(D. Cioranescu, V. Girault, K.R. Rajagopal (Auth.) (B-Ok - Xyz)

Advances in Mechanics and Mathematics 35
D. Cioranescu
V. Girault
K.R. Rajagopal
Mechanics and
Mathematics
of Fluids of the
Differential Type
Advances in Mechanics and Mathematics
Volume 35
Series Editors:
David Y. Gao, Virginia Polytechnic Institute and State University
Tudor Ratiu, École Polytechnique Fédérale
Advisory Board:
Ivar Ekeland, University of British Columbia
Tim Healey, Cornell University
Kumbakonam Rajagopal, Texas A&M University
David J. Steigmann, University of California, Berkeley
More information about this series at http://www.springer.com/series/5613

D. Cioranescu V. Girault
•
K.R. Rajagopal
Mechanics and Mathematics

of Fluids of the Differential
Type
123
D. Cioranescu K.R. Rajagopal
Laboratoire Jacques-Louis Lions Department of Mechanical Engineering
Université Pierre et Marie Curie Paris 06 Texas A&M University
Paris College Station, TX
France USA
V. Girault
Laboratoire Jacques-Louis Lions
Université Pierre et Marie Curie Paris 06
Paris
France
ISSN 1571-8689 ISSN 1876-9896 (electronic)

Advances in Mechanics and Mathematics
ISBN 978-3-319-39329-2 ISBN 978-3-319-39330-8 (eBook)
DOI 10.1007/978-3-319-39330-8
Library of Congress Control Number: 2016940108
Mathematics Subject Classification (2010): 35A01, 35A02, 35A15, 35A16, 35A23, 35B35, 35B45,
35D30, 35G60, 35Q30, 35Q35, 35R45, 76A05, 76A10, 76E30
© Springer International Publishing Switzerland 2016

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made.
Printed on acid-free paper
This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG Switzerland
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Introductory Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Non-Newtonian Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Shear-Thinning and Shear-Thickening . . . . . . . . . . . . . 13
2.2.2 Thixotropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.3 Stress Relaxation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.4 Creep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.5 Normal Stress Differences . . . . . . . . . . . . . . . . . . . . . . 16
2.2.6 Yield . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.2 Frames, Frame-Indifference and Restrictions
Due to Frame-Indifference. . . . . . . . . . . . . . . . . . . . . . 24
2.3.3 Balance Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Constitutive Relations for Fluids of the Differential Type . . . . . . 28
2.4.1 Special Motions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4.2 Secondary Flows in Pipes . . . . . . . . . . . . . . . . . . . . . . 39
2.4.3 Stability to Finite Disturbances . . . . . . . . . . . . . . . . . . 41
2.4.4 Instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5 Boundary Conditions for Fluids of the Differential Type. . . . . . . 47
2.6 Creeping Flows of Fluids of the Differential Type . . . . . . . . . . . 54
2.6.1 Creeping Flows of Fluids of Grade Two . . . . . . . . . . . . 55
2.7 Boundary Layer Theories for Fluids of the Differential Type. . . . 59
2.7.1 Boundary Layers in the Limit of Zero Reynolds
Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 60
2.7.2 Development of Boundary Layers in Flows
of Fluids of Grade Two . . . . . . . . . . . . . . . . . . . . ... 61
v
vi Contents
2.7.3 Inertial Boundary Layers. . . . . . . . . . . . . . . . . . . . . . . 63

2.7.4 Flows of Fluids of Grade Two with a Free Surface . . . . 64
2.7.5 Universal Flows. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.8 Rate Type Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.8.1 A Gibbs-Potential-Based Formulation for Obtaining
Rate Type Response Functions for Viscoelastic
Fluids. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.9 Bingham Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.10 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.10.1 Integral Type Constitutive Equations . . . . . . . . . . . . . . 85
2.10.2 Fractional Derivative Models of the Rate
and Integral Type . . . . . . . . . . . . . . . . . . . . . . . . ... 87
2.10.3 Free-Surface Flows . . . . . . . . . . . . . . . . . . . . . . . ... 90
3 Mathematical Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.1 Notation and Spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.1.1 Sobolev and Hölder Spaces. . . . . . . . . . . . . . . . . . . . . 93
3.1.2 Some Interpolation Inequalities . . . . . . . . . . . . . . . . . . 99
3.2 Some Theoretical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.2.1 The Laplace Equation. . . . . . . . . . . . . . . . . . . . . . . . . 101
3.2.2 The Stokes System. . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.3 Further Tools of Functional Analysis . . . . . . . . . . . . . . . . . . . . 109
3.3.1 Subdifferential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.3.2 Some Tools for Nonlinear and Time-Dependent
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4 Classical Non-Newtonian Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.2 Subdifferential Type Constitutive Equation . . . . . . . . . . . . . . . . 116
4.2.1 Formulation and Examples . . . . . . . . . . . . . . . . . . . . . 116
4.2.2 Functional Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.2.3 Existence Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.2.4 Biregularization and Galerkin Solutions . . . . . . . . . . . . 131
4.2.5 First Passing to the Limit . . . . . . . . . . . . . . . . . . . . . . 134
4.2.6 Second Passing to the Limit . . . . . . . . . . . . . . . . . . . . 137
4.2.7 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
4.2.8 Examples and Applications . . . . . . . . . . . . . . . . . . . . . 150
4.2.9 What About the Pressure?. . . . . . . . . . . . . . . . . . . . . . 153
4.3 Model Example: The Bingham Fluid . . . . . . . . . . . . . . . . . . . . 158
4.3.1 Definition of a Bingham Fluid . . . . . . . . . . . . . . . . . . . 158
4.3.2 Stress Threshold . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.4 Fluids with Nonlinear Viscosity Bounded at Infinity . . . . . . . . . 167
4.4.1 Variational Formulation. Examples. . . . . . . . . . . . . . . . 167
4.4.2 Formulation and Properties of the Main Operator. . . . . . 168
4.4.3 Existence and Uniqueness Results . . . . . . . . . . . . . . . . 174
Contents vii
5 Grade-Two Fluids: Some Theoretical Results . . . . . . . . . . . . . . . . . 179

5.1 Introduction and Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . 179
5.1.1 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
5.2 Flow of a Grade-Two Fluid with No-Slip Boundary Condition.
The Simply-Connected Case . . . . . . . . . . . . . . . . . . . . . . . . . . 182
5.2.1 Constructing Solutions: The Main Ideas . . . . . . . . . . . . 183
5.2.2 A First Variational Formulation . . . . . . . . . . . . . . . . . . 185
5.2.3 A Galerkin Solution . . . . . . . . . . . . . . . . . . . . . . . . . . 190
5.2.4 A Priori Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
5.2.5 Passing to the Limit . . . . . . . . . . . . . . . . . . . . . . . . . . 196
5.2.6 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
5.2.7 Global Existence in Time . . . . . . . . . . . . . . . . . . . . . . 201
5.3 Flow of a Grade-Two Fluid with No-Slip Boundary
Condition. The Case of a Multiply-Connected Domain . . . . . . . . 203
5.4 Flow of a Grade-Two Fluid with Tangential Boundary
Condition Prescribed. The Case of a Convex Corner Domain . . . 208
5.4.1 A Lifting; Reduction to Homogeneous Boundary
Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
5.4.2 Formal A Priori Estimates. . . . . . . . . . . . . . . . . . . . . . 211
5.4.3 Galerkin Solution and A Priori Estimates . . . . . . . . . . . 213
5.4.4 Passing to the Limit and Recovering the Pressure . . . . . 215
5.4.5 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Condition Prescribed. The Case of a Smoother Domain . . . . . . . 220
5.5.1 A Leray–Hopf Lifting and Its Consequences . . . . . . . . . 221
5.5.3 Local in Time Existence . . . . . . . . . . . . . . . . . . . . . . . 228
5.6 Splitting: The Steady System in 2-D . . . . . . . . . . . . . . . . . . . . 230
5.6.1 Split Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
5.6.2 A Simpler Leray–Hopf Lifting. . . . . . . . . . . . . . . . . . . 234
5.6.4 Construction of a Solution . . . . . . . . . . . . . . . . . . . . . 239
5.6.5 Uniqueness of the Steady 2-D Problem. . . . . . . . . . . . . 243
5.6.6 Limit as a Tends to Zero . . . . . . . . . . . . . . . . . . . . . . 247
5.7 Splitting: The Time-Dependent System in 2-D. . . . . . . . . . . . . . 249
5.7.2 Existence of Solutions by Semi-discretization
in Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
5.7.3 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
5.8 Splitting: The Steady System in 3-D . . . . . . . . . . . . . . . . . . . . 273
5.8.1 Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
5.8.2 Two Auxiliary Systems . . . . . . . . . . . . . . . . . . . . . . . 278
viii Contents
5.8.3 A Fixed-Point Argument. . . . . . . . . . . . . . . . . . . . . . . 282

5.8.4 The Splitting of J.M. Bernard . . . . . . . . . . . . . . . . . . . 285
5.9 Splitting: The Time-Dependent Problem in 3-D . . . . . . . . . . . . . 288
5.9.1 Splitting: Equivalence. . . . . . . . . . . . . . . . . . . . . . . . . 289
5.9.2 Splitting: A Semi-discrete Scheme . . . . . . . . . . . . . . . . 291
5.9.3 Splitting: A Priori Estimates . . . . . . . . . . . . . . . . . . . . 292
5.9.4 Splitting: Passing to the Limit . . . . . . . . . . . . . . . . . . . 305
5.10 Grade-Two Fluids: Some Open Problems . . . . . . . . . . . . . . . . . 308
5.10.1 Other Boundary Conditions. . . . . . . . . . . . . . . . . . . . . 308
5.10.2 Analysis of Numerical Schemes. . . . . . . . . . . . . . . . . . 309
5.10.3 Rough Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
5.10.4 Limit as a Tends to Zero . . . . . . . . . . . . . . . . . . . . . . 310
6 Short Survey on the Theory of Grade-Three Fluids. . . . . . . . . . . . . 311
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
6.1.1 Formulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
6.1.2 Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
6.2 Variational Formulation. The Time-Dependent Problem . . . . . . . 316
6.3 A Priori Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
6.4 Construction of Solutions by Galerkin’s Method . . . . . . . . . . . . 322
6.4.1 Existence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
6.4.2 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
6.5 The Steady-State Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
6.5.1 A Remarkable Monotonicity . . . . . . . . . . . . . . . . . . . . 327
6.5.2 Existence of a Solution. . . . . . . . . . . . . . . . . . . . . . . . 329
6.5.3 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
7.1 Korn’s Inequality in Lr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
7.2 The Steady Transport Equation . . . . . . . . . . . . . . . . . . . . . . . . 341
7.2.1 Some L2 Results on the Scalar Transport Equation
for Steady Flow. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
7.2.2 Extension to a Steady Transport System in Lp . . . . . . . . 350
7.3 The Time-Dependent Transport Equation . . . . . . . . . . . . . . . . . 356
7.3.1 A Scalar Equation: Existence of Solutions. . . . . . . . . . . 356
7.3.2 A Scalar Equation: Uniqueness . . . . . . . . . . . . . . . . . . 359
7.3.3 The Case of a System with a Lipschitz Velocity . . . . . . 365
7.3.4 Extension to a System in Lp . . . . . . . . . . . . . . . . . . . . 370
7.4 A Family of Riccati Integro-Differential Inequalities. . . . . . . . . . 372
List of Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
Chapter 1
Introduction
Many real fluids exhibit response characteristics that cannot be satisfactorily

described by the classical Navier–Stokes fluid model and such fluids are referred
to as non-Newtonian fluids.1 As the departure from “Newtonian” behavior occurs
in widely disparate ways,2 numerous models have been proposed to describe the
different aspects of non-Newtonian behavior. These models can be broadly classi-
fied under the heads: fluids of the differential, rate, and integral type.3 This book is
mainly devoted to a discussion of the mechanics and mathematics of a special sub-
1 Itwould have been more appropriate to have used the terminology non-Navier–Stokesian fluids
rather than non-Newtonian fluids as the theories of fluid advanced by Newton are quite far removed
from the Navier–Stokes theory. This is not surprising as the notion of a partial derivative had not
yet been invented. Also, Newton advanced more than one theory to describe the motion of fluids.
As Dugas [86] points out, Newton proposed two theories for fluids, the first “a schematic theory
of fluids, which he considered to be formed of an aggregate of elastic particles, which repelled
each other, were arranged at equal distances from each other, and were free”, and “In the second
theory the particles of the fluid are contiguous”. The two theories led to results to the resistance due
to the translation of a solid cylinder in a fluid body of infinite extent to differ by a factor of four.
In [272], Truesdell goes as far as to aver that “Newton’s theories of fluids are largely false”. We
shall use the terminology non-Newtonian fluid as this has become the terminology adopted by the
practitioners in the field. We shall also use the terminology “Newtonian fluid” and “Navier–Stokes
fluid” interchangeably.
2 We are interested in departures from the behavior exhibited by “Newtonian fluids” in what one
understands as laminar flows. Turbulent flows of fluids described in the laminar flow regime by the
Navier–Stokes model remain shrouded in mystery with numerous models being used to describe
different categories of turbulent flows.
3 Some fluid models can be cast as rate type or integral type models, and example of the same being
the eponymous model due to Maxwell [184]. Recently, Rajagopal in [227] (see also Prusa and
Rajagopal [219]) has introduced a new class of fluid models wherein the history of the deformation
gradient and the history of the stress are implicitly related. The class introduced by Rajagopal
in [227] includes most of the fluid models of the differential, rate, and integral type and introduces
new models that can describe phenomena that are not possible to characterize within the context of
models currently in use, an example being the dependence of the material moduli on the invariants
of both the stress, kinematical variables and mixed invariants of the stress and the appropriate
kinematical tensors.
© Springer International Publishing Switzerland 2016 1
D. Cioranescu et al., Mechanics and Mathematics of Fluids
of the Differential Type, Advances in Mechanics and Mathematics 35,
DOI 10.1007/978-3-319-39330-8_1
2 1 Introduction
class of fluids of the differential type. The principal thrust of the book is a discussion
of mathematical results concerning the flows of fluids of grade 2, fluids that form a
subclass of fluids of grade n (see Truesdell and Noll [276] for a definition of fluids
of grade n), with a short discussion of grade-three fluids and also the class of fluids
referred to as Reiner–Rivlin fluids and Bingham fluids.4
While the Navier–Stokes fluid model has been studied extensively, in fact possibly
more than any other fluid model, not much attention has been accorded to the rigorous
mathematical and numerical analysis of general fluids of grade n. These fluids present
interesting challenges to those interested in understanding the underlying physics as
well as those interested in establishing rigorous results concerning the equations
governing the flows of such fluids. While some progress has been made along both
these directions, much remains to be done and the book identifies and tries to fill
some of these lacunae. It is our hope that the book might instigate others to look into
the open issues.
With regard to fluids of grade 2 and grade 3, as well as the Bingham fluid, there
have been studies concerning the thermodynamics and stability of the flows of such
fluids and also some rigorous mathematical results have been established, and many,
but not all, of these results have been collected in this book. We also present some
new mathematical results concerning the existence and uniqueness of flows of these
fluids.
The Cauchy stress in a fluid of the differential type is completely determined
by knowing the current values of a set of frame-indifferent kinematical tensors,
referred to as the Rivlin–Ericksen tensors (Rivlin and Ericksen [246]), the first Rivlin–
Ericksen tensor being twice the symmetric part of the velocity gradient. The Cauchy
stress in fluids of grade n are determined by the first “n” Rivlin–Ericksen tensors,
and the classical Navier–Stokes fluid model is a fluid of grade 1. The Navier–Stokes
fluid model is however incapable of describing many phenomena exhibited by real
fluids such as shear-thinning or shear-thickening (due to the viscosity of the fluid
depending on the shear rate), the viscosity depending on the mean value of the stress,
stress relaxation, nonlinear creep, the development of normal stress differences in
simple shear flows, thixotropy, and a threshold stress before the fluid starts to flow
(referred to as the “yield-stress”), “spurt”, and the usual boundary conditions that are
assumed cannot characterize a variety of interesting phenomena at solid boundaries
such as stick-slip.
While fluids of grade 2 can describe normal stress differences in simple shear flow
and non-linear creep, and fluids of grade 3 possess shear dependent viscosity, and
while the Bingham fluid can describe the phenomenon of “yield”, few fluid models
can describe all the non-Newtonian phenomena that have been observed. Since flows
of most fluids are dominated by one or more of these phenomena, one can justify
studying fluid models which are only capable of describing a few of the plethora of
non-Newtonian phenomena that have been documented. Thus, the fluids considered
4 Ifby a fluid one means a body that cannot resist a shear stress, however small the time interval,
then the terminology “Bingham fluid” is an oxymoron as the appellation is supposed to describe a
body that can resist a shear force until a particular threshold is achieved for the shear stress. Since
the material flows like a fluid once the threshold is exceeded, it is considered a fluid.
1 Introduction 3
in this book are capable of describing some, but not all, of the phenomena that the
Navier–Stokes fluid is unable to do. We discuss the flow characteristics of the fluids
being considered in detail in the following chapters.
The book is divided into two parts. While each part can be considered as being
self-contained, the first being an introduction to non-Newtonian fluid mechanics in
general and fluids of grades 2 and 3 in particular, and the second a discussion of
mathematical and numerically oriented issues concerning the governing equations
for such fluids, the two parts complement and supplement each other to present a
discussion of the mechanics and mathematics of such fluids.
In order to facilitate an understanding of the thermomechanical underpinnings5 of
the models that are being studied, the first part of the book is devoted to a description
of the mechanics, thermodynamics, and the stability of flows of fluids of grade 2
and grade 3. In the first chapter, we describe and analyze a variety of phenomena
exhibited by many real fluids, that were mentioned in the previous paragraph. They
cannot be adequately described by the classical Navier–Stokes fluid model. Such
phenomena are not merely of academic interest but have important consequences
with regard to numerous technological applications. The phenomenon of “die swell”
is a consequence of the normal stress differences that develop during the flow of a
large class of non-Newtonian fluids and determines the design procedure for dies that
are used in the manufacture of practically all films and tapes, while the phenomenon of
“stick-slip” determines the surface roughness of most extruded polymeric products.
We shall not discuss the technological relevance of these phenomena in Chapter 2,
however suffice it is to say that each of the phenomena delineated in the previous
paragraph has significant technological consequences.
The second part of the book is concerned with the development of rigorous mathe-
matical results concerning the equations governing the motion of a family of fluids of
the differential type. We decided to restrict our attention to the presentation of math-
ematical results concerning Reiner–Rivlin fluids and fluids of grade 2 and grade 3,
in view of the fact that rate type fluids such as Oldroyd-B fluids have been the object
of a thorough discussion in a previous book by Fernández-Cara et al. [97], and very
little by way of rigorous mathematical results are available with regard to the equa-
tions governing the motion of non-linear fluids of the integral type (there are several
treatments of the linear viscoelastic fluid and thus we decided to not discuss the same
in this book).
After a separate chapter recalling, mostly without proof, the mathematical notions
required in this work, the second part starts with a chapter devoted to a discussion of
the mathematical issues of fluids that are defined through a generalization of fluids
of complexity 1 by defining constitutive relationship via a sub-differential and by
extending the analytical ideas developed by Duvaut and Lions in [91], Ladyzhen-
skaya in [155], Brezis in [49] and Cioranescu in [62]. Then the remainder of the book
5 Itwould be fair to say that the readers are introduced to a particular thermodynamic perspective
as there is considerable disagreement concerning in what form the second law of thermodynamics
is to be enforced. We shall not get into a discussion of the various perspectives, their merits and
demerits. We shall merely present the consequences of the Clausius-Duhem inequality being used
as the surrogate for the second law of thermodynamics.
4 1 Introduction
is devoted to the mathematical analysis of grade 2 and to a lesser extent of grade-three

fluids. Both grade-two and grade-three fluids are interesting examples of fluids of
the differential type that can be elegantly and efficiently analyzed using Galerkin’s
method with a special basis of eigenfunctions. This approach has been chosen for
exposition, rather than the reproduction of a more familiar fixed-point argument,
with the hope that the interested reader will favor this direction when studying com-
plex fluids. However, we have also kept in mind the practical aspects of numerical
approximation. As this special basis of eigenfunctions is not easily discretized, a
substantial part of the study of grade-two fluids is devoted to splitting the problem
into two simpler systems of equations, each one being easily approximated. We have
not expanded this idea for the grade-three model because the same splitting does
not lead to sufficiently simple equations. Thus, an efficient splitting for grade-three
fluids is still an open problem.
Finally, the proofs of a number of useful results are collected in an appendix.
These are mostly dedicated to systems of transport equations, under low regularity
assumptions on the domain and the driving velocity, that are frequently encountered
in the analysis of complex fluids.
To our knowledge, this is the first time, results concerning both the thermome-
chanics and mathematical analysis of Reiner–Rivlin fluids and fluids of grades 2 and
3 are collected under a single book and we hope that the material presented here will
encourage others to fill in the lacunae that exist with regard to both mathematical
and numerical results concerning the fluids studied in here.
Chapter 2
Mechanics
In this chapter, the reader will be introduced to a variety of non-Newtonian phenom-

ena exhibited by real fluids, namely stress relaxation, nonlinear creep, shear-thinning
and shear-thickening, thixotropy, development of normal stress differences in simple
shear flows, yield, etc. This is followed by a section dedicated to the basic kinematical
definitions. The next subsection deals with the definition of frame-indifference and
the restrictions that are a consequence of its requirement. In the following subsection
the balance laws, and the Clausius–Duhem inequality1 (which is interpreted in this
book as the second law of thermodynamics) are recorded and this is followed with a
subsection on constitutive relations wherein the reader is introduced to the fluid of
the differential type and to its special subclasses such as fluids of complexity n, and
fluids of grade n. In this chapter, the reader is also introduced to the three categories
of fluids, namely the differential, rate, and integral type fluids. The classical Navier–
Stokes model is a fluid of grade 1 and can be obtained as a special subclass of fluids
of grade 2 by setting appropriate terms that appear in the constitutive relation to be
zero.
2.1 Introductory Remarks
It is hard to define precisely what one means by a fluid. David Goldstein [124] aptly
remarks “Precisely what do we mean by the term liquid? Asking what is a liquid is
like asking what is life; we usually know it when we see it, but the existence of some
doubtful cases makes it hard to define precisely.” In fact, the situation is far worse
than that described by Goldstein. In the pitch drop experiment set up by Professor
1A thermodynamic framework has been put into place by Rajagopal and Srinivasa in [234] that
appeals to the maximization of the rate of entropy production to provide a basis for fluids of grade
2 and the same framework can be used to develop the model of the grade 3 fluid and the Bingham
fluid.

DOI 10.1007/978-3-319-39330-8_2
6 2 Mechanics
Parnell in 1927 that yet continues to date, pitch was poured into a funnel with a
sealed bottom and allowed to consolidate. The stem was cut in 1930. The first drop
fell in December 1938, the second in February 1947, and the sixth in April 1979 (see
Edgeworth et al. [93] for details). Thus, a person observing the pitch in the funnel
for even an hour would conclude that the material did no flow and was a solid, while
were one to wait 8 years one would conclude that the pitch was a fluid that flowed.
The same pitch, when hammered shatters like a brittle solid. Thus, whether a body
behaves in a fluid-like manner depends on the time and length scale of observation,
as well as the magnitude of the forces a body is subject to.
Maxwell [184] captures the quintessential character of a fluid when he writes: “In
the case of viscous fluid it is time which is required, and if enough time is given, the
very smallest force will produce a sensible effect, such as would require a very large
force if suddenly applied.” Maxwell’s comments bring into focus the most important
factor in defining fluid-like and solid-like behavior: the notion of time scales, length
scales and force scales. What are the time, length, and force scales of interest? Are we
interested in motions discernible to the naked eye in nanoseconds? Are we interested
in motions of the order of nanometers over a period of years? Based on which of the
two above questions we are interested in, the response of a body would be considered
fluid-like or solid-like. We shall not get into a lengthy discussion of this issue here.
By a fluid-like body, we mean a body which moves in a manner discernible to the
naked eye for a time scale of observation that is meaningful for the forces under
consideration.
The Oxford English Dictionary [1] defines a fluid as “Having the property of
flowing, …consisting of particles that move freely along themselves so as to give
way before the slightest pressure, ….” The first part of the above definition does not
convey much meaning as “flowing” is regarded as a property of a fluid, and with
regard to the second part, it is now commonly recognized that it is the inability to
withstand a shear stress rather than “pressure” that is characteristic of a fluid. Of
course, one cannot get rid of the time and length of scales of observation or the
magnitude of the shear stress in determining the inability of the body to withstand
the shear stress. We shall not get into a more detailed discussion of the nature of
fluids.
Details concerning the historical development of fluid mechanics can be found in
the authoritative treatises by Tokaty [269] and Truesdell [270]. Here, we shall briefly
discuss the main contributions of Newton and those that followed. In his immortal
Principiae in 1687, Newton [196] discusses in detail the resistance occurring due
to the motion of fluids. He states that “The resistance arising from the want of
lubricity in parts of fluids is, other things being equal, proportional to the velocity
with which the parts of the fluid are separated from one another.” Such a constitutive
description which provides the genesis for the fluid model that is popularly referred to
as a Newtonian fluid does not provide a clear mathematical definition of the model.
Navier [193] in 1823, Poisson [213] in 1831 and Stokes [259] in 1845, greatly
generalized the ideas of Newton to arrive at a model that bears the names of the first
and last of these authors. It is interesting to note that very different paths were taken
by these authors to arrive essentially at the same result. As Stokes [259] observes:
2.1 Introductory Remarks 7
“I afterwards found that Poisson had written a memoir on the same subject, and on
referring to it, I found that he had arrived at the same equations. The method which
he employed was however so different from mine that I feel justified in laying the
latter before this Society.” In a footnote he adds: “The same equations have also been
obtained by Navier in the case of an incompressible fluid (Mémoire de l’Académie,
t.VI, p. 389), but his principles differ from mine still more than Poisson’s.” The
model that Stokes derived in 1845, satisfied the requirements of frame-indifference
and isotropy.
Stokes’ work is all the more remarkable that he arrived at a model far more general
than that which bears his name partially, in that he recognized that the viscosity could
possibly depend on the pressure. He then provides a justification for a class of flows
where one could view viscosity as a constant: “Let us consider in what cases it is
allowable to suppose μ to be independent of pressure. It has been concluded by
Du Buat from his experiments on the motion of water in pipes and canals, that
the total retardation of the velocity due to friction is not increased by increasing
the pressure… I shall therefore suppose that for water, and by analogy for other
incompressible fluids, μ is independent of pressure.” On the basis of this additional
assumption, Stokes arrives at the equations of motion for incompressible fluids like
water, which in our current notation would read2
dv
μ Δv + b = grad p + , (2.1.1)
dt
div v = 0. (2.1.2)
The above equations of motion are a consequence of assuming that the fluid is in-
compressible and assuming furthermore a constitutive relation for the Cauchy stress
that is linear in the symmetric part of the velocity gradient. They are referred to as
the incompressible Navier–Stokes equations and they have been quite successful in
describing the laminar flow of many liquids such as water. However, the classical
linearly viscous fluid model (Newtonian fluid model, Navier–Stokes fluid model)
cannot adequately describe the laminar response of many polymeric liquids, biolog-
ical fluids, foams, and slurries. Their departure from the Newtonian fluid response
is manifest in a variety of ways, namely the ability of the fluid to
1. shear-thin or shear-thicken,
2. exhibit thixotropy,
3. allow stress relaxation,
4. creep in a nonlinear manner,
2 Thisis the vector version of equation (13) in Stokes’ paper [259]. We have not yet introduced the
notation that will be followed in this book. Here, μ is the viscosity, the density, v the velocity,
p the pressure, b the specific body force, and Δ denotes the Laplacian, grad the gradient, div the
d
divergence and dt the material time derivative, see Section 2.3 for more details.
8 2 Mechanics
5. develop normal stress differences,

6. and exhibit a threshold for the shear stress before it starts to flow (commonly
referred to as “yielding” behavior).
3
The last of the above characteristics of non-Newtonian fluids contradicts our earlier
motivation that a fluid cannot sustain a shear stress. Thus condition (6) should be
interpreted as there being a threshold for the flow with regard to some reasonable
time and length scale of observation, and some specified range of forces.
Many fluids exhibit one or more of these characteristics, and some all of them.
In order to describe the response of such complex fluids a plethora of fluid models
have been developed, all of them being referred to as non-Newtonian fluid models.
We shall discuss the different characteristics displayed by non-Newtonian fluids
in Section 2.2. The delineation of a viscoelastic material as a solid or a fluid is not an
easy matter. We can at best classify such materials as being fluid-like or solid-like. As
we remarked earlier, such categorizations depend on inherent time scales associated
with the material in comparison with the time scale of the observation. In this book,
we are concerned with the fluid-like response of materials. It is quite possible for a
viscoelastic liquid to display more “elastic response” than a viscoelastic solid, and
for a viscoelastic solid to display more “dissipative” response than a viscoelastic
liquid. Maxwell [184] discussed this issue with a clarity that is not to be found even
in the most current texts. He was able to draw a clear distinction between the response
of a soft solid and a viscoelastic fluid. He remarked: “What is required to alter the
form of a soft solid is a sufficient force, and this, when applied produces its effect at
once. In the case of a viscous fluid it is time which is required, and if enough time is
given, the very smallest force will produce a sensible effect, such as would require
a very large force if suddenly applied. Thus a block of pitch may be so hard that
you cannot make a dent in it by striking it with your knuckles; and yet in course
of time, it will flatten itself by its own weight, and glide down-hill like a stream of
water.” Maxwell is making the distinction between the instantaneous response of
pitch (asphalt/bitumen) and its long-time behavior.
We shall discuss the notion of stress relaxation in some detail later. A measure of
how the relaxation takes place is given by a characteristic time associated with the
body, referred to as the relaxation time. An important nondimensional number that
one comes across while studying the mechanics of viscoelastic fluids is the DEB-
ORAH4 number that is the ratio of a characteristic time associated with the body,
3 There are several phenomena associated with the response of non-Newtonian fluids such as “elas-
tic turbulence,” “drag reduction due to turbulence,” “secondary flows in turbulent flows of non-
Newtonian fluids,” etc. that are a consequence of some of the above-mentioned response charac-
teristics of non-Newtonian fluids coupled with the effects of turbulence. As turbulence, to date, has
defied proper understanding, even within the context of fluids that in the laminar range are described
by the classical Navier–Stokes fluid, we shall not discuss such characteristics in the book.
4 If one waits sufficiently long, then one could discern motion in nearly all bodies due to the effect of
gravity. Deborah [80] remarks in the Old Testament that “Even mountains quaked in the presence
of the Lord.” Usually, Deborah’s statement is translated to read “Even mountains flowed in the
presence of the Lord,” but this translation does not seem to be felicitous (see Rajagopal [228] for a
detailed discussion concerning this issue).
to a time interval associated with the process. In considering problems involving

non-Newtonian fluids, we will come across a variety of nondimensional numbers
in addition to the usual Reynolds number that one comes across while considering
isothermal flows of the Navier–Stokes fluid. Depending on which of these nondi-
mensional numbers are dominant, we can have a variety of response characteristics.
Many non-Newtonian fluids have memory, that is the stress in the fluid at current
time can depend on how the fluid has deformed in the past. The incompressible
Newtonian fluid on the other hand has no memory in that the stress in the fluid is
completely determined to within an indeterminate spherical stress by the current
value of the symmetric part of the velocity gradient.
Models that have been developed to describe the non-Newtonian response of fluids
can be broadly classified into three categories
1. fluids of the differential type,
2. fluids of the rate type,
3. fluids of the integral type.
Fluids in which the stress is determined by the velocity gradient and its various
higher time derivatives are called fluids of the differential type. In incompressible
fluids of the differential type, as soon as the motion ceases, the stress reduces to
an indeterminate spherical stress. Unlike fluids of the differential type, in which the
stress is given explicitly as a function of the velocity gradient and its higher time
derivatives, in rate type fluid models there is an implicit relationship between the
stress and its higher time derivatives.5 Integral type models are those in which the
Cauchy stress at the current time is given in terms of an integral over the past time
of the history of the relative deformation gradient. Some rate type models can be
expressed as integral type models, but not all rate type models can be expressed in
such a manner. In general, rate type models define a class of models rather than just
one model (cf. Truesdell and Noll [276], p. 95).
As we remarked earlier, Stokes recognized that the viscosity of a fluid could
depend upon the pressure. There is considerable experimental evidence indicating
that this is indeed the case and a thorough discussion of the experimental results prior
to 1930 can be found in the authoritative treatise by Bridgman [52] on the physics
of high pressure. As early as 1893, Barus [21] suggested that the viscosity is related
to the pressure through
μ( p) = μ0 exp(α p), α > 0.
Later, according to Bridgman [52], Andrade proposed the following relationship

between the viscosity, pressure, density, and temperature:

1 B
μ( p, , θ ) = A 2 exp ( p + D 2 ) ,
θ
5 Recently, it has been shown that the Cauchy stress in many rate type models can be expressed
as an elastic response from an evolving natural configuration, i.e., the stress can be expressed in
terms of a Cauchy–Green stretch tensor which obeys an evolution equation (see Rajagopal and
Srinivasa [235]).
10 2 Mechanics
where A, B, and D are constants, θ denotes the temperature and the density (see
also the paper by Andrade [11] wherein the pressure dependence of the viscosity is
discussed in detail).
Various other relationships between the viscosity and pressure have been pro-
posed. At very high pressures, experiments suggest that the fluid is close to under-
going glass transition and equation such as those suggested by Barus and Andrade
seem to be inappropriate. For a discussion of the relevant issues, we refer the readers
to some of the experimental work in the area (Cutler et al. [72], Griest et al. [126],
Johnson and Cameron [143], Johnson and Greenwood [144], Johnson and Tevaar-
werk [145], Bair and Winer [19], Roelands [248], Paluch et al. [209], Irving and
Barlow [140], Bender et al. [25] and Bair and Kottle [18]).
In liquids such as water, the change in density due to pressure seems to correlate
reasonably well with the relation (see Dowson and Higginson [84]),

0.6 p
= 0 1 + ,
1 + 1.4 p
where 0 is the reference density (density where the pressure is zero).

While the density changes by about 3 to 4 percent as the pressure changes from
2 to 3 GPa, the viscosity changes by approximately 108 percent. Thus, it would be
reasonable that fluids such as water and organic liquids be modeled as incompressible
fluids with pressure dependent viscosities when the pressure varies over a reasonably
wide range. However, under operating conditions as those encountered in flows in
pipes and channels, the viscosity varies ever so slightly that it can be considered a
constant.
When the viscosity of the fluid depends on the pressure, the balance of linear
momentum reduces to

∂v ∂v T dv
μ( p)Δv + b + + grad μ( p) = grad p + ,
∂x ∂x dt
div v = 0.
The variation of the viscosity with pressure is of great significance in a variety of

important technological applications such as elastohydrodynamics (see Szeri [261]).
Recently, Rajagopal and Szeri [238] have shown that the additional term involving
the gradient of the viscosity in the above equation cannot be neglected while con-
sidering problems such as elastohydrodynamics. It is possible that the viscosity of
an incompressible fluid can depend on both the pressure and the symmetric part of
the velocity gradient. These fluids cannot be characterized by an explicit relationship
between the stress and the symmetric part of the velocity gradient; the relationship
is implicit. Such fluid models are also non-Newtonian models and they could be of
the differential, rate, or integral type. We shall not consider these models here, the
interested reader can find a detailed discussion of the mathematical properties of
the equations governing the flows of such fluids in the review article by Málek and
Rajagopal [170].
In addition to the above types of models, other forms of models based on the
notion of fractional derivatives, the notion of the conformation tensor, etc. have been
proposed. Here, we shall be concerned with incompressible fluids of the differential
and rate type, wherein the stress is expressed in terms of the kinematical quantities.
In recent years, there has been considerable amount of interest in the study of
non-Newtonian fluids of the differential type (the Navier–Stokes fluid is also a fluid
of the differential type). Despite such extensive studies, a considerable amount of
confusion and lack of clarity seems to prevail in the interpretation of the results
concerning the flows of fluids of the differential type. Though repeated attempts at
clarifying the issues have not seemed to have put paid to erroneous and inaccurate
conclusions, we shall endeavor to set the records straight. As there has been much
progress in the past decade, our task might be easier in understanding the response
of these fluids.
There has been a great deal of controversy surrounding the thermodynamics of
such fluids. In the two decades that followed the work of Coleman and Noll [69], in
which they used the Clausius–Duhem inequality to obtain restrictions on the forms
of constitutive relations that are allowable, it became common practice in continuum
mechanics to use such a procedure. However, for a variety of reasons, serious reser-
vations were expressed against using the procedure, the most compelling objection
being the appeal to the body being subject to arbitrary processes to obtain these
restrictions. As no single constitutive relation is expected to hold for all arbitrary
processes that any body is subject to, any specific constitutive relation that is used
will be valid for only a certain class of processes beyond which the constitutive re-
lation is not expected to hold. We believe that such an objection is well taken. The
main problem stems from trying to obtain necessary and sufficient conditions that
will ensure that the Clausius–Duhem inequality should hold. If on the other hand,
one is only interested in sufficient conditions that will guarantee that the Clausius–
Duhem inequality holds, then one can overcome the objections that are raised. In
special cases, the procedure used by Coleman and Noll [69] led to sensible results
and in the absence of an alternate procedure to obtain restrictions on constitutive
relations, it gained credence. For instance, in the case of a Newtonian fluid, such a
procedure would lead to the viscosity being nonnegative, a perfectly reasonable re-
sult, reasonable in that such a fluid exhibited reasonable agreement with experiments,
its stability characteristics, and its mathematical properties.
Another serious drawback in using the Clausius–Duhem inequality, as advocated
by Coleman and Noll [69], stems from the fact that the radiation is eliminated from
the reduced dissipation equation by substituting for it from the energy equation.
Thus, the second law, as interpreted in the above approach, places no restriction
whatsoever on the radiation. While, this may not be of much consequence in many
problems involving the flow of fluids, it is of great significance in other problems.
From a philosophical standpoint, ignoring radiation altogether, or for that matter
being incapable of recognizing the restrictions that need to be placed on the radiation,
is an unacceptable situation. We shall not get into a detailed discussion of this issue
here. The interested readers can find details in Rajagopal and Tao [239].
12 2 Mechanics
However, controversy surfaced concerning the consequences of using Clausius–

Duhem inequality and the apparent predictions of experiments for a special subclass
of fluids of the differential type, namely fluids of grade two. We shall not get into
a discussion of these issues but refer the reader to Dunn and Rajagopal [88], where
the relevant issues are discussed at length. Suffice it to say that fluids of grade two
that satisfy the Clausius–Duhem inequality and whose specific Helmholtz potential
is a minimum in equilibrium exhibit reasonable response. Of course, it is perfectly
acceptable for one to yet object to the use of the Clausius–Duhem inequality on the
grounds that one cannot appeal to arbitrary processes, in which case we have to make
appropriate restrictions to the constitutive relation that guarantees that the second
law is met (i.e., we can a priori assume constitutive relations that guarantee that the
entropy production is nonnegative. Our constitutive assumptions are “sufficient” to
assume that the second law is met).
We shall not be wedded to the consequences of the restrictions due to the Clausius–
Duhem inequality. However, we shall show that noncompliance with thermodynamic
considerations leads to fluids that exhibit unacceptable physical characteristics: in-
stability of the rest state of the fluid. Thus, while from a general philosophical point
of view, the thermodynamic procedure can be rightfully called into question, in the
case of some special models of fluids of the differential type that are considered, they
do lead to models that exhibit reasonable physical characteristics.
This book is devoted to mechanics issues, and in some cases mathematical issues,
concerning the flows of incompressible fluids of differential and rate type. Some
numerical issues that arise during the solution of the equations that govern the flows
of these fluids are addressed in the reference by Girault and Hecht [112]. Among
others, we shall examine two important questions: the boundary conditions and the
development of boundary layers.
An important mathematical issue that surfaces when we consider the flows of
a Navier–Stokes fluid instead of an Euler fluid is the necessity for an additional
boundary condition. In the case of an Euler fluid, we need to only require that the
normal component of the velocity be zero at an impervious boundary while we need
the adherence condition in the case of a Navier–Stokes fluid. A similar situation,
i.e., a need to augment boundary conditions, presents itself when we consider fluids
of higher grade.
Unlike the flows of a classical Newtonian fluid which lead to second order partial
differential equations for which the “no-slip” boundary condition is sufficient for
well-posed problems, flows of fluids of grades two and three lead to partial differ-
ential equations of third order and in general the “no-slip” boundary condition is
inadequate to guarantee uniqueness of solutions. In fact, an infinite number of solu-
tions have been exhibited for several simple boundary value problems (cf. Rajagopal
and Gupta [230], Rajagopal and Kaloni [232]). Not only is the “no-slip” condition
insufficient for certain flows (those for which v · n = 0), but also there are many
flows of fluids wherein the fluid “stick-slips” at the boundary. It has been experi-
mentally observed (cf. Hatzikiriakos and Dealy [131], Ramamurthy [240], Kraynik
and Schowalter [152]) in the flow of many polymer melts that the fluid adheres to
the boundary, provided that the pressure gradient (the corresponding shear stress
at the wall) is below a certain critical value. However, when the pressure gradient
is increased beyond this critical value, the fluid starts to slip at the boundary. This
sticking and slipping at the boundary leads to surface instabilities in an extruded melt
producing a phenomenon popularly referred to as the “shark-skin effect” (cf. [131]).
One of the cornerstones of classical fluid mechanics is the boundary layer ap-
proximation developed by Prandtl [216, 217] for the Navier–Stokes equation. The
physical basis for the boundary layer approximation is the experimentally observed
fact that the vorticity, in the flow of fluids such as water along a solid boundary,
is confined in a narrow region adjacent to the boundary, this region being referred
to as the boundary layer. This confinement of vorticity adjacent to the solid wall
leads to a great simplification, namely the flow outside the boundary layer is approx-
imated as the flow due to an Euler fluid, while the flow inside the boundary layer
is that of a Newtonian fluid (the equations being further simplified). In the case of
a Newtonian fluid, boundary layers arise at sufficiently high Reynolds numbers, the
layer becoming more pronounced (narrower) as the Reynolds number increases. In
non-Newtonian fluids, confinement of vorticity adjacent to solid boundaries is not
necessarily a consequence of the Reynolds number being sufficiently large. In certain
non-Newtonian fluids, it is possible to have confinement of vorticity adjacent to the
boundary in the limit of Reynolds number tending to zero. In other non-Newtonian
fluids, boundary layers can develop much in the same manner as in the case of New-
tonian fluids, at high Reynolds number. However, a much richer “class” of boundary
layers are possible in non-Newtonian fluids: there could be confinement (concentra-
tion) of various kinematical quantities due to elastic effects, shear-thinning, etc. We
shall illustrate these issues via specific examples.
2.2 Non-Newtonian Behavior
Let us discuss the departure from Newtonian response that we mentioned in the
introduction (see also Huilgol [138] and Málek and Rajagopal [169]).
2.2.1 Shear-Thinning and Shear-Thickening
Consider a simple shear flow, i.e., the velocity field v has the form
v = u(y)i, (2.2.1)
in a cartesian coordinate system. In the case of a Newtonian fluid, the shear stress
Tx y is directly proportional to the rate of shear u (y), the constant of proportionality
being the viscosity μ
Tx y = μ κ,
where κ = u (y) .
14 2 Mechanics
Fig. 2.1 Shear dependent

behavior of non-Newtonian
fluids shear thicknening Navier-Stokes
shear thinning
shear stress (τ )
limit case
nonmonotone
shear rate (κ̇)
In a general simple fluid (cf. Noll [197]), it can be shown that
Tx y = τ (κ) := [μ̃(κ)]κ, (2.2.2)
where μ̃(κ) is the generalized viscosity and we shall see later that τ (κ) is one of
the three viscometric functions. It is possible that the variation of the shear stress
with the shear rate is such that the derivative of the generalized viscosity μ̃(κ) with
respect to κ is negative. Such fluids are called shear-thinning fluids. On the other
hand, if the derivative of the generalized viscosity with respect to κ is positive, it is
called a shear-thickening fluid. Thus, the variation of the shear stress with respect to
the shear rate in a simple shear flow has one of the forms depicted in Figure 2.1.
Fluids with shear rate dependent viscosity are used widely in the chemical en-
gineering field, ice mechanics, geology, and hemodynamics (see Bird et al. [39]).
Issues concerning the existence, regularity of solution and stability for fluids with
shear rate dependent viscosity can be found in Málek et al. [171].
2.2.2 Thixotropy
Thixotropy is not a very well-understood phenomenon and is usually confused with

shear-thinning. The phenomenon of thixotropy is exhibited by suspensions of col-
loidal and non-colloidal particles that form a flocculated system, and it is also ob-
served in cross-linked gels. The main characteristic of thixotropic fluids is the change
in material properties, such as viscosity and elasticity, with time. Whereas in a shear-
thinning fluid, the viscosity changes with the shear rate, in a thixotropic fluid the
viscosity can decrease with time while the shear rate remains constant. Such changes
in a thixotropic material are a consequence of structural changes in the underlying
microstructure of the fluid. In view of this, usually a scalar parameter referred to as
the “structure parameter” is introduced. The evolution of the “structure parameter”
2.2 Non-Newtonian Behavior 15
with time is given by a kinetic equation that is derived from microstructural consid-
erations, but is in most instances ad hoc. Many thixotropic fluids also exhibit “yield”
and the properties of such bodies are sensitive to the effect of temperature. In some
thixotropic fluids changes in temperature can lead to crystallization which in turn
affects the flow characteristics of the fluid. Another important aspect that needs to
be taken into account while describing thixotropic fluids is the effect of Brownian
motion, which in turn is affected by the temperature. Hence, in order to describe the
response of thixotropic fluids, one needs to consider a fully thermodynamic frame-
work.
The detailed reviews by Mewis and Wagner [190] and Barnes [20] provide a
description of the main features of thixotropic fluids and also copious references to
the subject. De Souza Mendes, Thomson and coworkers have developed models to
describe the response of thixotropic fluids (see de Souza Mendes [75, 76], de Souza
Mendes and Thomson [78, 79]). Many of these models, while they are very useful
in that they have been able to explain reasonably well several of the experimental
observations in one-dimensional shear flows, are ad hoc and mostly restricted to
one-dimension, and none of them have a rigorous thermodynamic underpinning.
Recently, de Souza Mendes et al. [77] developed a thermodynamic framework to
generate properly invariant three-dimensional models, which when restricted to one
dimension can capture the main features of the one-dimensional models developed
by de Souza Mendes [75, 76] and de Souza Mendes and Thomson [78, 79]. The
thermodynamic analysis by de Souza Mendes et al. [77] does not explain all the
features exhibited by thixotropic fluids, however, it can serve as the starting point
for the development of a more complete model. Moreover, most of the experiments
that have been carried out thus far are one-dimensional shear flow experiments, thus
before a complete model can be put into place, it is necessary to carry out experiments
in more complex domains as well as a more thorough experimental study of the
dependence of the response characteristic of thixotropic fluids on temperature.
We shall not discuss thixotropy in further detail as the fluids that will be considered
in this book do not exhibit thixotropy.
2.2.3 Stress Relaxation
Next, consider a fluid that is subject to a step shear strain of the following form:

0 for all t < 0,
εx y =
ε0 for all t ≥ 0.
In certain fluids, the shear stress that is required to maintain the stresses would
decrease and tend to zero with time. As the shear stresses required to maintain
the strain decrease, the fluid is said to stress relax. In the case of certain solids, a
similar type of response is elicited; however, the shear stress decreases to a nonzero
asymptotic value (Figure 2.2).
16 2 Mechanics
Fig. 2.2 Stress relaxation in viscoelastic bodies
Many materials which stress-relax also exhibit instantaneous “elastic” response.

A Newtonian fluid neither has an instantaneous “elastic” response nor does it have the
ability to stress relax. In a Newtonian fluid, in response to the step strain there would
be a Dirac measure for the shear stress. We shall see later that fluids of the differential
type share this feature in common with the Newtonian fluid; they are incapable of
stress relaxation and they are incapable of instantaneous elastic response.
2.2.4 Creep
Consider a fluid that is subject to a constant shear stress. If the fluid were Newtonian,
the shear strain would increase linearly with respect to time. However, there are many
materials in which the variation of the strain with respect to time is nonlinear. When
certain materials are subject to a shear stress of the form
⎧
⎨ 0 if t < 0,
Tx y = T0 if 0 ≤ t < α,
⎩
0 if t ≥ α,
they typically respond in the manner shown in Figure 2.3.

In a viscoelastic fluid capable of an instantaneous elastic response, the strain
will asymptotically tend to a constant value while in a viscoelastic solid capable of
an instantaneous elastic response, the strain will asymptotically tend to zero. This
phenomenon is called “creep”.
2.2.5 Normal Stress Differences
Once again, let us consider a flow of the form (2.2.2). In the case of incompressible
Newtonian fluids where the viscosity μ is constant, we have
Fig. 2.3 Nonlinear creep
Tx x = Tyy = Tzz = − p,
Tx y = μκ, Tx z = Tyz = 0.
Thus
Tx x − Tyy = 0, Tx x − Tzz = 0.
We will see later that in many fluids of differential type, we have
Tx x − Tyy = 0, Tx x − Tzz = 0,
and thus normal stress differences develop in a simple shear flow. These normal stress
differences are the cause for many interesting phenomena such as “die-swell,” “rod-
climbing” (see Figure 2.5), and secondary flows in pipes of noncircular cross-sections
(see Figure 2.4).
Fig. 2.4 Flow in a pipe of

noncircular cross-section:
vortices due to normal stress
differences
18 2 Mechanics
Fig. 2.5 Rod-climbing due

to normal stress differences
The development of normal stresses are common to many types of nonlinear

response of materials. In nonlinear solids, the Poynting effect, first studied by
Wertheim (see Bell [24]), and later by Poynting [214] and Thomson (Lord Kelvin)
and Tait [149], is a consequence of unequal normal stresses developing due to shear.
A similar phenomenon in non-Newtonian fluids is called the Weissenberg effect
(cf. Weissenberg [282]), and in granular materials such a phenomenon was observed
and characterized by Reynolds [243]. Most interestingly, simple shear in nonlinear
materials is characterized by the normal stresses as well as the shear stresses.
2.2.6 Yield
There are certain materials which flow only after a certain threshold for the shear
stress is reached; such materials are called “Bingham plastic fluids” and the threshold
value for the stress is called the yield stress. However, as a fluid is a material that
cannot resist a shear stress given sufficient amount of time, the existence of such a
threshold is merely due to the time scale of the experiment not being sufficiently
large. Since we are at times forced to work within certain time scales, it is found
convenient to describe the material as a fluid, but with a “yield” condition (see
Figure 2.6). But we must take great care to recognize that existence of a yield stress
is only a convenient approximation and the graph in this figure is that of a set-valued
function.
At this juncture, it is worthwhile recalling the distinction between a fluid and a
solid that was offered by Maxwell [184], as it is clearer than any of the distinctions
made in the works that followed. Maxwell states: “Thus, a tallow candle is much
softer than a sealing wax; but if the candle and the stick of sealing wax are laid
horizontally between two supports, the sealing wax will in a few weeks in summer
Fig. 2.6 A material with a

yield-stress
bend with its own weight, while the candle remains straight. The candle is therefore
a soft solid and the sealing wax a very viscous fluid.”
2.3 Preliminaries
2.3.1 Kinematics
Let κ R (B) and κt (B) denote a reference and current configuration of the abstract
body B, in a three-dimensional euclidean space. A motion χ κ R of the body is a one-
to-one mapping that assigns to each point X ∈ κ R (B) a point x ∈ κt (B) at time t,
i.e.,
x = χ κ R (X, t). (2.3.1)
If the choice of the reference configuration changes, the form of the motion will also
change. The velocity v is defined by
∂
v := χ , (2.3.2)
∂t κ R
and the deformation gradient F κ R is defined through
∂ ∂x
F κ R := χ = . (2.3.3)
∂ X κR ∂X
Now, for any function ϕ defined on B × RI , we can define a function ϕ̃ on κ R (B) × R

I
and a function ϕ̂ on κt (B) × R
I such that
ϕ( P, t) = ϕ̃(X, t) = ϕ̂(x, t). (2.3.4)

20 2 Mechanics
Suppose ϕ̃ : κ R (B) × R
I → F1 and ϕ̂ : κt (B) × R
I → F2 , where F1 and F2 are inner
product spaces. We use the following notation:
∂ϕ ∂ ϕ̃ ∂ ϕ̃
:= , Grad ϕ := , (2.3.5)
∂t ∂t ∂X
dϕ ∂ ϕ̂ ∂ ϕ̂
ϕ̇ = := , grad ϕ := . (2.3.6)
dt ∂t ∂x
The specification in terms of (X, t) is usually referred to as Lagrangian specification

(used earlier by Euler) and the specification in terms of (x, t) is usually referred to as
Eulerian specification (used earlier by d’Alembert). Thus, with the above notation:
d
v= χ , F κ R = Grad χ κ R . (2.3.7)
dt κ R
The derivative dϕ
dt
is called the material time derivative or the substantial time deriv-
ative of the quantity ϕ. The expression ∂ϕ ∂t
is called the local time derivative and
denotes the variation with time in the quantity at a fixed location x. The material
time derivative denotes the variation in the quantity as observed by a person moving
with the particle. A flow is said to be steady if ∂ϕ
∂t
= 0, for all quantities ϕ associated
with the flow.
The velocity gradient L is defined by

∂vi
L = grad v = . (2.3.8)
∂x j i, j
We also denote by
∂v1 ∂v2 ∂v3

div v = tr(grad v) = + + , (2.3.9)
∂ x1 ∂ x2 ∂ x3
Div v = tr(Grad v). (2.3.10)
We shall assume that F κ R is non-singular. For convenience, we shall suppress the

suffix κ R from all the quantities henceforth. It follows that
L = Ḟ F −1 . (2.3.11)
Let ξ denote the position occupied by X at time τ . Then
ξ = χ (X, τ ) = χ (χ −1 (x, t), τ ) := χ t (x, τ ). (2.3.12)
The mapping χ t is referred to as the relative deformation, namely the motion of the
body with respect to the current configuration as the reference. Thus,
2.3 Preliminaries 21
∂χ t
v= |τ =t . (2.3.13)
∂τ
The relative deformation gradient F t (x, τ ) is defined through
∂χ t
F t (x, τ ) = = grad χ t = F(X, τ )F −1 (x, t), (2.3.14)
∂x
and the gradient of the velocity L is related to the relative deformation gradient F t (τ )
through
L(x, t) = Ḟ t (x, t).
It follows from the Polar Decomposition Theorem that the deformation gradient F
can be expressed as
F = R U = V R, (2.3.15)
where R ∈ O, the group of orthogonal transformations6 and U, V are positive

definite and symmetric linear transformations; R is called the rotation and U and
V the stretch tensors. Similarly, the polar decomposition theorem of the relative
deformation gradient F t (x, τ ) leads to
F t (x, τ ) = Rt (x, τ )U t (x, τ ) = V t (x, τ )Rt (x, τ ),
U t (x, τ ) and V t (x, τ ) are the right and left relative stretch tensors and Rt (x, τ ) is
the relative rotation tensor.
The Cauchy–Green tensors C and B are defined through
C = FT F = U 2, (2.3.16)
B = F FT = V 2. (2.3.17)
The Green–Saint Venant and the Almansi–Hamel strains, E and e are introduced
through
1
E = (C − I), (2.3.18)
2
1
e= (I − B −1 ), (2.3.19)
2
where I is the identity tensor. We can now proceed to define the relative Cauchy-
Green tensors in a manner analogous to (2.3.16) and (2.3.17)
C t (x, τ ) := [F t (x, τ )]T [F t (x, τ )], (2.3.20)
6 More precisely, R ∈ O+ , i.e., det R = 1. This follows from the fact that, as a consequence of
the conservation of mass, det F > 0 at all times. However, some authors relax this constraint by
allowing F to become singular at some points in the presence of shocks.
22 2 Mechanics
and
B t (x, τ ) := [F t (x, τ )][F t (x, τ )]T . (2.3.21)
The Rivlin–Ericksen tensors An are defined through (cf. Rivlin and Ericksen [246])

∂n
An = C t (x, λ) |λ=t . (2.3.22)
∂λn
It immediately follows that

A1 = L + L T .
Usually, in the Navier–Stokes theory, the symmetric part of the velocity gradient is
denoted by D, i.e.,
1 1 1
D= A1 = grad v + (grad v)T = (L + L T ), (2.3.23)
2 2 2
and its skew part, called the spin tensor, is denoted by W , i.e.,
1
W= (L − L T ). (2.3.24)
2
Thus, the symmetric part of the velocity gradient D and the skew part of the velocity
gradient W are related to the relative stretch and the relative rotation through
D(x, t) = U̇ t (x, t) = V̇ t (x, t),
and
W (x, t) = Ṙt (x, t).
Let us define L n through

(n)
L n (x, t) := F t (x, t),
where the superscript n denotes n material time derivatives. It immediately follows

that
(n)
L n (x, t) = grad χ .
It is possible to express the Rivlin–Ericksen tensors An by
n−1

∂n
n
An = L n + L nT + L Tj L n− j + C t (x, λ) . (2.3.25)
j=1
j ∂λn |λ=t
It follows from the definition of the Rivlin–Ericksen tensors (2.3.22) that the follow-
ing recursive relation holds:
d
An = An−1 + An−1 L + L T An−1 , for n = 2, . . . (2.3.26)
dt
We shall see in the next subsection that the material time derivative of a frame-
indifferent quantity is not necessarily frame-indifferent. However, if A is a second-
order tensor, the quantity
d
A + A L + L T A,
dt
is frame-indifferent. Several other frame-indifferent derivatives are used in fluid me-
chanics; another that is particularly popular is the upper-convected Oldroyd derivative
defined by
∇ d
A:= A − L A − A LT , (2.3.27)
dt
where A is any second-order tensor. For instance,
∇ ∇∇ ∇
I= −2 D = − A1 , I = −2 D . (2.3.28)
The vorticity ω is the axial vector associated with the spin tensor W

T
∂v3 ∂v2 ∂v1 ∂v3 ∂v2 ∂v1
ω = curl v = ∇ × v = − , − , − , (2.3.29)
∂ x2 ∂ x3 ∂ x3 ∂ x1 ∂ x1 ∂ x2
and thus for all vectors a

1
Wa= ω × a. (2.3.30)
2
A source of confusion in fluid mechanics concerns the relationship between the
time rate of the rotation tensor Ṙ(t) and the spin tensor W : the spin tensor is not
the rate of rotation. Another interesting confusion that stems from imprecise use of
language is that of “irrotational flow.” Since R(t) is the rotation tensor, one would
be led to believe that in an “irrotational flow,” Ṙ(t) = 0. But this is not the case,
for in an “irrotational flow,” the vorticity ω (and thus the spin tensor W ) is zero. It
would be more appropriate to call such flows “vorticity-less flows.” This concept is
important to grasp, as we shall later develop boundary layer approximations based
on the concept that, in some flows, the vorticity is confined close to boundaries. We
now establish the relationship between the time rate of the rotation tensor and the
spin tensor. A straightforward computation leads to
1
W = Ṙ R T + R U̇ U −1 − U −1 U̇ R T ,
2
T
because Ṙ R T + R Ṙ = dtd (R R T ) = 0 since R(t) ∈ O. Thus, the spin tensor W is
clearly not the rate of rotation Ṙ. Also, Ṙ = 0 does not mean that W = 0.
24 2 Mechanics
The terminology “stretching tensor” that is usually used to describe D can also
be a source of confusion, as D is not the same as the rate of stretch U̇. Indeed, a
simple calculation yields
1
D= R U̇ U −1 + U −1 U̇ R T .
2
2.3.2 Frames, Frame-Indifference and Restrictions

Due to Frame-Indifference
Two frames {x, t} and {x ∗ , t ∗ } are said to be related by a change of frame, if the two
observers can agree on the measurement of length in the two frames, the measurement
of time intervals, and the sense of time. It then follows that these two frames are related
by
x ∗ = c(t) + Q(t)[x − x 0 ], (2.3.31)
t ∗ = t − a, (2.3.32)
where Q(t) ∈ O, x 0 is some fixed point in the frame {x, t}, and a is a scalar constant.
The term c(t) stands for a translation and Q(t)[x − x 0 ] stands for a rotation about
the fixed point x 0 if Q(t) ∈ O+ , i.e., det Q(t) = 1, and it stands for a reflection
if Q(t) ∈ O− , i.e., det Q(t) = −1. There is a considerable amount of controversy
if we should require restrictions due to Q(t) ∈ O or only to rigid body rotation,
i.e., Q(t) ∈ O+ . Here, we shall require restrictions to merely rigid body rotations.
It follows from (2.3.31), that a directed line segment transforms according to
∗ = Q(t) . (2.3.33)
We say that vectors that transform according to the above rule are Frame-Indifferent
or Objective. A scalar ψ that is such that
ψ ∗ = ψ, (2.3.34)
is said to be Frame-Indifferent, and a second-order tensor T that transforms according

to
T ∗ = Q(t) T Q(t)T , (2.3.35)
is said to be Frame-Indifferent.
Equations (2.3.31) and (2.3.32) imply that the velocity v transforms according to
v∗ = ċ + Q̇(t)[x − x 0 ] + Q(t)v, (2.3.36)

and thus, we immediately recognize that velocity is not frame-indifferent. Equation

(2.3.36) can be expressed as
v∗ − Q(t)v = ċ + A[x ∗ − c], (2.3.37)
where
A = Q̇(t) Q(t)T = − AT .
The tensor A is called the angular velocity of the starred frame with respect to the
unstarred frame. It follows from (2.3.37) that
a∗ − Q(t)a = c̈ + 2 Q̇(t)v + Q̈(t)[x − x 0 ],
where a = v̇ is the acceleration, and the above can be rewritten as
a∗ − Q(t)a = c̈ + 2 A(v∗ − ċ) + ( Ȧ − A2 )(x ∗ − c).
Next, we find that under a change of frame, the deformation gradient transforms
according to
F ∗ = Q(t) F(t); (2.3.38)
but since the deformation gradient is a second-order tensor, it follows from (2.3.35)
that the deformation gradient is not frame-indifferent. Furthermore, it follows from
(2.3.38) and (2.3.11) that
∗
L ∗ = Ḟ (F ∗ )−1 = Q̇(t) F + Q(t) Ḟ F −1 Q(t)−1
= Q̇(t) Q(t)T + Q(t)L Q(t)T ; (2.3.39)
here, it is understood that F depends also on t, but we have suppressed this de-
pendence to simplify the notation. Thus once again, the gradient of the velocity is
not frame-indifferent. Now, let us consider how the symmetric part of the velocity
gradient transforms. We have
1 ∗
D∗ = L + (L ∗ )T
2
1
= Q̇(t) Q(t)T + Q(t)L Q(t)T + Q(t) Q̇(t)T + Q(t)L T Q(t)T
2
1
= Q(t) D Q(t)T ,
2
because Q(t) ∈ O. Thus, the symmetric part of the velocity gradient is frame-
indifferent, and hence, the first Rivlin–Ericksen tensor is frame-indifferent. However,
the material time derivative of the symmetric part of the velocity gradient and the
material time derivative of the first Rivlin–Ericksen tensor are not frame-indifferent.
26 2 Mechanics
Similarly, it can be shown that
W ∗ = Q(t)W Q(t)T + A,
and hence the spin tensor is not frame-indifferent.

It is straightforward to verify that all the Rivlin–Ericksen tensors of order n, de-
fined by the recursive relation (2.3.26) are frame-indifferent. It is possible to introduce
a variety of frame-indifferent time derivatives, and one that has proved particularly
useful is called the upper-convected Oldroyd derivative, which was introduced in
(2.3.27).
As far as the Cauchy–Green stretch tensors are concerned, we have:
B ∗ = F ∗ (F ∗ )T = ( Q(t) F)(F T Q(t)T ) = Q(t) B Q(t)T ,
while
C ∗ = (F ∗ )T F ∗ = (F T Q(t)T )( Q(t) F) = C.
Thus, B is frame-indifferent while C is not. On the other hand, according to (2.3.14)

and (2.3.20), the relative-stretch tensor C t (τ ) can be expressed as (we have sup-
pressed the dependence on x)
T
C t (τ ) = F t (τ )T F t (τ ) = F(τ )F(t)−1 F(τ )F(t)−1
= F(t)−T F(τ )T F(τ )F(t)−1 ,
where we have used the notation superscript −T to denote the transpose of the
inverse. Thus,
C ∗t (τ ) = ( Q(t)F(t))−T ( Q(τ )F(τ ))T Q(τ )F(τ ) ( Q(t)F(t))−1

= Q(t)−T F(t)−T F(τ )T Q(τ )T Q(τ )F(τ )F(t)−1 Q(t)T

= Q(t) F(t)−T F(τ )T F(τ )F(t)−1 Q(t)T = Q(t)C t (τ ) Q(t)T ,
that is C t (τ ) is frame-indifferent. However, the Cauchy–Green stretch tensor B t (τ )

is not frame-indifferent.
A common mistake made in mechanics is to assume that all scalars are frame-
indifferent. This is not true. Consider, for instance, the euclidean norm of the velocity
|v|2 = v · v.
A glance at (2.3.36) shows that in general,
|v∗ |2 = v∗ · v∗ = |v|2 .
This point cannot be overemphasized as viscosities that depend on the norms of

the velocity have been used to describe the flows of fluids.
2.3.3 Balance Laws
In this section, we shall document the basic balance laws that a continuum has to
satisfy. Our discussion will be brief and a reader interested in details concerning the
same can find them in Truesdell [275], Jaunzemis [141], Malvern [174], or any other
standard text in continuum mechanics.
Let κ R denote the density in the configuration κ R (B) and denote the current
density in the current configuration κt (B). The local form of the balance of mass is
det F κ R = κ R , (2.3.40)
which, if F κ R is continuously differentiable with respect to time, takes the form
∂
+ div[v] = 0. (2.3.41)
∂t
The local form of the balance of linear momentum is
dv
= div T + b, (2.3.42)
dt
where T is the Cauchy stress and b the specific body force.

We shall assume that there are no body couples, and in the absence of body
couples, the balance of angular momentum implies that the stress is symmetric
We do not concern ourselves with thermodynamic issues in this book. But since we
consider models that are thermodynamically compatible in the sense that the fluids
meet the second law of thermodynamics expressed in the form of the Clausius–
Duhem inequality, in the motions that they undergo, and the requirement that their
specific Helmholtz potential be a minimum at equilibrium, for the sake of complete-
ness we shall record the balance of energy and the Clausius–Duhem inequality. The
local form of the balance of energy states that
dε
= T · L − div q + r, (2.3.43)
dt
where ε is the specific internal energy, q is the heat flux vector and r the specific
radiant heating. Finally, the Clausius–Duhem inequality reads, with η the specific
entropy
dη q r
≥ −div + . (2.3.44)
dt θ θ
Requiring that the motions of a fluid meet the Clausius–Duhem inequality leads
to restrictions on the signs of the coefficients. We shall see that when the material
coefficients satisfy the restrictions imposed by the Clausius–Duhem inequality and
the requirement that the specific Helmholtz free energy be a minimum when the
fluid is locally at rest, we have models that are well behaved in that they exhibit
28 2 Mechanics
physically reasonable characteristics. For instance, the state of rest of the fluid is
stable to disturbances and basic flows, provided that they are sufficiently “slow,” are
asymptotically stable. From the mathematical point of view, for such fluids, one is
able to prove existence of solutions. On the other hand, when the material moduli
do not satisfy the requirements of the Clausius–Duhem inequality, the fluids exhibit
unacceptable stability characteristics and anomalous behavior.
2.4 Constitutive Relations for Fluids

of the Differential Type
In this section, we introduce the constitutive relations for fluids of the differential
type, but to stress the difference with other fluids, we shall provide a brief discussion
of fluid models of the rate and integral types and postpone details to Section 2.8.
Let us first define the notion of history. The history of a scalar, vector or tensor-
valued function of time, ϕ, denoted by ϕ t (s), is defined through
ϕ t (s) = ϕ(t − s), for s ≥ 0.
A fluid is said to be an incompressible simple fluid (cf. Noll [198], Truesdell and
Noll [276]) if the Cauchy stress is determined to within a spherical part by the history
of the deformation gradient, i.e.,
T = − p I + F [F t (t − s)] . (2.4.1)
The constitutively determined part of the stress is referred to as the extra stress tensor.
While many of the popular models fall into the class of simple fluids, there are many
that do not. In fact, rate type materials do not fall into the class of simple materials.
A material is said to be a material of the rate type (in the sense of Truesdell and
Noll [276]) if
(n) (n−1) (n)
T = f (T , Ṫ , . . . , T , F, Ḟ, . . . , F), (2.4.2)
recall that the superscript (n) denotes n material time derivatives. Of course, frame
indifference will place restrictions on the possible forms that are allowable. As Trues-
dell and Noll [276] p. 95 remark, models of the class (2.4.2) do not model a simple
material, but a class of simple materials. It is also important to bear in mind that
relations such as (2.4.2) provide an implicit relation between the stress and the kine-
matical quantities, and it might not be possible to express the stress explicitly in
terms of the kinematical quantities. When this is possible, we get in fact a fluid of
complexity n as defined below.
The rate type models (2.4.2) are special subclasses of implicit constitutive relations
that relate the histories of the stress and the deformation gradient of the form

G (t − s)T (t − s), F(t − s) = 0. (2.4.3)
2.4 Constitutive Relations for Fluids of the Differential Type 29
Prusa and Rajagopal [219] studied fluids described through implicit constitutive
relations of the form
L T (t − s), F t (t − s) = 0. (2.4.4)
In the case of incompressible fluids, using the assumption of fading memory and
restricting themselves to retarded motions, they are able to obtain approximations of
(2.4.4) that have the same form as the Maxwell, Oldroyd-B, Rivlin–Ericksen, and
numerous other popular models that are used in non-Newtonian fluid mechanics.
The approximations used by Prusa and Rajagopal [219], just as those used earlier
by Coleman and Noll [68] in the case of a simple fluid, do not lead to models, as
these representations only hold in the special flows being considered (see Dunn and
Rajagopal [88] for a discussion as to why such approximations are not models in
their own right).
It is worth observing that (2.4.2) provides an explicit expression for the nth mate-
rial time derivative of the stress in terms of the n − 1 material time derivatives of the
deformation gradient. A more general implicit rate type relationship has the form
(n) (m)
g(T , . . . , T , F, . . . , F ) = 0, where n and m are integers. (2.4.5)
A special case of such a relationship is the implicit relation
g(T , F) = 0, (2.4.6)
for an elastic solid. Another special case is the relation
g(T , A1 ) = 0. (2.4.7)
The general representation for the implicit relationship between the stress and the
first Rivlin–Ericksen tensor can be found in [227]. This representation includes as
a special subclass the classical Stokesian fluid (see equation (2.4.12) that follows).
The implicit relation also includes as a special subclass fluids modeled by
D = f (, T ).
Recently, Perlacova and Prusa [211] have used such implicit models to describe the
response of colloids and slurries which cannot be described within the context of
classical Stokesian fluids, as the relationship between the stress and the shear rate is
multiple-valued.
In virtue of the constraint of incompressibility, the model
T = − p I + [μ( p, A1 )] A1 (2.4.8)
for an incompressible fluid, with a viscosity that depends on the pressure p and the
kinematical tensor A1 , can be expressed as
30 2 Mechanics

1 1
T =− tr(T ) I + μ( tr(T ), A1 ) A1 . (2.4.9)
3 3
We see that the above constitutive relation (2.4.9) falls into the class of models (2.4.7)
and does not belong to the class of materials defined through (2.4.2).
As we shall restrict our studies to purely mechanical issues in later chapters, that
is to issues concerning the existence, uniqueness and stability of flows, exclusively
with regard to the conservation of mass and the balance of linear momentum, we
shall not provide constitutive relations for such fluids within the context of a fully
thermodynamic framework. Therefore, the results that we shall establish cannot
be impugned on the basis of the thermodynamic considerations being faulty. The
interested reader can find a detailed and comprehensive treatment of the relevant
issues in a critique of the thermodynamics of the fluids of the differential type by Dunn
and Fosdick [87]. Results concerning the boundedness and stability of mechanical
and thermal quantities for fluids of the differential type of grade n can be found in
Rajagopal [222, 223].
An incompressible homogeneous isotropic7 fluid of the differential type of com-
plexity n is described by
(n−1)
T = − p I + f̂ (L, L̇, . . . , L ), (2.4.10)
where − p I is the constraint response stress due to the constraint of incompressibility.

The model (2.4.10) is not frame-indifferent. It follows from frame-indifference that
the Cauchy stress in fluids of complexity n is related to the fluid motion through
T = − p I + f ( A1 , . . . , An ), (2.4.11)
where Ak , k = 1, . . . , n, are the Rivlin–Ericksen tensors defined in (2.3.22). Such

fluids were first studied by Rivlin and Ericksen in [246].
As mentioned earlier, results concerning the boundedness of both thermal and
mechanical quantities for such fluids can be established (cf. Rajagopal [222]). We just
cite a rather weak result concerning the boundedness of the kinetic energy associated
with the flows of such fluids.
Theorem 2.4.1 Suppose an incompressible fluid of complexity n is undergoing a
process that is mechanically isolated8 and meets the second law; then
7 There is a popular misconception that by definition, a fluid is isotropic. If by fluid we mean a simple
fluid in the sense of Noll and furthermore a body whose symmetry group is the unimodular group,
then such fluids are isotropic. From such a narrow perspective one cannot have an anisotropic
liquid, say a fluid model for a liquid crystal. On the other hand, the common definition that a
fluid cannot support a shear stress indefinitely allows one to describe anisotropic fluids. For such
fluids, the symmetry group is not the unimodular group (cf. Rajagopal [225] and Rajagopal and
Srinivasa [235] for a discussion of the relevant issues).
8 A body is said to be mechanically isolated if there is no working due to either the tractions on the
boundary or due to the body forces.


0≤ |v(t)|2 dv ≤ |v(t )|2 dv,
Ωt 2 Ωt 2
where t ≤ t. (That is, the kinetic energy is not increasing.)

