 30, 38, 42, 118
Bochner, S. 113
Bohr, H. 113
Bronstein, I. U. 7
Budak, B. M. 7
Bushaw, D. 8, 112
Fuller, F. B. 118
Furstenberg, H. 2
Gantmacher, F. M. 164
Garcia, M. 42
Geiss, G. R. 113
Gottschalk, W. H. 2, 7, 30, 42, 118
Grabar, M. I. 7, 8
Green, L. W. 7
Grohman, D. M. 118
Haas, F. 30, 118
Hadamard, J. 162
Hahn, F. J. 7, 112
Hahn, W. 2, 163
Hajek, 0. 3, 7, 8, 30, 112, 118
Hale, J. K. 118
Hartman, P. 1, 8, 118
Hedlund, G. A. 2, 7, 30, 42
Hilmy, H. 42
Ingwerson, D.R. 164
Cellina, A. 164
Chetaev, N. 164
Chu, Hsin 2, 42
Clay, J.P. 112
Coddington, E. A. 1, 7, 8
Conley, C. C. 8
Conti, R. 1
D'Ambrosio, U. 113
Denjoy, A. 118
Desbrow, D. 112
Deysach, L. 113
Dirichlet, G. L. 111
Dowker, Y. N. 42
Dugundji, J. 55, 113
Ellis, R. 2, 42
El'sgol'c, L. E. 30
England, J. W. 42
Jones, G. S. 30, 113

Kakutani, Sh. 2
Kalman, R. E. 8
Kaplan, W. 118
Kayande, A. A. 113
Kent, J. F. 42
Kimura, I. 112, 118
Krasovskii, N. N. 112, 164, 165
Kudaev, M. B. 118, 160
Kurzweil, J. 112, 165
Lagrange, J. L. 111
Lakshmikantham, V. 113
LaSalle, J. P. 134, 163, 165
Lazer, A. C. 30, 112
Lefschetz, S. 1, 30, 112, 117, 162
Leighton, W. 164
222
Author Index
Levinson, N. 1, 7, 8
Lewis, D. C., Jr. 162
Liapunov, A. M. 1, 111, 112, 162, 163
Malkin, I. G. 112, 162
Markov, A. A. 2, 7, 113
Markus, L. 2, 55, 113, 162
Massera, J. L. 112, 165
Mendelson, P. 112
Miller, R. K. 8
Minkevich, M. I. 7
Minorsky, N. 117
Montgomery, D. 7
Moser, J. 2, 162
Nemytskii, V. V. 1, 2, 78, 30, 42, 55, 112,
113, 118, 160
Okamura, H. 183
Olech, Cz. 164
Onuchic, N. 162
Papush, N. P. 160
Peixoto, M. M. 2, 162
Perov, A. I. 118
Persidskii, K. P. 112, 165
Persidskii, S. K. 112
Plis, A. 162
Poincare, H. 1, 2, 7, 30, 112, 117, 162
Schwartz, A. J. 118
Seibert, P. 30, 112, 133, 162
Sell, G. R. 8, 113
Sibirskii, K. S. 30, 42, 118
Siegel, C. L. 118
Smale, St. 2, 162
Stepanov, V. V. 1, 2, 7, 8, 30, 42, 55, 112
Sternberg, Sh. 118
Stokes, A. P. 118
Strauss, A. 134, 162, 163
Sutti, C. 164, 183
Szego, G. P. 3, 7, 8, 30, 42, 112, 118, 134,
163, 164, 165, 183
Szmydt, Z. 162
Treccani, G. 3, 10, 30, 112, 113, 118, 164,
165, 183
Tumarkin, G. S. 42
Ura, T. 2, 7, 30, 55, 112, 118, 133
Vrkoc, I. 112, 165
Wazewski, T. 162
Whitney, H. 2, 7, 55
Whyburn, G. T. 7
Wilson, F. W., Jr. 112, 165
Roxin, E. 7, 112
Yorke, J. A. 30, 160, 162, 164, 165, 183

