CBMS-NSF REGIONAL CONFERENCE SERIES

IN APPLIED MATHEMATICS
A series of lectures on topics of current research interest in applied mathematics under the direction of
the Conference Board of the Mathematical Sciences, supported by the National Science Foundation and
published by SIAM.

GARRETT BIRKHOFF, The Numerical Solution of Elliptic Equations

D. V. LINDLEY, Bayesian Statistics, A Review
R. S. VARGA, Functional Analysis and Approximation Theory in Numerical Analysis
R. R. BAHADUR, Some Limit Theorems in Statistics
PATRJCK BILLINGSLEY, Weak Convergence of Measures: Applications in Probability
J. L. LIONS, Some Aspects of the Optimal Control of Distributed Parameter Systems
ROGER PENROSE, Techniques of Differential Topology in Relativity
HERMAN CHERNOFF, Sequential Analysis and Optimal Design
J. DURBIN, Distribution Theory for Tests Based on the Sample Distribution Function
SOL I. RUBINOW, Mathematical Problems in the Biological Sciences
P. D. LAX, Hyperbolic Systems of Conservation Laws and the Mathematical Theory of Shock
Waves
I. J. SCHOEKBERG, Cardinal Spline Interpolation
IVAN SINGER, The Theory of Best Approximation and Functional Analysis
WERNER C. RHEWBOLDT, Methods of Solving Systems of Nonlinear Equations
HANS F. WEINBERGER, Variational Methods for Eigenvalue Approximation
R. TYRRELL ROCKAFELLAR, Conjugate Duality and Optimization
SIR JAMES LIGHTHILL, Mathematical Biofluiddynamics
GERARD SALTON, Theory of Indexing
CATHLEEN S. MORAWETZ, Notes on Time Decay and Scattering for Some Hyperbolic Problems
F. HOPPENSTEADT, Mathematical Theories of Populations: Demographics, Genetics and Epidemics
RICHARD ASKEY, Orthogonal Polynomials and Special Functions
L. E. PAYNE, Improperly Posed Problems in Partial Differential Equations
S. ROSEN, Lectures on the Measurement and Evaluation of the Performance of Computing Systems
HERBERT B. KELLER, Numerical Solution of Two Point Boundary Value Problems
J. P. LASALLE, The Stability of Dynamical Systems - Z. ARTSTEIN, Appendix A: Limiting Equations
and Stability of Nonautonomous Ordinary Differential Equations
D. GOTTLIEB AND S. A. ORSZAG, Numerical Analysis of Spectral Methods: Theory and Applications
PETER J. HUBER, Robust Statistical Procedures
HERBERT SOLOMON, Geometric Probability
FRED S. ROBERTS, Graph Theory and Its Applications to Problems of Society
JURIS HARTMANIS, Feasible Computations and Provable Complexity Properties
ZOHAR MANNA, Lectures on the Logic of Computer Programming
ELLIS L. JOHNSON, Integer Programming: Facets, Subadditivity, and Duality for Group and Semi-
Group Problems
SHMUEL WINOGRAD, Arithmetic Complexity of Computations
J. F. C. KINGMAN, Mathematics of Genetic Diversity
MORTON E. GURTIN, Topics in Finite Elasticity
THOMAS G. KURTZ, Approximation of Population Processes

(continued on inside back cover)

Navier-Stokes Equations
and Nonlinear Functional
Analysis
This page intentionally left blank
Roger Temam
Université de Paris-Sud, Orsay, France
Indiana University, Bloomington, Indiana

Navier-Stokes Equations
and Nonlinear Functional
Analysis
Second Edition

SOCIETY FOR INDUSTRIAL AND APPLIED MATHEMATICS

PHILADELPHIA
Copyright © 1983, 1995 by the Society for Industrial and Applied Mathematics.

1098765432

All rights reserved. Printed in the United States of America. No part of this book may
be reproduced, stored, or transmitted in any manner without the written permission of
the publisher. For information, write to the Society for Industrial and Applied
Mathematics, 3600 University City Science Center, Philadelphia, PA 19104-2688.

Library of Congress Cataloging-in-Publication Data

Temam, Roger.
Navier-stokes equations and nonlinear functional analysis / Roger
Temam. — 2nd ed.
p. cm. — (CBMS-NSF regional conference series in applied
mathematics ; 66)
Includes bibliographical references.
ISBN 0-89871-340-4
1. Fluid dynamics. 2. Navier-Stokes equations—Numerical
solutions. 3. Nonlinear functional analysis. I. Title.
II. Series
QA911.T39 1995
532'.051'01515353—dc20 94-43733

is a registered trademark.
 Contents
Preface to the Second Edition ix

Introduction xi

Part I. Questions Related to the Existence, Uniqueness and Regularity

of Solutions

Orientation 1

1. REPRESENTATION OF A FLOW. THE NAVIER-STOKES

EQUATIONS

2. FUNCTIONAL SETTING OF THE EQUATIONS

2.1. Function spaces 7
2.2. The Stokes problem and the operator A 9
2.3. Sobolev inequalities. The trilinear form b 11
2.4. Variational formulation of the equations 13
2.5. Flow in a bounded domain 15

3. EXISTENCE AND UNIQUENESS THEOREMS (MOSTLY

CLASSICAL RESULTS)
3.1. A priori estimates 17
3.2. Existence and uniqueness results 21
3.3. Outlines of the proofs 23
3.4. Generic solvability of the Navier-Stokes equations 25

4. NEW A PRIORI ESTIMATES AND APPLICATIONS

4.1. Energy inequalities and consequences 29
4.2. Structure of the singularity set of a weak solution 32
4.3. New a priori estimates 32

5. REGULARITY AND FRACTIONAL DIMENSION

5.1. Hausdorff measure. Time singularities 35
5.2. Space and time singularities 36

6. SUCCESSIVE REGULARITY AND COMPATIBILITY CON-

DITIONS AT t=0 (BOUNDED CASE)
6.1. Further properties of the operators A and B 43
6.2. Regularity results 44
6.3. Other results 47
V
Vi CONTENTS

7. ANALYTICITY IN TIME
7.1. The analyticity result 51
7.2. Remarks 56

8. LAGRANGIAN REPRESENTATION OF THE FLOW

8.1. The main result 57
8.2. Proof of Theorem 8.1 58
8.3. Appendix 60

Part II. Questions Related to Stationary Solutions and Functional

Invariant Sets (Attractors)

Orientation 61

9. THE COUETTE-TAYLOR EXPERIMENT

10. STATIONARY SOLUTIONS OF THE NAVIER-STOKES

EQUATIONS
10.1. Behavior for t—»oo. The trivial case 67
10.2. An abstract theorem on stationary solutions 71
10.3. Application to the Navier-Stokes equations 72
10.4. Counterexamples 76

11. THE SQUEEZING PROPERTY

11.1. An a priori estimate on strong solutions 79
11.2. The squeezing property 79

12. HAUSDORFF DIMENSION OF AN ATTRACTOR

12.1. Functional invariant sets and attractors 85
12.2. Hausdorflf dimension of functional invariant sets 86
12.3. Other properties of functional invariant sets 89

Part III. Questions Related to the Numerical Approximation

Orientation 91

13. FINITE TIME APPROXIMATION

13.1. An example of space-time discretization 93
13.2. The convergence theorem 95
 CONTENTS Vii

13.3. A compactness theorem 97

13.4. Proof of Theorem 13.1 (conclusion) 100

14. LONG TIME APPROXIMATION OF THE NAVIER-STOKES

EQUATIONS
14.1. Long time finite dimensional approximation 105
14.2. Galerkin approximation 109

APPENDIX: INERTIAL MANIFOLDS AND NAVIER-STOKES

EQUATIONS 113
A. 1. Inertial manifolds and inertial systems 113
A.2. Survey of the main results 114
A.3. Inertial system for the Navier-Stokes equations 118
A.4. Flow around a sphere 124

Comments and Bibliography 127

Comments and Bibliography: Update for the Second Edition .... 129

References 131
This page intentionally left blank
 Preface to the Second Edition

Since publication of the first edition of this book in 1983, a very active area in
the theory of Navier-Stokes equations has been the study of these equations as
a dynamical system in relation to the dynamical system approach to turbulence.
A large number of results have been derived concerning the long-time behavior
of the solutions, the attractors for the Navier-Stokes equations and their approxi-
mation, the problem of the existence of exact inertial manifolds and approximate
inertial manifolds, and new numerical algorithms stemming from dynamical sys-
tems theory, such as the nonlinear Galerkin method. Numerical simulations of
turbulence and other numerical methods based on different approaches have also
been studied intensively during this decade.
Most of the results presented in the first edition of this book are still relevant;
they are not altered here. Recent results on the numerical approximation of the
Navier-Stokes equation or the study of the dynamical system that they generate
are addressed thoroughly in more specialized publications.
In addition to some minor alterations, the second edition of Navier-Stokes Equa-
tions and Nonlinear Functional Analysis has been updated by the addition of
a new appendix devoted to inertial manifolds for Navier-Stokes equations. In
keeping with the spirit of these notes, which was to arrive as rapidly and as
simply as possible at some central problems in the Navier-Stokes equations, we
choose to add this section addressing one of the topics of extensive research in
recent years.
Although some related concepts and results had existed earlier, inertial mani-
folds were first introduced under this name in 1985 and systematically studied for
partial differential equations of the Navier-Stokes type since that date. At this
time the theory of inertial manifolds for Navier-Stokes equations is not complete,
but there is already available a set of results which deserves to be known, in the
hope that this will stimulate further research in this area.
Inertial manifolds are a global version of central manifolds. When they exist
they encompass the complete dynamics of a system, reducing the dynamics of an
infinite system to that of a smooth, finite-dimensional one called the inertial sys-
tem. In the Appendix we describe the concepts and recall the definitions and some
typical results; we show the existence of inertial manifolds for the Navier-Stokes
equations with an enhanced viscosity. We also describe a tentative route for prov-
ing the existence of inertial systems for the actual two-dimensional Navier-Stokes
equations and for the two-dimensional version of the Navier-Stokes equations
corresponding to the flow around a sphere (flow of a thin layer of fluid around a
sphere), a subject of obvious interest for geophysical flows and climate problems.
As indicated earlier, another aspect of inertial manifolds not presented here is
the use of approximation of inertial manifolds for the development of new multi-
level algorithms adapted to the resolution of the many scales present in turbulent
flows. These aspects are addressed (and will be addressed further) in publications
more numerically or computationally oriented; some bibliographic references are
given in the Appendix.
ix
This page intentionally left blank
 Introduction

The Navier-Stokes equations are the equations governing the motion of

usual fluids like water, air, o i l , . . . , under quite general conditions, and they
appear in the study of many important phenomena, either alone or coupled
with other equations. For instance, they are used in theoretical studies in
aeronautical sciences, in meteorology, in thermo-hydraulics, in the petroleum
industry, in plasma physics, etc. From the point of view of continuum
mechanics the Navier-Stokes equations (N.S.E.) are essentially the simplest
equations describing the motion of a fluid, and they are derived under a quite
simple physical assumption, namely, the existence of a linear local relation
between stresses and strain rates. These equations, which are recalled in § 1,
are nonlinear. The nonlinear term (u • V)u contained in the equations comes
from kinematical considerations (i.e., it is the result of an elementary
mathematical operation) and does not result from assumptions about the
nature of the physical model; consequently this term cannot be avoided by
changing the physical model.
While the physical model leading to the Navier-Stokes equations is simple,
the situation is quite different from the mathematical point of view. In particular,
because of their nonlinearity, the mathematical study of these equations is
difficult and requires the full power of modern functional analysis. Even now,
despite all the important work done on these equations, our understanding of
them remains fundamentally incomplete.
Three types of problems appear in the mathematical treatment of these
equations. Although they are well known, we recall them briefly for the
nonspecialist.
1) Existence, uniqueness and regularity. It has been known since the work of
J. Leray [1] that, provided the data are sufficiently smooth (see § 3), the initial
value problem for the time-dependent Navier-Stokes equations possesses a
unique smooth solution on some interval of time (0, T^); according to J. Leray
[3] and E. Hopf [1], this solution can be extended for subsequent time as a
(possibly) less regular function. A major question as yet unanswered is whether
the solution remains smooth all the time. In the case of an affirmative answer
the question of existence and uniqueness would be considerably clarified. In
the case of a negative answer, then it would be important to have information
on the nature of the singularities, and to know whether the weak solutions are
unique and, should they be not unique, how to characterize the "physical"
ones.
All these and other related questions are interesting not only for mathemati-
cal understanding of the equations but also for understanding the phenomenon
of turbulence. Recall here that J. Leray's conjecture [l]-[3] on turbulence, and
his motivation for the introduction of the concept of weak solutions, was that
the solutions in three space dimensions are not smooth, the velocity or the
Xi
Xii INTRODUCTION

vorticity (the curl of the velocity) becoming infinite at some points or on some
"small" sets where the turbulence would be located.
All the mathematical problems that we have mentioned are as yet open. Let
us mention, however, some recent studies on the Hausdorff dimension of the
set of singularities of solutions (the set where the velocity is infinite), studies
initiated by B. Mandelbrot [1] and V. Scheffer [l]-[4] and developed by L.
Caffarelli-R. Kohn-L. Nirenberg [1] (see also C. Foias-R. Temam [5]). The
studies are meant to be some hopeful steps towards the proof of regularity if
the solutions are smooth, or else some steps in the study of the singular set if
singularities do develop spontaneously.
2) Long time behavior. If the volume forces and the given boundary values of
the velocity are independent of time, then time does not appear explicitly in
the Navier-Stokes equations and the equations become an autonomous infinite
dimensional dynamical system. A question of interest is then the behavior for
t —> oo of the solution of the time-dependent N.S.E. A more detailed description
of this problem is contained in § 9, but, essentially, the situation is as follows. If
the given forces and boundary values of the velocity are small then there exists
a unique stable stationary solution and the time-dependent solution converges
to it as t —» oo. On the other hand, if the forces are large, then it is very likely
from physical evidence and from our present understanding of bifurcation
phenomena that, as £—»°o, the solution tends to a time periodic one or to a
more complicated attracting set. In the latter case, the long time behavior of
the solution representing the "permanent" regime, could well appear chaotic.
This is known to happen even for very simple dynamical systems in finite
dimensional spaces, such as the Lorenz model (cf. also O. Lanford [1]) or the
examples of mappings of the unit interval of R into itself, discussed by M.
Feigenbaum [1] and O. Lanford [2]. Such chaotic behavior is another way to
explain turbulence; it is based on the ideas of dynamical systems and strange
attractors, following D. Ruelle [1], [2], D. Ruelle-F. Takens [1] and S. Smale
[2]. Actually, for the moment, these two ways for the description of turbulence
are not mutually exclusive, as singularities and long time chaotic behavior could
perhaps be both present in the Navier-Stokes equations. Let us mention also
that, as observed by D. Ruelle [3], the strange attractor point of view is not
sufficient to explain the chaotic structure in space of the physical flow1.
A great number of mathematical problems relating to the behavior for t —» o
of the solutions of the N.S.E. are open. They include convergence to a stable
stationary or time periodic solution in connection with bifurcation theory2 or
convergence to a more complicated attractor (cf. C. Foias-R. Temam [5], [8],
[9] and also J. Mallet-Paret [1]). Of course in three space dimensions the

1
We specify "physical flow" since the word "flow" is also used in the dynamical system context
to describe the trajectory of u(t) in the function space.
2
From the abundant literature we refer, for instance, to C. Bardos-C. Bessis [1], P. Rabinowitz
[1] and the references therein.
 INTRODUCTION xiii

difficulties are considerable, and the problem is perhaps out of reach for the
moment since we do not even know whether the initial value problem for the
N.S.E. is well posed.
3) Numerical solution. We mentioned at the beginning that the N.S.E. play
an important role in several scientific and engineering fields. The needs there
are usually not for a qualitative description of the solution but rather for a
quantitative one, i.e., for the values of some quantities related to the solution.
Since the exact resolution of the N.S.E. is totally out of reach (we actually
know only a very small number of exact solutions of these equations), the data
necessary for engineers can be provided only through numerical computations.
Also, for practical reasons there is often a need for accurate solutions of the
N.S.E. and reliance on simplified models is inadequate.
Here again the problem is difficult and the numerical resolution of the N.S.E.
will require (as in the past) the simultaneous efforts of mathematicians,
numerical analysts and specialists in computer science. Several significant
problems can already be solved numerically, but much time and effort will be
necessary until we master the numerical solution of these equations for realistic
values of the physical parameters. Besides the need for the development of
appropriate algorithms and codes and the improvement of computers in
memory size and computation speed, there is another difficulty of a more
mathematical (as well as practical) nature. The solutions of the N.S.E. under
realistic conditions are so highly oscillatory (chaotic behavior) that even if we
were able to solve them with a great accuracy we would be faced with too
much useless information. One has to find a way, with some kind of averaging,
to compute mean values of the solutions and the corresponding desired
parameters.
As mentioned before, some analytical work is necessary for the numerical
resolution of the N.S.E. But, conversely, one may hope that if improved
numerical methods are available, they may help the mathematician in the
formulation of realistic conjectures about the N.S.E., as happened for the
Korteweg-de Vries equations.
After these general remarks on fluid dynamics and the Navier-Stokes
equations, we describe the content of this monograph, which constitutes a very
modest step towards the understanding of the outstanding problems mentioned
above. The monograph contains 14 sections grouped into three parts (§§ 1-8,
9-12, 13-14), corresponding respectively to the three types of problems which
we have just presented.
Part I (§§ 1-8) contains a set of results related to the existence, uniqueness
and regularity of the weak and strong solutions. The material in Part I may
appear technical, but most of it is essential for a proper understanding of more
qualitative or more concrete questions. In § 1 the N.S.E. are recalled and we
give a brief description of the boundary value problems usually associated with
them. Sections 2 and 3 contain a description of the classical existence and
uniqueness results for weak and strong solutions. We have tried to simplify the
Xiv INTRODUCTION

presentation of these technical results. In particular, we have chosen to em-

phasize the case of the flow in a cube in (Rn, n = 2 or 3, with space periodic
boundary conditions. This case, which is not treated in the available books or
survey articles, leads to many technical simplifications while retaining the main
difficulties of the problem (except for the boundary layer question, which is not
considered here). Of course the necessary modifications for the case of the flow
in a bounded region of Rn are given. The results in §§ 2 and 3 are essentially
self-contained, except for some very technical points that can be found
elsewhere in the literature, (cf. in particular O. A. Ladyzhenskaya [1], J. L.
Lions [1] and R. Temam [6] to which we will refer more briefly as [RT]).
New (or recent) developments, related either to weak solutions or to strong
solutions, are then presented in §§4 to 8. New a priori estimates for weak
solutions are proved in § 4 following C. Foias-C. Guillopé-R. Temam [1].
They imply in particular, as noticed by L. Tartar, that the L°°-norm of a weak
solution is L1 in time, and this allows us in §8 to define the Lagrangian
representation of the flow associated with a weak solution. In § 5 we present the
result established in C. Foias-R. Temam [5] concerning the fractional dimen-
sion of the singular set of a weak solution.
In § 6 we derive, after R. Temam [8], the necessary and sufficient conditions
for regularity, at time t = 0 of the (strong) solutions to the N.S.E. The result
relates to the question of compatibility conditions between the given initial and
boundary values of the problem. This question, which is a well-known one for
other initial and boundary value problems including linear ones, apparently has
not been solved for the N.S.E.; of course the question of regularity at t = 0 has
nothing to do with the singularities which may develop at positive time.
Finally in § 7 we prove a result of analyticity in time of the solutions,
following with several simplifications C. Foias-R. Temam [5]. The proof, which
could be used for other nonlinear evolution problems, is simple and is closely
related to the methods used elsewhere in these notes.
Part II (§§ 9-12) deals with questions related to the behavior for t —> o° of the
solutions of the N.S.E. Section 9 explains the physical meaning of these
problems through the example of the Couette-Taylor experiment. Section 10 is
devoted to stationary solutions of the N.S.E.; contains a brief proof of
existence and uniqueness (for small Reynolds number), and the fact that the
solution to the N.S.E. tends, as t—*°°, to the unique stationary solution when
the Reynolds number is small. This section contains also a proof of a finiteness
property of the set of stationary solutions based on topological methods.
The results in §§ 11 and 12 (following C. Foias-R. Temam [5]) are related to
the behavior for r—>oo of the solutions to the N.S.E. at arbitrary (or large)
Reynolds numbers. They indicate that a turbulent flow is somehow structured and
depends (for the cases considered, see below), on a finite number of parame-
ters. These properties include a squeezing property of the trajectories in the
function space (§ 11) and the seemingly important fact that, as t —> <», a solution
to the Navier-Stokes equations converges to a functional invariant set (an
w-limit set, or an attractor), which has a finite Hausdorff dimension (§ 12).
 INTRODUCTION XV

Intuitively, this means that under these circumstances all but a finite set of
modes of the flow are damped.
Both results are proved for a space dimension n = 2 or 3, without any
restriction for n = 2, but for n = 3 with the restriction that the solution
considered has a bounded H^-norm for all time. It is shown (cf. § 12.3 and C.
Foias-R. Temam [13]) that if this boundedness assumption is not satisfied then
there exists a weak solution to the N.S.E. which displays singularities. Alterna-
tively, this means that the relevant results in §§11 and 12 (and some of the
results in Part III) fail to be true in dimension 3 only if Leray's conjecture on
turbulence is verified (existence of singularities).
Although the problems studied in Part II are totally different from those
studied in Part I, the techniques used are essentially the same as in the first
sections, and particularly in §§ 2 and 3.
Part III (§§ 13 and 14) presents some results related to the numerical
approximation of the Navier-Stokes equations. At moderate Reynolds num-
bers, the major difficulties for the numerical solution of the equations are the
nonlinearity and the constraint div u = 0. In §13 we present one of the
algorithms which have been derived in the past to overcome these difficulties,
and which has been recently applied to large scale engineering computations.
Section 14 contains some remarks related to the solution of the N.S.E. for
large time: it is shown (and this is not totally independent of the result in Part
II) that the behavior of the solution for large time depends on a finite number
of parameters and an estimate of the number of parameters is given. A more
precise estimate of the number of parameters in terms of the Reynolds number
and further developments will appear elsewhere (C. Foias-R. Temam [9], C.
Foias-O. Manley-R. Temam-Y. Trève [1], C. Foias-R. Temam [11], [12]).
At the level of methods and results, there is again a close relation between
Part III and Part I (§§ 3 and 13 in particular), and there is a connection as
already mentioned with Part II (§§ 12 and 14).
It is not the purpose of these notes to make an exhaustive presentation of
recent results on the Navier-Stokes equations. We have only tried to present
some typical results, and the reader is referred to the bibliographical comments
in the text and at the end for further developments. In particular we have
refrained from developing the stochastic aspect of the Navier-Stokes equa-
tions, which would have necessitated the introduction of too many different
tools. The interested reader can consult A. Bensoussan-R. Temam [1], C.
Foias [1], C. Foias-R. Temam [6] [7], M. Viot [1], M. I. Vishik [1], M. I.
Vishik-A. V. Fursikov [1], [2], [3].
Our aim while writing these notes was to try to arrive as rapidly and as
simply as possible at some central problems in the Navier-Stokes equations.
We hope that they can stimulate some interest in these equations. One can
hope also that the demand of new technologies and the accelerated improve-
ment of the opportunities offered by new (existing or projected) computers will
help stimulate further interest in these problems in the future.
In conclusion I would like to thank all those who helped in the preparation
Xvi INTRODUCTION

of these notes: C. Foias for his collaboration which led to several articles on
which these notes are partly based, C. Guillopé, Ch. Gupta, J. C. Saut, D.
Serre and the referee for reading the manuscript and for the comments they
made. Finally I would like to thank the mathematical secretary at Dekalb
University and Mrs. Le Meur at Orsay for kindly typing the manuscript.
PART I

Questions Related to the Existence,

Uniqueness and Regularity of Solutions
Orientation. In this first Part, which contains §§ 1 to 8, we present the
Navier-Stokes equations of viscous incompressible fluids, and the main bound-
ary value problems which are usually associated with these equations. Then we
study the case of the flow in a bounded domain with periodic or zero boundary
conditions, and we give in this case the functional setting of the equation, and
various results on existence, uniqueness and regularity of time-dependent
solutions. We emphasize the case of the flow with space periodic boundary
conditions, treating more briefly the zero boundary conditions which are much
more often considered in the literature; see for instance O. A. Ladyzhenskaya
[1], J. L. Lions [1], and R. Temam [6] to which we will refer more briefly as
[RT].
In § 1 we recall the Navier-Stokes equations and the corresponding bound-
ary value problems. In § 2 we present the appropriate functional setting. In
§ 3 we recall the main existence and uniqueness results (which are essentially
classical), with the details of various a priori estimates used frequently in the
sequel and we sketch the proof of existence and uniqueness. Section 4 contains
some new a priori estimates, used in particular in § 8. Section 5 includes some
results on the Hausdorff dimension of the singular set of a weak solution.
Section 6 presents the necessary and sufficient conditions of regularity of the
solution at time f = 0 (the compatibility conditions). Section 7 shows under
appropriate assumptions that the solution is analytic in time with values in
D(A). Finally, in § 8 we exhibit the Lagrangian representation of the flow
associated with a weak solution of the Navier-Stokes equations.
This page intentionally left blank
1 Representation of a Flow.
The Navier-Stokes Equations

Let us assume that a fluid fills a region fl of space. For the Eulerian
representation of the flow of this fluid, we consider three functions p = p(x, t),
p = p(x, t), u = u(x, t), x = (x1; x2, x 3 )eft, t e R , where p(x, t) (or p(x, 0) is the
density (or the pressure) of the fluid at point x at time t and u(x, t) = (uj(x, f)»
u2(x, r), u3(x, t)) is the velocity of the particle of fluid which is at point x at time
t. One may also consider Lagrangian representation of the flow, in which case
we introduce functions p = p(a, r), p = p(a, t}, u = u(a,t); here u(a, t) is the
velocity of the particle of fluid which was at point a e H at some reference time
f 0 , and the meanings of p(a, t}, p(a, t) are similar. The Lagrangian representa-
tion of a flow is less often used, but we will make some comments on it in § 8.
If the fluid is Newtonian, then the functions p, p, u are governed by the
momentum conservation equation (1.1) (Navier-Stokes equation), by the con-
tinuity equation (1.2) (mass conservation equation) and by some constitutive
law connecting p and p:

where JLL > 0 is the kinematic viscosity, A. another physical parameter and
f = /(x, r) represents a density of force per unit volume. If the fluid is
homogeneous and incompressible, then p is a constant independent of x and t,
and the equations reduce to

Usually we take p = 1, set v — /n and, using the differential operator V = (d/dXj,

d/3x2, 6/3x3) arrive at

We can also consider (1.5) as the nondimensional form of the Navier-Stokes

equation (1.3), which is obtained as follows. We set p = p*p', p = p#p', u -
Wstcii', x = L^x', t = T*t', / = (p4cl/*/T*)f, where p#, L*, T* are respectively a
reference density, a reference length and a reference time for the flow, and
t/* = L-jc/T^, p* = L^p*. By substitution into (1.3) we get (1.5) for the reduced
3
4 PART I. QUESTIONS RELATED TO SOLUTIONS

quantities u'(x', f'), P'(*', 0, /'(*', 0, but in this case the inverse of v
represents the Reynolds number of the flow:

The equations (1.4), (1.5) are our basic equations. We note that they make
sense mathematically (and in some way physically) if (1 is an open set in IR2,
u = (u1, u2), /—(fi,/2)- Since it is useful to consider this situation too, and in
order to cover both cases simultaneouslv, we assume from now on that
(1 is an open set of IR", n = 2 or 3, with boundary P.
(1.7) We assume furthermore that H is located locally on one side of F and that
F is Lipschitzian or sometimes of class C r for a specified r.
One of the first mathematical questions concerning the equations (1.4), (1.5)
is the determination of a well-posed boundary value problem associated with
these equations. This is still an open problem, but it is believed (and has been
proved for n = 2 ) that (1.4), (1.5) must be completed by the following initial
and boundary conditions (for flow for f > 0 , xeO).
Initial condition:

Boundary condition:

If H is unbounded (and in particular for fl = IR"), we add to (1.9) a condition

at infinity:

ty given1.
Instead of (1.9) (and (1.10)) it is interesting to consider another boundary
condition which has no physical meaning:

where e^,... ,en is the canonical basis of IR", and L is the period in the ith
direction; Q = (]0, L[)n is the cube of the period2. The advantage of the
boundary condition (1.11) is that it leads to a simpler functional setting, while
many of the mathematical difficulties remain unchanged (except of course those
related to the boundary layer difficulty, which vanish). In fact, in § 2 we will
describe in detail the corresponding functional setting of the problem (i.e. for
(1.4), (1.5), (1.8), (1.11)), and we will mention only briefly the case with a

1
For special unbounded sets H further conditions must be added to (1.9), (1.10); cf. J. G.
Heywood [1], O. A. Ladyzhenskaya-V. A. Solonnikov [1], [2].
2
Of course one may consider different periods L(,..., Ln in the different directions, and in this
case Q=W=iQO,Li[).
 REPRESENTATION OF A FLOW. THE NAVIER-STOKES EQUATIONS 5

boundary, which is presented in detail in the references already quoted

(Ladyzhenskaya [1], Lions [1], [RT]).
Remark 1.1. In the periodic case (i.e. (1.11)), it is convenient to introduce
the average of u on the cube of the period,

and to set

The average m is explicitly determined in terms of the data. By integration of

(1.5) on Q, using (1.4), the Stokes formula and the fact that the integrals on the
boundary 3Q of Q vanish because of (1.11), we get

so that

By substitution of (1.13) into (1.5) we find

The quantity mu being known, the study of (1.16) is very similar to that of
(1.5). Therefore in the periodic case we will assume for simplicity that the
average flow vanishes, mu = 0.
This page intentionally left blank
^7 Functional Setting of the Equations
In this section we describe the functional setting of the equations, insisting
more particularly on the space periodic case (boundary condition (1.11)).