The class of models described by (2.4.11) is too large to obtain sharper results. We
can say a lot more for special subclasses of fluids of the differential type, namely
fluids of grade n.
In this section, we shall be concerned with fluids of complexity one, i.e.,
T = − p I + f ( A1 ),
and fluids of complexity two, i.e.,
T = − p I + g( A1 , A2 ),
where the symbols f and g stand for functions in general. Let us consider first
the more general case of a compressible fluid of complexity one. Such fluids are
usually referred to as Stokesian fluids, as they were first studied in full detail from
phenomenological point of view by Stokes [259].9 The Cauchy stress in such fluids
is given by
T = f (, A1 ). (2.4.12)
Isotropy demands that
∀ Q ∈ O, f (, Q A1 Q T ) = Q f (, A1 ) Q T . (2.4.13)
It then follows from standard representation theorems (cf. Spencer [256]) that
f (, A1 ) = α0 I + α1 A1 + α2 A21 , (2.4.14)
where the αi , i = 0, 1, 2, are functions of the density and the principal invariants
of A1
1
I1 = tr A1 , I2 = (tr A1 )2 − tr( A21 ) , I3 = det A1 .
2
Reiner [242] developed a model of the class (2.4.14) in which α1 and α2 are
polynomials in the invariants, to describe the phenomenon of “dilatancy” (that prop-
erty of a material by which it increases volume when stirred), a term coined by
Reynolds [243] to describe the response of granular solids and granular solids in-
fused with fluids. This behavior exhibited by granular materials is very similar to
the swelling that takes place in non-Newtonian fluids due to shearing and the phe-
nomenon of rod-climbing due to stirring. Normal stress differences are at the root of
9 Reiner [242] derived and discussed special flows of a subclass of such fluids by using a power-
series expansion. We shall discuss mathematical issues concerning these fluids as well as a more
general subclass of those fluids that are referred to as Reiner–Rivlin fluids.
32 2 Mechanics
these phenomena in different types of materials. It is possible that α1 and α2 could be

general functions of the invariants, and such fluids are called Reiner–Rivlin fluids.
The Reiner–Rivlin model (2.4.14) does not predict distinct normal stress dif-
ferences in simple shear flow. While distinct normal stress differences have been
observed in flows of certain polymeric liquids, one should bear in mind that there are
many fluids that exhibit no normal stress differences, while exhibiting pronounced
shear-thinning or shear-thickening. Many geological, biological materials, as well as
food products and chemicals, are adequately described by the Reiner–Rivlin model
(2.4.14). This fact has been largely overlooked by the mathematical community in
view of the indictment of the model by Truesdell and Noll [276]: “Data was collected;
they seem to indicate that for polyisobutylene solutions, T <11> and T <22> are not
equal when κ becomes appreciably large. This experimental evidence was motive for
the rejection of the Reiner–Rivlin equation (119.4) as an adequate basis of a physical
theory and for the search of greater generality, in the last few years….” In a similar
vein, Coleman et al. [67] observe: “Note that the two normal stress functions (30.4)
resulting from the Reiner–Rivlin equation (30.2) are the same, which implies for
example that the normal stresses in simple shear flow must coincide (see E.11). A
relation of this kind can be compared with experiments. Data reported by Padden and
De Witt show that they are not the same in some polymer solutions. This observation
was a motive for the rejection of the Reiner–Rivlin theory. There is a large literature
on the Reiner–Rivlin theory which must be regarded as obsolete today.”
Unfortunately, these statements have been taken more seriously than they ought
to have been. One cannot indict the whole class of models on the basis of experi-
ments on polyisobutylene or for that matter on a few dilute polymeric liquids. While
the Reiner–Rivlin model might be unsuitable for fluids that have distinct normal
stress differences in simple shear flows, it is a perfectly reasonable model for shear-
thinning and shear-thickening fluids. In fact the most popular models in chemical
engineering, food rheology, glaciology and other areas are generalized power-law
models or generalizations of the Navier–Stokes model that belong to the class of
Reiner–Rivlin fluids. We notice that all the popular power-law fluid models and gen-
eralized Newtonian fluid models fall into the class of Stokesian fluids. Mathematical
issues concerning such fluids have been studied in great detail (cf. the book by Málek
et al. [168] for some of the recent results). We shall discuss these fluids later.
If we require the fluid to be linear in A1 , then (2.4.14) simplifies to

f (, A1 ) = − p() + λ()(tr A1 ) I + μ() A1 , (2.4.15)
where λ and μ are called the bulk and shear viscosities, and the negative sign in
front of p() is due to the sign convention that compressive stresses are negative. A
few remarks concerning the material moduli λ and μ are in order. While the material
modulus μ can be determined directly through an experiment, the material modulus
λ cannot be measured directly. It has to be inferred from measurements of μ and
(3 λ + 2 μ), the latter being the bulk modulus of the fluid.
In a seminal paper that has had tremendous influence on the development of fluid
dynamics, Stokes [259] suggested that there were several flows wherein the density
of the fluid remains nearly constant and wherein one could make the assumption
3 λ + 2 μ = 0.
This assumption, referred to as the “Stokes Assumption,” has become a central part
of fluid dynamics. Maxwell [182–184] claimed to have derived the “Stokes Assump-
tion” for monatomic gases within the context of the kinetic theory of gases, and this
gave further credibility to the “Stokes Assumption.” However, as Truesdell [271]
points out, Maxwell’s results are a direct consequence of the assumptions he makes
for the pressure and the temperature and thus the results are actually built into the
assumptions. Despite Maxwell’s work, Stokes [259] was far from convinced by his
own assumption. He remarks (Stokes [260]): “Although I have shown (Volume 1, p.
119) that on the admission of a supposition which Poisson would have allowed, the
two constants in his equation are reduced to one, and the equations take the form (1),
and although Maxwell obtained the same equations from his kinetic theory of gases
(Philosophical Transactions for 1867, p. 81) I have always felt that the correctness
of the value μ/3 for the coefficient in the last term of (1) does not rest on as firm
a basis as the correctness of the equations of motion of an incompressible fluid, for
which the last term does not come in at all.” In fact, the “Stokes Assumption” is
clearly wrong. There is overwhelming experimental evidence that contradicts the
assumption despite which the assumption continues to be used. A detailed discus-
sion as to why the “Stokes Assumption” is inapt can be found in the recent paper by
Rajagopal [229] wherein a new interpretation and development of the Navier–Stokes
constitutive relation is provided.
If we now require in addition that the fluid be incompressible, then as
1
tr A1 = div v = 0, (2.4.16)
2
and since is a constant, because the fluid is incompressible and homogeneous,10

we find that (2.4.15) reduces to
T = f ( A1 ) = − p I + μ A1 , (2.4.17)
where p is indeterminate due to the constraint of incompressibility. This is the popular

incompressible Navier–Stokes fluid model. This model approximates exceptionally
well the response of liquids like water, in certain flows. However, such a model fails
to describe the response of water through a wide range of pressures. As mentioned
at the beginning of the introduction, Stokes [259] recognized that the model (2.4.17)
10 Ifthe fluid is incompressible but inhomogeneous then the density is not a constant everywhere,
the density of a specific material point remains the same, i.e., = (x).
34 2 Mechanics
is obtained under certain assumptions that do not apply to all flows and this matter
warrants discussion.
In classical mechanics, one encounters a variety of constraints such as incom-
pressibility, inextensibility, etc. Of course no material is truly incompressible; it is
merely that the variations in the density are small enough to be negligible even under
large spherical stresses. Such materials are approximated as bodies that are incom-
pressible, and the stresses in such bodies are determined to within an indeterminate
constraint stress. In the case of (2.4.17), the indeterminate part is − p I. In classical
mechanics, it is assumed that the constraint force does no work, but it is important
to bear in mind that this is merely an assumption and a variety of other assumptions
could be made concerning the constraint (cf. Liu [166]). In this context, it is worth-
while to recognize an important result due to Gauss [110] where he makes the point
that the constraint response merely enforces the constraint and it may or may not do
work. As written in the introduction, Stokes [259] recognized that it is possible for
the material moduli to depend on the constraint stress; for instance, the viscosity μ
could depend on the pressure. This assumption of Stokes is valid for incompressible
lubricants used for elastohydrodynamic lubrication where the fluid is subject to a
wide range of pressures. Under such severe pressures, it is found that the viscosity
varies exponentially with the pressure! (cf. Andrade [11] and Bridgman [51]).
In the case of incompressible fluids, we saw that the material moduli can depend
on the constraint response. Thus, in the case of an incompressible fluid of complexity
one, the Cauchy stress could take the form
T = − p I + α1 ( p, I1 , I2 ) A1 + α2 ( p, I1 , I2 ) A21 , (2.4.18)
where
1
p = − (tr T ).
3
This model is fundamentally different from (2.4.14) in that (2.4.18) provides an
implicit constitutive relation, i.e., equation (2.4.18) is of the form
f (T , A1 ) = 0.
If one requires that the fluid of complexity one be incompressible and also that the
stress depend linearly on A1 , then (2.4.18) reduces to
T = − p I + μ( p) A1 .
There is a large body of experimental work that attests to the fact that, while the
density changes are very small (i.e., of the order of a percent or less), the viscosity
may change by a factor of 108 when the pressure changes considerably. In fact, if the
pressures are sufficiently high, as it is the case in elastohydrodynamic lubrication,
it is even possible that “glass transition” might take place (cf. Szeri [261]). A thor-
ough discussion of the early experimental work on the variation of viscosity with
pressure can be found in the authoritative text on the physics of high pressure by
Bridgman [52].
It is however possible that in constrained materials, the material moduli can de-
pend on the constraint response. In this case, the isotropy condition (2.4.13) and the
constraint will lead to a representation different from (2.4.14). This leads to a class
of models that have not been studied with the intensity that they deserve.
In the case of a compressible fluid of complexity two, isotropy requires that
∀ Q ∈ O, g(, Q A1 Q T , Q A2 Q T ) = Q g(, A1 , A2 ) Q T .
Once again, using results from representation theory, we find that
g(, A1 , A2 ) = α0 I + α1 A1 + α2 A21 + α3 A2 + α4 A22 + α5 [ A1 A2 + A2 A1 ]

+ α6 [ A21 A2 + A2 A21 ] + α7 [ A22 A1 + A1 A22 ],
where the coefficients α0 , . . . , α7 depend on the density and on the invariants
I1 = tr A1 , I2 = tr A21 , I3 = tr A31 , I4 = tr( A1 A2 ),
I5 = tr( A21 A2 ), I6 = tr A2 , I7 = tr A22 , I8 = tr A32 , I9 = tr( A22 A1 ).
We shall be concerned primarily with fluids of grade two, i.e., fluids for which
the Cauchy stress is
T = − pI + μ A1 + α1 A2 + α2 A21 , (2.4.19)
and a special subclass of fluids of grade three. In a general fluid of grade three, the
Cauchy stress has the form
T = − pI + μ A1 + α1 A2 + α2 A21 + β1 A3 + β2 [ A1 A2 + A2 A1 ] + β3 (tr A21 ) A1 .
If the above fluid of grade three is to be thermodynamically compatible (cf. Fosdick

and Rajagopal [101]), then it follows that the above constitutive relation reduces to
T = − pI + μ A1 + α1 A2 + α2 A21 + β3 (tr A21 ) A1 , (2.4.20)
where
μ ≥ 0, α1 ≥ 0, β3 ≥ 0, α1 + α2 ≤ 24μβ3 . (2.4.21)
When β3 = 0, the above restrictions reduce to
μ ≥ 0, α1 ≥ 0, α1 + α2 = 0, (2.4.22)
36 2 Mechanics
which are the restrictions obtained by Dunn and Fosdick [87] for fluids of grade
two.11 We immediately notice that thermodynamically compatible fluids of grades
two and three are fluids of complexity two.
In a much cited paper that has also been much misunderstood, Coleman and
Noll [68] showed that a model of the form (2.4.19) could be obtained as a truncation
at second order from the expansion for the stress in a simple fluid with a very special
fading memory within the context of what they termed as retarded (slowed down)
motions. The documentation of the mathematical procedure that they adopted and
the result that they obtained is referred to as the “Retardation Theorem.” Coleman
and Noll [68] did not obtain models for describing the response of fluids, nor did
they claim that they obtained models. They essentially showed that for very special
simple fluids undergoing very special motions, when the expression for the stress is
truncated using a special expansion procedure, at second order the truncation led to
an expression of the form (2.4.19). However, outside of all the special circumstances
that were assumed, their procedure remains mute with regard to the form the stress
takes on in the simplified structure. This notwithstanding, a great deal of misunder-
standing has resulted leading to overblown statements that the restrictions (2.4.22)
that are a consequence of the Clausius–Duhem inequality are incorrect as they are
contradicted by experiments, thereby impugning the Clausius–Duhem inequality. A
detailed discussion of all the relevant issues will take us far afield and so we refer the
reader to a critical review of the issues by Dunn and Rajagopal [88]. Here, we quote a
clear and succinct discussion of the issues that has unfortunately, by and large, gone
mostly unheeded (see Truesdell [274]):
“In the experimental literature the result of Coleman and Noll is sometimes given a
vastly exaggerated statement such as: “For sufficiently slow flows the second-order
fluid is a valid approximation to any simple fluid.” Coleman and Noll neither stated
nor proved any such thing. In particular, they never claimed that all simple fluids
have fading memory in the sense of Coleman and Noll, for it is easy to provide
examples of fluids that do not. Second, while their general theory concerns many
different kinds of approximation, their only application to “slow” flows refers to
those obtained by retardation of a given flow. Other definitions of “slow” lead to
somewhat different results. Finally, they never claimed nor proved any relation at all
between the solutions of differential equations of motion for the fluid of grade 2 and
the solutions of the equations of motion of the general simple fluid that the fluid of
grade 2 approximates. An example known from the kinetic theory of gases shows
that it is risky to jump to conclusions in matters of this kind.”
Recently, a thermodynamic framework has been put into place to describe bodies
undergoing entropy producing processes. It appeals to the following central idea:
the rate of entropy production (rate of dissipation for purely isothermal processes)
must be maximized, and this among a set of admissible constitutive relations that
automatically meet that the entropy production be nonnegative. It can be shown easily
within that framework that the classical Navier–Stokes model corresponds to a fluid
11 Ting [268] studied several flows of fluids of grade two, that is fluids modeled by (2.4.19). None
of the problems he studied had bounded solutions when α1 was negative.

whose specific Helmholtz potential ψ is given by
ψ = ψ(θ ),
and the rate of dissipation ξ is given by
μ(θ )
ξ = ξ(θ, A1 ) = | A1 |2 . (2.4.23)
2
Rajagopal and Srinivasa [234] have shown that rate type viscoelastic fluid models
such as the Maxwell fluid, the Oldroyd-B fluid, and a variety of other popular models
can be derived within such a framework (see Murali Krishnan and Rajagopal [154] for
a derivation of Burgers’ model). The model takes into account how energy is stored
by the body, how this energy that is stored is released (whether it can be recovered in
a purely mechanical process or whether one needs to use a thermodynamic process
to recover the energy), how the body produces entropy, conducts heat, etc. Rao and
Rajagopal [241] have shown that the integral K-BKZ model and generalizations of it
can also be obtained within that framework. However, fluids of the differential type
such as grade-two fluids do not seem to fall within the class of models that can be
obtained directly using the above thermodynamic framework. We need to make cer-
tain modifications to the theory in order to obtain the grade-two fluid (see Rajagopal
and Srinivasa [234]). These modifications lead to results that are completely in keep-
ing with the results that arise from enforcing the Clausius–Duhem inequality.
If one makes the choice that the Helmholtz potential ψ is given by
α1
ψ = ψ(θ, A1 ) = ψ̂(θ ) + | A1 |2 with α1 > 0,
4
and the rate of dissipation ξ is given by (2.4.23) then one obtains (see [234]) that the
stress T̄ is
T̄ = − pI + μ A1 + α1 Ȧ1 . (2.4.24)
We note that the model (2.4.24) for T̄ is not frame-indifferent as Ȧ1 is not frame-
indifferent. However, it is possible to obtain a properly frame-indifferent model, with
the same associated rate of dissipation as T̄ , that leads to the model for a grade-two
fluid. Indeed, on adding the term
T̂ = α1 ( A1 L + L T A1 ) − α1 A21 ,
to (2.4.24), we immediately obtain a thermodynamically compatible model for a

grade-two fluid wherein the stress is given by (2.4.19) with α1 = −α2 . It is important
to note that
T̂ · L = 0,
38 2 Mechanics
and thus there is no rate of dissipation associated with T̂ . Hence the stress T = T̄ + T̂
has the same associated rate of dissipation as the stress T which is the model for a
grade-two fluid.
When substituting (2.4.20) into the well-known balance of linear momentum:
dv
= div T + b, (2.4.25)
dt
after dividing by the density, making obvious changes in notation, and denoting the
gradient by ∇, we obtain the equation of motion of a grade-three fluid
∂
(u − α1 Δ u) − μ Δ u + curl(u − (2α1 + α2 )Δ u) × u
∂t

− (α1 + α2 ) Δ([∇ u]u) − 2[∇(Δ u)]u − β3 div(| A1 |2 A1 )
1 1
+ ∇ p − (2α1 + α2 )(u · Δ u + | A1 |2 ) + |u|2 = f . (2.4.26)
4 2
If we set β3 = 0, then (2.4.26) reduces to
∂
∂t
− (α1 + α2 ) Δ([∇ u]u) − 2[∇(Δ u)]u
1 1
+ ∇ p − (2α1 + α2 )(u · Δ u + | A1 |2 ) + |u|2 = f , (2.4.27)
4 2
which is the equation of motion of a grade-two fluid and of course can also be obtained
by substituting (2.4.19) into (2.4.25).
2.4.1 Special Motions
There are special flows in which it is impossible to distinguish a general simple fluid
from a fluid of complexity three, i.e., in such flows any simple fluid can be described
by an appropriate fluid of complexity three. One such class is called monotonous flows
(cf. Noll [198]) and many flows such as shear flows fall into this category. A motion is
said to be monotonous if and only if there is an orthogonal tensor Q(t), a scalar κ, and
a constant tensor N 0 such that (see (2.4.1) with τ̂ = t − s)
F 0 (τ̂ ) = Q(τ̂ )eτ̂ κ N 0 , Q(0) = I, |N 0 | = 1. (2.4.28)
In such flows, a simple fluid (2.4.1) has a much simpler representation. Its stress is
completely determined by the first three Rivlin–Ericksen tensors A1 , A2 and A3 ; other
information concerning the history of the relative deformation gradient is unnecessary.
A special subclass of monotonous flows are viscometric flows. These are monoto-
nous flows wherein N 20 = 0. In such flows, a simple fluid cannot be distinguished from
a fluid of complexity two, i.e., the first two Rivlin–Ericksen tensors are sufficient to
determine the stress. In fact, in viscometric flows, the stress is completely determined
by three functions referred to as viscometric functions (cf. Coleman et al. [67]). To
illustrate this fact, let us consider the steady linear flow (cf. (2.2.1))
v = u(y)i,
and recall that

κ := u (y).
Further, define the three viscometric functions by
τ (κ) := Tx y , σ1 (κ) := Tx x − Tzz , σ2 (κ) := Tyy − Tzz .
Then, it follows from (2.4.28) that the extra stress S is given by (see Noll [198] for
details)
S = τ (κ)(N + N T ) + σ1 (κ)N T N + σ2 (κ)N N T , (2.4.29)
where ⎛ ⎞
010
N = ⎝0 0 0⎠.
000
It is easy to show that (2.4.29) is equivalent to (cf. Criminale et al. [71])
τ (κ) 1 [σ1 (κ) − σ2 (κ)] σ2 (κ) 2

S= A1 + A2 + A1 .
κ 2 κ 2 κ2
It is obvious from the above expression for the extra stress that in viscometric flows,
the stress for a simple fluid takes the same form as that of a fluid of complexity two.
2.4.2 Secondary Flows in Pipes
Most non-Newtonian fluids cannot flow unidirectionally along an axis of a pipe of

noncircular cross-section, in marked contrast to the Navier–Stokes fluid. In fact, the
similarity in the secondary flow structure between the laminar flow of viscoelastic
fluids and the turbulent flow of a Navier–Stokes fluid have led to the development of
models for turbulent flows of the Navier–Stokes fluid (see Rivlin [245] and Huang and
Rajagopal [136]). The problem of the flows of fluids of grade n in noncircular pipes has
been studied at great length ever since the study of Ericksen [94] that first investigated
this issue.
40 2 Mechanics
Fosdick and Serrin [102] have obtained necessary and sufficient conditions for the
development of secondary flows in simple fluids. It immediately follows from these
conditions that a fluid of grade three cannot flow in a straight path along the axis of a
noncircular pipe while a fluid of grade two can. Fosdick and Serrin study the class of
simple fluids whose viscometric functions satisfy
1. μ̃(κ) is of class C2 ,
2. μ̃(κ) and σ1 (κ)/κ 2 are of class C3 near κ = 0 and their first derivatives and third
derivatives vanish at κ = 0,
3. κ μ̃(κ) is an increasing function of κ.
They show that if

σ1 (κ) = Cκ 2 μ̃(κ),
where C is a constant, then rectilinear flow along the axis of a cylindrical tube is possible
only in circular tubes. In addition, such a flow is possible in the annular region between
two circular concentric tubes. Ericksen [94] had originally conjectured that rectilinear
flows would be possible in a tube, the boundary of its cross-section being comprised of
parts of circles and straight lines. The analysis of Fosdick and Serrin [102] proved the
conjecture false.
In the case of a fluid of grade two, we have in simple shear flows
μ̃(κ) = μ, σ1 (κ) = (2α1 + α2 )κ 2 , σ2 (κ) = α2 κ 2 .
Thus
2α1 + α2 2
σ1 (κ) = κ μ.
μ
However, in the case of simple shearing flow of a fluid of grade three
μ̃(κ) = μ + 2(β2 + β3 )κ 2 , σ1 (κ) = (2α1 + α2 )κ 2 , σ2 (κ) = α2 κ 2 ,
thus
σ1 (κ) = Cκ 2 μ̃(κ),
and hence we will have secondary flow.

Assuming that the driving force, i.e., the pressure gradient along the axis of the pipe,
is small, Truesdell and Noll [276] study the structure of the secondary flow using a
perturbation approach, the perturbation parameter being the driving force. Thus, the
basic solution that is being perturbed is the null solution. It turns out that under such a
perturbation scheme, the secondary flow manifests itself only at fourth order. Normal
tractions, not required in a Navier–Stokes fluid are required in this case (see Pipkin and
Rivlin [212]).
We could ask a different question concerning secondary flows in pipes: what is the
nature of the secondary flow when the bounding surface of the pipe has a small departure
from circularity, the driving force being large? In this case, we cannot study the problem
via a perturbation of the driving force; we need to use a domain perturbation technique
(see Hadamard [129]). It then transpires that in this case, the secondary flow manifests
itself at first order for a fluid of grade three. The zeroth order solution in this case is the
Poiseuille flow in a pipe of circular cross-section (see Huang and Rajagopal [136] and
Mollica and Rajagopal [192]).
2.4.3 Stability to Finite Disturbances
As explained in the introduction, we are not overly concerned about thermodynamic

considerations. However, we find that the consequences of such considerations lead to
the fluid exhibiting reasonable response characteristics, while violation of such con-
siderations lead to unacceptable behavior. An important response characteristic is the
flow’s stability. In this section, we establish results that are similar in nature to the sem-
inal contributions of Orr, Hopf, Kampe de Feriet and Thomas concerning the stability
of flows of the Navier–Stokes fluid. The results we establish will leave no doubt as to
the proper signs of the material coefficients that appear in the model for fluids of grade
two and three. Before we can establish the stability results, we will have to put into
place certain identities.
The starting point for the study of stability to finite disturbances is the Power The-
orem. Recall the balance of linear momentum (2.4.25)
dv
= div T + b.
dt
On taking the scalar product of both the left-hand and right-hand sides of the above
equation with the velocity v and integrating over the flow domain, integrating by parts,
using the divergence theorem, and the conservation of mass, we obtain

d
t · v da + b · v dv = (|v|2 )dv + T · L dv,
∂Ωt Ωt Ωt 2 dt Ωt
where t = T T n is the surface traction vector, n is the unit outward normal to the
boundary, and recall that | · | denotes the Euclidean norm. The first term in the right-
hand side denotes the time rate of change of the kinetic energy, while the second term
denotes the stress power (the rate at which work is done by the stress on the part Ωt ).
This equality is usually referred to as the power theorem and is the starting point of all
nonlinear stability analysis.
Owing to the conservation of mass, we have for any function ϕ

ϕ dv = 0 ϕ |det F| dv,
Ωt Ω0
42 2 Mechanics
where 0 is independent of t.12 Therefore,

d d d
ϕ dv = 0 ϕ dv = ϕ dv.
dt Ωt Ω0 dt Ωt dt
Applying this equality with ϕ = |v|2 , we derive

d 2
t · v da − T · L dv + b · v dv = |v| dv. (2.4.30)
∂Ωt Ωt Ωt dt Ωt 2
If the body is mechanically isolated, then

t · v da + b · v dv = 0,
∂Ωt Ωt
and thus we find that (2.4.30) reduces to

d 2
− T · L dv = |v| dv. (2.4.31)
Ωt dt Ωt 2
In the case of the Navier–Stokes fluid, where T is given by (2.4.17), we immediately

obtain the rather weak result

1 d 1
|v| dv = −
2
μ| A1 |2 dv, (2.4.32)
2 dt Ωt 2 Ωt
(for a tensor | · | denotes the Frobenius norm), and we conclude that the kinetic energy is
nonincreasing provided μ ≥ 0. Of course, if μ < 0, we would conclude that the kinetic
energy is increasing, an absurd situation that leads us to conclude that the assumption
μ ≥ 0 is sensible. Our conclusions concerning the material modulus α1 can be based
on similar grounds, without any appeal to thermodynamic arguments.
We shall derive the energy identity (2.4.31) for the model (2.4.20) and by setting
β3 = 0, we shall recover the energy identity for a fluid of grade two. On substituting
(2.4.20) into the energy identity (2.4.31) and using the fact that the density is constant,
we obtain the energy identity (that can also be obtained by taking the scalar product of
both sides of (2.4.26) by v)

d α1 d μ
|v|2 dv + | A1 |2 dv + | A1 |2 dv
dt Ωt 2 dt Ωt Ωt

α1 + α2 β3
+ (tr A31 )dv + | A1 |4 dv = 0. (2.4.33)
Ωt Ωt
On setting β3 = 0 in the above equation, and using α1 + α2 = 0, we find that
12 More precisely, in the second integral ϕ stands for ϕ(x(X, 0), 0).

d α1 μ
|v|2 dv + | A1 |2 dv = − | A1 |2 dv. (2.4.34)
dt Ωt 2 Ωt Ωt
We hasten to add that even if α1 + α2 = 0, we can show that the rest state of the fluid
is unstable if α1 < 0. For the moment, let us suppose that α1 > 0. We now define a
functional E(t) by
α1
E(t) := |v|2 dv + | A1 |2 dv. (2.4.35)
Ωt 2 Ωt
We note that since α1 > 0 and > 0, then E(t) is zero if and only if v = 0. Thus,
(2.4.34) reads
d μ
E(t) = − | A1 |2 dv, (2.4.36)
dt Ωt
and since μ > 0 and > 0, we have that the energy is nonincreasing, a result similar to
that for the Navier–Stokes fluid (2.4.32). This result is not sufficiently strong. We next
show that the energy decays to zero with time, bounded by an exponential, a result due
to Dunn and Fosdick [87]. It is stated in a compact domain in which the flow satisfies
v(x, t) = 0 on ∂Ωt .
Within a purely mechanical context, this implies that there is no velocity on the bound-
ary, i.e., the body is not supplied energy from the environment.
Theorem 2.4.2 Let a homogeneous incompressible fluid of grade two for which
(2.4.22) holds,
μ ≥ 0, α1 ≥ 0, α1 + α2 = 0,
undergo a flow in a compact domain and meet the “no-slip” boundary condition. Then
t α1 + S0,2
0 ≤ E(t) ≤ E(0)e− τ with τ = ,
2μ
where S0,2 is the Poincaré constant for the domain Ω (see (3.1.5) in Chapter 3 below).
Proof By Poincaré’s inequality, we have

S0,2
|v|2 dv ≤ | A1 |2 dv.
Ωt 2 Ωt
Therefore
1 α1
E(t) ≤ (S0,2 + ) | A1 |2 dv.
2 Ωt
On the other hand, (2.4.36) gives

μ
Ė(t) + | A1 |2 dv = 0,
Ωt
44 2 Mechanics
so that,
2μ
Ė(t) + E(t) ≤ 0,
α1 + S0,2 Ωt
The result follows by integrating in time this last inequality.

We next ask the question if disturbances to a state of rest of a fluid die down in finite
time. It follows from (2.4.36) and (2.4.35) that for any λ > 0,

λα1 μ
Ė(t) + λE(t) = λ |v| dv + (
2
− ) | A1 |2 dv. (2.4.37)
Ωt 2 Ωt
In particular,
λα1 − 2μ
Ė(t) + λE(t) ≥ | A1 |2 dv,
2 Ωt
and thus, there exists λ = 2μ/α1 such that
Ė(t) + λE(t) ≥ 0.
Hence the energy cannot decay to zero in finite time! The energy E(t) is bounded both
above and below
−2 μ t
E(t) ≥ E(0)e α1 .
2.4.4 Instability
There are many reasons why α1 cannot be negative, thermodynamic issues aside. Here,
we discuss a compelling reason for the nonnegativity of the modulus α1 . We first
establish results for a fluid of grade two with μ ≥ 0, α1 < 0 and α1 + α2 = 0. We then
relax the condition α1 + α2 = 0 and consider fluids wherein μ > 0 and α1 < 0. Once
again, our starting point is (2.4.34) which now reads

d |α1 | μ
|v| dv −
2
| A1 | dv = −
2
| A1 |2 dv.
dt Ωt 2 Ωt Ωt
Let us define the quantity N (t) by

|α1 |
N (t) := | A1 |2 dv − |v|2 dv. (2.4.38)
2 Ωt Ωt
We recognize that N (t) is not a Lyapunov function; it is not necessarily nonnegative.

It follows from (2.4.38) that

μ
Ṅ (t) = | A1 |2 dv, (2.4.39)
Ωt
and thus
Ṅ (t) ≥ 0.
Of course, we do not know if N (0) is positive for the domain under consideration. Let
us investigate this more closely. From the definition of N (t) in (2.4.38), we have

|α1 | − S0,2 (Ωt )
N (t) ≥ | A1 |2 dv, (2.4.40)
2 Ωt
where S0,2 (Ωt ) is the maximum in time of the Poincaré constant for the domain. We
immediately see that if
|α1 |
S0,2 (Ωt ) < , then N (0) > 0.

The inequality (2.4.40) can be interpreted in two ways. Given a fluid with some values
for α1 and , we can always find a sufficiently small domain Ωt in which N (0) is
positive. Also, given a domain, it is always possible to find a fluid with sufficiently
large |α1 | such that N (0) is positive. We shall use the first interpretation, namely given
a fluid there are domains in which N (0) is positive. It follows from (2.4.38) and (2.4.39)
that

μ λ|α1 |
Ṅ (t) − λN (t) = | A1 | dv −
2
| A1 | dv + λ
2
|v|2 dv. (2.4.41)
Ωt 2 Ωt Ωt
In particular, for any λ > 0,

2μ − λ|α1 |
Ṅ (t) − λN (t) ≥ | A1 |2 dv.
2 Ωt
Thus, if we pick
2μ
λ= ,
|α1 |
then
Ṅ (t) − λN (t) ≥ 0,
and in domains wherein S0,2 (Ωt ) < |α1 |, then N (t) → ∞ at least exponentially with
time. Thus, the difference between the averaged stretching of the fluid (which is an
imprecise interpretation of the first integral in (2.4.38)) and the averaged kinetic energy
of the fluid becomes unbounded as t → ∞. This essentially implies that disturbances
do not die down with time.
A few words about the above instability analysis are in order, as it is not the usual
type of analysis that is carried out to show that the rest-state of a fluid is unstable. We
are not showing that certain “small disturbances” tend to blow up in time: in fact we
show that there exist domains where all disturbances blow up in time.
46 2 Mechanics
When μ > 0, α1 < 0 and α1 + α2 = 0, we cannot establish as strong a result as

the result above. We can show that for any fluid whose material moduli meet the above
restrictions, there exist rigid containers (filled with the above fluid) wherein, any initial
disturbance to the rest-state at initial time is such that

| A1 |3 dv
Ωt
must become larger than any pre-assigned positive real number M, provided the fluid
is sufficiently viscous. In fact, the larger the viscosity, the more the disturbance grows!
(cf. Fosdick and Rajagopal [100] for a detailed discussion of these issues). We can also
show that there exist domains in which any sufficiently smooth initial disturbance can
never subside in the sense that
lim { sup | A1 (x, t)|} > 0.

t→∞ x∈Ω
t
Here, we shall rest content proving the first of the above two results. In order to show
that Ωt | A1 |3 dv is unbounded, we need the following elementary lemma whose proof
can be found in [100].
Lemma 2.4.3 Let A be any symmetric traceless tensor and let α be any real number.
Then,
|α| |α|
− √ | A|3 ≤ α(tr A3 ) ≤ √ | A|3 . (2.4.42)
6 6
Moreover, there exist tensors A for which (2.4.42) reduces to an equality.

As α1 < 0 and α1 + α2 = 0, it follows from (2.4.33) with β3 = 0 that

d |α1 | μ α1 + α2
|v|2 dv − | A1 |2 dv = − | A1 |2 dv − (tr A31 )dv.
dt Ωt 2 Ωt Ωt Ωt
Let λ ∈ (0, |α2μ1 | ). Then, it follows from (2.4.38) and (2.4.42) that

|α1 + α2 |
Ṅ (t) − λN (t) ≥ − √ | A1 |3 dv. (2.4.43)
6 Ωt
Now, suppose that

∀t ∈ (0, ∞), | A1 |3 dv ≤ M < ∞. (2.4.44)
Ωt
Then (2.4.43) implies that
|α1 + α2 |
Ṅ (t) − λN (t) ≥ − √ M,
6
which in turn implies that,

2μ |α1 + α2 |M
∀λ ∈ (0, ), N (t) ≥ N (0) − √ eλt .
|α1 | λ 6
We know that there exist domains for which N (0) > 0. Now, if we can pick a λ
sufficiently large then the term multiplying the exponential is positive. In fact, the
larger the viscosity of the fluid, the larger the λ that we can pick and the faster N (t)
will blow up! Thus, (2.4.38) implies that

lim | A1 |2 dv = ∞.
t→∞ Ω
t
An application of Hölder’s inequality then implies that

lim | A1 |3 dv = ∞,
t→∞ Ω
t

but this violates our assumption (2.4.44); hence Ωt | A1 |3 dv cannot be less than any
preassigned number. Thus, even when α1 + α2 = 0, if μ > 0 and α1 < 0, the fluid
does not exhibit desirable stability characteristics.
2.5 Boundary Conditions for Fluids

While there has been intense activity in mechanics in the development of constitutive
equations for the bulk material, little attention has been paid to the “determination”
of boundary conditions which are also constitutive specifications. In fact, the nature
of the boundary, or a narrow region adjacent to the boundary, depends on the material
on either side of the boundary. Boundaries are rarely, if ever, sharp and what needs
to be prescribed as a boundary condition depends on the structure of the material that
is being enveloped by the boundary as well as the structure of the environment. That
Stokes [259] was fully cognizant of the need to arrive at boundary conditions based
on the nature of the bodies adjacent to the boundaries is made evident by his remark:
“Besides the equations which must hold in the interior of the mass, it will be necessary
to form the equations which must also be satisfied at its boundaries.” Stokes [259]
developed a variety of boundary conditions that ought to hold between two fluids, a
fluid and a solid, etc.
On the basis of experiments of “slow” flows in channels, Du Buat [85] offered the
opinion that the fluid adheres to the solid boundary adjacent to which it flows. How-
ever, this suggestion was not immediately accepted and there was considerable debate
concerning what one obtains at a boundary. On the basis of arguments, Navier [193]
obtained a formula for the “slip” that would take place at the wall, while Poisson [213]
48 2 Mechanics
suggested that a condition similar to that derived by Navier [193] would hold, not at
the wall, but at the edge of a narrow layer adjacent to the wall. The early debate con-
cerning what ought to be enforced at the boundary involved the likes of Navier, Girard,
Coulomb, Poisson, Stokes and many others. About the end of the nineteenth century,
the adherence (“no-slip”) boundary condition was adopted, supposedly on the basis of
experimental evidence pertaining to the flow of fluids under moderate pressures and
velocities. Stokes is generally credited with the notion that the no-slip boundary con-
dition is appropriate for fluids like water. But his own comments belie this belief when
he writes: “Du Buat found by experiment that when the mean velocity of water flowing
through a pipe is about less than one inch in a second, the water near the inner surface of
the pipe is at rest. If these experiments may be trusted, the conditions to be satisfied in
the case of small velocities are those which first occurred to me, and which are included
in those given by supposing ν = ∞ . . . .” This implies no-slip. He further states that:
“I have said that when the velocity is not very small, the tangential force called into
action by the sliding of water over the inner surface of a pipe varies nearly as the square
of the velocity ….”
Recent experiments, that are capable of measuring velocities very close to the bound-
ary, question the correctness of this conventional wisdom for at least a class of polymeric
fluids. Also, the use of such a boundary condition for the flows of thin films or droplets
seems inappropriate. We shall not get into a discussion of the merits or demerits of the
adherence condition, or detail the types of slips that can occur. The interested reader
can find a clear and succinct account of the early history of the same in Goldstein [123].
More recent developments are discussed at length in Le Roux [157] and in the review
article by Le Roux et al. [158].
We shall be concerned with classical boundary conditions. However, we document
below some of the nonstandard boundary conditions that might be useful. To our knowl-
edge, while there are few rigorous mathematical results concerning the flows of non-
Newtonian fluids under such boundary conditions, they have been considered in some
detail, within the context of the Navier–Stokes fluid.
On the basis of molecular considerations, Navier [193] proposed that the slip velocity
(i.e., the component of the velocity in the tangential direction at the boundary) is directly
proportional to the shear stress in the fluid
v · τ = −k(T T n · τ ) on ∂Ω, (2.5.1)
where τ and n are the unit tangent and normal vectors to the boundary and k > 0 is a
constant. The above slip condition is also written in the form
(1 − θ)(v · τ ) + θ(T T n · τ ) = 0 on ∂Ω,
where θ = k/(1 + k). The “no-slip” condition corresponds to θ = 0 (or k = 0) and

the “free-slip” condition corresponds to θ = 1 (or k = ∞).13 Threshold-type slip
conditions seem to be particularly suitable for the flows of polymeric melts. Such fluids
13 An interesting discussion of the difficulties inherent to the no-slip boundary condition can be
found in Frehse and Málek [104].

2.5 Boundary Conditions for Fluids of the Differential Type 49
seem to flow only when a certain threshold for the shear stress is overcome (we have to
bear in mind that by definition a fluid cannot resist shear, and thus, it is merely the time
scale of the experiment that prevents the observer from noticing the flow that is taking
place). The boundary condition that is appropriate for the flow of polymeric melts is
|T T n · τ | ≤ G|T T n · n| ⇒ v · τ = 0,
v·τ
|T T n · τ | > G|T T n · n| ⇒ T T n · τ = −N (|v · τ |, |T T n · n|) ,
|v · τ |
where G is a given constant threshold and N is a given nonlinear function. The above
boundary condition states that if the shear stress is below a certain value (which depends
on the normal stress), then the fluid will not flow in the direction tangent to the boundary,
i.e., the “no-slip” condition holds. However, if the shear stress exceeds this value, the
fluid will slip past the boundary, the slip velocity depending on the normal stress. We
expect that the higher the normal stress, the lower will be the slip velocity. Usually, a
simpler form of threshold-type slip is used
|T T n · τ | ≤ g ⇒ v · τ = 0, (2.5.2)
v·τ
|T T n · τ | > g ⇒ T T n · τ = − , (2.5.3)
|v · τ |
where g is a constant threshold that is independent of the normal stress.

Berker [27] derived an elegant expression for the surface traction t on a solid surface
to which the fluid adheres. He showed that
∂ω
t = [− p + (2α1 + α2 )|ω|2 ]n + μ ω × n + α1 × n, (2.5.4)
∂t
which, by virtue of α1 + α2 = 0, reduces in the case of a steady flow to
t = [− p + α1 |ω|2 ]n + μ ω × n. (2.5.5)
Later, we shall find this formula particularly useful.

In the case of a thermodynamically compatible fluid of grade three, the traction t on
a solid surface to which the fluid adheres is given by (cf. Rajagopal [221])
∂ω
t = [− p + (2α1 + α2 )|ω|2 ]n + (μ + β3 tr A21 )ω × n + α1 × n. (2.5.6)
∂t
One of the great breakthroughs in fluid mechanics was to determine what would be
appropriate as boundary condition at solid boundaries. In the case of the Euler fluid
(that is a perfectly elastic fluid), the requirement of impenetrability at an impervious
boundary, i.e., v · n = 0 on the boundary, is sufficient to render the problem well
posed. But an Euler fluid slips while it flows past a solid boundary, it was felt, based
on many experimental observations, that the requirements that a viscous fluid adheres
50 2 Mechanics
to the solid boundary surface would be a reasonable assumption; and this is indeed the
case for the classical Navier–Stokes fluid. Interestingly, this condition is sufficient to
render problems to be well-posed even in the case of a fluid of complexity one (such
as power-law fluids, etc.). However, careful experiments suggest that even in the case
of fluids that are considered to be of the Navier–Stokes type, the fluid may slip at the
boundary or stick-slip at the boundary. Also, if we do not have a solid boundary but
a porous boundary or a free surface, modifications have to be made to the adherence
boundary condition.
When we leave the domain of the Navier–Stokes theory (or the theory of fluids of
complexity one) and enter into the realm of fluids of higher complexity, we immediately
get bogged down into a quagmire. For instance, consider the existence of solutions to
fluids of grade greater than one, subject to the no-slip boundary condition. It is not
surprising that one is able to prove existence of solutions to equations of order greater
than two subject to the no-slip boundary condition, a condition which is sufficient for
establishing existence of solutions to second-order partial differential equations. To
make matters clear, it is best to consider the possibility that in special flows a third-
order partial differential equation (namely, the equation governing the flow of a fluid of
grade two) could reduce to a third-order ordinary differential equation, and the no-slip
boundary condition is not sufficient to fully determine the solution to the problem;
in fact an additional condition is necessary. In other words, a one-parameter family
of solutions might be possible.14 What is however remarkable is that for a class of
flows, not only can we show existence of solutions, but we can also show that such
solutions are unique. These uniqueness results are by and large possible because such
results have been established for small data. This is also true for the Navier–Stokes
equations, but there is a marked difference between considering small data for the
Navier–Stokes equations, versus those for a fluid of grade two. In the case of the
Navier–Stokes equations, linearizing the problem does not lead to a reduction in order
of the partial differential equation. However, in the case of the equations of motion for
a fluid of grade two, linearization leads to a reduction of the order of the equation. In
fact, this is the crux of the matter. By considering small data, the higher order nonlinear
terms are neglected, thereby allowing the no-slip boundary condition to be sufficient
for establishing uniqueness! However, if we are not able to “control” those higher order
nonlinearities, we cannot establish such results. We can understand these issues within
the context of specific examples wherein these higher order nonlinear terms reduce to
higher order linear terms.
Here are a couple of specific problems that illustrate certain subtle issues concerning
boundary conditions for fluids of grade two and higher. Let us consider the flow of a
grade-two fluid past an infinite porous plate, at which we can either inject or suck in
the fluid. Let us seek a special solution of the form
v(x, y, z, t) = u(y)i + v(y) j . (2.5.7)
14 More generally, in the flow of a fluid of grade n, the equations of motion are partial differential
equations of order n + 1. However, in view of thermodynamic restrictions, it is possible that the

equations are of lower order as is the case of third-grade fluids.
We shall show that the problem admits an infinity of solutions of the form (2.5.7) that
satisfy the no-slip boundary condition. Of course, it is possible that the equations might
admit solutions other than those of the form (2.5.7), but this emphasizes the fact that
the problem has nonunique solutions that satisfy the no-slip boundary condition. The
incompressibility constraint (2.4.16) gives
v(y) = v0 = a constant,
while (2.4.25) with b = 0 and (2.4.19) reduce to (cf. Rajagopal and Gupta [230]),
where the prime denotes derivation with respect to y
∂p
μ u + α1 v0 u − v0 u = ,
∂x
∂p
(2α1 + α2 )[(u )2 ] = ,
∂y
∂p
0= .
∂z
We recognize that if v0 > 0, we have injection, and if v0 < 0, we have suction at the
plate. On defining
p̂ = p − (2α1 + α2 )(u )2 ,
the problem reduces to
∂ p̂ ∂ p̂ ∂ p̂
μ u + α1 v0 u − v0 u = , = = 0. (2.5.8)
∂x ∂y ∂z
We observe that (2.5.8) is one order higher than that which corresponds to the Navier–
Stokes fluid. As it is linear, it is trivial to establish that its exact solution is
u(y) = C1 em 1 y + C2 em 2 y + C3 ,
where 1
1 μ μ 2 4 2
m1 = ± + .
2 2 α1 v0 α1 v0 α1
Since α1 > 0, > 0, μ > 0, and v0 > 0, we note that m 1 = m 2 ; without loss
of generality, let us assume that m 1 > m 2 . We use the boundary conditions in order
to determine the constants. For the problem under consideration, no-slip at the porous
plate implies that
u(0) = 0. (2.5.9)
Let us suppose that the free stream velocity is U , i.e.,
lim u = U. (2.5.10)
y→∞
52 2 Mechanics
In general, these two boundary conditions are not sufficient to completely determine
the solution to the problem, as is shown in Rajagopal and Gupta [230].
The above example considers a flow in an unbounded domain and there might be
a misapprehension that the difficulty is due to the unboundedness of the domain. In
fact, the results are worse in the case of a domain bounded in one dimension. If the
flow were to take place between two porous disks, then it transpires that the equation
governing the problem are exactly the same as (2.5.8) and once again, we have the
boundary condition (2.5.9). However, in contrast to the flow in an unbounded domain,
we cannot take recourse to any asymptotic condition such as (2.5.10). We are thus at an
impasse, as the boundary condition (2.5.9) is not sufficient to determine the solution to
the problem. A similar difficulty presents itself in the case of the radial flow of a fluid
of grade two in the annular region between two concentric cylinders (cf. Bernstein and
Fosdick [35], Rajagopal and Kaloni [232]).
Next, let us consider the steady radial flow of a fluid of grade two in an annulus
bounded by two porous co-axial cylinders. We shall assume a velocity field of the form
(in cylindrical polar coordinates)
v(r, θ, z, t) = u(r )er + v(r )eθ ,
where er is the unit vector in the radial direction and eθ in the angular direction. The
equations of motion in the θ direction reduce to (the prime denotes derivation with
respect to r )
v T v v D
(v − ) + (v − ) + Re = ,
r r r r μr 2
where we have appropriately nondimensionalized the equations (cf. Rajagopal and

Kaloni [232]), and
V R1 α1 V R1
Re = , T = ,
μ μ
where V is a characteristic velocity, R1 is the radius of the inner cylinder, Re is the

Reynolds number, and D is a constant which appears by virtue of our eliminating the
pressure and then integrating the equation. The no-slip boundary condition leads to
v(R1 ) = v(R2 ) = 0,
where R2 is the radius of the outer cylinder. If we could determine v(r ), then we could
use it to find u(r ) from the equations of motion in the radial direction. But the problem
is that we cannot determine v(r ). Indeed, a straightforward computation leads to, if
Re = 2:
r2 r2 r2 r2 D
v(r ) = C1r e−( 2T ) χ1 ( ) + C2 r e−( 2T ) χ2 ( ) + ,
2T 2T μ( Re
2 − 1)r
and if Re = 2,
r2
r2 r2 r2
v(r ) = C1r e−( 2T ) χ1 (
) + C2 r e−( 2T ) χ2 ( )
2T 2T
D 2T r2 r2 1 r2
+ 2 2
ln( ) − 2e−( 2T ) + e( 2T ) ,
2μr r 2T r
where
⎧
⎪
⎪ Deξ
⎪
⎨ if Re = 2,
4μT ( Re2 − 1)ξ
χ1 (ξ ) =
⎪
⎪ D
⎪
⎩ eξ ln(ξ )dξ if Re = 2,
4μT ξ
χ2 (ξ ) = χ1 (ξ ) ln(ξ ) + χ3 (ξ ) + χ4 (ξ ) ,
Γ (a + k)Hk Re 1
χ3 (ξ ) = , a= − 2, χ4 (ξ ) = Γ (a − 1)Γ (a) ,
k!(1 + k)!Γ (a) 2 ξ
k−1
−1 1 1
Hk = − − ,
a +r r +ν r +1
ν=0
and Γ denotes the usual Gamma function. We have to determine three constants C1 ,
C2 and D, but we have only two equations.
The examples considered thus far correspond to
v · n = 0 on ∂Ω,
and thus one might jump to the erroneous conclusion that the nonuniqueness of solutions
stems from the boundary not being impervious. Unfortunately, this is not the case, the
difficulties being far deeper. To illustrate the possibility of nonunique solutions even
when v · n = 0, let us consider the flow engendered above an elastic stretching sheet,
occupied by a grade-two fluid, due to the sheet being stretched such that the velocity at
a point (x, 0, z) is proportional to the x coordinate. Following Rajagopal et al. [233],
we seek solutions of the form
v(x, y, z, t) = (x f (y) + g(y))i − f (y) j . (2.5.11)
It follows from (2.5.11), (2.4.19) and (2.4.25) that

f + f f − ( f )2 + k 2 f f − ( f )2 − f f iv = 0,
(2.5.12)
g + f g − f g + k g f − ( f g − f g − f g iv ) = 0,
where prime denotes differentiation with respect to y and k is an appropriate nondi-

mensional variable
54 2 Mechanics
√
α1 C 2 Re U∞
k= , Re = √ ,
U∞ 2 Cμ
C is the proportionality constant for the stretching and U∞ is the free-stream velocity.
Thus, the equations reduce to those for the Navier–Stokes fluid when k = 0. Otherwise
when k = 0, the equations are of higher order than the equations for a Navier–Stokes
fluid. Let us consider the case g = 0; this implies that the sheet is stretched in a special
manner. The no-slip boundary condition at the stretching sheet implies
f (0) = 0, f (0) = 1, (2.5.13)
the asymptotic condition is

lim f = 0, (2.5.14)
y→∞
and we have a fourth-order equation with these three conditions: this problem admits
multiple solutions (cf. Chang et al. [60]). In the case of the Navier–Stokes equations,
we have third-order equations with these three conditions, and one can show that the
solution to the equations subject to these conditions is unique (cf. McLeod and Ra-
jagopal [189], Overman et al. [206]). Of course, we have used a semi-inverse method
and thus the full Navier–Stokes equations might admit other solutions.
2.6 Creeping Flows of Fluids of the Differential Type
By “creeping flows” of fluids of the differential type we mean flows of such fluids in
which inertial effects are being neglected. Unfortunately, this choice of terminology
leaves much to be desired as one associates “slowness” with the word “creeping.” In
general, this slowness of the velocity will require a variety of nonlinear terms which
depend on the velocity to be neglected, depending on the choice of the specific model of
the fluid of differential type. Here, we shall ignore the acceleration terms in the balance
of linear momentum while retaining all the nonlinear terms that may stem from the
divergence of the stress. Thus, the equations governing the creeping flow of a fluid of
the differential type have the form
− ∇ p + div[ f ( A1 , . . . , An )] = 0. (2.6.1)
Different choices for f lead to different governing equations.

In the case of a fluid of grade one, i.e., the Navier–Stokes fluid, we obtain the Stokes
equation
− ∇ p + μΔ v = 0. (2.6.2)
In 1851, in his influential paper concerning the motion of pendulums in a fluid,

Stokes [259] also studied the “slow” motion of a sphere in a fluid. He simplified the
Navier–Stokes equations by neglecting the inertial terms and obtained an exact solution
2.6 Creeping Flows of Fluids of the Differential Type 55
for the flow under consideration. Using this exact solution, Stokes derived the famous
formula for the drag due to the slow movement of a sphere in an infinite fluid medium.
In the case of an object moving in a Navier–Stokes fluid, the Stokes approximation
leads to the following expression for the drag:

Stokes drag = (− p n + μω × n) · κ da, (2.6.3)
∂Ω
where the object is moving along the κ direction, and p and ω = curl v correspond to
the solution ( p, v) for (2.6.2).
In the case of general fluids of complexity one, the creeping flow equations reduce
to
− ∇ p + div[ f ( A1 )] = 0. (2.6.4)
While solutions to (2.6.2) do not usually exhibit pronounced boundary layers, solutions
to (2.6.4) can develop sharp boundary layers depending on the form of f . We discuss
such layers in Section 2.7.
2.6.1 Creeping Flows of Fluids of Grade Two
Unlike the case for the flows of a Navier–Stokes fluid wherein the equations reduce
to the linear Stokes equation in the limit of the Reynolds number tending to zero, in
the case of a fluid of grade two, the equations reduce to a nonlinear partial differential
equation that can admit possible nonunique solutions. To see this, let us first appropri-
ately nondimensionalize the equations for the steady creeping flow of a fluid of grade
two. By taking the curl of (2.4.27), the equation becomes
μΔ ω + α1 curl(Δ ω × v) − curl(ω × v) = 0. (2.6.5)
Let us introduce nondimensional variables

x v
x̄ = , v̄ = , (2.6.6)
L V
where L and V are some characteristic length scale and velocity, respectively. Then, it
immediately follows that
Re
Δ x̄ ω̄ + curl x̄ (Δ x̄ ω̄ × v̄) − Re curl x̄ (ω̄ × v̄) = 0, (2.6.7)
Γ
where
VL L 2
Re := , Γ = , (2.6.8)
μ α1
and the subscript x̄ denotes the non-dimensional variable. In deriving the above, we
have supposed that μ > 0 and α1 > 0.
56 2 Mechanics
The nondimensional quantity Re is the usual Reynolds number. The nondimensional

Γ is a measure of the relative strengths of the inertial and normal stress effects. In
1964, Truesdell [273] realized that the term involving α1 determines the character of
the diffusion of vorticity from the boundary and discussed the change of absorption and
phase shift with the driving force for simple shearing flows.
We notice that even in the limit of Re tending to zero, if the ratio Re/Γ is finite and
possibly large, then the character of the solution will be determined by the terms which
this ratio multiplies, as they are the highest order terms in the equation as well as being
nonlinear.
Let us suppose that Re → 0 and Re/Γ is finite. Then (2.6.7) reduces to
Re
Δ x̄ ω̄ + curl x̄ (Δ x̄ ω̄ × v̄) = 0, (2.6.9)
Γ
which is referred to as the equation for the creeping flow of a fluid of grade two. As this
equation is nonlinear, it is possible that it could have more than one solution. On taking
the product of (2.6.9) with Δ x̄ ω̄ and dropping the overbars for the sake of notational
clarity, we obtain
Re 1
|Δ ω|2 + [ div(|Δ ω|2 v) − A1 Δ ω · Δ ω] = 0. (2.6.10)
Γ 2
On integrating (2.6.10) over the flow domain and using Green’s formula, we obtain

Re Re
|Δ ω|2 dv − A1 Δ ω · Δ ω dv + |Δ ω|2 v · n da = 0. (2.6.11)
Ω Γ Ω 2Γ ∂Ω
Now, let M > 0 denote the maximum of the eigenvalues of A1 over the flow
domain.Then
A1 Δ ω · Δ ω ≤ M |Δ ω|2 ,
and hence since Re and Γ are nonnegative, it follows that

Re Re
(1 − M) |Δ ω|2 dv + |Δ ω|2 v · n da ≤ 0. (2.6.12)
Γ Ω 2Γ ∂Ω
Now, if the flow is such that

R
(i) 1 > M, (ii) |Δ ω|2 v · n da = 0,
Γ ∂Ω
then, it immediately follows that

|Δ ω|2 dv = 0. (2.6.13)
Ω
Conditions (i) and (ii) are quite reasonable and apply to a sufficiently large class of
flows. If the absorption number is sufficiently large, then Γ > Re M. This condition
will also be met if the fluid is sufficiently viscous. The second condition (ii) will be met
when v · n = 0 on ∂Ω, in particular in all flows that satisfy v = 0 on ∂Ω.
If Δ ω is continuous, (2.6.13) implies that
Δ ω = 0, (2.6.14)
which one obtains by taking the curl of the equation for Stokes flow. This leads us to a
rather interesting history concerning the creeping flows of fluids of grade two. In 1966,
Tanner [263] remarked that the classical Stokes solution for plane flows is a solution to
the plane creeping flow of a fluid of grade two.15 He did not realize that the equations
are of different orders: the Stokes flow being of order two and that for a fluid of grade
two of order three. Nor did he realize that the equations for the creeping flow of a fluid
of grade two are nonlinear and might admit other solutions in addition to the Stokes
solution. Most importantly, he did not realize that it is possible that the equations for
the creeping flow of a fluid of grade two could admit a solution that is not a solution of
the corresponding Stokes problem (see Rajagopal [224]).
Under conditions (i) and (ii), Fosdick and Rajagopal [100]16 showed that the velocity
solution for the Stokes problem is the unique velocity solution for a fluid of grade two,
provided that the pressure field is appropriately modified. That is, if (v, p̂) solves the
Stokes problem, then (v, p) is the unique solution to the creeping flow of a fluid of
grade two, where
1
p̂ = p + α1 (v · Δ v + | A1 |2 ), (2.6.15)
4
provided v is such that the conditions (i) and (ii) are met. For more details, the interested
reader can refer to [100].
It is important to recognize that many flows do not meet conditions (i) and (ii). For
instance, v · n = 0 in many flows past a porous boundary, and it is highly unlikely that
(ii) is met. Similarly, (i) will not be met if the flow is sufficiently fast (but in this case,
the creeping flow equations themselves will not hold).
The drag on a body moving slowly in a fluid of grade two, the fluid adhering to its
boundary, is given by

1
Drag = Stokes Drag + α1 |ω|2 − | A1 |2 n · κ da, (2.6.16)
∂Ω 4
where ω and A1 are computed from a velocity field that satisfies the Stokes equations
and κ is the constant direction along which the body is moving. A straightforward
calculation establishes that
15 In a thermodynamically compatible fluid of grade two, Tanner’s comment apply to general three-
dimensional flows as α1 + α2 = 0. If in a plane flow, α1 + α2 = 0, it can be shown that the terms

that are multiplied by α1 + α2 reduce to the gradient of a scalar and hence can be absorbed in the
pressure.
16 Huilgol [137] showed that the Stokes solution is the only solution for the creeping flow equations
in the plane flow of a second grade fluid with α1 < 0.

58 2 Mechanics
1 1 3
|ω|2 − | A1 |2 = |∇ v|2 − tr[(∇ v)2 ]. (2.6.17)
4 2 2
It can be shown (cf. Fosdick and Rajagopal [99]) that if v is a sufficiently smooth vector
field defined on the complement Ω of a smooth open set Ω in R I 3 , satisfying div v = 0
in Ω and v = 0 on ∂Ω, then

tr[(∇ v)2 ]n da = 0. (2.6.18)
∂Ω
Then (2.6.16)–(2.6.18) immediately imply that

Drag = Stokes Drag + α1 |∇ v|2 n · κ da. (2.6.19)
∂Ω
Thus, for bodies that possess certain geometric symmetries, say for example
(x1 , x2 , x3 ) → (x1 , x2 , −x3 ),
it immediately follows that

|∇ v|2 n · κ da = 0,
∂Ω
and hence, the drag is the same whether the fluid is a Navier–Stokes fluid or a fluid
of grade two! In fact, if the body is star-shaped so that it is symmetric under the
transformation x → −x, once again, the drag will be the same in the two fluids.
In the case of a fluid of grade three, the expression for the drag takes the form

Drag = {− pn + μω × n + α1 |ω|2 n + β3 (tr A21 )ω × n} · κ da,
∂Ω
where κ is the constant direction along which the body is moving and the velocity field
v and the pressure p are obtained as solutions to the creeping flow problem for a fluid
of grade three, i.e.,
−∇ p + μΔ v + α1 Δ ω × v + β3 (tr A21 )Δ v + β3 A1 ∇(tr A21 ) = 0.
Early results concerning existence and uniqueness for fluids of grade two (cf. Cio-
ranescu and Ouazar [64, 65], Oskolkov [204], Dunn and Fosdick [87]) were for flows
that take place in compact domains. While there has been rigorous studies concerning
Stokes flows in exterior domains (cf. [120]), until recently, there have been no studies
concerning the creeping flow of a fluid of grade two in an exterior domain. Galdi and
Rajagopal [109] show the existence and uniqueness of the solution to the creeping flow
of a fluid of grade two in the exterior of a compact body when α1 ≥ 0 and α1 + α2 = 0.
A small perturbation of the “null” solution of the Navier–Stokes equations leads
to the Stokes flow equations, and such an approximation is called the Stokes flow
approximations. In 1927, Oseen [203] studied a different type of approximation to the
Navier–Stokes equations. He considered perturbations about a uniform solution. Other

than a passing reference to such an approximation, nothing has been done concerning
the equations within the context of higher grade fluids.
2.7 Boundary Layer Theories for Fluids

One of the great achievements of Prandtl was the development of the Boundary Layer
Theory for Navier–Stokes fluids that made much of the progress in aerodynamics pos-
sible. Prandtl recognized that at high Reynolds number, for the flow of a Navier–Stokes
fluid past a solid boundary, the vorticity was confined to a narrow region adjacent to
the boundary and outside this region the flow was similar to that of an inviscid fluid
(i.e., an Euler fluid). Here, we shall be concerned primarily with laminar boundary
layers in fluids of the differential type of grade two and three, and also for certain fluids
of complexity one.
While in Navier–Stokes fluids, boundary layers develop due to inertial effects at
sufficiently large Reynolds numbers, in nonlinear fluids it is possible for vorticity to
be confined adjacent to the boundary even in the limit of the Reynolds number tending
to zero by virtue of shear-thinning or shear-thickening. Furthermore, in fluids of grade
two and three, it is possible that we could have boundary layers with multiple deck
structures with different effects confined to different layers. For example in one layer,
we could have a concentration of vorticity (due to the effect of viscosity) and in another
layer, we could have a concentration of the effects of elasticity (or terms in which α1
appears).
It is important to discuss another mathematical issue related to boundary layers of
the Navier–Stokes fluid that has led to some misunderstanding in applied mathematics.
To illustrate the issue, let us consider the steady laminar flow of a Navier–Stokes fluid
in dimensionless form (we shall later introduce the nondimensionalization procedure
in some detail). It follows that
1 P
Δ x̄ v̄ − grad x̄ p̄ = [grad x̄ v̄]v̄,
Re V 2
where v̄ is the nondimensional velocity, p̄ the nondimensional pressure, V is some

characteristic velocity for the problem of interest, P is some characteristic pressure and
Re stands for the Reynolds number defined by (2.6.8):
V L
Re := .
μ
We now notice that if Re → ∞, then the equation loses the highest order terms and we
have a singular perturbation. In fact, many of the advances made in singular perturbation
theory can be directly traced to important applications in fluid mechanics. Boundary
Layer Theory for Navier–Stokes fluids goes far beyond the above observation that as
Re → ∞, the equation essentially reduces to the equation for an Euler fluid. This
60 2 Mechanics
equation holds outside a narrow region where the effects of viscosity dominate. We
see from the definition of the Reynolds number that it is a measure of the ratio of
inertial effects to viscous effects, and when viscous effects are dominant, the Reynolds
number is small. It is also important to bear in mind that in the limit of the Reynolds
number tending to infinity, the usual “no-slip” boundary conditions are more than
what is necessary. In Boundary Layer Theory, even in the narrow region adjacent to the
boundary layer where the effects of viscosity cannot be neglected, further simplifications
are made based on an order of magnitude analysis for the components of the velocity
field. In this section, we shall carry out such an analysis for fluids of grades two and
three. The important point that we would like to make is that the confinement of vorticity
in fluids flowing past solid boundaries can, in the case of nonlinear fluids, occur in the
limit of Re → 0.
2.7.1 Boundary Layers in the Limit of Zero

Reynolds Number
To illustrate the confinement of vorticity adjacent to solid boundaries in the limit of

zero Reynolds number, let us consider, the flow of a fluid of complexity one of the
following type:
T = − pI + μ(tr A21 )m A1 . (2.7.1)
This fluid is a shear-thinning fluid if m < 0; we shall see that, due to such shear-thinning,
very pronounced boundary layers can develop in the limit of the Reynolds number Re
tending to zero.
Let us consider the flow of a fluid whose stress is given by (2.7.1), in the absence
of body forces, and in the limit of Re tending to zero. As mentioned in the preceding
section, flows in which Re → 0 are usually referred to as “creeping flows”. Substituting
(2.7.1) into the balance of linear momentum yields
dv
= −grad p + μdiv (tr A21 )m A1 + b.
dt
We now have b = 0, and ignoring the Reynolds number (which is tantamount to

neglecting inertial effects) and considering the steady problem, leads to

− grad p + μdiv (tr A21 )m A1 = 0. (2.7.2)
We shall illustrate our main thesis by means of a special example. Let us consider
the Jeffrey–Hamel flow of a fluid modeled by (2.7.2) between two intersecting planes.
Following Jeffrey [142] and Hamel [130], we assume a velocity field, in a cylindrical
polar coordinate system, of the form
F(θ)
v= er . (2.7.3)
r
2.7 Boundary Layer Theories for Fluids of the Differential Type 61
We can get rid of the pressure from (2.7.2) by taking the curl of the equation and on
substituting (2.7.3) into the resulting equation, we obtain

(2 + 4m) 2F [8F 2 + 2(F )2 ]m + F [8F 2 + 2(F )2 ]m

+ 4m(2 + 4m)F [8F 2 + 2(F )2 ]m − 2F [8F 2 + 2(F )2 ]m

+ 8m F[8F 2 + 2(F )2 ]m + F [8F 2 + 2(F )2 ]m = 0. (2.7.4)
The appropriate boundary conditions for the problem are
F(α) = 0, F(−α) = 0. (2.7.5)
However, these boundary conditions do not suffice to solve (2.7.4), as it is a third-order

equation (in eliminating the pressure, we have increased the order of the equation). We
fill this lacuna by requiring that
α
F(θ) dθ = Q, (2.7.6)
0
where Q denotes the flow rate. Interestingly, when m = 0, the above equation reduces
to (cf. Birkhoff and Zarantanello [40]):
F + 4F = 0,
for which we have an exact solution if we impose the boundary conditions (2.7.5) and
the integral condition (2.7.6). We shall not get into a detailed discussion of the nature
of this solution. Suffice it is to say that the equations admit very pronounced boundary
layers for a certain range of values of m (cf. Mansutti and Rajagopal [176]). Recall
that we have neglected inertial effects in deriving (2.7.2), i.e., we have considered the
case where the Reynolds number Re = 0; yet we have sharp concentration of vorticity
adjacent to the boundaries.
2.7.2 Development of Boundary Layers in Flows

of Fluids of Grade Two
In a fluid of grade two, it is possible that a boundary layer with a two-deck structure
develops, where in one of the layers, the viscous and inertial effects are comparable,
while in the other, the effects of the viscous and elastic (to be more precise, the influence
of α1 ) are comparable; the flow outside this two-deck structure being essentially that of
an Euler fluid (cf. Rajagopal et al. [231]). Once again, let us introduce nondimensional
quantities through
x v p
x̄ = , v̄ = , p̄ = .
L V P
62 2 Mechanics
For steady flows, we find

2
μV α1 V 2 V α1 V 2 α1 P
Δ x̄ v̄ + (Δ x̄ ω̄ × v̄)+grad x̄ v̄ · Δ x̄ v̄ + | Ā1 | − grad x̄ p̄
2
L2 L3 L2 4L 2 L
V2
= [grad x̄ v̄]v̄.
L
We shall restrict ourselves to two-dimensional flows. On taking the curl of the above
equation, we obtain:
Re
Δ x̄ ω̄ + curl x̄ (Δ x̄ ω̄ × v̄) − Re curl x̄ (ω̄ × v̄) = 0,
Γ
where Re and Γ are defined by (2.6.8), ω̄ denotes the scalar curl
∂ v̄2 ∂ v̄1
ω̄ = − ,
∂ x̄1 ∂ x̄2
and ω̄ × v̄ = ω̄(−v̄2 , v̄1 ). The terms involving Γ can significantly affect the diffusion
of the vorticity from the boundary (cf. Truesdell [275]) and can therefore affect the
structure of the boundary layer. Dividing both sides by Re, we can rewrite the above
equation in the form:
1 1
Δ x̄ ω̄ + curl x̄ (Δ x̄ ω̄ × v̄) − curl x̄ (ω̄ × v̄) = 0. (2.7.7)
Re Γ
We now recognize that Γ1 is the term that multiplies the highest order term in the partial
differential equation. Thus boundary layers due to the dominance of α1 are possible as
Γ → ∞, and these equations are
1
curl x̄ (Δ x̄ ω̄ × v̄) = 0. (2.7.8)
We also notice that ω̄ = 0 automatically satisfies (2.7.8), and thus, we could have a
situation such that when Γ1 → ∞, we have a narrow region adjacent to the boundary
governed by (2.7.8) with the flow being that due to an Euler fluid outside this boundary
layer.
We realize issues are rather delicate in the case of equation (2.7.7). For instance, it is
possible that Re → 0, while the ratio Re Γ is O(1), in which case, (2.7.7) would reduce
to
Re
Δ x̄ ω̄ + curl x̄ (Δ x̄ ω̄ × v̄) = 0. (2.7.9)
Γ
We notice that these different equations are of different orders and great care must be
exercised, as the boundary conditions that are necessary to make these equations well
posed are quite different. We shall discuss these issues subsequently.
2.7.3 Inertial Boundary Layers
By inertial boundary layers, we mean boundary layers at high Reynolds number. To

illustrate the development of inertial boundary layers in fluids of grade two, we shall
consider the flow of a grade-two fluid past a wedge. This is the counterpart of the
problem studied by Falkner and Skan [96] for the Navier–Stokes fluid. Using the same
approximations as that used for the Navier–Stokes fluid, we can reduce the problem to
solving (cf. Mansutti et al. [175])
∂u ∂v
+ = 0,
∂x ∂y
(2.7.10)
∂u ∂v dV ∂ 2u ∂ ∂ 2u ∂u ∂ 2 v ∂ 3u
u +v =V + 2 +κ (u 2 ) + + u ,
∂x ∂y dx ∂y ∂x ∂y ∂ y ∂ y2 ∂ y3
where V is the free stream velocity, u and v are appropriately nondimensionalized

velocities in the x and y directions, respectively, and
α1
κ := √ ,
L 2 Re
where L is an appropriate length scale.

The boundary conditions to be enforced are
u = 0, v = 0 at y = 0,
(2.7.11)
u → V as y → ∞.
Here y = 0 is one of the faces of the wedge (see Figure 2.7). But here also these
boundary conditions are insufficient, as we have increased the order of the equation by
eliminating the pressure. We augment the above boundary conditions with the condition
that
Tx y → 0 as y → ∞, (2.7.12)
which reflects the fact that the shear stresses vanish as y tends to infinity. We find that
(2.7.10), subject to (2.7.11) and (2.7.12) admits solutions that have very pronounced
boundary layers adjacent to the wedge, and the boundary layers become more pro-
nounced (thinner) as Re increases.
In the previous section, we discussed the issue of the lack of boundary conditions for
general flows of fluids of grade two. We come across this difficulty here as we notice
that (2.7.10) has higher order than the Navier–Stokes equations by virtue of the term
κu ∂∂ yu3 . Many studies on such problems use a formal perturbation approach in which the
3
velocity is expressed as a power series in κ that is assumed to be small. The rationale

for such a formal approach is that in a “slightly” non-Newtonian fluid, α1 is small and
thus κ is small and the Reynolds number is large. However, it is important to recognize
that the small parameter κ multiplies the highest order derivative in the equation, and
thus, we have a singular perturbation. It is incorrect to treat it as a regular perturbation
64 2 Mechanics
Fig. 2.7 Flow past a wedge
(cf. Bourgin and Tichy [46] for a discussion of the delicate issues concerning the singular
nature of the perturbation). The additional boundary condition that we have introduced
helps us overcome this difficulty. However, we cannot always resort to such methods
to augment boundary conditions. We are able to use such an asymptotic assumption
based on “physical” reasoning as we have an unbounded domain. Were the flow taking
place past an object inside a pipe, say, we would not be able to augment our boundary
condition in such a manner.
2.7.4 Flows of Fluids of Grade Two with a Free Surface
Most of the mathematical results concerning fluids of grade two suppose that the fluid
adheres to a solid boundary. We have already discussed the difficulties associated with
the flow of fluids of grades two or greater past porous boundaries. We now consider
the problem of the flow of a fluid of grade two wherein a part of the boundary is free,
i.e., the fluid is free of traction (we are of course neglecting the ambient pressure).
In the case of a general free surface, we would have to express this traction-free
condition in terms of a local coordinate system with one coordinate normal to the
surface and the two others lying on the tangent plane. Let n, τ 1 and τ 2 denote the
unit normal and any two orthonormal vectors lying on the tangent plane at the point of
interest. The fact that the surface is free of traction at that point implies that
(T T n) · n = 0, (T T n) · τ 1 = 0, (T T n) · τ 2 = 0. (2.7.13)
If the surface were exposed to atmospheric pressure, then the first equation in (2.7.13)
would take the form
(T T n) · n = patm ,
where patm is the atmospheric pressure. These innocuous looking conditions turn out to
be quite complicated expressions, even if the surface is planar, say x3 = 0 in a cartesian
system (x1 , x2 , x3 ). In this case, the second and third equations in (2.7.13) take the
following forms (cf. Galdi and Rajagopal [108]):

∂u 1 ∂u 3 d ∂u 1 ∂u 3 ∂u 1 ∂u 1 ∂u 2 ∂u 1 ∂u 2
tx1 = μ( + ) + α1 ( + )+2 + ( + )
∂ x3 ∂ x1 dt ∂ x3 ∂ x1 ∂ x1 ∂ x3 ∂ x3 ∂ x1 ∂ x1

∂u 3 ∂u 1 ∂u 3 ∂u 1 ∂u 1 ∂u 3 ∂u 2 ∂u 2 ∂u 3 ∂u 3 ∂u 3
+ ( + )+ ( + )+ ( + )+2
∂ x3 ∂ x3 ∂ x1 ∂ x1 ∂ x3 ∂ x1 ∂ x1 ∂ x3 ∂ x2 ∂ x1 ∂ x3

∂u 1 ∂u 1 ∂u 3 ∂u 1 ∂u 2 ∂u 2 ∂u 3 ∂u 3 ∂u 1 ∂u 3
+ α2 2 ( + )+( + )( + )+2 ( + )
∂ x1 ∂ x3 ∂ x1 ∂ x2 ∂ x1 ∂ x3 ∂ x2 ∂ x3 ∂ x3 ∂ x1
= 0 on the free surface,
and

∂u 2 ∂u 3 d ∂u 2 ∂u 3 ∂u 1 ∂u 2 ∂u 1 ∂u 2 ∂u 2
tx2 = μ( + ) + α1 ( + )+ ( + )+2
∂ x3 ∂ x2 dt ∂ x3 ∂ x2 ∂ x3 ∂ x1 ∂ x2 ∂ x2 ∂ x3

∂u 3 ∂u 2 ∂u 3 ∂u 1 ∂u 1 ∂u 3 ∂u 2 ∂u 2 ∂u 3 ∂u 3 ∂u 3
+ ( + )+ ( + )+ ( + )+2
∂ x3 ∂ x3 ∂ x2 ∂ x1 ∂ x3 ∂ x1 ∂ x2 ∂ x3 ∂ x2 ∂ x2 ∂ x3

∂u 2 ∂u 1 ∂u 3 ∂u 1 ∂u 2 ∂u 2 ∂u 3 ∂u 3 ∂u 2 ∂u 3
+ α2 ( + )( + )+2 ( + )+2 ( + )
∂ x1 ∂ x2 ∂ x1 ∂ x3 ∂ x2 ∂ x3 ∂ x2 ∂ x3 ∂ x3 ∂ x2
= 0 on the free surface,
where the velocity v = (u 1 , u 2 , u 3 ) and n = ex3 is the unit vector in the x3 -direction.
On the free surface, we also need to meet the condition
u 3 = v · n = 0. (2.7.14)
We note that the expressions for tx1 and tx2 are evolution equations, a situation quite
unlike the situation one has in the Navier–Stokes theory. We first conclude from (2.7.14)
that
∂u 3 ∂u 3
= = 0 on the free surface. (2.7.15)
∂ x1 ∂ x2
Thus, the above equations for tx1 and tx2 simplify to

∂u 1 d ∂u 1 ∂u 1 ∂u 1
tx1 =μ + α1 ( )+3
∂ x3 dt ∂ x3 ∂ x1 ∂ x3

∂u 2 ∂u 1 ∂u 2 ∂u 3 ∂u 1 ∂u 2 ∂u 2
+ ( + )+ + (2.7.16)
∂ x3 ∂ x1 ∂ x1 ∂ x3 ∂ x3 ∂ x1 ∂ x3

∂u 1 ∂u 1 ∂u 2 ∂u 1 ∂u 2 ∂u 3 ∂u 1
+ α2 2 + ( + )+2 = 0,
∂ x1 ∂ x3 ∂ x3 ∂ x2 ∂ x1 ∂ x3 ∂ x3
66 2 Mechanics
and

∂u 2 d ∂u 2 ∂u 1 ∂u 2 ∂u 1
tx2 =μ + α1 ( )+ ( + )
∂ x3 dt ∂ x3 ∂ x3 ∂ x1 ∂ x2

∂u 2 ∂u 2 ∂u 2 ∂u 3 ∂u 1 ∂u 1
+3 + + (2.7.17)
∂ x2 ∂ x3 ∂ x3 ∂ x3 ∂ x3 ∂ x1

∂u 1 ∂u 2 ∂u 1 ∂u 2 ∂u 2 ∂u 2 ∂u 3
+ α2 ( + )+2 +2 = 0.
∂ x3 ∂ x1 ∂ x2 ∂ x2 ∂ x3 ∂ x3 ∂ x3
We notice that in the case of a Navier–Stokes fluid, (2.7.16) and (2.7.17) reduce to
∂u 1 ∂u 2
tx1 = μ = 0, tx2 = μ = 0,
∂ x3 ∂ x3
and then the free-surface boundary condition can be expressed in terms of the velocity
and its gradients, i.e.,
∂u 1 ∂u 2
= 0, = 0 on the free surface,
∂ x3 ∂ x3
and the first equation in (2.7.13).

It is interesting to note that if the first equation in (2.7.13) holds, then even in the
case of a fluid of grade three, the tractions tx1 and tx2 vanish. It can be shown that under
certain conditions, even in the case of a fluid of grade two, the first equation in (2.7.13)
is equivalent to the tractions tx1 and tx2 vanishing on the free surface (cf. Galdi and
Rajagopal [108]).
2.7.5 Universal Flows
A flow is said to be universal for a class of fluids if it is possible in every member of the
class, the body forces being held fixed. Thus, a universal flow of a fluid of grade n is a
universal flow for a fluid of grade n − 1. The converse may or may not be true. There
has been considerable amount of effort aimed towards determining all the universal
solutions of the Navier–Stokes fluid (see the work by Marris and his coworkers [177–
181]). The higher grade the fluid is, the fewer the universal flows that are possible in
that class of fluids (see Fosdick and Truesdell [103]).
Since Euler fluids are particular examples of fluids of grade n, it is necessary that
universal flows of fluids of grade n meet:
skw grad ẍ = 0, (2.7.18)
(where skw grad denotes the skew part of the tensor), i.e., the flows should preserve
circulation. For the case of grade-one fluids, i.e., the Navier–Stokes fluid, in addition
to (2.7.18), we need to meet
skw grad div A1 = 0. (2.7.19)
In the case of a fluid of grade two, it can be shown (cf. Fosdick and Truesdell [103],
Truesdell and Rajagopal [277]) that in addition to (2.7.18) and (2.7.19), one needs to
satisfy for a potential Ψ
div A2 = −grad Ψ (2.7.20)
for a flow to be universal.

We notice that the velocity field, in order that it be universal, has to meet ever
increasing partial differential equations. In the case of a fluid of grade three, in addition
to (2.7.18)–(2.7.20), one needs to meet
skw grad div A3 = 0,

(2.7.21)
skw grad div[(tr A2 ) A1 ] = 0,
in order for the flow to be universal.

The class of universal solutions for grade-three fluids has not been fully delineated,
let alone the Navier–Stokes fluid that has a far richer class of universal solutions.
2.8 Rate Type Fluids
The first one-dimensional rate type fluid model was developed in 1866 by Maxwell in
his celebrated paper on the dynamical theory of gases [184]. He recognized that there
is a concept of the “time of relaxation” that is inherent in every body. His rate type
fluid model is based on a superposition of viscous and elastic response, and while it
could be thought of in terms of a mechanical analogue consisting of a spring and a
dashpot in series, Maxwell himself gave no such analogy. His model allowed one to
describe the stress relaxation that is exhibited by many bodies. In 1874, Boltzmann [44]
also developed a linear one-dimensional model for describing the viscoelastic response
of fluids. Nearly, four decades later, in 1915, Jeffery [142] developed a rate type vis-
coelastic model that has enjoyed a great deal of popularity. Then in 1939, Burgers [55]
introduced the one-dimensional analogue of a model in which the Cauchy stress T has
the form:
T = − p I + S, (2.8.1)
∇ ∇∇ ∇
S + λ1 S +λ2 S = 2η1 D + 2η2 D, (2.8.2)
where S is the constitutively determined part of the stress, − p I is the indeterminate

spherical stress due to the constraint of incompressibility, λ1 , λ2 , η1 , η2 are constant,
and the superscript ∇ is defined by (2.3.27). Several models derived later on are partic-
ular cases of (2.8.2) with an appropriate choice of parameters. Indeed, if λ2 = 0, then
(2.8.1)–(2.8.2) is the Oldroyd-B model (2.8.6), if λ2 = η2 = 0, it is the Maxwell model
68 2 Mechanics
(2.8.4), and if λ1 = λ2 = η2 = 0, then it is the incompressible Navier–Stokes model

(2.4.17).
In 1950, Oldroyd [200] was the first to create a systematic framework for rate type
models describing the response of viscoelastic fluid within a three-dimensional context.
He introduced convective derivatives of appropriate physical quantities that ensured
their frame indifference and he allowed for the current stress of the body to depend
on the history of its deformation. He also provided explicit formulae for calculating
the evolution of material symmetry due to the deformation.17 This remarkable work of
Oldroyd’s drew a great deal from the earlier study of Frohlick and Sack [106] who had
developed three-dimensional models that were not properly frame-indifferent. Oldroyd
generated many models and some of these had more than one characteristic time scale
associated with their response. One that subserves several of the popular models is the
Oldroyd 8 constant model wherein the Cauchy stress T has the form (2.8.1) with
∇ μ0 μ1 ν1
S + λ1 S + (tr S) A1 + (S A1 + A1 S) + (tr(S A1 )I)
2 2 2

∇ ν2
= η0 A1 + λ2 D −μ2 A21 + (tr A21 )I . (2.8.3)
2
This model includes the Oldroyd-B, Maxwell and Navier–Stokes fluids as special cases.
While a great variety of rate type models have been used to describe viscoelastic
fluids, rigorous mathematical results have been established for only a few of them.
As this work is concerned with a rigorous mathematical treatment of non-Newtonian
fluids, we shall restrict our discussion to these few models, though we mention in
passing closely related models of the rate type.
Few rate type models are such that the differential equation for the extra stress can
be integrated to obtain an integral representation for this extra stress. In the case of the
Maxwell model which is given by (2.8.1) and
∇
S + λ1 S= 2η1 D, (2.8.4)
one can obtain the following equivalent integral representation for the stress:
∞
η1 − λS
S=2 e 1 [C −1
t (t − s) − I] ds. (2.8.5)
λ21 0
The expression for the extra stress in this form leads to an integro-differential equation.
An approximating scheme has been developed to express (2.8.3) as an integral model
under special flow conditions, but we shall not discuss such integral approximations
here (see Huilgol [138] for a detailed discussion).
17 Thisevolution of material symmetry was given with regard to physical constants that appear in
the constitutive equations and is different from the evolution of material symmetry of materials
with multiple natural configurations (cf. Rajagopal and Srinivasa [235] for a discussion of material
symmetry in anisotropic liquids).
2.8 Rate Type Fluids 69
The most popular model due to Oldroyd is the “upper convected Oldroyd-B model.”
In this model, the Cauchy stress T is related to the fluid motion through (2.8.1) and
∇ ∇
S + λ1 S= 2η1 ( D + λ2 D), (2.8.6)
where η1 is the viscosity and λ1 and λ2 , satisfying λ1 > λ2 > 0, are the relaxation
and retardation time, respectively. As mentioned above, when λ1 = λ2 = 0, the above
model reduces to the classical incompressible Navier–Stokes fluid model (2.4.17) and
when the retardation time λ2 is set to zero, the model reduces to the Maxwell model
(2.8.4).
There is an alternative way for expressing the Cauchy stress in a Maxwell fluid
that could prove to have some advantages. This alternate formulation stems from a
thermodynamic framework that has been developed for describing the response of
bodies such as the Maxwell fluids that are capable of instantaneous elastic response.
The stress within such a thermodynamic framework is expressed as
T = − p I + μB κ p(t) ,
where B κ p(t) is the Cauchy-Green tensor measured from an appropriate stress-free

configuration κ p(t) , that satisfies the evolution equation
∇
B κ p(t) = λ[B κ p(t) − γ I].
We shall not get into a detailed discussion of the thermodynamic framework for ma-
terials with multiple natural configurations within which most rate type models that
are capable of instantaneous response fit. The interested reader can find the details
in Rajagopal and Srinivasa [234]. Here, we provide a redacted version of the same.
The models that arise from the use of the framework automatically allow for all the
response characteristics of viscoelastic fluids, namely shear-thinning/shear-thickening,
creep, normal stress differences and stress relaxation. The methodology adopted in the
framework is in keeping with the original ideas of Maxwell in his development of a
model for viscoelastic fluids and automatically ensures that the second law is met.18
One can also make an identification between models developed within this framework
and those that use the notion of a conformation tensor. Depending on the choice for
the stored energy and the rate of dissipation, the relaxation time can depend on the
deformation, a feature that is exhibited by fluids such as blood (see Thurston [267]).
Most bodies have more than one natural (i.e., stress-free) configuration and as the
body deforms, the underlying natural configuration changes, as does the response from
18 It is worth emphasizing that Maxwell [184] developed his original model for a viscoelastic fluid to
describe the dynamic response of air, a fluid that can store as well as dissipate energy. The classical
Euler fluid is a perfectly elastic fluid incapable of dissipation; it can only store energy, and usually
gases are either modeled as ideal fluids or Van der Waal’s fluids. Once a choice is made for the
rate of dissipation, as it is nonnegative, it serves as a Lyapunov function and decreases with time
satisfying the minimum entropy production theorem of Onsager [201] (see also Prigogine [218])
that characterizes the steady states for special choices of the rate of entropy production (see [234]).
70 2 Mechanics
Fig. 2.8 Evolution of stress-free states
these natural configurations. An elastic body has one natural configuration modulo rigid
motions while a traditional “elastic-plastic” body has infinity of natural configurations,
the elastic response being possibly different from these natural configurations. From
now on, when we refer to natural configurations we always refer to it modulo rigid
motions.
A solid body that undergoes solid-to-solid phase transitions such as that from Austen-
ite to Martensite has a finite number of natural configurations. Viscoelastic fluids that
are capable of instantaneous elastic response can also be thought of as materials with
multiple natural configurations, infinity of them. The natural configuration correspond-
ing to a current deformed state is determined by allowing the body to instantaneously
stress relax to a stress-free state (see Figure 2.8). The viscoelastic body in question is
characterized by a stored energy function ψ which depends on the Cauchy–Green tensor
associated with the deformation from the natural configuration (i.e., the stretch associ-
ated with the elastic response), and a rate of dissipation function ξ which automatically
meets the requirement that it be nonnegative. Thus, the second law is automatically
enforced.
The evolution of the natural configuration is accompanied by dissipation and the
manner in which the natural configuration evolves is determined by maximizing the
rate of dissipation and in a full thermodynamic setting, the rate of entropy production
(see [234]). That is, among all possible admissible processes, the one that is chosen
is that which maximizes the rate of dissipation (see Rajagopal [225], Rajagopal and
Srinivasa [235] for a rationale for such a requirement). For viscoelastic fluids the rate of
dissipation depends upon the elastic stretch and the velocity gradient associated with the
evolution of the natural configuration. In order to make these ideas clear, we introduce
a few kinematical concepts.
As we shall not deal with one reference configuration, but several natural configu-
rations, let us carefully label the various functions based on the reference from which
they are measured. Let κ p(t) denote the natural configuration corresponding to the cur-
rent configuration κt (see Figure 2.8). If κ R is the reference configuration and if the
body has been subject to an inhomogeneous deformation, then it is not possible to un-
load to a stress-free configuration that is geometrically compatible (i.e., fits together)
in a three-dimensional Euclidean space. However, it is always possible to do so in a
non-Euclidean space (see Eckart [92, 235]). But if the deformation is homogeneous,
one can unload to a geometrically compatible stress-free state. Moreover, the notion of
configuration (see Noll [197]) is local, and hence, we can always locally unload to a
stress-free configuration.
Let F κ p(t) denote the gradient of the mapping from the configuration κ p(t) to κt (see
Figure 2.8).19 We define as before the Cauchy–Green tensors B κ p(t) and C κ p(t) through
B κ p(t) = F κ p(t) F κTp(t) , C κ p(t) = F κTp(t) F κ p(t) . (2.8.7)
We denote by G the mapping (see Figure 2.8)
G = F κ R →κ p(t) := F −1
κ p(t) F κ R .
Next, we define
C κ R →κ p(t) := G T G,
and it immediately follows that
B κ p(t) = F κ R C −1
κ R →κ p(t) F κ R .
T
The velocity gradient L κ p(t) is defined by
L κ p(t) := Ġ G −1 , (2.8.8)
and the symmetric part of L κ p(t) is defined by
1
Dκ p(t) = [L κ p(t) + L κTp(t) ]. (2.8.9)
2
As we shall not provide a detailed derivation but merely indicate the methodology for
generating rate type models, also we shall not define a variety of other kinematical
quantities and the relationships between them that are necessary for developing the
models. For our purposes, the above kinematical definitions suffice.
19 In
general, F κ p(t) is a mapping transforming the vectors belonging to the tangent space at a
material point of κ p(t) into the tangent space at the same material point in the configuration κt .
72 2 Mechanics
The stored energy ψ is assumed to depend on C κ p(t) . Since we shall be concerned

with isotropic fluids, it follows that ψ can depend on C κ p(t) only through its invariants,
I := IC κ p(t) = tr C κ p(t) = tr B κ p(t) ,

1
II := IIC κ p(t) = (tr C κ p(t) )2 − tr C 2κ p(t) = II B κ p(t) ,
2
III := IIIC κ p(t) = det C κ p(t) = III B κ p(t) .
Since we are interested in an incompressible fluid, ψ can only depend on I and II,
i.e.,
ψ = ψ̂(I, II).
We shall assume that the rate of dissipation ξ has the following dependence
ξ = ξ̂ (B κ p(t) , Dκ p(t) ).
Let us make the special choice

μ
ψ= (I − 3), ξ = η̄ Dκ p(t) · B κ p(t) Dκ p(t) ,
2
where μ and η̄ are constants. The above choices mean that we have a mechanical
analogue in which the spring stores energy like a neo-Hookean solid and a dashpot
which is in series with the spring that dissipates like a Navier–Stokes fluid. The fact
that we have a spring and dashpot in series leads to the rate of dissipation depending
on Dκ p(t) and not on D. The choice for the stored energy ensures that when there is no
elastic deformation, i.e., F κ p(t) = I, then I = 3 thus no energy is stored. Also, when G
is a constant, i.e., when the natural configuration does not change, then Ġ = 0 which
by virtue of (2.8.8) and (2.8.9) implies that the rate of dissipation ξ = 0. It then follows
(see [234] for details) that the stress in such a fluid is given by
T = − p I + μB κ p(t) , (2.8.10)

∇ 2μ
− B κ p(t) = B κ p(t) − λI , (2.8.11)
η̄
where
3
λ= . (2.8.12)
tr B −1
κ p(t)
Recall that the Cauchy stress in an upper-convected Maxwell fluid is given by (2.8.1)
with S defined through (2.8.4)
∇
T = − p I + S, S + λ1 S= 2η1 D.
By virtue of (2.3.28), (2.8.4) can be written as

∇ η1 ∇
S + λ1 S +2 I = 0.
λ1
Next, defining S̄ through

η1 ∇
S̄ = S + 2 I, (2.8.13)
λ1
we can rewrite (2.8.1) and (2.8.4) as
T = − p̄I + S̄, (2.8.14)
∇
η1
S̄ + λ1 S̄= 2 I, (2.8.15)
λ1
where
η1
p̄ = p + 2 .
λ1
We notice that (2.8.14), (2.8.15) can be re-written as
¯
T = − p̄I + S̄, (2.8.16)
∇
η1 ¯
S̄¯ + S̄= λ̄I, (2.8.17)
2μ
where
S̄¯ = μB κ p(t) . (2.8.18)
The equations (2.8.1), (2.8.4) and (2.8.16), (2.8.17) have exactly the same form, except
that in (2.8.17), λ̄ is not a constant.
If we make the additional assumption that the elastic strains are small in the sense
that
B κ p(t) − I = O(δ), δ 1, (2.8.19)
then it immediately follows that:
λ̄ = μ + O(δ 2 ),
and for this value of λ̄, the systems (2.8.1), (2.8.4) and (2.8.16), (2.8.17) are the same.
Thus, the rate type model reduces to the Maxwell model in the limit of small elastic
strains. That is, the classical Maxwell fluid stores energy like a linearized elastic spring
and dissipates energy like a Navier–Stokes fluid.
74 2 Mechanics
The above procedure can be used to generate a variety of properly frame-indifferent

and thermodynamically compatible models. It is worth noting that the tensor B κ p(t) bears
a striking similarity in the manner in which it appears in the constitutive relations, to
the conformation tensor C that is associated with the second moment of the end-to-
end molecular distance distribution (see Flory [98] and Beris and Edwards [26]). If
such an identification is made, then one can immediately see a similarity between the
constitutive theories. While C has some relation to configurations in view of it being a
distribution of molecular distance, there is yet no clear identifiable characteristic that C is
a measure of the stretch from a natural configuration, as end-to-end distance can change
without the stretch changing and vice-versa. Moreover, there is no clear meaning to the
evolution of the conformation tensor C. Also, while the enforcement of the constraint
of incompressibility is achieved by requiring that det B κ p(t) = 1, requiring the same of
the conformation tensor cannot have such a meaning though it is often interpreted in
such a manner.
Notice that the model defined through (2.8.10)–(2.8.12) has a relaxation time that
depends on the deformation. On the other hand, the models due to Maxwell, Oldroyd,
Burgers, and others have a constant relaxation time. In general, all the material moduli
such as the viscosity and relaxation time can depend not only on the manner in which
the body deforms but also on the pressure. Experiments on blood by Thurston [267]
show that the relaxation time of blood depends on the shear rate. Early experiments
on asphaltic bitumen by Saal and Koens [250] showed that the viscosity depended on
shear stress and the normal stress, while experiments by Bingham and Stephens [38]
showed the effect of pressure on the material properties.
Now, we proceed to derive the constitutive representation for the Oldroyd-B fluid
within the above thermodynamic framework. Let us keep the same choice of the stored
energy ψ:
μ
ψ = (I − 3),
2
and now choose the rate of dissipation ξ of the form
ξ = η̄ Dκ p(t) · B κ p(t) Dκ p(t) + η1 D · D;
this leads to the model (see [234])
T = − pI + μB κ p(t) + η1 D, (2.8.20)
∇ 2μ
− B κ p(t) = [B κ p(t) − λI], (2.8.21)
η̄
where λ is once again given by (2.8.12). When we assume that the elastic deformations
are small in the sense of (2.8.19), it can be shown (see [234]) that this model is equivalent
to
T = − pI + Ŝ, (2.8.22)
∇ ∇
Ŝ + λ1 Ŝ= 2η1 ( D + λ2 D). (2.8.23)
When the elastic deformations are not small, we have in place models (2.8.10)–(2.8.12)
and (2.8.20)–(2.8.21) capable of nonlinear elastic response that can shear thin. This is
an important advantage over the classical Maxwell and Oldroyd-B models that cannot
shear thin or shear thicken. Also, modifications to how the body stores energy and
dissipates energy lead to models in which the relaxation and retardation times depend
on the deformation.
Despite the equivalence of the models (2.8.10)–(2.8.12) and (2.8.22)–(2.8.23), there
is an important philosophical difference that is worth noting. In the model (2.8.22)–
(2.8.23), the extra stress Ŝ is introduced as a quantity that meets (2.8.23). However, no
physical meaning whatsoever is given to this quantity other than that it is the constitu-
tively determined part of the stress. It has been shown recently that a splitting of the stress
into a constrained part and a constitutively determined part has an inherent nonunique-
ness (see [234]). In the thermodynamic approach that leads to (2.8.10)–(2.8.12), a very
precise meaning is associated with the extra stress. It is the elastic response of the fluid
from the natural configuration κ p(t) at time t, to its current configuration κt at time t.
The configurational tensor approach also assigns a specific meaning to the extra stress
and this is to be expected as we have observed earlier that there is a correspondence
between the conformation tensor C and the tensor B κ p(t) .
We now turn our discussion toward the subtle issue of prescribing boundary condi-
tions for rate type fluids. We first recall that for fluids of grade n > 1, the balance of
linear momentum is of order n + 1. This is a direct consequence of An containing a
term that has the nth spatial gradient of the velocity v. In the case of a fluid of grade two,
by virtue of the stress depending on A2 , we have a partial differential equation of order
∇
three. Now, let us consider the Oldroyd-B fluid defined through (2.8.6). The term D that
appears in the equation also involves second spatial derivatives of the velocity, i.e., as
∇
far as spatial gradients are concerned, A2 and D are of the same order. In fact, both A2
∇
and D are different but properly invariant temporal derivatives of A1 or D. As we have
difficulties with prescribing boundary conditions in general for a fluid of grade two,
we anticipate similar difficulties for Oldroyd-B fluids. We also note that the definition
∇
of a Burgers fluid (2.8.2) also involves the term D which has spatial derivatives of the
same order as A2 ; thus we expect that Burgers’ model will require additional boundary
conditions than the Navier–Stokes model.
Let us consider a specific flow that of the flow of an Oldroyd-B fluid past a porous
plate. We seek a solution in a Cartesian coordinate system for the velocity and extra
stress of the form
v = u(y)i + v(y) j , S = S(y). (2.8.24)
The constraint of incompressibility implies that
v(y) = v0 a constant.
76 2 Mechanics
Substituting (2.8.24) into the balance of linear momentum (2.4.25), we obtain
∂p d Sx y du ∂p d S yy ∂p
− + = v0 − + = 0, − = 0. (2.8.25)
∂x dy dy ∂y dy ∂z
The constitutive relation (2.8.6) implies that
d Sx y du du d2 u
Sx y + λ1 v0 = λ1 S yy + η1 + λ2 η1 v0 2 ,
dy dy dy dy
(2.8.26)
d S yy
S yy + λ1 v0 = 0,
dy
plus an additional differential equation that determines explicitly Sx x once u is known.

