Elements of The Mathem

Elements of the Mathematical Theory of Multi-Frequency Oscillations
Mathematics and Its Applications (Soviet Series)
Managing Editor:
M. HAZEWINKEL
Centre for Mathematics and Computer Science, Amsterdam. The Netherlands
Editorial Board:
A. A. KIRILLOV. MGU. Moscow. U.S.S.R.
Yu. I. MANIN. Steklov Institute ofMathematics, Moscow. U.S.S.R.
N. N. MOISEEV. Computing Centre. Academy ofSciences, Moscow, U.S.S.R.
S. P. NOVIKOV. Landau Institute ofTheoretical Physics. Moscow. U.S.S.R.
M. C. POLYVANOV. Steklov Institute ofMathematics. Moscow. U.S.S.R.
Yu. A. ROZANOV. Steklov Institute ofMathematics. Moscow, U.S.S.R.
Volume 71
Elements of
the Mathematical Theory
of Multi-Frequency
Oscillations
by
A. M. Samoilenko
SPRINGER SCIENCE+BUSINESS MEDIA, B.V.
Library ofCongress Cataloging-in-Publication Data

Samollenko. A. M. (Anatolil Mlkhallovichl
[~lementy matematlcheskol teorii mnogochastotnykh kolebanil.
Englishl
Elements of the mathematlcal theory of mUltl-frequency
osclllatians I by A.M. Samoilenko.
p.
c -- (Mathematics and its applications (Soviet series)
v. 71>
Translation of, ~lB.enty matematicheskol teorii mnogochastotnykh

ko leban i 1.
Includes index.
ISBN 97894-010-5557-4
ISBN 978-94-011-3520-7 (eBook)
DOI 10.1007/978-94-011-3520-7
1. Nanlinear oscillatians.
r. Title. II. Ser ies, Mathematics
and jts applications (Kluwer Academic Publishers). Saviet serles
71.
OA8S7.5.S2S13 1991
515' . 352--dc20
91-29110
ISBN 978-94-010-5557-4
Printed an acid-free paper

Translatedfrom the Russian by Yuri Chapavsky
This is the translation of the original work
Mathematical Theory Elements of Multi-Frequency Oscillations
Published by Nauka Publishers, Moscow, 1987
An Rights Reserved
1991 Springer Science+Business Media Dordrecht
OriginalLy published by Kluwer Academic Publishers in 1991
Softcover reprint of the hardcover 1st edition 1991
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner,
SERIES EDITOR'S PREFACE
"EI moi. '.'f si favait su comment en revcnir.

je n'y serais point alli.'
Jules Verne
One service mathematics has rendered the

human race. II has put cammon sense back
where it belongs, on the topmost shelf next
to the dUlly canister labelled 'discarded non
The series is divergent; therefore we mlY be

able to do something with it.
O. Heaviside
sense'.
Eric T. Bell
Mathematics is a tool for thought. A highly necessary tool in a world where both feedback and nonlinearities abound. Similarly, all kinds of parts of mathematics serve as tools for other parts and for
other sciences.
Applying a simple rewriting rule to the quote on the right above one finds such statements as:
'One service topology has rendered mathematical physics ...'; 'One service logic has rendered computer science .. .'; 'One service category theory has rendered mathematics ...'. AU arguably true. And
all statements obtainable this way form part of the raison d'dre of this series.
This series, Malhemalics and liS Applicalions, started in 1977. Now that over one hundred
volumes have appeared it seems opportune to reexamine its scope. At the time I wrote
"Growing specialization and diversification have brought a host of monographs and
textbooks on increasingly specialized topics. However, the 'tree' of knowledge of
mathematics and related fields does not grow only by pUlling forth new branches. It
also happens, quite often in fact, that branches which were thought to be completely
disparate are suddenly seen to be related. Further, the kind and level of sophistication
of mathematics applied in various sciences has changed drastically in recent ycars:
measure theory is used (nontrivially) in regional and theoretical economics; algebraic
geometry interacts with physics; the Minkowsky lemma, coding theory and the structure
of waler meet one another in packing and covering theory; quantum fields, crystal
defects and mathematical programming profit from homotopy theory; Lie algebras are
relevant to filtering; and prediction and electrical engineering can use Stein spaces. And
in addition to this there are such new emerging subdisciplines as 'experimental
mathematics', 'CFD', 'completely integrable systems', 'chaos, synergetics and largescale
order', which are almost impossible to fit into the existing classification schemes. They
draw upon widely different sections of mathematics."
By and large, all this still applies today. It is still true that at first sight mathematics seems rather
fragmented and that 10 find, see, and exploit the deeper underlying intcrrelations more effort is
needed and so are books that can help mathematicians and scientists do so. Accordingly MIA will
continue to try to make such books available.
If anything, Ihe description I gave in 1977 is now an understatement. To the examples of
interaction areas one should add Siring theory where Riemann surfaces, algebraic geometry, modular functions, knots, quantum field theory, Kac-Moody algcbras, monstrous moonshine (and more)
all come together. And 10 the examples of things which can be usefully applied let me add the topic
'finile geometry'; a combination of words which sounds like it might not even exisl, let alone be
applicable. And yci it is being applied: to statistics via designs, 10 radar/ sonar detection arrays (via
finite projective planes), and to bus connections of VLSI chips (via difference sets). There seems to
be no part of (so-called pure) mathematics Ihal is not in immediale danger of being applied. And,
accordingly, the applied mathematician needs to be aware of much more. Besides analysis and
numerics, Ihe traditional workhorses, he may need all kinds of combinatorics, algebra, prObability,
and so on.
In addition, Ihe applied scientist needs to cope increasingly with the nonlinear world and the
vi
SERIES EDITOR'S PREFACE
extra mathematical sophistication that this requires. For that is where the rewards are. Linear
models are honest and a bit sad and depressing: proportional efforts and results. It is in the nonlinear world that infinitesimal inputs may result in macroscopic outputs (or vice versa). To appreciate what I am hinting at: if electronics were linear we would have no fun with transistors and computers; we would have no TV; in fact you would not be reading these lines.
There is also no safety in ignoring such outlandish things as nonstandard analysis, superspace
and anticommuting integration, p-adic and ultrametric space. All three have applications in both
electrical engineering and physics. Once, complex numbers were equally outlandish, but they frequently proved the shortest path between 'real' results. Similarly, the first two topics named have
already provided a number of 'wormhole' paths. There is no telling where all this is leading fortunately.
Thus the original scope of the series, which for various (sound) reasons now comprises five subseries: white (Japan), yellow (China), red (USSR), blue (Eastern Europe), and green (everything
else), still applies. It has been enlarged a bit to include books treating of the tools from one subdiscipline which are used in others. Thus the series still aims at books dealing with:
- a central conc~pt which plays an important role in several different mathematical and/ or
scientific specialization areas;
new applications of the results and ideas from one area of scientific endeavour into another;
- influences which the results, problems and concepts of one field of enquiry have, and have had,
on the development of another.
The world is full of periodic and quasi-periodic motions and oscillations. That is multifrequency
oscillations. Mathematically, to a large extent, that means the study of invariant tori of dynamical
systems. These constitute the more systematic part of the phase space of a dynamical system as
opposed to the chaotic parts and their study is just as (perhaps more) important as the study of
chaos.
This monograph constitutes a thorough and up-to-date treatment of invariant tori, including
existence, perturbation and stability results. Many of these results involve original work by the
author himself. These chapters are preceded by ones on periodic and quasi-periodic functions and
on stability theory, thus making the volume selfcontained.
The shortest path between two truths in Ihe
real domain passes through Ihe complex
N ever lend books, for no one ever returns

them; the only books I have in my library
domain.
are books that other folk have lent me.
J. Hadamard
Anatole Frana:
La physique ne nous donne pas seulemenl

I'occasion de re'soudre des problemes ... ene
The funaion of an expert is nol to be more

righl than other people, bUI 10 be wrong for
more sophisticated reasons.
David BUller
DOUS fait presscnlir 1a solution.
H. Poincare
Amsterdam, August 1991
Michiel Hazewinkel
Contents
xi
PREFACE
.xiii
INTRODUCTION .
CHAPTER 1. Periodic and quasi-periodic functions
1.1. The function spaces Cr(Tm ) and Hr(Tm ) . . . . .
1.2. Structure of the spaces Hr(Tm ). Sobolev theorems.
3
7
.
1.3. Main inequalities in Hr(Tm )
1.4. Quasi-periodic functions. The spaces Cr(w)
. .
9
13
1.6. First integral of a quasi-periodic function .
18
1.5. The spaces
H r (w)
and their structure
1.7. Spherical coordinates of a quasi-periodic vector function
25
1.8. The problem on a periodic basis in En . . . . . .
28
1.9. Logarithm of a matrix in C/(Tm ). Sibuja's theorem

1.10. Garding's inequality . . . . . . . . . . .
38
41
CHAPTER 2. Invariant sets and their stability
46
2.1. Preliminary notions and results . . . . . .
46
2.2. One-sided invariant sets and their properties
49
2.3. Locally invariant sets. Reduction principle
56
2.4. Behaviour of an invariant set under small perturbations of the system 65

2.5. Quasi-periodic motions and their closure . . . . . . . . . . .
70
2.6. Invariance equations of a smooth manifold and the trajectory flow
on it. . . . . . . . . . . . . . . . . . . . . . . . . . .
83
2.7. Local coordinates in a neighbourhood of a toroidal manifold. Stability

of an invariant torus . . . . . . . . . . . . . .
88
2.8. Recurrent motions and multi-frequency oscillations
96
CHAPTER 3. Some problems of the linear theory
99
3.1. Introductory remarks and definitions
99
.....
viii
CONTENTS
3.2. Adjoint system of equations. Necessary conditions for the existence

of an invariant torus . . . . . . . . . . . . . . . . . . .
101
3.3. Necessary conditions for the existence of an invariant torus of a

linear system with arbitrary non-homogeneity in C(Tm )
106
3.4. The Green's function. Sufficient conditions for the existence of an

invariant torus
113
3.5. Conditions for the existence of an exponentially stable invariant

torus
118
3.6. Uniqueness conditions for the Green's function and the properties
of this function . . . . . . . . . . . . . . . . . .
3.7. Separatrix manifolds. Decomposition of a linear system
124
134
3.8. Sufficient conditions for exponential dichotomy of an invariant torus 142

3.9. Necessary conditions for an invariant torus to be exponentially
dichotomous
. . . . . . . . . . . . . . . . . . . . . .
151
3.10. Conditions for the C'(Tm)-block decomposability of an exponentially

dichotomous system
. . . . . . . . . . . . . . . . . . . .
158
3.11. On triangulation and the relation between the C'(Tm)-block decomposability of a linear system and the problem of the extendability
of an r-frame to a periodic basis in En
. . . . . . . .
3.12. On smoothness of an exponentially stable invariant torus
169
178
3.13. Smoothness properties of Green's functions, the invariant torus

and the decomposing transformation of an exponentially dichotomous system. . . . . . . . . . . . . . . . . . . . . .
188
3.14. Galerkin's method for the construction of an invariant torus . .
194
3.15. Proof of the main inequalities for the substantiation of Galerkin's

method . . . . . . . . . . . . . . . . . . . . . . . . . . 203
CHAPTER
4. Perturbation theory of an invariant torus of a nonlinear system . . . . . . . . . .
211
4.1. Introductory remarks. The linearization process . . . . . . . ,
211
4.2. Main theorem
215
4.3. Exponential stability of an invariant torus and conditions for its

preservation under small perturbations of the system
.....
220
4.4. Theorem on exponential attraction of motions in a neighbourhood

of an invariant torus of a system to its motions on the torus
4.5. Exponential dichotomy of invariant torus and conditions for its pre-
226
CONTENTS
servation under small perturbations of the system . . . . . . .
ix
241
4.6. An estimate of the smallness of a perturbation and the maximal

smoothness of an invariant torus of a non-linear system
. 258
4.7. Galerkin's method for the construction of an invariant torus of a

non-linear system of equations and its linear modification . . .
4.8.
Proof of Moser's lemma
. . . . . . . . . . . . . . . . .,
270
281
4.9. Invariant tori of systems of differential equations with rapidly and

slowly changing variables
285
BmLIOGRAPHY .
297
AUTHOR INDEX
309
INDEX OF NOTATION
310
SUBJECT INDEX
311
Preface
It can be seen from the title that we shall be dealing here with a mathemati-
cal theory based on precise notions and definitions. The central object of this
theory is an invariant toroidal manifold of a dynamical system considered in a
Euclidean space En or in its product with an n-dimensional torus, that is, in
the space En
Tm . Taking into account the fact that a quasi-periodic solution
of a dynamical system in En "sweeps out" an invariant torus of this sytem, the

relation between the theory of invariant tori and the theory of multi-frequency
oscillations becomes clear; the existence of such tori is a necessary condition
for the existence of multi-frequency oscillations of quasi-periodic solutions of
dynamical systems. By defining multi-frequency oscillations as a motion of a
dynamical system, describing a recurrent trajectory on an invariant toroidal
manifold of the system, we make the invariant toroidal manifold into the main
subject of the mathematical theory of multi-frequency oscillations; the existence of such a manifold is sufficient for multi-frequency oscillations of the
system to exist.
The original part of the monograph -
the last two chapters, is devoted
to the following topics: the existence of invariant toroidal manifolds for linear systems in En
Tm , perturbation theory of such manifolds for non-linear

systems, smoothness and stability properties of these manifolds, the behaviour
of trajectories in a small neighbourhood, the study of separatrix manifolds in
the case of exponential dichotomy of toroidal manifolds, linear system block
decomposition under exponential dichotomy, and substantiation of Galerkin's
X
procedure for finding toroidal manifolds.

The first two chapters of the book contain a theory of quasi-periodic functions, expounded on the basis of the theory of periodic functions of several
variables, and some results from the theory of invariant sets and their stability.
The material expounded in these chapters is closely related to the theory of
xi
xii
PREFACE
multi-frequency oscillations and will be useful to specialists who are not well
acquainted with the above-mentioned topics.
Questions concerning asymptotic integration of multi-frequency oscillation
systems are excluded from the book. They are considered in a separate work.
The results given in the last two chapters were obtained mainly by the author. The results obtained in conjunction with V.L. Kulik and 1.0. Parasyuk
form an exception. The content of the book was partially covered in special
courses given to graduate students of the department of Mathematics and Mechanics of Kiev State University. The courses have been given regularly since
1974.
The book is designed for specialists working in the area of multi-frequency

oscillation analysis, and can be useful to many researchers, students and graduate students who in their work meet oscillatory phenomena.
A.M. Samoilenko.
Introduction
Originally, oscillation theory arose within celestial mechanics. A mathematical
idealization of processes and phenomena studied in oscillation theory reduced
the problem of describing a given physical system to the study of a linear
differential equation, and the oscillation itself to a solution of this equation of
"oscillatory" type. The principle of superposition reflected the main regularity
of the linear theory and the amplitude-frequency characteristic was the main
property of oscillations [54], [155].
The works by Poincare and Lyapunov [66], [69-71], [101] devoted to the
theory of periodic solutions of weakly non-linear systems have given researchers
a mathematical tool for studying oscillation processes of "non-linear" nature,
which are difficult to interpret in the linear theory. The needs of electrical and
electronic engineering contributed to a fast development of this ''new direction"
and were instrumental in its separation as an independent theory in non-linear
oscillation theory in mathematical physics. A fundamental contribution to the
solution of problems of the non-linear theory was made by N.M. Krylov and
N.N. Bogolyubov [19], [21], [23], [55-57].

Periodic solutions of non-linear systems of differential equations have
turned out to be a central object of study. For studying these, a great number
of methods have been developed, from numerical to topological; thus periodic
solutions have been studied from all sides [44], [51], [52], [66], [71], [80], [87],
[92], [97], [99], [122], [133], [136], [137], [140], [141], [144]. Many real processes
whose mathematical idealization allow one to regard them as "repeating themselves exactly" in the same periods of time, have been well interpreted within
oscillation theory [2], [4], [23], [31], [37], [38], [55-57], [71], [73], [76], [80], [92],
[97], [99], [136-138].

Starting in the sixties, a sharp change took place in oscillation theory
towards a study of oscillating processes characterized by their "almost exact" repetition in "almost the same" periods of time. Research in this theory has focused on quasi-periodic oscillations, quasi-periodic solutions of the
xiii
xiv
INTRODUCTION
corresponding equations; and the creation of KAM-theory, which is the theory

of quasi-periodic solutions of "almost integrable" Hamiltonian systems [6-8],
[48], [49], [89-91], [156-160] has been the greatest achievement. Quasi-periodic
solutions of "weakly non-linear" systems of general type have been studied
by N.N. Bogolyubov [20], [21], [24]. Many problems concerning the analysis
of systems which describe quasi-periodic oscillations were solved in papers by
Yu.A. Mitropol'skii [24], [77-79], [82],[83]. The papers [103-105] by the author
are closed related to the latter.
Quasi-periodic oscillations represent a rather complicated and "sensitive"
object of study. The practical effectiveness of finding such oscillations in an
individual dynamical system is negligible, for the reason that the frequency
basis is "unstable". This shows up well in examples and can be explained
theoretically using deep results from the qualitative theory of differential equations: a quasi-periodic oscillation can be easily "destroyed" and turned into
a periodic oscillation by a small change in the right-hand side of the system.
This fact forces one to look for a "coarser" object of study in the theory of
multi-frequency oscillation than a quasi-periodic solution.
In 1967 V.V. Nemytskii in his lectures at the Mathematical School in
Ujgorod, analyzed in detail possible answers to the following question: what
motion of a dynamical system in En should be considered as "oscillatory", and
concluded that it is recurrent motion [93]. With such a definition, a minimal
set "swept" by the trajectories of a recurrent motion becomes the main object
of study in the theory of auto-oscillation.
Until now, among the minimal sets, those "swept" by the trajectories of
quasi-periodic motions have been studied best. Following A.A. Andronov and
A.A. Witt [2], [3], we can assert that already Poincare knew that such a set is
a torus. Regardless of the fact that the invariant tori of a dynamical sytem in
En are "carriers" of auto-oscillations of a non-linear system, until recently they

were studied only occasionally. Thus, the first deep statements about invariant
toroidal manifolds of systems of non-linear mechanics, that is, integral manifolds of toroidal type, were obtained by N.N. Bogolyubov and N.M. Krylov [19],
[23], [57], while substantiating the asymptotic methods of non-linear mechanics. Only later were the ideas suggested in these works fully developed by Yu.A.
Mitropol'skii, giving rise to the method of integral manifolds used in non-linear
mechanics [22], [81]. This is a powerful and practically convenient mathemati-
INTRODUCTION
xv
cal device for studying integral manifolds of a wide class of non-linear systems
of differential equations. They influenced the nature of further developments in
the perturbation theory of toroidal manifolds, and led S.P. Diliberto [145-147],
T. Hale [138], [151], Y. Kurtsveil [61], J. Kyner [153], I. Kupka [154] and others
[11-13], [45], [95] to deep results.

The work of K.O. Friedrichs [149], [150] started a new area in this theory.
His theory of periodic solutions of positive symmetric systems of linear partial differential equations, worked out using methods of functional analysis as
applied in mathematical physics, but used for linear extensions of dynamical
systems on a torus, is an original and distinctive theory of existence of invariant
tori of these systems.
In the sixties, J. Moser developed his rapidly convergent iteration method
[89], [156-159], and applied it, in particular, to non-linear generalizations of

positive symmetric systems. This led to new results on the preservation of
an invariant torus under perturbations. These results were essentially different
from those obtained by the method of integral manifolds. The assumption that
the system of equations considered has a high order of smoothness, needed in
order to apply the method of Moser, were significantly weakened by R. Sacker
[163]. Subsequent studies by J. Moser and R. Sacker [90], [91], [160], [164] have
finished the creation of the perturbation theory of invariant tori of dynamical
systems.
The report given at the Fifth International Conference on non-linear oscillations [107] started a new series of papers on problems of the theory of
perturbation and stability of invariant toroidal manifolds of dynamical systems. The introduction of the notion of a Green's function in the problem of
existence of an invariant torus of a linear extension of a dynamical system on
a torus turned out to be very fruitful and encouraged development in different
areas of this theory. It seemed a return to the ideas of the method of integral
manifolds in the theory of perturbation and stability of invariant tori of systems of non-linear mechanics, and led to new results in this theory [58], [59],
[86], [106], [109], [111-117], [119], [123], [124].

Other work containing deep results on invariant tori of dynamical systems
includes the monographs of M. Hirsch, C. Pugh, M. Shub [152], N.1. Neimark
[92], and D.U. Umbetjanov [132], and papers of Yu.N. Bibikov and V.A. Pliss
[15], [16], LV. Bronshtein and V.P. Burdijaev [29-30], N. Fenichel [148] and
xvi
INTRODUCTION
G. Sell [165-168].
Some aspects of the theory given in this book were developed and applied
to different classes of equations [74], [75], [84], [85], [88], [96], [120], [125], [126].
Problems relating to the topic were also considered in [32], [33], [40], [41]' [45],
[53], [121], [130], [131], [133], [134], [135], [170], [171].
Chapter 1. Periodic and quasi-periodic functions
Let f( 1jJ) =
h (1jJ), . .. ,fn(ljJ)) be an Rn-valued function of a variable IjJ =
(ljJl, ...
. . . ,ljJm) that is continuous and periodic with period 271" with respect to each
variable ljJa (0: = 1, ... , m). The set of all such functions forms a linear space
which we denote by C(Tm ). This space can be turned into a complete normed
space by introducing the norm
If 1o
where
= ~~ IIf(IjJ)11,
n
IIfll = L
2
Ifi 1
i=1
is the Euclidean norm in the space Rn of the function f(IjJ).

In C(Tm ) we distinguish a subspace cr(Tm ) offunctions having continuous
partial derivatives up to order r inclusive with respect to the variables ljJa (0:
1, ... ,m). The set cr(Tm ) can be turned into a complete normed space by
introducing the norm
If(IjJ)lr
= 0::Sp::Sr
max IDPf(ljJ) 1o,
where DP is the pth partial derivative with respect to ljJa (0:
= 1, ... , m).
Let P(IjJ) be a trigonometric polynomial in C(Tm), that is, P(IjJ) is a finite

sum of the form
P(IjJ)=
aksin(k,IjJ)+hcos(k,IjJ)=
II k ll::SN
where k
= (k1, ... ,km) is
+ ... + kmljJm,
Pkei(k.~),
II k ll::SN
a vector with integer components, (k,ljJ)
= k1ljJl+
ak and bk are real and Pk complex coefficients, that is,
CHAPTER 1
(Pic is the complex conjugate of Pic), N is an arbitrary non-negative integer.

The space of all such polynomials forms a linear spa.ce which will be denoted
by P(Tm ).
According to Weierstrass's theorem [10], a function f( J) E C(Tm ) can be
uniformly approximated by trigonometric polynomials. This means that for
any function f(J) E C(Tm) there is a sequence of trigonometric polynomials
Pv(J) (v = 1,2, ...) such that

lim If(J) - Pv(J)lo
v-oo
= o.
It follows that the space C(Tm ) is the closure of the space P(Tm ) of trigono-
metric polynomials with respect to the norm

also be made for the spaces cr(Tm
),
I . 1o.
A similar conclusion can
namely, every such space is the closure of
the space of trigonometric polynomials P(Tm ) with respect to this norm.

Thus, using P(Tm ) we can form a chain of Banach spaces embedded in
each other:
C(Tm )
= CO(Tm ) J
where Coo(Tm)
C1(Tm ) J ... J cr(Tm ) J Cr+1(Tm ) J ... J Coo(Tm),

(1.1)
= n cr(Tm ).
00
r=O
Now we can define an inner product (-, ')0 for any two trigonometric polynomials from P(Tm )
P =
Plcei(Ic,r/,
IIlcll~N
Qkei(Ic,r/
(1.2)
IIlcll~N
by setting
(1.3)
Here {Pic, Q -k} =
'L-7=1 pt ~-k
is the usual inner product of the vectors Pic =
(P}, pl, .. , pr:) and Q-Ic (Q:Ic' Q:Ic"'" Q~Ic)'

The inner product (1.3) defines a norm II .11o in the space P(Tm
1
II P II5 = (P,P)o = (21l')m
27f
27f
lo . "lo
IIPWdJl ... dJm =
IIlcll~N
):
IIPIcW
Taking the closure of P(Tm ) with respect to this norm we obtain a Hilbert
space denoted by HO(Tm ). The elements of the space HO(Tm ) are given by
PERIODIC AND QUASI-PERIODIC FUNCTIONS
series LA: !A:ei(A:,,p) for which the sum LA: lilA: 11 2 is finite. Here LA: denotes
LA:EZ m , where zm is the set of all integer-valued vectors k = (k 1 , ... I km ).
For the polynomials (1.2) and any non-negative integer r, one can define
an inner product (-, ')r by setting, for example, as in [14], [89]
(P,Q)=I-Llt P,Q)o=
where Ll
= L;;'=1 k
(1+lI kWY(PA:,Q-A:),
(1.4)
IIA:II~N
is the Laplace operator.
1IlIr:
The inner product (1.4) induces in P(Tm ) a norm
II P II; =
(P, P)r =
(1
+ II k WrlJP.l:11 2
IIA:II~N
Taking the closure of the space P(Tm ) with respect to this norm we obtain a
Hilbert space which we shall denote by H r (Tm ). The elements of the space
Hr(Tm ) are series LA: !A:ei(k,q for which the sum LA:(1 + IIkll 2 Yllfkl1 2 is finite.
Thus, starting with P(Tm ) we can form together with the chain (1.1) the
following chain of Hilbert spaces:
where HOO(Tm ) =
n Hr(Tm ).
r=O
00
We now find out what the structure of the spaces H r (Tm ) is.
1.2. Structure of the spaces Hr(Tm ). Sobolev theorems
According to the Riesz-Fisher theorem [50], each series
Lk fJ.ei(A:,,p) for which
the sum L lilA: 11 is finite is a Fourier series of a function I(<p) which is square
summable on the cube Tm : 0 < <Pet :5 211" (0' = 1, ... ,m). From this it follows
2
that the space H(Tm ) can be identified with the space of functions periodic in
each variable <Pet (0'
= 1, ... , m) with period 211", and which are square summable
on the cube Tm of the periods. In other words, H(Tm ) can be identified with
the space .L: 2 (Tm ).
Now let the series LA: hei(A:,,p) define an element of Hr(Tm ). For any
integer-valued vector P
PI
+ P2 + ... + Pm :5 T,
(Pl,,Pm), Pet
L(ikYfkei(k,,p)
1:
~ 0 (0'
1, ... ,m) and
the series
= Lilplkflk~> ... kfnm/A:ei(k,q

k
Ipi =
CHAPTER 1
represents an element I(p) of Hr-lpl(Tm
).
This can be seen from the following
estimate:
II/(p)II;-lpl
= L(1 + IlkIl2r-lpllI(ikYlkW =
k
= L(1 + IlkIl2r-lpl(k)2PllikW ::; L(1 + IIkWnl/kll 2 = 11/11;

k
We consider the inner product (f(p) , P)o for an arbitrary polynomial P E
P(Tm ). We have
(f(p) , P)o == L
(ikY(lk, P-k)
IIkll~N
= (-lY
(fk, (-ikY P-k)
= (-l)lpl(f, DP P)o,
(2.1)
IIkll~N
where DP = a ipi / (a/{l ... a1>fnm
is the Iplth derivative of the function P with
respect to 1>. We say that a function
DP if there is a function I(p) in .c 2 (Tm )
.c2 (Tm ) has the generalized derivative

satisfying the equality
(2.2)
for any polynomial P E P(Tm ).

Equality (2.1) shows that an element I of Hr(Tm ) can be identified with a
function that belongs to .c(Tm ) and has all generalized derivatives with respect
to 1> up to the order r inclusive. The generalized derivative DP I = f(p) has the
following Fourier series
L(ikY h:ei(k,),
k
where
f2T
f2T
Ik = (21l')m Jo ... Jo
1(1))e- (k')d1>l.'' d1>m.
Indeed, if L:k IkP)ei(k,) is the Fourier series of the function f(p)
= D(p) f,
then according to equality (2.2) we have
(lip) - (ikY fk, P-k)
=0
IIkll~N
for any trigonometric polynomial P
= L:llkll~N[fiP) -
(ik)P fk]e-i(k,)
= L:llkll~N Pkei(k,) E P(Tm). By taking

we see that lip) = (ik)P Ik for any integer
k. Since the series L,k(ik)p fkei(k,</ is the Fourier series of the function f(p)
belonging to 2(Tm ), it satisfies Parseval's equality [67]:
E lI(iky fkll 2 = E(k) 2P lllkIl 2 = IIf(p)lI~ = E IIfk P)11 2.

k
This leads to the following estimate:
(k)2 PllhIl 2 = c E
E(1 + II k ll 2r1lfkW ~ c E
k
Ipl~r,k
which shows that the function
IIf~p)W
=c E
Ipl~r,k
IIDPfll~,
Ipl~r
f generates an element of the space Hr(Tm ).
It shows that the space Hr(Tm ) can be identified with the space of functions
periodic with respect to the variables Ja (a = 1, ... , m) with period 21T that are
square summable on the cube of periods Tm and have generalized derivatives
with respect to 4> up to order r inclusive.
The properties given by the Sobolev theorems [127] form an important
characterization of the spaces Hr(Tm ).
Theorem 1.
> m/2
For r
+p
the inclusion Hr(Tm ) C CP(Tm ) and the
estimate
IDP flo ~ cllfllr

hold, where
Ipl
p, c is a positive constant that does not depend on f and f
is an arbitrary function belonging to Hr(Tm ).

A proof of Theorem 1 follows from the following sequence of inequalities
II ~ cE II kll P IIfk II
E lI(iky hei(k,</
k
cE(1 + IlkWt- p)/2+ P /21Ifkll(1 + II kIl 2)(p-r)/2 ~

k
~ c(E(l
+ IlkWrllhIl 2)1/2(E(1 + II kIl 2)p-r)I/2 = cillfllr,
(2.3)
where, using the condition r > m/2 + p, we obtain the following estimate for
the constant:
ci = c2E(1 + IIkIl 2)p-r ~

k
E(1 + 1I 2 )p-r E 1 ~
v=1
E 1I2p-2r+m-1 = E 1I00
~ C2
00
C'
00
C2
v=1
v=1
I-
Ikl=v
2(r-p-m/2)
= c~ < 00.
CHAPTER 1
According to (2.3), the series obtained by differentiating the function f =

term by term not more than p times is uniformly convergent, so
L:k hei(k,)
that
E CP(Tm ). Moreover,
IDP flo
= 12:(ik Y hei(k,) I
where
Cl
cdlfllr,
is a constant independent of f.
Theorem 2. For s < r, every set of functions that is bounded in Hr(Tm ) is

compact in H'(Tm ).
To prove Theorem 2 we take a bounded sequence of functions f(j) (j

1,2, ...) in Hr(Tm
).
This means that

Ilf(j)IIr
< M,
= 1,2, ....
(2.4)
Let L:k f~j) ei(k,) be the Fourier series for the function f(j). Then taking (2.4)
into account we have
and for any k there is a convergent subsequence of the sequence of the coeffi-
cients f~j) (j = 1,2, ...). Without loss of generality we can assume that the
sequence f~j) itself converges as j ~ 00. We set
liP = 2:
IIkll~N
and estimate
f~j) ei(k,>,
RN f(j)
= 2:
he i (k,4
II kll>N
IIf(j) - f(v)II. for s < r and arbitrary
j,
l/
2: 1.
For each fixed N the convergence of the coefficients I~j),
IIkll ~ N,
--+ 00
implies that
as j,
l/ ~ 00.
On the other hand, for fixed j, l/ and s < r we have

IIRN f(j) - R~)II; ~ c( 1 + N 2 ).-r M 2
as N
--+
~ 00.
But then
as j,
l/ ~ 00.
This proves that the set f(j) (j = 1,2, ...) is compact in the space H'(Tm )
for s < r.
1.3. Main inequalities in Hr(Tm )
Following [62] it is easy to derive a number of inequalities satisfied by functions

belonging to the space Hr(Tm ).
Indeed, the equality
(f,9)r ==
2)1 + II k Wnh,9-k),
(3.1)
which defines an inner product of the elements f == Lk fkei(k,) and 9 ==

Lk 9J;e i(k,) from Hr(Tm ) makes sense even if f E Hr-6(Tm ), 9 E Hr+6(Tm ).
It is therefore natural to apply formulae (3.1) for elements, one of which is
taken from the space Hr-6(Tm ) and the other from Hr+6(Tm ). It follows from
the generalized Schwartz inequality for the inner product (" ')r that
IU,g)r! S IIfllr-6I1gllr+6'
for elements f E H r- 6(Tm ) and 9 E H r+6(Tm ). Thus we obtain the inequality
which gives an estimate of the r-norm of an element

of its (r - s)-norm and (r + s)-norm.
f of H r + (Tm ) in terms
6
The inequality
is obvious for any a
> 0 and r > s
~ roo For any
> 0, setting a == i 1/(r-6)(1 +
IIk1l 2 ) we obtain the estimate
This proves that when f E Hr(Tm ), r > s
ro and
> 0,
(3.2)
Setting in (3.2) r == 0,
== IIfll:/r-1Ilfll~-6/r, we obtain the inequality
which gives an estimate of the s-norm of f E Hr(Tm ) in terms of its r-norm

and O-norm with r > s > O.
CHAPTER 1
The next two inequalities which relate the norm of a function
1 =L
fJ.ei(k,) E
H r (Tm )
to the norm of its generalized derivatives DP 1 = L,k(ik)P Ikei(k,) for
Ipl
~ r
follow from considerations of the previous section:

IIDP 1lip ~ cll/ll,pl+p,
II/l1r ~ c
IIDP 1110
Ipl~r
for any 0 ~ p ~ r - Ipl and some constant c which does not depend on f.
The following more complicated inequality which gives an estimate of generalized derivatives of functions is due to Nirenberg [162]:
(3.3)
for any
Ipl
r and some c which is independent of f.
Inequality (3.3)
leads to Moser's estimate [89] for the composite of functions (f . g)(
I(>,g(>)), where I(>,y) E Cr(Tm x Tn), Tn is the unit ball in En, g( =

(g1(, ... ,gn() is a function from Hr(Tm ) n C(Tm) taking values in Tn,
IIgll ~
1. The estimate is as follows:
1I/(>,g(>))lIr ~
cl/i r (1 + IIgllr),
(3.4)
where c is a positive constant independent of 1 and g,
Iflr
= max
O~q~r
max
(,y)eTmxTn
liD" f(!/l,y)'!,
D" is any partial derivative with respect to >, y of order
(J'.
To prove (3.4) we write down the differentiation formula for the composite
of functions:
D';pl(>,g()
=L
p+,,<r
(D;D;/(>,y
= g(!/l))) x
(X
where
frl' ... ,frr ,
(J'
and p are non-negative positive numbers that satisfy:
D;
and
of order
D:
<T
are any partial derivatives with respect to the variables
> and
and p respectively, and the cqpa are non-negative constants. By
making an estimate of the terms in these formulae using Holder's inequality,

we find that
II D ;f(>,g(>))llo S;
II D D:f(>,y
= g(>)) II(Dg(a" 110 S;
1'=1
p+cr<r
a-
S;
"IID;D;f(>,y
= g()lIr/ ll"/2 II II II Dg(lI

q
1'=1
/"
110'
Inequality (3.3) now leads to Moser's estimate (3.4):
II D;f(>,g()lIo S;
S; c
1IIIDD;f(>,Y)lIr/qll~/r
p+cr~r
S;
clflr
II Igl~l-v/r)a"lIgll~val/)/rS;
r
,,=1
IlglI~-q/r S; cdflr (1
+ IIgllr).
p+q~r
Note that for a function
f that is linear with respect
y estimate (3.4) holds
when 9 E Hr(Tm ).
The next two equalities, which relate the inner product (" -)r and the
differential operator 1 - Ll = 1 -
(I,g)r
L:::'=1 -k =
K:
= (J(' f,g)r-. = (I, J('g)r-.,
11K' fllr
= IIfllr+2"
easily follow from equality (1.4) which defines the inner product in Hr(Tm
These equalities hold for elements
f and
).
9 of HP(Tm ) with the indices being
such that the corresponding inner products and norms are defined.
1.4. Quasi-periodic functions. The spaces Gr(w)
Let WI, W2, ... ,W m be positive numbers that are incommensurable in the sense
that
(k,w) = kIWI
for any integer vector k
+ k2W2 + ... + kmw m f:. 0
= (k 1, ... , k m ), IkI =f. o.
10
CHAPTER 1
A function of a real variable t:
F(t)
= j(wt) = j(WIt, ... ,wmt),
where j E C(Tm ), is called quasi-periodic, the set W
(4.1)
= (WI,.
,w m ) its frequency basis, and the numbers w" (v = 1, ... , m) the jrequencies of the function
00
F.
We give a number of properties of quasi-periodic functions. To derive them

from the similar properties of periodic functions we use Kronecker's statement
[67] on solutions of a system of inequalities. We give it as a separate lemma.
Lemma 1. Let WI,W2, ... ,W m be incommensurable real numbers. Then jar
any set oj real numbers t/JI, t/J2, ... ,t/Jm the system oj inequalities
Iw"t -t/Jvl < 6(mod211'), v
= 1, ... , m,
has a relatively dense set of solutions jar any arbitrary small positive 6.
From Lemma 1 and the definition of a quasi-periodic function one can
easily deduce Weil's theorem [172] on the density of the values of a quasiperiodic function F in the set of the values of the function f.
Theorem 1. The set of values oj a quasi-periodic function F(t)
= f(wt)
IS
dense in the set oj values oj the junction f( t/J).

Indeed, let t/J
(t/JI, ... t/Jm) be an arbitrary point of the cube Tm : 0::;
t/Ja::; 211' (0' = l,oo.,m), (an arbitrary small number. Since f(t/J) is periodic
and continuous, it is uniformly continuous. Therefore we can find a 6 = c(()
such that
IIj(t/J) - j(t/J')II < (
(4.2)
for any point t/J' = (t/J~, t/J2"'" t/J:") with It/J" -t/J~I < 6, v = 1, ... , m. It follows
from Lemma 1 that for t/J~, ... , t/J:" and such a 6 there exists r E R such that
Iwvr-t/J~I < 6(mod211'),
v= 1, ... ,m.
Then from (4.2) we have the estimate
IIf(t/J') - F(r)1I
which proves Theorem 1.
= IIf(t/J') -
j(wr)1I < (,
II
As a corollary of Theorem 1 we obtain the following equality

sup II F (t)1I = max
tER
.pETm
\I/(4J)11 = 1/10,
(4.3)
which relates the maximum of a quasi-periodic function
I . 10
with the norm
of a function
IIF(t)11 = 11/(wt)1I
E C(Tm ). Equality (4.3) allows one to
deduce a number of properties of quasi-periodic functions. They are known as

the approximation theorem, the mean value theorem, and the limit function
theorem [64].
Indeed, since the function
E C(Tm
can be uniformly approximated by
trigonometric polynomials PN(4J) = Lll kll:5N Pkei(k,.p), it follows from (4.3)

that the quasi-periodic function F(t) = I(wt) can be uniformly approximated
by quasi-trigonometric polynomials PN(wt) = Lllkll:5N Pkei(k,w)t such that
supIIF(t) - PN(wt)11 = max

tER
.pETm
for a given (.
1I/(4J) - PN(4J)11 = II - PNlo < (.
(4.4)
> 0 and some PN (4J) E P(Tm ). Inequality (4.4) is equivalent to
the following approximation theorem.
Theorem 2. Let F(t) be a quasi-periodic function with the frequency basis
w = (WI, ... , wm ). For any (. > 0 there exists a quasi-trigonometric polynomial

Piv(t) = Lll kll:5 N Pkei(k,w)t such that
sup IIF(t) - PN(t)1I
tER
< f-
Furthermore, for any quasi-trigonometric polynomial
P(wt)
Pkei(k,w)t
II kll:5 N
the mean value
1
lim -T
T-oo
+ P(wt)dt = Po = -()m
1
271"
exists and does not depend on
T.
1 1
2
"-
...
,.-
P(4J)d4JI ... d4Jm
Thus, there also exists a similar value for
a quasi-periodic function F which can be uniformly approximated by these

polynomials. This leads to the well-known mean value theorem.
12
CHAPTER 1
Theorem 3. Let F(t)
lim -1
T-oo
= f(wt)
+ F(t)dt
be a quasi-periodic function. Then the limit
1 1
2
1
..
..
= Fo = -()m
...
f(</J)d</Jl ... d</Jm
211"
(4.5)
exists and converges uniformly with respect to T E R.

The limit (4.5) is called the mean value of the function F. As can be seen
from (4.5) it does not depend on T E R.
The set of all quasi-periodic functions with a frequency basis w forms a
linear space which we shall denote by C(w). By introducing a norm 11 in C(w)
using
!Flo = sup IIF(t)lI,

tER
we turn C(w) into a normed space, which we shall denote by CO(w). Taking
into account the fact that the space CO(Tm ) is complete and using equality
(4.3) we see that the following theorem holds.
Theorem 4. The space CO(w) is a Banach space.
The inner product (/,g)o of functions f,g E CO(Tm ) induces an inner
= f(wt)
product on CO(w) of the functions F
and G
= g(wt) E CO(w).
Using
(4.5), this inner product is given by the following equality:
liT
(F,G)o = lim -T
T-oo
(F(t),G(t)dt = (/,g)o.
(4.6)
The inner product (4.6) induces on CO(w) the norm
I!FII~
= (F, F)o.
The family of functions {ei(k,w)t} forms a complete orthonormal set in C(w).

Any function F E C(w) can be represented as a Fourier series with respect to
this set:
F(t) ::::::
L Fkei(k,w)t,
k
Fk
liT
lim T-oo T
F(t)e-(k,w)ldt,
and it satisfies Parseval's equality
L l!FkW = IIFII~
k
(4.7)
13
In the space CO(w) of quasi-periodic functions we choose the subspace cr(w)

of all functions F for which representation (4.1) is defined by a function
Cr(Tm ). The function F E cr(w) has r continuous derivatives with respect

to t and each derivative is a quasi-periodic function. The space cr(w) can be
turned into a complete space if one defines a norm
I. Ir
by setting
IFlr = Iflr
whenever F(t)
= f(wt),
where f E Cr(Tm
).
1.5. The spaces Hr(w) and their structure
Suppose that the function Ek !ke i(k,4 belongs to the space Hr(Tm ), and let
= (Wl, ... ,wm )
be positive incommensurable numbers. We form a function F(t) by defining it to be the series Ek fkei(k,w)t. The set of functions of
this type forms a linear space which we denote by Hr(w). For any two functions F and G belonging to Hr(w) and given by the series Ek !kei(k,w)t and
Ek gkei(k,w)t, we can define an inner product and a norm

(F,G)r =
E (1 + IIkIl nfk,g-k),
2
IIFII: =
II . IIr:
(F,F):.
Since the space Hr (w) is the completion with respect to the norm
space P(w) generated by polynomials of the form pet)
of the
= Ellkll~N Pkei(k,w)t,
= n Hr(w).
r=O
is quasi-periodic with the frequency basis w = (Wl, ... ,wm ),
the space Hr(w) is a Hilbert space. We set Hoo(w)

If a function F
II .IIr
00
it follows from Parseval's equality (4.7) that its Fourier series Lk Fkei(k,w)t
belongs to the space HO(w). So, each such series can be identified with a quasiperiodic function for which it is the Fourier series. To find out what other
elements of the space HO(w) can be identified with a quasi-periodic function
one needs an extension of the above notion of a quasi-periodic function due to
P.Bol [25]. This extension leads to the notion of a quasi-periodic function in
the sense of Besicovich [142].
Definition. A function F(t) defined on the real line R and square summable
on any finite line interval is called a Besicovich quasi-periodic function with
CHAPTER 1
14
frequency basis w
= (w}, ... ,wm )
if there exists a sequence of quasi-periodic
polynomials Pv E P( w) such that
1
lim lim 2T
lI~ooT_oo
fT
-T
2 = O.
IIF(t) - Pv (t)1I dt
The set of all Besicovich quasi-periodic functions with frequency basis w

forms a metric space, where the distance between two functions F and G is
defined to be
{ liT
p(F, G) = }~ T
II F (t) - G(t)1I 2dt
}1/2 .
We denote this space by B 2 (w). The following theorem due to Besicovich [142]
gives an important property of this space.
Theorem 1. The space B 2 (w) is complete.
The proof of Theorem 1 is constructive and the interested reader can find
it in [64].
Since the space B 2 (w) is complete, the limit of a sequence of quasi-trigonometric polynomials Pv(t) = Lllkll:5 N v f~v)ei(k,w)t convergent with respect to the
mean square belongs to this space. Because the function Lk !kei(k,w)t in HO(w)
is obtained by taking closure of the space P(w) with respect to the norm 11110,
it corresponds to a Besicovich quasi-periodic function F(t). Conversely, for
any Besicovich quasi-periodic function F(t) E B 2 (w) there exists in B 2 (w) a
Cauchy sequence of quasi-trigonometric polynomials Pv(t) E P(w) converging
to F(t). The equality
which holds for the polynomials Pv(t) (1/ 1,2, ...), shows that the sequence
P v (1/ = 1,2, ...) is a Cauchy sequence in HO(w). Therefore the limit of the
sequence Pv (1/ = 1,2, ...) belongs to the space HO(w). Thus we see that there
is a correspondence between the functions of the space B 2 (w) and the space
HO(w). That is the space HO(w) can be identified with the space of Besicovich
quasi-periodic functions B 2 (w). This proves the following theorem.
Theorem 2. The space HO(w) is the space of Besicovich quasi-periodic func-
tions with frequency basis w.
15
Now the series Ek Ikei(k,w)t which represents a function of HO(w) can be

obtained from the Fourier series Ek Ikei(k,<p) of a function 1(4)) E HO(Tm ) by
formally replacing J by wt. Since the function I(J) is measurable on the cube
of periods Tm , it can be identified with every function of HO(Tm ) that is equal
to
1 almost
everywhere on Tm
The set of J = (wt) mod 271", t E R has zero
Lebesgue measure on Tm . Consequently, the function 1(4)) with its Fourier

series Ek hei(k,<p) E HO(Tm ) can be identified in HO(Tm ) with a function !(J)
which on the line J
= wt,
t E R, takes values equal to the values of the
Besicovich quasi-periodic function F(t) generated by the series Ek hei(k,w)t E
HO(w). This leads to the following equality

F(t) = !(wt),
(5.1)
which associates with the Besicovich quasi-periodic function F(t) a measurable square summable periodic function of many variables !( 4 on the cube of
periods Tm .
Equality (5.1) leads to the following relation
lim - 1
T-oo
2T
IT
-T
F(t)e-(k,w)tdt =
which shows that the series Ek !kei(k,w)t is a Fourier series of the Besicovich
quasi-periodic function F(t). It also leads to the equality:
L IIhl1
k
IT
2
1
= Tlim- o2T
IIF(t)1I dt,
o_T
which is Parseval's equality for the function F(t) E HO(w).

We say that a Besicovich quasi-periodic function F(t) has the generalized derivative d! /dt r if there exists a function F(r)(t) in HO(w) satisfying the
equality
(5.2)
for any trigonometric polynomial P(t) E P(w). The inner product in (5.2) will
be understood in the usual sense for quasi-periodic functions:
(F, P)o
1
= T-oo
lim -T
2
IT
-T
(F(t), P(t)}dt.
16
CHAPTER 1
Let the series Lk !kei(k,w)t define an element of the space Hr(w). The
function F(t) with the above Fourier series belongs to the space B 2 (w). The
function F(p)(t) with Fourier series Lk[i(k,w)]Pfkei(k,w)t also belongs to this
space for any 0 ::s p
::s r.
Moreover, since
(F(p),P)o
= L[i(k,w)jP(fk,P- k ) =
k
= (-1)P L(!k, [i( -k,w)jP P-k) = (-l)P(F, d PPjdtP)o,

k
F(p)(t) is the pth generalized derivative of the function F(t). This shows that
each function F(t) E Hr(w) can be identified with a Besicovich quasi-periodic
function having generalized derivatives. Thus the following theorem holds.
Theorem 3. The space Hr(w) contains the Besicovich quasi-periodic junctions
having generalized derivatives up to order r inclusive.

However, the space Hr(w) does not coincide with the entire space of Besicovich functions having generalized derivatives of order r. This can be seen by
considering the following example.
Consider a function
F(t)
~ cos(kn,w)t
= ~ II kn ll m+1
'
(5.3)
m ~ 2,
= (k?, . .. ,k;:') is an integer vector satisfying the conditions:
where kn
(5.4)
Since the series
converges for r
> 0 because
00
= CL
11=1
00
v-(m+1)
L
IIkll=1I
l::S Cl L
11=1
v-(m+1)+m-1
00
C1
L
11=1
1/-2
C2
< 00,
17
the function F(t) is infinitely differentiable. On the other hand, since
for 2r
r
m + 1, the
+ 1)/2 > 1.
(m
function F(t) does not belong to space Hr(w) when

Thus, the function F(t) has derivatives of any order
but does not belong to Hr(w) when r
(m + 1)/2 > 1.
Finally, we also need to see that conditions (5.4) hold for a sequence of
integer vectors k n (n = 1, 2, ...). This can be achieved by using the following
result from number theory [139].
Lemma.
For any irrational number a, there exists a sequence of integers
Pn,qn (qn ~ 2(n-l)/2 for n ~ 2) such that
1
1-.,..------:< a - Pn
- < -qn(qn+l + qn)
qn
qnqn+l'
(5.5)
As a consequence of this lemma we have inequality

n
+ kn2 W 2 I = k n2 w I I-W2
+ kki2n I < k 2n WI k n k1n+l
Ik IWl
WI
::;
2 2
(WI>
for j
(5.6)
ki = -Pn and k~ =
0). Inequality (5.6) leads to conditions (5.4) for m ~ rand kj = 0
which can be obtained from (5.5) by setting a

qn
2- n/2 WI,
= WdWI,
3.
Sobolev theorems similar to those given in 1.2 hold also for the spaces
Hr(w). The first theorem asserts that Hr(w) C CP(w) for r> m/2 + P, while
the second asserts that a bounded set of functions in Hr(w) is compact in
H3(W) for s < r.
Because the Sobolev theorems hold, the space Hr(w) has advantages over
other spaces of quasi-periodic functions that have derivatives up to order r
inclusive. For example, the spaces 71'r(w) which contain all quasi-periodic functions with frequency basis wand whose rth order derivatives are also quasiperiodic functions. The norm
1lr = 0:5P:5
maxr supl~1
tER dtp
turns 71'r(w) into a normed space, but the Sobolev theorems do not hold. This
can be seen by considering the sequence of functions
sin(kn,w)t,
n=I,2, ...
,m~2,
(5.7)
18
CHAPTER 1
which belong to space ?I"r (w) for any r
l(k",w)1
(k",w)
~ 1,
= (k?, ... , k:;') are chosen
0 only if k"
in such a way that

-+
(n
-+
00).
Even though the sequence (5.7) is bounded in ?I"r(w) it is compact in ?I"(w).

1.6. First integral of a quasi-periodic function
Let
F(t) = f(wt),
f(r/J) E C(Tm ),
be a quasi-periodic function with frequency basis w = (Wi> ... , w m ) having zero
mean value:
liT
F = lim T-+oo
Consider the first integral
J(t)
By considering the function
F(t)
F(t)dt.
(6.1)
sin(k",w)t ,
"=1
where k"
it
F(t)dt = 0.
(6.2)
n 2 sign(k n ,w)
= (kf, . .. ,k:;') is an integer vector satisfying the conditions

l(kn,w)l~n-2,
I/k"II#O, n=1,2, ... ,
we see that J(t) is not always a quasi-periodic function.

Indeed, since series (6.2) is uniformly convergent, it can be integrated term
by term. Thus,
J(t)
= ~ (1- cos(k",w)t) =2 ~ sin 2(k",w)tj2

~
n21(k",w)1
n21(k",w)1
and therefore
~ sin 2(k",w)tj2
supJ(t) ~ 2sup L..
21(k" )1 ~
tER
tER "=1
,w
19
The last inequality shows that the function J(t) is unbounded on R, therefore
J(t) is not quasi-periodic.

The following main result which gives the conditions for the function (6.1)
to be quasi-periodic is due to P. Bol [25], [43].
The first integral J(t) oj a quasi-periodic junction F(t) IS a

quasi-periodic junction ij and only ij it is bounded on R.
According to the definition of a quasi-periodic function, it is bounded on
R. So to prove Theorem 1 we need to show that if the function (6.1) is bounded
then it is quasi-periodic. Thus, let
Theorem 1.
IJ(t)1
M, t E R.
We define the function
which satisfies the obvious relation
4>T(wt) = J(t
+ T) -
J(t),
t, T E R.
(6.3)
Since the function J(t) is bounded on R, it follows that for some sequence of
numbers
the sequence
1
Nk
converges as k
iNk J(T)dT,
0
k=I,2, ...
-+ 00.
We define a sequence of functions in C(w) by setting
k=I,2, ....
According to equality (6.3) we have
~k(wt)
= -J(t) + Nk
iNk [J(t + T) - J(r)]dr,

0
k = 1,2, ... ,
(6.4)
20
CHAPTER 1
and after some obvious transformations we obtain
'ltk(wt)
+ J(t)
it
= ~k iNk
= ~ t jNk+'f(WT)dTdS = t
Nk Jo
Jo
J(N1:
f(ws
+ s) N1:
+ WT)dsdT =
J(s) ds,
= 1,2, ....
(6.5)
Since J(t) is bounded on R and in view of the relations (6.5), it follows that
the limit
lim ['It A;(wt)
1:-00
+ J(t)] = 0
(6.6)
converges uniformly on any bounded interval Ie R.

We claim that limit (6.6) is uniform on the whole of R. According to (6.4),
this is equivalent to the fact that
1 iNk
lim -N
k-oo
[J(t
+ s) -
J(s)]ds
=0
(6.7)
is uniform on the whole of R.

We shall prove (6.7) using ideas of H. Bohr [26], [143], S. Bochner [43].
Following [65], we choose an arbitrary sequence of numbers t n such that wt n
converges to modulo 271":
lim wt n
n ..... oo
Consider the sequence J(t
= mod
271",
E Tm .
+ t n ) of shifts of the function
J(t). Since
(6.8)
this sequence is compact in the sense of local convergence on R, that is, it is

uniformly convergent on any bounded interval of R. Let J(t) be a limit point
of sequence (6.8). Passing to the limit in (6.8) we obtain
J(t)
= J(O) +
it
f(ws
+ )ds.
(6.9)
Let us show that

sup J(t)
tER
= sup
J(t),
tER
inf J(t)
tER
= tER
inf J(t).
(6.10)
21
Indeed, since J(t) is a limit function of some subsequence of {J(t
+ tn)},
inf J(t) :5 inf J(t) :5 sup J(t) :5 sup J(t).
tER
tER
tER
tER
Consider the sequence J(t - tn). According to (6.9) we have
J(t - tn)
[t-tn
= J(O) + Jo
[t
+ c/J)ds = J(O) + Jo f(ws - wtn + c/J)ds+
f(ws
1-tn
f(ws
+ c/J )ds = J( -tn) +
and hence it is clear that the sequence
J (t -
it f(ws - wtn + c/J)ds,
t n ) is compact in the sense of local
convergence on R. Let J 1 (t) be a limit point of the sequence J (t - t n ). Passing

to the limit in (6.9) we see that
Now we have
J1 (0) + tER
inf J(t) = inf(J1 (0) + J(t))
tER
~ inf J(t) ~ inf J(t),

tER
tER
J1 (0) +supJ(t) = sup(J1 (0) + J(t)):5 supJ(t):5 supJ(t),

tER
tER
tER
tER
which is a contradiction if equalities (6.10) do not hold.

We now show that the convergence J(t
+ tn)
J(t) is uniform on the

whole of R. Suppose the contrary. Then there exists a sequence {Sn} C R such
-+
that
(6.11)
Without loss of generality we can suppose that the sequence Sn is such that
lim wS n
n-oo
= c/Jo mod
211",
c/Jo E Tm .
Consider the sequence of functions
J(t
+ Sn + tn) = J(sn + tn) +

J(t
+ Sn)
=J(Sn) +
it f(ws + wSn + wtn)ds,
f(ws
+ wS n + 4J)ds.
22
CHAPTER 1
Their local limits on Rare
Jt(t)
= Jt(O) +
J 2 (t) = J 2 (0)
1
t
f(ws
+ 4>0 + 4ds,
f(ws
+ 4>0 + 4ds,
1
t
where Jt(O) and J2 (0) are the corresponding limit points of J(sn
+ tn)
and
J(Sn). It follows from equalities (6.10) that

sup J1(t) = J1(0)
teR
sup J2 (t)
teR
+ M 1 = sup J(t),
teR
= J 2 (0) + M 1 = sup J(t) = sup J(t).

teR
teR
But then Jt(O) = J 2 (0), which contradicts inequality (6.11), according

to which the equality lim J(snv
v~oo
+ tnJ =
contradiction shows that the convergence
J(t
lim J(snJ is impossible. This
v~oo
+ tn) --+ J(t)
(6.12)
is uniform on the whole of R.

We now show that relation (6.12) implies that for any { > 0 there exists a
positive number I
= I({) such that any interval I = [T, T + I] of R contains at
least one number T
= T({) for which the inequality

I
J (t
+ T) -
J (t) 1 < {, t E R
(6.13)
>
0 such that one cannot
is true. For suppose not. Then there exists an {o
choose a corresponding I({o). Hence for any I > 0 there is an interval IT =
[T -I, T + I] of R such that for any point () E IT we have

sup IJ(t
teR
+(}) -
J(t)1
(6.14)
~ to
We construct a sequence of intervals I Tn = [Tn -In, Tn
+ In]
(n = 1,2, ...)
of this type by taking h to be arbitrary, and In, Tn satisfying the condition
In > max lTv I, n = 2,3, ....

lI<n
(6.15)
Condition (6.15) ensures that Tn - Til E I Tn for v < n. Consider the sequence
J(t+Tp ) (p= 1,2, ...) of shifts ofthefunction J(t). Let Tpj (j = 1,2'00') be a
23
Tn (n = 1,2, ...) such that WTpj converges modulo

= </> mod 211" for some </> E Tm The subsequence
subsequence of the numbers

211" as j
J(t
-+ 00 ;
.lim
WTp .
J-OO
1
+ Tpj ) (j = 1,2, ... ) is compact in the sense of uniform convergence on R:

J(t
+ Tpj ) -+ J(t)
(6.16)
uniformly on the whole of R. However, when Pv < Pn we have the following

estimate for sup IJ(t
tER
sup IJ(t
tER
+ Tpn ) -
+ Tpn ) J(t
J(t
+ Tpv)l:
+ Tpv)1 = sup IJ(t + Tpn tER
Tpv ) - J(t)1
(6.17)
~ fO.
This is true since Tpn - Tpv E I Tpn = [Tpn - lpn, T pn + lpn].

Inequality (6.17) means that the sequence J(t+Tpj ) is not compact in the
sense of uniform convergence on R. The latter statement contradicts relation
(6.16). This proves inequality (6.13). Now it is evident that the convergence
(6.7) is uniform on R. Indeed, for any f > 0 there exists I = l(f). such that on
any interval 1= [t,t + l(f)] there exists a point T = Ti(f) such that
IJ(s + T) - J(s)1 < f, s E R.
But then, for any t E R
1
1 fNk
N", 10 [J(t
+ s) -
I
11 iT
I
11 fNkH
I
J(s)]ds $ N", t J(s)ds + N", 1N +T J(s)ds +
k
+ 1N",
10fNk [J(s + T) -
2Ml
J(s)]ds $ N",
+ f,
which proves that the convergence of (6.7) is uniform on R.

Thus we have shown that
J(t)
=-
(6.18)
lim Ilf",(wt)
"'-00
in the sense of uniform convergence on R. It follows that the sequence Ilf'" (</ E
C(Tm ) of functions is Cauchy with respect to the norm of the space C(Tm ):
max IIlfk+p(</ - Ilf",( 0, P = 1,2, ... , k ~ N = N(f). Since this space is complete,
the sequence Ilf",(</ converges uniformly with respect to
</>
E Tm to a certain
function in the space C(Tm ):

lim Ilf",(</
"'-00
= Ilf(</,
Ilf(</ E C(Tm ).
(6.19)
24
CHAPTER 1
Relations (6.18) and (6.19) show that the function J(t) belongs to the
space C(w), as required. From the above proof we obtain the formula
J(t)
where j
= j + ~(wt),
= Tlim
~ Jd J; f(ws)ds dr is the mean value of the function
..... oo
J(t)
J~ f(ws)ds, ~(4)) is the function in C(Tm ) given by the formula
~(4)) = - Tlim
fT r
..... oo 1 1
0
f(ws
+ 4ds dr,
(6.20)
where the limit is uniform with respect to 4> E Tm . This formula gives a relation
between the functions f(4)) and
~(4))
which define the corresponding function
F(t) and its first integral J(t).

According to (6.20), the Fourier series of the function
~(4))
can be deter-
mined from the Fourier series of the function f(4)) ~ Ek;to !kei(k, by using
the relation
~(4)) ~ L>-~ ei(k,

k;to
t(k,w)
(6.21)
As is clear from (6.21), the "smoothness" property of the primitive, that is,
its membership of the space Cr(Tm ) or Hr(Tm ) with r 2: 1, depends essentially
on the arithmetic properties of the frequency basis w of the function being
integrated. In particular, this leads to the following result on the "smoothness"
of the primitive.
Theorem 2. Let F(t) = f(wt) E Hr(w) n C(w), F = 0 and suppose that the
frequency basis w (WI, ... ,wm ) satisfies the inequality
l(k,w)l2: Kllkll- d
for any integer vector k = (k 1 , ... ,km
),
(6.22)
k"# 0, and for constants K > 0 and
d > O. If the primitive J(t) = f~ F(t)dt E C(w) and r 2: d, then J(t) E

Hr-d(w) and the function ~(wt) = J(t) - J satisfies inequality
The assertions of the theorem follow from the chain of estimates
25
PERIODIC AND QUASI.PERIODIC FUNCTIONS

d
Z d
S IC Z2: II k W
(1 + II k ll y- llfk liZ S
k;tO
S g-Z 2:(1 + IIkll ZYllfkll Z= g-2I1f(4))II;.

k;tO
From this theorem and taking into account Sobolev's theorem on embedding the spaces H r (Tm ) into C'(Tm ) it follows that if the conditions (6.22) hold
for the function F(t)
= f(wt) E cr(w), then its primitive J(t) = f~ f(wt)dt =
J + ~(wt) is a member of C'(Tm }, where s is the largest integer satisfying the

inequality
r - d- mJ2
> s.
(6.23)
The value d + mJ2 determines the "loss of smoothness" of a quasi-periodic

function in the process of taking its first integral. It should be noted that
the set of frequencies (WI, ... ,wm ) which satisfy inequality (6.22) is a rather
"powerful" set even for d
= m + 1.
Theorem 3. In the ballllwil
In fact we have the following theorem [9].
S 1, the Lebesgue measure of the numbers w =
(WI, ... ,wm ) that do not satisfy (6.22) approaches to zero for d = m
+1
as
g -. O.
A frequency basis satisfying (6.22) for d
= m + 1 is
sometimes called
a "strongly incommensurable" basis. As is seen from (6.23), for a strongly

incommensurable frequency basis, the loss of smoothness of a function from
the space
cr (w)
due to taking the first integral is determined by the value of
3mJ2 + 1.
1.7. Spherical coordinates of a quasi-periodic vector function
Let the formulas
= Rcos8 l ,
Xz = Rsin8 l cos8 2 , ... ,
Xl
Xn-l
Xn
= Rsin8lsin8z,.sin8n_zcos8n_1>
(7.1)
+ Rsin 81 sin 8z ... sin 8n - z sin 8n - l
define the relations between Cartesian coordinates x = (X1> ... , x n ) and spher-
ical coordinates 8 = (81)'''' 8n-d, 0 S 8j S 11' (j = 1,,.., n - 2), -00 <

8n- l < 00 introduced on a sphere SR(C) of radius R and with centre at the
point Xo = (0, ... ,0, c).
26
CHAPTER 1
= (II (<b), ... ,fn (<b with the Cartesian

= /j(<b) E c:r(Tm ) for j = 1, . .. ,n. For
Consider a vector function f (<b)

coordinate functions
Xj
the spherical coordinates
for this function on the sphere S R( c) are defined and there arises the question of
how the coordinates O(<b) depend on <b. The answer is provided by the following
theorem.
Theorem. Let f(<b) E Cr(Tm ) and
(7.2)
Then the spherical coordinates O(<b)

satisfy the conditions
OJ(<b) E Cr(Tm ) Vj
= (0 1 (<b), ... ,On-I{ <b of the function
= 1, ... , n -
2,
8n- l (<b)
= (p, <b) + <fl(<b),
f( <b)
(7.3)
where p = (PI, ... ,Pm) is an integer vector, <fl(<b) E Cr(Tm ).