2.1. Function spaces. We denote by L2(ft) the space of real valued func-
tions on fl which are L2 for the Lebesgue measure dx = dxv • • • dxn; this space
is endowed with the usual scalar product and norm

We denote by Hm(fl) the Sobolev space of functions which are in L2(ft),

together with all their derivatives of order ^m. This is as usual a Hilbert space
for the scalar product and the norm

Ho(H) is the Hilbert subspace of H^O), made of functions vanishing on F. The

reader is referred to R. Adams [1] and J. L. Lions-E. Magenes [1] for the
theory of Sobolev spaces.
We denote by H^(Q), m e N , the space of functions which are in H^OR")
(i.e., u \G<=Hm(G) for every open bounded set 0} and which are periodic with
period Q:

For m = 0, H°(Q) coincides simply with L2(Q) (the restrictions of the functions
in H°(Q) to O are the whole space L2(Q)). For an arbitrary m eN, H^(Q) is a
Hilbert space for the scalar product and the norm

The functions in H^(O) are easily characterized by their Fourier series

expansion

and the norm |u|m is equivalent to the norm &kez" (l + |k| 2m ) kk|2}1/2. We also
set

7
8 PART I. QUESTIONS RELATED TO SOLUTIONS

We observe that the right-hand side of (2.2) makes sense more generally for
m e R , and we actually define H™(Q), m e I R , m ^ O , by (2.2); it is a Hilbert
space for the norm indicated above. For m e I R , we define H™(Q) with (2.3);
this is a Hilbert space for the norm {ZfceZ-> |fc|2m |ck|2}1/2, and H™(Q) and
H~ m (Q) are in duality for all m e I R .
Two spaces frequently used in the theory of Navier-Stokes equations are

where H™(Q) = {H™(Q)}n; more generally, for any function space X we denote
by X the space Xn endowed with the product structure. We equip V with the
scalar product and the Hilbert norm

This norm is equivalent to that induced by (Hp(Q)) n , and V is a Hilbert space

for this norm. It is easy to see that the dual V of V is

|| • ||v will denote the dual norm of || • || on V. We have

where the injections are continuous and each space is dense in the following
one. One can also show by mollification that the space of smooth functions,

is dense in V, H and V.
Remark 2.1. i) Using the trace theorem, one can show that u e V if and
only if its restriction «|Q to Q belongs to

where we have numbered the faces F l 5 . . . , F2n of Q as follows:

and u|r. is an improper notation for the trace of v on Fy. The characterization
of u|Q for u e H is more delicate and relies on a trace theorem given in [RT,
Chap. I., Thm. 1.2]1: ueH if and only if u belongs to

1
Let 0 be an open bounded set of Rn of class (€2. Then if v eL2(C) and div v e L2(6], we can
define yvv e H~l/2(d6), which coincides with v • v M if v is smooth (v denotes the unit outward
normal on d6). Furthermore, for every u in H l(G), we have the generalized Stokes formula
(v, Vu) + (div v, u) = (yvv, yuu), where •y()M = M lac denotes the trace of u on d€, and (•, •) is the
pairing between H'/2(dG) = y0H'(0) and its dual H~ 1/2 (d0).
 FUNCTIONAL SETTING OF THE EQUATIONS 9

ii) Let G be the orthogonal complement of H in H(p = (Lj(Q)/R)n. We have

2.2. The Stokes problem and the operator A. The Stokes problem as-
sociated with the space periodicity condition (1.11) is the following one.

It is easy to solve this problem explicitly using Fourier series. Let us introduce
the Fourier expansions of u, p and /;

Equations (2.9) reduce for every k ^ 0 to

Taking the scalar product of (2.10) with k and using (2.11) we find the p k 's:

then (2.10) provides the w k ? s:

By the definition (2.2) of H^(O), if /eHj(Q) then u EHj(Q) and p e H P (Q);

if / e H p ' C Q ) then uelHlJ(Q) and peHj(Q). Now if / belongs to H, then
k • fk = 0 for every k so that p = 0 and

We define in this way a one-to-one mapping / —> u from H onto

Its inverse from D(A) onto H is denoted by A, and in fact

If D(A) is endowed with the norm induced by Hp(Q), then A becomes an

isomorphism from D(A) onto H. If follows that the norm |Au on D(A) is
equivalent to the norm induced by Hp(Q).
The operator A can be seen as an unbounded positive linear selfadjoint
operator on H, and we can define the powers A", a elR, with domain D(A") in
10 PART I. QUESTIONS RELATED TO SOLUTIONS

H. We set

Va is a closed subspace of H£(Q), and in fact

In particular, V2 = D(A), Vl=V, V0 = H, V_! = V; A is an isomorphism from

Va+2 onto Va, D(A) onto H, V onto V, and so forth. The norm |Aa/2u| on Va
is equivalent to the norm induced by Hp(Q),

with2 c, c' depending on L and a.

The operator A is an isomorphism from Va+2 onto Va for all a e R . We
recall also that the injection of Va into V a _ e is compact for every a e!R, e >0.
Indeed if u^ is a sequence converging weakly to 0 in Va, then

a constant independent of m,

For every KelR, K>0,

and since this limsup is arbitrarily small, wm —»0 in V a _ e .

Eigenfunctions of A. The operator A"1 is linear continuous from H into
D(A), and since the injection of D(A) in H is compact, A"1 can be considered
as a compact operator in H. As an operator in H it is also selfadjoint. Hence it
possesses a sequence of eigenfunctions w,, j e N , which form an orthonormal
basis of H,

In fact the sequence of w,'s and A/'s is the sequence of functions w k a and
numbers A ka ,

2
The letters c, c', cf, c\ indicate various positive constants. The letters cf represent well-defined
constants while the constants represented by the letters c, c', c[ may be different in different places
in the text.
 FUNCTIONAL SETTING OF THE EQUATIONS 11

where k = (kl,. . ., k n )eZ", k ^ O , a = 1,. . . , n, and e\,...,en represents the

canonical basis of [R".

2.3. Sobolev inequalities. The trilinear fonn 6. We recall some Sobolev

inequalities and some basic properties of Sobolev spaces.
If G is an open set of R" and its boundary d€ is sufficiently regular (say
Lipschitzian), and if ^ — m / n > 0 , then

the injection being continuous. In particular, there exists a positive constant c

depending on m, n and L such that

For m>n/2, H™(Q)c: ^ p (Q) (the space of real continuous functions with
period Q) with a continuous injection.
If m l 5 m 2 e!R, ml^m2 and 0e]0,1[, the discrete Holder inequality 3 gives

If (1 — 0 ) m 1 + m 2 >n/2, the continuous imbedding of H^(O) into C€P(Q) shows

that there exists a constant c depending only on 6, m l 5 m 2 , n, L such that

Actually (cf. S. Agmon [1]), the inequality (2.21) is also valid if

In particular, if n = 2,

and because of (2.16)

12 PART I. QUESTIONS RELATED TO SOLUTIONS

if n = 3

The form b. We now show how to apply these properties of Sobolev spaces
to the study of the form b.
Let O be an open bounded set of R n which will be either H or Q. For
u, v, w ELl(®\ we set

whenever the integrals in (2.22) make sense. In particular, we have:

LEMMA 2.1. Let 0 = fl or Q. The form b is defined and is trilinear continuous
on H m '(0)xH m 2 +1 (<9)xH m3 (<!?) where m^O, and

Proof. If m, < n/2 for i = 1, 2, 3, then by (2.18) Hm-(C] c Lq>(0) where 1/q, =
^-mjn. Due to (2.28), (!Ah + l/q 2 + 1/^3)= 1, the product u^DjU^Wj isintegra-
ble and 6(u, u, w) makes sense. By application of Holder's inequality we get

If one or more of the m( 's is larger than n/2 we proceed as before, with the
corresponding q; replaced by +00 and the other q,'s equal to 2. If some of the
ra^'s are equal to n/2, we replace them by m|<mj, m; — m! sufficiently small so
that the corresponding inequality (2.29) still holds.
Remark 2.2. i) As a particular case of Lemma 2.1 and (2.29), ft is a tnlmear
continuous form on V m i xV m 2 + 1 x Vmi, mt as in (2.28) and

In particular, b is a trilinear continuous form on V x V x V and even on

Vx Vx V1/2.
ii) We can supplement (2.29)-(2.30) by other inequalities which follow from
(2.29)-(2.30) and (2.20)-(2.26). For instance the following inequalities combin-
ing (2.30) and (2.20) will be useful.
 FUNCTIONAL SETTING OF THE EQUATIONS 13

iii) Less frequently, we will use the following inequalities. We observe that
UjlDjUjOw, is summable if (for instance) 14 e L00^), D^,, w, e L2(€], and

In this manner, and using also (2.24), we get in the case n = 2

and in the case n = 3, using (2.26)

Finally we recall a fundamental property of the form b:

This property is easily established for w, v, w e V (cf. [RT, p. 163]) and follows
by continuity for u, u, w e V. With u = w, (2.33) implies

The operator B. For w, D, w e V we define B(u, v} e V and Bu e V by setting

Since b is a trilinear continuous form on V, B is a bilinear continuous operator

from V x V into V. More generally, by application of Lemma 2.1 we see that
B is a bilinear continuous operator from V mi xV m2+1 (or from
(2.36) H m '(C')xH m2+1 (<5 > )) into V^, where m 1? m2, m 3 satisfy the as-
sumptions in Lemma 2.1.
It is clear that various estimates for the norm of the bilinear operator B(% •)
can be derived from the above estimates for b.

2.4. Variational formulation of the equations. As indicated before, we are

interested in the boundary value problem (1.4), (1.5), (1.8), (1.11), when the
average of u on Q is 0; u0 and / are given, and we are looking for u and p.
Let T>0 be given, and let us assume that u and p are sufficiently smooth,
say we^ 2 (!R"x[0, T]), p e ^'(GT x[0, T]), and are classical solutions of this
problem. Let u(t) and p(t) be respectively the functions {x elR" •->• w(x, f)},
{x e(R n f-> p(x, t}}. Obviously u e L2(0, T; V), and if v is an element of V then
by multiplying (1.5) by u, integrating over Q and using (1.4), (1.11) and the
14 PART I. QUESTIONS RELATED TO SOLUTIONS

Stokes formula we find (cf. [RT, Chap. Ill] for the details) that

By continuity, (2.37) holds also for each v e V.

This suggests the following weak formulation of the problem, due to J. Leray
[1], [2], [3].
Problem 2.1 (weak solutions). For f and u0 given,

find u satisfying

and

If u merely belongs to L2(0, T; V), the condition (2.42) need not make
sense. But if u belongs to L2(0, T; V) and satisfies (2.41), then (cf. below) u is
almost everywhere on [0, T] equal to a continuous function, so that (2.42) is
meaningful.
We can write (2.41) as a differential equation in V by using the operators A
and B. We recall that A is an isomorphism from V onto V and B is a bilinear
continuous operator from V x V into V. If weL 2 (0, T; V), the function
Bu: {t^Bu(t)} belongs (at least) to L\Q, T; V). Consequently (2.41) is equi-
valent to

and since f-vAu-Bu£Ll(Q, T; V), u' = du/dt belongs to La(0, T; V), and

Furthermore, (cf. [RT, Chap. Ill, § 1]) u is almost everywhere equal to a

continuous function from [0, T] into V, and (2.42) makes sense.
We refer to [RT] (and the sequel) for more details and in particular for th
relation of Problem 2.1 to the initial problem (1.4), (1.5), (1.8), (1.11). One can
show that if u is a solution of Problem 2.1, then there exists p such that (1.4)
(1.5), (1.8), (1.11) are satisfied in a weak sense (cf. [RT, p. 307]).
Of course a weak solution of the Navier-Stokes equations (Problem 2.1) may
or may not possess further regularity properties. For convenience we will
introduce a class of more regular solutions which we call strong solutions.
 FUNCTIONAL SETTING OF THE EQUATIONS 15

Problem 2.2 (strong solutions). For f and u0 given,

find u satisfying

and (2.41)-(2.43).
Further regularity properties of strong solutions are investigated in § 4 (for
the space periodic case) and § 6 (for the bounded case). Let us observe here
that, by application of (2.32) and (2.41)-(2.42),

Hence if u is a strong solution, then for almost every t,

and the function Bu :{t >->Bu(0} belongs to L4(0, T; H). Since / and Au
belong to L2(0, T; H), u' = f- vAu~Bu belongs to this space too, and

The two conditions ueL 2 (0, T; D(A)), u'eL 2 (0, T; H) imply by interpolation
(cf. J. L. Lions-E. Magenes [1] or [RT, Chap. Ill § 1.4), that u is almost
everywhere equal to a continuous function from [0, T] into V:

2.5. Flow in a bounded domain. Although up to now we have concentrated

on the space periodic case, all of what follows applies as well to the case of flow
in a bounded domain with a fixed boundary (1.4), (1.5), (1.8), (1.9) with $ = 0,
ft bounded, provided we properly define the different spaces and operators.
In this case (cf. [RT],

where P is the orthogonal projector in L 2 (ft) onto H.

Most of the abstract results in § 2.2 remain valid in this case. We cannot use
16 PART I. QUESTIONS RELATED TO SOLUTIONS

Fourier series any more, so that neither can A'lf be explicitly written, nor can
the eigenfunctions of A be calculated. Still, A is an isomorphism from V onto
V and from D(A) onto H, but this last result is a nontrivial one relying on the
theory of regularity of solutions of elliptic systems (cf. S. Agmon-A. Douglis-
L. Nirenberg [1]), L. Cattabriga [1], V. A. Solonnikov [1], I. I. Vorovitch-V. I.
Yodovich [1]). The spaces Va = D(A°'/2), a >0, (which will not be used too
often) are still closed subspaces of H a (n), but their characterization is more
involved than (2.15) and contains boundary conditions on F. All the in-
equalities in § 2.3 apply to the bounded case, replacing just H™(Q) by H m (H)
(assuming that (1.7) is satisfied with an appropriate r). The proof of (2.20)4,
which was elementary, relies in the bounded case on the theory of interpola-
tion, as well as on the definition of Hm((l), m e [ R \ N (cf. J. L. Lions-E.
Magenes [1]). Finally, with these definitions of A, V, H , . . . , § 2.4 applies to
the bounded case without any modification.

4
In the bounded case, (2.20) becomes
3 Existence and Uniqueness Theorems
(Mostly Classical Results)

In this section we derive basic a priori estimates for the solutions of

Navier-Stokes equations, and we recall the classical existence or uniqueness
theorems of weak or strong solutions. The only recent result is the theorem of
generic solvability of Navier-Stokes equations, given in §3.4 and due to A. V.
Fursikov [1].

3.1. A priori estimates. We assume that u is a sufficiently regular solution of

Problems 2.1-2.2, and we establish a priori estimates on u, i.e., majori-
zations of some norms of u in terms of the data u{), / , . . . .
i) By (2.41), for every re(0, T) and v e V,

Replacing v by u(t) we get, with (2.34),

Hence

By integration in f from 0 to T, we obtain, after dropping unnecessary terms,

Then by integration in t of (3.3) from 0 to s, 0<s < T, we obtain

ii) Assuming again that u is smooth, in (3.1) we replace v by Au(t):

17
18 PART I. QUESTIONS RELATED TO SOLUTIONS

Since

this relation can be written

Now the computations are different, depending on the dimension,

iii) Dimension n = 2. We use the relation (2.31); (3.9) implies

The right-hand side can be majorized by

using Young s inequality in the form

with p = 5 and e = v/2. We obtain

Momentarily dropping the term v \A.u(t)\ , we have a differential inequality,

from which we obtain by the technique of Gronwall's lemma:

or
 EXISTENCE AND UNIQUENESS THEOREMS (MOSTLY CLASSICAL RESULTS) 19

With (3.4)-(3.7),

We come back to (3.11), which we integrate from 0 to T:

iv) Before treating the case n = 3, let us mention an improvement of the

preceding estimates in the periodic case, for n = 2. This improvement, which
does not extend to the case of the flow in a bounded domain or if n - 3, is based
on:
LEMMA 3.1. In the periodic case and if n = 2,

Proof. In the periodic case A(f> = — Ac/>, and

and, by integration by parts, using the Stokes formula,

Now both integrals vanish, the first one because

and the second one because the sum £?/tk = i D^D^D^ vanishes identically
(straightforward calculation).

1
c2 (and therefore c{) depends on the domain 0, i.e., on L if 0 = Q.
20 PART I. QUESTIONS RELATED TO SOLUTIONS

This lemma allows us to transform (3.9) into the simpler energy equality

from which we derive, as for (3.4)-(3.7),

v) Dimension n = 3. In the case n = 3, we derive results which are similar to

that in part iii) (but weaker).
After (3.9) we use (2.32) instead of (2.31), and we obtain

(by Young's inequality),

But

This is similar to (3.11), but instead of (3.12), the comparison differential

inequality is now

We conclude that
 EXISTENCE AND UNIQUENESS THEOREMS (MOSTLY CLASSICAL RESULTS) 21

as long as t < l/(2y(0) c 4 ), and thus

for

LEMMA 3.2. There exists a constant K6 (= 3/8c4) depending only on /, v, Q, T

such that

for

It follows that if n = 3 and u is a sufficiently regular solution of Problems

2.1-2.2 then (assuming 7^ ^ T)

Tj given by Lemma 3.2, and from (3.24),

3.2. Existence and uniqueness results. There are many different existence
and uniqueness results for Navier-Stokes equations. The next two theorems
collect the most typical results, obtained in particular by J. Leray [1], [2], [3],
E. Hopf [1], O. A. Ladyzhenskaya [1], J. L. Lions [1], J. L. Lions-G. Prodi [1],
and J. Serrin [1].
THEOREM 3.1 (weak solutions). For f and u0 given,

there exists a weak solution u to the Navier-Stokes equations (Problem 2.1)

satisfying

as well as (2.39) (or (2.43)) and (2.40).

Furthermore, if n = 2, u is unique and

2
t ^ T,(||u0!|) and obviously t ^ T.
22 PART I. QUESTIONS RELATED TO SOLUTIONS

If n - 3, u is weakly continuous from [0, T] into H:

and

THEOREM 3.2 (strong solutions, n = 2, 3). i) For n = 2, / and u0 given,

there exists a unique strong solution to the Navier-Stokes equations (Problem

2.2), satisfying3:

ii) For n = 3, f and u0 given, satisfying (3.38), there exists TH.= T^e(u0) =
min (T, T^UoH)), Ti(lluoll) given by (3.29), and, on [0, T%], there exists a unique
strong solution u to the Navier-Stokes equations, satisfying (3.39), (3.40) with T
replaced by T*.
Remark 3.1. i) The theory of existence and uniqueness of solutions is not
complete for n = 3: we do not know whether the weak solution is unique (or
what further condition could perhaps make it unique); we do not know
whether a strong solution exists for an arbitrary time T. See, however, § 3.4.
ii) We recall that as long as a strong solution exists (n = 3), it is unique in the
class of weak solutions (cf. J. Sather and J. Serrin in J. Serrin [1], or [RT, Thm.
III.3.9]).
iii) Due to a regularizing effect of the Navier-Stokes equations for strong
solutions, those solutions can have further regularity properties than (3.39)-
(3.40) if the data are sufficiently smooth: regularity properties will be investi-
gated in § 4 for the space periodic case and in § 6 for the bounded case.
Let us also mention that if n = 2 and in (3.38) we assume only that u0eH,
then the solution u is in L?oc ((0, T]; D(A)) and <g((0, T]; V).
Remark 3.2. The strong solutions (and the weak solution if n = 2) satisfy the
energy equality (3.2). For n - 3 we know only that there exists a weak solution
which verifies the energy inequality

It is not known whether all the weak solutions satisfy this inequality or whether
this inequality is actually an equality.
Remark 3.3. Let us assume that the conditions (3.38) are verified. Then for
n = 2, if u is a weak solution to the Navier-Stokes equations (Problem 2.1), u
is automatically a strong solution by uniqueness and Theorem 3.2. For n =3,

;
/eL 2 (0, T;H) is sufficient for n = 2.
 EXISTENCE AND UNIQUENESS THEOREMS (MOSTLY CLASSICAL RESULTS) 23

the same is true on (0, T#(u0)), but u will be a strong solution on the whole
interval [0, T] if u satisfies some further regularity property.
For instance, if a weak solution u belongs to L4(0, T; V), then u is a strong
solution: indeed u is a strong solution on some interval (0, T%), T*^= T, Let us
assume that the largest possible value of T# is T' < T. Then we must have

On the other hand, the energy inequality (3.24), which is valid on (0, T'-e) for
all e > 0, implies

and by the technique of Gronwall's lemma,

Thus \\u(t)\\ is bounded for t —> T' - 0, contradicting the assumption that T' < T.
3.3. Outlines of the proofs. The proofs of Theorem 3.1 and 3.2 can be
found in the original papers or in the books of O. A. Ladyzhenskaya [1], J. L.
Lions [1], [RT]. We finish this section with some outlines of the proofs which
we need for the sequel.
i) We implement a Galerkin method, using as a basis of H the eigenfunc-
tions w,-,/eN, of the operator A (cf. (2.17)). For every integer m, we are
looking for an approximate solution um of Problems 2.1-2.2,

The function um satisfies, instead of (2.39)-(2.40),

where
Pm is the orthogonal projector in H onto Wm.
The semiscalar equation (3.42) is equivalent to the ordinary differential
system

4
Due to the definition of the w,-'s, Pm is also the orthogonal projector on Wm, in
V, V, D ( A ) , . . . .
24 PART I. QUESTIONS RELATED TO SOLUTIONS

The existence and uniqueness of a solution wm to (3.42)-(3.45) defined on

some interval (0, Tm), T m >0, is clear; in fact the following a priori estimate
shows that Tm = T.
ii) The passage to the limit m —> °° (and the fact that Tm = T) is based on
obtaining a priori estimates on um. The first estimate, needed for Theorem 3.1,
is obtained by replacing v by um (= um(t)} in (3.42). Using (2.34) we get

This relation is the same as (3.2), and we deduce from it the bounds on u^
analogous to (3.4)-(3.7):

Since |u 0m | = |PmM0| = |MO|, we get for um exactly the same bounds as (3.4)-(3.7),
from which we conclude that

Then, as usual in the Galerkin method, we extract a subsequence wm, weakly

convergent in L2(0, T; V) and L°°(0, T; H),

The passage to the limit in (3.42), (3.43) allows us to conclude that u is a

solution of Problem 2.1 and proves the existence in Theorem 3.1. However,
this step necessitates a further a priori estimate and the utilization of a
compactness theorem; we will come back to this point in a more general
situation in § 13. Finally, we refer to [RT] for the proof of the other results in
Theorem 3.1.
iii) For n = 2, the only new element in Theorem 3.2 is (3.39)-(3.40). We
obtain that ueL 2 (0, T; D(A))nL°°(0, T; V) by deriving further a priori esti-
mates on um, in fact, a priori estimates similar to (3.14)-(3.17). We obtain them
by taking the scalar product of (3.45) with Aum. Since Pm is selfadjoint in H
and PmAum - APmum = Aum, we obtain, using (3.8),

This relation is similar to (3.9). Exactly as in §3.1, we obtain the bounds

(3.14)-(3.17) for um, with u0 replaced by w0m. Since u0m = Pmu0 and Pm is an
orthogonal projector in V,
 EXISTENCE AND UNIQUENESS THEOREMS (MOSTLY CLASSICAL RESULTS) 25

and we then find for um exactly the same bounds as for u in (3.14)-(3.17):

By extraction of a subsequence as in (3.48), we find that u is in

L2(0, T;D(A))nL~(0, T; V). The continuity property (3.40) results from
(3.39) by interpolation (cf. J. L. Lions-E. Magenes [1] or [RT]). The fact that
u'eL 2 (0, T;H) follows from (2.43),

and the properties of B;f and Au are obviously in L (0, T; H) and for B we
notice in the relation (2.31) that

so that Bu belongs to L4(0, T; H) and u' belongs to L2(0, T; H).

The proof of Theorem 3.2 in the case n = 3 is exactly the same, except that
we use the estimates (3.30), (3.31) valid on [0, T^WoU)] instead of the estimates
(3.14)-(3.17) valid on the whole interval [0, T]. Also, instead of using (2.31)
and obtaining (3.53), we use the relation (2.32) which gives us

Bu is still in L2(0, T; H), and u' is too.

Remark 3.4. Except for Lemma 3.1 and (3.20)-(3.23), the a priori estimates
and the existence and uniqueness results are absolutely the same for both the
space periodic case and the flow in a bounded domain with u — 0 on the
boundary (cf. §2.5). If u = < / > ^ 0 on the boundary and/or n is unbounded,
similar results are valid. We refer to the literature for the necessary modifica-
tions.

3.4. Generic solvability of the Navier-Stokes equations. We do not know

whether Problem 2.2 is solvable for an arbitrary pair u(), /, but this is generi-
cally true in the following sense (A. V. Fursikov [1]):
THEOREM 3.3. For n = 3, given v, 0 (= Q or fl) and u 0 belonging to V, there
exists a set F, included in L2(0, T; H) and dense in

such that for every /eF, Problem 2.2 corresponding to u 0 ,/, possesses a unique
solution (strong solution}.
Proof, i) Since L2(0, T;H) is dense in L q (0, T; V), l^q<|, it suffices to
show that every /eL 2 (0, T; H) can be approximated in the norm of
L q (0, T; V) by a sequence of /m's, fm e L2(0, T; H), such that Problem 2.2 for
u0, fm possesses a unique solution.
For that purpose, given /, we consider the Galerkin approximation um
described in § 3.3 above (cf. (3.41)-(3.45)). It is clear that u^ is in L2(0, T; W m )
and hence in L2(0, T; D(A)) and that um is continuous from [0, T] into Wm
and hence into D(A). Now for every m we consider also the solution vm of the
26 PART I. QUESTIONS RELATED TO SOLUTIONS

linearized problem

It is standard that the linear problem (3.56)-(3.57) possess a unique solution

satisfying in particular

We then set wm = um + vm and observe that vvm satisfies

and by adding (3.56) to (3.45) and (3.57) to (3.43),

The proof will be complete if we show that, for m —> o°,

ii) Since |(/-Pm)/(0|->0 for m^<x> for almost every t, and \(I-Pm)f(t)\^
\f(t)\, it is clear by the Lebesgue dominated convergence theorem that (I—
Pm)/-» 0 in L2(0, T; H) for m -* «.
Multiplying (3.56) by Avm we get

from which it follows that, for m -* °°,

Using (3.64), the estimate (3.47) for um and Lemma 2.1 (with m^O, m2 =
1, m3 = 1), we find

In a similar manner we prove that

for m —> oo, and the proof of (3.63) is reduced to that of

 EXISTENCE AND UNIQUENESS THEOREMS (MOSTLY CLASSICAL RESULTS) 27

iii) The proof of (3.65) is technical and we give only a sketch of it.
The sequence um converges to some limit u (cf. (3.48))5. Since Bu e
L (0, T; V')6, and since, by Lebesgue's theorem, (I~Pm) Bu -»0 in
4/3

L4/3(0, T; V), it suffices to show that

Due to (2.33) and Lemma 2.1,

therefore B is a bilinear continuous operator from V3/4 x V3/4 into V (see also
(2.36)) and

It follows from the proof of Theorem 3.1 (cf. the cited references) that um
converges to u strongly in L2(0, T; V^J and L 1/e (0, T;H) for all e >0. By
(2.20) with m j = 0 , m 2 = l - e , 0m2 + (l-0)m 1 = 0(l-e) = i and the Holder
inequality,

l/p = e ( l - 0 ) + 0/2. We conclude that um -> u in L p (0, T; V3/4), and with

(3.67), that Bi^-^Bu in L p/2 (0, T; V). Finally, since p<8/3, p-^ 8/3, as
e —» 0, we choose e sufficiently small so that p/2 = q
Remark 3.5. As indicated in Remark 3.1, the theory of existence and
uniqueness of solutions is not complete when n = 3, while it is totally satisfac-
tory for n = 2.
It was Leray's conjecture on turbulence, which is not yet proved nor
disproved, that the solutions of Navier-Stokes equations do develop sin-
gularities (cf. also B. Mandelbrot [1], [2]). It seems useful to study the properties
of weak solutions of Navier-Stokes equations with the hope of either proving
that they are regular, or studying the nature of their singularities if they are
not.
The results in §§ 4, 5 and 8 tend in this direction. Of course they (as would
Theorem 3.3) would lose all of their interest if the existence of strong solutions
were demonstrated.

In (3.48) a subsequencem <u of um converges to u, but this is sufficient.

Cf. [RT]; this is how (3.37) is proved.
This page intentionally left blank
A New A Priori Estimates and Applications
In this section we establish some properties of weak solutions of Navier-
Stokes equations. The results are proved for the space periodic case, but they
do not all extend to the bounded case (cf. the comments, following § 14). We
assume throughout this section that n = 3 and the boundary condition is space
periodicity.

4.1. Energy inequalities and consequences. We derive formal energy in-

equalities assuming that u0, /, u are sufficiently regular.
LEMMA 4.1. If u is a smooth solution of Problems 2.1-2.2 (space periodic case,
n = 3), then for each t >0, for any r^l,

where the constant Lr depends on the data, v, Q and N r _ t (/) = |/| L «. (O ,T ; V,_,)•
Moreover, for any r^3, we have

where L'r depends also on v, Q and N r _ t (/).

Proof, i) We take the scalar product in H of (2.43) with Aru, and we obtain

By the definition (2.42) of B and as Au = -Au in the space periodic case (cf.
(2.14)), we can write

The first term in the right-hand side of (4.3) is majorized by

The second term is a sum of integrals of the type

We can integrate by parts using Stokes' formula; the boundary terms on dQ

cancel each other due to the periodicity of u, and the integrals take the form

29
30 PART I. QUESTIONS RELATED TO SOLUTIONS

With Leibniz' formula, we see that these integrals are sums of integrals of
the form

and of integrals of the form

where 5 k is some differential operator Da with

The sum for i = 1, 2, 3 of the integrals (4.6) vanishes because of the condition
div u = 0. Then it remains to estimate the integrals (4.7).
ii) Proof of (4.1). By Holder's inequality, the modulus of (4.7) is less than or
equal to

From the Sobolev injection theorems

(2.18)), this term is less than or equal to

c { depending on k, r, Q.
We then apply the interpolation inequality (2.20) with

to get

Therefore the modulus of the integrals of type (4.7) is bounded by an

expression

Hence

(by Young's inequality

This relation with (4.3) and (4.4) gives (4.1).
iii) Proof of (4.2). We majorize b(u, u, A r u) in a slightly different way. We
write
 NEW A PRIORI ESTIMATES AND APPLICATIONS 31

by application of (2.20) with m^ = 1, m2 = m + l, 0 = l-l/r. Then (4.8) gives

This relation combined with (4.3)-(4.4) gives (4.2).