Let us consider the case of suction at the plate, i.e., v0 < 0. Since λ1 > 0, the second
equation in (2.8.26) implies that if we require S yy to be bounded, it has to be zero. It
then follows from the first equation in (2.8.26) and (2.8.25) that
λ2 v0 η1 u + (λ1 v02 − η1 )u − v0 u = 0. (2.8.27)
In arriving at (2.8.27), we have used the fact that, since we are looking for bounded
solutions, then S yy = 0 and (2.8.25) leads to ∂ p/∂ x = constant. But as the flow is
uniform at infinity, i.e., as u → U as y → ∞, then ∂ p/∂ x has to be zero, i.e., the
pressure field is a constant. Of course, it is possible that the equations do not admit
such a solution, but we shall show below that there do exist solutions with such an
asymptotic structure.
We notice that (2.8.27) is a third-order ordinary differential equation. Its solution is
u = C 1 + C 2 em 1 y + C 3 em 2 y .
The adherence boundary condition at the porous plate requires
u(0) = 0. (2.8.28)
Also, as the free stream velocity is U , we have
u → U as y → ∞. (2.8.29)
These two conditions are sufficient to find a bounded solution
u(y) = U [1 − emy ],
where

1
m := η1 − λ1 v02 + (η1 − λ1 v02 )2 + 4λ2 v02 η1 .
2λ2 v0 η1
We are able to obtain this solution with the two conditions above, although this is
a third-order differential equation, because we require boundedness of the solution.
However, suppose we seek a solution in a bounded domain, which is the case in the
flow between two porous parallel plates, one of which is at rest and the other moving
with constant speed U , the fluid being either sucked or blown through the porous plates.
Then, following the above procedure, we find that the boundary condition (2.8.28) and
u(h) = U,
are not sufficient to determine the solution. Of course, what we have shown is that we
are unable to find solutions of the form (2.8.24) with the boundary conditions (2.8.28)
and (2.8.29). It is possible that there could be other solutions to the general equations
that satisfy these boundary conditions. However, to our knowledge, we are unaware
of any rigorous existence results for the flow of an Oldroyd-B fluid when the normal
component of the velocity at the boundary is not zero.
Notice that in the case of a Maxwell model, λ2 = 0 and (2.8.27) reduces to
(λ1 v02 − η1 )u − v0 u = 0,
which is a second-order equation. Then, the conditions (2.8.28) and (2.8.29) are suffi-
cient to determine the solution u
|v0 |
− 2 y
u(y) = U 1 − e (η1 −λ1 v0 ) ,
provided λ1 v02 < η1 . If λ1 v02 ≥ η1 , we cannot satisfy the boundary condition and
solutions of the form that is sought cannot exist.
Unfortunately, issues concerning boundary conditions for rate type fluids are not as
simple as that. This will become clear in what follows. In the case of an Oldroyd-B
fluid, let us introduce a new tensor Ŝ through
S = Ŝ + γ D, η > γ > 0, (2.8.30)
where S satisfies (2.8.6). Then (2.8.1) takes the form
T = − pI + Ŝ + γ D, (2.8.31)
and, since λ1 > λ2 , by choosing

λ2
γ =η ,
λ1
Equation (2.8.6) can be written as
∇
Ŝ + λ1 Ŝ= (η − γ ) D. (2.8.32)
78 2 Mechanics
∇
We notice that the term D does not appear explicitly in the equations, and as far as S
or Ŝ are concerned, the spatial derivative of them are of the same order.
∇
The transformation (2.8.30) that allows us to get rid of the term involving D will
∇
not work in general for rate type fluids wherein D appears. For instance, consider the
rate type model whose extra stress S satisfies the equation
∇ ∇
S + λ1 S= f ( D, D), (2.8.33)
∇ ∇
where f is nonlinear in D. In this case, (2.8.30) will not allow us to eliminate D from
the equation for Ŝ.
The models (2.8.20)–(2.8.21) and (2.8.22)–(2.8.23) are equivalent, as both represent
the Oldroyd-B model; but they demand quite different initial and boundary conditions.
∇
Indeed, both D and D appear in the representation (2.8.22)–(2.8.23), while only D ap-
pears in the representation (2.8.20)–(2.8.21). Thus, the representation (2.8.20)–(2.8.21)
for the Oldroyd-B model seems much more suitable for solving problems, as it demands
less in terms of boundary conditions. Even in the representation (2.8.31)–(2.8.32), which
seems to demand less in terms of boundary conditions for the velocity, we seem to need
initial conditions on the extra stress. But a priori, we do not know what this extra stress
is other than that it satisfies (2.8.32). However, in the representation (2.8.20)–(2.8.21)
that arises from the thermodynamic framework, the extra stress given by (2.8.18) is
well-defined, and we have an evolution equation for B κ p(t) . Now, the initial condition
for B κ p(t) is known if we know the initial conditions for the displacement field from
the appropriate natural configuration. More importantly, we shall see that even for rate
∇
type models in which D can appear nonlinearly (see (2.8.33)) and in which we can-
∇
not eliminate D in the constitutive expression for the extra stress, the thermodynamic
framework will yield equivalent models wherein we will only have an evolution equa-
tion for B κ p(t) . Thus, within the thermodynamic framework we need to consider only
the initial condition for the displacement from the appropriate natural configuration.
We end with a discussion of the efficacy of the thermodynamic procedure by con-
sidering the status of Burgers’ model within the above framework. It will also serve to
illustrate the use of more than one natural configuration corresponding to the current
configuration of the body. When one considers a body such as asphalt that is essen-
tially composed of a mixture of constituents, then it is necessary to associate natural
configurations corresponding to each of the constituents (see Murali Krishnan and Ra-
jagopal [154]). Such models are also necessary to describe viscoelastic fluids whose
mechanical analogue involves several springs and dashpots, natural configurations be-
ing associated with the undeformed lengths of the various springs.
The mechanical analogue for the Burgers model is made up of two springs and two
dashpots; for the linear Burgers model in one dimension, the stress σ and the rate of
strain ε̇ follow the relation:
σ + p1 σ̇ + p2 σ̈ = q1 ε̇ + q2 ε̈,
Fig. 2.9 Bodies with multiple relaxation times
where p1 , p2 , q1 , q2 are material functions that are related to the properties of the two
springs and two dashpots.
We now proceed to derive the constitutive relation for Burgers’ model within the
thermodynamic framework for bodies that have multiple natural configurations. Recall
that the Cauchy stress for the Burgers model is given by (2.8.1)–(2.8.2)
T = −I + S,
∇ ∇∇ ∇
S + λ1 S +λ2 S = 2η1 D + 2η2 D .
Since we have two springs, each of them has an associated natural configuration. Let
κ p1 (t) and κ p2 (t) denote the two natural configurations associated with the configuration
κt at time t, and let F κ p1 (t) and F κ p2 (t) denote the deformation gradients from the
configurations κ p1 (t) and κ p2 (t) to the configuration κt , respectively (see Figure 2.9). We
can define as before the appropriate Cauchy-Green tensor B κ pi (t) , i = 1, 2 (see [154])
and the tensor
G i := F −1
κ p (t) F κ R , i = 1, 2.
i
The velocity gradient L κ pi (t) and its symmetric part Dκ pi (t) are defined through
L κ pi (t) = Ġ i G i−1 , i = 1, 2,
1
Dκ pi (t) = [L κ pi (t) + L κTp (t) ], i = 1, 2.
2 i
80 2 Mechanics
Let the stored energy associated with the i-th constituent (spring) depend only on
its deformation, i.e., the deformation gradient F κ pi (t) . As we are concerned with an
incompressible isotropic material, it immediately follows that:
ψi = ψ̂i (I B κ p (t) , II B κ p (t) ), i = 1, 2.

i i
Let the rate of dissipation associated with the ith constituent (dashpot) ξi be given
through
ξi = ξ̂i (B κ pi (t) , Dκ pi (t) ).
A full thermodynamic treatment that includes the effects of temperature, heat flux,
entropy, etc., that arrives at the appropriate form for the stress by requiring that the rate
of dissipation be maximized can be found in [154]. For instance, in the case where the
stored energy and the rate of dissipation have the special forms
μi
ψi = (I B κ p (t) − 3), i = 1, 2,
2 i
where μi , i = 1, 2, are constant and
ξi = ηi Dκ pi (t) · B κ pi (t) Dκ pi (t) , i = 1, 2,
where ηi , i = 1, 2, are constant, a lengthy analysis (see [154]) leads to
T = − pI + μ1 B κ p1 (t) + μ2 B κ p2 (t) , (2.8.34)
with
1 ∇ μi 3
− B κ pi (t) = B κ pi (t) − , i = 1, 2. (2.8.35)
2 ηi tr B −1
κ p (t)
i
In general μi and ηi are not constants. If the material under consideration strain
softens or strain stiffens, then it is necessary to allow the μi to be functions of the
Cauchy–Green tensors B κ pi (t) . Similarly, if the body shear thins or shear thickens, it
would be necessary that all viscosities ηi be functions of Dκ pi (t) . It is also possible
that the ηi depend on B κ pi (t) as one cannot always separate the elastic and dissipative
responses in the sense that the dissipation in the current configuration could depend on
the elastic response from the natural configuration to the current configuration. In fact,
materials such as asphalt concrete present very complicated response characteristics.
They have completely different responses to compressive and tensile loading. Moreover,
their elastic as well as dissipative response changes considerably with the deformation,
and they are very sensitive to temperature, an issue that we ignore in this book.
For a viscoelastic material such as asphalt, a simple deformation such as triaxial
response cannot be described by a Burgers model in which the material moduli are
constant. Murali Krishnan and Rajagopal [153] use shear moduli of the form
ni −1
bi
μi = μ̄i 1 + (tr B κ pi (t) − 3) , (2.8.36)
ni
and viscosities of the form

ni −1
ηi = η̄i 1 + N (tr B κ pi (t) − 3)m ) , (2.8.37)
where μ̄i , N and η̄i are constants. When n i = 1 in (2.8.36) and (2.8.37), we recover
Burgers’ model (2.8.1)–(2.8.2).
It can be shown that (2.8.34) and (2.8.35) are equivalent to
T = − pI + S,

2
ηi ∇
S= Si , Si + Si = λi I, i = 1, 2, (2.8.38)
2μi
i=1
where
3
Si = μi B κ p1 (t) , λi = , i = 1, 2. (2.8.39)
tr Si−1
Since Si is non-singular, it can be shown that the system (2.8.38)–(2.8.39) is equivalent

to
∇ ∇∇ ∇ ∇∇
S + α S +β S = λ̄ I +λ̄β I , (2.8.40)
where
η1 η2 η1 η2
α= + , β= .
2μ1 2μ2 4μ1 μ2
By virtue of (2.3.28), (2.8.40) reduces to

∇ ∇∇ ∇
S + α S +β S = γ1 D + γ2 D,
which is indeed equivalent to (2.8.2).

The framework within which the rate type models were derived here has very clear
thermodynamic underpinnings. Not only does such a framework seem to yield rea-
sonable rate type fluid models, it leads to sensible models in a whole variety of fields
including plasticity, twinning, solid-to-solid phase transitions, super plasticity, crystal-
lization of polymer melts, multi-network polymer mechanics, and mechanics of gran-
ular materials.
Finally, we describe a rate type model, the Johnson–Segalman fluid, that for a range
of values for the material parameters exhibits a marked difference from most other rate
type models and the classical Navier–Stokes fluid. It is well-known that for the Navier–
Stokes fluid, the shear stress increases monotonically with the shear rate. The Maxwell,
Oldroyd-B and Burgers fluids also exhibit such a monotone behavior. Unlike such
fluids, the Johnson–Segalman fluid exhibits a non-monotone relationship between the
82 2 Mechanics
Fig. 2.10 Spurt

phenomenon
shear stress and shear rate, for a certain range of parameters. Such a response has been
advanced as the cause for many interesting phenomena, such as “spurt,” that manifest
themselves during the flow of polymeric melts. We shall discuss this phenomenon later.
In a Johnson–Segalman fluid, the Cauchy stress has the form (cf. [146]),
T = − pI + Ŝ,
Ŝ = S + 2μ D,

DS
S+λ + S(W − a D) + (W − a D)T S = 2η1 D,
Dt
where D and W are the symmetric and skew part of the velocity gradient, μ and η are
viscosities, λ is the relaxation time and a is called the slip parameter, all the material
moduli being constant. When a = 1, the above model reduces to an Oldroyd-B model,
while if in addition, μ = 0, it reduces to a Maxwell model. When μ = 0 and λ = 0,
the model reduces to the classical Navier–Stokes model. For a certain range of values
of the parameter a, the relationship between the shear stress and shear rate in a plane
Couette or cylindrical Poiseuille flow, is non-monotonic. This lack of monotonicity
is cited as the reason for the phenomenon of “spurt”20 that has been observed during
the flow of some polymeric melts (cf. Kolkka et al. [151], Malkus et al. [172, 173]).
As the relationship is non-monotone in a very special manner, as depicted in Figure
2.10, a jump in the shear rate occurs at a constant shear stress at the value when it
loses monotonicity. This jump occurs at a critical pressure gradient, and the flow rate
versus the pressure gradient curve has a discontinuity in its derivative at the point, the
derivative having a pronouncedly larger value as the pressure gradient increases. At
those points, the velocity gradient is discontinuous.
20 The phenomenon of a large increase in the volume flow rate due to a small increase in the driving
pressure is referred to as “spurt.”

However, it seems more probable that spurt is a consequence of the stick-slip that
takes place at the boundary. Indeed, if the cause were the lack of monotonicity, then
a variety of other strange phenomena ought to manifest themselves in different flow
situations. Nonetheless, the model provides yet another dimension to the description of
behavior that does not conform to what is expected in a Navier–Stokes fluid.
Many exact solutions have been established for simple flows of rate type models
such as the Maxwell, Oldroyd-B, Burgers, and the Johnson–Segalman fluid undergoing
steady and unsteady unidirectional flows or planar flows. Also, some linearized stability
analysis of special flows have been carried out. These studies are too numerous to
document here, and moreover our interests lie in documenting mathematical results
concerning the existence and uniqueness of the flows of such fluids under general
conditions.
2.8.1 A Gibbs-Potential-Based Formulation for Obtaining

Rate Type Response Functions for Viscoelastic Fluids
Thus far, we have discussed the development of models within the context of a body
having a natural configuration that evolves. The procedure rests in choosing appropriate
forms for the Helmholtz potential and the rate of dissipation. Recently, Rajagopal and
Srinivasa [237] developed a different formulation, namely one based on a choice being
made for the Gibbs potential and the rate of dissipation, as one cannot obtain some of
the useful rate type fluid models by using the Helmholtz potential formulation. These
two ways of generating models go hand in hand and together yield practically all the
phenomenological models that are in use, and in addition many new models that can
describe experimental data better than those that are currently available. The approach
provides a rational means for describing anisotropic fluids without appealing to the use
of directors.
We shall not get into a discussion of the thermodynamic procedure other than men-
tioning that appropriate choices for the Gibbs potential and the rate of dissipation, and
the maximization of the rate of dissipation subject to the reduced energy-dissipation
equation as a constraint, leads to models proposed by Phan-Thien and Tanner, Met-
zner, White and Dunn, Giesekus, and several others. We refer to Rajagopal and Srini-
vasa [237] for details of the same.
2.9 Bingham Fluids
The one-dimensional response of a “perfect” Bingham fluid is portrayed in Figure 2.11

below.
After a threshold τth is reached for the shear stress, the body flows at constant shear.
Such a response is the counterpart to a rigid-perfectly plastic response of solids. Clearly,
with regard to the response depicted in Figure 2.11 below, the stress cannot be expressed
84 2 Mechanics
Fig. 2.11 1-D Bingham

fluid
shear stress (τ )
yield stress (τcr )
shear rate (κ̇)
as a function of the shear rate. In many instances one needs a truly implicit relationship
between the stress and the shear rate. Recently, Rajagopal and Srinivasa [236] have de-
veloped a thermodynamic framework within which such response can be described.21
The thermodynamic approach rests on making an appropriate choice for the rate of
entropy production and then carrying out a maximization of the rate of entropy produc-
tion subject to the reduced energy-dissipation equation holding (see [236] for details).
A choice of the rate of dissipation ξ of the form
τ ·τ
ξ= ,
2μ
where τ is the deviatoric part of the stress and μ > 0, leads to the constitutive relation
τ
D = D . (2.9.1)
2μ
Thus, if D = 0, then we can conclude that τ = 2 μ, and if τ < 2 μ, we can

conclude that D = 0. Hence τ = 2 μ is the threshold value for the stress, below
which there is no flow and when the threshold is reached the body flows. Therefore
(2.9.1) describes the response depicted in Figure 2.11. For the one-dimensional response
of a linear Bingham fluid see Figure 2.12.
Such a fluid withstands shear stresses up to the threshold level of τth , and beyond
the threshold flows like a linearly viscous fluid. Once again, we notice that τ is not a
function of κ, however, κ is a function of τ (see Figure 2.13).
21 Bulíčeket al. [53, 54] have provided a detailed mathematical study of the equations governing
the flows of such fluids.
2.10 Appendix 85
Fig. 2.12 1-D linear

Bingham fluid
shear stress (τ )
yield stress (τcr )
shear rate (κ̇)
Fig. 2.13 1-D linear

Bingham fluid, κ as a
function of τ
shear rate (κ̇)
yield stress (τcr ) shear stress (τ )
2.10 Appendix
2.10.1 Integral Type Constitutive Equations
Integral type constitutive equations can be traced back to the work of Boltzman [44].
In fact, though Maxwell [184] developed his model in the rate form, its equivalent
integral form can be easily established (see the integral model given by the equation
(2.8.5)). We shall not go into a detailed discussion of integral models, as little by way
of rigorous mathematical results have been established for properly frame-indifferent
nonlinear models.
86 2 Mechanics
The most popular integral model is called the general linear integral model, in which
the Cauchy stress takes the form:
t
T = − pI + G(t − τ ) D(τ ) dτ,
−∞
where G(t − τ ) is called the relaxation modulus. The above model, which was the
model of choice several decades ago, has been studied in great detail, and continues
to be widely used to describe the response of polymers within their linear range even
today. However, it is incapable of describing normal stress differences in simple shear
flows. Several mathematical results concerning the above model have been established
and we refer the reader to Hrusa et al. [135] for a discussion of, and references to, the
same.
Most of the integral models that are currently in vogue are special cases of the
K-BKZ model (cf. Kaye [147], Bernstein et al. [36]). The Cauchy stress in a K-BKZ
model takes the form
t
∂ U −1 ∂U
T = − pI + ( )C t (τ ) − ( )C t (τ ) dτ, (2.10.1)
−∞ ∂I ∂ II
where U is the stored energy that depends on t −τ and the principal invariants of C t (τ ),
i.e.,
U = Û (t − τ, I, II), I = tr C t (τ ),
1 (2.10.2)
II = (tr C t (τ ))2 − tr C 2t (τ ) = tr C −1
t (τ ).
2
The last equality in (2.10.2) holds because we are interested in incompressible fluids,
and thus, the flows are isochoric. This model is properly frame-indifferent and the
statistical mechanical approach to the modeling of polymers leads to models of the
form (2.10.2), cf. Doi and Edwards [83]. Different choices for the stored energy U
leads to different models. For instance, Wagner [280], based on experimental evidence,
made the choice
∂U ∂U 1 d G(s) −n √II−3
= 0, =− e ,
∂I ∂ II 2 ds
where G(s) is the shear stress relaxation function and n = 0.29, a constant value that
he fixed on the basis of experiments on polyethylene melts.
The Lodge model (cf. [167]) corresponds to a choice of U of the form:
∂U ∂U
= φ1 (t − τ ), = φ2 (t − τ ).
∂I ∂ II
In this case, the model reduces to

t
T = − pI + φ1 (t − τ )C −1
t (τ ) − φ2 (t − τ )C t (τ ) dτ.
−∞
2.10 Appendix 87
In addition to models such as (2.10.1) given in terms of a single integral, multiple

integral representations for the stress have been proposed (see Green and Rivlin [125])
but little if anything by way of specific initial-boundary value problems or rigorous
mathematical results have been established within the context of such models.
2.10.2 Fractional Derivative Models of the Rate

and Integral Type
Fractional derivative models have been used to describe the response of viscoelastic
materials such as asphalt. They have also been advanced as possible models for de-
scribing the response of amorphous polymers near the glassy state. Interestingly, they
were introduced to model viscoelastic materials on the basis of models in the field of
Psychology! (cf. Grement [111] for a more detailed discussion of fractional derivative
models for viscoelastic fluids).
The notions of fractional differentiation and integration were recognized by Leib-
nitz [159] in 1675 and studied by Euler [95] in 1730. It was given a somewhat rigorous
basis by Liouville [165] in 1832 and Riemann [244] in 1876.
Fractional derivatives can be derived in a variety of ways and these definitions are
unfortunately not equivalent, see for instance the definition of fractional Sobolev spaces
in Subsection 3.1.1. What makes the study of papers using models based on fractional
derivatives difficult is that these papers do not use fractional derivatives in the same
sense and hence one has to be careful in passing judgements concerning the usefulness,
or otherwise, of these models. Fractional derivatives or integrals can be defined without
making any explicit use of either derivatives or integrals (see (2.10.3) below). How-
ever, it is common to introduce fractional integrals in terms of Liouville or Riemann–
Liouville operators by using integrals explicitly. The fractional derivative based on
the Riemann–Liouville operator has the shortcoming that the fractional derivative of a
constant function is not zero (cf. Oldham and Spanier [199]). By slightly modifying its
definition, one can obtain the fractional derivative to be zero. But the price to pay is a
lack of consistency: the limit obtained by setting the fractional order of differentiation
or integration to zero is not the same. Nevertheless, some authors adopt this definition
of fractional derivative to develop rheological models (see Van Arsdale [278]). On the
other hand, some rheologists prefer to use a definition for fractional integration that
yields a nonzero fractional derivative of a constant (cf. Palade et al. [207]).
There has been no systematic study of any of the fractional derivative or integral
models from either a rigorous mathematical or numerical point of view. A few patchy
studies that are available do not provide any insight into the usefulness of these models
with regard to describing the behavior of viscoelastic fluids. It might be worthwhile to
carry out a careful and systematic assessment of the status of such models within the
class of models used to describe the response of viscoelastic fluids. There have been
several studies concerning fractional derivative models for viscoelastic solids from
an engineering standpoint (see Rossikin and Shitikova [249]). However, no rigorous
mathematical results have been established concerning these models.
88 2 Mechanics
In 1936, Gremant [111], used fractional derivative models to describe experimental

results for viscoelastic bodies. In 1947, Blair et al. [41] used fractional derivative models
to describe the non-Newtonian response of certain fluids. These early studies by Blair et
al. have been followed by numerous attempts at modeling the viscoelastic response of
materials to this date (the papers by Bagley [16], Bagley and Torvik [17], Koeller [150],
Stastna et al. [257], Palade et al. [207], Palade and Santo [208] and the references cited
therein will give the interested reader a reasonable picture of the successes and failures
of fractional derivative models).
The derivative or integral of order q of a function is given by (see [199])
x−a q N
dq f [ N ] −1 Γ ( j − q) x −a
= lim f (x − j ( )) , (2.10.3)
[d(x − a)]q N →∞ Γ (−q) Γ ( j + 1) N
j=0
where a denotes the lower limit of integration, q is arbitrary and Γ denotes the Gamma
function defined by

N !N x
Γ (x) = lim .
N →∞ x(x + 1) . . . (x + N )
In 1985, Van Arsdale [278] introduced a generalization of the Rivlin–Ericksen ten-

sors within the context of fractional derivatives through
Aα = F −T D α (F T F)F −1 .
This tensor is properly frame-indifferent, i.e., for all Q ∈ O, Aα = Q Aα Q T . Also,

these tensors satisfy the following recursive relation:
d
Aα+n = Aα+n−1 + Aα+n−1 L + L T Aα+n−1 .
dt
Clearly, one can develop constitutive models based on these Rivlin–Ericksen tensors
of fractional order. As an example, [278] discusses the model
S = μ0 A1 + μ1 Aα A1 Aα + μ2 ( A1 A2α + A2α A1 ),
that ensures that the stress power is nonnegative provided μ0 and μ2 are positive and
|μ1 | ≤ μ2 /2. A variety of such models can be constructed. We shall not get into a
discussion of these fluid models of the differential type.
A general three-dimensional constitutive model based on fractional derivatives that
has been used is of the form
2.10 Appendix 89

K

P

P

J
1+ ak D βk 1 + b p Dγ p T = 1 + b p D γ p λ0 + λ j D α j (tr ε) I
k=1 p=1 p=1 j=1

K

M
+2 1+ αk D βk μ0 + μm D δm ε,
k=1 m=1
where ε is the linearized strain and D denotes the time derivative. Even if one switches D
to be a frame-indifferent time derivative such as the upper-convected Oldroyd derivative,
since ε is neither frame-indifferent nor Galilean invariant, the above model is neither
frame-indifferent nor Galilean invariant.
A proper frame-indifferent model was introduced in [207] where the extra stress
tensor satisfies
t
∇
α −1 −T
S(t) + λ F(t) μ1 (t − τ )F (τ ) S (τ )F (τ ) dτ F T (t)
−∞
t
= Gλβ F(t) μ2 (t − τ )F T (τ ) A1 (τ )F(τ ) dτ F T (t), (2.10.4)
−∞
where μ1 and μ2 are memory kernels given in terms of fractional derivatives and λ is
a relaxation time. It is shown in this paper that if
1
μ1 (t − τ ) = (t − τ )−α ,
Γ (1 − α)
1
μ2 (t − τ ) = (t − τ )−β ,
Γ (1 − β)
then (2.10.4) reduces to a generalization of the Maxwell model in fractional integral

form
t t
∂ S(τ )
S(t) + λα μ1 (t − τ ) dτ = Gλβ μ2 (t − τ ) A1 (τ ) dτ.
−∞ ∂τ −∞
The constants α and β have physical significance. The constant β represents the slope
of the complex modulus G in the limit of zero frequency on a log–log plot while β − α
is related to the slope of the complex modulus G in the high frequency limit.
Palade et al. [207] have shown that the model (2.10.4), which helped describe the
linear viscoelastic behavior of polymers quite well, exhibits peculiar stability charac-
teristics calling into question the specific model that they have used. But this by no
means closes the door with regard to the use of models based on fractional derivatives
and integrals.
However, before one starts embarking on rigorous mathematical investigations one
has to grapple with an important physical issue, that of boundary and initial conditions
for fractional derivative models. For instance, when dealing with differential or rate
type fractional derivative models, say involving the fractional derivative α > n, where
90 2 Mechanics
n = 1, one has to first determine what boundary conditions need to be prescribed in

order to render the problem well posed. We already know for integral values of n > 1,
we have difficulties concerning fluids of the differential type. Even for the integral
type fractional derivative models, if the memory kernel is given in terms of fractional
derivatives, it is not clear that the usual initial conditions will suffice.
2.10.3 Free-Surface Flows
A very important class of fluids that is not considered in this book is free-surface flows.
Such flows, even within the context of the Navier–Stokes fluid are obviously difficult for
a variety of reasons. First, in most such flows the domain in which the flow takes place
changes with time. Second, one invariably finds certain incompatibilities with regard
to the specification of boundary conditions, for instance, traction may be specified on
one part of the boundary and the velocity on another part, leading to an incompatibility
where these two boundaries intersect. Third, the boundary of the domain in which
the flow takes place is usually not smooth. However, free boundary flows have very
important technological significance.
For instance, the problem of “fiber spinning,” which is how polyester yarn is pro-
duced, or “film blowing,” which is how most plastic bags (grocery bags, garbage bags,
etc.) are produced lead to free-surface problems. The classical experiments that best elu-
cidate the effect of normal stress differences are, namely “rod climbing” (see Figure 2.5
and “flow down a tilted trough” (see Figure 2.14), which are also used to experimentally
determine the material moduli of non-Newtonian fluids.
Another important “free-surface flow” that needs careful study is at the heart of the
manufacture of all manners of thin films, tapes, etc. This is the flow that is involved
in all manufacturing processes made via extrusion. The flow domain is exceedingly
complex, consisting of boundaries that are made up of both free and fixed surfaces (see
Figure 2.15). The engineer is in fact faced with the inverse problem of designing the
“die” so that the flow that emerges out of the “die” is flat (see Figure 2.16) and without
wrinkles (referred to in the engineering literature as the “shark skin effect”).
Fig. 2.14 Flow down a

tilted trough gravity
θ
2.10 Appendix 91
Fig. 2.15 Extrusion with die

swelling
Fig. 2.16 Extrusion without

swelling
Flat Surface
The “stress-free” (“traction-free”) boundary condition that needs to be specified in

the case of non-Newtonian fluids are far more complicated than their Navier–Stokes
counterpart. Even in the case of a second-grade fluid, it takes a complicated form.
The “traction-free” boundary condition for a third-grade fluid would include additional
nonlinear terms that would make it most daunting. This notwithstanding, it is neces-
sary to tackle such problems in view of their technological importance; “free-surface”
problems for non-Newtonian fluids constitute a very important class of problems that
have not received the rigorous mathematical scrutiny that they deserve. Most analysis
that has been carried out by the engineering community, in virtue of their technological
significance, while physically insightful, lack mathematical rigor and often times are
clearly incorrect.
Chapter 3
Mathematical Preliminaries
This short chapter collects most mathematical notions, definitions, and results that
will be used in the following chapters. Nearly all results are recalled without proof,
or are briefly established. Additional results with proofs will be found in Chapter 7.
3.1 Notation and Spaces
The following notation will be used in the sequel. We state them in dimension d = 3,
but it is easy to restrict them to d = 2. We denote by | · | the Euclidean norm of a
vector or the Frobenius norm of a tensor: |u|2 = u · u or |F|2 = tr(FFT ). The scalar
product of two tensors F and G is denoted by
F : G := tr(FGT ).
Unless otherwise specified, the domain of interest Ω will be bounded, connected,

with a boundary ∂Ω that is at least Lipschitz-continuous (cf. Grisvard [127]); to
simplify, we call it a Lipschitz domain.
3.1.1 Sobolev and Hölder Spaces
The main properties, for our purpose, of Sobolev and Hölder spaces are recalled in
this subsection. We denote by D(Ω) the space of all functions that are infinitely
differentiable and with compact support in Ω and by D (Ω) its dual space, i.e., the
space of distributions in Ω. Let k = (k1 , k2 , k3 ) be a triple of nonnegative integers
and set |k| = k1 + k2 + k3 . The partial derivative ∂ k of order |k| is then defined by

DOI 10.1007/978-3-319-39330-8_3
94 3 Mathematical Preliminaries
∂ |k| v
∂kv = .
∂x1k1 ∂x2k2 ∂x3k3
We denote by S the space of all C ∞ (IRd ) functions ϕ which satisfy for all triples
of nonnegative integers k and
lim |x| ∂ k ϕ(x) = 0.

|x|→∞
We call tempered distribution a distribution of D (IRd ) that can be extended to

a continuous linear functional on S. The space of all tempered distributions is
denoted S .
Recall the standard Sobolev spaces, for a nonnegative integer m and a number
r ≥ 1 (cf. Adams [3] or Nečas [195]),
W m,r (Ω) = {v ∈ L r (Ω) ; ∂ k v ∈ L r (Ω) 1 ≤ |k| ≤ m},
equipped with the seminorm

1r
|v|W m,r (Ω) = |∂ k v(x)|r dx ,
|k|=m Ω
and the norm (for which it is a Banach space)

1r
vW m,r (Ω) = |v|rW n,r (Ω) ,
0≤n≤m
with the usual modification when r = ∞. When r = 2, this space is the Hilbert
space H m (Ω) for the scalar product

((u, v))H m (Ω) = (∂ k u, ∂ k v),
0≤|k|≤m
where the scalar product of L 2 (Ω) is denoted ( ·, · ).

We refer to [127], Lions and Magenes [164] or [3] for defining fractional Sobolev
spaces W m+s,r (Ω) when m ≥ 0 is an integer and 0 < s < 1 is a real number,

|∂ k v(x) − ∂ k v(y)|r
W m+s,r (Ω) = {v ∈ W m,r (Ω) ; dxdy < ∞ ∀|k| = m},
Ω Ω |x − y|d+sr
equipped with the norm

|∂ k v(x) − ∂ k v(y)|r 1r
vW m+s,r (Ω) = vrW m,r (Ω) + dxdy ,
Ω Ω |x − y|d+sr
|k|=m
3.1 Notation and Spaces 95
for which it is a Banach space. These definitions are extended straightforwardly to

vector-valued functions, with the following exception: In the case of a vector u or
tensor F, we set
1r
uLr (Ω) = |u(x)|r dx .
Ω
Again, when r = 2 and 0 < s < 1, H m+s (Ω) stands for W m+s,2 (Ω). It is a Hilbert
space for the scalar product
((u,v))H m+s (Ω) = ((u, v))H m (Ω)

1
k

+ ∂ u(x) − ∂ k u(y) ∂ k v(x) − ∂ k v(y) dxdy. (3.1.1)
Ω Ω |x − y|
d+2s
|k|=m
We shall also use the Hölder spaces of continuous functions C m,σ for a nonnegative
integer m and a real number σ ∈]0, 1]: C m,σ (Ω) is the set of functions v in C m (Ω)
that satisfy for all k such that |k| = m,
∀x ∈ Ω, ∀y ∈ Ω, |∂ k v(x) − ∂ k v(y)| ≤ C|x − y|σ ,
with a constant C independent of x and y, equipped with the norm
|∂ k v(x) − ∂ k v(y)|
vC m,σ (Ω) = sup |∂ k v(x)| + sup | ,
|x − y|σ
|k|≤m x∈Ω |k|=m x,y∈Ω
where it is understood that x

= y.
For imposing vanishing boundary values on ∂Ω, we define
H01 (Ω) = {v ∈ H 1 (Ω) ; v|∂Ω = 0},
and more generally, for a number r ≥ 1, we define
W01,r (Ω) = {v ∈ W 1,r (Ω) ; v|∂Ω = 0}.
There are several definitions for imposing vanishing higher order traces on a Lipschitz
domain. For example, for m ≥ 2 and r ≥ 1, W0m,r (Ω) can be defined as the closure
of D(Ω) in W m,r (Ω). When the boundary ∂Ω is smooth, this amounts to imposing
that the traces of the functions and of their derivatives up to order m − 1 vanish. For

1 < r < ∞, the dual space of W01,r (Ω) is denoted by W −1,r (Ω), 1/r + 1/r = 1.
This space has the following characterization: A distribution belongs to W −1,r (Ω)
if and only if there exist (nonunique) functions fi ∈ L r (Ω), 0 ≤ i ≤ d, such that
d
∂fi
= f0 + . (3.1.2)
i=1
∂xi
This characterization motivates the following notation (see Maz’ya and

Rossmann [188], page 517): The space C −1,σ (Ω) is the space of distributions
of the form (3.1.2)
d
∂fi
= f0 + ,
i=1
∂xi
for functions fi ∈ C 0,σ (Ω), 0 ≤ i ≤ d.

We shall frequently use Sobolev’s imbeddings (cf. [3]): If p, q, s and t are real
numbers with 1 < p ≤ q < ∞, 0 ≤ t ≤ s, and t − dq ≤ s − dp , then W s,p (Ω) is
imbedded into W t,q (Ω) and there exists a constant C, depending only on Ω, p, q, s
and t, such that
∀v ∈ W s,p (Ω), vW t,q (Ω) ≤ CvW s,p (Ω) . (3.1.3)
This imbedding is compact if t < s and t − dq < s − dp . If s − dp is not an integer then

W s,p (Ω) is imbedded into C k,σ (Ω) where the integer k satisfies 0 ≤ k < s− dp < k+1
and σ = s − dp − k, and there exists a constant C, depending only on Ω, p, and s,
such that
∀v ∈ W s,p (Ω), vC k,σ (Ω) ≤ CvW s,p (Ω) . (3.1.4)
These imbeddings will mostly be used in the following particular situations: If d = 2

and 2 ≤ p < ∞ or d = 3 and 2 ≤ p ≤ 6, there exists a constant S0,p (that depends
only on p and the domain) such that
∀v ∈ H01 (Ω), vLp (Ω) ≤ S0,p |v|H 1 (Ω) . (3.1.5)
When p = 2, this is Poincaré’s inequality and S0,2 is called the Poincaré constant.
The value p = ∞ is reached for r > d: There exists a constant S0,∞ (that depends
only on r and the domain) such that
∀v ∈ W01,r (Ω), vL∞ (Ω) ≤ S0,∞ |v|W 1,r (Ω) . (3.1.6)
For arbitrary functions in H 1 (Ω) or W 1,r (Ω), we shall use the same relations for the
same range of values of p and r, with the full norm on the right-hand side
∀v ∈ H 1 (Ω), vLp (Ω) ≤ Sp vH 1 (Ω) , (3.1.7)

and
∀v ∈ W 1,r (Ω), vL∞ (Ω) ≤ S∞ vW 1,r (Ω) . (3.1.8)
This last inequality generalizes to W s,p for all real numbers s such that s − d
p
> 0.
In virtue of Poincaré’s inequality, the seminorm | · |H 1 (Ω) is a norm on H01 (Ω),

equivalent to the full norm. As it is directly related to the gradient operator, we choose
this seminorm as norm on H01 (Ω) and set
vH01 (Ω) = |v|H 1 (Ω) , (3.1.9)
which will be used as seminorm on H 1 (Ω) and norm on H01 (Ω). In particular, we
use it to define the dual norm on its dual space H −1 (Ω)
< , v >
∀ ∈ H −1 (Ω), H −1 (Ω) = sup , (3.1.10)
v∈H01 (Ω) vH01 (Ω)
where < ·, · > denotes the duality pairing between H −1 (Ω) and H01 (Ω).
For integers m ≥ 1 and real numbers r, 1 < r < ∞, the following Hardy
inequality (see [127]) valid on a Lipschitz domain, is closely related to Poincaré’s
inequality
f

∀f ∈ W0m,r (Ω), m p ≤ Cm,r f W m,r (Ω) , (3.1.11)
d L (Ω)
where d denotes the distance function to the boundary, and Cm,r are constants that
depend only on Ω, r and m.
For imposing tangential boundary conditions, we define
Hτ1 (Ω) = {v ∈ H 1 (Ω)3 ; v · n = 0 on ∂Ω}, (3.1.12)
where n is the unit outward normal vector to ∂Ω. An easy application of Peetre–
Tartar’s Theorem (cf. Peetre [210] and Tartar [264] or Girault and Raviart [116,
Chapter I, Theorem 2.1]) proves the analogue of Sobolev’s imbeddings in Hτ1 (Ω) for
any real number p in the range of validity of (3.1.7),
∀v ∈ Hτ1 (Ω), vLp (Ω) ≤ S̃p |v|H 1 (Ω) , (3.1.13)
and in W 1,r (Ω)d ∩ Hτ1 (Ω) for r > d,
∀v ∈ W 1,r (Ω)d ∩ Hτ1 (Ω), vL∞ (Ω) ≤ S̃∞ |v|W 1,r (Ω) . (3.1.14)
In particular, for p = 2, the mapping v → |v|H 1 (Ω) is a norm on Hτ1 (Ω), equivalent
to the H 1 norm and S̃2 is the analogue of Poincaré’s constant. Recalling that div v is
defined by (2.3.9), i.e.,
3
∂vi
div v = ,
i=1
∂xi
we shall also use the classical spaces for Navier–Stokes equations

V = {ϕ ∈ D(Ω)3 ; div ϕ = 0}, (3.1.15)
V = {v ∈ H01 (Ω)3 ; div v = 0 in Ω}, (3.1.16)

V ⊥ = {v ∈ H01 (Ω)3 ; ∀w ∈ V , ∇ v(x) · ∇ w(x) dx = 0}, (3.1.17)
Ω
where ∇ v denotes the gradient tensor defined by (2.3.8): (∇ v)ij = ∂vi /∂xj ,
W = {v ∈ Hτ1 (Ω) ; div v = 0 in Ω},

Lm (Ω) = q ∈ L (Ω) ;
r r
q(x) dx = 0 ,
Ω
(3.1.18)
H(div, Ω) = {v ∈ L 2 (Ω)3 ; div v ∈ L 2 (Ω)},
Hτ (div, Ω) = {v ∈ H(div, Ω) ; v · n = 0 on ∂Ω},
H(curl, Ω) = {v ∈ L 2 (Ω)3 ; curl v ∈ L 2 (Ω)3 },
where curl v is defined by (2.3.29)

∂v ∂v2 ∂v1 ∂v3 ∂v2 ∂v1 T
3
curl v = − , − , − .
∂x2 ∂x3 ∂x3 ∂x1 ∂x1 ∂x2
We recall that V is dense in V , see for instance [116, Chapter I, Corollary 2.5].
The dual space of V is denoted by V ; this space has to be used with caution because
it cannot be identified with a subspace of D (Ω)3 (see Simon [254]), but we can use
the dual norm
(v)
∀ ∈ V , V = sup . (3.1.19)
v∈V vH01 (Ω)
These definitions carry over to d = 2 with one exception when d = 2, the curl
operator is considered a scalar because it has only one component
∂v2 ∂v1
∀v = (v1 , v2 ), curl v = − , (3.1.20)
∂x1 ∂x2
and we define
H(curl, Ω) = {v ∈ L 2 (Ω)2 ; curl v ∈ L 2 (Ω)}. (3.1.21)
We also recall the following identity, valid in a Lipschitz domain of IRd , d = 2, 3:
∀v ∈ H01 (Ω)d , v2H 1 (Ω) = div v2L2 (Ω) + curl v2L2 (Ω) , (3.1.22)
0
where curl denotes the vector or scalar curl according to the dimension.
As usual, for handling time-dependent problems, it is convenient to consider

functions defined on a time interval ]a, b[ with values in a functional space, say X
(cf. [164]). More precisely, let · X denote the norm of X; then for any number r,
1 ≤ r ≤ ∞, we define
b
L (a, b; X) = f measurable in ]a, b[ ;
r
f (t)rX dt < ∞ ,
a
equipped with the norm
b 1r
f Lr (a,b;X) = f (t)rX dt ,
a
with the usual modification if r = +∞. It is a Banach space if X is a Banach space.

Here X is usually a Sobolev space. In particular, L 2 (a, b; H m (Ω)) is a Hilbert space
and L 2 (a, b; L 2 (Ω)) coincides with L 2 (Ω×]a, b[). In addition, we shall also use
spaces with derivatives in time, such as
∂f
H 1 (a, b; X) = f ∈ L 2 (]a, b[; X) ; ∈ L 2 (]a, b[; X) ,
∂t
equipped with the graph norm
∂f 2 21
f H 1 (a,b;X) = f 2L2 (a,b;X) + L2 (a,b;X) ,
∂t
for which it is a Hilbert space. Finally, to simplify the notation, we shall sometimes
denote the partial derivative in time with a prime.
3.1.2 Some Interpolation Inequalities
The following interpolation inequalities will be frequently used further on. The first
result, known under the name of “Ladyzhenskaya’s Inequality” plays a fundamental
role in the uniqueness theorems for fluid flows (see for example, Section 4.2.7 below).
The statement presented here is a sharper version established recently by Dupont and
Mogultay in [89].
Lemma 3.1.1 Let Ω be a bounded domain in IR2 . Then

1 1 1
v 2 2 v 2 1
vL4 (Ω) ≤ 1 L (Ω) H0 (Ω)
for v ∈ H01 (Ω). (3.1.23)
2 4
Proof Due to density properties, it is sufficient to prove (3.1.23) for u in D(Ω), and
by extending u by zero outside of Ω, it is actually sufficient for the proof to take u
in D(IR2 ).
The starting point is the following inequality, valid for all g ∈ D(IR):
+∞
g2L∞ (Ω) ≤ |g(x)g (x)| dx. (3.1.24)
−∞
By applying (3.1.24) first to u(x1 , ·) and next to u(·, x2 ), we obtain

+∞ ∂u

|u(x1 , x2 )| ≤
2
|(u(x1 , s)| (x1 , s) ds = ϕ(x1 ),
−∞ ∂x2
and +∞ ∂u

|u(x1 , x2 )| ≤
2
|u(s, x2 )| (s, x2 ) ds = ψ(x2 ).
−∞ ∂x1
Then it follows successively,

+∞ +∞ +∞ +∞
|u(x1 , x2 )| dx1 dx2 ≤
4
ϕ(x1 ) dx1 ψ(x2 ) dx2
−∞ −∞ −∞ −∞
+∞ +∞ +∞ +∞ ∂u
∂u
= |u| dx1 dx2 |u| dx1 dx2
−∞ −∞ ∂x2 −∞ −∞ ∂x 1
∂u ∂u
1
≤ u2L2 (IR2 ) 2 2 ≤ u2L2 (IR2 ) ∇ u2L2 (IR2 ) .
∂x1 L (IR ) ∂x2 L2 (IR2 ) 2

Remark 3.1.2 Inequality (3.1.23) is a particular case (namely for p = 4) of the
following convexity inequality:
1 1 1 1 1
vLp (Ω) ≤ C vL2 2 (Ω) vH2 1 (Ω) with = − , (3.1.25)
0 p 2 2d
for v ∈ H01 (Ω).

Lemma 3.1.1, which is not true in three dimensions, is replaced by the following
result:
Lemma 3.1.3 Let Ω be a bounded domain in IR3 . Then
1 3
vL4 (Ω) ≤ C vL4 2 (Ω) vH4 1 (Ω) for v ∈ H01 (Ω). (3.1.26)
0
Proof The proof is based on the interpolation inequality stating that if h is a scalar
function in L p (O) ∩ L q (O) for O an open set in IRd and 1 ≤ p ≤ q < ∞, then
h ∈ L m (O) for p ≤ m ≤ q, and moreover,
hLm (O) ≤ hαLp (O) hL1−α

q (O) , (3.1.27)
with
1 α 1−α
= + , 0 ≤ α ≤ 1.
m p q
Choosing
1
m = 4, p = 2, q = 6, α = ,
4
yields
3 1
vL4 (Ω) ≤ vL4 6 (Ω) vL4 2 (Ω) . (3.1.28)
Then (3.1.26) is a simple consequence of the Sobolev imbedding H01 (Ω) ⊂ L 6 (Ω)
for d = 3.
3.2 Some Theoretical Results
We recall here some theoretical results, mainly for the Laplace and Stokes equations
in dimension d = 2 or d = 3, that we will need.
3.2.1 The Laplace Equation
Let us start with the Laplace equation with a homogeneous Dirichlet boundary con-
dition in a bounded Lipschitz domain: For f given in H −1 (Ω), find u in H01 (Ω) such
that
− Δ u = f in Ω. (3.2.1)
It can be set into the following equivalent variational formulation: Find u in H01 (Ω)
solution of
∀v ∈ H01 (Ω), (∇ u, ∇ v) = f , v.
By Lax–Milgram’s Theorem (cf. Lax and Milgram [156]), this problem has one
and only one solution that depends continuously on f . Furthermore, increasing the
regularity of f , increases up to a certain extent, the regularity of u, but this is very
sensitive to the regularity of the boundary. The first theorem concerns domains with
smooth boundaries.
Theorem 3.2.1 Let the boundary of Ω be of class C m,1 , for some integer m ≥ 1,
and let s and r be two real numbers with 0 ≤ s ≤ m and 1 < r < +∞. If f belongs
to W s−1,r (Ω), then the solution u of (3.2.1) belongs to W s+1,r (Ω) with continuous
dependence on the data.
The proof can be found in Lions and Magenes [163] when the boundary is very
smooth, but the technique of Grisvard [127] permits to reduce the regularity to C m,1 .
The next results address the Laplace equation in corner domains. Theorem 3.2.2
is proved by Grisvard in [127] and the first part of Theorem 3.2.3 is proved by Dauge
in [74], while the second part is due to Costabel and Dauge [70].
Theorem 3.2.2 Let Ω be a polygon in IR2 or a Lipschitz polyhedron in IR3 . If f

belongs to L r (Ω) for some r with 1 < r ≤ 43 when d = 2, or 65 < r ≤ 43 when d = 3,
then the solution u of (3.2.1) belongs to W 2,r (Ω) with continuous dependence on f .
Theorem 3.2.3 Let Ω be a polygon in IR2 or a Lipschitz polyhedron in IR3 and let u
be the solution of (3.2.1).
(i) If f belongs to H s−1 (Ω) for some s with 0 ≤ s < 21 , then u belongs to H s+1 (Ω)
with continuous dependence on f .
3 3
(ii) If f belongs to L 2 (Ω), then u belongs to H 2 (Ω) with continuous dependence
on f .
When f is smoother than in the above statements, the solution is also smoother
provided the inner angles of ∂Ω are suitably restricted. For instance, it is well known
that the next regularity holds in a convex domain (cf. [127]).
Theorem 3.2.4 If f belongs to L 2 (Ω) and the domain is convex, then the solution u
of (3.2.1) belongs to H 2 (Ω), with continuous dependence on f . Furthermore, as the
inner angles of a convex polygon or the inner dihedral angles of a convex polyhedron
are always strictly smaller than π , there exists a real number r0 > 2 such that for
all r ∈ [2, r0 ], if f belongs to L r (Ω), then u belongs to W 2,r (Ω) and there exists a
constant C, depending only on r and Ω such that
uW 2,r (Ω) ≤ Cf Lr (Ω) . (3.2.2)
Another interesting result concerns the regularity of the solution u when f belongs
to W −1,r (Ω), see [74].
Theorem 3.2.5 In a polygonal or Lipschitz polyhedral domain, if 43 ≤ r ≤ 4 when
d = 2 or 23 ≤ r ≤ 3 when d = 3, then for all f in W −1,r (Ω), problem (3.2.1) has a
unique solution u that belongs to W 1,r (Ω), without restriction on the angles.
If r > 4 when d = 2 or r > 3 when d = 3, the angles must be progressively
restricted as the value of r increases. In a convex polygon or polyhedron, this result
holds for all 1 < r < ∞, see [74].
When d = 2, the case r = ∞ follows from Theorem 3.2.4 and Sobolev’s imbed-
dings. Indeed, by applying Sobolev’s imbedding (3.1.8) to the components of ∇ u,
Theorem 3.2.4 implies in particular that there exists another constant C such that
3.2 Some Theoretical Results 103
uW 1,∞ (Ω) ≤ Cf Lr (Ω) . (3.2.3)
When d = 3, (3.2.3) requires that r > 3 and this restricts the angles of ∂Ω beyond
convexity. However, a better result can be obtained in a convex domain by finding a
solution in C 1,σ (Ω) for some σ > 0. The following result can be found essentially
in the works of Maz’ya and Plamenevskii [185, 186].
Theorem 3.2.6 Let Ω be a convex polygon or polyhedron. If f belongs to C −1,σ (Ω)
for some σ > 0, related to the largest inner angle of Ω, then the solution u of (3.2.1)
belongs to C 1,σ (Ω) with continuous dependence on f .
Hence u belongs to W 1,∞ (Ω). As the functions of L r (Ω) have the form (3.1.2) with
d
f0 = 0 and fi ∈ W 1,r (Ω) and as W 1,r (Ω) is continuously imbedded into C 0,1− r (Ω),
provided r > d, Theorem 3.2.6 and (3.1.4) imply in particular, the three-dimensional
extension of (3.2.3), as follows:
Corollary 3.2.7 Let σ be the constant of Theorem 3.2.6 and let r be defined by
σ = 1 − 3r . If f ∈ L r (Ω), then the solution u of (3.2.1) belongs to W 1,∞ (Ω) and
there exists a constant C such that
uW 1,∞ (Ω) ≤ uC 1,σ (Ω) ≤ Cf Lr (Ω) . (3.2.4)
The above results can be extended to the Laplace equation (3.2.1) with a nonho-
mogeneous boundary condition
u = g on ∂Ω, (3.2.5)
provided g has a continuous lifting, say ug , with the same regularity as u. The fol-
lowing theorem addresses the existence of a lifting, see [127, Theorem 1.5.1.1]:
Theorem 3.2.8 Let m ≥ 1 be an integer and assume that the domain Ω has a
boundary of class C m−1,1 for a positive integer m. Let p and s be two real numbers
such that 1 < p < ∞ and m − 1 + 1p < s < m + 1p . The trace operator γm−1
∂v ∂ m−1 v
v −→ v|∂Ω , ,...,
∂n |∂Ω ∂nm−1 |∂Ω
m−1 s−k− 1 ,p
is continuous from W s,p (Ω) onto k=0 W p (∂Ω) and has a continuous right
inverse. Moreover the following characterization holds
s,p
W0 (Ω) = v ∈ W s,p (Ω) ; γm−1 v = 0 . (3.2.6)
On a polygonal or Lipschitz polyhedral domain (a particular case of m = 1), a higher

regularity than that given by the above statement can be achieved by prescribing
compatibility conditions on g at the corners, but this is outside the scope of this
work; the reader can refer to Bernardi et al in [34].
Finally, the last result concerns harmonic functions in the whole space IR3 .
Theorem 3.2.9 The only tempered distributions that solve the Laplace equation
in IR3 ,
Δu = 0 in IR3 , u ∈ S ,
are polynomial functions.
3.2.2 The Stokes System
Now we turn to the Stokes problem. Let Ω be a connected Lipschitz-continuous

domain in IRd , d = 2 or 3, with boundary ∂Ω; consider the steady Stokes problem
for laminar flow
−μΔ u + ∇ p = f , div u = 0 in Ω,
(3.2.7)
u = g on ∂Ω,
satisfying the compatibility condition

g · ndσ (x) = 0, (3.2.8)
∂Ω
where μ is the viscosity divided by the density, both assumed to be constant and f
is an external body force. From a mathematical point of view, for any given positive
constant μ and given vector functions f in H −1 (Ω)d , g ∈ H 2 (∂Ω)d , problem (3.2.7)
1
has one and only one solution (u, p) in H (Ω) × Lm (Ω). The well-posedness of the
1 d 2
Stokes problem stems from two properties: the term −μΔ u is elliptic for μ > 0 and
the divergence operator is an isomorphism from V ⊥ onto Lm 2
(Ω), cf. for instance
see [116] or Temam [265]. The ellipticity, which is immediate from Poincaré’s
inequality (3.1.5), and the Lax–Milgram Theorem yield directly the existence and
uniqueness of the velocity when g = 0. Existence and uniqueness of the pressure
follow from the above-mentioned isomorphism. It is a particular case of the solution
of the equation
div v = f ,
given in the next theorem. We omit the proof, as it is outside the scope of this book
and we refer the reader to Bogovskii [43], Durán and Muschietti [90] or Galdi [107].
Theorem 3.2.10 Let Ω be a connected bounded domain that is a finite union of open
sets with Lipschitz-continuous boundaries. There exists an operator T that belongs
to L(Lmr
(Ω); W01,r (Ω)d ) for all r ∈]1, ∞[ satisfying
∀f ∈ Lm
r
(Ω), div T (f ) = f , (3.2.9)
and there exists a constant Kr > 0 depending only on Ω and r, such that
1
∀f ∈ Lm
r
(Ω), |T (f )|W 1,r (Ω) ≤ f Lr (Ω) . (3.2.10)
Kr
This theorem is useful when dealing with different values of r. In the Hilbert case,
i.e., when r = 2, it may be more convenient to use the solution of div v = f that
belongs to V ⊥ .
Theorem 3.2.11 Let Ω be a connected bounded domain that is a finite union of open
sets with Lipschitz-continuous boundaries. There exists an operator T̃ that belongs
to L(Lm2
(Ω); V ⊥ ) satisfying
∀f ∈ Lm
2
(Ω), div T̃ (f ) = f , (3.2.11)
and there exists a constant K̃ > 0 depending only on Ω, such that
1
∀f ∈ Lm
2
(Ω), |T̃ (f )|H 1 (Ω) ≤ f L2 (Ω) . (3.2.12)
K̃
It is easy to prove that either one of these two theorems also prove well-posedness
of (3.2.7), (3.2.8) for nonvanishing boundary data g. In particular, the following
theorem, due independently to Babuška [15] and Brezzi [50] (cf. also [116, Chapter I,
Lemma 4.1]) relates the result of Theorem 3.2.11 with the pressure.
Theorem 3.2.12 Let X and M be two Hilbert spaces with norms · X and · M
and dual spaces X and M . Let B be an operator in L(X; M ) and let B in L(M; X )
be its dual operator. Let V be the kernel of B in X, V ⊥ the orthogonal of V in X and
PV the polar of V in X, i.e.,
PV = { ∈ X ; ∀v ∈ V , , v = 0}.
The following propositions are equivalent:

1. There exists a constant β > 0 such that
B v, q
inf sup ≥ β. (3.2.13)
q∈M v∈X vX qM
2. The mapping B is an isomorphism from M onto PV and with the same constant β,
∀q ∈ M, B qX ≥ βqM . (3.2.14)
3. The mapping B is an isomorphism from V ⊥ onto M and with the same constant β,
∀v ∈ V ⊥ , BvM ≥ βvX . (3.2.15)
Theorems 3.2.12 and 3.2.11 imply the inf-sup condition,

(div v, q)
inf sup ≥ K̃. (3.2.16)
2 (Ω)
q∈Lm v∈H01 (Ω)d vH01 (Ω) qL2 (Ω)
The statement of Theorem 3.2.12 carries over to reflexive Banach spaces X and
M with V ⊥ replaced by the quotient space X/V , see [116, Chapter I, Remark 4.2].
Hence, when combined with Theorem 3.2.10, it implies the more general inf-sup
condition, valid for 1 < r < ∞ and its dual exponent r , 1r + r1 = 1,
(div v, q)
inf sup ≥ Kr . (3.2.17)
r (Ω)
q∈Lm
v∈W01,r (Ω)d
|v|W 1,r (Ω) qLr (Ω)
Korn’s first inequality is closely related to Theorems 3.2.11 and 3.2.12. A proof can
be found for instance in Duvaut and Lions [91, Chapter 3, Section 3.3]. Traditionally,
it is stated in terms of the deformation rate tensor, see Section 2.3.1, as follows:
1
1
ε(v) = ∇ v + ∇ vT = A1 .
2 2
When combined with Poincaré’s inequality, Korn’s first inequality in H01 (Ω)d ,
d = 2, 3 reads: Under the assumptions of Theorem 3.2.11, there exists a constant
C2,κ , depending only on Ω, such that
∀v ∈ H01 (Ω)d , vH01 (Ω) ≤ C2,κ ε(v)L2 (Ω) . (3.2.18)
By applying Theorem 3.2.10 and the extension of Theorem 3.2.12 to reflexive Banach
spaces, (3.2.18) can be extended to W01,r (Ω)d for 1 < r < ∞: there exists a constant
Cr,κ , depending only on r and Ω, such that
∀v ∈ W01,r (Ω)d , |v|W 1,r (Ω) ≤ Cr,κ ε(v)Lr (Ω) . (3.2.19)
A proof in a smooth or polygonal domain can be found in Wang [281]. The reader will
find in the Appendix, Section 7.1 a straightforward extension to connected Lipschitz
domains.
Let us recall some regularity results of the solution of the Stokes problem. As is
the case of the Laplace equation, regularity of the solution depends closely on that of
the boundary. The first result, due to Cattabriga [59], holds in domains with smooth
boundaries, see also the reference by Amrouche and Girault [9] and the reference by
Serre [251] on the Navier–Stokes system.
Theorem 3.2.13 Let the boundary of Ω be of class C m,1 , for some integer m ≥ 1,
and let s and r be two real numbers with 0 ≤ s ≤ m and 1 < r < +∞. If f belongs
to W s−1,r (Ω)d and g to W s+1− r ,r (∂Ω)d then problem (3.2.7) has a unique solution
1
(u, p) that belongs to W s+1,r (Ω)d × W s,r (Ω) and depends continuously on f and g.
Now we turn to corner domains. To simplify, we state results for a homoge-
neous boundary condition, but they can be extended to a nonhomogeneous boundary
condition provided g has a continuous divergence-free lifting, say ug , with the same
regularity as u. The following result is now well known. When d = 2, we refer to
Kellogg and Osborn [148] or [127], and when d = 3 we refer to Dauge [73].
Theorem 3.2.14 If f belongs to L 2 (Ω)d and the domain Ω is convex, then the
solution (u, p) of (3.2.7) for g = 0 belongs to H 2 (Ω)d × H 1 (Ω), with continuous
dependence on f . As the inner angles of a convex polygon or the inner dihedral angles
of a convex polyhedron are always strictly smaller than π , there exists a real number
r0 > 2 such that for all r ∈ [2, r0 ], if f belongs to L r (Ω)d , then (u, p) belongs to
W 2,r (Ω)d × Lmr
(Ω) and there exists a constant C, depending only on r and Ω such
that
uW 2,r (Ω) + |p|W 1,r (Ω) ≤ Cf Lr (Ω) . (3.2.20)
In two dimensions, Sobolev’s imbedding (3.1.8) implies in particular that there exists
another constant C such that
uW 1,∞ (Ω) + pL∞ (Ω) ≤ Cf Lr (Ω) . (3.2.21)
As for the Laplace equation, the same approach for deriving maximum norm esti-
mates in three dimensions can only be applied when r > 3 and in this case, convexity
is not sufficient (see [73]): the largest inner dihedral angle of ∂Ω must be strictly
less than 2 π3 , the precise value depending on r. But here also, a better result can be
obtained in a convex domain by finding a solution in C 1,σ (Ω)3 for some σ > 0, see
Maz’ya and Rossmann [187].
Theorem 3.2.15 Let Ω be a convex polyhedron. If f belongs to C −1,σ (Ω) for some
σ > 0, related to the largest inner angle of Ω, then the solution (u, p) of (3.2.7)
with g = 0 belongs to C 1,σ (Ω)3 × C 0,σ (Ω) with continuous dependence on f .
Therefore (u, p) belongs to W 1,∞ (Ω)3 ×L ∞ (Ω) when f belongs to L r (Ω)3 for some
r > 3, and Theorem 3.2.15 and (3.1.4) imply in particular the three-dimensional
extension of (3.2.21) in a convex polyhedron. Thus we have the following corollary:
Corollary 3.2.16 Let σ be the constant of Theorem 3.2.15 and define r by σ =

1 − 3r . If f ∈ L r (Ω)3 , then the solution (u, p) of (3.2.7) for g = 0 belongs to
W 1,∞ (Ω)3 × L ∞ (Ω) and there exists a constant C such that
uW 1,∞ (Ω) + pL∞ (Ω) ≤ uC 1,σ (Ω) + pC 0,σ (Ω) ≤ Cf Lr (Ω) . (3.2.22)
Of course when Ω is convex, we obtain by interpolation for 0 ≤ s ≤ 1, that (u, p)

belongs to H s+1 (Ω)d ×H s (Ω), with continuous dependence on f , whenever f belongs
to H s−1 (Ω)d . But for small s, the restrictions on the angles can be substantially
relaxed. Indeed, the following theorem holds without restriction on the angles of
∂Ω; the first part can be found in [127] and the second one in [73].
Theorem 3.2.17 Let Ω be a polygon in IR2 or a Lipschitz polyhedron in IR3 , and let
(u, p) be the solution of (3.2.7) with g = 0.
(i) If f belongs to L r (Ω)d for some r with 1 < r ≤ 43 when d = 2, or 65 < r ≤ 43

when d = 3, then (u, p) belongs to W 2,r (Ω)d × W 1,r (Ω) with continuous
dependence on f .
(ii) If f belongs to H s−1 (Ω)d for some s with 0 ≤ s < 21 , then (u, p) belongs to
H s+1 (Ω)d × H s (Ω) with continuous dependence on f .
3 3 1
(iii) If f belongs to L 2 (Ω)d , then (u, p) belongs to H 2 (Ω)d ×H 2 (Ω) with continuous
dependence on f .
The result in part (iii) is due to Costabel and Dauge [70] and can be found in Girault
and Lions [113]. The proof is written in the three-dimensional case, but its conclusion
is valid in two dimensions.
The regularity results in the case when f belongs to W −1,r (Ω)d are similar to
those for the Laplace equation, see [73].
Theorem 3.2.18 In a polygonal or Lipschitz polyhedral domain, if 43 ≤ r ≤ 4 when
d = 2 or 23 ≤ r ≤ 3 when d = 3, then for all f in W −1,r (Ω)d , problem (3.2.7) with
g = 0 has a unique solution (u, p) that belongs to W01,r (Ω)d × Lmr
(Ω).
If r > 4 when d = 2 or r > 3 when d = 3, the angles must be progressively
restricted as the value of r increases. In a convex polygon or polyhedron, this result
holds for all 1 < r < ∞. Whatever the angles, the nonhomogeneous problem with
boundary data g ∈ W 1− r ,r (∂Ω)d satisfying (3.2.8) follows easily by extending g
1
inside owing to Theorem 3.2.8 and applying Theorem 3.2.10 for reducing to zero the
divergence of the extended function.
Finally, in the case of higher regularity, there are several results for handling the
Stokes problem with nonzero divergence. We shall use the following one established
by Bogovskii in [43, 42] and Amrouche and Girault [9, Corollary 3.1] (see also [148]
for the two-dimensional case):
Theorem 3.2.19 Let Ω be a Lipschitz-continuous domainof IRd , m a nonnegative
integer, and 1 < r < ∞. For each g in W0m,r (Ω) satisfying Ω g dx = 0, there exists
v in W0m+1,r (Ω)d such that
div v = g, vW m+1,r (Ω) ≤ C gW m,r (Ω) . (3.2.23)
For any integer k ≥ 1 and real number r ∈]1, ∞[, let Vk,r (Ω) denote the kernel
of the divergence in W0k,r (Ω)d ,
Vk,r (Ω) = {v ∈ W0k,r (Ω)d ; div v = 0}, (3.2.24)
and let PVk,r (Ω) be the polar space of Vk,r (Ω),

PVk,r (Ω) = { ∈ W −k,r (Ω)d ; ∀v ∈ Vk,r (Ω), , v = 0}. (3.2.25)
Theorem 3.2.19 implies that the divergence operator is an isomorphism from
W0m+1,r (Ω)d /Vm+1,r (Ω) onto W0m,r (Ω)⊥IR.
By duality, this is equivalent to

Corollary 3.2.20 Under the assumptions of Theorem 3.2.19, the gradient operator

is an isomorphism from W −m,r (Ω)/IR onto PVm+1,r (Ω).
3.3 Further Tools of Functional Analysis
We collect here additional functional analysis tools that will be used in this work. In
the first subsection, we recall some notions used in the study of variational inequalities
and monotone operators. In particular, we recall the definition of the subdifferential
of a functional and its properties. In the second subsection, we present some essential
tools for the treatment of nonlinear and time-dependent problems.
3.3.1 Subdifferential
Let B be a Banach space.

Definition 3.3.1 A functional Φ : v ∈ B −→ Φ(v) ∈ IR is Gâteaux-differentiable
or simply G-differentiable, at the point u, if there exists one element in B , denoted
Φ (u) such that for all v in B,
Φ(u + λv) − Φ(u)

lim = Φ (u), vB ,B .
λ→0 λ
The element Φ (u) is called the gradient of Φ at u.

Let us now list some relevant results concerning the notion of G-differentiability.
Proposition 3.3.2 Let Φ be a G-differentiable functional on B (i.e., at all points of

B). Then Φ is convex if and only if
∀u, v ∈ B, Φ(v) − Φ(u) ≥ Φ (u), v − uB ,B . (3.3.1)
Proof Suppose that Φ is convex, then by definition

Φ(u + θ (v − u)) ≤ Φ(u) + θ Φ(v) − Φ(u) , θ ∈ [0, 1], (3.3.2)
from which we have

Φ(u + θ (v − u)) − Φ(u)
Φ(v) − Φ(u) ≥ .
θ
The G-differentiability of Φ allows to pass to the limit as θ → 0 and yields (3.3.1).
Conversely, assume now (3.3.1) and let
w = θ u + (1 − θ )v = θ (u − v) + v,
with θ as before. By hypothesis
Φ(v) − Φ(w) ≥ Φ (w), v − wB ,B = −θ Φ (w), u − vB ,B ,
as well as
Φ(u) − Φ(w) ≥ Φ (w), u − wB ,B = (1 − θ )Φ (w), u − vB ,B .
Multiply the first inequality by 1 − θ , the second by θ and add the results; this gives
(1 − θ )Φ(v) + θ Φ(u) − Φ(w) ≥ 0,
which (recalling the definition of w) says that Φ is convex.
The next proposition recalls the mean value theorem for G-differentiable func-
tionals.
Proposition 3.3.3 If Φ is G-differentiable on B, then there exists θ ∈]0, 1[ such that
∀u, v ∈ B, Φ(u + v) = Φ(u) + Φ (u + θ v), vB ,B .
Proof Consider the function of a real variable λ −→ ϕ(λ) = Φ(u + λv). Then
ϕ (λ) = Φ (u + λv), vB ,B .
The result follows by applying to ϕ the mean value theorem for functions of a real
variable, from which we infer that ϕ(1) − ϕ(0) = ϕ (θ ) for some θ ∈ (0, 1). The
proposition is proved since ϕ(1) = Φ(u + v) and ϕ(0) = Φ(u).
Proposition 3.3.4 Let Φ be a G-differentiable functional on B. Then the operator
u ∈ B −→ Φ (u) ∈ B ,
is monotone if and only if Φ is convex.

3.3 Further Tools of Functional Analysis 111
Proof Suppose that Φ is convex. By Proposition 3.3.2, we have
Φ (u), u − vB ,B ≥ Φ(u) − Φ(v),

−Φ (v), u − vB ,B ≥ Φ(v) − Φ(u),
so that Φ (u) − Φ (v), u − vB ,B ≥ 0, i.e., Φ is monotone.

Conversely, suppose that Φ is monotone. Proposition 3.3.3 implies successively,
Φ(v) = Φ(u) + Φ (u + θ (v − u)), v − uB ,B

= Φ(u) + Φ (u), v − uB ,B
1

+ Φ (u + θ (v − u)) − Φ (u), u + θ (v − u) − u B ,B ,
θ
and the result follows from Proposition 3.3.2. Indeed, since Φ is monotone, the last
term in brackets is positive.
Remark 3.3.5 Let Φ be convex and G-differentiable on B. Then from Proposi-

tion 3.3.4, we infer that Φ (u), uB ,B ≥ 0 for any u ∈ B.
Definition 3.3.6 One says that ζ ∈ B belongs to the subdifferential of Φ at the

point u ∈ B if and only if
∀v ∈ B, φ(v) − Φ(u) ≥ ζ, v − uB ,B .
Usually, the subdifferential of Φ at the point u is denoted ∂Φ(u). By definition,

∂Φ(u) is a set of points in B . It is clear from Definition 3.3.1 that if ∂Φ(u) is a set of
more than one point in B , then Φ is not G-differentiable at u. The converse is also
true, it is given in the following important result due to Rockafellar [247]:
Theorem 3.3.7 If Φ is a convex function such that the subdifferential of Φ at the

point u is reduced to (only) one point ξ ∈ B , then Φ is G-differentiable at u and
ξ = Φ (u).
Let now recall another notion, extensively used in Chapter 4, Subsection 4.4.3.
Definition 3.3.8 The functional Φ is lower semi-continuous (l.s.c.) for the weak
topology of B if
lim inf Φ(un ) ≥ Φ(u) if un u in the weak topology of B.

n→∞
Examples
• The functional u −→ u is l.s.c. in a Banach space.
• Let Ω be a bounded domain in IRd and let B = H 1 (Ω). For u ∈ H 1 (Ω), define the
functional,

|∇ u|2 dx + |∇ u| dx if u ∈ H01 (Ω),
φ(u) = Ω Ω
+∞ otherwise.
Then φ is convex and l.s.c on H 1 (Ω).

Let V = V (Ω) be a Banach space and Φ a functional on V . Suppose that for
f ∈ V , we have to solve the variational inequality

Find u ∈ V such that
(3.3.3)
a(u, v − u) + Φ(v) − Φ(u) ≥ f , v − uV ,V , ∀v ∈ V ,
where
a(u, v) = ∇ u · ∇ v dx.
Ω
Then the following result holds true:

Lemma 3.3.9 If Φ is a convex G-differentiable functional on V , then (3.3.3) is
equivalent to solving the equation
a(u, v) + Φ (u), vV ,V = f , vV ,V ∀v ∈ V . (3.3.4)
Proof Let us test (3.3.3) with u + λw for λ > 0 and w arbitrary in V . After dividing
by λ, we get
1
a(u, w) + Φ(u + λw) − Φ(u) ≥ f , wV ,V .
λ
Letting λ → 0, yields
a(u, w) + Φ (u), wV ,V ≥ f , wV ,V ,
and replacing w by −w, it follows that u solves (3.3.4).

Now suppose that u is a solution of (3.3.4) that we rewrite as

a(u, v − u) + Φ(v) − Φ(u) = f , v − uV ,V + Φ(v) − Φ(u) − Φ (u), v − uV ,V .
Using the fact that the term in brackets is positive by Proposition 3.3.2, we get that
u solves (3.3.3).
Remark 3.3.10 In view of Definition 3.3.6, the variational inequality (3.3.3) may be
written as
−Δu − f ∈ ∂Φ(u),
if for instance, V = H01 (Ω).

3.3 Further Tools of Functional Analysis 113
3.3.2 Some Tools for Nonlinear and Time-Dependent

Problems
In order to analyze nonlinear problems, we shall use the following variant of

Brouwer’s Fixed-Point Theorem (cf. for instance [115]) and Schauder’s Fixed-Point
Theorem.
Lemma 3.3.11 Let H be a finite dimensional Hilbert space, with scalar product and
associated norm denoted, respectively, by (·, ·)H and | · |H . Let Φ be a continuous
mapping from H into H satisfying the following property: There exists a constant
μ > 0 such that
∀f ∈ H with |f |H = μ, (Φ(f ), f )H ≥ 0.
Then Φ has at least one zero in H satisfying
Φ(f ) = 0 and |f |H ≤ μ.
In other words, if (Φ(f ), f ) ≥ 0 (or if (Φ(f ), f ) does not change sign) on the
surface of the ball B(0; μ), then Φ has at least one zero in B(0; μ).
Theorem 3.3.12 Let C be a nonempty, convex, and compact subset of a normed

vector space and let F be a continuous mapping from C into C. Then F has at least
one fixed point.
For handling time-dependent problems, we shall use Gronwall’s Lemma.
Lemma 3.3.13 Let κ be a nonnegative integrable function defined a.e. on an interval

]0, T [. Let C ≥ 0 be a constant and ϕ ∈ C 0 ([0, T ]) a function satisfying the inequality
t
∀t ∈ [0, T ], 0 ≤ ϕ(t) ≤ C + κ(s)ϕ(s) ds. (3.3.5)
0
Then ϕ is bounded by

t
κ(s) ds
∀t ∈ [0, T ], 0 ≤ ϕ(t) ≤ Ce 0 . (3.3.6)
The next result establishes compact imbeddings in space and time and generalizes
the Aubin-Lions Lemma, see Aubin [13], Lions [161,162]. Its proof, due to Simon,
is written in [252].
Theorem 3.3.14 Let X, E, Y be three Banach spaces with continuous imbeddings:
X ⊂ E ⊂ Y , the imbedding of X into E being compact. Then for any number
q ∈ [1, ∞], the space
∂v
v ∈ L q (0, T ; X) ; ∈ L 1 (0, T ; Y ) (3.3.7)
∂t
is compactly imbedded into L q (0, T ; E).