It should be noted that for n = 2 and 2 can be reduced to the case of n
= 2.
Indeed, suppose that
the theorem is true for any pair of functions fn-l(<b) and fn(<b) - c satisfying
condition (7.2). Then it follows from the representation
fn-l(<b)
fn(<b)
= Rl(<b) COSOn-l(<b),
= + R 1(<b)sin8n- 1(<b)
C
=J
that R l (<b)
f~-l (<b) + (fn( <b) - c)2 E Cr (Tm ) and 8n - 1 (<b) (p, <b) + <fl(<b),
where P = (Pll'" ,Pm) is an integer vector, <fl(<b) E Cr(Tm ). Relation (7.2)
also holds for the pair of functions fn-2(<b) and Rl(<b), which leads to the
representation
= R2(<b)COSOn-2(<b),
R1(<b) = R 2(<b) sin On-2(<b),
fn-2(<b)
where R2(4)) =
27
JfL2(4)) + R?(4)) E C"(Tm) and the function On-2(4)) has the
form of (7.3):
Since the functions R I and R2 are positive, sin On-2(4)) = R I (4))/ R 2(4)) > 0
for all 4> E Tm and thus p(l) = 0 in (7.~). This proves that the function On-2(4))
belongs to the space C'"(Tm ). By considering further pairs offunctions fn-3(4))
and R 2(4)).!n-4(4>) and R3(4)) =

fL3(4)) + !q(4)), etc., we see that the
theorem holds assuming that it is true for n 2. Thus it remains to prove the
theorem for n = 2. In the proof we shall follow the reasoning of Jensen [65].
We set
and consider the polar coordinates R(4)),O(4>) for the pair 91(4)),92(4>). They
are defined by the relations:
91(4))
= R(4)) cos 0(4)),
92(4))
= R(4)) sin 8(4)).
In view of inequality (7.2), R(4)) E C"(Tm ) and fixing a continuous branch of

the function tan- I (92(4/91 (4))), for example, by the condition
_~ < 0(0) = tan- 1 92(0) < ~

2 -
91(0) - 2'
we can take it to be 0(4)).

Since the functions 91 (4)),92(4)) are periodic and 0(4)) is continuous, it
follows that for every 1 ~ j :5 m there exists an integer Pi such that
We set P
= (PI, ... , Pm) and write 0(4)) in the form

0(4))
= (p,4 + 4>(4)).
It follows from (7.5) that 4>(4)) E C(Tm
).
From the relations sinO(4)) E C"(Tm), cosO(4)) E C"(Tm ) and 4>(4)) E

C(Tm ) it becomes evident that 4>( E C"(Tm ).
28
CHAPTER 1
1.8. The problem on a periodic basis in En

Let
(8.1)
= 1, ... , r)
be a n x r-matrix formed by the columns Uj(4)) (j
that belong to
the space C/(Tm ) and are linearly independent for all 4> E Tm so that
rankU(4)) = r,
For r
4> E Tm
= n, the vectors Uj (4)) (j = 1, ... , n) form a basis in the Euclidean space
En for any 4> E T m . This basis is called periodic and orthonormal if the vectors
'Uj(4)) (j = 1, ... , n) are orthonormal for every 4> E Tm :
where
Oij
For 1
is the Kronecker symbol.
s: r < n there arises the problem whether there exists an n x (n -
r)
matrix
(the columns Vj(4)) (j = 1, ... , n - r) of which belong to the space C/(Tm ) and
are linearly independent for all 4> E Tm ) such that the system of vectors
forms a periodic basis in En such that
det(U(4, V(4))) :/; 0, 4> E Tm
This problem is called the problem of extending a periodic r-system of vectors

U( 4 to a periodic basis in En, or the problem on a periodic basis in En.
The problem on a periodic basis in En has an elementary solution of

n=r+l.
In this case a complementary vector V(4))
= (v~(4, ... ,v~(4))) can be formed
using elements of the matrix U (4)) by setting

Vj(4))=(-l)j-l~j(4>), j=l, ... ,n,
29
where tlj (4)) is the determinant of the matrix obtained from U (4)) by taking out
the jth row. With such a choice of the vector V(4)), the matrix (U(4, V(4)))
satisfies the condition
ensuring that the (n - I)-system (8.1) can be extended to a periodic basis in
En by the vector V(4)).

Generally, the solution of the problem on a periodic basis in En depends
on relations between r, nand m. We give a solution based on the author's
paper [110]. First we prove some auxiliary statements.
Let the column vector
have the coordinate functions
= sin 4>1 sin 4>2'" sin 4>j-1 cos4>j, j = 1, ... , n e~ ( 4>1> ... , 4>n- d =sin 4>1 sin 4>2 ... sin 4>n- 2sin 4>n-1
ej(4)1'''' ,4>j)
1,
which give the formulae for transition between Cartesian and spherical coordinates introduced on the sphere S 1(0). For j = 2, ... , n we form the column
vectors ej
= ej( 4>j -1, 4>j, ... ,4>n- d by setting
(8.3)
Lemma 1. The system of vectors (el, e2, ... , en) forms a periodic basis in En.
Indeed, according to formula (8.3), the vector ej has the form
ej = (0, ... ,0, - sin 4>j-l, cos 4>j-l cos 4>j, cos 4>j-l sin 4>j cos 4>j+l, ... ,
COS4>j_l sin 4>j ... sin4>n_2cos4>n_1>Cos4>j_lsin4>j ... sin4>n-l),
for j
(8.4)
2. This leads to the equality
(ej,ej)
= 1,
j=2, ... ,n.
(8.5)
30
CHAPTER 1
Moreover, if j
2 then, as follows from formulae (8.4), we have the chain
of inequalities for ej:
1
~_
e,._- Sm'f'lSm'f'2sm'f'j-2
.
U'f'j-l
A.
A.
A.
f;;A.
02 el
- cos<Plsin<P2 ... sin<pj-2 O<PlO<Pj-l - ...
1
1
oj-leI
... - COS <PI cos <P2 ... cos <Pj -2 O<PlO<P2 ... O<Pj -1
But then
ej+l
1
oe'
= cos <Pj
-',
-1 oJj
ej +v =
1
oe
..
'
,
sm
<PHv-2 OJHv-l
Cos<Pj_lsmJj' ..
1/
= 2, ... ,
so that the value of(ej,ej+v) is proportional to the value of (ej,oej/oJj+v_l)

which is equal to
j
( ej, oJ'oe
,+v-l
And so the vectors ej (j
= 2' oJ.
ej,ej = O.
,+v-l
= 1, ... , n) are orthogonal.
We now return to the matrix (8.1). Schwartz's orthogonalization process

enables us to represent it in the form
U(J)
= O(J) T(J),
(8.6)
where O(J) is an n x r-matrix with orthonormal columns and T(J) is an r x r

non-singular lower triangular matrix. The elements of the two matrices belong
to the space C/(Tm ). We take an arbitrary n x (n - r)-matrix V(J) and form
the square matrix
cJ>(J) = (u(J), V(J)),
by extending U(J) to an n x n-matrix.
Relation (8.6) shows that
(U(J), V(J)) = (O(J)T(J), V(J))
= (O(<p), V(<p)) (T~J) ~),
(8.7)
31
where E is the (n - r)(n - r) identity matrix.

Taking into consideration equality (8.7) it is easy to see that the following
statement holds.
Lemma 2. For the r-system of vectors (8.1) to be extendable to a periodic

basis in En it is necessary and sufficient that the r-system of vectors 0(4)),
obtained by orthogonalization of the columns of the matrix U (4)), be extendable
to an orthonormal periodic basis in En.
Indeed, if the matrix V(4)) E C/(Tm ) extends U(4)) to a periodic basis in
En, then V (4)) also extends 0(4)) to a periodic basis in En because the matrix
(0(4, V (4))) = (U(4, V(4))) =
[T-~(4))
;]
is periodic and non-singular.

Making the columns of the matrix (0(4, V(4))) orthonormal, we find a
periodic matrix 0 1 (4)) which extends 0(4)) to a periodic orthonormal basis in
En.
Lemma 2 relates the problem on a periodic basis in En to the problem of
finding linearly independent periodic solutions of the system of linear equations
x'U(4)) = 0,
(8.8)
where x' is a row vector.

Indeed, according to (8.6), system (8.8) is equivalent to the system
x'O(4)) = 0,
(8.9)
and the matrix 0 1 (4)) which extends 0(4)) to an orthonormal basis in En,
satisfies the identity
0~(4))0(4>)
= 0,
(8.10)
where denotes the matrix transpose. Consequently, the columns of the matrix
0 1 (4)) define a system of linearly independent periodic solutions of equation
(8.9).
Conversely, after making any n - r linearly independent solutions of equation (8.9) belonging to the space C/(Tm ) orthonormal, they form a matrix
0 1 (4)) satisfying identity (8.10). Consequently, they form a matrix which extends 0(4)) to an orthonormal periodic basis in En.
CHAPTER 1
32
Lemma 3. Suppose that for any integer p satisfying inequality
(8.11)
n-r+l~p~n
and any vector V(4))
= (Vl(4, ... , v
(4))), IIV(4))1I
= I,
belonging to C'(Tm
),
the equation
(8.12)
has p - 1 linearly independent solutions belonging to C'(Tm ).
system of vectors U(4)) = (U1(4, ... , u r (4))) in C'(Tm
Then any r-
can be extended to a
periodic basis in En.
To prove Lemma 3 it is sufficient to find n - r linearly independent periodic solutions of system (8.8) or the equivalent system (8.9). We denote the
orthonormal columns of the matrix 0(4)) by ej (4)) (j
= 1, ... , r)
and rewrite
system (8.9) in the form

(8.13)
Consider the first equation of (8.13). It follows from our assumptions that
this equation has p - 1
=n -
1 linearly independent solutions belonging to
the space C'(Tm ). We denote them by V2(4)), ... , v n (4)) and make the system
of vectors e1(4)), V2(4)), ... , v n (4)) orthonormal. This gives the orthonormal
matrix
which forms a periodic basis in En. Using the matrix
4)1 (4))
we make a change
of variables in (8.13) by introducing y' = (Y1, Zl), Zl = (Zl' ... ,Zn-l) according
to the formulae
Zl
y ' 4)i (4)).
Since the vector e 1 (4)) is orthogonal to the vectors ej (4)) (j
= 2, ... , r) and
to the vectors which form the matrix V1(4)), the system of equations (8.13) in
the new coordinates will take the form
Y1
or, equivalently,
Y1
=0,
=0,
=1
z'U1 (4))
= 0,
(8.14)
r ~ 2,
where the matrix U1 (4)) defined by the expression
(8.15)
33
is an (n - 1) x (r - I)-matrix of rank r - 1 belonging to the space C'(Tm ).

The unit basis vectors (0, 1,0, ... ,0), ... ,(0, ... ,0,1) clearly form n - r
= 1) linearly independent solutions of system (8.14).

the lemma for r = 1.
n - 1 (for r
To prove Lemma 3 for r
This proves
2 consider the system
(8.16)
which consists of r - 1 equations with n - 1 unknowns.
Making the matrix U1 (<! orthonormal, we obtain a system of equations of
the form (8.9), equivalent to (8.16):
(8.17)
Since the conditions of the lemma apply to the first equation of this system, it
has p-l = n-2linearly independent solutions belonging to the space C'(Tm ).
Using them, we transform system (8.17) to the form similar to (8.14):
= 0,
fh
r -
= 1,
or to the form similar to (8.15):
j72
= 0,
z'U2 ()
= 0,
r - 1 ~ 2,
where U2(<! is an (n - 2) x (r - 2) matrix belonging to C'(Tm ) or rank r - 2.

The composite of the above transformations reduces the initial system of
equations (8.8) to the system
Yl
= 0,
jh
= 0,
=2
(8.18)
or to the system
Yl
= 0,
j72
= 0,
Z'U2(<!
= 0,
r ~
3.
In the first case the unit basis vectors (0,0,1,0, ... ,0), ... ,(0,0, ... ,0,1)
form n - r = n - 2 linearly independent solutions of system (8.18), and this
proves the lemma for r = 2. In the second case, the transformations similar to
the ones performed for system (8.16) lead to the system of equations
34
CHAPTER 1
Continuing this transformation process applied to system (8.8) r times, we

reduce it to the system
Y1
= 0, fh = 0, ... , Yr = 0.
(8.19)
The unit basis vectors
(0, ... ,0,1,0, . .. ,0), (0, ... ,0,0,1,0, ... ,0), ... , (0, ... ,1)
clearly form n - r linearly independent solutions of system (8.19). This proves
the lemma for an arbitrary r, 1
n.
Lemma 4. Suppose that the set of values of the function
is not everywhere dense on the unit sphere Sl (0):
Then the equation
(8.20)
has n - 1 linearly independent solutions belonging to the space C'(Tm
).
The proof of Lemma 4 follows from the statements of Lemma 1. Indeed,

if the set of values of the function u( 4 is not everywhere dense on 51 (0), then
there is a a-neighbourhood of a point p E 5 1 (0) containing no values of the
function u(4)). Without loss of generality we can assume that the point p has
the coordinates (0,0, ... ,0,1). Under this assumption the coordinate un (4)) of
the vector u( 4 satisfies the inequality
where a = const
< 1. Thus the coordinates
Un
-1(4)) and u n (4)) of the vector
u(4)) satisfy the condition

(8.21)
for an arbitrary c E (a, 1) and some I<
= Ic) > O.
35
Let
denote the spherical coordinates of the vector u( 4 defined on the sphere with
centre (D,D, ... ,e) and radius
In view of inequality (8.21) we can apply the theorem from 1.7 to tP(4)) and
thus we see that the coordinates of this function are appropriately smooth and
periodic. The transition formulae relating Cartesian and spherical coordinates
allow one to write the vector u( 4 in the form
(8.22)
where el(4)) is the vector (8.2) and eo
= (0, ... ,0,1) is a unit basis vector.
Using
representation (8.22) we now give linearly independent solutions of equation

(8.20) belonging to C/(Tm
).
Let us start with the following system of vectors:

(8.23)
where the
ej (4))
are the orthonormal vectors (8.3).
We show that the system (8.23) forms a periodic basis in En. Since the
functions tPj(4)) satisfy conditions (7.3) (if in (7.3) we replace 0 and r by
tP and
1 respectively), the functions (8.23) belong to the space C/(Tm ). It remains to

show that vectors (8.23) are linearly independent. To see this, we evaluate the
determinant of the matrix
Taking into account (8.22), we have

det q>
= R( 4 + c det( eo, e2( tP( 4>)), ... , en ( tP( 4)).
Taking into account the form of the vector ej (4)) and continuing the calculation
we find that
36
CHAPTER 1
Using formula (8.22) and the form of the vector el(<p) we now have
= R(<p) + csin lh(<p) sin 1P2(<p) ... sin 1Pn-l ( =

2
_c_( ( _ ) = R ( + cun(<p) - c = 1 - Ctln(<p)
det <l>
= R(
+ R(
tin
R(
R( >)
From the last relation it follows that
Making the vectors (8.23) orthogonal we obtain n - 1 linearly independent

solutions of equation (8.20) belonging to the space C'(Tm ).
The following lemma will be used to verify whether the conditions of
Lemma 4 hold.
Lemma 5. Ifu(
inequality
= (Ul(, ... ,un (<p)) E C'(Tm ),
lIu(1I
= 1,
I ~ 1 and the
(8.24)
n>m+l
holds, then the set of values of the function u( <p) is not everywhere dense on
the unit sphere 5 1 (0).
Indeed, the function u( defines a smooth map U of the m-dimensional
torus Tm : 0 S >i
S 271" (j
= 1, ... , m) into the n-l-dimensional sphere 5 1 (0).
Because of (8.24), U is a map of a manifold of smaller dimension to a manifold

oflarger dimension. Under such a map, all the image points UTm are "critical"
values of the map U and their measure equals zero (see Sard's theorem [129]).
Being closed, the set UTm cannot "cover" the entire sphere 5 1 (0).
The above statements enable one to prove the following result on the extendability of a periodic r-system of vectors to a periodic basis in En.
r-matrix U(<p) = (Ul(<P), ... , ur(<p) be a member of the

and have rank r for all > E Tm . If the inequality
Theorem. Let n
space
C I (Tm )
(8.25)
n> m+r
holds, then the matrix U ( ca n be extended to a periodic basis in En.

To prove the theorem, first suppose that I
that for any integer p satisfying the condition

n -
r+ 1 S p S n,
1. Since it follows from (8.25)
37
the inequality
p>
holds, the equation

Vl(rjI)Xl
m+ 1,
+ ... + vp(rjI)x p = 0
(where, according to Lemmas 4 and 5, the coefficients v(rjI) =
(VI (rjI),
... , V p ( rP))
belong to the space C/(Tm )) has exactly p - 1 linearly independent solutions

which belong to the space C/(Tm ). Then Lemma 3 asserts that the r-system
of vectors U(rP) = (Ul(rP), ... , ur(rP)) in the space C'(Tm ) can be extended to a
periodic basis of En.
Now let 1 = O. We represent U(rP) in the form (8.6):
U(rP) = O(rP)T(rP)
where O(rP) is an n x r orthonormal matrix belonging to the space C(Tm ).
Applying the approximation theorem to the matrix O(rP) , we construct an
orthogonal matrix 01(rP) belonging to the space C1(Tm ) and arbitrarily close
to O(rP) with respect to the metric of C(Tm ). Let
where
c is a positive number of suitable magnitude.
As was proved above, the matrix 0 1 (rP) can be extended to an orthonormal

periodic basis in En. Consequently, there exists a matrix V1 (rP) E C 1 (Tm ) such
that
The matrix VI (rP) extends O( rP) to a periodic basis in En since
for sufficiently small c > O. This is sufficient to enable us to extend U (rP) to a

periodic basis in En.
Note that the theorem given above cannot be improved, since if inequalities
(8.25) fail to hold, then this can result in the inability to extend a periodic rsystem of vectors from C/(Tm) to a periodic basis in En. The reasons for this
and examples are given in [34], [35]. Note also that the problem on a periodic
basis in En has a well known interpretation in topology, and the fact stated in
the theorem has, in the latter context, a more elegant proof than the one given
above.
38
CHAPTER 1
1.9. Logarithm of a matrix in C'(Tm ). Sibuja's theorem

Let P(<p) be an n x n-matrix whose elements are functions in C'(Tm ). A matrix
X = X (<p) that satisfies the equation

(9.1)
is called a logarithm of the matrix P( <p). Since the exponential of any matrix
is non-singular, the condition
det P(<p) ;C 0
(9.2)
is necessary for a logarithm of the matrix P(<p) to exist.

Suppose that condition (9.2) holds, let
be a Jordan form of the matrix P(<p), and let T(<p) be a matrix that reduces
P( <p) to this form:
Let C(<p) be an arbitrary non-singular matrix commuting with the matrix :J(<p).
A general formula giving all solutions of equation (9.1), that is, giving all
the logarithms of the matrix P(<p) has the form [39]:
(9.3)
where the matrix In :J(<p) defined by the expression:
is quasi-diagonal. The diagonal blocks of the matrix In :J(<p) can be found

using the formula:
In(Aj(<p)Ej
+ Zj) = In AjEj + -y:::Zj + ... =
/ '(,\.)
= /(,\.)E- + __
J Z.
J
J
I!
J
/"(A')
+ __
J_ Z.
2!
J
.. .
+ /(pr
1)(,\.)
(Pi _ I)!
Z~j-l
J
'
39
where
1
Z) -- [00.
f(>') = In >.,
...
o ...
~ .~o]
1
and Pi is the dimension of the matrix Zi'

In the above formulae q and Pi (j = 1, ... ,q) are functions of <p.
As one can see from (9.3), the logarithm of a non-singular matrix is multivalued. We shall be interested in the real branch of the logarithm of a nonsingular real matrix P(<p). As follows from [63], such a branch exists if and only
if the matrix P(<p) either does not have elementary divisors which correspond
to negative eigenvalues, or each such divisor is repeated an even number of
times. The latter is true if, in particular,
where Q(<p) is a real matrix.

Suppose that the matrix P(<p) has a real logarithm. We denote it by
X(<p) = In P(<p)
and find out when we can choose it so that it will belong to the space C/(Tm ).
The example of the matrix
Po(<p)
COS 
0] [ i
-i
0 e-it/>
1]
1
(9.4)
shows that this cannot always be achieved. Indeed, formula (9.3) defines the
real branch of the logarithm of matrix (9.4):
In Po(<p)
= [ -<p +0 2k1r
<p - 2k1r]
0
'
(9.5)
where k is an arbitrary number. For a fixed number k, matrix (9.5) is not

periodic with respect to <p; for k equal to the integer part of <p/(21r), matrix
(9.5) is periodic with respect to <p but it is not continuous. Thus for the matrix
P(<p) given by (9.4), its logarithm does not belong to C(TI).

To find out the conditions for a matrix In P( <p) to belong to the space
C/(Tm ), we write down Cauchy's integral formula for In P(<p) [42]:
In P(J)
1
= -2
. f (>.E if'
'Tn
P(J-lln >.d>.,
(9.6)
40
CHAPTER 1
where r is a closed contour in the complex A-plane bounding a simply connected

region G containing the eigenvalues Al (4)), ... ,An (4)) of P(4)) and not containing
the point A = 0, and In Ais a branch of the logarithm that is regular in G. Since
the boundary
of the region G in (9.6) can be chosen separately for each 4>,
we see that formula (9.6) is a local one.

The totality of values of the eigenvalues Al(4)), ... , An (4)) of the matrix
P(4)) when 4> covers Tm is called the spectrum of the matrix P(4)). We shall
say that the spectrum of the matrix P( 4 does not contain the origin if it
is contained in a simply connected region G of complex plane that does not
enclose the origin. For the matrix P(4)) the contour
in (9.6) can be chosen to
be the same for all 4> E Tm if its spectrum does not contain the origin. In this
case, formula (9.6) becomes global since it defines an integral over
which is
the same for all 4>; therefore the integral (9.6) preserves the properties of the
matrix P(4)) considered as a function in the space C/(Tm ). The following result
summarizes the above remarks.
Theorem 1. Let P(4)) be a non-singular matrix in C/(Tm ) with spectrum not
containing the origin. Then a branch of logarithm of the matrix P(4)) can be
chosen in such a way that it belongs to the space C'(Tm ).
Note that the above branch of the logarithm turns out to be real when P(4))
is real and its eigenvalues satisfy the conditions given above for the logarithm
to be real.
Having finished our discussion of matrices in C/(Tm ) we now give a result
on the block diagonalization of a matrix P(4)) E C 1(7jJ due to Y. Sibuja [36],
[169].
Theorem 2. Let P( 4 be an n x n square matrix of a scalar variable 4> I periodic
with respect to 4> with the period 21T and I times continuously differentiable.
Suppose that the characteristic polynomial p(4),A) of the matrix P(4)) can be
represented in the form
p(4), A) = Pl(4), A) . P2(4), A),

where Pl and P2 are polynomials in A of the form
= (_A)m, + Ql(4))A m ,-l + + Qm,(4)),

P2( 4>, >') = (_>.)m, + Tl (4))A m,-l +
+ r m,( 4
Pl(4), A)
41
with I times continuously differentiable periodic coefficients with period 271".

Suppose that the polynomials PI (4J, A) and P2( 4J, A) have no common roots for
any 4J E
11.
Then there exists a non-singular matrix T( 4J) that is periodic with
respect to 4J with period 271", I times continuously differentiable and such that
P(4J) = T(4J)diag{E(4J), F(4J)}T- 1 (4J),

where E( 4J) and F( 4J) are the corresponding (ml x mI) and (m2 x m2) square
matrices with characteristic polynomials Pl(4J, A) and P2(4J, A). If the matrix
P( 4J) and coefficients of the polynomials PI (4J, A) and P2( 4J, A) are real then the
matrix T( 4J) can be chosen to be real and periodic with period 471".
The example considered in [169] of the matrix
where Po(4J) is matrix (9.4), D
= diag{d1 ,d2},
d1 ,d2 are different, shows that
the doubling of the period of the matrix T( 4J) in choosing it to be real is essential
for Theorem 2 to hold. We also note that the analogue of Sibuja's theorem is
not known for m
> 1.
1.10. Garding's inequality
Let
be a differential operator, the coefficients of which are square matrices A vp , a v , b

of order n belonging to C 1(Tm ), C(Tm ), C(Tm ) respectively. We require that
the operator 2 be elliptic, that is, the inequality
m
vp ,1Jv, 1Jp} ~
L(A
L II1JvIl
v,p
v=1
II
must hold for any collection "11, ... , "1m of n-dimensional vectors 1Jv
for some
II
(10.1)
= (1J~, ... , 1J:)
> O.
The ellipticity 2 enables us to prove an important inequality for the inner

product (2U, u)p. In the literature it is known as Carding's inequality (14].
(10.2)
42
CHAPTER 1
where I and fJ are positive constants which depend on the coefficients of L 2 but
do not depend on u.
Proof Let u E HOO(Tm). Then L 2 u E cP(Tm ) C HP(Tm ) and the expression
on the left hand side of (10.2) makes sense. We now derive inequality (10.2).
Let p = O. Then
We make an estimate for the functional
</10' Using the estimation technique
given in 1.3 we find that
I~ol ::; ~II aa~: :</1~ I oli ulio + ~llav :;v Iioliullo+ IIbulioliulio ::;
::; 2c(l: Iaa~1' 10 + l: la

V.I'
I'
10 + Iblo) lIulh lI ulio ::; c' (fJ~lIulii + fJ\ lIulI~)
(10.3)
where c' = c (Lv.1' &&~e 10 + Lv la v10 + Ibl o), fJ 1 is an arbitrarily small number,
and c is a positive constant which does not depend either on u or on the
coefficients of L 2 .
We make an estimate of the product Lv.1' (Avl' :;v'

ity (10.1) we obtain the estimate:
~ 11 l:
(21l")m
v=l
m
1 1 II au II
2
:11"
.. .
:11"
(fi:"""
'!'v
:~u",) o'
Using inequal-
d</11'" d</1m =
(10.4)
where
Cl
and
C2
are positive constants independent of u and A v !,"
Using both inequalities (10.3) and (10.4) we arrive at inequality (10.2) for
p= 0:
(L 2 u, u)o
~ (11 cl - c' fJf)lIulir -
(11 c2
;:)
1
lIull~ = Iliulir - fJllull~
43
Let p = 1. Then
(L 2 u, uh = (L 2 u, u)o +(L 2 u, -6.u)o 2:
~(
{;;t
ou OU)
L OJi 'OJi
~(O~i L u, : ; ) 0 +1'lIull~ - 611ull~ =
~(",OAv" 02 u
+ {;;t
OJi oJvoJ" +
f::
'" oa v ou
oJv
+ L.: OJi
2: l'
~II
:;J:- ~II :;i II:

6
ob
ou )
1 112
2
+ OJi
U, OJi 0 + l' lu 1 - 611ull o 2:
+ ~ (~iU, :;) 0 + 1'IIull~ -
611ull~ 2:
1 2
112 ~ (
ou )
2: 1'ctllulb2- 1'C211ullo2- 611ul1 2
1 +1'1 ulll - 611u 0 + {;;t ~iU, OJi 0 2:
2:
where
ClJ C2
1'Clllull~ - 6c2I1ull~ + rllull~ - 611ull~ + ~(~iU, :;i )0'

are positive constants independent of
We now estimate the product
1(~iU, :;)01 ~ [2:ll

v,,,
(~iU, OU/OJi)O.
u and
L 2
We have:
:;i" 0~2;J" 110+

+
2:11 ~~: :;v 110 +

v
":;i ullol1I :;i 110 ~
~ C[2: IAv,,1t + 2: la v It + Iblt] lI ull211 ulit = ~ lI ull2l1 ullt,

VIP
where
(10.5)
is a positive constant independent of u and L 2
Taking into account estimate (10.5) we obtain the following inequality:
which, in view of the inequalities
211ull211ullt ~ 6?lIull~ + 61211ullr ~ 6?lIull~ + ;2l1ull2l1ullo ~

1
44
CHAPTER 1
takes the form of inequality (10.2):
where
1(1)
and
0(1)
are positive constants which depend only on
L:v lavh + Ibh, and do not depend on u.

Now suppose that inequality (10.2) has been proved for p
L:v,,,, IAv",h +
:$ r. Consider
the product (L 2 u, U)r+1. We have
2 I C111 ullr+2
OC2
Ilull12+ Illullr+1
2 - olullo
I 2+ {;;:
~ (1>.u, 01>.
OU ) r ~
~ IC1I1ull~+2 - (6~2) .2I1ullr+2I1ullr+1 - 611ull~ + ~ ( ~.U, %;i) r

Making an estimate of (~iU, OU/01>i)r, we have:
I(~u, %;)J:$ [Lf:i 0~2;4>Jr +LII~~ :;JL+

v ,I-'
+11
'"
:;i ulU 11 :;i Ilr :$
2c
[~IAv"'lr+1 + ~ la vlr+1 + Iblr+l] x
But since by (3.2)
2l1ullr+2l1ul Ir+1 :$ 6121Iullr2+2+ 521 lI u12Ir+l

1
:$
2 + 62I1ullr+2I1ullr+l
1
6121 lullr+2
:$
1

2
45
= 28 1 11 u llr+2 + 2+4(r+l) lI u Ir+211 Ullo ::;

81
::; 28rllull;+2
202+~cr+l) (O~(r+2)lIulI;+2 +
1
04C!+2) lI u lin
1
~8rllulI;+2 + 282[3+~cr+l)]lIulI~,
1
it follows from (10.6) that
aU) r I::;;;;c' [5281211 U112r+2 + 28i[3+4(r+l)]

II Il5] .
I(~iU, a</>i
U
(10.7)
Using inequality (10.7) we make an estimate of the product (L 2u, U)r+l as

follows:
(L2 U,U)r+l
[r c l -
8C2
- [ ( ""2
+
,)
~(8~2 +c')8r]lI u ll;+21

]
2
II 1 2
2
28 2[3+4(r+l)1 + 8 lIulio ~ r(r+l) u Ir+2 - 8(r+l)ll u llo,
1
where r(r+l) and 8Cr + l ) are positive constants which are dependent only on
L:v,1l IAvlllr+l + L:v la v Ir+l + Iblr+l but do not depend on u.
By assuming that inequality (10.2) holds for p ::; r we have proved it
for p = r
+ 1.
Hence by induction, inequality (10.2) holds for any p and
u E W>O(Tm ).
Now let u E HP+l(Tm). Then L 2u E HP-l(Tm ) and the product (L 2u, u)p
makes sense. For u E HP+l(Tm ), inequality (10.2) follows from the following
chain of inequalities
~ rllu1:II;+l - 811u1:W -IIL 2(u - u1:)IIp_ll1ullp+l-
-IIL 2u1:lIp-dlu - u1:lIp+l,
(10.8)
where U1: is a sequence of functions in Hoo(Tm ) convergent to u in HP+l(Tm ).

Indeed, passing to the limit as k
-> 00
in (10.8) we have
(L 2u, u)p ~ rllullp\1 - ollull~ - lim IIL 2(u - U1:)lIp-lliullp+l1:-+00
- lim IIL2ul:llp-1 x lim Ilu - U1:lIp+l

k-oo
k-+oo
where the constants rand 8 depend only on

do not depend on u.
= rllull:+l - ollull5,
L:v,1l IAvlli p + L:v lavlp +
Iblp, but
Chapter 2. Invariant sets and their stability

2.1. Preliminary notions and results
We shall be looking at the following system of differential equations:
dx
(1.1)
dt = X(x),
the right hand side of which is defined, continuous and satisfies a Lipschitz
condition in a domain D of n-dimensional Euclidean space. Under these assumptions, for every Xo E D there exists a unique solution x(t, xo) of system
(1.1) such that x(O,xo)
= Xo.
This solution is defined and belongs to the do-
main D for a maximal interval (T-, T+) in R, where -00 ~ T- < < T+ ~
+00, T- = T-(xo), T+ = T+(xo). Since the system of equations (1.1) is
autonomous, the function x(t, xo) satisfies the group condition with respect to
t:
x(t + T,XO)
for any Xo E D and any t, T and t
= X(t,X(T,XO))
+ Tin (T-, T+).
Following conventional terminology, the solution x(t, xo) for a fixed Xo

will be called the motion. The set {x(t,xo), t E (T-,T+)} will be called
the trajectory of the motion x(t, xo), and when T-00, T+
+00 it will
be denoted by x(t,xo;R). The set {x(t,xo),t E [O,T+)} for T+

+00 will
be called the positive semi-trajectory and the set {x(t,xo), t E (T-,Oj) for
T-
= -00 will be called the negative semi-trajectory of the motion x(t, xo) and
will be denoted by x(t,xo;R+) and x(t,xojR-) respectively. Finally, the set of

points {x(t, xo), T 1 ~ t ~ T2 } for T- < T 1 < T 2 < T+ will be called a finite
arc of the trajectory of the motion x(t, xo) with the time length T 2 - T 1 which
we denote by x(t,xo;[T1 ,T2 ]).
A set M C D is called an invariant set of system (1.1) if it consists of
points of the trajectories of this system.
46
INVARIANT SETS AND THEIR STABILITY
47
A set D l C D is said to be compact if any infinite sequence of points of

it contains a convergent subsequence and the limit point of this subsequence
belongs to D.
A set 11 C En is called connected if it cannot be represented as a union of
two non-empty sets in such a way that neither set contains limit points of the
other.
Let
o ~ tl < t2 < ... < t n < ... ,
lim t n
n-oo
= +00
(1.2)
be a sequence of values of t such that

lim x(tn,xo)
n-oo
= x O E D.
(1.3)
Then x O is called an w-limit point of the semi-trajectory x(t, Xo; R+). Similarly,
one defines an a-limit point of a semi-trajectory x(t,xo;R-).
The set of all w-limit points of a semi-trajectory x(t, xo; R+) is denoted by
11"0' the similar set of a-limit points of a semi-trajectory x(t, xo; R-) is denoted
by A.,o.
For subsets A and B of En we define the distance p(A, B) between them
by setting
p(A,B)
= .,eA
inf p(x,y) = inf IIx - yll.
.,eA
!leB
!leB
The common properties of the limit sets 11"0 and A.,o of compact semitrajectories are known from the qualitative theory of differential equations [17],
[94] and are given for 11"0 in the following theorem.

Theorem. If x(t,xo;R+) is a compact semi-trajectory of system (1.1), then
11"0 is a closed invariant compact connected set satisfying the condition
lim p(x(t,xo), 11"0) = O.
1-+00
(1.4)
To prove the theorem we shall follow the reasoning of [94]. Let x O E 11"0'
We choose a sequence (1.2) such that relation (1.3) holds and consider the
+ tn, xo) (n = 1,2, ...), each of which is defined for

t E [-tn, +00).
It follows from the continuity of the solution x(t, xo) with respect to Xo
that
sequence of functions x(t
48
CHAPTER 2
for each t E R. Because the semi-trajectories x(t, Xo; R+) are compact, the
values ofthe functions x(t+t n , xo), n = 1,2, ... , belong to some closed bounded
subset D 1 of the domain D. But then x(t, Xo) E D 1 for any t E R. Hence, in
view of (1.5), we have x(t,xO) C 0"0 for all t E R.
By passing to the limit in the equality
x(t + tn, xo) = x(t n , xo) +
it
X(x(t
+ tn, xo))dt
(1.6)
we prove that the function x(t, xO) defines a solution of system (1.1). This is
sufficient for the set 0"0 to be invariant.
To show that the set 0"0 is closed we take a sequence of points x~ (n =
1,2, ...), x~ E 0,,0'
lim x~ = xO, and show that xO E 0,,0' Because xO is
n .... oo
a limit point of the sequence x~, for any
> 0 there exists
n ;::: 1 such that
< f/2. Since x~ is an w-limit point of the semi-trajectory x(t, xo;R+),

it follows that there exists t n > 0 such that p(x(tn,xo),x~) < f/2. But then
p(x(t n , xo), xO) < f, and therefore xO is an w-limit point of the semi-trajectory
p(x~, xO)
x(t,xo;R+). This proves that the set 0"0 is closed. It is obvious that 0"0 is
compact.
We show that the set 0"0 is connected. Assume the contrary. Then there
exist two non-empty disjoint closed sets A and B such that 0"0
= AU Band
p(A, B) = d > O. Since A C 0"0 and B C 0"0' one can find t~ and t~ arbitrarily
large such that x(t~, xo) lies in a (d/3)-neighbourhood of the set A and x(t~, xo)
lies in a (d/3)-neighbourhood of the set B. Choosing subsequences
{t~}
{t~}
and
such that the inequalities
o< t~ < t~ < t~ < t~ < ... < t~ < t~ < t~+l < ...
hold, we find that
p(x(t~, xo),
A) < d/3,
p(x(t~, xo),
A) ;::: p(A, B) -
p(x(t~,
xo), B) > 2d/3.
(1.7)
Since p(x(t, xo), A) is a continuous function with respect to t, we can find

t~
< Tn < t~ such that

p(x(Tn,xo),A)
= d/2.
(1.8)
Because the set X(t,xoiR+) is compact in D, we can choose a subsequence

from the sequence X(Tn,XO) (n
= 1,2, ...) that converges in D.
According to
49
the definition, the limit point of this subsequence XO belongs to the set 0,,0.
On the other hand, by passing to the limit in relation (1.8) we obtain
p(xO,A)
= d/2,
p(xO,B);::: p(A, B) - p(A,xo)
= d/2,
that is, 0"0 f:. Au B. This contradiction shows that 0"0 is connected.
Finally, we prove limit equality (1.4). Suppose that it does not hold. Then
one can find a sequence of positive numbers {t n }, lim t n =
n-oo
a > 0 such that
+00 and a number

(1.9)
Because the semi-trajectory x( t, Xo; R+) is compact m D, the subsequence
{x(tn,xo)} has a limit point xo in D. On passing to the limit in (1.9), we

obtain for this point the inequality p(XO, 0"'0) ;::: a. The latter contradicts the
definition of the set 0"'0' according to which the point xO must lie in 0"'0.
This contradiction proves relation (1.4) and this finishes the proof of the
theorem.
2.2. One-sided invariant sets and their properties
A set M C D consisting of positive semi-trajectories of system (1.1) is called

a positively invariant set of this system. Similarly, a set consisting of negative semi-trajectories of system (1.1) is called a negatively invariant set of this
system. Both of these sets are called one-sided invariant sets of system (1.1).
A closed compact positively invariant set M of system (1.1) is called stable
if for any f> 0 there exists 6 = 6(f) > 0 such that x(t, U6) C (U, U M) for all
t ;:::
o.
If M is stable and satisfies the limit relation:

lim p(x(t, xo), M) = 0
(-+00
for all Xo E U6 0 and some 60 > 0, then the set M is said to be asymptotically
stable for system (1.1). If M is not stable, then it is called unstable. The term
"unstable" will also sometimes be used for sets that are not positively invariant.
Here and in what follows, U,
= U,(M) will denote the f-neighbourhood of
the set M consisting of all the points x E En for which 0 < p(x, M)
<L
Let D 1 be a bounded region that is contained in D together with a neighbourhood of it, let 8D 1 be the boundary of D 1 , and D 1
= D 1 u8D 1 the closure
50
CHAPTER 2
of D 1 A function Vex) that is defined and has continuous partial derivatives

in some region containing D} is said to be continuously differentiable in D 1 .
Let V
= Vex)
be a function that is continuously differentiable in D 1 We
say that V is of constant sign on D 1 if for all x E D} the non-zero values of
Vex) have the same sign.

Consequently, if a function V is of constant sign on D} then either V (x)
or Vex) ::; 0 for all x E D}. A function V of constant sign on D} is said to

be sign-definite on D 1 if the set of zeros of the function V is not empty and is
compact in D 1 .
Consequently, if a function V is sign-definite, then the set No of all xED}
No =I 0, No n aD} = 0, and the

values of V(x) have the same sign on the complement of No.
A sign-definite function on D} taking the value zero at a single point x = 0
such that Vex)
= 0 satisfies the conditions:
is a sign-definite Lyapunov function [68].

We denote by VI' the set of points D} for which
0< lV(x)1 ::; J-l,

and by No the set of zeros of the function V in D}.
Lemma. Let V
there exist J-l
= Vex)
be a sign-definite function on D}. Then for any
= J-l(f) > 0 and fJ = fJ(f)
>0
>0
such that
(2.1)
Indeed, for a sign-definite function V on D} the function
IVI
is positive-
definite on D 1 and the set VI' is not empty for any J-l > O.
Suppose that the left inclusion in relation (2.1) does not hold for some
sufficiently small
> O. We can suppose that
is so small that UNo) is
contained in D 1 . This can always be achieved because No is compact in D 1 .

Under these assumptions, one can find a point X n in V}/n that does not belong
to UNo) U No for any n
1,2,.... For the sequence {x n } the following
inequalities hold:
0< V(x n )
< l/n,
p(x n , No)
~ f,
n = 1,2, ....
For a limit point x O of the sequence {x n }we have then the following relations
51
But they contradict each other because of the definition of the set No. This
contradiction shows that for any
left inclusion in (2.1) holds.
Suppose that for some JJ
> 0 there exists JJ
= JJ( c) > 0 such that the
= JJ( c) > 0 there does not exist 8 = 8(c) > 0 such
that the right inclusion in (2.1) holds. Then there exists a point X n in U1 !n(No)
= 1,2, .... For the sequence {x n }
that does not belong to VI' U No for any n

the following relations hold:
Then for a limit point x O of the sequence {x n } we have

o
p(x ,No)
= 0,
x E D 1,
which, as follows from the definition of No, contradict each other. The contra-
diction shows that for each JJ = JJ(c) > 0 we can find a 8 = 8(JJ) = 8(c) > 0 such
that the right inclusion in (2.1) holds. This finishes the proof of the lemma.
Remark. The value of JJ = JJ(c) played no role in the proof of the right-hand
inclusion in (2.1). Hence it follows from this proof that for any JJ > 0 there
exists 8 = 8(JJ) > 0 such that
The function
V(x)
= aV(x) X(x) = ~ aV Xv (x)

ax
~ ax v
v=1
is called the total derivative of the function V(x) along the solutions of system
(1.1) .
Theorem 1. If V
tive V
= V(x)
= V(x)
is a sign-definite function on D 1 whose total deriva-
along the solutions of system (1.1) is sign-definite, then the set

V(x)
= 0,
x E D1
(2.2)
is a one-sided invariant set of system (1.1). More precisely, this set is negatively
invariant if the signs of V and
same and is positively invariant (and
V are different)!
We regard the case when V(x) = 0 for all x E D 1 as the case of different signs
stable) if the signs of V and
V are the
of V and
V.
52
CHAPTER 2
To prove Theorem 1, we assume that V(x) 2: 0 for x E D I . Let V(x) ~ 0

for x E D I . We claim that the set No consisting of points of DI such that
V(x) = 0 is a positively invariant stable set of system (1.1). To see this, we

consider a solution x(t,xo) of system (1.1) passing through Xo E No when
t O. Since No is compact in DI, it follows that x(t, xo) E D I for t in some
interval (hI, h 2 ), where hI < 0 and h 2 > O. By hypothesis, dV(x(t, xo))/dt =
V(x(t, xo)) ~ 0 for t E (hI, h 2 ). But then the function V(x(t, xo)) is nonincreasing for t E [0, h 2 ), and this leads to the inequality
V(x(t, xo))
V(xo)
= 0,
t E [0, h 2 ),
which proves that x(t, xo) E No for all t E [0, h 2 ).

Since there is a neighbourhood of No lying inside D I , so that there is a
"gap" between No and the boundary oD 1 of the region D I , it follows that the
point x(t, xo) in changing its position continuously in the region D I cannot leave
the set No for t > O. This means that No consists of positive semi-trajectories
ofsystem (1.1) and therefore No is a positively invariant set of system (1.1).
We now prove that the set No is stable. Let
positive number such that U,
p.
= U,(No)
be a sufficiently small
C D I . By the lemma we can choose
= P.(f) > 0 and 6 = 6(f) > 0 such that the inclusions

(2.3)
hold, where U6 = U6(No).

Taking into account (2.3) we see that the solution x(t,xo) of system (1.1),
which starts at a point Xo E U6 for t
= 0, belongs to the set VI' U No for all t
in some interval (h l ,h 2 ), where hI < 0, h 2 > O. Because the function V(x) is

of constant sign in D I , we have the inequality:
V(x(t,xo))
V(xo)
~ p.
(2.4)
for all t E [0,h 2 ). Hence x(t,xo) E VI' U No for all t E [0, h 2 ].

Let h 2 be the largest value in R+ U {+oo} for which inequality (2.4) holds.
If h 2 is finite, then we immediately reach a contradiction. Thus x(t, xo) E VI'
for all t 2: O. Since U, :> VI' (see (2.3)), we have x(t,xo) E U, for all t 2: O.
This proves that x(t, U6) C U, for t E R+. And this means that the set No is
stable.
53
Now let V V(x) ~ 0 for x E D 1 The function -V -V(x) is the total

derivative of the function V(x) along solutions of the system
-dx
= -X(x)
dt
(2.5)
and satisfies the inequality -V(x)::; 0 on D 1 .

Applying the above result to the system (2.5), we see that No is a positively
invariant stable set of system (2.5). Because system (2.5) was obtained from
(1.1) by changing t to -t, No is a negatively invariant set of system (1.1). This
completes the proof.
It should be noted that it is not sufficient that the signs of V and
V in the
conditions of Theorem 1 be the same in order that the set (2.2) be unstable.
However, if the signs of V and
V are the same then set (2.2) can be stable only
if it is invariant and contained in a "larger" invariant set of system (1.1); the

closure of the set x(t, U~o U No) for some 60 > 0 and all t E R is an example.
Theorem 2. If the function V = V(x) and its total derivative V = V(x) along
solutions of system (1.1) are sign-definite on D1 and their zero sets in D1 are
the same, then the one-sided invariant set (2.2) is asymptotically stable if the
signs of V and V are different and is unstable otherwise.
To prove the theorem we suppose that V(x) ~ 0 for x E D 1 . Let V =
V(x) ::; 0 for x E D 1 . It follows from Theorem 1 that the zero set No in
D 1 of the function V(x) is a positively invariant stable set of system (1.1).
Consequently, for a sufficiently small i > 0 there exists 6 = 6( i) > 0 such that
x(t, U~) C (Uf UNo) C D 1 for t ~ 0, where U~ = U~(No), Uf = Uf(No). We
now show that
lim p(x(t, xo), No)
t_+oo
for any point Xo E

Indeed, let 6
U~o
=0
(2.6)
= U~o(No), where 60 is a sufficiently small number.
= 6(i)
be chosen such that x(t, U~) C (U f UNo) C D 1 for
t ~ O. For a point Xo E U~ we consider the semi-trajectory x(t, Xo; R+). If

x(r,xo) E No for some r E R+, then the semi-trajectory x(t,x(r,xo);R+) lies
inside No for t ~ 0 because the set No is positively invariant, and so relation
(2.6) holds for x(t, xo).
Let x(t,xo) (/. No for all t ~ O. Then dV(x(t,xo))/dt = V(x(t,xo))
for t
<0
0 and, consequently, the function V(x(t, xo)), which is monotonically
54
CHAPTER 2
decreasing, tends to a limit 11-0

suppose that 11-0>
o.
0 as t
Then V(x(t, xo))
+00. We claim that
--+
for t
~ 11-0
11-0
= O.
For
O. Using the remark made
in the proof of the lemma, we take 60 = 60(11-0) > 0 such that the inclusion
V/Jo :) U6 0 holds. From the definition of the set V/Jo and the above inclusion it
follows that x(t, xo) E U. and x(t, xo) f/:. U6 0 for t ~ O. But then 60 < (. and
x(t,xo) E U.\U6 0 for t ~ O. We set
sup V(x)
rEU,\U6 o
= -a.
Since V(x) takes only
negative values on the closure of the set U. \ U6 0 ' it follows that a > O. But
then dV(x(t,xo))/dt::; -a for t
O. This leads to the inequality
V(x(t,xo))::;V(xo)-at,
tER+.
(2.7)
Inequality (2.7) contradicts the fact that the function V(x) is positive definite
on D 1 This contradiction shows that 11-0 = O. Thus, if x(t, xo) f/:. No for t
then
lim V(x(t, xo))
1-+00
Consider the limit set

lim p(x(t, xo), Oro)
t-+oo
Suppose that
Oro
Oro
= o.
0,
(2.8)
of the semi-trajectory x(t, Xo; R+). Now since
= 0, to obtain (2.6) it is sufficient to prove that Oro C No.
i. No
Let y E
Oro \No.
Since y is an w-limit point of the
semi-trajectory x(t,xo;R+), we can find a sequence 0 < tl < t2 < ... < t n <
... , lim t n +00 such that lim x(tn,xo) y. But then lim V(x(tn,xo)) =
"-00
n-oo
n-oo
V(y) # 0, which contradicts (2.8). This contradiction shows that Oro C No for
any Xo E U6. This leads to relation (2.6) for all Xo E U6 and therefore proves
that the set No is asymptotically stable.
Now let V(x) ~ 0 for x E D 1 . Then -V(x) ::; 0 for x E D 1 and the system
of equations (2.5) satisfies the conditions of Theorem 2 which ensure that the
set No is asymptotically stable as an invariant set of system (2.5). This leads
to the relation
lim p(x(t,xo),No)=O
t--oo
for all Xo E U6 0 ' where 60 is a positive number. But then the limit set A ro
of any semi-trajectory x(t,xo;R-), Xo E U6 0 ' belongs to the set No.
11-1
> 0 be so small that
V/JI
Let
C U6 0 Consider the semi-trajectory x(t,xo;R-)
for Xo E U60 \ V/JI Since A ro C No, we can find a sequence of numbers
o < t 1 < t 2 < ... < t n < ... ,
lim t n =
n-oo
+00
55
such that
lim p(x(-tn,xo), No)
n-oo
= O.
This means that there is at least one point of the sequence

(2.9)
in any neighbourhood of No. Each solution x(t, x( -tn, xo (n = 1,2, ...) takes
the value x(t n , x( -tn, xo) = Xo rt VI'I for t = t n > o. If now we fix fO > 0 such
that the inclusion UfO C VI'I holds, then in any neighbourhood of No one can
find a point of (2.9) that leaves the fo-neighbourhood of No when t > O. Thus
the set No cannot be a stable set of system (1.1) and this completes the proof
of Theorem 2.
It should be noted that the parts of the given theorem that relate to the
stability of the set No are similar to the results on stability theory of invariant
sets of system (1.1) [46], [47) and, in particular, are generalizations of the wellknown Lyapunov theorems on stability of equilibrium points of system (1.1)
when No consists of a single point [68], [72).
We end our discussion of one-sided invariant sets and their stability by
giving one further result [47) which is a quantitative characterization of trajectories of system (1.1). It has an obvious extension to positively invariant sets
of system (1.1).
Theorem 3. In order that a closed invariant set of system (1.1) having a
a sufficiently small compact neighbourhood not containing entire trajectories
be stable, it is necessary and sufficient that there does not exist a motion
x(t,xo), Xo
rt
M, having a-limit points in M.
The proof of the theorem is based on the fact that if the conditions of the
theorem are satisfied, then the quantity
inf
t<O
p(Xo)vl)=f
IS
positive for sufficiently small f >
o.
p(x(t, xo), M)
And by virtue of the definition of
stability, this is sufficient for the set M to be stable.
56
CHAPTER 2
2.3. Locally invariant sets. Reduction principle
A set M C D is called a locally invariant set of system (1.1) if it consists of

arcs of trajectories of this system. In other words, M is a locally invariant
set of system (1.1) if for any point Xo E M there exists T 1
T 2(xo), T 1T2 50 such that x(t,xo;[Tl,T2])
= TI (xo)
< T2 =
eM.
Any arc of a trajectory and any region D 1 contained in D are obvious

examples of invariant sets of system (1.1).
General criteria for finding locally invariant sets of system (1.1) are given
in the following theorem.
= V(x)
Theorem 1. If a function V
and its total derivative V
= V(x)
along
solutions of system (1.1) are of constant sign on D 1 then the set

V(x)=O,
(3.1)
xED I
is a locally invariant set of system (1.1) if it is non-empty.
To prove the theorem, suppose, for example, that V(x) ~ 0 and V(x) 5 0
for x E D 1 . We denote set (3.1) by N.
Let Xo E N. Then Xo E D 1 and consequently, x(t, xo) E D 1 for t E (hI, h 2 ),
where hI < 0 < h 2 . Then the function V(x(t, xo satisfies the inequality V(x(t, xo
V(xo)
=0
for all t E [0, h 2 ).
x(t,xo;[0,T2]) belongs to the set N for T 2
<
This shows that the arc
h 2. This is sufficient for N
to be a locally invariant set of system (1.1).

Necessary conditions for the set (3.1) to be a locally invariant set of system
(1.1) are given in the following theorem.
= V(x)
Theorem 2. If a function V
is continuously differentiable and its
zero set (3.1) in D 1 is not empty and is locally invariant with respect to system
(1.1), then its total derivative V
= V(x)
along solutions of system (1.1) takes
zero values on the set (3.1).
Corollary. If the functions V
= V(x)
and V
= V(x) = (8V(x)j8x)X(x)
are
of constant sign on D 1 then in D 1 the zero set of the function V is contained

in the zero set of the function V.
The statement of the Corollary follows from Theorem 1 and Theorem 2.
57
Under certain additional conditions, a locally invariant set of system (1.1)

contains an invariant set, or is itself an invariant set of this system. An example
of these conditions is given in the theorem given below.
Theorem 3. Let V = V(x) be a sign-definite function on D 1 the total deriva-
tive V = V(x) of which along solutions of system (1.1) is of constant sign. If

V(x) and X(x) are analytic on D 1 , then the set (3.1) is an invariant set of
system (1.1).
Thus an additional condition that the functions V(x) and X(x) be analytic
is one that guarantees that the locally invariant set (3.1) is in fact an invariant
set of system (1.1).
Turning to the proof of the theorem, we denote set (3.1) by N and consider
a solution x(t, xo) of system (1.1) for Xo E N. According to Theorem 1 of 2.2,
this solution is defined on the semi-axis and takes values in N for t E R+. We
suppose for definiteness that x(t, Xo; R+) C N. Then
V(x(t, xo)) = 0
for all t
(3.2)
O. Because the functions V(x) and X(x) are analytic on D 1 , we
have the decomposition

(3.3)
which converges for t in some interval
ItI < f-
It follows from equalities (3.2) and (3.3) that
VA:
= 0,
= 0,1,2, ....
But then
V(x(t, xo)) == 0 for all
It I <
f-
This leads to the inclusion x(t,xo;[-t,t]) eN. Since x(-t,xo) belongs to N,

some arc that starts at this point, namely, the arc x(t, x( -t, xo); [-tl' 2t)) for
some t also belongs to N. Continuing in this manner we see that the entire
semi-trajectory x(t, xo;R-) belongs to the set N. Since Xo is an arbitrary point
of N, this means that x(t, N) = N for all t E (-00,00). The theorem is proved.
Let N be a locally invariant set of system (1.1), M a positively invariant
set of this system such that N
M. We shall say that M is stable in N if for
58
CHAPTER 2
any neighbourhood U< of M there exists a neighbourhood U6 of it such that

x(t, Ui
n N) c (U 0 such that
in N if M is stable in N and there is a 60
lim p(x(t, xo), M)
1_+00
for all
Xo
U60
=0
(3.4)
n N.
It is clear that stability of a set M in N does not always imply that this
set is stable; however, if M is unstable in N it follows that M is unstable. Thus

there arises the problem of giving conditions under which stability of M will
follow from its stability in a locally invariant set N. This problem has been
solved by V.A. Pliss [98] when M is a point. His result is known in the literature
on stability theory as the reduction principle. The result for the general case
given in Theorem 4 is similar to this principle and is due to the author [108].
Theorem 4. Suppose that a locally invariant set N of system (1.1) contains a
closed positively invariant set M that is asymptotically stable in N. If N is a
set of type (3.1) for some function V
= V(x)
derivative of which V
= V(x)
of constant sign on D 1 , the total
along solutions of system (1.1) is of constant
sign and has the opposite sign of V, then M is a stable set of system (1.1).
If, in addition, the zero sets in D 1 of V and V are the same, then M is an
asymptotically stable set of system (1.1).
Before going to the proof of the theorem we establish some auxiliary results.
Lemma 1.
Under the conditions of Theorem 4, the limit relation (3.4) is
uniform with respect to
Xo
E U6 , n N, where 61 is a positive number.
Suppose that the lemma does not hold. Then for any 61 > 0 one can find
a number
j1.
> 0 such that there is a sequence of numbers t n and a convergent
sequence of points
Xn
in U6 , n N satisfying
p(xn,M)
< 61 ,
p(x(tn,xn),M)
x(t,xn;[O,tnD C D 1 ,
2:
j1.,
tn
-+
+00.
(3.5)
Since M is closed and is contained in the bounded region D 1 , some neighbourhood of M is contained in D 1 The number 61 can be chosen so small that
the limit of the sequence
Xn
belongs to the 6o-neighbourhood of the set M,
59
n N)
n N. Let
where 60 is chosen in such a way that both the points x(t, U60
N for t
0 and limit relation (3.4) holds for points of U6 0
Since xO E D 1 and V(x n)

XO
E (U6 0 U M)
n N.
= 0,
belong to
it follows that xO E N and consequently,
But then
lim p(x(t,xO),M) =0,
1-+00
and we can find T > 0 such that
p(x(t,xO),M) < 6(p)J2
T, where 6(p) is a positive constant such that x(t, U6(p) n N) C

U M) for all t ~ O.
for all t
(U p / 2
(3.6)
Since the solutions of system (1.1) depend continuously on the initial data,
we can find N > 0 such that the inequality
(3.7)
holds for all t E [0, T] and n > N. We can suppose that N is so large that
t n > T when n > N.
(3.8)
It then follows from inequalities (3.6) and (3.7) that
p(x(T, x n ), M) :::; p(x(T, xO), M)+

+p(x(T,xn),x(T,xo)) < 6(p)J2 + 6(p)J2 = 6(p).
The last inequality means that x(T, x n ) E (U6(p) U M) for n > N. Since
the function
V has
the opposite sign of V for all x E D 1 and, according to
(3.5) and (3.8), x(t,xn,[O,T)) C D 1 , it follows that V(x(T,xn)) = 0 and

x(T, x n ) E (U6(p) U M) n N for n > N. The last inclusion shows that the value
of the function x(t,x(T,x n )) belongs to the set Up / 2 U M for all t ~ 0 and
n> N, which yields the inequality

p(x(t,x(T,xn)),M) < pJ2
(3.9)
60
CHAPTER 2
for t
? 0 and n > N. We set t
= tn -
T in (3.9). By virtue of inequality (3.8)
this is always possible. As a result we obtain

p(x(t n , x n ), M)
< p./2 when
> N.
(3.10)
Inequality (3.10) contradicts one of the inequalities (3.5). The contradiction

proves the lemma.
Remark. By setting V
=0 in the conditions of Theorem 4, it will follow from
Lemma 1 that the limit (3.4) is uniform for Xo E U6, (M) and for any closed
compact asymptotically stable positively invariant set M of system (1.1). This
property for an invariant set is well known [47]. It is the basis of the definition
of uniform asymptotical stability of an invariant set.
We denote by It:5,11 the set of points x E Db that belong to the h-neighbourhood of the set M and satisfying the inequality
IVI :::; p..

Lemma 2. If the conditions of Theorem 4 are satisfied then
hI
= h 1(p.)-neighbourhood U6,(/l) of M
contains some
V6,/l
which contracts to the set
V6,O
= U6 n N
as J1. ...... 0, so that
(3.11)
First we suppose that

any
hI
> O.
V6,1'
does not contain a hI-neighbourhood of M for
Then one can find a convergent sequence of points
Xn
such that
Since the functional p is continuous and the set M is closed, the limit of
the sequence
Xn ,
belongs to M. But then
which contradicts the relation
61
Now suppose that the limit relation (3.11) does not hold. Then one can
find convergent sequences JJn and Yn such that
The limit of the sequence Yn,

Yo = n_oo
lim Yn
satisfies the inequalities
which are contradictory since the first one means that Yo E N and the second
that Yo fI- N.
We now turn to the proof of Theorem 4. Using the statements of Lemma 1
and asymptotical stability of Min N we choose positive numbers ( and 6 = 6( ()
such that the relations
p(x(t, xo), M) < (/2,

hold for all t
(U< U M) C D 1
0 and all Xo E V6 ,o and such that the limit

lim p(x(t, xo), M)
1-+00
is uniform with respect to Xo E
=0
(3.12)
V6,O'
Let T be a positive number which is determined so that the condition
p(x(T,xo),M) < 6/2

holds for all Xo E
V6,O'
We can choose such T because the limit relation (3.12)
is uniform. Now choose JJo
= J-lo(T, ()
> 0 in such a way that the trajectory
arc x(t, xo; [0, T]) belongs to the set U< U M for any Xo E
V6,l'o'
Such a choice
of J-lO is possible because of the following considerations.

Suppose that J-lO cannot be chosen in the indicated manner. Then one can
find convergent sequences {tn}, {Yn}, {J-ln} such that 0
T,
lim Yn
n-oo
= Yo,
lim J-ln
n-oo
= 0 and
tn
T, lim t n
n-oo
62
CHAPTER 2
The last relation is equivalent to the equality
p(x(t n ,Yn), M)
f,
from which, on passing to the limit, we obtain
p(x(r,yo), M)
(3.13)
f.
According to Lemma 2, the set Vb,l'" contracts to Vb,O as J-ln
-+
0. Therefore
Yo E Vb,O and, as follows from the choice of 1>,

p(x(r,yo),M) < f/2.
We have arrived at a contradiction with equality (3.13). Since solutions of
system (1.1) depend continuously on initial conditions, we have the inequality
p(x(t,xo),x(t,xd) < 1>/2,
t E [O,T]
(3.14)
for any two points xo, Xl E (Vb ,I'O U M) such that
p(xo, xd :5 d,
(3.15)
where d is sufficiently small.

According to Lemma 2, the set Vb,l' contracts to Vb,o when Jl
can always choose Jl
= Jl(1,
-+
< Jl( 1 < Jlo, such that for any point
0. So, one
Xl
E \!c'1'
there will be a point Xo in Vb,O satisfying inequality (3.15). Suppose that Jl has
been chosen in this manner. Let y be an arbitrary point of the set Vb,!' and xO
a point of Vb,O for which
Inequality (3.14) considered for Xo = xO E Vb,O and
p(x(t,xo),x(t,y)) < 1>/2:5 f/2,
Xl
= Y leads to the relation
t E [O,T],
(3.16)
which proves that
p(x(t, y), M) :5 p(x(t, y), x(t, xo)) + p(x(t, x), M) < f/2 + f/2
= f,
t E [0, T].
Moreover, taking into consideration the choice of T, relation (3.16) leads

to the inequality
p(x(T, y), M) :5 p(x(T, y), x(T, XO))
+ p(x(T, XO), M) < 1>,
63
proving that
:r;(T, y) E (U6 U M).

Since the functions V and
(3.17)
V have different signs in D l and :r;(t, y) E (U6 UM)
D l for t E [0, T], it follows that
lV(x(t, y))1 ~ lV(y) I ~ IJ,
t E [0, T].
The last inequality leads to the inclusion
x(t, y) E (V',J.' U M),
t E [0, T]
which, together with relation (3.17), proves that the trajectory which starts
= T passes through a point x(T, y)

belonging to the set (U6 UM)n(V.,J.' UM) = V6,J.'UM without leaving V',J.' UM.
at the point y E
v.s,J.'
at the moment t
If x(T,y)tM, then x(t,y) E M for all t
~ T because the set M is positively

M, then x(T,y) E V6,J.' , so that x(t,x(T,y)) = x(t +
T,y) E (V',J.' UM) for all t E [O,T]. This leads to the inclusion
invariant. If x(T,y)
x(t, y) E (V',J.' U M)
for t
O. This proves that
x(t, V6,J.') C (V',J.' U M) for t ~
o.
(3.18)
= ol(IJ)-neighbour> 0 one can find 61 = 01 (IJ) = 61 (IJ( 0)) =
According to Lemma 2, the set V6,J.' contains some 01

hood of the set M. But then for any
01(IJ(0(C)))
= 02(C) > 0 such that
(3.19)
for all t
> O. Relation (3.19) proves that the set
M is stable.
Assume now that the zero sets of the functions V and
V coincide in
Dl
Choose 5 and IJ = J-l(5) such that relation (3.18) holds. Then for a point
xo E V6,J.' either x(t, xo) E M for t = T or x(t, xo) E Vc,J.' for all t > O. In the
second case lV(x(t, xo))1 is monotonically decreasing and as t -+ +00 tends to
a limit which, according to considerations similar to those given in the proof of
Theorem 2 of 2.2, is equal to zero:
lim V(x(t, xo)) = O. But then the limit
1-+00
set Oxo of the semi-trajectory x(t, Xo; R+) belongs to the set V.,o. We claim
64
CHAPTER 2
that it is possible to choose 00 > 0 such that the limit set 0"0 lies inside M for
any point
Xo
E V 60 ,p(6 0 )' This will prove the limit relation

lim p(x(t,xo),M) = 0
t~+oo
for points Xo E V 60 ,p( 60 ) and hence the asymptotic stability of the set M will
be established.
In fact, suppose that it is impossible to choose such a 00 > O. Then for
any
fO
> 0 no matter how small, there will be a point
y, y E V 60 ,p(60)' such
that Oy C V <o,p(60) and M does not contain Oy, where 00
= oo( f)
~ fO
is
determined in such a way that x(t, V 60 ,p(6 0 ) C (V<0,p(6 0 ) U M) for all t ~ O.