Yoshizawa, T. 112, 113, 165, 183
Saito, T. 118
Sansone, G. 1
Zippin, L. 7
Zubov, V. I. 3, 112, 113, 163, 165
Subject Index
Absolutely stable compact set 124-128
Almost periodic motion 106 - 111
Asymptotic stability for compact sets 58,
61-64, 102
for compact minimal sets 65
- and Liapunov functions 66
for closed sets 88, 91
for compact sets relative to a set 100
- and Liapunov functions for differential equations 134, 135, 141, 142,
146-150, 156-160
Attraction for compact sets 58, 61, 62
for closed sets 87, 97
for compact sets relative to a set 99
- and Liapunov functions for
ordinary differential equations 156,
170, 176, 177
Axioms 5
Equi-attraction for closed sets 87

- and Liapunov functions 92, 93
Equilibrium points 15
Equi-stability of closed sets 84, 85, 97,
128
First positive (negative) prolongation:
D+(x) 24-30
of dispersive flows 47
of stable sets 61
of weak attractors 64
First positive (negative) prolongational
limit set: J+(x) 24-30, 60
- of dispersive flows 4 7
- of limit poin ts 60
First positive relative prolongation 29,
101, 102
First positive relative prolongational
limit set 29, 102
Fundamental period 18
Bebutov dynamical system 8

Generalized recurrence 131
Global (uniform) (weak) attractors 58,
82
- , asymptotically stable set 58, 73
- , unstable set 61
Global extension theorem 148, 149
Group axiom 5
c - c map 121, 122

Central motions 36
Classification theorems 114- 118
Cluster map 121
Compactly based tube 52
Completely unstable dynamical system
43
Component-wise stable compact set 104
Continuity axiom 5
Critical points 15-17, 82, 83
Higher prolongations: D't (x) 128-130

Higher prolongational limit sets: J! (x)
125, 130
Definite real-value function 138

Dispersive dynamical system 44-48,
53, 54
Dynamical system 5
defined by ordinary differential equations 6, 7, 8
on a torus 9, 33
Identity axiom 5
Improper saddle points 55
Instability for compact sets 58
- - and Liapunov functions for ordinary differential equations 144
Invariant sets 12-14
- , compact 41
224
Subject Index
Lagrange stability 41, 42, 109

Lagrange unstable motion (dynamical
system) 43
Liapunov functions for dynamical
systems 66-79
- for ordinary differential equations
134-184
Limit sets: A+(x), A-(x) 19-24, 27, 32,
60, 110
- VJ and f3 118
Local extension theorem 146-148
Minimal set 36 - 40
- , compact 38, 39, 41, 108, 109
- , non-compact 40
Motion 6
Negative definite real-valued function
138
Negative Lagrange stability 41
Negative limit set 19
Negative semi-trajectory 14
Negatively invariant sets 12-14
Negatively Lagrange unstable motion 43
Negatively Poisson unstable point 43
Non-wandering point 34-36, 131
Orbit 30
Ordinary autonomous differential systems 6, 7, 8
Ordinary non-autonomous differential
systems 8
Parallelizable dynamical system 48- 54,
46, 83, 91
Period 17
Periodic point 17-19, 32, 34, 51, 81
-, motion 17
-, trajectory 17
Phase map 5
Phase space 5
Poisson stability 31- 36
Poisson unstable point (dynamical
system) 43
Positive definite real-valued functions
138
Positive Lagrange stability point (motion) (trajectory) 41, 109, 110
Positive limit set 19
Positive semi-trajectory 14
which uniformly approximates its
limit set 109
Positively compact set 60, 103

- homeomorphic to the closed unit fall
in Rn 82
Positively invariant sets 12-14
Positively Lagrange unstable motion
(dynamical system) 43, 44
Positively (negatively) Poisson stable
point (motion) (trajectory) 31, 32
Positively Poisson unstable point (dynamical system) 43, 44
Positively recursive set 31
Primitive period 18
Prolongation (first), see First (positive)
(negative) prolongation
- (higher), see Higher prolongations
Prolongational limit set (first), see First
(positive) (negative) limit set
Recurrent point (motion) (trajectory)
38-40, 108
- generalized 131
Region of attraction of the compact
set M:.A (M) 56, 57, 70
- and Liapunov functions 74, 75
of the closed set M 88
Region of instability 61, 180-183
Region of uniform attraction of the set
M: Au(M) 56, 57
Region of weak attraction of the compact set M: A 00 (M) 56, 57, 82
- of the closed set M 88
Relative asymptotic stability 100
attraction 99
stability 99, 100
uniform attraction 100
weak attraction 99, 103
Relative prolongations 29, 101, 102
Relatively dense set of real numbers 39
Rest points 15
Section of dynamical system 48
Self posi ti vel y recursive set 31
Semi-asymptotically stable closed set 88
Semi-attraction for closed sets 8 7, 97, 98
Semi-weak attraction for closed sets 87,
97
Stability for compact sets 58, 60, 62, 83,
96, 104
for closed sets 84, 85, 97, 98
for compact sets relative to a set 99,
100
Subject Index
Stability, component-wise 104
of a motion 106
- - and Liapunov functions for ordinary differential equations 139, 140
Stability-additive pair 105
Trajectory 6, 14
Transition 6
Transitive map 121
Tube 49
Ultimate boundedness 73, 82
Uniform approximation of a limit set
109
Uniform asymptotic stability for closed
sets 88
- - and Liapunov functions 92, 93
Uniform attraction for compact sets 58,
62
225
Uniform attraction for closed sets 8 7