Finally, we recall a useful Green formula in time. Let X be a Hilbert space and X
its dual space; consider the following generalization of H 1 (a, b; X).
dv
W (0, T ; X) = v ∈ L 2 (0, T ; X) ; ∈ L 2 (0, T ; X ) , (3.3.8)
dt
which is a Hilbert space for the norm
dv 21
vW (0,T .X) = v2L2 (0,T ;X) + 2L2 (0,T ;X ) . (3.3.9)
dt
The next properties of W (0, T ; X) are well known; the proofs can be found for
instance in [284].
Lemma 3.3.15 The space C ∞ ([0, T ]; X) is dense in W (0, T ; X).
Theorem 3.3.16 Let V and H be two Hilbert spaces such that V ⊂ H ⊂ V with
continuous and dense imbeddings. Then
W (0, T ; V ) ⊂ C 0 ([0, T ]; H),
with continuous and dense imbedding and all u and v in W (0, T ; V ) satisfy the
following Green formula:
T du(t) dv(t)
, v(t) + ,u(t) dt
0 dt V ,V dt V ,V
= (u(T ), v(T ))H − (u(0), v(0))H . (3.3.10)

Chapter 4
Classical Non-Newtonian Fluids
4.1 Introduction
The aim of this section is to discuss the mathematical properties of the govern-
ing equations of some non-Newtonian fluids introduced in Chapter 2, Section 2.4,
namely, the Reiner-Rivlin fluid and in particular, the Bingham fluid. By compress-
ible Reiner-Rivlin fluid, we refer to the class of fluids defined by (2.4.14) where the
material moduli α0 , α1 , and α2 are arbitrary functions of the density and the principal
invariants of the symmetric part of the velocity gradient (any appropriate integrity
basis). A compressible Reiner fluid is one wherein the material moduli are polyno-
mials of the principal invariants (or polynomials of an appropriate integrity basis).
We shall, however, only consider incompressible fluids and by an incompressible
Reiner-Rivlin fluid we refer to the fluids defined by (4.2.1) below, where the material
moduli ϕ1 and ϕ2 are arbitrary functions of the second and third principal invariants,
while an incompressible Reiner fluid is defined by (4.2.1) where the material moduli
ϕ1 and ϕ2 are polynomial functions of the second and third principal invariants.
Two classes of fluids will be studied. The first corresponds to what we shall refer
to as Reiner fluids, the second to some particular examples of Reiner-Rivlin fluids.
The coefficients in the constitutive equation of Reiner fluids are in general poly-
nomials of the integrity basis (say the principal invariants), but we shall only consider
them to be polynomials of the second principal invariant D2 of the symmetric part
of the velocity gradient D. Several fluid models that have been proposed are spe-
cial subclasses of the models discussed in Chapter 2, Section 2.4. Some examples of
the same are the dilatant fluids or the pseudo plastic fluids (see for other examples
Wilkinson [283], Aziz and Govier [14]), all of them will be treated in the last part
of the first section below. We will give a general constitutive equation covering not
only the classical Reiner fluids but also those with a threshold, like the Bingham
fluid presented in Section 2.9. Strictly speaking, this fluid does not fall in the class
of Reiner-Rivlin fluids since its constitutive function is not really a polynomial as

DOI 10.1007/978-3-319-39330-8_4
116 4 Classical Non-Newtonian Fluids
can be seen in Section 4.3. Nevertheless, it turns out to be a model example for our
general constitutive relation.
The general constitutive relation we define in Section 4.2 is of subdifferential
type and therefore, the flow of a fluid so defined is characterized by a variational
inequality, for which general existence and uniqueness results will be given. An
abridged version of the results we present here is contained in Cioranescu [62]. They
extend and generalize earlier results due to Brezis [49], Ladyzhenskaya [155] and
Duvaut and Lions [91].
Let us now turn to our model, the Bingham fluid, characterized by the fact that
it has a stress threshold g ≥ 0, for more details see Section 2.9 and Section 4.3
below. The equations governing the flow of such a fluid were studied in Duvaut and
Lions [91] where it was shown for the first time (to the best of our knowledge) that
its flow is given by a variational inequality if g > 0 (g = 0 corresponding actually
to the Newtonian fluid and so, to the Navier–Stokes equations). Let us point out that
the existence and uniqueness of its flow are obtained as simple applications of the
general results mentioned above. In particular, since g = 0 leads to the Navier–Stokes
equations, the case of Navier–Stokes fluids is also covered by the same results.
As said before, we are also concerned by a particular class of Reiner-Rivlin fluids.
This is the object of Section 4.4. For this class, the coefficients in the constitutive
relation are still nonlinear functions of the second invariant but not anymore poly-
nomials. In the class we consider here, the main feature of the coefficients is that
they have a finite limit at infinity. This situation occurs for quite a large class of
non-Newtonian fluids, some very popular models for example, the Carreau fluid, the
Eyring–Prandtl, or the Cross fluids, their definitions being recalled at the beginning
of Section 4.4. We will again give some mathematical results for this class, which are
rather close to those obtained for the Navier–Stokes equations as regarding unique-
ness, stability, and asymptotic behavior in time. Some of these results are contained
in Boujena [45] (see also Sobolevsky [255] for another approach).
4.2 Subdifferential Type Constitutive Equation
4.2.1 Formulation and Examples
We start by recalling the definitions of Reiner and Reiner-Rivlin fluids given in

Section 2.4. The constitutive relation for a Reiner fluid (2.4.14) [242], discussed in
Section 2.4 is of the form
T = − p I + ϕ1 D + ϕ2 D2 , (4.2.1)
where D is the symmetric part of the velocity gradient defined by (2.3.23), i.e.,
1
D= grad v + (grad v)T ,
2
4.2 Subdifferential Type Constitutive Equation 117
and ϕ1 and ϕ2 are polynomial functions of D2 and D3 . Recall that D1 , D2 , D3 are

the principal invariants of D:
1 1
D1 = tr D, D2 = Di j Di j , D3 = Di j D jk Dki ,
2 3
where summation is implied by repeated indices. As noticed in Chapter 2, D1 = 0
in the incompressible case. With the notation of Section 3.1, D2 and D3 can also be
written as
1 1 1
D2 = tr( D2 ) = | D|2 , D3 = tr( D3 ),
2 2 3
where | · | denotes the Frobenius norm.
The Reiner-Rivlin fluids obey the same constitutive equation (4.2.1) but with
general functions ϕ1 and ϕ2 (not polynomial) of the same invariants.
Remark 4.2.1 The constitutive equation (4.2.1) was obtained by appealing to the
Rivlin-Ericksen representation theorems for isotropic functions (and the use of the
Hamilton-Cayley theorem). It says that if F is an isotropic function of a symmetric
tensor A, it has necessarily the form
F( A) = f 1 (A1 , A2 , A3 )I + f 2 (A1 , A2 , A3 ) A + f 3 (A1 , A2 , A3 ) A2 ,
where A1 , A2 and A3 are the principal invariants of A and the functions f 1 , f 2 , and
1 , A
f 3 are isotropic (i.e., f i (A1 , A2 , A3 ) = f i ( A 2 , A
3 ) where the tilde denotes the

corresponding invariant of A = Q A Q for any orthogonal tensor Q). So, (4.2.1)
T
is the most general constitutive relation that can be obtained in this framework, as
function of D. As explained in Chapter 1, the fluids defined by a constitutive equation
of this type do not have distinct normal stress differences in simple shear flows, nor
can they describe stress relaxation.
We are concerned here with both constitutive relations restricted to ϕ1 = ϕ(D2 )
and ϕ2 ≡ 0. We will only consider the incompressible case, i.e., D1 = 0 (see
Chapter 2), so that (4.2.1) reads
T = − p I + ϕ(D2 ) D.
In view of the strong nonlinearity of the constitutive relations for Reiner and
Reiner-Rivlin fluids, there is no mathematical study for their full general forms,
that is to say, for ϕ2 ≡ 0.
Let Ω be a connected bounded domain in IR d (d = 2 or d = 3 will be of interest
in our applications below), with a sufficiently smooth boundary (see for more details
Chapter 3, Section 3.1). Let F be a proper convex function on L 2 (Ω)d×d , i.e.,
F : L 2 (Ω)d×d −→] − ∞, +∞], F convex, F ≡ +∞.

Denote by τ the deviatoric part of the stress tensor T , i.e., τ = p I + T . We now

introduce the following constitutive relation:
τ ∈ ∂ F( D(u)). (4.2.2)
Here, ∂ F( D(u)) is the subdifferential of F at the point D(u) ∈ L 2 (Ω)d×d with

coordinates Di j (u), i, j ∈ {1, 2, . . . , d} (for details and properties of this notion, see
Section 3.3.1).
Recalling the definition 3.3.6 of the subdifferential with B = B = L 2 (Ω)d×d , it
follows that τ ∈ L 2 (Ω)d×d , satisfies

F(w) − F( D(u)) − τ , w − D(u) ≥ 0 , ∀w ∈ L 2 (Ω)d×d . (4.2.3)
Let v ∈ H01 (Ω)d and write formally (4.2.3) at the point w = D(v) ∈ L 2 (Ω)d×d ,

F( D(v)) − F( D(u)) − τ (u) · D(v − u) d x ≥ 0, (4.2.4)
Ω
where as usual, the colon denotes the scalar product of two tensors.
Now, recall the equation of motion (written for the components of the velocity u)
of an incompressible fluid, derived from the balance of the linear momentum given
in Section 2.3.3) where, to simplify, we take the density = 1,
∂u i ∂u i ∂ Ti j
+ uj = + f i in Ω,
∂t ∂x j ∂x j (4.2.5)
div u = 0 in Ω,
with f a body force. We will add to this system a homogeneous boundary condition
on the wall ∂Ω, i.e.,
u = 0 on ∂Ω.
Observe that
∂u i
uj = [∇ u]u i ,
∂x j
for which reason the product [∇ v]u is often denoted u · ∇ v.

Multiplying the equation in (4.2.5) by v − u where v is a function such that
div v = 0 in Ω,
v = 0 on ∂Ω,
and integrating by parts yield

∂u
T : D(v − u) d x = f− − [∇ u]u · (v − u) d x.
Ω Ω ∂t
Using (4.2.4) we obtain

∂u
F( D(v)) − F( D(u)) ≥ f− − [∇ u]u · (v − u) d x. (4.2.6)
Ω ∂t
From (3.3.3) and Lemma 3.3.9 it is obvious that if F is not G-differentiable (see
Definition 3.3.1 for this notion), then u is a solution of a variational inequality. But
if F is G-differentiable, by Lemma 3.3.9, the variational inequality (4.2.6) reduces
to the following equation:

(F ◦ D) (u), v = F ( D(u)), D(v)

∂u
= f− − [∇ u]u · v d x, (4.2.7)
Ω ∂t
where F is the G-derivative of F.

A particular case. Suppose that

F( D(u)) = φ(D2 (u)) d x, (4.2.8)
Ω
where φ : s −→ φ(s) is a function of one variable, proper convex, l.s.c. and
differentiable. It is with this kind of functional that we will be concerned in the
sequel.
Remark 4.2.2 Taking into account the definition of G-derivative, an easy computa-
tion shows that in the case of (4.2.8),

F ( D(u)), D(v) = φ (D2 (u)) D(u) : D(v) d x, (4.2.9)
Ω
dφ
where φ stands for the derivative of φ with respect to s.
ds
In view of this remark, equation (4.2.7) with F defined by (4.2.8) becomes

∂u
φ (D2 (u)) D(u) : D(v) d x = f− − [∇ u]u · v d x. (4.2.10)
Ω Ω ∂t
Relation (4.2.10) means that the constitutive equation (4.2.1) for the stress tensor T
is
T = − p I + φ (D2 (u)) D(u).
Examples
1. Newtonian fluid
Since a Newtonian fluid is defined by
T = − p I + 2μ D(u),
where μ > 0 is the viscosity, the function φ in (4.2.8), denoted here φ N , is
φ N (s) = 2μ s. (4.2.11)
Thus, φ N = 2μ.
From (4.2.7), we derive the well-known Navier–Stokes equations for an incom-
pressible fluid
∂u
− μΔu + C(u, u) = f − ∇ p in Ω×]0, T [,
∂t
div u = 0 in Ω×]0, T [, (4.2.12)
u = 0 on ∂Ω×]0, T [,
u(0) = u0 in Ω,
where we are given the exterior force f defined on Ω×]0, T [, and the initial condition
u0 . The nonlinear term C(u, u) is defined as
∂v
C(u, v) = [∇ v]u = u j . (4.2.13)
∂x j
2. Dilatant fluids
They are characterized by the function φ denoted here by φ D , defined either by
φD = φN + ν ψD , (4.2.14)
with φ N defined by (4.2.11), or simply by
φD = μ ψD ,
where μ and ν are positive constants, and
2 p
ψ D (s) = s2, p > 2,
p
so that
2 2p
ψ D (D2 (u)) d x = D2 (u) d x.
Ω p Ω
Inserting this definition into (4.2.7), one can easily see that the system of equations
for the motion of this fluid is of the same form as the Navier–Stokes equations but
with the Laplacian Δ replaced by the following operator, known as the “ p-Laplacian”

Δ p u = div |∇u| p−2 ∇u . (4.2.15)
Therefore, equation (4.2.10) takes the form

∂u
|∇u| p−2
D(u) : D(v) d x = f− − [∇ u]u · v d x. (4.2.16)
Ω Ω ∂t
3. Pseudoplastic fluids
They are characterized by several definitions, the most popular being the Sisko equa-
tion,
η−1
T = − p I + 2μ D(u) + α D2 (u) 2 D(u), 0 < η < 1,
where μ and α are positive constants. Then the corresponding function φ, denoted
here by φ P , reads
φP = φN + ψP ,
with φ N defined by (4.2.11) and
2α η+1
ψ P (s) = s 2 , 0 < η < 1,
η+1
so that
2α η+1
ψ D (D2 (u)) d x = D2 (u) 2
d x, 0 < η < 1.
Ω η+1 Ω
4.2.2 Functional Setting
In this section, we introduce the mathematical framework where problem (4.2.10)

will be studied. To begin with, let us give its variational formulation.
With w a smooth test function such that div w = 0 and w = 0 on ∂Ω, (4.2.6)
formally reads
(u , w − u) + c(u; u, w − u) + F( D(w)) − F( D(u))

≥ ( f , w − u) in Ω×]0, T [,
div u = 0 in Ω×]0, T [ , (4.2.17)
u = 0 on ∂Ω×]0, T [,
u(0) = 0 in Ω,
where the prime denotes the derivative with respect to time, and c is the trilinear
form (classical in the theory of Navier–Stokes equations) defined by

∂v j
c(u; v, w) = ([∇ v]u, w) = C(u, v) · w d x = ui w j d x, (4.2.18)
Ω Ω ∂ xi
with C given by (4.2.13).

A first task is to make precise the functional spaces chosen for the solutions of
the variational inequality (4.2.17). Some of them were introduced in Chapter 3, as
the space V defined by (3.1.16), i.e.,
V = {v ∈ H01 (Ω)d ; div v = 0 in Ω}, (4.2.19)
and the space H ,
H = {v ∈ L 2 (Ω)d ; div v = 0 in Ω, v · n = 0 on ∂Ω}, (4.2.20)
which is the closure of V defined by (3.1.15), with respect to the norm of L 2 (Ω)d .
Denoting dual spaces with a prime, we recall that
V ⊂ H ⊂ V ,
with dense and continuous imbeddings, and compactness for the first one.
Notice that, due to Sobolev’s imbeddings (see for more details, Section 3.1.1),
H 1 (Ω) ⊂ L ρ (Ω), (4.2.21)
with
1 1 1
= − if d>2; ρ arbitrary if d = 2. (4.2.22)
ρ 2 d
Now, let us look at the nonlinear form c and list some of its properties that will
be extensively used in the sequel. The following ones are straightforward:
• c(u, v, w) is linear with respect to u, v, and w separately,
• by the usual Green formula, the form c satisfies
c(u; v, w) = −c(u; w, v), ∀u ∈ V, v, w ∈ H 1 (Ω)d . (4.2.23)
This entails in particular, that
c(u; u, u) = 0,
a crucial identity in the mathematical modeling of fluid flows,

• c is continuous on
H 1 (Ω)d × H 1 (Ω)d × H 1 (Ω)d for d ≤ 4,

H 1 (Ω)d × H 1 (Ω)d × (H 1 (Ω) ∩ L d (Ω))d for d > 4.
This is easily seen by applying the Hölder inequality with ρ given by (4.2.22) to
get
|c(u; v, w)| ≤ u L ρ (Ω) w L d (Ω) v H 1 (Ω) ,
and then the continuity is a consequence of (4.2.21) observing that H 1 (Ω) ∩

L d (Ω) = H 1 (Ω) for d ≤ 4.
To go further, let us recall some important results concerning the spaces V and
H , they will enable us to sharpen the properties of the form c. We will see that these
properties strongly depend on the dimension d of the space. Indeed, Lemmas 3.1.1
and 3.1.3 imply that H01 (Ω)d ⊂ L 4 (Ω)d for d = 2 and d = 3 (the cases of interest)
with

1

1
v L 4 (Ω) ≤ C
v
L2 2 (Ω)
v
H2 1 (Ω) for d = 2,
0

41

34 (4.2.24)

v L 4 (Ω) ≤ C v L 2 (Ω) v H 1 (Ω) for d = 3,

0
for all v ∈ H01 (Ω)d and consequently, for all v ∈ V . These inequalities follow by
applying Lemmas 3.1.1 and 3.1.3 component by component to the vectorial function
v. Let us mention that for d = 4, one still has v ∈ L 4 (Ω)4 (see for instance,
Temam [265]).
The following result is a straightforward consequence of (3.1.23):
Lemma 4.2.3 Let d = 2. Then for all u ∈ H01 (Ω)2 and v ∈ H 1 (Ω)2 ,
1
|c(u; u, v)| ≤ √ u L 2 (Ω) u H01 (Ω) v H01 (Ω) . (4.2.25)
2
Proof Since by the antisymmetry of c,

2

|c(u; u, v)| = | − c(u; v, u)| ≤

u
L 4 (Ω)
v
H 1 (Ω) ,
0
the estimate is a simple application of (3.1.23) to each component of u and v.
Recalling definition (4.2.13) and formula (4.2.18), i.e.,

c(u; v, w) = C(u, v) · w d x,
Ω
we state the following result whose proof is obvious in view of the antisymmetry of
c, (4.2.23) and (4.2.24):
Proposition 4.2.4 Let
u ∈ L 2 (0, T ; V ) ∩ L ∞ (0, T ; H ) and v ∈ L 2 (0, T ; H01 (Ω)d ) ∩ L ∞ (0, T ; L 2 (Ω)d ).
Then
C(u, v) ∈ L 2 (0, T ; H −1 (Ω)d ) for d = 2,

L 3 (0, T ; H −1 (Ω)d ) for d = 3,
4
C(u, v) ∈
C(u, v) ∈ L 1 (0, T ; H −1 (Ω)d ) for d = 4.
A last result concerning the form c is related to its convergence for bounded
sequences in appropriate spaces.
Proposition 4.2.5 Let d = 2 or 3, and let vn be a sequence in L 2 (0, T ; V ) ∩
L ∞ (0, T ; H ) such that
lim vn = v weakly in L 2 (0, T ; V ),

n→∞
lim vn = v weakly in L ∞ (0, T ; H ),
n→∞
and
lim vn = v strongly in L 2 (0, T ; H ). (4.2.26)
n→∞
Then, strongly in L 1 (0, T ),
lim c(vn ; vn , w) = c(v; v, w) for w ∈ H01 (Ω)d .

n→∞
Proof Since div vn = 0 implies that formula (4.2.23) holds, i.e.,
d
∂(w) j
c(vn ; vn , w) = − (vn )i (vn ) j d x,
i, j=1 Ω
∂ xi
we only have to consider the convergence of products of the form (vn )i (vn ) j .
If d = 2, due to (4.2.26) and to Lemma 3.1.1, we have
lim (vn )i = vi strongly in L 2 (Ω×]0, T [),

n→∞
lim (vn )i = vi weakly in L 4 (0, T ; L 4 (Ω)).
n→∞
But by the Hölder inequality, applied for any α < 4, one has
lim (vn )i = vi strongly in L α (0, T ; L α (Ω)).

n→∞
Therefore, for β < 2,
lim (vn )i (vn ) j = vi v j strongly in L β (0, T ; L β (Ω)),

n→∞
hence weakly in L 2 (Ω×]0, T [), and this gives the convergence of c(vn ; vn , w).
If d = 3, the following convergences hold by assumption for the first one and by
Sobolev’s imbeddings for the second one:
lim (vn )i = vi strongly in L 2 (Ω×]0, T [),

n→∞
lim (vn )i = vi weakly in L 2 (0, T ; L α (Ω)),
n→∞
for any α ≤ 6. This time, by Hölder’s inequality, we get for any α < 6,
lim (vn )i = vi strongly in L 2 (0, T ; L α (Ω)),

n→∞
and so, for β < 3
lim (vn )i (vn ) j = vi v j strongly in L 1 (0, T ; L β (Ω)),

n→∞
and this convergence again entails that of c(vn , vn , w).
In the sequel, we will need the following spaces defined for p ≥ 1 and s ≥ 0:
1, p
W p = closure of V in W 1, p (Ω)d = {v ∈ W0 (Ω)d ; div v = 0 in Ω},
(4.2.27)
Vs = closure of V in H s (Ω)d ,
with s integer or not (see [164] for more details on H s (Ω)-spaces). The space H s (Ω)d
is endowed with the Hilbert scalar product (see (3.1.1)),

d
((v, w))s = ((vi , wi )) H s (Ω) .
i=1
Notice that space Vs for s > 1

2
coincides in a Lipschitz domain with
{v ∈ H0s (Ω)d ; div v = 0 in Ω}.
Notice also that by definition, one has W 2 = V . Furthermore, observe that by Korn’s
1, p
inequality (3.2.19), for 1 < p < ∞, the space W0 (Ω)d can be endowed with the
norm

2p 1p

v
= D 2 (v) dx , (4.2.28)
p
Ω
which is equivalent to that induced by W 1, p (Ω)d and coincides with v H01 (Ω) when
p = 2.
The equivalence of norms is also valid in W p when p = 1 or p = ∞. By
identifying H with its dual space, one has the inclusions
W p ⊂ W p−1 ⊂ ... ⊂ W 2 = V ⊂ H ⊂ V ⊂ ... ⊂ (W p−1 ) ⊂ (W p ) , ∀ p ≥ 3.
d
Let s > 1 + . If v ∈ H s (Ω), by Sobolev’s imbeddings (3.14),
2
1 s−1
Di v ∈ H s−1 (Ω) ⊂ L ∞ (Ω) since − < 0. (4.2.29)
2 d
This implies that in particular, Di v ∈ L p (Ω) so that
Vs ⊂ W p ⊂ H ⊂ (W p ) ⊂ (Vs ) , ∀ p ≥ 2.
4.2.3 Existence Theorems
Let us go back to our variational inequality (4.2.17). We now make the following
assumptions:
H.1. The function F has the form
F( D(u)) = μ(u) + gΨ (u),
where

(u) = φ(D2 (u)) d x, Ψ (u) = ψ(D2 (u)) d x, (4.2.30)
Ω Ω
and μ ≥ 0, g ≥ 0. The functions of one variable φ and ψ and the functions φ ◦ D2

and ψ ◦ D2 are assumed to be proper convex and l.s.c., and ψ is non differentiable.
H.2. The function φ is differentiable and is such that there exist q ≥ p > 1 and
α > 0 with
qp

q
(D2 (v)) 2 d x = α
v
p .
p

φ (D2 (v)) D(v) : D(v) d x ≥ α (4.2.31)
Ω Ω
Moreover, we assume that
1 1
φ (D2 (v)) D(v) ∈ L p (Ω)d×d ,
+ = 1,
p p
and satisfies

q−1

q−1

φ (D2 (v)) D(v)
p ≤ C (D 2 (v))
p
2 dx
p
= C
v
p , (4.2.32)
L (Ω)
Ω
where C is a positive constant.

H.3. There exists a family of convex differentiable functions of one variable ψε , such
that ψε ◦ D2 is convex and the function Ψε , defined by

Ψε (v) = ψε (D2 (v)) d x,
Ω
satisfies
T T
lim Ψε (v(t)) dt = Ψ (v(t)) dt, ∀v ∈ L 2 (0, T ; H01 (Ω)d ). (4.2.33)
ε→0 0 0
We also assume that

ψε (0) = 0.
H.4. For every sequence {vε } such that
lim vε = v weakly in L 2 (0, T ; V ),

ε→0
and
T

Ψε (vε (t)) dt ≤ C (C constant independent of ε),
0
one has
T T
lim Ψε (vε (t)) dt ≥ Ψ (v(t)) dt.
ε→0 0 0
H.5. We assume that

1 1
ψε (D2 (v)) D(v) ∈ L p (Ω)d×d , + = 1,
p p
and one has the estimates

q−1

q−1

ψ (D2 (v)) D(v)
p ≤C
p
(D2 (v)) 2 d x
p
= C
v
p ,
ε L (Ω)
Ω
where C is a strictly positive constant.

Remark 4.2.6 Hypothesis H.2 is obviously verified by polynomial functions for

appropriate values of p and q.
Finally, we introduce the operator A defined as follows:

A(v), w = A(v), w W −1, p (Ω)d ,W 1, p (Ω)d = μ (D(v)), D(w)
0
(4.2.34)
=μ φ (D2 (v)) D(v) : D(w) d x,
Ω
where we used Remark 4.2.2. From (4.2.31) in hypothesis H.2, for real p > 1, the
1, p
operator A is coercive on W0 (Ω)d , and

q
A(v), v ≥ μ α
v
p . (4.2.35)

On the other hand, due to (4.2.32), A maps W0 (Ω)d into W −1, p (Ω)d . Moreover,
1, p
A is a monotone hemicontinuous operator. Indeed, this follows from Proposition 3.3.4

since u −→ A(u) corresponds to the gradient of the functional
u −→ (u)
with convex and G-differentiable, by hypotheses H.1 and H.2.
Remark 4.2.7 At this point, one can make a few comments on the above hypothe-
ses. Hypothesis H.2 says exactly that the operator A defined by (4.2.34) is coer-
1, p 1, p
cive on W0 (Ω)d , is monotone and hemicontinuous acting from W0 (Ω)d into
−1, p
W (Ω) . Apparently, what plays also an essential role for establishing these prop-
d
erties of A, is the special form of the functional given by H.1, namely (4.2.30).
As a matter of fact there are other possibilities for defining the form of , as long as
they guarantee that the corresponding operator A enjoys the same properties. One of
them is the following one:
β
(u) = φ(D2 (u)) d x , (4.2.36)
Ω
for some β > 1. In this case, it is easily seen that instead of (4.2.9) one has

1 β−1
F ( D(u)), D(v) = φ(D2 (u)) d x
β Ω

× φ (D2 (u)) D(u) : D(v) d x, (4.2.37)
Ω
and obviously, hypothesis H.2 has to be formulated accordingly, i.e.,

(H.2)’. The function φ is differentiable and is such that there exist q ≥ p > 1 and
α > 0 with
β−1
φ(D2 (u)) d x φ (D2 (u)) D(u) : D(v) d x
Ω Ω
qp

q
(D2 (v)) 2 d x = α
v
p .
p
≥α
Ω
Moreover, we assume that

β−1 1 1
φ(D2 (v)) d x φ (D2 (v)) D(v) ∈ L p (Ω)d×d ,
+ = 1,
Ω p p
and satisfies

β−1
q−1

p
φ (D2 (v)) D(v)
p

φ(D2 (v)) d x ≤C (D2 (v)) 2 d x
Ω L p (Ω) Ω

q−1
= C
v
p , (4.2.38)
where C is a positive constant.
Now define

Uad = {v ; v ∈ L q (0, T ; Vs ), v ∈ L q (0, T ; H ), v(0) = 0},
the set of “admissible” test functions for problem (4.2.17) in the functional frame-
work introduced above.
For v in Uad , take w = v(t) a.e. in ]0, T [ as test function in (4.2.17). Since
1 d
(v − u , v − u) = v − u2L 2 (Ω) ,
2 dt
and so, T
d
v − u2L 2 (Ω) dt = v(T ) − u(T )2L 2 (Ω) ≥ 0, (4.2.39)
0 dt
we derive, again formally, by integrating over ]0, T [

T
(v , v − u) + A(u), v − u + c(u; u, v − u) + gΨ (v) − gΨ (u) dt
0
T
≥ f , v − udt, ∀v ∈ Uad . (4.2.40)
0
Here we have used Proposition 3.3.2 and the convexity of u −→ (u) by Hypothesis
H.1. Summing up, we are looking for u satisfying (4.2.40) and
div u = 0 in Ω×]0, T [,
u = 0 on ∂Ω×]0, T [, (4.2.41)
u(0) = 0 in Ω.
We take as definition of our problem the variational inequality (4.2.40) together with
conditions (4.2.41).
T
Remark 4.2.8 One might wonder why the integral 0 (u , v−u) dt has been replaced
T
by the integral 0 (v , v − u) dt in the “weak” formulation (4.2.40). As can be seen
in the proof of the existence theorem, this is related to the fact that u and u belong to
spaces of vector-valued functions which are not in duality, so that the product (u , u)
may not have sense (and not belong to L 1 (0, T )), in contrast to some particular cases
when d = 2, see Subsection 4.2.8.
Remark 4.2.9 If Ψ ≡ 0 inequality (4.2.40) is an equation (see Lemma 3.3.9) that
reads in a “strong” formulation
(u , v) + A(u), v + c(u; u, v) = f , v, ∀v ∈ Uad .
This equation with the “ p-Laplacian” (see (4.2.15)) for operator A, was studied in
Lions [162]. As seen above, this case corresponds to dilatant fluids.
Theorem 4.2.10 Let q ≥ p ≥ 2 and

f ∈ L q (0, T ; W −1, p (Ω)d ).
Under hypotheses H.1–H.5, there exists u satisfying (4.2.40), and such that
u ∈ L q (0, T ; W p ) ∩ L ∞ (0, T ; H ),
d
u ∈ L q (0, T ; (Vs ) ) for s > 1 + ,
2
u(0) = 0.
The proof is written in the next two subsections. It consists of the following steps:
1. Biregularization of problem (4.2.40) and Galerkin solutions.
2. First passing to the limit.
3. Second passing to the limit, back to variational inequality (4.2.40).
Remark 4.2.11 Let us point out that the data f is taken here with values in the space

L q (0, T ; W −1, p (Ω)d ), whereas in several references (see for instance Lions [162] or

Temam [265]), it is taken in L q (0, T ; (W p ) ), the dual of L q (0, T ; W p ), the natural

space where u is searched. Theorem 4.2.10 still holds with f in L q (0, T ; (W p ) )
without any change in its proof. However, we choose to take the values of f in the

smaller space L q (0, T ; W −1, p (Ω)d ) in order to be able to recover the pressure, as
it can be seen in Section 4.2.9.
4.2.4 Biregularization and Galerkin Solutions
In order to regularize the variational inequality (4.2.40), we first replace the non
differentiable functions ψ by the differentiable functions ψε from hypotheses H.3–
H.5. As this is not sufficient to guarantee that the solution belongs to a space of
the type L r (0, T ; H s (Ω)d ) instead of L a (0, T ; W 1,b (Ω)d ) with b < 2, we stabilize
the equation by using an artificial viscosity method. This is achieved by means of a
monotone hemicontinuous operator A from H0s (Ω)d to H −s (Ω)d such that

∀w ∈ H0s (Ω)d , A(w), w ≥ β(w H s (Ω) )q for some β > 0,

| A(w), v| (4.2.42)
sup ≤ C(w H s (Ω) )q−1 .
v∈H0s (Ω)d v H s (Ω)
This double regularization leads to the variational equation

u εδ , v + A(uεδ ), v + c(uεδ ; uεδ , v) + g Ψε (uεδ ), v
εδ ), v = f , v, ∀v ∈ Vs ,
+ δ A(u (4.2.43)
where δ is positive and (see Remark 4.2.2)

Ψε (uεδ ), v = ψε (D2 (uεδ )) D(uεδ ) : D(v) d x.
Ω
Our first task now is to solve the biregularized equation (4.2.43). As will be done
in Chapter 5 for second grade fluids, (4.2.43) will be solved by Galerkin’s method
with a special basis that we shall define below.
Observe first that the injection Vs ⊂ H is compact. This is obvious since Vs ⊂ V
and the injection V ⊂ H is compact. It follows that the spectral problem: Find
w ∈ Vs ,
((w, v))s = λ(w, v) L 2 (Ω) , ∀v ∈ Vs ,
has a set of eigenfunctions {w j } j≥0 corresponding to the eigenvalues {λ j } j≥0 such

that
((w j , v))s = λ j (w j , v) L 2 (Ω) , ∀v ∈ Vs ,
(4.2.44)
w j L 2 (Ω) = 1.
This special basis is orthogonal in Vs and orthonormal in H .

Let Vm be the m dimensional subspace of H , spanned by w1 , . . . , wm . What is

essential in the analysis of the Galerkin method are the following classical facts.
On the one hand, the projection operator Pm from H to Vm and its restriction to Vs ,
defined respectively by

m
Pm v = (v, wi ) L 2 (Ω) wi , ∀v ∈ H,
i=1

m
Pm v = (v, wi ) L 2 (Ω) wi , ∀v ∈ Vs ,
i=1
are such that
lim Pm v = v strongly in H, ∀v ∈ H, (4.2.45)

m→∞
lim Pm v = v strongly in Vs , ∀v ∈ Vs . (4.2.46)
m→∞
On the other hand, Pm can be extended to (Vs ) by

m
Pm v = v, wi (Vs ) ,Vs wi , ∀v ∈ (Vs ) .
i=1
Finally, all norms of Pm as a linear and continuous operator from H to H , from V

to V , and from (Vs ) to (Vs ) are less than or equal to one.
The Galerkin method reads: Find um = umεδ of the form

m
um (x, t) = g mj (t)w j (x),
j=1
satisfying the finite-dimensional approximate problem
(u m , wi ) + A(um ), wi + c(um ; um , wi )

m ), wi = f , wi ,
+ g Ψε (um ), wi + δ A(u 1 ≤ i ≤ m, (4.2.47)
um (0) = 0,
a system of m ordinary first order differential equations with m unknowns g1m , . . . , gmm .
It reads
dgim
+ A(um ), wi + (C(um ; um ), wi ) + g Ψε (um ), wi
dt
m ), wi = f , wi , (4.2.48)
+ δ A(u
gim (0) = 0,
for any i = 1, . . . , m. The local existence of g1m , . . . , gmm on an interval [0, Tm ],

is insured by the Cauchy theorem. So we have a local solution um on [0, Tm ], for
each m. In the next step, we will obtain a priori estimates (independent of m) for um ,
which will entail that actually Tm = T .
Multiply (4.2.48) by gim and sum with respect to i. Using (4.2.35), (4.2.42), and
(4.2.23), one has
1 d

q
um 2L 2 (Ω) + α μ
um
p + g (ψε (D2 (um )) D(um ), D(um ))
2 dt
+ δ βum qH s (Ω) ≤ f , um .
Since by hypothesis H.3 and Remark 3.3.5,

(ψε (D2 (um )) D(um ), D(um )) = (ψε ◦ D2 ) (um ), um ≥ 0,
by integrating in time, it is immediate that
um is in a bounded set of L ∞ (0, T ; H ) ∩ L q (0, T ; W p ),

1 (4.2.49)
δ q um is in a bounded set of L q (0, T ; Vs ),
with bounds independent of ε, δ, and m.

One also has, thanks to the definitions (4.2.18) and (4.2.13) of c and C, and Green’s
formula (4.2.23),

c(um ; um , wi ) = (C(um , um ), wi ) = − [∇ wi ]um · um d x,
Ω
so that, in virtue of (4.2.49),
C(um , um ) is in a bounded set of L ∞ (0, T ; H −s (Ω)d ) and L ∞ (0, T ; (Vs ) ).
Due to (4.2.32) and the fact that s > 1 + d/2,

A(um ) is in a bounded set of L q (0, T ; W −1, p (Ω)d ) ⊂ L q (0, T ; H −s (Ω)d ),

A(um ) is in a bounded set of L q (0, T ; (Vs ) ).
imply
Similarly, hypotheses (4.2.42) on A
m ) is in a bounded set of L q (0, T ; H −s (Ω)d ),

A(u
while hypothesis H.5 on ψε entails that

Ψε (um ) is in a bounded set of L q (0, T ; W −1, p (Ω)d ) ⊂ L q (0, T ; H −s (Ω)d ),
and in particular,

Ψε (um ) is in a bounded set of L q (0, T ; (W p ) ) ⊂ L q (0, T ; (Vs ) ).
Then the density of the basis {w j } j≥0 in Vs implies

u m is in bounded set of L q (0, T ; (Vs ) ), (4.2.50)
independently of m, ε, δ, for δ bounded by a fixed constant, eg., δ ≤ 1.
4.2.5 First Passing to the Limit
The above a priori estimates imply that there is a subsequence, say {un }, of the
sequence {um } such that,
lim un = uεδ weakly in L q (0, T ; W p ),

n→∞
lim un = uεδ weakly in L ∞ (0, T ; H ),
n→∞

lim u n = u εδ weakly in L q (0, T ; (Vs ) ).
n→∞
We complete these weak convergences with a strong one, as we are in position to

apply Theorem 3.3.14 due to estimates (4.2.49) and (4.2.50) to get
lim un = uεδ strongly in L q (0, T ; H ). (4.2.51)

n→∞
One also has the convergences

lim A(un ) = χ εδ weakly in L q (0, T ; W −1, p (Ω)d ) and L q (0, T ; (W p ) ),
n→∞

lim Ã(un ) = ξ εδ weakly in L q (0, T ; H −s (Ω)d ) and L q (0, T ; (Vs ) ),
n→∞

lim Ψε (un ) = θ εδ weakly in L q (0, T ; W −1, p (Ω)d ) and L q (0, T ; (W p ) ),
n→∞
1 1
lim δ q un = δ q uεδ weakly in L q (0, T ; Vs ).
n→∞
This last limit means that for each δ, uεδ belongs to Vs . Finally, recalling the
reasoning of the proof of Proposition 4.2.5, we have the following convergence for
any i ∈ IN + :
lim c(un ; un , wi ) = c(uεδ ; uεδ , wi ) for instance in D (]0, T [). (4.2.52)

n→∞
These convergences allow to pass to the limit in (4.2.47) for i ∈ IN + , to get
u εδ , wi + χ εδ , wi + c(uεδ ; uεδ , wi ) + g θ εδ , wi + δ ξ εδ , wi = f , wi ,
uεδ (0) = 0,
and since {w j } j≥0 is complete in Vs , we have for all v ∈ Vs ,
u εδ , v + χ εδ , v + c(uεδ ; uεδ , v) + g θ εδ , v + δ ξ εδ , v = f , v,

(4.2.53)
uεδ (0) = 0.
As noted above, c(uεδ ; uεδ , v) belongs to L ∞ (0, T ). Since the remaining terms

of (4.2.53) belong to L q (0, T ), it follows that (4.2.53) holds in L q (0, T ).
It remains to show that
εδ ).
χ εδ + g θ εδ + δ ξ εδ = A(uεδ ) + g Ψε (uεδ ) + δ A(u (4.2.54)
This is done by the classical Minty procedure (see for instance, Minty [191]). For
the reader’s convenience, we will recall it briefly.
Let
ṽ ∈ L q (0, T ; Vs ),
with

ṽ ∈ L q (0, T ; (Vs ) ), ṽ(0) = 0.
Set
T
Xn = u n − ṽ , un − ṽ + A(un ) − A(ṽ), un − ṽ
0

n ) − A(ṽ),
+ g Ψε (un ) − Ψε (ṽ), un − ṽ + δ A(u un − ṽ dt.
and Ψε (by Proposition 3.3.4),

Observe that because of the monotonicity of A, A
X n ≥ 0.
As δ is fixed and un is in L q (0, T ; Vs ), by using (4.2.53), X n can be rewritten in the

form
T T
Xn = f ,un dt − (u n , ṽ) + ṽ , un − ṽ + A(un ), ṽ
0 0

+ A(ṽ), un − ṽ + g Ψε (un ), ṽ + g Ψε (ṽ), un − ṽ

n ), ṽ + A(ṽ),
+ δ A(u un − ṽ dt,
where we can pass to the limit as n → ∞ due to the convergences (4.2.51)–(4.2.52),

to get
X n → X εδ ,
with
T T
X εδ = f ,uεδ dt − u εδ , ṽ + ṽ , uεδ − ṽ + χ εδ , ṽ
0 0
+ A(ṽ), uεδ − ṽ + g θ εδ , ṽ + g Ψε (ṽ), uεδ − ṽ

+ δ ξ εδ , ṽ + A(ṽ), uεδ − ṽ dt.
Moreover, as X n is positive, X εδ is positive too.

Writing equation (4.2.53) with v = uεδ , we have

T
X εδ = u εδ −ṽ , uεδ − ṽ + χ εδ − A(ṽ), uεδ − ṽ

0

+ g θ εδ − Ψε (ṽ), uεδ − ṽ + δ ξ εδ − A(ṽ), uεδ − ṽ dt.
Take in this identity

ṽ = uεδ − λ
w,
with

w ∈ L q (0, T ; Vs ),
λ > 0, w ∈ L q (0, T ; (Vs ) ),
w(0) = 0.
Then, upon dividing by λ,

T T

0 ≤ X εδ = λ
w ,
w dt + χ εδ − A(uεδ − λw),
w
0
0

+ g θ εδ − Ψε (uεδ − λw), εδ − λ
w + δ ξ εδ − A(u w),
w dt.
Let λ → 0. Thanks to the hemicontinuity of each term

T
0≤ χ εδ + g θ εδ + δ ξ εδ , εδ ),
w − A(uεδ ) + g Ψε (uεδ ) + δ A(u w dt,
0
whence (4.2.54), by virtue of the arbitrariness of

w. Hence the limit function uεδ
satisfies for all v ∈ Vs , a.e. in ]0, T [,

εδ ), v + c(uεδ ; uεδ , v) = f , v.
u εδ , v + A(uεδ ) + g Ψε (uεδ ) + δ A(u
4.2.6 Second Passing to the Limit
Take v ∈ Uad in (4.2.53), all the terms herein making sense. Moreover, due to a priori
estimates, all terms can be integrated from 0 to T since
t → c(u(t); u(t), v(t)) ∈ L 1 (0, T ) for u ∈ L q (0, T ; W p ) ∩ L ∞ (0, T ; H ), v ∈ Uad .
This is an immediate consequence of the properties of u and v and of the inequality

T T

|c(um ; um , v)| dt = [∇ v]um · um d x dt
0 0 Ω
T

≤ um 2L 2 (Ω)

∇v
L ∞ (Ω) dt.
0
By using (4.2.53) et (4.2.54), it makes sense to write

T
u εδ , v + A(uεδ ), v + c(uεδ ; uεδ , v) + g Ψε (uεδ ), v
0

T
εδ ), v dt =
+ δ A(u f , v dt, ∀v ∈ Uad . (4.2.55)
0
We will now pass to the limit in this equation as ε and δ tend to zero. Let us set
T
Yεδ = v , v − uεδ + A(uεδ ), v − uεδ
0
+ c(uεδ ; uεδ , v − uεδ ) + g Ψε (v) − g Ψε (uεδ )

εδ ), v − uεδ − f , v − uεδ dt.
+ δ A(u (4.2.56)
Taking into account (4.2.55), we can rewrite Yεδ in the form

T T
Yεδ = v − u εδ , v − uεδ dt + g Ψε (v) − Ψε (uεδ ) − Ψε (uεδ ), v − uεδ dt,
0 0
which satisfies, due to Proposition 3.3.2, Yεδ ≥ 0. Then from (4.2.56) and the antisym-
metry of c, it follows that

T
εδ ), v
v , v − uεδ + A(uεδ ), v + c(uεδ ; uεδ , v) + g Ψε (v) + δ A(u
0

T T
− f , v − uεδ dt ≥ A(uεδ ), uεδ dt + g Ψε (uεδ ) dt, (4.2.57)
0 0
since, by (4.2.42)
εδ ), uεδ ≥ 0.
δ A(u
In virtue of the a priori estimates, one can extract a subsequence still denoted εδ,
lim uεδ = u weakly in L q (0, T ; W p ),

lim uεδ = u weakly in L ∞ (0, T ; H ),

lim u εδ = u weakly in L q (0, T ; (Vs ) ),
lim uεδ = u strongly in L q (0, T ; H ),
and also T
lim δ εδ ), v dt = 0
A(u in IR .
0
With these convergences we can pass to the limit in (4.2.57) to get

T
v , v − u + A(u), v + c(u; u, v) + g Ψ (v) − f , v − u dt
0
T T
≥ lim inf A(uεδ ), uεδ dt + lim inf g Ψε (uεδ ) dt
0 0
T T
≥ A(u), u dt + g Ψ (u) dt,
0 0
where, for the last limit above, we used the monotonicity of the operator A and hypoth-
esis H.4 on the functionals Ψε . So, the final result is the variational inequality
T
v , v − u+A(u), v + c(u; u, v) + g Ψ (v) − g Ψ (u) dt
0
T
≥ f , v − u dt, ∀v ∈ Uad , (4.2.58)
0
and the proof of Theorem 4.2.10 is complete.

Theorem 4.2.10 can be extended to the more general situation where instead of the
operator A, one has the following operator A:

k
A= μi Ai , (4.2.59)
i=1
where, for i = 1, . . . , k, the operators Ai are of the same type as A, and μi are positive
constants. In the examples listed above, this is precisely the case of dilatant fluids.
Hypotheses H.3–H.5 being the same, by simply asking that every Ai be of the form
(4.2.34), we reformulate the corresponding hypotheses H.1 and H.2 as follows:
H.1. The function F in (4.2.2) is of the form

k
F( D(u)) = μi i (u) + gΨ (u),
i=1
where μi ≥ 0 and g ≥ 0,

i (u) = φi (D2 (u)) d x, ∀i ∈ {1, . . . , k}, Ψ (u) = ψ(D2 (u)) d x.
Ω Ω
The functions φi and ψ are convex proper and l.s.c., and ψ is non differentiable.
H.2. For any i ∈ {1, ..., k}, the corresponding function φi is differentiable and such
that there exist qi ≥ pi > 1 and αi > 0 with
qi
pi pi

q
φi (D2 (v)) D(v) : D(v) d x ≥ αi (D2 (v)) 2 dx = αi
v
pi .
i
Ω Ω
Moreover,
1 1
φi (D2 (v)) D(v) ∈ L pi (Ω)d×d , + = 1,
pi pi
qi −1

pi pi

q −1

φ (D2 (v)) D(v)
p ≤ Ci (D2 (v)) 2 dx = Ci
v
pi ,
i L i (Ω) i
Ω
where Ci is a strictly positive constant.

Let us now introduce the set of admissible functions corresponding to this setting,

k
k

Uad = v ; v ∈ L qi (0, T ; Vs ), v ∈ L qi (0, T ; H ), v(0) = 0 .
i=1 i=1
Then, recalling (4.2.59), we have to solve the variational inequality

T
v , v − u + A(u), v − u + c(u; u, v − u) + gΨ (v) − gΨ (u) dt
0
T
≥ f , v − u dt, ∀v ∈ Uad . (4.2.60)
0
The analogue of Theorem 4.2.10 for (4.2.60) is stated as follows:

k

Theorem 4.2.12 Let qi ≥ pi ≥ 2 and f ∈ L qi (0, T ; W −1, pi (Ω)d ). Under
i=1
hypotheses H.1, H.2 and H.3–H.5 above, there exists u satisfying (4.2.60) and such
that
k
u∈ L qi (0, T ; W pi ) ∩ L ∞ (0, T ; H ),
i=1

k
d
u ∈ L qi (0, T ; (Vs ) ) for s > 1 + ,
2
i=1
u(0) = 0.
Let us revert to our original problem (4.2.17). We introduced the weak formulation
(4.2.40) for which we showed, under appropriate hypotheses, the existence of solutions.
We can now ask whether these solutions solve also the stronger formulation (4.2.17).
A positive answer given in the next result, concerns the case of a “true” equation, that
is to say, (4.2.40) with Ψ = 0,

T T
u , v + A(u), v + c(u; u, v) dt = f , vdt, (4.2.61)
0 0
for all v ∈ L q (0, T ; Vs ) with v(0) = 0.

In view of the properties of the operator A and of the functional framework that we
introduced above (see in particular, (4.2.27)), a natural “strong” formulation of (4.2.61),
is the following one:
u , w + A(u), w + c(u; u, w) = f , w, ∀w ∈ W p . (4.2.62)
The existence Theorem 4.2.10 provides a solution of (4.2.61) without any condition
relating p, q, and d. As can be seen below, the price to pay for having solutions of
(4.2.62) is reflected in conditions (4.2.63) from the statement below.
Theorem 4.2.13 Let

f ∈ L q (0, T ; W −1, p (Ω)d ),
and q ≥ p satisfy
3d 2p
p≥ , q≥ . (4.2.63)
d +2 p−1
Under hypotheses H.1 and H.2, there exists u satisfying (4.2.62), and such that
u ∈ L q (0, T ; W p ) ∩ L ∞ (0, T ; H ),

u ∈ L q (0, T ; (W p ) ),
u(0) = 0.
Proof We follow the lines of the proof of Theorem 4.2.10. No regularization is needed
here, so we will apply the Galerkin method with the same basis {w j } as in Subsection
4.2.4, directly to
u , v + A(u), v + c(u; u, v) = f , v, ∀v ∈ Vs .
We get the approximate solutions um for which a priori estimates are straightforward
from Subsection 4.2.4. From Subsection 4.2.5 we have, up to a subsequence, the con-
vergences
lim um = u weakly in L q (0, T ; W p ),

m→∞
lim um = u weakly in L ∞ (0, T ; H ),
m→∞

lim u m = u weakly in L q (0, T ; (Vs ) ),
m→∞
lim um = u strongly in L q (0, T ; H ),
m→∞
as well as the following one, due to the hemicontinuity assumption on A:

lim A(um ) = χ weakly in L q (0, T ; W −1, p (Ω)d ).
m→∞
Passing to the limit as in Subsection 4.2.5, we end up with the equation
u , v + χ , v + c(u; u, v) = f , v, ∀v ∈ Vs ,
(4.2.64)
u(0) = 0.
To go further, we will follow an argument in [162] (when treating a variant of Navier–

Stokes equations).
Let us show first that the form c(u; v, w) is continuous on W 1, p (Ω)d × W 1, p (Ω)d ×
W p (Ω)d . For the proof observe that, due to Sobolev’s imbeddings (see for more
1,
details, Section 3.1.1),

W 1, p (Ω) ⊂ L ρ (Ω),
with
1 1 1 1 1
= − − > 0,
if
ρ p d p d
(4.2.65)
1 1
ρ arbitrary if − ≤ 0.
p d
Therefore, the continuity of c follows from the estimate
|c(u; v, w)| ≤ u L ρ (Ω) ∇v L p (Ω) w L ρ (Ω) ,
which holds if
2 1
+ ≤ 1.
ρ p
It is easily seen that this is true if the first part of (4.2.63) is verified.
Consequently, (4.2.64) can be rewritten in the form
u , v = F (u), v, (4.2.66)
with
|F (u), v| ≤ C vW 1, p (Ω) .
This allows us to say that,
Equation (4.2.64) holds true by continuity, ∀v ∈ W p , (4.2.67)
so that,
u , v + χ, v + c(u; u, v) = f , v, ∀v ∈ W p ,
(4.2.68)
u(0) = 0.

The next step is to take v ∈ L q (0, T ; W p ) L ∞ (0, T ; H ) as test function in
(4.2.68), all the terms herein making sense, and afterwards integrate in time from 0 to
T . This is possible if we show that
t −→ c(u(t); u(t), v(t)) ∈ L 1 (0, T ),

(4.2.69)
for u, v ∈ L q (0, T ; W p ) ∩ L ∞ (0, T ; H ).
Here is the only place in the proof where both parts of hypothesis (4.2.63) will intervene.
In fact, the zero divergence condition will not be used and W p can be replaced by
W 1, p (Ω)d .
Recall that W 1, p (Ω) ⊂ L ρ (Ω) with ρ defined in (4.2.65). The case p ≥ d being
obvious, suppose that p < d. Then
L q (0, T ; W 1, p (Ω)) ⊂ L q (0, T ; L ρ (Ω)d ).
(i) If q ≤ ρ, L q (0, T ; L ρ (Ω)d ) ⊂ L q (0, T ; L q (Ω)d ), and c(u; u, v) ∈ L 1 (0, T ) if
1 2
+ ≤ 1,
p q
which requires that
2p dp 3d
≤q≤ρ= hence p≥ .
p−1 d−p d +2
(ii) If q > ρ, L q (0, T ; L ρ (Ω)d ) ⊂ L ρ (0, T ; L ρ (Ω)d ). Therefore, one must have
1 2
+ ≤ 1,
p ρ
that is to say
3d dp
p≥ and q≤ρ= ,
d +2 d−p
which are nothing else than hypotheses (4.2.63).

Therefore, from (4.2.68), we have

T T

u , v + χ , v + c(u; u, v) dt = f , v dt, ∀v ∈ L q (0, T ; W p ).
0 0
Now we can apply the Minty procedure to show that
χ = A(u).
The proof is skipped because the argument is the same as in Subsection 4.2.5.
The result is that u solves
T T
u , v + A(u), v + c(u; u, v) dt = f , v dt, ∀v ∈ L q (0, T ; W p ),
0 0
u(0) = 0.
The fact that this equation is equivalent to (4.2.62) follows by standard arguments
in the theory of variational problems (see, for instance [91, 162]).
Remark 4.2.14 The essential step in the above proof is the extension by continuity
(4.2.67). Let us point out that this procedure cannot apply to the variational inequality
(4.2.40). This is obvious when examining the equality (4.2.53) in the proof of Theorem
4.2.10. Indeed, because of the presence of too many terms (not continuous or not
designed to be continuous in W p ), one cannot write (4.2.53) in the form (4.2.66).
kIn a similar way, we can also consider the general case of the operator A =
i=1 μi Ai . Theorem 4.2.12 above gives the existence of solutions for the weak for-
mulation (4.2.60) corresponding to this A. The strong formulation in this case is
u , w + A(u), w + c(u; u, w) = f , w, ∀w ∈ W p . (4.2.70)
The counterpart of Theorem 4.2.13 is the following result whose proof is straight-
forward:

k

Theorem 4.2.15 Let q ≥ p and f ∈ L qi (0, T ; W −1, pi (Ω)d ). Under hypotheses
i=1
H.1 and H.2, there exists u satisfying (4.2.70) such that

k
u∈ L qi (0, T ; W pi ) ∩ L ∞ (0, T ; H ),
i=1

k
d
u ∈ L qi (0, T ; (Vs ) ) for s > 1 + ,
2
i=1
u(0) = 0,
with
3d 2 pi
pi ≥ , qi ≥ , ∀i ∈ {1, . . . , k}. (4.2.71)
d +2 pi − 1
One may ask what happens if pi = qi for all i ∈ {1, . . . , k}, in hypotheses H.1 and
H.2. As a matter of fact, in this case we have a much “better” result (in the sense of no
constraints on d and on the exponents ( pi )).
Theorem 4.2.16 The existence result stated in Theorem 4.2.15 holds true for any d ≥ 2
and for
pi = qi ≥ 2, ∀i ∈ {1, . . . , k}. (4.2.72)
Proof We have seen in Theorem 4.2.13 that relations (4.2.71) were obtained in order
to guarantee (4.2.69). Let us go back to the term c(u; u, v); we have to show that it is
in L 1 (0, T ).
Now assume that pi = qi for all i ∈ {1, . . . , k}. We start by weakening the con-
straints of Theorem 4.2.13. By interpolation
u ∈ L qi (0, T ; W pi ) ∩ L ∞ (0, T ; H ) ⊂ L qi (0, T ; L γ (Ω)d ) ∩ L ∞ (0, T ; H )

⊂ L ρ (0, T ; L σ (Ω)d ),
where for all i in {1, . . . , k},
1 1 1
= − ,
γ pi d
(4.2.73)
1 θ 1−θ 1 θ 1−θ
= + and = + ,
ρ ∞ qi σ 2 γ
and of course, the first equality holds for pi < d, γ is arbitrary if pi = d, and γ is
infinite if pi > d.
On the other hand, as v ∈ L pi (0, T ; Vs ), one knows by (4.2.29) that Di v ∈
L ∞ (Ω)d×d , and consequently v ∈ L pi (0, T ; W α ) for any α > 0. We are now able to
evaluate the L 1 -norm of c(u; u, v).
To do so, observe first that
T T
|c(u; u, v))| dt ≤ u2L σ (Ω) ∇v L σ (Ω) dt,
0 0
with
1 2

+ = 1.
σ σ

But as seen before, v ∈ L pi (0, T ; W σ ). Thus, by Hölder’s inequality
2 1
T T

ρ T

β
|c(u; u, v)| dt ≤
u
σ dt
ρ

v
dt β ,
L (Ω) Wσ
0 0 0
with
1 2 ρ
+ =1 so that β = .
β ρ ρ−2
Taking β = pi for i fixed in {1, . . . , k}, yields
2 pi
ρ= .
pi − 1
Going back to (4.2.73), we finally have
pi − 1 1−θ
= ,
2 pi pi
which implies that

3 − pi
θ= .
2
Since θ ∈]0, 1[, in order for (4.2.69) to hold, one does have
2 ≤ pi < 3 for any d, (4.2.74)
and this has to be true for every i in {1, . . . , k} (by virtue of the arbitrariness of i).
Let us now apply Theorem 4.2.15 in the particular case pi = qi . The second inequal-
ity in (4.2.71) becomes
2 pi
pi ≥ , i ∈ {1, . . . , k},
pi − 1
whence
pi ≥ 3 i ∈ {1, . . . , k}.
Consequently, the existence of a solution is insured for
3d
pi = qi ≥ Max 3, = 3 for any d ≥ 2 and i ∈ {1, . . . , k}.
d +2
This together with (4.2.74), give the claimed result.
We end this section with a remark concerning the case of general initial conditions
u0 ≡ 0.
Remark 4.2.17 In this section we considered zero initial condition. In the case of an
equation, i.e., when Ψ ≡ 0, it is easily seen that the results we proved above are
still valid with a nonzero initial condition. Indeed, consider equation (4.2.62) with the
initial condition u(0) = u0 ∈ H . As formerly, we again apply the Galerkin method but
with a nonzero initial condition in the approximate equation corresponding to problem
(4.2.47),
m ), wi = f , wi ,
(u m , wi ) + A(um ), wi + c(um ; um , wi ) + δ A(u 1 ≤ i ≤ m,

m
um (0) = Pm (u0 ) = (u0 , wk ) L 2 (Ω) ,
k=1
where Pm u0 is the projection of u0 on Vm . This leads to the following system of ordinary

differential equations:
dgim
m ), wi = f , wi , 1 ≤ i ≤ m,
+ A(um ), wi + (C(um ; um ), wi ) + δ A(u
dt
gim (0) = (u0 , wi ) L 2 (Ω) , 1 ≤ i ≤ m.
Recalling that (see (4.2.45)),
lim Pm u0 = u0 strongly in H,
m→∞
it is obvious that nothing changes in the proof of the existence theorems given above.
The situation is much more complicated in the case of variational inequalities
because of their “very weak” formulation (4.2.40) which was derived by using inequal-
ity (4.2.39). But this inequality holds only under the assumption u(0) = 0. As men-
tioned in Remark 4.2.8, there are examples (essentially in the two-dimensional case),
where one has no need to pass from (4.2.17) to (4.2.40), the main example being the
two-dimensional Bingham fluid which is the subject of Section 4.3 below. For this
particular case, one can take u0 = 0 in Theorem 4.2.10. As far as we know, there is no
result in the general case.
4.2.7 Uniqueness
We now turn to the question of uniqueness. We can only expect partial results when
speaking about it. This is already a hard problem for the Navier–Stokes equations,
solved for d = 2 but still open for d ≥ 3.
One of the main difficulties is the estimation of the nonlinear term c(u; u, v). We
will see further on that with appropriate estimates of this term, in some particular cases
one is able to use the Gronwall inequality in order to show uniqueness of the solution
given by Theorem 4.2.10.
In the two-dimensional case, what is crucial is the Ladyzhenskaya inequality (3.1.23)
which leads to estimate (4.2.25) of c(u; u, v) in Section 4.2.2. Inequality (3.1.23) is not
true for d ≥ 3. In this case, with the use of the Hölder inequality and of (3.1.25), what
is obtained is the following estimate:
Lemma 4.2.18 Let d ≥ 3. Then

2

1+ dr
|c(u; u, v)| ≤ Cv L r (Ω)
u
LS 2 (Ω)
u
H 1 (Ω) , (4.2.75)
0
where
2 d
+ = 1, r > d. (4.2.76)
S r
Proof For ρ such that

1 1 1
= − ,
ρ 2 r
one has
|c(u; u, v)| ≤ v L r (Ω) u H 1 (Ω) u L ρ (Ω) . (4.2.77)
0
We now apply the interpolation inequality (3.1.27) in Ω to each component u i (i =

1, . . . , d) of u with
2d 2
m = ρ, p = 2, q = , α= ,
d −2 S
so, by (4.2.76), 1 − α = dr . We get
2 d
u i L ρ (Ω) ≤ u i LS 2 (Ω) u i Lr 2d/(d−2) (Ω) .
But by Sobolev’s imbeddings H01 (Ω) ⊂ L 2d/(d−2) (Ω), therefore
2 d
u i L ρ (Ω) ≤ C u i LS 2 (Ω) u i Hr 1 (Ω) .
0
Inserting this estimate into (4.2.77) yields (4.2.75).
Remark 4.2.19 Let us point out that Lemma 4.2.18 is true with (4.2.76) replaced by
2 d
+ ≤ 1, r > d.
S r
For more details, we refer the reader for instance, to [162, 265].
It is estimate (4.2.75) that is used in the uniqueness theorem that we will prove below.
Before stating this result, let us mention that in our general setting, other difficulties
have to be treated. One comes from the fact that the partial differential operator is not
linear in general, so one cannot always use a Gronwall type inequality. This difficulty
disappears if among the operators Ai there is −Δ. On the other hand, u and u do not
belong to spaces that are in duality, and therefore one cannot choose u as test function
in the variational formulation. This difficulty is present even in the case when ψ ≡ 0,
which means that formally u solves the equation
u + A(u) + C(u; u) = f − ∇ p in Ω,
div u = 0 in Ω,
(4.2.78)
u = 0 on ∂Ω,
u(0) = 0 in Ω.
The above hypotheses on Ai entail that

k

A(u) ∈ L qi (0, T ; W −1, pi (Ω)d ),
i=1
and from Theorem 4.2.12, it follows that if u is a solution of (4.2.60), then

k

u ∈ L qi (0, T ; (W pi ) ).
i=1
Consequently, we are allowed to take u as test function, and we can state the following
uniqueness theorem:
Theorem 4.2.20 Suppose that ψ = 0 and p1 = q1 = 2. Assume furthermore that
d 2 pi
pi > , qi ≥ , ∀i ∈ {2, ..., k}. (4.2.79)
2 2 pi − d
Then the solution given by Theorem 4.2.10 is unique.
Proof Let u1 and u2 be two solutions of (4.2.78) and set
u∗ = u1 − u2 .
Owing to the antisymmetry of c, it satisfies
(u∗ ) , u∗ + A(u1 ) − A(u2 ), u1 − u2 + c(u∗ ; u1 , u∗ ) = 0.
With the above assumptions on Ai , we obtain
1 d

u∗
2 2

2
L (Ω)
+
u∗
H 1 (Ω) + c(u∗ ; u1 , u∗ ) ≤ 0. (4.2.80)
2 dt 0
Let us now show that in order to have uniqueness it is sufficient that
2 d
u ∈ L S (0, T ; L r (Ω)d ) with + = 1, r > d. (4.2.81)
S r
Suppose for the moment that (4.2.81) holds. Then, recalling (4.2.75), one has the
estimate

2

1+ dr
|c(u∗ ; u1 , u∗ )| ≤ C1
u1 (t)
L r (Ω)
u∗ (t)
LS 2 (Ω)
u∗ (t)
H 1 (Ω)
0

∗
2
∗
2

≤ u (t) H 1 (Ω) + C2 M(t) u (t) L 2 (Ω) ,

(4.2.82)
0
where M(t) = (u1 (t) L r (Ω) ) S . Note that M is in L 1 (0, T ) since u1 ∈ L S (0, T ;
L r (Ω)d ). Then from (4.2.80) we get
d ∗ 2
u L 2 (Ω) ≤ 2 C2 M(t)u∗ (t)2L 2 (Ω) ,
dt
and since u∗ (0) = 0, Gronwall’s inequality yields u∗ ≡ 0, whence uniqueness.

To complete the proof, it remains to check that (4.2.81) holds. Recall the imbeddings
of Theorem 4.2.16,
u ∈ L qi (0, T ; W pi ) ∩ L ∞ (0, T ; H ) ⊂ L qi (0, T ; L γ (Ω)d ) ∩ L ∞ (0, T ; H )

⊂ L ρ (0, T ; L σ (Ω)d ),
with the parameters given by (4.2.73). By combining the equations of (4.2.73), we

obtain 2 d − pi d d
(1 − θ) + − = 1 − ≤ 0, (4.2.83)
qi pi 2 2
or, by changing sign

d 2 d − pi d
(1 − θ) − + = − 1 ≥ 0.
2 qi pi 2
Now, the discussion splits into cases: d ≥ 3 or d = 2. In the first case, the second
part of (4.2.79) implies that
d 2 d − pi d
− + ≥ − 1 > 0,
2 qi pi 2
and we see immediately that (4.2.83) holds with 0 < 1 − θ ≤ 1. In the second case, the
second part of (4.2.79) states simply that qi ≥ pi , the dual exponent of pi . Therefore,

it suffices to consider that u is in L pi (0, T ; L γ (Ω)2 ) with
1 1 1
= − .
γ pi 2
Then we have
2 2
+ = 1,
pi γ
which is precisely (4.2.81) with S = pi , r = γ , and d = 2. The proof of Theorem

4.2.20 is now complete.
4.2.8 Examples and Applications
Newtonian fluid and Navier–Stokes equations

We saw in the first section that a Newtonian fluid is defined by the general constitutive
equation in the case where φ, convex l.s.c. and differentiable, is given by
φ N (s) = 2μs,
and of course, φ N (s) = 2μ. By definition,

N (u) = 2μ D2 (u) d x,
Ω
so that hypothesis H.2 holds with p = q = 2. Therefore, we are in a position to apply

Theorem 4.2.16 to get the existence of a solution u of the Navier–Stokes equations
(4.2.12), and since W 2 = V ,
u ∈ L 2 (0, T ; V ) ∩ L ∞ (0, T ; H ). (4.2.84)
It should be pointed out that in the two-dimensional case, one does not need to introduce
the spaces Vs . We work directly in the space V where the test functions are taken and
the special basis for the Gakerkin method consists of the eigenfunctions related to the
compact inclusion V ⊂ H . Moreover, it is easily seen from the proof of Theorem
4.2.10 applied to problem (4.2.12), that the term C(u, v) determines the space to which
u belongs. The answer is given by Proposition 4.2.4 stating in particular, that
u ∈ L 2 (0, T ; V ) the dual space of L 2 (0, T ; V ) for d = 2,

4
u ∈ L 3 (0, T ; V ) for d = 3.
By Theorem 4.2.20 this solution is unique for d = 2 without any additional condi-
tion since the solution is in the “natural” space L 2 (0, T ; V ) ∩ L ∞ (0, T ; H ) where
Gronwall’s inequality works.
Let us be more precise. Consider the Navier–Stokes equations (4.2.12), i.e.,
u , vV ,V + μ(∇u, ∇v) + c(u; u, v) = f , vV ,V , ∀v ∈ V,

(4.2.85)
u(0) = u0 .
which is the strong formulation (4.2.62) in this case.

Suppose that d = 2 and let u1 and u2 be two solutions corresponding to the same
initial data u0 . Set u∗ = u1 − u2 and test (4.2.85) with u∗ (which is allowed),
(u∗ ) , u∗ V ,V + μ(∇u∗ , ∇u∗ ) + c(u∗ ; u1 , u∗ ) = 0,

u∗ (0) = 0.
Integrating in time (allowed since all the terms are in L 1 (0, T )), yields
1 d

u∗
2 2

2
L (Ω)
+ μ
u∗
H 1 (Ω) ≤ |c(u∗ ; u1 , u∗ )|. (4.2.86)
2 dt 0
Recalling estimate (4.2.25) from Lemma 4.2.3, and applying Young’s inequality, we
can write

2

2

2
|c(u∗ ; u1 , u∗ )| ≤ μ
u∗
H 1 (Ω) + C
u1
H 1 (Ω)
u∗
L 2 (Ω) ,
0 0
where C is a positive constant that depends on μ. Inserting this inequality into (4.2.86)
gives the analogue of (4.2.82),
d ∗ 2
u L 2 (Ω) ≤ 2C M(t) u∗ 2L 2 (Ω) ,
dt
with M(t) = u1 2H 1 (Ω) ∈ L 1 (0, T ), because of (4.2.84). As u∗ (0) = 0, Gronwall’s
0
inequality immediately implies that u∗ ≡ 0.
As mentioned before, the situation is dramatically different for dimensions higher
than two. Inequality (4.2.86) is still valid and one has to treat the term c(u∗ ; u1 , u∗ ).
This was already done in the proof of Theorem 4.2.20 based on Lemma 4.2.18. The
key point in this proof is that uniqueness holds in the space L S (0, T ; [L r (Ω)]d ) with
S and r given by (4.2.81). But, the existence of a solution in this space is not known
(that is one of the “challenging open problems” for Navier–Stokes equations).
In problem (4.2.78) above, we could play with the numbers pi and qi in order to
prove that (4.2.81) is true for appropriate choices of such exponents. If the operator in
(4.2.78) is just −Δ as in the case of Navier–Stokes equations, there is no possibility for
playing this game.
Ladyzhenskaya’s example
A variant of Navier–Stokes equations, proposed by Ladyzhenskaya [155], corre-
sponds to the case of F given by
2
α
F( D(u)) = μ D2 (u)d x + D2 (u) d x , α ≥ 0,
Ω 2 Ω
which means that

F( D(u)) = 1 (u) + 2 (u),
with
2
α
1 (u) = N (u) = μ D2 (u) d x, 2 (u) = D2 (u)d x .
Ω 2 Ω
In this case the equation to be solved (compare with Navier–Stokes equation (4.2.12))
is
∂u
− μ + αu2H 1 (Ω) Δu + C(u, u) = f − ∇ p in Ω×]0, T [,
∂t 0
div u = 0 in Ω×]0, T [, (4.2.87)

u=0 on ∂Ω×]0, T [,
u(0) = u0 in Ω,
recalling that
u2H 1 (Ω) = |∇u|2 d x.
0 Ω
This situation fits in the context of Remark 4.2.7 with 2 given by (4.2.36) with
β = 2. We have here an operator of the form (4.2.59) with two elements, corresponding
respectively to N and to 2 . From (4.2.37) and (4.2.38), it is then easily seen that
hypotheses H.1 and H.2 are satisfied for p1 = q1 = 2 and p2 = 2, q2 = 4. There-
fore, Theorem 4.2.15 can be applied and gives existence of a solution of (4.2.87) in
L 4 (0, T ; V ) ∩ L ∞ (0, T ; H ) for d ≤ 4 as can be checked from (4.2.63).
According to Theorem 4.2.20, since p1 = q1 = 2, uniqueness holds if
d 4
p2 = 2 > , q2 ≥ .
2 4−d
The first inequality is true if d < 4, whereas the second one requires d ≤ 3. Thus, the
solution above is unique for d ≤ 3.
Dilatant fluids
As seen before, the deviatoric tensor of a dilatant fluid is defined by the constitutive
equation
p
2
τi j = 2μ Di j + α D22 Di j with p > 2,
p
corresponding to the case where
φD = φN + ψD ,
with φ N defined by (4.2.11), i.e., φ N (s) = 2 μ s, and
2 p
ψ D (s) = αs2,
p
so that
1 (u) = φ N (D2 (u)) d x = 2μ D2 (u) d x,
Ω Ω

2 p
D (u) = ψ D (D2 (u)) d x = α (D2 (u)) 2 d x.
Ω p Ω
When checking hypotheses H.1 and H.2, we have
p1 = q1 = 2 and p2 = q2 = p.
It is now Theorem 4.2.16 that applies and provides a solution in L p (0, T ; W p ) ∩

L ∞ (0, T ; H ) for p ≥ 2 in any dimension d ≥ 2. By Theorem 4.2.20, due to inequalities
(4.2.79), this solution is unique for d ≤ 2( p − 1).
In the literature, one can also find the following definition:
p
−1
τi j = α D22 Di j .
Theorem 4.2.10 guarantees again the existence of a solution in L p (0, T ; W p ) ∩

L ∞ (0, T ; H ) if p ≥ 2.
4.2.9 What About the Pressure?
The existence theorems from the preceding section were only concerned with the veloc-
ity field u, assumed to be divergence-free. In the variational formulations we studied,
the pressure, present as an unknown in the general constitutive equations for incom-
pressible fluids, disappears by integration by parts because of the condition div u = 0
and of the no-slip condition u = 0 on ∂Ω. The question is : how does one recover the
pressure?
This is a delicate and hard question, that has not yet been completely answered.

Whatever the situation, the data f has to be taken in L q (0, T ; W −1, p (Ω)d ) instead of

L q (0, T ; (W p ) ) because otherwise, there is no way to recover the pressure, as already
pointed out in Remark 4.2.11. This will be made clear in the derivation below.
Let us discuss first this point for the Navier–Stokes equations (4.2.85). We have seen
at the beginning of Section 4.2.8 that this system admits a solution u in L 2 (0, T ; V ) ∩
L ∞ (0, T ; H ) with u in L r (0, T ; V ) where r = 43 when d = 3 and r = 2 when
d = 2, in which case the solution is unique. When d = 3, for each v ∈ L 2 (Ω)d , the
mapping t → (u(t), v) belongs to L ∞ (0, T ), while when d = 2, since u belongs to
C 0 ([0, T ]; H ) according to Theorem 3.3.16, this mapping belongs to C 0 ([0, T ]). Now
integrating (4.2.85) in time over the interval ]0, t[, for 0 < t ≤ T , using the fact that
(see for instance Simon [253]),
d
u , vV ,V = (u, v),
dt
and integrating by parts in time we obtain for all v ∈ V , considering that u(0) = 0
t
(u(t), v)+ {μ(∇ u(s), ∇ v)+(C(u(s), u(s)), v)− f (s), v H −1 (Ω)d ,H 1 (Ω)d }ds = 0.
0
0
Next, for all v ∈ H01 (Ω)d and a.e. in ]0, T [, we define the mapping
t
L(v, t) = (u(t), v) + {μ(∇ u(s), ∇ v) + (C(u(s), u(s)), v)
0
− f (s), v H −1 (Ω)d ,H 1 (Ω)d }ds.
0
Since for each v ∈ H01 (Ω)d , the mapping t → (u(t), v) belongs to C 0 ([0, T ]) when
d = 2 or L ∞ (0, T ) when d = 3, it follows from Proposition 4.2.4 that the mapping
t → L(v, t) has the same properties. Moreover a.e. in ]0, T [, L is a continuous linear
functional on H01 (Ω)d that vanishes on V . In other words, L belongs to P V , the polar
space of V defined in Theorem 3.2.12. Hence Theorems 3.2.12 and 3.2.11 imply that
if Ω is a connected bounded domain that is a finite union of open sets with Lipschitz-
continuous boundaries, then a.e. in ]0, T [ there exists a unique function P(t) in the
space L 2m (Ω) defined by (3.1.18), such that
∀v ∈ H01 (Ω)d , L(v, t) = ∇ P(t), v H −1 (Ω)d ,H 1 (Ω)d ,

0
and the above properties of L imply that P ∈ C 0 ([0, T ]; L 2m (Ω)) when d = 2, or

P ∈ L ∞ (0, T ; L 2m (Ω)) when d = 3. Hence we have for all v ∈ H01 (Ω)d , a.e. in
]0, T [,
t
(u(t), v) + {μ(∇ u(s), ∇ v) + (C(u(s), u(s)), v)
0
− f (s), v H −1 (Ω)d ,H 1 (Ω)d }ds = −(P(t), div v). (4.2.88)
0
By differentiating (4.2.88) in the sense of distributions on ]0, T [, we find that for all v
in H01 (Ω)d ,
d d
(u(t), v) + μ(∇ u(t), ∇ v) + (C(u(t), u(t)), v) + (P(t), div v)
dt dt
= f (t), v H −1 (Ω)d ,H 1 (Ω)d .
0
Thus, setting p(t) = −P (t), the following holds for all v ∈ H01 (Ω)d :
d
(u(t), v) + μ(∇ u(t), ∇ v) + (C(u(t), u(t)), v) − ( p(t), div v)
dt
= f (t), v H −1 (Ω)d ,H 1 (Ω)d .
0
Moreover, since in both cases P belongs to L ∞ (0, T ; L 2m (Ω)), the product

T
(P(t), ϕ (t))dt
0
is well defined for all ϕ ∈ W01,1 (0, T ; L 2m (Ω)). Hence p = −P can be identified with
an element of the dual of W01,1 (0, T ; L 2m (Ω)), i.e., in W −1,∞ (0, T ; L 2m (Ω)). Therefore
the existence theorem for the Navier–Stokes equations should be completed as follows:
there exists a solution (u, p) of (4.2.12) with u satisfying (4.2.84) and u ∈ L r (0, T ; V )
where r = 2 when d = 2, r = 4/3 when d = 3, and p ∈ W −1,∞ (0, T ; L 2m (Ω)).
Let us now turn our attention to equation (4.2.62). In view of the properties of its
solution u, we can rewrite it in the form (see (4.2.85))

u , v (W p ) ,W p + A(u) + C(u, u) − f , v W −1, p (Ω)d ,W 1, p (Ω)d = 0, (4.2.89)
0
for all v ∈ W p . Then proceeding as above, we see that there exists a pressure p∗ in (at
least) D (Ω×]0, T [).
In the case of a variational inequality, the argument used above does not work, since
there is no possibility to write down an identity of the form (4.2.89). We face the same
difficulty as that evoked in Remark 4.2.17 when dealing with nonzero initial conditions,
related to the presence in (4.2.40) of the weak term (v , v − u) with v ∈ Uad . However,
in the case of the Bingham fluid in two dimensions, one can get rid of difficulties
by interpreting the variational inequality as an equation with the help of Lagrange
multipliers. This result is due to Duvaut and Lions [91]. For the reader’s convenience
(and also in order to show the complexity of the topic), let us recall it.
Suppose that d = 2 and to simplify, let p = 2 and q = 2 in hypotheses H.1 and H.2,
which means that the operator A is the Laplacian −μΔ. Moreover, we will consider a
particular functional Ψ , corresponding to
√
ψ(s) = 2g s, g ≥ 0. (4.2.90)
Therefore
1
Ψ (u) = 2g j (u), where j (v) = (D2 (v)) 2 d x. (4.2.91)
Ω
Obviously, the functional Ψ is non G-differentiable. The variational inequality to

be solved is
u , v − uV ,V + μ(∇u, ∇(v − u)) + c(u; u, v − u) + g j (v) − g j (u)

≥ f , v − u, ∀v ∈ V, μ > 0, (4.2.92)
with
div u = 0 in Ω×]0, T [,
u = 0 on ∂Ω×]0, T [,
u(0) = u0 in Ω.
Notice that we take here a nonzero initial condition, this is possible since we are in the
situation mentioned in Remark 4.2.17.
Remark 4.2.21 The variational inequality (4.2.92) models the flow of the Bingham
fluid which will be studied in details in the next section.
By a variant of Theorem 4.2.10, we have existence and uniqueness of a solution u
in L 2 (0, T ; V ) ∩ L ∞ (0, T ; H ) with u ∈ L 2 (0, T ; V ) (for more details, see Section
4.3.1 below).
Theorem 4.2.22 Assume that Ω is a connected bounded domain in IR 2 that is a finite
union of open sets with Lipschitz-continuous
boundaries. Then, under the hypotheses
of Theorem 4.2.10, there exist M = m i j 1≤i, j≤2 and p ∈ D (Ω×]0, T [) such that for
i, j ∈ {1, 2},
(i) m i j ∈ L ∞ (Ω×]0, T [), m i j = m ji , m 11 + m 22 = 0,

(ii) M : M ≤ 1 a.e. on Ω×]0, T [,

(iii) M : D(u) = D(u) : D(u),
∂u i ∂u i √ ∂m i j ∂p
(iv) − μΔ u i + u j −g 2 = fi − in D (Ω×]0, T [). (4.2.93)
∂t ∂x j ∂x j ∂ xi
Proof First of all, observe that inequality (4.2.92) can be written in the form

F(t), vV ,V + g j (v) − F(t), uV ,V + g j (u) ≥ 0, ∀v ∈ V, (4.2.94)
where
F = u − μΔu + C(u, u) − f .
Taking in (4.2.94) v = ±λw, for λ ≥ 0 and w ∈ V and using the fact that j is
positively homogeneous of degree 1 (easy to check), yields

λ ± F(t), wV ,V + g j (w) − F(t), uV ,V + g j (u) ≥ 0,
and if w ∈ L 2 (0, T ; V ), we integrate in time to get
T T T
λ ± F(t), wV ,V dt + g j (w) dt − F(t), uV ,V + g j (u) dt ≥ 0.
0 0 0
An immediate consequence of this inequality is the following one:

T T

F(t), wV ,V dt ≤ g j (w) dt. (4.2.95)
0 0
Also, by choosing successively v = 0, v = 2u and using again the positive homogeneity

of j, yield
T

F(t), uV ,V + g j (u) dt = 0. (4.2.96)
0
Let now introduce the space Z defined as

Z = Q = qi j 1≤i, j≤2 , qi j = q ji , qi j ∈ L 1 (Ω×]0, T [) ,
endowed with the norm

1
QZ = Q: Q 2 d xdt.
Ω×]0,T [
Denote by F the map
w ∈ L 2 (0, T ; H01 (Ω)2 ) −→ D(w) ∈ Z .
In this setting, inequality (4.2.95) reads

T √

F(t), wV ,V dt ≤ g 2F (w)Z .
0
We can apply the Hahn–Banach theorem to state that there exists some M in Z such
that m 11 + m 22 = 0 (as D11 (w) + D22 (w) = 0) and satisfying for all w ∈ V ,
T T
F(t), wV ,V dt = g M : D(w) d xdt, (4.2.97)
0 0 Ω
with
MZ ≤ 1. (4.2.98)
Notice that, as (4.2.97) is an equality, it defines a pressure as in the Navier–Stokes

system, and then it is nothing else than (4.2.93)(iv), while (4.2.98) corresponds to
(4.2.93)(ii). On the other hand, (4.2.96) together with (4.2.97) give
T T
M : D(u) dt = D(u) : D(u) dt,
0 0
which is equivalent to (4.2.93)(iii), and this ends the proof.
Remark 4.2.23 Via some technicalities, one can extend Theorem 4.2.22 to the case of
a general operator A instead of −Δ, which entails the replacement of V by the space
W p in the statement of the result.
4.3 Model Example: The Bingham Fluid
4.3.1 Definition of a Bingham Fluid
The Bingham fluid (see Section 2.9) was introduced in 1922 by Bingham [37] in order
to describe what he called viscoplastic fluids. He observed that some paints have an
unusual behavior, they behave as rigid bodies until the stress reaches a certain value,
called yield stress, but flow afterwards as viscous fluids.
The mathematical study of the Bingham fluid started with the book of Duvaut and
Lions [91] where it was shown that this fluid is characterized by a variational inequality.
We will follow along the lines of this book for the derivation of the specific constitutive
equation of this fluid. To do so, we have to go back to Section 2.3.3, more precisely
to the energy balance (2.3.43). Let us focus on the term T : L, usually called rate of
dissipation, where as usual, the scalar product of two tensors is denoted by a colon.
Following Prager [215], we require that this function (which by symmetry equals T : D)
depends only on the tensor D, i.e.,
T : D = D1 ( D) + D2 ( D), (4.3.1)
where D1 and D2 , are positively homogeneous functions, of order, respectively, 1 and

2, with respect to the components of D. This means that
∂ D1 1 ∂ D2
D1 = Dkl , D2 = Dkl .
∂ Dkl 2 ∂ Dkl
Taking into account that div u = 0, it follows that
∂ D1 1 ∂ D2
Ti j = − pδi j + + , (4.3.2)
∂ Di j 2 ∂ Di j
where p is the (scalar) pressure. Following again Prager, a Bingham fluid is defined by

D1 = 2g D2 , D2 = 4μD2 ,
where g ≥ 0 is the yield stress and μ ≥ 0 is the viscosity. Then (4.3.2) reads
D
τ = 2μ D + g √ . (4.3.3)
D2
Obviously, if D2 = 0, T is undetermined.
At this point, and continuing the discussion about the Bingham fluid from Section
2.9, let us quote from Rajagopal and Srinivasa [236]: “For such a material, it is not
possible to write the stress as a function of the velocity gradient since for this material,
the stress does not take a specific value when D is zero. However, the Bingham fluid is
a perfect example of a model where it is more natural to consider the symmetric part of
4.3 Model Example: The Bingham Fluid 159
the velocity gradient as a function of the stress.” What was shown in [236] is that the
indetermination of the stress can be resolved by the use of the rate of dissipation of the
fluid.
It is precisely what we will do next, invert the relation (4.3.3) by using the rate of
dissipation.
First, let us give the expression of τ2 , the second principal invariant of τ . A simple
computation in (4.3.3) yields
1 2
τ2 = τ : τ = 2μ D2 + g , (4.3.4)
2
which immediately shows that

√
τ2 ≥ g.
Consequently, we are allowed to invert (4.3.3) to get
1 g
Di j = 1− √ τi j ,
2μ τ2
which makes sense.