Let z E Oy, z rt. M. Then because the set Oy is invariant it follows that
x(t,z) E Oy for all t E (-00,+00). Thus the semi-trajectory x(t,z;R-) lies in

V <0,0, But this contradicts the fact that M is stable in N.
Indeed, because x(t, z; R-) C V <0,0 and fo > 0 is arbitrarily small, the 0'limit points of the semi-trajectory x(t, z; R-) lie either in M or in (NnU <o)\M.
In either case, we can take a point on the semi-trajectory x(t, z; R-) so close to
the set A z that the time of passage from this point to the point z for t > 0 is
greater than a preassigned fixed number T. Since this point belongs to N n U <0
and the limit relation (3.4) is uniform for points in NnU<o, it follows, in view
of the arbitrary smallness of fO, that the time of passage from a given point to
the point z for t > 0 is determined by the distance from z to M and cannot
be greater than the number T. This contradiction proves relation (3.18). Thus
the asymptotic stability of the set M has been proved.
We make two remarks concerning Theorem 4.
Remark 1. If the condition on asymptotic stability of M on N is dropped this
may lead to instability of the set M. The following system is an example of
this:
dXI _
dt -
Here the equilibrium in M
= {(O, On
d.t = XIX2
dX2
(3.20)
is unstable. By taking V
= xi we see
that the set M is stable in N and that all the conditions of Theorem 4 are
satisfied except for the condition of asymptotic stability of M in N.
Remark 2. If we drop the condition that the zero sets of V and
V in D 1 coincide
then it may happen that the set M will only be stable but not asymptotically
stable.
65
The system
-dXl
= 0,
dt
dX2
= Xl - X2,
dt
gives an example of the situation where the equilibrium position M = {(O, On
is stable but not asymptotically stable. For the same function V as in the
previous example all the conditions of Theorem 4 are fulfilled.
2.4. Behaviour of an invariant set under
small perturbations of the system
Let M o be a closed invariant set of system (1.1). We shall find out the behaviour
of M o under small perturbations of system (1.1). The magnitude of a small
perturbation will be defined by a positive parameter Jl. Using it, we write the
perturbed system of equations in the form
dx
di = X(x) + JlY(x).
(4.1)
We shall assume that the functions X and Yare defined, continuous and satisfy
Lipschitz condition for all x in D.
Denote the region of attraction of Mo by II(Mo). It consists of all the
points Xo E D such that
n.,o eM.
Theorem 1. If M o is a closed compact asymptotically stable invariant set of
system (1.1) then one can find 6 > 0 and Jlo
= Jlo( 6)
> 0 such that for all
Jl < Jlo the system of equations (4.1) has a closed asymptotically invariant set
M = M(Jl) for which lim p(Mo, M)
1'-0
= 0 and U6(Mo) C II(M).
Theorem 1 determines the behaviour of an asymptotically stable set M o

under small perturbations. It establishes that small perturbations merely "reconstruct" the trajectories of the perturbed system in a neighbourhood of M o
by forming from them an asymptotically stable set M(Jl) which is drawn to
Mo for Jl
-+
0 and that the region of attraction of any of these sets contains
some neighbourhood of M o common to all the sets M(Jl) for Jl < Jlo.
Turning to the proof of the theorem, we denote by x(t, Xo,Jl) the solution
of the system of equations (4.1) for which x(O,XO,Jl)
= Xo.
We set x(t,xo,O) =
x(t, xo). Since x(t, xo, Jl) is continuous with respect to all the arguments t, Xo, Jl
for t E [0, Tl, Xo E U6(Mo) and Jl E [0, Jlo] for any T
it follows that the relation
lim p(x(t,xo,Jl), x(t,xo))
1'-0
> 0 and small 8 and Jlo,
=0
(4.2)
CHAPTER 2
66
holds uniformly with respect to (t, xo) E [0, T] x U6(Mo).

Since the set M o is asymptotically stable, the limit
lim p(x(t, xo), M o)
1-+00
is uniform for all
Xo
=0
E U6(Mo) for sufficiently small 6 >
(4.3)
o.
Equalities (4.2) and (4.3) enable us, using a sufficiently small 61 > 0, to
choose 6
= 6(6d > 0,
= T(6d
> 0, 1J(6d > 0, lim 1J(6 1 )

6,-0
= 0 monotonically,
hold for all Xo E U6 (Mo), IJ < 1J(6d, t E [0,+00).

Indeed, fixing 61 > 0 and using (4.3) we choose the constants 6 = 6(6 1 ) > 0
and T = T( 6) > 0 such that
for all
Xo
p(x(t, xo), Mo) <
1
2'
p(x(t, xo), Mo) <
2'
t 2: 0,
(4.5)
t >T
E U6 (Mo). Taking into consideration (4.2) and using the above 61 ,6
and T we find 1J(6d
= 1J(6 1 ,6, T) > 0 such that
p(x(t,xo,IJ), x(t,xo <
2'
t E [O,r]
(4.6)
for all Xo E U6(Mo) and alllJ < 1J(6d.

Inequalities (4.5) and (4.6) lead to the estimates
p(x(t, Xo,IJ), M o) ~ p(x(t, xo, IJ), x(t, xo)) + p(x(t, xo), M o) < 61 ,
p(x(T,xo,IJ), M o) < 6
for all Xo E U6(Mo), IJ < 1J(6d, t E [O,T]. The last inequality means that the
motion x(t, Xo,lJ) starting at the point Xo E U6(Mo) remains in U 6, (Mo) all
the time for t E [0, T] and at t
=T
comes back to a point in U6(Mo). This

is sufficient for a positive semi-trajectory of the motion x( t, xo, IJ) to belong to
the set U6,(Mo) and for the point x(T, xo, IJ) to belong to the set U6(Mo) for
all Xo E U6(Mo), IJ < 1J(6d
67
Let M i be union of all positive semi-trajectories of the motion x(t, xo, p)

for which Xo E U6(Mo), and let M i be its closure. Both M i and M i are
positively invariant sets of system (4.1) and the following inclusion relations
hold for them:
(4.7)
We define the set M to be the union of all trajectories of system (4.1) lying
in Mi. It is clear that this set is a closed compact invariant set of system (4.1).
Applying (4.7) to M we find that U 6, (Mo) :J M. In view of this, the inequality
p(Mo, M) ::; 6i holds, which proves the limit relation lim p(Mo, M) = 0 since
1'-0
61 was chosen arbitrarily. We now prove the inclusion U 6 (Mo) C II(M). To do

this, we choose ( =
(i
(6d > 0, T 1 = T 1 (6d >
analogous to (4.4) hold for sufficiently small 6i >
and all Xo E U
2~, (Mo),
p <
p(6d, t E [0, +00).

The set M 2 formed by the positive semi-trajectories of the motion x(t, xo, p),
where Xo E U 26, (Mo), satisfy the inclusions
analogous to (4.7).
Moreover, since X(Ti,XO'P) E U6(M o) for all Xo E Uu,(Mo), it follows
that x(t,xo,p) E U6,(Mo) for all t ~ T i . Therefore U6,(Mo) contains an wlimit set of the motion x(t,xo,p) for any Xo E U 2 6,(Mo), p < p(6d. Because
of the inclusion x(Ti,xo,p) E U6(Mo), this set is an w-limit set of the motion
x(t,x(Ti,xo,p),p) which starts at a point of the set U6(Mo). Therefore it

belongs to the set M which by definition contains all the trajectories of motions
which start at a point in
U~(Mo).
for all p < p(6d, so that

inclusion is proved.
This proves that
U6, (Mo)
C II(M) for p < p(6d. The required
We now prove that the set M is asymptotically stable. Since p( M o , M)

as p
-+
0, we can find 60 >
-+
and Po > such that the 6 -neighbourhood

0
CHAPTER 2
68
U60 (M) of the set M is contained in the set U6(Mo) for any j.l < [.to. But then
for j.l < rnin([.to,j.l(cSd) we have the inclusion U6 o (M) C II(M), which proves
that
lim p(x(t,xo,j.l),M)
f-++oo
=0
for all Xo E U6 o (M), j.l < j.lo < min([.to,j.l(cSd)

Now to show that the set M is asymptotically stable for j.l
< j.lo
it is
sufficient to prove that this set is stable.

Since U6 o (M) C II(M), there are no whole trajectories of the motions
x(t, XO,j.l) of system (4.1) that start in U6 o (M) and remain entirely in U6 o (M).
Otherwise, these trajectories would belong to M. But this is impossible since
U6 o (M) does not contain points of the set M. So, by applying Theorem 3 of
2.2 to show that the set M is stable it is sufficient to prove that there are no
a-limit points in M of motions x(t,xo,fl) that start at points Xo E U6 o(M) for
j.l < j.lo

We now prove that M does not contain any a-limit points of motions
x(t, Xo, j.l) starting in U 26. (Mo) outside the set M. Thus let Xo E U 26. (Mo)\M.
Then either Xo E U 26. (Mo)\M 1 or x E M 1 \M. Consider the motion x(t, Xo,fl)
for t ~ O.
First suppose that Xo E U26.(Mo)\M1 . We show that this motion leaves
U26.(Mo) for some t = -h, where 0 ~ h ~ T 1 . Indeed, let x(t,xo,j.l) E
U 26.(Mo) for all t ~ O. Then x(t,xo,fl) E U26.(Mo)\M1 for t ~ 0, since otherwise x(-tl,xO,fl) E M 1 for tl > 0 and x(t,x(-tl,xO,fl),j.l) E M 1 for all t ~ 0,
which is impossible in view of the equality x(t I, x( -t I, Xo, fl), j.l) = Xo and the
relation Xo ct. MI. We take a point x(-T,xO,fl) of the motion x(t,xo,j.l). Because it belongs to the set U26(Mo), according to (4.8) it must at time T1 belong
to the set U6(M o). But this is impossible since x(T1 , x(-T1 , XO,j.l), fl) = Xo and
Xo ct. M 1 :::> U6(M o). This contradiction shows that x(t,xo,j.l) ct. U26.(Mo) for
t = -h, where h > O. But then the motion x(t,xo,j.l) does not have a-limit
points in the set U 26. (Mo).
Now let Xo E M1\M. If the motion x(t,xo,fl) leaves M 1 for t < 0, then
it enters U26(Mo) and leaves it at some time t = -h < O. Consequently, the
motion x(t,xO,fl) does not have a-limit points in U26.(Mo).
If the motion x(t, Xo, fl) E M 1 for all t ~ 0, then the entire trajectory
of this motion lies in M 1, which contradicts the assumption that Xo ct. MI.
This contradiction shows that the inclusion x(t, Xo,fl) E M 1 for all t ~ 0 is
69
impossible.
It follows from the above remarks that the set M does not contain alimit points of the motions x(t, xo, Jl) starting in U26, (Mo) for sufficiently small
Jl. Consequently, the set M is asymptotically stable for sufficiently small Jl.
Theorem 1 is proved.
Theorem 1 enables us to investigate the behaviour of the stable set M o of
system (1.1) at the moment when it loses stability as a result of small perturbations of the system. Namely, we have the following theorem.
Theorem 2. Suppose that the system of equations (4.1) has the same closed
compact invariant set M o for all sufficiently small J.l E [O,J.lo], If M o is asymp-
totically stable for J.l = and unstable for J.l E (0, Jlo], then for sufficiently small
J.lo the system of equations (4.1) has an invariant set M p for all J.l E (0, Jlo] such
0, lim p(Mo, M p ) 0.
that M o n M p
1'-0
In other words, Theorem 2 states that when an asymptotically stable set
M o becomes unstable a new invariant set M p , Jl E (O,J.lo], of system (4.1) is
"born" .
To prove Theorem 2 it is sufficient to set M p equal to
(4.9)
where M
= M(J.l) is the asymptotically stable set of system (4.1) of Theorem 1.
Since M contains all the trajectories that start in U6, (Mo), it follows that
M o C M. Because M is an asymptotically stable set and M o is an unstable
set of system (4.1) when Jl E (O,Jlo], we have that M o # M. But then set
(4.9) is non-empty, invariant, and together with M satisfies the limit relation
lim p(Mo, M p )
1'-0
= 0.
It should be noted that small perturbations of system (1.1) may have a
severe "destructive" effect on the invariant set Mo of Theorem 1. For example,

for equation (4.1) with X and Y defined by the expressions
={
-(x + 1),
x::;
0,
-(x - 1),
-1 ::; x ::; +1, Y = 1,
-1,
x 2: +1,
the set M o consists of the interval -1 ::; x ::; +1, while M = M(J.l) consists
merely of the point
xl'
= 1 + IL.
The set M p born from M o may also differ
70
CHAPTER 2
significantly from Mo. For example, in the system
where p2 = xi + x~, the set M o consists of the single point Xl = 0, X2

while M p consists of all the points of the disc p2 ::; p. without the centre.
= 0,
Note that a change from stability to instability of the set M o does not
necessarily lead to the "birth" from Mo of a new invariant set of system (4.1)
for!J E (O,!Jo], This is the case, for example, for the point M o
equation
dx
-d
t
=X
3'
sm x
= {O} in the
+ J1.x.
2.5. Quasi-periodic motions and their closure

Suppose that the system of equations (1.1) has a quasi-periodic motion
X
= x(t,xo) = f(wt),
f(4)) E C(Tm ),
(5.1)
the trajectory x(t, xo; R) of which is compact in D. The closure of the trajectory
x(t, xo; R) belongs to the region D and consists of the points defined by the
equation
X
= f(4)),
4> E Tm .
(5.2)
According to equation of motion (1.1) we have
f(wt)
= f(O) +
it
X(f(wsds,
t E R,
so that, passing to the limit in the sequence offunctions f(wt+wt n ), lim wtn =
4> mod 271", we obtain the identity
f(wt
+ 4
= f(4)) +
n-+oo
it
X(f(ws
+ 4)ds,
(5.3)
which proves that
x(t,f(4)
= f(wt + 4,
t E R, 4> E Tm
Consequently, the set (5.2) is an invariant set of system (1.1), the trajectories
of which are quasi-periodic with the same frequency basis.
71
We denote set (5.2) by M and determine its properties as a subset of En.

Lemma 1. /fx(t,xo) E C'(w) and x(t,xo) E C(Q), then x(t,xo) E C(Q).
According to the lemma, the number
5,
describing smoothness of M, is an
invariant of the motion x(t,xo).

Thrning to the proof of the lemma, we write down the identities
= f(wt), IE C'(Tm ),
x(t,xo) = h(Qt), hE C(Tm ),
x(t, xo)
(5.4)
where m and m1 are the dimensions of the bases wand Q.

According to the definition of a basis, we can choose a basis Q, Q', using
the numbers (Q,w) by extending Q to a basis with numbers from w. With such
a choice
w = ~(P1 Q + PzQ'),
rank[P1 , Pz] = m,
(5.5)
where d is an integer, P1 and Pz are integer matrices. From identities (5.4) it

follows that
~~clll(~P1Qt+ ~PzQ't) -
= (l/J,v)ET
max
h(Qt)11 =
III(P11/> + Pzv) - h(d1/11
= 0,
that is, we have the identity
h(d1/
= I(P11/> + Pzv),
(1/>, v) E Tp,
(5.6)
where p is the dimension of the basis Q, Q'. But then

(5.7)
Relation (5.7) proves that 11 E C(Tm .), hence x(t, xo) E C' (Q).
The number m such that x(t,xo) E C(w) for some basis w = (W1, ... ,w m)
and x(t, xo) f/:. C(Q) for any basis Q
= (Q 1, ... , Qm,)
with m1 < m will be
called the real dimension of the frequency basis of the quasi-periodic function
x(t, xo). The real dimension of the frequency basis describes the dimension of
the set M at each point of it.
This fact can easily be expressed in mathematical terms when x( t, xo) E
C'(w) for
5 ~
1. In fact, we have the following lemma.
72
CHAPTER 2
Lemma 2. Let x(t,xo) = f(wt) E C'(w), where s 2: 1, w = (WI, ... ,wp ).
Then
rank
where
a~~) = m,
4J E Tp,
(5.8)
is the real dimension of the frequency basis of the x(t, xo).
To prove the lemma, first suppose that p = m, that is, the dimension of
the frequency basis w is real. Suppose that relation (5.8) does not hold for
some 4J ::::: 4Jo E Tm , so that
af(t/Jo)
rank~
= ml < m.
Differentiating (5.3) we see that
af(wt + t/Jo)
8t/J;
for all t E R (i
= 8f(t/Jo) + t
8t/J;
Jo
= 1, ... , m).
8X(f(ws + t/Jo)) af(ws + t/Jo) ds

8x
8t/J;
This proves that the functions
8f(wt + t/Jo)
af(wt + 4>0)
04>1
' ... ,
a4Jm
(5.9)
are solutions of the same system of linear differential equations
dy
dt
8X(f(wt + 4Jo))
ax
y,
(5.10)
which consists of n equations.

When t
=0 the functions (5.9) are linearly dependent, therefore they are
linearly dependent for all t E R. Consequently, we can find constants Cl,.., Cm,
I:::l Icd ::f:. 0 such that

~ of(wt + t/Jo) . _ 0
ot/J'
c, - ,
t E R.
LJ
;=1
(5.11)
'
Without loss of generality we can assume that
Cm
::f:. O. Because (5.11) is homo-
equal to wm . Let Cm W m in (5.11). Differentiating

(5.3) and subtracting identity (5.11) we find that
geneous we can make
Cm
where OJ
Wj -
(i
Cj
= 1, ... , m -
73
1). On passing to the limit we obtain from
(5.12) the new identity:
m- 1 8f()
~ 8j OJ
= X(f()),
E Tm,
from which it follows that X = f(Olt, ... ,Om_It, m) is a solution of the

system of equations (1.1). By virtue of the uniqueness of the solution of the
Cauchy problem for system (1.1), we now have
f(wt)
= f(Ot, 0),
tER
where we use the notation Q = (Q 1 , ... ,Om-d. This contradicts the assumption that m is the real dimension of the frequency basis of the x(t, xo)
= m.
This contradiction proves relation (5.8) for p
= f(wt).
Now let p > m. Then x(t, xo) E C(Q), where Q = (0 1 , ... , Q m ). By

Lemma 1, x(t,xo) E C'(Q) and the h E C'(Tm
x(t, xo)
= h(Qt)
satisfying the equality
is related to f E C'(7,,) by formula (5.7). Upon differen-
tiating this relation we find that

m
8!I()
. (8f()
= rank ~
:5 mm rank a '
rank PI .
&j(4))
Thus, ran k &4> 2: m.

On the other hand, we have the relation
analogous to (5.7). But then
8f()
8fl()
-)
ranka:5 min rank~, rankP l
= min(m, rank -Pd.
From this and the previous inequalities it follows that
8 f( )
rank a
= m,
E 7".
We denote the image of the set U6(0) = {III - oll

map f :
-+
< 6} under the
f() by M6(f(O)). The following statement is the analogue of
Lemma 2 in the general case.
74
CHAPTER 2
Lemma 3. Let x(t, xo) = f(wt)
C(w), w = (WI, ... ,wm ) and let m be the
real dimension of the frequency basis of the x(t, xo). Then there exists 6
>0
such that the map
(5.13)
is a homeomorphism for any o E Tm
By [1], the map (5.13) is a homeomorphism if it is one-tl>-one. Suppose

that the map (5.13) is not one-tl>-one. Then there exist sequences of points
~
E Tm and
E Tm (v = 1,2, ...) such that

f(~) = f(~),
f:.
and
II~ - ~II-+ 0, as v -+
00.
Then x(t,f(~ = x(t,f(~ for all t E R (v = 1,2'00')' which leads to

the system of identities
f(wt
+ ~) = f(wt + ~),
= 1,2, ... ,t E R.
(5.14)
Passing to the limit in (5.14) we obtain the identities

f(
+ ~) = f( + ~),
1/
= 1,2, ... ,
which lead to a similar system of identities for the Fourier coefficients of the
function f():
!k(ei(k ..p~-.p~)
1)
=0, =1,2, ....

1/
(5.15)
Considering (5.15) on the spectrum of the f(), that is, for those k such that
!k
:f. 0,
we arrive at the equalities:

(k,~-~)=Omod 211',
Since Iit/>~
~" -+
as
v -+
00,
1/=1,2, ....
there exists I/o
(k,~ - ~)
= 0,
Dividing each equality in (5.17) by 1It/>~

v -+ 00 we obtain the equality
(k,{)
1/
~ II
= 0,
(5.16)
= I/o ( k) > such that

(5.17)
VO.
f:.
and passing to the limit as

(5.18)
75
where { is the limit of the sequence (4): - 4>~)/II4>: - 4>~11 (v

1,2, ...).
Because II{II = 1 we can assume that the mth coordinate {m of the vector
{=
(6, ... ,{m) is non-zero. Solving relation (5.18), we find that

k
= _k 16 + ... + km-1{m-l .
{m
(5.19)
Using equality (5.19) we prove that
for all
4> E Tm .
To do this we set
PN(4))
Ikei(k, ,
IIkll~N
where 1(4)) :::::
L:k Ikei(k,
is the Fourier series of the function 1(4)). Because
relation (5.19) holds for the values of k in the spectrum of 1(4)), it also holds
on the spectrum of PN (4)). This leads to the identity
where 4>' = (4)1,''' ,4>m-d,

But then
= ({1"" ,{m-d
--> 00 we see that J = O.

Consequently,
= 0 almost everywhere on Tm And because 1(4))
and passing to the limit as N
1I/(4 - 1(4)' - (f /{m)4>m, 011
76
CHAPTER 2
is continuous for </> E Tm , this equality holds for all </> E Tm . Relation (5.20) is
proved.
It follows from (5.20) that
f(wt)
= f(nt, 0),
t E R,
where 0 = (Oi,"" Om-d, OJ = Wj -(jfm)w m . But then the real dimension

of the frequency basis of the function x(t,xo) = f(wt) does not exceed m - 1
which contradicts the assumption.
The map I : </>
--+
I(</ does not always define a homeomorphism between
the torus Tm 51 (</>d x ... X 5 i (</>m) and the manifold M, where 5 i (rPj) is
a circle with the angular coordinate system. This becomes evident if we take
a function I E C(Tm) that is periodic with respect to one of the variables rPv
with period 27r f d, where d is an integer, d ~ 2. Then every point of M is the
image of d different points of Tm , namely, the points </> + 18 (l = 1, ... ,d), where
= ... =
= ... =
Bv - l
BV + 1
8m
0, Bv
27r fd. The non-uniqueness of
8i
the inverse image of a point f(</>o) E M under the map I : Tm --.. M is related
to the fact that it is possible to choose a frequency basis of a quasi-periodic
x(t, xo) in different ways.

For x(t,xo) E C(w) we introduce the notion of maximal frequency basis
by defining a frequency basis w = (W1,'" ,Wm) to be maximal if any other
frequency basis 0 = (0 1 , ... ,Om,) of this is related to W via the relation
W
=PO,
(5.21)
where P is an integer matrix of rank m.

From this definition it becomes clear that the number of frequencies in
a maximal basis is equal to the dimension of the real frequency basis of the
function x(t, xo), and two maximal bases wand 0 satisfy (5.21), where both
P and p- 1 are integer-valued. A maximal basis of a periodic function of one
variable is equal to the largest frequency of the function x(t,xo) calculated
from the smallest of its periods.
= f(wt)
The map I : Tm --.. M is a

homeomorphism if and only if the frequency basis w = (Wi,'" ,wm ) is maximal.
To prove the theorem first assume that the mapping I : Tm --.. M is a
homeomorphism. Then I(</:j; 1(0) if </>:j; 0 mod 27r. Let 0 = (0 1 , ... ,Om,)
Theorem 1. Let x(t, xo)
E C(w).
77
be a basis of the function x(t,xo). Then
x(t,xo)
= !I(Qt),
t E R,
where h(1fJ) = !I(1fJ1, ... ,1fJm, E C(Tm,). It follows from the arguments of the
proof of Lemma 1 that the bases wand Q satisfy equality (5.5) and the functions
1 and h satisfy identities (5.6) and (5.7). Because the mapping 1 : Tm

is one-to-one, the matrix Pz = 0 in equality (5.5).
Indeed, if Pz ::j:. 0 then it follows from identity (5.6) that !I(O)
-+
= I(Pzv) =
1(0) for all v E 'Tp-m, , where P > m1. For small v this contradicts the inequality 1(0) ::j:. 1(1/, where I/> ::j:. 0 mod 21T. Thus the bases wand Q satisfy the
equality:
(5.22)
where d is an integer, P 1 is an integer matrix of rank m and the functions
and
!I satisfy identity (5.7). Since h is periodic, it follows from (5.7) that

(5.23)
where Pi is the jth column of the matrix Pl. Because 1(0) ::j:. 1(1/ for I/> ::j:.
o mod 21T, equalities (5.23)

Pi
can hold only if
= 0 mod d,
= 1, ... , m1'
Then the matrix (1/d)P1 = P is an integer matrix of rank m and equality

(5.22) is equivalent to equality (5.21). Thus, any basis of the function x(t, xo)
different from w is related to w by relation (5.21) and consequently, the basis
w is maximal.
Suppose now that the frequency basis w = (W1,'" ,wm ) of x(t,xo) =

I(wt) E C(w) is maximal. Suppose that the map 1 : Tm -+ M is not a
homeomorphism. Since 1(1/ is continuous and periodic with respect to 1/>11 (v =
1, ... , m) with period 21T, the assumption made is equivalent to the property
that the inverse image under the map
1 of a point of the set M is not unique.
Consequently, there exist two points 1/>1 E Tm and z E Tm , 1 ::j:. z such that
I(d = I(z). It follows from the arguments given in the proof of Lemma 3
that
(5.24)
78
CHAPTER 2
Given a vector <p and a number 0', we shall denote by <p mod 0' the vector
<Po with non-negative components of the smallest length such that <p - <Po = O'p,
where p is an integer vector. We set 6 (<P2 - <pdf(27r) mod 1. Then it follows
from identity (5.24) that
f(<p)
= f(<p + 27r6),
Equality (5.25) shows that the function F(t)
<p E Tm .
(5.25)
= f(M)
is periodic with respect
to t with period 27r. From this it necessarily follows that
6
where d is an integer, P
= pfd,
= (PI, ... ,Pm)
(5.26)
is an integer vector, 0:S Pi < d, j
1, ... ,m, P f O.
Indeed, by choosing the basis (1, (>,') from the numbers (1,6) we obtain the
following representation for 6:
(5.27)
where d is an integer, PI is an integer vector, P2 is a matrix of integers,
rank [PI ,P2] :2:
ml -
1,
being the number of frequencies of the basis (1, (>").
ml
From (5.27) and the identity F(dt)
= f(Plt + P2(>,'t),
t E R, it follows
that
(5.28)
For
ml
> 1 the vector d6 cannot be integer-valued for any integer d and so it
follows from (5.27) that P2
O. But if P2
0, the identity (5.28) leads to the
equality
(5.29)
Since f(wt) has a frequency basis w
map f : U60 (0)
-+
= (WI,'"
,wm
of real dimension, the
M 60 (J(O is a homeomorphism for small 60 > O. This
contradicts equality (5.29) considered for those v satisfying P2 v E U 60 (0). The

contradiction shows that
ml
= 1.
Consequently the basis of numbers (1,6)
consists of the number 1, and representation (5.27) has the form (5.26), where
PI
=p. The inequalities for the coordinates of the vector p follow from similar
inequalities for the coordinates of the vector 6. This finishes the proof of (5.26).
79
Suppose that the jth coordinate Pj of the vector P is non-zero. We define

the numbers
n = (n I , ... , nm )
_ PI AHj
d
WI -
+ (1)10"
-
where the integers
IIi,
by the relations:
HI, ... ,
i- j,
A
Wj -_ dPj AHj , ... ,Wm _- Pm
dHj
+ (l)V
- mH m ,
A
ni (i = 1, ... , m)
the numbers n are
are chosen so that the numbers
are positive. Since the numbers ware incommensurable,

also incommensurable.
Consider the function
This is periodic with respect to the variables t/Jv,
II
i- j,
with period 2'11" since f
is periodic, and it is periodic with respect to t/Jj with period 2'11" since it follows
from (5.24) that
h(lPI,oo.,l/Jj +2kj 'll", ... ,t/Jm)
= f(Plt/Jj
+ (-lY't/JI+
2 PI k
Pi .1.
2 Pi k
Pm .1.
()Vm.l.'l'm+ 2'Tr Pm kj )
+ 'll"d j'''''d'l'j+ 'Tr d j""'d'l'j+ -1
d
for any integer ki . Since II (t/J) is periodic, the function II (nt) is quasi-periodic,
and from the formulae relating wand n, f and II it follows that
lI(nt) = f(wt) = x(t,xo),
t E R.
This shows that x(t, xo) E C(n).

We set (lId)?
= (( -It'el, ... ,pld, ... , (-ltmem), where

e v = (0, ... ,1, ... ,0)
is the 11th unit vector,
II
i- j.
The relations between wand
n can be expressed
in terms of the matrix (lid)? in the form

w
= (lld)?n.
(5.30)
Since the matrix (lid)? in (5.30) is not integer-valued, the basis w of the
function x(t, xo) is not maximal.
80
CHAPTER 2
Thus we have arrived at a contradiction, which proves that the map
f : Tm
M is a homeomorphism. Theorem 1 is proved.
Next we consider the problem of the existence of a maximal frequency

basis of the function x(t,xo) E C(w). To do this we first study the properties
of the map f : T m
M when x(t, xo)
-.
f(wt)
C(w) and the basis w is not
a maximal basis of the function x(t, xo).
Let r d be a subgroup of the group zm ofall integer vectors k
= (kl>""
km )
with the property that for a vector P the vector kd + P also belongs to it for
'Vk E zm, and let d be an integer 2: 2. We define T d to be the set of all vectors
rd.
P mod d, where P runs over the subgroup
zero vectors of T d by Td. If Td

all its elements PI, ... ,PN, pv
Td
{PI, .. . , PN }.
The set of vectors Td
:f:.
:f:.
We denote the set of all non-
0, then we define the set Td by prescribing
0 for
II
= 1, ... , N,
= {PI, ... , PN}
2:
1 and write it as
has the following property: if we
arrange the positive numbers in the set of jth coordinates Pi v of all vectors
'T'
.
.
d
l'
h
(1) (2)
(lj) ( (i)
.L d III Illcreasmg or er to lorm t e sequence Pi
,Pi , ... , Pi
Pi <
(i+l) .
1
I
1)
h
(1) .
d"
f
th
b
d
d
(i)
. (1)
Pi
' Z
i , t en Pi IS a IVlsor 0
e num er an Pi
zPi
Pv E
= ,... ,
for i = 1, ... ,Ii and Ii
+ 1 = d/p~I).
This property can be easily derived from the definition of the set Td .
Lemma 4. If the frequency basis w
= (WI,'"
,wm ) of the x(t,xo)
= f(wt)
C(w) is not maximal but m is the real dimension of the frequency basis, then
Jor the Junction J( 4 there is a set Td
= {PI, ... , PN}
such that
(5.31)
Jor any pv E Td and
(5.32)
Jor any p such that p mod d
rt. Td.
To prove the lemma, we consider the set {rPl,' .. ,rP N} of all the roots of
the equation
where
rPi
= rPv
if j
:f:.
II,
f(rP)
= f(O),
.p E Tm ,
.pj :f:.
0 for j
= 1, ... , N.
(5.33)
From the arguments in
the proof of Lemma 3 it follows that this set is finite and from the arguments
given in the proof of Theorem lone can see that 1/>",

where 6", has the form of (5.26) with P =
P~
81
= 21f'6", for II = 1... N.
and d = d",. By taking the
I.c.m. of all denominators in the expression for 6", in the form (5.26) we find
a representation 6",
= p",/d (v = 1... N) for
6",. where d is the I.c.m. Then.
the vector P", = (PI", ... Pm",) satisfies the inequality 0 ~ Pi'"
1, ... ,m, v= 1... N.
<d
for j
We set T d = {Pl ..... PN }. It follows from identity (5.24) considered for
1/>1
= 0 and 1/>2 = 21f'p",/d that identity (5.31) holds for v = 1, ... N. The prop-
erties of the set Td clearly follow from identities (5.31) because the function
/(1/ is periodic with respect to 4>",. II = 1... m. with period 21f'. Inequality
(5.32) also follows because the set {4>1 ... I/> N} contains all the roots of equation (5.33) and consequently T d contains all the vectors P = P mod d for which
equality (5.31) holds. Lemma 4 is proved.

Lemma 4 indicates a simple procedure for transforming an arbitrary basis
= (WI, ... wm ) of the function x(t. xo) = /(wt) E C(w) to a maximal one.
We describe one step of this procedure. Suppose that m is the real dimension of
the frequency basis of the function x(t. xo). Consider the set Ttl = {Pl ... PN}.
Suppose that the first coordinate of at least one of the vectors from T d is not
zero. Then the first coordinates of the vectors P", (II = 1... N) are multiples
of the least positive number p~l). Without loss of generality we can assume
that this is the vector PI
= (Pu ... Pmd which has p~l) as its first coordinate.

= 0 by setting
We define the numbers (0 1 ... Om)

(1)
Wi
PI
= dHi.
i'"\
Wj+1
(1)'"J+IHi+i.
= -pj+i,i
-HI + d
i'"\
i'"\
= 1, ... ,m -
1. (5.34)
and choosing integers IIj+! so that the numbers 0 1 ... ,Om are positive. Because W is a basis. the numbers 0 1 ... Om are incommensurable.
Consider the function
(5.35)
By identity (5.31) the function ft(1/J) turns out to be periodic with respect
to 1/J", (II
= 1..... m)
with period 27f'. Thus ft(Ot) E C(O). It follows from
formulae (5.34) and (5.35) that
h(Ot) = /(wt) = x(t. xo).
t E R.
82
CHAPTER 2
therefore x(t,xo) E C(Q).

We define the matrix P by setting
(5.36)
where ell = (0, ... ,1, ... ,0) is the v-unit vector taken as a column vector.
According to formulae (5.34)-(5.36) we have the equalities
(5.37)
Consider the roots of the equation
ft(1/J)
= ft(O),
1/J E Tm.
(5.38)
According to formulae (5.37) these roots are defined in terms of the values
<Pk = 21rk and <P1I,k

from the formulae
= 27r(PII/d+ k)
(where k
PII
27r ( d
+k
= (kl ... ,km) E zm,
PII E Td )
P
P
= d:1/JII.k
= d:1/J1I
(5.39)
as 1/Jk mod 27r and 1/J1I mod 27r.

From (5.39) we select the equalities for the first coordinates of the vectors
1/Jk = (1/J1/c, ... ,1/Jmk) and 1/J1I
= (1/JIII, ... ,1/Jmll)' We have
(I)
27rk l =
P~ 1/J1k.
27r (P~"
(I)
+ k 1) = P~
1/Jll1
(5.40)
Since p~l) is a divisor of the numbers d and PIli it follows from (5.40) that
1/Ja
= 0 mod 27r,
1/JIII
= 0 mod 27r.
Hence 1/Jk mod 27r and 'ljJ1I mod 27r are vectors that have their first coordinates
equal to zero for arbitrary k E
zm
and any v = 1, ... , N. But then the set
Td,
= {p~,
, PN'} constructed using the function It (<p) consists of the vectors
P~
(v = 1,
, N') with first coordinate equal to zero. This finishes the first
step of the procedure of transforming a basis w into a maximal basis Q.

Suppose that the first coordinates of all the vectors in T d
= {PI, ... , PN }
are equal to zero and the second coordinate of at least one of the vectors from T d
is not equal to zero. In this case we have to carry out the same transformations
83
with the function f(ifJ) and basis w as in the case considered above except
that the transformations must be applied to the variable ifJ' = (ifJ2, ... ,ifJm)
and the basis w'
WI
= (W2, ... ,wm ), leaving the coordinate ifJI
and the frequency
unchanged. In this way we find a new II('I/J) and a new basis
h(nt)
n such that
= f(wt) = x(t, xo) E C(n), and the set Td' = {pL ,PN'} constructed
using the function
II (ifJ)
consists of the vectors p~ (v
= 1, ... , N') with the first
and second coordinates equal to zero.

Continuing this transformation process with the function f( ifJ) and basis
w
we can construct in a finite number of steps a function f' and a basis
that the map
f' : Tm
-+
M is a homeomorphism and the basis
basis of the function x(t, xo)
= f'(n't) = f(wt).
n' such
n' is a maximal
As a result we have the following statement.

Theorem 2. If the motion x(t, xo)
= f(wt)
is quasi-periodic, then the function
x(t, xo) has a maximal frequency basis.
Theorems 1 and 2 together with Lemma 1 yield the following result.

Theorem 3. If the motion x(t,xo) is quasi-periodic and x(t,xo) E C'(w) with
s
2: 0 then the set M
= x(t, Xo; R)
is C' -homeomorphic to an m-torus, where
m is the real dimension of the frequency basis of the function x(t, xo).
A set that is C' -homeomorphic to an m-dimensional torus is usually
called an m-dimensional toroidal manifold of smoothness s or simply an mdimensional torus of smoothness s.
Theorem 3 therefore states that the closure of a quasi-periodic orbit of
system (1.1) is an m-dimensional s-smooth torus.
Because M is homeomorphic to an m-dimensional torus and belongs to
the space En it necessarily follows that m < n [1]. At the same time the value
m
n - 1 is possible; it is easy to construct an analytic submanifold of the
space Em+l that is analytically diffeomorphic to an m-dimensional torus.
2.6. Invariance equations of a smooth manifold and
the trajectory flow on it
Let M be an m-dimensional smooth compact manifold, a neighbourhood U(M)
of which lies in a domain DC En. We form a finite covering of the set M:
84
CHAPTER 2
= U M6(Xi)
i=1
by choosing a sufficiently dense net of points Xi E M (i
1, ... , N) and
defining
where
lII~(xi)
is an n-dimensional ball in En with centre at the point Xi and
redius fl.
For a sufficiently small fJ > 0, the set M6(Xi) is diffeomorphic to an mdimensional ball 1II1'(0) of the space Em with a certain coordinate system
</> = (</>1, ... ,</>m) and consequently, it consists of the points

the equation
X
= Fi (</,
E En given by
</> E lII1'(O),
where the function Fi (</ is continuously differentiable for </> E lII1'(O) and
oF;(</
rank ~ = m,
</> E 1II1'(0).
We define a multivalued function F( </ by setting

N
= UFi (</,
F(</
i=1
</>
=1II1'(0),
and use it to write the equation of the manifold M in the form

X
= F(</,
</> E 1II1'(0).
(6.1)
We write down the invariance conditions for M with respect to the system of
equations (6.1) by using the analytic expression for the equation of the manifold
M.
Let Xo E M, so that Xo Fi(</>o) for some i, 1 ~ i ~ N, and </>0 E lII1'(O).

The set M is an invariant set of system (1.1) if and only if the vector X(xo)
defining the right side of this system at the point
Xo
belongs to the tangent
hyperplane to M at the point xo. This leads to the equality

(6.2)
for some ri
= ri(</>o).
85
Upon multiplying (6.2) by the transpose (8Fi (4>0)/84* of the matrix
8Fi (4)0)/84>, we obtain the equality

ri(4)O)ri
*
= ( 8Fi(4>0)
84>
X(Fi(4>O,
(6.3)
where we have set

(6.4)
Matrix (6.4) is the Gramm matrix of linearly independent vectors, therefore
det r i (4))
"# 0, 4> E 1II6"(0).
The vector ri is uniquely defined from equality (6.3):
ri = r i-1 (4)0) (
8Fi(4>0) *
84>
X(fi(4>o.
Substituting this value of ri into equality (6.2) we find that the function Fi (4))
satisfies the relation
for every 4> E 1II6" (0).

Consequently, the manifold M is an invariant set of system (1.1) if and
only if the function F( 4 satisfies the equation
(FcP r- 1 F; - E)X(F) = 0
(6.5)
for all 4> E 1II6"(O). In (6.5) we have introduced the notation
and the expression "F( 4 satisfies" means that each function Fi (4)) for i =
1, ... ,N satisfies equation (6.5).
Equation (6.5) will be called the invariance equation of the manifold M
with respect to system of equations (1.1).
Suppose that the function F(4)) satisfies the invariance equation (6.5).
We consider the differential equation
(6.6)
86
CHAPTER 2
= 1, ... , N) on M, every point

M is an interior point of the ball III:;(x;), so that Xo is not a boundary
point of M6(Xi).
Because boundary points are mapped into boundary points under a diffeomorphism, cPo = F;-l(xO) is an interior point of the ball III:S(O). Hence the solution cP = cP;(t, cPo) of equation (6.6) is defined for t in some interval 10 = Io(cPo)
containing the point t O. Differentiating the function F;( cPi(t, cPo and using
relations (6.5) and (6.6) we find that
For a sufficiently dense net of points Xi (i
Xo E
But then
(6.7)
where x(t, xo) is the solution of system (1.1), x(O, xo)
= xo.
Using the standard
notation, relation (6.7) means that
(6.8)
Consequently, any trajectory of system (1.1) that starts at a point of the
set M, can be obtained by using equality (6.8) from equation (6.1) of the
manifold and the solutions of the equation
~~ = r- 1 (cP)F; (cP)X(F(cP,
understood as the set of equations (6.6) for i
cP E III:S(O)
= 1, ... , N.
(6.9)
Having in mind this
possibility we shall say that equation (6.9) defines a trajectory flow for system
(1.1) on the invariant manifold M, or that it is the equation of this flow.
We make some remarks concerning equations (6.5) and (6.9).
Let B(cP)
= U Bi(cP),
;=1
<jJ
E III:S(O), be an n x (n-m) matrix that extends
the matrix Ftf>(<jJ) to a non-singular matrix [Ftf>(<jJ), B(4))] : det[F>(4)), B(4))] ::j;
0, cP E III:S(O). Multiplying equation (6.5) by the matrix [Ftf>(cP) , B(cP)]* we
obtain the equation
B*(4))(F>r- 1 F; - E)X(F)
=0
equivalent to (6.5) and containing n - m equations instead of the initial n

equations.
Let P(</
= U p.(</ ,
;=1
87
</> E III6'(O) , be an m x n matrix such that its
product with the matrix Fq.(</ is non-singular:
# 0,
det P(</Fq.(</
</> E ill6(O).
The following identity follows from invariance equation (6.5):
P(</Fq.(</r-1(</F;(</X(F(</) = P(</X(F(</>)), </> E III6'(O),

and we can use it to write equation (6.9) in the following form:
~~
= (P(</Fq.(</>))-l P(</X(F(</), </> E III6'(O),
where there is "certain degree of freedom" in choosing the right hand side.
Finally we note that the non-uniqueness involved in the description of the
equation of M in the form of (6.1) disappears whenever the manifold equation
is given by a single-valued function F( </ for all </> in a compact set
J{.
In this
case each of the equations (6.5) and (6.9) is to be understood as a separate

equation for all </> E
J(.
In particular, the invariance equation (6.5) for the m-dimensional torus M

defined by equation (5.2) with the function /(</ satisfying the conditions
rank 8/(</
8</>
=m
T/</> E Tm
(6.10)
has the form
(6.11)
while the trajectory flow equation (6.9) for system (1.1) on M takes the form
d</>
dt
= r- 1 (</(8/(</)* X (f(</>))
8</>
'
</> E Tm
(6.12)
For example, for a torus defined by the equations
(6.13)
where p~ = const
> 0 for
v = 1, ... , m, invariance equation (6.11) has the form
of the system
cos </>v(sin </>vXZv-1
+ cos </>vXzv) =
0,
v = 1, ... , m
88
CHAPTER 2
equivalent to the system
(6.14)
where Xj (j
= 1... ,2m) denotes the expression
If system of equalities (6.14) is satisfied for all 4J E Tm then the torus
(6.13) is an invariant set of system (1.1) with n
= 2m.
In the latter case the
trajectory flow of system (1.1) on the torus (6.13) is defined by equation (6.12)
which takes the form of the system
d4Jv
1(
.)
-d = 0 cos4JvX2v-l -sm4Jv X 2v ,
t
Pv
II
= 1... , m,
4J E Tm .
2.1. Local coordinates in a neighbourhood of a toroidal manifold.

Stability of an invariant torus
Let M be a toroidal manifold defined by equation (6.1) with the function
F(4J)
= f(4J)
satisfying conditions (6.10). Suppose that the matrix 8f(4J)If}4J
can be extended to a periodic basis in En and that B( 4J) E C+l (Tm ) is the
extending matrix so that
det[8f(4J)/84J, B(4J)]"# 0,
4J E Tm .
(7.1)
In studying the trajectories of system (1.1) which start in a neighbourhood of the manifold M it is convenient to go from the Euclidean coordinates
x = (Xl X n ) to local coordinates 4J = (4Jlo ... 4Jm). h = (h1 ... hn - m )
introduced in a neighbourhood of M in such a way that the manifold equation
(6.1) takes the form

h
= o.
4J E Tm
(7.2)
Under the above assumptions concerning the manifold M. the local coordinates
4J, h can be introduced by giving equations relating the variables x with the
variables 4J. h by means of equalities
x = f(4J)
+ B(4J)h.
(7.3)
89
Lemma 1. For each sufficiently small 6

61 (6)
-->
> 0 one can find 61
= 61 (6)
> 0,
0 as 6 --> 0 such that every point x satisfying the condition
p(x,M) < 6,
(7.4)
has local coordinates 4>, h such that
(7.5)
Indeed, since
p(x,M)
for some tP
= yEM
infp(x,y)= inf
.pET
IIx-f(4))II=lIx-f(tP)11
= tP(x) E Tm , inequality (7.4) means that

Ilx -
f(tP)1I < 6.
We set
x=f(tP)+z
and show that for a sufficiently small 6 > 0 the equation
f(tP)
+ z = f(4)) + B(4))h
(7.6)
can be uniquely solved with respect to 4>, h in the domain

(7.7)
where 61 = 61 (6) is a sufficiently small number, 61 (6)
-->
0 as 6 --> O.
To do this we write equation (7.6) in the form
a~~) (4) -
tP) + B(tP)h = z - [f(4)) - f(tP) -
a~~) (4) -
tP)] - [B(4)) - B(tP)]h

(7.8)
and, taking into account inequality (7.1), solve (7.8) with respect to 4> - tP,h.
As a result we obtain the system of equations
af( tP)
] -[B(tP+(4>-tP-B(tP)]h } ,
4>-tP = L 1 {z- [f(tP+(4>-tP-f(tP)-~(4>-tP)
h=
L2{
Z -
[f(tP + (4) - tP - f(tP) -
a~~) (4) -
tP)]-
-[B(tP + (4) - tP - B(tP)]h},
(7.9)
90
CHAPTER 2
where 1
= 1 (.,p) and 2 -
2(.,p) are matrices of appropriate diemensions.
Using this system of equations one can see how to fix 61
= 61 (6)
> 0 for
a sufficiently small 6 > 0 in such a way that in the domain (7.7) the operator
defined by the right hand side of this system will be contracting in the space
of the vectors </J - .,p,h. This is sufficient to solve uniquely equations (7.9)
with respect to </J - .,p, h in the domain (7.7) for the corresponding choice of
61 = 61 (6) > 0 and sufficiently small 6 > O. But then any point x satisfying
condition (7.4) has local coordinates </J, h satisfying inequality (7.5).
Consider the matrix
fo(</J)
= B(</J)B(</J),
</J E Tm
It is the Gramm matrix of linearly independent vectors and so the eigenvalues of
this matrix are positive for all </J E Tm . Since the matrix fo(</J) is periodic with
respect to </J one can find upper and lower estimates
')'0, ')'0
for these numbers
that are independent of </J. The quadratic form (r o( </J )h, h) can be estimated
by the inequality
(7.10)
for all </J E Tm
hE E n -
Inequality (7.10) leads to the estimate

(7.11)
which relates the Cartesian coordinates x = (Xl, ... , x n ) and the local coordinates </J
= (</J1, ... ,m), h = (h 1, . .. , hn - m ) of any point of a neighbourhood
of the manifold M.
It is clear from Lemma 1 and inequality (7.11) that a 6-neighbourhood of
M considered as a subset of points in En is a certain domain contained in a 61 -
neighbourhood of manifold (7.2) considered as a subset of points in Tm x En-m

and vice versa.
We use this fact to write in local coordinates the equations of the motion of
system (1.1) which starts in a neighbourhood of M. To do this we differentiate
relation (7.3) as a formula of change of variables in system (1.1) and we obtain
instead of (1.1) the system
f (</J)
[ 8 8</J
+ 8B(</J)h] d</J + B(</J) dh = X(J(</J) + B(</J)h).

8</J
dt
dt
(7.12)
91
Condition (7.1) enables us to solve system (7.12) with respect to d</>/dt, dh/dt
for all h in the domain (7.5) for sufficiently small 61 > O. This leads to the
system of equations:
~~ = L1(</>,h)X(f(</ + B(</h)
~~ = L 2 (</>, h)X(f(</ + B(</h)
(7.13)
where L 1(</>, h), L 2 (</>, h) denote the blocks of the matrix
L(</>, h)
= column(L 1 (</>, h), L 2 (</>, h
which is the inverse of the matrix (8f(<//8, B(</).
The right hand side of the system of equations (7.13) is defined on the
domain (7.5), is periodic with respect to </>v (/I = 1, ... , m) with period 211"
and has degree of smoothness with respect to </>, h one less than the degree
of smoothness of the manifold M and the smoothness of the right hand side
of system (1.1) in a neighbourhood of M. The system of equations (7.13) is
equivalent to system of equations (1.1) considered on the inverse image of the
domain (7.5) under the mapping x - (,h) given by the change of variables
formula (7.3).
Note that the expressions for the matrices L 1 (, h) and L 2 (, h) in terms of
the matrices f,p = 8 f( )/8 and B B() can be found by using the Frobenius
formula [39] for inverting a four-block matrix. This leads, for example, to the
following values for L 1 (, h) and L 2 ( </J, h):
where
r 1 = r 1 ( </J , h)
8Bh)
= ( f,p + 8Bh)*(
8</J
f,p + 8 .
92
CHAPTER 2
Suppose that the manifold M is an invariant set for system (1.1). Then
set (7.2) which is an m-dimensional torus in Tm x E n -
is an invariant set
of system (7.13). Estimate (7.11) shows that the nature of the stability of the
manifold M is uniquely determined by that of the torus (7.2): if the manifold
M is asymptotically stable, stable or unstable then torus (7.2) is asymptotically
stable, stable or unstable, and conversely.

Consider in equations (7.13) the terms linear with respect to h. The invariance condition for the torus (7.2) is expressed by the identity
(7.14)
therefore, to within terms of order
IIh1l2, equations
dt = a() + ~()h,
(7.13) have the form
dh = P()h.
dt
To find the values of the function a() and matrices
~(),
(7.15)
P() it is enough
to substitute the expressions (7.15) into equation (7.12) and equate the terms
of order
IIhllo and IIhll in
relating a = a(),
the equalities so obtained. This leads us to equations
~()
and P = P():
8~~) a = X(J(,
O~~)~ +
B()P
= 8XCfx(
B() _
8~~) a().
(7.16)
The first equation of (7.16) defines a( ) by the expression
a()
The second defines
= r-1()(O~~)" X(J( ,
~()
where r()
(7.17)
and P( ) as
~( = L~()(OXCfx(
P()
E Tm .
= LO() (OX(J(
2
8x
B() _
O~~) a() ,
(7.18)
B() _ oB( a()

8'
= 1;1", and L~(), Lg() are the blocks of the inverse matrix LO()
of the matrix
[I""
B]. In particular,
L~()
= [/;(E -
Lg()
= [B"(E - l",r- 1 1;)B]-l B"(E - l",r- 1 I;).
Brill B")/",]-l I;(E - Brill B"),
93
The system of equations
dh
dt/J
di = a(t/J), di =
P(t/J)h,
(7.19)
where a(t/J) is the function (7.17), P(t/J) is matrix (7.18), will be called the
system of variation equations of the invariant torus (7.2). The first of equations
(7.19) describes the "tangent" and the second the "normal" components of
the trajectory flow of system (7.13) in a neighbourhood of torus (7.2). It is
clear that the variation equations are defined subject to the condition that the
function X(x) has continuous derivatives with respect to x in a neighbourhood
of the manifold M.
Theorem 1. Suppose that system of equations (1.1) and the toroidal manifold
M satisfy the above-mentioned smoothness and invariance conditions, and that
local coordinates can be introduced. If there is a positive-definite quadratic form
v = (S(t/J)h,h) :::: ,llhW, ,= const > 0,

defined by the symmetric matrix S(t/J) E C1(Tm
(7.20)
such that its total derivative
given by virtue of variation equations (7.19) as
v = (S(t/J)h, h), S(t/J) = S(t/J)P(t/J) + P*(t/J)S(t/J) + a~~) a(t/J)
(7.21)
is negative definite:
v::; -.BllhIl
.B = const > 0,
(7.22)
then M is an asymptotically stable invariant set of system (1.1).

To prove the theorem it is sufficient to prove that the invariant torus (7.2)
of system (7.13) is asymptotically stable. Let f be a fixed number in the interval
(0,6d. From identity (7.14) and the smoothness conditions for the function
X( x) it follows that
IIL1(t/J, h)X(J(t/J) + B(t/J)h) - a(t/J)II ::; M(f),
IIL 2 (t/J, h)X(J(t/J) + B(t/J)h) -
(7.23)
P(t/J)hll ::; M(f)lIhll
for all t/J, h in the domain

(7.24)
94
CHAPTER 2
where M(f) is a positive constant, M(f)
-t
0 as f
-t
O.
Consider the derivative of the function V taken along the solutions of

system (7.13).
We have
~~
V + ({)~~) (L 1 (4>, h)X(f(4 + B(4))h) -
a( 4>))h, h)+
+2(5(4(L 2 (4>, h)X(f(4

where
V is the function (7.21).
+ B(rP)h) -
P(4))h),h},
From inequalities (7.22) and (7.23) we have the
estimate for dVjdt:

(7.25)
where
J(
is a positive constant chosen to satisfy the condition
max 8S{)(4))
ETm
rP
11+ 2 ET
max 115(4))11:::; K.
m
This estimate holds for all 4>, h in the domain (7.24). For sufficiently small
f
> 0 it follows from estimate (7.25) that

dV
{3
2
/3
-dt <
--llhll
< -V
2
21'1 '
(7.26)
where 1'1 is a constant determined from the inequality
Since the function V is positive definite we can write inequality (7.26) in

the form
where (rPt, h t ) is the solution of system (7.13) that takes the value (rPo, h o) when
t
0, V(t) is the value of V when 4>
4>t, h = ht, and T is a non-negative
number defined to be the maximal value for which (4)t, h t ) belongs to domain
(7.24) for any t E [0, T). We choose
4>0, ho in the domain

(7.28)
95
Then T will be positive and inequality (7.27) takes the form
Ilhdl < (e-({3/(4 .,.t,

o
t E [O,T).
(7.29)
Since T is positive, it follows from inequality (7.29) that the curve (4)t, ht)
does not hit the boundary of the domain (7.24) at any finite t 2: O. Hence it
follows that T
= +00 and
(7.30)
The arbitrariness of the choice of {and inequality (7.30) prove that any solution
(4)t, ht) of system (7.13) with initial conditions in the domain (7.28) satisfies
the conditions needed for the torus (7.2) to be asymptotically stable:
Ilhtll < (,
t E [0, +00),
lim
t-+oo
Ilhtll = O.
This proves the theorem.

We conclude this section by noting that it is always possible to introduce
local coordinates <P, h in a neighbourhood of our manifold M. This follows from
the theorem on periodic basis in En for n = m
+ 1 or n > 2m.
Otherwise it is
necessary to embed the system of equations (1.1) in an "extended" system with

the required properties in order to introduce local coordinates. For example,
system (1.1) and the equation
(7.31)
where p > 2m - n, A = diag{Al"'" Ap }, Aj = const < 0 (j
be taken as the extended system of equations.
= 1, ... ,p), can
However it is possible to introduce local coordinates 4>, h and write motion

equations (1.1), (7.31) in the form of system (7.13) in a neighbourhood of the
set formed by the points x E M, Y
= O.
Here the stability characteristics
of M considered as an invariant set of system (1.1) can be transferred to the

manifold (7.2) considered as an invariant set of system (7.13). Consequently,
to study it we can use the various statements proved for the manifold (7.2)
and, in particular, the theorem on stability with respect to variation equations
(7.19) given before.
96
CHAPTER 2
2.8. Recurrent motions and multi-frequency oscillations

A motion x(t,xo) of system (1.1) is called recurrent if for a given ( > 0 and
any to, T E R, there exist T
= T({) > 0 and 8 such that to < 8 <to + T
IIx( T, xo) - x(8, xo)II
and
< (.
According to the definition, recurrent motion has the property that any arc of
time length T
=T({) approximates the whole arc to within (:

x(t,xo;R) C U,(x(t,xo;[to,to +TJ)),
where U(x(t, Xoi [to, to
+ TJ))
71) of the recurrent trajectory
(8.1)
is an (-neighbourhood of the arc x(t, xo; [to, to +

x( t, xo; R).
Intuitively, the oscillations described by system (1.1) must be related to
(8.1) by requiring that it should hold for the trajectories of the motion. This
is the case, in particular, for periodic, quasi-periodic and almost periodic oscillations of system (1.1), the trajectories of which are recurrent.
We now explain the relation between recurrent motions and invariant sets
of system (1.1).
A set M C D is called a minimal set of system (1.1) if it is not empty,
closed, invariant and does not have a proper subset with these properties.
Using Baire's theorem on the power of a totally ordered sequence of distinct closed sets ordered by inclusion [94], one can prove by contradiction the
following theorem on a minimal set [94].
Theorem 1. Every invariant closed compact set contains a minimal set.

The following theorems due to Birkhoff [17] hold for trajectories of a minimal set.
Theorem 2. Every motion of system (1.1) which starts in a compact minimal

set is recurrent.
Theorem 3. If the recurrent motion of system (1.1) is compact, then the

closure of its trajectory is a compact minimal set.
Birkhoff's theorems can be proved by reductio ad absurdum: the assumption that the theorem does not hold necessarily contradicts the characteristic
97
property of a minimal set by virtue of which any trajectory of a minimal set is

everywhere dense in it and vice versa [94].
Suppose that the manifold M given by the equation
= f(.p),
.p E Tm ,
(8.2)
with the function f(.p) E C' (Tm ) defining a local homeomorphism from the
torus Tm onto its image f(Tm )
the property that
f : Tm
relation
= M,
is an invariant set of system (1.1). Here,
M be a local homeomorphism is ensured by the
of(.p) = m,
rank ~
for s
1 and by the assumption that
.p E Tm
(8.3)
f satisfies conditions of Lemma 5.3 for
s = O.
As follows from the results of 2.5, the existence of such a manifold for
system (1.1) is a necessary condition for its quasi-periodic oscillations with
a frequency basis containing m frequencies to exist. We associate the multifrequency oscillations of system (1.1) with the manifold M by defining each
oscillation to be a recurrent motion the trajectory of which belongs to the minimal set M. The existence of minimal sets on M follows because the manifold
M is compact and closed, and Birkhoff's theorems imply that the motions that
start at the points of minimal sets are recurrent and they belong to a "certain
family" of similar motions each of which "winds round" the minimal set.
With such a definition of a multi-frequency oscillation the number m charactrizes how many frequencies there are by giving an upper bound m for the
dimension of a frequency basis for any quasi-periodic motion of system (1) contained in multi-frequency oscillations. Periodic oscillations whose trajectories
form M as well as stationary states of system (1.1), the points of which lie in M
and form there a closed set with connected components joined by separatrices
of system (1.1) in M are considered as a "single frequency" oscillation.
The behaviour of "double frequency" oscillations corresponding to the
value m
= 2 is determined
for s
2 by the Poincare-Denjoy theory [100]
which gives a qualitative analysis of trajectories of a dynamical system on a

two-dimensional torus T2 . For m
= 2 and s ~ 2 the trajectory flows of system
(1.1) on M defined by the system of equations (6.12) is a smooth dynamical

system on
'Tz.
98
CHAPTER 2
It follows from the results of Poincare-Denjoy that the minimal sets on
M are formed either by the trajectories of the quasi-periodic motions with a
double frequency basis or by the trajectories of periodic motions or stationary

states. In this case every trajectory of a quasi-periodic motion "sweeps" the
manifold M, and the trajectories of periodic motions are either limit cycles or
cover all of M.
The properties of multi-frequency oscillations of system (1.1) for m
are not well studied. It is known from Pugh's closure lemma [102] that a
recurrent motion of smooth system (1.1) can be turned into a periodic one by
a small change of the right hand side of the system, and from the results in [5],
[89], [103] it follows that if the right hand side of system (6.12) is sufficiently
smooth and close to a constant, then quasi-periodic oscillations are "typical"
of the multi-frequency oscillations of system (1.1).
Chapter 3. Some problems of the linear theory

3.1. Introductory remarks and definitions
We shall be considering a system of equations of the form
d>
di
dx
= a(, di = P(x + /(,

> = (>l,' .. ,>m), x = (Xl, ... ,xn ).
(1.1)
where a,P,! E Cr(Tm ),

Underlining its
linearity with respect to the variable x we shall call it a linear non-homogeneous
system of equations defined on the direct product of the m-dimensional torus
Tm and the Euclidean space En. An invariant manifold of system of equations

(1.1) of the form
x
where u E C (Tm
),
= u(,
> E Tm
(1.2)
will be called an m-dimensional s times continuously dif-
ferentiable invariant torus of system (1.1). For s = 0 an invariant manifold of

system (1.1) of the form (1.2) will be called an invariant torus of this system.
The invariant torus
x
= 0,
J E Tm
(1.3)
of the homogeneous system of equations
d>
di = a(J) ,
dx
dt
= P(x,
(1.4)
which is sometimes called a linear extension of a dynamical system on a torus

[27] will be called exponentially stable if for any solution J
Xt(>o, xo) of system (1.4) satisfying the condition

Jo(>o) = Jo,
xo(>o,xo) = xo,
the inequality
99
= >t (Jo),
CHAPTER 3
100
holds for all t
T, arbitrary,po E Tm , Xo E En, T E R and some positive K"
that are independent of ,po, xo, T.

A system of equations (1.4) for which the invariant torus (1.3) is exponentially stable will be called exponentially stable.
An invariant torus (1.3) of the system of equations (1.4) will be called
exponentially dichotomous and the system of equations itself exponentially di-
chotomous whenever for any ,po E Tm the Euclidean space En can be represented as a direct sum of two complementary subspaces E+ and E- of the
dimensions rand n - r in such a way that any solution ,pt(,po), Xt(,po,xo) of
system (1.4) with Xo E E+ satisfies the inequality
and any solution ,pt(>o), Xt(,po, xo) of(1.4) with Xo E E- satisfies the inequality
for arbitrary
E R, ,po E T,n and some positive K", K 1 , "n independent of
>o,xo, T.
An invariant torus (1.2) of the non-homogeneous system of equati()n3 (1.1)
will be called exponentially stable (or dichotomous) whenever the lfivariant
torus (1.3) of the homogeneous system (1.4) corresponding to system (1.1) is
exponentially stable (or dichotomous).
We shall say that the system of equations (1.4) is block CS(Tm)-decomposable if there exists a non-singular matrix 4> E C'(Tm ) such that the change of
variables
x = 4>(,p)y
transforms this system into the system
d,p
di = a(,
dy
dt
= Q(y,
(1.5)
where the matrix Q is block-diagonal:
Here the property of being block CO(Tm)-decomposable means that systems

(1.4) and (1.5) are topologically equivalent in the sense that
SOME PROBLEMS OF THE LINEAR THEORY
for all t E R, 4>0 E Tm , Xo E En, Yo
101
= (f>-l(4>O)xO'
The system of equations (1.1) is the simplest among the systems obtained
by taking a linearization of the equation of motion of a dynamical system in the
Euclidean space, considered in a neighbourhood of a smooth torus in a local
coordinate system. This explains the deep relation between these systems with
the theory of multi-frequency oscillations, the elements of which are considered
here.
Some problems that we shall be dealing with in this chapter are finding
existence conditions and methods of construction for the invariant tori of system
(1.1), studying their exponential stability and dichotomous properties, finding

the degree of smoothness, and finding conditions for the system (1.4) to be
block C (Tm )-decomposable.
3.2. Adjoint system of equations.
Necessary conditions for the existence of an invariant torus

We shall be considering system of equations (1.1) under the assumption that
a E cr+l(Tm ), P E Cr(Tm ), r 2: O. We write down the homogeneous system
of equations
d4>
dt
=a(4)),
dx = P(4))x
dt
(2.1)
corresponding to (1.1) and use it to construct the operator L

HO(Tm ), by setting
au
Lu=Eav(4))o4>v +b(4))u,
HI(Tm ) - .
(2.2)
v=l
where (al(4, ... ,a m (4))) = a(),
b(4)) = -P().
The operator L is related to the formal adjoint operator L : Hl(Tm )
-.
HO(Tm ) defined by the expression
Lu
= - E av() o;v + (b() m
p.(4)))u,
v-I
where b (4)) is the matrix adjoint to b( 4 and
A.)
P. ('I'
= ~ oa v(4)) =
LJ o4>v
v=l
tr
oa(4))
04> .
(2.3)
102
CHAPTER 3
Lemma 1. For any u, v E H 1(Tm
the following equality holds:
(Lu, v)o
=(u, L*v)o.
(2.4)
Indeed, according to the definition of the operator L,
(Lu, v)o
f21r
f2 1r
= (21r)m io ""lo
m
au()
[?;(a,,() a" ,v(
+ (b()u(),
v( ]d,
where (u, v) = L~=1 UiVi is the usual inner product of vectors u, v considered
as elements of n-dimensional Euclidean space. Since
au
L: (a" a" ,v) + (bu, v) = ,,=1

L: a" (a"u, v) ,,=1
m
p(u, v}-
-( u, fa" : ; ) + (u,b*v)
,,=1
=f
,,=1
a: (a"u,v)
"
+ (u,L*v),
(2.5)
it follows from (2.5) that
(Lu, v)o
= (u, Lv)o.
We write the system of ordinary differential equations
dx
dt = -a(), dt
= [P*() + p()]x
(2.6)
in terms of L in such a way that the operator L can be written using system
(2.6) in the same way as the operator L is written using system (2.1). We call
system (2.6) the adjoint of system (2.1).
We denote by G'(Tm ) the subspace of C(Tm ) consisting of all those functions u for which u(t(q,o has continuous derivatives with respect to t such
that
dU(t(q,o = u(q,t(o
dt
for all t E R, 410 E Tm and some function u(</ E C(Tm).