An immediate consequence of (4.2) is that u remains in Vr as long as \\u(t)\\
remains bounded. This is expressed in
LEMMA 4.2. If u 0 e Vr and /eL°°(0, T; V^), r ^ l , then the solution u to
Problem 2.2 given by Theorem 3.2 ii) belongs to <g([0, T*]; Vr).
If u 0 e V and /eL°°(0, T; V r _j), r ^ l , then we<g((0, T*]; Vr).
Proof, i) We consider the case u 0 e Vr, and we first show that u belongs to
i-°°(0, T*; Vr). For that it suffices to prove that the Galerkin approximation i^
of u constructed in § 3 remains bounded in L°°(0, T^; Vr) as m -»<». We take
the scalar product in H of (3.45) with ATum = (—\)rb.rum, and since Pm is
self ad joint in H and PmArum = A'^, we get

This is similar to (4.3) and thus, exactly as in Lemma 4.1, we get the
analogue of (4.2):

due to (3.51) we conclude that

Now Pm is a projector in Vr too, and |PmWo|r^|w0|r, so that um remains in a

bounded set of L°°(0, T*; Vr) and L2(0, T*; Vr+1), and

We then check that Bu, and therefore u = f-vAu-Bu, belongs to

L2(Q, T%; V r _j); the continuity of u from [0, T^] into Vr follows by interpolation
(cf. J. L. Lions-E. Magenes [1] or [RT, §111.1]).
ii) For u 0 e V, we observe that the solution u of Problem 2.2 belongs to
L2(0, T*; D(A)). Thus u(t}e D(A) = Vr almost everywhere on (0, T*), and we
can find ^ arbitrarily small such that u(t 1 )eV 2 . The first part of the proof
shows that a e ^ d ^ , T*]; V2]nL2(tl; T*; V3). Hence u(f 2 )e V3 for some r 2 e
[*i, 7*], t2 arbitrarily close to t a , and ue^([t2, T*]; V 3 )nL 2 (f 2 , T*; V4). By
induction we arrive at u < # ([*,_!, T*]; V r )nL 2 (<,._!, T*; V^) and, since f r _ 1 is
arbitrarily close to 0, the result is proved.
32 PART I. QUESTIONS RELATED TO SOLUTIONS

4.2. Structure of the singularity set of a weak solution. Let m ^ 1. We say

that a solution u of Problem 2.1-2.2 is Hm-regular on (f 1} r2) (O^f^fz) if
u e <#((*!, f 2 );H m (Q)). We say that an Hm-regularity interval (r l5 r2) is maximal
if there does not exist an interval of B-Om-regularity greater than (f l 5 1 2 ).
The local existence of an Hm-regular solution is given by Lemma 4.2: if
UQ<= Vm and /eL°°(0, T; V m _ t ), then there exists an (HP-regular solution of the
Navier-Stokes equations defined on some interval (0, r0). Also, it follows easily
from Lemma 4.2 that if ( r l 5 t 2 ) is a maximal interval of Hm-regularity of a
solution M, then

We are now able to give some indications on the set of Hm-regularity of a

weak solution.
THEOREM 4.1 (n = 3, space periodic case). We assume that u 0 eH, /e
L°°(0, T; Vm_l), m=Sl, and that u is a weak solution of the Navier-Stokes
equations (Problem 2.1). Then u is Mm-regular on an open set of (0, T) whose
complement has Lebesgue measure 0.
Moreover, the set of Hm-regularity of u is independent of m, i.e. is the same for
r = 1 , . . . , m.
Proof. Since u is weakly continuous from [0, T] into H, u(t] is well defined
for every t and we can define

It is clear that Or is open for every r.

For r= 1, since weL 2 (0, T; V), Si has Lebesgue measure 0. If t0 belongs to
O,i and not to Ol then, according to Theorem 3.2 ii), f 0 is the left end of an
interval of H1-regularity, i.e. one of the connected components of GV Thus
n^O1! is countable and [0, TJX^j has Lebesgue measure 0.
The theorem is proved for m = 1. We will now complete the proof by
showing that Om =6^.
If (t\, t2) is a connected component of Ol (a maximal interval of H1-
regularity), then for every t{ in this interval, u(t'i)e V and, according to Lemma
4.2, there exists a unique IH!m-regular solution defined on some interval (f(, t2),
t'1<t2^=t2. Since uniqueness holds also in the class of weak solutions (cf.
Remark 3.1 ii)) this solution coincides with u, i.e., ( t { , t 2 ) is an interval of
Hm-regularity of u. According to (4.2), u remains bounded in H m as long as u
is bounded in V. Therefore, using Lemma 4.2 also, we see that t2 = t2, and
since t[ is arbitrarily close to f 1? (^, t2) is an interval of Hm-regularity. This
proves that Om =Ol} and

4.3. New a priori estimates.

THEOREM 4.2 (n = 3, space periodic case). We assume that u 0 eH, fe
L2(0, T; V m _ t ) and that u is a weak solution of the Navier-Stokes equations
 NEW A PRIORI ESTIMATES AND APPLICATIONS 33

(Problem 2.1). Then u satisfies

r = l , . . . , m + l, where the constants cr depend on the data v, Q, u0, /, and the

a/s are given by

Proof, i) Let (ah ft), i e N , be the connected components of 0^, which are
also maximal intervals of H m -regularity of u.
On each interval (ah ft), the inequalities (4.1) are satisfied, r = 1,. . . , m, and
we write them in the slightly stronger form

Then we deduce

By integration in t from «j to ft, we get

From (4.12), the first term in the left-hand side of this inequality vanishes,
since (a f , ft) is a maximal interval of IH m -regularity. Thus

By summation of these relations for i e N , we find, since u e L2(0, T; V),

ii) The proof of (4.13) is now made by induction. The result is true for
r ~ 1. We assume that it is true for 1, . . . , r, and prove it for r +1 (r^i m).
We have

(by Holder's inequality),

34 PART I. QUESTIONS RELATED TO SOLUTIONS

where

Therefore (4.13) follows for r + 1, due to (4.16) and the induction

assumption.
Remark 4.1. As indicated at the beginning of this section, all the results
have been established in the periodic case. We do not know whether they are
valid in the bounded case (although it is likely that they are) except for the
following special results: (4.1) and (4.2) for r = 1, which coincide with (3.24); in
Theorem 4.1 the fact that the complement of Oi in [0, T] has Lebesgue
measure 0 and is closed; and (4.13) for r = 2(«r=|) (same proofs)1.
The following interesting consequence of Theorem 4.2, which relies only on
(4.13) with r-2, is valid in both the periodic and the bounded cases.
THEOREM 4.3. We assume that n = 3 and we consider the periodic or bounded
case. We assume that uQeH and /eL°°(0, T; H). Then any weak solution u of
Problem 2.1 belongs to L\Q, T;l°°(0)), O = fl or Q.
Proof. Because of (2.26)

and by Holder's inequality

and the right-hand side is finite, thanks to (4.13) (r = 2, cf. Remark 4.1).
1
Cf. also § 6, and in particular Remark 6.2, for a complete extension of Theorem 4.1 to the
bounded case.
j Regularity andFractionalDimension

If the weak solutions of the Navier-Stokes equations develop singularities, as

Leray has conjectured, then a natural problem is to study the nature of the
singularities. B. Mandelbrot conjectured in [l]-[2] that the singularities are
located on sets of Hausdorff dimension <4 (in space and time), and V. Scheffer
has given an estimate of the dimension of the singularities which he succes-
sively improved in [l]-[4]. A more recent (and improved) estimate is due to L.
Caffarelli-R. Kohn-L. Nirenberg [1]. In this section we present some results
concerning the Hausdorff dimension of the singular set of weak solutions,
following (for one of them) the presentation in C. Foias-R. Temam [5].

5.1. Hausdorff measure. Time singularities. We recall some basic defini-

tions concerning the Hausdorff dimension (cf. H. Federer [1]). Let X be a
metric space and let D >0. The D-dimensional Hausdorff measure of a subset
Y of X is

where

the infimum being taken over all the coverings of Y by balls Bj such that
diam By ( = diameter of B,)^e.
It is clear that ^ D-e (Y)^ De .(Y) for e^e' and fj, D (Y)e[0,+<»]. Since
/* D , e (Y)^e D -% Doe (Y) for D>D 0 , then if ^Da(Y)<™ for some D 0 e(0,«>)
then fx D (Y) = 0 for all D>D 0 . In this case the number

is called the Hausdorff dimension of Y. If the Hausdorff dimension of a set Y is

finite, then Y is homeomorphic to a subset of a finite dimensional Euclidean
space. Finally, let us also mention the useful fact that JJLD(-) is countably
additive on the Borel subsets of X (see Federer [1]).
The first result on the Hausdorff dimension of singularities concerns the set
of t in [0, T] on which u(t) is singular, u(t)£ V. Actually this is a restatement
by V. Scheffer [1] of a result of J. Leray [3] (cf. also S. Kaniel-M. Shinbrot [1]).
THEOREM 5.1. Let n = 3, C = fl or Q, and let u be a weak solution to the
Navier-Stokes equations (Problem 2.1). Then there exists a closed se
whose ^-dimensional Hausdorff measure vanishes, and such that u is (at least)
continuous from [0, T]\<£ info V.
Proof, i) The proof of this theorem is partly contained in that of Theorem
4.1 (cf. also Remark 4.1). We set # = [0, TJXC1!, and we only have to prov
that the ^-Hausdorff measure of <£ is 0.
35
36 PART I. QUESTIONS RELATED TO SOLUTIONS

Let (ah ft), i/ , be the connected components of Gl. A preliminary result,
due to J. Leray [3], is

Indeed, let (ah ft) be one of these intervals and let te(di, ft). According to
Theorem 3.2 and (3.29),

where K6 depends only on /, v and C (= fl or Q). Thus

Then we integrate on (ah ft) to obtain

and we add all these relations for i e I to get

ii) The proof now follows that given in V. Scheffer [1, §3]. For every e >0,
we can find a finite part Ie of / such that

The set [0, T]\Uiei e («i» ft) is the union of a finite number of mutually disjoint
closed intervals, say B,, / = 1,. . . , N. It is clear that U/li Bj ^ $, and since the
intervals (ah ft) are mutually disjoint, each interval (at, ft), i£le, is included in
one, and only one, interval Bj. We denote by 7, the set of fs such that
BJ =>(«<, ft). It is clear that /E, 7 l 5 . . . , / N , is a partition of I and that J3, =
(IU (<*, ft)) U (ft n *) for all / = 1 , . . . , N. Hence
diam

and

Letting e -» 0, we find

5.2. Space and time singularities. The second result concerns the Hausdorff
dimension in space and time of the set of (possible) singularities of a weak
 REGULARITY AND FRACTIONAL DIMENSION 37

solution (cf. V. Scheffer [l]-[4], C. Foias-R. Temam [5]):

THEOREM 5.2. Let n - 3, G = £l or Q; let u be a weak solution to the Navier-
Stokes equations (Problem 2.1) and assume moreover that

Then there exists a subset 00<^0 such that

(5.7) the Hausdorff dimension of

Before proving Theorem 5.2 we present some preliminary material of
intrinsic interest.
Lemma 5.1 is borrowed from V. Scheffer [1]; we reproduce it for the
convenience of the reader 1
LEMMA 5.1. For a > 0 and feL'([R"), let A a (f) be the set of those xeUn such
that there exists mx with

Then the Hausdorff dimension of IR"\A a (/) is ^a.

Proof. By definition of A a (/), for any e >0 and x e!R" \ A a (_f) there exists a
ball B F (x) centered in x such that

By a Vitali covering lemma (see E. M. Stein [1]), there exists a system

{B e (x,):/eJ}c={B F (x):xe[Fr\A a (/)} such that the Be(x,-)'s are mutually dis-
joint, J is at most countable and

where B'E(XJ) (j&J) denotes the ball centered in xf and with diameter
5(diamB E (x,-)). By virtue of (5.1), (5.9), (5.10), it follows that

arid the Hausdorff dimension of [R n \ A a (/) does not exceed a.

1
In Lemma 5.1 and Theorem 5.3 the dimension of space n is any integer ^1 (and is not
restricted as everywhere else to 2 or 3).
38 PART I. QUESTIONS RELATED TO SOLUTIONS

THEOREM 5.3. Let G be an open subset in Un and let Te (0, <»), K p <°° and
/eL p (Gx(0, T)) be given. Let moreover

be such that

in the distribution sense in G x (0, T)

and

Then there exists a subset G0 <= G such that

and the Hausdorff dimension of G0 is Simax{n + 2 —2p, 0}.

Proof. We infer from (5.11)-(5.12) and the regularity theory for the heat
equation (cf. O. A. Ladyzhenskaya-V. A. Solonnikov-N. N. Ural'ceva [11])
that:

(where W 2 ' P (G) is the usual Sobolev space of functions defined on G, which
together with all their derivatives up to order 2 belong to L P (G)). If ^ e Co(G)
and vl = (f>v, then u t will satisfy (5.11)-(5.13), and also

where /t is a suitable function which also belongs to L p (Gx(0, T)).

Therefore, taking into account that the Hausdorff measures are Borel meas-
ures, we can consider only the special case

where K is a certain compact set <=(R" while to is of Lebesgue measure =0. It is

clear that in this case v(x, t) coincides almost everywhere in J?"x(0, T) with
the function

2
It is well known (cf. for instance J. L. Lions [1] or R. Temam [RT, Chap. Ill, Lemma 1.2]) that if
v satisfies (5.11) and (5.12) then v is equal for almost every f e ( 0 , T) to a function in
«([0, T]; L2(G)), and therefore u(0) makes sense.
 REGULARITY AND FRACTIONAL DIMENSION 39

where

and

Therefore we can assume that, for any t e (0, T)\o>, v(x, t) = w(x, t) everywhere
on IRn. It follows that for such f's and for all x 0 e[R n , we have

(by (4.16) and the change of variables

where, for all

From the classical theorem on the maximal functions (see E. M. Stein [1]), we
have

Since for t e [0, T]\o>, v(x, t)- w0(x, t) is continuous in x, from (5.17) we finally
infer that

But, setting
40 PART I. QUESTIONS RELATED TO SOLUTIONS

we have for all xeR", fe(0, T), and because of standard estimates on E:

where c'n,c'n,... are suitable constants (with respect to (jc, t}) and

Therefore A a (g) (see Lemma 5.1) makes sense, and if

obviously

Consequently (since w0 is continuous on [R n x(0, T)), by virtue of (5.19) we

have
 REGULARITY AND FRACTIONAL DIMENSION 41

Therefore the set

is included in R"\A a (g). From Lemma 5.1 we infer that the Hausdorff
dimension of G0 is ^a. The conclusion (5.14) is now obtained by letting
a —» max {0, n + 2 - 2p}.
Remark 5.1. The proof of Theorem 5.2 shows that the result remains valid
if G = Q and we replace (5.11) by

We now turn to the proof of Theorem 5.2.

Proof of Theorem 5.2. Obviously it is sufficient to prove the result for a fixed
Te (0, o°). Also it is clear that each component v(x, t) = w,(x, f) (/ = 1, 2, 3) of
u(t) satisfies (5.11)-(5.12). Since

we find by interpolation that weL r (0, T; L6r/(3r~4)(n)) for every ri=2 (see J. L.
Lions-J. Peetre [1]), whereupon, for r = -y, we obtain

It follows that

since

We then get

where (see (5.21))

By referring to the regularity theory for the Stokes system (5.22) (cf. K. K.
Golovkin-O. A. Ladyzhenskaya [1], K. K. Golovkin-V. A. Solonnikov [1], V.
A. Solonnikov [2]), we see that

3
This boundary condition must be replaced by the periodicity condition when G = Q; see
Remark 5.1.
42 PART I. QUESTIONS RELATED TO SOLUTIONS

for any Ka <|. Thus for v(x, t) = u,(x, t) we obtain

on so that by virtue of Theorem 5.3, we obtain that the set

has the Hausdorff dimension ^5-2a. The conclusion follows by letting

6 Successive Regularity and Compatibility
Conditions at t = 0 (Bounded Case)

In this section we assume that n = 2 or 3 and consider the flow in a bounded

domain jQcilR". We are interested in the study of higher regularity properties
of strong solutions, assuming that the data u0, /, O possess further regularity
properties. While this question is essentially solved in § 4 for the space periodic
case (through Lemmas 4.2 and 4.1), the situation in the bounded case is more
involved. In particular, we derive in this section the so-called compatibility
conditions for the Navier-Stokes equations, i.e., the necessary and sufficient
conditions on the data (on dfl at t = 0) for the solution u to be smooth up to
time t = 0.

6.1. Further properties of the operators A and B. We recall that if (1

satisfies (1.7) with r = m + 2 , and if m is an integer >n/2 ( m ^ 2 for n — 3, 2),
then H m (fl) is a multiplicative algebra:

For every integer m we define the space

which is a Hilbert space for the norm induced by H m (n) = H m (fl) n . It is clear
that Em+i<=Em, for all m and that E0 = H:

The orthogonal projection from L2(H) onto H being denoted by P as before,

we introduce the operator s&

which is linear continuous from JEm, mi?2, into E0 = H. Actually the operator
P is linear continuous from H m ((l) into H m (O) Pi H = Em (cf. [RT, Chap. I,
Remark 1.6]), and therefore
s£ is linear continuous from Em+2 into Em.
Now if u e E m + 2 n V, m ^ O , and s£u=feEm, then A u + / = (7-P)Au be-
longs to H1, which amounts to saying (cf. §2.5) that there exists peH 1 ^)
such that A u + / = Vp. Hence

43
44 PART I. QUESTIONS RELATED TO SOLUTIONS

and u is the solution to the Stokes problem associated with /. The theorem on
the regularity of the solutions of Stokes problem mentioned in § 2.5 implies
then that
s& is an isomorphism from E onto
2
It is clear from (6.6) that

Operator B. Lemma 2.1 shows us that b(u, v, w) is a trilinear continuous

form on H m '(n)xH m 2 + 1 (n)xH if ml + m2>n/2 (or ml + m2^n/2 if m^Ofor
all 0- This amounts to saying that the operator B defined in (2.41) is bilinear
continuous from H m '(ft)x[Hl m2+1 (n) into H, under the same conditions on
m l 5 m 2 . Clearly (u • V)u belongs in this case to L2(O) and

We have
LEMMA

Proof. If m = 1, u, v eH 2 (fl), then obviously (u • V}v eL2(H) and

belongs to L2(H), since the first derivatives of u and v are in L6(H) (at least) and
u and v are continuous on H. The operator P mapping fl-flr(^) into H r (n)n
H,B(u, v) = P((u - V ) u ) is in H 1 (n)nH=E 1 .
If m ^ 2, then ut and DtVj are in H2(O), and their product is too, due to
(6.1). Hence (u • V)ueH m (H) and P((u • V)u) as well3.

6.2. Regularity results. If u() and / are given, satisfying

then Theorem 3.2 asserts the existence of a (strong) solution u of Problem 2.2
defined on some interval [0, T"], T" = T if n — 2, T'^ T if n = 3. Changing our
notation for convenience, we write T instead of T', and

1
If $ is the inverse of ^| Em42 nv ^^M = u' f°r a" u e ^ m + 2 n ^ but in general ^stfu^ u for an
arbitrary u e Em+2, m ^ 0.
2
Because of (6.7) and (6.8), Vm is a closed subspace of Em, for all m e N, the norm induced by
£m being equivalent to that of Vm. In the space periodic case, the analogue of Em —H™(Q)r\H is
equal to Vm.
3
The proof shows also that B(Em x£ m + 1 )c Em for m ^ 2 .
 SUCCESSIVE REGULARITY AND COMPATIBILITY CONDITIONS AT t = 0 45

Our goal is to establish further regularity properties on M, assuming more

regularity on w 0 , /.
We introduce the space

where m is an integer and / = [m/2] is the integer part of m/2, i.e. m =21 or
m = 2 i + l.
We begin by assuming that for some m ^ 2

and we make the following observation:

LEMMA 6.2. If u is a strong solution to the Navier-Stokes equations and u{)
and f satisfy (6.15) for some m ^ 2 , then the derivatives at
(d]u/dt')(Q), / = ! , . . . , / , can be determined "explicitly''' in terms of u0 and /(O),
and

Proof. By successive differentiation of (2.43), we find

where we set <£ (/) = d'<J>/dfJ', 8(r) = Bu(r), so that

Now by (6.5), (6.10), (6.15), j3(0) = Bu(0)e£ m _ 1 , and u'(0) =

Similarly,

The proof continues by induction on / using (6.5), (6.10), (6.15), and shows
that

fo
It is known that for an initial and boundary value problem, the solutions may
not be smooth near f = 0, even if the data are c$co. For Navier-Stokes
equations, the following theorem gives the necessary and sufficient conditions
on the data for regularity up to t = 0, the compatibility conditions (cf. (6.21)).
46 PART I. QUESTIONS RELATED TO SOLUTIONS

THEOREM 6.1. We assume that n — 2 or 3, H satisfies (1.7) with r =

m + 2, m ^2, u0 and f satisfy (6.15) and

and that u is a (strong) solution of Problem 2.2.

Then a necessary and sufficient condition for u to belong to Wm is that

Proof, i) We first show that the condition (6.20) (which is void for m = 2) is
necessary if m ^3.
We know that u is in <g([0, T]; V). If u belongs to Wm, then u e
^([0, T];E m ), u'e<g([0, T];E m _ 2 ), and necessarily u' takes its values in
E m _ 2 H V, so that u' e <g([0, T]; E m _ 2 n V) and u'(0) 6 V. The proof is the same
for the other derivatives.
We now prove that under conditions (6.15), (6.20) and (6.21) u is in Wm.
ii) For / = !, the equation

together with u (1) (0)e V given by (6.19), allows us to show that 4

We continue by induction; once we establish that

we consider (6.17), (6.18) with 7 = / and u (n (0) given by (6.19) and belonging to
V (if m = 2J + 1) or to H (if m =21). In a similar manner, we show that

iii) The next step in the proof consists in showing that

4
The fact that u ( 1 ) eL 2 (0, T; D(A))nL°°(0, T; V) is obtained by deriving one more a priori
estimate for the Galerkin approximation um of u (cf. (3.42)-(3.45)): that u^ belongs to a bounded
set of L2(0, T; D(A))nL°°(0, T; V). As (3.47), (3.52), this estimate is obtained by differentiating
(3.45) with respect to t and taking the scalar product in H of the differentiated equation with u^,
and Au^,.
The fact that u e <£([(), T]; V) follows by interpolation, like (2.50), after we show that u" =
f-i/Au'-B(u', u)-B(u, u')eL 2 (0, T; H). It is clear that /' and vAu' are in L2(0, T; H). For the
quadratic terms, it follows from (2.47), (2.48) that B maps (L2(0, T; D(A))nL°°(0, T; V))2 into
L4(0, T;H).
 SUCCESSIVE REGULARITY AND COMPATIBILITY CONDITIONS AT t = 0 47

For that purpose we write (6.17), (6.18) in the form

f(i) and u ()+1) are in <g([0, T]; H) (at least). Because of Lemma 2.1 and (2.36),
B is bilinear continuous from V x V into V_1/2. Hence by (6.24), for i =
0 , . . . , / , and / = 0 , . . . , / - 1,

Since A is an isomorphism from V3/2 onto V_1/2, (6.27) implies then that
u (j) e^([0, T]; V 3/2 ),/ = 0,. . . ,1-1. Using again Lemma 2.1, B is a bilinear
continuous operator form V3/2 x V3/2 into H. Thus, for the same values of i and
/ as in (6.28),

and (6.26) follows,

iv) Finally we show that u e Wm, i.e.,

For j = I, this is included in (6.25). For j = 1 — 1, using (6.10) and (6.26) we find
that

and then (6.5), (6.27) and (6.25) show that u (( " 1} e <g([0, T]; E m _ 2l+2 ). The
proof continues by induction for j = 1 — 2,... ,0.
Theorem 6.1 is proved.

6.3. Other results. We particularize Theorem 6.1 to the case m = 3.

THEOREM 6.2. We assume that n = 2 or 3, fl satisfies (1.7) with r = 5,u 0 e
H 3 (n)nV,/e<£([0, rfcHX^HH) and d//dreL 2 (0, T;H). Then the solution
u of Problem 2.2 defined by Theorem 3.2 on some interval [0, T*], 0< T^T,
belongs to <£([0, T*];H3(n)n V), if and only if

The only condition left in (6.21) is (6.30). Since u'(G)&El =H 1 (H)nH,

u'(0) • v = 0 on F, and (6.30) is a restrictive assumption for the tangential value
on F of

Another way to formulate (6.30) is this: , where (cf. [RT, Chap.

48 PART I. QUESTIONS RELATED TO SOLUTIONS

I, Thm. 1.5]) q0 is a solution of the Neumann problem:

^0 must be such that the tangential components on F of ^o and Vq0 coincide:

(6.32), (6.33) constitute an overdetermined boundary value problem (Cauchy

problem on F for the Laplacian) which does possess a solution.
If the compatibility conditions (6.21) are not satisfied, the other hypotheses
of Theorem 6.1 being verified, we get a similar result of regularity on
Ox(0, T]:
THEOREM 6.3. We make the same hypotheses as in Theorem 6.1, but the
conditions in (6.21) are not satisfied5. Then

Proof. Thanks to (3.39), there exists 0< t 0 < T%, tQ arbitrarily close to 0, such
that u(f 0 )eD(A). Equation (6.19) for t0 and j = l shows that w'(f 0 )eH. We
conclude as in Theorem 6.1 or 3.2 that u'eL 2 (f 0 , T*; V)n^([f 0 ,1*]; H).
We choose r l s t0<t1<T^, t1 arbitrarily close to tQ, such that w'^^e V, and
conclude that u ' e L 2 ( f l 9 T*; D(A)}n(€([tl, T*]; V). We then choose f 2 > f i <
t2<T%, t2 arbitrarily close to tl, such that u'(t2)eD(A), etc.
Finally we get that u e <g([f,, T*]; E m ), u'e <£((>,, T*]; E m _ 2 ),. . ., for f, arbi-
trarily close to 0, and the result is proved.
Remark 6.1. If F is ^°°, u () e c€eo(£l)n n H, f e ^°°(nx[0, T])", then u e
«°°(n x [0, T])". The ^~ regularity in O x (0, T] was proved in O. A.
Ladyzhenskaya [1]. Of course, by combining Theorems 6.1 and 6.2 with the
imbedding theorems of Sobolev spaces into spaces of continuously differenti-
able functions, one can get partial results of regularity in spaces of ^ fc
functions.
Remark 6.2. Let u be a weak solution of the Navier-Stokes equations
(Problem 2.1) in the three-dimensional case. Let Ol be the set of H1-regularity
of u defined in Theorem 4.1 and Remark 4.1, and let <£ = [0, TJXOV According
to Theorem 6.3, if u 0 eH, f e Wm^2 and hypothesis (6.20) is verified, Oi is also
an interval of H r -regularity, r^m,i.e., ue^C^; H m (ft)n V). This is the
analogue of Theorem 4.1 for the case of a bounded domain.
Remark 6.3. Let u be a weak solution of Navier-Stokes equations (Problem
2.1) in the three-dimensional case. We assume that M 0 eIHl,f eH m - 2 (H)n
H, m^3, is independent of t. Then for every teO,u'(t) makes sense and

5
It suffices also to assume that u n e V instead of u n e Em D V.
 SUCCESSIVE REGULARITY AND COMPATIBILITY CONDITIONS AT t = 0 49

belong to V. If we introduce the analogue i/>(0 of (6.31)

and write that the overdetermined boundary value problem similar to (6.32),
(6.33) possesses a solution, we conclude that u(t) belongs, for t in C^ to a
complicated "manifold" of V: i// depends on u(t] (and /); q(t) =
jV(div i/f(0, <MO • v), where N is the "Green's function" of the Neumann
problem (6.32): the condition (6.33) is the "equation" of the manifold.
A similar remark holds for every t >0 for a strong solution; similar remarks
follow from the other conditions (6.21) if m ^5.
Remark 6.4. If we introduce pressure, then we clearly get regularity results
for pressure:

In the situation of Theorem 6.1,

and in the situation of Theorem 6.3,

This page intentionally left blank
7 Analyticity in Time

In this section we prove that the strong solutions are analytic in time as
D(A)-valued functions. We assume for simplicity that f e H is independent of
t: we could as well assume that / is an H-valued analytic function in a
neighborhood in C of the positive real axis. The interest of the proof given
below is that it is quite simple and relies on the same type of method as that
used for existence. The method applies also to more general nonlinear evolu-
tion equations with an analytic nonlinearity.

7.1. The analyticity result. The main result is the following one:
THEOREM 7.1. Let there be given u0 and f, u 0 e V, /e H independent of t.
If n=2, the (strong) solution u of Problem 2.2 given by Theorem 3.2 is
analytic in time, in a neighborhood of the positive real axis, as a D(A}-valued
function.
If n = 3, the (strong) solution u of Problem 2,2 given by Theorem 3.2 is
analytic in time, in a neighborhood in C of the interval (0, T#), as a D(A)-
valued function.
Proof, i) Let C denote the complex plane and Hc the complexified space of
H, whose elements are denoted u + iv, u, v e H, i = V/-^T. Similarly Vc, V'c are
the complexified V, V, and Xc is the complexified space of a real space X. By
linearity, A (resp. Pm, resp. B) extends to a selfadjoint operator in Hc (resp. to
the orthogonal projection in Hc, Vc, Vc, onto the space C w , + • • • +Cw m ,
resp. a bilinear operator from V c x Vc into Vc).
Consider now the complexified form of the Galerkin approximation of
Navier-Stokes equations, i.e., (compare with (3.41)-(3.45)) the complex
differential system in PmHc:

where £eC, and um maps C (or an open subset of C) into PmHc =

Cw, + • • •-f Cw m . The complex differential system (7.1), (7.2) possesses a
unique solution um defined in C in some neighborhood of the origin.
It is clear that the restriction of u m (£) to some interval (0, Tm) of the real axis
coincides with the Galerkin approximation u m ( t ) defined in the real field by
(3.41H3.45).
ii) As in the real case, we now prove some a priori estimates on um. We take
the scalar product in Hc of (7.1) with Au m (£)

51
52 PART I. QUESTIONS RELATED TO SOLUTIONS

We multiply this relation by eie, with £ = seie, and for fixed 0, |0| < ir/2, we get:

Therefore

The absolute value of the term involving / is less than or equal to

For the term involving B, we use the inequality (2.32), which clearly extends to
the complex case1,

This allows us to write

(by Young's inequality, see (3.10)),

where c' depends on the data v, u0, ft.
If we take into account this majorization and (7.4), then (7.3) becomes

The inequality (3.25) is still valid in the complex case

and thus

1
This relation, which is not so good as the first relation (2.31), is valid for both dimensions, n = 1
and 3.
 ANALYTICITY IN TIME 53

This differential inequality is of the same type as (3.27), setting y(s) =

\\u,,,(se'"}\\\ c'4 = ( l / ( c o s 3 0 ) ) ( ( 2 / i > ) \ f \ 2 + c'). We conclude, as in Lemma
3.2, that there exists K'(, which depends only on / , v and il (or (2), such
that

for

This shows that the solution um of (7.1)-(7.2), which was defined and
analytic in a neighborhood of £ = 0, actually extends to an analytic solution of
this equation in an open set of C containing (see Fig. 7.1):

The estimate (7.8)-(7.10) shows that

iii) The analyticity of um and Cauchy's formula allow us to deduce from

(7.12) a priori estimates on the derivatives of um (with respect to £) on compact
subsets of A(u ( ) ). Indeed, for £ e A ( w ( ) ) and k e N , k ^ l .

where d = d(£, dA(u,,)) is the distance of £ to the boundary dA(u 0 ) of A(u 0 ).