What about the indeterminacy of T when D2 = 0 ? If D2 = 0, even if T is
undetermined, from (4.3.4) we infer that necessarily τ2 ≤ g. Indeed, τ2 > g, by (4.3.4)
would imply D2 > 0.
To conclude, we are now able to give the constitutive equation for a Bingham fluid
√
τ2 < g ⇐⇒ D = 0,
√ 1 g
τ2 ≥ g ⇐⇒ Di j = 1− √ τi j ,
2μ τ2
or equivalently,
√
τ2 ≤ g ⇐⇒ D = 0,
√ D (4.3.5)
τ2 > g ⇐⇒ T = − p I + 2μ D + g √ .
D2
If g = 0, this equation is nothing else than the one defining the Newtonian fluid
with viscosity μ. As remarked above, D = 0 means that the material behaves like a
rigid body, but flows like a fluid as the stress is bigger than the yield.
Let us consider the constitutive relation (4.3.5). Because of the last term in (4.3.5),
one cannot really say that this constitutive equation is that of a Reiner-type fluid. How-
ever, it turns out that it is a perfect model for the general relation. From (4.3.5), we see
that the functional φ B for the Bingham fluid is defined by
φ B = φ N + ψ,
where φ N is the Newtonian functional introduced in (4.2.11) and ψ is given by

√
ψ(s) = 2g s.
Hence,
φ B (D2 (u)) d x = 2μ D2 (u) d x + 2g j (u),
Ω Ω
where j, already defined by (4.2.91) in Section 4.2.9, is

1
j (v) = (D2 (v)) 2 d x.
Ω
It is easily seen that j is convex and continuous on H01 (Ω)d (hence weakly l.s.c.
for the weak topology of H01 (Ω)d ). The continuity is obvious; regarding the convexity,
one has to use the following inequality (easy to prove):
1 1
D(u) : D(v) ≤ 2(D2 (u)) 2 (D2 (v)) 2 .
In contrast to the preceding examples, the functional φ B is not differentiable because

of the presence of ψ. Therefore, for this fluid we have to solve the variational inequality
u , v − u + μ(∇u, ∇(v − u)) + c(u; u, v − u)

+ g j (v) − j (u) ≥ f , v − u, ∀v ∈ V. (4.3.6)
In order to satisfy hypotheses H.3–H.5, j is approximated by

2 1+ε
jε (v) = (D2 (v)) 2 d x, ε > 0.
1+ε Ω
The first regularization leads to the equation
u ε , v + μ(∇uε , ∇v) + c(uε ; uε , v) + g jε (uε ), v = f , v, ∀v ∈ V,
where
ε−1
jε (uε ), v = (D2 (uε )) 2 D(uε ) : D(v) d x.
Ω
As in a Newtonian fluid, one has p1 = q1 = 2 and so, when d = 2, Theorems 4.2.12

and a variant of 4.2.20 give existence and uniqueness of the solution in L 2 (0, T ; V ) ∩
L ∞ (0, T ; H ) (which are also the spaces for the solution of Navier–Stokes equations).
Following along the lines of the proof of Theorem 4.2.10, it is easily seen that one
obtains u ∈ L 2 (0, T ; V ), therefore the second regularization is not needed and what
we solve in this case is the strong formulation (4.3.6).
When d = 3, the variational inequality to be solved is the corresponding weak

formulation of (4.2.40), i.e.,
T
(v , v − u) + μ(∇ u, ∇(v − u)) + c(u; u, v − u) + g j (v) − j (u) dt
0
T
≥ f , v − udt, ∀v ∈ Uad , (4.3.7)
0
where
3
Uad = v ; v ∈ L q (0, T ; Vs ), v ∈ L q (0, T ; H ), v(0) = 0, s = .
2
We are in the same situation as that of the Navier–Stokes equations: existence (but
not uniqueness) in L 2 (0, T ; V ) ∩ L ∞ (0, T ; H ) and uniqueness (but not existence) in
the space L S (0, T ; L r (Ω)3 ) with
2 3
+ = 1, r > 3.
S r
4.3.2 Stress Threshold
In the sequel we suppose that d = 2. Now, we define a “stress threshold”. To do that,

the following estimates of j are essential:
K Ω v L 2 (Ω) ≤ j (v) ≤ CΩ v H 1 (Ω) , (4.3.8)

0
where the constants K Ω and CΩ depend only on Ω. The lower bound is due to Nirenberg
1
and Strauss (see for example, [91]), the upper bound is obvious with CΩ = |Ω| 2 , where
|Ω| denotes the measure of the set Ω. Notice that these inequalities imply the following
one:
CΩ
S0,2 ≤ , (4.3.9)
KΩ
with S0,2 the Poincaré constant of the domain Ω (see (3.1.5)).
Proposition 4.3.1 Let f ∈ L ∞ (0, +∞; L 2 (Ω)d ) and denote by f ∞ its norm, i.e.,

f ∞ = inf C ; f (t) L 2 (Ω) ≤ C a.e. on ]0, +∞[ .
If
f ∞
g > g1 = ,
KΩ
there exists t0 ∈ [0, +∞[ such that the solution of (4.3.6) satisfies
u(t) ≡ 0 for t ≥ t0 .
Proof Taking v = 0 in (4.3.6) gives
1 d
u2L 2 (Ω) + μu2H 1 (Ω) + g j (u) ≤ f ∞ u L 2 (Ω) . (4.3.10)
2 dt 0
Dividing by the L 2 norm of u, and applying the first part of (4.3.8), we get
d
u L 2 (Ω) + μλ21 u L 2 (Ω) ≤ f ∞ − gK Ω , (4.3.11)
dt
where to simplify, we set

1
λ1 = . (4.3.12)
S0,2
Then by Gronwall’s inequality, for 0 ≤ t0 < t < +∞,
f ∞ − gK Ω 2
u(t) L 2 (Ω) ≤ exp−μ λ1 t u(t0 ) L 2 (Ω) + 1 − exp−μ λ1 t . (4.3.13)
2
μ λ1
2
Suppose first that there is a t0 > 0 such that u(t0 ) = 0. By hypothesis f ∞ −

gK Ω < 0. Then using (4.3.13), u(t) = 0 for any t > t0 and the proposition is true.
Suppose now that for any t ≥ 0, u(t) = 0. Observe that (4.3.13) implies that
u(t) L 2 (Ω) ≤ C1 exp−μ λ1 t −C2 ,

2
where C1 and C2 are positive constants, which contradicts the assumption that u(t) = 0
for any t.
Proposition 4.3.2 Let f ∈ L ∞ (0, +∞; L 2 (Ω)d ). If there exists some t0 < +∞ such
that u(t) = 0 for all t ≥ t0 , then
S0,2
g ≥ g2 = f ∞ . (4.3.14)
CΩ
Proof Observe that by assumption and (4.3.6),
g j (v) ≥ ( f (t), v) ∀v ∈ V, ∀t ≥ t0 .
Then the second inequality in (4.3.8) gives on the one hand
|( f (t), v)| ≤ g CΩ v H 1 (Ω) . (4.3.15)

0
On the other hand, the sharpest estimate of |( f (t), v)| is obtained by using the
Poincaré inequality as follows:
|( f (t), v)| ≤ f ∞ v L 2 (Ω) ≤ S0,2 f ∞ v H 1 (Ω) . (4.3.16)

0
This, together with (4.3.15) yield
S0,2 f ∞ ≤ g CΩ ,
which gives (4.3.14)

Remark 4.3.3 Notice that owing to (4.3.9),
g2 ≤ g1 .

This remark, as well as Propositions 4.3.1 and 4.3.2, justify the following definition:
Definition 4.3.4 The quantity defined as
f ∞
gc = , (4.3.17)
KΩ
is called flow (or stress) threshold of problem (4.3.6).

Finally we recall some other results concerning the behavior of the solution of
(4.3.6) with respect to the yield g. A first one, given in Duvaut and Lions [91], states
the continuous dependence on g as follows:
Proposition 4.3.5 If u g1 and u g2 are the solutions of (4.3.6) with respectively, g1 ∈
[0, g0 ] and g2 ∈ [0, g0 ] for some g0 finite, then there is a constant C(g0 ) such that
u g1 − u g2 L 2 (0,T ;V ) + u g1 − u g2 L ∞ (0,T ;H ) ≤ C(g0 )|g1 − g2 |.
One may ask what happens when g is small, or when g is very large. It is interesting
to observe that the behavior of a Bingham fluid is highly influenced by the presence of
the yield stress g. In [91] two mathematical results deal with the limits as g → +∞
and g → 0. For the reader’s convenience, we will recall here these results and sketch
their proofs.
For g > 0, it was observed experimentally that in the fluid there are zones of rigidity
(i.e., zones where D = 0), their size increases as g increases and can even block the
motion. From a mathematical point of view, this is still a conjecture. However, the next
proposition [91] treating the case of g → +∞, gets closer to the answer:
Proposition 4.3.6 Let u g be the solution of (4.3.6) with g > 0. Then
lim u g = 0 weakly in L 2 (0, T ; V ).

g→+∞
Proof Recall inequality (4.3.10) obtained in the proof of Proposition 4.3.1 by testing
(4.3.6) with v = 0,
1 d
u2L 2 (Ω) + μu2H 1 (Ω) + g j (u) ≤ f ∞ u L 2 (Ω) .
2 dt 0
Integrating it in time, gives

T

2 T 1

2
μ
u g
dt+g j (u g ) dt +
u g (T )
L 2 (Ω)
H 1 (Ω)
0 2
0 0
T
1

2
≤ f ∞ u g L 2 (Ω) dt +
u(0)
L 2 (Ω) , (4.3.18)
0 2
which implies
T C
j (u g ) dt ≤ . (4.3.19)
0 g
In addition, (4.3.10) yields
1 d 1
u g 2L 2 (Ω) + μu g 2H 1 (Ω) ≤ f ∞ u g H 1 (Ω) .
2 dt 0 λ1 0
Integrating in time as above implies

2

ug
2 ≤ C,
L (0,T ;V )
with a constant independent of g. Consequently, up to a subsequence,
lim u g = U weakly in L 2 (0, T ; V ).

g→+∞
T
Then, by the l.s.c. of the functional v −→ 0 j (v) dt,
T T
lim inf j (u g ) dt ≥ j (U) dt,
g→∞ 0 0
which together with (4.3.19) imply

T
j (U) dt = 0,
0
so j (U) = 0 a.e. in time. Thus, U ≡ 0.

Remark 4.3.7 Proposition 4.3.6 says that at the limit the fluid behaves like a rigid
body.
We now turn to the question of g → 0. We assume that d = 2.
Proposition 4.3.8 Let u g be the solution of (4.3.6) with g > 0, and u the solution
of the Navier–Stokes equations (4.2.12) (i.e., (4.3.6) with g = 0) with the same initial
value u0 . Then,
lim u g = u weakly in L 2 (0, T ; V ),

g→0
lim u g = u weakly in L 2 (0, T ; V ).

g→0
Proof As in Section 4.2.3, following along the lines of the procedures used to establish
a priori estimates and using also (4.3.10), one has for g > 0 and bounded,
u g L 2 (0,T ;V ) ≤ C, u g L 2 (0,T ;V ) ≤ C, u g L ∞ (0,T ;H ) ≤ C,
independently of g. Thus, we have the convergences
lim u g = w weakly in L 2 (0, T ; V ), (4.3.20)

g→0
lim u g = w weakly in L 2 (0, T ; V ), (4.3.21)

g→0
lim u g = w weakly in L ∞ (0, T ; H ). (4.3.22)
g→0
Moreover, due to Theorem 3.3.14,
lim u g = w strongly in L 2 (0, T ; L p (Ω)2 ),

g→0
for any finite p, since d = 2.

Testing (4.3.6) with v ∈ L 2 (0, T ; V ) and integrating in time from 0 to T gives,
T
u g , v + μ(∇u g , ∇v) − c(u g ; v, u g ) + g j (v) − j (u g ) − f , v − u g dt
0

1

u g (T )
2 2 1

2 T

2
≥ L (Ω)
−
u0
L 2 (Ω) + μ
u g
H01 (Ω)
dt.
2 2 0
The above convergences are sufficient to pass to the limit in this inequality. For the term
c(u g ; u g , v), observe that (4.3.20)–(4.3.22) are exactly what is required for applying
Proposition 4.2.5. Consequently,
lim c(u g ; u g , v) = c(u; u, v) in L 1 (0, T ).

g→0
Since (see the examples at the end of Section (3.3.1))

2

2 T

2 T

2
lim inf
u g
L 2 (Ω) ≥
w
L 2 (Ω) and lim inf
ug
H 1 (Ω)
dt ≥
w
H01 (Ω)
dt,
g→0 g→0 0 0 0
we have at the limit

T
w , v + μ(∇w, ∇v) + c(w; w, v) − f , v − w dt
0

1

w(T )
2 2 1

u0
2 2
T

2
≥ L (Ω)
− L (Ω)
+ μ
w
H 1 (Ω) dt,
2 2 0 0
which reads

T T
w , v − w + μ(∇w, ∇(v − w)) + c(w; w, v) dt ≥ f , v − w dt.
0 0
Both w and v belong to L 2 (0, T ; V ); therefore it is readily seen that a.e. for t ∈]0, T [,
w (t), v − w(t) + μ(∇w(t), ∇(v − w(t))) + c(w(t); w(t), v) ≥ f (t), v − w(t).
Taking v = w(t) ± h with h ∈ V , shows that w solves the Navier–Stokes equations,

and by uniqueness (we are in dimension 2!) w = u.
We end this section with a result comparing the solution of the variational inequality
(4.3.6) with the solution of the Navier–Stokes equations.
Proposition 4.3.9 Let u g be the solution of (4.3.6) with g > 0, and u the solution of
Navier–Stokes equations (4.2.12) with the same initial value u0 . Then at least one of
the two assertions is satisfied:
(i) u g (t) L 2 (Ω) ≤ u(t) L 2 (Ω) a.e. for t > 0.

(ii) There exists a set S of points ti with u g (ti ) L 2 (Ω) = u(ti ) L 2 (Ω) and then
ti ti

μu g 2H 1 (Ω) − f , u g dt ≤ μu2H 1 (Ω) − f , u dt, ∀ti ∈ S .
0 0 0 0
Proof Taking into account that j (u) ≥ 0, in a neighborhood of the origin one has
|u g | < |u|, whence assertion (i).
Assertion (ii) follows from the following relations:
1 d

u g
2 2

2
L (Ω)
+ μ
u g
H 1 (Ω) + g j (u g ) ≤ f , u g ,
2 dt 0
1 d

2

u
2

2
L (Ω)
+ μ
u
H 1 (Ω) = f , u,
2 dt 0
obtained by testing (4.3.6) with v = 0 and the Navier–Stokes system (4.2.12) with
v = u.
Integrating form 0 to ti leads to

ti

2 1

2 1

2
μ
u g
H 1 (Ω) − f , u g dt + g j (u g ) ≤ −
u g (ti )
L 2 (Ω) +
u0
L 2 (Ω) ,
0 0 2 2
ti

2 1

2 1

2
μ
u
H 1 (Ω) − f , u dt = −
u(ti )
L 2 (Ω) +
u0
L 2 (Ω) ,
0 0 2 2
whence

ti

2 ti

2
μ
u g
H01 (Ω)
− f , u g dt − μ
u
H 1 (Ω) − f , u dt ≤ 0,
0 0 0
and assertion (ii) follows as we assumed that u g (ti ) L 2 (Ω) = u(ti ) L 2 (Ω) .
Remark 4.3.10 Numerical computations show that it is actually assertion (i) which
seems to be always true (see Begis [22]).
4.4 Fluids with Nonlinear Viscosity Bounded at Infinity
4.4.1 Variational Formulation. Examples
Let us consider again the constitutive equation for generalized Newtonian fluids
T = − p I + μ(s(u)) D, (4.4.1)
with the notation

s(u) = 2D2 (u) = | D(u)|.
We assume that the function μ satisfies the following hypotheses:

H.1. μ is continuously differentiable from IR + to IR + .
H.2. There exists μ∞ > 0 such that
lim μ(z) = μ∞ . (4.4.2)

z→∞
H.3. There exists a set E ⊂ IR + such that
μ (z) ≥ 0, ∀z ∈ E ,
(4.4.3)
z |μ (z)| ≤ μ(z), ∀z ∈ IR + \ E .
Remark 4.4.1 Hypothesis (4.4.3) does not exclude the case E = IR + or E = ∅.

Examples
Eyring–Prandtl fluids
Such fluids are defined by
1
(μ0 − μ∞ ) Log[λz + (λ2 z 2 + 1) 2 ]
μ(z) = μ∞ + ,
λz
where μ0 , μ∞ , and λ are strictly positive constants.

Cross fluids
The constitutive equation for these fluids is
μ0 − μ∞
μ(z) = μ∞ + 2
,
1+ z3
where μ0 and μ∞ are positive constants with μ∞ ≥ μ0 > 0. One has E = IR + .

Williamson fluids
These fluids are defined by
λ(μ0 − μ∞ )
μ(z) = + μ∞ , (4.4.4)
λ+z
where μ0 , μ∞ , and λ are positive constants satisfying μ0 > μ∞ > 0. In this case
E = ∅.
Observe that for these three fluids, if μ0 = μ∞ , then μ(z) is constant and we recover
the equation of a Newtonian fluid.
Carreau fluids
They are defined by the constitutive equation
r −2
μ(z) = μ∞ + μ0 − μ∞ )(1 + λz 2 2 , (4.4.5)
where μ0 > μ∞ ≥ 0, 1 < r ≤ 2, λ > 0. Here again E = ∅. When r = 2, μ(z) is

constant and we recover again the equation of a Newtonian fluid.
4.4.2 Formulation and Properties of the Main Operator
Let introduce the quantities M and m defined as
sup μ(z) = M,
z∈IR +
(4.4.6)
inf μ(z) = m.
z∈IR +
4.4 Fluids with Nonlinear Viscosity Bounded at Infinity 169
Observe that these numbers exist by virtue of hypotheses H.1–H.2 and satisfy, due to
(4.4.3),
0 ≤ m ≤ M < +∞.
The functional spaces where we will work are the classical spaces for Navier–Stokes
equations, namely V and H , whose definitions are respectively, (4.2.19) and (4.2.20).
The variational equation corresponding to (4.4.1) is
u , v + Au, v + c(u; u, v) = f , v, ∀v ∈ V,

(4.4.7)
u(0) = u0 ,
where the operator A is defined by

Av, w = 2 μ(s(v)) D(v), D(w) , ∀v, w ∈ V, (4.4.8)
and f is an exterior force.

It is easily seen that owing to hypotheses H.1–H.3, the operator A maps the space
H01 (Ω)d into H −1 (Ω)d . Moreover, one has the following result:
Lemma 4.4.2 The operator A has the following properties:

(i) A is hemicontinuous,
(ii) A is monotone.
Proof (i) For the hemicontinuity, one has to show that for u, v, w ∈ H01 (Ω)d the
real-valued functional
h : −→ A(u + λv), w
is continuous with respect to λ ∈ IR .

To do so, let {λn }n∈N be a real sequence such that λn → λ. By definition

A(u + λn v), w = 2μ s(u + λn v) D(u) : D(w) d x
Ω

+ λn 2μ s(u + λn v) D(v) : D(w) d x. (4.4.9)
Ω
When λn → λ, since by H.1 μ is continuous,

μ s(u + λn v)) → μ s(u + λv)) as n → ∞.
So, pointwisely

2μ s(u + λn v)) D(u) : D(w) → 2μ s(u + λv)) D(u) : D(w).
On the other hand

|2μ s(u + λn v)) D(u) : D(w)| ≤ 2M| D(u) : D(w)|,
with D(u) : D(w) ∈ L 1 (Ω), since u, w ∈ V . One can now apply the Lebesgue
dominated convergence theorem to the two terms of (4.4.9) to get the result.
(ii) For the monotonicity, we have to show that
Au − Av, u − v ≥ 0, ∀u, v ∈ H01 (Ω)d , (4.4.10)
where

Au − Av, u − v = 2μ s(u) D(u) : D(u − v) d x
Ω

− 2μ s(v) D(u) : D(u − v) d x.
Ω
This proof is the only delicate point of this section. Let α ∈ [0, 1] and consider the
following function:

ψ(α) = μ s(v + α(u − v)) D(v + α(u − v)) : D(u − v) d x. (4.4.11)
Ω
Notice that ψ is continuous on [0, 1] and
Au − Av, u − v = 2[ψ(1) − ψ(0)]. (4.4.12)
The proof of the monotonicity is done in two steps. We start by showing that ψ
is differentiable on the interval ]0, 1[. Afterwards, we apply the Taylor formula to the
right-hand side of (4.4.12) for some λ ∈ [0, 1], and finally we estimate it by making
use of hypothesis (4.4.3).
For x ∈ Ω and α ∈ [0, 1] set

F(x, α) = μ s(v + α(u − v)) D(v + α(u − v)) : D(u − v), (4.4.13)
therefore,
ψ(α) = F(x, α) d x.
Ω
We will show that F is differentiable with respect to α at any x ∈ Ω, and moreover,

that its modulus is bounded by an integrable function on Ω, independently of α. The
Lebesgue dominated convergence theorem will then give the result.
To achieve this, consider the following two subsets of Ω:
Ω1 (ξ ) = {x ∈ Ω ; 0 < s(v + ξ(u − v)) < +∞ at the point x},

(4.4.14)
Ω2 (ξ ) = {x ∈ Ω ; s(v + ξ(u − v)) = 0 at the point x}.
Obviously,
Ω1 (ξ ) ∩ Ω2 (ξ ) = ∅,
and the measure of the set Ω \ (Ω1 (ξ ) ∪ Ω2 (ξ )) is zero.

We will now verify that for x ∈ Ω1 (ξ ) the derivative of F with respect to ξ is given
by the formula
∂ F(x, ξ ) μ (s(v + ξ(u − v))) 2

= D(v + ξ(u − v)) : D(u − v)
∂ξ s(v + ξ(u − v))
+ μ(s(v + ξ(u − v))) D(u − v) : D(u − v), (4.4.15)
and that for x ∈ Ω2 (ξ ),
∂ F(x, ξ )
= μ(0) D(u − v) : D(u − v). (4.4.16)
∂ξ
Regarding (4.4.15), let x ∈ Ω1 (ξ ). Then
∂ F(x, ξ )
= μ(s(v + ξ(u − v))) D(u − v) : D(u − v),
∂ξ
∂s(v + ξ(u − v))
+ μ (s(v + ξ(u − v))) D(v + ξ(u − v)) : D(u − v). (4.4.17)
∂ξ
But
s(u) = 2D2 (u) = | D(u)|,
where | · | is the Frobenius norm, so that
∂s(v + ξ(u − v)) ∂s(v + ξ(u − v)) ∂ D2 (v + ξ(u − v))

=
∂ξ ∂ D2 (v + ξ(u − v)) ∂ξ
1
= D(v + ξ(u − v)) : D(u − v), (4.4.18)
s(v + ξ(u − v))
which, when substituted in (4.4.17) gives formula (4.4.15).

In the case when x ∈ Ω2 (ξ ), one cannot apply (4.4.18) because s(v+ξ(u−v)) = 0.
Instead, this implies that
D(v) = −ξ D(u − v).
This can happen only in the following situations:
(a) D(u) = D(v) = 0,

(b) ∃k, l with Dkl (u − v) = 0.
In the case (a), from definition (4.4.13) of the function F,
F(x, α) = 0, ∀α ∈]0, 1[,
∂F
so = 0 and formula (4.4.16) is valid.
∂ξ
In the case (b) the situation is more complicated. Of course, the fact that D(v+ξ(u−
v)) = 0 implies again that F(x, ξ ) = 0, but nevertheless for any α in a sufficiently
small neighborhood of ξ , F(x, α) = 0.
Let {αk }k∈IN be a sequence of elements in ]0, 1[ such that αk = ξ for any k ∈ IN and
limk→∞ αk = ξ . From the remark above, the function α → F(x, α) is differentiable
with respect to αk , and its derivative is given by (4.4.15). Now, write (4.4.15) at the
point (x, αk ) and let k → ∞. Since μ and s are continuous,
∂ F(x, αk )
lim = μ(0) D(u − v) : D(u − v) + lim Ik ,
k→∞ ∂ξ k→∞
where
μ (s(v + αk (u − v))) 2
Ik = D(v + αk (u − v)) : D(u − v) .
s(v + αk (u − v))
Notice that formula (4.4.16) is completely established if we can show that
lim Ik = 0. (4.4.19)
k→∞
But since s(v + αk (u − v)) = | D(v + αk (u − v))|, we immediately derive from the
properties of the Frobenius norm
|Ik | ≤ |μ (s(v + αk (u − v)))| | D(v + αk (u − v))|| D(u − v)|2

≤ |μ (s(v + αk (u − v)))| s(v + αk (u − v)) | D(u − v)|2 . (4.4.20)
If k → ∞, then αk → ξ and by the continuity of s, s(v + αk (u − v)) → s(v +

ξ(u − v)) = 0 as well as μ (s(v + αk (u − v))) → μ (0). It follows that the right-hand
side of (4.4.20) goes to zero; then this gives (4.4.19) and consequently, (4.4.16).
∂F
It is easily seen now that is bounded by an integrable function. Indeed, by
∂α
(4.4.15) and (4.4.16)
∂F 2

∂α ≤ C D(u − v) .
This is a consequence of the fact that in the right-hand side of the two formulas (4.4.15)
and (4.4.16), the factors multiplying | D(u − v)|2 are all compositions of the functions
μ, μ and s. But by assumptions, these are continuous functions in the closed interval
[0, 1], hence bounded herein. This ends the proof of the first step.
We now apply the Taylor formula to the function ψ from (4.4.12) to get

dψ ∂ F(x, α)
Au − Av, u − v = 2 (λ) = 2 (λ) d x,
dα Ω ∂α
for some λ ∈]0, 1[. Consequently,


dF dF
Au − Av, u − v = 2 (x, λ) d x + 2 (x, λ) d x,
ΩE dξ Ω\ΩE dξ
where
ΩE = {x ∈ Ω ; s(v + λ(u − v)) ∈ E }.
For the first integral, from hypothesis (4.4.3), we have

dF
2 (x, λ) d x ≥ m u − v2H 1 (Ω ) ≥ 0,
ΩE dξ 0 E
as μ (z) ≥ 0 if z ∈ E .
Next, we split the second integral into two integrals and show that each one is
nonnegative. More precisely, we write

dF dF dF
(x, λ) d x = (x, λ) d x + (x, λ) d x,
Ω\ΩE dξ Ω1 ∩(Ω\ΩE ) dξ Ω2 ∩(Ω\ΩE ) dξ
where Ω1 and Ω2 are defined by (4.4.14).

The nonnegativity of the second integral in the right-hand side follows immediately
from (4.4.16) and the nonnegativity of μ.
On the other hand, by formula (4.4.15),

dF
(x, λ) d x ≥ μ(s(v + λ(u − v))) D(u − v) : D(u − v) d x
dλ
Ω1 ∩(Ω\ΩE ) Ω1 ∩(Ω\ΩE )

− |μ (s(v + λ(u − v)))| s(v + λ(u − v)) | D(u − v)|2 d x ≥ 0,
Ω1 ∩(Ω\ΩE )
(4.4.21)
where we used the inequality
μ(z) − z |μ (z)| ≥ 0, ∀z ∈ IR + \ E ,
which is a consequence of hypothesis (4.4.3). This ends the proof of the lemma.
Remark 4.4.3 One can improve the preceding estimates by making more restrictive
assumptions on μ as for instance, the following one instead of H.3 :
H.3’. There exists a set E ⊂ IR + such that
μ (z) ≥ 0, ∀z ∈ E ,
(4.4.22)
μ(z) − z |μ (z)| ≥ μ1 > 0, ∀z ∈ IR + \ E .
Notice that (4.4.22) holds with E = ∅ in the case of Williamson and Carreau fluids,
defined respectively, by (4.4.4) and (4.4.5).
With this new hypothesis, going back to estimate (4.4.21), one easily shows

2

2
(Au − Av, u − v) ≥ m
u − v
H 1 (Ω ) + m
u − v
H 1 (Ω \(Ω\Ω ))
0 E 0 2 E

2

2

+ μ1 u − v H 1 (Ω \(Ω\Ω )) > C0 u − v H 1 (Ω) .

0 1 E 0
Remark 4.4.4 If μ satisfies H.1–H.3 with E = IR + , the above proof leads to

2
(Au − Av, u −v) ≥ m
u −v
H 1 (Ω) .
0
4.4.3 Existence and Uniqueness Results
Theorem 4.4.5 Let f ∈ L 2 (0, T ; L 2 (Ω)d ) with 0 < T < +∞, u0 ∈ H and μ
satisfying hypotheses H.1–H.3. Then there exists a solution u of (4.4.7) such that
u ∈ L 2 (0, T ; V ) ∩ L ∞ (0, T ; H ), u ∈ L 2 (0, T ; V ). (4.4.23)
Proof The proof follows the lines of Theorem 4.2.10, obviously without needing any
regularization. Let us sketch it.
We discretize again by Galerkin’s method with the special basis: w j ∈ V ,
((w j , v))V = λ j (w j , v), ∀v ∈ V.

n
The approximate solution un = j=1 c jn (t)w j , solves

(u n , wi ) − div [2μ(s(un )) Dun ], wi + c(un ; un , wi ) = ( f , wi ),
(4.4.24)
un (0) = u0n ,
for i = 1, 2, . . . , n, with
u0n ∈ H, u0n → u0 strongly in H.
We get a system of ordinary differential equations for cin , i = 1, . . . n, of the form
dcin
= G i (t, c1n , . . . , cnn ), 1 ≤ i ≤ n,
dt
which by the Cauchy theorem has a solution un on an interval [0, Tn ]. As usual, we

have to obtain a priori estimates independent of n, in order to prove that Tn = T .
It is immediate from (4.4.24) that

1 d
un (t)2L 2 (Ω) + 2 μ(s(un (t))) D(un (t)) : D(un (t)) d x
2 dt Ω

= f (t) · un (t) d x. (4.4.25)
Ω
With the notation introduced at the beginning of this section,
D(un (t)) : D(un (t)) = s 2 (un (t)),
so that

μ(s(un (t))) s 2 (un (t)) d x = μ(s(un (t))) − μ∞ s 2 (un (t)) d x
Ω Ω

+ μ∞ s 2 (un (t)) d x, (4.4.26)
Ω
where μ∞ is defined by (4.4.2). Then from (4.4.25)
1 d

un (t)
2 2

2

L (Ω)
+ 2μ∞
un (t)
H 1 (Ω) ≤
f (t)
L 2 (Ω)
un (t)
L 2 (Ω)
2 dt 0

−2 μ(s(un (t))) − μ∞ s 2 (un (t)) d x. (4.4.27)
Ω
Set μ1 = μ(s) − μ∞ . By hypothesis, μ1 (s) → 0 when s → ∞. Therefore for

ε > 0, there exists β > 0 such that if |s(un (t))| > β, then
|μ1 (s(un ))| < ε for all (x, t) ∈ Ω×]0, Tn [. (4.4.28)
For any t ∈ [0, Tn ], consider the following two subsets of Ω:
Ω1t = {x ∈ Ω ; s(un (x, t)) ≤ β},

Ω2t = {x ∈ Ω ; s(un (x, t)) > β}.
With the notation (4.4.6) and using (4.4.28), we get

|μ1 (s(un (x, t)))|s 2 (un (x, t)) d x ≤ β 2 |μ1 (s(un (x, t)))| d x
Ω Ω1t

2
+ε s 2 (un (x, t)) d x ≤ (μ∞ + M)β 2 |Ω| + ε
un
H 1 (Ω) .
Ω2t 0
When substituted into (4.4.27), this yields
1 d

un (t)
2 2

2

L (Ω)
+ 2μ∞
un (t)
H 1 (Ω) ≤
f (t)
L 2 (Ω)
un (t)
L 2 (Ω)
2 dt 0

2
+ 2(μ∞ + M)β 2 |Ω| + ε
un (t)
H 1 (Ω) .
0
1
Choose ε = μ∞ and set
2
C(Ω, μ∞ , M) = 2(μ∞ + M)β 2 |Ω|.
Then
1 d

un (t)
2 2

L (Ω)
+ 2μ∞
un (t)2H 1 (Ω)
2 dt 0
2
S0,2
2

2
≤
f (t)
2 + μ∞
un (t)
H 1 (Ω) + C(Ω, μ∞ , M),
4μ∞ L (Ω) 0
where S0,2 is the Poincaré constant of the domain Ω (see (3.15)). Integrating in t we
get the a priori estimate
T
1
un (T ) L 2 (Ω) + μ∞
2
un (t)2H 1 (Ω) dt
2 0 0
2
S0,2 1
≤ f 2L 2 (0,T ;H ) + u0 2L 2 (Ω) + T C(Ω, μ∞ , M) ≤ C1 (T ),
4μ∞ 2
uniformly in n for any finite T , and where obviously, the constant C1 (T ) depends on
T . Thus,
un is in a bounded set of L ∞ (0, T ; H ) ∩ L 2 (0, T ; V ).
The remainder of the proof follows the argument of the proof of Theorem 4.2.10.
For the a priori estimates, the only additional information is the following one, straight-
forward from the properties of the operator A defined in (4.4.8):
Aun is in a bounded set of L 2 (0, T ; H −1 (Ω)d ),
and thus,
u n is in a bounded set of L 1 (0, T ; V ).
It follows that, up to a subsequence,
lim un = u weakly in L 2 (0, T ; V ),

n→∞
lim un = u weakly in L ∞ (0, T ; H ),
n→∞
lim un = u strongly in L 2 (0, T ; H ).
n→∞
Moreover,
lim A(un ) = χ weakly in L 2 (0, T ; H −1 (Ω)d ).
n→∞
With these convergences and using Proposition 4.2.5, we can pass to the limit in (4.4.24)
to get
(u (t), v) + χ (t), v + c(u(t); u(t), v) = f (t), v, ∀v ∈ V.
Finally, as A is monotone, the Minty method used in the proof of Theorem 4.2.10,
allows us to identify χ as χ(t), v = A(u(t)), v for all v ∈ V . Hence u is solution
of (4.4.7) and this ends the proof of Theorem 4.4.5.
In the particular case where the lower bound m > 0 (see (4.4.6)), one has global
existence for the solution. The following result holds true:
Theorem 4.4.6 Let f ∈ L 2 (0, +∞; L 2 (Ω)d ), u0 ∈ H and μ satisfying hypotheses
H.1–H.3. Assume furthermore that
m > 0.
Then problem (4.4.7) has a solution u such that
u ∈ L 2 (0, +∞; V ) ∩ L ∞ (0, +∞; H ).
Proof Looking at the above proof, it is clear that inequality (4.4.27) does not allow to
take T = +∞. We will now estimate more sharply the second term in (4.4.7) in order
to have a priori estimates independent of T , the remainder of the proof being the same
as above.
Observe that

|μ(s(un (x, t)))|s (un (x, t)) d x ≥ m
2
s 2 (un (x, t)) d x.
Ω Ω
Then, from (4.4.7)

1 d

un (t)
2 2
L (Ω)
+ 2m s (un (x, t)) d x ≤
2
f (t) · un (x, t) d x.
2 dt Ω Ω
Since m > 0, the contribution of un to the right-hand side can be absorbed by the second
term in the left-hand side. As f belongs to L 2 (0, +∞; L 2 (Ω)d ), the same computation
as in the proof of Proposition 4.3.1, yields the estimates
un L ∞ (0,+∞;L 2 (Ω)d ) ≤ C and un L 2 (0,+∞;H 1 (Ω)d ) ≤ C,

0
where the constant C depends only on u0 , m and f L 2 (0,+∞;L 2 (Ω)d ) .
We end this section with a few remarks on the uniqueness of the solutions of equa-
tion (4.4.7). Let us point out that the fluids defined by the constitutive equation (4.4.1)
with μ satisfying hypotheses H.1–H.3, are very close to the Newtonian fluid which
corresponds to A = −Δ. Glancing at the proof of the uniqueness Theorem 4.2.20,
we see that in order to be able to apply Gronwall’s inequality, we need to estimate
Au1 − Au2 for two solutions of (4.4.7). But, from (4.4.10) and hypotheses H.1–H.3’,
(A(u1 ) − A(u2 ), u1 − u2 ) ≥ C0 u1 − u2 2V ,
with C0 = m if E = IR + .
Consequently, as seen in the proof of Theorem 4.2.20, the uniqueness depends on
the estimates of the form c. Thus, we have the following result:
Theorem 4.4.7 Let Ω ⊂ IR 2 and μ satisfying H.1, H.2 and H.3’ with m > 0. Then
the solution u ∈ L 2 (0, T ; V ) ∩ L ∞ (0, T ; H ) of (4.4.7), given by Theorem 4.4.5, is
unique.
Chapter 5
Grade-Two Fluids: Some Theoretical Results
5.1 Introduction and Preliminaries
In this chapter, we present the essential ideas of the mathematical analysis of the
equations modeling the flow of grade-two fluids.
Starting with a first attempt by Oskolkov [204], a large number of mathematicians
have worked on the theoretical analysis of grade-two fluids. In some sense, the theo-
retical results established so far are fairly satisfactory, but there still remain important
open questions such as the problem posed by fully nonhomogeneous Dirichlet bound-
ary conditions or that posed by a rough exterior force, such as a force in L 2 to mention
just these two “simple” questions. In Section 2.5, we discussed the important issue
of nonhomogeneous Dirichlet boundary conditions or other boundary conditions.
In this chapter, we shall treat flows assuming tangential Dirichlet boundary con-
ditions, i.e., with no inflow or outflow. If there is an inflow or outflow, the problem
is ill-posed and we yet have no theoretical foundation for imposing an additional
boundary condition to make the problem well posed. As far as the exterior force is
concerned, from a practical point of view, one is seldom faced with a rough exte-
rior force, because this force is usually gravitational. But from a mathematical point
of view, it would be interesting to be able to handle cases where the exterior force
belongs only to L r with r ≤ d, the dimension. This last point has been addressed
recently when free-slip is prescribed on the whole boundary but such boundary con-
ditions are not encountered in practical situations.
5.1.1 Formulation
Recall that the Cauchy stress for a grade-two fluid is given by (2.4.19)
T = − p I + μ A1 + α1 A2 + α2 A21 .

DOI 10.1007/978-3-319-39330-8_5
180 5 Grade-Two Fluids: Some Theoretical Results
We shall refer to the Lagrange multiplier that enforces the constraint of incom-
pressibility as “pressure”. Unlike the Newtonian fluid where the Lagrange multiplier
is also the mean value of the stress, for a grade-two fluid, the Lagrange multiplier is
not the mean value of the stress. Usually the term pressure refers to the mean value
of the stress (see Rajagopal1 for a detailed discussion of relevant issues).
By expanding formula (2.3.26) for A2
d ∂ ∂
3
A2 = A1 + A1 L + L T A1 = A1 + ui A1 + A1 L + L T A1 ,
dt ∂t i=1
∂ x i
introducing the spin tensor W defined in (2.3.24)
1
W= (L − L T ),
2
and observing that
A1 L + L T A1 = A1 W − W A1 + A21 ,
we obtain an alternative expression for T
∂ 3
∂
T = − p I +μ A1 +α1 A1 + ui A1 + A1 W −W A1 +(α1 +α2 ) A21 . (5.1.1)
∂t i=1
∂ xi
When substituted into the balance of linear momentum

d
u = div T + f , (5.1.2)
dt
where is the density and f is an exterior force (such as gravity) and dividing by
the density, (5.1.1) yields the momentum equation of a grade-two fluid. For this, we
use the following formulas:

3
∂
div ui A1 + A1 W − W A1
i=1
∂ xi
1
= 2 u · Δ W + ∇ u · Δ u + | A1 |2
4
1
= curl(Δu) × u + ∇ u · Δ u + | A1 |2 , (5.1.3)
4
1 K.
R. Rajagopal, Remarks on the notion of “pressure”, International Journal of Non-linear
Mechanics, 71, 165–172 (2015).
5.1 Introduction and Preliminaries 181
and
1
div( A21 ) = Δ([∇ u]u)−2[∇(Δ u)]u+curl(Δu)×u+∇ u·Δ u+ | A1 |2 , (5.1.4)
4
where we recall that | · | is the euclidian or the Frobenius norm
| A1 |2 = A1 : A1 = tr A21 .
Thus, denoting by the same symbols the viscosity and normal stress moduli divided
by the density, the equation of motion reads
∂
∂t
− (α1 + α2 )Δ [∇ u]u + 2(α1 + α2 )[∇(Δ u)]u + ∇ p = f , (5.1.5)
where, for the sake of simplicity, we still denote by p all the terms that appear as
gradients
1 1 1
p := p + |u|2 − (2 α1 + α2 )(u · Δ u + | A1 |2 ).
2 4
For the fluid to be thermodynamically compatible (cf. [87]), the viscosity μ and
the material moduli α1 and α2 (all assumed to be constant) must satisfy (2.4.22)
μ ≥ 0 , α1 ≥ 0 , α1 + α2 = 0.
As α1 + α2 = 0, we set α = α1 and (5.1.5) simplifies to
∂
(u − αΔ u) − μ Δ u + curl(u − αΔ u) × u + ∇ p = f , (5.1.6)
∂t
with
1 1 1
p := p + |u|2 − α (u · Δ u + | A1 |2 ),
2 4
that must be complemented by the incompressibility condition, an initial condition

at time t = 0 and a suitable boundary condition. In the next section, we analyze the
problem on a simply-connected domain, with a no-slip boundary condition.
Remark 5.1.1 When α = 0, (5.1.6) coincides with the Navier–Stokes equations, in

virtue of the identity
1
[∇ u]u = curl u × u + ∇(|u|2 ).
2

Condition. The Simply-Connected Case
In this section, we concentrate on the three-dimensional case; the simpler two-

dimensional problem is discussed in Section 5.6. Let Ω be a bounded domain of
IR 3 with a Lipschitz-continuous boundary ∂Ω and let ]0, T [, T > 0, be a time inter-
val. The system of equations we propose to solve is Given a vector-valued function
f defined on Ω×]0, T [, and material constants α > 0, μ > 0, find a vector-valued
function u = (u 1 , u 2 , u 3 ) and a scalar function p defined on Ω×]0, T [, satisfying
∂
(u − αΔ u) − μ Δ u + curl(u − αΔ u) × u + ∇ p = f in Ω×]0, T [, (5.2.1)
∂t
div u = 0 in Ω×]0, T [, (5.2.2)
with homogeneous Dirichlet boundary conditions
u = 0 on ∂Ω×]0, T [, (5.2.3)
and initial data

u(0) = ut 0 in Ω, (5.2.4)
satisfying the compatibility conditions
div ut 0 = 0 in Ω and ut 0 = 0 on ∂Ω. (5.2.5)
Even in this simplified form, problem (5.2.1)–(5.2.4) is difficult because its non-
linear term involves a third-order derivative whereas its elliptic part is only a Laplace
operator. In this respect, it behaves much as a hyperbolic system. There are several
techniques for showing that it has solutions, but the best approach, in the sense that
it imposes the least restrictions on the size of the data and time interval, is due to
Cioranescu and Ouazar and dates back to 1981. With this approach, local existence
in time, in a simply-connected domain with smooth boundary, was first established
in the thesis of Ouazar [205] and later in the works of Cioranescu and Ouazar [64,
65]. By the same approach, Cioranescu and Girault established global existence in
time in [63] for small data. J.-M. Bernard in his thesis [28] and subsequent publi-
cations [29, 31, 32], generalized and extended this approach for proving existence
and uniqueness in domains with less smooth boundaries and in multiply-connected
domains.
5.2 Flow of a Grade-Two Fluid with No-Slip Boundary Condition … 183
5.2.1 Constructing Solutions: The Main Ideas
The following is a synthetic, formal description of the construction of solutions by

Cioranescu and Ouazar; rigorous details are given in the next subsection. Beforehand,
recall the notation (3.1.9)
v H01 (Ω) = |v| H 1 (Ω) .
The construction is based on a special Galerkin method and a priori estimates; they
are derived in five steps.
(1) First, observe that (5.2.1)–(5.2.3) yield the “energy” equality
1 d α d
u(t)2L 2 (Ω) + u(t)2H 1 (Ω) + μu(t)2H 1 (Ω) = ( f (t), u(t)). (5.2.6)
2 dt 2 dt 0 0
It shows in particular that, if a solution u exists, then it is unconditionally bounded

in L ∞ (0, T ; H 1 (Ω)3 ) by the data f and initial velocity ut 0 .
(2) Now, set
z = curl(u − α Δ u). (5.2.7)
This choice is crucial, because if curl(u − α Δ u) belongs to L 2 (Ω)3 , Ω is simply-

connected and ∂Ω is sufficiently smooth, then u is in W 1,∞ (Ω)3 , with continuous
dependence on z (see Lemma 5.2.5), that is
∇ u L ∞ (Ω) ≤ C∞ z L 2 (Ω) , (5.2.8)
with a constant C∞ that depends only on α and on the domain.

(3) Next, take formally the curl of (5.2.1). By virtue of the identity
curl(z × u) = [∇ z]u − [∇ u]z + zdiv u − udiv z, (5.2.9)
that reduces to
curl(z × u) = [∇ z]u − [∇ u]z,
because both u and z are divergence-free, we obtain a transport equation, (that we

multiply here by α)
∂
α z + μ z + α([∇ z]u − [∇ u]z) = μ curl u + α curl f in Ω×]0, T [. (5.2.10)
∂t
Formally multiplying (5.2.10) by z yields the inequality
α d
z(t)2L 2 (Ω) + (μ − α∇ u(t) L ∞ (Ω) )z(t)2L 2 (Ω) ≤ (μ curl u(t) L 2 (Ω)
2 dt
+ α curl f (t) L 2 (Ω) )z(t) L 2 (Ω) . (5.2.11)
By substituting (5.2.8) into the left-hand side of (5.2.11) and the estimate deduced
from (5.2.6) to bound curl u(t) L 2 (Ω) in its right-hand side, we find that z(t)2L 2 (Ω)
solves a Riccati differential inequality; hence it exists on the time interval [0, T ∗ ],
for some T ∗ , 0 < T ∗ ≤ T . This shows that, if a solution u exists, then it is bounded
in L ∞ (0, T ∗ ; W 1,∞ (Ω)3 ) in terms of the data.
(4) On multiplying formally (5.2.1) by u := ∂∂tu and substituting the previous
bounds for u and z into

(z × u, u ) ≤ S0,6 S0,3 z L 2 (Ω) u H 1 (Ω) u H 1 (Ω) ,
0 0
we infer that u is also bounded in L 2 (0, T ∗ ; H 1 (Ω)3 ).

(5) Finally, using the isomorphism of Theorem 3.2.11 and applying the above
bounds for u, z, and u , we deduce that p is bounded in L 2 (Ω×]0, T ∗ [).
Remark 5.2.1 Although the grade-two fluid model is more complex than the Navier–
Stokes model, it satisfies stronger a priori estimates. In particular, estimating the
pressure is much easier.
Remark 5.2.2 At first sight, one might think that the “energy” equality (5.2.6) is
discardable since it gives a bound in H 1 (Ω)3 , whereas (5.2.11) and (5.2.7) give a
bound in W 1,∞ (Ω)3 . In fact it is vital for estimating curl u(t) L 2 (Ω) in the left-hand
side of (5.2.11) in terms of the data f . If we discard (5.2.1) and hence discard (5.2.6),
and replace it by (5.2.11), then the unknown quantity z appears in the right-hand side
instead of the data f , and the existence results are no longer optimal. In particular,
if this strategy is applied to the two-dimensional problem (or any other strategy that
discards (5.2.1)) we can no longer establish existence of solutions without restrictions
on the data. And, much worse, we do not know how to handle the numerical analysis
of discrete schemes that solve (5.2.1)–(5.2.4).
These estimates only hold provided a solution exists, but constructing a solution
by making use of (5.2.1), (5.2.7), and (5.2.10) is very difficult because these three
equations are redundant and no fixed point can use all three at the same time. The orig-
inality and power of the construction by Cioranescu and Ouazar lie in that it does use
all three equations. This is achieved by discretizing (5.2.1) using a Galerkin method
(see (5.2.31)) in the basis of the eigenfunctions of the operator curl curl(u − α Δ u),
a process similar to that used in Section 4.2.4. This special basis has the effect that, on
multiplying the ith equation of (5.2.31) that discretizes (5.2.1) by the eigenvalue λi
and on summing over i, it yields a discrete version of the transport equation (5.2.10)
(see (5.2.33)). This allows one to recover (5.2.11) in the discrete case. As a con-
sequence, the discrete solution um is bounded uniformly in L ∞ (0, T ∗ ; W 1,∞ (Ω)3 )
with um bounded in L 2 (0, T ∗ ; H 1 (Ω)3 ). Note that all the above steps (which were
hitherto formal), are justified because the basis functions are sufficiently smooth (see
Lemma 5.2.8). Furthermore, passing to the limit is standard, because there are enough
a priori bounds and the limit is only taken in the discrete version of (5.2.1). This
proves local existence in time of a solution. Finally, global existence for small data
can be established by taking better advantage of the damping effect of the viscous
term −μΔ u.
5.2.2 A First Variational Formulation
In view of (5.2.7), we define the space for α > 0
V α = {v ∈ V ; α curl Δ v ∈ L 2 (Ω)3 }, (5.2.12)
and for α = 0, we set V 0 = V . In the sequel, unless otherwise specified, we shall

take α > 0. The space V α is a Hilbert space equipped with the norm
1
vV α = v2H 1 (Ω) + α curl Δ v2L 2 (Ω) 2 . (5.2.13)
0
Further on, it will also be convenient to use the following scalar product on V α :
((u, v)) = (u, v) + α(∇ u, ∇ v) + (curl(u − αΔ u), curl(v − αΔ v)). (5.2.14)
Its associated norm is clearly equivalent to · V α on V α .

The first important result shows that, under adequate conditions on Ω, V α is
imbedded into C1,σ (Ω)3 for some σ > 0. Its proof is based on the following con-
struction of vector potentials.
Proposition 5.2.3 Let Ω be a bounded, Lipschitz-continuous open set of IR 3 and

let Γi , 0 ≤ i ≤ R, denote the connected components of its boundary ∂Ω. Every
function w ∈ H (div, Ω) that satisfies
div w = 0 in Ω ,
w · n, 1Γi = 0 0 ≤ i ≤ R, (5.2.15)
has a solenoidal vector potential ϕ ∈ H 1 (Ω)3
w = curl ϕ , div ϕ = 0 in Ω, (5.2.16)
and ϕ can be constructed so that the mapping w → ϕ is linear and continuous:

there exists a constant c, depending only on Ω, such that
ϕ H 1 (Ω) ≤ cw L 2 (Ω) . (5.2.17)
Proof We just sketch the main steps of the proof; details can be found in [116,
Chapter I, Theorem 3.4]. The idea is to suitably extend w to IR 3 and then construct
ϕ by Fourier transforms.
First, by solving a Laplace equation with Neumann boundary conditions match-
ing w · n, w is extended to IR 3 in such a way that it stays divergence-free and has
compact support. This extension is possible if and only if the boundary conditions
in (5.2.15) hold. Let w̃ denoted the extended function. Then the Fourier transform
of ϕ is constructed from the Fourier transform of w̃ by solving a system of algebraic
equations in IR 3 and the compact support of w̃ implies that the restriction of ϕ to
Ω belongs to H 1 (Ω)3 . It is easy to check that these two steps define a linear and
continuous mapping and (5.2.17) holds.
In order to apply this result to curl Δ v, in the case where ∂Ω is not connected,
i.e., R ≥ 1, we must check the boundary conditions in (5.2.15). This result, which
is not completely straightforward, is the object of the following proposition:
Proposition 5.2.4 Let Ω be a bounded, Lipschitz-continuous open set of IR 3 and
let Γi , 0 ≤ i ≤ R, denote the connected components of its boundary ∂Ω. Let
w ∈ H −1 (Ω)3 satisfy div w ∈ L 2 (Ω) and curl w ∈ L 2 (Ω)3 . Then

curl w · n, 1Γi = 0 , 0 ≤ i ≤ R. (5.2.18)
Proof The proof proceeds in two steps.

(i) First, we prove that (5.2.18) holds when w ∈ L 2loc (Ω)3 and curl w ∈ L 2 (Ω)3 . This
is established by localizing w in a neighborhood of each Γi . For each i, 0 ≤ i ≤ R,
we choose a smooth function ϕi such that
ϕi = δi j in a neighborhood of Γ j , 0 ≤ j ≤ R,
and consider curl(ϕi w). On one hand, since any distribution T satisfies div(curl T ) =
0, we have div (curl(ϕi w)) = 0. On the other hand, the identity
curl(ϕi w) = ϕi curl w + (∇ ϕi ) × w,
the assumption on w, and the properties of ϕi near the boundary imply that
curl(ϕi w) ∈ L 2 (Ω)3 .
Hence curl(ϕi w) belongs to H (div, Ω). As a consequence curl(ϕi w) · n is well-

defined on ∂Ω (cf. [116, Chapter I, Theorem 2.5]); furthermore curl w · n =
curl(ϕi w) · n on Γi . Therefore, Green’s formula yields

0= div (curl(ϕi w)) d x =
curl(ϕi w) · n, 1Γi =
curl w · n, 1Γi .
Ω
(ii) Next, we prove that indeed w belongs to L 2loc (Ω)3 . To this end, we localize w
inside Ω. First, as w belongs to H −1 (Ω)3 , it is easy to extend it by zero outside Ω,
see (3.1.2), and the resulting distribution, still denoted by w belongs to H −1 (IR )3 .
Next, we fix two arbitrary compact sets K 1 K 2 Ω, and we define a cutoff
function μ ∈ D(Ω) such that 0 ≤ μ ≤ 1 in IR 3 , μ = 1 in K 1 and μ = 0 in IR 3 \K 2 .
We shall prove that μ w belongs to L 2 (Ω)3 ; as μ w = w in K 1 and K 1 is arbitrary,
this will show that w ∈ L 2loc (Ω)3 .
For proving that μ w ∈ L 2 (Ω)3 , we solve the Laplace equation in IR 3 : Find v such
that
Δ v = −curl curl (μ w) + ∇ div (μ w) = Δ (μ w). (5.2.19)
It is easy to check that the right-hand side of (5.2.19) belongs to H −2 (IR 3 )3 and has
compact support. Therefore, (5.2.19) has a unique solution with components in the
following weighted space:
1
L 21 (IR 3 ) = v ∈ D (IR 3 ); v ∈ L 2 (IR 3 ) ,
ρ
1
where ρ = (1 + |x|2 ) 2 (see for instance Giroire [122, Theorem 2.17, p. 100]). Hence
v ∈ L 2 (Ω)3 and
Δ(v − μ w) = 0 in IR 3 .
As both v and μ w are tempered distributions, Theorem 3.2.9 implies that they
differ by a polynomial vector function. Therefore, the restriction to Ω of μ w belongs
to L 2 (Ω)3 .
With these preliminary propositions, we can establish the following crucial results:
Lemma 5.2.5 Let Ω be a bounded, simply-connected open set of IR 3 that is either a

convex polyhedron or has a boundary ∂Ω of class C1,1 . Then any v in V α belongs to
2 3
H (Ω) ∩ C1,σ (Ω) for some σ > 0 depending on ∂Ω, and there exists a constant
γα depending only on α and Ω and a constant γσ,α depending only on σ , α, and Ω
such that
∀v ∈ V α , v H 2 (Ω) ≤ γα curl(v − αΔ v) L 2 (Ω) , (5.2.20)
∀v ∈ V α , vC1,σ (Ω) ≤ γσ,α curl(v − αΔ v) L 2 (Ω) . (5.2.21)
Proof Let v belong to V α and set
y = curl(v − αΔ v).
Then w := v − αΔ v satisfies the assumptions of Proposition 5.2.4; hence y satisfies

the assumptions of Proposition 5.2.3 and therefore there exists ϕ ∈ H 1 (Ω)3 such
that
y = curl ϕ , div ϕ = 0 in Ω , ϕ H 1 (Ω) ≤ ccurl(v − αΔ v) L 2 (Ω) ,
with the constant c of (5.2.17). Thus
curl(αΔ v − v + ϕ) = 0,
and the fact that Ω is simply-connected implies that there exists p in L 2 (Ω) such
that
v − αΔ v + ∇ p = ϕ in Ω. (5.2.22)
Dividing both sides of this equation by α, we see that v is the velocity part of the
solution of a Stokes problem with viscosity coefficient 1 and right-hand side α1 (ϕ −v)
in H 1 (Ω)3 . The above regularity assumptions on the domain imply on one hand that v
belongs to H 2 (Ω)3 . On the other hand, we observe that Sobolev’s imbedding implies
that H 1 (Ω) ⊂ L 6 (Ω) and W 1,6 (Ω) ⊂ C0,σ (Ω) for any σ in the range ]0, 21 ]. When
Ω is a convex polyhedron, Theorem 3.2.15 and Corollary 3.2.16 imply that there
exists a real number σ ∈]0, 21 ], depending on the angles of ∂Ω such that v is in
C1,σ (Ω)3 . When ∂Ω is C1,1 , then, owing to Theorem 3.2.13, v belongs to W 2,6 (Ω)3
1
that is imbedded into C1, 2 (Ω)3 .
To derive (5.2.20), let C2 denote the continuity constant in H 2 of the homogeneous
Stokes operator with unit viscosity, i.e., if h belongs to L 2 (Ω)3 and w is the solution
in H01 (Ω)3 of
−Δ w + ∇ q = h , div w = 0 in Ω,
then
w H 2 (Ω) ≤ C2 h L 2 (Ω) . (5.2.23)
Thus
C2
v H 2 (Ω) ≤ ϕ − v L 2 (Ω) .
α
But we easily derive from (5.2.22) that
v L 2 (Ω) ≤ ϕ L 2 (Ω) .
Therefore,
2C2
v H 2 (Ω) ≤ ϕ L 2 (Ω) ,
α
whence (5.2.20) with γα = 2cC2 /α, where c is the constant of (5.2.17).
To derive (5.2.21), we apply Corollary 3.2.16
Cσ
vC1,σ (Ω) ≤ ϕ − v L r (Ω) ,
α
where r = 3/(1 − σ ), and Cσ is a continuity constant analogous to C2 in (5.2.23).
Therefore, (3.1.5), (3.1.7), and an easy manipulation yield

Cσ Cσ S0,r
vC1,σ (Ω) ≤ Sr ϕ H 1 (Ω) + S0,r v H01 (Ω) ≤ Sr + √ ϕ H 1 (Ω) .
α α 2 α
This implies (5.2.21) with

Cσ S0,r
γσ,α =c Sr + √ .
α 2 α
Note that γσ,α is of the order of α − 2 .

3

Remark 5.2.6 When ∂Ω is smoother, v has more regularity. In particular, if ∂Ω is

of class C2,1 , we recover the full regularity of v, namely, v ∈ H 3 (Ω)3 .
In this section, we assume that the domain Ω satisfies the hypotheses of Lemma 5.2.5.
A consequence of (5.2.21) is that if u belongs to V α and z is defined by (5.2.7),
then u satisfies (5.2.8) with C∞ = γσ,α .
In view of (5.2.21) and the a priori estimates announced in Section 5.2.1, we pro-
pose the following formulation of (5.2.1)–(5.2.4): Given f ∈ L 2 (0, T, H (curl, Ω))
and ut 0 ∈ V α , find u ∈ L ∞ (0, T ; V α ) with u ∈ L 2 (0, T ; V ) and p ∈
L 2 (Ω×]0, T [), solution of
∀v ∈ H01 (Ω)3 , (u , v) + α (∇ u , ∇ v) + μ (∇ u, ∇ v) + (z × u, v)

− ( p, div v) = ( f , v), (5.2.24)
with the initial condition (5.2.4) u(0) = ut 0 in Ω , where z ∈ L ∞ (0, T ; L 2 (Ω)3 )

and u are related by (5.2.7)
z = curl(u − α Δ u),
z satisfies the system

α z + μ z + α [∇ z]u − [∇ u]z = μ curl u + α curl f , (5.2.25)
z(0) = curl(ut 0 − α Δ ut 0 ) in Ω. (5.2.26)
It is easy to check that system (5.2.24), (5.2.4), (5.2.7), (5.2.25), and (5.2.26) is
equivalent to (5.2.1)–(5.2.4) when the solutions belong to the above spaces.
Remark 5.2.7 Let u ∈ L ∞ (0, T ; V α ) be a solution of (5.2.24). Since div u and div z
are both zero, we have the following identities:
[∇ z]u = div(z ⊗ u) [∇ u]z = div(u ⊗ z).
Therefore, the products [∇ z]u and [∇ u]z are well-defined in the sense of distribu-
tions as soon as the components of z ⊗ u belong to L 1loc (Ω). More precisely, owing to
Lemma 5.2.5, the divergence of these two products belong to L ∞ (0, T ; H −1 (Ω)3 )
and it follows from (5.2.25) that z belongs to H 1 (0, T ; H −1 (Ω)3 ). Hence the initial
condition (5.2.26) has meaning.
5.2.3 A Galerkin Solution
Following [63, 64, 65, 205], we define the basis functions w j , for j ≥ 1, by solving
the eigenvalue problem: Find w j in V α such that
∀v ∈ V α , ((w j , v)) = λ j {(w j , v) + α(∇ w j , ∇ v)}, (5.2.27)
where the scalar product ((·, ·)) is defined in (5.2.14). Owing to Lemma 5.2.5, the
imbedding of V α into V is compact, therefore (5.2.27) defines a unique set of positive
eigenvalues λ j and a unique set of othonormalized eigenfunctions w j that constitutes
a basis of V α . Furthermore, we have the following extra regularity result:
Lemma 5.2.8 Under the assumptions of Lemma 5.2.5, the eigenfunctions w j defined
by (5.2.27) satisfy
curl(w j − αΔ w j ) ∈ H 1 (Ω)3 .
Proof For the sake of simplicity, we drop the index j. Let w ∈ V α and λ ∈ IR + solve
(5.2.27). Expanding (5.2.27) and applying Green’s formula, we derive the following
equation for w:
∀v ∈ V α , (curl(w − αΔ w), curl(v − αΔ v)) = (λ − 1)(w, v − αΔ v). (5.2.28)
Considering the regularity assumptions of Lemma 5.2.5, for any ϕ ∈ H 1 (Ω)3 , there
exists a unique v ∈ V ∩ H 2 (Ω)3 and p ∈ H 1 (Ω) ∩ L 2m (Ω) such that
v − αΔ v + ∇ p = ϕ in Ω.
Clearly, v belongs to V α and observing that (w, ∇ p) = 0, (5.2.28) implies
∀ϕ ∈ H 1 (Ω)3 , (curl(w − αΔ w), curl ϕ) = (λ − 1)(w, ϕ). (5.2.29)
Therefore, in the sense of distributions, w satisfies
curl(curl(w − αΔ w)) = (λ − 1)w,
and hence curl(curl(w − αΔ w)) is in L 2 (Ω)3 , i.e., curl(w − αΔ w) belongs to

H (curl, Ω). Of course, div(curl(w − αΔ w)) = 0. Furthermore, (5.2.29) yields
curl(w − αΔ w) × n = 0 on ∂Ω. Owing to the regularity assumptions on Ω,
these three properties imply that curl(w − αΔ w) belongs to H 1 (Ω)3 (cf. [116] or
Amrouche et al. [6]).
For any positive integer m, let Vm be the vector space spanned by the first m
eigenfunctions {w j }mj=1 and let Pm denote the orthogonal projection operator on Vm
for the scalar product of V α . Note that, owing to (5.2.27), Pm satisfies for all v ∈ V α
and for all j, 1 ≤ j ≤ m,
(Pm (v), w j ) + α(∇ Pm (v), ∇ w j ) = (v, w j ) + α(∇ v, ∇ w j ),

curl(Pm (v) − αΔ Pm (v)), curl(w j − αΔ w j ) (5.2.30)

= curl(v − αΔ v), curl(w j − αΔ w j ) .
Then we discretize the problem (5.2.24), (5.2.4) with z given by (5.2.7) as follows:
Find
m
um (t) = c j,m (t) w j ,
j=1
solution, for 1 ≤ j ≤ m, of
(um (t), w j ) + α(∇ um (t), ∇ w j ) + μ(∇ um (t), ∇ w j )

+ (curl(um (t) − α Δ um (t)) × um (t), w j ) = ( f (t), w j ), (5.2.31)
with
um (0) = Pm (ut 0 ). (5.2.32)
Since the functions w j are known, the unknown of this problem is the vector of
IR m with coefficients c j,m (t) and hence problem (5.2.31), (5.2.32) is a system of m
nonlinear ordinary differential equations of order one and degree two, with constant
coefficients in the left-hand side and with an initial condition at time t = 0. In
addition, the factor of the vector with coefficients cj,m (t) is a nonsingular matrix
and the right-hand side ( f (t), w j ) belongs to L 2 (0, T ) for each j. It is well known
(cf. for example Coddington and Levinson [66]) that such a system has a solution um ,
unique and continuous on [0, Tm∗ ] with um in L 2 (0, Tm∗ )3 , for some number Tm∗ > 0,
i.e., um belongs to H 1 (0, Tm∗ )3 .
The next theorem shows how the particular choice of basis functions w j allows
us to recover a discretization of (5.2.25). To simplify, set
z m (t) = curl(um (t) − αΔ um (t));
we know from Lemma 5.2.8 that z m is in C0 ([0, Tm∗ ]; H 1 (Ω)3 ) and its derivative z m
belongs to L 2 (0, Tm∗ ; H 1 (Ω)3 ), i.e., z m belongs to H 1 (0, Tm∗ ; H 1 (Ω)3 ).
Theorem 5.2.9 Under the assumptions of Lemma 5.2.5 and if
f ∈ L 2 (0, T ; H (curl, Ω)),
the pair of functions (um , z m ) satisfies a.e. in ]0, Tm∗ [, for 1 ≤ j ≤ m,
α(z m (t),curl(w j − α Δ w j )) + μ(z m (t), curl(w j − α Δ w j ))

+ α [∇ z m (t)]um (t) − [∇ um (t)]z m (t), curl(w j − α Δ w j )
= (μ curl um (t) + α curl f (t), curl(w j − α Δ w j )). (5.2.33)
Proof First, following [64], we define the vector-valued function F m by
∀t ∈ [0, Tm∗ ], F m (t) = −μ Δ um (t) + z m (t) × um (t). (5.2.34)
Observe that

α curl F m (t) = μ (z m (t) − curl um (t)) + α [∇ z m (t)]um (t) − [∇ um (t)]z m (t) ,
and hence, in view of Lemmas 5.2.5 and 5.2.8,
curl F m ∈ L ∞ (0, Tm∗ ; L 2 (Ω)3 ).
By definition, for 1 ≤ j ≤ m,
(um (t), w j ) + α(∇ um (t), ∇ w j ) + (F m (t), w j ) − ( f (t), w j ) = 0. (5.2.35)
Next for a.e. t in ]0, Tm∗ [, let vm (t) ∈ V be the velocity part of the solution of the
Stokes problem
vm (t) − α Δ vm (t) + ∇ qm (t) = F m (t) − f (t). (5.2.36)
It is a steady-state problem, but its solution depends on m and t because the right-
hand side depends on m and t. In addition, the assumption on f and the above
considerations yield that curl(vm − α Δ vm ) belongs to L 2 (Ω×]0, Tm∗ [)3 . Therefore,
vm (t) belongs to V α and hence to C1,σ (Ω)3 for some σ > 0. Now, as w j belongs in
particular to V , we can take the scalar product of both sides of (5.2.36) by w j and
we obtain
(F m (t) − f (t), w j ) = (vm (t), w j ) + α(∇ vm (t), ∇ w j ).
Substituting this expression into (5.2.35), multiplying the resulting equation by λ j

and using (5.2.27), yields for any j with 1 ≤ j ≤ m,
d
((um (t), w j )) + ((vm (t), w j )) = 0.
dt
In order to see that this is indeed a discretization of the transport equation (5.2.25),
let us expand the above scalar products and compare with (5.2.35); we obtain, for
1 ≤ j ≤ m,
(z m (t), curl(w j − α Δ w j )) + (curl(vm (t) − α Δ vm (t)), curl(w j − α Δ w j )) = 0.
Then (5.2.33) follows easily by substituting into this equation the curl of (5.2.36).
The proof is now complete.
5.2.4 A Priori Estimates
The a priori estimates of this section are derived for the Galerkin solution of (5.2.31),
(5.2.32), but the statement of the first lemma is valid for any choice of sufficiently
smooth basis functions. To simplify, we introduce the following norm for any v in
H 1 (Ω):
v2α = v2L 2 (Ω) + αv2H 1 (Ω) . (5.2.37)
0
Lemma 5.2.10 Let f be given in L 2 (Ω×]0, T [)3 and ut 0 in V α . Then any solution
um of (5.2.31)–(5.2.32) satisfies the a priori estimate for all t in [0, Tm∗ ],
2 t
S0,2
um (t)2α ≤ e−μ K t ut 0 2α + e−μ K (t−s) f (s)2L 2 (Ω) ds, (5.2.38)
μ 0
where S0,2 is the constant of Poincaré inequality, i.e., (3.1.5) with p = 2, and
1
K = . (5.2.39)
α + S0,2
2
Proof Let t ∈]0, Tm∗ ] and take s ∈]0, t[. By multiplying (5.2.31) with c j,m , summing
over j and dropping the index m for simplification, we obtain a.e. in ]0, t[,
1 d
u(s)2α + μ u(s)2H 1 (Ω) = ( f (s), u(s)). (5.2.40)
2 dt 0
On one hand, for any ε > 0, we have
|( f (s), u(s))| ≤ f (s) L 2 (Ω) S0,2 u(s) H01 (Ω)

2
1 S0,2
≤ ε u(s)2H 1 (Ω) + f (s)2L 2 (Ω) .
2 0 ε
Choosing ε = μ and substituting into (5.2.40) gives

2
d S0,2
u(s)2α + μ u(s)2H 1 (Ω) ≤ f (s)2L 2 (Ω) . (5.2.41)
dt 0 μ
On the other hand, (3.1.5) yields
1
u(s)2α ≤ u(s)2H 1 (Ω) ,
K 0
where K is defined by (5.2.39). Therefore, substituting into (5.2.41), we obtain

2
d S0,2
u(s)2α + μ K u(s)2α ≤ f (s)2L 2 (Ω) . (5.2.42)
dt μ
Then (5.2.38) is derived by multiplying both sides of (5.2.42) by eμ K s , integrating

with respect to s from 0 to t, and applying the first part of (5.2.30).
The second a priori estimate is derived from (5.2.33) and relies on the following
Green formula, valid for all functions z ∈ H 1 (Ω) and u ∈ W :

1
(u · ∇ z)z d x = u · ∇(|z|2 ) d x
Ω 2 Ω

1 1
=− (div u)|z|2 d x + (u · n)|z|2 ds = 0. (5.2.43)
2 Ω 2 ∂Ω
The integrals in the second line vanish because u belongs to W .
Proposition 5.2.11 Let f be given in L 2 (0, T ; H (curl, Ω)) and ut 0 be given in V α .

Under the assumptions of Lemma 5.2.5, any solution um of (5.2.31)–(5.2.32) is such
that the quantity ζ := z m 2L 2 (Ω) , with
z m (t) = curl(um (t) − αΔ um (t)),
satisfies the differential inequality a.e. in ]0, Tm∗ [,
α ζ (t) + μ ζ (t) − 2 α γσ,α ζ 2 (t) ≤ G m (t),

3
(5.2.44)
where for a.e. t ∈]0, Tm∗ [,

2 μ −μ K t 2
2 S0,2 t
0 ≤ G m (t) ≤ e ut 0 2α + e−μ K (t−s) f (s)2L 2 (Ω) ds
α α 0
2 α2
+ curl f (t)2L 2 (Ω) , (5.2.45)
μ
and γσ,α is the constant of (5.2.21).
Proof By multiplying (5.2.33) with c j,m , summing over j, dropping the index m,
applying (5.2.43) that is valid here in view of Lemma 5.2.8, and using (5.2.21), we
obtain for any ε > 0 and λ > 0,
α d
z(t)2L 2 (Ω) + μz(t)2L 2 (Ω) ≤ α∇ u(t) L ∞ (Ω) z(t) L 2 (Ω)
2 dt
+ μcurl u(t) L 2 (Ω) + αcurl f (t) L 2 (Ω) z(t) L 2 (Ω)
μ 1
≤ α γσ,α z(t)3L 2 (Ω) + ε z(t)2L 2 (Ω) + curl u(t)2L 2 (Ω)
2 ε
1 α 2
+ μλ z(t)2L 2 (Ω) + curl f (t)2L 2 (Ω) .
2 μλ
Hence setting ζ (t) = z(t)2L 2 (Ω) , the choice ε = λ = 1

2
yields that ζ satisfies the
differential inequality
α μ 3
ζ (t) + ζ (t) − α γσ,α ζ 2 (t) ≤ μcurl u(t)2L 2 (Ω)
2 2
α2
+ curl f (t)2L 2 (Ω) . (5.2.46)
μ
Therefore, defining the function G m a.e. in ]0, Tm∗ [ by
α2
G m (t) := 2 μcurl u(t)2L 2 (Ω) + curl f (t)2L 2 (Ω) ,
μ
we obtain trivially (5.2.44); (5.2.45) follows from
curl u(t) L 2 (Ω) ≤ u(t) H01 (Ω) ,
which holds because u belongs to H01 (Ω)3 (see (3.1.22)), and from the substitution
of (5.2.38) into (5.2.46).
We adopt the assumptions of Proposition 5.2.11. The constants in the left-hand

side of (5.2.44) are independent of t and m and owing to (5.2.45), the assumptions on
f imply that the upper bound for G m belongs to L 1 (0, T ), with norm independent of
m. Thus any positive W 1,1 solution of the Riccati-type differential inequality (5.2.44)
with initial value z(0)2L 2 (Ω) exists over a time interval [0, T ∗ ], with T ∗ > 0 inde-
pendent of m. By construction, z m 2L 2 (Ω) belongs to W 1,1 (0, Tm∗ ), but since it satisfies
(5.2.44), Proposition 5.2.11 shows that in fact Tm∗ = T ∗ and z m L ∞ (0,T ∗ ;L 2 (Ω)3 ) is
bounded uniformly with respect to m; consequently, by virtue of Lemma 5.2.5, um
3
is bounded uniformly with respect to m in L ∞ (0, T ∗ ; H 2 (Ω) ∩ C1,σ (Ω) ).
Finally, the third a priori estimate given below is derived for the Galerkin solution
of (5.2.31), (5.2.32) and is independent of the choice of basis functions.
Proposition 5.2.12 We retain the notation of Proposition 5.2.11. Under the assump-
tions of Lemma 5.2.10, the time derivative um of the solution um of (5.2.31)–(5.2.32)
satisfies the following a priori estimate for all t in [0, T ∗ ]:
um 2L 2 (Ω×]0,t[) + α um 2L 2 (0,t;H 1 (Ω)3 ) + μum (t)2H 1 (Ω)
0 0
1
≤ 2 (S0,3 S0,6 )2 z m 2L ∞ (0,t;L 2 (Ω)3 ) (1 − e−μ K t )ut 0 2α
α μK
2 t
S0,2
+ (1 − e−μ K (t−s) ) f (s)2L 2 (Ω) ds
μ 0
+ μut 0 2H 1 (Ω) + f 2L 2 (Ω×]0,t[) . (5.2.47)
0
Proof Let t ∈]0, T ∗ ] and s ∈]0, t[. By multiplying (5.2.31) with cj,m , summing over
j and dropping the index m, we easily derive
d
u (s)2α + μ u(s)2H 1 (Ω) ≤ f (s)2L 2 (Ω)
dt 0
1
+ (S0,3 S0,6 )2 z(s)2L 2 (Ω) u(s)2H 1 (Ω) .
α 0
Then (5.2.47) follows by substituting (5.2.38) into this inequality and integrating
with respect to s from 0 to t.
The next theorem collects these estimates.
Theorem 5.2.13 Let Ω be a bounded, simply-connected open set of IR 3 that is
either a convex polyhedron or has a boundary ∂Ω of class C1,1. Let f be given
in L 2 (0, T ; H (curl, Ω)) and ut 0 be given in V α .
Then there exists a time T ∗ with 0 < T ∗ ≤ T , and constants C1 , C2 , C3 , depending
only on Ω, α, μ, ut 0 , and f such that for all m, the Galerkin solution um of (5.2.31)–
(5.2.32) exists on the time interval [0, T ∗ ] and is uniformly bounded with respect to
m,
um L ∞ (0,T ∗ ;(H 2 (Ω)∩C1,σ (Ω))3 ) ≤ C1 ,
z m L ∞ (0,T ∗ ;L 2 (Ω)3 ) ≤ C2 , (5.2.48)
um L 2 (0,T ∗ ;H 1 (Ω)3 ) ≤ C3 ,
where σ > 0 depends only on ∂Ω.