Lemma 2. Let wE C'(T,n). Then
f21r
o
f271
""lo
w(</> )#1 ... dq,m
f21r
f21r
=- i o ""lo
p()w( </#1 ... dq,m'
(2.7)
103
Proof. Since the function dw(ljJt(IjJ/dt is continuous with respect to t,ljJ, it

follows that
2..
dIet)
(2.8)
...
W(IjJ)dljJl" .dljJm =
o
0
t t=O
12.. .
-d-I '
where I(t) denotes the integral
= Jro ""lor
27r
I(t)
27r
w(ljJt(</J)dljJl" .dljJm.
Consider this integral. We make a change of variables by introducing the

variables t/J as follows:
The Jacobian of the transformation IjJ
t/J is the function
-->
Since the matrix fN-t(t/J)/81/J satisfies both the matrix equation
dz
dt
and the condition &t.-;,t
.,..
theorem that
t=O
=-
= E,
J(t,1/J)
Jot
= exp{ -
tr
8a(ljJ_t(1/J
81jJ
it follows from the Ostrogradskii-Liouville
= det 8</J-t(t/J) =
81/J
8a(1jJ (1/J
;;
dt} = exp{ -
Jot
Jl(IjJ-t(1/Jdr}.
Thus we have the following expression for I:
1=
",:,.
"':
r
...
r
w(1/J)IJ(t, 1/J)ldt/Jl .. .d1/Jm =
J",~ J",~
",:,.
1/1:
=l~ l~
1
t
w( 1/J) ex p { -
Jl( <P-T (t/J dr}d1/Jl ... dt/Jm,
where 1/JJ, 1/J] (j = 1, ... , m) are the new limits of integration.
104
CHAPTER 3
Taking into account the fact that Jt(J) - J is periodic with respect to Jv
with period 211" for
the equalities
II
= 1, ... ,m, we obtain via the change of variable formulae

tP] = 211' + tPJ,
j = 1,.,., m.
Since the functions under the integral sign are periodic, the above equalities
yield the following expression for the integral 1:
(2.9)
Differentiating equality (2.9) with respect to t we find that
11
dl
1. = -d
= t t=o
...
11'
11'
w(tP)tt(1/J)d1/Jl,. .d1/Jm.
This relation together with (2.8) leads to the identity (2.7).

We now define the action of the operators Land L * on functions u E
C'(Tm ) by setting
= u(if + b(J)u(if,
L*u(J) = -u(J) + [b*(J) Lu(J)
(2.10)
tt(J)u(J).
It is clear that Land L * defined by relation (2.10) are the same as the operators
given by equalities (2.2) and (2.3) when u E C'(Tm ) n H 1 (Tm
).
Lemma 2 enables us to prove that these operators are adjoint in the space
C'(Tm ). In fact we have the following statement.

Lemma 3. For any functions u, v E C'(Tm
equality (2.4) holds.
To prove the lemma we use the identity
(dU( ~~(if)
+ b(Jt(J ))u(Jt(J)), v(Jt(J))) =
= ~ (u( Jt(J , v(Jt (if> )-( u(Jt (J, dv( ~~( 4>)) )+(u(Jt(J, b* (4)t (4)) )v(Jt( 4)
to obtain
(Lu,v)o=
i '''i
o
11'
11'
W(ifd4>l ... d4>m-(u,V)o+(u,b*v)o.
105
= (u(t/J),v(t/J)}. Since the function w as well as the functions c and

v belong to the space G'(Tm ), according to Lemma 2 we have
where w(t/J)
{k {k
Jo .. '}o W(t/J)dt/Jl'" dt/Jm
fW J.l(t/J)W(t/J)dt/Jl ... dt/Jm =
{k
=- Jo
.. '}o
= -(J.l(t/J)u(t/J) , v(t/Jo.
But then
(Lu,v)o
= (u,-ti + b*v -
jlv)o
= (u, L*v)o,
as required.
Lemma 4. Suppose that the system of equations (1.1) has an invariant torus
t/JETm .
x=u(t/J),
(2.11)
Then u E G'(Tm ) and Lu(t/J) f(t/J).

Indeed, if equality (2.11) defines an invariant torus of system (1.1) then
1.1 E G(Tm ) and
(2.12)
for all t E R, t/Jo E Tm . Because the right hand side of identity (2.12) is
continuous and periodic in t/J, the function u belongs to the space G'(Tm ) and
the equality obtained from (2.12) for t
for all t/Jo E Tm
= 0 implies the identity Lu(t/Jo) = f(t/Jo)
Using Lemmas 3 and 4 we formulate necessary conditions for the existence

of an invariant torus of system (1.1) in the form of the following statement.
Theorem 1. For system of equations (1.1) to have an invariant torus it is

necessary and sufficient that the equality
(f,v)o=O
holds for any function v that defines the invariant torus x
the adjoint system of equations (2.6).
Proof. Let
x=u(t/J),
t/JETm
(2.13)
= v(t/J) (t/J E Tm) of
106
CHAPTER 3
be an invariant torus of the system of equations (1.1). By Lemma 4, u E C'(Tm )

and Lu(4)) = 1(4))
Because the function v belongs to the space C'(Tm ) and satisfies the identity L*v(4))
= by virtue of the same lemma,
we obtain from Lemma 3 the
chain of equalities:
(J,v)o
= (Lu,v)o = (u,L*v)o = (u,O)o = 0,
which leads to relation (2.13).

The example of the system of equations
d4>
-=w
where
= (WI,'"
dt
,W m )
dx
dt =1(4)),
'
(2.14)
are constants, shows that the necessary condition for
existence of an invariant torus for system (1.1) given in Theorem 1 is not, in

general, a sufficient condition.
Indeed, for the homogeneous system of equations corresponding to (2.14)
the adjoint system has the form
d4>
dt =
-W,
dx
dt
= O.
Invariant tori for this system are given by the equations

x
= c, c = const.
The necessary condition in Theorem 1 is that the function f( 4 should

have the mean value equal to zero:
1
(21r)m
t"
[2.-
Jo ."Jo
f(c/J)dc/JI ... d4>m
= O.
(2.15)
The system of equations (2.14) has invariant tori only if the first integral of the
function f(wt) is quasi-periodic. But for this, as follows from Borel's theorem,
it is not sufficient that relation (2.15) holds.
3.3. Necessary conditions for the existence of an invariant torus of

a linear system with arbitrary non-homogeneity in C(Tm )
An invariant torus x = u( 1jJ), c/J E Tm , of the system of equations (1.1) will be
called degenerate if u( J)
'. 0 and
(3.1)
107
for any point
tPo
E Tm
An invariant torus x
rPo
a point
= u(), E T
m ,
=t 0 and there exists
for which u()
E T m such that (3.1) does not hold, is called a non-degenerate
invariant torus of the system of equations (1.1). The trivial invariant torus
of the homogeneous system of equations (3.1) will not be considered either as
degenerate or as non-degenerate.
Theorem 1. For the system of equations (1.1) to have an invariant torus

for an arbitrary function f E C(Tm
it is necessary that the homogeneous sys-
tem (2.6) does not have invariant non-degenerate tori and that adjoint system
of equations (2.1) should not have invariant tori other than the trivial one,
x
= 0,
E Tm .
Proof. Suppose that for an arbitrary function! in C(Tm
the system of equa-
tions (1.1) has an invariant torus (1.2). By Theorem 1 of 3.2, the relation
(I, v)o
holds, where the function v E C(Tm
system of equations (2.6). If v
dictory:
IIvll6 = 0 for v =t O.
= 0,
(3.2)
defines an invariant torus of the adjoint
=t 0 then
for!
= v the relation (3.2) is contra-
This contradiction shows that the adjoint system
of equations cannot have an invariant torus other than the trivial one.
Suppose that the homogeneous system (2.1) has a non-degenerate invariant
torus
x = uo(), E Tm
Let
x
= u;() ,
ETm
(3.3)
j=1, ... ,n,
be invariant tori of the system of equations (1.1) for

!=Uj-l,
j=1, ... ,n.
The following relation holds for these tori:
where n~() denotes the matricent of the linear system of equations
dx
dt = P(tPt())x,
(3.5)
108
CHAPTER 3
that is, the fundamental matrix of solutions of system (3.5) which is equal to
the identity matrix E when t
= O.
Indeed, since the torus (3.3) is invariant, it follows that
But then
UlI>t(4J
1n~(4J)uo(4JT(4JdT] =
= n~(4J)
+ 1n~(4J)ul(4JT(4JdT]
= nt(4J) [Ul(4J) +
nt(4J)[Ul(4J) + uo(4J)t],
U2(4Jt(4J
[U2(4J)
= n~(4J) [U2(4J) + Ul(4J)t + uo(4J) ~~].

Similarly one can prove (3.4) for j = 3, ... , n.
Using relations (3.4), we express the functions nM4J)uj(4J) in terms of the
Uj(4Jt(4J. We set
U = column(uo(4Jt(4J)), Ul(4Jt(4J), ... , Un (4Jt(4J))) ,
it
= column(n~(4J)uo(4J),
nt(4J)Ul(4J), ... ,n~(4J)un(4J
and rewrite relations (3.4) as the matrix equality

U
= Tit,
where the matrix T has the form
T=
tE
t 2E
2!
tE
00 Zktk
=eZt=L~'
k=O
.................................
tn
t n- 2
t n- 1
-E
E
E E
N!
(n-I)!
(n-2)!
where E is the n-dimensional identity matrix and
z=
o.
... E 0
109
Then
which is equivalent to the system of equalities
.t i
... +(-l)l"'7juo(4>t(4>)),
j=l, ... ,n.
J.
(3.6)
We use this system to prove that the torus (3.3) is degenerate. To do this,
taking into consideration the linear dependence of the vectors
we divide Tm into two subsets No and N 1 = T m \No by attributing to No all

the points 4> E Tm for which the vectors
(3.7)
are linearly independent. For a fixed point 4>0 in No one can find constants
00,01, ... ,on-l such that the vector u n(4)o) decomposes with respect to the
system of vectors (3.7) as:
Using relations (3.6) we obtain the representation

n~(4)o)un(4>O) = oouo(4)t(4>o))
+ 01 [U1(4)t(4>0) -
tuo(4)t(4>o))] + ...
tn - 1
... + 0n-l [un-1 (4)t( 4>0)) - tUn-2( 4>t (4)0)) + ... + (_1)n-1 (n _ I)! uo(4>t( 4>0))] ,
which together with equality (3.6) for j = n, = o leads to the identity
n 1
1 t [00 - t01 + ... + (-It- (n _ 1)!on-1] uo(4)t(o))+
tn -
+[01 - t02 + ... + (-It- 2(n
_ 2)!On-1]u1(t(4>0))
+ ...
tn
... + 0n-1 Un-l (t(4>o)) = (_l)n ,uo(4)t(O))+

n.
tn -
+(-It- 1 (n _
1)!u1(4)t(0)) + ... + (-1)tun-1(4>t(4>0)) + un(t(4>o)).

(3.8)
110
CHAPTER 3
Letting
It I tend
to
+00 in (3.8) we obtain the limit relation

lim l!uo(<pt(<Po))1! = O.
Itl-+oo
(3.9)
Thus, equality (3.9) holds for points <Po E No.

Let <Po E N 1 . This means that the functions (3.7) are linearly independent
for <P
= <Po
We choose those points <P E N 1 for which the system of vectors
uo(<p), U1(<P), ... ,Un -2(<P)

is linearly independent. For each such point <P
(3.10)
= <Po the vector Un -1(<PO) has a
decomposition in the vectors (3.10) similar to the decomposition of the vector
(<Po) in the vectors (3.7) for the point <Po E No. This leads to an identity
analogous to (3.8) and proves limit relation (3.9) for those points of N 1 for
Un
which vectors (3.10) are linearly independent. Continuing this reasoning we

see that relation (3.9) holds for all those points <Po E T m for which at least one
of the systems of vectors
uo(<Po), U1(<PO), ... ,Uj(<Po),
= 1, ... , n -
1,
(3.11)
is linearly independent.
Let <Po be a point for which all systems of vectors (3.11) are linearly dependent. Then there exist constants
If 0'1
f.
0'0,0'1
such that 0'6 + ai
f. 0 and
0 then U1 (<Po) = - ~; uo(<Po) and from equalities (3.6) for j = 1 we
obtain
- 0'0
0'1
uo(<pt(<Po))
= U1(<Pt(<PO)) -
On passing to the limit in the last equality as
tuo(<pt(<Po)),
It I - +00
we obtain relation
(3.9). This proves limit relation (3.9) for all points <Po except those for which
uo(<Po)
O.
Let N be the set of such points. It follows from the existence and unique-
ness of a solution of the Cauchy problem for system of equations (3.5) that a
solution of this system x(t) satisfies the identity x(t)
x(O)
= O.
Thus if <Po E N, then x
= 0 for t
E R whenever
= u(<pt(<Po)) = 0 for all t E R.
limit relation (3.9) for all points <Po EN.
This proves
111
Consequently, (3.9) holds for all points o E Tm . Therefore the torus (3.3)
turns out to be degenerate, which contradicts our assumption. This contradiction completes the proof of the theorem.
The following example of the system of equations
d
dt
=0
dx
dt = P()x
'
+ !(),
det P() = const
(3.12)
shows that the necessary conditions for existence of an invariant torus given by
Theorem 1 may also be sufficient. Indeed, an invariant torus of system (3.12)
for an arbitrary function! E C(Tm ) exists if and only if
det P()
# 0,
E Tm .
(3.13)
The necessary conditions given by Theorem 1 also reduce to the requirement

that (3.13) should hold.
Consequently, they are the same as the sufficient
condition for the existence of an invariant torus of system (3.12) for an arbitrary
function! E C(Tm ).
The example of the system of equations
d! = Sill,!"
A-
dx
d! = (cos)x + !()
(3.14)
shows the extent to which the necessary conditions are "best possible" by proving that it is impossible to replace them by the requirement that the homogeneous system have no invariant tori. Indeed, the homogeneous system of
equations corresponding to the system of equations (3.14) has an invariant
torus given by the equation
x
= sin ,
E 7i.
Since
where
c=
{~an(/2),
the torus (3.15) is degenerate:
=o, =1r,
o < < 1r, 1r < < 21r,
(3.15)
CHAPTER 3
112
for every 4> E
7i. Consequently, the homogeneous system corresponding to
system (3.14) has a degenerate invariant torus.

We show that in spite of this, the non-homogeneous system of equations
(3.14) has an invariant torus for any function
E C(Tt}. To do this, we
consider the function u defined for 4> E [0,211") by the relation
- 1(0),
sin 4> 1
u(4)) =
4>
tP
~(t),
1f/2 sm
.,p
I( 11"),
= 0,
o < 4> < 11",

4>=11",
. 4> 1</> 1(1j;) d1/J

. 2
'
3"/2 sm 1/J
sm
11"
< 4> < 211",
and extended to all 4> by periodicity. Since

.
.
1</> 1(1/J)
_.
J;/2(f(1/J)/sin
hm sm
- ' - 2 - d t/J hm
. -1
</>-+00
,,/2 sm .,p
</>_+0
sm 4>
lim I( 4/ sin 4>

</>-+0 - cos / sin 2 4>
.
1(4))
</>-1f0 - cos,/,
= hm --",=/(11"),
.,p)d1/J_
-
=_1(0),
k=21,
lim sin r</> ~(~')d1/J= lim

1(4)),,,=-1(211")=-1(0),
</>-2,,-0
J3"/2 sm 1/J
</>-2,,-0 - cos '/'
the function u is continuous for E (-00,00). Consequently, u belongs to the
space C(T1 ).
Now we have
so that x
= u(4)t(O)) and x = u(t(1I")) are solutions of the equation

dx
dt = (COSt())x + f(4)t())
for eP
= 0 and eP =
1r.
The function x
113
= u(ePt(eP)) also satisfies this equation for
eP E (0,11") and eP E (11",211") since, by (3.16),

du( ePt(eP)) _ du oeP _
A.. (A..) . A.. (A..)
dt
- deP
cos 'f't 'f' sm 'f't 'f' X
at -
From what has been said it follows that the equation

x
= u(eP),
eP E 1i,
defines an invariant torus of the system of equations (3.17) for an arbitrary

function f E C(TI).
Finally we note that the set N of zeros of the function u defining an
invariant torus of the homogeneous system of equations (1.4) is an invariant
set of the system
d4>
di = a(4)).
(3.17)
Since N is closed and in view of the relation
which follows from (3.1), the set N contains all non-wandering points of system (3.17). Thus the homogeneous system of equations (1.4) can have nondegenerate invariant tori only if the set of wandering points of system (3.17) is
not empty in Tm . In particular, it follows that the homogeneous quasi-periodic
system of equations-system (1.4) cannot have degenerate invariant tori when
a
= w, w = const.
3.4. The Green's function. Sufficient conditions for

the existence of an invariant torus
As before, let
n~ (4))
system of equations
denote the fundamental matrix of the solutions of the
dx
dt = P( <Pt( 4>))x,
(4.1)
CHAPTER 3
114
which takes the value of the identity matrix E for t
T.
Because the functions a, P belong to the space C r (Tm ) it immediately

follows that ifJt (ifJ) - ifJ and n~ (ifJ) also belong to the space C r (Tm ) for all t, T E R
and any r 2: 1. We claim that
(4.2)
for any t, T, (J E R. To prove this, we replace ifJ by ifJlJ(ifJ) in the identity defining
n~(ifJ):
and taking into account the group property of the function ifJt(ifJ) expressed by
the relation
we obtain
dn~ (:tlJ (ifJ = P( ifJt+lJ(ifJ n~ (ifJlJ (ifJ.
The last identity holds for arbitrary t, T, (J E R, ifJ E Tm and means that
n~(ifJlJ(ifJ
is the fundamental matrix of the solutions of the system of equations
dx
dt
= P(ifJt+o(ifJx,
(4.3)
which is equal to the identity matrix E for t = T. The matrix n~~~ (ifJ) also has
the same property. This is possible only if the matrices n~ (ifJlJ (ifJ and n~~~ (ifJ)
are the same, and this proves the required equality (4.2).
Let C(ifJ) be a matrix belonging to the space C(Tm ). We set
T:S 0,
T>O
(4.4)
and call G o( T, ifJ) a Green's function for the system of equations
difJ = a(ifJ)
dt
dx = P(ifJ)x
dt
(4.5)
in the case when the integral f~oo IIGo( T, ifJ )lIdT is uniformly bounded with
respect to ifJ:
I:
IIGo( T, ifJ)lldT :S I< < 00.
(4.6)
115
We give the simplest properties of the Green's function (4.4). It follows

from its definition that Go(1', </ E C(Tm ) for each
l'
and
Go(O,</ - Go(+O,</ = E.
Let Gt ( 1',1 be the matrix defined by the relation
l'
t( ,</ -
t ;::: 1',
{n~(</C(</>,.(</),
-n~(</[E - C(</>,.(</)],
(4.7)
< 1'.
Then
(4.8)
This follows immediately from the property of the matrix n~ (</ expressed by
equality (4.2)
l'
{n~(</>t(</)C(</>,.(>t(</>))),
-n~(1)t(</>))[E - C(>,.(</>t(1>)))],
o( , </>t(1))) -
n~+t(C(>T+t(</>)),
1'~O
1'>0
1'+t~t
- { -n~+t(</>)[E - C(>,.+t(</>))],
1'+t>t
=
= Gt (1'+t,1.
It follows from equality (4.8) that the matrix
t ;::: 0,
t <0
consists of solutions of the homogeneous system (4.1) considered for t ;::: 0 and
t < 0 respectively.
Let I E C(Tm ). Consider the integral
i:
(4.9)
GO(1', >)I(1),.(</>))d1'.
I:
This integral is majorized by a convergent integral, as follows from the estimate
Ill:
GO(1',<//(</>T(</>))d1'11
We set
u(
and show that
E C(Tm ).
I:
IIGo(1',</lI d1'
Go(T,<//(</>,.(</>))d1'
1/10.
116
CHAPTER 3
Indeed, let
Ui(tP)
i
i Go(T,tP)f(tPT(tPdT,
= 1,2, ....
The sequence of functions Ui(tP) obviously belongs to the space C(Tm

satisfies the estimate
IIU(tP) - ui(</I)II
[1: IIG
:s
o(T, </I)lIdT +
and
JOO IIGo(T, </I)lI dT ]Iflo.
The uniform convergence of integral (4.6) with respect to tP ensures that the
limit relation
.lim Ui(tP) = u(tP)
'-00
is uniform with respect to </I E Tm . This means that the sequence Ui(tP) (i =
1,2, ...) converges with respect to the C(Tm
space,
E C(Tm ).
1:
norm. Since C(Tm ) is a complete
Integral (4.9) defines an operator T on the function space C(Tm
Tf(tP)
Go(T,</I)f(tPT(tPdT.
):
(4.10)
We show that the set
= u(tP) = Tf(tP),
tP E Tm ,
(4.11)
defines an invariant torus of the system of equations
dx
dtP
dt = a(tP), dt = P(tP)x + f(</I).

For this purpose we consider the function
x(t,tP)
1:
= u(</It(</I.
According to equalities (4.8) and (4.10) we have the representation
U(tPt(tP)
1:
GO(T,tPt(tP)f(tPT(tPl(</I)))dT =
GI(T + t,tP)f(tPT+!(</IdT
1:
Gt(T,</I)f(tPT(tPdT
(4.12)
117
for u(>t(>)). Differentiating it with respect to t, we find that
dU(~~() = C(>t()f(>t() + (E +P(tPt(tP
i:
C(>t(>)))J(tPt(tP))+
Gt(r, tP)f(tPr(tP))dr = P(>t(>))u(tPt(>)) + f(tPt()
for any t E R. This shows that u E C'(Tm ) and the set (4.11) is an invariant
torus of the system of equations (4.12).
Thus, if there exists a Green's function for system of equations (4.5), then
there is an invariant torus of the system of equations (4.12) for an arbitrary
function f E C(Tm ).
It should be noted that in the above argument we have used the fact that
the functions a, P belong to the space C T (Tm ) only to prove that the matrix
n~(
is continuous with respect to >. Thus, everything said remains true for
a E CLip(Tm ),
P E C(Tm ). This leads to the following statement which gives
sufficient conditions for the existence of an invariant torus of system (4.12).

Theorem 1. Suppose that the right hand side of the system of equations (4.5)
satisfies the conditions a E CLp(Tm ), P E CT(Tm ), r ~ 0, and that the system

itself has a Green's function Go(r,tP). Then for any f E CT(Tm ) the system of
equations (4.12) has an invariant torus defined by relation (4.11) and satisfying
the estimate
julo
K
= ITfio ::::; Klflo
= .pET
max
m
00
-00
IIGo(r,lIdr.
We now give an example showing that the invariant torus of Theorem 1

may only be continuous even in the case when the functions a, P and fare
analytic with respect to >.
Consider the system of equations from [86]:
dJ = _ sin Adx
di
dt
'f',
-X+Slll'f'.
A-
By definition (4.4), the Green's function Go(r,tP) with the matrix C(tP) == 1
has the form:
118
CHAPTER 3
The invariant torus (4.11) is defined by the expression
where
sinT()
={
0,
2e T tan(/2)
2T
e + tan 2(/2)'
= br, k = 0,1,2, ... ,
Calculation of the function u leads to the expression
0,
{
u() = tan(/2) lnsin2(/2),
>
= br
>:f:. br
which shows that the function u does not have a finite derivative for
2
due~
_ I'1m (InSin (/2)
I1m
- </>-0 d
</>-0 2cos 2(/2)
= 0:
+ 1)- -00.
3.5. Conditions for the existence of an exponentially

stable invariant torus
Suppose that the matrix
nbC~
for the system of equations (4.1) satisfies the
inequality:
(5.1)
for all E Tm and some positive
J{
and / independent of . We show that in
this case the function
Go(r, )
= {n~(),
0,
r ::; 0,
r>O
(5.2)
is a Green's function for the system (4.5) 5.2) is obtained from (4.4) by setting
C() == E in (4.4)).
Indeed, it follows from inequality (5.1) that
Thus, we have for function (5.2) the inequality
119
which implies estimate (4.6) for the integral of IIGo(T, > )11. The function (5.2)
is therefore a Green's function for the system of equations (4.5).
The set
x = u( = T!(,
> E Tm ,
(5.3)
where T is the operator (4.10), is an invariant torus of system (4.12). We show

that this torus is exponentially stable.
We denote the general solution of the system of equations (4.1) by x =

x( t, >, xo) = 0b(> )xo. Using the properties of the matrix 0b( for this solution
we obtain the estimate
IIx(t,>,xo)1I = 1I0~(xoll = 1I0~(Oo(xoll =
1I0~-T+r(x(T,>,xo)1I ~ 1I0~-T(>r(>))lIl1x(T,>,xo)1I S;
S;
which holds for all t 2:
J( e-"Y(t-T)lIx( T,
>, xo)11.
(5.4)
and arbitrary > E Tm . The estimate (5.4) is sufficient
for the invariant torus (5.3) to be exponentially stable.

Consequently, if the matrix OM satisfies inequality (5.1) then the system
of equations (4.12) has an exponentially stable invariant torus which is given
by the relation
and satisfies the estimate

(5.6)
We introduce an index which enables us to check whether inequality (5.1)

holds for the matrix OM of the system of equations (4.1). We denote by 9lo
the set of n x n positive"definite matrices S( belonging to the space C'(Tm ).
As before we denote by
8(
the matrix
lim dS(>t() = 8(.
t-O
dt
We set
inf max S(P(
.E9t o 11"11=1
+ ~8(>))x,x) < -(3(
(S( x, x)
120
CHAPTER 3
and define the number 130 as

130
= 4>ET
inf
f3(tfJ).
(5.7)
Lemma 1. For any J.l > 0 one can find [{ = K(J.l) > 0 such that the solution
x(t, tfJ, xo) = OM tfJ )xo o[ the system o[ equations (4.1) satisfies the inequality
[or all t
T and arbitrary tfJ E Tm .
It follows from the lemma that inequality (5.1) always holds when
130> o.
(5.8)
We now turn to the proof of the lemma. If the infimum is attained on the
set 9l 0 , then there exists a matrix S(tfJ) E 91 0 such that
max (S(tfJ)P(tfJ) + ~S(tfJ))x,x} < -f3(tfJ).
IIxll=l
(S( tfJ)x, x)
But then
/
dx(t,tfJ,xo)
)
\S(tfJt(tfJ))
dt
' x(t,tfJ,xo)
= (S(tfJt(tfJ))P(tfJt(tfJ))x(t,tfJ,xo),
x(t,tfJ,xo)),
which leads to the inequality
~ (S( tfJt(tfJ ))x(t, tfJ, xo), x(t, tfJ, x o)) ~

~
-2f3(tfJt (tfJ) )(S(tfJt(tfJ ))x(t, tfJ, xo), x(t, tfJ, xo)).
Taking the integral of the last inequality, we obtain
~ exp {
for all t
-21
(S(tfJt(tfJ))x(t,tfJ,xo),x(t,tfJ,xo))
t
f3(tfJ$(tfJ))ds } (S(tfJr(tfJ))X(T, tfJ,xo), X(T,tfJ, xo)}
(5.9)
~ T.
Since the matrix S( tfJ) is periodic and positive definite, we have the estimate:
Ko(x,x) ~ (S(tfJ)x, x) ~ [{O(x,x)
for some positive
[{o, /{o
121
which are independent of > E Tm , x E En.
Therefore inequality (5.9) leads to the estimate
for t
~ T,
where
whence the required estimate follows:
[{2
= [{o/K o does not depend on >, xo, T.
If the infimum is not attained on the set 9't o, then there exists a sequence
of matrices Sv(, II 1,2, ... ,Sv(J) E 9't o, such that
+ tSv()x, x} < -{3(J).

(Sv(x, x)
-
lim max ((Sv(P(
v....oo IIxll=!
Then for an arbitrary p. > 0 there exists a matrix S JJ ( E 9'to such that
+ !SIJ(J))x, x} < -f3(

(SIJ(x, x)
-
max ((SIJ(P(J)
IIxll=l
+ p..
By replacing SIJ(J) by S( and f3( - p. by f3(J) in the above reasoning,

we obtain the required estimate for x(t, >, xo). This completes the proof of
Lemma 1.
Corollary. If inequality (5.8) holds then there exists an exponentially stable

invariant torus of system (4.12) for an arbitrary function f E Cr(Tm ).
We prove the converse. Suppose that the invariant torus x
= 0,
> E Tm
of the system of equations (4.5) is exponentially stable. Then
for arbitrary > E Tm ,
Xo
E En,
E R and some positive [{, r independent of
J, xo, T.
Since
estimate (5.1) for the matrix
n~(
follows from (5.11). We use this estimate
to prove inequality (5.8). Denote by S( the matrix defined by the expression

(5.12)
122
CHAPTER 3
We show that S E 91 0 and that
max S()P() + ~S(x,x} <_fj

11"11=1
(S()x,x)
-
where fj is a positive constant.

Since II(O(j(*11
= II0O()II,
inequality (5.1) ensures that the integral
(5.12) converges uniformly, so that the matrix S() belongs to the space C(Tm
).
Because the integrand matrix in formula (5.12) is symmetric, the matrix S()
is symmetric.
Now we have
S(t(
(O~(*
= faoo (O:+T(*O:+T()dT =
1 (O~(*O~()dTO~(),
00
(5.13)
which shows that the function S(t( is differentiable with respect to t for
t E R. This means that S E C'(Tm
).
We now show that the matrix S() is
positive definite. Using the inequality
where o ,(,0 are fixed values in the set Tm
(S()x, x)
II{II = 1, we write the estimate
~ (S(o){o,~o)llxW = (faoo (O~(o*O~(O)dT{O,{O
)llxW
= fa (O~(o){o, O~(o){o)dTllxW=
00
faoo 1I0Q(0){oWdTllxIl2 = Adlxll2,
where we have set
A1
(5.14)
= faoo 1I0~(0){01l2dT.
Since 0li(o){o is a solution of the homogeneous system of equations (4.1)

taking the value {o
i- 0 for
=0, it follows that
123
for all
E R. Thus Al > 0 and this shows that the matrix S(4)) is positive
definite. This proves that S E vto

We now find 5(4)). Taking the derivative of (5.13) with respect to t we
obtain
therefore
S( 4
is defined by the expression:
5(4))
= -P*(4))S(4>) -
S(4))P(4>) - E.
Now the equalities
(S(4P(4
+ t5(4)))x,x)
= t(S(4P(4
+ P*(4))S(4>)-
-P*(4))S(4>) - S(4))P(4>) - E)x, x) =
-tllxll2
together with inequalities (5.14) lead to the estimate:

max
IIxll=1
(S(4P(4 + tS(4)))x, x)
-lIxW
= max
=
(S(4x,x)
IIxll=12(S(4x,x)
-2" max (S(4x,x) - - 2A2'
II xll=1
where A2
Al
> O. Consequently
max (S(4P(4 + t S(4)))x,x) < -/3
IIxll=!
(S(x,x)
-,
where /3
(5.15)
= 1/(2A2) is a positive constant. Inequality (5.15) leads to the estimate

/30
/3
~ 0
for the index (5.7), as required.

Summing up, we have the following statement.
Theorem 1. Suppose that the right hand side of the system of equations (4.5)
satisfies the conditions: a E q,ip(Tm ), P E Cr(Tm ), r
O. Then the system of
equations (4.12) has an exponentially stable invariant torus for any

if and only if the index (5.7) is positive.
E Cr(Tm )
124
CHAPTER 3
In practice, to see whether
Po
is positive one chooses S in !X O so that the
derivative along the solutions of the system of equations (4.5) of the form
v = {5(4))x,x}
is negative definite. Since this derivative has the form
v = 2{(5(4P(t/J) + ~5'(t/Jx,x} = (S(t/J)x, x),

where S(t/J) is the symmetric matrix given by
5(t/J)
it follows that
matrix.
V is
= 5(t/J)P(t/J) + ?*(4))S(4>) + 5'(4)),
a negative definite form when 5(4)) is a positive definite
Using Sylvester's criterion [39] to determine whether a symmetric
matrix is positive definite we conclude that
Po
is positive whenever there exists
a symmetric matrix 5(4)) E C'(Tm ) such that its principal minors and the
principal minors of the matrix -5(4)) corresponding to it are positive for all
t/J E Tm
These conditions hold, in particular, when the principal minors of the
symmetric constant matrix 5 and the matrix -5(t/J)
= -(5P(4 + P*(4))5) are
positive for all 4> E Tm

3.6. Uniqueness conditions for the Green's function and
the properties of this function
It follows from the necessary conditions for the existence of an invariant torus of
the system of equations (4.12) for an arbitrary function
f E cr (Tm ) and The-
orem 4.1 that a necessary condition for the existence of a Green's function for
the homogeneous system of equations (4.5) is that there be no non-degenerate
invariant tori of this system. The conditions under which the system of equations (4.5) that has degenerate invariant tori would have a Green's function are
not clear at the present time. However, the peculiar properties of the latter due
to the existence of degenerate invariant tori of system (4.5) have been studied
to some extent. We consider some of them. First we show that the absence
of degenerate tori of system (4.5) ensures that the Green's function for this
system is unique.
125
Theorem 1. Let a E CLip(Tm ), P E C(Tm ) and suppose that the system of
equations (4.5) has a Green's function G o( T, Q). Then the torus x = 0, 4> E Tm
is a unique invariant torus of system (4.5) if and only if this system does not
have any Green's functions other than GO(T,4.
Suppose that the system of equations (4.5) has a Green's function
Prool.
G 1 (T, 4 besides the function G o( T, 4. We set
where
is an arbitrary function in C(Tm ).
It is clear that the set
defines an invariant torus of the system of equations (4.5). We prove that
(6.2)
for some function
E C(Tm
).
It follows from the definition of the Green's function and formula (6.1)
that
where R is a non-zero matrix in C(Tm ). To prove inequality (6.2) it is sufficient

to show that an operator To on the function space C(Tm ) defined by the relation
T ol(4)) = [ : G 1(T, </J)/(</JT(</JdT - [ : GO(T, 4/(</JT(4>))dT =
=[: n~(</J)R(</JT(</J/(</JT(4>))dT
(6.3)
is not the zero operator on this space.

Suppose this is not the case, that is,
Tol == 0 fIl E C(Tm ).
(6.4)
We show that (6.4) implies the equality
(6.5)
126
CHAPTER 3
thus contradicting the assumption that G o( r, J)
G 1 (r, J) which implies that
R(J) " 0, J E Tm .
Turning to the proof of equality (6.5), we first consider the values of the
matrix R at the points of trajectories of system (4.5) that are periodic on
Tm . Let <Po be such a point, J
Jt+T(JO)
<Pt(Jo) mod 211", t E R, the
corresponding trajectory, T its period. It follows from the Floquet-Lyapunov
theory that
where ct> is a periodic non-singular matrix with period T and A is a constant

matrix. But then
so that
+00
= pJ;ool
e-ApTe-ATct>-l(r)R(JT(Jo!(JT(Jo))dr =
+00
=L
p=-oo
(e-AT)p Bj,
where
Bj = iT e- AT ct>-l (r)R(JT(JO!(JT (Jo))dr.
Since the series I::~oo(e-AT)p is divergent for any matrix A, it follows

that To!(Jo)
= 0 only if
(6.6)
Consider equality (6.6). We distinguish the cases when Jo is a stationary
point of the dynamical system on Tm and when Jo is not such a point. In the
first case equality (6.6) holds for any T> O. This is possible only if
127
But then
and, taking into account the identity tPT(tPO) =: tPo and the fact that the function
is arbitrary, we see that this can hold only if
R(tPo)
= o.
(6.7)
In the second case there exists a number T > 0, the real period of the
periodic trajectory on the torus, such that
for any 0
<
< T.
Then there is one-to-one correspondence between a point
tPt(tPo) on the torus Tm and a point tPt
= vt, v = 21r/T, on the circle 'Ft.
In
view of this correspondence, for any function F E C(TI) we can find a function
f E C(Tm ) such that

f(tPt(tPo))
= F(vt),
t E R.
In this case equality (6.6) becomes

(6.8)
Since equality (6.8) holds for an arbitrary function Fin C(1'i), this is possible
only if
Because the matrices e- AT and /fl-l(T) are non-singular, this leads to the identity
R(tPT(tPO)) == 0,
E [O,T].
(6.9)
Indeed, we set
F(vt)
rij (t)
= {O, ... ,Fj(vt), ... ,O},
= {e- AT ~-1 (T)R( <PT (<Po)) lij
(i,i
= 1, ... , n)
and rewrite equality (6.8) for the chosen function F(vt). We have:
(6.10)
CHAPTER 3
128
In (6.10) we replace Fj by the function in C(71 ) defined for 1/J E [0,211"] by the
equality
where
fij
is the complex conjugate of
rij,
and
/0
is a function in C(1i) equal
to zero for 1/J = 0 and positive for 1/J = (0,211'):
/0(0)
= /0(211') = 0,
/0(1/J) > 0 when 1/J E (0,271').
As a result we obtain the equality
This is possible only if
which is equivalent to
rjj(T)
Since
rjj
=0,
(6.11)
TE[O,T].
is an arbitrary element of the matrix e- AT cI>-1(T)R(r(0, identity
(6.11) is equivalent to (6.9). This proves (6.7) for any point o that belongs to
a trajectory of system (4.5) that is periodic on 7m .
We now consider the values of the matrix R() at the other points of the
torus 7 m . Let o be such a point, and = t( o), t E R, the trajectory
corresponding to it. Because the integral
is uniformly bounded, the integral
is also uniformly bounded. But then one can find a sequence of positive numbers
E"n,
monotonically decreasing to zero
E"n+1
< E"n, n-oo

lim E"n
= 0 and a sequence of
= +00
positive numbers t n monotonically increasing to +00, t n+1 > tn, lim t n

n-+oo
such that the following inequality holds:
129
But since T ol(4)o)
= 0, it follows that
II Itt: n~(4)o)R(4>r(4>o))I(4>r(4>o))drllS
s
Ill: n~(4)o)R(4>r(4>o))f(4>r(4>o))drll+
l:n IIn~(4)o)R(4>r(4>o))lIdrlllo+
+
The arc 4>
1:
00
IIn~(4)o)R(4>r(4>o))lIdrllloS (nl/lo
= 4>t(4)o), -tn S t S t n of the trajectory on Tm does not
intersect itself, therefore for any continuous function F(t), defined for t E R,
one can find a function In E C(Tm ) related to F as follows:
Suppose that R(4)t(4>o))

for t E R.
Let rij(t)
n~(4)o)R(4>t(4>o)) satisfying
We set
rij(t)
=t
0 for t E R. Then n~(4)o) x R(t(o))
the inequality rij(t)
= {r~j(t!,
slgnrij(t),
=t
be an element of the matrix
{n~(4)o)R(4>t(o))};j
=t 0, t E R.
Irij(t)1 S 1,
Irij(t)l> l.
It is obvious that the function rij(t) is continuous for t E Rand %1tx 17\j(t)1 S l.
We choose scalar functions In E C(Tm ) such that the conditions
hold and consider inequality (6.12) for the functions In (4))ej , where ej
{O, ... , 1, ... , O} is the jth unit vector. We have:
Il-tn
tn
rij(r)7\j(r)drl
Since rij (t)rij (t)
=t
= ltn
-tn
rij(r)rij(r)dr S (n
for t E R, it follows that
(6.13)
CHAPTER 3
130
for some sufficiently large N. But then, since the function rij(t)rij(t) is nonnegative, it follows that
Vn 2: N,
which contradicts inequality (6.13) and the condition
lim
n_oo
En
= O.
This contradiction derived from the assumption that
proves that
n~(rf>o)R(t(o)) == 0,
t E R.
(6.14)
For any point o E Tm not belonging to a trajectory of system (4.5) that is

periodic on Tm , relation (6.7) follows from equality (6.14).
The above arguments in combination prove equality (6.5). But this, as
has been noted, contradicts the assumption that the Green's function is not
unique. Thus if the Green's function is not unique it will necessarily imply
that there exists an invariant torus of system (4.5) different from the trivial
one x
= 0,
E Tm . The "only if" part of Theorem 1 is proved. We now prove
the reverse implication. To do this, we suppose that uniqueness of the Green's

function does not imply the uniqueness of an invariant torus of system (4.5).
First we establish two important equalities that follow from the assumption
that the Green's function is unique.
Lemma 1. Suppose that the function Go( r, ) is the only Green's function for
the system of equations (4.5). Then the matrix C( )
= Go(O, )
satisfies the
equalities
C(rf>t())
C 2 ()
= n~()C(rfn~()
= C()
V E Tm
Vt E R, rf> E Tm
(6.15)
(6.16)
Indeed, suppose that the first relation does not hold, that is, for some t 1
C(rf>t, ()) ~ n~l(rfC()n~, (),
E Tm .
i- 0
131
We set
and consider the function
1:
1:
1:
For this function we have the chain of inequalities
~ I< +
IIG o(T,)lIdT +
IIn~()R(T())dT ~
IIn~+lt ()C(T+d)) - n~()C(T())n~1 (T())lI dT ~
1: IIQ~+ll
1: IIQ~(4))C(T())lIdTIQ~,lo
~ K+
+
1:
IIG 1 (T,)lldT
()C(T+II (4)))lIdT+
xQ~, (T())lI dT ~ 2K +
+
1~ IIn~()[E -
1~ IIQ~+11 ()[E-C(4>T+l
())]lIdT+
1~ IIQ~+t,() - n~()C(T())X
[1:' IIQ~(4))C(4>T())lldT+
C(T())J1l dT] IQ~llo
~ 2!{ + f{IQ~llo + [11~, IIQ~()C(T())lIdTI+
+11-
11
IIQ~(4))[E -
c(4)T(4>))lI dT I]
IQ~llo ~
from which it follows that
uniformly with respect to E Tm
But then the function G 1 (T,) satisfies
all the conditions necessary for it to be a Green's function for the system of
equations (4.5). This contradicts the assumption that the Green's function is
unique. The contradiction proves equality (6.15).
132
CHAPTER 3
Now let
Then the function
satisfies the chain of inequalities
i : IIG 1 (r, )lIdr:::; [{ + iOoo
1 Iln~()[E
IIn~(4))C(r())lIdrIC
00
- C(r(]dTIClo:::; K[1
- Elo+
+ max(IC -
Elo, IClo)]
= K1 ,
which shows that it is also a Green's function for the system of equations (4.5).
The latter contradicts the assumption that the Green's function is unique and
thus proves equality (6.16).
We now use equality (6.15) to complete the proof of Theorem 1. Suppose
that the Green's function GO(T,) is unique, but that the torus
(6.17)
is an invariant torus of system (4.5) different from the trivial one. Then
(6.18)
Consider the function Ul(4)) E C(Tm ) defining the invariant torus
of the system of equations
d4>
-;It
= a(4)),
dx
-;It
= P(4))x + uo(4)).
(6.19)
Since the torus (6.17) of the homogeneous system of equations corresponding to the system (6.19) is invariant, it follows that uo(4)t(4>))
= nh(4))uo(4>).
Therefore
oo
Ul(4)) = i : Go(T,4uO(4>r(4>))dT = i :
Go(T,4n~(4uo(4dT=
=i~ n~(4))C(4>r(4>))n(j(4>)UO(4>)dT-
-1
00
n~()[E -
C(r(4>))]fl(j()uo()dT,
(6.20)
133
which together with (6.15) leads to the relation

(6.21)
This is possible only if
= 0, C(</uo(</ = uo(</
But then uo(</ = 0, V</> E Tm , which contradicts inequality
C(</uo(</
simultaneously.
(6.18).
The main properties of the Green's function that follow from the assumption that it is unique are defined by relations (6.15) and (6.16). The first
means that the matrix G(</ belongs to the space G'(Tm ) and is a solution of
the matrix equation
= P(</G -
GP(</,
while the second means that the matrix G( </ is a projection.

We shall show that equality (6.15) is a characteristic property for a Green's
function to be unique, namely, if the system of equations (4.5) has more than
one Green's function then none of them satisfies equality (6.15). We formulate
this in the form of the following statement.
Lemma 2. Let a E GLip(Tm ), P E G(Tm
and suppose that the system of

equations (4.5) has an invariant torus different from the trivial one x = 0,
</> E Tm Then none of the Green's functions for system (4.5) satisfies relation
)
(6.15).
Thrning to the proof of the lemma we suppose that the torus (6.17) is an
invariant torus of system (4.5) different from a trivial one. Let G o(T, </ be a
Green's function for system (4.5) satisfying relation (6.15). Then the system
of equations (6.19) has an invariant torus defined by the function (6.20) which,
according to equality (6.21), can only be equal to zero:
But then Ul( </ == 0, </> E Tm , and it follows from the second equation of (6.19)
that uo() == 0, </> E Tm
But this is a contradiction since the torus (6.17)
is trivial. The contradiction proves that the function G o( T, ) cannot satisfy

relation (6.15).
CHAPTER 3
134
Finally we note that the condition that there be no invariant tori of system
(4.5) is sufficient for the Green's function of this system to be unique. Therefore
the quasi-periodic system of equations, system (4.5) with a(</
= w = const can
have only one Green's function such that relations (6.15), (6.16) hold.
3.7. Separatrix manifolds. Decomposition of a linear system

Consider the system of equations (4.5):
d</>
dt = a(</,
dx = P(</x
dt
(7.1)
under the assumption that a E CLip(Tm ), P E C(Tm ) and that this system does
not have an invariant torus other than the trivial one x = 0, </> E Tm . Suppose
that there exists a Green's function G o( T, </ for the system of equations (7.1).
These assumptions introduce certain properties of the behaviour of trajectories of system (7.1). We now discuss this behaviour.
First we note that the matrix C(</ = Go(O,</ satisfying equality (6.16)
is a projection for all </> E Tm . It follows from this that all the eigenvalues of
the matrix C( </ equal 1 or O. Since the matrix C( </ is continuous in </>, it
follows that its rank does not depend on </> and is equal to the multiplicity of
1 considered as an eigenvalue of the matrix C(</. Thus the Jordan form of
the matrix C( </ is the matrix E l
= diag{ E, O}, where E is the r x r identity
matrix. We denote by 5( </ the matrix converting C( </ to Jordan form:
Let M+ be a set of the phase space of system (7.1) defined as the set of
points (</>, x) satisfying the equation
C(</x
= x,
(7.2)
and M- the analogous set of points (</>, x) satisfying the equation
C(</x
= O.
(7.3)
We now discuss the geometric structure of the set M+. For each fixed </>,
equation (7.2) defines in En a plane Et spanned by the vectors 51 (</, ... ,5r (</
135
which are the first r columns of the matrix S(t/J). Since the matrix C(t/J) is
periodic, it follows that
E '"+ -E+
"'mod 2..
Because the matrix S( t/J) is non-singular for all t/J E Tm it follows that Et is an
r-dimensional plane. We define on M+ a neighbourhood system by taking a
c-neighbourhood of a point Po
= {t/Jo, xo}
E M+ to be the intersection of the
set M+ with a e5-neighbourhood of the point Po in Tm x En,
Since the Jordan form of the matrix C(t/J) does not depend on t/J and is
C-continuous with respect to t/J, it follows from a result in [18] that the matrix
S( t/J) converting C( t/J) to Jordan form E 1 can be taken to be continuous with
respect to t/J for t/J E U6(t/Jo), where U6(t/JO) is understood as a c-neighbourhood

of the point t/Jo on the torus Tm .
Having chosen S(t/J) to be continuous, we make a change of variables in
the system of equations (7.2) by setting
= S(t/J)y.
(7.4)
As a result we obtain the system of equations

(7.5)
which is equivalent to the system
Yj
= 0, j = r + 1, ... ,n.
Hence it follows that the last n - r coordinates of the vector S-1 (t/J)x are equal
to zero for any point P
= {t/J, x} E U6(P
O/
M+). In view of all this, the change of
variables (7.4) defines a continuous one-to-one mapping S of the neighbourhood
U6(Po/M+) onto a neighbourhood of the point S(Po) E Tm x E r , where E r is

the r-dimensional Euclidean space defined in En by the system (7.5).
Let SU6(Po/M+) be the image of the neighbourhhood U6(Po/M+) under
the map S, and Q a point in SU6(Po/M+). We define its coordinates (t/J,c)
using the coordinates (t/J, x) of the inverse image P
= S-IQ by setting:
136
CHAPTER 3
We take a point PI
= {4>1' Xd
E M+ close to Po so that a neighbourhood
U6(PI/M+) of it has a non-empty intersection with U6(Po/M+):
Let SI (4)) be a continuous matrix in 4> for 4> E U6 (4)1) converting C( 4 to Jordan
form, and SI the map from the neighbourhood U6(PI/M) into Tm x E r induced
by this matrix. Every point P
= {4>, x} in the set D has the coordinates (4) , c)
defined by the map S and the coordinates (4), cd defined by the map SI' To
find out the relation between the coordinates c and
Cl
we prove the following
lemma.
Lemma 1. Let S( 4 and Sl (4)) be matrices transforming the projection matrix

c(4)) to the Jordan form El. Then
(7.6)
where D l (4)) and D 2 ( 4 are non-singular matrices of dimensions rand n - r
respective/yo
To prove the lemma we split the matrix St l (4) )S( 4 into blocks
and write down the equality relating Sl, C and E l
According to this equality,
so that (using the form of the matrix
This is possible only if D 12

l
Ed
we have the relation
= 0 and D = O.
21
Consequently the matrix
St (4))S(4>) has a quasi-diagonal form. Since St (4))S(4>) is non-singular, the

diagonal blocks of diag{D l (4)), D 2 (4))} are also non-singular. This completes
the proof of the lemma.
Using the lemma we express the relation between the coordinates
137
Cl
and
c. By definition,
[~] = S-l(J)x, [~] = Sll(J)x.

Thus it follows from equality (7.6) that
This is equivalent to the equality

(7.7)
Since the matrices Sll(J) and S(J) are continuous for J E V, the matrix D 1 (J)
is also continuous for J E V, where V = U6(JO) n U6(ifJI).
Relation (7.7) shows that the local coordinates of the point P considered
as. a point of U6(P/M+) can be expressed using a continuous and invertible
change of variables in terms of the local coordinates of this point considered
as a point of U6(Po/M+). This suffices for M+ to be a topological manifold of
dimension m
+ r.
To emphasize that a small neighbourhood of any point of the manifold
M+ is homeomorphic to some neighbourhood of the manifold Tm x E r we shall

say that the manifold M+ is locally homeomorphic to the product Tm x E r .
The set M- has a similar structure. To prove this we only need to write
equation (7.3) using the matrix C1(cf = E - C(cf in the form (7.2) and apply
the above reasoning to it.
We now show that the manifolds M+ and M- are invariant sets of the system of equations (7.1). Suppose that (ifJo,xo) E M+. Let ifJt(Jo),Xt(ifJo,xo) =
nh(ifJo)xo, t E R be a trajectory of system (7.1) that passes through the point

(ifJo,xo) at t = O. The condition (4)o,xo) E M+ is equivalent to
C(4)o)Xo
= Xo
Thus, to prove that the manifold M+ is invariant it is sufficient to prove the

equality analogous to the one given above for any point (4)t(4>o),Xt(4>o,xo)) for
any t E R.
Taking into consideration relation (6.15) we have
C(cf>t(cf>oXt(4>o,xo)
= C(4)t(cf>o))n~(4>o)xo =
= nh(4)o)C(4>o)xo = Xt(4)o,xo)
\It E R.
138
CHAPTER 3
This means that (</It(</lo), Xt( </lo, xo E M+ for any t E R. Thus we have proved
that the set M+ is invariant.
The proof that M- is invariant is similar.
Theorem 1. Under the assumptions made above on the system of equations
(7.1), the sets M+ and M- are invariant manifolds of this system that are
locally homeomorphic to the products Tm x Er and Tm x E n - r respectively.
We give some further properties of the manifolds M+ and M-.

First we note that
(7.8)
The first relation states that the intersection of the manifolds M+ and M- is
the torus
(7.9)
while the second asserts that their direct sum is the whole phase space of the
dynamical system (7.1).
The first relation of(7.8) follows because the inclusion (</l,x) E M+nMnecessarily implies equality (7.9) under the assumption that (7.2) and (7.3)
hold simultaneously. The second follows since V</l E Tm the planes Et and E;
by defining the decomposition of En into the direct sum of subspaces E r = Et
and E n - r = E; uniquely define the decomposition of Tm x En into a direct

sum of M+ and M- .
Let us find out the properties as It I ~ +00 of the motions of system (7.1)
that start on the manifolds M+ and M- respectively.
Suppose that the Green's function Go(r,</l) is damped with respect to r
as Irl
+00, in other words,

(7.10)
where d(t) is a positive function that is monotonically decreasing to zero as

t ~ +00. The motions that start on the manifold M+ at t = 0 are damped
as t ~ +00 and the motions that start on the manifold M- are damped as
t ~ -00 and have the same damping character as the function Go( r, </l):
IIXt(</l,X)1I S; d(t - r)llx T (</l,x)lI,
t 2: r, (,x) E M+,
IIxt(,x)lIS;d(r-t)llxT(,x)'I,
tS;r, (,x)EM-.
(7.11)
139
Indeed, suppose that inequality (7.10) holds and (, x) E M+. The chain
of relations
IIxl(,x)1I
= IIQh()xll = IIQ~()C()xll = IIn~()n(j(4C(4xll =
= IIn~(4))C(T(4>))n~(4>)xll = IIn~(4))C(4>T())xT(4>,x)1I =
= IIn~_I(I(4)C(4>T-I(t()))xT(4>, x)11 ~
~
IIG o(T - t,t())lllIx T(,x)1I
which hold for t 2:
T,
d(t - T)llx T(,x)lI,
proves the first inequality of (7.11).
The similar chain of relations, which hold for t
~ T
and (4), x) E M-,
proves the second inequality of (7.11).

Now let (4)o,xo)
f/. M+ U M-.
We claim that a motion 1(4)0), XI(O,XO)
that starts at the point (4)0, xo) leaves any neighbourhood of torus (7.9) both
for t > 0 and for t <
o.
Indeed, let
x~
where C 1 (4))
Ilx~11
=E
= C(4)o)xo,
x~
= C 1 (0)xo,
- C(). Then (o,x~) E M+, (o,x~) E M-, IIx~1I
0,
- O.
In fact, since the matrices C( 4 and C1 (4)) are projections, it follows that
= C 2 (0)xo = C(o)xo = x~,

C(4)o)x~ = C(4)0)C1 (0)xo = 0,
C(o)x~
which proves that the point
(o,x~)
belongs to the manifold M+ and the point
(o, x~) to the manifold M-. Now we have the decomposition
Xo
= x~ + x~,
which proves the inequalities Ilx~1I

(o,xo) f/. M+ and (o,xo) f/. M-.
0, IIx~1I
(7.12)
0 since we assumed that
Consider the function Xt(4)o,xo). According to decomposition (7.12), we

have
For t > 0 this leads to the estimate
CHAPTER 3
140
which proves that lim IIXt(<Po, xo)1I =

For t
<
+00.
we have the similar estimate

1-+00
which proves that lim IIXt (<Po, Xo )11 = +00.

..... -00
Combining the obtained estimates we have the needed relation
Since the manifolds M+ and M- are invariant and in view of the behaviour
of the trajectories of system (7.1), which start on these manifolds and outside of
them, the manifolds M+ and M- are called sepamtrix manifolds of the system
of equations (7.1). The above discussion can be summarized as follows.
Theorem 2. Suppose that the system of equations (7.1) satisfies the conditions:
1)
a E CLip(Tm
),
P E C(Tm );
0, <P E Tm is a unique invariant torus of this
2) the trivial torus x
system;
3) there is a Green's function G o( T, <p) for this system satisfying inequality
(7.10).
Then equations (7.2) and (7.3) define separatrix manifolds of system (7.1)
on which the damping properties of the motions are determined by relations
(7.11) as Itl--+ +00.

We now consider the question of the C S -block decomposability of system
(7.1) by relating this problem to the question of global homeomorphism between
the manifolds M+, M- and the products Tm x E r , Tm x E n - r respectively.
Suppose that such a homeomorphism exists, so that the matrix S( <p) defining it belongs to the space CS(Tm ) n C'(Tm ).
We subdivide a vector Y into two blocks Yl and Y2 of dimensions rand
n - r respectively: Y = (Yl; Y2) and write the equation for the manifold M+ in
terms of the variables Y
S-l(<p)X. Using the change of variables formulae,
the equation for M+ takes the form
Y2
= 0,
</>
E Tm
(7.13)
Since the change of coordinates from x to Y is made by means of a "global"

change of variables, the invariance of the manifold M+ is not violated.
141
The transformed system of equations is the system
d<p
di = a(<p),
dV
dt
= Q(<p)V,
(7.14)
where the matrix Q( <p) is defined by the expression

(7.15)
and thus the coordinate plane (7.13) is invariant.
But this is possible only if the block Q21 (<p) of the block decomposition of
the matrix Q(<p), which corresponds to the decomposition of V into VI and V2:
IS
zero:
The manifold M- is given in terms of the variables (VI, V2) by the equation:
VI=O,
<PETm .
(7.16)
But then the plane (7.16) is also invariant with respect to system (7.14). This
necessarily implies the equality
From what has been said one can see that the transformed system of
equations, system (7.14) has the block diagonal form:
The properties of the motions of system (7.1) do not change under the
transformation to the form (7.14); therefore the solution <Pt(<Po), Vi (t, <Po, V?),
V2(t, <Po, V~) of system (7.14) that takes the value <Po, V~, vg at t = 0 satisfies the
inequalities
IIvl (t, <Po, vr)1I $

IIv2(t, <Po, vg)1I $
J( d(t
J( d( T -
T)lIvl (T, <Po, vr)11, t ~ T,

t)lIv2( T, <Po. vg)11, t $ T,
142
CHAPTER 3
for any 4>0 E Tm
y~ E
Er, y~ E En-r,
E R, where
J{
is a positive constant.
Theorem 3. Suppose that conditions 1 and 2 of Theorem 2 hold and that

there exists a Green's function G o( T, 4 for system (7.1) such that the matrix
C(4))
= Go(O, 4
is reduced to Jordan form by the matrix

S(4)) E C'(Tm
n C'(Tm ).
Then the change of variables
x = 5(4))y
splits system (7.1) into CS(Tm ) blocks thus reducing it to system (7.14).
The problem whether the system of equations (7.1) can be split into
C' (Tm)-block form is reduced by Theorem 3 to the algebraic problem of the
membership of the eigenvectors of the projection C(4)) to the space CS(Tm ).
We shall return to a discussion of this point later. For the moment we merely
note that this problem is completely solved only for m
C(4)) that has rank r
=n -
= 1 and for
a matrix
1 [118].
3.8. Sufficient conditions for exponential dichotomy

of an invariant torus
To find out the conditions under which the trivial torus x
= 0,
4> E Tm , of
system (7.1) is exponentially dichotomous it is convenient to use the machinery

of sign-changing Lyapunov functions that are quadratic forms of the variable
x. This leads to the following result which gives necessary conditions for the
invariant torus to be exponentially dichotomous [117].
Theorem 1. Let a E CLip(Tm ), P E C(Tm ) and suppose that there exists a
non-singular symmetric matrix 5(4)) E C'(Tm ) such that the matrix
8(4))
= 5(4)) + 5(4))P(4>) + P*(4))5(4>)
is negative-definite for all 4> E T m . Then the trivial torus x
= 0,
4> E T m of
system (7.1) is exponentially dichotomous.
To prove the theorem, we assume that the matrix 5(4)) has r positive and
n - r negative eigenvalues.
143
Let
and let
J(f
be the set of points x of En satisfying the inequality
(SfX,X)
where 4J and t are fixed numbers, 4J E Tm
We set
J(+
0,
(8.1)
O.
J(,
,~o
and prove that
J(+
is an r-dimensionallinear subspace of En, which plays the
role of the space E+ in the definition of the dichotomy of the trivial torus for
system (7.1).
Using the expressions for matrices Sf and S( 4J) we find that
Thus for every x E En,
-# 0,
d(SfX,X)
dt
= (S(4J,(4Jn~(4J)x,n~(4J)x) =
= (S(4Jf( 4J))x,(4), x), x,(4J, x),

and, sin ce the matrix
S(4))
is negative definite, this leads to the inequality
(Sf x, x) < (STx, x)
Vt>
It follows from inequality (8.2) that the set

T
: J(,
T,
J{f
VT E R.
(8.2)
is contained in
J{ T
for t >
C K T and that the intersection of the boundary oK, of K, with the
boundary oKT of K T contains only the point zero: oK,
n oKT = {O}.
Thus, K+ is the intersection of sets I<f that are totally ordered by inclusion. We show that I<+ has r linearly independent vectors.
Since the matrix Sf is continuous with respect to t and non-singular, it
has the same number of positive eigenvalues as the matrix So
= S(4J).
ing symmetric, it is orthogonally similar to its Jordan form
which can be
represented in the form
~,
Be-
144
CHAPTER 3
where DI(t) = diagpI(t), ... ,Ar(t)}, D 2 (t) = diagPr+i(t), ... ,An(t)}, and
the Aj (t) are positive numbers.
But then
(8.3)
where O(t) is an orthogonal matrix, and E I and E 2 are the (r
r) and
(n - r) x (n - r) identity matrices respectively,
Under the linear coordinate transformation according to the formulae

x
= By,
detB:f 0,
the form (StX,x) will be evidently transformed into the form (StY,Y) with
matrix
By making an appropriate choice of the coordinate system in En we can transform the quadratic form with matrix So
So
= diag{ E I , -
= S( </
to the form with matrix
E 2 }. Without loss of generality we can suppose that the
coordinate system x has already been chosen in such a way that the quadratic
form (S(</x, x) can be written as
can suppose that
xi + ... + x; -
X;+l -
... -
x~. That is, we
We now consider the intersection of the set I<t and the plane Po which is
given by the equations
Xl
= 0, ... ,
Xj
= 1, ... ,
Xr
= O.
(8.4)
The set I<o n Po consists of the points satisfying the inequality
and so it is a closed ball ilio in an (n - r)-dimensional subspace of En which

is the plane Po.
The set I<t n Po consists of the points satisfying simultaneously inequality
(8.1) and equalities (8.4). We claim that the set I<t n Po is not empty. To see
this we write inequality (8.1) defining

x by the formula
Y
](t
145
in the coordinate system Y related to
= O(t)z,
where O(t) is the matrix from the representation of St in the form (8.3).
We have
(StZ,z)
= (~tO(t)z,O(t)x) = (~tY,Y) =
= A1(t)yi + ... + Ar(t)y; - Ar+1(t)Y;+1 - ... - An(t)y~ ~ 0,

or
Since the functions Aj(t) are positive, the values
Y1 = C1,, Yr
where
Cl, .. , Cr
= Cr, Yr+1 =
... = Yn = 0,
are arbitrary constants, satisfy inequality (8.5).
Thus the set
](t
contains the subspace E r (t) defined by the equation
= O(t) [~]
= column(c1' ... ,c
r )
or by the equation
01(t) ]
= [ 021(t)
c,
where 0 1 and 0 21 are the r x rand n x r matrices respectively formed by the

first r columns of the matrix O(t).
Using the inclusion
]{t
C [(o we find that the linear subspace Er(t) is
contained in the set
But then the matrix 0 1(t) is non-singular:
= 0 (which,
t- 0, entails the inequality 02l(t)C t- 0) does
since otherwise, a point of Er (t) defined by a non-zero C with 0 1 (t)c

in view of the relation O(t)[~l
not belong to the set
](0:
n-r
o ~ 2)021(t)C)J,
j=l
146
CHAPTER 3
which is impossible.
Because the matrix 01(t) is non-singular, there exist a vector Oll(t)ej,
where ej
= (0, ... , 1, ... ,0) is the r-dimensional unit vector, and a point
belonging to the intersection of the planes Po and Er(t) so that it also belongs
to the set I<t n Po. Thus the set I<t n Po is non-empty. We show that I<t n Po
either consists of a single point or it is homeomorphic to a closed ball in an
(n - r)-dimensional subspace of En, namely, the plane Po.

To do this we write inequality (8.2) for T = 0:
xi + ... + x; -
x;+l - ... - x~ ~ (5 t x, x), t > O.
(8.6)
If we write 5 t in the block form
and regard 51 as an r x r matrix, then using inequality (8.6) we see that the
matrix (52
+ 52)/2 is negative-definite:
= (xr+l, ... ,x
0/; O.
But then the matrix (52 + 52)/2 can be represented in the form
for each vector X2
n )
where O 2 is a non-singular matrix and the change of variables
reduces the quadratic form (52X2, X2) to the form
Setting x = (x 1, X2), Xl = (x 1, ... , X r ), we rewrite the inequality defining

K t in the form
147
Thus, the set K t
n Po consists of the points satisfying the inequality

(8.8)
In the coordinate system on Po chosen according to (8.7) inequality (8.8)

becomes
or
(8.9)
where {3l, ... , {3n-r,
f are functions of t depending on ej' Since the set
K t n Po
is non-empty, there is at least one value of Y2 satisfying inequality (8.9), and

this is possible only if f
Let
O.
= 0 for some t = to.
Then the set K to n Po consists of the single
point
Xl=ej,
Yr+v={3v,
v=I, ... ,n-r.
{3v={3v(to),
(8.10)
> to, it follows that (I{t n Po) C (K to n Po) when

t > to, so that the set K t n Po contains the point (8.10) and only this point
when t > to.
IC f > 0, then the set (8.9) is a closed ball in the space Po, so that K t n Po
Since K t C K to when t
is homeomorphic t.o a closed ball III t of an (n - r )-dimensional subspace of En,

namely, the plane Po.
Since III t C III T when t >
T,
the set
n III
1>0
consists either of a single
point (8.10) or is the intersection of a family of closed non-empty "balls" that

is totally ordered by inclusion and therefore contains at least one point
In both cases the set K+ n Po contains at least one point. By changing the
unit vector ej and setting j
= 1,2, ... , r
we obtain r points belonging to the
set K+.
Let
Xl,""
Xr
(8.11)
be these points. Since Xj = (ej, xL (j = 1, ... , r), it follows that the vectors
joining the origin and points (8.11) are linearly independent. Moreover, since
148
CHAPTER 3
the segments joining the origin and points (8.11) are formed by the points of
the form JlX i, 0 S Jl S 1, they belong to the set ]{+.
Let x be a point of ]{+. We prove that the function
Xt = n~(4))x
satisfies the inequality
(8.12)
where ]{ and I are positive constants not dependent on the choice of the point
4>.
It follows from the inclusion x E
(StX,x)
that
J{+
0 'lit> O.
This inequality together with inequality (8.2) implies that
'litER.
(StX,x)~O
To obtain estimate (8.12) we set
considering
to be a small positive constant. For an arbitrary function V.(t)
we then have the estimate
dV.(t)
= (S(4>t(4)Xt,Xt)
+ tP(4>t(4>)) + P (4)t(4>)))Xt,Xt) S
S -/dlxtl1 2 + tMllxtlI 2 = -(/1
tM)ll xtl1 2 = -/2I1 x tIl 2,
where
M=
max
4>ETm .ll r ll=l
1P)4
+ P*(4)))x,x)l,
12
= 11 - tM > O.
Since also
V.(t) S (StX, x)
= (S(4>t(4)Xt,Xt)
II x tW S
(St X , x)
+ tllXtW =
+ tllxtlI 2 S
+ tlixtlI 2
(M 1 + t)llxtlI 2 ,
= V.(t) ,
t
(8.13)
where M 1
149
= 4>ETmax
I(S(4x,x)1. we have
,II"'II=l
m
Upon integrating this inequality we obtain
or, by using inequalities (8.13),
IIXtIlZ::; V<~t)
::;
(V.~T)e-Z-r(t-r) ::; (1 + ~l )e-Z')'(t-r) x
xllxrW ::; I<ze-z-r(t-r)llxrW
'Vt
> T, 'VT E R.
Since M 1 and I are positive constants and do not depend on the choice of 4>,
the last inequality is the required inequality (8.12).
We denote the subspace of En spanned by the vectors (8.11) by Er. We
claim that E r C I<+.
For if not, then let
y= L:>jXj
j=l
be a linear combination of vectors (8.11) not belonging to I<+. It follows from
Y ft. K+ that one can find to > 0 such that
(StY, y) < 0 'Vt 2: to

But then using inequality (8.2) we find that
= const > 0
(-SIY'Y) > (-SloY'Y)
'Vi> to.
(8.14)
On the other hand,

r
(StY,Y) = LPj(StXj,Xj) = LPj(S(4>I(4>))Xt(Xj), Xt(Xj),

j=l
where Xt(Xj)
= nM4xj
j=l
satisfies estimate (8.12) because Xj E I<+ and thus

(8.15)
150
CHAPTER 3
Because the matrix S(>t(>)) is bounded for t > 0, by using relation (8.15)
we find that
lim (-StY, y) =
t-+oo
o.
This contradicts inequality (8.14). The contradiction proves that E r C j{+.