FIG. 7.1. The region A(u ; ) ) in C.

54 PART I. QUESTIONS RELATED TO SOLUTIONS

Therefore

and by (7.12), for any compact set K<=A(M O )

Considering (7.1), we find that

Thus

and it follows from (7.14) that for every compact subset K of A(u0)

where c'4<™, depends on K and the data,

Using again Cauchy's formula (7.13) and (7.16), we obtain also for every
£eK and keN:

where K' (containing K] is the set

iv) We now pass to the limit, m—»°°. Since the set {ve Vc, \\v\\^p} is
compact in Hc for any 0 < p < o°, we can apply to the sequence um the vector
 ANALYTICITY IN TIME 55

version of the classical Vitali's theorem. Thus we can extract a subsequence um,
which converges in Hc, uniformly on every compact subset of A(u 0 ) to an
Hc-valued function u*(£) which is analytic in A(u 0 ) and satisfies obviously:

Since the restriction of u^ to the real axis coincides with the Galerkin
approximation in R + of the Navier-Stokes equation, it is clear that the
restriction of u*(£) to some interval (0, T') of the real axis coincides with the
unique (strong) solution u of the Navier-Stokes equations given by Theorem
3.2. Hence u* is nothing else than the analytic continuation to A(u 0 ) (at least)
of u, and we will denote the limit w(£) instead of w*(£). Secondly the whole
sequence u m (-) converges to u(-) in the above sense (i.e. uniformly on compact
subsets of A(u 0 ), for the norm of Hc).
Since the injection of D(A) in V is compact, it also follows from (7.16) and
Vitali's theorem that the sequence um converges to u in V uniformly on every
compact subset of A(u 0 ), and that

with the same constant c^K) as in (7.17). Finally the majorizations (7.14),
(7.18) imply that dkujd£,k converges to dku/d£k in V uniformly on every
compact subset K of A(u 0 ), and

K' defined in (7.19).

To conclude, we observe that the reasoning made at t = 0 can be made at any
other point t0e (0, o°) such that u(r 0 )e V. We obtain that u is a D(A)-valued
analytic function at least in the region of C:

where the union is for those f 0 's for which (£ 0 e (0, °°) and u(t 0 )e V). Finally we
observe that, actually, A(u 0 ) = A(||u0||) depends only on the norm in V of u0 and
decreases as ||u0|| increases. Therefore if u is bounded in V on some interval
[a,^],sup te[a ^ ] ||u(r)||^R, then

and this guarantees that the domain of analyticity of u contains the region
mentioned in the statement of Theorem 7.1 (See Fig. 7.2.) The proof of
Theorem 7.1 is complete.
56 PART I. QUESTIONS RELATED TO SOLUTIONS

Fir, 7.2. Uloe(a,3,{'o + A(R)}.

7.2. Remarks.
Remark 7.1. Of course the result of analyticity given by Theorem 7.1
applies as well to a weak solution of the Navier-Stokes equations (Problem 2.1,
n = 3) near a point t0 belonging to the set of (HI1-regularity (cf. § 4). If f e H is
independent of t, and (a, /3) is a maximal interval of H1-regularity of a weak
solution u, then u is a D(A)-valued analytic function in

Remark 7.2. Theorem 7.1 and Remark 7.1 extend easily to the case where /
is an He-valued analytic function of time in a neighborhood of the positive real
axis: it suffices to replace everywhere in the proof A(w () ) (or f() + A(u(f 0 ))) by its
intersection with the domain of analyticity of /.
We conclude this section with the following:
PROPOSITION 7.1. Under the same hypotheses as in Theorem 7.1, for 0<t^
4-* 1 u0||), Ti(||woll) given by (7.10), the following relations hold:
3 ,
T

where

Proof. It suffices to apply (7.22), (7.23) with K = {t}. The distance d(t, dA(u0))
appears in the right-hand side of these inequalities, and an elementary calcula-
tion shows that for 0 ^ t ^ { T\( I I M O I I ) , this distance is
Remark 7.3. When the compatibility conditions (6.21) are not satisfied, u is
not smooth near t = 0 and the H-norm of the derivatives u (k)t tends to
infinity as t —» 0. The relations (7.21), (7.22) give some indications on the way
in which they tend to infinity. A similar result has been obtained by totally
different methods by G. looss [1] and by D. Brezis [1].
X Lagrangian Representation of the Flow

We assume that the fluid fills a bounded region O of 1R3. The Lagrangian
representation of the flow of the fluid, mentioned in § 1, is determined by a
function <£:{a, r}eHx(0, T)—»4>(a, t)e£l, where O(a, t) represents the posi-
tion at time t of the particle of fluid which was at point a at time t - 0 (flow
studied for time r, O ^ r ^ T ) . This also amounts to saying that {t >-»<I>(a, t}} is
the parametric representation of the trajectory of this particle.
The Lagrangian representation of the flow is not used too often because the
Navier-Stokes equations in Lagrangian coordinates are highly nonlinear. It
plays an important role, however, in two cases at least: it is used in the
numerical computation of a flow with a free boundary, and, in the mathemati-
cal theory of the Navier-Stokes equations, it is the starting point of the
geometrical approach developed by V. I. Arnold [1], D. Ebin-J. Marsden [1],
among others.
If we are given u, a strong solution of the Navier-Stokes equations, then it is
easy to determine the trajectories of the particles of fluid for the corresponding
flow: for every a e f l , the function t •-» 3>(a, t), also denoted £ or £a, is a
solution of the ordinary differential equation

with initial (or Cauchy) data

Our goal in this section is to show that, using one of the new a priori
estimates mentioned in § 4, it is possible to determine the trajectories of the
fluid (in some weak sense), even if u is a weak solution of the Navier-Stokes
equations.

8.1. The main result. We assume for convenience that

and that we are given u, a weak solution of the Navier-Stokes equations

(Problem 2.1) associated with u() and /, and that Cl satisfies (1.7) with r = 2.
We have first to give meaning to (8.1), (8.2) since u is not regular.
LEMMA 8.1. If ^ is a continuous function from [0, T] into O and u is a weak
solution of Problem 2.1, then u(£(f), t} is defined for almost every t e[0, T], the
function f >->u(|(0, t} belongs to L J (0, T;R 3 ) and (8.1) and (8.2) make sense.
Proof. It follows from Theorem 4.1, Remark 4.1 and Theorem 5.1, that u is
continuous from [0, T]\£ into D(A), where  has Lebesgue measure 0. Due
to (1.7), H2(H) is included in <g(A) (cf. §§2.3 and 2.5, dimension n = 3), and
D(A)c=^(fi) 3 . Hence u(£(f), t) is well defined for every fe[0, T]\£ and is
57
58 PART I. QUESTIONS RELATED TO SOLUTIONS

continuous from [0, T]\ £ into R 3 . The function t •-> w(£(f), t) is measurable. If
we show that this function is integrable, then (8.1) will make sense in the
distribution sense and (8.2) obviously makes sense. But,

and according to Theorem 4.3, t ^ |w(0|ii_-(n) is

The main result can be stated:
THEOREM 8.1. Let £1 be an open bounded set of !R3 which satisfies (1.7) with
r = 2, and let u be a weak solution of the Navier-Stokes equations corresponding
to the data u0, f, which satisfy (8.3).
Then, for every a eft, the equations (8.1), (8.2) possess (at least] one
continuous solution from [0, T] into H. Moreover, we can choose this solution in
such a way that the mapping 4>:{a, f}|-»£a(0, belongs to L°°((lx(0, T))3 1 and
d<f>/dteLl({lx(Q,T))3.

8.2. Proof of Theorem 8.1. i) Let wj} / e N, denote the eigenfunctions of the
operator A (cf. (2.17)) and, as in (3.44) we denote by Pm the orthogonal
projector in H on the space spanned by w l 5 . . . , vvm. The function u being
continuous from [0, T] into V with u'eL4/3(0, T; V), um is, in particular,
continuous from [0, T] into D(A), with iCeL4/3(0, T; D(A)).
Due to Agmon's inequality (2.21) (cf. §2.5), there exists a constant c{
depending only on f l , and such that

Pm being a projector in V and D(A),

for every fe[0, T]\£, where

0eL a (0, T) by Theorem 4.2 and Remark 4.1.

Now as m -^ oo

Indeed the convergence in L2(0, T; V) is a straightforward consequence of the

properties of Pw and the Lebesgue dominated convergence theorem. The
convergence in LJ(0, T;L°°(ft)) is proved in a similar manner, using the
following consequence of (8.5):

1
and even
 LAGRANGIAN REPRESENTATION OF THE FLOW 59

and by Holder's inequality2

ii) For every m, we define in a trivial manner an approximation £m of £,

where |m is solution of (6.1), (6.2) with u replaced by um. The usual theorem
on the Cauchy problem for ordinary differential equations (O.D.E.'s) asserts
that J;m is defined on some interval of time [0, T(m, a)], 0< T(m, a)^= T. But
um ifi—>IR 3 vanishes on dft, and if we extend M™ by 0 outside n, we see that a
trajectory £ m (t) starting at time t = 0 from a en (condition (8.2)) cannot leave
n at_a time <T. Hence T(m,A) = T, and £ma(t) = £ m (f)eft for all fe[0, T],
(a en).
The theorem on the continuous dependence on a parameter of the solution
of an O.D.E., shows that the function <I>m : ft x [0, T] -> O defined by

is in ^(nxfO, T])3. Since d i v u m = 0 , the Jacobian of 3>m,

det (D4> m (-, t)IDd) is equal to one for every f and therefore 4> TO (-, r): n >-»• n is
locally invertible for every r. Furthermore, if aedfl, | m (f) — a is a trivial
solution of (8.1), (8.2) (with u replaced by um), and hence <I>m(a, i) — a for all
a e 80 and for all t. The classical theorems on the global invertibility of <# 1
mappings (cf. F. Browder [1]) imply that, for every f, 4> m (-, f) is a C S 1
diflfeomorphism from n onto itself. Finally, as div um = 0, $ m (-, t) preserves the
areas.
iii) We now pass to the limit m —» o°, /or a fixed a G n. Due to (8.6),

and since 0 e L^O, T), the functions ^m = ^ma are equicontinuous. Thus, there
exist a subsequence m' (depending on a) and a continuous function £ = £a from
[0, T] into n, such that £„•-*•£, as m'-*o°, uniformly on [0, T]. We will
conclude that ^ is a solution of (8.1), (8.2), provided we establish that

But, for

2
Since the smooth functions are not dense in L°°, and L2/3(0, T;D(A)) is not a normed space,
there is not much flexibility in the construction of a sequence of smooth functions um approximat-
ing u in the norm of L^O, T;L°°(ft)).
60 PART I. QUESTIONS RELATED TO SOLUTIONS

Because of (8.8), (8.9), it remains, to prove (8.11), to show that

For every t££, u(-,t)eD(A) is a continuous function on H; hence

u
(£m'(0, 0—*• u(£(t), 0, and iw(|m'(0, 0| = |w(0|i_~(n), and Lebesgue's theorem
gives us the result (8.12).
iv) We infer from iii) that, for every fixed a e H, there exists a solution £ = £a
(nof necessarily unique) of (8.1), (8.2). We denote by A(a) the set of continuous
functions from [0, T] into O which satisfy (8.1), (8.2). According to (8.1), (8.5),
a function £ in A(a) belongs to the Sobolev space W M (0, T) = W ia (0, T)3, and
|£'(0| = 0(0 almost everywhere (for all t<£ &). We consider now the set-valued
mapping

where Y is the subset of W M (0, T) of functions 77 taking their values in H and

such that |V(0| = 0(0 almost everywhere; Y is closed in W1'1^, T): it is a
complete metric space. For every a eH, A(a) is not empty because of part iii)
of the proof. As for (8.12), one can show easily that A(a) is closed in Y an
also that the graph of A is closed.
By the von Neumann measurable selection theorem (recalled in the Appen-
dix to this section), A admits a measurable section L, i.e., a measurable
mapping L : H •-> Y, with L(a) e A(a) for all a e H. Thus £a = L(a) is a solution
of (6.1), (6.2), and | a (f)en for all re[0, T]. We set 0>(a, 0 = 4(0, and clearly,
<i> possesses the desired properties.
Remark 8.1. It would be interesting to prove further properties of <I>: for
instance that 4> preserves the areas, that <£(a, •) is unique at least for almost
every a e H, etc.

8.3. Appendix. For the convenience of the reader, we recall here the
measurable selection theorem (cf. C. Castaing [1]):
THEOREM 8.A. Let X and Y be two separable Banach spaces and A a
multiple-valued mapping from X to the set of nonempty closed subsets of Y, the
graph of A being closed.
Then A admits a universally Radon measurable section, i.e., there exists a
mapping L from X into Y, such that

and L is measurable for any Radon measure defined on the Borel sets of X.
PART II

Questions Related to Stationary Solutions and

Functional Invariant Sets (Attractors)

Orientation. The question studied in Part II pertains to the study of the

behavior of the solutions of the Navier-Stokes equations when t —> Q°, and to
the understanding of turbulence. Of course, before raising such questions, we
have to be sure that a well-defined solution exists for arbitrarily large time;
since this question is not solved if n = 3, we will often restrict ourselves to the
case n = 2 or, for n - 3, we will make the (maybe restrictive) assumption that a
strong solution exists for all time for the problem considered.
In § 9 we start by describing the classical Couette-Taylor experiment of
hydrodynamics, in which there has recently been a resurgence of interest. We
present here our motivations for the problems studied in Part II. In § 10 we
study the time independent solutions of the Navier-Stokes equations. In § 11,
which is somewhat technical, we present a squeezing property of the trajec-
tories of the flow (in the function space) which seems important. In § 12 we
show that, if the dimension of space is n = 2, then every trajectory converges
(in the function space) to a bounded functional invariant set (which may or may
not be an attracting set), and that the HausdorfT dimension of any bounded
functional invariant set is finite. This supports the physical idea that, in a
turbulent flow, all but a finite number of modes are damped.
Throughout §§10 to 12 we are considering the flow in a bounded domain (11
or Q) with indifferently either 0 or periodic boundary conditions. We will not
recall this point in the statements of the theorems.

61
This page intentionally left blank
U The Couette-Taylor Experiment

We recall what the Couette flow between rotating cylinders is: the fluid fills a
domain H of R3 which is limited by two vertical axisymmetric cylinders with the
same axis (radii r l5 r2, 0<r 1 <r 2 ) and by two horizontal planes separated by a
distance h. The outer cylinder is at rest and the inner one is rotating with an
angular velocity o^. This corresponds to the flow in a fixed bounded Lipschitz
domain n of !R3 with vanishing forces per unit volume1 and nonzero boundary
conditions on F = dfl: <£ / 0 in (1.9) (but <£ • v = 0). This case was not studied in
Part I to avoid purely technical difficulties, but its mathematical treatment is
very similar to that considered in Part I (cf., e.g., C. Foias-R. Temam [3], [4],
[5]; J. C. Saut-R. Temam [2]). When o>1(0 is constant, o>1(0 = w1, a Reynolds
number can be defined for this flow (cf. (1.6)), Re = d31r1d/v, d - r2 — rl, which
corresponds for instance to the choice of L% = d as a characteristic length for
the flow, and to U* = ailrl = the velocity on the inner cylinder as a characteris-
tic velocity.
The Taylor experiment on the Couette flow is as follows. We start with the
fluid at rest and we increase the angular velocity of the inner cylinder from 0 to
some value o^:
i) If MI (more precisely Re, which has no dimension) is sufficiently small,
then after a very short transient period we observe a steady axisymmetric flow.
The trajectories of the particles of fluid are essentially circles with the same
axis as the cylinders.
ii) If oi^Re) is larger than some threshold A', but not too large, then, after
the transient period, another steady state appears, different from that in i). A
cellular mode now appears: the trajectory of the particles of fluid is now the
superposition of the motion around the axis and a roll-like motion in the
azimuthal plane. The steady flow (see Fig. 9.1) remains axisymmetric, and in
adjacent cells the fluid particles move in counterrotating spiral paths.
With changing (i.e., increasing) aj1, a cellular structure may lose its stability
and another cellular steady motion may appear, with more cells.
iii) When w^Re) is larger than a higher threshold value, the flow observed
after the transient period becomes unsteady with its cellular structure per-
turbed by circumferential travelling waves. At this point the flow is periodic in
time.
As Re increases, the temporal variations of the flow become more complex,
quasiperiodic perhaps, with two or more incommensurate frequencies. Finally
the flow becomes totally turbulent with no apparent structure at all.
This is a somewhat simplified description of the phenomena, the flow
observed depending in a sensitive way on the ratio d/l and/or on how the

1
The gravity forces are of the form grad (—pgx 3 ), where Ox3 is a vertical axis pointing upward,
and we can incorporate them in the pressure term.
63
64 PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

constant angular velocity a)l is attained; different situations may also appear if
the flow does not start from rest. We refer the reader to T. B. Benjamin [1], T.
B. Benjamin-T. Mullin [1], [2], P. R. Fenstermacher-H. L. Swinney-J. P.
Gollub [1], H. L. Swinney [1], J. P. Gollub [1]. Let us also mention the
remarkable fact reported in T. B. Benjamin [1] that different stable steady
motions have been observed for the same geometry and the same boundary
velocity (same values of v, r l5 r2, h, Wj).
Although the above description of the experiment is over-simplified, it is
quite typical, and similar phenomena are observed for other experiments, such
as Benard flow and flow past a sphere. Let us interpret, for the flows studied in
Part I, what the corresponding mathematical problems are. We start from rest
(u 0 = 0), and we can imagine that f(t) is "increased" from 0 to some value / e H
through some complicated path, or through a linear one:
A ( O e K , increasing from 0 to A^ The parameter2

can play the role of the Reynolds number, L being a characteristic length of
the domain (H or Q), n the dimension of space and |/| the H-norm of feH.
Then the conjectures on bifurcation and onset of turbulence are as follows (cf.
D. Ruelle [1], D. Ruelle-F. Takens [1]):
a) If R is sufficiently small, the solution u(t) of Problem 2.2 tends, for t —»<*> 5
to u a time independent solution of the Navier-Stokes equations.
b) For larger values of R, u may lose its stability and u(i) converges for

2
R has no dimension if p = 1; if p^ 1, the number without dimension is
M, = V.
 THE COUETTE-TAYLOR EXPERIMENT 65

t—*cc to another stationary solution u'; this can be repeated for higher values
of R, u ( t ) - » u " , . . . .
c) After a higher threshold, u(f) converges, for f—»Q°, to a time periodic
solution of the Navier-Stokes equations u n (f), or to a quasiperiodic solution.
d) For higher values of R, u(t) as r —»<*> tends to lie on a "strange attractor,"
such as the product of a Cantor set with an interval. This would explain the
chaotic behavior of turbulence.
The small contributions presented below to this list of outstanding problems are:
• The proof of a), which is elementary and has been known for a long time;
• Some properties of the set of stationary solutions of the Navier-Stokes
equations in connection with b) (§ 10);
• Existence and a property of the limit set if n = 2 in the case d) (§ 12).
This page intentionally left blank
10 Stationary Solutions of the
Navier-Stokes Equations

Assuming that the forces are independent of time, we are looking for time
independent solutions of the Navier-Stokes equations, i.e., a function u = u(x)
(and a function p = p(x)) which satisfies (1.4), (1.5), and either (1.9) with (/> = 0,
or (1.10):

or

In the functional setting of § 2, the problem is

Given / in H (or V), to find u <= V which satisfies

or

10.1. Behavior for r —»°°. The trivial case. We start by recalling briefly the
results of existence and uniqueness of solutions for (10.5)-(10.7).
THEOREM 10.1. We consider the flow in a bounded domain with periodic or
zero boundary conditions (<9 = O or Q), and n = 2 or 3. Then:
i) For every f given in V and v>0, there exists at least one solution of
(10.5)-(10.7).
ii) /// belongs to H, all the solutions belong to D(A).
iii) Finally, if

where cl is a constant depending only on Gl, then the solution of (10.5)-(10.7) is

unique.
Proof. We give only the principle of the proof and refer the reader to the
literature for further details.
For existence we implement a Galerkin method (cf. (3.41)-(3.45)) and look,

l
cl is the constant in (2.30) when m 1 = m 3 = 1, m 2 = 0. Since |/| = >/Aj|/||v. if f e H , a sufficient
condition for (10.8) is v2>cl-J\~l\f\, which can be compared to (9.1).
67
68 PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

for every m e N , for an approximate solution um,

such that

for every v in Wm = the space spanned by w 1 ? . . ., w m . Equation (10.10) is also

equivalent to

The existence of a solution um of (10.10)-(10.11) follows from the Brouwer

fixed point theorem (cf. [RT, Chap. II, § 1] for the details). Taking v = um in
(10.10) and taking into account (2.34), we get

and therefore

We extract from um a sequence u m <, which converges weakly in V to some

limit u, and since the injection of V in H is compact, this convergence holds
also in the norm of H:
u m - —» u weakly in V, strongly in H.
Passing to the limit in (10.10) with the sequence m'. we find that u is a
solution of (10.6).
To prove ii), we note that if u e V, then Bu e V-i/2 (instead of V) because of
(2.36) (applied with ml = 1, m 2 = 0, m3 = ^). Hence u = v~lA~l(f-Bu) is in
V3/2. Applying again Lemma 2.1 (with ml =|, m2 = 2> m3 = 0), we conclude
now that BueH, and thus u is in D(A).
We can provide useful a priori estimates for the norm of u in V and in
D(A). Setting v = u in (10.6) we obtain (compare to (10.12)-(10.13))

For the norm in D(A) we infer from (10.7) that

(by the first inequality (2.32))

(by the Schwarz inequality)

(with (10.6)).
 STATIONARY SOLUTIONS OF THE NAVIER-STOKES EQUATIONS 69

Finally

For the uniqueness result iii), let us assume that u 1 and u2 are two solutions
of (10.6):

Setting v = u}-u2, we obtain by subtracting the second relation from the first
one:

and U j - u2 = 0 if (10.8) holds.

Concerning the behavior for t —> °c of the solutions of the time-dependent
Navier-Stokes equations, the easy case (which corresponds to point a) in § 9) is
the following:
THEOREM 10.2. We consider the flow in a bounded domain with periodic or
zero boundary conditions (0 = fl or Q) and n = 2 or 3. We are giuen f e H, v>0
and we assume that

where c 2 , c2 depend only on O.

Then the solution of (10.7) (denoted ux] is unique. If u(-) is any weak
solution2 of Problem 2.1 with u () e H arbitrary and f(t} = f for all t, then

Proof. Let w ( f ) = u(t) — u^. We have, by differences,

2
If n = 2, u(0 is unique and is a strong solution (at least for t >0). If n = 3, u ( - ) is not necessarily
unique, and we must assume that u ( - ) satisfies the energy inequality (see Remark 3.2).
70 PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

and, taking the scalar product with w(t),

Hence, with (2.34),

Using (2.30) with mv = 5, m 2 = 1, m3 = 0, we can majorize the right-hand side of

this equality by

which, because of (2.20), is less than or equal to

with (3.10) this is bounded by

where c^ = 3(cl/2)4/3. Therefore,

If

then (10.22) shows that |w(r)| decays exponentially towards 0 when *—><»:

w(0) = u 0 -u 00 . Using the estimation (10.17) for w^, we obtain a sufficient

condition for (10.23), which is exactly (10.18).
If we replace u(t] by another stationary solution u* of (10.7) in the
computations leading to (10.22), we obtain instead of (10.22)

Thus (10-23) and (10.18) ensure that u* = u^; i.e., they are sufficient conditions
for uniqueness of a stationary solution, like (10.8).
The proof is complete.

3
The proof that <w'(t), w(t)) = \(dldt) |w(t)| 2 is not totally easy when n = 2, and relies on [RT,
Chap. Ill, Lemma 1.2]. If n = 3 we do not even have an equality in (10.20), but an inequality ^,
which is sufficient for our purposes: a technically similar situation arises in [RT, Chap. Ill, § 3.6].
 STATIONARY SOLUTIONS OF THE NAVIER-STOKES EQUATIONS 71

Remark 10.1. Under the assumptions of Theorem 10.2, consider the linear
operator si from D(A) into H defined by

whose adjoint si* from D(A) into H is given by

Then with the same notation as in the proof of Theorem 10.2, we have

and (10.20) is the same as

or

The operator ((si + si*}l1) l is self adjoint and compact from H into itself, and
the conclusions of Theorem 10.2 will still hold if we replace (10.18) by the
conditions that the eigenvalues of (si+si*}l2 are >0.

10.2. An abstract theorem on stationary solutions. In this section we derive

an abstract theorem on the structure of the set of solutions of a general
equation

this theorem will then be applied to (10.7).

Nonlinear Fredholm operators. If X and Y are two real Banach spaces, a
linear continuous operator L from X into Y is called a Fredholm operator if
i) dim ker L<°°,
ii) range L is closed,
iii) coker L= Y/ rangeL has finite dimension.
In such a case the index of L is the integer

For instance, if L = Ll + L2 where L^ is compact from X into Y and L2 is an

isomorphism (resp. is surjective and dim ker L2 = q), then L is Fredholm of
index 0 (resp. of index q). For the properties of Fredholm operators, see for
instance R. Palais [1], S. Smale [1].
Now let to be a connected open set of X, and N a nonlinear operator from co
into Y; N is a nonlinear Fredholm map if N is of class (€} and its differential
N'(u) is a Fredholm operator from X into Y, at every point u eco. In this case
it follows from the properties of Fredholm operators that the index of N'(u) is
independent of w ; we define the index of N as the number i ( N ' ( u ) ) .
72 PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

Let N be a <# l mapping from an open set co of X into Y, X, Y being again

two real Banach spaces. We recall that u e X is called a regular point of N if
N'(M) is onto, and a singular point of N otherwise. The image given by N of
the set of singular points of N constitutes the set of singular values of N. Its
complement in Y constitutes the set of regular values of N. Thus a regular
value of N is a point /e Y which does not belong to the image N(o>), or such
that N'(u) is onto at every point u in the preimage N~l(f).
Finally we recall that a mapping N of the preceding type is proper if the
preimage N~l(K] of any compact set K of Y is compact in X.
We will make use of the following infinite dimensional version of Sard's
theorem due to S. Smale [1] (see also K. Geba [1]).
THEOREM. Let X and Y be two real Banach spaces and o> a connected open set
of X. If N: a) -> Y is a proper ^ k Fredholm map with k > max (index N, 0), then
the set of regular values of N is a dense open set of Y.
We deduce easily from this theorem:
THEOREM 10.3. Let X and Y be two real Banach spaces and o) a connected
open set of X, and let N: to —> Y be a proper <# fc Fredholm map, k =s 1, of index
0.
Then there exists a dense open set o}l in Y and, for every fea)^ N~l(f) is a
finite set.
If index N = q>0 and k^q, then there exists a dense open set a)l in Y and,
for every /e a)l, N~l(f) is empty or is a manifold in a) of class (&k and dimension
q.
Proof. We just take u>l = the set of regular values of N which is dense and
open by Smale's theorem. For every fea)1, the set N~l(f] is compact since N is
proper. If index (N) = 0, then for every f e a>l and u 6 N~l(f], N'(u) is onto (by
definition of oj^ and is one-to-one since
dim ker N'(u) = dim coker N'(u) = 0.
Thus N'(u) is an isomorphism, and by the implicit function theorem, u is an
isolated solution of N(v) =/. We conclude that N~l(f) is compact and made of
isolated points: this set is discrete.
If index (N) = q^k, for every /ea> 1 , and every u e N ~ l ( f ] , N'(u) is onto and
the dimension of its kernel is q: it follows that N-1(/) is a manifold of
dimension q, of class (€k like N.
Applications of this theorem to the stationary Navier-Stokes equation (and
to other equations) will be given below.

10.3. Application to the Navier-Stokes equations. We are going to show

that Theorem 10.3 applies to the stationary Navier-Stokes equation (10.7) in
the following manner:

It follows from (2.36) that B(-, •) is continuous from D(A)xD(A) and even
V3/2x V3/2 into H and N makes sense as a mapping from D(A) into H.
 STATIONARY SOLUTIONS OF THE NAVIER-STOKES EQUATIONS 73

LEMMA 10.1. N:D(A)-^-H is proper.

Proof. Let K denote a compact set of H. Since K is bounded in H, it follows
from the a priori estimation (10.17) that N~\K) is bounded in D(A) and thus
compact in V3/2. As observed before, B(-, •) is continuous from V 3/2 x V3/2 into
H, and thus B(N~\K)) is compact in H.
We conclude that the set N~ 1 (K) is included in

which is relatively compact in D(A), and the result follows.