An a priori estimate for the pressure cannot be directly deduced from the Galerkin
system (5.2.31)–(5.2.32) because its basis functions are divergence-free. The pres-
sure solution and its estimate will be recovered after passing to the limit.
5.2.5 Passing to the Limit
Theorem 5.2.13 readily implies that there exists a function u ∈ L ∞ (0, T ∗ ; V α ) and
a subsequence m k , such that
lim um k = u weakly in L ∞ (0, T ∗ ; V α ),

k→∞
lim um k = u weakly in L 2 (0, T ∗ ; V ).
k→∞
As V α is continuously imbedded into H 2 (Ω)3 (see Lemma 5.2.5), then its imbedding
is compact into W 1,3 (Ω)3 ∩ V , for instance. Therefore, Theorem 3.3.14 applied with
X = V α , E = W 1,3 (Ω)3 ∩ V , Y = V , and q = 2 implies
lim um k = u strongly in L 2 (0, T ∗ ; W 1,3 (Ω)3 ∩ V ).

k→∞
Theorem 5.2.14 Under the assumptions of Theorem 5.2.13, the limit function u
satisfies the same bounds as um
u L ∞ (0,T ∗ ;V α ) ≤ C2 , u L 2 (0,T ∗ ;H 1 (Ω)3 ) ≤ C3 . (5.2.49)
Moreover, a function p ∈ L 2 (Ω×]0, T ∗ [) can be associated with u so that the

pair (u, p) is a solution of problem (5.2.24), (5.2.4), and (5.2.7) and there exists a
constant C4 , depending only on the constant K̃ of (3.2.12) and the Sobolev imbedding
constants S0,2 and S0,6 such that

p L 2 (Ω×]0,T ∗ [) ≤ C4 u L 2 (Ω×]0,T ∗ [) + z × u L 2 (0,T ∗ ;L 65 (Ω)3 )

+ f L 2 (Ω×]0,T ∗ [) . (5.2.50)
Proof First, the above weak convergences imply immediately (5.2.49). Next, to sim-
plify, we drop the index k and still denote the subsequence by m. Choose a fixed
integer j0 , 1 ≤ j0 ≤ m and let ϕ be an arbitrary function in H 1 (0, T ∗ ). The system
(5.2.31) implies
T∗ T∗
(um (t), w j0 )ϕ(t) dt + α (∇ um (t), ∇ w j0 )ϕ(t) dt
0 0
T∗
+ (curl(um (t) − α Δ um (t)) × um (t), w j0 )ϕ(t) dt
0
T∗ T∗
+μ (∇ um (t), ∇ w j0 )ϕ(t) dt = ( f (t), w j0 )ϕ(t) dt.
0 0
Let m tend to infinity. The above convergences (the strong convergence is used in
the nonlinear term) readily imply that the limit function u satisfies
T∗ T∗

(u (t), w j0 )ϕ(t) dt + α (∇ u (t), ∇ w j0 )ϕ(t) dt
0 0
T∗
+ (curl(u(t) − α Δ u(t)) × u(t), w j0 )ϕ(t) dt
0
T∗ T∗
+μ (∇ u(t), ∇ w j0 )ϕ(t) dt = ( f (t), w j0 )ϕ(t) dt.
0 0
As j0 is arbitrary and the set {w j } j≥1 is a basis for V α , we easily recover a.e. in
]0, T ∗ [
∀v ∈ V α , (u (t), v) + α (∇ u (t), ∇ v) + μ (∇ u(t), ∇ v) + (z(t) × u(t), v)

= ( f (t), v), (5.2.51)
with z defined by (5.2.7). But V α is dense in V . Therefore, (5.2.51) holds for all
v ∈ V . Furthermore, (5.2.32) and the continuity of u in time imply that u satisfies
the initial condition (5.2.4).
It remains to recover the pressure. For this, observe that
u − α Δ u − μ Δ u + z × u − f
belongs to L 2 (0, T ∗ ; H −1 (Ω)3 ) and (5.2.51) is equivalent to the following equation:
∀v ∈ V,
u (t)−α Δ u (t)−μ Δ u(t)+ z(t)× u(t)− f (t), v = 0 a.e. in ]0, T ∗ [,
where
·, · denotes the duality pairing between H −1 (Ω)3 and H01 (Ω)3 .
Then Theorem 3.2.11 and the Babuška–Brezzi Theorem 3.2.12 part 2, yield the
existence of a function p in L 2 (Ω×]0, T ∗ [) such that for a.e. t in ]0, T ∗ [,
∀v ∈ H01 (Ω)3 (u (t), v) + α (∇ u (t), ∇ v) + μ (∇ u(t), ∇ v)

+ (z(t) × u(t), v) − ( p(t), div v) = ( f (t), v),
which is precisely (5.2.24) on ]0, T ∗ [. Finally, (5.2.50) follows from the preceding
equation and (3.2.15) with β = K̃ by choosing v in V ⊥ , solution of div v = p.
Remark 5.2.15 Note that (5.2.50) can be replaced by a sharper estimate where f and
u are measured in L 2 (0, T ∗ ; H −1 (Ω)3 ). But even in the weaker form presented here,
it is much sharper than what can be established for the Navier–Stokes equations. The
difference comes from the fact that deriving a bound for u in L 2 (0, T ; H −1 (Ω)3 ) is
much harder for the Navier–Stokes equations.
5.2.6 Uniqueness
In this section, we retain the assumptions of Theorem 5.2.13. Suppose that problem
(5.2.24), (5.2.4) and (5.2.7) has two solutions, say u1 and u2 , on a time interval
[0, T ∗ ] with 0 < T ∗ ≤ T . Set w = u1 − u2 . Then an easy calculation shows that w,
u1 and u2 satisfy a.e. in ]0, T ∗ [,
∀v ∈ V, (w (t), v)+α (∇ w (t), ∇ v) + μ (∇ w(t), ∇ v)

+ (curl(w(t) − α Δ w(t)) × u1 (t), v)
+ (curl(u2 (t) − α Δ u2 (t)) × w(t), v) = 0, (5.2.52)
w(0) = 0. (5.2.53)
The choice v = w gives a.e. in ]0, T ∗ [

1 d
w(t)2α +μ w(t)2H 1 (Ω) = −(curl(w(t)−α Δ w(t))× u1 (t), w(t)). (5.2.54)
2 dt 0
Clearly, we must find an estimate for this right-hand side that involves at most first-
order derivatives of w. This makes use of the following lemma:
Lemma 5.2.16 All w and u in V α satisfy the identity
3
∂u ∂w ∂∇ u ∂w
(curl(Δ w) × u, w) = [∇ w] − [∇ u] −[ ]w, . (5.2.55)
k=1
∂ xk ∂ xk ∂ xk ∂ xk
Proof First note that, since w and u are both in V α , all terms in (5.2.55) are well-
defined and all operations below are justified. Following [28], any vector fields u
and v satisfy formally the identity
∂u
(curl u × v)i = [∇ u]v i − v · .
∂ xi
Hence
(curl(Δ w) × u, w) = ([∇(Δ w)]u, w) − ([∇(Δ w)]w, u).
In other words, recalling the familiar trilinear form of the Navier–Stokes operator,
see (4.2.18) and (4.2.13),
c(u; v, w) = ([∇ v]u, w), (5.2.56)
we have
(curl(Δ w) × u, w) = c(u; Δ w, w) − c(w; Δ w, u). (5.2.57)
Recall that c satisfies the analogue of (5.2.43) for all u ∈ W and for all v ∈ H 1 (Ω)3
c(u; v, v) = 0. (5.2.58)
This antisymmetry and Green’s formula yield
3
∂ ∂w
c(w; Δ w, u) = −c(w; u, Δ w) = [∇ u]w , .
k=1
∂ xk ∂ xk
By expanding, this gives
3
∂w ∂w ∂ u ∂w
c(w; Δ w, u) = c ; u, + c w; , . (5.2.59)
k=1
Similarly,
3
∂ ∂w
c(u; Δ w, w) = −c(u; w, Δ w) = [∇ w]u , ,
k=1
∂ xk ∂ xk
and by expanding and using again (5.2.58), we obtain
3 ∂u ∂w
c(u; Δ w, w) = c ; w, . (5.2.60)
k=1
∂ xk ∂ xk
Then (5.2.55) follows by substituting (5.2.59) and (5.2.60) into (5.2.57).
Because of the third term in the right-hand side of (5.2.55), the second derivative
of u1 must be bounded in L 3 (Ω)3 for deriving uniqueness; this condition must be
added to the assumptions. It holds for instance if ∂Ω is smooth or if its angles are
suitably restricted.
Theorem 5.2.17 In addition to the assumptions of Theorem 5.2.13, we suppose that

problem (5.2.24), (5.2.4) and (5.2.7) has a solution u in L 1 (0, T ∗ ; W 2,3 (Ω)3 ). Then
this problem has no other solution on [0, T ∗ ].
Proof We adopt the notation of (5.2.54) and drop the superscript on u. Lemma 5.2.16
implies that for a.e. t in [0, T ∗ ],

(curl(w(t) − α Δ w(t)) × u(t), w(t)) ≤ w(t)2 1 S0,3 S0,6 u(t) H01 (Ω)
H0 (Ω)

+ 2α ∇ u(t) L ∞ (Ω) + α S0,6 |u(t)|W 2,3 (Ω) ,
that we substitute into the left-hand side of (5.2.54). Then, integrating the resulting
differential inequality from 0 to t ≤ T ∗ , and using the fact that w(0) = 0, we derive
in particular,
t
∀t ∈ [0, T ∗ ] αw(t)2H 1 (Ω) ≤ 2 w(s)2H 1 (Ω) κ(s)ds,
0 0
0
where κ(t) is the factor of w(t)2H 1 (Ω) in the right-hand side of the equation above.
0
Owing to the first estimate in (5.2.49) and to the assumptions, the nonnegative func-
tion κ belongs to L 1 (0, T ). Then Gronwall’s Lemma 3.3.13 implies that w = 0 on
[0, T ∗ ].
Remark 5.2.18 Formula (5.2.55) is not the only expression for (curl(Δ w) × u, w).
For example, by keeping the curl operator, instead of expanding it, we can show that
3
∂u ∂w
(curl(Δ w) × u, w) = (curl w, Δ u × w) + 2 curl w, ×
i=1
∂ xi ∂ xi
+ ([∇ u]curl w, curl w). (5.2.61)
In this case, the assumption for uniqueness on the second derivatives of u can be
somewhat relaxed: it suffices that Δ u belong to L 2 (0, T ; L 3 (Ω)3 ).
5.2.7 Global Existence in Time
So far, we have only constructed a local in time solution. In view of the Riccati
inequality (5.2.44), it is clear that we cannot prove existence of a global solution
unless we show the following uniform bound for the Galerkin solution:
μ
∀t ∈ [0, T ], curl(um − α Δum )(t) L 2 (Ω) ≤ . (5.2.62)
2 α γσ,α
Obviously, such a bound cannot be achieved without restriction on the data. The
next theorem states sufficient conditions for global existence in time. To simplify,
we introduce the notation
21
M = S0,2
2
f 2L 2 (Ω×]0,T [) + α 3 K curl f 2L 2 (Ω×]0,T [) , (5.2.63)
where K is defined in (5.2.39).
Theorem 5.2.19 In addition to the hypotheses of Theorem 5.2.13, assume that the
domain Ω and the data α, μ, ut 0 and f satisfy: There exists θ in ]0, 1[ such that
1 2 μ2
ut 0 2α + M < θK ,
μ 8 γσ,α
2
(5.2.64)
μ 2
curl(ut 0 − α Δut 0 )2L 2 (Ω) < (1 − θ ) .
2 α γσ,α
Then the solution um of (5.2.31)–(5.2.32) exists on [0, T ] and satisfies (5.2.62).
Proof With the notation of Proposition 5.2.11, let us prove that any nonnegative
continuous solution ζ of the Riccati differential inequality (5.2.44) with starting
value z m (0)2L 2 (Ω) exists on [0, T ] and satisfies
μ 2
∀t ∈ [0, T ], 0 ≤ ζ (t) < .
2 α γσ,α
As z m 2L 2 (Ω) is a nonnegative continuous solution of (5.2.44), this will imply global
existence of um .
The choice (5.2.27) of basis functions and the choice of Galerkin initial data um (0)
imply that by virtue of (5.2.30),
z m (0) L 2 (Ω) ≤ curl(ut 0 − α Δut 0 ) L 2 (Ω) .
Hence by assumption,
μ 2
0 ≤ ζ (0) < ,
2 α γσ,α
and by continuity, this is true for ζ (t) in some interval [0, t˜], with t˜ > 0.
To show that t˜ = T , we proceed by contradiction. Suppose there exists t ∗ > 0
such that
μ 2 μ 2
∀t < t ∗ , 0 ≤ ζ (t) < and ζ (t ∗ ) = . (5.2.65)
2 α γσ,α 2 α γσ,α
Then on [0, t ∗ ], the expression μ ζ (t) − 2 α γσ,α ζ 2 (t) is nonnegative and α ζ is

3
bounded by the right-hand side of (5.2.45). Thus, integrating in time from 0 to t ∗ ,

we obtain
t∗
2μ
α ζ (t ∗ ) ≤ α ζ (0) + ut 0 2α e−μ K s ds
α 0
2 t∗ τ
2 S0,2 2 α2
+ e−μ K (τ −s) f (s)2L 2 (Ω) ds dτ + curl f 2L 2 (Ω×]0,t ∗ [)
α 0 0 μ
2 ∗
= α ζ (0) + (1 − e−μ K t )ut 0 2α
αK
2 ∗
2 S0,2 t ∗
+ f (s)2L 2 (Ω) (1 − e−μ K (t −s) )ds + α 2 curl f 2L 2 (Ω×]0,t ∗ [) .
μ αK 0
Bounding the terms involving exponentials by one and applying (5.2.64) yields
μ 2
ζ (t ∗ ) < ,
2 α γσ,α
thus contradicting the above equality.
In this case, the arguments of Sections 5.2.5 and 5.2.6 prove that problem (5.2.24),
(5.2.4) and (5.2.7) has a global solution on [0, T ], unique if Δ u is smoother.
Of course, the conditions (5.2.64) are only satisfied when the data f and ut 0 are
sufficiently small, or when α is small and μ is large. Finally, if f belongs to
L 2 (IR + ; H (curl, Ω)), then all constants involved in (5.2.64) are independent of T and
consequently problem (5.2.24), (5.2.4) and (5.2.7) has a global solution on [0, +∞].

Condition. The Case of a Multiply-Connected Domain
Simple connectedness of the domain plays a fundamental part in the proof of Lemma
5.2.5. More precisely, in deriving (5.2.22), we use the fact that on a simply-connected
domain, the solution v of the equation
curl v = 0,
is necessarily of the form v = ∇ q, and this particular form enables us to use the
regularity of a Stokes problem. This is no longer true in a multiply-connected domain,
but of course it holds in any simply-connected region of the domain. Therefore, we
propose to revert to simply-connected regions by covering the domain with simply-
connected subdomains and using a partition of unity to recover a homogeneous
Dirichlet boundary condition. To this end, we make the following assumption.
Hypothesis 5.3.1 The domain Ω has a finite open covering
Ω ⊂ ∪i=1
N
Oi ,
such that the intersections Ωi = Oi ∩Ω, 1 ≤ i ≤ N , are simply-connected Lipschitz

open sets.
This assumption is satisfied by all the examples we have in mind.
Lemma 5.3.2 Let Ω be multiply-connected and satisfy Hypothesis 5.3.1; in addi-

tion, we suppose that either ∂Ω is of class C1,1 or if part of it has corners, these are
3
not reentrant. Then any v in V α belongs to H 2 (Ω) ∩ C1,σ (Ω) for some σ > 0
depending on ∂Ω and there exists a constant γ̃α depending only on α and Ω and a
constant γ̃σ,α depending only on σ , α, and Ω such that

∀v ∈ V α , v H 2 (Ω) ≤ γ̃α curl(v − αΔ v) L 2 (Ω) + v H01 (Ω) , (5.3.1)
and

∀v ∈ V α , vC1,σ (Ω) ≤ γ̃σ,α curl(v − αΔ v) L 2 (Ω) + v H01 (Ω) . (5.3.2)
Proof Let v belong to V α and set
y = curl(v − αΔ v).
Arguing as in the proof of Lemma 5.2.5, there exists ϕ ∈ H 1 (Ω)3 such that
y = curl ϕ , div ϕ = 0 in Ω , ϕ H 1 (Ω) ≤ ccurl(v − αΔ v) L 2 (Ω) ,

with the constant c of (5.2.17). Thus
curl(αΔ v − v + ϕ) = 0,
but as Ω is not simply-connected, we cannot conclude that v − αΔ v + ∇ p = ϕ.

Instead, we use the covering Oi , 1 ≤ i ≤ N , guaranteed by Hypothesis 5.3.1. Let
ψi ∈ D(Oi ), 1 ≤ i ≤ N , be a partition of unity subordinated to this covering, i.e.,

N
∀x ∈ Ω, ψi (x) = 1 and 0 ≤ ψi (x) ≤ 1, 1 ≤ i ≤ N .
i=1
Let i =
Oi be an adequate subset of Oi containing the support of ψi and set Ω Oi ∩Ω.
There is no loss of generality in assuming that Ω i is simply-connected and that its
boundary has the same regularity as that of Ω. Therefore for each i, 1 ≤ i ≤ N ,
i ) such that
there exists a function q ∈ L 20 (Ω
i .
v − αΔ v + ∇ q = ϕ , div v = 0 in Ω (5.3.3)
Furthermore,

∇ q H −1 (Ω i ) ≤ Ci ϕ L 2 (Ω i ) + v L 2 (Ω i ) + αv H01 (Ω i ) ,
where Ci is the Poincaré constant of Ω i . Therefore, the gradient isomorphism (see

Theorem 3.2.12, Part 2) implies that there exists another constant Ci depending only
i such that
on Ω

q L 2 (Ω i ) ≤ Ci y L 2 (Ω) + v L 2 (Ω i ) + αv H01 (Ω i ) . (5.3.4)
Next, multiply the first equation of (5.3.3) by ψi , set vi = ψi v and qi = ψi q;

observe that this makes sense and vi , qi have the same regularity, respectively, as v,
q because ψi ∈ C∞ (Ω i ). Then a standard manipulation yields
i ,
vi − αΔ vi + ∇ qi = F i , div vi = gi in Ω (5.3.5)
and
i ,
vi = 0 on ∂ Ω
where
3
∂ψi ∂v
F i = ψi ϕ − 2α − αΔ ψi v + q∇ ψi ,
j=1
∂x j ∂x j
gi = v · ∇ ψi .
The above boundary condition on vi holds because either v = 0 or ψi = 0 on

∂Ω i . Therefore, the pair (vi , qi ) solves a Stokes problem in Ω i , and the fact that
v ∈ H 1 (Ω i )3 , q ∈ L 2 (Ω
i ) implies that F i ∈ L 2 (Ω
i )3 and gi ∈ H 1 (Ω i ). More
i ) because either v = 0 or ∇ ψi = 0 on ∂ Ω
precisely, gi ∈ H01 (Ω i . Then we can use
Theorem 3.2.19 to revert to a zero divergence condition, and the regularity of ∂ Ω i
i )3 and qi ∈ H 1 (Ω
implies that vi ∈ H 2 (Ω i ) with

vi H 2 (Ω i ) + qi H 1 (Ω i ) ≤ C F i L 2 (Ω i ) + gi H01 (Ω i )

≤ C y L 2 (Ω) + v H 1 (Ω i ) , (5.3.6)
where we have used (5.3.4) to derive the last bound. On one hand vi is defined in
Ω and it is easy to prove that it belongs to H 2 (Ω)3 with the same norm. On the
other hand, qi can be extended by zero in Ω and it is easy to prove that the extended
N to H (Ω), also with
1
N still denoted qi , belongs
function, the same norm. Recall that
v = i=1 vi , and set q = i=1 qi . We have v ∈ H 2 (Ω)3 , q ∈ H 1 (Ω) and as the
above sums are finite
v H 2 (Ω) + q H 1 (Ω)

N

N

≤ vi H 2 (Ω) + qi H 1 (Ω) ≤ C y L 2 (Ω) + v H 1 (Ω i )
i=1 i=1

≤ C y L 2 (Ω) + v H 1 (Ω) ≤ C y L 2 (Ω) + v H01 (Ω) ,
i )3 and gi ∈ H 2 (Ω
whence (5.3.1). In turn, this implies that F i ∈ H 1 (Ω i ). By
r 3 1,r
Sobolev’s imbedding, F i ∈ L (Ωi ) and gi ∈ W0 (Ωi ) for any r ≤ 6, with

F i L r (Ω i ) ≤ C y L 2 (Ω) + vW 1,r (Ω i ) + q L r (Ω i ) ,
(5.3.7)
gi W 1,r (Ω i ) ≤ CvW 1,r (Ω i ) .
We apply again Theorem 3.2.19 in order to use zero divergence. Then the regularity
i and Corollary 3.2.16 imply that vi ∈ C1,σ (Ω
of ∂ Ω i )3 and qi ∈ C0,σ (Ω i ) for some

σ ∈]0, 2 ] depending on the regularity of ∂ Ωi . Furthermore, in view of (5.3.7) and
1
(5.3.6)

vi C1,σ (Ω i ) + qi C0,σ (Ω i ) ≤ C y L 2 (Ω) + vW 1,r (Ω i ) + qi L r (Ω i )

≤ C y L 2 (Ω) + v H 1 (Ω i ) . (5.3.8)
As above, vi belongs to C1,σ (Ω)3 , qi extended by zero belongs to C0,σ (Ω) each with
the same norm, hence v ∈ C1,σ (Ω)3 , q ∈ C0,σ (Ω) and

N

vC1,σ (Ω) +qC0,σ (Ω) ≤ C y L 2 (Ω) +v H 1 (Ω i ) ≤ C y L 2 (Ω) +v H01 (Ω) ,
i=1
which is (5.3.2).
Remark 5.3.3 In general, the term |v| H 1 (Ω) in the right-hand sides of (5.3.1) and
(5.3.2) cannot be eliminated. As an example, consider the case when Ω is a smooth
torus, Σ a smooth plane that cuts Ω so that Ω ◦ = Ω\Σ is connected and simply-
connected (Σ is called a cut). Choose a unit normal n on Σ. Let ϕ ∈ L 2 (Ω)3
satisfy
curl ϕ = 0 in Ω. (5.3.9)
Since Ω ◦ is simply-connected, there exists q ∈ H 1 (Ω ◦ ) such that
ϕ = ∇ q a.e. in Ω ◦ .
As ∇ q ∈ L 2 (Ω ◦ )3 , it can be extended to Ω; we denote the extended function by

q and ∇
∇ q belongs to L 2 (Ω)3 . Furthermore,
q a.e. in Ω.
ϕ=∇
Now, for a sufficiently smooth function r , let [r ]Σ denote the jump of r through Σ
in the direction of n, and define the space
Θ = {r ∈ H 1 (Ω ◦ ); [r ]Σ is constant}.
r ) = 0.
In Lemma 3.11 of Reference [6], it is proven that r ∈ Θ if and only if curl(∇
Applying this result to q, we see that (5.3.9) implies that
q
ϕ=∇ a.e. in Ω , q ∈ Θ. (5.3.10)
Reverting to V α , let v ∈ V α satisfy
curl(v − α Δ v) = 0.
The above considerations show that there exists q ∈ Θ such that
q
v − αΔv = ∇ a.e. in Ω.
But in contrast to the case of a simply-connected domain, this does not mean that
v = 0. In fact, simple applications of Green’s formula yield

v2α = (v · n)[q] ds = c (v · n) ds,
Σ Σ
and this integral does not necessarily vanish.

Because of this extra lower order term, the stability estimates in Lemma 5.3.2
are not as favorable as those of Lemma 5.2.5, but the regularity results are the same.
Therefore, under the assumptions of Lemma 5.3.2, the results of Sections 5.2.3–5.2.7
for a simply-connected domain are also valid for a multiply-connected domain, up to
two modifications. The first one concerns the a priori estimates for z, and the second
one concerns global existence in time. More precisely,
1. Proposition 5.2.11 is replaced by
Proposition 5.3.4 Let f be given in L 2 (0, T ; H (curl, Ω)) and ut 0 be given in V α .

Under the assumptions of Lemma 5.3.2, any solution um of (5.2.31)–(5.2.32) is such
that the quantity ζ := z m 2L 2 (Ω) , with
z m (t) = curl(um (t) − αΔ um (t)),
satisfies the differential inequality a.e. in ]0, Tm∗ [,

√
α ζ (t) + μ − 2 γ̃σ,α α M̃ ζ (t) − 2 α γ̃σ,α ζ 2 (t) ≤ G m (t),
3
(5.3.11)
where for a.e. t ∈]0, Tm∗ [, G m (t) is bounded as in (5.2.45), γ̃σ,α is the constant of
(5.3.2), and
2
S0,2 21
M̃ = ut 0 2α + f 2L 2 (Ω×]0,T [) . (5.3.12)
μ
Proof As in the proof of Proposition 5.2.11, we obtain
d
α z(t)2L 2 (Ω) + μz(t)2L 2 (Ω) ≤ 2 α∇ u(t) L ∞ (Ω) z(t)2L 2 (Ω)
dt
α2
+ 2 μu(t)2H 1 (Ω) + 2 curl f (t)2L 2 (Ω) .
0 μ
Applying (5.3.2), the first term in the above right-hand side is bounded by
2 α γ̃σ,α z(t)3L 2 (Ω) + 2 α γ̃σ,α u(t) H01 (Ω) z(t)2L 2 (Ω) .
So, by taking the maximum with respect to t of estimate (5.2.38) for |u(t)|2H 1 (Ω) ,
substituting above, and using the definition of M̃, we easily derive (5.3.11).
2. Theorem 5.2.19 is replaced by
Theorem 5.3.5 Let Ω be multiply-connected, satisfy Hypothesis 5.3.1, and either

∂Ω is of class C1,1 or if part of it has corners, these are not reentrant. Let f be given
in L 2 (0, T ; H (curl, Ω)) and ut 0 be given in V α . Moreover, assume that the domain
Ω and the data α, μ, ut 0 , and f satisfy
μ
M̃ ≤ √ , (5.3.13)
4 α γ̃σ,α
and there exists a number θ in ]0, 1[ such that

1 2 μ2
ut 0 2α + M < θK ,
μ 32 γ̃σ,α
2
μ 2
curl(ut 0 − α Δut 0 )2L 2 (Ω) < (1 − θ ) , (5.3.14)
4 α γ̃σ,α
where M is defined by (5.2.63). Then the solution um of (5.2.31)–(5.2.32) exists on

[0, T ] and satisfies
μ
∀t ∈ [0, T ], curl(um − α Δum )(t) L 2 (Ω) ≤ . (5.3.15)
4 α γ̃σ,α
Proof By substituting the bound (5.3.13) for M̃ into (5.3.11), we obtain

μ
α ζ (t) + ζ (t)
1
− 2 α γ̃σ,α ζ 2 (t) ≤ G m (t).
2
From there the argument is the same as in proving Theorem 5.2.19.
Comparing with the conditions in the statement of Theorem 5.2.19, conditions

(5.3.13) and (5.3.14) restrict more strongly the data. But nevertheless, if f belongs
to L 2 (IR + ; H (curl, Ω)) and these conditions are satisfied, then problem (5.2.24),
(5.2.4) and (5.2.7) still has a global solution on [0, +∞].

Condition Prescribed. The Case of a Convex Corner
Domain
In this section, we consider problem (5.2.1)–(5.2.2)
∂
(u − αΔ u) − μ Δ u + curl(u − αΔ u) × u + ∇ p = f in Ω×]0, T [,
∂t
div u = 0 in Ω×]0, T [,
with a tangential Dirichlet boundary condition
u = g on ∂Ω with g · n = 0 on ∂Ω, (5.4.1)
and initial condition (5.2.4)

u(0) = ut 0 in Ω,
satisfying the compatibility condition (5.2.5)
div ut 0 = 0 in Ω and ut 0 = g(0) on ∂Ω.

5.4 Flow of a Grade-Two Fluid with Tangential Boundary Condition … 209
The reason for treating only the tangential boundary condition (5.4.1) is that this
grade-two fluid model with a fully nonhomogeneous Dirichlet boundary condition
is not always well posed. There are examples in which the resulting problem has
multiple solutions (cf. Section 2.5, the work of Rajagopal and Gupta [230] and of
Rajagopal and Kaloni [232]), thus implying that additional boundary conditions
should be imposed. However, it is not yet known what boundary conditions could be
prescribed in order to ensure that the problem is well posed. The reader can refer to
the publication of Rajagopal [226] for a discussion on the boundary conditions for
such fluids.
It is well known that the analysis of a boundary value problem with a nonhomoge-
neous Dirichlet boundary condition relies on a suitable lifting of the boundary data
g, but constructing a good lifting is complicated by the incompressibility condition
(5.2.2) and the high nonlinearity. In this section, we shall analyze the nonhomoge-
neous grade-two fluid model in a convex corner domain by means of a straightforward
lifting used in the Stokes problem. In the next section, we will study a Leray–Hopf’s
lifting, see Leray [160], Hopf [132], or for instance [116, Chapter IV, Lemma 2.3]
or Temam [265], that is used in the Navier–Stokes system in a Lipschitz domain.
Its extension to problem (5.2.1), (5.2.2), (5.2.4), (5.4.1) in a domain with corners
is problematic because of the higher order derivatives and therefore we shall only
present it in smooth domains.
In view of the second condition of (5.4.1), we define
W α = {v ∈ W ; α curl Δ v ∈ L 2 (Ω)3 }, (5.4.2)
where W is defined in (3.1.18). It is a Hilbert space equipped with the norm

1
vW α = (v2H 1 (Ω) + α curl Δ v2L 2 (Ω) ) 2 , (5.4.3)
0
where we have used (3.1.13) to eliminate the L 2 norm of the function. Of course,
when α = 0, W α reduces to W .
5.4.1 A Lifting; Reduction to Homogeneous Boundary

Conditions
Note that in general, the functions of W α do not share the regularity of those of V α
stated in Lemma 5.2.5, because the definition of W α does not completely specify
their boundary values. To achieve sufficient regularity for constructing solutions, we
shall make adequate assumptions on the boundary data g.
1
Let g be given in H 1 (0, T ; H 2 (∂Ω)3 ), and let S denote any plane face of ∂Ω. In
addition to
g · n = 0 in ]0, T [,
3
we assume that g ∈ H 1 (0, T ; [H 2 (S)∩C1,σ (S)]3 ) and that at the dihedral and conical
angles of ∂Ω, g satisfies compatibility conditions (see Maz’ya and Rossmann [188])
that ensure that it is the trace of a function u g in H 1 (0, T ; [H 2 (Ω) ∩ C1,σ (Ω)]3 ) that
solves in [0,T],
−Δ u g + ∇ q g = 0, div u g = 0 in Ω, u g = g on ∂Ω, (5.4.4)
where the associated pressure q g belongs to H 1 (0, T ; H 1 (Ω)∩C0,σ (Ω)), both veloc-
ity and pressure depending linearly and continuously on g. Here 0 < σ < 1 depends
on the angles of ∂Ω. Note that
u g (t) = u g(t) , q g (t) = q g(t) , (u g (t)) = u g (t) .
Defining the broken norms

21
[g] 23 = g2 3 ,
H (S)
2
S∈∂Ω
(5.4.5)
[g]1,σ = gC1,σ (S) ,
S∈∂Ω
the continuous dependence of (u g , q g ) on g reads
u g H 1 (Ω) ≤ C1 g H 21 (∂Ω) ,

u g H 2 (Ω) ≤ C2 [g] 23 , (5.4.6)
u g C1,σ (Ω) ≤ C3 [g]1,σ ,
with similar inequalities for q g . Define
z g = curl(u g − αΔ u g ) = curl u g ,
owing to the first part of (5.4.4). Therefore, u g belongs to H 1 (0, T, W α ) for any α,
and z g belongs to H 1 (0, T ; [H 1 (Ω) ∩ C0,σ (Ω)]3 ) with
√ √
z g H 1 (Ω)) ≤ 2C2 [g] 23 , z g C0,σ (Ω) ≤ 2C3 [g]1,σ .
The next straightforward lemma reduces the original problem to one with homo-
geneous boundary conditions; it makes use of the orthogonality properties in (5.4.4).
As usual, we set z = curl(u − α Δ u).
Lemma 5.4.1 Let u solve (5.2.1), (5.2.2), (5.4.1), (5.2.4), and let u g be defined by
(5.4.4). Then the function u0 = u − u g ∈ V α solves a.e. in ]0, T [, for all v ∈ V
(u0 (t), v) + α(∇ u0 (t), ∇ v) + μ(∇ u0 (t), ∇ v) + (z(t) × u0 (t), v)

+ (z(t) × u g(t) , v) = (H(t), v), (5.4.7)
u0 (0) = ut 0 − u g (0), (5.4.8)
where
H = f − u g ,
and u g (0) is the lifting of g(0) defined by (5.4.4).
5.4.2 Formal A Priori Estimates
The a priori estimates in this paragraph are established under the assumption that
a solution exists during the time interval [0, T ] for some 0 < T ≤ T . First, the
argument of the proof of Lemma 5.2.10 readily gives the following result:
1
Lemma 5.4.2 Let f be given in L 2 (Ω×]0, T [)3 , g in H 1 (0, T ; H 2 (∂Ω)3 ), sat-
isfying all assumptions of Section 5.4.1, and ut 0 in W α . Then any solution (u0 , z)
of (5.4.7) in L ∞ (0, T ; H01 (Ω)3 ) × L 2 (Ω×]0, T [)3 satisfies the following a priori
estimate for all t in [0, T ∗ ]:

3 t −μ K (t−s)
u0 (t)2α ≤ e−μ K t ut 0 2α + e
μ 0

× G 1 (s) + S0,62
S̃32 C12 g(s)2 1 z(s)2L 2 (Ω) ds, (5.4.9)
H 2 (∂Ω)
where
G 1 (t) = S0,2
2
f (t)2L 2 (Ω) + C12 g (t)2 1 ,
H (∂Ω)
2
C1 is the constant of the first inequality in (5.4.6), and K is defined by (5.2.39).
Note that by assumption, G 1 belongs to L 1 (0, T ).

The next proposition gives an a priori estimate for z. We set z 0 = curl(u0 −
αΔ u0 ).
Proposition 5.4.3 Let Ω be as in Lemma 5.2.5, and in addition to the assumptions

of Lemma 5.4.2, suppose that f belongs to L 2 (0, T ; H (curl, Ω)). Then any solu-
tion (u0 , z) of (5.4.7)–(5.4.8) in L ∞ (0, T ; V α ) × L 2 (Ω×]0, T [)3 is such that the
quantity
ζ := z2L 2 (Ω)
satisfies the following integro-differential inequality a.e. in ]0, T ∗ [:

t
α ζ (t) + μ ζ (t) − 3α γσ,α ζ 2 (t) − G 4 e−μ K (t−s) ζ (s)ds
3
0
≤ G 2 (t) + G 3 (t), (5.4.10)
where
3α 2
G 2 (t) = 9μC12 g(t)2 1 + curl f (t)2L 2 (Ω)
H 2 (∂Ω) μ

128 C33
+ α γσ,α C1 g(t) 1
3 3
+ 2 [g(t)]1,σ , 3
(5.4.11)
27 H 2 (∂Ω) γσ,α

3μ −μK t 3 t −μ K (t−s)
G 3 (t) = e ut 0 2α + e G 1 (s) ds , (5.4.12)
α μ 0
3 2 2 2
G4 = S S̃ C g2 ∞ 1 . (5.4.13)
μ 0,6 3 1 L (0,T ;H 2 (∂Ω))
Proof It is shown in Section 7.3.3 that, for given u in L 2 (0, T ; W ∩ W 1,∞ (Ω)3 ),
the transport equation (5.2.10) has a unique solution and with the above notation, ζ
satisfies
α
ζ (t) + μ ζ (t) ≤α∇ u(t) L ∞ (Ω) ζ (t)
2
1
+ μcurl u(t) L 2 (Ω) + αcurl f (t) L 2 (Ω) ζ 2 (t).
By writing u = u0 +u g , z = z 0 + z g and applying (5.2.21) and (5.4.6), the inequality

becomes
α 1
ζ (t) + μ ζ (t) ≤ αζ (t) γσ,α ζ 2 (t) + γσ,α z g(t) L 2 (Ω) + C3 [g(t)]1,σ
2
1
+ ζ 2 (t) μcurl u0 (t) L 2 (Ω) + μcurl u g(t) L 2 (Ω)

+ αcurl f (t) L 2 (Ω) . (5.4.14)
By suitably applying Young’s inequality, we can write

128 C3
αζ (t) + μ ζ (t) − 3γσ,α ζ 2 (t) ≤
3
α γσ,α C13 g(t)3 1 + 23 [g(t)]31,σ
27 H 2 (∂Ω) γσ,α
α 2
+ 3 μcurl u0 (t)2L 2 (Ω) + 3μC12 g(t)2 1 + curl f (t)2L 2 (Ω) ,
H 2 (∂Ω) μ
and (5.4.10) follows by substituting (5.4.9) into this inequality.
Note that G 2 belongs to L 1 (0, T ), G 3 belongs to L ∞ (0, T ), and G 4 is constant.

These three quantities depend only on the domain and on the data.
5.4.3 Galerkin Solution and A Priori Estimates
Now, we construct a solution. As problem (5.4.7) is set in V , we discretize it in the

basis (5.2.27): Find

m
u0m (t) = c j,m (t) w j ,
j=1
((u0m ) (t), w j ) + α(∇ (u0m ) (t), ∇ w j ) + μ(∇ u0m (t), ∇ w j )

+ (z m (t) × u0m (t), w j ) + (z m (t) × u g(t) , w j )
= (H(t), w j ), (5.4.15)
u0m (0) = Pm (ut 0 ). (5.4.16)
Recall that Pm is the orthogonal projection on Vm for the scalar product of V α . The
analysis of (5.4.15)–(5.4.16) involves more terms, but is similar to that of (5.2.31)–
(5.2.32), and hence we can skip most details. First, it has a unique solution on an
interval [0, Tm ], for some Tm > 0 depending on the dimension m. Next, it satisfies
the estimate of Lemma 5.4.2 a.e. in ]0, Tm [ with z m instead of z. But the statement
of Proposition 5.4.3 does not apply directly to z m because we cannot take the curl of
(5.4.15). An equation for the curl is recovered by the technique of Cioranescu and
Ouazar. Following the proof of Theorem 5.2.9, we define the function F m (t) for all
t ∈ [0, Tm ] by (5.2.34) with u0 instead of u in the Laplace operator
F m (t) = −μ Δ u0m (t) + z m (t) × um (t).
Then the proof of Theorem 5.2.9 carries over and yields, with η j = curl(w j −
α Δ w j ),

α((z 0m ) (t), η j ) + α [∇ z m (t)]um (t) − [∇ um (t)]z m (t), η j + μ(z 0m (t), η j )
= (μ curl u0m (t) + α curl f (t) − α curl u g (t) , η j ), (5.4.17)
for 1 ≤ j ≤ m.
Of course, we suppose that f belongs to L 2 (0, T ; H (curl, Ω)). Since z 0m is a linear
combination of the functions η j , (in contrast to z m ), we can derive from (5.4.17), by
expanding z m and um , that z 0m (t)2L 2 (Ω) satisfies an integro-differential inequality
in [0, Tm ] of the same form as (5.4.10)
t
α ζ (t) + μ ζ (t) − 3α γσ,α ζ 2 (t) − G 4 e−μ K (t−s) ζ (s)ds ≤ G 2 (t) + G 3 (t),
3
0
but in terms of other quantities G 2 , G 3 , and G 4 . They have the same properties, i.e.,
G 2 is bounded by a function in L 1 (0, T ) and G 3 by a function in L ∞ (0, T ), and
G 4 is constant, with possibly higher bounds since they involve more terms, but their
bounds are independent of m. As all factors of the terms related to ζ in the left-hand
side of (5.4.10) are independent of t and m, and the right-hand side is bounded by
a function in L 1 (0, T ), with bound independent of m, it follows that any positive
W 1,1 solution of the Riccati-type integro-differential inequality (5.4.10) with initial
value z 0 (0)2L 2 (Ω) exists over a time interval [0, T ∗ ], with T ∗ > 0 independent of m
(see the proof in Section 7.4). Since z 0m 2L 2 (Ω) belongs to W 1,1 (0, Tm∗ ) and satisfies
(5.4.10), Proposition 5.4.3 implies that Tm∗ = T ∗ . Therefore, z 0m L ∞ (0,T ∗ ;L 2 (Ω)3 ) is
bounded uniformly with respect to m, and Lemma 5.2.5 implies that u0m is bounded
3
uniformly with respect to m in L ∞ (0, T ∗ ; H 2 (Ω) ∩ C1,σ (Ω) ).
Here also, the a priori estimate for the time derivative is independent of the choice
of basis functions.
Proposition 5.4.4 Under the assumptions of Lemma 5.4.2, the time derivative (u0m )
of the solution u0m of (5.4.15)–(5.4.16) satisfies the following a priori estimate for
all t in [0, T ∗ ]:
(u0m ) 2L 2 (Ω×]0,t[) + α (u0m ) 2L 2 (0,t;H 1 (Ω)3 ) + μu0m (t)2H 1 (Ω) ≤ μut 0 2H 1 (Ω)
0 0 0
+ 2 f 2L 2 (Ω×]0,t[) + 2( S̃2 C1 ) 2

g 2 2 1
L (0,t;H 2 (∂Ω)3 )
1
+ ( S̃3 S0,6 )2 z m 2L ∞ (0,t;L 2 (Ω)3 ) um 2L 2 (0,t;H 1 (Ω)3 ) , (5.4.18)
α
where
um (t) = u0m (t) + u g(t) , z m (t) = curl(um (t) − αΔ um (t)).
Proof Let t ∈]0, T ∗ ] and s ∈]0, t[. By multiplying the jth line of (5.4.15) with cj,m ,
summing over j and dropping the index m, we obtain
μ d
(u0 ) (s)2α + u0 (s)2H 1 (Ω) = − z(s) × u(s) + f (s) − u g (s) , (u0 ) (s) .
2 dt 0
Then applying Young’s inequality, we easily derive
d 1
(u0 ) (s)2α + μ u0 (s)2H 1 (Ω) ≤ ( S̃3 S0,6 )2 u(s)2H 1 (Ω) z(s)2L 2 (Ω)
dt 0 α 0
+ 2 f (s) L 2 (Ω) + 2( S̃2 C1 )2 g (s) H 21 (∂Ω) ,

2
and (5.4.18) follows by integrating with respect to s from 0 to t.
Then the conclusions of Theorem 5.2.13 hold and we can state the following
result:
Theorem 5.4.5 Let Ω be a bounded, simply-connected open set of IR 3 that is

either a convex polyhedron or has a boundary ∂Ω of class C1,1 . Let f be given
1
in L 2 (0, T ; H (curl, Ω)), g be a given tangential field in H 1 (0, T ; H 2 (∂Ω)3 ), that
has a lifting u g in H (0, T ; [H (Ω) ∩ C (Ω)] ) satisfying (5.4.4) in [0, T ], and let
1 2 1,σ 3
ut 0 be given in W α . Then there exists a time T ∗ with 0 < T ∗ ≤ T , and constants C1 ,

C2 , C3 , depending only on Ω, α, μ, ut 0 , f , and g, such that for all m, the Galerkin
solution u0m of (5.4.15)–(5.4.16) exists on the time interval [0, T ∗ ] and is uniformly
bounded with respect to m
u0m L ∞ (0,T ∗ ;(H 2 (Ω)∩C1,σ (Ω))3 ) ≤ C1 ,

z 0m L ∞ (0,T ∗ ;L 2 (Ω)3 ) ≤ C2 , (5.4.19)
(u0m ) L 2 (0,T ∗ ;H 1 (Ω)3 ) ≤ C3 ,
where C1 , C2 , C3 are constants independent of m, and σ > 0 depends only on ∂Ω.

With the assumptions on g, this implies that um and z m are also bounded uniformly
with respect to m.
5.4.4 Passing to the Limit and Recovering the Pressure
As in Section 5.2.5, the uniform bounds (5.4.19) imply weak convergence up to

subsequences of u0m to u0 , z 0m to z 0 = curl(u0 − αΔ u0 ), and (u0m ) to u0 . This
allows us to pass to the limit in (5.4.15) and the limit functions
u = u0 + u g , and z = curl(u − αΔ u)
solve on ]0, T [
∀v ∈ V α , (u (t), v) + α (∇ u (t), ∇ v) + μ (∇ u(t), ∇ v) + (z(t) × u(t), v)

= ( f (t), v), (5.4.20)
and satisfy the same bounds as um and z m . In addition, u satisfies the initial condition
(5.2.4)
Similarly, we can deduce from (5.4.20) existence of a unique pressure p, and the
triple (u, p, z) satisfies on ]0, T [
∀v ∈ H01 (Ω)3 , (u (t), v) + α (∇ u (t), ∇ v) + μ (∇ u(t), ∇ v)

+ (z(t) × u(t), v) − ( p(t), div v) = ( f (t), v). (5.4.21)
To derive a bound for p, we choose v in V ⊥ unique solution of div v = p; recall it

satisfies (3.2.15) with β = K̃. As v ∈ V ⊥ , (5.4.21) reduces to
( p(t), div v) = (u (t), v) + α (∇ u g (t) , ∇ v) + μ (∇ u g(t) , ∇ v)

+ (z(t) × u(t), v) − ( f (t), v).
Moreover, (5.4.4) yields
(∇ u g(t) , ∇ v) = (q g(t) , div v) = (q g(t) , p),

(∇ u g (t) , ∇ v) = (q g (t) , div v) = (q g (t) , p).
Thus we infer from (3.2.15)
1
p(t) L 2 (Ω) ≤ S0,2 u (t) L 2 (Ω) + αq g (t) L 2 (Ω) + μq g(t) L 2 (Ω)
K̃
+ S0,6 z(t) × u(t) L 65 (Ω) + S0,2 f (t) L 2 (Ω) . (5.4.22)
These results are collected in the following theorem:
Theorem 5.4.6 Under the assumptions of Theorem 5.4.5, the limit function u sat-
isfies the same bounds as um , i.e.,
u L ∞ (0,T ∗ ;W α ) ≤ C2 , u L 2 (0,T ∗ ;H 1 (Ω)3 ) ≤ C3 . (5.4.23)
Moreover, a unique pressure p ∈ L 2 (Ω×]0, T ∗ [) can be associated with u so that

the pair (u, p) is a solution of problem (5.2.1), (5.2.2), (5.4.1), and (5.2.4), and
there exists a constant C4 , depending only on α, μ, on the constant K̃ of (3.2.12), the
Sobolev imbedding constants, and the first continuity constant of (5.4.6), such that

p L 2 (Ω×]0,T ∗ [) ≤ C4 u L 2 (Ω×]0,T ∗ [) + z × u L 2 (0,T ∗ ;L 65 (Ω)3 )

+ f L 2 (Ω×]0,T ∗ [) + g H 1 (0,T ;H 21 (∂Ω)) . (5.4.24)
5.4.5 Uniqueness
For addressing uniqueness, we start as in Section 5.2.6. The difference w ∈ V α of

any two solutions u1 = u10 + u g and u2 = u20 + u g of (5.2.1), (5.2.2), (5.4.1), and
(5.2.4) satisfies (5.2.52)–(5.2.53) a.e. in ]0, T ∗ [,
∀v ∈ V, (w (t), v) + α (∇ w (t), ∇ v) + μ (∇ w(t), ∇ v)

+ (curl(w(t) − α Δ w(t)) × u1 (t), v)
(5.4.25)
+ (curl(u2 (t) − α Δ u2 (t)) × w(t), v) = 0,
w(0) = 0.
However, deriving an H 1 estimate for w is not straightforward because u1 and u2 do

not vanish on the boundary. When d = 2, this is done in [177]. But we present here
a more general proof, valid in two and three dimensions, that is due to Bernard [32].
We retain all the assumptions of Theorem 5.4.5, so that u1 and u2 are smooth
enough. Reverting to the proof of Lemma 5.2.16, we see that all terms in the expansion
of (curl(Δ w)×u, w) as given by formula (5.2.55) carry over here, with the exception
of c(u1 ; w, Δ w) because u1 does not vanish on the boundary; but it would apply if w
had compact support. This suggests to argue by appealing to density; it is the object
of the following lemma.
Lemma 5.4.7 All w in V α and all smooth enough u in W α satisfy the identity
3

∂ u ∂w
([∇ w]u, Δ w) = − [∇ w] , , (5.4.26)
k=1
∂ xk ∂ xk
i.e., identity (5.2.60) is still valid here.

Proof As D(Ω) is dense in H01 (Ω), there exists a sequence {wn }, {wn } in D(Ω)3
such that
lim wn − w H 1 (Ω) = 0.
n→∞
On one hand, since all functions involved here are sufficiently smooth,
lim ([∇ wn ]u, Δ w) = ([∇ w]u, Δ w). (5.4.27)

n→∞
On the other hand, the compact support of wn and Green’s formula imply
3
∂u ∂wn ∂w
([∇ wn ]u, Δ w) = − [∇ wn ] + [∇ ]u, .
k=1
∂ xk ∂ xk ∂ xk
But
3
∂w ∂w ∂wn
3
∂wn
− [∇ ]u, = [∇ ]u, ,
k=1
∂ xk ∂ xk k=1
∂ xk ∂ xk
because u ∈ W α . Then, the convergence of wn and the fact that u belongs to W α

imply that
3
∂w ∂wn
lim [∇ ]u, = 0.
n→∞
k=1
∂ xk ∂ xk
Hence
3
∂ u ∂w
lim ([∇ wn ]u, Δ w) = − [∇ w] , ,
n→∞
k=1
∂ xk ∂ xk
and (5.4.26) follows from (5.4.27).

As a consequence, identity (5.2.55) holds for all w in V α and all sufficiently

smooth u in W α , and hence we have the analogue, with the same proof, of Theorem
5.2.17.

problem (5.2.1), (5.2.2), (5.4.1), and (5.2.4) has a velocity solution
u ∈ L 1 (0, T ∗ ; W 2,3 (Ω)3 ).
Then this problem has no other solution on [0, T ∗ ].
We have constructed a solution in an interval [0, T ] for some 0 < T ≤ T . There-

fore, we can apply Proposition 5.4.3: the quantity ζ := z2L 2 (Ω) satisfies the integro-
differential inequality (5.4.10) a.e. in ]0, T ∗ [,
t

e−μ K (t−s) ζ (s)ds ≤ G 2 (t) + G 3 (t),
3
α ζ (t) + μ ζ (t) − 3α γσ,α ζ (t) − G 4
2
with G 2 , G 3 , and G 4 given by (5.4.11)–(5.4.13). Here the difficulty is that the integral
term in (5.4.10) cannot be compensated by the positive damping term μ ζ . As we
shall see in the next theorem, this integral term restricts the data much further than
in the homogeneous case.

the data are such that, for some number θ ∈]0, 1[,
1
curl(ut 0 − αΔut 0 )2L 2 (Ω) + G 2 L 1 (0,T )
α
2 μ 2
+ G 3 L ∞ (0,T ) < θ , (5.4.28)
μ 6αγσ,α
and
2
G 4 < 1 − θ, (5.4.29)
μ2 K
where G 2 , G 3 , and G 4 are defined in (5.4.11)–(5.4.13). Then the solution u of (5.2.1),

(5.2.2), (5.4.1), and (5.2.4) constructed in Section 5.4.4 exists on the interval [0, T ]
and satisfies the bound
μ
∀t ∈ [0, T ], z(t) L 2 (Ω) < . (5.4.30)
6αγσ,α
Proof We have seen in Proposition 5.4.3 that ζ (t) = z(t)2L 2 (Ω) is a positive solution
in W 1,1 (0, T ) of the integro-differential inequality (5.4.10)
t
α ζ (t) + μ ζ (t) − 3α γσ,α ζ 2 (t) − G 4 e−μ K (t−s) ζ (s)ds ≤ G 2 (t) + G 3 L ∞ (0,T ) .
3
0
(5.4.31)
Note that (5.4.28) implies (5.4.30) at time t = 0, and it follows from the continuity
of ζ with respect to time that (5.4.30) holds on an interval [0, t˜], t˜ > 0.
Let us prove that t˜ = T . We argue by contradiction as in Theorem 5.2.19. Suppose
that there exists t > 0 such that
μ 2

∀t < t ∗ , 0 ≤ ζ (t) < ,
6 α γσ,α
μ 2 (5.4.32)
ζ (t ∗ ) = .
6 α γσ,α
When substituted into (5.4.31), this gives a.e. on ]0, t [

μ 1 t
1
ζ (t) + ζ (t) ≤ G 4 e−μ K (t−s) ds + G 2 (t) + G 3 L ∞ (0,T ) .
2α α 0 α
μ
By multiplying both sides with exp( 2α t), this inequality becomes

d μt 1 t
μ 1 μ t
e 2α ζ (t) ≤ G 4 e 2α t e−μ K (t−s) ζ (s)ds + e 2α G 2 (t) + G 3 L ∞ (0,T ) .
dt α 0 α
Therefore, by integrating with respect to t on an interval ]0, τ [ with τ ≤ t , and

μ
multiplying with exp(− 2α τ ), we derive
τ t
1μ
− 2α τ μ
ζ (τ ) ≤ e ζ (0) + G 4 e 2α (t−τ ) e−μ K (t−s) ζ (s)dsdt
α
0 0
1 τ μ (t−τ )
+ e 2α G 2 (t) + G 3 L ∞ (0,T ) dt.
α 0
But
τ
μ
t
2α μ 1 −μτ
e 2α (t−τ ) e−μK (t−s) dsdt = 1 − e− 2α τ + e 2α − e−μK τ .
0 0 μ K
2 1 − 2α K
As the contribution of the last two terms in the right-hand side is negative, this
simplifies to
τ
μ
t
2α μ 2α
∀τ ∈]0, t [, e 2α (t−τ ) e−μK (t−s) dsdt ≤ 1 − e− 2α τ ≤ 2 . (5.4.33)
0 0 μ K
2 μ K
Therefore, by substituting the bound (5.4.32) into the factor of G 4 , we obtain
2 G4 1 μ 2 2
ζ (τ ) ≤ ζ (0) + + G 2 L 1 (0,T ) + G 3 L ∞ (0,T ) . (5.4.34)
μ K
2 α 6 α γσ,α μ
This is valid for any τ ≤ t ; in particular, when t = t , (5.4.28) and (5.4.29) imply
that μ 2
ζ (t ∗ ) < ,
6 α γσ,α
thus contradicting the equality in (5.4.32).

Condition Prescribed. The Case of a Smoother Domain
This section is devoted to problem (5.2.1), (5.2.2), (5.4.1), (5.2.4), in a domain Ω

with a boundary of class C2,1 (that does not allow corners)
∂
∂t
div u = 0 in Ω×]0, T [,
u = g on ∂Ω with g · n = 0 on ∂Ω,
u(0) = ut 0 in Ω,
with the compatibility condition (5.2.5)
div ut 0 = 0 in Ω and ut 0 = g(0) on ∂Ω.
To simplify the discussion, we assume that Ω is simply-connected; the analysis in a

multiply-connected domain follows by the arguments of Section 5.3.
By comparing the H 1 estimates of (5.2.38) and (5.4.9) with u replaced by um ,
in the homogeneous and nonhomogeneous cases, we see that the presence of g
brings an unfavorable term in the right-hand side involving z m multiplied by a factor
that is large if g is large. An inspection of (5.4.7) shows that this term arises from
(z(t) × u g(t) , u0 (t)), whose upper bound depends on the lifting of g. Therefore,
we ask the question: Can we use the following ideas of the Leray–Hopf’s Lemma
(see for instance [116, Chapter IV, Section 2.1]) in order to construct a more suitable
lifting u g that makes the factor multiplying z m arbitrarily small? Since u g is not
differentiated in this term, its contribution can be made small by lifting g with a
function whose support is located in a neighborhood of ∂Ω. This is easily obtained
by truncating the lifting defined in (5.4.4), but since truncation does not preserve
the zero divergence, the idea of Leray [160] and Hopf [132] consists in writing the
lifting as the curl of a vector potential and truncating the potential. In a Lipschitz
domain, it is not difficult to bound the corresponding lifting in H 1 , as is the case
of the Navier–Stokes equations. But in order to control z g , we also need to bound
a fourth-order derivative of the lifting, and this explains why we need a smoother
domain. The necessity of controlling this higher order derivative makes this approach
less attractive (see Remark 5.5.8), but it is included here for the sake of completeness.
Throughout this section, we assume that
5
g ∈ W 1,∞ (0, T ; H 2 (∂Ω)3 ). (5.5.1)
Theorem 3.2.8 guarantees that g has a continuous lifting in W 1,∞ (0, T ; H 3 (Ω)3 ).
5.5.1 A Leray–Hopf Lifting and Its Consequences
Let us start with the standard lifting defined in (5.4.4) (the notation is changed because
the lifting will be modified)
−Δ v g + ∇ s g = 0 , div v g = 0 in Ω,
v g = g on ∂Ω. (5.5.2)
In view of (5.5.1), Theorem 3.2.13 implies that this problem has a unique solu-
tion (v g , s g ) in W 1,∞ (0, T ; H 3 (Ω)3 ) × W 1,∞ (0, T ; H 2 (Ω) ∩ L 2m (Ω)) that depends
continuously on g, i.e.,
v g W 1,∞ (0,T ;H 3 (Ω)3 ) + s g W 1,∞ (0,T ;H 2 (Ω)) ≤ CgW 1,∞ (0,T ;H 25 (∂Ω)3 ) . (5.5.3)
The first lemma constructs a vector potential of v g . To simplify, we state it without

dependence on time.
5
Lemma 5.5.1 Let ∂Ω be of class C2,1 . For each g ∈ H 2 (∂Ω)3 satisfying g · n = 0,
there exists a function ϕ g in H 4 (Ω)3 such that
v g = curl ϕ g ,
(5.5.4)
div ϕ g = 0 in Ω,
and the mapping g → ϕ g is continuous: There exists a constant C depending only

on Ω, such that
ϕ g H 4 (Ω) ≤ Cg H 25 (∂Ω) . (5.5.5)
Proof The proof proceeds in two steps. The general ideas are fairly standard.
(1) We extend v g outside Ω so that the extended function, w g has zero divergence,
compact support in IR 3 , belongs to H 3 (IR 3 )3 , and depends continuously on v g . For
this, we pick a large enough ball B so that Ω B; then v g has a standard exten-
sion, say v g , in H03 (B)3 , depending continuously on v g . Moreover div
v g belongs in
particular to H02 (B\Ω) and Theorem 3.2.19 guarantees the existence of a function
z ∈ H03 (B\Ω)3 , depending continuously on div v g , such that
div z = div
v g in B\Ω.
We extend z by zero outside B\Ω, and the difference w g = v g − z satisfies all the
above requirements.
(2) The vector potential is constructed by solving the Laplace equation in IR 3
− Δ ϕ g = curl w g . (5.5.6)
As w g has compact support, each component of curl w g belongs to a weighted

Sobolev space with arbitrary weight at infinity, and it follows from Proposition 6.5
in Amrouche et al. [10] that problem (5.5.6) has a unique solution ϕ g in H44 (IR 3 )3 :
H44 (IR 3 ) = {v ∈ L 2 (IR 3 ); ρ k D k v ∈ L 2 (IR 3 ), 1 ≤ k ≤ 4},

1
where ρ(x) = (1 + |x|2 ) 2 . It is easy to prove that
ϕ g H 4 (Ω) ≤ Ccurl w g H 2 (Ω) .
It remains to prove that w g = curl ϕ g . To this end, we first observe that
div Δ ϕ g = 0.
Therefore, div ϕ g is harmonic and applying Theorem 3.2.9, we easily derive that
div ϕ g = 0. Then (5.5.6) easily implies that curl ϕ g − w g is also harmonic and the
same argument yields that curl ϕ g − w g = 0.
To proceed with the truncation of ϕ g , we use the following classical function
defined on IR for any ε > 0:
⎧
⎪ if − ∞ ≤ x ≤ e− ε ,
2
⎪
⎪ 1
⎪
⎨ − 1ε
θε (x) = ε log e if e− ε < x ≤ e− ε ,
2 1
(5.5.7)
⎪
⎪ x
⎪
⎪
⎩
if e− ε ≤ x < ∞,
1
0
To simplify, we set η = 21 e− ε . As θε is not smoother than W 1,∞ (IR ), it is regularized

2
by convolution with the following mollifier ρε :
1 1 x
ρε (x) = 1 ρ( ), (5.5.8)
η η
−1 ρ(x) d x
where ρ ∈ D(IR ) is the basic mollifier

− 1−x
1
ρ(x) = e 2
if x ∈ [−1, 1],
0 elsewhere.
Since ρε L 1 (IR ) = 1 and θε L ∞ (IR ) = 1, the product of convolution θε ρε satisfies

∀x ∈ IR , θε ρε (x) = θε (x − y)ρε (y) dy ≤ 1,
IR
and the supports of the two factors imply that

1 if x ≤ η,
θε ρε (x) =
0 if x ≥ η + e− ε .
1
As a consequence all derivatives of θε ρε vanish on ] − ∞, η] ∪ [η + e− ε , ∞[.

1
Moreover, it is easy to check that outside this range, the derivatives of any order k
satisfy

(θε ρε )(k) (x) ≤ ck ε , (5.5.9)
xk
with a constant ck that depends only on k. Finally, the truncation function is obtained
by composing θε ρε with Stein’s regularized distance function Δ(x) ∈ C∞ (Ω) that
satisfies (see Stein [258], Chapter 6, Section 2.1, Theorem 2),
∀x ∈ Ω, c2 d(x) ≤ Δ(x) ≤ c1 d(x), (5.5.10)
and for any derivative k = (k1 , k2 , k3 )
Ak
|D k Δ(x)| ≤ , (5.5.11)
d(x)|k|−1
with constants c1 , c2 , and Ak independent of x, where d denotes the usual distance

to the boundary ∂Ω. We set
∀x ∈ Ω, χε (x) = θε ρε (Δ(x)). (5.5.12)
Clearly, χε belongs to C∞ (Ω), χε L ∞ (Ω) = 1, and it satisfies

1 if c1 d(x) ≤ η,
χε (x) = (5.5.13)
0 if c2 d(x) ≥ η + e− ε .
1
This function will be used with sufficiently small ε so that its support, as defined in
(5.5.13), is a tubular neighborhood of ∂Ω; in particular, it consists of mutually disjoint
neighborhoods of each connected component of ∂Ω when ∂Ω is not connected. This

tubular neighborhood is denoted Ωε ,
1 1 2
Ωε = x ∈ Ω; d(x) ≤ CΩ e− ε + e− ε , (5.5.14)
2
where here CΩ = c12 with the constant c2 in (5.5.10).

A straightforward calculation yields for any derivative k
Bk
∀x ∈ Ω, |D k χε (x)| ≤ ε , (5.5.15)
d(x)|k|
with a constant Bk depending only on k. Then we define the lifting u g,ε by
u g,ε := curl(χε ϕ g ) = χε v g + ∇ χε × ϕ g . (5.5.16)
By construction, the regularity of u g,ε is determined by that of v g ; thus u g,ε belongs to

H 3 (Ω)3 . However, owing to (5.4.4), we have curl(Δ vg ) = 0, and hence curl(Δ u g,ε )
is in H 1 (Ω)3 although it involves a third-order derivative. Therefore, the function
z g,ε defined by
z g,ε = curl(u g,ε − α Δ u g,ε ), (5.5.17)
is also in H 1 (Ω)3 . The following lemma collects useful properties of u g,ε . We skip
the proof, which is straightforward.
5
Lemma 5.5.2 Let ∂Ω be of class C2,1 and let g ∈ H 2 (∂Ω)3 satisfy g · n = 0. The
function u g,ε belongs to H 3 (Ω)3 and satisfies
u g,ε |∂Ω = g , support(u g,ε ) ⊂ Ωε ,

ε
∀x ∈ Ω, |u g,ε (x)| ≤ |v g (x)| + C0 |ϕ (x)|, (5.5.18)
d(x) g
2(m+1)
|u g,ε | H m (Ω) ≤ Cm εe ε g H m− 21 (∂Ω) , 1 ≤ m ≤ 3,
with constants Ci that depend only on Ω, 0 ≤ i ≤ m. Furthermore, z g,ε satisfies the

bounds
2(m+4)
|z g,ε | H m (Ω) ≤ Cm (α) εe ε g H m+ 23 (∂Ω) , 0 ≤ m ≤ 1, (5.5.19)
with constants Cm (α) that depend only on α and Ω.

5
Now, we assume that g is given in H 1 (0, T ; H 2 (∂Ω)3 ). For any ε > 0 sufficiently
small, both u0 = u − u g,ε and its time derivative u0 = u − u g ,ε belong to V α and
solve (5.4.7) a.e. in ]0, T [ for all v ∈ V
(u0 (t), v) + α(∇ u0 (t), ∇ v) + μ(∇ u0 (t), ∇ v)

+ (z 0 (t) × u0 (t), v) + (z g(t),ε × u0 (t), v) + (z 0 (t) × u g(t),ε , v) = (H(t), v),
with a different expression for H because u g,ε is not in V ⊥ ,
(H(t), v) = ( f (t) − u g (t),ε − z g(t),ε × u g(t),ε , v)

− α(∇ u g (t),ε , ∇ v) − μ(∇ u g(t),ε , ∇ v), (5.5.20)
but the same initial condition (5.4.8),
u0 (0) = ut 0 − u g (0).
Let us examine the effect of the Leray–Hopf lifting on the only nonlinear term in
the left-hand side of (5.4.7) that does not vanish when v = u0 , i.e., (z 0 (t)×u g(t),ε , v).
Considering that z 0 can only be bounded in L 2 , estimating this term amounts to
deriving a bound in L 2 for the product of any components of u g,ε and v, i.e., u g,ε ⊗ v.
Lemma 5.5.3 Under the assumptions of Lemma 5.5.2, for any δ > 0, there exists a
real number ε > 0 depending on g H 23 (∂Ω) such that
∀v ∈ V, u g,ε ⊗ v L 2 (Ω) ≤ δv H01 (Ω) . (5.5.21)
Proof Take for the moment any ε > 0 sufficiently small. Considering the support of
u g,ε and the definition (5.5.16), we can write for any v ∈ V
u g,ε ⊗v L 2 (Ω) = curl(χε ϕ g )⊗v L 2 (Ωε ) ≤ χε v g ⊗v L 2 (Ωε ) +(∇ χε ×ϕ g )⊗v L 2 (Ωε ) .
Consider the first term. On one hand, Cauchy–Schwarz’s inequality, Sobolev’s

imbedding, and the definition (5.5.14) of Ωε imply
3 13
e− 3ε v H01 (Ω) .
1 1
v L 2 (Ωε ) ≤ S0,6 |Ωε | 3 v H01 (Ω) ≤ S0,6 |∂Ω|CΩ
2
Then the H 2 regularity of vg (see Theorem 3.2.13) yields
χε v g ⊗ v L 2 (Ωε ) ≤ v g L ∞ (Ωε ) v L 2 (Ωε )

3 1 1
≤ C1 S0,6 |∂Ω|CΩ 3 e− 3ε g H 23 (∂Ω) v H01 (Ω) . (5.5.22)
2
For the second term, we use (5.5.16) with k = 1; this gives
v
(∇ χε × ϕ g ) ⊗ v L 2 (Ωε ) ≤ B1 εϕ g L ∞ (Ω) L 2 (Ωε ) .
d
As v vanishes on ∂Ω, Hardy’s inequality implies (cf. [127])
(∇ χε × ϕ g ) ⊗ v L 2 (Ωε ) ≤ C2 B1 εϕ g L ∞ (Ω) v H01 (Ω)

≤ C3 B1 εg H 23 (∂Ω) v H01 (Ω) , (5.5.23)
with a constant C3 that depends only on Ω. Summing the two inequalities (5.5.22)
and (5.5.23), we obtain
3 1 1
u g,ε ⊗ v L 2 (Ω) ≤ C1 S0,6 |∂Ω|CΩ 3 e− 3ε + C3 B1 ε g H 23 (∂Ω) v H01 (Ω) ,
2
and as e− 3ε is negligible compared to ε for small enough ε, this reduces to (5.5.21)

1
with
δ
δ ≥ C εg H 23 (∂Ω) i.e., ε ≤ , (5.5.24)
Cg H 23 (∂Ω)
with a constant C depending only on Ω.
Strictly speaking, the proof of Lemma 5.5.3 only uses the L ∞ norm of v g . There-
fore, the statement of Lemma 5.5.3 is valid in Lipschitz polyhedra with boundary
data g in W 1− r ,r (∂Ω)3 for some r > 3, close to 3, and practically no restriction on
1
the angles. However, the statement is written in a more regular situation because we
need to control z g,ε in H 1 , see Remark 5.5.6.
With a similar proof, we get the following estimate for (u g (t),ε , v):
3 56
e− 6ε g H 21 (∂Ω) v H01 (Ω)
5
|(u g (t),ε , v)| ≤ C1 S0,6
2
|∂Ω|CΩ
2
+ C4 B1 εg H 21 (∂Ω) v H01 (Ω) ≤ C εg H 21 (∂Ω) v H01 (Ω) , (5.5.25)
with another constant C that depends only on Ω.
Again the a priori estimates in this paragraph are established under the assumption
that a solution exists during the time interval [0, T ] for some 0 < T ≤ T . The
situation is fairly similar to that of Section 5.4.2 except that curl Δ u g(t),ε does not
vanish. Thus (5.4.7) is replaced by

+ (z(t) × u g(t),ε , v) = (H ε (t), v), (5.5.26)
with
H ε (t) = f (t) − u g (t),ε + α Δ u g (t),ε + μ Δ u g(t),ε .
Then arguing as in Lemma 5.2.10, we easily derive

3
Lemma 5.5.4 Let Ω be C2,1 , f in L 2 (Ω×]0, T [)3 , g in H 1 (0, T ; H 2 (∂Ω)3 ), and
ut 0 in W α . Pick an arbitrary ε > 0 and let δ satisfy (5.5.24). Then any solution
(u0 , z) of (5.5.26), (5.4.8) in L ∞ (0, T ; H01 (Ω)3 ) × L 2 (Ω×]0, T [)3 satisfies the
following a priori estimate for all t in [0, T ∗ ]:

−μ K t 5 t
u0 (t)2α ≤e ut 0 2α + e−μ K (t−s) G 1,ε (s)+δ 2 z(s)2L 2 (Ω) ds, (5.5.27)
μ 0
with

G 1,ε (t) = S0,2
2
f (t)2L 2 (Ω) + K 1 (ε)2 α 2 g (t)2 1 + μ2 g(t)2 1
H (∂Ω)
2 H (∂Ω)
2
+ C 2 ε2 g (t)2 1 , (5.5.28)
H (∂Ω)
2
6
where K 1 (ε) = C2 εe ε is the factor in (5.5.18), C is the factor of (5.5.25), and K is
defined by (5.2.39).
Note that by assumption, G 1,ε belongs to L 1 (0, T ).

As far as z is concerned, the inequality (5.4.14) holds with a slight modification,
α 1

ζ (t) + μ ζ (t) ≤ αζ (t) γσ,α ζ 2 (t) + γσ,α z g(t),ε L 2 (Ω) + ∇ u g(t),ε L ∞ (Ω)
2 1
+ ζ 2 (t) μcurl u0 (t) L 2 (Ω) + μcurl u g(t),ε L 2 (Ω)

+ αcurl f (t) L 2 (Ω) . (5.5.29)
Then proceeding as in Proposition 5.4.3, we easily derive the next result.
Proposition 5.5.5 In addition to the assumptions and notation of Lemma 5.5.4, we

suppose that f belongs to L 2 (0, T ; H (curl, Ω)) and g is in L ∞ (0, T ; H 2 (∂Ω)3 ).
5
Then any solution (u0 , z) of (5.5.26), (5.4.8) in L (0, T ; V ) × L (Ω×]0, T [)3

∞ α 2
is such that the quantity ζ := z2L 2 (Ω) , satisfies the integro-differential inequality
a.e. in ]0, T ∗ [
t
15
α ζ (t) + μ ζ (t) − 3α γσ,α ζ 2 (t) − δ 2 e−μ K (t−s) ζ (s)ds
3
α 0

15 t −μ K (t−s)
≤ G 2,ε (t) + e G 1,ε (s)ds, (5.5.30)
α 0
where
3α 2 3μ −μK t
G 2,ε (t) = 9μK 1 (ε)2 g(t)2 1 + curl f (t)2L 2 (Ω) + e ut 0 2α
H 2 (∂Ω) μ α
128 6
+ α γσ,α (C(α)εe ε g(t) H 21 (∂Ω) )3
27
1 8
+ C3 εe ε g(t) H 25 (∂Ω) )3 , (5.5.31)
γσ,α
Note that G 2,ε ∈ L ∞ (0, T ) depends only on ε, the domain and the data.
5.5.3 Local in Time Existence
In this section, ε > 0 is fixed but arbitrary and δ satisfies (5.5.24). Therefore,
the situation is much the same as in the case of a convex corner domain, except
that the calculations are more technical since more terms are involved. As the ideas
are the same, this section is kept very short to avoid repetitions.
The discretization of u0 in the basis (5.2.27) is straightforward: Find

m
u0m (t) = c j,m (t) w j ,
j=1
((u0m ) (t), w j ) + α(∇ (u0m ) (t), ∇ w j ) + μ(∇ u0m (t), ∇ w j ) + (z 0m (t) × u0m (t), w j )
+ (z g(t),ε × u0m (t), w j ) + (z 0m (t) × u g(t),ε , w j ) = (H ε (t), w j ),
u0m (0) = Pm (ut 0 ).
This system has a unique solution on an interval [0, Tm ], for some Tm > 0 depend-
ing on the dimension m, and the argument of Lemma 5.5.4 shows that it satisfies the
same a priori estimate as in this lemma. As expected, the estimate of z 0m does not
follow from Proposition 5.5.5, but by arguing as in Section 5.4.3, we can show that
it satisfies an integro-differential inequality of the same form as (5.5.30), but with a
different function G 2,ε , that takes into account the additional terms in the decompo-
sition of z m . This function is independent of m and can be bounded independently of
t in the interval [0, T ]. From here, the arguments used in the convex corner domain
yield existence of a solution in an interval [0, T ], and owing to the boundary’s
smoothness, this solution belongs to L ∞ (0, T ; H 3 (Ω)3 ). Therefore, Theorem 5.4.8
implies that this solution is unique.
Remark 5.5.6 The function F m used in recovering an equation for the variable z 0m
has the form
F m (t) = −μ Δ u0m (t) + z 0m (t) × u0m (t) + z g(t),ε × u0m (t) + z 0m (t) × u g(t),ε .
5
Its curl involves the gradient of z g,ε , which explains why we assume an H 2 regularity
on g and a C2,1 boundary ∂Ω. If we could bypass this regularity and only assume g
in W 2− r ,r (∂Ω) for some r greater than 3, then a C1,1 boundary would be sufficient.
1
However, we have not explored this possibility.
To begin with, we choose δ to be

√
θαK
δ=μ , (5.5.32)
30
for some number θ ∈]0, 1[, and we associate with it ε satisfying (5.5.24):
δ
ε≤ .
Cg H 23 (∂Ω)
Then we have the following theorem.
Theorem 5.5.7 In addition to the assumptions of Proposition 5.5.5, we suppose that

the data are such that
30 2
curl(ut 0 − αΔut 0 )2L 2 (Ω) + G 1,ε L 1 (0,T ) + G 2,ε L ∞ (0,T )
αμ μ
μ 2
< (1 − θ ) , (5.5.33)
6αγσ,α
where G 1,ε and G 2,ε are defined, respectively, in (5.5.28) and (5.5.31). Then the
(unique) solution u of (5.2.1), (5.2.2), (5.4.1), and (5.2.4) exists on the interval
[0, T ] and satisfies the bound
μ
∀t ∈ [0, T ], z(t) L 2 (Ω) < . (5.5.34)
6αγσ,α
Proof Let ζ (t) = z(t)2L 2 (Ω) . It stems from Proposition 5.5.5, that ζ is a positive
solution in W 1,1 (0, T ) of the integro-differential inequality (5.5.30)
t
15
e−μ K (t−s) ζ (s)ds
3
α ζ (t) + μ ζ (t) − 3α γσ,α ζ (t) − δ 2
2
α 0

15 t −μ K (t−s)
≤ G 2,ε (t) + e G 1,ε (s)ds.
α 0
The continuity of ζ in time and (5.5.33) imply that (5.5.34) holds on an interval
[0, t˜], t˜ > 0. The argument is now familiar: We prove by contradiction that t˜ = T .
Suppose that there exists t > 0 such that

μ 2 μ 2
∀t < t ∗ , 0 ≤ ζ (t) < and ζ (t ∗ ) = .
6 α γσ,α 6 α γσ,α
By substituting into (5.5.30) and applying the expression (5.5.32) for δ, this gives
a.e. on ]0, t [,

μ 1 2 μ 2 t −μ K (t−s) 1
ζ (t) + ζ (t) ≤ μ Kθ e ds + G 2,ε (t)
2α 2α 6 α γσ,α 0 α

15 t −μ K (t−s)
+ 2 e G 1,ε (s)ds.
α 0
By proceeding as in the proof of Theorem 5.4.9 and using (5.4.33), we easily derive
on [0, t ]
μ μ 2 30
ζ (τ ) ≤ e− 2α τ ζ (0) + θ + G 1,ε L 1 (0,T )
6 α γσ,α αμ
2 μ 2
+ G 2,ε L ∞ (0,T ) < ,
μ 6 α γσ,α
by assumption on the data. This strict inequality contradicts the supposed equality
of ζ (t ∗ ).
Remark 5.5.8 When ε is small, the coefficients G 1,ε and G 2,ε are very large and
(5.5.33) drastically restricts the data.
5.6 Splitting: The Steady System in 2- D
The material in this section is mostly taken from [117].

The construction of a solution by a Galerkin expansion in the special basis (5.2.27)
is useless for numerical purposes because this basis is not available; furthermore, its
discretization is expensive and will hardly preserve its properties. Therefore, we ask
the question: Can the special basis be bypassed? Going back to Section 5.2.1, we
observe that
1. The original equation of motion (5.1.6) is fundamental in deriving an priori energy
estimate for u (see (5.2.6));
2. The third derivative on u can be eliminated by substituting z into (5.1.6) (see
(5.2.7));
3. The transport equation (5.2.10) is fundamental in deriving an a priori estimate for
z in terms of u provided u is sufficiently smooth (see (5.2.11)). The smoothness
of u stems from (5.2.7) (see Lemma 5.2.5).
5.6 Splitting: The Steady System in 2-D 231
Of course the numerical difficulty arises from Eq. (5.2.7). However, this equation
is not totally essential because some smoothness can be recovered from (5.1.6),
provided the boundary of Ω is suitable. It turns out that this regularity is sufficient
for solving the problem. This has been the object of several works by authors such
as Girault and Scott [117, 118], Abboud and Sayah [2], Amara et al. [4].
The two-dimensional case is fairly straightforward. We start with the steady
problem.
5.6.1 Split Formulation
In this section, unless otherwise specified, Ω is a bounded connected domain of IR 2 ,

possibly multiply-connected, with a Lipschitz-continuous boundary ∂Ω. We denote
by Γi , 0 ≤ i ≤ J , the connected components of ∂Ω. In two dimensions, the vector
u is written in the form u = (u 1 , u 2 , 0) in order to define the curl and the vector
product. We have curl u = (0, 0, curl u), where curl u is defined by (3.1.20),
∂u 2 ∂u 1
curl u = − .
∂ x1 ∂ x2
In the steady regime, the equation of motion of a grade-two fluid reduces to
− μ Δ u + curl(u − α Δ u) × u + ∇ p = f in Ω. (5.6.1)
The incompressibility condition (5.2.2) is unchanged,
div u = 0 in Ω,
as well as the tangential boundary condition (5.4.1),
u = g on ∂Ω with g · n = 0 on ∂Ω.
The assumptions on the data are: f is a given function in H (curl, Ω), g is a given
1
tangential vector field in H 2 (∂Ω)2 , and μ > 0 and α > 0 are two given constants.
α
Recall the space W defined by (5.4.2),
W α = {v ∈ W ; α curl Δ v ∈ L 2 (Ω)2 },
with W defined in (3.1.18). Let (u, p) ∈ W α × L 2m (Ω) be a solution of (5.6.1),

(5.2.2), (5.4.1), and introduce the auxiliary variables
z = curl(u − αΔ u), z = (0, 0, z).
Note that z ∈ L 2 (Ω), z ∈ L 2 (Ω)3 , and
div z = 0. (5.6.2)
For the sake of convenience, in what follows, we shall make no distinction between
the scalar z and vector z. Then the system (5.6.1), (5.2.2), (5.4.1) becomes a gener-
alized Stokes equation
− μΔu + z × u + ∇ p = f in Ω, (5.6.3)
with z × u = (−zu 2 , zu 1 ). Next, we observe that the distributional curl of (5.6.3)

simplifies, because owing to the structure of z and u, we have [∇ u]z = 0 and hence
curl(z × u) = (0, 0, u · ∇ z), (5.6.4)
where

d
∂z
u·∇z = ui , (5.6.5)
i=1
∂ xi
is the scalar analogue of [∇ u]z. Thus z satisfies the transport equation
μ z + α u · ∇ z = μ curl u + α curl f . (5.6.6)
A quick inspection shows that (5.6.3) makes sense with z in L 2 (Ω), u in H 1 (Ω)2 ,
p in L 2 (Ω). With this regularity, the meaning of (5.6.6) follows from the fact that
the product u · ∇ z is well-defined as a distribution as soon as u is in H (div, Ω) and
z in L 2 (Ω). More precisely, denoting by
·, · X the duality pairing between X and
X , we have by definition
∀ϕ ∈ D(Ω),
u · ∇ z, ϕD(Ω) =
∇ z, u ϕ Hτ (div,Ω)

= −
z, div(u ϕ) L 2 (Ω) = − z div(u ϕ)d x,
Ω
and u ϕ belongs to Hτ (div, Ω) provided ϕ belongs to W01,∞ (Ω). In particular, when

1, p
u belongs to H 1 (Ω)2 , this holds whenever ϕ is in W0 (Ω), for some p > 2.
Furthermore, this expression simplifies when div u = 0 and we obtain for all ϕ in
D(Ω), for all z in L 2 (Ω) and all u in H (div, Ω) with div u = 0,

u · ∇ z, ϕD(Ω) = − z(u · ∇ ϕ)d x. (5.6.7)
Ω
The system (5.6.3), (5.2.2), (5.6.6) falls into the category of mixed formulations,
where an auxiliary variable (here, z) is introduced so as to reduce the order of dif-
ferentiation, and facilitate discretization. This is used for example in the case of
biharmonic equations or the stream function–vorticity formulation of the Stokes
problem, see for instance the work of Ciarlet and Raviart in [61], or [116].
Let us check that this system is equivalent to the original problem.
Proposition 5.6.1 Problem (5.6.1), (5.2.2), (5.4.1) with (u, p) in W α × L 2m (Ω) is

equivalent to: Find (u, p, z) in Hτ1 (Ω)× L 2m (Ω)× L 2 (Ω) solution of the generalized
Stokes problem (5.6.3), (5.2.2), (5.4.1) and the transport equation (5.6.6), namely
−μ Δ u + z × u + ∇ p = f in Ω, with z × u = (−zu 2 , zu 1 ),
div u = 0 in Ω,
(5.6.8)
u = g on ∂Ω,
μ z + α u · ∇ z = μ curl u + α curl f in Ω.
Proof We have seen that the original problem implies (5.6.8). Conversely, if (u, p, z)
solves (5.6.8), we only need to recover the relation between z and u. First, z =
(0, 0, z) satisfies (5.6.2) and taking the curl of (5.6.3) in the sense of distributions
yields, in view of (5.6.4),
−μ Δ(curl u) + u · ∇ z = curl f .
Then multiplying by α and comparing with (5.6.6), we obtain
−μ α Δ(curl u) = μ z − μ curl u,
i.e., z = curl(u − αΔ u). Therefore, u belongs to W α and substituting the expression

for z into (5.6.3) shows that (u, p) is a solution of the original problem.
Note that if (u, p, z) ∈ W × L 2m (Ω) × L 2 (Ω) solves (5.6.8), then necessarily

αu · ∇ z belongs to L 2 (Ω). Hence Problem (5.6.8) has the following equivalent
variational formulation: Find (u, p, z) in Hτ1 (Ω) × L 2m (Ω) × L 2 (Ω) solution of
∀v ∈ H01 (Ω)2 , μ(∇ u, ∇ v) + (z × u, v) − ( p, div v) = ( f , v), (5.6.9)
∀q ∈ L 20 (Ω) , (q, div u) = 0, (5.6.10)
u = g on ∂Ω, (5.6.11)
∀θ ∈ L 2 (Ω), μ(z, θ ) + α(u · ∇ z, θ ) = (μ curl u + α curl f , θ ). (5.6.12)
In order to study (5.6.8), it is convenient to write (5.6.9)–(5.6.11), for fixed f

and g, as an equation for (u, p) with given z. It is easy to prove (see Proposition
5.6.7 below) that for each given z in L 2 (Ω), z = (0, 0, z), there exists a unique pair
(u(z), p(z)) in Hτ1 (Ω) × L 2m (Ω) solution of
∀v ∈ H01 (Ω)2 , μ(∇ u(z), ∇ v) + (z × u(z), v) − ( p(z), div v) = ( f , v), (5.6.13)
∀q ∈ L 20 (Ω) , (q, div u(z)) = 0, (5.6.14)

u(z) = g on ∂Ω.
Remark 5.6.2 When α = 0, the generalized Stokes equation (5.6.3) is unchanged

whereas the transport equation (5.6.6) reduces to z = curl u. In this case, z is simply
the vorticity of u and problem (5.6.8) can be interpreted as a velocity–vorticity
formulation of the Navier–Stokes equations.
5.6.2 A Simpler Leray–Hopf Lifting
In two dimensions, the Leray–Hopf lifting described in Section 5.5.1 substantially

simplifies because the vector potential (or stream function) vanishes on ∂Ω. This has
been used by Temam in [266] in the case of smooth tangential boundary conditions
on a smooth boundary. Here we shall use a variant of the lifting in [266], developed
by Girault and Scott in [118], that applies to a Lipschitz boundary. It operates on
1
boundary data in H 2 (∂Ω)2 , but as the calculations are simpler when g is bounded,
we shall assume that g is in W 1− r ,r (∂Ω)2 for some r with 2 < r ≤ 4. Indeed, the
1
imbedding (3.1.4) with k = 0, s = 1, p = r , d = 2, implies that g is bounded.