Since E r C j{+, the solution >t(, Xt(>,x) = nh(x of system (7.1)
with x E E r satisfies inequality (8.12) and thus E r can be considered as the
space E+ featuring in the definition of the exponential dichotomy of the trivial
torus for system (7.1).
A similar argument for the function -(StX, x) establishes without difficulty
that there exists an (n - r)-dimensional space E n the sets
Kt
in the intersection
J{-
of
formed by the points x E En for which
For each point x E
J{-
o.
(StX, x) :5 0,
t:5
the function Xt
= nb(x satisfies the inequality

(8.16)
This is true because the derivative of the function
satisfies the following inequality for small positive
f:
dV (t)
----Jt
= -(S(>t()Xt,Xt) + f((P(>t()+
+P*(>t(>)))Xt,Xt) ~ en - fM)IIXtI12 = "Y211 x tl1 2 ~
1
~ Ofi2
1>1+ f
v/(t)
= 2iV/(t)
"It E R,
which on integration yields the estimate
which in turn leads to inequality (8.16):
IIXtl12 :5 v/}t) :5
:5 (1 + ~1 )e2"(t-T)lIxTW
(V/ f(T)e 2"(t-T) :5
= I<2e2"(t-T)lIxTW
Vt:5 T, "IT E R.
151
Because of inequality (8.16), the space E n -
can be considered relative to
the solutions of system (7.1) as the space E- featuring in the definition of the
exponential dichotomy of the trivial torus for system (7.1).
Since (S(4x,x) > for points x E W, x =F and (S(4x,x) < for

points x E E n- r , x =F 0, it follows that the sets E r and En-r intersect in a
single point
n En -
The equalities
[(+
= Er,
= {OJ.
= En-r,
J(-
En
= [(+ EB [(-
now follow from
the equality En = E r EB E n - r , inequalities (8.12), (8.16) and the relation

lim
Itl-+oo
lI0h(4))xll
= +00,
rt. ErUEn-r
which can be obtained by applying an
argument similar to that used to obtain the relation

for (4)0, xo)
rt.
M+ U M- in 3.7.
Since the choice of 4> E Tm was arbitrary and [("
lim
Itl-+oo
IIxt(4)o, xo)11 = +00
do not depend on 4>, the
invariant torus x = 0, 4> E Tm of system (7.1) is exponentially dichotomous.

This completes the proof of the theorem.
Remark. It follows from the proof of the theorem that the manifold M+ 14>=const
lies in the "cone" in the space En defined by the inequality
(S(4x, x) > 0,
for every 4> E Tm , while the manifold
En, defined by the inequality
M-I...
Y'=cons
(8.17)
t
lies in the "cone" in the space
(S(4x, x) < 0.
3.9. Necessary conditions for an invariant torus
to be exponentially dichotomous
We show that the condit.ions given in the preceding section as sufficient are
also necessary for exponential dichotomy of the trivial torus x
= 0,
4> E Tm , of
system (7.1). Thus, we shall make the assumption that the torus x = 0, 4> E
Tm is an exponentially dichotomous invariant torus of system (7.1). For each

4> E Tm
, the space En can be decomposed into a direct sum of the spaces

E r (4)) and En-r(4)): En = Er(4))EBE n- r (4>) such that the solution Xt(4),xo) =
nh(4))xo of the system of equations
dx
di
= P(4)t(4>))x,
(9.1)
152
CHAPTER 3
taking the value Xo E E r (4)) at t = 0, satisfies the estimate
while the solution taking the value
Xo
E E n - r (4)) satisfies the estimate
where K, r are positive constants independent of 4> E Tm .

The above property of solutions of system (9.1) characterizes it as an
exponentially dichotomous system on the entire axis R = (-00,00).
Denote the projections corresponding to the decomposition of En into the
direct sum of the spaces E r (4)) and E n - r (4)) by C(4)) and C 1 (4))
=E -
C(4)).
Then the function

t
2:
T,
t<T
(9.2)
satisfies the estimate [42]

(9.3)
We use this function to determine the properties of the matrix C(4)).
Lemma 1. Let a E CLip(Tm ), P E CLip(Tm ) and suppose that the function
(9.2) satisfies (9.3). Then the matrix C( 4 belongs to the space C(Tm ).
To prove this, we consider the matrix
t 2: 0,
t < 0,
which equals the matrix C(4)) - C(~) at t
= O.
(9.4)
It is continuous with respect to
t, satisfies the inequality

(9.5)
and is a solution of the differential equation
dZ
dt = P(4)t(4>))Z + [P(4)t(4>)) - P(4)t(4>))]G t (O, 4.
(9.6)
153
Because system (9.1) is exponentially dichotomous, a solution of system

(9.6) that is bounded on R is unique and can be represented in the form of the
integral
Hence it follows that
We find an estimate for the integrand in formula (9.7). Let
lI a( for any
>, 4> E Tm
a(4))11 Sail> - 4>11, II P (

and some positive
S,BII> - 4>11
- P(4))1I
a and ,B.
One can easily derive from system of equations (7.1) the following estimates
4>11,
>t(4))11 S II> - 4>11 + 2l aloltl,
lI>t( - >t(4))11 S ea1tlll> -
lI>t( as a result of which we obtain
for all t E R and arbitrary v 2:
o.
Applying inequality (9.3) to the function Gt(T,, this leads to the estimate
II Zt(>,41I
s ,BKill> - 4>W/(v+l)
1:
ex p { -1(lt -
TI + ITI)+
+_a-ITI}(2IaloITI + II> v+1
4>lIy/(v+l)dT,
(9.8)
which implies that the integral (9.7) is convergent for a sufficiently large v:
v + 1 > af-y and that the inequality for Zo( >, 4
= C(
- C( 4
(9.8)
holds. Here M is a positive constant equal to the value of the integral for
t = 0 in inequality (9.8) and
II> - 4>11
= 21r-..fiTi. Inequality (9.9) proves that
CHAPTER 3
154
the matrix C(<jJ) is continuous with respect to <jJ for <jJ E Tm . It follows from
the definition of the matrix C( <jJ) that it is periodic with respect to <jJ. Hence,
C(<jJ) E C(Tm
),
as required.
We now consider the function

T
0,
T> 0,
(9.10)
where C( <jJ) is taken to be the matrix in formula (9.2). It can be seen from
formulae (9.2) and (9.10) that
and so the inequality

(9.11)
holds.
Because C(<jJ) E C(Tm ), the function (9.10) is a Green's function for system
(7.1). Because system (9.1) does not have bounded solutions on R other than
the trivial one, this function is unique, and the equality
TSO
T>O
holds, which proves that Go( T, <jJ) can be obtained from function (9.2) for t
= O.
This proves that the notations of (9.2) and (9.10) agree.

Thus,the exponential dichotomy of the trivial torus x = 0, <jJ E Tm , of
system (7.1) implies that a Green's function for this system exists and is unique,
and is exponentially damped as
ITI -+ +00, in
accordance with (9.11).
We shall use properties of the functions Gt ( T, <jJ) and G o(T, <jJ) to prove the
main statement of this section.
Theorem 1. Let a E CLip(Tm ), P E CLip(Tm ) and suppose that the trivial
torus of system of equations (7.1) is exponentially dichotomous. Then there
exists a non-singular symmetric matrix S(<jJ) E C'(Tm ) such that the matrix
S(<jJ)
= S(<jJ) + S(<jJ)P(<jJ) + P*(<jJ)S(<jJ)
is negative-definite for all <jJ E Tm
155
To prove the theorem we define the matrix

(9.12)
by setting
SI(IjJ)
100
C*(IjJ)(O(;(IjJ))"OQ(IjJ)C(IjJ)dT,
(9.13)
S2(1jJ)
= 100 C;(IjJ)(O(; (1jJ*O(; (IjJ)C1 (IjJ)dT.
Inequality (9.3) ensures that the integrals (9.4) converge uniformly. And this
leads to the inclusions S;(IjJ) E C(Tm
for i
= 1,2 because the matrices C(IjJ)
and QQ(IjJ) belong to the space C(Tm ) for every fixed
E R.
We show that S;(IjJ) E C'(Tm ). To do this we use the fact that the matrix
C(1jJ) satisfies equality (6.15) (this follows because the Green's function G o( T, 1jJ)
is unique) to transform the expression for SI(ljJt(IjJ) in the following way:
=
00
SI(ljJt(ljJ
1 (n~+T(IjJ)C(IjJ)O?(IjJ)rO~+T(IjJ)C(IjJ)O?(IjJ)dT
100
=
=
100 C*(ljJt(IjJ)(O:+T(IjJ)*O~+T(IjJ)C(ljJt(ljJdT =
(O?(IjJ*C' (1jJ)(O(; (1jJ*O(; (1jJ)C(IjJ)O?(IjJ)dT.
=
(9.14)
It follows from formula (9.14) that the function SI(ljJt(ljJ is continuously
differentiable with respect to t, therefore SI belongs to the space C'(Tm ). For

the derivative of SI(ljJt(IjJ we have the expression
One can similarly transform S2(IjJt(1jJ) to the form
CHAPTER 3
156
from which it follows that S2 E C'(Tm ) and
Equalities (9.15), (9.16) imply that the matrix S(J) is negative definite.
Indeed, since
S = S + SP + r
= Sl -
=-CC - rSl
=-CC - C;C
1 -
= -CC -
+ SP + r S =
SlP- C;C1 + PS2 + S2P +SP + rs =
P(Sl - S2) - (Sl - S2)P + p. S + SP =
S2
CiCl,
it follows that
which means that the matrix S( J) is negative definite for all J E Tm .

To complete the proof we need to show that the matrix S( J) is nonsingular for J E Tm . Taking into consideration the fact that C2 (J)
= C(J) and
IIC(J)II
IIn~(J)no(J)C(J)1I $ IIn~(J)1I x IIn(j(J)C(J))1I $ IIn~(J)IIf{e--YT
for T ~ 0, we have the following estimates for (Sl(J)X,X):
(Sl(J)X, x) =
$1
$
00
1<>0 IIn~(J)C(J)xI12dT
IIn~(J)C(J)1I2dTIIC(J)xIl2
~; 1IC(J)xW $
A2I1C(J)xIl 2,
[<>0
dT
(Sl(J)X, x) ~ Jo IIn~(J)1I211C(J)xIl2 ~
By combining these estimates we obtain the inequality
which holds for all x E En, J E Tm and some positive Al and A2.
157
In similar fashion we obtain an inequality for (S2(t/J)X,X) in the form:
which holds for all x E En, t/J E Tm and some positive ~3 and ~4'
Let T(t/J) be the matrix that reduces C(t/J) to diagonal form:
where E 1 is the r x r identity matrix.

By applying to inequalities (9.17) the change of variables according to the
formulae
x
= T(t/J)y
and taking into account the fact that
we obtain from inequalities (9.17) the new inequalities
A1I1 T - 1 11- 2I1JyI12 ~ AdlCTyW ~

~ (T" SlTy, y) ~ A211CTyW ~ A211TWIIJyll2,
from which it follows that the matrix T" S1 T has the block-diagonal form
T"S1T
= diag{D 1 ,O},
where D 1 is an r x r positive-definite matrix for any t/J E Tm .

Similar reasoning for the matrix S2( ) leads to the inequalities
from which it follows that the matrix T" S2T also has the block-diagonal form
where D 2 is an (n - r) x (n - r) positive-definite matrix for any t/J E Tm

The matrix S( ) = S1 () - S2( ) can then be represented as
CHAPTER 3
158
and has exactly r positive and n - r negative eigenvalues for each tjJ E Tm . This
proves that the matrix S(tjJ) is non-singular.
Remark. It is clear from what has been said that if the trivial torus of system (7.1) is dichotomous, then the matrix G(tjJ) defining the Green's function
Go(tjJ, T) and the matrix S(tjJ) defined by equalities (9.12) and (9.13) are related
by the property that any matrix T(tjJ) that reduces G(tjJ) to Jordan form can
be used to represent S(tjJ) in the form (9.18) with the dimensions of the blocks
D1(tjJ) and D 2 (tjJ) equal to rand n - r respectively.
We now combine the results of this and the previous two sections in the
form of a single assertion on the exponential dichotomy of the trivial torus of
system (7.1), making it a separate theorem.
Theorem 2. Let a E GLip(Tm) and P E GLip(Tm). The trivial torus of system
of equations (7.1) is exponentially dichotomous if and only if there exists a
non-singular symmetric matrix S(4)) E G'(Tm ) for which the matrix S(tjJ) =
S(tjJ) + S(tjJ)P(tjJ) + p(tjJ)S(tjJ) is negative-definite for all tjJ E Tm or if the
Green's function (9.10), (9.11) for system (7.1) exists and is unique.
It follows from Theorem 2 that the necessary and sufficient conditions given
there are equivalent for the trivial torus of system (7.1) to be exponentially
dichotomous.
3.10. Conditions for the G'(Tm)-block decomposability of
an exponentially dichotomous system

We shall be considering system (7.1) under the assumption that its trivial
torus is exponentially dichotomous. We shall give conditions under which such
a system is G'(Tm)-block decomposable. If the system in question is G'(Tm )block decomposable, then we shall require that the separatrix manifolds be
"straightened", that is, they can be turned into the subspaces E r and E n -
for any tjJ E Tm .

Theorem 1. For the exponentially dichotomous system of equations (7.1) to
be G'(Tm)-block decomposable it is necessary and sufficient that there exist a
non-singular symmetric matrix S(tjJ) E G'(Tm ) for which the matrix S(tjJ) =
+ S(tjJ)P(tjJ) + p(tjJ)S(tjJ)is negative definite for all tjJ E Tm and can be
S(tjJ)
159
= Q()diag{D 1(), -D 2()}Q"(),
(10.1)
represented as
5()
where Q() is a non-singular matrix in G'(Tm

ric positive-definite matrices for all E Tm
and D 1(), D 2(4)) are symmet-
Necessity. Suppose that the system of equations (7.1) is exponentially dichoto-
mous and G'(Tm)-block decomposable. Suppose that the change of variables

x = T()y, T E G'(Tm ),
reduces system (7.1) to the split form

(10.2)
where y
= (Y1,Y2),
Q1(4)) is an r
r matrix and Q2() an (n - r) x (n - r)
matrix. Here Qi(4)) (i = 1,2) are such that

IIn~(Qdll ~ Ke-"'I(t-T),
t 2:
T,
IIn~(Q2)1I ~ Ke"'l(t-T),
t ~
T,
(10.3)
where K and 'Yare positive constants independent of and r is an arbitrary

number in R. Because of inequalities (10.3), the system (10.2) has a Green's
function in the form
G (T ) _
o , -
{n~()3",
T~ 0,
-n~(4))(E - 3"),
where n~(4))
diag{n~(Q1),n~(Q2)}' 3"
identity matrix.
> 0,
= diag{E1>O}, and E
The initial system of equations (7.1) has a Green's function
the projection C() = Go(O,) of which admits the representation
where 3" is its Jordan form.
is the r x r
160
CHAPTER 3
According to the Remark of the previous section, the matrix S() constructed in the proof of Theorem 1 of 3.9 admits the representation (9.18):
with blocks D 1 () and D 2 ( ) of dimensions rand n - r respectively that are

positive-definite for all E Tm . By setting Q( ) = (T- 1 (
n we obtain for
S() representation (10.1).

Sufficiency. Suppose that there exists a matrix S( ) that satisfies the condi-
tions of Theorem 1. Then by Theorem 2 of 3.9 there exists a unique Green's

function Go(r, ) for system (7.1). Thus, the matrix C()
projection for any E Tm and belongs to the space C'(Tm
).
= Go(O, )
is a
This is sufficient
to reduce the C'(Tm)-block decomposability problem to one of converting the

matrix C( ) to Jordan form
C'(Tm
= diag{ E 1 , O} using a matrix T( ) in the space
).
Since C( ) is a projection, the columns of the matrix T( ) reducing C( )

to Jordan form are composed of linearly independent solutions of the equations
C()x
= 0,
(lOA)
C()x
=x
(10.5)
defining the eigenvectors of the matrix C( ).

Equation (lOA) is the same as equation (7.3) which defines the separatrix
manifold M- and equation (10.5) is the same as equation (7.2) which defines
the separatrix manifold M+ of system (7.1). According to the Remark in 3.8,
the non-zero solutions of equation (lOA) satisfy inequality
(S()x, x) < 0,
(10.6)
while the non-zero solutions of equation (10.5) satisfy the inequality
(S()x, x) > O.
(10.7)
Finally, a maximal family of linearly independent solutions of equations (lOA)

and (10.5) forms a basis in En for any E Tm .
We now show that a maximal family of linearly independent solutions of
equations (lOA) and (10.5) can be chosen so that they belong to the space
C'(Tm ).
161
To prove this we make a change of variables in equations (10.4) and (10.5)

by setting
Y = Q*(J)x.
(10.8)
Multiplying the result by Q*(J) we have, instead of equations (10.4) and

(10.5), the equations
A(J)y
= 0,
(A(4 - E)y
where A(J)
= Q*(J)C(J)(Q*(J))-l
(10.9)
= 0,
(10.10)
is a matrix in C'(Tm ) with rank rand E
is the n x n identity matrix.

After the transformation (10.8), inequalities (10.6) and (10.7) become
(D1(J)YI,YI) < (D 2(4Y2,Y2),

(D1(J)YI,YI) > (D 2(J)Y2,Y2),
where (YI,Y2)
Let
(10.11)
= y.
A(4))
= [A 1(4))
A 21 (J)
A I2 (J)]
A 2(J)
be the decomposition of A( 4 into blocks corresponding to the decomposition

of Y into YI and Y2' We write system (10.9) in the form of two equations
+ A 12 (J)Y2 = 0,
A 2J (J)YI + A2(4))Y2 = O.
Al (4) )YI
(10.12)
We claim that the rank of the matrix

(10.13)
is equal to r for any J E Tm . For suppose, on the contrary, that the rank of
matrix (10.13) is equal to rJ
< r for some value J E Tm . Then system (10.12)
has a non-zero solution of the form

YJ = y~(J),
Y2 = O.
This solution satisfies inequality (10.11); therefore
162
CHAPTER 3
which is impossible since the matrix D I ( is positive-definite.

Thus the rank of matrix (10.13) is equal to r. An orthogonalization process
enables us to represent matrix (10.13) in the form
where &( is a matrix with orthogonal columns, T I ( is an r x r non-singular

matrix, and &( and T I ( belong to the space C'(Tm ).
We transform equation (10.12) by setting
As a result we obtain the equations
+ AdY2 = 0,
&2I(ZI + A 2(Y2 = O.
&I(ZI
(10.14)
Since the rank of the coefficient matrix of system (10.14) is equal to that
. &(, the columns of [AI2(] are

of A( and hence to that of the matrIX
A2(
linear combinations of the columns of &(. Because the columns of &( are
orthogonal, we can write the coefficients of the decomposition of columns of the
matrix
[AA
12
( ]
2 (
with respect to columns of &( explicitly and we see that
these coefficients are also elements of the space C'(Tm ). The representation of
. [Ad].In t he rwrm
t he matrIX
A 2 (
where R I ( is an r X (n - r) matrix belonging to C'(Tm ), follows from the

above remarks. We now write system (10.14) in the form
and, on multiplying the left hand side of this equality by the matrix
we obtain the following system of equations equivalent to (10.14):
163
It is clear from this that the system of functions
in the space C'(Tm ), where ej is the (n - r)-dimensional unit vector, forms

a system of linearly independent solutions. A system of linearly independent
solutions of equation (lOA) is formed by the system of functions in C'(Tm )
defined by the equalities
(10.15)
where Yj(4J) is the n-dimensional vector
This proves that the system of equations (lOA) has n - r linearly independent
solutions in C'(Tm ).
Applying similar reasoning to equation (10.10) we see that the rank of the
matrix
Ad4J) ]
[ A 2 (4J) - E 2
'
where E2 is the (n - r) x (n - r) identity matrix, is equal to n - r for any
4J E Tm
This is sufficient to transform system of equations (10.10) to an equivalent
system of equations
and to prove that equation (10.5) has r linearly independent solutions in the
space C'(Tm ). These solutions are the functions
where Yv(4J) is an n-dimensional vector of the form

(10.16)
e v is the r-dimensional unit vector.

This proves that the system of equations (10.5) has r linearly independent
solutions in C'(Tm ).
164
CHAPTER 3
The matrix whose columns are formed by the solutions (10.15), (10.16)
transforms the matrix C(4J) to the Jordan
form~.
It belongs to the space
C'(Tm ) and, by Theorem 3 of 3.7, splits the system (7.1) into C'(Tm)-blocks.
The theorem is proved.
Remark 1. It can be seen from the proof of the theorem that the matrix T(4J)
that decomposes system (7.1) into C'(Tm)-blocks has the same smoothness as
the matrices Q(4J) and C() = Go(O,). Consequently, this system is CS(Tm )block decomposable when Q(4J) E CS(Tm ) and C() E CS(Tm ).
Remark 2. It follows from formulae (10.15) and (10.16) that when the matrix
S( 4J) has the block diagonal form
(10.17)
the matrix T(4J) that splits system (7.1) into C'(Tm)-blocks has the form
where E 1 and E 2 are the identity matrices of dimensions rand n-r respectively.
In this case the matrices T 1 () and T 2 () are solu tions of the matrix Riccati
equations
1'1
= PiTi - T i P2 - T i P2i T 1 + P 12 ,
T 2 = P2T2 - T 2P1
(10.18)
T2P12 T 2 + P21 ,
and the split system of equations (10.2) has the form
d4J
dY1
di = a(4J), dt
dY2
dt
= [P () + P12 (4J)T2(4J))Yi,
i
= [P2() + P2i ()Ti (4J)]Y2,
where Pi, P 12 , P21 , P 2 are the blocks of the matrix P = [Pi

priate dimensions.
P2i
P12] of appro-
P2
The requirements imposed on the matrix (10 .17) by the conditions of Theorem 1 of 3.8 turn out to be sufficient for the solvability of the Riccati equations
(10.18) in C'(Tm ).
Remark 3. It can be seen from the proof of the theorem that ifin representation
(10.1) the matrix Q() has a period equal to a multiple of 211", then the matrix
reducing system (7.1) to split form is periodic with the same period as Q().
165
The problem on the representation of matrices S(<!J) in G'(Tm ) in the form

of (10.1) has been little studied. The following statement completely solves it
only for two-dimensional matrices.
Lemma 1. Let S(<!J) be a two-dimensional matrix in G'(Tm ) with eigenvalues
).1(<!J) and ).2(<!J) of different signs.
Then there exists a non-singular matrix
Q(<!J) that is periodic and continuous with respect to <!J with period 471", and
continuously differentiable with respect to t for <!J
= <!Jt(<!J)
such that
(10.19)
where dl(<!J) and d2 (<!J) are positive functions for all <!J E Tm .
To prove the lemma, we first note that if representation (10.19) is possible

with matrix Q(<!J) satisfying the conditions of the lemma, then it is possible
with dl(<!J) = ).l (<!J) , d2 (<!J)
= -).2(<!J) if we assume that ).1(<!J) and ).2(<!J) are
the positive and negative eigenvalues of the matrix S(<!J). But then
= [detQ(<!JWd 1(<!J)(-d2(<!J =
= [detQ(4))F>'1(4>)>'2(4>) = [detQ(4))]2detS(4>),
detS(<!J)
and consequently
[detQ(4))F
= 1.
If det Q(<!J) = -1, then the matrix

Q(4))diag{l,-i}
brings about the representation (10.19) and has determinant equal to 1. Thus
if representation (10.19) is possible then the representation
(10.20)
is also possible with the periodic matrix Q( <!J) of period 471" satisfying the conditions of the lemma and such that
det Q(<!J)
= 1.
(10.21)
Thus we seek a matrix Q(4)) effecting the transformation (10.20) and satisfying condition (i 0.21). Let
Q( 4
= [X
Xl
Y] .
Yl
166
CHAPTER 3
Condition (10.21) takes the form of the equation

(10.22)
while equality (10.20) takes the form of the system
aYI - by = AIX,
bYI - dy
+ bx = A2Y,
-bxI + dx = A2YI
-axi
= AIXl,
(10.23)
Substituting the values of X, Xl found from the first two equations of system
(10.23) into the two remaining equations, we obtain the identities
Consequently, in order to find Q() it is sufficient to choose
and
Xl
using
y and YI according to the first equalities of system (10.23) such that relation
(10.22) holds. The latter, by virtue of the choice for x, Xl, takes the form of
the equality
(10.24)
The problem has been reduced to that of finding two functions y( ), YI ()
that are continuous and periodic in and which turn equality (10.24) into an
identity for all E Tm . Since two arbitrary functions y( ), Yl () satisfying
(10.24) cannot become zero simultaneously, we can set
Y
= Rcos'I/J,
Yl = Rsin'I/J
(10.25)
and reduce equality (10.24) to the form
and, after obvious transformations, to the form
(d - a) cos 21/J - 2bsin 21/J
= ~; - (a + d).
(10.26)
Since the eigenvalues of the matrix S( ) have different signs, it follows

that
167
which enables us to rewrite equality (10.26) as

2A1 - (a + d)R 2
cos(2'IjJ + a) = R2[(d _ a)2 + 4b2]1/2'
(10.27)
where a is determined by the equalities

d-a
cosO'
= [(d _ a)2 + 4b2J1/2'
.
sma
= [(d _ a)2 + 4b2]1/2'
2b
Since A1 (4)) > 0, A2( 4 < 0, it follows that
Thus, by choosing R( 4 sufficiently large we can make the absolute value of the
right hand side of equality (10.27) less than one. With such a choice of R(4))
the function
1
'IjJ
= '2 cos
-1
2A1-(a+d)R2
R2[(d _ a)2 + 4b2]1/2
a
2
is a solution of equation (10.27).

Find out the properties of this solution considered as a function of 4>. For
a sufficiently large R
= R(4)) from the space G'(Tm ) the function
cos
-1
belongs to the space G'(Tm
2A1 - (a + d)R 2
R2[(d _ a)2 + 4b2]1/2
= 2~(4))
).
By applying the argument principle to the pair of functions

d(4)) - a(4))
[(d(4 - a(4)))2 + 4b 2(4))p/2'
2b( 4
[( d( 4
- a( 4)2 + 4b 2 ( 4> )]1/2
we see that
where k
= (k 1, . .. ,k
then
'It
where
~o
m )
is an integer vector and
~1
a function in G'(Tm ). But
= ~(4)) - ~1(4)) - (k,4 = ~o(4)) _ (k,4,

2
2
E G'(Tm ).
168
CHAPTER 3
It follows from formulae (10.25) that the functions

y = Y(4)) = R(4)) cos ( <1>0(4)) _
(k,./)) ,
(k;4)
VI = VI (4)) = R(4))sin(<1>0(4)) -
satisfy equality (10.24) for all 4> E Tm

respect to 4>v (v
= 1, ... , m),
These functions have period 41T with
are continuous with respect to 4>, and have
continuous derivatives with respect to t for 4> = 4>t(4)). Then the matrix Q(4))
has similar properties because its elements are the functions y( 4, VI (4)) and
the functions x(4)), Xl(4)) which are expressible in terms of them.
To show that period doubling in the representation of the matrix S(4)) in
form (10.1) can take place, we consider this representation for the matrix
_ [- cos 4> sin 4> ]
(4)) -
. A..
sm,/,
where 4> is a scalar.

The eigenvalues of this matrix are Al
for the matrix S( 4 we can take:
A..
(10.28)
cos,/,
= 1 and A2 = -1.
(sin(4>/2), cos(4)/2))
for Al
=1
(cos(4>/2), - sin(4)/2))
for A2
= -1,
As eigenvectors
and consequently the eigenvectors for the matrix S( 4 cannot be chosen to be

periodic with period 21T.
We shall call system (7.1) two-dimensional if x
= (x 1, X2) is a two-dimen-
sional vector, and C'(Tm)-block decomposable with period doubling if the decomposition is effected by a matrix T( 4 that is continuous with respect to 4>,
periodic with respect to 4> with period 41T and has continuous derivative with
respect to t for 4> 4>t(4)).

The following result is a consequence of Lemma 1 and Theorem 1.
Theorem 2. Let a E CLip(Tm ), P E C(Tm ) and suppose that system (7.1)

is two-dimensional and exponentially dichotomous. Then it is C'(Tm)-block
decomposable with period doubling.
For n
3, period doubling is not sufficient for the C'(Tm)-block decom-
posability of an exponentially dichotomous system. The reasons for this will

be discussed in the following section.
169
3.11. On triangulation and the relation between the
C'(Tm)-block decomposability of a linear system

and the problem of the extendability
of an r-frame to a periodic basis in En
We continue our discussion of the system of equations
dx
di = P(4))x,
(11.1)
where a and P are functions in CLip(Tm) and C(Tm ) respectively.

We set
Lu
= 11. -
P(4))u,
thus defining an operator L on the function space C'(Tm ).
A scalar function A = A(4)) in C(Tm ) is called an eigenvalue of the operator
L if the equation
Lu = Au
(11.2)
= u(4)) t. 0,
4> E Tm , in C'(Tm ). An eigenvalue of

the operator L is called an eigennumber if it does not depend on 4>. A non-
has a non-trivial solution u
trivial solution in C'(Tm ) of equation (11.2) will be called an eigenfunction of

L corresponding to the eigenvalue A.
Theorem 1. Suppose that the operator L has eigenvalues
and let
(11.3)
be the eigenfunctions of the operator L corresponding to these eigenvalues.
Then the surface
(11.4)
is an invariant set of system (11.1) and on this set system (11.1) is equivalent
to the following system of equations
d4>
di = a(4)),
(11.5)
170
CHAPTER 3
To prove the theorem we fix a point (4)0, xo) on the surface (11.4) by setting
P
Xo
= I>j(4)o)yJ,
yJ E R
j=l
and consider the motion 4>t(4)o), Xt(4)o,xo) of system (11.1), that starts at this
point. Let
x(t,4>o,xo) =
2: Uj (4)t (4)o))Yj(t, 4>0, yJ),

P
j=l
where 4>t( 4>0), Yj (t, 4>0, yJ) is the solution of system (11.5) with initial conditions
4>0, yJ. By differentiating the function x(t, 4>0, xo) with respect to t, we see that
dx(t,~o,XO) = t[Uj (4)t(4>o))Yj (t, 4>0, yJ) -
Aj(4)t(4>O)) x
j=l
P
XUj(4)t(4>o))Yj(t,4>o, yJ)l
= P(4)t(4>o)) 2: Uj (4)t(4>o))Yj (t, 4>o,YJ) =

j=l
=P(4)t(o))x(t,o,xo).
Hence it foliows that 4>t(4)o), x(t, o, xo) is the solution of system (11.1)
defined by the initial condition o, Xo. But then
This means that any motion t(o), Xt(4)o,xo) for which (o,xo) belongs
to the surface (11.4) is defined for any t E R by formula (11.4) if
4>, Yj are
replaced there by the solution t( o), Yj (t, o, yJ) of system of equations (11.5).
This proves Theorem 1.
Note that if the vectors (11.3) form a p-frame in En, then the surface
(11.4) is a manifold diffeomorphic to the direct product of the torus Tm and

the plane EP. In particular, it can be a separatrix manifold of an exponentially
dichotomous system.
Theorem 2. Under the change of variables
= U(4))Y + V()z,
det[U(), V()] ::/= 0,
4> E Tm
(11.6)
171
the system of equations (11.1) is reduced to the block-triangular form

dz
dt
dtP
dt = a(tP),
dy
dt = -D(tP)y + Ql(4J)Z,
= Q(tP)z,
(11.7)
where U( tP) = (Ul (tP), ... ,up ( tP)), V (tP) = (Vl (4J), ... ,vn _ p ( 4J)) are matrices in
C'(Tm ), D(4J) = diag{>'1(4J), ... , Ap (4J)} E C(Tm ), if and only if the functions
Aj(tP) (j = 1, ... ,p) are eigenvalues of the operator Land Uj(tP) (j = 1, ... ,p)
are the eigenfunctions of L corresponding to them.
To prove the theorem, first we suppose that the Ai (4J) (j = 1, ... , p) are
eigenvalues of L and the Uj(4J) (j = 1, ... ,p) are the eigenfunctions of L corresponding to them, so that
Equalities (11.8) are equivalent to the single matrix equality
LU()
= U(tP)D(cP) ,
tP E Tm .
(11.9)
Let V(tP) be a matrix satisfying the conditions of Theorem 2. We show that

the change of variables (11.6) reduces system (11.1) to the form (11.7).
To do this, we show that the motion tP = t(tPo), :I: = :l:t(4Jo, :1:0) of system
(11.1) is defined by formula (11.6) for any tPo E Tm , :1:0 E En, t E R, by setting
4J
= tPt(4Jo), y = Yt, Z = Zt
and choosing Yt, Zt such that they are solutions of

some system of equations (11.7). For this purpose we set
.
U(cPt(tP))Yt
dYt
dZ
+ U(tPt())di + V(tPt(tP))Zt + V(tPt(tP))di:t =
= P(tPt(tP))U(t(tP))Yt + P(tPt(4J))V(tPt(tP))Zt
for all t E R, 4J E Tm . Using (11.9) we obtain the following relation for dytldt
and dztldt
dYt
UPt(tP)) ( di:
dZ t
+ D(t())Yt ) + V(tPt(tP))di:
=
= [P(t())V(tPt()) - V(t())]Zt.
We denote the inverse of the matrix 4>( tP) = [U(tP), V( tP)] by
(11.10)
172
CHAPTER 3
Then the identity ct>-l (tP )ct>( tP)
= E can be written as:
U1(tP)U(tP) = E,
V1(4))U(4>)
= 0,
U1(tP)V(tP) = 0,
V1(4))V(4>)
= E,
where E and 0 are the unit and zero matrices of appropriate dimensions.
Multiplying equality (11.10) on the left by ct>-l(tPt(tP)) we obtain the following system of relations for dYt/dt and dztldt:
dYt
dt = -D(tPt(4>))Yt + U1(tPt(4>))[P(tPt(tP))V(tPt(tP)) - V(tPt(4>))]Zt,

dZ
dtt = V1(4)t(tP))[P(4>t(4>))V(tPt(tP)) - V(tPt(tP))]Zt,

which proves that the change of variables (11.6) reduces the initial system of
equations (11.1) to the form of (11.7):
dtP
di = a(tP),
dy
dt
dz
.
dt = VI (tP)[P(tP)V(tP) - V(tP)]z,
= -D(tP)y + U1(tP)[P(tP)V(tP) -
.
V(4))]z.
Conversely, suppose that the system (11.1) is reduced to system (11.7)

under the change of variables (11.6). The motion 4>t(tP), Xt(4), xo) is related to
the solution tPt( tP), Yt, Zt of system of equations (11.7) by the formula
Differentiating it with respect to t we obtain the identity:
U(4)t(4>))Yt + U(4)t(4>))[-D(4>t(4>))Yt + Ql(4)t(4>))ztl+

+V(4)t(tP))Zt
+ V(tPt(4>))Q(4>t(tP))Zt
= P(4)t(tP))U(tPt(tP))Yt+
+P(4)t (4)))V(4>t (tP))Zt,
t E R, tP E Tm
If we consider this identity for the solution Yt, Zt with Zt == 0, we obtain

the new identity
173
from which it follows for t = 0 that

U() - P()U()
= U()D(),
E Tm
in view of the arbitrariness of the choice of the initial condition Yo. But then
U () satisfies the matrix equation (11.9) and consequently the
1, ... ,p) are eigenvalues and the Uj() (j
>.} () (j
= 1, ... ,p) are the eigenfunctions of
L corresponding to them.
The following statement, which clarifies the relation between the C'(Tm )block decomposability of the linear system (11.1) and the problem of the extendability of an r-frame to a periodic basis in En, easily follows from Theorem 2.
Theorem 3. The system of equations (11.1)is reduced to the block-triangular
form (11.7) by transformation (11.6) with the matrix [U(), V()] E C'(Tm ) if
and only if the p-frame (11.3) can be extended to a periodic basis in En.
By Theorem 3, system (11.1), the eigenfunctions of which form a frame

that cannot be extended to a periodic basis in En, cannot be transformed to
the block-diagonal form (11.7) by transformation (11.6). System (11.7) has
the general form of a block-triangular system if y is a one-dimensional vector.
Therefore if the operator L only has eigenvalues such that the eigenfunctions
corresponding to them cannot be extended to a periodic basis in En, then
system (11.1) cannot be transformed to a block-triangular form with 1- and
(n - I)-dimensional blocks by a transformation with coefficients in the space

C'(Tm ). This can be used to construct exponentially dichotomous systems
that are not C'(Tm)-block decomposable. Such examples are constructed in
[35] for systems of equations with quasi-periodic coefficients. We shall say
that the exponentially dichotomous system (11.1) is C'(Tm)-block decomposable
with separatrix manifolds that can be straightened if the separatrix manifolds
of the decomposed system are its coordinate planes, and the motion on them
is determined by the block subsystems of the decomposed system.

Theorem 4.
Suppose that the system of equations (11.1) is exponentially
dichotomous. Suppose further that the eigenvalues of the operator L of the

system contain r positive eigenvalues and that the eigenfunctions corresponding
to them form an r-frame in En.
174
CHAPTER 3
Then system (11.1) is C' (Tm)-block decomposable with separatrix manifolds

that can be straightened if and only if this system of vectors can be extended to
a basis in En.
To prove the theorem, we denote the positive eigenvalues of L by

(11.11)
and the eigenfunctions of this operator corresponding to them by

(11.12)
Suppose that functions (11.12) form an r-frame of vectors that can be extended
to a periodic basis in En. Then the transformation (11.6) with matrix V(J)
in C'(Tm ) which extends U(J)
= (Ul(J), ... , ur(J)) to a periodic basis in
En
transforms system (11.1) to the form (11.7). Under the transformations (11.6)
the exponential dichotomy of system (11.1) is preserved, so that system (11.7)
is exponentially dichotomous.
As follows from the properties established earlier, the number r determines
the dimension of the separatrix manifold M+ of the exponentially dichotomous
system. Since the eigenvalues Aj (J) (j
= 1, ... ,r) are positive, it follows from
the form of system (11.7) that its separatrix manifold M+ is given by the
equation
z=o,
JETm .
(11.13)
Using the variables J, x the manifold M+ can be given by equation (7.2)
C(J)x
= x,
J E Tm
(11.14)
where C(J) is a projection in C'(Tm ).

By expressing x in terms of y and z using (11.6) and substituting it into
(11.14) we can write the equation that defines the manifold M+ of system
(11.7) in the form:
C(J)U(J)y + C(J)V(J)z
= U(J)y + V(J)z.
(11.15)
The points of the manifold (11.13) satisfy equation (11.15), but since y was
chosen arbitrarily, this is possible only if
C(J)U(
= U(J),
J E Tm .
(11.16)
We write down the equation for the separatrix manifold M- (7.3):
C(J)x
=0,
J E Tm
175
Under the transformation (11.6) it becomes the separatrix manifold M- of

system (11.7) with equation
C(4))[U(4>)Y + V(4))z]
= 0,
4> E Tm
which, in view of identity (11.16), takes the form
U(4))Y + C(4))V(4>)z
=o.
(11.17)
Since the matrix U(4))U(4>) is a Gram matrix of linear independent vectors, it is non-singular for all 4> E Tm . Multiplying equation (11.17) first by
U(4)) and then by [U (4))U(4>)]-l, we arrive at the equality

(11.18)
which is satisfied by any solution of equation (11.17). For a fixed 4> E Tm ,
equality (11.18) defines an (n - r)-dimensional plane in the y,z space. Since
the manifold M- is an (n - r)-dimensional plane in the Y, z space, it follows
that for a fixed 4> E Tm , M- is the plane (11.18). This proves that equality
(11.18) is the equation of the separatrix manifold M- of system (11.7).
By expressing the variables y in equations (11.7) in terms of the variables
Yt according to the formula
we transform system (11.7) into a system of the same form but with a separatrix
manifold with equation
YI =0,
4>ETm .
The latter is possible only if the system of equations for 4>, z, YI has the form
(11.7) with Y = YI and QI == 0 and therefore has the form of a block diagonal
system with diagonal blocks Q(4)) and -D(4)) = -diagPl (4)), ... ,A2( 4>
n. The
C'(Tm)-block decomposability of system (11.1) has been proved. From the form
of the decomposed system it follows that the separatrix manifolds M+ and Mhave been "straightened".
Suppose now that the eigenfunctions (11.12) of the operator L form an
r-frame which cannot be extended to a periodic basis in En. Suppose that
system (11.1) can nevertheless be reduced to the block-diagonal form
d4>
di = a(4)),
(11.19)
176
CHAPTER 3
with straightened separatrix manifolds under the change of variables
with the matrix ~(4))
= [Ul(4)), VI (4))] E C'(Tm ),
det~(4))"# 0
for 4> E Tm
Suppose that the equation
defines the separatrix manifold M+ of system (11.19). Then the surface
(11.20)
is an invariant set of system (11.1), the motion of which is defined by the
first two equations of system (11.19) and consequently they are exponentially
damped as t
-4
+00. The surface

x=U()y,
yEE r ,
(11.21)
ETm
is also an invariant set of system (11.1) by Theorem 1. The motions on it are

defined by system (11.5), and since the values Aj() (j
they are also exponentially damped as t - 4 +00.
= 1, ... ,r) are positive,
For any fixed E Tm both surfaces (11.20) and (11.21) define r-dimensional
planes in x-space. The motion of system (11.1) on surfaces (11.20) and (11.21)
shows that these surfaces are one and the same separatrix manifold M+ of
system (11.1) written in the different coordinate systems (,yt) and (4),y) respectively. For a fixed E Tm there are defined two Euclidian coordinate
systems Yl and y in the same Euclidian space E r (4)) which have the same
origin. Consequently, there is a non-singular r x r matrix R( ) defined for
every E Tm , such that the transition from the coordinate system Yl, to the
coordinate system y is given by the relation
Yl
= R(4))y.
For equalities (11.20) and (11.21) to define the same plane for any 4> E Tm ,
it is necessary that the matrix R( ) satisfy the identity
(11.22)
177
Using the matrix
~(</
we can rewrite identity (11.22) as
and by taking the inverse of the matrix
~(t/
we see that R(</ E C'(Tm ).
This proves that the matrix R( </ in identity (11.22) belongs to the space
C'(Tm ). Then we have the representation
det[U(t/, VI ()] ::j; 0,
</>
E Tm
(11.23)
since the matrices () == [Ud4, Vd4] and R() are non-singular.
Since VI () E C'(Tm ), it follows that inequality (11.23) ensures that the
r-frame U () can be extended to a periodic basis in En. Thus we have a
contradiction, and this completes the proof of Theorem 4.
The statement of Theorem 4 can be used to construct exponentially dichotomous systems (11.1) that are not C'(Tm)-block decomposable with an
arbitrary size of diagonal blocks.
The following result on C'(Tm)-diagonalization of system (11.1) is a trivial
consequence of Theorem 2.
Corollary.
In order that the system of equations (11.1) be reduced to the
diagonal form
by the transformation
x=U(4))y,
UEC'(Tm ),
detU()::j;O, ETm ,
it is necessary and sufficient that the operator L of the system have n eigenvalues AI(), ... , An() the eigenfunctions of which form a basis in En.
We note that necessary conditions for C'(T,n)-diagonalization of system

(11.1) are given in [118].
178
CHAPTER 3
3.12. On smoothness of an exponentially stable invariant torus
In our discussion of the invariant torus
= u(<I,
 E Tm
(12.1)
dx
d<l>
7ft = a(<I,
7ft
= P(<Ix + f(<I
(12.2)
there immediately arises the problem of the dependence of the smoothness of

the invariant torus on that of the right hand side of system (12.2). As was
noted before, this dependence is non-trivial. Therefore it needs to be studied
independently.
In our study of the smoothness of the torus (12.1) we start with the following statement.
Theorem 1. Let a,P E C/(Tm
),
2:: 1. Then if for an arbitrary function
E C/(Tm ) a Green's function of system (11.1) satisfies the inequality
(12.3)
where <l>t( E Tm , and J( and I are positive constants independent of <1>, then the invariant torus (12.1) of system (12.2) belongs to the space C'(Tm ) and satisfies the
inequality
(12.4)
The proof of the theorem is obvious if we use estimate (12.3) and the
integral representation for the function u(.,.(<I)dT.
(12.5)
By Theorem 1 the smoothness of invariant torus (12.1) of system (12.2)

depends essentially only on the properties of the Green's function G o( T, <1 and
the solutions <l>t( <1 of the system defining the trajectory flow (12.2) on torus
f belong to the space C'(Tm )

2:: 1 and look at conditions ensuring that inequality (12.3) holds.
(12.1). We assume that the functions a, P and

with finite I
179
We shall consider here the case when the torus (12.1) is exponentially stable. According to results of 3.5, this will be the case if the following condition
holds:
= (30) 0,
inf (3(<1
</>ETm
(12.6)
where (3(<1 is defined by the inequality
+ (1/2)5()]x,x) < -(3(<1,

(S()x,x)
-
inf max ([S(<IP()
SE!R ollxll=1
(12.7)
91 0 is the set of n x n-dimensional positive-definite symmetric matrices belonging to the space C'(Tm
).
We define the value of OI( ) by setting
. ([S1() ~</> + 2S1()]1I>, 11

()
sup mm
> 01 ,
SIE!R 1 1ltJ>1I=1
(S1()1I>, 1I
l'
(12.8)
where 91 1 is the set of m x m-dimensional positive-definite symmetric matrices

belonging to the space C 1 (Tm
).
Theorem 2. Suppose that inequality (12.6) holds and
inf [(3()
</>ETm
+ 10l()] > O.
(12.9)
Then the invariant torus (12.1) of system (12.2) belongs to the space C1(Tm
).
To prove the theorem we first exploit the coarseness of its conditions,

which ensure that there exist matrices S() E !R o, S1() E 91 1 and functions
(3(<1, 01(<1 satisfying inequalities of the form (12.7), (12.8):

max ([S(<IP() + (1/2)5()]x, x) < -(3(<1,
IIxll=1
(S()x,x)
-
(12.10)
min ([S1()oa(<I/o + (1/2)51(<1]11>, 11 > 01(<1

1ItJ>1I=1
(S1(<I1I>, 11
and conditions of the form (12.6), (12.9):
(3(<1 2: (30,
(3()
where (30 and 1 are positive constants, (30
+ 101(<1 2: 1,
= </>ET
inf
(3(), 1
(12.11)
= <l>ET
inf
[(3(<I+la(<I].
If the first equality of (12.11) is satisfied, then the second is of independent
180
CHAPTER 3
significance only for those values of > for which O'(
< O. Therefore we may
assume that the function 0'( satisfies the condition
O'(
0, > E Tm .
(12.12)
Assuming that inequality (12.12) holds, we prove the estimate

(12.13)
for any 1
~ 5 ~
I and arbitrarily small { > 0, and some !( that depends on
but does not depend on t,>. Here D:>t( is any derivative of order
function >t( with respect to the variables >
= (>1, ... ,>m).
5, (
of the
We carry out the proof by induction. To do this we write the system of

equations
(12.14)
which is satisfied by the matrix of partial derivatives 8>t( / O of the function
>t( . Setting 11':
= 8>t (/ 84Ji and considering the function
we find by using inequality (12.10) for SI( that
it 2: 20'( >t( >)) v.
(12.15)
Upon integrating this inequality and making the standard estimates we obtain
from (12.15) the inequality:
which proves estimate (12.13) for
= 1.
Suppose that inequality (12.13) holds for all

holds for
= 11 + 1.
To prove this we differentiate
by replacing the variable
11'
by its value
result in the form of the equality:
11':
5 ~
/1
h. We prove that it then
times the identity obtained
in equation (12.14), and write the
where
R( tPt (tP))
181
= D~ [Oa~~~ tP)) VJ;] _ oa~~~ tP)) D~ VJ:.
Since R( tPt (tP)) is a differential expression which contains the terms

l,-j
D,p
(oa(tPt(tP))) n.i (f)q)t(tP))

8tPt
1.f,p ~'
._
)-,1, ... ,1 1 -1, v-l, ... ,m
with constant coefficients, inequality (12.13) taken for s $ 11 enables us to

estimate R( tPt (tP )) as follows:
where
Therefore
which leads to the estimate
$ K 1 K ex p {
-1
$ I< 1 I< ex p {
Il
1-
,p
11<
oa(tPt(tP)) . vi. OtPl(tP)

otPt
,p otP
-
[(It - j) + j + l]a(tPT(tP))dT} exp{ -(/1 - j + l)d} $
-1
(11
+ l)a(tPT(tP))dT -
(11
This proves for R( tPl( tP)) the following estimate:
where K 2 is a positive constant independent of tP.

Since
VJb = (0, ... ,0,1,0, ... ,0),

'--v--"
i-I
'--v--"
m-i
+ l)d}, t
$ 0.
CHAPTER 3
182
it follows that [D~1/Jnt=o

satisfies the relation
= 0 for
II 2: 1. Consequently the derivative D~1/J;

(12.17)
where n~(oalo<! denotes the fundamental matrix of solutions of system (12.14),
which becomes the identity matrix when t T.

We now estimate the integral (12.17). Clearly, the columns of the matrix
1/J;(T,T(<!)dT}, t ~ T,
n~(oalo<! are the solutions
the values of unit vectors
ej
-1
which shows that
n~ (oal o<!
satisfies an inequality of the form
This inequality allows one to obtain an estimate for integral (12.17) and
to obtain for D~l1/J; an estimate of the form:
!(J{21 -1
P{ -1
+
1 {-1
IID~1/J:1I ~
~ J{3 ex
T
O:(<!>T(<!dT-1(/l+l)O:(T(<!>))dT-(/l+ 1)fT }dT
exp {
O:(<!>T(<!>))dT - (/1
x
l)d}X
eX P
11O:(<!>T())dT}dT,
~ O.
(12.18)
On estimating the integral
It =
we find that
It
= -texp {
~ -tex p {
-1
-1
1 -1
ex p {
1Io:(T(<I)dT }dT'
~ 0,
11O:(<!>T(<!>))dT}-
-1
11 O:(<!>T(<I)dT }
11To:(<!>T(<!>))ex p {
~ J{(f)exp{
-1
-1
11 0: (<!>t (<!>))dT}dT
hO:(<!>T(<!)dT -
ft}, (12.19)
183
where K(i) is a constant dependent on
i.
Using (12.18), (12.19) we can write the estimate for the function D~ltN as
whence it follows that
for some positive K

proves (12.13) for s
= Iid and arbitrary fixed i1 > O. Inequality

= It + 1; hence it proves inequality (12.13).
(12.20)
Now consider the function
where
n~ (cP)
is the fundamental matrix of solutions of the system
dx
dt = P(cPt(cP))x,
which becomes the identity matrix for t
(12.21)
= T.
Let us prove the estimate
IID~xt(T,cP,xo)11 S J{ ex p { - i t {3(cPT(cP))dT
-1
sa(cPT(cP))dT - iT }lIxoll
(12.22)
for arbitrary
S t S 0, 0 S s S
f, arbitrarily small
> 0 and some
K which
depends on s, i but does not depend on t, T, cP.

We proceed by induction. For s
= 0 inequality (12.22) follows from condi-
tion (12.10) for the matrix S(cP) by applying standard estimates relating to the
function V
to t.
= (S(cPt(cP))Xt(T,cP,XO),
Xt(T,cP,Xo)) and its derivative with respect
Assuming that estimate (12.22) holds for all 0 S s

it - 1, we prove
it for s
fl. To do this, we differentiate 11 times the identity obtained by
substituting Xt(T,cP,XO) for x in equation (12.21) and write the result as:
where
184
CHAPTER 3
The function R 2 (4)t(4>),Xt(T,4>,XO)) is a differential expression which contains the terms D~-ip(4)t(J))D{Xt(T,J,X)(j
= O, ... ,h -1)
with constant
coefficients. Inequalities (12.13) and (12.22) enable us to estimate R2( Jt (J),
Xt(T,,XO as follows.
We have
l,-i
D~-i P(Jt(
=L
D~IP(Jt(J
L coa(Dq,Jt(Ja, x
a
0=1
This leads to the estimate
But then for any 0 :::; j :::; 11 - 1 and
T :::;
t :::; 0
1
0
IIDj-ip(t(J)D{Xt(T,,xo)ll:::; I<4J(eXP{-(lI-j)
x exp{ - i t {3(Jr(JdT -
1
1
a(r(4)dT} x
a(Jr(Jdr - f(t
:::; J(4J( exp{ - i t {3(Jr(JdT} exp{ -11
+ T) }lIxoli :::;
a(Jr(JdT - 2fT }lIxoll,
IIR2 (t(J),Xt(T,,xo1I
:::; K sexp { - i t {3(Jr(JdT -11
1
0
:::;
a(Jr(Jdr - 2fT }lIxoli
(12.23)
t :::; 0 and some positive I<s independent of t, J, T.

Since [D4,xt(r,J,xO)]t=r = 0 for any s ;::: 1, it follows from the equation
for Di x.(T, J, xo) that
for all
T :::;
(12.24)
185
Using inequality (12.23) we can obtain the estimate for integral (12.24):
IID~Xt(T, tP,xo)1I ~
K611
ex p {
-1~ (3(tPT(tPdT
x ex p { -11
= K6(t -
-1
T)exp{
-I
T'
(3(tPT(tPdT} X
Q(tPT(tPdT - 2fT} dTtllxo II
(3(tPT(tPdT} x
x exp{ -11
Q(tPT(tPdT - 2fT }lIxoll, T ~ t
~0
(12.25)
It follows from inequality (12.25) that
IID~xl(T,tP,xo)1I~
~ K(ft) ex p { - i t (3(tPT(tPdT
-1
I 1 Q(tPT(tPdT - fIT }lIxoll
0, arbitrarily small f1 > 0 and some K(ft) > O. This inequality

S = 11 and consequently, proves it for any 0 ~ S ~ I.
If we set t = 0 in inequality (12.22) we obtain
for all
T ~
proves estimate (12.22) for
IID~xo(T,tP,xo)1I ~
~ K ex p { -l[(3(tPT(
for all
T ~
+ SQ(tPT(tP]dT -
fT }lIxoll
(12.26)
O. Since we are considering an exponentially stable invariant torus
of system (12.2), the homogeneous system of equations corresponding to (12.2)

has a Green's function C o( T, ) defined by the matrix n~ (tP) for
the zero matrix for
T ~
> O. Thus for
~ 0 and by
> 0 inequality (12.3) holds, while for
0 it becomes:
(12.27)
We now prove (12.27).

We have:
D~XO(T,,J(tPT(<jJ))) =
D', 8Pxo~;~,f) x
p+q$'
Ql
+ Q2 + ... + QI
= P,
Q1
+ 2Q2 + ... + IQI + U
(12.28)
186
CHAPTER 3
Now since
qf(4JT(4J))
=L
D;~f(4JT(4J))
L co1a (D4JT(4J))a 1 ... (~4JT(4J))a"

a
Ol~j
we have
(12.29)
for all
:$; O.
The function xo( T, 4J, f) is linear with respect to
liD; (}pxo~~~4J, f)
where
ej
II :$; const [IID;xo( T, 4J, f)llh op +
f, therefore
IID;xo( T, 4J, ej )lIh1P] ,
}=1
is a unit vector, and hap is equal to 0 for
0'
=F p and 1 for
0'
= p.
In view of inequality (12.26), we have the estimate
II
:$; KlIifllhop + hlP] exp{
-1
DO (}P xo( T, 4J,
{}fP
f)
I<
-
(3(4JT(4J))dT} X
x exp{ -(J'
O'(4JT(4J))dT - fT},
T:$; O.
(12.30)
Inequalities (12.29), (12.30) lead to the estimate
II D ,xo( T, 4J,f(4JT (4J )))11

:$; [{ ex p { -l[(3(4JT(4J))
for any s :$; I and
:$;
+ sO'(4JT(4J))]dr -
(r }Ifl.
:$; O. It follows from the second inequality of (12.11) that
which is sufficient for inequality (12.3) to hold for ( <

proof of Theorem 2.
"y.
This completes the
187
By looking at the proof of Theorem 2 we see that it holds if the fundamental matrices of the solutions n~(4)) and n~(oa(4/o4 of systems (12.21)
and (12.14) satisfy the inequalities
IIn~(4))11 ~ I< exp{- jt (3(4>T (4>))dT} ,

Iln~ (O~~) II ~ K exp
{l
T S t,
a(4)T(4>))dT} , T ~ t
(12.31)
respectively, where (3(4 and a(4)) are subject to conditions (12.6), (12.9).
Since the matrices nM4, n&( oa( 4> )/04 satisfy the identities ofform (4.2):
n~(4)I/(4>)) = n~+I/(4)), n~(oa(:~(4) = n~+I/(O~~),

it follows from the inequalities
{-l
IIn~(4))1I ~ J{ exp
(3(4>T(4>))dT},
0 S t,
Iln~(O~~)11 S Kex P {l a(4)T(4)))dT},
~0
(12.32)
that inequalities (12.31) hold. Indeed, according to the given identities and
inequalities (12.32) we have, for example, the estimate for n~ (4)):
~ J( exp{
_I
Iln~(4))11
t
-
= IIn~-T+T(4))11 = IIn~-T(4)T(4>))1I ~
T
(3(4>T,(4>T(4dTl}
= f{ exp{ -
= J( exp{
jt (3(4>T(4)dT},
which is the same as estimate (12.31) for
_I
t
-
T
(3(4>T+T,(4>))dTl}
t - T 2: 0,
n~(4)).
This gives rise to the following statement.

Corollary 1. The statements of Theorem 2 remain true if the matrices n&(4))
and n~(oa/o4 satisfy inequalities (12.32), where the values (3(4 and a(4)) are
subject to conditions (12.6) and (12.9).
In particular, if the inequalities

IIn~(4))1I ~ Ke- fJot ,
t 2: 0,
IIn~(oa/o41I ~ f{e-exot,
t ~0
188
CHAPTER 3
hold, where
130 = const > 0, ao = const ~ 0, then the conditions (12.6), (12.9)
are fulfilled for I satisfying I < 130/ao. Then the invariant torus (12.1) of system
(12.2) belongs to the space C/(Tm ) with 1<
13o/ao.
It should be noted that the number I defined by the conditions of Theorem 2 as the minimal possible smoothness may be attained. This can easily be
seen from the example of the system
dfjJ
dt
= -sin'",/"
dx = -bx + sin b '"

dt
,/"
(12.33)
which consists of two scalar equations. As was shown in [89], the invariant
torus (12.1) of this system is defined by the function
u(fjJ)
2b-
1t
fjJ 1~/2 cos
.
dt,
,,/2
Sillt
= - tan -2
which behaves as _cfjJb log fjJ as fjJ
-+
0< fjJ < 211",
O. Taking b to be a positive integer and
taking the bth derivative of u( fjJ) we see that this derivative is not bounded.
Inequality (12.9) for system (12.33) has the form inf [b-lcosfjJ]
~ETI
> 0 and holds
for 1= b - 1. Consequently, the smoothness of torus (12.1) of system (12.33),

which is defined by inequality (12.9), is the same as its real smoothness.
3.13. Smoothness properties of Green's functions,
the invariant torus and the decomposing transformation

of an exponentially dichotomous system
We shall be considering the system of equations (11.1) under the assumption
that it is exponentially dichotomous. Let GO(T,fjJ) be a Green's function for
this system. Suppose that G o(T, fjJ) satisfies the inequality
IIGo( T, fjJ)1I :S J( e -..,.ITI,
E R,
(13.1)
where K, "Yare positive constants independent of fjJ.

In order to elucidate the smoothness properties of the function Go(T, fjJ)
with respect to fjJ, we introduce the constant a
> 0 defined by the inequality

(13.2)
where nMIJa/IJfjJ) is the fundamental matrix of the solutions of system (12.14).
189
Theorem 1. Let I ~ 0, P E C'(Tm ) n CLip(Tm ) and

'Y> lao
(13.3)
Then GO(T,</ E C/(Tm ) for any T E Rand

(13.4)
where = 'Y - C, 0 ~ s ~ I, c is an arbitrarily small positive number and

K 1 = K1(c) is a positive constant independent of</>, </> E Tm .
Because estimate (13.1) holds and the matrix C(</ = Go(O, <p) belongs to
the space C'(Tm ), the theorem holds for I = o. We therefore suppose that
I ~ 1. Consider the differences
where Gt(T,</ = GO(T-t,<pt(<P, ~ = </>+Doe;, e; = (0, ... ,1, ... ,0) is the
unit vector, and Do is a scalar constant. Since Gt(T, </ is a Green's function for
system (12.21)' Zt( T,~, </ satisfies the matrix equation
This has a unique bounded solution on R given by the expression

(13.5)
Inequality (13.1) ensures that the function Zt( T,~, </ is bounded on R, so
that the function Zt(T,~,</ is equal to the function (13.5). But then
GO(T,</ - GO(T,~)
+00
-00
Go(s,~)[P(</>.(</>))- P(</>.(~]G.(T,</ds.
Since
(13.6)
190
CHAPTER 3
uniformly with respect to t/> E Tm and (T,S) E D 2, it follows that
' Go( T, t/ - Go( T, 4J)

I1m
D.
.0.-0
_ j+oo J( 5,T,'I'A-.)d5
(13.7)
-00
when the integral on the right hand side is uniformly convergent. Here D2 is
any bounded domain of the T,5 plane and J(5, T, t/ is the right hand side of
equality (13.6). For J(S,T,t/ we have the estimate
from which we see that integral (13.7) converges uniformly with respect to
(T,t/ E R x Tm and therefore defines the derivative {)Go(T,t//{)t/>;.

By formulae (13.7), (13.8) we have the estimate
T, t/ II s. J{ 1(_2_ + ITI) e-(-Y-a)ITI,

I ()Go(
{)t/>;
2r - a
T E R, 4> E T m ,
(13.9)
for ()GO(T,t//{)t/>;, and this proves the theorem for 1=1.

It follows from equality (13.7) that
()GO(T,t/
{)t/>i
= j+oo GO(5,t/
-00
v=1
()P(t/>.(t/>)) {)(t/>.(t/)v G.(T,t/ds

{)(t/>.)v
{)t/>;
(13.10)
for all (r,t/ E R x Tm. Using identity (13.10) and taking derivatives we prove
Theorem 1 for any I ;::: 1.
In particular, for I
inequality
= 2 by
using estimates (13.1)-(13.3), (13.9) and the
which follows from (13.2) by reasoning similar to that in 3.11, we obtain the
following estimate for the deri vative of the integrand expression J (5, T, t/ in
formula (13.10):
II D J(s,T,t/11
s.
s. J{2[e-('Y- 2a -J!'!e-'YIT-'1 + e-(-y-2aJI'le-('Y-aJIT-'lj S.

<
_ J( 2 [e-(-Y-~dl'le-'YIT-'1+ e-(-y-~tll'le-'Yll'-TI] ,
where we have set iiI
= 20' + t,
rl
=r -
a.
(13.11)
191
Estimate (13.11) is of the form (13.8). Consequently the integral
+00
-00
D~J(s, T, <jJ)ds
converges to the function D~aGo(r, /a>i uniformly with respect to (T, E

R x Tm . This implies the estimate
for all t E R, > E Tm . Because the number

the function
D~aGO(T,<jJ)/a>i
>
0 is arbitrary we have that
can be estimated by inequality (13.4) for 1=2.
This completes the proof of the theorem for L 2.