We have the following generic properties of the set of stationary solutions to
the Navier-Stokes equations.
THEOREM 10.4. We consider the stationary Navier-Stokes equations in a
bounded domain (ft or Q) with periodic or zero boundary conditions, and n = 2
or 3.
Then, for every v > 0, there exists a dense open set €V^H such that for every
ftOv, the set of solution of (10.5)-(10.7) is finite and odd in number.
On every connected component of Ov, the number of solutions is constant, and
each solution is a ^x function of f.
Proof, i) We apply Theorem 10.3 with the choice of X, Y, N indicated in
(10.26) (and w = X). It is clear that N is a ^ mapping from D(A) into H and
that

It follows from (3.20) that for every u e D ( A ) , the linear mappings

are continuous from V into H and they are therefore compact from D(A) into
H. Since A is an isomorphism from D(A) onto H, it follows from the
properties of Fredholm operators (recalled in § 10.2) that N ' ( u ] is a Fredholm
operator of index 0.
We have shown in Lemma 10.1 that N is proper: all the assumptions of
Theorem 10.3 are satisfied. Setting Ov = the set of regular values of N = N», we
conclude that Ov is open and dense in H and, for every f £ @ v , N H/), which is
the set of solutions of (10.7), is finite.
ii) Let (<D'i)j e / be the connected components of Ot, (which are open), and let
/0, /) be two points of (?, for some i. Let u ( ) e N ~'(/ 0 ). There exists a continuous
curve

and the implicit function theorem shows the existence and uniqueness of a
continuous curve s^>u(s) with

Since f(s) is a regular value of N, for all s e[0, 1], u(s) is defined for every
s, O ^ s S i l , and therefore u(l}eN '(/i). Such a curve {s>—»w(s)} can be con-
74 PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

structed, starting from any ukeN~l(f0). Two different curves cannot reach the
same point u^eN" 1 ^) and cannot intersect at all, since this would not be
consistent with the implicit function theorem around U* or around the intersec-
tion point. Hence there are at least as many points in N"1^) as in N~l(f0). By
symmetry the number of points is the same.
It is clear that each solution uk = uk (f) is a ^ °° function of / on every O.
iii) It remains to show that the number of solutions is odd. This is an easy
application of the Leray-Schauder degree theory.
For fixed v>Q and feCL, we rewrite (1.7) in the form

Therefore the Leray-Schauder degree d(Tv, Ag, BR] is well defined, with BR
the ball of D(A) of radius JR. Also, when Ag is a regular value of Tv, i.e., A/ is a
regular value of N, the set T~l(\g) is discrete = {u^..., uk}, and
d(Tv, Ag, BR) = Xf=i i(Uj) where 1(11,-) = index Uj.
It follows from Theorem 10.1 that there exists A^efO, 1], and for O ^ A ^
A*, N~l(\f) contains only one point ux. By arguments similar to that used in
the proof of Theorem 10.1, one can show that N"'(w x ) is an isomorphism, and
hence for these values of A, d(Tv, Ag, BR) = ±1. By the homotopy invariance
property of degree, d(Tv, g, BR) = ±1 and consequently k must be an odd
number.
The proof of Theorem 10.4 is complete.
Remark 10.2.
i) The set G is actually unbounded in H; cf. C. Foias-R. Temam [13].
ii) Similar generic results have been proved for the flow in a bounded
domain with a nonhomogeneous boundary condition (i.e., c/>7^0 in (1.9)):
generic finiteness with respect to / for </> fixed, with respect to </> for / fixed and
with respect to the pair /, </>; see C. Foias-R. Temam [3], [4], J. C. Saut-R.
Temam [2], and for the case of time periodic solutions, J. C. Saut-R. Temam
[2], R. Temam [7].
iii) When H is unbounded, we lack a compactness theorem for the Sobolev
spaces H m (H) (and lack the Fredholm property). We do not know whether
results similar to that in Theorem 10.4 are valid in that case; cf. D. Serre [1]
where a line of stationary solutions of Navier-Stokes equations is constructed
for an unbounded domain H.
We now present another application of Theorem 10.3, with an operator of
index 1, leading to a generic result in bifurcation. We denote by
the set of solutions of (1.5)-(1.7) and
 STATIONARY SOLUTIONS OF THE NAVIER-STOKES EQUATIONS 75

FIG. 10.1

THEOREM 10.5. Under the same hypotheses as in Theorem 10.4, there exists a
Gs-dense set 0<^H, such that for every feO, the set S(f) defined in (10.29) is a
*% °° manifold of dimension 1.
Proof. We apply Theorem 10.3 with X = D ( A ) x R , Y = H , <o = <om =
D(A)x(l/m,t»)cX, m e N , N ( u , v) = vAu + Bu, for all (u, v)eX. It is clear that
N is ^ from w into Y and

For is the sum of the operator

which is compact4, and the operator

which is onto and has a kernel of dimension 1. Thus JV'(w, v) is a Fredholm

operator of index 1 and N is a nonlinear Fredholm mapping of index 1.
The proof of Lemma 10.1 and (10.17) shows that N is proper on D(A)x
(i>0,°°), for all v 0 >0 and in particular v0=\/m. Hence Theorem 10.3 shows
that there exists an open dense set <9 m <=H, and for every f e O m , N m l ( f ) is a
manifold of dimension 1, where Nm is the restriction of N to wm. We set
®- Hmisi ^m> which is a dense G8 set in H, and, for every f e O , S ( f ) =
Umsi Nml(f] is a manifold of dimension 1.
Remark 10.3. i) By the uniqueness result in Theorem 10.1, S(f) contains an
infinite branch corresponding to the large values of v.
ii) Since S(f) is a ^ manifold of dimension 1, it is made of the union of
curves which cannot intersect. Hence the usual bifurcation picture (Fig. 10.1) is
nongeneric and is a schematization (perfectly legitimate of course!) of generic
situations of the type shown in Fig. 10.2.
Remark 10.4. Other properties of the set S(f, v} are given in C. Foias-R.
Temam [3], [4] and J. C. Saut-R. Temam [2]. In particular, for every
v, f, S(f, v} is a real compact analytic set of finite dimension.

4
Same proof essentially as in Theorem 10.4.
76 PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

10.4. Counterexamples. A natural question concerning Theorem 10.4 is

whether Cv is the whole space H or just a subset. Since Theorem 10.4 is a
straightforward consequence of Theorem 10.3, this question has to be raised at
the level of Theorem 10.3. Unfortunately, the following examples show that w
cannot answer this question at the level of generality of the abstract Theorem
10.3, since for one of the two examples presented Ov = H, while for the second
one Gv + H.
Example 1. The first example is the one-dimensional Burgers equation
which has been sometimes considered in the past as a model for the Navier-
Stokes equations: consider a given v > 0 and a function /:[0, 1]—>1R which
satisfies

For the functional setting we take H = L"(0, 1), V = Ho(0, 1), D(A) =
H10(0, 1)DH 2 (0, 1), Au=-d2u/dx2, for all u e D ( A ) ,

Then, given f in H (or V), the problem is to find ueD(A) (or V) which
satisfies

We can apply Theorem 10.3 with X = D(A), Y = H, N(u) = vAu +Bu. The
mapping N is obviously c£°° and

N'(u), as the sum of an isomorphism and a compact operator, is a Fredholm

operator of index 0. Now we claim that every u e D(A) is a regular point, so
that w t = H.
In order to prove that u is a regular point, we have to show that the kernel
of N'(u) is 0 (which is equivalent to proving that N'(u) is onto, as index
N'(u) = Q). Let v belong to N'(w); v satisfies

By integration — w' + uv = constant = a, and by a second integration taking into

account the boundary conditions we find that v = 0.
Since the solution is unique if / = 0 (u =0), we conclude that
(10.33) (10.30)-(10.31) possesses a unique solution V v, V/.
Example 2. The second example is due to Gh. Minea [1] and corresponds to
a space H of finite dimension; actually H = !R3.
 STATIONARY SOLUTIONS OF THE NAVIER-STOKES EQUATIONS 77

Theorem 10.3 is applied with X =-- Y = [R3, N ( u ) = vAu + Bu for all w =

(u l 5 u2, u 3 )eR 3 . The linear operator A is just the identity, and the nonlinear
(quadratic) operator B is defined by Bw = (^(u^ + uf), —duiU2, —8uiU3) and
possesses the orthogonality property Bu • u =0. The equation N(u) = f reads

It is elementary to solve this equation explicitly. There are one or three

solutions if |/2| + \fi\ / 0. In the "nongeneric case", /2 = /3 = 0, we get either one
solution, or one solution and a whole circle of solutions: <o, ^ Y in this case.
This page intentionally left blank
J^ J^ The Squeezing Property

In this section we establish a squeezing property of the semigroup of

operators associated with the time-dependent Navier-Stokes equations. This
property, which will be useful for the next section, is also interesting by itself
and has found other applications (see in particular Foias-Temam [7]): the
squeezing property shows that, up to some error which can be made arbitrarily
small, the flow is essentially characterized by a finite number of parameters.

11.1. An a priori estimate on strong solutions. The following a priori

estimate is valid for all time; it is a particular case of the more general results
stated in § 12.3 (Lemma 12.2); we refer the reader to that section for more
details.
LEMMA 11.1. Assume that u 0 eH, and f satisfies
f is continuous and bounded from [0, <») into H,
f is continuous and bounded from [0, oo) into V,
and let u be the strong solution to the Navier-Stokes equations given by Theorem
3.2, defined on [0, oo) i f n = 2, on [0, T^UQ)] if n = 3.
Then u is bounded in V on [0, oo) if n=2, on [0, T\(u0)] if n — 3, and for every
a >0, Au is bounded on [a, oo) if n = 2, on [a, T\(u0)] if n = 3,

where c' depends on ||u0||, /, v, fl (or Q), and c" depends on the same data and
furthermore a.
Remark 11.1. i) If n = 3, (11.2) follows from (3.28)-(3.29), but (11.2) must
be proved for large time if n = 2.
ii) If u is defined on some interval (TO, T]) and supTo<,STi||u(t)|| = ci<oo then,
by inspection of the proof of (11.3), we can see that, for every <*>0,
supT +«<tsT |Au(0| = c", where c" depends on c[ and the data u0, /, v, H (or
Q).

11.2. The squeezing property. We assume that n = 2 or 3. Let u0 and t>0 be

given in V, ||u0||^K, ||u0|| = -R, and let / be given satisfying
/ is continuous and bounded from [0, oo) into H,
/' is continuous and bounded from [0, oo) into V.

1
0<a^T 1 (u 0 ) if n = 3, and in this case the supremum is for O ^ t g T ^ U o ) in (11.2) and for
a ^ t ^ T j ( u 0 ) in (11.3).
79
80 PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

We denote by u and v the strong solutions of Navier-Stokes equations

corresponding respectively to (w0, /) and to (u0, /), which are given by Theorem
3.2, and are defined on (0, <») if n = 2 and on [0, T^CR)] if n = 3, where

(see (3.29)).
We recall that vvm, Am denote the eigenfunctions and eigenvalues of A and
that Fm is the projector in H (or V, V, D(A)) on the space Wm spanned by
wlt. .., wm.
THEOREM 11.1. Under the above hypotheses, for every a >0, there exist two
constants c5, c6, which depend on a, R, /, v, H (or Q) and such that, for every
t^a, for every m sufficiently large, i.e. satisfying

we have either

or

Proof. We assume in this proof that

i) We set

We have, by taking the difference of the equations (2.43) for u and v,

Taking the scalar product, in H, of (11.10) with Pmw(t], we obtain

We apply Lemma 2.1 (i.e. (2.29)) respectively with m l 5 m2, m3 = 2, 0, 0; |, 1, |;

0, 1, 1, and we use (11.3) for u and v, with a replaced by a/2:
 THE SQUEEZING PROPERTY 81

With these lower bounds, (11.11) becomes

Since A L ^ A m + 1 and

we have

In a similar manner, taking the scalar product in H of (11.10) with

we obtain

But

and thus

We set A = Am + 1, p = max (c{\\"1/4, c2, c^ c^AT 1 ' 4 )- Thus (11.14HH.16) take

the form:

and if

then for almost every t2

ii) We now want to prove the alternative (11.7)-(11.8). Let us consider a

2
The first relation of (11.19) follows easily from the first relation of (11.17) at a point (n where
p(t ( ) )>0. If p(t () ) = 0 then either p is not differentiable at t0 (which can only occur on a set of points
t () of measure 0) or dp(f ( ) )/dt§0. The differentiability almost everywhere of p follows trivially from
the differentiability almost everywhere in V of u and v, and thus w = u — v.
82 PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

specific point t0, r0 = « if n = 2, a ^ t o ^ T ^ R ) if n = 3. If q(f 0 ) = p(fo) then

(11.7) is satisfied at t0 and we have nothing to prove. Therefore we assume that

and on the other hand we assume that m is sufficiently large so that

This implies that

in a neighborhood of f 0 -
At this point two possibilities can occur: either (11.22) is valid on the whole
interval [r0-a/2, t0], or (11.22) is valid for t e(f l 9 f0), with t^to-a/2 and

The second possibility is discussed in point iii) of the proof; in the first case,
since (11.18) and (11.19) are satisfied on [t0 — a/2, to], we have

and consequently

Due to (11.2),

and (11.8) follows at t0 with appropriate constants c5, c6.

iii) The last step consists in proving (11.7) at
assuming that (11.22) is satisfied on an interval (f 1} r0), and (11.23) is satisfied
at
We denote by (11.19)' the differential system obtained by replacing the
inequalities in (11.19) by equality signs. It is natural to formally associate with
(11.19)' the differential system

We prove ve by elementary calculations that remains constant when

are solutions of (11.19) and
 THE SQUEEZING PROPERTY 83

By elementary calculations we also check that

when p(-), q(-) are solutions of (11.19), in particular on [f l5 1 0 \. Thus

By

and on the other hand, with (11.20),

Since

and p(to)<q(t0), we obtain an estimate of the type (11.8) for

(p200) + q 2 (r 0 )) 1/2 =iV2q(r 0 ).
The proof is complete.
Remark 11.2. i) As in Remark 11.1 ii), the squeezing property is valid for
any pair of strong solutions u, v of the Navier-Stokes equations, defined on
some interval (TO, r a ), and satisfying

so that

ii) Of course we can replace the coefficient V2 in the right-hand side of

(11.6) by any constant (3 > 1. It is clear that in such case c5 and c6 depend also
on |3.
Remark 11.3. It is reasonable to consider as physically "undistinguishable"
two strong solutions u, u, which satisfy (11.8) for some m sufficiently large.
Mathematically, we can consider an equivalence relation among the strong
solutions which are defined on (TO, TJ) (and bounded in V on this interval): two
solutions u, u are said to be equivalent if there exist two constants c5, c6 such
that (11.6)-(11.8) hold on (TO, TJ). It is easy to check that this relation is indeed
an equivalence one. The squeezing property implies that Pmu characterizes the
equivalence class of u, i.e., characterizes u up to an error exp (-cA^+j).
Remark 11.4. Let us mention an alternate form of the squeezing property
which is useful for later purposes.
84 PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

Let M and v be two strong solutions of the Navier-Stokes equations, defined

and bounded in V on some finite interval r0 Si t ^ rl < <». For every a > 0, there
exist two constants c*, c* which depend on TO, r l5 a, R, f, v, ft3, such that, on
[TQ + CX, rj, we have either (11.7) or

for every m satisfying

The proof is exactly the same, except that in (11.25) and (11.27) we replace
the majorizations

by the estimate (S=C"|W(T O )|) which follows from the next lemma.
LEMMA 11.2. // u and v are two strong solutions defined and bounded in V on
a finite interval TQ^^T!, n = 2 or 3, then there exists a constant c" which
depends on TO, r l5 /, v, ft (or Q), and R, with

such that

Proof. It follows easily from the energy inequality (3.24), valid for strong
solutions, and from the assumptions on u and u, that Au and Av belong to
JL2(r0, T! ; H) with their norm in this space bounded by a constant depending on
TO, TI, f, v, fl (orQ) and R.
We take the scalar product in H of (11.10) with w(t), and using (2.29),
(2.33), (2.34) and (11.30) and Remark 11.1 ii), we get, as for (11.13):

and since |Au| 4/3 eL^O, T), (11.31) follows from Gronwall's lemma.

3
Here
12 Hausdorff Dimension of an Attractor
In this section we derive an important property of functional invariant sets and
attractors which are bounded in V, i.e., in the //'-norm. We show that these sets
have a finite Hausdorff dimension.
Definitions are given below but it is well known that attractors encompass the
long-time behavior of solutions of the Navier-Stokes equations. As explained
below, attractors are always bounded in the //'-norm in space dimension n = 2,
but it is not known whether this is true in space dimension n = 3.
The main result of this section was first proved in C. Foias-R. Temam [5] and
the proof presented below is based on this article. Although this result was sub-
sequently improved in different ways, the proof in this article relies mostly on the
Navier-Stokes equation techniques presented in this book, while the subsequent
proofs depend more heavily on dynamical systems techniques (cf. the comments
in the Comments and Bibliography section).
This result on the Navier-Stokes attractors shows that the long-time behavior
of the solutions of these equations \sfinite dimensional although these equations
have infinite dimension. As indicated before, this is true without restriction for
two-dimensional flows and, in space dimension three, this is true for flows which
remain smooth for all time. Other aspects of finite dimensionality of flows are
mentioned in the Comments and Bibliography section.

12.1. Functional invariant sets and attractors. Throughout this section, we

assume that f(t) = f is independent of t and belongs to H. Let S(t) be the
semigroup associated with the strong solution of the Navier-Stokes equation

i.e., the mapping w 0 = u(0)>-»w(f), which is defined on V for all ti^O if n = 2,

and for re[0, T^UQ)] (at least), if n = 3.
DEFINITION 12.1. A functional invariant set for the Navier-Stokes equations
is a subset X of V which satisfies the following properties:
i) For every u f) e X, S(t}uQ is defined for every f > 0 .
ii) S(t)X = X for all f > 0 .
We recall that an attractor (in V or H) is a functional invariant set X which
satisfies furthermore the condition:
iii) X possesses an open neighborhood w (in V or H), and for every
u () e o>, S(t)u 0 tends to X, in V or H, as t -»°°.
If n=2, we have the following:
LEMMA 12.1. Let us assume that n = 2, and that u0 and f are given in H. Then
there exists a functional invariant set X <= V, compact in H, such that the distance
in H of u(t) to X tends to 0 as t —> o°, where u(-) is the solution of Problem 2.1.
85
86» PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

Proof. We set

where the closures are taken in H. An element </> e H is in X if and only if

It is easy to see that X is also equal to

Since for t0>Q, the set

is bounded in V (cf. Lemma 11.1), so is its closure in V and so is X. By

Theorem 3.2, since n = 2, S(t)<f> is defined for all t >0, for every $ e X, and the
first condition i) in Definition 12.1 is satisfied. For condition ii) we observe that
S(OX<=X follows easily from the fact that S(t)u(s) = u(s + t), for all s, f ^ O .
In order to show that S(t)X = X, we note that S(t) is infective because of the
time analyticity property (see § 7) and therefore since the set (12.4) is relatively
compact in H(t0>0], the sequence u ( S j ) , j e N , converges in H whenever the
sequence S(t)u(Sj) converges. Indeed, if S(t)u(Sj) converges in H to some limit
<f>, as /—»oc (and s,-»o°), then by (12.2), </>eX. The sequence w(s,), being
bounded in V, is relatively compact in H and possesses cluster points in H; if iff
and i// are two cluster points, S(t)if/ = S(t)^' -- </> so that i// = i/>, the cluster
point is unique and the whole sequence M(S ; ) converges to it in H as j —*• °°.
Finally we show that the distance in H of u(t) to X tends to 0 as t -» oo. We
argue by contradiction. If this were not true, we could find e 0 >0, and a
sequence s, —»<», such that

Since «(«,) is bounded in V we can, by extracting a sequence, assume that u(s y )

converges in H to some limit if/. By (12.2), «/>eX, in contradiction with

The lemma is proved.

12.2. Hausdorff dimension of functional invariant sets.

THEOREM 12.1. Let n = 2 or 3 and let f e H. We assume that X is a functional
invariant set bounded in V. Then the Hausdorff dimension of X is finite and this
dimension is bounded by a number c, which depends on £1 (or Q), i/, |/| and1

Proof, i) We are going to make use of the squeezing property, in the

conditions indicated in Remark 11.4: we fix a > 0 and choose T0 = 0, T1 = 2a,
1
The bound given in the proof is, in fact, an increasing function of v \ |/| and R.
 HAUSDORFF DIMENSION OF AN ATTRACTOR 87

and te[a, 2a]2. We then choose m sufficiently large so that (11.29) is verified
and (see (11.28))

and the value (or one value) of T) which is appropriate below is

We set S(t) = S, then Theorem 11.1 and Remark 11.4 show us that for a
<£,(// e X, we have either

Given r > 0, since X is relatively compact in H, it can be covered by a finite

number of balls of H of radius <r, say Bl,. . . , BM. By definition, SX = X;
hence

and (12.7) implies that for all k = 1,. . . , M, and for all u, v e S(Bk DX) we
have either

Let <$> be an arbitrary point of S(Bk DX) and let Yk be the largest subset of
S(Bk OX), containing <£ and such that

of course Yk may be reduced to <£3.

ii) We now define a reiterated covering of X.
Let BE1!,. . . , B^ be a covering of PmS(Bk HX) by balls of PmH of radius
^r k = 5diam Bk. For those /'s such that PmYk HB^-^ 0, we choose arbitrarily
an ukj & Yk, such that P m w kj e PmYk nB£], and we denote by Bkj the ball of H
centered at uk. and of radius

Setting Bk] =0 if B£nP m Y k =0, we have

Indeed, if u e S(Bk OX), then there exists u'e Yk (u' = u if u e Y k ) such that,
by (12.9) and the definition of Yk,

2
Since, by the definition of a functional invariant set, S(t)<f> is defined for every (=SO, for every
<£eX we are free to choose TO, T L in Remark 11.4, O^T O ^T]<+°°.
3
If Y k ^ S ( B k n X ) then sup^^ n x ) x n dist («fc Y k )=ST] diam Bk.
88 PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

Then

Before proceeding to the next step, let us observe that we can derive an
upper bound for Mk.
By Lemma 11.2,

hence

Since PmS(Bk HX) is included in a ball of PmH of radius c" diam Bk, we have

where / m (<r) is, in IR m , the minimum number of balls of radius ^cr which is
necessary to cover a ball of radius 1, / m (or)^2~~ m/2 cr~ m .
iii) We have

Since the Bki, k = 1 , . . . , M, j — 1 , . . . , Mk, for all k, constitute a covering of

X by balls of radius ^er, we infer from the definition of ju, TS (X) in (5.2) that

This last inequality, valid for any covering of X by balls of radius ^r,
implies in its turn

where A = / m (l/4c")e\ By reiteration,

 HAUSDORFF DIMENSION OF AN ATTRACTOR 89

Since e < 1, if 7 is sufficiently large, then A < 1 and A ; —» 0 as / —» °o, so that

Therefore, provided

the Hausdorff dimension of X is ^y.

The proof is complete.
Remark 12.1. We do not know, even if n = 2, whether every functional
invariant set is bounded in V. But Lemma 12.2 shows us that this is the case
for those associated with the limit of u(t) as t—><x>.
The question of the boundedness of X in V is totally open if n = 3.

12.3. Other properties of functional invariant sets. We conclude this sec-

tion by indicating some regularity results and a priori estimates which are valid
for long times, and we infer from these estimates some properties of functional
invariant sets; in particular, if / is sufficiently regular and n = 2, a functional
invariant set is necessarily carried by the space of ^°° functions.
LEMMA 12.2. Let n = 2 or 3, and consider the time-dependent Navier-Stokes
equations with the data u0, / satisfying

is continuous and bounded from

is continuous and bounded from

where m is an integer
We assume that the solution u to Problem 2.2 satisfies

Then u is continuous and bounded from [T], °°) intoHm(O), d!u/dt' is continu-
ous and bounded from [TJ, <») into Hm~2'(€), j = 1, . . . , I, V TJ >0, and the cones-
ponding norms of u and d'u/dt' are bounded by constants which depend only on
the data, t0, TJ, and |u| L ~ ( , 0)00; v >-
We do not give the proof of these a priori estimates, which is rather long; we
refer the reader to C. Guillope [1].
We recall that, if n = 2, then (12.19) is automatically satisfied for every £ 0 >0
and the result holds for 17 >0 arbitrarily small. When n = 3, we do not know
whether (12.19) is true; however, the boundedness of ||u(t)|| for t —> ™ is closely
related to Leray's conjecture on turbulence.
90 PART II. QUESTIONS RELATED TO STATIONARY SOLUTIONS

Let us recall that Leray's hypothesis, and his motivation in introducing the
concept of weak solution was that singularities may develop spontaneously
over a finite interval. We know that \u(t)\ remains bounded even for weak
solutions but the enstrophy !|n(f)|| may perhaps become infinite. We formulate
Leray's assumption in a more precise way:
There exist 12^IR3, T> 0, i»0, u0e K,/e L^O, T; H) such that
(12.20) MOll becomes infinite at some time t e (0, T), u being the weak
solution of the Navier-Stokes equations.
Let us assume now that /e L°°(0, o°; H) and is not "too chaotic" at infinity, i.e.,
There exists a > 0 such that for any sequence t, —> +°°, the
(12.21) sequence of functions fj(t) = f(t + tj}, 0 < f <a possesses at least
one cluster point in L2(0, a; H).
This property is trivially satisfied if / is independent of time, or more generally
if

with

for all s >0, a >0, where c(f) is independent of s. In such a case we have the
following result proved elsewhere (cf. C. Foias-R. Temam [13]):
THEOREM 12.2. Given fteIR3, T, v and /eL°°(0,^;H) satisfying (12.21),
then either (12.19) or (12.20) holds; i.e., there exists u 0 e V for which \\u(t)\\
becomes infinite on the interval [0, a], a >0 given in (12.21).
Finally we infer from Lemma 12.2 the following regularity result for a func-
tional invariant set (or attractor):
THEOREM 12.3. Let us assume that f is independent of t, and f e c€°°(C)n DH5.
Then any functional invariant set X, bounded in V, for the corresponding Navier-
Stokes equations, is contained in £#°°(<j?)n.
Proof. We consider the semigroup operator S(t) defined at the beginning of
§ 12.1. Since S(t)X = X, for all f > 0 , and S(t) is one-to-one by the analyticity
property, every <£ eX is of the form $ = S(t)u0, with u 0 eX, and therefore by
Lemma 12.2, <£ is in H m (0), for all m.
Remark 12.2. i) This result is comparable to regularity results for the
solutions of stationary Navier-Stokes equations, but it applies also to periodic,
and quasi periodic solutions, among others.
ii) The conclusion of Theorem 12.2 is valid if we assume only that X is
bounded in V1/2+e for some e>0. See C. Guillope [1] for the details.

5 If we consider the flow in fl(0 = fl), we also need (1.7) with r = °o.
PART III

Questions Related to
the Numerical Approximation

Orientation. In Part III, containing §§ 13 and 14, we develop some results

related to the numerical approximation of the Navier-Stokes equations. This is
far from being an exhaustive treatment of a subject which is developing
rapidly, due to the needs of modern sophisticated technologies and the appear-
ance of more and more powerful computers.
Part III does not discuss any practical developments concerning implemen-
tation of computational fluid dynamics algorithms1 (though these are impor-
tant). It is limited to the presentation of two questions related to the numerical
analysis of the Navier-Stokes equations. The first is the question of stability
and convergence of a particular nonlinear scheme; this is developed in § 13.
The particular scheme was chosen almost arbitrarily among several schemes
presently used in large scale computations and for which convergence has been
proved. The second question, considered in § 14, is that of the approximation
of the Navier-Stokes equations for large time: in the context of the questions
studied in § 12 (see also § 9), the problem of numerical computation of
turbulent flows is connected with the computation of u(t) for t large (while the
force, or more generally the excitation, is independent of time). Section 14
contains some useful remarks pertaining to this question.
1
See the comments following § 14.

91
This page intentionally left blank
1 ^ Finite Time Approximation

In this section we study the convergence of a space and time discretization

scheme for the evolution Navier-Stokes equations. This scheme combines a
discretization in time by an alternating direction (or decomposition) method
with a discretization in space by finite elements. It belongs to a series of
efficient schemes which have been introduced and studied from a theoretical
point of view in the past (cf. A. J. Chorin [2], [3], [4], R. Temam [1], [2], [5],
[6]), and which have recently regained interest and have been implemented in
large scale computations for industrial applications (cf. e.g., H. O. Bristeau-R.
Glowinski-B. Mantel-J. Periaux-P. Perrier-O. Pironneau [1], R. Glowinski-B.
Mantel-J. Periaux [1]).
The proof of convergence is the same as that appearing in [RT, Chap. Ill,
§§5, 6, 7] and in R. Temam [1], [2], [4], [5] for closely related schemes.
However, for the proof of the strong convergence result which is necessary for
the passage to the limit, we use a compactness argument different and slightly
simpler than that considered in those references. This compactness result is
given in § 13.3.
13.1. An example of space-time discretization. We consider the flow in a
bounded domain with periodic or zero boundary condition (i.e., 0 =Cl or Q),
and n = 2 or 3. We are given u() and / as in Theorem 3.2 and we want to
approximate the strong solution u to Problem 2.2 defined in Theorem 3.2, on
[0, T] if n = 2, on [0, T*] if n = 3.
We are given a family of finite dimensional subspaces Vh, he^f, of V, such
that

For instance we can take ffl = N, V h =the space spanned by the eigenvectors
wi, . . . , wh, and the discretization would correspond to a Galerkin method
based on the spectral functions. Also Vh may be a more general subspace
corresponding to a more general Galerkin method, but the most interesting
case we have in mind is that in which Vh is a space of finite element functions
with the set ffl properly defined: we refer to [RT, Chap. I, § 4] for the precise
definition of a possible finite element space (Approximation APX4).
Actually, in order to be able to treat more finite element schemes we will
consider the following more general situation:
is a family of finite dimensional subspaces of
H^O) or HJ(O), such
he9e W k isthat
dense in W.
For every h, Vh is a subspace of Wh, such that the family Vh,
h 6 $?, constitutes an external approximation1 of V.
1
For the precise definition of an external approximation cf. [RT, Chap. I, § 3.1]. The reader who
so wishes may concentrate on the first, simpler case: V h c=V with (13.1).
93
94 PART III. QUESTIONS RELATED TO THE NUMERICAL APPROXIMATION

The cases covered by (13.2), (13.3) contain the finite element approxima-
tions of V called APX2, APX3, APX3' in [RT, Chap. I, §4] and others
published in the literature (cf. M. Bercovier-M. Engelman [1], V. Girault-P.
A. Raviart [1], among others).
For every h, let u0h be the projection of u0 in W, on Vh, i.e.,

Let N be an integer2, k - T'/N. For every h and k we recursively define a

family w™ +1/2 of elements of Vh, m = 0 , . . . , N-1, i = 1, 2. We start with

Assuming that u° (or more generally u™, m^O) is known, we define u™+l/2
and then u!T+1 as follows:

where

and

Here3 b(u, v,w) is the modification of the nonlinear term which was intro-
duced in R. Temam [1], [2] and which guarantees the existence of a solution to
(13.8), and the stability of the approximation for (13.8) or similar schemes:

The existence and uniqueness of a solution w™ +1/2 for (13.6) follow from the
Riesz representation theorem (or Lax-Milgram theorem). The existence of a
solution U ™ + I G Wh of (13.8) (a nonlinear finite dimensional problem) follows
from the Brouwer fixed point theorem, using [RT, Chap. II, (2.34) and Lemma
1.4]. Problem (13.6) is just a variant of the linear stationary Stokes problem,
and (13.8) is a relatively standard nonlinear Dirichlet problem. We refer to
Glowinski, Mantel and Periaux [1] and to F. Thomasset [1], for the practical

2
With the notation of Theorem 3.2, T = T if n = 2, T" = T* = min (T, T^UQ)) if n = 3.
3
b(u,v, w) may be ^0, while b(u, v, w) = 0, for u, uelHp(ft) (or H^Q)), by lack of the free
divergence property.
 FINITE TIME APPROXIMATION 95

resolution of these problems. We note that, as usual (cf. R. Temam [3], [RT,
Chap. Ill, § 7]), the alternating direction or splitting-up method allows us to
decompose, and treat separately, the difficulties related to nonlinearity and to
the incompressibility constraint div u = 0 (contained of course in (13.6)).
We associate with this family of elements u™+i/2 of Wh the following
functions denned on [0, T] (cf. R. Temam [3], [RT]):
• u k l) is the piecewise constant function which is equal to u™+l/2 on
[mk, (m + l)k), i = 1, 2, m = 0 , . . ., N-1.
• u ( k r) is the continuous function from [0, T'] into Wh, which is linear on
(mk, (m + l)k) and equal to u™ +i/2 at mk, i = 1, 2, m = 0 , . . . , N-1.
By adding (13.6) and (13.8) we obtain a relation which can be reinterpreted
in terms of these functions as

where

Similarly, by adding (13.8) to the relation (13.6) for m + 1, we arrive at an

equation which is equivalent to

13.2. The convergence theorem. We now discuss the behavior of these

functions u(kl), u k l) , as h and k —> 0.
THEOREM 13.1. Under the above assumptions, the functions
belong to a bounded set of
As k and h —>• 0, u k l) and u k ° converge to the solution u of Problem 2.2 m
and
Proof, i) We first derive the a priori estimates on the functions. We set
UK = Uh+V2 in (13.6) and, observing that

we get
96 PART III. QUESTIONS RELATED TO THE NUMERICAL APPROXIMATION

whence

Similarly, taking vh = u™ + 1 in (13.8) and taking into account (2.34), we obtain

By adding all the relations (13.14), (13.15) for m = 0 , . . . , JV-1, we find

(by the Schwarz inequality).