- for compact sets relative to a set 100
Uniform (first) (positive) prolongation of
a compact set M: Dt (M) 128
Uniform stability for closed sets 84, 86,
98
- for motions 106
Uniformly unbounded function 70
Wandering point 43, 52, 53
Weak attraction for compact sets 58, 61,
62, 64
for closed sets 8 7, 9 7
for compact sets relative to a set 99,
103
- and Liapunov functions for ordinary differential equations 165, 170,
179-183
Weak attractor theorem 64

Stability

Uploaded by

Copyright:

Available Formats

Stability

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stability

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What topics are covered in the book?

What topics are covered in the book?

CLASSICS IN MATHEMATICS

Nam Parshad Bhatia Giorgio P. Szego

Springer-Verlag Berlin Heidelberg New York 1970

Professor Nam Parshad Bhatia

Professor George Philip Szego

Professor Dr. B. Eckmann

Professor Dr. B. L. van der Waerden

To Sushiela and Emilia

1. Definition and Related Notation . .

1. Definition of Higher Prolongations .

IX. Non-continuous Liapunov Functions

Other standardly used set theoretic notations are:

the set {y: (} (y, M) ~ X}.

Notation Pertaining to Dynamical Systems

dynamical system on a space X (I, 1.1, p. 5).

positive (negative) limit set of x (II, 3.1, p. 19).

Bhatia/Szego, Stability Theory

of the class of all second order differential equations. By introducing

bringing in the direct method of Liapunov was made by V.I.ZuBov [l].

1. Definition and Related Notation

n(n(x, t1), t2)

for every x EX (identity axiom),

for every x EX and t1 , t2 in R

n is continuous (continuity axiom).

Given a dynamical system on X, the space X and the map n are

for every x EX,

xt1 (t2) = x(t1

for every x EX and t1 , t2 in R.

In line with this notation, if MC X and A C R, then MA is the set

Theorem. For each tE R, nt is a homeomorphism of X onto

Proof. For any t ER the transition 11/ is continuous as n is such.

n'(n8 (x)) = nt(xs) = xs(t) = x(t

2. Examples of Dynamical Systems

2. Examples of Dynamical Systems

<p (t, x) exists which is defined on R and satisfies <p (0, x)

for all x, v in Rn.

where/: D ~Rn is a continuous function on some open set D (Rn, and

where !! II is the euclidean-distance norm. If D =f= Rn, then oD =f= 0 and

where is a parameter in a metric space M, and f: Rn X M -+ Rn is

where f: RX Rn -+ Rn is continuous. Assume that for each (t0 , x0) E RX Rn,

where y = (t, x), and g (y)

g) = max {II f (t) - g (t)

2m l?m (/, g).

One may now verify that e defines a metric on X (i.e., X is a metric

2. Examples of Dynamical Systems

On the metric space X defined above we define a dynamical system

lfgn(T) - g(-r) If< !lln(tn + T) - ln(t + T) I!+ llln(t + T) - l(t + T) If

where I: R 2 ~ R2 is continuous and the conditions of example 2.1 are

It is customary to think of the equation 2.6.1 satisfying condition 2.6.2

Notes and References

Notes and References

differential equations without uniqueness may define a semigroup of multivalued

1. Invariant Sets and Trajectories

Definition. A set M C X is called invariant whenever

for all xE Mand tE R.