By applying standard estimates for higher derivatives of the integrand
function J(S,T,<jJ) in formula (13.10) and using induction we see that the theorem holds for any I> 2.
Theorem 2. Let the conditions of the previous theorem be satisfied and f E
G'(Tm ). Then the invariant torus (12.1) of system (12.2) belongs to the space
G'(Tm ) and admits the estimate
lui,
Klfl,
To prove the theorem we need to use estimate (13.1) for the Green's function Go(r, and the estimate for the function >t( in the form
which follows from condition (13.2). These estimates clearly lead to the inequality
(13.12)
for all (T, E R x Tm . If we choose
> 0 sufficiently small, it will follow from
inequality (13.3) that the difference ::y - La is positive, and this is sufficient for
Theorem 2 to hold.
We now find a relation between the index I of exponential decay of the
Green's function GO(T,<jJ) for the exponentially dichotomous system (11.1) and
the parameters of the matrices S(<jJ) and 5( which make the system exponentially dichotomous.
192
CHAPTER 3
Lemma 1. Let a E CLip(Tm ), P E C(Tm ). If the non-singular matrix S( J) E

C/(Tm ) satisfies inequalities
max I(S(J)x,x)1
II r ll=l
S p.,
(13.13)
max ([~S(J) + S(J)P(J)]x,x) S -{3

II r ll=l
with positive constants p. and (3 then the Green's function Go( T, J) for system
(11.1) satisfies inequalities (13.1) with index
'Y
= (3/p..
(13.14)
To prove the lemma, we first of all note that its conditions imply that
system (11.1) is exponentially dichotomous. Denote the solution of this system
belonging to the separatrix manifold M+ by Jt(J), xt(J,xo). Since, according
to the Remark in 3.8, this manifold lies in the cone (8.17) for fixed J, we have
(13.15)
for some positive > O. The function V

dV = 2 (
dt
.
[~S(Jt(J
= (S(Jt(J4(J,xo),xt(J,xo
+ S(Jt(JP(Jt(J]xt(J,xo),4(J, xo) )
s -2f3l!xt(J,xo)1I
::;
and therefore
Integrating this inequality, we obtain
for all t
T and any (T,J) E R x Tm . Using condition (13.15) and inequality
(13.16) we have:
IIxt(J,xo)1!
s ~ ex p { -~(t -
T) }l!x T (J, xo)l!,
T.
(13.17)
193
In similar fashion we can prove the following estimate for the motion t/Jt(t/J),
xt"(t/J,xo) of system (11.1) belonging to the separatrix manifold M-:
(13.18)
But then the Green's function Gt(T,t/J) for system (12.21) defined by relation (9.2) for solutions satisfying inequalities (13.17) and (13.18), satisfies the
inequality
(13.19)
with the constant f{ independent of t/J. Since Go(T, t/J) = Gt ( T, t/J)lt:o' it follows
from (13.19) that estimate (13.1) holds for the function GO(T,t/J) with I equal
to the value given in (13.14).
The index of exponential growth of the matrix n~(oa/o) can easily be
related to the parameters of the matrix 51 (t/J) in the space 9't 1. Indeed, if the
positive constant 0' is chosen such that the inequality
. f
sup m max
<pETm S,E!lt, '1/>11=1
I([~ S't (t/J) + 51 ( t/J) a~~) ]1P,1P) I <

(51 (t/J)1P, 1P)
(13.20)
0'
holds, then estimate (13.2) holds for the matrix nb(oa/ot/J) with
0'
defined by
condition (13.20). This follows at once by making standard estimates of the
derivative of the function V1 = (51 (t/Jt( t/J,Pt{t/J, 1Po), 1Pt( t/J, 1Po, where 1Pt( t/J, 1Po)
is a solution of system (12.14).
The next statement follows from what has been said as well as Theorem 1
and Theorem 2.
Corollary 1. Let a,f,? E C1(Tm
),
2 1.
Suppose that the non-singular
matrix 5() E C'(Tm ) satisfies inequalities (13.13) with positive constants j.t
and {3. If
{3
where
0'
> 1j.t0',
(13.21)
is chosen from condition (13.20), then the Green's function Go(r, t/J)
for system (11.1) and the invariant torus (12.1) of system (12.2) belong to the
space CI(Tm ).
It is now easy to answer the question on smoothness of a decomposing
transformation of an exponentially dichotomous system when this transformation is decomposing.
194
CHAPTER 3
Theorem 3. Let a, P E G/(Tm ), I :::: 1. Suppose that the non-singular symmet-
ric matrix S(J) E G/(Tm ) satisfies inequalities (13.13) with positive constants
p., f3 and admits the representation
(13.22)
with a non-singular matrix Q(J) E G/(Tm ) and positive definite blocks D1(J)
and D 2 (J) for all J E Tm . If inequality (13.3) holds then system of equations
(11.1) is G/ (Tm)-block decomposable.
The proof of Theorem 3 follows from the theorem on the G'(Tm)-block
decomposability of system (11.1) and the properties of the matrices G( J) =
Go(O, J) and Q(J), namely: G(J) E G/(Tm ) and Q(J) E G/(Tm ).

In conclusion, we consider two particular forms of system (11.1) when the
smoothness of the Green's function, the invariant torus and the decomposing
matrix are the same as that of the functions a, P and
The first relates to the case when a
arbitrarily small number
Q'
f.
= w = const.
Then we can take an
and inequality (13.3) will be satisfied by any finite
number I. In this case the smoothness of the Green's function and the invariant
torus of the system in question are the same as that of the coefficients a, P and
of the system, while the smoothness of the decomposing transformation is
the same as that of the functions a, P, f and the matrix Q( J) in representation
(13.22).
The second case relates to an exponentially dichotomous system such that
its function G( J) = Go(O, J) does not depend on J. The smoothness of an
invariant torus of such a system is equal to that of the functions a and f, and
the decomposing transformation does not depend on J. It is easy to see that
we have the second case if the matrix P( J) has the following form
where S is a constant matrix.
3.14. Galerkin's method for the construction of an invariant torus

The construction of an invariant torus of system (12.2) requires approximation
methods. One such method which is convenient for practical realization is
Galerkin's iteration method. According to this method, the function u( J) that
195
defines the invariant torus (12.1) of system (12.2) is constructed as a limit of

successive Galerkin approximations of a periodic solution of the equation
Lu(J)
= !(J),
(14.1)
where L is the differential operator

m
Lu
=L
(14.2)
o;v av(J) - P(J)u
v=l
defined on the function space Hr(Tm ). Galerkin's approximations are given by

the trigonometric polynomials
WN()
WiN)ei(k,cP)
(14.3)
IIkll~N
the coefficients WiN) of which are solutions of the system of algebraic equations
(14.4)
The conditions under which the Galerkin's approximations exist for any
N
2: 0 and converge to a solution of equation (14.1) as N
+00
can be
obtained from the following statement.

Lemma 1. Suppose that the following conditions hold:
1) the functions a(J) and P(J) belong to the space Cr(Tm );

2) the matrix
b (.4.) = _ P(J) + P*(J) = ~ ~ oav(J) E
o 'I'
2
2 ~ oJv
'
(14.5)
v=l
where E is the identity matrix, is positive-definite for all J E Tm
3) for each s, r 2: s > 1 and arbitrary u E H'+l(Tm ) the inequality

(14.6)
holds, where rand 8 are positive constants independent of u.
= 0, 1,2, ... ) is defined for any function! E Hr(Tm ) and converges in H'(Tm } for 1 :s s < r to
a function u = uo( J} which satisfies the inequality
Then Galerkin's approximation sequence {WN(J)} (N
(14.7)
196
CHAPTER 3
where C is a positive constant independent of Nand f.
To prove the lemma, we denote by S N f( 4 the partial sum of the Fourier

series of the function f(4)) :: Lk fkei(k,<p:
SNf(4))
fkei(k,<p).
II k ll:S N
In this notation system (14.4) is equivalent to the single equation
and has a solution if the equation
has the only solution WN(4)) == O.

Since for any function (14.3) we have
(SN LWN(4)) , W N (4)))o
= (LWN(4, SN W N (4)))0 =
+ L* WN(4)), WN(4)) ) 0 =
=(LWN(4, WN(4)))O = ( -L 2-
=(b O(4WN(4, WN (4)))o

and the matrix bo(4)) is positive-definite, (bo(O)x,x) ~ 10llxll2 (we can assume
that 'Yo
= 'Y), we have the inequality
from which it follows that system (14.4) can be solved for any integer N
and an arbitrary function
f E HO(Tm ).
Moreover, inequality (14.8) leads to the following estimate for the Galerkin
approximations WN:
which by Schwarz's inequality yields
We use inequalities (14.6) to estimate the Galerkin approximations WN(4))

with respect to the Hr(Tm ) norm. Since
197
it follows from (14.6) and Schwarz's inequality that
(14.9)
Solving inequality (14.9) we obtain the estimate
(14.10)
where
rl
= 2r/(1 + VI +4fJh)
By Sobolev's compactness theorem, inequality (14.10) ensures that the sequence of Galerkin approximations is compact in the space H'(Tm ) for s < r.
Let {WNj} (j
1,2, ...) be a subsequence of {WN} that converges in
H6(Tm ) to the limit Wo:
Consider the product
(14.11)
where 9 is a trigonometric polynomial in P(Tm ). If j is sufficiently large, then
SNj9
9 and equality (14.11) takes the form
But then
and by taking the limit we find that
(14.12)
For r
>
s 2: 1 H'(Tm ) C H 1 (Tm ), therefore WO E H 1 (Tm ). It then follows
from (14.2) that (LWO ,g)o
=(f, g)o. Since the function 9 E P(T
m )
by virtue of the theorem on the completeness of the space HO(Tm
is arbitrary,
),
it follows
that this relation can hold only if
LWo
=f
a.e.,
(14.13)
198
CHAPTER 3
1 as elements of the space HO(Tm ).
that is, when LWo is equal to
We shall be using equality (14.13) to prove that the Galerkin approximations converge. To this end we consider an arbitrary subsequence of {WN} that
is convergent in H'(Tm ). Let Wo E H'(Tm ) be its limit. Now equality (14.13)
holds for W 0, therefore
L(Wo - WO) = 0
almost everywhere. But then
which is possible only if

(14.14)
almost everywhere. Relation (14.14) shows that the sequence of Galerkin approximations has only one accumulation point in HO(Tm ). Denote it by uo().
Then lim IIWN - uollo
N-oo
= 0 and, since the sequence {WN} (N = 0,1,2, ...) is
compact in H'(Tm ), we have the limit relation
To complete the proof of the lemma, it remains to find an estimate for
WN
Un.
To do this, using the equality of the form (14.13) for
write down the value of the operator L at WN -
L(WN - un)
= (E -
Un.
= uo(), we
We have:
SN )LWN - (E - SN )/,
where E is the identity operator.

Then, for 0
r - 1,
(L(WN - uo), WN - uo),
= (E -
~ (II(E - SN )LWNII,
For any function
II(E - SN )/11,
+ II(E -
1 E Hr(Tm ), s
= II L
SN)LWN - (E - SN)/, WN - uo),
SN )/II,)IIWN -
uoll,
(14.15)
r, we clearly have the estimate
/kei(k,cP)11. ~ N,-rll(E - SN)/lir ~ N,-rll/llr,
IIkll>N
from which it follows that inequality (14.15) can be written in the form
199
Taking into account the form of the operator L we obtain the following
estimate for IILWNllr-1:
~ C(I: lavlr-dlWNlir + IPlr-dIWNllr-1) ~ C11IWNllr,

v=l
where C1= CO:::'=l

of WN , a v and P.
lav Ir-1 +IPlr-1), C being a positive constant independent
Now by using estimate (14.10), we find that
and from this it follows that we can rewrite inequality (14.16) as:
Setting s
= 0 in
(14.17) and using the fact that
(L(WN
we have that IIWN
uollo
uo), WN - uo)o ;::: 'YIIWN - uoll~,

~
C' N1-rllfllr. If we take 1 ~ s ~ r - 1 in (14.17)
and apply estimate (14.6) to the left hand side of (14.17), we find after simple
calculations that
where C z is a positive constant which depends only on CI,'Y1,'Y,C. Inequality
(14.7) is proved.
By Sobolev's embedding theorem the limit function
UQ
belongs to C/(Tm )
if
s> mJ2 + l.
If we take I;::: 1 in inequality (14.18), we obtain the inclusion
follows from the fact that

LUQ(Ij;)
(14.18)
UQ
E C1(Tm ). It
= f(lj;)
(14.19)
holds almost everywhere, that the latter equality holds for all J E Tm . This
means that uo(J) is a solution of equation (14.1) and belongs to C'(Tm
).
Thus,
if the conditions of Lemma 1 are satisfied for

r
> mJ2 + I,
I;::: 1,
(14.20)
200
CHAPTER 3
then the sequence of Galerkin approximations {WN (4))} (N

fined for any
space C/(Tm )
= 0,1, ...) is de-
E Hr(Tm } and is convergent with respect to the norm of the

n H'(Tm ), s < r, to the solution of equation (14.1) uo(4)) with
convergence rate defined by inequality (14.7).

Sufficient conditions for inequalities (14.6) to hold are known from the
papers of Friedrichs [149] and Moser [89]. If the assumptions of Lemma 1 on
the smoothness of the functions a and P hold, then these conditions can be
replaced by the requirement that the inequalities
(30(4
+ soo(4)) -
tJl(4)) ~ /,
s = 0,1, ... , r,
(14.21)
hold, where / is an arbitrarily small number,
. (oa(4
)
mill
~1/J,1/J ~ 00(4)),
1I,p1l=1
u'/'
max (P(4x,x)
II x /l=l
-(30(4,
(14.22)
We shall prove this statement in the next section, but here we use it to
substantiate Galerkin's procedure for constructing an invariant torus (12.1) of
system (12.2).
Theorem 1. Let a, P E cr(Tm ) and suppose that the inequalities (14.21) hold
lor s
0 and s
r. If r satisfies inequality (14.20) then the sequence of
Galerkin approximations {WN(ifJ)} (N
= 0,1, ...)
is defined for any function
Hr(Tm ) and converges in C/(Tm } n Hr-1(Tm ) to an invariant torus (12.1)

of system (12.2) such that the inequality
IE
(14.23)
holds for any 0
and
~ r -
1 and some positive constant C independent of N
I.
Proof Taking into account the form of the matrix bo( 4 and using inequality
(14.22) we can write
201
But then it follows from inequality (14.21) considered for s = 0 that
(b o(4x,x) ~
I'lI x ll 2
This proves that the matrix bo(4)) is positive-definite for all 4> ETm and this is
sufficient for all the conditions of Lemma 1 to be satisfied.
According to Lemma 1, the sequence of Galerkin approximations {WN (4))}
(N = 0,1, ...) is defined for any function f E Hr(Tm ) and converges with respect to the norm of the space H$(Tm ), s < r, to a function uo( satisfying
inequality (14.7). The limit function uo(4)) satisfies equation (14.1) almost everywhere. If inequalities (14.20) are satisfied, then uo( E C'(Tm ) C C 1 (Tm ),
so that the function uo(4)) is a solution of equation (14.1). Consequently, the
function u = uo( defines an invariant torus (12.1) of system (12:2). The limit
relation
lim
N-oo
Iluo -
WNII$
= 0,
0:$ s :$
r-
implies the convergence of the sequence {WN(>)} (N = 0,1, ...) in the space
CI(Tm ) n H$ (Tm ) to a function uo(4)). This completes the proof of Theorem 1.

Inequality (14.23) gives an estimate of the deviation of the Nth Galerkin
approximation from the exact solution of equation (14.1) with respect to the
norm of the space H$(Tm ).
By using Sobolev's embedding theorem we can estimate the difference
between WN and Uo in the uniform metric as follows:
IWN - uol$ :$ CN-(I-$-I)lIfllr,
0:$ s :$ I.
It should be noted that condition (14.20) is engendered by the method used
to substantiate Galerkin's method. It is typical for functional methods used in

mathematical physics [14], [62]. It can be weakened by making it independent of
the dimension m of the invariant torus by combining the convergence conditions
for the sequence of Galerkin approximations with the conditions for solvability
in C(Tm ) of equation (14.1). This leads us to the following statement.
Theorem 2. Let a, P E C 2 (Tm ) and suppose that the inequalities
inf 13(4)) > 0,
~ETm
inf [f3(4)) + a(4))] > 0,
~ETm
inf [130(4)) - tll(] > 0,
~ETm
inf [130(4))
~ETm
+ 2ao(4)) - ~1l(4))] > 0
(14.24)
202
CHAPTER 3
hold, where !3(Iji), a(lji) , !3o (Iji) , ao(lji) and J1.(Iji) are the quantities defined by inequalities (12.7), (12.8), (14.22).
Then the sequence of Ga/erkin approximations {WN(Iji)} (N = 0, 1, ...) is

defined for any function f E H 2 (Tm ) and converges in H 1 (Tm ) to a function
u(lji) which defines an invariant torus (12.1) of system (12.2).

Proof. According to Theorem 12.2, the first two inequalities of (14.24) imply
that there exists an invariant torus (12.1) of system (12.2); furthermore, it belongs to the space C 1 (Tm ). The last two inequalities of (14.24) imply that the
sequence of Galerkin approximations {WN(Iji)} (N
1
= 0,1, ...) exists and con-
verges in H (Tm ) for any function f E H (Tm ). The limit function uo(lji) of the
= 0,1, ...) satisfies equation (14.1) almost everywhere,

as follows from the statements of Lemma 1 for r = 2 and s = 1.
sequence {WN(Iji)} (N
Since the torus (12.1) is smooth, it follows that
Lu(lji)
= f(lji) , Iji E Tm
Then the difference uo(Iji) - u( Iji) satisfies the equality
L(uo(lji) - u(lji))
=0
almost everywhere. By the third inequality of (14.24),
(L(u( - uo(), u( - uo()o
=(bo(Iji)(u(lji) where
= const > O.
uo(Iji)), u(lji) - uo()o ~ rllu(lji) - uo(Iji)1I6,
But then
Ilv - vollo = 0,
which implies that the functions
and
Uo
are equal for all > E Tm . This proves
Theorem 2.
It should be noted that the inequality
which determines the rate of convergence of the Galerkin approximations
{WN(Iji)} (N
= 0,1,2, ...) as
-+ 00
under the conditions of Theorem 2,
203
follows from estimate (14.7) taken for s = 0 and r = 2 together with the equality uo(4))
= u(4)),
4> E Tm , which holds if the conditions of Theorem 2 are
satisfied.
We also note that the first two inequalities of (14.24) can be replaced by
the requirement that an invariant torus (12.1) of system (12.2) should exist.
This does not affect the validity of Theorem 2.
3.15. Proof of the main inequalities for the
substantiation of Galerkin's method
To render the statements of the preceding section mathematically rigorous, it
remains to show that condition 3 of Lemma 1 of 13 holds when inequality
(14.21) is true for s
= 0 and s = r.
To prove this fact we shall assume that the
functions a and P belong to the space CT(Tm ).

For an arbitrary function u E H'+l(Tm
For s
= 0 we have
L+L'
(Lu,u)o = ( --2- u ,u
) 0 = ((P+P'
2
+ '21
consider the product (Lu,u),.
oa E
L o4>v
v
u,u
= (bou,u)o.
v=l
As was shown in the proof of Theorem 1 of 3.13, it follows from inequality

(14.21) considered for s
= 0 that the matrix bo(4))
is positive-definite for all
4> E Tm
Using the relation given at the end of 1.3 we have for s
(Lu, uh = (Lu, Ku)o = (Lu, u)o + (Lu,
=(bou, u)o + L
)
-~u)o
= 1:
=
(O:j ,Lu, :~) = (bou, u)o+

0
where Evj denotes the n x n matrix equal to E for v
=j
and 0 for v
=I j.
204
CHAPTER 3
We estimate the functional <1>1' Taking its form into account, we have:
(15.1)
where Jl is a sufficiently small positive constant and C is a constant independent
of u and the coefficients of L.
We impose the condition that the inequality
(15.2)
holds for any collection
(II
7]1, ... , 7]m
= 1, ... ,m) and some '1'1
of n-dimensional vectors,
7]v
= (7]~, ... , 7]~)
> O.
We defer until the end of the section the proof that by taking '1'1 sufficiently
small we can make the above inequality hold .
If inequality (15.2) holds, t.hen we have the following lower estimate for 1}:
= '1'1 2::11
v
:;v II: 2:
'1'ICdlulli -
'1'IC21Iull~,
(15.3)
where C1 and C2 are positive constants independent of u and the coefficients

of L.
Using both inequalities (15.1) and (15.3) we obtain estimate (14.6) for
= 1:
205
where r
rlCl - C'JJ2 > 0, /j rlC2 + C'JJJ2 > 0.
For s = 2 we use the relation given at the end of 1.3 to obtain
(Lu,uh
= (Lu,K 2 u)o = (I<Lu,I<u)o = (LKu,I<u)o+

+K L - LK)u, Ku)o
= (boKu, Ku)o + K L -
LK)u, Ku)o.
But
so that we have the following representation for (Lu,
uh:
where L 2 is the second order differential operator:
and
L~
is the first order differential operator, the coefficients of which are the
functions all, P and their derivatives up to second order.

Taking into account the form of the operator
211ull211ulh
given in 1.10, we estimate (L~u, uh
L~
and the inequality for
= c11 2(u) as follows:
IIcll 2(u)1 =1(L~u,uhl ~ IIL~ullollul12 ~ 2C(2: la 12 + 1P12)lIu11211ulh =

ll
II
(15.5)
206
CHAPTER 3
where J.l is a sufficiently small positive constant and C is a positive constant

independent of u and the coefficients of L.
We require that the operator L2 be elliptic. For this it is sufficient that
the inequality
holds for any collection 771, ... , 7]m of n-dimensional vectors
(v
= 1, . .. , m) and some /2 > O.
7]v
= (7]~, ... , 77~)
We defer to the end of this section the proof that this inequality is satisfied
by taking /z sufficiently small.
Then by using Garding's inequality we can make an estimate for (Lzu, uh
from below as follows:
(15.6)
where /1 and 81 are positive constants that depend only on 2:::'=1 lavlz + IFlz.
Combining inequalities (15.5) and (15.6), we obtain an estimate for (Lu, uh
in the required form:
where / and 8 are positive constants independent of u.

Passing to the general case, we denote by L 2 the operator
and prove that for s 2: 2
(Lu, u).
= (L;u, U).-1 + .(u),
(15.7)
where .(u) is a functional admitting the estimate

(15.8)
Here C' = C(2:v Javl. + IFI.), C being a positive constant independent of
and of the coefficients of the operator L.
Since representation (15.7) and estimate (15.8) have been proved for
s = 2, we can assume that formula (15.7) and estimate (15.8) are proved for
all s
= 2, ... ,p because (15.7) is equal to (15.4) and (15.8) to (15.5).
207
Consider
the product (Lu, U)P+l under this assumption. Using the relation given at the
end of 1.3, we have:
(Lu, u)P+l
= (Lu, J{ u)p = (Lu, u)p + (Lu, -Llu)p =
LU ,
=L-CJ:
j
1
:;J + ~~+l(U),
1
where
1~~+l(U)1
= I(Lu,u)pl::; IILullp-lllullp+l ::;
: ; (L-Ila v:;v IIp-l + II Pullp-l) Ilullp+l ::;

v
::; C1(L-lavlp-1 + IPlp-l )lI ullp+lllulip = C'llullp+lllullp

v
Since the following chain of equalities holds:
where
208
CHAPTER 3
I:II
+ .
1
IIII
{)P
{)u
{)Jj u p {)Jj p
II:(
.
11
{)bo
{)Jj 1
u
Ll + PI:. {)Jj{)2 av{)Jj {)Jv{)Jj
{)2
{)U)
I
, {)Jj, p-l ::;
U
1',1
:s c~II%~ IIJ%~ IIp-l + C2 I: lavlp+lllullpliullp+l+

1
+C2 1PIp+dl u llpllull p+1 + C2IPlpliullp+dlullp+

+C2
I: lavlp+lliullp+dlullp ::;
v
we finally have for (Lu, u)P+l the following expression:
-(L~U,U)P_l + ~+l(U) + ;+l(U)
= ((
L~ - I: ~:; {)J~~Jj )u, u)p + ~+l(U) + ~+l(U)+

1',1
where P+l (u) is a functional which can be represented as the sum of three
terms:
~+l(u) + ;+l(u)
+ ;~l(U).
To estimate the functional p+l (u) it only remains to estimate the term
~+l (u)
= -( L~u, U )p-l.
Since
1~~l(U)1
:s IIL~ullp-2I1ullp+l
:s C3 (L lavlp-l +
::;
IPlp-2)lI u llp+lllullp,
it follows from the estimates for the functionals ~+l (u), ~+l (u) that
209
where C is a positive constant independent of u and the coefficients of the

operator L.
By induction we see that representation (15.7) and estimate (15.8) hold
for any 2 :::;
8 :::;
r.
We require that the operator L 2 be elliptic. For this it is sufficient that

the inequality
(15.9)
holds for any collection
(II
1]1,., 1]m
= 1, ... ,m) and some I'
of n-dimensional vectors
1]/1 -
(1]t, ... ,
1]~)
> O.
The proof that this inequality can be satisfied by taking I' sufficiently
small will be given at the end of the section.

Under the assumption that inequality (15.9) holds we apply Garding's
inequality to the product (L 2u,u).-1, which leads to the estimate
where I' and 6. are positive constants that depend on L:" la" I. + IPI.
But then by applying the inequality for 2I1ull.llull.-1 given in 1.10, we
obtain:
(Lu, u). = (L;u, U)._1
+ ~.(u)
I.llull; -
6.lIull~
-2C'llull.llull._1 ~ I.llull; - 6.lIull~ -
c'GJl
2 u
1I ll;+
+ 2Jl2(~'-1) lIulI~)
as required. Thus, if inequality (15.9) holds for
= 0,1, ... , r then condition 3
of Lemma 13.1 holds.

It remains to show that inequalities (15.9) hold for
inequalities (14.21) hold for s

For any 0 :::;
8 :::;
= 0 and s = r.
~ ,llull; - 6I1ull~,
= 0, 1, ... , r whenever
r we have:
" aa"
i i
+s '"
~ aa"
a</;. (1]",1]j ) = "'(b
~ 01]",1]" } +8 '~
a</;_ '"
~1]/l1]j
"j)
"
/lj
CHAPTER 3
210
E<b o7]v,1]v) + s E(E ~:v 7]h~) =

v
vJ
= E<b 0 1]v, 1]v) + s E<~: Wi, Wi),

v
where Wi
= (7]l, ... ,7]:") is an m-dimensional vector.
2: (.80(1/1) - ~Il(<p)
But then
E 1I1]vll 2 + sao(l/I) E II w;l1 2 =

i
= (.80(1/1) + sao(l/I) -
!1l(tP) E l17]vW,
v
and this leads to inequalities (15.9) whenever inequalities (14.21) hold. It is

clear that if inequalities (14.21) hold for s = 0 and s = r, then they also hold
for any 0 < s < r. This completes the proof of the required statement.
Chapter 4. Perturbation theory of an invariant torus of a

non-linear system
4.1. Introductory remarks. The linearization process
The problems of the linear theory considered in the previous chapter are also
important for the general theory of invariant tori of non-linear systems. This
deals with the invariant surface
(1.1)
dJ
dh
di = a(J,h), di = F(J, h),
(1.2)
the right hand side of which is defined, continuous in J, h in the domain

(1.3)
and periodic in Jv (II = 1, ... , m) with period 271". The study of system (1.2)
is complicated by the fact that the functions a and F are non-linear with
respect to the variable h and are accessible by modern methods of investigation
orily within the framework of perturbation theory, where it is assumed that
the quantities lIa( J, h) - a( J, 0)11 and IIF(J, h) - F( J, 0)11 are small in domain
(1.3). Here progress is achieved by using iteration procedures that linearize the
problem at each stage of the iteration.
We shall consider a linearization procedure that enables one to find an
invariant surface (1.1) of system (1.2). To do this we separate the "linear
terms" in the second equation of system (1.2) by writing this system in the
form
dh
dt
= P(J, h)h + 1(<1,
211
(1.4)
212
CHAPTER 4
where
f(4J)
= F(4J,O),
P(4J h)
,
= fi
BF(4J,th) dt
B(th)
.
Jo
We define the zeroth iteration of the procedure by a function u O E G(Tm )

which takes values in the domain (1.3) such that
and define a sequence of tori

(1.5)
each of which is an invariant torus of the system
d4J
dt =
.
a(4J, u'(4J,
dh
dt
= P(4J,u'(4Jh + f(4J)
Lemma 1. Let the functions a, P and f be defined and continuous in
(1.6)
4J, h in
the domain (1.3), and periodic with respect to 4Jv (1/ = 1, ... , m) with period
211'. Suppose that for any i = 0,1, ... system (1.6) has an invariant torus (1.5)
lying in the domain (1.3).
If
uniformly with respect to 4J E Tm , then the limit function u(4J) defines an
invariant surface (1.1) of system (1.4).
Proof We denote by
the solution of system (1.6) that takes the value
at t = 0. Then
4J:
=J +
it
a( J~, ui (J~ dT,
ui+i(JD
= ui+1(J) + (P(J~, ui(J~ui+l(J~) + f(4J~dT
(1.7)
NON-LINEAR PERTURBATION THEORY FOR AN INVARIANT TORUS
213
for all t E R.
Taking into account the fact that the functions u i (4)) are continuous and
periodic and that their values lie in the domain (1.3) for all i = 1,2, ... , we see
that the first equation of system (1.7) implies that the sequence of functions
4>L 4>;, ... ,
4>~,
...
(1.8)
is uniformly bounded and equicontinuous for t from any bounded interval I

on the R axis. This is sufficient for the sequence (1.8) to be compact in the
space of continuous functions on I. Hence we can choose a subsequence that is
uniformly convergent on I. Let 4>~v (II
4>t
its limit:
lim 4>~v
v-oo
= 1,2, ...) be such a subsequence and
= 4>t,
tEl.
Replacing i by i v in (1.7) and then passing to the limit, we find that
4>t = 4> + it a( 4>."

u( 4>d
= u( 4 + it (P( 4>."
u( 4>., ))dT,
u( 4>., ))u( 4>.,)
+ f( 4>., ))dT
(1.9)
for all tEl. Because the interval I was chosen arbitrarily, equality (1.9) holds
for any t E R. Hence the surface (1.1) is an invariant set of system (1.4), as
required.
To realize the given linearization procedure, we need to find conditions
under which the system of equations
dh
di = (P(4 + P1 (4)))h + f(4))

has an invariant torus
= u(4)),
4> E Tm
for arbitrary sufficiently small functions
(1.10)
(1.11)
al (4)),
PI (4)) (with respect to the
Cr(Tm ) norm). These conditions naturally presuppose a certain coarseness

of the "generating" system of equations
d4> = a (A.)
di
'i",
ddht
= P(A.)h,
'i"
(1.12)
which is expressed in terms of the "coarseness" of a Green's function G o(T, 4

for system (1.12).
214
CHAPTER 4
We shall call a Green's function GO(T,I/ for system (1.12) coarse ifthere
exist a constant 6
(where ro
> 0 and an integer r 2: 0 such that whenever
= max(I,r and
al
E CTo(Tm )
(1.13)
the system of equations
dl/>
dt
= a (-I.) + al (-I.)
'I"
'I" ,
ddht
= P(-I.)h,
'I"
(1.14)
has a Green's function G o( T, 1/, for which

(1.15)
where f is an arbitrary function in CT (Tm ), I/>t (I/ a solution of the first equation
and
r are positive constants independent of 1/>, 6 and f.
of (1.14), and
f{
Lemma 2.
Suppose that the system (1.12) has a coarse Green's function
p(6) > 0, p(6) - 0 as 6 GO(T,I/. Then we can find p

for any ai E CTo(Tm ), Pi E CT (Tm ) satisfying the condition
0 such that
(1.16)
and an arbitrary function f E cr(Tm ), the system of equations (1.10) has

invariant torus (1.11) satisfying the condition
(1.17)
where
f{i
is a positive constant independent of p and f.
Proof We choose p so small that inequality (1.13) follows from inequality (1.16)
and system (1.14) has a Green's function G o( T, 1/ that satisfies condition (1.15).
Using the operator G defined on functions f E CT(Tm ) by the inequality
Gf(l/
roo GO(T,I/f(l/>r(I/)dT,
=i- oo
we define the equation

(1.18)
For sufficiently small p, the norm of the operator GPi in the space CT(Tm )
2cf{
IGPdT ~ -IPlI T ~
2cK
-p::; d < 1,
'Y
215
therefore equation (1.18) has the unique solution

00
UJ
= ~(GPd:Gf
k=O
in Cr(Tm ), and this solution satisfies the inequality

(1.19)
It follows from the definition of the operator G that the torus (1.11) with
function
= uJ
is an invariant set of system (1.10). Inequality (1.19) implies
that the estimate (1.17) holds for it.

As in the linear case, an invariant surface (1.1) of system (1.2) will be
called an m-dimensional invariant torus of this system. By the smoothness of
torus (1.1) we mean the smoothness of the function
on the right hand side of
its equation (1.1). The number r in the definition of a coarse Green's function
will be called the smoothness index of this function.
4.2. Main theorem
We shall be considering the following system of differential equations:

d<jJ
di = a(<jJ, h, i),
(2.1)
dh
di = P(<jJ, h,i)h + f(<jJ, i),

where a, P and f are periodic functions in 4>v (11
= 1, ... , m) with period 271"
which are defined and continuous with respect to all the variables <jJ, h, i in the
domain
(2.2)
and are such that
(2.3)
Condition (2.3) ensures that there exists a trivial invariant torus
h
of system (2.1) for
= O.
= 0,
<jJ E Tm
(2.4)
CHAPTER 4
216
We write down the variational equation corresponding to torus (2.4). It

has the form
d4>
dt
dh
= Po(4))h,
= ao(4)), di
(2.5)
where
ao(4)) = a(4),O,O),
For a function u E CLip(Tm
),
Po(4)) = Po(4),O,O).
the quantity
luI. + /{
will be denoted by
lul.,Lip where /{ is the Lipschitz constant for derivatives of order s of the

function u( 4.
Theorem 1. Suppose that the functions a, P and f have partial derivatives
with respect to 4>, h up to order r which are continuous with respect to 4>, h, e
in the domain (2.2). Suppose further that system (2.5) has a coarse Green's
function with smoothness index r.
If r
2: 1, then there exists a sufficiently small eo > 0 such that for any
e E [0, eo] system (2.1) has an invariant torus

(2.6)
where the function u belongs to the space C[;;,I(Tm ) and satisfies the inequality
lulr-I,Lip:S
Klflr,
(2.7)
/{ being a positive constant independent of (.

Proof. We write the system of equations (2.1) in the form
where
al(4),h,e)
P1 (4),h,e)
= a(4),h,e) -
= P(4),h,e) -
a(4),O,O),
P(4),O,O),
and apply the iteration procedure discussed in the previous section to find an
invariant torus (2.6) of this system. We define the zeroth iteration by the
function U o( 4 == 0, 4> E T.n We define the first approximation for the system
of equations
(2.8)
217
as the invariant torus

(2.9)
We choose to > 0 so small that the inequality
(2.10)
holds for any functions al = al(1),u(1>,t),t), PI = PI (1), u(1),c), c), where
u(1),c) is an arbitrary function in Cr(Tm ) satisfying the inequality
(2.11)
Here p and
[(I
Such a choice of the value to is possible since the right hand side
= 1(1),t).
are the positive constants defined by Lemma 2 of 1, and
of system (2.1) is smooth and condition (2.3) holds.

Then the right hand side of the system of equations (2.8) satisfies the
conditions of Lemma 2 of 1 for all c E [0, to]. Consequently an invariant torus
(2.9) of this system exists and satisfies the inequality
(2.12)
where I< 1 is the constant in estimate (2.11). This is sufficient to find the second
iteration satisfying estimate (2.12) by using the first one (2.9).
Suppose that for the chosen co > 0 all the iterations have been found
for i
1, ... ,p - 1, all of them satisfying inequality (2.12). The next pth
approximation will then be defined from the system of equations
d1>
di
= ao(1)) + al(1), Up -I(1), t),t),
(2.13)
dh
dt = [Po (1) ) + PI (1), Up-l (1), c), t)]h
+ 1(1), c)
as the invariant torus

(2.14)
Since the functions al
inequality (2.10) for all
= al(1),up -I(1>,i),c),
i
PI
P1 (1),U p -I(1>,i),t) satisfy
E [0, co], the torus (2.14) of system (2.13) exists and

(2.15)
218
CHAPTER 4
where
[(I
is the constant from estimates (2.11), (2.12).
The mathematical induction principle now guarantees that for all f E

[0, fO], any iteration of the procedure is defined and satisfies inequality (2.15).
We now prove the convergence of the iteration procedure. To this end, we
consider the difference
Since the functions Ui(J, f) belong to the space Cr(Tm ) with r
1, the invariant
torus of the corresponding system of equations determined by these functions

is smooth. Therefore
OUi(J,f)(
oJ
ao (J)+ al(J,ui-l ( J,f )
,f
= [Po (J) + PI (J, ui-tC J, f), f)]Ui( J, f) + f( J, f)

for every J E Tm and i
= 1,2, ....
Hence it follows that the function Wi+l(J, f)
satisfies the equation
ow
oJ
= (uo(J)
+ al(J, Ui(J, f),f =
= [Po(J) + PI (J, Ul(J, f), f)]W + MJ, f),

where
Therefore the torus given by the equation
is an invariant torus of the system of equations
(2.16)
for any
219
E [0, fO]. Since the functions
satisfy inequality (2.10) for f E [0, fO] and Ii E Cr-1(Tm), it follows that the
system of equations (2.16) satisfies all the conditions of Lemma 1.2 with r equal
to the value of r - 1. Estimate (1.17) for
For r
Wi+l (4),
f) takes the form
= 1, this estimate leads to the inequality

(2.17)
where ]{ is some constant independent of i and
f.
Inequality (2.17) and esti-
mates (2.15) for the iterations show that
for all
E [0, fO] and sufficiently small
fO
> O. But then
Iwilo ::; (1/2)i- 1 Iwdo,
= 1,2, ... ,
which ensures that that the sequence Ui(4), f), i

Consequently for each
= 1,2, ... , converges in C(Tm ).
E [0, fO] there exists a function u(4), f) in C(Tm ) such
that
lim Ui(4), C)
1-00
uniformly with respect to 4> E Tm
= u(4),c)
(2.18)
c E [0, co].
Since the sequence Ui(4), C) (i = 1,2, ... ) is bounded in cr(Tm ), it is compact in the space cr-l(Tm ) and this leads to the relation
,
.lim \Ui(4),C) - u(4),C)lr-l = 0,
1-00
c E [O,co].
On passing to the limit in inequality (2.15) with r equal to the value of r - 1

we obtain
lu(4), C)lr-l ::; KI!/lr-l
(2.19)
Finally, since for any (r - 1)th order derivative D r - 1 I( 4>, c) of the function
cr (Tm )
the estimate
IIDr - 1 I( <P, c)
D r - 1 I( <p', c)1I
::; clll r 114> - 4>'11
220
CHAPTER 4
holds with a constant c independent of f and 4>,4>' E Tm , it follows that
for all 4>,4>' E Tm , t E [0, to]. Passing to the limit in inequality (2.20), we find
that
Inequalities (2.19), (2.20) are equivalent to estimate (2.7). Theorem 1 is true

by virtue of Lemma 1.1.
It should be noted that inequality (2.7) implies the limit relation
lim lulr-I,Lip
(-0
= 0,
which proves that the functions u(J, f) and its derivatives with respect to 4> are
continuous with respect to
at the point
= 0 uniformly in 4> E Tm .
4.3. Exponential stability of an invariant torus and conditions for

its preservation under small perturbations of the system
We consider the system of equations (1.2). Suppose that it has an invariant
torus
= u(4)),
4> E Tm ,
(3.1)
a neighbourhood of which lies in the domain (1.3). We denote by 4>t,h t the

solution of system of equations (1.2), which at t
= 0 takes the value 4>0, ho.
We set
(3.2)
IIlJillTmthe Euclidian length of the

: IIlJillTm = min IIlJi mod 21r1l
where we have denoted by

torus Tm
vector lJi on the
In the direct product of the m-dimensional torus Tm and n-dimensional

Euclidian space En, where n is the number of components of the vector h =
(hI, ... ,h n ), the function dt gives the distance between the point (4)t, ht) and
the torus (3.1). Using this distance one can define in the usual way stability,
asymptotic stability, instability of the torus (3.1) considered as an invariant set
of system (1.2).
221
An invariant torus (3.1) of system of equations (1.2) will be called exponentially stable if it is stable and there exists 60
> 0 such that for any solution
Jt, h t satisfying the condition do < 60 we have the inequality
t
where
and, are positive constants independent of
J(
arbitrary number of the half-axis R+
(3.3)
~ T,
T
and do, and
is an
= [0,+00). It follows from the definition
that exponential stability of torus (3.1) implies its asymptotic stability.

We denote the expression
= Pt(Jo, ho).
by Pt
It is clear that dt ~ Pt and that dt
= Pt whenever u(J) =
const.
Lemma 1. If u E CLip(Tm ), then
Pt ~
where
J(
V2(1 + K)d t ,
(3.4)
is the Lipschitz constant for the function u(J).
Indeed, since Tm is compact, it follows that
for some value J; E Tm . But then
IIh t -
Pt ~
u(J;)1I
+ lIu(J;) -
x[llh t
u(Jt)11 ~ (1
u(J;)11 +
x[llh t
+ K)x
114>; - 4>t1IT
u(J;)1I 2
l~
+ IIJ;
V2(1 + K)x
-
4>t1lt P/2 = V2(1 + K)dt ,

m
which proves estimate (3.4).

The number, in inequality (3.3) is called the index of exponential attraction of the trajectories to the torus (3.1).
Lemma 2. Let u E CLip(Tm
4>t,h t
and suppose that for some 60
of system (1.2) for which Po
< 60
>0
any solution
(3.5)
222
CHAPTER 4
with constants
J( 1
> 0 and 1 > 0 independent of Po. Then the invariant torus
(3.1) of system (1.2) is exponentially stable with index of exponential attraction

of the trajectories to the torus (3.1) equal to 1.
To prove the lemma we consider the solutions of system (1.2) for which
Po
< 60 / J(1. For these solutions P,
~ 1<1 e--YT Po
< 60 for any T E R+. Therefore
the solution >t+T' h +T of system (1.2) that takes the value >T' h T at t = 0
'
satisfies the conditions of Lemma 2 for any t E R+, and consequently, estimate
(3.5):
(3.6)
holds for it.
Since PI(>T, h T)
in the form
= IIh' +T-u( >'+T)II = P'+T, inequality (3.6) can be written

t +T ~
and, on replacing t
+T
T, T
E R+,
by t, we obtain
(3.7)
which holds for all t ?:
T,
any
E R+ and those solutions >t, hi of system (1.2)
< 60/1<1.
lt follows from Lemma 1 considered for t
for which Po
which do < 60 /(V2J{1(1
= 0 that all the values >o, ho for
+ K)) satisfy the condition
Po
< 60/1<1. Therefore
inequality (3.8) holds for all solutions >" hi of system of equations (1.2) such
that
(3.9)
and this proves Lemma 2.
Lemma 2 simplifies the verification that the invariant torus (3.1) of system
(1.2) is exponentially stable by reducing it to the verification of inequality (3.5).
We now return to our consideration of system of equations (2.1). We
denote by (3( and o( the quantities defined by inequalities (12.7) and (12.8)
of Chapter 3, where we now set a(J)
= ao(J),
P(
= Po(
and the inf and
223
sup are taken over the subsets fRo and fR l consisting of matrices in the space
Cl(Tm ).
Theorem 1. Suppose that functions a, P and f have partial derivatives with
respect to f/J, h up to order r, which are continuous relatively to f/J, h, ( in the
domain (2.2). Suppose that the inequalities
{30
= ,pETm
inf {3(f/J) > 0,
inf [{3(f/J)
4>ETm
(3.10)
+ ra(f/J)] > 0
hold.
I, then we can find a sufficiently small (0 > such that for any
( E [0, (0] system (2.1) has an exponentially stable invariant torus (2.6) with
If r
the function belonging to the space CD~/(Tm) and satisfying inequality (2.7).
Proof. Since the quantities {3( f/J) and 0'( f/J) are defined in terms of the matrices
Po, oa%f/J, S, SI in the space C l (Tm ), we see that the similar quantities
{36(f/J) and 0'6 (f/J), defined in terms of the matrices Po and oa%f/J+ oaI/of/J for
an arbitrary al E C'"(Tm ) satisfying inequality (1.14), are related to {3( and
0'( f/J) via the inequality
1{3( f/J) - {36 (f/J) I + IO'( f/J) - 0'6 (4)) I < ]{ 6,

where ]{ is a constant independent of 6. For sufficiently small 6
(3.11)
> 0,
inequali-
ties (3.10) and (3.11) lead to inequalities for {36(4)) and O'6(f/J) similar to (3.10).
This suffices for any of the systems (1.13) with a = ao, P = Po to be exponentially stable and for its Green's function G o( T, f/J) to satisfy inequality (12.3) of
Chapter 3. The latter is equivalent to the condition that the Green's function
for system (2.5) be coarse with smoothness index r.
This means that the conditions of our theorem satisfy all the assumptions
of the main theorem, so that there exists an invariant torus (2.6) of system
(2.1) for all ( E [0, (0], the function u defining this torus belongs to the space
C[,i;,I(Tm ) and equality (2.7) holds.

We now prove that the torus under consideration is exponentially stable.
Since the torus (2.6) is, in general, merely Lipschitzian, it is impossible to write
down the variational equation and to conclude on the basis of Theorem 7.1 of
224
CHAPTER 4
Chapter 2 and the estimates obtained in its proof that it is exponentially stable.
This forces us to use in our proof properties of the invariant torus (2.6) related
to the iteration procedure for its construction. According to this procedure:
U(~,i)
= .-00
,lim Ui(~,i)
uniformly with respect to ~ E Tm , f E [0, fO], where
Iudl
Ui E Cl(Tm ),
Klih,
= 1,2, ... ,
OUi~:' i) a(~, Ui-l(~, i), f) = P(~, Ui-l (~, f), i)Ui(~, i) + i(~, f).
We use the properties of the functions
Ui(~, f)
(3.12)
to estimate the quantity
Pt
II h t - u(~t, i)lI, where ~t, h t is the solution of system of equations (2.1),
that at t = 0 takes the value (~o, h o) in the domain
Po
We choose
iO
> 0 and 60 >
< 60 .
so small that for all
E [0, iO] we have the
inequality
60
+ Klflo S d/2,
where d defines the domain (2.2) and
f{
is the constant in inequality (3.12).
With such a choice, a neighbourhood of the point

(2.2), and the solution
~t, h t
is defined for t
(~o,ho)
lies in the domain
> 0 until the moment when (~t, h t )
leaves the domain (2.2).

We fix an arbitrary 6, 60 < 6 < d and define T
>
0 to be the greatest
value such that
Pt < 6 "It E [O,T).
Consider the function Zt

ht - U(~t,f) for t E [O,T). Since U E CLip(Tm ), Zt
satisfies a Lipschitz condition with respect to the variable t and therefore it has
a derivative with respect to t for almost all t E [0, T). But then for almost all
t E [0, T):
,1m
I' [OUi(~O,f) a (A.
(A.)
I1m
= c.t-O
';'0, U ';'0, i + Zo,
'-00
u~
J::>
)] ,
where () = t
+ Tilt,
225
T E [0,1]. We set
It follows from equality (3.12) that
dU(t,t)
dt
= I'1m
~t-O
I' [OUi(8,t) a(".-00

1m
>l"-
V'I'
'l'8,Ui-l
("-))
...... ("'l'8,t ,t +'i
'l'8,Z8,t )]
for almost all t E [0, T).

Now we have
with the obvious value of the constant !( 1. From this it follows that
(3.14)
for almost all t E [0, T). Using relation (3.13), we can write for dztldt the
expression:
(3.15)
where
R(t, u( t, t), Zt, t)
= [P( t, u( <PI> t) + Zt, t) -
P( t, 0, O)]Zt+
+[P(t,U(<Pt,t) + Zt,t) - P(<pt,U(t,t),t)]U(<Pt,t) - lim .Iim
~t-OJ-OO
~i(8,Z8,t).
The inequalities (3.12), (3.14) enable us to obtain the following estimate

for the function R= R(t,u(t,t),Zt,t):
where p(6, to) is monotonically decreasing to zero as 6 -+ 0, to

Inequality (3.16) enables us to estimate
using the condition
(5.8) of Chapter 3:
Po >
IIZtll
= Pt
-+
0.
in standard fashion by
and, as a result, obtain an estimate in the form of

< J( e-(Po-31'(6,cot p 0
Pt _
(3.17)
226
CHAPTER 4
where
/30
is an arbitrary fixed number in the interval (0,,80)' When band fO
are sufficiently small, inequality (3.17) leads to the estimate (3.5) with
(3.18)
But then the conditions of Lemma 2 are satisfied and the invariant torus
(2.6) of system (2.1) is exponentially stable.
It should be noted that the index of exponential attraction of the trajectories to torus (2.6), as can be seen from formula (3.18), depends on the distance
from the point (4)0, h o) to the torus (2.6) and increases as this distance decreases.
We further note that inequalities (2.12), (3.17) together with inequality
(3.9) guarantee that for a sufficiently small bo
fO
> 0 it is possible to choose
= fo(bo) > 0 such that for all f E [O,fO] the inequality

lIu(4), f)1I
< bo/2,
4> E Tm
holds and that any solution 4>1> h t of system (2.1) for which
II h oll < bo
is attracted to the torus (2.6) as t -- +00 according to the exponential law
(3.3).
4.4. Theorem on exponential attraction of motions in a
neighbourhood of an invariant torus of a system
to its motions on the torus
The notions of asymptotic and exponential stability of an invariant torus of
a dynamical system presuppose that the attraction of semi-trajectories of the
system from a small neighbourhood of the torus to the torus itself is "coarse",
namely, the attraction to the torus is considered as an attraction to a set of
points in the phase space of the system. Being invariant, the torus is, apart
from anything else, a set of trajectories of the dynamical system. In this connection there arises the question of when the attraction to the torus has the
characteristics of an attraction between the motions of the system that start
on the torus and in a small neighbourhood of it.
We shall answer this question for system (1.2), assuming that the torus
(1.1) is an invariant set of it.
227
We set
ao(4))
= a(4),u(4>)),
p, (4))
= 8F(4),u(4>)) _
8h
8u(4)) 8a(4),u(4>))
84>
8h
(4.1)
and define the quantities 13(4)) and a( 4 in terms of the functions ao and Po,
as in the previous section.
Let
We adjoin to the system of equations (1.2) the equation

(4.2)
which defines the trajectory flow of this system on the torus (1.1).
Theorem 1. Let the functions a, F, u and their first order derivatives with
respect to 4>, h be defined and satisfy a Lipschitz condition in the domain (1.3).
Suppose that the torus (1.1) along with a do -neighbourhood of it lies in the
domain (1.3) and is an invariant set of system (1.2). If the inequalities
inf 13(4))
ETm
= 130> 0,
inf [13(4))
ETm
+ 0:(4))) >
(4.3)
hold, then for any sufficiently small ( > 0, P > 0 and p > 0 the solution >t, h t
of system (1.2), for which
d(>o,h o) ~ pp,
where
t/Jt
= t/Jt (t/Jo)
114>0 - t/Joll ~
J( pp,
is a solution of system (4.2) for which t/Jo
/J( t/J) = 13(t/J) -
(-
J( jl, J(
= t/Jo(tPo, h o),
being a positive constant inde-
pendent of p and p.
Proof. We make a change of variables in system (1.2), (4.2) by setting
>
= t/J + pO,
= u( + ppz.
(4.5)
228
CHAPTER 4
Equations (1.2) of system (1.2), (4.2) are written in terms of the new variables
as:
dO
P dt
= a(tf; + pO,u(tf; + pO) + ppz) -
a(tf;,u(tf;)),
dz
pp dt = F(tf; + pO,u(1/J + pO) + ppz)-
We set
A(A. 0) =
'/',
P(4), z)
84>
a(tf;+pO,u(tf;+p8)+ppz).
(4.6)
8a o( 4> + TO) d
84>
T,
10
A 1(4),z) =
8u( 1/J + pO)
118a(4),U~~)+TZ)dT,
1
1
8F(4), u~:)
(4.7)
+ TZ) dT _ 8~~) A 1(4), z)
and write system of equations (4.6) in the form:
dO
dt
dz
dt
= A(tf;,pO)O + A
(tf;
+ p,8,ppz)pz,
(4.8)
= P(1/J + pO,ppz)z.
It follows from the assumptions of the theorem and formulae (4.7) that
the functions A, Al and P are defined and continuous and satisfy a Lipschitz
condition with respect to the variables 1/J,O,z,P,P in the region
11011 ~ 1,
IlzlI
~ 1, tf; E Tm
0 ~ p ~ Po, 0 ~ P ~ po,
where Po and Po are sufficiently small constants, Po

be seen from formulae (4.7) that
A(A. 0)
= 8 a o( 4
P(4), 0)
= Po(4)),
,/"
84>'
where ao and Po are the functions (4.1).
1, Po
(4.9)
1. It can also
229
We shall denote by n~(p) the fundamental matrix of solutions of the

homogeneous linear system of equations in normal form, the coefficient matrix
of which is P = P(t), such that
n~(p)
= E, where E is the identity matrix.
Lemma 1. Let the matrices P(t) and P1(t) be defined and continuous for
t
2: T
(or t ~ T). If for any t
2: T (or
t ~ T) and Xo the following inequality
holds:
(4.10)
where is a constant and ;3(t) is an integrable function for t
2: T
(or t
T),
then
IIP11I = max
x
IIP1xll.
II ll=1
We prove this lemma as follows. Using the function
for any t
2:
(or t ~ r), where
which satisfies inequality (4.10), we construct a sequence of functions xn(t, T, xo)
(n = 1,2, ...) by setting
Xn+l(t, T, xo)
= n~(p)xo + jl n~(p)PI(S)Xn(s,r, xo)ds.
Suppose that for n
holds for all t
t 2:
(4.12)
= 1,2, ... ,k the inequality
2: r (or t
r). Then it follows from formula (4.12) that for
(or t ~ r)
IIXl:+I(t, T, xo)1I ~ 12exp{jl ;3(s)ds }lIxoll+
{1
+Ijt 12exp
;3(s)ds }IIPI(s)II12exp{j' ;3(s)ds} x
230
CHAPTER 4
The inequality obtained for I!xk+l(t,T,Xo)1! has the form of inequality

(4.13) for n = k
+ 1.
This is sufficient for inequality (4.13) to hold for any
n ~ 1. For all t ~ T (or t :S T), n 1,2, ... , we can estimate the differences
Xn+l (t, T, xo) - xn(t, T, xo) as follows:
I! X2(t, T,XO) - Xl(t, T,XO)I! :S
:S
Il I!n~(p)Pl(S)Xl(S,
t
T,xo)l! ds l :S
texp{lt fJ(s)ds }It lt IIPl(s)l!dsll!xoIL
Il IIn~(p)Pl(s)(x2(s,T,xo)t
I! X3(t,T,XO) -X2(t,T,XO)I!:S
It follows from this that the sequence of functions xn(t, T, xo) (n = 1,2, ...)
converges uniformly with respect to t, T, Xo.

We set
X(t, T, xo)
= n-oo
lim xn(t, T, xo).
On passing to the limit in equality (4.12) we obtain
X(t, T, xo)
= n~(p + Pdxo.
Similarly we use inequality (4.13) to find that
for all t
lemma.
T (or t
:S T). The last two inequalities prove inequality (4.11) of the
231
From the conditions of the theorem it follows that
IIn:(po)zoll ::; ex p {
IIn~(~~)ooll::;
-1"
1 ex
.B(1/>r )dr + f(S - t) }lIzoll,
p{l a(1/>r)dr+f(s-t)}1I001l
(4.14)
2:: t 2:: 0, arbitrary f > 0, some 2:: 1, 1 2:: 1 and any Zo E En, 00 E
Em and Po = Po(1/>t). Here, oa%r/> = oao(1/>d/or/>, where 1/>t is the solution of
system (4.2) that takes the value 1/>0 E Tm at t = 0.
For arbitrary 1/>, z, j.L, p in the region (4.9) the functions A(1/>,j.LO) and
for all
P(1/>
+ j.L0, j.Lpz)
satisfy the estimate
IIA(1/>,j.LO) - A(1/>,O)II::; Kj.L,
II P (1/> + j.L0, j.Lpz) - P( 1/>,0)11 ::; K j.L(1 + p),

where K is a positive constant independent of J.L and p. Therefore, for any
functions O(t), z(t) that are continuous for t 2::
1I0(t)11 ::; 1,
IIz(t)ll::; 1,
and satisfy the condition
t 2:: 0,
(4.15)
the matrices n:(po + P) and n~(oa%r/> + A), where
= P(1/>t + j.LO(t),
j.Lpz(t - Po(1/>t),
A = A(tPt , j.LO(t - A(1/>t,O),

admit the estimates
IInHPo + P)zoll
~ ex p {
Iln~ (~~ + A) 00II ::;

for all S 2:: t 2::
-1"
{l
1 ex p
{3(1/>t)dr + (f + [(1 + p)j.L)(s - t) }lIzoll,
a(1/>r )dr + (f + 1 [(j.L)(S - t)} 11 0011

(4.16)
and arbitrary Zo,Oo,j.L,p in the region (4.9).
We define the sequences of functions On(t), zn(t) by setting Oo(t)
zo(t)
Zo
On(t)
= 0,
and
= -p
1+
00
n~(An-dA1(1/>" + j.L On-l(S), j.LPZn(szn(s)ds,
Zn(t) = n~(Pn-dzo, n = 1,2, ... ,
t 2:: 0,
(4.17)
232
CHAPTER 4
where
We now show that for an appropriate choice of the values flo, Po the
functions (4.17) are defined for any n
o ::; fl ::; flo,
1 and allfl, P, Zo in the region
0::; P ::; Po,
(4.18)
IIzoll::; 1/ .
For n = 1 it follows from formulae (4.17) and inequalities (4.14) that
if J.l, P, Zo belong to region (4.18) and
< f3o.
For n = 1 we have the following estimate for the integrand in the first of
equalities (4.17):
I
IIn~(Ao)Al(1P"flPZl(Szl(s)lI::; .c.c1M1 exp

xex p
{-l'
{1
a(1PT)dr} x
f3(1PT)dr+f(s-t)+fs}lIzoll,
s~t2:0,
where M 1 is a constant chosen from the condition that
in the domain (4.9).

For s
it
0 the inequality
a(1PT )dT -
l'
f3( 1PT )dr
-1'
::; -1'(8 - t)
holds, where l'
= 4>ET
inf
[f3()
[a(1PT) + f3( 1PT )]dT -
-it
+ a()] > O.
f3(1PT)dT
Therefore
it
f3( 1PT )dT ::;
233
for s 2: t 2: 0, which leads to the uniform convergence of the integral (4.17) for
( < r/2,
= 1, and to the following estimate for its value Ol(t):
(4.19)
We set
l =
+ J(J-Lmax(2.c,.cd, r' = r -
and require that the quantities
2l
J-Lo, Po satisfy the conditions
( + J( J-Lo max(2.c,.cd ::; 130/ 2,
(4.20)
for all t 2: 0 and all J-L,P, Zo in the region (4.18), (4.20).

Suppose that the functions On(t), zn(t) are defined and satisfy the inequalities
IlOn(t)lI:s;
for t 2: 0, n
2.c.c 1M 1P
{
exp -}o13(tPT)dT+lt
}II zoll,
IIzn(t)1I :s; .cexp{
-it
= 1,2, ... , k and
all J-L, p, Zo in the region (4.18), (4.20). Then
(4.21)
13(tPT)dT + a }llzoli
IIOk(t)11 :s; 1, IIzdt)1I :s; 1 for t 2: 0 and J-L, P, Zo in the region (4.18), (4.20),
and the function Zk+l(t) is defined and satisfies the inequality
while the function Ol;+l(t) is defined and satisfies the inequality
II Ok+l(t)11
::; .c.c1M1P
+00
exp{}o o{rpT)dT + ( + .c1J(J-L)(S - t)}x
x ex p { -
i'
13(tPT )dT + ( + 2.cK J-L)S }dsllzoII
::;
234
CHAPTER 4
+00
ex p { -[1 - 2{ - (.e l
{1
I
::; 2IM P exp 1
for t
+ 2)Kp](s - t) }dsllzoll ::;
{3(tPT)dr+ft } Ilzoll
0 and all p, p, Zo in the region (4.18), (4.20).
According to the mathematical induction principle, it is sufficient that the

functions On(t), zn(t) be defined and satisfy inequalities (4.21) for t
0, any
1 and all p, P, Zo in the region (4.18), (4.20).

We now prove that sequence (4.17) converges. To do this, we consider the
difference
Since rn+llt:o= 0 and
dTn+1
= PnZn+l(t) - Pn-Izn(t) = PnTn+l + (Pn - Pn-dzn,
we have the following representation for rn+l:
Tn+l
it n~(Pn)(Pn
Pn-dn~(Pn-dzods,
from which, using estimates for the matrices in the integrand, we obtain the
inequality
IITn+l1l ::; 21
t
exp{
-1
{3(tPT )dr + ({ + 2Kp)(t - s) }llPn - Pn-t1lx

xex p
::; 2 t ex p
{-l
{3(tPT )dr +
{-l'
{3(tPT)dr+({+2J<p)s} ds lizoll::;
n} max IlPn 'E(O,t]
Pn- dlllzoli.
~ O.
(4.22)
According to our assumptions and the adopted notation,
IlPn -
Pn-t11 = II P (tPt
+ j.dJ,,(t), PPZn (t)) - P(tPt + pOn-l (t), PPZn-l (tnll ::;
::; /(Jl[llO,,(t) - On-l(t)1I
+ pllrnlll
235
This inequality enables us to obtain from (4.22) the estimate
IIrn+11l :s 2 f{p.t exp{
-it
f3(1/Jt)dr
+ ft} x
[lIOn - On-Illo + II rnllo]lI zolL t ~ 0,
(4.23)
where 11110 = sup 1111, R+ = [0, +00).

tER+
We set
and consider the difference 'II n+!-'II n
'II n]t=.
= 0 and
= [n~(An)-n~(An-d]Oo.
we have the following representation for 'II n +1
Since ['lin+!-
'lin:
from which, using estimates for the matrices in the integrand, we obtain
lI'11 n +1 -
'lin II
xex p
:s r
{l
I' {l
ex p
o:(1/Jddr + (c
+ If{p,)(r -
T
O:(1/JT)dr+(c+lJ{p.)(s-r)}drllOoll,
t) }IIA n
s~t~O.
An-Illx
(4.24)
Since
we obtain from inequality (4.24) the final estimate:
lI'11 n+1 -Wnll:S rKp,(s -
t)ex p
{l
o:(1/JT)dr+ (c+
+If{P,)(S - t) } liOn - On-dlollOoll
for s
O. By using this in combination with (4.23), we estimate the
difference en+! = On+l(t) - On(t). For t
0 we have:
236
CHAPTER 4
1+
xAI(tP,
00
+ fl Bn(S),flP Zn+I(sznH(s)IIds+
IIn~(An-dll IIA (tfJ, + flBn(S),flPZn+I(S(Zn+I(S) 1
1+
00
+p
Zn(sII ds +
IIn~(An-I)II IIA I(tfJ, + flBn(s),flPZnH(S-AI(tP,
+ flBn_I(S),flPZn(slIllzn(s)llds.
(4.25)
The first term It in inequality (4.25) has the following estimate for t 2: 0:
It ;:; pCiCl< MIfl
1+
00
exp{lt a(tfJT )dT + ({ + C1l<fl)(S x exp{
;:; pciCl<MIfl
1+
00
-1'
,B(tfJT )dT + ({
exp{ -/'(s -
2CiCl< M1Pfl
{it
exp -
+ 2Cl<fl)S }dsIlCn+dlollzoli ;:;
t) }dsx
x exp{
;:;
t)} x
-it
,B(tfJT )dr + (t }IICnHllollzoII
,B(tfJT)dT+(t
}IICn+dlollzoll,
;:;
where
max IIA I (,flPZ)II
IIzll~I
< MI.
-
For the second term 12 in inequality (4.25) we have the following estimate
for t 2: 0:
h;:; PCIC2KMIfl1+00 sexp{lt a(tPT)dT+({+CIKfl)(S-t)}x

x exp{
-1'
,B(tPT )dT + ({ + 2Cl<fl)S }dsx
roo sexp{
x[IlCnllo + II rnllolll zoll ;:; pC I C2K MIfl 1

t
xexp{
;:;
2C C2 l<M
2
Iflp(-+t)exp{-
-/'(s - t) }dsx
-it + }mCnllo + Ilrnllolllzoll ;:;

it ,B(tPT)dT+(t}lIICnllo+llrnllolllzoll.
,B(tfJT )dr
(t
237
The third term 13 in inequality (4.25) has the following estimate for t ;::: 0:
where
](1
is the Lipschitz constant for the function A 1 (1/l, z).
Inequality (4.25) now leads to the estimate
IICn+dl S IlPC(1 + t) exp{ - i t .B( 1/'T )dr + it }lIzoll x

x(IJCn il o + II rnllo + pllrn+dlo), t;::: 0,
(4.26)
where C is a constant independent of Il, P, f chosen in obvious fashion.

Let J = sup (1 + t)e-(!3 /2)t. It follows from inequality (4.23) that
tER+
(4.27)
On applying (4.26) and (4.27) to
IICn+dl
we obtain a final estimate:
Inequalities (4.27), (4.28) prove that
for all Il, P, Zo in the region (4.18) with sufficiently small J1.0 and PO.
Inequality (4.29) holds for any 1/'t = 1/'1(1/'0). Thus it follows from this that
the sequences of functions Bn(t) and z,,(t) (n = 1,2, ...) converge uniformly
with respect to t,1/'o,zo,Il,P in the region
t E R+, 1/'0 E Tm ,
II zoll S 1/,0 S Il S J1.0,
S P S Po
(4.30)
We set
B(t) = lim Bn(t),

n--+oo
z(t) = lim zn(t)

n-oo
(4.31)
238
CHAPTER 4
and look at certain properties of the functions O(t), z(t). First of all by taking
the limit in inequalities (4.21) we find that
{1
1
1I0(t)lI:S 2lM P exp -
IIz(t)1I
:s exp{
-1
f3(tPT)dr
+ it } IIzoll,
(4.32)
f3(tPT )dr + {t }lIzoll
for all t,tPo,Zo,J.l,P in the region (4.30).