Due to (13.4), ||u0J^||u0||, and therefore the sequence uoh is bounded in H.

We then conclude that

where

By adding the relations (13.14) and (13.16) for m = 0 , . . . , p, we obtain, after

simplification,

Adding the relations (13.14) for m = 0 , . . . , p and (13.16) for m = 0 , . . . , p —1,

we find, after dropping unnecessary terms,

The relations (13.18)-(13.21) amount to saying that the functions Uk\ wk°,
i = 1, 2, belong to a bounded set of L°°(0, T'; G), and that w(fc1}, w k 2) , belong to a
bounded set of L2(0, T'; W). In order to show that u(k° also belongs to a
bounded set of L2(0, T'; W), we observe by direct calculation (cf. [RT, Chap.
 FINITE TIME APPROXIMATION 97

III, Lemma 4.8]) that

ii) We want to pass to the limit as k —» 0, h —»• 0. Due to the previous a priori
estimates, there exist a subsequence (denoted k, h) and w (1) , u <2) in
L2(0, T'; W)nL°°(0, T'; G) such that

The relations (13.17) and (13.22) imply that the same is true for u^\ i = 1, 2.
Now we infer from (13.17) that

and therefore

so that u (2) = u(1). Also we infer from the properties of the Vh's (which
constitute an external approximation4 of V) that w m belongs in fact to
L2(0, T'; V)nL°°(0, T';H):

The last (and main) step in the proof is to show that u% is a solution to
Problem 2.2. We need for that purpose a result of strong convergence which
will be proved with the help of the compactness theorem in § 13.3.

13.3. A compactness theorem.

THEOREM 13.2. Let X and Y be two (not necessarily reflexive) Banach spaces
with
the injection being compact
Let <& be a set of functions in Ll(U; Y)HL P (IR;X), p > l , with

uniformly for g e ^;
the support of the functions e e $ is included in a fixed compact of

Then the set ^ is relatively compact in LP(!R; X).

4
For the details see [RT].
98 PART III. QUESTIONS RELATED TO THE NUMERICAL APPROXIMATION

Proof, i) For every a >0, and for every geL p (R; X] we define the function
Jag: U i-» X, by setting

It follows from the first relation in (13.30) that the function s •->• (Jag)(s) is
continuous from [R into X; on the other hand / a geL p (IR; X), and furthermore

by the Holder inequality

The same reasoning shows that JageLl(U; Y) and

ii) It is easy to see that Jag -> g in LP([R; X) (resp. L1^; Y)) as a -» 0 for
every geL p (!R;X) (resp. I/flR; Y)). Due to (13.31), it suffices to prove this
point for functions g which are ^°° with values in X (resp. Y) and have a
compact support, and this is obvious.
We then observe that, due to (13.28),

In order to prove (13.33), we write

and (13.28) implies that this last quantity converges to 0, uniformly for g e $ as
a^O.
 FINITE TIME APPROXIMATION 99

iii) We now show that

/••i-3-7 4 \ F°r every a > 0 fixed, the set ^a = {/ag, ge <&} is uniformly
equicontinuous in ^(R;X).
Indeed, for every s,teU

Hence we have (13.34).

We then apply Ascoli's theorem (cf. Bourbaki [1]) to show that the set ^a is
relatively compact in ^(1R;X). All the functions Jag have their support
included in a fixed compact, say [—L — 1, L +1] (assuming that |a| ^ 1). Due to
(13.34), the only assumption of Ascoli's theorem which remains to be checked
is that
(13.35) For every s eR, the set |/ag(s), ge $} is relatively compact in X.
But according to (13.26), it suffices to show that this set is bounded in Y, and
we have with (13.27)

The set <3a is therefore relatively compact in ^(R; X), and due to (13.29) this
set is, for every fixed a, relatively compact in L P (R; X).
iv) Finally we prove that the set ^ itself is relatively compact in L P (IR;X).
For instance, we have to prove (cf. Bourbaki [1]) that for every e >0 there
exist a finite number of points g l 5 . . ., gN, in L P (IR; X), such that ^ is included
in the union of the balls centered at & and of radius e.
According to (13.33), for every e > 0 there exists a such that

Since ^a is compact, it can be covered by a finite number of balls of

LP(R; X), centered at J a g l5 . . . , JagN, and of radius e/3; finally it is clear with
100 PART III. QUESTIONS RELATED TO THE NUMERICAL APPROXIMATION

(13.36) that the balls of LP(U; X) centered at g l 9 . . . , gN, and of radius e cover
«.
The proof of Theorem 13.2 is complete.
We have a similar result for functions g defined on a bounded interval, say
[o, n
THEOREM 13.3. We assume that X and Y are two Banach spaces which
satisfy (13.26). Let % be a set bounded in L\0, T; Y) and L p (0, T; X), T>
0, p > 1, such that

Then ^ is relatively compact in L q (0, T; X) for any q, l^=q<p.

Proof. This is proved by applying Theorem 13.2 in the following manner.
We consider the set ^ of functions g: g(s) = g(s) if s E [0, T] and = 0 otherwise.
The set <$ is bounded in L'flR; Y)DL q ([R; X) and satisfies (13.28) for p
replaced by q, as

and this goes to 0 uniformly with respect to g e ^ as a —>• 0.

Remark 13.1. Of course, under the assumptions of Theorem 13.2 we can
obtain that ^ is relatively compact in L p (0, T; X) itself if we assume, instead of
(13.37), that

13.4. Proof of Theorem 13.1 (conclusion), i) By application of Theorem

13.3, we show that the functions u(k°, M(k° converge to u strongly in L2(0, T'; G)
or even Lq(0, T';G), l^q«*.
By integration of (13.12),
 FINITE TIME APPROXIMATION 101

We majorize the absolute value of the right-hand side of this equality as

follows:

where c{ depends only on the data, u(), /, v, fl (or Q), T'. Similarly,

For the term involving b, we use Lemma 2.1 with

by holders os inequality and since

Finally we obtain the majorization

Since u(kn is a Vh-valued function 5 , we can take u h = u(k\t + a)- u(k\t), and
we find after integration with respect to t:

5
This is not the case for u ( k 2) (a \Vh-valued function).
102 PART III. QUESTIONS RELATED TO THE NUMERICAL APPROXIMATION

We apply Theorem 13.3, as follows. ^ is the family of functions uk1} which is

bounded in L2(0, T; W)nL°°(0, T; G), (Y = W, X = G), p = 2, so that (13.37)
follows from (13.40). We conclude that uk1} is relatively compact in
L q (0, T; G), l ^ q < 2 , and since u(k1} converges weakly in L q (0, T'; G) to u*
(see (13.23)-(13.25)), this means that u^ converges to u*, strongly in
Lq(0, T'; G), l ^ q < 2 . The sequence uk1} being bounded in L°°(0, T'; G), we
conclude from the Lebesgue dominated convergence theorem that we actually
have

As for (13.22), it is easy to check by direct calculation that

and with (13.17)

This, together with (13.24) and (13.41), shows that

(13.44) w(k° and u(k° converge to u strongly in Lq(0, T; G) V q, 1 ^q <oo.

ii) Using (13.44) and the weak convergences obtained in § 13.2 (in particular
(13.23)), the passage to the limit in (13.12) is standard. We will just give the
main lines of the proof. Let $ be any ^l scalar function on [0, T'] which
vanishes near T'; we multiply (13.11) by tKO» integrate in t and integrate by
parts the first term to get

Let v be an arbitrary element of V. We choose for vh an approximation of v,

and passing to the limit in (13.45) we get
 FINITE TIME APPROXIMATION 103

(The passage to the limit in the nonlinear term necessitates the strong con-
vergence (13.41); cf. [RT].) We deduce from (13.46) that u* is a solution to
Problems 2.1-2.2, and since this solution is unique in the present situation, we
conclude that u% = u. By uniqueness, the entire sequences u(^\ u^ converge to
u as k and h tent to 0. The last step of the proof is to obtain strong
convergence in L2(0, T"; W).
This page intentionally left blank
14 Long Time Approximation of the
Navier-Stokes Equations

The study of the long time behavior of the solutions to the Navier-Stokes
equations is directly related to an understanding of turbulent flow, as can be
seen from Part II. In this section we present two results which tend to indicate,
as do those in §§11 and 12, that the flow has, for time large, a finite
dimensional structure. The first result is related to the flow itself (§ 14.1); the
second to its Galerkin approximation (§ 14.2).

14.1. Long time finite dimensional approximation. Let / and g be two

continuous bounded functions from [0, °°) into H, let w0 and u() be given in V,
and let u and v denote the corresponding solutions to Problem 2.2:

We assume that u and v are defined and bounded for all time in V:

We recall that this property is automatically satisfied if n = 2 (cf. Lemma 12.2).

We recall also that for every a > 0

where c" depends on a, c' and the data (Lemma 12.2).

The space E. Let us now consider a finite dimensional subspace E of V. We
denote by P(E) the orthogonal projector in H onto E, and Q(E) = /-P(E).
Since P(E) is not a projector in V, it may happen that ((</>,(/>)) ^ 0, if
$eE, i / > e V and P(E)i/> = 0. However (see Lemma 14.1 below), there exists
p ( E ) , Q ^ p ( E ) < l , such that

We also associate with E the two numbers A(E), ju,(E),

106 PART III. QUESTIONS RELATED TO THE NUMERICAL APPROXIMATION

so that

When it is not necessary to mention the dependance on E, we will write

simply P, Q, p, A, /a instead of P(E), and so forth.
We will now prove (14.5).
LEMMA 14.1. Under the above assumptions, there exists p = p(E} such that
(14.5) holds.
Proof. If this were not true, we could find two sequences <£>,-, j^-,/^1, <£,-e
E, 0;. G V, Pty = 0 such that

We can extract a subsequence, still denoted /, such that </>) converges to some
limit <}>, \\(j>\\ =l,<j>eE (E has finite dimension), and ^,' converges weakly in V
to i/r, i/f e V, W^ 1, P»A = 0. At the limit, (14.8) gives

so that ||i//|| = 1, $ = k<t> ^ 0, in contradiction with

An inequality. We consider the two solutions u, v of (14.1)-(14.2), and we
set

(P = P(E),Q = Q(E) = /-P(E)).

We apply the operator Q to the difference between (14.1) and (14.2). We
obtain

We then take the scalar product in H with q,

Using Lemma 2.1 (or (2.36)) and (14.5)-(14.7), we find that the right-hand
side of this inequality is less than or equal to
 LONG TIME APPROXIMATION OF THE NAVIER-STOKES EQUATIONS 107

If f=^a, a > 0 arbitrary, then this expression is in turn less than or equal to

where e > 0 is arbitrary.

If

then we set

and we have established:

LEMMA 14.2. If (14.3)-(14.5)-(14.7) and (14.9) hold then, for t ^ a > 0 ,

with v', e as in (14.10).

We now prove
THEOREM 14.1. We assume that the space dimension is n = 2 or 3 and that u
and v are solutions of (14.1)-(14.2) uniformly bounded in V. Let E be a finite
dimensional subspace of V such that (14.9) is satisfied.
Then, if

we have also

Remark 14.1. Theorem 14.1 shows that if the condition (14.9) is satisfied
the behavior for t — » < » of u(t] is completely determined by that of P(E)u(t]
(although we do not know how to recover u(f), knowing P(E}u(t)). Lemma
14.3 below shows that (14.9) is satisfied if E is "sufficiently large".
108 PART III. QUESTIONS RELATED TO THE NUMERICAL APPROXIMATION

Proof of Theorem 14.1. We infer from (14.6) and (14.11) that

where

whence for

Given 5>0, there exists M (which we can assume ^a) such that for t ^M

Therefore, for t^t0 + M, (14.11) implies

where

We let 5 -> 0 and then M^ °o and we obtain (14.14).

As mentioned before, the following lemma shows that condition (14.9) is
always satisfied if E is sufficiently large.
LEMMA 14.3. // Ej} j ^ 1, is an increasing sequence of subspaces of V such that
\JjEj is dense in V, then A(£/)—> +°° as /—»<».
Proof. In fact, we will prove that
(14.17) For every integer m there exists jm and, for /i?/ m , we have
A(E Jm )i=A m ,
where A m is the mth eigenvalue of A.
 LONG TIME APPROXIMATION OF THE NAVIER-STOKES EQUATIONS 109

By assumption, for every k,

Hence, for given m and S > 0, there exists /m such that

Thus, for every /i^; m , there exist w l 5 . . . , wm in £,-, with ||w, — w,-||^=5. There-
fore if $ E V, (/-?(£;))(/> = 0, we have

This implies

and the results follow by letting

14.2. Galerkin approximation. We are going to establish a result similar to

that in Theorem 14.1, for the Galerkin approximation of Navier-Stokes
equations when the space dimension n = 2. For simplicity we restrict ourself to
the Galerkin approximation using the eigenfunctions vvm of A (as in § 3.3), and
we will show that if m is sufficiently large, the behavior as t —> °° of the
Galerkin approximation um is completely determined by the behavior as t —> °°
of a certain number m* of its modes, i.e. the behavior of Pm^um, m^<m.
We start with some complementary remarks on Galerkin approximations.
For fixed m, the Galerkin approximation um of the solution u of (14.1) is
defined (see § 3.3) by

where Pm is the projector in H, V, V, D(A),. . . , onto the space spanned by

the first m eigenf unctions of A, w l 5 . . . , vvm.
We have established in § 3.3 some a priori estimates which are valid on a
finite interval of time [0, T]; we can obtain a priori estimates valid for all time
as follows.
110 PART III. QUESTIONS RELATED TO THE NUMERICAL APPROXIMATION

We infer from (3.46) that

where N(f) = suptg£0 |/(f)|. Thus for every r ^ s ^ O

and |um(0| is bounded for all time:

The following a priori estimate is also verified by um.

LEMMA 14.4. 11^(011 is bounded independently of m and t.
Proof. For T = a > 0, this is proved on [0, a] in (3.59). In order to obtain the
result for t^a, we consider the analogue of (3.11) for i^ (obtained by taking
the scalar product of (14.18) with Au m ; see § 3):

Therefore with (14.21)

and for 0 ^ s ^ t, we can show by integration that

If t ^ a > 0, we integrate in s from t — a to t and we find

Using (14.20), we see that the right-hand side of this inequality is bounded
by a constant depending on a and the data, but independent of t and m. The
lemma follows.
 LONG TIME APPROXIMATION OF THE NAVIER-STOKES EQUATIONS 111

We can now state the result announced.

Let u^ and vm be the Galerkin approximations to the solution u and v of
(14.1)-(14.2),

THEOREM 14.2. We assume that n = 2 and that m ^ m*, m* sufficiently large

so that the condition (14.29) below is verified.
Then, if

we have

i.e.,

Proof. We have to obtain an analogue of (14.11); we then proceed as for

Theorem 14.1.
We set for m ^ ^ m (m* to be determined):

(Qm = I-P m ). Then by subtracting (14.21) from (14.23) and applying Q,^ to
the equality which we obtain, we arrive at

Consequently,,

The right-hand side is equal to

Because of (2.31) and Lemma 14.4, this quantity is bounded by

112 PART III. QUESTIONS RELATED TO THE NUMERICAL APPROXIMATION

If

we set

and we bound the last quantity by

Finally we find

This inequality is totally similar to (14.11), and starting from (14.31) the
proof of Theorem 14.2 is the same as that of Theorem 14.1.
Remark 14.2. We did not establish above any connection between the
behavior for t —>°° of u(t) and of its Galerkin approximation um(i). This
remains an open problem. See however P. Constantin-C. Foias-R. Temam [1].
APPENDIX
Inertial Manifolds and
Navier-Stokes Equations

Inertial manifold is a new concept introduced in 1985 after the publication of

the first edition of these notes. Although related concepts and results existed
earlier, inertial manifolds were introduced under this name in 1985 and system-
atically studied for nonlinear dissipative evolution equations since that time. It
would be beyond the scope of this appendix to make a thorough presentation of
inertial manifolds; we refer the interested reader to more specialized books and
articles (see references below). In this appendix we recall only the concept, the
main definitions, and the main results and concentrate on some results specifically
related to the Navier-Stokes equations as they appear in the articles of C. Foias-
G. Sell-R. Temam [1], [2], M. Kwak [1], and R. Temam-S. Wang [1]. We restrict
ourselves to space dimension two. Some of the results are not in their final form;
a number of problems are still open but we believe they are of sufficient interest
to be presented in their current form.

A.I. Inertial manifolds and inertial systems. It was shown in § 12 that attrac-
tors for Navier-Stokes equations, which are bounded in the //'-norm, have finite
dimension. As indicated in the introduction of § 12, this shows that the long-time
behavior of the solutions to these equations is finite dimensional. The concept of
inertial manifold is an attempt to go further in the reduction of the dynamics of
infinite-dimensional equations such as the Navier-Stokes equations to that of a
finite-dimensional differential equation.
Consider a semigroup of continuous operators {S(t)}t^o in a Hilbert space //
(scalar product (•, •), norm | • |). This could be, for instance, the semigroup defined in
§ 12 for the Navier-Stokes equations in space dimension two when /(?) = / E H
is independent of t\ S(t) is then the nonlinear mapping «(0) — u(t) in //, whose
existence is guaranteed by Theorem 3.1 (in particular (3.34)).
DEFINITION Al. Given a semigroup of operators {S(t)},^o in a Hilbert space
//, an inertial manifold for this semigroup (or for the corresponding evolution
equation) is a Lipschitz finite-dimensional manifold M in H such that

9A.1)

for every «o G //, S(t)uQ converges to M

at an exponential rate, i.e.,
(A 2)
dist
where c\,C2 depend boundedly on \UQ\ .
113
114 APPENDIX

When an inertial manifold exists, the restriction of S(t) to M defines a finite-

dimensional semigroup of operators {L(t)}t^o,

which fully reproduces the dynamics of the initial equation (or semigroup). This
system is called an menial system. More generally an inertial system (or an inertial
form) for a semigroup {S(t)}t^o is a finite-dimensional system (or semigroup)
{Z(r)}rgo which fully reproduces the dynamics of the initial system.1
Having defined inertial manifolds (and inertial systems), we now make some
comments and comparisons with attractors.
Remark A.I. When an attractor j^ exists as well as the inertial manifold M
then obviously d C M and, if Theorem 12.1 applies, then stf like M has finite
dimension. However, there are differences between attractors and inertial mani-
folds:
i) Attractors may be complicated sets, even fractals, while inertial manifolds
are required to be smooth, at least Lipschitz manifolds, usually ^ manifolds.
ii) The convergence of orbits to the attractors may be slow, like for instance a
negative power of t or even slower. However the convergence of the orbits to an
inertial manifold is required to be exponential. Hence, after a short transient pe-
riod, the orbits essentially lie on the inertial manifold M and most of the dynamics
takes place on M.
Remark A.2. See Remark A.4 for some comments on the connection between
inertial manifolds and turbulence.

A.2. Survey of the main results. In this section we survey some typical results
on inertial manifolds for general equations and give bibliographical references.
We consider in the Hilbert space H an evolution equation

This could be the Navier-Stokes equation written in the form (2.43), with A re-
placed by z/A and R(u) = f — Bu. We assume here that A is an unbounded self-
adjoint closed positive operator in H with domain D(A) C H, and that R is a %l
mapping from the domain D(Ay] in H of A7, for some y (0 ^ y < 1), into H.
The function u = u(t) is a mapping from IR+ (or some interval of 1R) into D(A}.
We assume that the initial value problem (A.3), (A.4) is well posed; i.e., we
assume that for every UQ e H there exists a function u continuous from R+ into
1
We realize that this definition is not very precise but prefer to avoid technicalities; see the references
given below for the precise statements.
 APPENDIX 115

H which satisfies (A.3), (A.4) in some sense. According to Theorem 3.1, this
is true for the Navier-Stokes equations in space dimension two (space periodic
case or flow in a bounded domain). In that case we denote obviously by S(t) the
operator

Assuming also that A"1 is compact, we see that there exists an orthonormal
basis of H, {w/}/ e N which consists of eigenvectors of A, i.e., exactly as in (2.17),

Then any u e H (or u e D(A^) for some /3 ^ 0) can be expanded in the form

where the series (A.6) converges in H (or in D(Af3)).

For any N  IN, let PN denote the orthogonal projector in H onto the space
spanned by w\,..., WN, and let QN = I -P^ be the orthogonal projector in H onto
the space spanned by the vv/, j ^ N + 1. Then for u e //, we write

It is easy to see that equation (A.3) is equivalent to the system

As in (A.8), (A.9) we will omit the indices N and write F, Q, v, z instead of PN, QN,
y^, ZN, when N is fixed and no confusion can arise.
There are by now many ways to construct an inertial manifold for equation
(A.3)(A.4) (or the semigroup {SU)}rso); m general they produce the inertial man-
ifold M as the graph of a function $ from P^H into <2/v//, for some N:

The following result proved in C. Foias-G. Sell-R. Temam [1], [2], is a general
and typical result of existence of inertial manifolds. To avoid technicalities we
state it in a fuzzy way without all the assumptions; the interested reader is referred
to the references above for the precise statements.
116 APPENDIX

THEOREM A.I. The assumptions are those given above in this Appendix. We
also assume
i) some technical hypotheses on A and R,
ii) X/v+i ^ K i , f o r some N 6 IN,

for some K\,K2 > 0, and some a, 0 ^ a < 1, which depend on A and R.
Then there exists a %l mapping $ from P^H into Q^H such that the graph M
of 4> is an inertial manifold for equation (A.3), (A.4) (or the semigroup {S(t)}t^o)-
The technical hypotheses mentioned in i) are satisfied by many equations includ-
ing the Navier-Stokes equations in space dimension two. Hypothesis ii) is easily
satisfied for TV sufficiently large, but the most restrictive assumption is hypothesis
iii), also called the spectral gap condition. For instance, for the two-dimensional
Navier-Stokes equations, a = 1/2 and since,

(see R. Courant-H. Hilbert [1], G. Metivier [1]) we do not know if iii) is satis-
fied. This hypothesis is satisfied and Theorem A. 1 applies for reaction-diffusion
equations for which a = 0 (dimension one or two), for certain dissipative evolu-
tion equations of mathematical physics (such as the Kuramoto-Sivashinsky, the
Cahn-Hilliard, or the complex Ginzburg-Landau equations). It applies more gen-
erally to "very dissipative" equations, including the Navier-Stokes equations with
enhanced viscosity described below.
Before describing these equations of Navier-Stokes type and proceeding with
recent results and open problems, we make a few comments.
Remark A.3 (asymptotic completeness). Another interesting property of inertial
manifolds is the asymptotic completeness property proved in C. Foias-G. Sell- E.
Titi [1], and P. Constantin-C. Foias—B. Nicolaenko-R. Temam [1]; namely,

Remark A.4 (inertial manifolds and turbulence). Consider an orbit u = u(t)

lying in M; then in view of (A.6), (A.7), and (A. 10), we have, at all times,

In the physical language of turbulence, (A. 12) means that the high frequency
component of the flow, z, is slaved by its low frequency component, y. Other
comments on the relevance of inertial manifolds to turbulence appear in R. Temam
[11], [12]. See also the concept of approximate inertial manifolds in C. Foias-O.
Manley-R. Temam [2]-[5],
 APPENDIX 117

Remark A.5. Detailed study of inertial manifolds for dissipative evolution equa-
tions appears in the references quoted above; see also J. Mallet-Paret-G. Sell [1]
for reaction-diffusion equations in higher dimension, A. Debussche-R. Temam [2]
for a general result of existence of inertial manifolds, including the case where
the operator A is not self-adjoint and an estimate of their dimensions. Many
more results and references appear in the books by P. Constantin-C. Foias-B.
Nicolaenko-R. Temam [1], R. Temam [13], and the books edited by C. Foias-
B. Nicolaenko-R. Temam [1] and C. Foias-G. Sell-R. Temam [3]. See also the
books quoted above and the references therein about approximation of inertial
manifolds and attractors. The relations of inertial manifolds with the concept of
slow manifolds in meteorology is addressed in A. Debussche-R. Temam [1], [2]
and discussed from a nontechnical point of view in R. Temam [12].
Remark A.6. Relations like (A. 12) open the way for the use of approximate
inertial manifolds for the development of multilevel methods for the numerical
approximation of the Navier-Stokes equations; see the references quoted above
and R. Temam [14].
We conclude § A.2 with the Navier-Stokes-related example.
Example A. 1: Navier-Stokes equation with enhanced viscosity. We consider the
Navier-Stokes equations in space dimension n with a higher-order viscosity term
(see J. L. Lions [1]):

The functions u = u(x, t) and p = p(x, t) are defined on IR" X 1R+, taking values
in IR" and IR, respectively, u = (u\,...,un); r 6 IN, and //,v are strictly posi-
tive numbers. For // = 0, equations (A. 13), (A. 14) reduce to the incompressible
Navier-Stokes equations.
We consider the space periodic case; hence u and p are periodic in each direction
x\,.. .,xn, with period L > 0, i.e., (1.11) holds. We also assume as in Remark 1.1
and in the preceding chapters that

where g = (0,L)'!.
We consider the same space H as in § 2.1 and set

with domain D(A) = V n H^(£>) where V and H^r(Q) are defined as in § 2.1.
Writing
118 APPENDIX

where n is the orthogonal projector in IL2(<2) onto H, equation (A.3) is the same
as (A. 13), (A.14).
We consider the same trilinear form b as in (2.27) and we infer from Lemma
2.1 that b is trilinear continuous on Hm'(0 X Hm2+1(e) X 1L2(0 provided m\ =
m.2 + 1 > n/4. In particular for m\ — mi + 1 = r, b is trilinear continuous on
D(Al/2) X D(Al/2) X H, A as in (A. 15), provided

e.g., r ^ 1 in space dimension 2 ^ n ^ 5. Using the methods of § 2, it is very

easy to see that (1.11), (A.13), (A.14) define a well-posed initial value problem
and that all hypotheses of Theorem A. 1 are satisfied except iii) (see the technical
details in C. Foias-G. Sell-R. Temam [1], [2]).
Now we turn our attention to hypothesis iii) in Theorem A.I. By R. Courant-D.
Hilbert [1]-G. Metivier [1],

where c is an appropriate constant. Assume momentarily that X# = cN2r/n\ then

for a = 1/2,

and the spectral gap condition is satisfied for N large provided

It can be shown that this condition, established by assuming that \N = cN2r/n,

remains partly valid if we only assume that \# ~ cN2r/n as N -* oo; namely we
obtain that the spectral gap condition is satisfied for a sequence Af, of Ns,Nj — oo
as / — oo.
In conclusion, under assumption (A. 17), equation (A.13), (A.14) with space
periodicity boundary condition (1.11) possesses an inertial manifold. A better
(smaller) value of r can be obtained by choosing a more carefully (a smaller).