The key ideas are the same as in Section 5.5.1, but the function θε ρε is replaced
by a piecewise polynomial. First, we define the function λ ∈ C1 (IR + )
⎧
⎪
⎨1 for 0 ≤ t ≤ 21 ,
λ(t) = 2t (t − 23 )2 for 21 ≤ t ≤ 23 , (5.6.15)
⎪
⎩
0 for t ≥ 23 ,
and next, we scale λ as follows:

Δ(x)
∀x ∈ Ω, λε (x) = λ .
ε
As (5.5.10) and (5.6.15) imply that λε (x) vanishes for all x ∈ Ω such that d(x) ≥ 2c
3ε
2
,
we define
3
Ωε = {x ∈ Ω; d(x) ≤ CΩ ε} with CΩ = . (5.6.16)
2c2
Again we assume that ε > 0 is sufficiently small so that Ωε consists of mutually

disjoint neighborhoods Ωε,i of each Γi ,
Ω ε,i ∩ Ω ε, j = ∅ , for j = i. (5.6.17)
Therefore, support(λε ) ⊂ Ωε ,
C C
λε L ∞ (Ω) = 1 , λε W 1,∞ (Ω) ≤ , λε W 2,∞ (Ω) ≤ 2 , (5.6.18)
ε ε
where here and in the sequel, all constants C are independent of ε.

As we are in two dimensions, the vector potential of v g constructed in Lemma
5.5.1 is in fact a stream function. Indeed, the procedure of Lemma 5.5.1 constructs
a function ϕ g ∈ H 1 (Ω) such that
v g = curl ϕ g , (5.6.19)
but since in two dimensions the vector curl determines the gradient, the regularity
of ϕ g is determined by that of v g . The regularity of the Stokes problem (5.5.2) for
2 < r ≤ 4 implies that v g belongs to W 1,r (Ω)2 with continuous dependence on g,
see Theorems 3.2.18 and 3.2.10. Hence ϕ g belongs to W 2,r (Ω) with
ϕ g W 2,r (Ω) ≤ CgW 1− r1 ,r (∂Ω) . (5.6.20)
Furthermore, (5.6.19) and g · n = 0 on ∂Ω imply that ϕ g is constant on ∂Ω: There

exist constants Ci such that
ϕ g |Γi = Ci , 0 ≤ i ≤ J.
ϕ g defined in each Ωε,i by

Thus the function ϕ g = ϕ g − Ci , 0 ≤ i ≤ J , vanishes on
∂Ω and is single valued in Ωε owing to (5.6.17). We take
u g,ε = curl(λε
ϕ g ). (5.6.21)
Owing to the support of λε , this function is completely defined in Ω. By construction,

u g,ε ∈ W 1,r (Ω)2 and support(u g,ε ) ⊂ Ωε . Since ϕ g vanishes on ∂Ω, we can check
that it satisfies the following Poincaré inequality, for any number s with 1 ≤ s ≤ ∞:
ϕ g L s (Ωε ) ≤ Cε∇
ϕ g L s (Ωε ) . (5.6.22)
Then we have the following propositions that express the dependence of u g,ε upon
ε and g.
Proposition 5.6.3 Let g ∈ W 1− r ,r (∂Ω)2 for some r > 2. For any number s with
1
1 ≤ s ≤ ∞, we have
1
u g,ε L s (Ωε ) ≤ Cε s gW 1− r1 ,r (∂Ω) . (5.6.23)
Proof Expanding u g,ε = curl(λε

ϕ g ) and using (5.6.18), (5.6.22) and the regularity
of
ϕ g , we derive
1
u g,ε L s (Ωε ) ≤ Ccurl
ϕ g L s (Ωε ) ≤ Cε s curl
ϕ g L ∞ (Ωε ) ,
and (5.6.23) follows from (3.1.4) with k = 0, s = 1, p = r and (5.6.20).
Proposition 5.6.4 Let g ∈ W 1− r ,r (∂Ω)2 for some r > 2. We have for any s ∈ [2, r ],
1
∇ u g,ε L s (Ωε ) ≤ Cε s −1 gW 1− r1 ,r (∂Ω) .

1
(5.6.24)
Proof Expanding u g,ε = curl(λε

ϕ g ) and applying (5.6.18) and (5.6.22), we derive
1 1
∇ u g,ε L s (Ωε ) ≤ C ϕ g L s (Ωε ) + | ϕ g |W 1,s (Ωε ) + |
ϕ g |W 2,s (Ωε )
ε 2 ε
1
≤ C | ϕ g |W 2,s (Ωε ) .
ϕ g |W 1,s (Ωε ) + |
ε
Hence, the regularity of
ϕ g yields
1
εs
∇ u g,ε L 2 (Ωε ) ≤C | ϕ g |W 2,s (Ωε ) ,
ϕ g |W 1,∞ (Ωε ) + |
ε
and (5.6.24) follows from (3.1.8) applied to the gradient of

ϕ g and (5.6.20).
Proposition 5.6.5 Let g ∈ W 1− r ,r (∂Ω)2 for some r > 2 and let u g,ε be defined by
1
(5.6.21). We have
∀v ∈ H01 (Ω)2 , u g,ε ⊗ v L 2 (Ωε ) ≤ Cεv H01 (Ωε ) gW 1− r1 ,r (∂Ω) . (5.6.25)
Proof For v ∈ H01 (Ω)2 , we can write
u g,ε ⊗ v L 2 (Ωε ) ≤ u g,ε L ∞ (Ωε ) v L 2 (Ωε ) .
Since v ∈ H01 (Ω)2 , we can apply (5.6.22) with s = 2; then (5.6.23) with s = +∞
yields (5.6.25).
Remark 5.6.6 The comparison between (5.6.24) with s = 2 and (5.5.18) with m =
√ of u g,ε in L (Ω), is striking: as ε tends
2
1, that are both upper bounds for the gradient
to zero the first one tends to infinity as 1/ ε whereas the second one tends to infinity
as exp(4/ε). The difference between these two results, the first one when d = 2 and
the second one when d = 3, arises exclusively from the fact that in the first case, the
stream function is constant on the boundary.
The a priori estimates in this section are derived under the assumption that Problem
(5.6.9)–(5.6.12) has a solution. Solutions will be constructed in the next section. The
first estimate concerns the auxiliary problem (5.6.13), (5.6.14), (5.6.11).
Proposition 5.6.7 Let Ω be bounded, connected, and Lipschitz-continuous, μ > 0,

f ∈ L 2 (Ω)2 and g ∈ W 1− r ,r (∂Ω)2 for some r > 2, satisfying g · n = 0. For an
1
arbitrary ε > 0, let u g,ε denote the lifting defined in (5.6.21). For any z in L 2 (Ω),
the generalized Stokes problem (5.6.13), (5.6.14), (5.6.11) has a unique solution
(u(z), p(z)) in Hτ1 (Ω) × L 2m (Ω). This solution satisfies the following bounds in
H 1 (Ω)2 × L 2 (Ω):
S0,2 Cε
u(z) H01 (Ω) ≤ f L 2 (Ω) + 2u g,ε H01 (Ω) + gW 1− r1 ,r (∂Ω) z L 2 (Ω) ,
μ μ
(5.6.26)
1
p(z) L 2 (Ω) ≤ S0,2 f L 2 (Ω) + μu g,ε H01 (Ω) + S0,4 S̃4 u(z) H01 (Ω) z L 2 (Ω) ,
K̃
(5.6.27)
where K̃ is the isomorphism constant of the divergence operator in (3.2.12), S0, p and
S̃ p are defined in (3.1.5) and (3.1.13), respectively, and C is the constant in (5.6.25).
Proof Set u0 = u − u g,ε , z = (0, 0, z), and let a z denote the bilinear form
a z (v, w) = μ(∇ v, ∇ w) + (z × v, w).
Then (5.6.13), (5.6.14), (5.6.11) is equivalent to: Find u0 ∈ V such that
∀v ∈ V, a z (u0 , v) = ( f , v) − a z (u g,ε , v). (5.6.28)
For fixed z in L 2 (Ω), the bilinear form a z is elliptic on H01 (Ω)2 × H01 (Ω)2 since
(z × v, v) = 0, and it is continuous on H 1 (Ω)2 × H 1 (Ω)2 since
∀u, v ∈ H 1 (Ω)2 , |(z × u, v)| ≤ z L 2 (Ω) u L 4 (Ω) v L 4 (Ω) .
Therefore by Lax–Milgram’s Theorem, (5.6.28) has a unique solution u0 ∈ V and the

divergence isomorphism (Theorem 3.2.11) implies that (5.6.13), (5.6.14), (5.6.11)
has a unique solution (u(z), p(z)) in Hτ1 (Ω) × L 2m (Ω).
The bound (5.6.26) follows immediately by choosing v = u0 in (5.6.28) and
applying (5.6.25). This gives first the intermediate estimate
S0,2 Cε
u0 H01 (Ω) ≤ f L 2 (Ω) + u g,ε H01 (Ω) + gW 1− r1 ,r (∂Ω) z L 2 (Ω) , (5.6.29)
μ μ
and then (5.6.26). The bound for p(z) follows by choosing v = T̃ ( p(z)) defined in
Theorem 3.2.11 and using the orthogonality of v.
Remark 5.6.8 (1) As u(z) is uniquely defined by z, it does not depend on the lifting,
but of course we obtain different bounds according to the choice of lifting. If we
choose u g defined in (5.4.4), then u(z) can be bounded by

S0,2 S0,4 S̃4

u(z) H01 (Ω) ≤ f L 2 (Ω) + Cg H 21 (∂Ω) 1 + z L 2 (Ω) , (5.6.30)
μ μ
where C is the continuity constant of u g with respect to g.

(2) It is interesting to note that the mapping z → u(z) is locally Lipschitz-

continuous in L 2 (Ω), in the sense that for all pairs (z, z̄) in the ball B(0; R) of
L 2 (Ω), the difference u(z) − u(z̄) satisfies
S0,4 S̃4
u(z) − u(z̄) H01 (Ω) ≤ Az − z̄ L 2 (Ω) , (5.6.31)
μ
where
S0,2 S0,4 S̃4
A= f L 2 (Ω) + Cg H 21 (∂Ω) 1 + R .
μ μ
The second a priori estimate concerns the auxiliary variable z. It is readily obtained
from the transport equation (5.6.12) and the following Green formula (an extension
of (5.2.43)) established in the Appendix, Proposition 7.2.7:

(u · ∇ z)z d x = 0, (5.6.32)
Ω
valid for all u ∈ W and z in L 2 (Ω) such that u · ∇ z belongs to L 2 (Ω).
Proposition 5.6.9 In addition to the assumptions of Proposition 5.6.7, we suppose

that f belongs to H (curl, Ω). Then the z part of any solution of Problem (5.6.9)–
(5.6.12) satisfies
α
z L 2 (Ω) ≤ curl u L 2 (Ω) + curl f L 2 (Ω) . (5.6.33)
μ
By combining this estimate with (5.6.26) and (5.6.27), we derive an a priori bound
for z, u, and p.
Theorem 5.6.10 Under the hypotheses of Proposition 5.6.9, any solution (u, p, z)
of Problem (5.6.9)–(5.6.12) satisfies, with constant C1 , C2 , and C3 independent of
μ, α, and g,
2S0,2 C1 3 2α
z L 2 (Ω) ≤ f L 2 (Ω) + √ g 2 1− 1 ,r + curl f L 2 (Ω) , (5.6.34)
μ μ W r (∂Ω) μ
2S0,2 C2 3 α
u H01 (Ω) ≤ f L 2 (Ω) + √ g 2 1− 1 ,r + curl f L 2 (Ω) , (5.6.35)
1 √ 3
p L 2 (Ω) ≤ S0,2 f L 2 (Ω) + C3 μg 2 1
K̃ W 1− r ,r (∂Ω)

+ S0,4 S̃4 u H01 (Ω) z L 2 (Ω) . (5.6.36)
Proof Let (u, p, z) be any solution of Problem (5.6.9)–(5.6.12). Then u = u(z),

and by writing curl u = curl u0 + curl u g,ε , and substituting (5.6.29) into (5.6.33),
we obtain for any small ε > 0
√ α
z L 2 (Ω) ≤ u0 H01 (Ω) + 2u g,ε H01 (Ω) + curl f L 2 (Ω)
μ
S0,2 √ α
≤ f L 2 (Ω) + (1 + 2)u g,ε H01 (Ω) + curl f L 2 (Ω)
μ μ
Cε
+ gW 1− r1 ,r (∂Ω) z L 2 (Ω) . (5.6.37)
μ
Let us choose ε such that

Cε 1 1 μ
gW 1− r1 ,r (∂Ω) = i.e., ε = . (5.6.38)
μ 2 2 CgW 1− r1 ,r (∂Ω)
Then
2S0,2 √ 2α
z L 2 (Ω) ≤ f L 2 (Ω) + 2(1 + 2)u g,ε H01 (Ω) + curl f L 2 (Ω) ,
μ μ
and (5.6.34) follows by substituting (5.6.24) with s = 2 into this inequality. In turn,
the choice (5.6.38) of ε gives
S0,2 1
u H01 (Ω) ≤ f L 2 (Ω) + 2u g,ε H01 (Ω) + z L 2 (Ω) ,
μ 2
and (5.6.35) follows by substituting (5.6.34) into this inequality. The proof of (5.6.36)
is similar.
Remark 5.6.11 When g = 0, the above bounds simplify to
S0,2 α
z L 2 (Ω) ≤ f L 2 (Ω) + curl f L 2 (Ω) , (5.6.39)
μ μ
S0,2
u H01 (Ω) ≤ f L 2 (Ω) , (5.6.40)
μ
1
p L 2 (Ω) ≤ S0,2 f L 2 (Ω) + S0,4
2
u H01 (Ω) z L 2 (Ω) . (5.6.41)
K̃
5.6.4 Construction of a Solution
Now, we turn to existence of a solution of the mixed Problem (5.6.9)–(5.6.12). To

simplify, we note that the form c defined in (5.2.56) reduces to
2
∂z
c(u; z, θ ) = ui θ d x.
i=1 Ω ∂ xi
It satisfies the analogue of (5.2.58)
∀u ∈ W, ∀z, ∈ H 1 (Ω) , c(u; z, z) = 0. (5.6.42)
Existence of a solution is established by discretizing z by Galerkin’s method in a basis

of H 2 (Ω), say {wi }i≥1 . Let Z m be the vector space spanned by wi for 1 ≤ i ≤ m.
For each z m in Z m , we denote by u(z m ) the unique solution of the generalized Stokes
problem (5.6.13), (5.6.14), (5.6.11), and we discretize the transport equation (5.6.12)
by: Find z m in Z m such that, for 1 ≤ i ≤ m,
μ(z m , wi ) + αc(u(z m ); z m , wi ) = μ(curl u(z m ), wi ) + α(curl f , wi ). (5.6.43)
As (5.6.43) is set into finite dimensions, we can solve it by the variant of Brouwer’s
fixed point theorem stated in Lemma 3.3.11.
Proposition 5.6.12 Let Ω be bounded, Lipschitz-continuous, and connected. For

all integers m ≥ 1, all μ > 0, all α > 0, all f ∈ H (curl, Ω) and g ∈ W 1− r ,r (∂Ω)2
1
for some r > 2, satisfying g · n = 0, the Galerkin system (5.6.43) has at least one
solution z m in Z m and z m satisfies the estimate (5.6.34), which is uniform in m,
2S0,2 C1 3 2α
z m L 2 (Ω) ≤ f L 2 (Ω) + √ g 2 1− 1 ,r + curl f L 2 (Ω) , (5.6.44)
with the constant C1 of (5.6.34).
Proof For any λm in Z m , we define F(λm ) in Z m by
∀1 ≤ i ≤ m , (F(λm ), wi ) = μ(λm , wi ) + αc(u(λm ); λm , wi )

− μ(curl u(λm ), wi ) − α(curl f , wi ). (5.6.45)
This square system of linear equations defines F(λm ) ∈ Z m uniquely and, owing to
the finite dimension and the continuity of u(z) (see (5.6.31)), the mapping λm →
F(λm ) is continuous. Let us evaluate (F(λm ), λm ). As the basis functions are smooth,
(5.6.45) and (5.6.42) yield
(F(λm ), λm ) = μλm 2L 2 (Ω) − μ(curl u(λm ), λm ) − α(curl f , λm )

α
≥ μλm L 2 (Ω) λm L 2 (Ω) − curl u(λm ) L 2 (Ω) − curl f L 2 (Ω) .
μ
Arguing as in the proof of Theorem 5.6.10, the term in brackets is bounded below
by the quantity
1 S0,2 √ α
λm L 2 (Ω) − f L 2 (Ω) − (1 + 2)u g,ε H01 (Ω) − curl f L 2 (Ω) ,
2 μ μ
with ε defined by (5.6.38). By substituting (5.6.24) with s = 2, the lower bound

becomes
1 S0,2 C1 3 α
λm L 2 (Ω) − f L 2 (Ω) − √ g 2 1− 1 ,r − curl f L 2 (Ω) .
2 μ 2 μ W r (∂Ω) μ
Hence (F(λm ), λm ) is nonnegative as soon as the term in brackets is zero, i.e.,
2S0,2 C1 3 2α
λm L 2 (Ω) = f L 2 (Ω) + √ g 2 1− 1 ,r + curl f L 2 (Ω) .
Thus, we infer from Lemma 3.3.11 that F(λm ) has at least one zero in the ball centered
at zero, with radius
2S0,2 C1 3 2α
R= f L 2 (Ω) + √ g 2 1− 1 ,r + curl f L 2 (Ω) .
Therefore, (5.6.43) has at least one solution z m and this solution satisfies the uniform
bound (5.6.44).
Of course, (5.6.44) and Proposition 5.6.7 immediately imply that both u(z m ) and
p(z m ) are bounded uniformly with respect to m, the first one in H 1 (Ω)2 and the
second one in L 2m (Ω). Therefore, we can pass to the limit and prove existence of a
solution of problem (5.6.9)–(5.6.12).
Theorem 5.6.13 Let Ω be bounded, connected, Lipschitz-continuous. For all μ > 0,

all α > 0, all f ∈ H (curl, Ω) and g ∈ W 1− r ,r (∂Ω)2 for some r > 2, satisfying
1
g·n = 0, problem (5.6.9)–(5.6.12) has at least one solution (u, p, z) and all solutions
satisfy the estimates of Theorem 5.6.10.
Proof In view of the uniform bounds for
(u(z m ), p(z m ), z m ) ∈ H 1 (Ω)2 × L 2 (Ω) × L 2 (Ω),
there exist a subsequence, still denoted by the index m, and three functions z ∈
L 2 (Ω), u ∈ H 1 (Ω)2 and p ∈ L 2 (Ω) such that
lim z m = z weakly in L 2 (Ω),

m→∞
lim u(z m ) = u weakly in H 1 (Ω)2 ,
m→∞
lim p(z m ) = p weakly in L 2 (Ω).
m→∞
The fact that div u(z m ) = 0 in Ω implies that div u = 0 in Ω, the fact that u(z m )−u g,ε
belongs to H01 (Ω)2 implies that u − u g,ε belongs to H01 (Ω)2 , the fact that p(z m )
belongs to L 2m (Ω) implies that p is in L 2m (Ω) and the weak convergence of u(z m )
in H 1 (Ω)2 together with the compactness of H 1 (Ω) in L q (Ω) for any real number
q < ∞ imply that
lim u(z m ) = u strongly in L q (Ω)2 .
m→∞
This allows us to pass to the limit in the generalized Stokes problem (5.6.13), (5.6.14),
(5.6.11), associated with z m , and it yields (5.6.13), (5.6.14), (5.6.11) associated with
z.
Next, to pass to the limit in (5.6.43), we apply (5.6.42) with z = wi − z m , we
rewrite (5.6.43) as
∀1 ≤ i ≤ m , μ(z m , wi )−αc(u(z m ); wi , z m ) = μ(curl u(z m ), wi )+α(curl f , wi ),
and we choose an arbitrary index i. Owing to the strong convergence of u(z m ) and
the regularity of wi , the product u(z m ) · ∇ wi converges strongly to u · ∇ wi in L 2 (Ω)
and we obtain for all i ≥ 1,
μ(z, wi ) − αc(u; wi , z) = μ(curl u, wi ) + α(curl f , wi ).
Then the density of the finite linear combinations of the functions wi in H 2 (Ω)
implies
∀ϕ ∈ H 2 (Ω) , μ(z, ϕ) − αc(u; ϕ, z) = μ(curl u, ϕ) + α(curl f , ϕ).
But, in view of (5.6.7), we have for all ϕ in D(Ω),
−c(u; ϕ, z) =
u · ∇ z, ϕD(Ω) .
Therefore, the pair (u, z) is a solution, in the sense of distributions, of the transport
equation (5.6.12).
The bounds are, of course, an immediate consequence of the a priori estimates of
Theorem 5.6.10 satisfied by all solutions of problem (5.6.9)–(5.6.12).
By equivalence (see Proposition 5.6.1) the statement of Theorem 5.6.13 extends to

Problem (5.6.1), (5.2.2), (5.4.1) with (u, p) in W α × L 2m (Ω).
Remark 5.6.14 The estimates (5.6.34)–(5.6.36) hold on a Lipschitz-continuous

domain, without restriction on the size of the data. But their derivation, and in par-
ticular deriving an unconditional estimate for z depends drastically on the choice
of this auxiliary variable and the space to which it should belong. With our choice,
that dates back to [205], the transport equation (5.6.6) for z has only one nonlinear
term and z L 2 (Ω) can be bounded unconditionally by curl u and curl f . There are
of course several possibilities for splitting the original equation, but no other choice
seems to produce this happy result. In this respect, the splitting achieved by (5.6.8)
is optimal.
5.6.5 Uniqueness of the Steady 2- D Problem
As is the case of the steady Navier–Stokes equations, uniqueness of the solution of

the steady grade-two fluid model cannot be expected for all data. But the comparison
stops here because, in contrast to the Navier–Stokes equations, uniqueness of the
solution of a grade-two fluid is not known without assuming more regularity on the
solution. As stated at the beginning of this section, the split formulation is obtained at
the expense of regularity, but, considering that the original and split formulations are
equivalent, now that existence of a solution is established, we can use the original
formulation. Similarly, as far as regularity is concerned, we can use the standard
lifting u g defined in (5.4.4).
Thus, we revert to the situation of Section 5.4 in a convex corner domain Ω ⊂ IR 2 ,
but since we are in two dimensions, we can aim at a W 2,r regularity for some r > 2,
depending on the largest inner angle of ∂Ω, rather than a C1,σ regularity. Let g be
1
given in H 2 (∂Ω)2 , satisfying g · n = 0, and let S denote any straight line segment
of ∂Ω. We assume that g ∈ W 2− r ,r (S)2 on all S and that at the corners of ∂Ω, g
1
satisfies compatibility conditions (see Arnold et al. [17, Theorem 7.1], or Girault and
Scott [117, Theorem 5.2]) that ensure that it is the trace of a function u g in W 2,r (Ω)2
that solves (5.4.4), with the associated pressure q g in W 1,r (Ω), both velocity and
pressure depending continuously on g. Here r is a real number with 2 < r < r0 ,
where r0 depends on the largest inner angle of ∂Ω. Note that the inner angles of ∂Ω
are always strictly smaller than π , and we can always assume that r0 ≤ 4 since we
are only interested in r slightly larger than two. Adapting the notation for broken
norms in (5.4.5),
21
[g]2− r1 ,r = g2 2− r1 ,r ,
W (S)
S∈∂Ω
the continuous dependence of (u g , q g ) on g reads as in (5.4.6)
u g H 1 (Ω) ≤ C1 g H 21 (∂Ω) , u g W 2,r (Ω) ≤ Cr [g]2− r1 ,r , (5.6.46)
with similar inequalities for q g . Then z g = curl u g ∈ W 1,r (Ω) with

√
z g W 1,r (Ω)) ≤ 2Cr [g]2− r1 ,r ,
and of course u g belongs to W α for any α. Therefore, the regularity of u = u0 + u g

is determined by that of u0 , and we know that u0 belongs to V α . Since we are in two
dimensions, the proof of Lemma 5.2.5 gives the analogue of (5.2.21)
∀v ∈ V α , vW 2,r (Ω) ≤ γr,α curl(v − αΔ v) L 2 (Ω) , (5.6.47)
where the constant γr,α is similar to γσ,α in (5.2.21). In particular, γr,α is of the order
of α − 2 . The regularity of u is summarized in the next proposition.
3
Proposition 5.6.15 Let Ω be a convex polygon. There exists a real number r0 > 2,
1
depending on the inner angles of ∂Ω, such that: if g ∈ H 2 (∂Ω)2 satisfies g · n = 0,
g ∈ W 2− r ,r (S)2 for all straight line segments S of ∂Ω, 2 < r < r0 , and suitable
1
compatibility conditions so that its lifting u g belongs to W 2,r (Ω)2 , then any solution
u ∈ W α of (5.6.1), (5.2.2), (5.4.1) belongs to W 2,r (Ω)2 with
α √
uW 2,r (Ω) ≤ γr,α curl u L 2 (Ω) + curl f L 2 (Ω) + 2C1 g H 21 (∂Ω)
μ
+ Cr [g]2− r1 ,r , (5.6.48)
where γr,α , Cr , and C1 are the constants of (5.6.47) and (5.6.46), respectively.
Proof In view of (5.6.47), we can write
uW 2,r (Ω) ≤ γr,α curl(u0 − αΔ u0 ) L 2 (Ω) + u g W 2,r (Ω)

≤ γr,α z L 2 (Ω) + curl u g L 2 (Ω) + u g W 2,r (Ω) .
Then (5.6.48) follows by substituting the inequalities of (5.6.46) into the above
bound.
Remark 5.6.16 A possibly sharper upper bound can be obtained by proceeding as

in the proof of Lemma 5.2.5, arguing directly with u, and using the regularity of
1
the Stokes problem with nonhomogeneous boundary conditions. Then the H 2 norm
of g can be avoided, but the constants are different. Also, a slightly different upper
bound can be obtained if we argue in C1,σ (Ω)2 . Whether we work in this last space
or in W 2,r (Ω)2 , the outcome is that u belongs to W 1,∞ (Ω)2 .
Now we investigate uniqueness of the solution of (5.6.1), (5.2.2), (5.4.1). It can

be obtained by the approach of [32] described in Section 5.4.5, but since the situation
is much simpler in two dimensions, we prefer to present it directly here. It can be
found in [117] and in the earlier work of Ouazar in [205].
Recall the trilinear form c applied to vectors defined in (5.2.56),
2
2
∂v j
c(u; v, w) = ui w j d x.
i=1 j=1 Ω ∂ xi
As in Section 5.4.5, if (u1 , p 1 ) and (u2 , p 2 ) are two solutions of (5.6.1), (5.2.2),
(5.4.1) in W α × L 2m (Ω) and w = u1 − u2 , q = p 1 − p 2 , then (w, q) satisfies
a.e. in Ω
− μ Δ w + curl(w − α Δ w) × u1 + curl(u2 − α Δ u2 ) × w + ∇ q = 0. (5.6.49)
We could use (5.2.55) or (5.2.61) to simplify the nonlinear terms, but we prefer to
take into account the present problem’s two-dimensional structure.
Proposition 5.6.17 Let u1 and u2 be any pair of velocity solutions of (5.6.1), (5.2.2),
(5.4.1) in W α , set w = u1 − u2 , and assume that u1 and u2 belong to W 2, 3 (Ω)2 .
4
Then
μ w2H 1 (Ω) + c(w; u1 , w) + α c(w; curl u1 , curl w)

0

−2α curl w(∇ u 11 · ∇ w2 − ∇ u 12 · ∇ w1 ) d x = 0. (5.6.50)
Ω
Proof Note that the regularity assumptions on u1 and u2 guarantee the validity of
all operations below. As in Lemma 5.2.16, we observe that
(curl w × u1 , w) = c(u1 ; w, w) − c(w; w, u1 ) = c(w; u1 , w).
Thus
μ w2H 1 (Ω) + c(w; u1 , w) − α(curl Δ w × u1 , w) = 0. (5.6.51)
0
Now, we argue in two dimensions. Expanding the last term and setting w⊥ =
(w2 , −w1 ), we obtain
(curl Δ w × u1 , w) = (Δ(curl w), u1 · w⊥ )

⊥ ∂ 1 ⊥
= (curl w, Δ(u · w )) −
1
(curl w) (u · w ) ds
∂Ω ∂n

∂w⊥
= (curl w, Δ(u1 · w⊥ )) − (curl w) g · ds,
∂Ω ∂n
since on the boundary, w vanishes and u1 = g. Now,
Δ(u1 · w⊥ ) = (Δ u1 ) · w⊥ + 2(∇ u1 · ∇ w⊥ ) + u1 · (Δ w⊥ ).
Using the fact that div w = 0 and div u1 = 0, we have
u1 · (Δ w⊥ ) = u1 · ∇ curl w , w⊥ · (Δ u1 ) = −w · ∇ curl u1 .
But
(curl w) u1 · (Δ w⊥ ) d x = c(u1 ; curl w, curl w) = 0,
Ω
and
(curl w) w⊥ · (Δ u1 ) d x = −c(w; curl u1 , curl w).
Ω
Hence (5.6.50) will follow from (5.6.51), provided we prove that

∂w⊥
(curl w) g · ds = 0.
∂Ω ∂n
Let us write g = (g · n)n + (g · t)t, where t is the tangent vector:
t = (n 2 , −n 1 ).
Expanding componentwise, one can show that
∂w⊥ ∂w
·n=− · n + curl w,
∂n ∂t (5.6.52)
⊥ ⊥
∂w ∂w
·t = · n + div w.
∂n ∂t
Therefore,
∂w⊥ ∂w ∂w⊥
g· = (g · n)(− · n + curl w) + (g · t)( · n + div w) = 0,
∂n ∂t ∂t
since div w = 0, g · n = 0 and w = 0 on the boundary. This proves (5.6.50).
As w is measured in the H 1 norm, the last two terms in (5.6.50) mean that u1
must belong to W 2,r (Ω)2 for some r > 2. Thus, if we want to prove uniqueness of
the solution of (5.6.1), (5.2.2), (5.4.1) by this technique, we must work in a convex
domain. We do not know if this problem has a unique solution when the domain has
a reentrant corner.
The following theorem is a straightforward consequence of (5.6.50).
1
Theorem 5.6.18 Let Ω be a convex polygon, and let g ∈ H 2 (∂Ω)2 satisfy g·n = 0,
g ∈ W 2− r ,r (S)2 for all straight line segments S of ∂Ω, for some r ∈]2, r0 [, and
1
suitable compatibility conditions so that its lifting u g belongs to W 2,r (Ω)2 , where
r0 > 2 is the constant of Proposition 5.6.15. If a solution u of problem (5.6.1), (5.2.2),
(5.4.1) is such that
√
μ > S0,4
2
u H01 (Ω) + α 2S0,r |u|W 2,r (Ω) + 2α∇ u L ∞ (Ω) , (5.6.53)
where
2r
r = ,
r −2
then problem (5.6.1), (5.2.2), (5.4.1) has no other solution.
When substituted into (5.6.53) the bounds (5.6.35), (5.6.34), (5.6.46), (5.6.47)
give the following result:
Corollary 5.6.19 With the assumptions of Theorem 5.6.18, problem (5.6.1), (5.2.2),
(5.4.1) has a unique solution in W α × L 2m (Ω) if the data satisfy
2 √
μ2 > S0,4 + αγr,α (2S∞ + 2S0,r ) 2S0,2 f L 2 (Ω)
√ 3
+ C4 μg 2 1 + 2αcurl f L 2 (Ω) , (5.6.54)
W 1− r ,r (∂Ω)
where C4 = max(C1 , C2 ), the constants of (5.6.34) and (5.6.35), and γr,α and S∞
are the constants of (5.6.48) and (3.1.8), respectively.
Remark 5.6.20 When g = 0, the sufficient condition for uniqueness (5.6.54) sim-
plifies and becomes
2
μ2 > S0,4 + 2S0,2 αS∞ γr,α f L 2 (Ω) + 2α 2 S∞ γr,α curl f L 2 (Ω) . (5.6.55)
Similarly, when f = 0, (5.6.54) reduces to
3 2 √ 3
μ 2 > S0,4 + αγr,α (2S∞ + 2S0,r ) C4 g 2 1 .
H 2 (∂Ω)
5.6.6 Limit as α Tends to Zero
Now, we examine the behavior, as α tends to zero, of the solution of (5.6.1), (5.2.2),
(5.4.1). These solutions depend on α and strictly speaking we should denote them
by (uα , p α , z α ). With this notation, (u0 , p 0 ) solves the Navier–Stokes equations and
z 0 = curl u0 .
Theorem 5.6.21 Under the assumptions and with the notation of Theorem 5.6.13,
let (uα , p α , z α ) in W × L 2m (Ω) × L 2 (Ω) be any solution of problem (5.6.9)–(5.6.12).
Then, we can extract a subsequence, still denoted (uα , p α , z α ), such that
lim uα = u0 strongly in H 1 (Ω)2 , (5.6.56)

α→0
lim p α = p 0 strongly in L 2 (Ω), (5.6.57)
α→0
lim z α = z 0 strongly in L 2 (Ω), (5.6.58)
α→0
where (u0 , p 0 ) satisfies the Navier–Stokes equations and z 0 is the curl of u0
−μ Δ u0 + curl u0 × u0 + ∇ p 0 = f in Ω,
div u0 = 0 in Ω, (5.6.59)
u = g on ∂Ω,
0
z 0 = curl u0 in Ω. (5.6.60)
Proof Since each solution satisfies the estimates of Theorem 5.6.10, the sequences
{uα }, { p α } and {z α } are bounded in the corresponding spaces, independently of
α when α tends to zero. Therefore, modulo a subsequence, the argument used in
the proof of Theorem 5.6.13 allows us to pass to the limit as α → 0 to get the
convergences
lim z α = z 0 weakly in L 2 (Ω),

lim uα = u0 weakly in H 1 (Ω)2 ,
lim p α = p 0 weakly in L 2 (Ω).
Similarly, we prove that u0 belongs to W , p 0 is in L 2m (Ω), u0 = g on ∂Ω, uα

converges to u0 strongly in any L r (Ω)2 and passing to the limit in (5.6.9), we obtain
that (u0 , p 0 , z 0 ) satisfies the first three equations in (5.6.8).
To pass to the limit in the transport equation (5.6.12), we choose ϕ in D(Ω) for
test function and apply (5.6.7)
μ(z α , ϕ) − α (uα · ∇ ϕ, z α ) = μ(curl uα , ϕ) + α(curl f , ϕ).
The above weak and strong convergences allow us to pass to the limit in this equation,
as α tends to zero, and we find (5.6.60). Then the equation in (5.6.59) follows by
substituting (5.6.60) into the first equation of (5.6.8).
It remains to establish the strong convergence. As far as the velocity is concerned,
the difference uα − u0 satisfies
∀v ∈ V, μ(∇(uα − u0 ), ∇ v) + ((z α − z 0 ) × uα , v) + (z 0 × (uα − u0 ), v) = 0.
Here we can choose v = uα − u0 ∈ V and we are left with
μ uα − u0 2H 1 (Ω) = −((z α − z 0 ) × uα , uα − u0 ).

0
Then the weak convergence of z α in L 2 (Ω) and the strong convergence of uα in

L 4 (Ω)2 imply that
lim uα − u0 H01 (Ω) = 0.
α→0
Next, the difference z α − z 0 satisfies
μ(z α − z 0 ) + α uα · ∇ z α = μ curl(uα − u0 ) + α curl f .
Then taking the scalar product of this equation with z α and using (5.6.32), we obtain
μ(z α − z 0 , z α ) = μ(curl(uα − u0 ), z α ) + α(curl f , z α ).

The strong convergence of uα in H 1 (Ω)2 and the weak convergence of z α imply that
lim μ(z α − z 0 , z α ) = 0.
α→0
Then expanding z α − z 0 2L 2 (Ω) , we readily find that
lim z α − z 0 L 2 (Ω) = 0.
α→0
Finally, choosing the test function v = T̃ ( p α − p 0 ) ∈ H01 (Ω)2 (see Theorem 3.2.11),
we find that p α − p 0 satisfies
p α − p 0 2L 2 (Ω) = ((z α − z 0 ) × uα , v) + (z 0 × (uα − u0 ), v).
Then the strong convergences of z α and uα and the boundedness of v imply that
lim p α − p 0 L 2 (Ω) = 0,
α→0
and this ends the proof.
5.7 Splitting: The Time-Dependent System in 2- D
As in Section 5.6.1, the domain of interest Ω is a bounded connected open set of

IR 2 , possibly multiply-connected, with a Lipschitz-continuous boundary ∂Ω. For f
1
given in L 2 (Ω×]0, T [)2 and g given in H 1 (0; T ; H 2 (∂Ω)2 ) satisfying the second
part of (5.4.1),
g · n = 0 on ∂Ω,
problem (5.2.1), (5.2.2), (5.4.1), (5.2.4) reads as follows: Find u in H 1 (0, T ; W )

with curl(u − α Δ u) in L ∞ (0, T ; L 2 (Ω)), and p in L 2 (0, T ; L 2m (Ω)), solution of
(5.2.1)
∂
(u − αΔ u) − μΔ u + curl(u − αΔ u) × u + ∇ p = f in Ω×]0, T [,
∂t
with the boundary condition (5.4.1)
u=g on ∂Ω×]0, T [,
and the initial condition (5.2.4)
u(0) = ut 0 in Ω,
satisfying the compatibility conditions (5.2.5)
div ut 0 = 0 in Ω
ut 0 = g(0) on ∂Ω.
The incompressibility condition (5.2.2), div u = 0 in Ω×]0, T [, is embodied in the

space W .
Remark 5.7.1 The fact that u belongs to H 1 (0, T ; W ) implies that u ∈ C0 ([0, T ]; W )
and therefore the initial condition (5.2.4) makes sense.
Following Section 5.6.1 and setting
z = curl(u − α Δ u) in Ω×]0, T [ , z = (0, 0, z), (5.7.1)
the triple (u, p, z) solves the problem: Find u ∈ H 1 (0, T ; W ), p ∈ L 2 (0, T ; L 2m (Ω))
and z ∈ L ∞ (0, T ; L 2 (Ω)) solution of
∂
(u − αΔ u) − μΔ u + z × u + ∇ p = f in Ω×]0, T [, (5.7.2)
∂t
where z × u = (−zu 2 , zu 1 ), with the incompressibility condition (5.2.2), the tan-

gential boundary condition (5.4.1), and initial condition (5.2.4), and the transport
equation
∂z
α + μ z + α u · ∇ z = μ curl u + α curl f , (5.7.3)
∂t
with initial condition
z(0) = z t 0 = curl(ut 0 − α Δ ut 0 ) in Ω. (5.7.4)
Remark 5.7.2 If z ∈ L ∞ (0, T ; L 2 (Ω)) satisfies (5.7.3), then the initial condition
(5.7.4) makes sense because z belongs to C0 ([0, T ]; W −1, 3 (Ω)). Indeed, observe
4
4
that on one hand the product zu belongs in particular to L 2 (0, T ; L 3 (Ω)2 ), and on
the other hand
div(zu) = u · ∇ z,
as div u = 0. Thus, u · ∇ z belongs to L 2 (0, T ; W −1, 3 (Ω)) and the regularity of the
4
remaining terms in (5.7.3) implies that
∂z
∈ L 2 (0, T ; W −1, 3 (Ω)).
4
∂t
Hence z belongs to H 1 (0, T ; W −1, 3 (Ω)).

4

The following proposition shows that problem (5.7.2), (5.2.2) (5.4.1), (5.2.4),
(5.7.3), (5.7.4) is equivalent to the original problem.
5.7 Splitting: The Time-Dependent System in 2-D 251
Proposition 5.7.3 The original problem (5.2.1), (5.2.2), (5.4.1), (5.2.4) and the split
problem (5.7.2), (5.2.2), (5.4.1), (5.2.4), (5.7.3), (5.7.4) are equivalent.
Proof It suffices to establish that if (u, p, z) solves the split problem (5.7.2), (5.2.2),
(5.4.1), (5.2.4), (5.7.3), (5.7.4), then (u, p) is a solution of (5.2.1), (5.2.2), (5.4.1),
(5.2.4). In view of (5.7.2), we only need to recover the relation (5.7.1) between z and
u. To this end, let us take the curl of (5.7.2)
∂
curl (u − αΔ u) − μΔ u + z × u + ∇ p = curl f .
∂t
Multiplying by α and using (5.6.4), this yields in Ω×]0, T [
∂
α curl(u − αΔ u) − α μ Δ(curl u) + α u · ∇ z = α curl f .
∂t
Then comparing with (5.7.3), we obtain in Ω×]0, T [,
∂
α z − curl(u − αΔ u) + μ z − curl(u − αΔ u) = 0.
∂t
Therefore, z − curl(u − αΔ u) is a solution of the ordinary differential equation in

time, where x plays the part of a parameter: Find ϕ ∈ H 1 (0, T ; H −2 (Ω)) such that
∂ϕ
α + μ ϕ = 0 in ]0, T [,
∂t
ϕ(0) = 0.
Note that since z − curl(u − αΔ u) belongs to L ∞ (0, T ; H −2 (Ω)) then this equation
implies that z − curl(u − αΔ u) belongs at least to H 1 (0, T ; H −2 (Ω)). But then
μ
ψ = exp t ϕ,
α
is a solution of ψ (t) = 0, in [0, T ], ψ(0) = 0, which has the unique solution ψ = 0;

hence ϕ = 0. Therefore, z = curl(u − αΔ u).
In semi variational form, the split system (5.7.2), (5.2.2), (5.4.1), (5.2.4), (5.7.3),
(5.7.4) reads: Find u ∈ H 1 (0, T ; W ), p ∈ L 2 (0, T ; L 2m (Ω)), z ∈ L ∞ (0, T ; L 2 (Ω))
satisfying a.e. in ]0, T [
∀v ∈ H01 (Ω)2 , (u (t), v) + α(∇ u (t), ∇ v)+μ(∇ u(t), ∇ v) + (z(t) × u(t), v)
− ( p(t), div v) = ( f (t), v), (5.7.5)
α z (t) + μ z(t) + α u(t) · ∇ z(t) = μ curl u(t) + α curl f (t), (5.7.6)

with the boundary condition (5.4.1)

u(t) = g(t) on ∂Ω,
and initial conditions (5.2.4) and (5.7.4)
u(0) = ut 0 , z(0) = z t 0 = curl(ut 0 − α Δ ut 0 ) in Ω .
Clearly, this formulation is equivalent to the split system.
The a priori estimates in this section are derived provided problem (5.2.1), (5.2.2),
(5.4.1), (5.2.4) has a sufficiently smooth solution.
In order to handle the nonhomogeneous boundary condition, we proceed as in
Section 5.6.3, namely we suppose that g belongs to H 1 (0, T ; W 1− r ,r (∂Ω)2 ) for
1
some r > 2. For any ε > 0, small enough (so that (5.6.17) holds), we lift g with u g,ε
defined in (5.6.21). The difference u0 = u − u g,ε ∈ V solves for all v in H01 (Ω)2
the following:

−( p(t), div v) = ( f (t), v) − (u g (t),ε , v) − α(∇ u g (t),ε , ∇ v)
− μ(∇ u g(t),ε , ∇ v) − (z(t) × u g(t),ε , v), (5.7.7)
u0 (0) = ut 0 − u g(0),ε . (5.7.8)
This problem is equivalent to (5.7.5), (5.2.2), (5.4.1), (5.2.4). Recall the norm vα
defined in (5.2.37)
v2α = v2L 2 (Ω) + αv2H 1 (Ω) .
0
The first two propositions state conditional estimates, valid for any small enough
ε > 0.

α > 0, f ∈ L 2 (Ω×]0, T [)2 , and g ∈ H 1 (0, T ; W 1− r ,r (∂Ω)2 ) for some r > 2 satis-
1
fying g·n = 0. Let (u, p, z) in L ∞ (0, T ; W )× L 2 (0, T ; L 2m (Ω))× L 2 (Ω×]0, T [) be

any solution of problem (5.7.7)–(5.7.8). Then for any arbitrary ε > 0 small enough
for (5.6.17), u0 satisfies a.e. in ]0, T [
t
u0 (t)2α + μ u0 (s)2H 1 (Ω) ds
0
0
4 2 2
≤ Aε (t) + C ε g2 ∞ 1 z2L 2 (Ω×]0,t[) , (5.7.9)
μ L (0,T ;W 1− r ,r (∂Ω)2 )
where
t
Aε (t) = u0 (0)2α + 4μ u g(s),ε 2H 1 (Ω) ds
0
0
t
4 2
+ S0,2 f 2L 2 (Ω×]0,t[) + (S0,2
2
+ α) u g (s),ε 2α ds , (5.7.10)
μ 0
and C is the constant of Proposition 5.6.5.
Proof The choice v = u0 (t) ∈ V in (5.7.7), Proposition 5.6.5, and the inequality
2
∀v ∈ H01 (Ω), v2α ≤ S0,2 + α v2H 1 (Ω) ,
0
yield
1 d 2 1
u0 (t)2α + μu0 (t)2H 1 (Ω) ≤ S0,2 f (t) L 2 (Ω) + S0,2 + α 2 u g (t),ε α
2 dt 0

+ μu g(t),ε H01 (Ω) + Cεg(t)W 1− r1 ,r (∂Ω) z(t) L 2 (Ω) u0 (t) H01 (Ω) .
Then (5.7.9) follows by suitable applications of Young’s inequality and integration

over ]0, t[.
Proposition 5.7.5 In addition to the assumptions and notation of Proposition 5.7.4,

we suppose that f belongs to L 2 (0, T ; H (curl, Ω)) and z ∈ L ∞ (0, T ; L 2 (Ω))
solves (5.7.6) and (5.7.4). Then z satisfies a.e. in ]0, T [
t
αz(t)2L 2 (Ω) + μz2L 2 (Ω×]0,t[) ≤ Bε (t) + 3μ u0 (s)2H 1 (Ω) ds, (5.7.11)
0
0
where
t
3α 2
Bε (t) = αz t 0 2L 2 (Ω) +6μ u g(s),ε 2H 1 (Ω) ds + curl f 2L 2 (Ω×]0,t[) . (5.7.12)
0 0 μ
Proof Note that (5.7.6) is a particular case of (7.3.1) with β = α and f = μ curl u +
α curl f . Therefore, according to Theorem 7.3.8, it has a unique solution and (7.3.7)
gives
1
αz(t)2L 2 (Ω) + μz2L 2 (Ω×]0,t]) ≤ αz t 0 2L 2 (Ω) + μ curl u + α curl f 2L 2 (Ω×]0,t]) .
μ
Then (5.7.11) is an easy consequence of the following bound, valid a.e. in ]0, T [:
μ curl u + α curl f 2L 2 (Ω) = μ curl(u0 + u g,ε ) + α curl f 2L 2 (Ω)

≤ 3 μ2 u0 2H 1 (Ω) + 2μ2 u g,ε 2H 1 (Ω) + α 2 curl f 2L 2 (Ω) ,
0 0
and the proof is complete.

By combining these two propositions, we obtain first a pointwise bound for z

and a conditional bound for u0 . To simplify, we introduce the following abbreviated
notation:
g∞ = g L ∞ (0,T ;W 1− r1 ,r (∂Ω)2 ) , g2,t = g L 2 (0,t;W 1− r1 ,r (∂Ω)2 ) . (5.7.13)
Proposition 5.7.6 Under the assumptions and notation of Proposition 5.7.5, the
parts u0 and z of any solution
(u, p, z) ∈ L ∞ (0, T ; W ) × L 2 (0, T ; L 2m (Ω)) × L ∞ (0, T ; L 2 (Ω))
of (5.7.5), (5.7.6), (5.4.1), (5.2.4) and (5.7.4) satisfy a.e. in ]0, T [
αz(t)2L 2 (Ω) ≤ 3A(t) + B(t), (5.7.14)
and
t
μ
u0 (t)2α + μ u0 (s)2H 1 (Ω) ds ≤ A(t) + z2L 2 (Ω×]0,t[) , (5.7.15)
0 0 3
with
4 2
A(t) = u0 (0)2α + S f 2L 2 (Ω×]0,t[) + C1 g∞ g22,t
μ 0,2
1 α
+ C2 (S0,2
2
+ α)g 22,t + 2 g∞ , (5.7.16)
g∞ μ
3α 2
B(t) = αz t0 2L 2 (Ω) + curl f 2L 2 (Ω×]0,t[) + C3 g∞ g22,t , (5.7.17)
μ
where Ci , 1 ≤ i ≤ 3, denote various constants independent of t, μ, and α.
Proof From (5.7.9), we infer

t
12 2 2
3μ u0 (s)2H 1 (Ω) ds ≤ 3Aε (t) + C ε g2∞ z2L 2 (Ω×]0,t[) ,
0 0 μ
and by substituting this bound into (5.7.11), we obtain that
12 2 2
αz(t)2L 2 (Ω) + μz2L 2 (Ω×]0,t[) ≤ Bε (t) + 3Aε (t) + C ε g2∞ z2L 2 (Ω×]0,t[) .
μ
Then, choosing in inequalities (5.6.23) and (5.6.24), both with s = 2,
μ 1
ε= √ , (5.7.18)
2 3 Cg∞
we get (5.7.14). Estimate (5.7.15) follows immediately by substituting (5.7.18) into

(5.7.9).
Next, we derive an unconditional bound for u0 .
Proposition 5.7.7 Under the assumptions and notation of Proposition 5.7.6, the part
u0 of any solution (u, p, z) ∈ L ∞ (0, T ; W )× L 2 (0, T ; L 2m (Ω))× L ∞ (0, T ; L 2 (Ω))
of (5.7.5), (5.7.6), (5.4.1), (5.2.4) and (5.7.4) satisfies
t
u0 (t)2α + μ u0 (s)2H 1 (Ω) ds ≤ 3A(t) + C(t), (5.7.19)
0
0
where, with a constant C4 independent of t, μ, and α,
4α 2
C(t) = αz t 0 2L 2 (Ω) + curl f 2L 2 (Ω×]0,t[) + C4 g∞ g22,t . (5.7.20)
μ
Proof If the pointwise bound (5.7.14) for z is substituted into (5.7.9) for estimating
u0 , it leads to a suboptimal bound. A sharper bound can be obtained by decreasing
the coefficient of u0 in estimate (5.7.11). For this, since we cannot change ε, we
modify (5.7.11) as follows by grouping terms differently,
4α 2
αz(t)2L 2 (Ω) +μz2L 2 (Ω×]0,t[) ≤ αz t 0 2L 2 (Ω) + curl f 2L 2 (Ω×]0,t[)
μ
t

+ 2μ u0 (s)2H 1 (Ω) + 4u g(s),ε 2H 1 (Ω) ds.
0 0
0
After substituting (5.7.15), this becomes

μ t
αz(t)2L 2 (Ω) + z2L 2 (Ω×]0,t[) ≤α z t 0 2L 2 (Ω) + 2A(t) + 8μ u g(s),ε 2H 1 (Ω) ds
3 0 0
4α 2
+ curl f 2L 2 (Ω×]0,t[) ,
μ
and (5.7.18) and (5.6.24) with s = 2 imply

μ
αz(t)2L 2 (Ω) + z2L 2 (Ω×]0,t[) ≤ 2A(t) + C(t), (5.7.21)
3
where C is defined by (5.7.20). Then (5.7.19) follows immediately from (5.7.21) and
(5.7.15).
Next, we obtain a bound for the time derivative of u0 .
Proposition 5.7.8 In addition to the assumptions and notation of Proposition 5.7.7,

we assume that u ∈ L 2 (0, T ; W ). Then u0 satisfies the bound,
t
u0 (s)2α ds + μu0 (t)2H 1 (Ω) ≤ D1 (t) + D2 (t), a.e. in ]0, T [, (5.7.22)
0
0
where D1 and D2 are given by
1 μ2 3 2
D1 (t) = z2L 2 (Ω×]0,t[) + √ S0,4 (3A(t) + C(t)) ,
α 2 2α
μ
D2 (t) = μu0 (0) H 1 (Ω) + 2 f L 2 (Ω×]0,t[) + C5 g∞ g22,t
2 2 (5.7.23)
0
μ α
2 α
+ C6 g 2,t + g∞ ,
g∞ μ
with constants C5 and C6 independent of t, μ and α.

Proof The regularity assumption on u allows us to choose v = u0 (t) in (5.7.7). This
leads to a straightforward bound apart from the treatment of (z(t) × u0 (t), u0 (t))
that is not zero. Since we are in two dimensions, we can use the interpolation bound
(3.1.23), that is to say
1 1 1
v 2 2 v 2 1 ,
∀v ∈ H01 (Ω), v L 4 (Ω) ≤ 1 L (Ω) H0 (Ω)
24
to obtain
1
|(z(t) × u0 (t), u0 (t))| ≤ 2− 4 S0,4 z(t) L 2 (Ω) u0 (t) L2 2 (Ω)
1
1
× u0 (t) H2 1 (Ω) u0 (t) H01 (Ω) = H(t).
0
Then in view of (5.6.25), we get
μ d
u0 (t)2α + u0 (t)2H 1 (Ω) ≤ H(t) + f (t) L 2 (Ω) u0 (t) L 2 (Ω)
2 dt 0
+ u g (t),ε α u0 (t)α + μu g(t),ε H01 (Ω) u0 (t) H01 (Ω)

+ Cεz(t) L 2 (Ω) g∞ u0 (t) H01 (Ω) .
Next, using the straightforward bound
1
∀v ∈ H 1 (Ω), v L 2 (Ω) v H01 (Ω) ≤ √ v2α ,
2 α
repeated applications of Young’s inequality, and (5.7.18), we derive
d 1 μ2 3 2
u0 (t)2α + μ u0 (t)2H 1 (Ω) ≤ z(t)2L 2 (Ω) + √ S0,4 u0 (t)2α
dt 0 α 2 2α
μ2
+ 2 f (t)2L 2 (Ω) + 6 u g(t),ε 2H 1 (Ω) + 2u g (t),ε 2α .
α 0
Finally, using in this inequality the monotonicity of A and C, (5.7.19), (5.6.23) and
(5.6.24) both with s = 2, (5.7.22) follows readily.
This last bound induces an estimate for the pressure.

Proposition 5.7.9 We retain the assumptions of Proposition 5.7.8. Then p satisfies
the following bound for any t ∈]0, T [:
5 3 2 μ2 2
S0,4
p2L 2 (Ω×]0,t[) ≤ 1 + S0,2 + √ (3A(t)
K̃2 α 6 2α
2
+ C(t)) z L 2 (Ω×]0,t[) + E(t) , (5.7.24)
with

E(t) = S0,2
2
f 2L 2 (Ω×]0,t[) + D2 (t) + C7 μg∞ g22,t
μ α
+ C8 (S0,2
2
+ α)g 22,t + g∞ , (5.7.25)
g∞ μ
where D2 is given by (5.7.23), C7 and C8 are constants independent of t, μ and α,

and K̃ is the isomorphism constant of the divergence operator in (3.2.12).
Proof Recall the operator T̃ defined in Theorem 3.2.11, and choose v = T̃ ( p(t)) in
(5.7.5). The orthogonality of v implies
( p(t), div v) = (u (t), v) + α(∇ u g (t),ε , ∇ v) + μ(∇ u g(t),ε , ∇ v)

+ (z(t) × u0 (t), v) + (z(t) × u g(t),ε , v) − ( f (t), v),
and therefore,

( p(t), div v) ≤ v L 2 (Ω) f (t) L 2 (Ω) + u (t) L 2 (Ω)
0
+ u g (t),ε α vα + μu g(t),ε H01 (Ω) v H01 (Ω)
S0,4 1 1
+ z(t) L 2 (Ω) v H01 (Ω) 1 u0 (t) L2 2 (Ω) u0 (t) H2 1 (Ω) + Cεg∞ .
24 0
Then inequality (3.2.12) and the choice (5.7.18) of ε yield
1
p(t) L 2 (Ω) ≤ S0,2 f (t) L 2 (Ω) + u0 (t) L 2 (Ω)
K̃
1
+ (S0,2
2
+ α) 2 u g (t),ε α + μu g(t),ε H01 (Ω)
S0,4 1 1 μ
+ z(t) L 2 (Ω) 1 u0 (t) L2 2 (Ω) u0 (t) H2 1 (Ω) + √ .
24 0 2 3
By squaring both sides and integrating over ]0, t[, we obtain
5
p2L 2 (Ω×]0,t[) ≤ 2
S0,2 f 2L 2 (Ω×]0,t[) + u0 2L 2 (Ω×]0,t[)
K̃2
t t
+ (S0,2
2
+ α) u g (s),ε 2α ds + μ2 u g(t),ε 2H 1 (Ω) ds
0
0 0
2
2
S0,4 μ
+ z2L 2 (Ω×]0,t[) √ sup u0 (s)2α + .
2α s∈]0,t[ 6
Substituting (5.7.19) and (5.7.22) into this inequality, gives (5.7.24).
Of course, z is eliminated from (5.7.22) and (5.7.24) by substituting (5.7.14) into

these inequalities. The next theorem summarizes the a priori bounds of this section.
Theorem 5.7.10 Let Ω be bounded, connected and Lipschitz-continuous. Let f in
L 2 (0, T ; H (curl, Ω)), and g satisfying g · n = 0 in H 1 (0, T ; W 1− r ,r (∂Ω)2 ) for
1
some r > 2. Assume that μ > 0 and α > 0.

Then any solution (u, p, z) of problem (5.7.2), (5.2.2), (5.4.1), (5.2.4), (5.7.3),
and (5.7.4), belonging to H 1 (0, T ; W ) × L 2 (0, T ; L 2m (Ω)) × L ∞ (0, T ; L 2 (Ω)), is
bounded by the data,
u H 1 (0,T ;W ) + p L 2 (Ω×]0,T [) + z L ∞ (0,T ;L 2 (Ω)) ≤ C, (5.7.26)
where C depends only on Ω, α, μ, f L 2 (0,T ;H (curl,Ω)) , g H 1 (0,T ;W 1− r1 ,r (∂Ω)2 ) and

ut 0 .
5.7.2 Existence of Solutions by Semi-discretization in Time
We shall construct a solution
(u, p, z) ∈ H 1 (0, T ; W ) × L 2 (0, T ; L 2m (Ω)) × L ∞ (0, T ; L 2 (Ω))
of Problem (5.7.5), (5.7.6), (5.4.1), (5.2.4), (5.7.4) by semi-discretization in time.

Let N > 1 be an integer, define the time step k by
T
k= ,
N
and the subdivision points t n = n k. For each n ≥ 1, we approximate f (t n ) and

g(t n ) by averages defined almost everywhere in Ω, respectively in ∂Ω,
tn tn
1 1
f (x) =
n
f (x, t)dt , g (s) =
n
g(s, t)dt. (5.7.27)
k t n−1 k t n−1
We set
u0 = ut 0 , z 0 = z t 0 = curl(ut 0 − αΔ ut 0 ) , g 0 = g(0). (5.7.28)
Note that the condition g · n = 0 carries over to g i , 0 ≤ i ≤ N . Then, our semi-

discrete problem reads: Starting from u0 and z 0 , find sequences (un )n≥1 , (z n )n≥1 and
( p n )n≥1 such that un ∈ W , z n ∈ L 2 (Ω) and p n ∈ L 2m (Ω) solve
1 n 1
(u − un−1 ) − α Δ(un − un−1 ) − μ Δ un + z n−1 × un + ∇ p n = f n , (5.7.29)
k k
un |∂Ω = g n , (5.7.30)
1
α (z n − z n−1 ) + μ z n + α un · ∇ z n = μ curl un + α curl f n , (5.7.31)
k
where z n = (0, 0, z n ). In variational form, these two problems read: Find un ∈ W ,

satisfying (5.7.30), z n ∈ L 2 (Ω) and p n ∈ L 2m (Ω) solution of
1 n α
∀v ∈ H01 (Ω)2 , (u − un−1 , v) + (∇(un − un−1 ), ∇ v)
k k
+ μ (∇ un , ∇ v) + (z n−1 × un , v) − ( p n , div v)
= ( f n , v), (5.7.32)
and
α n
∀θ ∈ L 2 (Ω), (z − z n−1 , θ ) + μ (z n , θ ) + α (un · ∇ z n , θ )
k
= μ (curl un , θ ) + α (curl f n , θ ). (5.7.33)
For each n ≥ 1, given z n−1 , (5.7.32) is essentially a steady Stokes problem and it is
easy to check that it has a unique solution (un , p n ). In turn, given un , (5.7.33) is a
steady transport equation, and owing to Proposition 7.2.6, it has a unique solution,
since curl f n belongs to L 2 (Ω).
5.7.2.1 A Priori Estimates for the Sequences
The solutions of these systems satisfy a priori estimates that are very similar to the
formal estimates of Section 5.7.1, and we revert to the setting of this section. Thus,
we assume that g belongs to H 1 (0, T ; W 1− r ,r (∂Ω)2 ) for some r > 2. For any ε > 0,
1
small enough (so that (5.6.17) holds), we lift g n with u g n ,ε defined in (5.6.21). The
difference un0 = un − u g n ,ε ∈ V solves for all v in H01 (Ω)2 ,
1 n α
(u0 − u0n−1 , v) + (∇(un0 − u0n−1 ), ∇ v) + μ (∇ un0 , ∇ v) + (z n−1 × un0 , v)
k k
1 α
− ( p , div v) = ( f n , v) − (u g n ,ε − u g n−1 ,ε , v) − (∇(u g n ,ε − u g n−1 ,ε ), ∇ v)
n
k k
− μ (∇ u g n ,ε , ∇ v) − (z n−1 × u g n ,ε , v), (5.7.34)
starting from
u00 = u0 − u g 0 ,ε . (5.7.35)
Then, defining the analogues of Aε (t) and Bε (t),

n
4 2
Anε = u00 2α + 4μ ku gi ,ε 2H 1 (Ω) + S f 2L 2 (Ω×]0,t n [)
i=1
0 μ 0,2

n
1
+ (S0,2
2
+ α) u gi ,ε − u gi−1 ,ε 2α , (5.7.36)
i=1
k

n
3α 2
Bnε = αz t 0 2L 2 (Ω) + 6μ ku gi ,ε 2H 1 (Ω) + curl f 2L 2 (Ω×]0,t n [) , (5.7.37)
i=1
0 μ
we have the same conditional estimates, valid for any small enough ε > 0, as in
Propositions 5.7.4 and 5.7.5. We skip the proofs, since they are much the same.

α > 0, f ∈ L 2 (Ω×]0, T [)2 , and g ∈ H 1 (0, T ; W 1− r ,r (∂Ω)2 ) for some r > 2
1
satisfying g · n = 0. The sequence un0 defined by (5.7.34) and (5.7.35) satisfies, for
1 ≤ n ≤ N , for any arbitrary ε > 0 small enough for (5.6.17),

n
n
un0 2α + ui0 − ui−1
0 α + μ
2
kui0 2H 1 (Ω)
0
i=1 i=1
4 n−1
≤ Anε + C 2 ε2 g2 ∞ n 1− r1 ,r kz i 2L 2 (Ω) , (5.7.38)
μ L (0,t ;W (∂Ω) )
2
i=0
where · α is defined by (5.2.37), Anε is defined by (5.7.36), and C is the constant

of Proposition 5.6.5.
Proposition 5.7.12 In addition to the assumptions and notation of Proposition

5.7.11, we suppose that f belongs to L 2 (0, T ; H (curl, Ω)). The sequence z n defined
by (5.7.33) and (5.7.28) satisfies for 1 ≤ n ≤ N ,

n
n
αz n 2L 2 (Ω) + α z i − z i−1 2L 2 (Ω) + μ kz i 2L 2 (Ω)
i=1 i=1

n
≤ Bnε + 3μ kui0 2H 1 (Ω) . (5.7.39)
0
i=1
As in Section 5.7.1, these two propositions can be combined to yield an uncondi-

tional bound for z n . First, we estimate the terms in Anε that involve u gi ,ε . To begin
with, we easily derive from (5.6.24) with s = 2,

n
C2
ku gi ,ε 2H 1 (Ω) ≤ g22,t n , 1 ≤ n ≤ N , (5.7.40)
i=1
0 ε
where we have used the notation (5.7.13) for · 2,t . The next proposition bounds
the differences u gi ,ε − u gi−1 ,ε .
Proposition 5.7.13 Let g ∈ H 1 (0, T ; W 1− r ,r (∂Ω)2 ) for some r > 2. For 1 ≤ n ≤

1
N , we have
n
1 4 α
u gi ,ε − u gi−1 ,ε 2α ≤ C ε + g 2,t n . (5.7.41)
i=1
k 3 ε
Proof On one hand, Propositions 5.6.3 and 5.6.4 with s = 2 imply

α
u gi ,ε − u gi−1 ,ε 2α ≤ C ε + g i − g i−1 2 1− r1 ,r . (5.7.42)
ε W (∂Ω)2 )
On the other hand, a simple calculation gives for all 1 ≤ i ≤ N ,

ti ti
1 1
g(t) dt = g(t ) − i
(t − t i−1 )g (t) dt. (5.7.43)
k t i−1 k t i−1
Therefore for 2 ≤ i ≤ N ,

1 ti

t i−1
g −g
i i−1
= (t − t)g (t) dt −
i
(t − t i−2 )g (t) dt . (5.7.44)
k t i−1 t i−2
By applying (5.7.43) with i = 1, we extend (5.7.44) to i = 1 by

1 k
1 k
g 1 − g 0 = g(k) − g(0) − t g (t) dt = k − t g (t) dt. (5.7.45)
k 0 k 0
Then (5.7.41) is a straightforward consequence of (5.7.42), (5.7.44), (5.7.45), and

several applications of the Cauchy–Schwarz inequality.
Now, with the notation (5.7.13), we choose ε through (5.7.18), that is to say
μ 1
ε= √ ,
2 3 Cg∞
and with this choice, estimates (5.7.40) and (5.7.41) yield the following bounds for
Anε and Bnε :
Proposition 5.7.14 The choice (5.7.18) for ε gives
4 2
Anε ≤ An := u00 2α + S f 2L 2 (Ω×]0,t n [) + C1 g∞ g22,t n
μ 0,2
4 1 α
+ C2 (S0,2
2
+ α)g 22,t n + 2 g∞ , (5.7.46)
3 g∞ μ
3α 2
Bnε ≤ Bn := αz 0 2L 2 (Ω) + curl f 2L 2 (Ω×]0,t n [) + C3 g∞ g22,t n , (5.7.47)
μ
with the constants Ci , i = 1, 2, 3, of (5.7.16) and (5.7.17).
The above results yield the analogue of Proposition 5.7.6, with the same proof.
sequences z n and un0 satisfy the following bounds for 1 ≤ n ≤ N :

n
μ
n−1
un0 2α + μ kui0 2H 1 (Ω) ≤ An + kz i 2L 2 (Ω) , (5.7.48)
i=1
0 3 i=0
and
αz n 2L 2 (Ω) ≤ 3An + Bn + μ kz 0 2L 2 (Ω) . (5.7.49)
Similarly, defining Cn by
4α 2
Cn = αz 0 2L 2 (Ω) + curl f 2L 2 (Ω×]0,t n [) + C4 g∞ g22,t n , (5.7.50)
μ
with the constant C4 of Proposition 5.7.7, we have the following unconditional

bounds:
sequences z n and un0 satisfy, for 1 ≤ n ≤ N ,

n
μ
n
α z n 2L 2 (Ω) + α z i −z i−1 2L 2 (Ω) + kz i 2L 2 (Ω)
i=1
3 i=1
2
≤ 2An + Cn + μ kz 0 2L 2 (Ω) , (5.7.51)
3

n
un0 2α + μ kui0 2H 1 (Ω) ≤ 3An + Cn + μ kz 0 2L 2 (Ω) . (5.7.52)
0
i=1
The next proposition, analogous to Proposition 5.7.8 estimates the sequence of

divided differences k1 (ui0 − ui−1
0 ). Beforehand, we introduce the quantity
μ
Dn = μu00 2H 1 (Ω) + 2 f 2L 2 (Ω×]0,t n [) + C5 g∞ g22,t n
0 α
4 2
μ α
+ C6 g 2,t n + g∞ , (5.7.53)
3 g∞ μ
with the constants C5 and C6 of (5.7.23).
Proposition 5.7.17 In addition to the assumptions and notation of Proposition

5.7.15, we suppose that g belongs to H 1 (0, T ; W 1− r ,r (∂Ω)2 ) for some r > 2.
1
Then we have for 1 ≤ n ≤ N ,

n
1
n
ui0 − ui−1
0 α + μu0 H 1 (Ω) + μ
2 n 2
ui0 − ui−1
0 H 1 (Ω) ≤ D
2 n
i=1
k 0
i=1
0
1 μ2
n−1
3 2
+ kz i 2L 2 (Ω) + √ S0,4 (3An + Cn + μ kz 0 2L 2 (Ω) ) . (5.7.54)
α i=0 2 2α
Proof By taking the scalar product of both sides of (5.7.34) with un0 − u0n−1 and
proceeding as in the proof of Proposition 5.7.8, we infer
1 n
u0 − u0n−1 2α + μ un0 2H 1 (Ω) − u0n−1 2H 1 (Ω) + un0 − u0n−1 2H 1 (Ω)
k 0 0 0
k n−1 2 μ2 3 2
≤ z L 2 (Ω) + √ S0,4 un0 2α + 2k f n 2L 2 (Ω)
α 2 2α
6μ2 2
+ ku g n ,ε 2H 1 (Ω) + u g n ,ε − u g n−1 ,ε 2α .
α 0 k
Then (5.7.54) readily follows by summing over n and substituting (5.7.52), the def-
inition (5.7.53), and (5.7.42) into the above inequality.
Finally, we deduce from (5.7.54) the following bound for the pressure. The proof
is skipped because it follows the lines of that of Proposition 5.7.9.
Proposition 5.7.18 We retain the assumptions and notation of Proposition 5.7.17.

Then we have for 1 ≤ n ≤ N ,
5 3 2 μ2
n 2
S0,4
k pi 2L 2 (Ω) ≤ 1 + S0,2 + √ (3An + Cn + μ kz 0 2L 2 (Ω) )
i=1 K̃2 α 6 2α

n−1
× kz i 2L 2 (Ω) + En , (5.7.55)
i=0
where

En = S0,2
2
f 2L 2 (Ω×]0,t n [) + Dn + C7 μg∞ g22,t n
4 2 μ α
+ C8 (S0,2 + α)g 22,t n + g∞ , (5.7.56)
3 g∞ μ
with the constants C7 and C8 of (5.7.25) and the constant K̃ of (3.2.12).

The next theorem summarizes the a priori bounds of this subsection.
Theorem 5.7.19 Let Ω be bounded, connected and Lipschitz-continuous, μ > 0,
α > 0. Let f and g, satisfying g · n = 0, be given in L 2 (0, T ; H (curl, Ω)), and
H 1 (0, T ; W 1− r ,r (∂Ω)2 ) for some r > 2, respectively. Then any solution (un , p n , z n )
1
of (5.7.28)–(5.7.31) is bounded by the data as follows:

n
n
un 2H 1 (Ω) + k pi 2L 2 (Ω) + z n 2L 2 (Ω)) + z i − z i−1 2L 2 (Ω)
i=1 i=1

n
1
+ ui − ui−1 2H 1 (Ω) ≤ C, (5.7.57)
i=1
k
where C depends only on Ω, α, μ, f L 2 (0,T ;H (curl,Ω)) , g H 1 (0,T ;W 1− r1 ,r (∂Ω)2 ) , and

ut 0 .
5.7.2.2 Passing to the Limit
Here, it is convenient to transform the sequences {un }, { p n }, {z n }, { f n } and {g n }

into functions. Since {un }, {z n } and {g n } need to be “differentiated”, we define the
piecewise linear functions in time
t − t n−1 n
uk (t) = un−1 + (u − un−1 ),
k
t − t n−1 n
z k (t) = z n−1 + (z − z n−1 ),
k
t − t n−1 n
g k (t) = g n−1 + (g − g n−1 ),
k
for all t ∈ [t n−1 , t n ] and n, 1 ≤ n ≤ N .

Next, in view of the other terms in (5.7.29), (5.7.31), we define the step functions
in time for all t ∈]t n−1 , t n ] and n, 1 ≤ n ≤ N ,
f k (t) = f n , wk (t) = un , pk (t) = p n , ζk (t) = z n ,
and for all t ∈ [t n−1 , t n [, 1 ≤ n ≤ N ,
λk (t) = z n−1 .
More generally, we denote by m( f ) the mean value of any function f in L 2 (0, T ),

tn
1
∀ f ∈ L (0, T ), ∀t ∈]t
2 n−1
, t ], m( f )(t) =
n
f (t) dt , 1 ≤ n ≤ N .
k t n−1
The mean value satisfies
∀ f ∈ L 2 (0, T ), lim m( f ) − f L 2 (0,T ) = 0. (5.7.58)

k→0
As a consequence, we have
∀ f ∈ L 2 (Ω×]0, T [)2 , lim f k − f L 2 (Ω×]0,T [) = 0. (5.7.59)

k→0
In addition, we have the following lemma:

Lemma 5.7.20 Let g, satisfying g · n = 0, be given in H 1 (0, T ; W 1− r ,r (∂Ω)2 ) for
1
some r > 2 and let u g,ε be defined by (5.6.21) for any small enough parameter ε.
Then
lim u g k ,ε − u g,ε H 1 (0,T ;W ) = 0. (5.7.60)
k→0
Proof On any interval ]t i−1 , t i ], 1 ≤ i ≤ N , we can write a.e on the set ∂Ω
t − t i−1 i t − ti i
(g k − g)(t) = g i−1 − g(t) + g − g i−1 = g i − g(t) + g − g i−1 .
k k
By substituting (5.7.44), respectively (5.7.45), into this equality, for 2 ≤ i ≤ N ,

t − ti ti

t i−1
(g k − g)(t) = m(g) − g(t) + 2 (t − t)g (t) dt −
i
(t − t i−2 )g (t) dt ,
k t i−1 t i−2
and respectively for i = 1,

t −k k
(g k − g)(t) = m(g) − g(t) + 2 k − t g (t) dt.
k 0
This yields the bound
g k − g2 2 1 ≤ 2m(g) − g2 2 1

L (0,T ;W 1− r ,r (∂Ω)2 ) L (0,T ;W 1− r ,r (∂Ω)2 )
8
+ k 2 g 2 2 1 ,
9 L (0,T ;W 1− r ,r (∂Ω)2 )
whence
lim g k − g L 2 (0,T ;W 1− r1 ,r (∂Ω)2 ) = 0. (5.7.61)
k→0
Regarding the time derivative, we write
1 i
g k − g = g − g(t i ) − g i−1 − g(t i−1 ) + m(g ) − g .
k
Thus, in view of (5.7.58), it suffices to investigate the first term. To this end, we use
the following expansion, valid for smooth enough g:

1 i 1 ti
g − g(t i ) = 2 g(t) − g(t i ) dt
k k t i−1
ti t
1
= (t − t i )g (t i ) + (t − τ )g (τ ) dτ dt
k2 t i−1 ti
ti
1 1
= − g (t i ) + (t i−1 − τ )2 g (τ ) dτ.
2 2 k2 t i−1
Hence we have, for 2 ≤ i ≤ N ,

1 i 1 ti 1 ti
g −g(t i )− g i−1 −g(t i−1 ) = − g (t) dt+ (t i−1 −t)2 g (t) dt
k 2 t i−1 2 k2 t i−1
t i−1
1
− (t i−2 − t)2 g (t) dt,
2 k2 t i−2
with a simpler expression when i = 1. Collecting these expressions, we obtain

the following bound (to simplify, we do not exhibit the constant) for all g ∈
H 2 (0, T ; W 1− r ,r (∂Ω)2 ):
1
g k − g 2 2 1
L (0,T ;W 1− r ,r (∂Ω)2 )

≤ C m(g ) − g 2 2 1 + k 2 g 2 2 1 ,
L (0,T ;W 1− r ,r (∂Ω)2 ) L (0,T ;W 1− r ,r (∂Ω)2 )
and the density of H 2 (0, T ) into H 1 (0, T ) implies that
lim g k − g L 2 (0,T ;W 1− r1 ,r (∂Ω)2 ) = 0. (5.7.62)

k→0
Finally, (5.7.60) follows from the linear dependence of u g,ε on g, (5.6.23), and
(5.6.24).
Regarding the unknowns, we have the following convergences.
Lemma 5.7.21 There exist functions u ∈ H 1 (0, T ; W ), p ∈ L 2 (0, T ; L 2m (Ω)) and

z ∈ L ∞ (0, T ; L 2 (Ω)) such that a subsequence of k, still denoted by k, satisfies
lim uk = lim wk = u weakly in L ∞ (0, T ; W ),

k→0 k→0
lim z k = lim ζk = lim λk = z weakly in L ∞ (0, T ; L 2 (Ω)),
k→0 k→0 k→0
lim pk = p weakly in L 2 (0, T ; L 2m (Ω)),
k→0
∂ ∂
lim uk = u weakly in L 2 (0, T ; W ).
k→0 ∂t ∂t
Furthermore,
lim uk = lim wk = u strongly in L 2 (0, T ; L 4 (Ω)2 ). (5.7.63)

k→0 k→0
Proof By virtue of the uniform estimates of Theorem 5.7.19, we can extract a sub-
sequence (still denoted by k) such that
lim uk = u weakly in L ∞ (0, T ; W ),

k→0
lim z k = z weakly in L ∞ (0, T ; L 2 (Ω)),
k→0
and
lim pk = p weakly in L 2 (0, T ; L 2m (Ω)),

k→0
∂ ∂
lim uk = u weakly in L 2 (0, T ; W ),
k→0 ∂t ∂t
lim wk = w weakly in L ∞ (0, T ; W ),
k→0
lim ζk = ζ weakly in L ∞ (0, T ; L 2 (Ω)),
k→0
lim λk = λ weakly in L ∞ (0, T ; L 2 (Ω)).
k→0
As far as the functions w, ζ and λ are concerned, observe that
tn − t n
∀t ∈]t n−1 , t n ], wk (t) − uk (t) = (u − un−1 ) , 1 ≤ n ≤ N ,
k
tn − t n
∀t ∈]t n−1 , t n ], ζk (t) − z k (t) = (z − z n−1 ) , 1 ≤ n ≤ N ,
k
t − t n−1 n
∀t ∈ [t n−1 , t n [, λk (t) − z k (t) = − (z − z n−1 ) , 1 ≤ n ≤ N .
k
Therefore
k n
N
wk − uk 2L 2 (0,T ;W ) = u − un−1 2H 1 (Ω) , (5.7.64)
3 n=1 0
ζk − z k 2L 2 (Ω×]0,T [) = λk − z k 2L 2 (Ω×]0,T [)

k n
N
= z − z n−1 2L 2 (Ω) . (5.7.65)
3 n=1
Then, owing to (5.7.57) and uniqueness of the limit, we have w = u and ζ = λ = z.

It remains to prove the strong convergence (5.7.63); in view of (5.7.64), it suffices
to prove the strong convergence of uk . But in particular, {uk } is bounded uniformly
in the space
∂v
v ∈ L 2 (0, T ; W ); ∈ L 2 0, T ; L 4 (Ω)2 ,
∂t
and since the imbedding of H 1 (Ω) into L 4 (Ω) is compact, Theorem 3.3.14 implies
that uk converges strongly to u in L 2 (0, T ; L 4 (Ω)2 ).
Theorem 5.7.22 Let Ω ⊂ IR 2 be bounded and Lipschitz-continuous. Then, for any

α > 0, μ > 0, f in L 2 (0, T ; H (curl, Ω)), g in H 1 (0, T ; W 1− r ,r (∂Ω)2 ) for some
1
r > 2, satisfying g · n = 0, and ut 0 ∈ W with curl(ut 0 − α Δ ut 0 ) ∈ L 2 (Ω),

problem (5.7.5), (5.2.2), (5.7.6), (5.4.1), (5.2.4), and (5.7.4) has at least one solution
u ∈ H 1 (0, T ; W ), p ∈ L 2 (0, T ; L 2m (Ω)), z ∈ L ∞ (0, T ; L 2 (Ω)). Moreover, all
solutions satisfy the bounds of (5.7.26).
Proof Let k be a subsequence satisfying the convergences of Lemma 5.7.21. It is

easy to check that the functions uk , pk , z k , wk , ζk , and λk satisfy the following
formulations:
∀v ∈ H01 (Ω)2 , ∀ϕ ∈ C0 ([0, T ]),

T
∂ ∂
( uk (t), v) + α( ∇ uk (t), ∇ v) + μ(∇ wk (t), ∇ v)
0 ∂t ∂t
T

+ (λk (t) × wk (t), v) − ( pk (t), div v) ϕ(t)dt = ( f k (t), v)ϕ(t)dt, (5.7.66)
0
and
∀θ ∈ W 1,4 (Ω), ∀ψ ∈ C1 ([0, T ]) with ψ(T ) = 0,

T T

− (z k (t), θ )ψ (t)dt + μ(ζk (t), θ ) − α(wk (t) · ∇ θ, ζk (t)) ψ(t)dt
0 0
T

− (z 0 , θ )ψ(0) = μ(curl wk (t), θ ) + α(curl f k (t), θ ) ψ(t)dt, (5.7.67)
0
(uk )|∂Ω = g k a.e. in ∂Ω×]0, T [. (5.7.68)
On one hand, the above weak convergences imply the convergences of all the lin-
ear terms in (5.7.66) and (5.7.67). On the other hand, it stems from (5.7.59) and
(5.7.60) that the terms involving f also converge. Therefore, it suffices to check the
convergence of the nonlinear terms. Now, for each indices i and j, 1 ≤ i, j ≤ 2,
lim (wk )i v j ϕ = u i v j ϕ strongly in L 2 (Ω×]0, T [).

k→0
Then since
lim λk = z weakly in L 2 (Ω×]0, T [),
k→0
we have
T T
lim (λk (t) × wk (t), v)ϕ(t)dt = (z(t) × u(t), v)ϕ(t)dt.
k→0 0 0
Similarly,
lim (wk · ∇ θ )ψ = (u · ∇ θ )ψ strongly in L 2 (Ω×]0, T [).

k→0
Therefore
T T
lim (wk · ∇ θ, ζk )ψ(t)dt = (u(t) · ∇ θ, z(t))ψ(t)dt.
k→0 0 0
Hence we can pass to the limit in (5.7.66) and (5.7.67) and we obtain
∀v ∈ H01 (Ω)2 , ∀ϕ ∈ C0 ([0, T ]),

T
∂ ∂
u(t), v + α ∇ u(t), ∇ v + μ(∇ u(t), ∇ v)
0 ∂t ∂t
T

+ (z(t) × u(t), v) − ( p(t), div v) ϕ(t)dt = f (t), v ϕ(t)dt, (5.7.69)
0
respectively,
∀θ ∈ W 1,4 (Ω), ∀ψ ∈ C1 ([0, T ]) with ψ(T ) = 0,

T T

− z(t), θ ψ (t)dt + μ(z(t), θ ) − α(u(t) · ∇ θ, z(t)) ψ(t)dt
0 0
T

− (z 0 , θ )ψ(0) = μ(curl u(t), θ ) + α(curl f (t), θ ) ψ(t)dt. (5.7.70)
0
By choosing v ∈ D(Ω)2 , ϕ and ψ ∈ D(]0, T [) and θ ∈ D(Ω), we easily recover

(5.7.2) and (5.7.3). Moreover Lemma 5.7.20 gives the boundary condition (5.4.1).
It remains to recover the initial data. First, observe that for any v ∈ L 2 (Ω)2 and any
ϕ ∈ H 1 (0, T ) satisfying ϕ(T ) = 0, we have

T ∂ T
uk (t), v ϕ(t) dt = − (uk (t), v)ϕ (t) dt − (u0 , v)ϕ(0).
0 ∂t 0
Passing to the limit in this equality yields

T ∂ T
u(t), v ϕ(t) dt = − (u(t), v)ϕ (t) dt − (u0 , v)ϕ(0),
0 ∂t 0
whence u0 = u(0). As far as the initial value of z is concerned, we recall that,

owing to (5.7.3) and (5.7.4), z belongs to H 1 (0, T ; W −1, 3 (Ω)) (see Remark 5.7.2).
4
Therefore for θ ∈ D(Ω) and ψ ∈ D([0, T ]) satisfying ψ(T ) = 0,

T T
∂
− (z(t), θ )ψ (t)dt = (z(t), θ )ψ(t)dt + (z(0), θ )ψ(0)
0 0 ∂t
T ∂
= z(t), θ ψ(t)dt + (z(0), θ )ψ(0),
0 ∂t
and comparing with (5.7.70), we obtain z(0) = z 0 . This gives global existence in
time.
Finally, Theorem 5.7.10 implies that all solutions are bounded by the data as in
(5.7.26).
5.7.3 Uniqueness
Considering that the split and original formulations are equivalent, we establish
uniqueness of the solution of the latter, i.e., of (5.2.1), (5.2.2), (5.4.1), (5.2.4), and
proceed as in Section 5.6.5. In particular, we use the canonical lifting function u g
defined in (5.4.4). We recall the regularity of the solutions: (u, p) ∈ W where

W = L ∞ (0, T ; W α ) ∩ H 1 (0, T ; W ) × L 2 (0, T ; L 2m (Ω)). (5.7.71)
As in Section 5.6.5, a proof of uniqueness requires more regularity. This is the object
of the following propositions.
4
Proposition 5.7.23 If Ω is a bounded polygon, ut 0 belongs to W 2, 3 (Ω)2 , and g
4
is the trace of a function that belongs to H 1 (0, T ; W 2, 3 (Ω)2 ), then any solution
(u, p) ∈ W of (5.2.1), (5.2.2), (5.4.1), (5.2.4) satisfies
u ∈ L ∞ (0, T ; W 2, 3 (Ω)2 ) , p ∈ L 2 (0, T ; W 1, 3 (Ω)).