It follows from formulae (4.17) that
dOn(t)
-;It = A(tPt,J.l0.._l(tO.. (t) + pA 1 (tPt + J.l 0n-l(t),J.lp zn(tz..(t),

(4.33)
dzn(t)
-;It = P(tPt + J.l0.. _l(t),J.lPZn_l(tZ.. (t), t
~ O.
Since the right hand side of equalities (4.33) has a limit as n
(Xl
uni-
formly with respect to t, tPo, zo, J.l, P in the region (4.30), the functions O(t), z(t)
have derivatives with respect to t given by:
dO(t)
dt
dz(t) _ lim dzn(t).

dt
.. _00 dt
lim dOn(t),
"-00 dt
By taking the limit in equalities (4.33) we see that the functions O(t), z(t) are
solutions of equations (4.8) for
tP
= tPt,
Let us study the dependence of the functions 8(t), z(t) on the parameters
J.l, P and the values tPo, zoo Since the limit (4.31) is uniform with respect to
t,tPo,Zo,J.l,P in the region (4.30), the functions O(t),z(t) are continuous with
respect to these same variables t,tPo,Zo,J.l,P in the region (4.30) whenever the
functions On(t), Z.. (t) are continuous with respect to t, tPo, zo, J.l, P in the region
(4.30) for n
= 1,2, ....
Under our assumptions concerning the torus (1.1) and the right hand side
of system (1.2), the function tPt = tPt(tPo) is continuous in (t, tPo) E R+ x

Tm . But then the matrix Po = P(tPt(tPo),J.lpzo) is continuous with respect to
t, tPo, Zo, J.l, P in the region (4.30). It follows from the theorem on the continuous
dependence of solutions of differential equations on the initial conditions and
parameters that the function Zl(t) is continuous with respect to t,tPo,Zo,J.l,P
in the region (4.30). Formulae (4.17) taken for n = 1 imply that the integrand
239
function defining 8 1 (t) is continuous with respect to t, tPo, zo, Jl, p in the region
(4.30) and s
~ 1.
Since the convergence of integral (4.17) is uniform for n
= 1,
the function 81 (t) is continuous with respect to the variables t,tPo,zO,Jl,P in

the region (4.30).
Continuing in this manner, we prove by induction that the functions
8n(t),zn(t) are continuous with respect to the variables t,tPo,zO,Jl,P in the

region (4.30) for n ~ 1.
We set
0(0)
= 8(tPo,zo,Jl,p).
(4.34)
As was proved before, the solution </J(t) , h(t) of system (1.2) for which
</J(O)
= tPo + Jl8( tPo, Zo, Jl, p),
h(O)
= u(</J(O)) + Jlpzo,
(4.35)
satisfies the inequalities
II</J(t) - tPtll
~ .c2JlP exp{
IIh(t) - u(</J(t))11
-I
-I
t
f3( t/JT )dr +
~ .cJ.lpex p{
{t} II zoll,
t
f3(t/JT)dr
+ ft }lIzoll. t
~0
(4.36)
where 2 = 2.clMI!'Y.
Using the fact that the function u( </J) in (4.36) is smooth we obtain the
new inequality
II h(t) - u(t/Jdll ~ II h(t) - u(</J(t))11
+ J(211</J(t) -t/Jtli
~(.c+J(22)JlPexp{-ltf3(tPT)dr+ft}lIzoll. t~O.
(4.37)
According to the second equality of (4.35) we have
JlPIiZoli
= IIh(O) -
u(</J(O))II:s II h(O) - u(t/Jo)11 + lIu(</J(O)) - u(tPo)lI:s
:s (1 + K 2 )[lIh(0) - u(t/Jo)1I + II</J(O) - tPolI].
(4.38)
Inequalities (4.36)-(4.38) lead to the estimate
1I(t) -t/Jdl
~ (1 + J{2)(.c + .c 2+ [{ 2.c2) exp{
+ IIh(t) -
-I
u(tPt)11 ~
f3( t/JT )dr + ft} x
x [II </J(O) -t/Joll + II h(O) - u(t/Jo)1IJ
(4.39)
240
CHAPTER 4
for all t,t/;o,Zo,J.l,P in the region (4.30).

We fix the values J.l, P and take any tPo, h o such that
IIh o - u(tPo)11 s J.lp/C-.
(4.40)
We claim that for such a choice of tPo, ho the equation
tPo
= t/; + J.l(} ( t/;, ho - J.lPu(tPo) ,J.l,P)
(4.41)
has a solution. To prove this, we replace t/; by the variable
1')
= tPo -
t/; and
rewrite equation (4.41) as
1J=J.lB ( tPo-1'),
ho - u(tPo)
)
J.lp
,J.l,p.
(4.42)
It follows from inequalities (4.32) and (4.41) that
II (J (tPo-1J,
Therefore for
1')
ho - J.lP
u( tPo) ,J.l,p)
II S.c 2 p.
in the region
(4.43)
the right hand side of equation (4.42) defines an operator which maps the
ball (4.43) into itself. Since this operator is continuous in
1'),
it follows from
Brouwer's theorem [1] that it has a fixed point in (4.43). This is equivalent to
the fact that equation (4.42) has a solution in the region (4.43). We denote it
by
1')0
= 1J o(tPo,h o).
The quantity
is now a solution of equation (4.41), which satisfies the condition
241
But then each value <Po, h o for which inequality (4.40) holds admits the
representation
<Po =
tPo + 11,{}(tPo, Zo,J-l, p), ho = u(<Po) + J-lpZo,
where tPo <Po -1J o(<Po, ho), Zo [ho - u(<Po)]/(J-lp).

Thus the solution <Pt, ht of system (1.2), for which <Po, ho satisfies inequality
(4.40), satisfies the inequality of the form (4.39):
lI<pt -
tPtll + II ht - u(tPt)1I ::; (1 + 1(2)(.c +.c 2 + 1< 2.c 2) x
x exp{
-it
f3(tPT(dr
+ (t }m<po - tPoll + II ho- u(tPo)lIJ,
where tPt is the solution of system (4.2) for which
tPo
= <Po -
2: 0,
(4.44)
1Jo(<Po, ho).
It follows from inequality (3.4) that all points (<Po, ho) in the neighbourhood
of torus (1.1) defined by the inequality
d(A.. h) <
'/'0,
v'2(1
J-lP
+ /{2).c
(4.45)
satisfy estimate (4.40). But then any solution <Pt, h t of system (1.2), for which
(<Po,h o) belongs to domain (4.45) satisfies inequality (4.44). Obvious transformations of the constants J-l, p,.c,.c 1 , .c2, K 2 complete the proof of the theorem.
An example of a system of equations satisfying the conditions of the above
theorem is system (2.1) when it satisfies the conditions of Theorem 1 in 3 for
r
= 2 and {o sufficiently small.
There is an attraction described by inequality
(4.4) between the motions of such a system that start on the torus (2.6) and
those that start in a small neighbourhood of it.
4.5. Exponential dichotomy of invariant torus and conditions for
its preservation under small perturbations of the system
Ail invariant torus (1.1) of system (1.2) will be called exponentially dichoto-
mous if for any sufficiently small { > 0 there exists 6 = 6({) > 0 such that
the {-neighbourhood of torus (1.1) contains one-side invariant sets M+ and
M- of system (1.2) (where for at least one of them its intersection with the
n-dimensional plane <P
lowing property.
= <Po
is non-empty for each <Po E Tm ) with the fol-
Let <pt,h t be a solution of system (1.2), that starts in a
242
CHAPTER 4
t5-neighbourhood of torus (1.1). Then: if (o, h o) E M+, then (t, ht) E M+

for all t 2: U and
(5.1)
if(o,h o) E M-, then (t,hd E M- for all
t:s 0 and
(5.2)
while if (o, h o) tt. M+ U /,,f-, then

(5.3)
for some t == T 1 <
and some t == T 2 > 0,
\Ii
here R-
= (-00,0], J{ and rare
positive cOllstants independent of o, h o , f, and
is the distaJlce from (t, ltd to the torus (1.1).

We write down the
SYStl~1I1
of variational equations corresponding to the
invariant torus (1.1) of system (1.2):
doh
di = a(1/>, 11('11')), dt =
where
P. (.1. (.1.))
o 0/, U 0/
= of(lP,u(1/>))
&h
Po( 1/" u( 'II' t5h,
_ Du(v') &a(lP,u(1/J))
&1/J
ah
(5.4)
Theorem 1. Suppose that the right hand side of system (1.2) is defined and
continuous and has continuous first order partial derivatives with respect to
, h in the region (1.3) which satisfy a Lipschitz condition. Suppose further

that the torus (1.1) along with a neighbourhood of it lies in the region (1.3) and
is an invariant set of system (1.2), and that u E CLip(Tm ). If there exists a

non-singular symmetric matrix S() E C 1 (7;,,) for which the matrix
S()
= &~~) a(, u()) + S()Po(, u()) + P;(, u(S()
const1'ucted in accordance with the variational equation (5.4) is negative-definite

for all E T m
then torus (1.1) is exponentially dichotomous.
243
Proof. By means of the change of variables
h = u(4)) + z
(5.5)
we transform system (1.2) to the form
d4>
di = a(4),u(4>) + z),
dz
di =
(5.6)
P(4), u(4)), z)z,
where
A.. (A..) ) = II [8F(4), u(4))

P( 'I',U'I' ,z
10
8h
+ tz)
_ 8u(4)) 8a(4), u(4))

84>
8h
+ tZ)]d
t.
The right hand side of system (5.6) is defined and continuous and satisfies
a Lipschitz condition with respect to 4>, z in the domain
(5.7)
where 60 is a sufficiently small positive number. We choose a function Xl (OJ)

from the conditions that
(5.8)
and set X(O) = (xI(Od, .. ,XI(On, 0 = (0 1 " " ,0,,).

Instead of system of equations (5.6) we consider its 'extension":
dO
dt
= P(4),u(4>), X(OX(O),
d4>
= a(4), u(4)) + X(O,
di
(5.9)
dz
di = P(4), u(4)), X(Oz.

This extension has the locally invariant set
O=z,lIzll<{I'
(5.10)
244
CHAPTER 4
the motion of which is given by system of equations (5.6).

It follows from the conditions of the theorem that the trivial torus 6h =
0, t/J E Tm of system (5.4) is exponentially dichotomous.

Since
II P (4>, u( 4, x(O))x(O)1I + lI a(4>, u( 4
+ X(O - a( 4>, u( 4> ))11+
+IIP(4), u(4)), X(O - Po(4), u(4)))11 <
J{ 60 ,
4> E Tm , 0 E Tm
where K does not depend on 4>,0,60, the matrix 5(4),0) constructed from S(4))
according to equations (5.9) and having the form
S(4),O)
A
oS(4))
= ar[a(4),u(4>)
+ x(O)) -
a(4),u(4>))]+
+S(4))[P(4>, u(4)), x(O)) - Po(4),u(4>))]+

+[P*(4), u(4)), x(O - P;(4),u(4>))]S(4>)
+ 5(4)),
is negative-definite for all 4> E Tm , 0 E Tn and any sufficiently small 60 > O. By

Theorem 1 of 3.8, this implies that the trivial torus
= 0,
4> E Tm , 0 E Tn of
system (5.9) is exponentially dichotomous.

It then follows from Theorem 2 of 3.9 that there exists a unique Green's
function G o(T, 4>, 0) for system (5.9), which satisfies the estimate
where
J{ 1
and I are positive constants independent of (4),0) E Tm x Tn. The
matrix C(4), 0) = Go(O, 4>, 0) E C(Tm x Tn) is a projection of rank r where r is

the number of positive eigenvalues of the matrix S( 4. The separatrix manifolds
Mit and Me; of system (5.9) are defined by the equations C( 4>, O)z = z and
C( 4>, O)z = 0, 4> E Tm , 0 E Tn respectively. The solutions 4>t, Ot, Zt of system
(5.9) satisfy for some T = T( 4>0,00 , Zo, cd > 0:
IIZtll
~ K1e--r(t-T)lIzTII, t 2:
IIztll
J{le-r(t-T)lIzTII,
IIztll 2: el, ItI 2: T
t ~
T, T
T, T
for (4)0,0 0 , zo)
E R+ for (4)0,0 0, zo) E Mit,
(5.11)
E R- for (4)0,0 0, zo) E Me;,
(5.12)
f/. Mit U Me;.
(5.13)
245
The restriction of these manifolds to the locally invariant set (5.10) gives
the sets
Mt
and M I consisting of the points (4J, z) of the region

IIzll < <I, 4J E Tm ,
defined by the equation
(5.14)
=Z
(5.15)
C(4J,z)z=O
(5.16)
C(4J, z)z
and the equation
respectively. It follows from inequalities (5.11) and (5.12) that the solution
= Zt,
Zt of system (5.9) for which IIzoll < <I1K l takes values in the region
(5.14) for t ~ 0 if (4Jo, ()o = Zo, zo) E Mri, and for t :5 0 if (4Jo, ()o = Zo, zo) E
Mo. But then (4Jt,()t = Zt,Zt) E Mri for all t ~ 0 if (4Jo,()o = zo,zo) E Mri,
and (4Jt,()t
Zt,Zt) E Mo for all t:5 0 if (4Jo,()o zo,zo) E Mo
4Jt,
()t
lt follows from this that the following statements hold for the solutions
4Jt,Zt of system (5.6) for which IIzoli < <I1Kl:

1) if(4Jo,zo) E Mt, then (4Jt,Zt) E Aft for all t ~ 0 and 4Jt,Zt satisfies
inequality (5.11),
2)
if (4Jo,zo) E M I
, then (4Jt,Zt)
E M I for all
t:5 0 and 4Jt,Zt satisfies
inequality (5.12).
Since u(4J) E CLip(Tm ), it follows from Lemma 3.1 that the inequality
IIzolI <
f1
holds whenever
where K z is the Lipschitz constant for the function u( 4J).

We set < = <d(V2(1
+ [{z)).
Then the <-neighbourhood of torus (1.1) is
contained in the region IIzoli < <1, 4J E Tm . We denote by M+ and M- the

restrictions of the sets Mt and M I to this i-neighbourhood. The sets M+
and M- are locally invariant sets of system (1.2). This follows from the fact
that some trajectory arc of the motion 4Jt, h t of system (1.2) which starts in
an i-neighbourhood of torus (1.1) belongs to this neighbourhood and hence, in
view of the local invariance of Mt and M I , it belongs to M+ if (4Jo, ho) E M+
and to M- if (4Jo, ho) E M-.
We set 8(t)
= i/(V2(I+[{z)K 1) and consider the solutions 4Jt, ht of system
(1.2), for which (4Jo, ho) belongs to the domain do < 8(f). Since for such
246
CHAPTER 4
IIzoll = Po
IIzoll <
(IK l < (t/Kt, it follows

from the inclusion (o,h o) E M+ that (t,h t ) E M+ for all t 2:: 0 and
solutions
(5.17)
while it follows from the inclusion (o,h o) E M- that (t,h t ) E M 1 for all
0 and
(5.18)
Inequalities (5.17) and (5.18) are sufficient for (o, ho) E M+ to imply
the inclusion (t,h t ) E M+ for all t 2:: 0, and for (o,h o) E M- to imply the
inclusion (t,h t ) E M- for all t ~ O. Then inequalities (5.11) and (5.12) lead
to the estimates (5.1) and (5.2) with
f{
= \1'2(1 + J(2)K l .
The required properties of the motions t, ht of system (1.2) that start in

a 8()-neighbourhood of torus (1.1) on the manifolds M+ and M- are established.
Let t,h l be a solution of system (1.2) for which (o,h o) M+ U M-,
do < 8. We consider a solution 1(O,ZO,zo), Ot(o,zo,zo), Zt(o,zo,zo) of
system of equations (5.9), that takes the value o, Zo, Zo at t
O.
Since
(o, Zo, zo) Met U Me; and IIzoll < 6, there exist numbers T l < 0 and T 2 > 0
such that
IIZt(o,zo,zo)1I < (1,
t E (Tl ,T2)'
IIzT,(o,zo,zo)1I = IIzT.(o,zo,zo)1I =
(5.19)
(1
The functions 1(O'ZO'zo), ZI(O,ZQ,zo) define a solution t,ZI of system of

equations (5.6) for t E (Tl , T2)' The inequality
together with relations (5.19) proves that
for t = Tt, and t = T 2 and hence establishes relation (5.3).

To complete the proof it remains to show that at least one of the sets
M+, M- is non-empty for any E Tm . Consider the set Mt for an arbitrary

fixed value of = o E Tm . Let S( tPo, z) be the matrix transforming C( o, z)
to the Jordan canonical form
= diag{El,O},
where El is the r x r identity
247
matrix and 0 < r < n. For asufficient.ly small <] in the region (5.14) the matrix
S( 4>0, z) is continuous in z and satisfies the inequality

IIS(4)o, z)1I
where M
:s M,
= </>ETm.llzll$<,
max
IIS(4),z)1I does not depend on 4>0.
We set
= S-](4)o, z)z
(5.20)
and write equation (5.15) in terms of the variables x = (x], X2), where x] is an
r-dimensional and X2 an (n - r)-dimensional vector. Using the representation
C(4)0, z) = S(4)o,z)~S-](4>o, z) we obtain instead of (5.15) the equation

~x
=x
or, taking into account the form of the matrix

X2
~,
= o.
In the x coordinates the set Mt for 4>
(5.21)
= <Po
consists of points satisfying
:s
equation (5.21) and belonging to the image D of the ball IIz'l

<] under the
map z -> x given by equality (5.20). These points x have the form x = (c,O)
where c
= (c], ... , c
r )
is an r-dimensional vector. We show that Mt contains
the set of the M] which is the inverse image of the set of points x
the hall
IIcll < 6], under the map
->
= (c, 0) of
given by equality (5.20), and that this
set AI] is non-empty for sumeiently small 6] > O.
Accordingly, the set Af l consists of points z belonging to the ball II z ll
< <]
that are solutions of equation (5.20) for x
= (c, 0), where IIcli < 6].
:s
We
write the equation for the definition of the points of the set M I in the form
(5.22)
where S](4)o,z) is the block of the matrix S(4)o,z) = (SI(<PO,Z), S2(<P0,Z of
appropriate dimension.
We choose 6]
:s (IiM.
a continuous map z
->
Then the right hand side of equation (5.22) defines
z of the ball
IIzll :s (]
into itself, since
248
CHAPTER 4
= z(c) in the ball IIzll ::;
This suffices for equation (5.22) to have a solution z

(1.
z(e)
Since the matrix S(<Po, z) is non-singular, Sl (<Po, z)
'# 0 for e,# O.
'# 0
and consequently,
The set M 1 contains the points z(e) for all lIell ::; 61 . These
points satisfy the condition IIz(c)1I ::;
and the inequality z(c)
(1
'# 0 for e,# O.
Thus Mt is non-empty for any J <Po E Tm

We set h(e) = u(<Po) + z(c) and consider the point (<Po, h(c)). The distance
do = do(Jo,h(c)) from this point to the torus (1.1) satisfies the estimate
For
(1
<
the point (<po,h(e)) belongs to the i-neighbourhood of torus (1.1).
This proves that the set M 1 considered for

Since h(c)
'# u(Jo),
it follows that 111+
By writing equation (5.16) for J
where C1 (<p,z)
=E-
The cases r
=0
< i is contained in the set M+.
for any J
= <Po in the form
=<Po E Tm .
'#
0 for <P = <Po we see that M- is non-
= nand r = 0 are included in the case under consideration
with the sole difference that for

and for r
(1
C(<p,z), (E is the identity matrix) and repeating the
argument used to prove that M+

empty for any J E Tm
'# 0
=n
the equality C1(Jo, z)
we have the equality C(Jo, z)
= E.
a neighbourhood of the torus (1.1) and M-
= E,
In the first case M+ contains
= 0,
while in the second case
M+ = 0 and M- contains a neighbourhood of the torus (1.1).
Next we consider system (2.1). We shall assume that the invariant torus
(2.4) of the generating system (2.1) is exponentially dichotomous for
= 0 by
virtue of the variational equations (2.5) and that this is established by using a
non-singular symmetric matrix S(J) E C 1 (Tm ) for which the matrix
S(<p) =
O~~) ao(J) + S(J)Po(<P) + P;(<p)S(J),
formed by using the variational equations (2.5), is negative-definite for all <P E
Tm . We define the positive constants 13, Jl, a from the conditions

max (S(J)h,h) ::; -213,
IIhll=l
max I(S(J)h,h)1
< Jl,
IIhll=l-
1([Sl(<p)'~a;l) +! 8S~~)ao(<p)ltP,11')1
<a
S,E9I,III/JII=l
(Sl(J)tP,tP)
- ,
. f
III
max
(5.23)
249
where 91 1 is the set of m x m positive-definite symmetric matrices belonging

to the space CI(Tm).
Theorem 2. Suppose that the functions a, P and f have rth order partial
derivatives with respect to , h that are continuous with respect to , h, f in the

region (2.2). Suppose further that the inequality
f31J.t>
holds. Ij r
f
(5.24)
ra
> 1, then there exists a sufficiently small fO > 0 such that for any
E [0, fO] system (2.1) has an invariant torus (2.6) with junction u belonging
to the space C~~I(Tm) and satisfying inequality (2.7). For r ~ 2 this torus is
exponentially dichotomous, while for r = 1 it is unstable if the matrix S( ) has
negative eigenvalues and is asymptotically stable otherwise.
Proof. By Lemma 1, of 3.13, the system of variational equations (2.5) has a
Green's function G o( T, ) satisfying the inequality
(5.25)
where 'Y f3I J.t and J( canst does not depend on E Tm

Using the inequality (5.23) which defines the constant a, one can derive via
a standard argument an estimate for the fundamental matrix of the solution,
n~( aaol 8)
of this system of linear homogeneous equations with the coefficient
matrix aaola
= aao(t())18, where t() is a solution of the first equation
of system (2.5):
(5.26)
where Ii
on
fl (fl
= a + fl
and /{ I
= /{ I (fd
is a positive constant which depends only
is an arbitrarily small positive number).
In view of Theorem 1 of 3.13, inequalities (5.24)-(5.26) lead to the estimate
(5.27)
for any function
f E
cr(Tm
which depends only on
),
l'
= 1- fl,
and some constant K 2
=K
2 (fd
fl.
We introduce a perturbation into the first equation of system (2.5) by

adding to its right hand side the function al() E Gr(Tm ) satisfying the inequality
CHAPTER 4
250
Then the matrix .5\(4))
= 5(4)) + (oS(4/o4al(4
max (Sl(4h,h) < -2136

II h ll=l
with constant 136
= 13 -
I< 6, where I<
Replacing ao by ao
+ a1
= ~}~ 118~~) II.
in the left hand side of inequality (5.23), we
obtain a similar inequality with
0'
equal to
positive constant independent of (1, 6, & =

the constants 136,
a6
that 6 and
(1
0'6
0'
= & + ]{16,
where
J{1
is a
+ (1. It follows from the form of
can be fixed so small that the inequality

(5.28)
holds. Then system (2.5) perturbed by the term al(4)) has a Green's function
G o(r,4 that satisfies inequality (5.25) with 1 equal to 16 = 136/J.I. The matrix
oto (~
8,p + lli)
8,p' for which
oao
04>
oa1
+ tii
oao(t(4)
04>
oal(t(4>))
04>
'
where t( 4 is a solution of the first equation of system (2.5) perturbed by the

a1 (4))
term
to
&6
0'6
on its right hand side, now satisfies inequality (5.26) with & equal
+ (1
This leads to estimate (5.27) for Go(r, 4 with
and 0- equal to 0-6
equal to
76
+ (1.
For sufficiently small
(1,
inequality (5.28) ensures that the exponent in
inequality (5.27) for the Green's function is negative. This is sufficient for the
Green's function G o(r,4 for the system (2.5) to be coarse with smoothness
index r . The main theorem ensures that there exists an invariant torus (2.6) of
system (2.1) with the function u belonging to the space Cf.i;,1(Tm ) and satisfying
inequality (2.7).
For r
2: 2, system (2.1) and the torus (2.6) satisfy conditions sufficient
for us to write down the variational equations corresponding to the torus (2.6)
considered as an invariant set of system (2.1).
Since luh,Lip
I<Ifl2
-->
as (
-->
0, the variational equations for the torus
(2.6) for ( E [0, (0] with a sufficiently small
(0
are defined by the functions a, P
which are close (with respect to the C(Tm ) norm) to the functions ao, Po
defining the variational equations (2.5). This suffices for the matrix
,
SI
oS
= S + 04> (a -
ao) + S(P - Po) + (P - Po)" S
= S( <f
constructed by using the matrix S
according to the variational equa-
tions for the torus (2.6) to be negative-definite for all <f> E Tm and any
where
fO
251
E [0, fO],
is a sufficiently small positive number.
Thus, for r
2 all the conditions of Theorem 1 are satisfied and the
invariant torus (2.6) of system (2.1) turns out to be exponentially dichotomous

for every
Let r
E [0, fO],
= 1.
We consider the "extended" system of equations:
d<f>
di = a(<f>,x(O),f),
dO
dt = P(<f>,X(O),f)x(O)
dh
+ f(<f>,f),
(5.29)
di = P(<f>,X(O),f)h+f(<f>,f),
where X(O)
= (Xl (Od, ... , Xl (<f>n
is the function (5.8).
> 0 and Co > 0 system of equations (5.29) satisfies
the conditions for the existence of an exponentially dichotomous torus and it

has such a torus
(5.30)
for all
E [0, fO]' The function
Ul
in (5.30) belongs to the space C(Tm x Tn)
and satisfies the inequality

(5.31)
where K(f)
-+
0 monotonically as f
-+
O.
Since the torus (5.30) is exponentially dichotomous, it follows that all the
motions of system (5.29) that are bounded for t E R are motions on the torus
(5.30).
The system of equations (5.29) has the locally invariant set
(5.32)
the motions on which are defined by the system of equations (2.1). On this set
the invariant torus (5.30) of system (5.29) becomes the locally invariant set of
system (2.1) defined by the relations
(5.33)
252
CHAPTER 4
for all f E [0, fO]' By virtue of the properties of the torus (5.30), the trajectory
of any motion >t, ht of system (2.1) that is bounded for t E R and for which
IIhtll < fl for all t E R lies in the invariant set (5.33) for t E R.
Since the motion >t,h t = U(>t, f) of system (2.1) that starts on the invariant torus (2.6) satisfies the inequality IIhtll = lIu(>t,f)1! ~ Kl/h ~ K(f) for all
t E R, it follows that for sufficiently small fO, IIh t II < fl for all t E R, f E [0, fO]'
This suffices for the torus (2.6) to satisfy the relations (5.33) for all f E [O,fO]'
Consequently, U(>,f) = Ul(>,U(>,f),f) for all > E Tm , f E [O,fO] for a sufficiently small fO > 0.
We now show that equation (5.33) has no solutions other than h = U(>, f)
in the domain Ilhll < fl, for any > E Tm and f E [0, fO] provided that fO is
sufficiently small. Indeed, suppose that for some > = >o E Tm , f = f' E
[0, fO], equation (5.33) has a solution h = hij = h o(>o, fl) f u( >o, fl) for which
/Jholl < fl
Then the motion
f
f'
= >t(>'o, h'o, fl), h = ht(>'o, h'o, fl) of system (2.1) for
that starts at the point (<p'o, ho) satisfies the identity
for all those t for which IIh t (<po, ho,fl)11
< fl.
It follows from (5.34) that for
these t the inequality
holds. It does not allow the motion <P
= <Pt (<p'o , h'o, fl),
= ht(<p'o,ho,f' ) to
leave the domain IIhl! < fl for any t E R. This proves inequality (5.35) for all
t E R. We set
(5.36)
The function
Zt
is defined and absolutely continuous for t E R and satisfies the
inequality
0< sup IIZtll ~ 2K(fO)'
(5.37)
tER
It follows from the arguments given in the proof of Theorem 1 of 4.3 for
= 1 that
Zt
is a solution of the system of linear non-homogeneous equations
(3.15) with <Pt
= <Pt(<Po, h'o, fl),
= f'.
Since the conditions of Theorem 2
ensure that the system of homogeneous equations corresponding to (3.15) is
exponentially dichotomous, it follows that
Zt
253
is the unique solution of system
(3.15) that is bounded on R and the following representation holds for it:
(5.38)
where G t (r) is a Green's function for the homogeneous system of equations
corresponding to system (3.15). The function G t ( r) satisfies the following
estimate which is the analogue of inequality (5.25):
where K
> 0 and
'Y
> 0 are independent of i, rand
f'.
lt follows from formula (5.38) and the above estimate for the function
Gt(r) that
II Z tll ~ K~(f')
'Y
where K(f)
-+
supllztll,
0 and is monotonically decreasing as f
fO
(5.39)
tER
O.
> 0, inequality (5.39) leads to the estimate
sup IIztll ~
tER
-+
! sup
IIztll,
tER
which contradicts inequality (5.37).

This contradiction shows that only the point (<p, h) which belongs to the
invariant torus (2.6) of system (2.1) satisfies relations (5.33).
Denote by p the number of positive eigenvalues of the matrix S(tP). Let
p
< n. On the locally invariant set (5.32) of system (5.29) the separatrix set
Mij of invariant torus (5.30) of this system becomes the locally invariant set
M- of system (2.1) defined by the relations
(5.40)
for all f E [0, fo]. It has the property that a solution tPt, ht of system (2.1) for
which (<Po, ho) E M- satisfies the inequality
(5.41)
for all i
0 as long
a..,
Ilhtll < fl.
254
CHAPTER 4
= >o
We fix > by setting >

belonging to the domain IIhll
and consider a solution of equation (5.40)
< <1, for > = >o. Let T( >o, h, <) be a matrix that
transforms C(>o, h, <) to Jordan form. Since C(>, h, <) is continuous in >, h for
>, h, < in the domain > E Tm , IIhll
< <1,
< E [0, <0] and is a projection with rank
p, we see that the matrix T(>o,h,<) can be chosen to be continuous in h when
h, f belong to the region IIhll < fl, f E [0, fO], By introducing the variables
x = (Xl, X2) related to h by the equality
equation (5.40) takes the form

Xl =0, >ETm
We define h = h( e, <) to be a solution of the equation

(5.42)
where e
= (el' ... , en - p) is an (n -
p )-dimensional vector in the ball lIell
< 01,
and S2(>,h,f) the block of the matrix T(>,h,f) = (Sl(>,h,f), S2(>,h,f)) of

appropriate dimension. We choose fO
> 0 and 01 > 0 such that for f E [0, fO]
the inequality
(5.43)
where
f{
is the constant in the estimate (5.41) and
M=
sup
IIS2(>,h,f)lI.
,lI h lll
1>ETm
Then the right hand side of equation (5.42) defines a continuous operator
h mapping the ball IIhll < fl into itself. This suffices for equations
h(e,f) for all f E [O,fO] and allllell < 01' This
(5.42) to have a solution h
solution satisfies the inequality IIh(e, c)1I < fl, for these f and e. Moreover,
since S2(>,h,c)e = 0 only for c = 0, it follows that h(e,f):f; ul(>o,h(e,f),f)
for e:f; 0 and consequently, h(e,c):f; U(>o,f) for e:f; O. The solution > =
>t(>o,h(e,c),f), h = ht(>o,h(e,f),c) of system (2.1) that takes the value >o,
h(e,f) at t = 0 satisfies the condition (>o,h(e,f)) E M- and consequently, it
h -
satisfies inequality (5.41) for
t:::;
0 as long as IIht(>o,h(c,<),<)1I
< <I.
255
Since IIh(e,f) - uI(4)o,h(e,f),f)11 < Mh l , it follows from inequality (5.41)

that
for t
0 as long as IIh t (4)o,h(e,f),f)1I
< fl. But then ht (4)o,h(e,f),f) satisfies
the inequality
(5.44)
for these t ~ 0 and cannot leave the domain IIhll
follows that inequality (5.44) holds for all t ~ O.
Letting t tend to
-00
< fl for t < O. Hence it
in inequality (5.44) we find that the a-limit points
of the semi-trajectory of the motion 4>t(4)o,h(e,f),f), ht (4)o,h(e,f),f) belong

to the set (5.33) and consequently, they belong to the invariant torus (2.6) of
system (2.1). At the same time the motion 4>t(4)o, h(e, f), f), h t (4)o, h(e, f), f)
itself does not start on invariant torus (2.6) of system (2.1) for e
=/: O.
There-
fore, the motion of system (2.1) that starts on the negative semi-trajectory of
the motion
4>t( 4>0, h( e, f), f),
h t ( 4>0, h( e, f), f) lying in an arbitrarily small neigh-
bourhood of torus (2.6), leaves a certain fixed neighbourhood of the latter at

some time t = T > O. According to the definition of an unstable invariant set
of the dynamic system, the torus (2.6) is unstable.
Let p
= 1l.
The locally invariant set M+ of system (2.1) is then defined
by the relations
for all f E [0, foJ and has the property that the solution
4>t, ht of system (2.1),
for which IIh oll < fl, 4>0 E Tm , satisfies inequality
for t ~ 0, as long as IIhtll < fl.

We fix 4> by setting 4>
4>0 and consider a solution h
= h( e, f)
of the
equation
(5.46)
where e = (el, ... , en) is an n-dimensional vector in the ball lIell < hI. If fO > 0
and hI > 0 are chosen such that for f E [0, foJ inequality (5.43) holds with
M
= IIEII. then the solution h(e,f) of equation (5.46) exists for
all f E [O,fO],
256
CHAPTER 4
IIcll < 61 , and satisfies inequality IIh(c, l-)" < i1 for these i and c.
inequality (5.45) and arguing as for the case p
Using
< n, we derive the inequalities
IIht( 4>0, h( c, i), i) - U1 (4)t( 4>0, h( c, i), i),

h t (4)o,h(c,i),i),i)1I < ~e-"'Yt,
IIh t (4)o,h(c,i),i)lI:s [{(CO)
i1
t
+ "2e-"'Y
< i1
(5.47)
for all t ~ 0, i E [O,iO], IIc" < 61 For a fixed i E [0, iO] the function U1 (4), h, i) is uniformly continuous in
4>, h as it is continuous on the compact set 4> E Tm , "hll :S 61 < i1. Therefore
we can choose 62 = 62 (6t} > such that the inequality
holds for arbitrary 4> E Tm and any hand
We set 62 < 61 /2 and choose iO >
holds for all 4> E Tm
For an arbitrary 4>
= 4>0 E Tm
h satisfying the inequalities:
such that the inequality
and any h
= h o in the domain
(5.48)
we now have the inequality
which proves that an arbitrary point (4)0, h o) in the domain (5.48) coincides
with the point (4)o,h(C,i)) when c = ho - u1(4)0,h o,i) in the ballllcll < 61 .
Thus, for any point (4)0, ho) satisfying condition (5.48), the solution 4>t =
4>t(4)o,h O,i), h t = ht (4)o,h o,i) of the system of equations (2.1) satisfies the
inequalities (5.47):
(5.49)
for all t
257
o.
Since u( >, t:) E CLip(Tm ) for t: E [0, t:o] and the function u( >, t:) satisfies
inequality (2.7) for r = 1, the set of points (>o, h o) satisfying condition (5.48),
where >o ranges over Tm contains for some d = d(h 2 ) > 0 ad-neighbourhood
of the torus h = u(>,t:), > E Tm . Thus, inequalities (5.49) hold for solutions
>t, h t that start at points of this neighbourhood. Letting t tend to +00 in
inequalities (5.49) we see that the w-Iimit points of the semi-trajectory of the
motion >t, h t belong to the invariant torus (2.6) of system (2.1).
None of the a-limit points of the semi-trajectory of the motion >t, h t
belongs to the torus (2.6). Indeed, if this is not so, then there exists a sesuch that all the points (>t n , h tn ) of the motion >t = >t(>o,ho,t:), h t = ht(>o,ho,t:) with fixed initial point (>o,h o) in
the d2 (h 2 )-neighbourhood of torus (2.6) belong to this neighbourhood for all
quence t n
-+ -00
as n
-+ 00
n = 1,2,.... But then the positive semi-trajectories of the motions >t
>t(>tn,htn,t:), h t
= ht(>tn,htn,t:) satisfy inequalities (5.49) for n = 1,2, ....
Since >l(>tn,htn,t:)
follows that
for all t
= >l+ln(>o,ho,t:),
= 1,2,....
> 0 and n
hl(>tn,htn,t:)
= ht+tn(>o,ho,t:) it
Consequently, the trajectory of the motion
>t(>o,ho,t:), ht(>o,ho,t:) satisfies the inequality
for t E [tn,+oo) (n
= 1,2 ...).
Thus the motion >t
= >t(>o,ho,t:),
ht =
ht(>o, h o, t:) of system (2.1) satisfies the inequality IIhtll < t:l, for all t E R. As
was proved earlier, such a motion takes place on the invariant torus (2.6). This
contradicts the fact that the point (>o, h o) does not belong to this torus.
Thus torus (2.6) has a compact neighbourhood that contains neither whole
trajectories nor motions having a-limit points on this torus. By Theorem 3 of
2.2, this means that the torus (2.6) considered as an invariant set of system
(2.1) is stable. The limit relation necessary for torus (2.6) to be asymptotically
stable follows from the fact that all the w-limit points of the motions >t, h t
that start in the d(h 2 )-neighbourhood of torus (2.6) belong to this torus. This
completes the proof that invariant torus (2.6) of system (2.1) is asymptotically
invariant for p
= n.
258
CHAPTER 4
It should be noted that the condition of Theorem 2 can be weakened
by replacing the left hand side of inequality (5.24) by a positive number /

determined from the inequality
(5.50)
This follows from the proof of Theorem 2 by taking into account the fact that
inequality (5.50) leads to estimate (5.25) with 'Y equal to the value in (5.50).
However, since (-(3) is the greatest of the eigenvalues of the matrix S(4J)/2,
4J E Tm , and
Jl is the greatest of the absolute values of the eigenvalues of the
matrix S(4J), 4J E Tm , we see that finding the ratio (3/Jl is simpler in practice
than finding the number /.
We set
0'0
equal to the greatest of the absolute values of the eigenvalues
+ (&a3~cf r], 4J E Tm. It is easy to see that inequality

(5.24) holds whenever (3/Jl > rO'o for r ~ 1. For example, for system (2.1), the
of the matrix
! [&a~~
cf
matrix Po(4J) of which, when taken in the block form
Pd4J) + P;l(J)]
-P2 (J) - P;(J)
(5.51)
necessary for it to be a negative-definite matrix, condition (5.24) holds whenever (3 > rO'o for r
1, where (- (3) is the greatest eigenvalue of matrix (5.51)
for J E Tm
4.6. An estimate of the smallness of a perturbation
and the maximal smoothness of
an invariant torus of a non-linear system
The estimate on how small a perturbation must be so that the statements of
perturbation theory of an invariant torus for the non-linear system (1.2) remain
true is a problem of quantitative analysis of this theory. At present, it is not
well studied and is not very effective from the point of view of applications.
Therefore we shall give only one result of such analysis, namely, conditions for
259
the preservation of an invariant torus involving an estimate of the smallness of

a perturbation admitting practical verification.
We start with the system of equations
d
di
where a, P and
dh
= a(, h), di = P(, h)h + f(),
(6.1)
f are periodic functions in v (v = 1, ... , m) with period
211',
defined in the domain
(6.2)
IIhll:::; d, E Tm
and satisfying Lipschitz conditions with respect to , h:
lIa( , h) - a( ', h')11 :::; O'oll - 'II + 0'111h - h'll,

IIP(. h) - P(', h')11 :::; ,811 - 'II + ,8111 h -
h'IL
(6.3)
IIf() - f(')11 :::; A6 oll - 'IL

where 0'0, 0'1, ,8, ,81 and A are non-negative constants, 60
hand h' are arbitrary constants in the region (6.2). Let
= .pErm
max IIf()IL , ',
ma.x (P(, h)7], 7]) :::; -,80

111111=1
(6.4)
for all , h in the region (6.2) and some constant ,80
> O. Finally, we require
that the inequality
,80 - 0'0
= 11 > 0
(6.5)
hold.
We set
f-lo
= ,8 +,81A,8o
~o
11
= ( f-lo + 0'1 f-lo
) 1/2
- f-lo
(6.6)
Theorem 1. Let conditions (6.3)-(6.5) hold for the system of equations (6.1).
If
00 :::; ,80 min
(d, ;:
c: -
2~0) ),
then the system of equations (6.1) has an invariant torus
h=u(), ETm
(6.7)
260
CHAPTER 4
where the function u belongs to the space CLip(Tm
l/u()1/
and satisfies the inequalities
IIu() - u(')11 ~ ~oll - 'II
~ 60/(30,
for arbitrary , ' E Tm .

Proof We denote by
C(~)
the set offunctions F() in CLip(Tm ) satisfying the
inequalities
!Flo
IIF() - F(')II ~ ~II - 'II
d,
for any , ' E Tm . By introducing the distance p(F1 , F 2 ) between the elements
F 1 , F 2 of the set C(~) according to the formula
we turn
C(~)
into a complete metric space.
Following the method of integral manifolds [22], [81] we define on
C(~)
an integral operator S (we shall write S",(F) instead of S(F)( by setting
= l~ n~(, F)f(~(dT,
S",(F)
(6.8)
where f() is a solution of the first equation of system (6.1) for h = F()
and
n~(, F)
is the fundamental matrix of solutions of the system

dh
di
= P(tF (),
that is equal to the identity matrix for t
S has a fixed point in
C(~).
F(t ()))h
T.
(6.9)
We now show that the operator
Let us make the necessary estimates.
We estimate the difference f()-['(), where F, F are any two functions

in C(~) and , are arbitrary points of Tm . We have:
I/f() - ['()II
~ II -
I/ +
F(f'()))l/dtl
~ II -
II +
Il
Il
~ II -
I/a(f(), F(f() - a(['(),
II + Il(aollf() - f'()II+
+ad!F(f( - F(['(IDdtl ~
(ao +
al~)IIf() -
f'()l/dtl + adF() - F()loltl,
261
whence, using the Gronwall-Bellman inequality [43], we obtain
114>;(4)) - 4>[(~)1I ::; e(ao+o,A) ltl ll4> - ~II+

+
0'1
[e<ao+a,A)ltl _
0'0 + 0'1~
1] IF(4)) -
We estimate the difference n~(4), F) - n~(~, F)
= z~.
F(4))lo.
(6.10)
To do this we use
equations (6.9) to write the system of equations for Z::
d~:
P(4);(4>),F(4>;(4>)))n~(4>,F) - P(4)[(~),F(4>r(~)))n~(~,F) =
= P(4>; (4)), F( 4>;(4))))Z:+[P( 4>;(4)), F( 4>;(4))))- P(4>[( ~), F( 4>[( ~)))]n~(~, F),
from which (since
Z~ =
1n~(4),
Z; = 0)
it follows that
F)[P(4); (4)), F(4); (4)))) -
p(i;(~), F( i;( ~)))]n~(~, F)ds.
Iln~(4), F)holl ::; exp{

for t 2:
T,
IIZ~holi ::;
-it
f30 ds }lIholl = exp{ -f3o(t - T)}lIh oli
as a result of which we obtain
1
0
e.B [.BII4>;(4)) O
4>!'(~)1I + f311IF(4);(4>)) - F(4);(~))II]x
xe-.Bo('-T)dsllholl ::; 1[(.B +
f31~)114>;(4)) - 4>;(~)II+
+f3t1F(4)) - F(4))lo]dse.Bo Tllh oll
for
T ::;
O.
The last estimate together with inequality (6.10) shows that
IIZ~holl ::; IIholle.BoT {(.B + .B1~)[1I4> - ~II +
0'11F(4)) - F(4))lo] [0 e-/"ds+

~
iT
+ (.B1 - 0'1(f3; f31~) IF(qi) - F(4))lo( -T) }
=e.BoT{.B + .B1~[1I4> _ ~II +

~
0'1IF(4)) - F(4))lo](e-/l T - 1)+

~
262
CHAPTER 4
+(.8 + ~~)0:1 -.81) IF(4)) - 1'(4))loT }lI ho ll =

= {.8 +:1~ (e(f3o-/J)T _ ef30T )114> _ ~II+
+ [(.8 + ~~~)0:1 (e(iJo-/J)T _ eiJOT) + (.8 + ~~)0:1 -.81) TeiJOT]

X
for T :::; 0 and p, =
0:0
IF(4)) - 1'(4))10 }lIholl
+ 0:16..
We now estimate IIS",(F)II and IIS",(F) - S~(F)II. We have:
IIS",(F)II :::;
[00 IIn~(4), F)f(4)~(4>))lldT:::;
: :; 1-00
eiJOT dTlf( 4> )10 :::;
:0 ,
PO
and similarly.
IIS",(F) -
S~(F)II :::; [00 IIZ~f(4)~(4>))lIdT + [00 IIn~(~, 1')(f(4>~(4)

-f(4)?,(~)))dT:::; {.8+.816.(~_; )1I4>-;II+
p,
PO -
P,
PO
+[(.8+.816.)0: 1 (_1__ ~) + (.8+.816.)0: 1 _ .81) (-~)]

p,2
.80 - P,.80
p,
.8'5
x IF(4)) - 1'(4))10 }If(4>)lo +
[[00 ef3OTII4>~(4)) - 4>?'(~)lIdT] ADo,
(6.11)
provided that
.80>
J.l.
Since
[~ eiJoTII4>~(4)) - 4>?'(~)lIdT S [~ eiJOT [e-/JT 114> - ~II+

+ 0:1 (e-/JT - 1)1F(4)) - 1'(4))lo]dT = ~1I4> - ~II+
J.l
PO -
J.l
0:1 (1
+- - - -1)IF(4)) -
J.l
.80 -
p,
.80
F(4))lo,
(6.12)
263
we have from (6.11), (6.12) the estimate

-
IIS.p(F) - S~(F)II :$ 130 _ fJ
(13+ 131 1i.

)
130 60 + A60 II> - 4>11+
+[(130-2ao)13I+13aI60+
al
>.60]IF(-F(4lo=
130(130 - fJ)
130(130 -fJ)
1
= 130 -
fJ
[13 + 131 Ii. 0

130 6
+ A60 II> x
1
4>11 + 130(130 _ fJ) x
[(130 -
aO~:l + 13a160 + al,Mo] IF(4)) - F(lo.
When the inequalities
(6.13)
hold, the operator S maps the space C(Ii.) into itself and is a contraction.
It follows from the contraction mapping principle that there exists in C(Ii.)
a unique fixed point F(
= u(
of the operator S such that the following
identity holds:
Here
Ilu(4))II:$ 11(4))10/130,
lI u ( - u(4)')11 :$ li.1I4> - 4>'11
for any 4>,4>' E Tm

We now examine the inequalities (6.13). For this purpose we rewrite them
in the form
60
130 :$d,
a l li.<'Y1,
60
1i.(-y1 - alii.)
130:$ 13+ 131 Ii. + A130 =g(Ii.),
60
13o(-Y1 - alii.)
130 < 11131 + 13a1 + >'13oal
(6.14)
Since g(O) = g(-Ydad = 0, g(li.) attains its maximum on the interval 0 < Ii. <
Idal'
Differentiating g( Ii.), we find that
'(Ii.) = (-Y1 - 2a 11i.)(13 + >'130 + 1311i.) - 1311i.(-y1 - alii.)

9
(13 + >'130 + 1311i.)2
=0
264
CHAPTER 4
if ~ is a solution of the equation
Solving this equation we see that
imum is equal to the number
defined by equalities (6.6).
~o
at which the function
g(~)
takes its max-
Now we have
We show that
.80(r1 - 0'1~0)
r1.81 + .80'1 + >".800'1
~0(r1 - 0'1~0)
> (~o + PO).81
(6.15)
To do this we use (6.6) to convert the denominator on the left hand side of
inequality (6.15) to the form
and write the inequality

.8oPo
0'1(~0
+ po)
>~
(6.16)
0,
equivalent to (6.15). On substituting the expression for
~o
into (6.16) and
making elementary calculations we obtain the inequality:

.80 - r1
0'1
- - - > Po -
(2
Po
r1
) 1/2
+ -Po
,
0'1
which clearly holds for

.80
Since r1
= .80 -
0'0, inequality (6.17) always holds.
Considering the inequalities (16.14) for

when
(6.17)
2: r1
= ~o,
we see that they hold
265
that is, when inequalities (6.7) hold. Thus inequalities (6.13) hold if the conditions of Theorem 1 are satisfied.
Consider the torus
h
= u(<p) = S</>(u),
<P E Tm.
(6.18)
We claim that it is an invariant set of system (6.1). To see this, consider the
function
where
u(<p~(<p)) = {~ n~(<p~(<p), u)f(<p~(<p~(<p)))dT,
<p~(<p)
is the solution of the first equation of system (6.1) for h
= u(<p).
Since
the equality
u(<p~(<p)) = {I n~(<p, u)f(<p~(<p))ds

oo
holds, therefore
dU(~(<P)) = P(<p~(<p), u(<p~(<p)))u(<p~(<p)) + f(<p~(<p))

for all t E R. And this implies that the pair of functions <PI
u(<p~(<p))
= <p~(<p),
hi
is a solution of system (6.1) for t E R, so that the torus (6.18) is an
invariant set of system (6.1). Since the function u belongs to the space C(Ll o),
inequalities of Theorem 1 hold for the function u( <p). This completes the proof
of Theorem 1.
We note the two limiting cases for inequality (6.7):
60 $;)omin(d,;:),
a1=0,
Z
60 $ ;)0 min
(d, 4a1 (;)'Y~ ).;)0)) ,
;)1 = 0,
for which it is significantly simpler to check whether inequality (6.7) holds.

As an example illustrating Theorem 1 we consider the system of three
equations
d~l = W1 + sin(<P1 -
<Pz)lhl,
d<P2
dt
= W2 + cos( <P1 + <P2 )h,
~~ = -2h + t sin <P1 cos <P2.
(6.19)
266
CHAPTER 4
The conditions of Theorem 1 hold for this system if the positive parameter
t

(6.20)
for some d E (0, 1).

The greatest value of t for which inequality (6.20) holds for d E (0,1) is
given by
(6.21)
and is attained for
d = do = to/2.
(6.22)
If we choose to and do according to equalities (6.21), (6.22), then we obtain

the expression
~o = ~(1 Consequently, for all t :::;
fO
do) =
VI + h
1.
system (6.19) has an invariant torus
lying in the region

Ihl :::; do
= fo/2 ~ 0.05
and satisfying the inequality
lI u ( 1>, f) - u(1)', f)1I :::;
~oll1> - 1>'11 = (VI + h - 1)111> - 1>'11 ~ 0.511q1 - qI'lI
The smoothness of an invariant torus of system (6.1) will be called maximal

if it is equal to the smoothness of the functions a, P, f on the right hand side of
this system. Theorem 1 differs from the other theorems of this chapter in that
it ensures that the invariant torus, the existence of which also follows from this
theorem, has maximal smoothness. In spite of the smallness of the difference of
smoothness between the functions a, P, f and that of an invariant torus in the
systems considered before, this difference in the theorems considered before is
natural and is explained by an important difference between the conditions of
those theorems and those of Theorem 1 of the present section.
267
Indeed, the conditions of the theorems in which there is a "loss of smoothness" of the invariant torus, the existence of which is affirmed by these theorems, assume that the appropriate argument is carried out for a bounded subset
of the space cr(Tm ), while the conditions of Theorem 1 of the present section
assume that we work in a bounded subset of the space CLp(Tm). A bounded
set of functions in cr(Tm ) is compact in Cr-1(Tm ), whereas a bounded set of
functions in CLp(Tm ) is compact in Cr(Tm ) and any limit function of this set
belongs to the space CLp(Tm ). Since an invariant torus of the systems under
consideration is defined by a limit function of bounded sets in spaces Cr(Tm )
and CLp(Tm ) respectively, it loses smoothness in the first case but not in the
second.
The above remarks not only explain but also define the conditions under
which it is possible to achieve a maximal smoothness of an invariant torus of the
considered systems. Namely, we have the following statement supplementing
Lemma 1 of 4.1.
Lemma 1. Suppose that the functions a, P and f are defined and periodic
4Jv (v = 1, ... , m) with period 211" in the region (6.2) and have continuous
partial derivatives with respect to 4J, h up to the order r which satisfy a Lipschitz
in
condition with respect to , h. Suppose that for any i
= 0, 1, ...
the system of
equations (1.6) has an invariant torus (1.5) belonging to the space CLip and
satisfying the inequalities
lu i+l ()lr,Lip~A,
where d1 and A are positive constants, d 1
i=0,1,2, ... ,
< d.
If
for every E T m then the limit function u() defines an invariant torus (1.1)
of system (6.1) belonging to the space CLp and satisfying the inequalities
Ilu()/1 ~ d1 ,
lu(4J)lr,Lip ~ A.
The following theorem is similar to Theorem 1 of 4.3 and Theorem 2

of 4.5 which define the conditions for preservation of an invariant torus of
maximal smoothness under small perturbations.
268
CHAPTER 4
Theorem 2. Suppose that the functions a, P and
have partial derivatives
with respect to 4>, h up to order I that are continuous in 4>, h, { and that these partial derivatives satisfy Lipschitz conditions with respect to 4>, h for all 4>, h, ( in
the region (2.2), where the Lipschitz constants K/
= K/({) of the Ith derivatives
with respect to 4>, h of the functions a( 4>, h, () - a( 4>, h, 0), P( 4>, h, () - P( 4>, h, 0)
and 1(4),{) tend to zero as
{->
O.
Suppose further that the inequalities (3.10) and (5.24) hold for r = 1+ 1.
If I ~ I, then there exists a sufficiently small (0
> 0 such
that for any { E [0, co]
system (2.1) has an invariant torus (2.6) with function u belonging to the space
CLp(Tm ) and satisfying the inequality:
luh,Lip
:s Kl/h,Lip,
where K is a positive constant independent of f.

Proof. We "smooth out" the right hand side of system of equations (2.1) by
means of the operator T NN defined by the Nash kernel [161] and obtain the
system of equations
(6.23)
where
for each ( E [0, (0], N being a large natural number.

It follows from the form of the operator TNN that the functions aN, PN
and IN are defined in the region
II h ll
= d'(N)
:s d',
4> E Tm , {E [O,fO],
(6.24)
d and is monotonically increasing as N -> +(0), are

periodic in 4>11 (1/ = 1, ... , m) with period 211" and have partial derivatives with
(where d'
->
respect to 4>, h of arbitrary order that are continuous in 4>, h, f. Furthermore,

the sequences aN, PN, IN together with their derivatives with respect to 4>, h
up to order I are uniformly convergent with respect to 4>, h, ( in the region
(6.24) to the functions a, P, f and their respective derivatives. Moreover,
269
and the partial derivatives of the functions aN, PN and IN with respect to l/J, h
up to order 1+ 1 are continuous in l/J, h, ( in the region (6.24) and satisfy the
inequalities
D~;~aN(l/J,h,()1I + IID~;~ PN(l/J,h,()1I ~ GK,
(6.25)
D~+l IN(l/J,()1I ~ GI/II,Lip,

where G is a constant independent of a, P,
I and
N, and K is the Lipschitz
constant of the derivatives of a and P with respect to l/J, h up to order I.

These properties of the functions aN, PN and IN are sufficient for us to
apply Theorem 1 of 4.3 or Theorem 2 of 4.5 to system (6.23) with r = 1+ 1
and to establish that for all ( E [0, (0] and all N 2: No with sufficiently small
(0
= (o(No) > and sufficiently large No, there exists an invariant torus
(6.26)
of system (6.23) with function UN belonging to the space GLp(Tm ) and satisfying the inequality
where
J( 2
is a positive constant independent of ( and N.
Let uo(l/J,() be the limit of the sequence of functions UN(l/J,() (N
No, No
+ 1, ...).
It follows from inequality (6.27) that
for all ( E [0, (0], while it follows from the identity
that the torus

h=uo(,(),
ETm
is an invariant torus of system (2.1) for all ( E [0, (0].

Since an invariant torus of system (2.1) is unique in a neighbourhood of
the trivial torus h = 0, l/J E Tm , as follows from the exponential dichotomy of
the invariant torus (2.6) of system (2.1), we conclude that
270
CHAPTER 4
This suffices to prove that the invariant torus (2.6) of system (2.1) satisfies the
conditions of Theorem 2, and so completes its proof.
Remark. The theorems on the compactness in Gr-l(D) of any bounded subset
of Gr(D) and the compactness in Gr(D) of any bounded subset of GLp(D),

where D is a region in Tm or a bounded region in Rm are similar to Sobolev's
Theorem 2 of 1.2. They are proved by means of the Arzela-Ascoli criterion
[50].
4.7. Galerkin's method for the construction of an invariant torus of
a non-linear system of equations and its linear modification
We denote by L( Ul )U2 the quasi-linear operator on the set of functions u E
Gr(Tm ) defined by the inequalities
lulo < d, lulr:S; K,

that is given by the expression
(7.1)
The function u(4)) defining the smooth invariant torus (1.1) of system (1.2) is
a solution of the equation
(7.2)
L(u)u=f,
which belongs to the space G 1 (T,,,).

Galerkin's method defines the Nth approximation of the solution u(4)) E
G1(Tm ) of equation (7.2) via the expression

WN(4))
=L
W~N)ei(k,</,
4> E Tm ,
(7.3)
\Ik\lSN
the coefficients W~N) of which are solutions of the system of non-linear algebraic
equations:
To define a linear modification of Galerkin's method we start with the

initial approximation uo(4))
Nj (j
= 0 for 4> E Tm and choose a collection of integers
= 0,1, ...) subject to the condition

Nj::::Nj_h
i=I,2, ... ,
(7.5)
271
and take the Njth linear Galerkin's approximation to the solution u(t/
C (Tm ) of equation (7.2) to be
vj{t/
(7.6)
Uk ei (k.4,
II k ll:;Nj
where the coefficients
Uk
= u~) are solutions of the system of linear algebraic
equations
[t is dear that the Njth linear Galerkin approximation to the solution
u(t/ E CI('Tn,) of equation (7.2) is the Njth Galerkin approximation to the

solution in C 1 (Tm ) of the equation
L(llj_l(lfJ))U()
= f(lfJ)
(7.8)
We shall suppose that tlle C r (Tm)-norm of the function f( lfJ) is sufficiently

small. 'Vith this assumption the conditions for the existence and convergence
of the G akrkin approximations W N(t/ and Uj (t/ are defined by the principal
part of the operator L(u), which is
m
L(O)
{).
= I>v(t/>,O) 8v
- P(t/>,O),
11=1
and they t.urn out to be inequalities of the form (14.21) or (14.21) for the
coefficients of the operator L(O) in the previolls chapter.
The proof of this fact relies on Moser's lemma [89], a particular case of
which is given below.
Lemma 1. Let a() and P() belong to the space Cr(Tm ). If inequality (14.21)
of Chapter 3 holds, then for an arbitrary u E Cr+l(Tm ) we have the following
inequalities
(Lu,u)o
2 'YlluI16.
(Lu,u).
2'Ylll ull;-6(1+ 2':lIavll.+IIPII.) .
",
(7.9)
v==l
2 L:~~I lavb +
2 2, 1:S s :s r.
where 'YI lInd 6 are positive constants which depend only on Co
IPh + luli,
L is the operator (14.2) of Chapter 3, and r
272
CHAPTER 4
The proof of Lemma 1 will be given later, but here we use it to substantiate
the linear modification of Galerkin's method.
Theorem 1. Let the functions a and P be defined in region (1.3) and have
continuous partial derivatives with respect to 4J, h up to order r. Suppose that
the inequalities (14.21) of Chapter 3 hold for a(4J) = a(4J,O), P(4J) = P(4J,O),
and for s
=and s = r.
If
r>m/2+1,
then for any M
>1
1~2,
there exists a sufficiently small p,
f E Hr(Tm ) and
IIfllr ~
(7.10)
= p,(M) > such that if

(7.11)
p"
then thl'. linear Galerkin approximations Uj (4J) exist for any j = 1,2, ... and
converge in H'(Tm ) n C/(Tm) for s
<r
to the function u(4J) which defines an
invariant torus (1.1) of system (1.2) such that

IIU-ujllo~C(fj,
where C
= const,
(fi
(7.12)
j=I,2, ... ,
N-(r-l)M-v
= "'=
LJi=i LJv=O
i-v
.
",i
Proof In equation (7.2) we replace u by
tu
and
f by {f, taking { to be a small
positive parameter. Equation (7.2) then takes the form

L(tu)u
= f(4J).
(7.13)
Let w(4J) be a function in cr(Tm ) satisfying the relation
Consider the linear operator

L( {w)
=L
m
av(4J, {w( 4J 84Jv - P( 4J, {w( 4J
v=!
as an operator in C=(Tm ). Since inequalities (14.21) of Chapter 3 are coarse,

they imply similar inequalities for the coefficients of the operator L({w) for all
> 0. Hence, using Lemma 1 we obtain the estimates
{ E [0, {oj for some {o
(L({w)u, u)o ~ illull~,
(L({w)u, u), ~ illulI;-
+ L lIa v(4J,{w(4JII, + II P (4J,{w(4JII,)

m
-60 (1
v=!
= 1, ... , r,
where, and 80 are positive constants independent of w, U and
f.
273
By applying
estimate (3.4) of Chapter 1 to the functions a v ( 1', fW( 1' and P( 1', fW( 1', the
above inequalities become
(L(fW)U, u)o ~
(L(fW)U,U).
,lIulI5,
~ ,lIulI; -
where 8 does not depend on w, U and
8(1
(7.14)
+ fllwll.)2,
f.
Consider the first linear Galerkin approximation
UI(1').
Its coefficients u~l)
are determined from the system of equations equivalent to the equation

(7.15)
where SN is the truncation operator for the Fourier series of the function
Lk !kei(k,q,) defined by the equality SN!(1')
= Lllkll~N !kei(k,q,).
!(1') :=
Inequalities
(7.14) for the operator L(O) guarantee that equation (7.15) can be solved and
that its solution
UI
(1')
has an estimate of the form
lIulli.
cll!II.,
= 0, ... , r,
where c is a positive constant independent of UI and
(7.16)
f.
Let r > p > m/2 + 2. By the first Sobolev theorem given in 1.2 and the
estimates in 1.3 we obtain from inequality (7.16) the estimate
where
is a positive constant independent of UI and
CI
f.
We require that f( 1') satisfy inequality

(7.17)
Then
lUI
b~
1 and the operator L( wI) is defined for all
E [0, fO] and satisfies
inequalities (7.14) for W = Ul' This guarantees that the approximation U2(
Lllkll~N, u~2)ei(k,q,) exists and that it has an estimate of the form
where s
= 1, ... , r.
274
CHAPTER 4
Solving this inequality we find that

(7.18)
for all ( E [0, (0] if
(0
1(
2"
1+
fy)-l ,
:;
(7.19)
Taking into account the estimate (7.16), the second inequality of (7.18) leads
to the inequality
(7.20)
if 61 satisfies not only inequality (7.19) but also the inequality
(7.21)
IU212 ~ cdlu211p ~ 2c1I1u211~-p/rllu211~/r ~

< 2c1I1fllb-p/r fyp/r(l + ()p/r < 1
-
,1-p/r
provided that the function f( satisfies the inequality
Then the operator L(W2) is defined for all ( E [0, (0] and satisfies in-
equality (7.14) for w

U3(ci
= EllkllSN
U2.
From this it follows that the approximation
u~3)ei(k,) exists and admits an estimate of the form
for all ( E [0, (0]' But then
IIU3110 ~
IIflloh,
IIU3112 ~ c111u311p
lIu311;
61(1 + ( + (2),
~ 2c1I1u311~-p/rOllu311~/r ~
< 2c11Ifll~-p/r fyp/r(l + (+ (2)p/r

-
,1-p/r
275
which leads to the estimates
provided that the function /( <jJ) satisfies the inequality
pl
2ClCr/r(1 + fO + (5) r II/Ill-pir < l.
l
plr
I 0It is easy to prove by induction that the approximation
any j 2: 3 and that
Uj (t/J)
Uj (t/J)
exists for

j-l
lIuili. ~
for all
Cl
Lf
v=o
E [0, fol if the function [(<jJ) satisfies the inequalities
l!Plr (2c 1 6r /r
j-l
(L f~ fir) II/II~-plr ~ l.
v=o
(7.22)
Since fO < 1/2, inequalities (7.19), (7.21) and (7.22) hold provided that
the function /( t/J) satisfies the inequalities
11/110 ~
I[2l+ plr clcr/T/(p-r),

[)1/2
II/II. ~ max{~,
where s
= 1, ... , r, and Cl
[Cl -
26
r( 1 + r) ]I
(7.23)
2
},
is a constant chosen so that
2c ( 1+ C) <6 1 .
-:;
(7.24)
When inequalities (7.23), (7.24) hold, the approximations Uj(<jJ) are defined for all
E [0, fol when
fO
> 0 is sufficiently small, and they satisfy the
inequalities
(7.25)
for all j = 1,2, ... and s = 1, ... , r.
(<jJ)
Uj+l(<jJ) - Uj(<jJ) satisfies the
We now tackle the question of the convergence of the sequence
(j
= 1,2, ... ).
equation
Since the function Wj+l(t/J)
Uj
276
CHAPTER 4
it satisfies the inequality
IllwHdlo 5 II [L( CUj) -
L({Uj - d]uj 110+

+1I(SNj+l - SNj)[L(cuj-duj -
1]110.
(7.26)
The first term on the right hand side of inequality (7.26), which we denote
by II, admits the estimate
1I 5 f=11[all(lCUj) 11=1
all(,{Uj_l)]~;: 110 +
The second term (denoted by 12 ) has the estimate
IlIr-l 5 c3(1 + II{Uj-dlr-l+
125 Nj-(r-l)Il L (CUj_d uj -
+IIUjllr + 1I/IIr_d N j-(r-l) 5 c4(1 + 8d N j-(r-l),

where
C2, C3
and
C4
are positive constants independent of I' { and 81 , The
inequalities for II and 12 lead to the estimate for

.11
11 W;+1 0
Let
0'
<{C2
_ -
= c4(1 + bdh.
'
11 w; 11 0
IlwHdlo in the form
+ c4(1+bI)N-(r-l)
j '
1
(7.27)
Choose {o so small that

(7.28)
For all { E [0, {oJ we obtain from inequality (7.27) the estimate
which on integration yields

j-l
II .
~ N-:(r-l)M+ II wM;lllo -< 0' L.J
~ N-:-(r-l)MIl ;+1 II 0 -< 0' L.J
;-11
;-11
w'
1I
11=0
11=0
The last inequality proves that the sequence Uj (j

with respect to the H(Tm ) norm whenever the series
+00
(j
= L.J
~ ~ N-:(r-l) M-II
L.J ;-11
j=O 11=0
= 1,2, ...) is convergent
converges. Here the difference of Uj and Uo
= 1.lim
.....
00
277
satisfies the estimate
Uj
(7.29)
where
is the remainder of the series
Uj
defined by the expression given in
Theorem 1. Inequalities (7.25) ensure that the sequence

compact in the space H'(Tm ) for s
For s
Uj
( (j
= 1,2, ...) is
< r. Then
> m/2+1 we have the inclusion H'(Tm ) C C/(Tm ) and the estimate
(7.30)
with the constant Cs independent of j. The conditions r

for s
= p.
>
> m/2 + I hold
Inequality (7.30) leads to the limit relation
where the number I is defined by the conditions of Theorem 1.