A.3. Inertial system for the Navier-Stokes equations. In this section we

present a tentative approach for proving the existence of an inertial form for the
space periodic Navier-Stokes equations in space dimension two, as in M. Kwak
[1]. Similar results for the flow around a sphere, as in R. Temam-S. Wang [1],
are given in § A.4.2
2
Despite the statements made in these references and in the first printing of the second edition of
this book, the existence of an inertial form for the Navier-Stokes equations in space dimension two is
not yet fully proved. The difficulty is explained below. Nevertheless we thought the method proposed
by M. Kwak sufficiently interesting and promising to warrant presentation in its present form. The
reduction of the Navier-Stokes equations to a system of reaction-diffusion equations is also interesting
on its own.
 APPENDIX 119

As observed before, Theorem A. 1 does not apply to the two-dimensional Navier-

Stokes equations with space periodic boundary conditions. The program of M.
Kwak consists of imbedding the Navier-Stokes equations, through what we will
call the Kwak transform, into a reaction-diffusion system for which Theorem
A.I might apply and then deducing the existence of an inertial system for the
Navier-Stokes equations themselves.
The following presentation, based on R. Temam-S. Wang [1], is much simpler
than the original proof; other results contained in R. Temam-S. Wang [1] are
presented in § A.4. Of course we will not be able to give all the details but we
will emphasize the main points. All the details can be found in R. Temam-S.
Wang [1], hereafter referred to as [TW].
It is convenient, in view of § A.4, to introduce the function

Taking the curl of (1.5), we obtain the curl equation

Space periodicity is also required for ^ (£(•, t) <E Hlp(Q) at each time t)3 and from
^ we recover u by the equations

where <// is the stream function. It is clear that (A. 18), (A. 19) are equivalent to the
space periodic Navier-Stokes equations in dimension two.
The appropriate Kwak transform in this case consists of considering the func-
tions

By straightforward calculation, one can check that /(Q is solution of the system
(</?' = d(p/dt):

Here A"1 is the inverse of the Laplace operator from L2(Q) into
the contracted product

3
For the function spaces and norms the notations are those of § 2.
120 APPENDIX

The embedded system. The embedded system (reaction-diffusion type system)

consists of looking for functions £, v, w which are space periodic with period Q,
and which satisfy the following equations:

Here v = v/L2, and u, v, w are still defined by (A. 19) and (A. 19') while v and w are
now functions independent of £; r > 0 sufficiently large will be choosen hereafter.
Except for the underlined terms, equations (A.21) are the same as (A.20) if we
replace v and w by v and w. We call (A.21) the embedded system.
For simplicity we write U - (£, v, w). The linear operator in (A.21) is the
nonself-adjoint operator U — At/,

We consider it an unbounded operator from W2 into Wo, where for s ^ 0, W5 =

HSP(Q) X HSP(Q)2 X HSP(Q). Then, the notation being obvious, we write (A.21) like
(A.3) in the form

One can derive a priori estimates for the solutions of (A.21) and prove theorems
of existence and uniqueness of solutions, similar to Theorem 3.1. As in § 3 we
will establish a priori estimates on the solutions U of (A.21) assuming that they are
sufficiently regular, skipping the details of the proof of existence and uniqueness;
this is done in two steps:

i) The first step, a key one, is to show that (A.21) converges in some sense to
(A.20) as t — oo; this is shown by proving that v(t) - v(t) and w(t) - w(t) converge
to 0 as t —• oo. Note that if £, v, w are solutions of (A.21) then

Subtracting this equation from the second equation (A.21) we find

Hence
 APPENDIX 121

and if \JL\ = 4?r2/L2 is the first eigenvalue of the operator

The proof is more involved for w — w. As for the derivation of the third equation
(A.20), we first observe (using div w = v • u) that w satisfies

Subtracting this equation from the third equation (A.21) we obtain

Upon taking the scalar product in L2(Q) of each side of (A.21) with w - w, we
find

Using appropriate estimates (see [TW]) we infer from (A.28) that

where c\ is a suitable constant independent of r. We now choose r = c\. Dividing

then (A.29) by z/2 and adding this relation to (A.24) we obtain
122 APPENDIX

In particular

and this shows that

This implies that the embedded system "converges" to the original one (i.e., (A. 18)
or (A.20)) when t —• oo.
ii) In a second step we derive further a priori estimates.
Estimates for £, v = grad £, and w = £w are derived as follows. We take the
scalar product in L2(Q) of the first equation (A.21) with £; we find

Therefore

We multiply (A.32) by //i/4 (m = 47r2/L2) and add this relation to (A.30):

 APPENDIX 123

From (A.33) we infer some estimates similar to (3.4M3.7). In particular

which yields

where K\ depends on (/o, |/|, ^ and L.

Further estimates are derived as in (3.9M3.17) by considering higher norms.
Recall that, for any s ^ 0, W, = HSP(Q) X HSP(Q)2 X HSP(Q) and that D(A) =
W2. Then, based on these estimates one proves (see [TW] and compare to
Theorem 3.2) the following theorem.
THEOREM A.2. For f given in HP(Q)2 = Hlp(Q)2 and U0 given in D(A), there
exists a unique solution U = U(t) of (A.21) such that (7(0) = UQ, and

Furthermore
i) // U0 = /(Co), then U(t) = 7(C(f)) V t i= 0.
ii) // I/o + ACo), then

Application of Theorem A.I. We would now like to apply Theorem A.I or, more
precisely, a version of Theorem A.I adapted to the nonself-adjoint case (see A.
Debussche-R. Temam [2], [3], hereafter referred to as [DT2,3]). Remember that
A is not self-adjoint; however its eigenvalues are the numbers (47r2v/L2)(k2 + k2),
&i,&2 e IN and its eigenfunctions are proper combinations of sines and cosines.
One of the main points in applying Theorem A. 1 is to check the spectral gap
condition, i.e., hypothesis iii) of Theorem A.I. Here the \# are the numbers
(4-K2i'/L2)(k2 + k2), k\,k2 G IN numbered in increasing order and according to
their multiplicity.
Considered in D(A), equation (A.21) is of reaction-diffusion type, i.e., its non-
linear terms are continuous functions of U in D(A); for this reason a = 0 in
hypothesis iii). Thus the spectral gap condition reduces to a well-known problem
in number theory; namely that there are arbitrarily large gaps among the integers
which are sum of two squares (of integers). Such gaps indeed exist.
Because we need the nonself-adjoint version of Theorem A.I appearing, for
instance, in [DT2,3], a new difficulty arises which is not yet resolved. Indeed some
of the requirements in [DT2,3] involve bounding, independently of N, the norms of
124 APPENDIX

the operators eA'P and e~^Q where t ^ 0 and P = PN and Q = QN are the spectral
projectors as above. Although A is not self-adjoint, its generalized eigenvectors
(root vectors) are orthogonal and the projectors P and Q are orthogonal. However,
the Jordan blocks of A produce some contributions to eAr, e~At, which we are not
able to control. The verification of this hypothesis (hypothesis (HI) in [DT3]) has
been overlooked in the references using the Kwak transform.
If an inertial manifold and an inertial system for the embedded equation exist
then, due to statement i) in Theorem A.2, the inertial form of the embedded
system is also an inertial form of the two dimensional Navier-Stokes equation.
Write U = Y + Z and write

the equation of the inertial manifold for the embedded system as in (A.7) and
(A. 10). Let P be the corresponding operator P = PN; then from (A.8MA.10) and
(A.22) we infer the inertial system

That is, we would have proved the following result:

(A.35') There exists a finite-dimensional differential system,

namely (A.35), which produces the same dynamics
as the two-dimensional space periodic Navier-Stokes equations.

Remark A.7. i) We assumed that the flow has the same period L in each direction
xi,X2- If the periods Lj,L 2 are different, then the existence of an inertial mani-
fold hinges on the fact that there are arbitrarily large gaps in the set of numbers
((L%/L])k\ + £2), k\,&2 e IN. This remains true if Li/L\ is rational and Theorem
A.3 is valid in this case as well (see C. Foias-G. Sell-R. Temam [1], [2]).
ii) For the flow in a bounded domain (see § 2.5), the embedded system is not
of reaction-diffusion type and the previous method no longer applies.
iii) If j/o is the global attractor for the Navier-Stokes equations and s$ the
attractor for the embedded system then s$ = J(^o) and, as far as we know,
(A.21) is the simplest imbedded system for which this property is valid.
iv) We could have considered the Navier-Stokes equations in vector form in-
stead of the curl equation, using a suitable embedded system. However, for some
geometric reason this is not straightforward for flows around a sphere considered
in § A.4 and, for the sake of simplicity, it was better to start from the curl equation
in both cases.

A.4. Flow around a sphere. In this section we consider the flow of an incom-
pressible fluid around a two-dimensional sphere; we follow [TW].
4
That is, the same equilibrium points, same time-periodic or time quasi-periodic orbits, same at-
tractors, and so on.
 APPENDIX 125

There are two reasons for considering such flows:

• The first reason is the obvious interest of such flows in meteorology and
climate type problem.
• The second reason is related to the spectral gap condition arising in Theorem
A.2. The Laplace operator is now replaced by the Laplace-Beltrami operator on
the sphere which we still denote by A. The eigenvalues and eigenfunctions of A
are known; the eigenfunctions are spherical harmonics and the eigenvalues are the
numbers n(n + 1), n E IN, repeated according to their multiplicity. Then the X./v are
numbers of the form vn(n +1), and there are indeed large gaps in the spectrum.
The corresponding Navier-Stokes equations for the flow around the unit sphere
S2 are written

Here A is the Laplace-Beltrami operator on the unit sphere S2 (for vector func-
tions), V«v is the covariant derivative of v in the direction u, grad and div are the
gradient and divergence operators on S2. The analytic expressions using spherical
coordinates can be found in [TW] and in many places in the literature (see, e.g.,
T. Aubin [1]). We have

Here the curl of a scalar function i// or a vector function u are defined by

where n is the unit outward normal on S2.

Now set

and observe that u is completely determined by £ using div u = 0, which implies

Taking the curl of each side of (A.36) we obtain

126 APPENDIX

This equation is the same as (A. 18) except that A is now the Laplace-Beltrami
operator and div, curl have to be properly defined. In fact, except for some minor
and obvious modifications, all that was said in § A.3 from (A. 18) to (A.35') applies
here. The embedded system is the same as (A.21) (see [TW]). The only modifi-
cation necessary occurs when we apply Theorem A.2 and check the spectral gap
condition: here we do not need to invoke the result in number theory. Instead we
observe that there are indeed arbitrarily large gaps in the spectrum which consists
of the numbers vn(n + 1). If \# and \N+I are two different consecutive eigenvalues
\N ~ vn(n +1), \N+i = v(n + \)(n + 2), then

Hence all hypotheses of Theorem A. 1 are easily verified, in particular the spectral
gap condition. In fact, (A.43) would give the value of N (expressed in terms of
the data) for which the spectral gap condition is satisfied and hence an estimate of
the dimension of the inertial manifolds of the embedded system and of the inertial
form of the Navier-Stokes equations for the flow around a sphere. However, here
we need again the nonself-adjoint version of Theorem A.I of [DT3] and in that
respect we meet the same difficulty as in § A.3. Instead of verifying hypothesis
(HI) of [DT3] we might perhaps consider different constructions of the inertial
manifold of the embedded system, using another function space. Also the reduc-
tion to the reaction-diffusion system (A.21) is not canonical; one might obtain a
different reaction-diffusion system for which the operator A is self-adjoint. These
problems are open.
Remark A.8. It would be very easy to incorporate the Coriolis force k X u in
the model above.
 Comments and Bibliography

Most of the results in §§ 2 and 3 are classical and, as mentioned, can be

found in the books of O. A. Ladyzhenskaya [1], J. L. Lions [1] and R. Temam
[6], and in J. Serrin [1]. The only difference is that we have emphasized a little
more the space periodic case which is not treated in detail in the literature;
also, at the end of § 3 (§ 3.4) we give a simple but interesting result of generic
solvability in the large, due to A. V. Fursikov [1].
The new a priori estimates in § 4 are borrowed from C. Foias-C. Guillope-
R. Temam [1]. The results on the Hausdorff dimension of the singular set of a
weak solution to the Navier-Stokes equations in § 5 are two typical results, if
not the most recent. They rely on the ideas of B. Mandelbrot [1], and on the
work of V. Scheffer [1], [4] and C. Foias-R. Temam [5]. The result in §5.1
(time singularities) is just a restatement by V. Scheffer [1] of an old result of J.
Leray [3], and also generalizes a result of S. Kaniel and M. Shinbrot [1]. The
result in § 5.2, leading to a Hausdorff dimension of singularity ^| in space and
time (n = 3), is an extension due to C. Foias-R. Temam [5] of a result by V.
Scheffer [3], [4]. The best result presently available along those lines is that of
L. Caffarelli-R. Kohn-L. Nirenberg [1] (the one-dimensional Hausdorff meas-
ure of the singular set vanishes).
The set of compatibility conditions given in § 6 was derived in R. Temam [8].
The first compatibility condition is mentioned in J. G. Heywood [2]; for the
compatibility conditions, cf. also O. A. Ladyzhenskaya-V. A. Solonnikov [3].
In § 7 we give a simple proof of time analyticity of the solutions to the
Navier-Stokes equations (with values in D(A)), which is a "complexification"
of the standard Galerkin approximation procedure; this presentation is slightly
simpler than the original one given in C. Foias-R. Temam [2], [5]. For other
results on time analyticity, cf. also C. Foias-C. Guillope-R. Temam [1], and for
a proof based on semigroup operators which apparently applies to N.S.E., cf.
F. J. Massey III [1].
Finally, in the last section of Part I (§ 8), we use an a priori estimate of § 4, to
determine the Lagrangian representation of the flow associated with a weak
solution of the N.S.E.; for further developments on this new result, cf. C.
Foias-C. Guillope-R. Temam [2].
The properties of the set of stationary solutions to the N.S.E. based on the
Sard-Smale theorem, and derived in § 10, are among the results proved in C.
Foias-R. Temam [3], [4]. Results of genericity with respect to the boundary
conditions are not presented here; they use somewhat more sophisticated tools
in analysis and topology, and can be found in J. C. Saut-R. Temam [1], [2]. For
other recent results on the set of stationary solutions, cf. also C. Foias-J. C.
Saut [3] and Gh. Minea [1].
The results in §§11 and 12 are taken from C. Foias-R. Temam [5]. The
squeezing property of the trajectories is used in § 12, and has played a role in
127
128 COMMENTS AND BIBLIOGRAPHY

the proof of other results on the structure of a turbulent flow: cf. in particular
C. Foias-R. Temam [7], [8]. The results mentioned in § 12.3 are proved in C.
Guillope [1]; a similar result is also announced in J. G. Heywood-R. Ran-
nacher [1].
For the numerical approximation of the N.S.E. on a finite time interval (and
for stationary solutions), see the references mentioned in § 13; see also R.
Temam [6], and M. Fortin-R. Peyret-R. Temam [1], R. Peyret [1], R. Temam
[1], [2], [4], [5], F. Thomasset [1], and the references therein. The compactness
theorem used in the proof of convergence is new. The results in § 14 are part of
a work in progress; cf. C. Foias-R. Temam [9], C. Foias-O. Manley-R.
Temam-Y. Treve [1] and the references there.
 Comments and Bibliography:
Update for the Second Edition

A new result published since the first edition of these notes is the Gevrey class
regularity of the space periodic solutions of the Navier-Stokes equations which
appeared in C. Foias-R. Temam [13]. This implies a result of space analyticity
of the solutions, providing an exponential decay rate of the Fourier coefficients
when the force / itself is in the same class. The proof, which is short and rather
elementary, is based on appropriate forms of energy estimates.
The results of Chapter 3 on the existence and uniqueness of solutions in space
dimension three have been partly improved by G. Raugel-G. Sell [1], who consider
three-dimensional flows in a thin domain (domain thin in the third direction).
Namely, if the thickness e and the data satisfy some set of inequalities, then the
solution exists for large times and remains smooth.
The "new a priori estimates" in Chapter 4 have been extended in several di-
rections by G. F. D. Duff [l]-[4] who addresses the case of a bounded domain
and of nonzero (nonpotential) volume forces. See also in D. Chae [1], [2] results
extending those of § 4 to Gevrey classes, combining the methods of C. Foias-C.
Guillope-R. Temam [1] and C. Foias-R. Temam [13]. The energy inequalities in
Lemma 4.1 (derived here as a preliminary result for the "new a priori estimates"
in § 4.3) have been improved by W. D. Henshaw-H. O. Kreiss-L. G. Reyna [1].
They make explicit the dependence of the bounds on v and on associated nondi-
mensional numbers and their proof leads to better bounds than those obtained if
one makes explicit the bounds from Lemma 4.1.
New results partly related to those of Chapter 8 (Lagrangian Representation of
the Flow) were proved by R. Coifman-P. L. Lions-Y. Meyer-S. Semmes [1].
The problems considered in Part II have been the object of extensive research
during the past decade; a number of new results have appeared concerning the
behavior for t — oc of the solutions, the concept of attractor, and the idea of fi-
nite dimensionality of flows. For the attractors, the result of finite dimensionality
of Navier-Stokes attractors presented in Chapter 12 (and based on C. Foias-R.
Temam [5]) have been improved using the concept of Lyapunov exponents and
ideas from dynamical systems theory. New bounds have been derived on the
dimension of attractors which are physically relevant and are related to and con-
sistent with the conventional theory of turbulence of Kraichnan (in dimension two)
and Kolmogorov (in dimension three); see P. Constantin-C. Foias-O. Manley-R.
Temam [1], P. Constantin-C. Foias [1], P. Constantin-C. Foias-R. Temam [2], [3].
A thorough description of the results and many more results on attractors for
dissipative evolution equations can be found in the following books and in the
references therein: A. V. Babin-M. I. Vishik [1], C. Foias-O. Manley-R. Temam
129
1 30 COMMENTS AND BIBLIOGRAPHY

[5], J. Hale [1], O. A. Ladyzhenskaya [2], E. Lieb [1], D. Ruelle [4], [5], M. I.
Vishik [2]; see also the books edited by C. Foias-B. Nicolaenko-R. Temam [1]
and C. Foias-G. Sell-R. Temam [3].
Another result pertaining to finite dimensionality of flows is the concept of
inertial manifolds addressed in the Appendix. The corresponding comments and
bibliographical references are given in the Appendix itself. Finally results on
finite dimensionality of flows also appear as part of the concepts of determining
modes and determining points. Determining modes were introduced and studied
in C. Foias-O. Manley-R. Temam-Y. Treve [1]; determining points (or nodes)
were introduced and studied in C. Foias-R. Temam [11], [12]. New developments
appear in D. A. Jones-E. S. Titi [l]-[3]; reference [3] contains the latest results.
In relation to Part III and in particular Chapter 14, we would like to mention
the development of new multiresolution algorithms based on the use of approx-
imate inertial manifolds. This includes the nonlinear Galerkin method and the
incremental unknown method. See C. Foias-O. Manley-R. Temam [2], [3], M.
Marion-R. Temam [1], [2], C. Foias-M. Jolly-I. F. Kevrekidis-G. R. Sell-E. S.
Titi [1], T. Dubois-F. Jauberteau-R. Temam [1], [2], R. Temam [15], [16], J. G.
Heywood-R. Rannacher [2], D. A. Jones-L. G. Margolin-E. S. Titi [1], and B.
Garcia-Archilla-J. de Frutos [1], among others; see also R. Temam [16].
 References

R. A. ADAMS,
[1] Sobolev Spaces, Academic Press, New York, 1975.
S. AGMON,
[1] Lectures on Elliptic Boundary Value Problems, Elsevier, New York, 1965.
S. AGMON, A. DOUGLIS AND L. NIRENBERG,
[1] Estimates near the boundary for solutions of elliptic partial differential equations satisfying
general boundary conditions, I, II, Comm. Pure Appl. Math., 12 (1959), pp. 623-727; 17
(1964), pp. 35-92.
V. I. ARNOLD,
[1] Mathematical Methods in Mechanics, Springer-Verlag, New York, 1978.
T. AUBIN,
[1] Nonlinear Analysis on Manifolds, Monge-Ampere Equations, Springer-Verlag, New York, Hei-
delberg, Berlin, 1982.
A. V. BABIN AND M. I. VISHIK,
[1] Attractors of Evolution Equations, North-Holland, Amsterdam, 1992.
C. BARDOS AND C. BESSIS,
[1] Bifurcation Phenomena in Mathematical Physics and Related Topics, Nato Advanced Study
Institutes Series, D. Reidel, Dordrecht, Boston, London, 1979.
T. B. BENJAMIN,
[1] Bifurcation phenomena in steady flows of a viscous fluid, I. Theory, II. Experiments, Proc.
Roy. Soc. London Ser. A, 359 (1978), pp. 1-26; 27-43.
T. B. BENJAMIN AND T. MULLIN,
[1] Transition to oscillatory motion in the Taylor experiment, Nature, 288 (1980), pp. 567-569.
[2] Anomalous modes in the Taylor experiment, Proc. Roy. Soc. London Ser. A, 377 (1981),
pp. 221-249.
A. BENSOUSSAN AND R. TEMAM,
[1] Equations stochastiques du type Navier-Stokes, J. Funct. Anal., 13 (1973), pp. 195-222.
M. BERCOVIER AND M. ENGELMAN,
[1] A finite element for the numerical solution of viscous incompressible flows, J. Comp. Phys., 30
(1979), pp. 181-201.
C. BESSIS,
See C. Bardos and C. Bessis.
N. BOURBAKI,
[1] Elements de mathematiques, livre III: topologie generale, Hermann, Paris, 1964.
D. BREZIS,
[1] Interpolation de semi-groupes non lineaires, These, Universite de Paris,
M. O. BRISTEAU, R. GLOWINSKI, B. MANTEL, J. PERIAUX, P. PERRIER AND O. PIRONNEAU,
[1] A finite element approximation of Navier-Stokes equations for incompressible viscous fluids.
Iterative methods of solutions, in Approximation Methods for Navier-Stokes Problems, R.
Rautman, ed., Lecture Notes irt Mathematics 771, Springer-Verlag, New York, 1980, pp.
78-128.
F. BROWDER,
[1] Nonlinear operators and nonlinear equations of evolution in Banach spaces, Proc. Symposia
in Pure Mathematics, vol. XVIII, part 2, American Mathematical Society, Providence, RI,
1976.
L. CAFEARELLI, R. KOHN AND L. NIRENBERG,
[1] Partial regularity' of suitable weak solutions of the Navier-Stokes equations, Comm. Pure Appl.
Math., 35(1982), pp. 771-831.
C. CASTAING,
[1] Sur les multi-applications mesurables, Rev. Franc. Inf. Rech. Oper., 1 (1967), pp. 91-126.
131
132 REFERENCES

L. CATTABRIGA,
[1] Su un problema al contomo relative al sistemo di equazioni di Stokes, Rend. Sem. Mat. Univ.
Padova, 31 (1961), pp. 308-340.
D. CHAE,
[1] Remarks on the regularity of weak solutions of the Navier-Stokes equations, Comm. Partial
Differential Equations, 17 (1992), pp. 1267-1285.
[2] Some a priori estimates for weak solutions of the 3D Navier-Stokes equations, J. Math. Anal.
Appl., 167 (1992), pp. 236-244.
A. J. CHORIN,
[1] Lectures on Turbulence Theory, Publish or Perish, Inc., Boston, 1975.
[2] Numerical method for solving incompressible viscous flow problems, J. Comp. Phys., 2 (1967),
pp. 12-26.
[3] Numerical solution of the Navier-Stokes equations, Math. Comp., 23 (1968), pp. 341-354.
[4] Numerical solution of incompressible flow problems. Stud. Numer. Anal., 2 (1968), pp.
64-71.
R. COIFMAN, P. L. LIONS, Y. MEYER AND S. SEMMES,
[1] Compensated compactness and Hardy spaces, J. Math. Pures Appl., 72 (1993), pp. 247-286.
P. CONSTANTIN AND C. FOIAS
[1] Global Lyapunov exponents, Kaplan-Yorke formulas and the dimension of the attractors for
two-dimensional Navier-Stokes equations, Comm. Pure Appl. Math., 38 (1985), pp. 1-27.
P. CONSTANTIN, C. FOIAS, O. MANLEY AND R. TEMAM,
[1] Determining modes and fractal dimension of turbulent flows, J. Fluid Mech., 150 (1985), pp.
427-440.
P. CONSTANTIN, C. FOIAS, B. NICOLAENKO AND R. TEMAM,
[1] Integral Manifolds and Inertial Manifolds for Dissipative Partial Differential Equations, Applied
Mathematical Sciences Series, Vol. 70. Springer-Verlag, New York, 1988.
P. CONSTANTIN, C. FOIAS AND R. TEMAM,
[1] On the large time Galerkin approximation of the Navier-Stokes equations, SI AM J. Numer.
Anal., 21 (1984), pp. 615-634.
[2] Attractors representing turbulent flows, Mem. Amer. Math. Soc., 53 (1985), pp. 1-67.
[3] On the dimension of the attractors in two-dimensional turbulence, Physica D, 30 (1988), pp.
284-296.
R. COURANT AND D. HlLBERT,
[1] Methods of Mathematical Physics, Intersciences Publishers, New York 1953.
J. DE FRUTOS,
See B. Garcia-Archilla and J. De Frutos.
A. DEBUSSCHE, T. DUBOIS AND F. JAUBERTEAU,
[1] The nonlinear Galerkin method: A multiscale method applied to the simulation of homogeneous
turbulent flows, Theoret. Comput. Fluid Dynamics, (1995), to appear.
A. DEBUSSCHE AND R. TEMAM,
[1] Inertial manifolds and slow manifolds, Appl. Math. Lett., 4 (1991), pp. 73-76.
[2] Inertial manifolds and the slow manifolds in meteorology, Differential Integral Equations, 4
(1991), pp. 897-931.
[3] Inertial manifolds and their dimension, in Dynamical Systems, Theory and Applications, S.I.
Andersson, A.E. Andersson, and O. Ottoson, eds., World Scientific Publishing Co., Singapore,
1993.
[4] Convergent families of approximate inertial manifolds, J. Math. Pures Appl., 73 (1994), pp.
485-522.
A. DOUGLIS,
See S. Agmon, A. Doughs and L. Nirenberg.
T. DUBOIS,
See A. Debussche, T. Dubois and F. Jauberteau
T. DUBOIS, F. JAUBERTEAU AND R. TEMAM,
[1] Solution of the incompressible Navier-Stokes equations by the nonlinear Galerkin method, J.
Sci. Comput., 8 (1993), pp. 213-240.
[2] Article in preparation for Computational Mechanics Advances.
 REFERENCES 133

G. F. D. DUFF,
[1] Navier—Stokes derivative estimates in three space dimensions with boundary values and hodv
forces, C. R. Math. Rep. Acad. Sci. Canada, 11 (1989), pp. 195-200.
[2] On derivative estimates for the Navier-Stokes equations in IR3, C. R. Math. Rep. Acad. Sci.
Canada, 12 (1990), pp. 81-86.
[3] Derivative estimates for the Navier-Stokes equations in a three dimensional region, Acta Math.,
164(1990), pp. 145-210.
[4] Navier-Stokes derivative estimates in three dimensions with boundary values and body forces,
Canad. J. Math., 43 (1991), pp. 1161-1212.
D. EBIN AND J. MARSDEN,
[1] Groups of diffeomorphisms and the motion of an incompressible fluid, Ann. Math., 92 (1970),
pp. 102-163.
M. ENGELMAN.
See M. Bercovier and M. Engelman.
H. FEDERER,
[1] Geometric Measure Theory, Springer-Verlag, New York, 1969.
M. J. FEIGENBAUM,
[1] Some formalism and predictions of the period doubling onset of chaos, in Nonlinear Problems:
Present and Future, A. R. Bishop, D. K. Campbell and B. Nicolaenko, eds., North-
Holland, Amsterdam, 1982, pp. 379-394.
P. R. FENSTERMACHER, H. L. SWINNEY AND J. P. GOLLUB,
[1] Dynamical instabilities and the transition to chaotic vortex flow, J. Fluid Mech., 94 (1979),
pp. 103-128.
C. FOIAS,
f 1] Statistical study of the Navier-Stokes equations, I & II, Rend. Sem., Mat. Univ. Padova, 48
(1972), pp. 219-348; 49 (1973), pp. 9-123.
See also P. Constantin and C. Foias; P. Constantin, C. Foias, O. Manley and R. Temam; P. Con-
stantin, C. Foias, B. Nicolaenko and R. Temam; P. Constantin, C. Foias and R. Temam.
C. FOIAS AND C. GUILLOPH,
]1] On the behavior of the solutions of the Navier-Stokes equations lying on invariant manifolds,
J. Differential Equations, 61 (1986), pp. 128-148.
C. FOIAS, C. GUILLOPE AND R. TEMAM,
[1] New a priori estimates for Navier-Stokes equations in dimension 3, Comm. Partial Differen-
tial Equations, 6 (1981), pp. 329-359.
[2| Lagrangian representation of a flow, J. Differential Equations, 57 (1985), pp. 440-449.
C. FOIAS, M. JOI.LY, I. F. KKVKHKIDIS, G. R. SELL AND E. TITI,
[ 11 On the computation of inertial manifolds, Phys. Lett. A, 131 (1988), pp. 433-436.
C. FOIAS, O. MANLFY AND R. TKMAM,
[ 1 ] Attractors for the Benard problem: Existence and physical bounds on their dimension, J. Non-
linear Anal.-T.M.A., 11 (1987), pp. 939-967.
|2| Stir I'interaction des petils et grands tourbillons dans les ecoulements turbulents, C. R. Acad.
Sci. Paris Ser. I Math., 305 (1987), pp. 497-500.
[31 On the interaction of small and large eddies in two-dimensional turbulent flows, Math. Model.
Numer. Anal.. 22 (1988). pp. 93-114.
|4| Approximate inertial manifolds and effective viscosity in turbulent flows, Phys. Fluids A, 3
(1991), pp. 898-911.
[5| Navier-Stokes Equations and Turbulence, Cambridge University Press, London, in preparation.
C. FOIAS, O. MANLEY, R. TEMAM AND Y. TREVE,
[1] Asymptotic analysis for the Navier-Stokes equations, Physica D, 9D (1983), pp. 157-188.
G. FOIAS AND G. PRODI,
[1] Sur le comportement global des solutions non stationnaires des equations de Navier-Stokes en
dimension 2, Rend. Sem. Mat. Padova, 39 (1967), pp. 1-34.
C. FOIAS AND J. C. SAUT,
11 ] Limite du rapport de I 'enstrophie sur I 'energie pour line solution faible des equations de Navier-
Stokes, C. R. Acad. Sci. Paris Ser, I, 293 (1981), pp. 241-244.
134 REFERENCES

[2] Asymptotic properties of solutions of the Navier-Stokes equations, expose au Seminaire de