From inequalities (7.25) we obtain the analogous inequalities for the limit
function:
Thus the operator L( w) is defined for the function w
= uO and the difference
L(wO)uO - / has the estimate

IL(wO)uO - /10 $ IL(wo)uo - L(wj )uj+do+
+I(E - SNj+.)[L(wj)Uj+l -
/llo
$ C6[1Uj - uOlo+
+IUj+l - uOb + Nj-;:-l)(llujll._l + lI uj+lll. + II/II.)] $

$ c6(luj - uOlo + IUj+l - uOll + C7Nj-;:-l)e51),
for any j
= 1,2, ....
Consequently, the function U
solution of equation (7.13) for all the values of

We choose
and
C7 = 4 + r 2
= UO( = UO(>,)
is a
considered by us.
> 0 so small and 15 1 so large that all the conditions in the
above argument, in particular, inequalities (7.19), (7.24), (7.28), are satisfied.

Then, taking
equal to
0,
we can satisfy inequalities (7.11) for the right hand
278
CHAPTER 4
side of equation (7.13) with < = <0 by making inequalities (7.11) hold for the
right hand side of equation (7.2) with some small positive fl = fl(M). Then the
= <OUj(>,<o) = <OUj(
functions Uj(
define linear Galerkin approximations
to a solution of equation (7.2) and satisfy all the conditions of Theorem 1 since
(initial u) = < (new u).
To complete the proof of Theorem 1 it remains to verify the convergence
of the series
(7'
for any collection of integers
arbitrary M
> 1.
Nj
satisfying inequality (7.5) and
Since the sequence of numbers N j is strictly monotone, we have

Nj
and so
(7'
j=O,I,oo.,
~j+l,
can be majorized by the series

1
00
""
"" (j + 1 ~~
1I)r-1M"
= (7'0,
which converges since
00
00
<
- ""
LJ -Mil ""
~ (j + l)r-l -< -M
11=0
)=0
for any n
Moo
-
1 ""
LJ (j + 1)1+m/2 -)=0
M
M -c
1
> 1. This completes the proof of the
1 and an arbitrary M
theorem.
It should be noted that estimate (7.12), which characterizes the rate of
convergence of the approximations Uj( to the solution u( of equation (7.2),

essentially depends on the choice of the sequence of numbers
Nj
(j = 1,2, ...)
and the number AI. By changing them we can control the rate of convergence.
In particular, if
then the rate of convergence of Uj to
is given by the inequality
j+l
lIu - ujllo:::; 20 N(r-l)j'
= 1,2, ....
279
This is a consequence of the following estimate of Uj for the chosen values of

Nj
and M:
. _
uJ
~ ~ N-II(r-l) N(II-i)(r-l)
L..J L..J
i=j 11=0
- L..J
i=j
i+1
< 2(j + 1)
Ni(r-l) -
N(r-l)j'
Note also that inequality (7.12) leads to the estimate
which characterizes the rate of convergence of

G'(Tm ) norm. Here r
Uj
to
with respect to the
> p > m/2 + I and c is a constant independent of j.
Theorem 1 simplifies the problem of the justification of Galerkin's procedure for the construction of the invariant torus (1.1) of system (1.2). The
following statement holds [119].
Theorem 2. Suppose that the conditions of the previous theorem hold. Then
> 0 such that if f E H r (Tm ) and in-
there exists a sufficiently small Jl
equality (7.11) holds, then the Galerkin approximations WN() exist for any
= 1,2, ...
and converge in H$(Tm
n G'(Tm ) for s <
r to a function
u()
which defines an invariant torus (1.1) of system of equations (1.2); furthermore,
(7.31)
where G = const.
Proof. Consider equation (7.13). The Galerkin procedure defines the Nth ap-
proximation WN() to its solution u() E G1(Tm ) as the trigonometric polynomial (7.3) satisfying the equation
(7.32)
To solve equation (7.32) we use a linear modification of Galerkin's procedure, according to which we define the approximation Wk() to the solution
WN() of equation (7.32), starting with the initial approximation

the polynomial
wk()
II k ll$N
W~~1ei(k,4,
w,Z. = 0,
by
280
CHAPTER 4
satisfying the equation

.
SNL({Wlv- )Wfv()
= SN f(),
= 1,2,....
(7.33)
Equation (7.33) has the form of the equation considered in the proof of
the previous theorem. According to this theorem, the approximations Wj".()
exist for any j
= 1,2, ... and any fixed N = 1,2, ....
They satisfy inequalities
similar to (7.25):
for all j = 1,2, ... , s = 1, ... ,r, { E [0, (0] when (0 > 0 and IIfllo are sufficiently
small. An estimate of the difference Vj,/I() = Wj".+I() - Wj".() satisfying
the equation
shows that
II Vk+ 1 1l0 :S c{llVkllo:S ~llVkllo,

for all { E [0, {a] and sufficiently small
(0
= 1,2, ... ,
> O.
Inequalities (7.35) imply t.hat the sequence W!A) (j

in the HO(Tm ) norm as j -sequence Wj".() (j
j
-> 00
WN()
00.
= 1,2, ...)
(7.35)
= 1,2, ...) converges
Then inequalities (7.31) ensure that the
converges in the H'(Tm ) n C/(Tm
norm as
when s < r. On passing to the limit we see that the limit function
= .lim
;->00
Wj".() is a solution of equation (7.32) satisfying the inequalities
similar to (7.34):
for s
= 1, ... , rand ( E [0, (0] when (0 > 0 and Ilflio are sufficiently small.
Estimating the difference Wj".+I() - Wj".() with respect to the H(Tm )
norm, we obtain
. ,
rll Wlv+ I-
Wlvllo:S
'-1
'-I
'-I
cI(II Wlv+l - Wt" Ilo+c211(E-SN)[L((Wlv )Wt" - !l1l0:S
i-1
- II'V
<
_ CI(
Y N+I
Wi-III
- N-(r-I) ,
N
a + C3
where the constants CI and C3 do not depend on j and {. The last inequality
shows that
(7.37)
where
(;4
does not depend on Nand e E [0, eo].
281
It follows from inequality (7.37) that the sequence WN (4)) (N

1,2, ...)
converges to the function uO(4)) = lim WN(4)) as N - 00. Inequalities (7.36)
N ..... oo
guarantee that this sequence converges to the function uO(4)) with respect to
the H'(Tm ) n C/(Tm ) norm for s
< r. As can be seen by passing to the
limit in identity (7.32) with WN = WN(4)), the limit function is a solution of

equation (7.2). The estimate (7.31) for the difference uO(4)) - WN(4)) follows
from inequalities (7.37). We now complete the proof of Theorem 2 by setting
u( 4 = uO( 4 = uO( 4>, e) and taking into account the relationship between the
solutions of equations (7.13) and (7.2).
As for the case of Theorem 1, inequality (7.31) implies the following estimate for the difference u(4)) - WN(4)) with respect to the C/(Tm ) norm:
lu where r
I<
W N /_ CN-(r-1)(1-p/r) ,
> p > m/2 + I
and
C is
= 1,2, ... ,
a constant independent of N.
4.8. Proof of Moser's Lemma

In effect, we have already started the proof of Moser's lemma in the version
given in 3.15. To complete this proof it remains to determine more precisely
how the parameters I and {) in the inequality for the product (Lu, u), given by
(14.6) in Chapter 3 depend on the coefficients of the operator
m
()
= 2: av (4))"" v=l
P(4)).
It follows from inequalities (15.1) and (15.3) in Chapter 3 that for s
=1
(8.1)
where I and {) are positive constants which depend only on I:~1
la v 11 + IPh.
The following estimate clearly follows from estimate (8.1):
(Lu,
uh 2: Iliulli -
{)
(2: lIa
lh + 11P11i + lI ullo) lIulh-
v=l
Consider the product (Lu, u), for s 2: 2. Write it in the form
(Lu,u),
= (Liu,u)'_l +4>,(u),
(8.2)
282
CHAPTER 4
where
L~
is an elliptic operator of the form
-Hp
+ P* + L::~1 ~E), E vi is the identity matrix E for v = j

where bo =
and the zero matrix for v i j, and 4>.(u) is the functional equal to the difference
(Lu, 1.). - (L~u, 1.).-1-
Let s - 1 = 2p. Using standard transformations we find that
where 8' is a positive constant which depends on L:::;'=1 la v l2 + 1P11, and Gis
a positive constant independent of u and the coefficients of L 2 The operator
contains terms of the form D,-i Alii' Diu (i
= 0, _. _, s -
1) which admit the
estimate
m
II D - i Avi Diullo
~ G (2:
vJ=l
IAvi 10 + lulo)
v=1
v,i=l
(2:
vJ=l
IIAvi II.
+ /lull.) ~
~ G' (2: lavii + IPlo + IUlo) (2: /lav/l.+1 + /lPII, + /lu/l.) ,

m
/I \lip (u)IIo ~ G"Go(2:
v,i=l
II av 112p+1
+ 1IPI12p + IIull2P) '
where Go ~ L~l la v l2 + IPII + luloFrom thi~ we obtain the following inequality
(L 2 u, U).-l
~ r; lI ull; -
81
(2: Ilavll. + II P II.-1 + IIull.-1) lI ull,,

v=l
(8.3)
where
1~
283
and 61 are positive constants that depend only on Co.
Let s - 1 = 2p + 1. Then
-6" (l: Ilavl12P+l
v=1
+ II P II2p + lI u112P) lI u Il2p+2,
and by using Garding's inequality we obtain
-6" (l: lI avll2p+l
v=1
~ 12I1ull~p+2 -
+ 1IP1I2p + lI ulb) lIulb+2 ~
62
(2: lIavl12p+1 + 1IP1I2p + lIull2P) lIuI12p+2,

v=1
where the positive constants 12 and 62 depend only on Co.

Thus estimate (8.3) holds for both even and odd s 2: 2.
Let us estimate the functional 4.(u). Suppose that for s
= 2, ... ,p the
following estimate holds
14.(u)1 ~ c[ l: (l: IIDiavIYulio + IIDiPDjullo) +

i+j~.
j'F'
+ l: l: IIDi(D2av)Dj(Du)lIo] lIull.,
2+i+j~.
where c
= const.
We now prove that a similar estimate holds for s
(8.4)
= p + 1.
It follows from the definition of the functional 4.(u) that for s =
p+ 1 we
have
l4p+1(U)1 ~ [l:lla v:;Jp-1

v
+II p ull p-1 +l:(II~av :uJp_1 +II:~ ull p+

v
+11:;: ~ullp_J +~II O~~~~j O::;i IIp-J lIullp+1 +l:14p (:uJ I~

v,t,1
1/
CHAPTER 4
284
2
L II Di (D a,,)lY(Du)lIo
i+j~p-l "
L
i
IID P Dju llo+
L
i+j~p+l
i~lJ~2
i
IID PlYullo+
i+j~p+l
i~l
]lI
i 2
L
LIID (D a,,)lY(Du)lIo u IIP+1'
i+j<p-l "
j~l
whence it follows that
+IIDiPLYullo)
L
i+j~p-l
LIIDi(D2a,,)LY(Du)lIo]lIullp+l'
"
(8.5)
Using inequality (8.4) for fPp(ou/o</J,,), we find that
i+j~p-2
~c[
2:
LIIDi(D2a,,)Dj(DZu)lIo]lIullp+1~
"
(2:IIDia"Djullo+IIDiPLYullo)+
"
i+j~p+l
jp+I
L
i+j~p-l
i
2
LIID (D a,,)lY(Du)lIo]lI u llp+l'
"
(8.6)
Estimate (8.4) for the functional fP p+ 1 (u) follows from inequalities (8.5)
and (8.6).
It follows from the calculations in 3.15 for s = 2 that
~ oP ou
+2 LJ o</J' o</J.
j
+ (~P)u,u
)
1
and thus inequality (8.4) holds for fP 2 ( u):

IfPz(u)1
c[ L
i+j~Z
#2
( L IID a"LYullo
"
+ IIDi PlYullo) +
2: IIDza"Dullo] lIullz,
"
where c
= const.
We see by induction that estimate (8.4) holds for all s
285
~
2.
From the inequalities
[IiDiavLYullo
+ IIDi PLYullol:::;
i+i~'
[IiDiavLYull+
i+i~.-l
#.
+IIDi PDiuliol +
L:
i+i=-2
[II Di (Dav)LY(Du)lIo+
+IIDi Di(tw)lIo1+:::; Cd(lavlo + IPlo + lulo)(lI avll.-l+

+IIPII.-l + lI ull.-d + (Iavh + IPh + luh)(llavll'-l + II PII.-l + lI ull.-l)]'
L:
IID i (D 2av )Di(Du)lIo:::; C2(lav b + luld(lIavll.
+ lIull.-d,
i+j~.-2
we obtain for
<fl. (u)
the estimate
1<fl.(u)1 :::; C. (L: la v 12 + IPh + lull) (L: lI avll. + IIPII. + lIull.-l) lI ull.,
v=l
v=l
where C. does not depend on a, P or u.
The last estimate together with estimate (8.3) proves that
(Lu, u).
where i and
~ illull; -
5(L: lIavll. + IIPII. + lIull.-l) IIUIl.,

v=l
5 are positive constants which depend only on Co.
Now if we take into consideration the fact that for any
> 0,
(L: lI a vll. + IIPII. + lI ull.- l ) :::;

v=l
: :; (L: Ilavll. + IIPII.)

v=l
/(2f) + lI uI15/(2f2. - l ) + 2fllull;,
we finally obtain for (Lu, u). estimate (7.9) of the lemma.

4.9. Invariant tori of systems of differential equations
with rapidly and slowly changing variables
We shall apply the results of the perturbation theory of an invariant torus of a
system of differential equations to systems of equations that contain slow and
rapid variables. We consider several types of such equations.
286
CHAPTER 4
10 Equations with slowly changing phase
By this we mean a system of equations of the form

d
dt
= w(,X,t"),
dx
di =
X(,X,f),
(9.1)
the right hand side of which is defined in the domain

xED,
ETm ,
= 1, ... , m)
and is periodic in v (v
n-dimensional Euclidean space En,

small positive parameter.
We shall assume that for f
equilibrium points x
= u():
fE[O,fO]
(9.2)
with period 211". Here D is a domain in
T.n is an m-dimensional torus and ( is a
= 0 system (9.1) has an m-parameter family of
X(, u(),O)
= 0,
E Tm
(9.3)
which cover the torus

x
= u(),
E Tm ,
u E C(Tm
(9.4)
).
By applying the perturbation theory of an invariant torus to system (9.1) we

can obtain the following result on the invariant tori of this system of equations.
Theorem 1. Let the functions a and X have partial derivatives with respect
to , x up to order r which are continuous with respect to , x, ( in the domain
(9.2). Suppose that torus (6.4) consists of equilibrium points of system (9.1)
when ( = 0, that a neighbourhood of it lies in the domain D, and that u E
Cr(Tm ).
Ifr
2 and the matrix P()
= 8X(,u(),0)/8x
admits the representa-
tion
(9.5)
where T() E cr-1(Tm
and the eigenvalues of the matrices D 1(), D 2 () have
positive real parts for all E T m
then there exists a sufficiently small
such that for any ( E [0, (0) system (9.1) has an invariant torus
(0
>0
287
satisfying the condition

limlu(>,E)-u(1_2L
=0,
r
Ip
-0
(9.6)
and this torus is exponentially stable when PC~ = -T(D1 (T- 1 (, and
is unstable when PC~ ::j; -T(D 1 (T-l(. When r ~ 3, this torus is
exponentially dichotomous.
h,
Proof Under the above assumptions on the variables > and x - u(
system (9.1) takes the form of system (2.1) in which the functions a, f and the
matrix P defined in the domain (2.2) for sufficiently small d > 0 and EO > 0 have
partial derivatives with respect to >, h up to order r-l which are continuous in
>, h, (. Here the system of variational equations corresponding to the invariant
torus (9.4) of the generating system of equations has the form (2.5) with ao = 0
and
poe~
pc~
= oX(>,u(,O)jox.
In this case the fundamental matrix n~ ( of the system of variational

equations is equal to the matrix eP(.p)(t-T). We set
where It = diag{E 1 ,0}, E 1 is the identity matrix of dimension equal to that

of the matrix D 1 (. The matrix C( is a projection and belongs to the space
cr-l (Tm ) C Cl (Tm ). Moreover, under the assumptions concerning the matrix
PC~
we have:
lI eP(.p)(t-T)C(1I
for all t
for all t
~ T
and
~ T
and
= IIT()diag{e- DM )(t-T),0}T- 1 (1I
I(l e -""'Il(t-T)
E R,
E R, where
1(1,1<2,/1
and /2 are positive constants in-
dependent of E Tm . Thus the trivial torus of the system of variational

equations corresponding to the invariant torus (9.4) of system (9.1) for
exponentially dichotomous (stable if P( )
= - T( > )D 1 (> )T-l (
= 0 is
and unstable
otherwise) .
The derivative of the matrix eP(.p)t with respect to the variable >" (/I
1, ... , m) satisfies the identity
oeP(.p)t
-I
a>" -
P(.p)(t-T)OP(
e
o>" e
P(,p)Td
T
(9.7)
288
CHAPTER 4
for all t E Rand </J E Tm . If we replace the matrix P(</J) by the matrix - D 1(</J)
in (9.7) we find that
II {)e~:~()t II ~ I<;
<
-
it
e--r.<t-T) II
y21 ()D1(</J)
Ite--r,t
()</Jv
\1
()~~:</J) Ile--r'T dr ~
t 2::
'
o.
(9.8)
Similarly, we find that

(9.9)
Inequalities (9.8) and (9.9) imply that the matrix S(</J) defined by relations
(9.13), (9.14) of Chapter 3 is continuously differentiable with respect to </Jv
= 1, ... ,m).
Since for this S(</J) the matrix S(</J) S(</J)P(</J) + P*(</J)S(</J)

is negative-definite for all E Tm , all the conditions of Theorem 1 of 4.3 or
(II
Theorem 2 of 4.5 hold for system (9.1) written in terms of the variables ,
x - u()
= h. The assertions of these theorems prove Theorem 1.
It should be noted that the statements of Theorem 1 remain true when the
matrix P(</J) has the representation (9.5) with the matrix T(</J) whose period
= 1, ... ,m) is a multiple of 27r. Indeed, if the period of

with respect to v (v = 1, ... , m) is equal to 27rp, where p
with respect to v (II

the matrix T(</J)
is an integer, p
2:: 2, then the substitution
= p'IjJ
transforms this system of
equations to a system of equations of the same kind, but now satisfying all the
conditions of Theorem 1.
Then the system of equations (9.1) has an invariant surface
x
( A.)
U '1',(,
A.
'I'
E Rm ,
(9.10)
where the function u( </J, () has period 27rp with respect to </Jv (v
Since the right hand side of system (9.1) is periodic in v (v
= 1, ... , m).
= 1, ... , m) with
period 27r, it follows that the surfaces given by the equations
(9.11)
where k
= (k 1 , ... , km ) is an integer vector for which Ikl = l::;':llkvl < p, are
invariant surfaces of this system.
289
All the surfaces (9.11) lie in a b-neighbourhood of surface (9.10) since
lI u(,()-u(+2k7r,()1I
~
where 1J()
lIu(, () - u()11
---+
0 as (
---+
+ lIu( + 2k7r, () -
u( + 2k1r)1I
1J(),
O. As follows from the proofs of Theorem 1 of 4.3
and Theorem 2 of 4.5, the invariant set (9.10) of system (9.1) has the stability
or instability property that there are no whole trajectories of system (9.1) in
some b-neighbourhood of it which does not depend on (. For sufficiently small
(0
> 0 we have the inequality 7]() < b for all ( E [0, (0], and this leads to the
identities
u(,()=u(+2k1r,(),
for all integer k
= (k
1 , . .
ER m
(9.12)
,k m ) and all ( E [0, (01. Identities (9.12) show that
u(,t) E C(Tm ) for all (E [0,(0], as required.

It follows from this remark and Sibuja's Theorem given in 1.9 that even
without the requirement expressed by relation (9.5) the statements of Theorem 1 remain true for m = 1. In particular, this leads to the following statement
for systems of the form
dx
di
where r
= X(r,x),
(9.13)
= a.
Theorem 2. Let the right hand side of the system of equations (9.13), defined
in the domain
XED,
r E R,
(E [0, (0],
be periodic in r with period 211" and have derivatives up to order r which are
continuous in r, x. Suppose that the equation
X(r,x)=O
has a solution
x=u(r), rE[O,211"1
that is periodic in r with period 211", some neighbourhood of which lies in the
domain D, the function u(r) having continuous derivatives with respect to r
up to order r.
290
CHAPTER 4
If r
2 and the real parts of the eigenvalues of the matrix per) =
{}X(r,u(rj{}x are non-zero for all r E [0,211"], then there exists a sufficiently
small {o >
such that for any { E (0, {oj system (9.13) has a solution
x=x(r,{),
which is periodic in r
=d
r={t
with period 211" and satisfies the condition

limlx(r,t)-u(r)1 r - 1=0.
(9.14)
(-+O
Furthermore, this solution is asymptotically stable if the real parts of the eigenvalues of the matrix P( r) are negative and is unstable otherwise.
To prove Theorem 2 we further note that x( T, {) is a solution of the equation obtained from (9.13) by substituting d
T for t. Thus, the function
x( r, {) is r
+ 1 continuously differentiable
with respect to r for any { E (0, {oj.
This enables us to replace in relation (9.6) the norm
I. Ir-2,Lip
by
I . Ir-1
for
the difference x( r, t) - u( r), and this completes the proof of Theorem 2.

To illustrate the statements of Theorem 2, we consider the system defined
by Van der Pol's equation
(9.15)
where a(T), j3( T), f( T) are twice continuously differentiable functions which are
= d is slowly changing time. By introducing

= dxjdt, equation (9.15) takes the form of system (9.13) which
periodic in T with period 211", and r

the variables x, y
consists of two equations. The function u(r) is defined by the expression
= fer),
= 0,
= [0,211"],
and the matrix P( r) by the equation
per) = [ 0
-1
1
] .
a(r)(1 - j3(T)f2(r
The eigenvalues of the matrix P( T) have negative real parts if
a(r)(I- j3(r)f2(T < 0,
r E [0,211"],
(9.16)
T E [0,211"].
(9.17)
and positive real parts if
a(r)(1- j3(r)fZCT > 0,
291
Consequently, if inequalities (9.16) hold, then for all sufficiently small c
>0
Van der Pol's equation has a solution
X=X(1',C),
1'=d
which is asymptotically stable periodic in
(9.18)
= ct with period 27T, while if
l'
inequalities (9.17) hold, then it has an unstable solution (9.18), and this solution
=0
lim IU(1',C) - /(1')11
<-0
in both cases.
2. Periodic systems
0/ equations
By this we mean a system
with rapidly and slowly changing time
dx
di =
X(t, 1', x),
(9.19)
the right hand side of which is defined in the domain

xED,
and periodic with respect to t,
l'
t E H,
l'
(9.20)
ER
= ct with period 27T.
We shall suppose that the system of equations (9.19) considered as a system

of equations that depends on the parameter
l'
has a solution
X=U(t,1')
for all
E [0,27T] that is periodic in t and
l'
(9.21)
with period 27T. We write down
the variational equation for solution (9.21) of system (9.19) for

dy
dt
and denote by
n~( 1')
8X(t, T, u(t, T))

8x
y,
l'
= const:
(9.22)
the fundamental matrix of its solutions which is equal to
the identity matrix when t
= O.
Theorem 3. Let the right hand side of system (9.19) be r times continuously
differentiable with respect to t,
T
T, X
in the domain (9.20) and periodic in t and
with period 27T. Suppose that when
= const
some neighbourhood of the
solution (9.21) of system (9.19) lies in the domain D for all (t, 1') E R x Rand
that
E C r (72).
292
CHAPTER 4
If r
2: 2 and the eigenvalues of the matrix n5"(r) lie on curves that do
not intersect the unit circle and do not encircle the origin of the complex plane
when r E [0,271"], then there exists a sufficiently small {o
> 0 such that for any
{ E [0, to] system (9.19) has a solution

x
that is periodic in t and r
= tt
= x(t, r,{), r = tt
with period 271" and satisfies the condition
lim IX(t,T,t) - u(t,r)lr_l
(-+0
= 0;
furthermore this solution is asymptotically stable if the eigenvalues of the matrix
nij" (T) lie inside the unit circle with centre at the origin and is unstable if there
are eigenvalues lying outside this circle.
Proof Using the coordinates <fJ = (0, T), x - u(O, T) = h, we rewrite system
(2.1) in the form of system (9.19), where a
= (l,t), f = -au(O,r)/aT and P
is a function defined in the domain (2.2) for sufficiently small d
> 0 and to > 0
and having partial derivatives with respect to <fJ, h up to order r - 1 which are
continuous in <fJ, h, {.
The system of variational equations corresponding to the invariant torus
of system (9.19) for
= 0, written in the coordinates <fJ,
x - u(<fJ)
= h has the
form (2.5) with ao = (1,0) and Po(<fJ) = aX(<fJ, u(<fJ/ax. Consequently, it is

equivalent to the system
dh
dt
where
= aX(t + 0, T, u(t + 0, T h
ax
(9.23)
and T are parameters. The fundamental matrix nMo, T) of solutions
of system (9.23) is clearly related to the fundamental matrix
nM r) of solutions
of system (9.22) by the formula
nt(O, T)
= n~+9(T)
for all (t, 0, T) E R 3 .

The right hand side of system (9.22) is periodic in t with period 271". Therefore we can apply to it the Floquet-Lyapunov reducibility theorem, according
to which,
293
where 4>(t, r) is a real periodic matrix with respect to t with period 411", 411" H( r)
is the real branch of the logarithm of the matrix [n~"'(r)j2:
H( r)
1
=~
In[n~'' (r)]2 = -4
In n~'" (r).
411"
11"
Since n~"'(r) E Cr-1(T1), we see that under the above assumptions it

follows from the lemma on the logarithm given in 1.9 that H(r) is also a
matrix in Cr-1(Td. Moreover, according to Sibuja's theorem also given in
1.9, the matrix H(r) can be represented as
where T( r) is a real matrix that is periodic in r with period 411" and has continuous derivatives withe respect to r up to order r -1, and D1(r) and D 2(r)
are matrices, the real parts of the eigenvalues of which are positive for all
r E [0,411"].
The matrix eH(T)t can be represented as
eH(T)t = T(r)diag{e- D1 (T)t,e D ,(T)t}T- 1(r),
where for all r E [0,411"]
lIeD2(T)tll
:S K 2e "Y2 t , t:S 0,
and K}, [(2, rl and r2 are positive constants independent of T. If we set

<p = r, then the derivatives of the matrices e- D1 (T)1 and eD,(T)t with respect
to r satisfy inequalities (9.8) and (9.9).
We set
where h = diag{ E 1 , O} and E 1 is the identity matrix of the same dimension as
D1(r). The matrix C(O, r) is a projection which is periodic in 0, r with period

411" and has continuous partial derivatives up to order r - 1 with respect to 0, r
for all (0, r) E R 2 .
Since
294
CHAPTER 4
it follows from the estimates of the matrix
e-D1(T)t
and its derivatives with
respect to r that the matrix SI(B, r), defined in terms of
n~(B, r)C(B,
r) by
relation (9.14) in Chapter 3, is continuously differentiable with respect to B, r

and periodic in to B, r with period 47r.
Similar reasoning shows that the matrix Sz(B, r), defined by relation (9.14)
in Chapter 3 in terms of nh(O, r)C1(B, r), where C1(0, r) = E - C(B, r), is

continuously differentiable with respect to B, r and differentiable and periodic
with respect to B, r with period 411". Then the matrix
5(0, r)
= 8S~, r) + S(B, r)Po(B, T) + Pci(O, r)S(B, T)
corresponding to S(0, r) = Sl (e, T) - Sz (e, T) is negative-definite for all (e, r) E
[0,411"1 x [0,411"].
This is sufficient for syst.em (9.19), written in terms of the variables tj> =
(B,T),
X -
u(B,r)
= h,
to satisfy all the conditions of Theorem 1 of 4.3 or
Theorem 2 of 4.5. We complete the proof of Theorem 3 by taking into account

the reasoning given after the proof of Theorem 1 and the assertions of those
theorems.
As an example illustrating Theorem 3 we consider the system given by the
equation
d2x
dx
(9.24)
dt 2 +V.(r)dt"+n (r)x=F(r)coswt,
where A( r), n( T) and F( T) are periodic functions in T = ft with period 211" which
# 0,
n( T) # for
all T E [0,211"]. In terms of the variables x, y = dx/dt we can write equation
are twice continuously differentiable and are such that A( T)
(9.24) in the form of system (9.19), which consists of two equations.

The function u(t, T) defined by the relations
x=u(t,r)=
a(r)
= tan
-I
F(T)Cos(wt-a(r
,
.j[n 2 (r) - wZ]2 + 4W 2 A2 (T)
2wA(r)
2()
2'
T -w
r E [0,211"],
(9.25)
is a periodic solution of equation (9.24) for constant r. The variational equation

corresponding to the torus (9.25) is the same as equation (9.24) when F == 0;
consequently the quantities
AI,::!
= -A(r) .jV(r) -
n 2 (r),
r E [0,211"],
295
are the eigenvalues of the matrix H(T).

When ..\( T) < 0 or ..\( T) > 0 for all T E [0,211"] the conditions of Theorem 3
hold; consequently for all sufficiently small
> 0 equation (9.24) has a solution
X=X(t,T,f)
that is periodic in t and
=d
with period satisfying the condition
lim Ix(t,
<-0
T,
f) - u(t, T)ll = O.
This solution is asymptotically stable for ..\(T) < 0 and unstable for ..\(T) > O.
Bibliography
1. p.s. Aleksandrov, Combinatorial topology, Gostekhizdat, Moscow, 1947 (in
Russian).
2. A.A. Andronov, Collected works, Izdat. Akad. Nauk. Moscow, 1956 (in
Russian).
3. A.A. Andronov and A.A. Witt, 'On quasi-periodic motions'. In: Collected
works of A.A. Andronov, Izdat. Akad. Nauk SSSR, Moscow, 1956 (in Rus-
sian).
4. - , -
and S.E. Haikin, Oscillation theory, Fizmatgiz, Moscow, 1959 (in
Russian).
5. V.1. Arnol'd, 'Small denominators I: On the mappings of a circle onto
itself', Izv. Akad. Nauk SSSR Ser. Mat. 25:1 (1961),21-86.
6. - , 'Small denominators. A proof of A.N. Kolmogorov's theorem on conservation of conditionally periodic motions under a small change of the
Hamiltonian', Uspekhi Mat. Nauk 18:5 (1963), 13-40.
7. - , 'Small denominators and the stability problem in classical and celestial
mechanics', Uspekhi Mat. Nauk 18:6 (1963),91-192.
8. - , Mathematical methods in classical mechanics, Nauka, Moscow, 1974
(in Russian).
9. -, Supplementary chapters to the theory of ordinary differential equations,

Nauka, Moscow, 1978 (in Russian).
10. N.!. Akhiezer, Lectures on approximation theory, Nauka, Moscow, 1965 (in
Russian).
11. A.S. Bakai and Yu.P. Stepanovskii, Adiabatic invariants, Naukova Dumka,
Kiev, 1981 (in Russian).
12. Ya.S. Baris, 'On the integral manifold of a non-regularly perturbed differential system, Ukrain. Mat. Zh. 20 (1968), 439-448.
297
BIBLIOGRAPHY
298
13. -
and V.I. Fodchuk, 'Study of bounded solutions of non-linear non-regu-
larly perturbed systems using the integral manifold method', Ukrain. Mat.
Zh. 22 (1970), 3-11.
14. L. Bers, F. John and M. Schechter, Partial differential equations, Wiley,

New York, 1964.
15. Yu.N. Bibikov, 'On the existence of invariant tori in a neighbourhood of
an equilibrium position of a system of differential equations', Dok/. Akad.
Nauk SSSR 185 (1969), 9-13.
16. -
and V.A. Pliss, 'On the existence of invariant tori in a neighbourhood
of the zero solution of a system of ordinary differential equations', Differentsial'nye Uravneniya 3 (1967),1864-1881.
17. G.D. Birkhoff, Dynamical systems, American Mathematical Society, New

York, 1927.
18. Yu.S. Bogdanov, 'On the transformation of a variable matrix to canonical
form', Dok/. Akad. Nauk IJSSR 7:3 (1963), 152-154.
19. N.N. Bogolyubov, On some statistical methods in mathematical physics,
Izdat. Akad. Nauk Ukrain. SSR, Kiev, 1945 (in Russian).
20. -,'On quasi-periodic solutions in problems of non-linear mechanics, In:
Proc. First Summer Math. School, Naukova Dumka, Kiev, 1964, pp. 11-
102 (in Russian).

21. - , Collected works, Vol. 1, Naukova Dumka, Kiev, 1969 (in Russian).
22. - and Yu.A. Mitropolskii, 'The method of integral manifolds in non-linear
mechanics', In: Proc. Internal. Conf. on Non-linear Oscillations Part I,
Analytic methods, Inst. Mat. Akad. Nauk Ukrain. SSR, Kiev, 1963, pp. 93-
154 (in Russian).

23. -
and -, Asymptotic methods in the theory of non-linear osci//ations,

24. - , -
and A.M. Samoilenko, The accelerated convergence method in non-
linear mechanics, Naukova Dumka, Kiev, 1969 (in Russian).
25. P.G. Bol, Collected works, Akad. Nauk Latv. SSR, Riga, 1961 (in Russian).
26. H. Bohr, Almost periodic functions, Gostekhizdat, Moscow-Leningrad, 1934
(in Russian).
27. I.Yu. Bronshtein, Extensions of minimal transformation groups, Shtiintsa,
Kishinev, 1975 (in Russian).
BIBLIOGRAPHY
299
28. - , 'Weak regularity and Green's functions of a linear extension of dynamical systems', Differentsial'nye Uravneniya 19 (1983), 2031-2038.
29. -
and V.P. Burdaev, 'Invariant manifolds of weakly linear extensions of
dynamical systems', Preprint Akad. Nauk Moldavsk. SSR, Kishinev, 1983.

30. -
and V.F. Chernii, 'Linear extensions satisfying the exponential di-
chotomy condition', Izv. Akad. Nauk Moldavsk. SSR 3 (1976), 12-16.

31. N.V. Butenin, Elements of the theory of non-linear oscillations, Sudpromgiz., Leningrad, 1962 (in Russian).
32. B.F. Bylov, 'On the structure of solutions of a system of linear differential
equations with almost-periodic coefficients', Mat. Sb. 66 (1965), 215-229.
33. - , 'On the reduction of a system of differential equations to diagonal
form', Mat. Sb. 66 (1965), 338-344.
34. - , R.E. Vinograd, V.Ya. Lin' and O.V. Lokutsievskii, 'On the topological
reasons for the abnormal behaviour of certain almost periodic systems, In:
Problems of asymptotic theory of non-linear oscillations, Naukova Dumka,
Kiev, 1977, pp. 54-61 (in Russian).

35. - , - , -
and - , 'On topological obstacles to block diagonalization of
exponentially split almost-periodic systems', Preprint No. 58, Inst. Prikl.

Mat. Akad. Nauk SSSR, Moscow, 1977.
36. V. Vazov, Asymptotic expansions of solutions of ordinary differential equations, Mir, Moscow, 1968 (in Russian).
37. B. Van der Pol, Non-linear theory of electric oscillations, Svyaz'tekhizdat,

Moscow, 1935 (in Russian).
38. V.M. Volosov and V.l. Morgunov, The averaging method in the theory of
non-linear oscillatory systems, Moscow University Press, Moscow, 1971 (in
Russian).
39. F.R. Gantmakher, Theory of matrices, Nauka, Moscow, 1967 (in Russian).
40. V.L. Golets, 'On the perturbation problem for a stable invariant torus of
a dynamical system', Ukmin. Mat. Zh. 23 (1971), 130-137.
41. Eu.A. Grebenikov and Yu.A. Ryabov, Constructive methods of analysis of
non-linear systems, Nauka, Moscow, 1979 (in Russian).
42. Yu.L. Daletskii and M.G. Krein, Stability of solutions of differential equations in Banach space, Nauka, Moscow, 1970 (in Russian).
43. B.P. Demidovich, Lectures on the mathematical theory of stability, Nauka,

300
BIBLIOGRAPHY
44. P.P. Zabreiko and S.O. Strygina, 'Cesari's equalities and Galerkin's procedure for finding periodic solutions of ordinary differential equations', Dokl.
Akad. Nauk Ukrain. SSR Ser. A 7 (1970), 583-586.
45. K.V. Zadiraka., 'On the integral manifold of a system of differential equations containing a small parameter', DoH. Akad. Nauk SSSR 115 (1957),
646-649.
46. V.1. Zubov, Lyapunov's methods and their applications, Leningrad State
Univ. Press., Leningrad, 1957 (in Russian).
47. - , Stability of motion, Vysshaya Shkola, Moscow, 1973 (in Russian).

48. A.N. Kolmogorov, 'On the preservation of conditionally periodic motions
under a small change of the Hamiltonian', DoH. Akad. Nauk SSSR 98
(1954), 527-530.
49. - , 'General theory of dynamical systems and classical mechanics', In:
International Math. Congress in Amsterdam, Fizmatgiz, Moscow, 1961,
187-208 (in Russian).
50. -
and S.V. Fomin, Elements of the theory of functions and functional
analysis, Nauka, Moscow, 1968 (in Russian).
51. M.A. Krasnosel'skii, 'On t.he theory of periodic solutions of non-autonomous
differential equations', Uspekhi Mat. Nauk 21:3 (1966), 53-74.
52. - , The shift operator along trajectories of differential equations, Nauka,

53. - , V.Sh. Burd and Yu.S. Kolesov, Non-linear almost-periodic oscillations,

Nauka, Moscow (1970) (in Russian).
54. A.N. Krylov, On some differential equations of mathematical physics that

have applications in technical problems, Izdat. Akad. Nauk SSSR, Moscow,
1950 (in Russian).
55. -
and N.N. Bogolyubov, Application of methods of non-linear mechanics
to the theory of stationa7'Y oscillations, Izdat. Akad. Nauk Ukrain. SSR,
Kiev, 1934 (in Russian).
56. -
and -, New methods in non-linear mechanics, ONTI, Moscow-Lenin-
grad, 1934 (in Russian).
57. - and - , Introduction to non-linear mechanics, Izdat. Akad. Nauk Ukrain.

SSR, Kiev, 1937 (in Russian).
58. V.L. Kulik, 'Quadratic forms and dichotomy of solutions of linear differential equations', Ukrain. Mat. Zh. 34 (1982), 43-49.
BIBLIOGRAPHY
301
59. - , 'Converse of a theorem on the decomposability of linear extensions of

dynamical systems on a torus', Ukrain. Mat. Zh. 35 (1983), 67-72.
60. A.G. Kurosh, A course in higher algebra, Nauka, Moscow, 1968 (in Russian).
61. Y. Kurtsveil', 'Invariant manifolds of differential systems', Differentsial'nye
Uravneniya 4 (1968), 785-797.
62. O.A. Ladyzhenskaya, Boundary-value problems in mathematical physics,

63. LA. Lappo-Danilevskii, Application of matrix functions to the theory of
linear systems of ordinary differential equations, Gostekhizdat, Moscow,
1957 (in Russian).

64. B.M. Levitan, Almost-periodic functions, Gostekhizdat, Moscow, 1953 (in
Russian).
65. -
and V.V. Zhikov, Almost-periodic functions and differential equations,
Moscow State Univ. Press, Moscow, 1978 (in Russian).

66. D.K. Lika and Yu.A. Ryabov, The methods of iteration and majorizing
Lyapunov's equations in the theory of non-linear oscillations, Shtiintsa,
Kishinev, 1974 (in Russian).

67. L.A. Lyusternik and V.1. Sobolev, Elements of functional analysis, Nauka,
68. A.M. Lyapunoy, The general problem on stability of motion, Gostekhizdat,
69. - , Collected works, Vol. 2, Izdat. Akad. Nauk SSSR, Moscow, 1956 (in
Russian).
70. LG. Malkin, The methods of Lyapunov and Poincare in the theory of nonlinear oscillations, Gostekhizdat, Moscow, 1949, (in Russian).
71. - , Some problems of the theory of non-linear oscillations, Gostekhizdat,

Moscow (1956) (in Russian).
72. - , The theory of stability of motion, Nauka, Moscow, 1966 (in Russian).
73. L.1. Mandel'shtam, Lectures on oscillation theory, Nauka, Moscow, 1972
(in Russian).
74. D.L Martynyuk and A.M. Samoilenko, 'Invariant tori of systems with aftereffect', Godishnik Vissh. Uchebn. Zaved. Prilozhna Mat. 11 (1976),47-53.
75. - and N.S. Tsiganovskii. 'Invariant manifolds of a system with delay under
the action of impulses', Differentsial'nye Uravneniya 15 (1979), 1783-1795.
302
BIBLIOGRAPHY
76. Yu.A. Mitropol'skii, Non-stationary processes in non-linear oscillatory systems, Izdat. Akad. Nauk Ukrain. SSR, Kiev, 1955 (in Russian).
77. - , 'On the construction of the general solution of non-linear differential

equations using an "accelerated" convergence method', Ukrain. Mat. Zh.
16 (1964), 475-501.
78. - , 'An accelerated convergence method in problems of non-linear mechanics', Funkcialaj Ekvacioj 9 (1966), 27-42.
79. - , 'On the construction of solutions and the reducibility of differential
equations with quasi-periodic coefficients', In: Third USSR Congress on
Theoretical and Applied Mechanics, Nauka, Moscow, 1968, pp. 212-213
(in Russian).
80. -, The averaging method in non-linear mechanics, Naukova Dumka, Kiev,
1971 (in Russian).
81. -
and O.B. Lykova, Integral manifolds in non-linear mechanics, Nauka,

82. -
and A.M. Samoilenko, 'On the structure of trajectories on toroidal
manifolds', DoH. Akad. Nauk Ukrain. SSR, Ser. A 8 (1964), 984-985.

83. -
and -, 'On the method of accelerated convergence in non-linear mech-
anics', In: Methods of integral manifolds in non-linear differential equations, Inst. Mat. Akad. Nauk Ukrain. SSR, Kiev, 1973, 5-31.
84. - and - , 'Some problems in the theory of multi-frequency oscillations', In:

VII Internationale Konferenz iiber Nichtlineare Schwingungen 1, Academic-
Verlag, Berlin, 1977, No.4, pp. 107-116.

85. -
and - , 'Some problems in the theory of multi-frequency oscillations',
Preprint No. 14, Inst. Akad. Nauk Ukra in. SSR, Kiev, 1977.
86. - , -
and V.L. Kulik, 'The study of linear differential equations using
quadratic forms', Preprint No. 10, Inst. Mat. Akad. Nauk Ukrain. SSR,
Kiev, 1982.
87. - , - and D.l. Martynyuk, Systems of evolutionary equations with periodic
and conditionally periodic coefficients, Naukova Dumka, Kiev, 1984 (in
Russian).
88. - , -
and K.B. Tsidylo, 'On invariant toroidal manifolds of non-linear
systems with delay', In: Differential equations with deviating arguments,

Naukova Dumka, Kiev, 1977, 207-214 (in Russian).
89. J. Moser, 'A rapidly convergent iteration method and non-linear differen-
BIBLIOGRAPHY
303
tial equations', Uspekhi Mat. Nauk 23:4 (1968), 179-238.
90. - , 'On the decomposition of conditionally periodic motions into convergent power series', Uspekhi Mat. Nauk 24:2 (1969), 165-211.
91. - , Lectures on Hamiltonian systems, Mir, Moscow, 1973 (in Russian).
92. Yu.I. Neimark, The method of point mappings in the theory of non-linear
oscillations, Nauka, Moscow, 1972 (in Russian).
93. V.V. Nemytskii, 'Oscillations in autonomous systems', In: Fifth Summer
Math. School, Naukova Dumka, Kiev, 1968,436-472 (in Russian).
94. - and V.V. Stepanov, Qualitative theory of differential equations, OGIZ
GITTL, Moscow-Leningrad, 1947 (in Russian).
95. G.S. Osipenko, 'Perturbations of dynamical systems near invariant manifolds' I, Differentsial'nye Uravneniya 15 (1979), 1967-1979; II 16 (1980),
620-628.
96. N.A. Perestyuk, 'Invariant sets of one class of discontinuous dynamical
systems', Ukrain. Mat. Zh. 1 (1984),63-69.
97. A.I. Perov, Variational methods in the theory of non-linear oscillations,
Voronezh Univ. Press, Voronezh, 1981 (in Russian).
98. V.A. Pliss, 'The reduction principle in the theory of stability of motion',
Izv. Akad. Nauk SSSR Ser. Mat. 28 (1964),1297-1324.
99. - , Non-local problems in oscillation theory, Nauka, Moscow-Leningrad,
1964 (in Russian).
100. H. Poincare, On curves defined by differential equations, OGIZ GITTL,
Moscow-Leningrad, 1947 (in Russian).
101. - , Collected works, I, II, Nauka, Moscow, 1971, 1972 (in Russian).
102. C. Pugh, 'A closure lemma', Matematika 12 (1968), 80-135 (in Russian).
103. A.M. Samoilenko, 'On the structure of trajectories on a torus', Ukrain.
Mat. Zh. 16 (1964),769-782.
104. - , 'On the reducibility of a system of ordinary differential equations in a
neighbourhood of a smooth toroidal manifold', Izv. Akad. Nauk SSSR Ser.
Mat. 30 (1966), 1047-1072.
105. -, 'On the reducibility of systems of linear differential equations with
quasi-periodic coefficients', Ukrain. Mat. Zh. 20 (1968), 279-281.
106. - , 'On the preservation of an invariant torus under perturbations', Izv.
Akad. Nauk SSSR Ser. Mat. 34 (1970), 1219-1240.
107. - , 'On perturbation theory of invariant manifolds of dynamical systems',
304
BIBLIOGRAPHY
In: Proc. Fifth Internat. Congress on Non-linear Oscillations I, Analytical

Methods, Inst. Mat. Akad. Nauk Ukrain. SSR, Kiev, 1970, pp. 495-499
(in Russian).
108. - , 'Study of dynamical systems by using functions of constant sign',
Ukrain. Mat. Zh. 24 (1972), 374-384.
109. - , 'On the exponential stability of an invariant torus of a dynamical system', Differentsia/'nye Uravneniya 11 (1975), 820-834.
110. - , 'Quasi-periodic solutions of systems of linear algebraic equations with
quasi-periodic coefficients'. In: Analytical methods of study of solutions of
non-linear differential oscillations, Inst. Mat. Akad. Nauk Ukrain. SSR,
Kiev, 1975,5-26 (in Russian).

111. - , 'Invariant toroidal manifolds of systems with slowly changing variables',
In: Problems of the asymptotic theory of non-linear oscillations, Naukova
Dumka, Kiev, 1977, 181-191 (in Russian).
112. - , 'Green's function for a linear extension of a dynamical system on a
torus, its uniqueness conditions and properties following from these conditions', Ukrain. Mat. Zit. 32 (1980),791-797.
113. - , 'Necessary conditions for the existence of invariant tori of linear extensions of dynamical systems on a torus', Differentsial'nye Uravneniya 16
(1980), 1427-1437.
114. - , 'Separatrix manifolds and decomposability of linear extensions of dynamical systems on a torus', Ukrain. Mat. Zh. 33 (1981), 31-38.
115. -
and A.V. Dvorak, 'Splitting of a dynamical system in a neighbourhood
of a conditionally stable invariant manifold', Ukra in. Mat. Zh. 29 (1977),

557-562.
116. -
and V.L. Kulik, 'On t.he problem of the existence of a Green's function
for the invariant torus problem', Ukrain. Mat. Zh. 27 (1975), 348-359.
117. -
and -, 'Exponential dichotomy of an invariant torus of dynamical
systems', Differentsial'nye Uravneniya 15 (1979),1434-1444.

118. -
and -, 'On decomposability of linearized systems of differential equa-
tions', Ukrain. Mat. Zit. 34 (1982), 587-597.

119. - and 1.0. Parasyuk, 'On Galerkin's procedure in the perturbation theory
of invariant tori', DoH Akad Nauk Ukrain. SSR, Ser. A, 1977, No.2, 112115.
120. -
and N.A. Perestyuk, 'On the existence of invariant toroidal sets of
BIBLIOGRAPHY
305
systems with instantaneous change', In: Integral manifold method in nonlinear differential equations, Inst. Mat. Akad. Nauk Ukrain. SSR, Kiev,
1973, 262-273.
121. -
and LV. Polesya, 'Birth of invariant manifolds in a neighbourhood of
a position of equilibrium', Differentsial'nye Uravneniya 11 (1975), 14091416.

122. -
and N.J. Ronto, Numerical-analytic methods of studying periodic solu-
tions, Vishcha Shkola, Kiev, 1976 (in Russian).
123. - and V.1. Tkachenko, 'Decomposability of linear extensions of dynamical

systems on a two- or three-dimensional torus', DoH. Akad Nauk Ukrain.
SSR, Ser. A, 1983, No.2, 22-25.
124. - , V.A. Mal'kov and SJ. Trofimchuk, 'Splitting of a linear extension of

a dynamical system on a torus', DoH. Akad Nauk Ukrain. SSR, Ser. A,
1982, No. 11, 19-22.
125. - , OJ. Martynyuk and N.A. Perestyuk, 'Existence of invariant tori of
systems of difference equations', Differentsial'nye Uravneniya 10 (1973),
1904-1910.
126. - , Yu.V. Teplinskii and N.S. Tsyganovskii, On invariant tori of countable
systems of differential equations, Preprint Inst. Mat. Akad. Nauk Ukrain.
SSR, Kiev, 1983.
127. S.L. Sobolev, Some applications of functional analysis in mathematical'
physics, Leningrad State Univ. Press, Leningrad, 1950 (in Russian).
128. - , Equations of mathematical physics, Nauka, Moscow, 1966 (in Russian).

129. S. Sternberg, Lectures on differential geometry, Prentice-Hall, Englewood
Cliffs, N.J., 1964.
130. V.1. Tkachenko, 'On block diagonalization of almost periodic systems',
DoH. Akad Nauk Ukrain. SSR, Ser. A, 1983, No.6, 18-20.
131. S.1. Trofimchuk, 'Necessary conditions for the existence of an invariant

manifold of a linear extension of a dynamical system on a compact manifold', Ukrain. Mat. Zh. 36 (1984),390-393.
132. D.U. Umbetzhanov, Almost multiperiodic solutions of partial differential
equations, Nauka, Alma-Ata, 1979 (in Russian).
133. A.N. Filatov, Asymptotic methods in the theory of differential and integrodifferential equations, FAN, Tashkent, 1974 (in Russian).
134. V.Kh. Kharasakhal, Almost periodic solutions of ordinary differential equa-
306
BIBLIOGRAPHY
tions, Nauka, Alma-Ata, 1970 (in Russian).
135. F. Hartman, Ordinary differential equations, Wiley, New York, 1964.

136. T. Hayasi, Forced oscillations in non-linear systems, Inost. Lit., Moscow,
1967 (in Russian).
137. - , Non-linear oscillations in physical systems, Mir, Moscow, 1968 (in Russian).
138. J .K. Hale, Oscillations in non-linear systems, Mir, Moscow, 1966 (in Russian).
139. A.Ya. Khinchin, Continuous fractions, Gostekhizdat, Moscow-Leningrad,
1949 (in Russian).
140. A.Ya. Khokhryakov, 'On the periodic boundary-value problem for the differential equation y(n) + !(t,y,y', ... ,y(n-l = 0, Differentsial'nye Uravneniya 3 (1967), 1643-1655.
141. L. Cesari, Asymptotic behaviour and stability problems in ordinary differential equations, Springer-Verlag, Berlin, 1963.
142. A. Besicovitch, Almost periodic functions, Cambridge Univ. Press, Cambridge, 1932.
143. H. Bohr, Collected mathematical works. II. Almost periodic functions,
Copenhagen, 1952.
144. L. Cesari, 'Functional analysis and periodic solutions of non-linear differential equations', In: Contributions to differential equations, New York,
1963, pp. 148-187.
145. S.P. Diliberto, 'An applicat.ion of periodic surfaces', In: Contributions to
the theory of non-linear oscillations, Princeton Univ. Press, Princeton,
N.J., 1956, pp. 257-261.
146. - , 'Perturbation theorems for periodic surfaces I, II, Rend. Circolo Mat.
Palermo 9 (1960),265-299; 10 (1961), 111-161.
147. - , 'Application of periodic surfaces to dynamical systems', In: Proc. Fifth

Internat. Conf. on Non-linear Oscillations Vol. 1: Analytical methods of
the theory of non-linear oscillations, Inst. Mat. Akad. Nauk Ukrain. SSR,
Kiev, 1970,257-264.
148. N. Fenichel, 'Persistence and smoothness of invariant manifolds for flows',
Indiana Univ. Math. J. 21 (1971), 193-226.
149. K.O. Friedrichs, 'Symmetric positive linear differential equations', Comm.

Pure Appl. Math. 11 (1958),333-418.
BIBLIOGRAPHY
307
150. - , 'Symmetric hyperbolic linear differential equations', Comm. Pure Appl.

Math. 7 (1954), 345-392.
151. J .K. Hale, 'Integral manirolds of perturbed differential systems', Ann. of

Math. 73 (1961), 496-531.
152. M.W. Hirsch, C.C. Pugh and M. Shub, Invariant manifolds, Lecture Notes
in Math. 583, Springer, Berlin, 1977.
153. J. Kyner, 'Invariant manifolds', Rend. Circolo Mat. Palermo 9 (1961), 98110.
154. I. Kupka, 'Stabilite des varictes invariantes d 'un champ de vecteurs pour
les petites perturbations', C.R. Acad. Sci., Paris, 258 (1964),4197-4200.
155. Lord Rayleigh, 'On maintained vibrations', Philosophical magazine 15
(1883), 229.
156. J. Moser, 'A new technique for the construction of solutions of non-linear
differential equations', Proc. Nat. Acad. Sci. 47 (1961), 1824-2831.
157. - , 'On invariant curves or area-preserving mappings of an annulus', Nachr.
Akad. Wiss. Gottingen, lIfath .. Phys. lla (1962), 1-20.
158. - , 'On invariant surfaces and almost periodic solutions for ordinary differential equations', Notices Amer. Math. Soc. 12 (1965), 124.
159. - , 'A rapidly convergent iteration method and non-linear differential equations', Ann. Scuola Norm Sup. Pisa 20:3 (1966), 499-535.
160. - , 'Convergent series expansions for quasi-periodic motions', Math. Ann.
169 (1967), 136-176.
161. J. Nash, 'The imbedding problem for Riemannian manifolds', Ann. of
Math. 63 (1956), 20-63.
162. L. Nirenberg, 'On elliptic partial differential equations', Ann. Scuola Norm.
Sup. Pisa 13:11 (1959), 115-162.
163. R.J. Sacker, ' A new approach to perturbation theory of invariant surfaces',
Comm. Pure Appl. Math. 18 (1965),717-732.
164. - , 'A perturbation theorem for invariant manifolds and Holder continuity',
J. Math. Mech 18 (1969),705-761.
165. -
and G.R. Sell, 'Existence of dichotomies and invariant splittings for
linear differential syst.ems', J. Differential Eqations 27 (1978),106-137.

166. - and - , 'A spectral theory for linear differential systems', J. Differential
Eqations 27 (1978), 320-358.
308
BIBLIOGRAPHY
167. G.R. Sell, 'The Floquet problem for almost periodic linear differential equations', Lecture Notes in Math. 415, Springer, Berlin, 1974,239-251.
168. F.R. Sell, 'Bifurcation of higher dimensional tori', Arch. Rational Mech.
Anal. 69 (1979), 199-230.
169. Y. Sibuja, 'Some global properties of matrices of functions of one variable',
Math. Ann. 161 (1965),67-77.
170. M. Vrabe, 'Existence theorems of quasi-periodic solutions to non-linear
differential systems', Funkcialaj. Ekvacioj. 15 (1972), 75-100.
171. - , 'On a modified Galerkin's procedure for non-linear quasi-periodic differential systems', In: Equations difTerentielles et fonctionelles non-lineaires, Paris, 1973, 223-258.
172. H. Weyl, '"Uber die Gleichvereilung von Zahlen Mod Eins', Math. Ann. 77
(1916),313-352.
Author Index
A.A. Andronov xii
J. Kyner xiii
A. Besicovich 13, 14
M. Lyapunov xi, 50, 55
Yu.N. Bibikov xiii
Yu.A. Mitropol'skii xii
G.D. Birkhoff 97
J. Moser xiii, 8,9,200,271
S. Bochner 20
J. Nash 268
N.N. Bogolyubov xi, xii
V.V. Nemytskii
P.G. Bol 19
N.I. Neimark
H. Bohr 20, 26
I.V. Bronshtein
Xli
Xlll
L. Nirenberg 8
Xlll
1.0. Parasyuk x
V.P. Burdijaev xiii
V.A. Pliss xiii, 58
S.P. Diliberto xiii
H. Poincare xi, 97
N. Fenichel xiii
C. Pugh xiii, 98
K.O. Friedrichs xiii, 200

L. Garding 41
R. Sacker xiii
T. Hale xiii
G. Sell xiv
M. Hirsch xiii
M. Simb xiii
N.M. Krylov xi, xii
Y. Sibuja 38, 41
V.L. Kulik x
S.L. Sobolev 3, 5
I. Kupka xiii
D.U. Umbetjanov
Y. Kurtsveil xiii
A.A. Witt xii
309
Xlll
Index of Notation
C"(Tm ) 1
H"(Tm
P(Tm) 2
H(Tm ) 3
L'.2(Tm ) 3
cr(w) 9
B 2 (w) 14
En ix
Tm ix
Ax. 47
Ox. 47
IT(Mo) 65
zm 80
O~()
O~()
107
113
CLp(Tm ) 117
!J{o 119
Jt 143
1.1+ 134
M- 134
310
Subject Index
adjoint
frame 169
formal 101
extendability of 173
matrix 101
frequency basis 10
system of equations 101, 102
maximal 76
real dimension of 71
Birkhoff theorem 97
function
block CS(Tm)-decomposable system
Besicovich 13
100, 158
Borel's theorem 106
Lyapunov 50
Brouwer's fixed point theorem 240
of constant sign 50
periodic 1
Cauchy problem 73, 110
quasi-periodic 1, 9-37
Cauchy's integral formula 39
sign-definite 50
compact set 47
spherical coordinates 25
connected set 47
total derivative of 51
contraction operator 263

critical value 36
Galerkin
decomposing transformation 93
approximations 195, 270-281
differential operator 41
iteration method 194, 270-281

linear modification 270
elliptic 41
Garding's inequality 41, 206, 209
eigenfunction 169
Green's function 113
eigennumber 169
coarse 214
eigenvalue 169
smoothness index 215

uniqueness 124
first integral 18
Floquet-Lyapunov theorem 292
Gronwall-Bellman inequality 261

311
SUBJECT INDEX
312
invariant set 46, 65

asymptotically stable 49, 57
uniformly 60
locally 56, 245
negatively 49
manifold 83
asymptotically stable 92, 220
invariance equations 83, 85
separatrix 134, 140, 244
straightened 173, 175, 176
one-sided 49
stable 92, 220
positively 49
topological 137
stable 49, 57
toroidal 83
unstable 49
invariant torus 88,92,99, 113, 117,
220
degenerate 106, 124
exponential attraction index 221
exponentially dichotomous 100,
142, 151-168, 241-258

exponentially stable 99, 18, 119,
121-123,200
non-degenerate 107, 124
smoothness of 258-270
system of variation equations of
93
linear system of equations 99
decomposition of 134
homogeneous 99
invariant torus of 99
non-homogeneous 99
phase space of 134
linearization process 211
Lipschitz constant 216
local convergence 20
loss of smoothness 25
Lyapunov
function 142
theorem 55
local coordinates 88
unstable 92, 220
matricent 107
matrix
block-diagonal form 157
characteristic polynomial of 40
exponential of 38
fundamental 107, 113, 260, 291
Gramm 85,90,175
Jordan form of
134-136, 143,
159, 160, 164, 254

logarithm of 38
quasi-diagonal form 136
spectrum of 40
minimal set 96
Moser's lemma 271,281-285
motion 46, 83
damping of 140
exponential attraction 226
flow 86
quasi-periodic 70, 83
recurrent 96
trajectory 40
Nash kernel 268
orthonormal basis 28
SUBJECT INDEX
oscillation
313
Sard's theorem 36
double-frequency 97
Schwartz orthogonalization process
multi-frequency 97
30
semi-trajectory 46
single-frequency 97
Ostrogradskii-Liouville theorem 103
a-limit point 47, 257

w-limit point 47, 257
periodic basis 28
Sibuja's theorem 38, 289, 293
perturbation 65, 211
slowly changing variables 285-295
Poincare-Denjoy theory 97
Sobolev
polynomial
quasi- trigonometric 11
trigonometric 1
compactness theorem 197

embedding theorem 3, 5, 25, 199
stationary point 126
Sylvester's criterion 124
rapidly changing variables 285-295

reduction principle 56, 57
triangulation 169
region of attraction 65
Van der Pol equation 291
Riccati equation 164
variational equation 216, 242

Elements of The Mathem

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Elements of The Mathem

Transféré par

Droits d'auteur :

Formats disponibles

Elements of the Mathematical Theory of Multi-Frequency Oscillations

Mathematics and Its Applications (Soviet Series)

SPRINGER SCIENCE+BUSINESS MEDIA, B.V.

Library ofCongress Cataloging-in-Publication Data

Translation of, ~lB.enty matematicheskol teorii mnogochastotnykh

Printed an acid-free paper

SERIES EDITOR'S PREFACE

"EI moi. '.'f si favait su comment en revcnir.

One service mathematics has rendered the

The series is divergent; therefore we mlY be

SERIES EDITOR'S PREFACE

N ever lend books, for no one ever returns

are books that other folk have lent me.

La physique ne nous donne pas seulemenl

The funaion of an expert is nol to be more

DOUS fait presscnlir 1a solution.

Amsterdam, August 1991

1.1. The function spaces Cr(Tm ) and Hr(Tm ) . . . . .

1.2. Structure of the spaces Hr(Tm ). Sobolev theorems.

1.6. First integral of a quasi-periodic function .

1.5. The spaces

and their structure

1.7. Spherical coordinates of a quasi-periodic vector function

1.8. The problem on a periodic basis in En . . . . . .

1.9. Logarithm of a matrix in C/(Tm ). Sibuja's theorem

CHAPTER 2. Invariant sets and their stability

2.1. Preliminary notions and results . . . . . .

2.2. One-sided invariant sets and their properties

2.3. Locally invariant sets. Reduction principle

2.4. Behaviour of an invariant set under small perturbations of the system 65

2.6. Invariance equations of a smooth manifold and the trajectory flow

2.7. Local coordinates in a neighbourhood of a toroidal manifold. Stability

2.8. Recurrent motions and multi-frequency oscillations

CHAPTER 3. Some problems of the linear theory

3.1. Introductory remarks and definitions

3.2. Adjoint system of equations. Necessary conditions for the existence

3.3. Necessary conditions for the existence of an invariant torus of a

3.4. The Green's function. Sufficient conditions for the existence of an

3.5. Conditions for the existence of an exponentially stable invariant

3.8. Sufficient conditions for exponential dichotomy of an invariant torus 142

3.10. Conditions for the C'(Tm)-block decomposability of an exponentially

3.12. On smoothness of an exponentially stable invariant torus

3.13. Smoothness properties of Green's functions, the invariant torus

3.14. Galerkin's method for the construction of an invariant torus . .

3.15. Proof of the main inequalities for the substantiation of Galerkin's

4. Perturbation theory of an invariant torus of a nonlinear system . . . . . . . . . .

4.1. Introductory remarks. The linearization process . . . . . . . ,

4.2. Main theorem

4.3. Exponential stability of an invariant torus and conditions for its

4.4. Theorem on exponential attraction of motions in a neighbourhood

servation under small perturbations of the system . . . . . . .

4.6. An estimate of the smallness of a perturbation and the maximal

4.7. Galerkin's method for the construction of an invariant torus of a

Proof of Moser's lemma

4.9. Invariant tori of systems of differential equations with rapidly and

Tm . Taking into account the fact that a quasi-periodic solution

of a dynamical system in En "sweeps out" an invariant torus of this sytem, the

the last two chapters, is devoted

Tm , perturbation theory of such manifolds for non-linear

procedure for finding toroidal manifolds.

The book is designed for specialists working in the area of multi-frequency

N.N. Bogolyubov [19], [21], [23], [55-57].

[97], [99], [136-138].