Mathematiques Appliquees du College de France, Pitman, London, 1982.
[3] Remarques sur les equations de Navier-Stokes stationeries, Ann. Scuola Norm. Sup. Pisa,
Serie IV, 10 (1983), pp. 169-177.
C. FOIAS, G. SELL AND R. TEMAM,
[1] Varietes inertielles des equations dijferentielles dissipatives, C. R. Acad. Sci. Paris Ser. I Math.,
301 (1985), pp. 139-142.
[2] Inertial manifolds for nonlinear evolutionary equations, J. Differential Equations, 73 (1988),
pp. 309-353.
[3] Turbulence in fluid flows: A dynamical system approach, IMA Volumes in Mathematics and Its
Applications, Vol. 55, Springer-Verlag, New York, 1993.
C. FOIAS, G. SELL AND E. Tin,
[1] Exponential tracking and approximation of menial manifolds for dissipative equations, J. Dy-
namics Differential Equations, 1 (1989), pp. 199-244.
C. FOIAS AND R. TEMAM,
[I] Sur certaines proprietes generiques des equations de Navier-Stokes, C. R. Acad. Sci. Paris,
280 (1975), p. 563-565.
[2] On the stationary statistical solutions of the Navier-Stokes equations and turbulence, Publica-
tions Mathematiques d'Orsay, no. 120-75-28, 1975.
[3] Structure of the set of stationary solutions of the Navier-Stokes equations, Comm. Pure Appl.
Math., 30 (1977), pp. 149-164.
[4] Remarques sur les equations de Navier-Stokes stationnaires et les phenomenes successifs de
bifurcation, Ann. Scuola Norm. Sup. Pisa, Ser. IV, 5 (1978), pp. 29-63.
[5] Some analytic and geometric properties of the solutions of the Navier-Stokes equations. J.
Math. Pures Appl., 58 (1979), pp. 339-368.
[6] Homogeneous statistical solutions of Navier-Stokes equations, Indiana Univ. Math. J., 29
(1980), pp. 913-957.
[7] A specifically nonlinear property of the semi-group operator of the Navier-Stokes equations,
Comm. Pure Appl. Math., 35 (1982), pp. 197-207.
[8] Finite parameter approximative structure of actual flows, in Nonlinear Problems: Present and
Future, A. R. Bishop, D. K. Campbell and B. Nicolaenko, eds., North-Holland, Amster-
dam, 1982.
[9] Asymptotic numerical analysis for the Navier-Stokes equations, in Nonlinear Dynamics and
Turbulence, G. I. Barenblatt, G. looss, D. D. Joseph, eds., Pitman, London, 1983, pp. 139-155.
[10] Self-similar universal homogeneous statistical solutions of the Navier-Stokes equations, Comm.
Math. Phys., 90 (1983), pp. 187-206.
[ I I ] Sur la determination d'un ecoulement fluide par des observations discretes, C. R. Acad. Sci.
Paris Ser. I Math., 295 (1982) pp. 239-241; Sur la determination d'un ecoulement fluide par
des observations discretes: Principe de la prediction d'un ecoulement, C. R. Acad. Sci. Paris
Ser. I Math., 295 (1982), pp. 523-525.
[12] Determination of the solutions of the Navier-Stokes equations by a set of nodal values, Math.
Comp., 43 (1984), pp. 117-133.
[13] Gevrey class regularity for the solutions of the Navier-Stokes equations, J. Funct. Anal., 87
(1989), pp. 359-369.
M. FORTIN, R. PEYRET AND R. TEMAM,
[1] Resolution numerique des equations de Navier-Stokes pour un fluide incompressible, J.
Mecanique, 10 (1971), pp. 357-390.
A. V. FURSIKOV,
[1] On some problems of control, Dokl. Akad. Nauk SSSR, (1980), pp. 1066-1070. (In Russian.)
See also M. I. Vishik and A. V. Fursikov.
B. GARCIA-ARCHILLA AND J. DE FRUTOS,
[1] Time integration of the nonlinear Galerkin method, Applied Mathematics and Computation
Reports, Report 1993/7, Universidad de Valladolid, Spain, July 1993.
 REFERENCES 135

K. GEBA,
[1] The Leray-Schauder degree and framed bordism, in La theorie des points fixes et ses
applications a Fanalyse, Seminaire de Math. Superieures 1973, Presses de 1'Universite de
Montreal, Montreal 1975.
V. GlRAULT AND P. A. RAVIART,
[1] Finite Element Approximation of Navier-Stokes Equations, Lecture Notes in Mathematics
749, Springer-Verlag, New York, 1979.
R. GLOWINSKI,
See M. O. Bristeau, R. Glowinski, B. Mantel, J. Periaux, P. Perrier and O. Pironneau.
R. GLOWINSKI, B. MANTEL AND J. PERIAUX,
[1] Numerical solution of the time dependent Navier-Stokes equations for incompressible viscous
fluids by finite element and alternating direction methods, in Proc. Conference on Numer-
ical Methods in Aeronautical Fluid Dynamics, Reading, England, March 30-April 1,
1981.
J. P. GOLLUB,
[1] Recent experiments on connective turbulence, in Nonlinear Problems: Present and Future, A.
R. Bishop, D. K. Campbell and B. Nicolaenko, eds., North-Holland, Amsterdam, 1982,
pp. 403-408.
See also P. R. Fenstermacher, H. L. Swinney and J. P. Gollub; H. L. Swinney and J. P. Gollub.
K. K. GOLOVKIN AND O. A. LADYZHENSKAYA,
[1] Solutions of non-stationary boundary value problems for the Navier-Stokes equations, Trudy
Mat. Inst. Steklov, 59 (1960), pp. 100-114. (In Russian.)
K. K. GOLOVKIN AND V. A. SOLONNIKOV,
[1] On the first boundary value problem for non-stationary Navier-Stokes equations, Dokl.
Akad. Nauk. SSSR, 140 (1961), pp. 287-290. (In Russian.)
C. GUILLOPE,
[1] Comportement a I'infini des solutions des equations de Navier-Stokes et propriete des ensembles
fonctionnels invariantes (ou attracteurs), C. R. Acad. Sci. Paris Sen I. 294 (1982), pp. 221-224,
and Ann. Inst. Fourier (Grenoble), 32 (1982), pp. 151-180.
[2] Remarques a propos du comportement, pour t tendant vers +00, des solutions des equations de
Navier-Stokes associees a une force nulle, Bull. Soc. Math. Fr., 111 (1983), pp. 151-180.
See also C. Foias and C. Guillope; C. Foias, C. Guillope and R. Temam.
J. HALB,
|1] Asymptotic Behavior of Dissipative Systems, Mathematical Surveys and Monographs, Vol. 25,
American Mathematical Society, Providence, 1988.
W. D. HENSHAW, H. O. KREISS AND L. G. RKYNA,
[1] Smallest scale estimates for the Navier-Stokes equations for incompressible fluids, Arch. Ra-
tional Mech. Anal., 112 (1990), pp. 21-44.
J. G. HEYWOOD,
[1] On the uniqueness questions in the theory of viscous flow, Acta Math., 136 (1976), pp.
61-102.
[2] Classical solutions of the Navier-Stokes equations, in Proc. IUTAM Symposium Paderbon,
1979, R. Rautmann, ed., Lecture Notes in Mathematics 771, Springer-Verlag, New York,
1980.
J. G. HEYWOOD AND R. RANNACHER,
[1] Finite approximation of the nonstationary Navier-Stokes problem. Part I: Regularity of
solutions and second-order error estimates for spatial discretization, SIAM J. Numer. Anal.,
19 (1982) pp. 275-311.
[2] On the question of turbulence modeling by approximate inertial manifolds and the nonlinear
Galerkin method, SIAM J. Numer. Anal., 30 (1993), pp. 1603-1621.
D. HiLBERT,
See R. Courant and D. Hilbert,
M. HIRSCH,
[1] Differential Topology, Springer-Verlag, New York, 1976.
136 REFERENCES

E. HOPF,
[1] Uber die Aufgangswertaufgabe fur die hydrodynamischen Grundliechungen, Math. Nachr., 4
(1951), pp. 213-231.
G. looss,
[1] Bifurcation of Maps and Applications, North-Holland, Amsterdam, 1979.
F. JAUBERTEAU,
See A. Debussche, T. Dubois and F. Jauberteau; T. Dubois, F. Jauberteau and R. Temam.
M. JOLLY,
See C. FOIAS, M. JOLLY, I. F. KEVREKIDIS, G. R. SELL AND E. TITI.
D. A. JONES, L. G. MARGOLIN AND E. S. Tin,
[1] On the effectiveness of the approximate inertial manifold: A computational study, Theor. Com-
put. Fluid Dynamics, 7 (1995), to appear.
D. A. JONES AND E. S. TITI,
[\]On the number of determining modes for the 2D-Navier-Stokes equations, J. Math. Anal. Appl.,
168(1992), pp. 72-88.
[2] Determining finite volume elements for the 2D-Navier-Stokes equations, Physica D, 60 (1992),
pp. 165-174.
[3] Upper bounds on the number of determining modes, nodes and volume elements for the Navier-
Stokes equations, Indiana Univ. Math. J., 42 (1993).
A. V. KAJIKOV,
[1] Resolution of boundary value problems for nonhomogeneous viscous fluids, Dokl. Akad.
Nauk SSSR, 216 (1974), pp. 1008-1010. (In Russian.)
S. KANIEL,
[1] A kinetic model for a mono-atomic gas, Univ. of Jerusalem, 1981, preprint.
S. KANIEL AND M. SHINBROT,
[1] Smoothness of weak solutions of the Navier-Stokes equations, Arch. Rational Mech. Anal.,
24 (1967), pp. 302-324.
R. KOHN,
See L. Caffarelli, R. Kohn and L. Nirenberg.
I. F. KEVREKIDIS,
See C. Foias, M. Jolly, I. F. Kevrekidis, G. R. Sell and E. Titi.
H. O. KREISS,
See W. D. Henshaw, H. O. Kreiss and L. G. Reyna.
K. KURATOWSKY AND C. RYLL-NARDZEWSKI,
[1] General theorem on selector, Bull. Acad. Polon. Sci. Ser. Sci. Tech. Mec. Theor. 13 (1965),
pp. 273-311.
M. KWAK,
[1] Finite dimensional inertial form for the 2D-Navier-Stokes equations, Indiana Univ. Math. J.,
41 (1993), pp. 927-981.
O. A. LADYZHENSKAYA,
[1] The Mathematical Theory of Viscous Incompressible Flow, 2nd ed., Gordon and Breach,
New York, 1969.
[2] Attactors for Semigroups and Evolution Equations, Accademia Nazionale dei Lincei Series,
Cambridge University Press, Cambridge, 1991.
See also K. K. Golovkin and O. A. Ladyzhenskaya.
O. A. LADYZHENSKAYA AND V. A. SOLONNIKOV,
[1] On some problems of vector analysis and generalized formulations of Navier-Stokes equa-
tions, Sem. Leningrad, 59 (1976), pp. 81-116. (In Russian.)
[2] On the solution of boundary and initial value problems for the Navier-Stokes equations in
domains with non-compact boundaries, Leningrad Universitet Vestnik, 13 (1977), pp.
39-47. (In Russian.)
[3] Linear and quasi-linear parabolic equations, Moscow, 1967; English trans., American
Mathematical Society, Providence, RI.
 REFERENCES 137

O. A. LADYZHENSKAYA, V. A. SOLONNIKOV AND N. N. URAL'CEVA,

[1] Linear and Quasi Linear Elliptic Equations of Parabolic Type, Vol. 23, Trans. Math.
Monographs 23, American Mathematical Society, Providence, RI, 1968.
O. LANFORD,
[1] The Lorenz attraction, in Turbulence and the Navier-Stokes Equations, R. Temam, ed.,
Lecture Notes in Mathematics 565, Springer-Verlag, New York, 1976.
[2] A computer-assisted proof of the Feigenbaum conjecture, Bull. Amer. Math. Soc., 6
(1982), pp. 427-434.
J. LERAY,
[1] Etude de diverses equations integrales non lineaires et de quelques problemes que pose
Vhydrodynamique, J. Math. Pures Appl., 12 (1933), pp. 1-82.
[2] Essai sur les mouvements plans d'un liquide visqueux que limitent des parois, J. Math. Pures
Appl., 13 (1934), pp. 331-418.
[3] Essai sur le mouvement d'un liquide visqueux emplissant Vespace, Acta Math., 63 (1934),
pp. 193-248.
E. LiliB,
[1] On characteristic exponents in turbulence, Comm. Math. Phys., 92 (1984), pp. 473^83.
J. L. LIONS,
[1] Quelques methodes de resolution des problemes aux limites non lineaires, Dunod, Paris, 1969.
[2] On some problems connected with Navier-Stokes equations, in Non Linear Evolution
Equations, M. Grandall, ed., Academic Press, New York, 1978.
J. L. LIONS AND E. MAGENES,
[1] Nonhomogeneous Boundary Value Problems and Applications, Springer-Verlag, New York,
1972.
J. L. LIONS AND J. PEETRE,
[1] Sur une classe d'interpolation, Publ. Inst. Hautes Etudes Sci., 19 (1964), pp. 5-68.
J. L. LIONS AND G. PRODI,
[1] Un theoreme d'existence et d'unicite dans les equations de Navier-Stokes en dimension 2, C.
R. Acad. Sci. Paris, 248 (1959), pp. 3519-3521.
P. L. LIONS,
See R. Coif man, P.L. Lions, Y. Meyer and S. Semmes.
E. MAGENES,
See J. L. Lions and E. Magenes.
J. MALLET-PARET,
[1] Negatively invariant sets of compact maps and an extension of a theorem of Carrwright, J.
Differential Equations, 22 (1976), p. 331-348.
J. MALLET-PARET AND G. R. SELL,
[ 1 ] Inertial manifolds for reaction diffusion equations in higher space dimensions, J. Amer. Math.
Soc., 1 (1988), pp. 805-866.
B. MANDELBROT,
[1] Intermittent turbulence and fractal kurtosis and the spectral exponent 5/3+ B, in Turbulence
and Navier-Stokes Equations, R. Temam ed., Lecture Notes in Mathematics 565,
Springer-Verlag, New York, 1976, pp. 121-145.
[2] Fractals: Form, Chance and Dimension, Freeman, San Francisco, 1977.
O. MANLKY,
See P. Constantin, C. Foias, O. Manley and R. Temam; P. Constantin, C. Foias, O. Manley and Y.
Treve; C. Foias, O. Manley and R. Temam; C. Foias, O. Manley, R. Temam and Y. Treve.
B. MANTEL,
See M. O. Bristeau, R. Glowinski, B. Mantel, J. Periaux, P. Perrier and O. Pironneau; R.
Glowinski, B. Mantel and J. Periaux.
L. G. MARGOLIN,
See D. A. Jones, L. G. Margolin and E. S. Titi.
138 REFERENCES

M. MARION AND R. TEMAM,

[1] Nonlinear Galerkin methods, SIAM J. Numer. Anal., 26 (1989), pp. 1137-1157.
\2] Nonlinear Galerkin methods: The finite elements case, Numer. Math., 57 (1990), pp. 205-226.
J. MARSDEN,
See D. Ebin and J. Marsden.
G. M£TIVIER,
[1] Valeurs propres d'operateurs de finis sur la restriction de systemes variationnels a des sous-
espaces, J. Math. Pures Appl., 57 (1978), pp. 133-156.
Y. MEYER,
See R. Coifman, P. L. Lions, Y. Meyer and S. Semmes.
G. H. MINEA,
Remarques sur Vunicite de la solution stationnaire d'une equation de type Navier-Stokes, Revue
Roumaine Math. Pures Appl., 21 (1976), pp. 1071-1075.
T. MULLIN,
See T. B. Benjamin and T. Mullin.
B. NlCOLAENKO,
See P. Constantin, C. Foias, B. Nicolaenko and R. Temam.
L. NlRENBERG,
See S. Agmon, A. Douglis and L. Nirenberg; L. Caffarelli, R. Kohn and L. Nirenberg.
R. PALAIS,
Seminar on the Atiyah-Singer Index Theorem, Ann. Math. Studies 57, Princeton Univ. Press,
Princeton, NJ, 1965, Lecture 7.
J. PEETRE,
See J. L. Lions and J. Peetre.
J. PERIAUX,
See M. O. Bristeau, R. Glowinski, B. Mantel, J. Periaux, P. Perrier and O. Pironneau; R.
Glowinski, B. Mantel and J. Periaux.
P. PERRIER,
See M. O. Bristeau, R. Glowinski. B. Mantel, J. Periaux, P. Perrier and O. Pironneau.
R. PEYRET,
See M. Fortin, R. Peyret and R. Temam.
R. PEYRET AND T. D. TAYLOR,
[1] Computational Methods for Fluid Flow, Springer-Verlag, New York, 1982.
O. PIRONNEAU,
See M. O. Bristeau, R. Glowinski, B. Mantel, J. Periaux, P. Perrier and O. Pironneau.
G. PRODI,
See C. Foias and G. Prodi, J. L. Lions and G. Prodi.
P. RABINOWITZ, ed.,
[1] Advanced Seminar on Applications of Bifurcation Theory, Academic Press, New York,
1977.
R. RANNACHER,
See J. G. Heywood and R. Rannacher.
G. RAUGEL AND G. SELL,
[1] Navier-Stokes equations on thin 3D domains. J. Amer. Math. Soc., 6 (1993), pp. 503-568.
P. A. RAVIART,
See V. Girault and P. A. Raviart.
L.G. REYNA,
See W. D. Henshaw, H. O. Kreiss and L. G. Reyna.
D. RUELLE,
[1] Turbulence and Axiom A attractors, in Proc. Intern. School of Math. Phys., Univ. di
Camerino, 1974, pp. 162-181.
[2] Lecture at the International Conference on Theoretical Physics, Rome, June 1977.
 REFERENCES 139

[3] Differentiable dynamical systems and the problem of turbulence, Bull. Amer. Math. Soc., 5
(1981), pp. 29^2.
[4] Large volume limit of distribution of characteristic exponents in turbulence, Comm. Math.
Phys., 87 (1982), pp. 287-302.
[5] Characteristic exponents for a viscous fluid subject to time-dependent forces, Comm. Math.
Phys., 92 (1984), pp. 285-300.
D. RUELLE AND F. TAKENS,
[1] On the nature of turbulence, Comm. Math. Phys., 20 (1971), pp. 167-192; Note, 23 (1971),
pp. 343-344.
C RYLL-NARDZEWSKI,
See K. Kuratowsky and C. Ryll-Nardzewski.
J. C. SAUT,
See C. Foias and J. C. Saut.
J. C. SAUT AND R. TEMAM,
[1] Generic properties of nonlinear boundary value problems, Comm. Partial Differential Equa-
tions, 4 (1979), pp. 293-319.
[2] Generic properties of Navier-Stokes equations: Genericity with respect to the boundary values,
Indiana Univ. Math. J., 29 (1980), pp. 427-446.
V. SCHEFFER,
[1] Turbulence and Hausdorff dimension, in Turbulence and Navier-Stokes Equations, R.
Temam, ed., Lecture Notes in Mathematics 565, New York, 1976, pp. 94-112.
[2] Partial regularity of solutions to the Navier-Stokes equations, Pacific J. Math., 66 (1976), pp.
535-552.
[3] Hausdorff measure and the Navier-Stokes equations, Comm. Math. Phys. 55 (1977), pp.
97-112.
[4] The Navier-Stokes equations in space dimension four, Comm. Math. Phys., 61 (1978), pp.
41-68.
G. SELL,
See G. Raugel and G. Sell.
S. SEMMES,
See R. Coifman, P. L. Lions, Y. Meyer and S. Semmes.
D. SERRE,
[1] Droites de solutions pour I'equation de Navier-Stokes stationnaire bidimensionnelle, Applicable
Anal., 13 (1982), pp. 297-306.
[2] Equation de Navier-Stokes stationnaire avec donnees peu regulieres. Application aux problemes
de Taylor et de la cavite, Ann. Scuola Norm. Sup. Pisa, 10 (1983), pp. 543-559.
[3] Thesis, Universite de Paris Sud. June 1982.
[4] Chute libre d'un solide dans unjiuide visqueux incompressible, Japan J. Appl. Math., 4 (1987),
pp. 99-110.
J. SERRIN,
[1] The initial value problem for the Navier-Stokes equations, in Nonlinear Problems, R. E.
Langer, ed., Univ. of Wisconsin Press, Madison, 1963, pp. 69-98.
[2] On the interior regularity of weak solutions of Navier-Stokes equations, Arch. Rational
Mech. Anal., 9 (1962), pp. 187-195.
M. SHINBROT,
See S. Kaniel and M. Shinbrot.
J. SIMON,
[1] Ecoulement d'un fluide non homogene avec une densite initiale s'annulant, C. R. Acad. Sci.
Paris, 287 (1978), pp. 1009-1012.
S. SMALE,
[1] An infinite dimensional version of Sard's theorem, Amer. J. Math., 87 (1965), pp. 861-866.
[2] Differentiable dynamical systems, Bull. Amer. Math. Soc., 73 (1967), pp. 747-817.
140 REFERENCES

V. A. SOLONNIKOV,
[1] Estimates of the Green tensor for some boundary problems, Dokl. Akad. Nauk. SSSR, 130
(1960), pp. 988-991. (In Russian.)
[2] Estimates of the solutions of a non-stationary linearized system of Navier-Stokes equations,
AMS Translation Ser. 2, 75, American Mathematical Society, Providence, RI, 1968, pp.
2-116.
See also O. A. Ladyzhenskaya and V. A. Solonnikov; O. A. Ladyzhenskaya, V. A. Solonnikov
and N. N. Ural'ceva.
E. M. STEIN,
[1] Singular Integrals and Differentiability Properties of Functions, Princeton Univ. Press,
Princeton, NJ, 1970.
H. L. SWINNEY,
[1] Doubly-periodic circular Couette flow in experiments compared with theory, in Nonlinear
Problems: Present and Future, A. R. Bishop, D. K. Campbell and B. Nicolaenko, eds.,
North-Holland, Amsterdam, 1982.
See also P. R. Fenstermacher, H. L. Swinney and J. P. Gollub.
H. L. SWINNEY AND J. P. GOLLUB, eds.,
[1] Hydrodynamic Instabilities and the Translation to Turbulence, Springer-Verlag, New York,
1981.
F. TAKENS,
See D. Ruelle and F. Takens.
G. I. TAYLOR,
[1] Stability of a viscous liquid contained between two rotating cylinders, Philos. Trans. Roy. Soc.
London Ser. A, 223 (1923), pp. 289-343.
R. TEMAM,
[I] Sur rapproximation des solutions des equations de Navier-Stokes, C. R. Acad. Sci. Paris Ser.
A, 262 (1966), pp. 219-221.
[2] Une methode d'approximation de la solution des equations de Navier-Stokes, Bull. Soc.
Math. France, 98 (1968),*pp. 115-152.
[3] Sur la stabilite et la convergence de la methode des pas fractionnaires, Ann. Mat. Pura Appl.,
74 (1968), pp. 191-380.
[4] Sur r approximation de la solution des equations de Navier-Stokes par la methode des pas
fractionnaires (I), Arch. Rational Mech. Anal., 33 (1969), pp. 135-153.
[5] Sur r approximation de la solution des equations de Navier-Stokes par la methode des pas
fractionnaires (II), Arch. Rational Mech. Anal., 33 (1969), pp. 377-385.
[6] Navier-Stokes Equations, Theory and Numerical Analysis, 2nd ed. North-Holland, Amster-
dam, 1979. (Cited in text as [RT].)
[7] Une propriete generique de Vensemble des solutions stationnaires ou periodiques des equations
de Navier-Stokes, in Proc. of the France-Japan Symposium, Tokyo, 1976, H. Fujita, ed.,
J.S.P.S., 1978.
[8] Behaviour at time t = 0 of the solutions of semi-linear evolution equations, J. Differential
Equations, 43 (1982), pp. 73-92.
[9] (ed.) Turbulence and the Navier-Stokes Equations, Lecture Notes in Mathematics 565,
Springer-Verlag, New York, 1976.
[10] On the Euler equations of incompressible perfect fluids, J. Funct. Anal., 20 (1975), pp. 32^43.
[II] Do menial manifolds apply to turbulence? Physica D, 37 (1989), pp. 146-152.
[12] Inertial manifolds, Math. Intelligencer, 12 (1990), pp. 68-74.
[13] Infinite Dimensional Dynamical Systems in Mechanics and Physics, Applied Mathematical
Sciences Series, Vol. 68, Springer-Verlag, New York, 1988.
[14] Navier-Stokes equations: Theory and approximation, in Handbook of Numerical Analysis, P.
G. Ciarlet and J. L. Lions, eds., Elsevier, New York, in preparation.
[15] New emerging methods in numerical analysis: Applications to fluid mechanics, in Incompress-
ible Computational Fluid Dynamics—Trend and Advances, M. Gunzburger and N. Nicolaides,
eds., Cambridge University Press, Cambridge, 1993.
 REFERENCES 14 1

[16] Incremental unknowns infinite differences, in Proc. of the ICASE/NASA LaRC workshop on
Algorithmic Trends in CFD for the 90s, September 1991, Hampton VA, M. Y. Hussaini, A.
Kumar, and M. D. Salas, eds., Springer-Verlag, New York, 1992.
[17] (editor), Inertial manifolds and their application to the simulation of turbulence, special issue,
Theoret. Comput. Fluid Dynamics, 7 (1995), to appear.
See also A. Bensoussan and R. Temam; P. Constantin, C. Foias, O. Manley and R. Temam; P.
Constantin, C. Foias and R. Temam; A. Debussche, T. Dubois and R. Temam; A. Debussche
and R. Temam; T. Dubois, F. Jauberteau and R. Temam; C. Foias, C. Guillope and R. Temam;
C. Foias, O. Manley and R. Temam; C. Foias, O. Manley, R. Temam and Y. Treve; C. Foias, G.
Sell and R. Temam; M. Fortin, R. Peyret and R. Temam; M. Marion and R. Temam; J. C. Saut
and R. Temam.
R. TEMAM AND S. WANG,
[1] Inertial forms of Navier-Stokes equations on the sphere, J. Funct. Anal., 117 (1993), pp. 215-
242.
F. THOMASSET,
[1] Implementation of Finite Element Methods for Navier-Stokes Equations, Springer-Verlag,
New York, 1981.
E. S. Tin,
[1] On a criteronfor locating stable stationary solutions to the Navier-Stokes equations, Nonlinear
Anal. 11 (1987), pp. 1085-1102.
[2] Un critere pour I'approximation des solutions periodiques des equations de Navier-Stokes,
C. R. Acad. Sci. Paris Sen I Math., 312 (1991), pp. 41-43.
See also C. Foias, M. Jolly, I. F. Kevrekidis, G. R. Sell and E. Titi; D. A. Jones, L. G. Margolin
and E. S. Titi; D. A. Jones and E. S. Titi.
Y. TREVE,
See C. Foias, O. Manley, R. Temam and Y. Treve.
N. N. URAL'CEVA,
See O. A. Ladyzhenskaya, V. A. Solonnikov and N. N. Ural'ceva.
M. VIOT,
[1] Solutions faibles d'equations aux derivees partielles stochastiques non linearies, These, Univ.
de Paris, 1982.
M. I. VISHIK,
[1] The Cauchy problem for the Hopf equation of quasilinear parabolic equations, Soviet Math.
Doklady, 16 (1975), pp. 1126-1130.
[2] As\rnptotic Behaviour of Solutions of Evolutionary Equations, Academia Nazionale dei Lincei
Series, Cambridge University Press, Cambridge, 1992.
See also A.V. Babin and M.I. Vishik.
M. I. VISHIK AND A. V. FURSIKOV,
[1] Les equations de Hopf, les solutions statistiques, les moments correspondent aux sysfemes
d'equations paraboliques quasi-lineaires, J. Math. Pures Appl., 56 (1977), pp. 85-122.
[2] Solutions statistiques homogenes des systemes differentiels paraboliques et du systeme de
Navier-Stokes, Ann, Scuola Norm. Pisa, Ser. 44 (1977), pp. 531-576.
[3] Translationally homogeneous statistical solutions and individual solutions with finite energy
of a system of Navier-Stokes equations, Siberian Math. J., 19 (1978), pp. 710-729.
I. I. VOROVICH AND V. I. YUDOVICH,
[1] Stationary flows of incompressible viscous fluids, Mat. Sb., 53 (1961), pp. 393-428. (In
Russian.)
S. WANG,
See R. Temam and S. Wang.
V. I. YUDOVICH,
See I. I. Vorovich and V. I. Yudovich.
(continued from inside front cover)

JERROLD E. MARSDEN, Lectures on Geometric Methods in Mathematical Physics

BRADLEY EFRON, The Jackknife, the Bootstrap, and Other Resampling Plans
M. WOODROOFE, Nonlinear Renewal Theory in Sequential Analysis
D. H, SATTINGER, Branching in the Presence of Symmetry
R. T^MAM, Navier-Stokes Equations and Nonlinear Functional Analysis
MIKLOS CSORGO, Quantile Processes with Statistical Applications
J. D. BUCKMASTER AND G. S. S. LuDFORD, Lectures on Mathematical Combustion
R. E, TARJAN, Data Structures and Network Algorithms
PAUL WALTMAN, Competition Models in Population Biology
S. R. S. VARADHAN, Large Deviations and Applications
KIYOSI Ird, Foundations of Stochastic Differential Equations in Infinite Dimensional Spaces
ALAN C. NEWELL, Solitons in Mathematics and Physics
PRANAB KUMAR SEN, Theory and Applications of Sequential Nonparametrics
LASZL6 LOVASZ, An Algorithmic Theory of Numbers, Graphs and Convexity
E. W. CHENEY, Mitltivariate Approximation Theory: Selected Topics
JOEL SPENCER, Ten Lectures on the Probabilistic Method
PAUL C. FIFE, Dynamics of Internal Layers and Diffusive Interfaces
CHARLES K. CHOI, Multivariate Splines
HERBERT S. WILF, Combinatorial Algorithms: An Update
HENRY C. TUCKWELL, Stochastic Processes in the Neurosciences
FRANK H. CLARKE, Methods of Dynamic and Nonsmooth Optimization
ROBERT B. GARDNER, The Method of Equivalence and Its Applications
GRACE WAHBA, Spline Models for Observational Data
RICHARD S. VARGA, Scientific Computation on Mathematical Problems and Conjectures
INGRID DAUBECHBBS, Ten Lectures on Wavelets
STEPHEN F. MCCORMICK, Multilevel Projection Methods for Partial Differential Equations
HARALD NIEDERREITER, Random Number Generation and Quasi-Monte Carlo Methods
JOEL SPENCER, Ten Lectures on the Probabilistic Method, Second Edition
CHARLES A. MICCHELLI, Mathematical Aspects of Geometric Modeling
ROGER TEMAM, Navier-Stokes Equations and Nonlinear Functional Analysis, Second Edition
GLENN SHAFER, Probabilistic Expert Systems
PETER J, HUBER, Robust Statistical Procedures, Second Edition
J. MICHAEL STEELE, Probability Theory and Combinatorial Optimization