4 4
(5.7.72)
Proof By equivalence, (u, p) also satisfies (5.7.2) with z = curl(u − α Δ u). Thus,
we set
∂u ∂u ∂g
w=α +μu , = f − −z×u, k=α + μ g,
∂t ∂t ∂t
and the pair (w, p) is the solution of the following steady Stokes system, almost
everywhere in ]0, T [:
− Δ w + ∇ p = , div w = 0 in Ω , w = k on ∂Ω. (5.7.73)

4
But belongs at least to L 2 (0, T ; L 3 (Ω)2 ), as f ∈ L 2 (Ω×]0, T ])2 and (u, p) ∈ W .
4
And by assumption, k is the trace of a function in L 2 (0, T ; W 2, 3 (Ω)2 ). Then part
4
(i) of Theorem 3.2.17 implies that w belongs to L 2 (0, T ; W 2, 3 (Ω)2 ) and p belongs
4
to L 2 (0, T ; W 1, 3 (Ω)). In turn,
∂ 4
α Δ u + μ Δ u = Δ w ∈ L 2 (0, T ; L 3 (Ω)2 ).
∂t
Therefore
∂ μt 1 μ 4
e α Δ u = e α t Δ w ∈ L 2 (0, T ; L 3 (Ω)2 ),
∂t α
and by integration over ]0, t[, the regularity assumption on the initial function ut 0
implies that Δ u belongs to L ∞ (0, T ; L 3 (Ω)2 ). Then (5.7.72) follows from Theorem
4
3.2.2 and the assumption on the regularity of g.
Proposition 5.7.24 Let Ω be a convex polygon and 2 < r ≤ r0 , where r0 is the

parameter of Theorem 3.2.14. If ut 0 belong to W 2,r (Ω)2 , and g is the trace of a
function that belongs to H 1 (0, T ; W 2,r (Ω)2 ) then the velocity part of any solution
(u, p) ∈ W of (5.2.1), (5.2.2), (5.4.1), (5.2.4) satisfies
u ∈ L ∞ (0, T ; W 2,r (Ω)2 ). (5.7.74)
Proof The proof is a simpler version of that of Lemma 5.2.5. By assumption, curl(u−
α Δ u) belongs to L ∞ (0, T ; L 2 (Ω)). It is easy to prove that any function in L 2 (Ω)
can be written under the form curl ϕ for some function ϕ ∈ H 1 (Ω)2 . Thus
curl(u − α Δ u) = curl ϕ , with ϕ ∈ L ∞ (0, T ; H 1 (Ω)2 ),
i.e., curl(u − α Δ u − ϕ) = 0 a.e. in Ω×]0, T [. When Ω is simply-connected, this

implies that there exists a function q in L 2 (Ω) such that
u − α Δ u − ϕ = ∇ q.
As div u = 0 and u = g on ∂Ω, (u, −q) is the solution of a Stokes problem with
interior data ϕ − u in L ∞ (0, T ; H 1 (Ω)2 ) and boundary data g that is the trace of a
function in H 1 (0, T ; W 2,r (Ω)2 ). Then (5.7.74) follows from Theorem 3.2.14.
Propositions 5.7.23 and 5.6.17 readily imply the next result. We omit the proof
which is straightforward. In view of the Sobolev imbeddings (3.1.3) and (3.1.6), we
see that all quantities in (5.7.75) below are well-defined.
Proposition 5.7.25 Under the assumptions of Proposition 5.7.23, any two solutions
(u1 , p 1 ) and (u2 , p 2 ) in W of (5.2.1), (5.2.2), (5.4.1), (5.2.4) satisfy
1 d
w2α + μw2H 1 (Ω) + c(w; u1 , w) + α c(w; curl u1 , curl w)
2dt
0
−2α curl w(∇ u 11 · ∇ w2 − ∇ u 12 · ∇ w1 ) d x = 0, (5.7.75)

Ω
where w = u1 − u2 .
As mentioned above, all terms in (5.7.75) make sense, but unfortunately, without
additional regularity, (5.7.75) does not seem to imply that w = 0. Indeed, the last
two terms in (5.7.75) have no particular sign and in order to be controlled by the
first two terms, they must be bounded in terms of the H 1 norm of w. This is the
case if we assume that u1 belongs to W 2,r (Ω)2 for some r > 2. Since by Sobolev’s
imbedding, w belongs to L p (Ω)2 for any p > 2, then we can choose p so that the
product (curl w)w belongs to L r (Ω)2 , the dual exponent of r , i.e., p = 2r/(r − 2).
Hence, with the notation of (3.1.6),
|c(w; curl u1 , curl w)| ≤ S0, r2r−2 w2H 1 (Ω) Δ u1 L r (Ω) , (5.7.76)
0

curl w(∇ u 11 · ∇ w2 − ∇ u 12 · ∇ w1 ) d x ≤ w2H 1 (Ω) |u1 |W 1,∞ (Ω) . (5.7.77)
0
Ω
With these remarks, we infer the following theorem from Propositions 5.7.24 and
5.7.25:
Theorem 5.7.26 Assume that Ω is a convex polygon and let 2 < r ≤ r0 , where
r0 is the parameter of Theorem 3.2.14. Then for any α > 0, μ > 0, f in
L 2 (0, T ; H (curl, Ω)), ut 0 ∈ W α , and g in H 1 (0, T ; H 2 (∂Ω)2 ) satisfying g · n = 0
1
and such that it is the trace of a function u g in H 1 (0, T ; W 2,r (Ω)2 ), problem (5.2.1),
(5.2.2), (5.4.1), (5.2.4) has exactly one solution (u, p) ∈ W .
Proof By substituting (5.7.76) and (5.7.77) into (5.7.75) and applying (3.1.5), we
obtain
1 d
w2 + μ w2H 1 (Ω)
2 d t α 0

√
≤ S0,4 u H01 (Ω) + 2α S0, r2r−2 |u1 |W 2,r (Ω) + 2α|u1 |W 1,∞ (Ω) w2H 1 (Ω) .
2 1
0
Since w(0) = 0, Gronwall’s Lemma 3.3.13 implies that w = 0.
The above convexity condition is restrictive, but we do not know if the solution
is unique when the domain has a reentrant corner.
5.8 Splitting: The Steady System in 3- D
In this section, we split the steady three-dimensional grade-two problem, with no-slip
boundary condition (5.6.1), (5.2.2),
−μ Δ u + curl(u − α Δ u) × u + ∇ p = f in Ω, (5.8.1)
div u = 0 in Ω, (5.8.2)
u = 0 on ∂Ω, (5.8.3)
into a Stokes-like system and a transport system. As in the two-dimensional case, the
split problem is well adapted to numerical approximation, but in contrast, the theory
developed here is not optimal, in the sense that the auxiliary variable z is measured
in L r with r > 3 instead of L 2 . This theory can be readily extended to a Dirichlet
tangential boundary condition, but the discussion is more technical. The reader will
find in the reference [32] by Bernard a somewhat different splitting, for the grade-two
model with a tangential boundary condition, that leads to optimal estimates, but does
not lend itself easily to discretization, see Section 5.8.4.
As in the two-dimensional case, we substitute (5.2.7)
z = curl(u − α Δ u),
into (5.8.1) and we consider the steady version of (5.2.10)
μ z + α([∇ z]u − [∇ u]z) = μ curl u + α curl f .
This gives the system of equations

−μ Δ u + z × u + ∇ p = f in Ω,
div u = 0 in Ω,
(5.8.4)
u = 0 on ∂Ω,
μ z + α([∇ z]u − [∇ u]z) = μ curl u + α curl f in Ω.
The analysis of (5.8.4) is much more difficult than that of (5.6.8) in two dimensions.
The first difficulty concerns the choice of spaces for the velocity u. Observing that
the last equation of (5.8.4) has the same structure as (7.2.37) in the Appendix, we
see that for a given u, the stability of this equation requires u sufficiently small in
W 1,∞ (Ω)3 , more precisely (compare with (7.2.35)),
1
|u|W 1,∞ (Ω) ≤ (μ − δ), (5.8.5)
α
for some 0 < δ < μ. In view of Theorem 3.2.15, this regularity can only be attained
if z belongs to L r (Ω)3 for some r > 3 related to the largest inner angle of ∂Ω.
Therefore, the relevant spaces for (u, p, z) is W 1,∞ (Ω)3 × L ∞ (Ω) × L r (Ω)3 .
The second difficulty lies in proving equivalence between (5.8.4) and problem
(5.6.1), (5.2.2). Indeed, in order to establish equivalence, we must show that any
solution z of (5.8.4) satisfies (5.2.7). This is easy provided we show that div z = 0.
This last property, which is by no means straightforward, is the object of the next
section.
5.8.1 Equivalence
Let us consider here the more general problem with f given in H (curl, Ω) and v
given in W ∩ W 1,∞ (Ω)3 (see (3.1.18)), satisfying (5.8.5): Find z ∈ L 2 (Ω)3 solution
of
μ z + α([∇ z]v − [∇ v]z) = μ curl v + α curl f . (5.8.6)
This is a particular case of (7.2.24) with C = μ I −α ∇ v. Observe that by assumption

C belongs to L ∞ (Ω)d×d and is uniformly positive definite: it satisfies (7.2.25) with
c0 = δ since, in view of (5.8.5),

(C z, z) ≥ μ − α ∇ v L ∞ (Ω) z2 ≥ δ z2 .
Therefore, according to Theorem 7.2.16 with p = 2, this problem has a unique

solution. As div v = 0, (5.2.9) implies
curl(z × v) = [∇ z]v − [∇ v]z − v div z and div(v div z) = v · ∇(div z), (5.8.7)
and so (5.8.6) reads

μ z + α(curl(z × v) + v div z) = μ curl v + α curl f , (5.8.8)
and taking the divergence of this equation leads to the following equation for div z:
μ div z + αv · ∇(div z) = 0.
This is a scalar transport equation analogous to (7.2.1), but the theory developed
in Section 7.2.1 does not directly apply here because the solution div z is only in
H −1 (Ω). The following lemma, inspired by Bernard [32], proceeds by regularizing
(5.8.6).
Lemma 5.8.1 Let Ω be a convex polyhedron. For F given in H (div, Ω) with
div F = 0 and v given in W ∩ W 1,∞ (Ω)3 , satisfying (5.8.5) for some 0 < δ < μ,
the unique solution z ∈ L 2 (Ω)3 of
μ z + α([∇ z]v − [∇ v]z) = F (5.8.9)
satisfies div z = 0.
Proof Following [32], we introduce the closed subspace of H 1 (Ω)3
X = {v ∈ H 1 (Ω)3 ; div v = 0, v × n = 0 on ∂Ω}.
Since Ω is convex, the mapping v → curl v L 2 (Ω) is a norm on X , equivalent to

the H 1 norm and there exists constants C1 and C2 such that (see for example [116,
6])
∀w ∈ X, w H 1 (Ω) ≤ C1 ∇ w L 2 (Ω) ≤ C2 curl w L 2 (Ω) . (5.8.10)
Then we consider the following regularization of (5.8.9): Find z ε ∈ X solution of

∀ϕ ∈ X, ε(curl z ε , curl ϕ) + μ(z ε , ϕ) + α [∇ z ε ]v − [∇ v]z ε , ϕ = (F, ϕ).
(5.8.11)
Note that when ϕ belongs to V (see (3.1.15)), the first term in (5.8.11) satisfies
(curl z ε , curl ϕ) = −
Δ z ε , ϕ,
but due to the fact that only the tangential components of z ε vanish on ∂Ω, (5.8.11)
does not have the structure of a Stokes-like problem. The reason for regularizing with
a (curl·, curl·) term instead of the more natural choice of a (∇·, ∇·) term is that with
the first choice, (5.8.11) is unchanged when the set of test functions ϕ is extended to
H 1 (Ω)3 with zero tangential components, thus implying that the gradient of pressure
associated with the zero divergence constraint is in fact zero. The proof of this result
is recalled in Proposition 5.8.2.
The system (5.8.11) is a linear variational problem with bilinear form

a(z, ϕ) = ε(curl z, curl ϕ) + μ(z, ϕ) + α [∇ z]v − [∇ v]z, ϕ ,
that is continuous and elliptic on X × X ,
|a(z, ϕ)| ≤ εcurl z L 2 (Ω) curl ϕ L 2 (Ω) + μz L 2 (Ω) ϕ L 2 (Ω)

+ α v L ∞ (Ω) ∇ z L 2 (Ω) + ∇ v L ∞ (Ω) z L 2 (Ω) ϕ L 2 (Ω)
≤ C(v)curl z L 2 (Ω) curl ϕ L 2 (Ω) ,
and

a(z, z) ≥ εcurl z2L 2 (Ω) + z2L 2 (Ω) μ − α∇ v L ∞ (Ω)
≥ εcurl z2L 2 (Ω) + δ z2L 2 (Ω) ,
where C(v) depends on ε, vW 1,∞ (Ω) , μ, α, and on the constants from (5.8.10). For
the ellipticity, we have applied (5.8.5) and the standard Green formula (5.2.43) to
(v · ∇ z i , z i ) for each i, that is valid because v belongs to W . As the right-hand side is
a continuous linear functional on X , Lax–Milgram’s Theorem implies that (5.8.11)
has a unique solution z ε ∈ X . Moreover, z ε satisfies the following bounds:
√ 1 1
ε curl z ε L 2 (Ω) ≤ √ F L 2 (Ω) , z ε L 2 (Ω) ≤ F L 2 (Ω) . (5.8.12)
2 δ δ
Since X is reflexive, we infer from the second inequality in (5.8.12) that there exists
a subsequence of ε and a function z in L 2 (Ω)3 such that
lim z ε = z weakly in L 2 (Ω)3 .

ε→0
Furthermore, we observe that div z = 0; indeed,
∀θ ∈ H01 (Ω), 0 = (div z ε , θ ) = −(z ε , ∇ θ ) → −(z, ∇ θ ) =

div z, θ .
Now, we pass to the limit in (5.8.11) with an arbitrary test function ϕ in V. By

applying Green’s formula to the first factor of α, we write it as

∀ϕ ∈ V, ε(curl z ε , curl ϕ) + μ(z ε , ϕ) − α (z ε , [∇ ϕ]v) + ([∇ v]z ε , ϕ) = (F, ϕ).
Owing to the first estimate in (5.8.12), the first term in the above left-hand side tends
to zero, and the remaining terms tend to

μ(z, ϕ) − α (z, [∇ ϕ]v) + ([∇ v]z, ϕ) .
Hence z satisfies

∀ϕ ∈ V, μ(z, ϕ) + α [∇ z]v − [∇ v]z, ϕ = (F, ϕ).
As v belongs to W 1,∞ (Ω)3 , the term [∇ z]v belongs to H −1 (Ω)3 and all other terms
involving z belong to L 2 (Ω)3 , and since V is dense in V , the above equation holds
for all ϕ ∈ V . Then the result of Proposition 5.8.2 below implies that

∀ϕ ∈ H01 (Ω)3 , μ(z, ϕ) + α [∇ z]v − [∇ v]z, ϕ = (F, ϕ).
Thus z satisfies on one hand div z = 0 and on the other hand (5.8.9). As (5.8.9) has
only one solution, this proves the lemma.
Proposition 5.8.2 We keep the assumptions and notation of Lemma 5.8.1. Then
problem (5.8.11) is equivalent to: Find z ε ∈ X solution of

ε(curl z ε , curl ϕ) + μ(z ε , ϕ) + α [∇ z ε ]v − [∇ v]z ε , ϕ = (F, ϕ), (5.8.13)
for all ϕ in H 1 (Ω)3 satisfying ϕ × n = 0 on ∂Ω.
Proof The ideas of the proof can be found for instance in [116, 6], but we recall
them for the reader’s convenience. Let ψ be an arbitrary function in H 1 (Ω)3 with
ψ × n = 0 on ∂Ω and let λ ∈ H01 (Ω) solve Δ λ = div ψ in Ω. Then λ belongs to
H 2 (Ω) because Ω is convex, and the function ϕ = ψ −∇ λ belongs to X . Therefore,
if z ε ∈ X solves (5.8.11), it also solves
ε(curl z ε , curl(ψ − ∇ λ)) + μ(z ε , ψ − ∇ λ)

+ α [∇ z ε ]v − [∇ v]z ε , ψ − ∇ λ = (F, ψ − ∇ λ).
But, owing that div z ε = div v = 0, we have
[∇ z ε ]v − [∇ v]z ε = curl(z ε × v).
Therefore, considering that div F = 0, all terms above involving ∇ λ vanish and z ε
solves (5.8.13). The converse is obvious.
The convexity assumption in the statement of Lemma 5.8.1 is not satisfactory. In

the case when v vanishes on ∂Ω, it can be relaxed.
Proposition 5.8.3 Let v be given in V ∩ W 1,∞ (Ω)3 satisfying (5.8.5) for some
0 < δ < μ, and assume that F ∈ L 2 (Ω)3 has the form F = curl G for some
function G ∈ H −1 (Ω)3 . Then the statement of Lemma 5.8.1 holds on a bounded,
connected, Lipschitz domain.
Proof Let B be a ball such that Ω B. As F = curl G for G ∈ H −1 (Ω)3 , it has an

extension, say F̃, in H (div, B) such that div F = 0 in B. Indeed, Propositions 5.2.4
and 5.2.3 guarantee that F is the curl of a function of H 1 (Ω)3 and this function has
an extension to B. Next, let ṽ be the extension of v by zero outside Ω; then ṽ belongs
to W 1,∞ (B)3 , div v = 0 in B, and clearly v vanishes on ∂Ω. Furthermore,
1
|ṽ|W 1,∞ (B) = |v|W 1,∞ (Ω) ≤ (μ − δ).
α
Now, we consider the transport system in B,

μ Z + α [∇ Z]ṽ − [∇ ṽ]Z = F̃.
As B is convex, all assumptions of Lemma 5.8.1 are satisfied and hence div Z = 0.
But of course, the restriction of Z to Ω satisfies

μ Z + α [∇ Z]v − [∇ v]Z = F,
and, since z is the unique solution of this problem, we conclude that div z = 0.
In the case of (5.8.8), F = μ curl v + α curl f satisfies the hypothesis of Propo-

sition 5.8.3, and the next theorem wraps up the results of this section.
Theorem 5.8.4 In a bounded connected Lipschitz domain Ω, problems (5.8.4) and

(5.8.1)–(5.8.3) are equivalent in the sense that any solution (u, p, z) ∈ W 1,∞ (Ω)3 ×
L 2m (Ω) × L 2 (Ω)3 of (5.8.4) with u satisfying (5.8.5) for some 0 < δ < μ, solves
(5.8.1)–(5.8.3), and z satisfies (5.2.7), i.e.,
z = curl(u − α Δ u).
Conversely, any solution (u, p) ∈ V α × L 2m (Ω) of (5.8.1)–(5.8.3), with z defined by

(5.2.7), solves (5.8.4).
Proof Let (u, p, z) ∈ W 1,∞ (Ω)3 × L 2m (Ω)× L 2 (Ω)3 be any solution of (5.8.4) with
u satisfying (5.8.5) for some δ > 0. Then Proposition 5.8.3 shows that div z = 0.
Hence taking the curl of the first line of (5.8.4) and comparing with the third line,
we derive (5.2.7). The converse is immediate.
5.8.2 Two Auxiliary Systems
In this paragraph, we study separately the two systems in (5.8.4); the results will be
used in the next section for constructing a solution.
From now on, we assume that Ω is a convex polyhedron. Let σ ∈]0, 1[ be the
parameter of Theorem 3.2.15, related to the largest inner angle of ∂Ω, let r > 3 be
defined by
3
σ =1− , (5.8.14)
r
and let f belong to L r (Ω)3 with curl f in L r (Ω)3 . There is no loss of generality in
assuming that 3 < r ≤ 4.
5.8.2.1 Transport
First, we consider the transport part of (5.8.4). Let v be a given function in the set
1
B = v ∈ V ∩ W 1,∞ (Ω)3 ; ∇ v L ∞ (Ω) ≤ (μ − δ),
α
S0,2
∇ v L 2 (Ω) ≤ f L 2 (Ω) , (5.8.15)
μ
where 0 < δ < μ is some fixed quantity. Note that B is a closed, convex subset of
V ∩ W 1,∞ (Ω)3 . For the sake of brevity, we shall say that a sequence of functions
{vk }k≥1 of B converges weakly to a function v ∈ B if
lim ∇ vk = ∇ v weakly in L 2 (Ω)3×3 and lim ∇ vk = ∇ v weakly in L ∞ (Ω)3×3 .

k→∞ k→∞
For each v in B, consider the problem: Find z = z(v) ∈ L r (Ω)3 solution of
μ z + α([∇ z]v − [∇ v]z) = μ curl v + α curl f . (5.8.16)
We have seen at the beginning of Section 5.8.1 that (5.8.16) is a particular case of
(7.2.24) with
F = μ curl v + α curl f ∈ L r (Ω)3 ,
and
C = μI − α ∇ v.
Therefore, Theorem 7.2.16 with p = r implies that problem (5.8.16) has a unique
solution z = z(v) ∈ L r (Ω)3 and this solution satisfies the upper bound
curl(μ v + α f ) L r (Ω) 1
z(v) L r (Ω) ≤ ≤ curl(μ v + α f ) L r (Ω) . (5.8.17)
μ − α∇ v L ∞ (Ω) δ
Furthermore, since div F = 0, it follows from Lemma 5.8.1 that div z(v) = 0.
Lemma 5.8.5 The mapping v → z(v) is weakly continuous in the sense that if
{vk }k≥1 is a sequence of functions of B that converges weakly to a function v ∈ B,
then there exists a subsequence of k, still denoted k, such that
lim z(vk ) = z(v) weakly in L r (Ω)3 . (5.8.18)

k→∞
Proof If {vk }k≥1 is a sequence in B, then by definition it is bounded in W 1,∞ (Ω)3 ,

and owing to (5.8.17), z(vk ) is also bounded,
1 √ 1
z(vk ) L r (Ω) ≤μ 2|Ω| r ∇ vk L ∞ (Ω) + α curl f L r (Ω)
δ
1 √ 1 1
≤ μ 2|Ω| r (μ − δ) + α curl f L r (Ω) .
δ α
Hence there exists a function ζ in L r (Ω)3 and a subsequence of k, still denoted by

k, such that
lim z(vk ) = ζ weakly in L r (Ω)3 .
k→∞
Moreover, div ζ = 0 because, as noted above, div z(vk ) = 0. It remains to pass to the
limit in (5.8.16) written in terms of vk and set in variational form, with r1 + r1 = 1,

∀θ ∈ L r (Ω)3 , μ(z(vk ), θ )+α([∇ z(vk )]vk −[∇ vk ]z(vk ), θ) = curl(μ vk +α f ), θ .
In particular, this holds for all θ ∈ D(Ω)3 . Owing to the weak convergence of vk
to v in B, passing to the limit is straightforward except in the factor of α. For this,
we use (5.8.7) and the fact that z(vk ) and vk have zero divergence; thus for any θ in
D(Ω)3 ,

([∇ z(vk )]vk − [∇ vk ]z(vk ), θ ) = curl(z(vk ) × vk ), θ = (z(vk ) × vk , curl θ ).
Then the weak convergence of z(vk ) in L r (Ω)3 and the strong convergence of vk in

L r (Ω)3 (in fact in L q (Ω)3 for any finite q) imply that for any θ in D(Ω)3 ,
lim k→∞ ([∇ z(vk )]vk − [∇ vk ]z(vk ), θ ) = (ζ × v, curl θ ) = ([∇ ζ ]v − [∇ v]ζ , θ ),
where in the last equality we have used div ζ = 0. This implies that ζ satisfies in the
sense of distributions
μ ζ + α([∇ ζ ]v − [∇ v]ζ ) = curl(μ v + α f ).
As this problem has a unique solution z(v) ∈ L r (Ω)3 , and since ζ is in L r (Ω)3 , the
conclusion is ζ = z(v).
5.8.2.2 Generalized Stokes System
Now, for given f and ζ , both in L r (Ω)3 , we examine the Stokes part of (5.8.4): Find
a pair u = u(ζ ) ∈ V and p = p(ζ ) ∈ L 2m (Ω) solution of
− μ Δ u + ζ × u + ∇ p = f in Ω. (5.8.19)
As in the two-dimensional case, it is easy to prove that (5.8.19) has one and only one
solution without restriction on the domain and the data; in particular, this is valid

for f and ζ in L 2 (Ω)3 . Therefore, (5.8.19) defines a mapping ζ → u(ζ ), p(ζ )
from L 2 (Ω)

3
into V × L 2m (Ω). Furthermore, when Ω is a convex polyhedron, the
solution u(ζ ), p(ζ ) has the following additional regularity:
Proposition 5.8.6 Let Ω be a convex polyhedron, and let f and ζ be both in L r (Ω)3 ,
with r >3 defined at the beginning of this subsection. Then the solution of (5.8.19)
satisfies u(ζ ), p(ζ ) ∈ C1,σ (Ω)3 × C0,σ (Ω).
Proof Assume first that both f and ζ are in L 3 (Ω)3 . Since we already know that u
belongs to V , hence to L 6 (Ω)3 , the product ζ × u is in L 2 (Ω)3 . Therefore, regarding
(5.8.19) as a Stokes system with right-hand side f −ζ ×u, the convexity of Ω implies
that u belongs to H 2 (Ω)3 , whence to L ∞ (Ω)3 . Then, in view of the full L r (Ω)3
regularity, r > 3, of f and ζ , we have f − ζ × u in L r (Ω)3 , and the statement of
the proposition follows from Theorem 3.2.15.

Thus the mapping ζ → u(ζ ), p(ζ ) is compact in the following sense:

Lemma 5.8.7 The mapping ζ → u(ζ ), p(ζ ) defined by (5.8.19) is compact in
the sense that if (ζ k )k≥1 is a sequence of functions of L r (Ω)3 that converges weakly
to a function ζ ∈ L r (Ω)3 , then there exists a subsequence of k, still denoted k, such
that

lim u(ζ k ), p(ζ k ) = u(ζ ), p(ζ ) strongly in C1 (Ω)3 × C0 (Ω). (5.8.20)
k→∞
In particular, this implies strong convergence in W 1,∞ (Ω)3 × L ∞ (Ω).

Proof Without assumptions on ζ k , u(ζ k ) is uniformly bounded in H01 (Ω)3 ,
1
∇ u(ζ k ) L 2 (Ω) ≤ f H −1 (Ω) .
μ
Next, the weak convergence of ζ k in L r (Ω)3 implies that ζ k is uniformly bounded

in L r (Ω)3 . Then, arguing as in the proof of Proposition 5.8.6, we first infer from the
convexity of Ω that u(ζ k ) is uniformly bounded in L ∞ (Ω)3 . This implies that the
product ζ k × u(ζ k ) is uniformly bounded
in L r (Ω)3 , and hence, by virtue of The-
orem 3.2.15, the pair u(ζ k ), p(ζ k ) is uniformly bounded in C1,σ (Ω)3 × C0,σ (Ω).
As a consequence, u(ζ k ) is uniformly bounded and equicontinuous, as well as the
sequence of its gradients. Therefore by Ascoli–Arzela’s Lemma (cf. for example
[105]) and the completeness of C1 , there exists a function u ∈ C1 (Ω)3 ∩ V , and a
subsequence, still denoted by k, such that
lim u(ζ k ) − uC1 (Ω) = 0.

k→∞
Similarly, there exists a function p in C0 (Ω)∩ L 2m (Ω), such that, up to a subsequence,
lim p(ζ k ) − pC0 (Ω) = 0.

k→∞

Clearly this also implies strong convergence of u(ζ k ), p(ζ k ) in W 1,∞ (Ω)3 ×
L ∞ (Ω).
It remains to pass to the limit in (5.8.19) written in terms of ζ k . To this end, we
write (5.8.19) in variational form

∀v ∈ H01 (Ω)3 , μ ∇ u(ζ k ), ∇ v + ζ k × u(ζ k ), v − ( p(ζ k ), div v = ( f , v).
Passing to the limit gives immediately

∀v ∈ H01 (Ω)3 , μ ∇ u, ∇ v + ζ × u, v − ( p, div v) = ( f , v), (5.8.21)
and we recover (5.8.19) in the sense of distributions. Therefore, the pair (u, p) solves
(5.8.19) and the uniqueness of its solution yields u = u(ζ ) and p = p(ζ ).
5.8.3 A Fixed-Point Argument
Consider the following mapping. For v given in B, we define first z(v) in L r (Ω)3
solution of (5.8.16) and next u z(v) in C (Ω)3 ∩ V , solution of
1,σ

∀v ∈ V, μ ∇ u, ∇ v + z × u, v = ( f , v), (5.8.22)

i.e., u z(v) is the velocity part of the solution of (5.8.21). Clearly, the uniqueness
of the solutions of (5.8.16) and of (5.8.22) imply that the relation

Z : v −→ u z(v) , (5.8.23)
defines a mapping in B. Furthermore, it stems from Lemmas 5.8.5 and 5.8.7 that Z
is a compact mapping.
Proposition 5.8.8 The mapping (5.8.23) is compact.
Proof Indeed, let (vk )k≥1 be a sequence of functions of B that is bounded in B.
Therefore, there exists a subsequence of k, still denoted k, and a function v ∈ B with
lim ∇ vk = ∇ v weakly in L 2 (Ω)3×3 ,

k→∞
lim ∇ vk = ∇ v weakly in L ∞ (Ω)3×3 .
k→∞
Hence Lemma 5.8.5 implies that, up to subsequences,
lim z(vk ) = z weakly in L r (Ω)3 ,

k→∞
and z solves (5.8.16), i.e., z = z(v). In turn, Lemma 5.8.7 implies that, again up to
subsequences,

lim u z(vk ) = u strongly in C1 (Ω),
k→∞

lim p z(vk ) = p strongly in C0 (Ω),
k→∞

and (u, p) solves (5.8.21), i.e., u = u z(v) and p = p z(v) .These strong conver-
gences imply strong convergence in W 1,∞ (Ω)3 × L ∞ (Ω), whence compactness of
(5.8.23).
It remains to establish that B is stable under Z, i.e., Z(B) ⊂ B.
Lemma 5.8.9 If the data satisfy
Cσ C2 1 √ 1 1 α
f L r (Ω) 1 + S0,∞ S6 |Ω| 3 2 |Ω| r (μ − δ) + curl f L r (Ω)
μ δ α μ
1 r −3 √ 1 1 α
× 1 + S0,2 S0,6 |Ω| 3r 2 |Ω| r (μ − δ) + curl f L r (Ω)
δ α μ
1
≤ (μ − δ), (5.8.24)
α

then for all v in B, the function u = u z(v) defined by (5.8.16) and (5.8.22) belongs
to B.
Proof We proceed in three steps.
1. If v ∈ B, then according to (5.8.17), the definition (5.8.15) of B, and Hölder’s
inequality, z(v) satisfies (see the proof of Lemma 5.8.5),
1
z(v) L r (Ω) ≤ curl(μ v + α f ) L r (Ω)
δ
1 √ μ 1

≤ 2 |Ω| r (μ − δ) + α curl f L r (Ω) . (5.8.25)
δ α
Another application of Hölder’s inequality yields
1
r −3 √ μ 1

z(v) L 3 (Ω) ≤ |Ω| 3r 2 |Ω| r (μ − δ) + α curl f L r (Ω) . (5.8.26)
δ α

2. Now, we use a bootstrap argument to derive a bound for u z(v) in L ∞ . First,
in view of (5.8.22) and Poincaré’s inequality (3.1.5) in H 1 , we recover the second
condition in the definition of B,
S0,2
∇ u z(v) L 2 (Ω) ≤ f L 2 (Ω) .
μ
Therefore, by Sobolev’s imbedding (3.1.5)
1
u z(v) L 6 (Ω) ≤ S0,2 S0,6 f L 2 (Ω) . (5.8.27)
μ
Then a combination of (5.8.26) and (5.8.27) yields
1 r −3
z(v) × u z(v) L 2 (Ω) ≤ S0,2 S0,6 |Ω| 3r f L 2 (Ω)
μδ
√ μ 1

× 2 |Ω| r (μ − δ) + α curl f L r (Ω) . (5.8.28)
α
The convexity of Ω implies that
C2
u z(v) H 2 (Ω) ≤ f L 2 (Ω) + z(v) × u z(v) L 2 (Ω) ,
μ
where C2 is the continuity constant of the Stokes operator in H 2 . With (5.8.28), this
gives

u z(v) H 2 (Ω)
C2 1 r −3 √ 1 1 α
≤ f L 2 (Ω) 1 + S0,2 S0,6 |Ω| 3r 2 |Ω| r (μ − δ) + curl f L r (Ω) .
μ δ α μ
Therefore, considering that (see (3.1.6))

6−r
u z(v) L ∞ (Ω) ≤ S0,∞ |u z(v) |W 1,r (Ω) ≤ S0,∞ |Ω| 6r ∇ u L 6 (Ω)
6−r
≤ S0,∞ S6 |Ω| 6r u H 2 (Ω) ,
we obtain
6−r C 2
u z(v) L ∞ (Ω) ≤ S0,∞ S6 |Ω| 6r f L 2 (Ω)
μ
1 r −3 √ 1 1 α
× 1 + S0,2 S0,6 |Ω| 3r 2 |Ω| r (μ − δ) + curl f L r (Ω) . (5.8.29)
δ α μ
3. Finally,
(5.8.29) and (5.8.25) allow to sharpen (5.8.28): they imply that z(v) ×
u z(v) belongs to L r with

z(v) × u z(v) L r (Ω)
C2 6−r √ μ 1
≤ S0,∞ S6 |Ω| 6r f L 2 (Ω) 2 |Ω| r (μ − δ) + α curl f L r (Ω)
μδ α
1 r −3 √ 1 1 α
× 1 + S0,2 S0,6 |Ω| 3r 2 |Ω| r (μ − δ) + curl f L r (Ω) .
δ α μ
Here it is more convenient to measure f in L r ,


z(v) × u z(v) L r (Ω)
C2 1
√ 1 1 α
≤ S0,∞ S6 |Ω| 3 f L r (Ω) 2 |Ω| r (μ − δ) + curl f L r (Ω)
δ α μ
1 r −3 √ 1 1 α
δ α μ
Then the convexity of Ω and Theorem 3.2.15 yield

∇ u z(v) L ∞ (Ω) ≤ u z(v) C1,σ (Ω)
Cσ C2 1 √ 1 1 α
≤ f L r (Ω) 1 + S0,∞ S6 |Ω| 3 2 |Ω| r (μ − δ) + curl f L r (Ω)
μ δ α μ
1
r −3 √ 1 1 α
δ α μ
Thus, due to (5.8.24), we recover the first condition in the definition of B.
Remark 5.8.10 Condition (5.8.24) holds if for example f and its curl are sufficiently
small in L r .
From Schauder’s fixed point theorem, Lemma 5.8.9 and Proposition 5.8.8, we
immediately deduce the existence of a fixed point of the mapping v → u z(v) in B
whence the existence of a solution of the split problem (5.8.4).
Theorem 5.8.11 Let Ω be a convex polyhedron, let σ ∈]0, 1[ be the parameter of

Theorem 3.2.15, related to the largest inner angle of ∂Ω, let r > 3 be defined by
3
σ =1− ,
r
and let f belong to L r (Ω)3 with curl f in L r (Ω)3 . If (5.8.24) holds for some real
number 0 < δ < μ, then the split problem (5.8.4) has at least one solution in the
ball B defined by (5.8.15).
5.8.4 The Splitting of J.M. Bernard
The splitting presented here is based on the following variant of Proposition 5.2.3,
see [116, Chapter I, Section 3.5].
Lemma 5.8.12 Let Ω be a convex polyhedron, or a simply-connected domain with

a connected boundary of class C1,1 . For each divergence-free vector field w in
H (div, Ω), there exists a unique function ϕ in H 1 (Ω)3 such that
w = curl ϕ, div ϕ = 0 in Ω, ϕ · n = 0 on ∂Ω, (5.8.30)

and the mapping Φ : w → ϕ is continuous: There exists a constant C depending

only on Ω, such that for all divergence-free vector fields w in H (div, Ω)
ϕ H 1 (Ω) ≤ Cw L 2 (Ω) . (5.8.31)
We skip the proof; it is based on a correction of the vector potential constructed

in Proposition 5.2.3. The correction is designed to satisfy the boundary condition
and is obtained by adding a term derived from the solution of a Laplace equation
with a Neumann boundary condition. The convexity or regularity assumptions on
the domain guarantee that this correction belongs to H 1 (Ω)3 .
Let Z be the space of divergence-free vector fields,
Z = {v ∈ L 2 (Ω); div v = 0}.
In his work [32], Bernard proposes the following splitting for the steady grade-two
fluid model with a tangential Dirichlet boundary condition specified; to simplify the
discussion, we write it for a homogeneous boundary condition. Find (u, p, v, z) in
H01 (Ω)3 × L 2m (Ω) × H01 (Ω)3 × L 2 (Ω)3 such that
−μ Δ u + z × u + ∇ p = f in Ω,
div u = 0,
μ
−μ Δ v + ∇ π = Φ(PZ (z)) − u in Ω, (5.8.32)
α
div v = 0,
μ z + α([∇ z]v − [∇ v]z) = μ curl u + α curl f in Ω,
where PZ is the orthogonal projection from L 2 (Ω)3 onto Z and Φ is the mapping
defined in Lemma 5.8.12. Since Φ takes its values in H 1 (Ω)3 , it follows immediately
from Theorem 3.2.15 that, under the above regularity assumptions on the domain, v
belongs to W 1,∞ (Ω)3 without requiring z in L r (Ω)3 for some r > 3. But on the other
hand, the numerical discretization of Φ(PZ (z)) is a complex impractical operation.
The equivalence between (5.8.32) and the original system is a consequence of the
following lemma:
Lemma 5.8.13 Let Ω be a convex polyhedron. If (u, p, v, z) is a solution of (5.8.32)
with v ∈ W 1,∞ (Ω)3 satisfying (5.8.5), i.e.,
1
|v|1,∞,Ω ≤ (μ − δ), (5.8.33)
α
for some 0 < δ < μ, then (u, p, z) solves (5.8.4) and v = u. Conversely, if
(u, p, z) solves (5.8.4), with u ∈ W 1,∞ (Ω)3 satisfying (5.8.5), then (u, p, u, z)
solves (5.8.32).
Proof Let (u, p, v, z) be like above. First we observe from Lemma 5.8.1 and the
above assumption on v that div z = 0, and hence PZ (z) = z. Next, we take the curl
of the third line in (5.8.32). Considering that by definition curl Φ(z) = z, this gives
μ
−μ curl(Δ v) = z − curl u .
α
But recall that the last line in (5.8.32) reads
μ
z − curl u = curl f − curl(z × v),
α
since both div v = 0 and div z = 0. Therefore

−μ curl(Δ v) = curl f − z × v .
As Ω is simply-connected, there exists π ∈ L 2m (Ω) such that
−μ Δ v + ∇ π + z × v = f .
Therefore, (v, π ) ∈ V × L 2m (Ω) is the unique solution of (5.8.19) with ζ = z.

Comparing with the first two lines of (5.8.32), this implies that v = u and π = p.
Conversely, if (u, p, z) solves (5.8.4), with u ∈ W 1,∞ (Ω)3 satisfying (5.8.5),
then div z = 0 and z satisfies (5.2.7). Thus
curl(u − α Δ u) = z = curl(Φ(z)) = curl Φ(PZ (z)).
Hence there exists q ∈ L 2m (Ω) such that
u − α Δ u + ∇ q = Φ(PZ (z)),
i.e.,
μ μ
−μ Δ u + ∇ q = Φ(PZ (z)) − u).
α α
μ
This yields the third line of (5.8.32) with v = u and pressure q.
α
As noted above, v defined by the third line in (5.8.32) has sufficient regularity to
give meaning to the transport equation (i.e., the last line) in (5.8.32), with z only in
L 2 (Ω)3 . As a consequence, existence of a solution of (5.8.32) can be established by
a Galerkin method. We sketch here the construction.
Let (θ i )i≥1 be a smooth basis of L 2 (Ω)3 and
let X m be the space spanned by the
first m basis functions. For any z m ∈ X m , let u(z m ), p(z m ) ∈ V × L 2m (Ω) be the
unique solution of (5.8.19) with ζ = z m ,
−μ Δ u(z m ) + z m × u(z m ) + ∇ p(z m ) = f in Ω,


and let v(z m ), π(z m ) ∈ V × L 2m (Ω) be the unique solution of the third and fourth
lines of (5.8.32),
μ
−μ Δ v(z m ) + ∇ π(z m ) = Φ(PZ (z m )) − u(z m ) in Ω.
α
Then (5.8.32) is discretized by: Find z m ∈ X m such that
μ z m + α([∇ z m ]v(z m ) − [∇ v(z m )]z m ) = μ curl u(z m ) + α curl f in Ω.
For suitably small data, Brouwer’s fixed point can be used to prove existence of a
Galerkin solution satisfying uniform a priori bounds that suffice to pass to the limit;
the reader will find all details in [32].
5.9 Splitting: The Time-Dependent Problem in 3- D
Here, we split the time-dependent three-dimensional grade-two model, with homoge-

neous Dirichlet boundary conditions (5.2.1)–(5.2.5). Homogeneous boundary condi-
tions are chosen to simplify the discussion: They avoid the additional terms brought
in the a priori estimates by tangential conditions, but the analysis below can be
extended to deal with them.
Recall the original problem:
Find (u, p) ∈ H 1 (0, T ; V ) ∩ L ∞ (0, T ; W 1,∞ (Ω)3 ) × L 2 (0, T ; L 2m (Ω)) such that
∂
∂t
div u = 0 in Ω×]0, T [,
u=0 on ∂Ω×]0, T [,
u(0) = ut 0 in Ω, (5.9.1)
div ut 0 = 0 in Ω,
ut 0 = 0 on ∂Ω.
As expected, we shall split it into a Stokes-like system and a transport system, both
unsteady. Again, the splitting is chosen so that the split problem is well adapted to
numerical approximation. Following Section 5.8, we set
z = curl(u − αΔ u),
we take the curl of the first equation in (5.9.1) and multiply the result by α. This
gives the coupled system: Find (u, p, z) ∈ H 1 (0, T ; V ) ∩ L ∞ (0, T ; W 1,∞ (Ω)3 ) ×
L 2 (0, T ; L 2m (Ω)) × L ∞ (0, T ; L 2 (Ω)3 ) satisfying
5.9 Splitting: The Time-Dependent Problem in 3-D 289
∂
(u − αΔ u) − μ Δ u + z × u + ∇ p = f in Ω×]0, T [,
∂t
div u = 0 in Ω×]0, T [, (5.9.2)
u=0 on ∂Ω×]0, T [,
u(0) = ut 0 in Ω satisfying (5.2.5),
and
∂z
α + μ z + α [∇ z]u − [∇ u]z = μ curl u + α curl f in Ω×]0, T [,
∂t (5.9.3)
z(0) = z t 0 = curl(ut 0 − αΔ ut 0 ) in Ω.
The analysis follows the pattern of Section 5.7, but is more intricate due to the
dimension. Beforehand, we must establish equivalence of the split formulation with
the original problem.
5.9.1 Splitting: Equivalence
As previously, equivalence between (5.9.1) and (5.9.2)–(5.9.3) is achieved by proving

that div z = 0. More precisely, we have the analogue of Lemma 5.8.1, also stated in
a nonhomogeneous setting. The proof is a straightforward consequence of Lemma
5.8.1 and the material of Section 7.3.3.
Lemma 5.9.1 Let Ω be a convex polyhedron. For F given in L 2 (0, T ; H (div, Ω))
with div F = 0, z t 0 given in H (div, Ω) with div z t 0 = 0 and u given in L ∞ (0, T ; W ∩
W 1,∞ (Ω)3 ), the unique solution z ∈ L ∞ (0, T ; L 2 (Ω)3 ) of
∂z
α + μ z + α [∇ z]u − [∇ u]z = F in Ω×]0, T [,
∂t (5.9.4)
z(0) = z t 0 in Ω,
satisfies div z = 0.
Proof The system (5.9.4) is a particular case of (7.3.23) in Section 7.3.3 with β = α,
f = F, and
C = μ I − α [∇ u] ∈ L ∞ (Ω×]0, T [)3×3 .
Therefore, owing to Theorem 7.3.13, it has a unique solution z ∈ L ∞ (0, T ; L 2 (Ω)3 )

and this solution is the weak * limit in L ∞ (0, T ; L 2 (Ω)3 ) of the sequence {z n }n=0
N
in L (Ω) defined by
2 3
α n
(z − z n−1 ) + μ z n + α [∇ z n ]un − [∇ un ]z n = F n , 1 ≤ n ≤ N ,
k (5.9.5)
z 0 = z t 0 in Ω,
with F n ∈ L 2 (Ω)3 and un ∈ W ∩ W 1,∞ (Ω)3 defined almost everywhere in Ω by

tn tn
1 1
F (x) =
n
F(x, t) dt, 1 ≤ n ≤ N , u (x) = n
u(x, t) dt, 1 ≤ n ≤ N .
k t n−1 k t n−1
It is easy to check that F n ∈ H (div, Ω) satisfies div F n = 0 and un ∈ W ∩W 1,∞ (Ω)3

satisfies
Max1≤n≤N ∇ un L ∞ (Ω) ≤ ∇ u L ∞ (Ω×]0,T [) .
At each step n, (5.9.5) is a steady-state system of the form (5.8.9) with the following
correspondence:
α n−1 α
v := un , F := F n + z , μ := μ + .
k k
Since div z 0 = 0, we can proceed by induction and assume that div z n−1 = 0. Hence
α n−1
div(F n + z ) = 0.
k
Moreover, un satisfies (5.8.5) for sufficiently small k. Indeed, upon dividing by α,
(5.8.5) is replaced by
1 μ
∇ un L ∞ (Ω) ≤ + − δ,
k α
for some δ > 0. This holds for all n provided k < k0 where

μ −1
∇ un L ∞ (Ω) − α
if α∇ un L ∞ (Ω) > μ
k0 = (5.9.6)
is arbitrary if α∇ un L ∞ (Ω) ≤ μ.
Thus all the hypotheses of Lemma 5.8.1 are satisfied and hence div z n = 0. By
induction, this is valid for all n. As a consequence, the weak convergence of z n
implies that div z = 0.
In the homogeneous setting, the convexity assumption can be relaxed, and we

have the analogue of Proposition 5.8.3.
Proposition 5.9.2 Let u in L 2 (0, T ; V ∩ W 1,∞ (Ω)3 ). If F ∈ L 2 (0, T ; H (div, Ω))
has the form F = curl G for some function G ∈ L 2 (0, T ; H −1 (Ω)3 ) and z t 0 =
curl ζ for some function ζ ∈ H −1 (Ω)3 , then the statement of Lemma 5.9.1 holds on
a bounded, connected, Lipschitz domain.
Proof Consider equation (5.9.5). We define almost everywhere in Ω

tn
1
G n (x) = G(x, t) dt, 1 ≤ n ≤ N ,
k t n−1
then for all 1 ≤ n ≤ N , G n ∈ H −1 (Ω)3 and F n = curl G n . Also, un ∈

V ∩ W 1,∞ (Ω)3 . Now, the proof proceeds by induction on n. Assume that z n−1 =
curl ζ n−1 for some ζ n−1 ∈ H −1 (Ω)3 . This is true for n = 1 since ut 0 − αΔ ut 0 is in
H −1 (Ω)3 . Then
α n−1 α α
F := F n + z = curl(G n + ζ n−1 ) , G n + ζ n−1 ∈ H −1 (Ω)3 .
k k k
Therefore, by applying Proposition 5.8.3 and the argument of Lemma 5.9.1, we
deduce that div z n = 0 for k sufficiently small. Hence
[∇ z n ]un − [∇ un ]z n = curl(z n × un ) , z n × un ∈ L 2 (Ω)3 ,
and we infer from (5.9.5) that

α n α
μ+ z = curl G n + ζ n−1 − α z n × un ,
k k
where the expression in parentheses belongs to H −1 (Ω)3 . This ends the proof.
5.9.2 Splitting: A Semi-discrete Scheme
In what follows, the assumptions on Ω, r , and σ are those of Section 5.8.2 (see
(5.8.14)). Regarding the data, we suppose that f as well as curl f belong to
L 2 (0, T ; L r (Ω)3 ), ut 0 is in V α (see (5.2.12)) and z t 0 in L r (Ω)3 . Recall that Lemma
5.2.5 guarantees that ut 0 belongs also to C1,σ (Ω)3 .
As in the two-dimensional situation, we approximate (5.9.2)–(5.9.3) by the fol-
lowing semi-discrete scheme: Let N > 1 be an integer, define the time step k, by
T
k= ,
N
and the subdivision points t = n k. For each n ≥ 1, f (t n ) is approximated by
n
tn
1
f (x) =
n
f (x, t)dt, a.e. in Ω, (5.9.7)
k t n−1
and the sequences are initialized by
u0 = ut 0 , z 0 = z t 0 = curl(ut 0 − αΔ ut 0 ). (5.9.8)
Then, our semi-discrete problem reads: Starting from u0 and z 0 , find sequences
{un }n≥1 , {z n }n≥1 and { p n }n≥1 such that un ∈ V , z n ∈ L 2 (Ω)3 and p n ∈ L 2m (Ω)
solve,
1 n 1
(u −un−1 )−α Δ(un −un−1 )−μ Δ un + z n−1 ×un +∇ p n = f n in Ω, (5.9.9)
k k
1
α (z n − z n−1 ) + μ z n + α([∇ z n ]un − [∇ un ]z n )
k
= μ curl un + α curl f n in Ω. (5.9.10)
The divergence-free constraint on un is included in V . In variational form, these

two problems read: Find un ∈ V , z n ∈ L 2 (Ω)3 and p n ∈ L 2m (Ω) solution of
1 α
∀v ∈ H01 (Ω)3 , (un − un−1 , v) + (∇(un − un−1 ), ∇ v) + μ (∇ un , ∇ v)
k k
+ (z n−1
× u , v) − ( p n , div v) = ( f n , v),
n
(5.9.11)
respectively,
α n
∀θ ∈ L 2 (Ω)3 , (z − z n−1 , θ ) + α([∇ z n ]un − [∇un ]z n , θ )
k
+ μ (z n , θ ) = μ (curl un , θ ) + α (curl f n , θ ). (5.9.12)
For each n ≥ 1, given z n−1 in L 2 (Ω)3 , (5.9.9) is a steady Stokes-like system that has a
unique solution (un , p n ) in V × L 2m (Ω). In addition, if z n−1 ∈ L r (Ω)3 with r defined
by (5.8.14), then the straightforward bootstrap argument below shows that the pair
(un , p n ) belongs to C1,σ (Ω)3 × C0,σ (Ω), thus implying that un is in W 1,∞ (Ω)3 . In
turn, given un in W 1,∞ (Ω)3 , the beginning of the proof of Lemma 5.9.1 shows that,
for small enough k, (5.9.10) has a unique solution z n ∈ L 2 (Ω)3 . Moreover, as the
right-hand side of (5.9.10) belongs to L r (Ω)3 , it stems from Lemma 7.3.14 that z n
belongs to L r (Ω)3 .
In the subsequent sections, uniform a priori estimates are established for un , p n ,
z , and the difference quotient of un , under appropriate conditions on the data. These
n
estimates permit to pass to the limit and prove existence of a solution of the mixed
formulation.
5.9.3 Splitting: A Priori Estimates
By testing (5.9.11) with un , a simple application of Young’s inequality yields imme-

diately an unconditional basic bound, for all n, 1 ≤ n ≤ N ,

n
n
un 2α + ui − ui−1 2α +μ k ui 2H 1 (Ω)
0
i=1 i=1
2
S0,2
≤ ut 0 2α + f 2L 2 (Ω×]0,t n [) , (5.9.13)
μ
where, for the sake of conciseness, we use the notation · α defined in (5.2.37). To
simplify, we introduce the notation
2
S0,2 21
Bnu = ut 0 2α + f 2L 2 (Ω×]0,t n [) . (5.9.14)
μ
Thus (5.9.13) implies

n 21
un α ≤ Bnu , ui − ui−1 2α ≤ Bnu ,
i=1
21 √ (5.9.15)
n
Bn
n
tn
k ui 2H 1 (Ω) ≤ √u , k u H01 (Ω)
i
≤ √ Bnu .
i=1
0 μ i=1
μ
These basic bounds are unconditional in the sense that they hold without restriction
on the data. All other bounds are derived recursively. They can take different forms
involving different constants, but for the sake of simplicity, we present only one set
of bounds.
Let us make the following assumption for some n, 1 ≤ n ≤ N , and for all i with
0≤i ≤n−1
α ∇ ui L ∞ (Ω) ≤ μ − δ, (5.9.16)
where 0 < δ < μ is independent of i. Then the matrix of (5.9.10) with index i,
C i = μI − α ∇ ui , satisfies

∀w ∈ IR 3 , (C i w, w) = μ |w|2 − α([∇ ui ] w, w) ≥ |w|2 μ − α ∇ ui L ∞ (Ω) ≥ δ |w|2 .
Hence formula (7.3.30) applied to (5.9.10) gives

n−1
n−1
α z n−1 L 2 (Ω) + δ k z i L 2 (Ω) ≤ α z t 0 L 2 (Ω) + k μ curl ui + α curl f i L 2 (Ω)
i=1 i=1
1
≤ α z t 0 L 2 (Ω) + (μ t n−1
) 2 B un−1 + α curl f L 1 (0,t n−1 ;L 2 (Ω)3 ) .
Similarly, an application of (7.3.37) with p = r yields

n−1
n−1
α z n−1 L r (Ω) +δ kz i L r (Ω) ≤ α z t 0 L r (Ω) + k μ curl ui + α curl f i L r (Ω)
i=1 i=1

n−1
≤ α z t 0 L r (Ω) + μ k curl ui L r (Ω) + α curl f L 1 (0,t n−1 ;L r (Ω)3 ) .
i=1
Again to simplify, we set
B0z =α z t 0 L 2 (Ω)
C0z =α z t 0 L r (Ω) ,
1
Bnz =α z t 0 L 2 (Ω) + (μ t n ) 2 Bnu + α curl f L 1 (0,t n ;L 2 (Ω)3 ) , (5.9.17)

n
Cnz =α z t 0 L r (Ω) + μ k curl ui L r (Ω) + α curl f L 1 (0,t n ;L r (Ω)3 ) .
i=1
It is worth noting that the values of Bnu and Bnz are bounded without restriction on the
data, in contrast to those of Cnz , since we do not yet have an a priori bound for curl ui
in L r (Ω)3 ; this bound will be the object of Lemma 5.9.4 below. We also note that
Bnu , Bnz , and Cnz are monotonic increasing in n. Therefore, as long as (5.9.16) holds,
we have
Bn−1
z Cn−1
z
sup z i L 2 (Ω) ≤ , sup z i L r (Ω) ≤ , (5.9.18)
1≤i≤n−1 α 1≤i≤n−1 α
and of course
B0z C0
z 0 L 2 (Ω) ≤ , z 0 L r (Ω) ≤ z .
α α
All this does not imply a useful estimate for the difference between two consecutive
values z i − z i−1 that we shall need for passing to the limit. A bound of this type can
be derived from formula (7.3.29) applied to (5.9.10),

n−1
α z i − z i−1 2L 2 (Ω) ≤ α z t 0 2L 2 (Ω)
i=1
2 n−1 2
+ μ Bu + α 2 curl f 2L 2 (Ω×]0,t n−1 [) . (5.9.19)
δ
At this stage, we can derive a bound on the divided difference k −1 ui − ui−1 2α .
Lemma 5.9.3 Under assumption (5.9.16), the solution of problem (5.9.8)–(5.9.10)
satisfies

n
1
n
ui − ui−1 2α + μ un 2H 1 (Ω) + μ ui − ui−1 2H 1 (Ω)
i=1
k 0
i=1
0
1
n 2
≤ μ ut 0 2H 1 (Ω) + S0,3 S0,6 Bn−1
z Bu + f 2L 2 (Ω×]0,t n [) . (5.9.20)
0 α μ
3
Proof By testing (5.9.11) with un − un−1 and applying (5.9.18) and Young’s inequal-
ity, we easily derive
1 n μ n 2
u − un−1 2α + u H 1 (Ω) − un−1 2H 1 (Ω) + un − un−1 2H 1 (Ω)
k 2 0 0 0
≤ S0,3 S0,6 z L 2 (Ω) u H01 (Ω) u − u H01 (Ω)

n−1 n n n−1
+ f n L 2 (Ω) un − un−1 L 2 (Ω)

1
≤ S0,3 S0,6 Bn−1 un H01 (Ω) un − un−1 H01 (Ω)
α z
+ f n L 2 (Ω) un − un−1 L 2 (Ω)
11 n 1 1 2
≤ u − un−1 2α + S0,3 S0,6 Bn−1 k un 2H 1 (Ω)
2 k α α z 0
1
+ k f n 2L 2 (Ω) .
2
Then summing over n and using (5.9.15), we obtain

n
1
n
ui − ui−1 2α + μ un 2H 1 (Ω) + μ ui − ui−1 2H 1 (Ω)
i=1
k 0
i=1
0
1 2 n
≤ μ ut 0 2H 1 (Ω) + S0,3 S0,6 Bn−1 k ui 2H 1 (Ω) + f 2L 2 (Ω×]0,t n [)
0 α 3 z
i=1
0
1
n 2
≤ μ ut 0 2H 1 (Ω) + S0,3 S0,6 Bn−1
z Bu + f 2L 2 (Ω×]0,t n [) ,
0 α3μ
and this ends the proof.
To simplify once again, we set

1 2 21
Dnu = μ ut 0 2H 1 (Ω) + 3 S0,3 S0,6 Bn−1 Bnu + f 2L 2 (Ω×]0,t n [) , (5.9.21)
0 α μ z
which depends only on the data and is monotonic increasing in n. Then Lemma 5.9.3
leads to
n
1 i 21
u − ui−1 2α ≤ Dnu . (5.9.22)
i=1
k
As usual, this bound enables us to derive an estimate for the pressure. By testing
(5.9.11) with the function v = T̃ ( p n ) ∈ V ⊥ defined in Theorem 3.2.11, we obtain
1
p n 2L 2 (Ω) = −( f n , v) + (un − un−1 , v) + (z n−1 × un , v).
k
Therefore
1 S0,2 n
p n L 2 (Ω) ≤ S0,2 f n L 2 (Ω) + u − un−1 L 2 (Ω) + S0,3 S0,6 Bn−1
z u n
H 1
(Ω) ,
K̃ k 0
where K̃ > 0 is the constant of (3.2.12). Hence
3
n
k pi 2L 2 (Ω) ≤ 2
S0,2 f 2L 2 (Ω×]0,t n ]) + (Dnu )2
i=1
K̃ 2
1
n 2
+ S0,3 S0,6 Bn−1 B . (5.9.23)
μ z u
5.9.3.1 Bootstrap
The following bootstrap argument will lead to the conclusion that (un , p n ) belongs
to C1,σ (Ω)3 × C0,σ (Ω). For this, we set
α n 1
wn = (u − un−1 ) + μ un , g n = f n − (un − un−1 ) − z n−1 × un , (5.9.24)
k k
and express (5.9.9) as a steady Stokes problem: Find (wn , p n ) ∈ V × L 2m (Ω) such
that
−Δ wn + ∇ p n = g n in Ω.
The first relation (5.9.24) between wn and un can be written as

μ −1 n−1 k μ −1 n
un = 1 + k u + 1+ k w ,
α α α
i.e.,
ρ
un = 1 − ρ k un−1 + k wn , 1 ≤ n ≤ N , (5.9.25)
μ
where
μ μ −1
ρ= 1+ k , 0 < ρ k < 1.
α α
By induction, (5.9.25) reads
n ρ n−i
n
un = 1 − ρ k ut 0 + k wi 1 − ρ k , 1 ≤ n ≤ N. (5.9.26)
μ i=1
This identity gives rise to a number of upper bounds for un expressed in an arbitrary
norm · X . First un satisfies
1 21
n
un X ≤ ut 0 X + √ k wi 2X , 1 ≤ n ≤ N , (5.9.27)
2α μ i=1
where we have used the formulas

ρ n
n−i ρ
n 1 n
2(n−i) 21
i 2 2
kw 1−ρk
i
≤ k w X k 1−ρk
μ i=1 X μ i=1 i=1
ρ 1 21
n
≤ √ k wi 2X ,
μ 2ρ i=1
ρ 1 1
√ ≤√ .
μ 2ρ 2α μ
We can also write
2n ρ
n
un 2X ≤ 2 1 − ρ k ut 0 2X + 2 k wi 2X , 1 ≤ n ≤ N ,
μ i=1
and therefore

n
1 ρ
n i

k ui 2X ≤ ut 0 2X + 2 k k w j 2X .
i=1
ρ μ i=1 j=1
Considering that

n
i n
n
k k w j 2X = k 2 (n + 1 − i)wi 2X ≤ t n k wi 2X ,
i=1 j=1 i=1 i=1
the above inequality yields

n
α tn
n
k ui 2X ≤ 1 + ut 0 2X + k wi 2X , (5.9.28)
i=1
μ α μ i=1
where we have used the fact that k ≤ 1 for estimating the factor ρ −1 in the first term.
Similarly, (5.9.26) also implies

n
α ρ
n i

k ui X ≤ 1 + ut 0 X + k k w j X
i=1
μ μ i=1 j=1
α tn
n
≤ 1+ ut X +
0 k wi X . (5.9.29)
μ α i=1
Finally, we can solve (5.9.24) for the divided difference

α n
(u − un−1 ) = wn − μ un .
k
It stems from this relation that for any norm

n
1 21 √
n
n 21
α ui − ui−1 2X ≤ 2 k wi 2X + μ2 k ui 2X ,
i=1
k i=1 i=1
and (5.9.28) leads to the conclusion

n
1 21
α ui − ui−1 2X
i=1
k
√ 2 α μ
n 21
≤ 2 μ 1+ ut 0 2X + 1 + t n k wi 2X . (5.9.30)
μ α i=1
The bootstrap proceeds in four steps. All constants below are either continuity
constants or Sobolev’s constants, hence independent of n.
(1) Since by assumption, f n belongs to L r (Ω)3 and the divided difference has
the same regularity as un − un−1 , the regularity of g n is determined by that of the
product z n−1 × un . To begin with, by Sobolev’s imbeddings in three dimensions,
3
z n−1 × un belongs to L 2 (Ω)3 and
z n−1 × un L 23 (Ω) ≤ z n−1 L 2 (Ω) un L 6 (Ω) ≤ S0,6 z n−1 L 2 (Ω) un H01 (Ω)
S0,6 n−1 n
≤ B u H01 (Ω) ,
α z
where we have used the first part of (5.9.18) to bound the maximum of z i L 2 (Ω) for
0 ≤ i ≤ n − 1.
3 3
Therefore, as Ω is convex, the pair (wn , p n ) is in W 2, 2 (Ω)3 × W 1, 2 (Ω) and
satisfies the following estimate:
wn W 2, 23 (Ω) + | p n |W 1, 23 (Ω)

1 S0,6 n−1 n
≤ C 23 f n L 23 (Ω) + un − un−1 L 23 (Ω) + Bz u H01 (Ω)
k α
1 1 S0,6 n−1 n
≤ C 23 |Ω| 4 f n L 2 (Ω) + un − un−1 L 2 (Ω) + Bz u H01 (Ω) .
k α
3
By substituting this inequality into (5.9.27) with X = W 2, 2 (Ω)3 , using (5.9.22) and
the first line of (5.9.15), we derive
un W 2, 23 (Ω) ≤ ut 0 W 2, 23 (Ω)

√
3 1 S0,6 1
n 2 2
+√ C 23 |Ω| 2 f 2L 2 (Ω×]0,t n [) + (Dnu )2 + √ Bn−1z Bu .
2α μ α μ
It is convenient to define the following quantities, bounded in terms of the data, and
monotonic increasing in n,
1
Enu = f 2L 2 (Ω×]0,t n [) + (Dnu )2 2 ,
S0,6 1 (5.9.31)
1
n 2 2
Fu,z
n
= |Ω| 2 (Enu )2 + √ Bn−1 z Bu ,
α μ
and obtain a simpler expression

√
3
u W 2, 23 (Ω)
n
≤ ut 0 W 2, 23 (Ω) + √ C 3 Fn . (5.9.32)
2α μ 2 u,z
Similarly,

n
n

n 2
k wi 2 2, 23
+ k | p i |2 1, 23
≤ 3 C 23 Fu,z , (5.9.33)
W (Ω) W (Ω)
i=1 i=1
3
and (5.9.28) with X = W 2, 2 (Ω)3 yields

n
α 3 tn
n 2
k ui 2 2, 23
≤ 1+ ut 0 2 2, 3 + C 23 Fu,z .
i=1
W (Ω) μ W 2 (Ω) αμ
We introduce another notation

α 3 tn 1
n 2 2
Gnu,z = 1+ ut 0 2 2, 3 + C 23 Fu,z , (5.9.34)
μ W 2 (Ω) αμ
a quantity that depends only on the data and is monotonic increasing in n. With this,
the above inequality simplifies

n
2
k ui 2 2, 23
≤ Gnu,z . (5.9.35)
W (Ω)
i=1
(2) Next, by Sobolev’s imbeddings in three dimensions,

3
W 2, 2 (Ω) ⊂ W 1,3 (Ω) ⊂ L q (Ω),
for any finite q. Let us pick for instance q = 12,
un L 12 (Ω) ≤ S̃12 un W 2, 23 (Ω) .
Then
S̃12 n−1 n
z n−1 × un L 127 (Ω) ≤ z n−1 L 2 (Ω) un L 12 (Ω) ≤ B u W 2, 23 (Ω) .
α z
12 12
Again, the convexity of Ω implies that (wn , p n ) is in W 2, 7 (Ω)3 × W 1, 7 (Ω) with
1 1
wn W 2, 127 (Ω) + | p n |W 1, 127 (Ω) ≤ C 127 |Ω| 7 f n L 2 (Ω) + un − un−1 L 2 (Ω)
k
S̃12 n−1 n
+ B u W 2, 23 (Ω) , (5.9.36)
α z
12
and we infer from (5.9.27) with X = W 2, 7 (Ω)3 , (5.9.22), and (5.9.31) that
un W 2, 127 (Ω) ≤ut 0 W 2, 127 (Ω)

√ S̃ n 21
3 2 12 n−1 2
+√ C 127 |Ω| 7 (Enu )2 + Bz k ui 2 2, 3 .
2α μ α i=1
W 2 (Ω)
Then by substituting (5.9.35), we deduce

√
3 2 S̃12 n−1 n 2 21
u W 2, 127 (Ω) ≤ ut 0 W 2, 127 (Ω) + √
n
C 127 |Ω| 7 (Enu )2 + B Gu,z .
2α μ α z
We use a final notation for a quantity that is again bounded only in terms of the data
and is monotonic increasing in n.
2 S̃12 n−1 n 2 21
Hu,z
n
= |Ω| 7 (Enu )2 + B Gu,z . (5.9.37)
α z
Then the previous expression simplifies
√
3
u n
12 ≤ u
t0 12 +√ C 12 Hn . (5.9.38)
W 2, 7 (Ω) W 2, 7 (Ω) 2α μ 7 u,z
Likewise, (5.9.36) and the first part of (5.9.31) imply

n
n
k wi 2 2, 12 + k | p i |2 12
W 7 (Ω) W 1, 7 (Ω)
i=1 i=1
2 S̃12 n−1 2 n
≤ 3 C 212 |Ω| 7 (Enu )2 + Bz k ui 2 2, 3 .
7 α i=1
W 2 (Ω)
By applying (5.9.37) and (5.9.35), this inequality reduces to

n
n

n 2
k wi 2 2, 12
+ k | p i |2 1, 12
≤ 3 C 127 Hu,z .
W 7 (Ω) W 7 (Ω)
i=1 i=1
12
Therefore, (5.9.28) with X = W 2, 7 (Ω)3 gives

n
α 3 tn
n 2
k ui 2 2, 12
≤ 1+ ut 0 2 2, 12 + C 127 Hu,z . (5.9.39)
i=1
W 7 (Ω) μ W 7 (Ω) αμ
12
(3) As W 2, 7 (Ω) ⊂ W 1,4 (Ω) and there is no loss of generality in assuming that
r ≤ 4, formula (5.9.38) permits to estimate curl un in L r (Ω)3 , thus providing a
bound for Cnz in terms of the data. To this end, we use the Sobolev inequality
curl un L r (Ω) ≤ S̃r un W 2, 127 (Ω) , (5.9.40)
and we establish the following lemma:
Lemma 5.9.4 Let Ω be convex and assume that (5.9.16) holds. Then
α μ 1
Cnz ≤μ S̃r 1 + ut 0 W 2, 127 (Ω) + αz t 0 L r (Ω) + S̃r t n C 127 |Ω| 7 f L 1 (0,t n ;L 2 (Ω)3 )
μ α
μ 3 S̃12 n−1 n
+ αcurl f L 1 (0,t n ;L r (Ω)3 ) + S̃r (t n ) 2 C 127 Dnu + B Gu,z . (5.9.41)
α α z
12
Proof According to (5.9.29) with X = W 2, 7 (Ω)3 , we have

n
α tn
n
k u i
12 ≤ 1+ ut W 2, 7 (Ω) +
0 12 k wi W 2, 127 (Ω) .
i=1
W 2, 7 (Ω) μ α i=1
Now it stems from (5.9.36), the Cauchy–Schwarz inequality, and (5.9.35) that

n
k wi 12
W 2, 7 (Ω)
i=1
1 √ S̃12 n−1
n
1
≤ C 12 |Ω| 7 f L 1 (0,t n ;L 2 (Ω)3 ) + t n D nu + Bz k ui 2 2, 3 2
7 α W 2 (Ω)
i=1
1 √ S̃12 n−1 n
≤ C 12 |Ω| 7 f L 1 (0,t n ;L 2 (Ω)3 ) + t n D nu + B Gu,z .
7 α z
Hence

n α
k curl ui L r (Ω) ≤ S̃r 1+ ut 0 2, 12
μ W 7 (Ω)
i=1
tn 1 √ S̃12 n−1 n
+ C 12 |Ω| 7 f L 1 (0,t n ;L 2 (Ω)3 ) + t n D nu + B Gu,z .
α 7 α z
From here (5.9.41) readily follows by substituting this inequality into the definition
of Cnz in (5.9.17) and reordering terms.
(4) Finally, since W 2, 7 (Ω) ⊂ L ∞ (Ω), we have un in L ∞ (Ω)3

12
un L ∞ (Ω) ≤ S̃∞ un W 2, 127 (Ω) ,
and the product z n−1 × un is in L r (Ω)3 with
S̃∞ n−1 n
z n−1 × un L r (Ω) ≤ z n−1 L r (Ω) un L ∞ (Ω) ≤ C u W 2, 127 (Ω) .
α z
Hence we infer from Theorem 3.2.15 that (wn , p n ) belongs to C1,σ (Ω)3 × C0,σ (Ω)
and an application of Sobolev’s imbedding yields
wn C1,σ (Ω) + | p n |C0,σ (Ω)

1 n S̃∞ n−1 n
≤ C∞ f L r (Ω) + u − u L r (Ω) +
n n−1
C u W 2, 127 (Ω)
k α z

S0,r n S̃∞ n−1 n

≤ C∞ f n L r (Ω) + u − un−1 H01 (Ω) + Cz u W 2, 127 (Ω) .
k α
With (5.9.22), this readily implies

n
n
k wi 2C1,σ (Ω) + k | pi |2C0,σ (Ω) ≤ 2
3 C∞ f 2L 2 (0,t n ;L r (Ω)3 )
i=1 i=1
2 S̃ n
S0,r ∞ n−1 2
+ (Dnu )2 + Cz k ui 2 2, 12 . (5.9.42)
α α i=1
W 7 (Ω)
By inserting this inequality into (5.9.27) with X = C1,σ (Ω)3 , we obtain

√ 2
3 S0,r
u C1,σ (Ω) ≤ ut 0 C1,σ (Ω) + √
n
C∞ f 2L 2 (0,t n ;L r (Ω)3 ) + (Dnu )2
2α μ α
S̃ n 21
∞ n−1 2
+ Cz k ui 2 2, 12 . (5.9.43)
α i=1
W 7 (Ω)
Furthermore, by substituting (5.9.42) into (5.9.30), we derive

n
1 21
α ui − ui−1 2C1,σ (Ω)
i=1
k

√ α μ
≤ 2 μ2 1 + ut 0 2C1,σ (Ω) + 3 C∞2
1 + t n | f 2L 2 (0,t n ;L r (Ω)3 )
μ α
2
S0,r S̃ n 21
∞ n−1 2
+ (Du ) +
n 2
C k u 2, 12
i 2
,
α α z i=1
W 7 (Ω)
and (5.9.39) gives

n
1 21 √ α
α ui − ui−1 2C1,σ (Ω) ≤ 2 μ2 1 + ut 0 2C1,σ (Ω)
i=1
k μ

μ 2
S0,r
+ 3 C∞ 2
1 + t n f 2L 2 (0,t n ;L r (Ω)3 ) + (Dnu )2
α α
S̃
! 21
∞ n−1 2 α 3 tn
n 2
+ C 1+ ut 0 2, 12
2
+ C 12 H . (5.9.44)
α z μ W 7 (Ω) α μ 7 u,z
The next lemma is a straightforward consequence of (5.9.43) and (5.9.39).
Lemma 5.9.5 If Ω is convex, (5.9.16) holds, f and curl f belong to L 2 (0, T ;

L r (Ω)3 ), ut 0 is in V α , and z t 0 in L r (Ω)3 , then the velocity pressure part (un , p n )
of the solution of the semi-discrete scheme (5.9.8)–(5.9.10) belongs to C1,σ (Ω)3 ×
C0,σ (Ω) and
√
3
∇ u L ∞ (Ω) ≤ u C1,σ (Ω) ≤ ut 0 C1,σ (Ω) + √
n n
C∞ f 2L 2 (0,t n ;L r (Ω)3 )
2α μ

21
2
S0,r n 2 S̃∞ n−1 2 α 3 tn n 2

+ (Du ) + C 1+ ut 0 2, 12
2
+ C 12 H .
α α z μ W 7 (Ω) α μ 7 u,z
5.9.3.2 Uniform Estimates
Lemma 5.9.5 shows that (5.9.16) is valid at level n if the data satisfy
√ 2 S̃
3 S0,r ∞ n−1 2
α ut 0 C1,σ (Ω) + √ C∞ f 2L 2 (0,t n ;L r (Ω)3 ) + (Dnu )2 + Cz
2α μ α α

21 !
α 3t
n
n 2

× 1+ ut 0 2 2, 12 + C 127 Hu,z ≤ μ − δ. (5.9.45)
μ W 7 (Ω) αμ
Let Cz , Du , and Hu,z denote the maximum with respect to n of respectively Cnz , Dnu ,
and Hu,z
n
. The following theorem collects the results of this section.
Theorem 5.9.6 Let Ω be convex with r and σ related to the largest inner angle
of ∂Ω by (5.8.14). Let f be given in L 2 (0, T ; L r (Ω)3 ) with curl f also in
L 2 (0, T ; L r (Ω)3 ), and let ut 0 and z t 0 be respectively given in V α and L r (Ω)3 .
Suppose that the data satisfy
√ 2
3 S0,r
α ut 0 C1,σ (Ω) + √ C∞ f 2L 2 (0,T ;L r (Ω)3 ) + (Du )2
2α μ α

1 !
S̃∞ 2 α 3T 2 2
+ Cz 1+ ut 0 2, 12
2
+ C 12 Hu,z ≤ μ − δ. (5.9.46)
α μ W 7 (Ω) αμ 7
Then the solution (un , p n , z n ) of the semi-discrete scheme (5.9.8)–(5.9.10) satisfies

the following uniform bounds for all n, 1 ≤ n ≤ N :
21
1 21
un C1,σ (Ω) + ui − ui−1 2α + ui − ui−1 2C1,σ (Ω) ≤ C,
1≤i≤n 1≤i≤n
k
21 21
k pi 2L 2 (Ω) + k | pi |2C0,σ (Ω) ≤ C,
1≤i≤n 1≤i≤n
21
z n L r (Ω) + z i − z i−1 2L 2 (Ω) ≤ C. (5.9.47)
1≤i≤n
The data condition (5.9.46) holds provided the force f and initial data ut 0 and z t 0
are sufficiently small.
Proof On one hand, the condition (5.9.46) implies that (5.9.16) is valid when n = 1
(i.e., at the initial step). On the other hand, in view of Lemma 5.9.5, (5.9.46) implies
that if (5.9.16) is true for all i ≤ n − 1, then it is also true for i = n. Therefore by
induction, it holds for all n ≤ N . Furthermore, a simple inspection of the definitions
of Bnu , Bnz , Dnz , Enu , Fu,z
n
, Gnu,z , and Hu,z
n
, and the bound (5.9.41) for Cnz , easily shows
that (5.9.46) is valid when f , ut 0 , and z t 0 are sufficiently small. Finally, as all these
quantities are bounded independently of n ≤ N , the estimates (5.9.47) immediately
follow from (5.9.15), (5.9.18), (5.9.19), (5.9.22), (5.9.23), (5.9.42), Lemma 5.9.5,
(5.9.44), and of course (5.9.45).
5.9.4 Splitting: Passing to the Limit
Form now on, we adopt the assumptions of Theorem 5.9.6, thus guaranteeing suffi-
cient a priori estimates for passing to the limit in (5.9.11) and (5.9.12). The procedure
is much the same as in the two-dimensional situation of Section 5.7.2, but the homo-
geneous boundary condition simplifies the discussion.
The sequences (un ), ( p n ), (z n ), and ( f n ) are transformed into functions, either
continuous piecewise linear or constant in each subinterval. We set for all t ∈
[t n−1 , t n ] and 1 ≤ n ≤ N ,
t − t n−1 n
uk (t) = un−1 + (u − un−1 ),
k
t − t n−1 n
z k (t) = z n−1 + (z − z n−1 ),
k
and for all t ∈]t n−1 , t n ], 1 ≤ n ≤ N ,
f k (t) = f n , wk (t) = un , pk (t) = p n , ζ k (t) = z n ,
and for all t ∈ [t n−1 , t n [, 1 ≤ n ≤ N ,
λk (t) = z n−1 .
The functions uk and z k satisfy the same formulations as in (5.7.66) and (5.7.67),
up to minor dimensional changes and an additional nonlinear term in the transport
equation,
∀v ∈ H01 (Ω)3 , ∀ϕ ∈ C0 ([0, T ]),

T
∂ ∂
( uk (t), v) + α( ∇ uk (t), ∇ v) + μ(∇ wk (t), ∇ v)
0 ∂t ∂t
T

+ (λk (t) × wk (t), v) − ( pk (t), div v) ϕ(t)dt = ( f k (t), v)ϕ(t)dt, (5.9.48)
0
respectively,
∀θ ∈ W 1,4 (Ω)3 , ∀ψ ∈ C1 ([0, T ]) with ψ(T ) = 0,

T
− (z k (t), θ )ψ (t)dt − (z 0 , θ )ψ(0)
0
T

+ μ(ζ k (t), θ) − α [∇ θ]wk (t), θ − α [∇ wk (t)]ζ k (t), θ ψ(t)dt
0
T

= μ(curl wk (t), θ ) + α(curl f k (t), θ ) ψ(t)dt. (5.9.49)
0
They converge in stronger norms than in Lemma 5.7.21.

Lemma 5.9.7 There exist u ∈ H 1 (0, T ; C1,σ (Ω)3 ), p ∈ L 2 (0, T ; C0,σ (Ω)3 ∩
L 2m (Ω)) and z ∈ L ∞ (0, T ; L r (Ω)3 ) such that up to subsequences,
lim uk = u strongly in C0 (0, T ; C1 (Ω)3 ), (5.9.50)

k→0
lim uk = u weakly in H 1 (0, T ; W 1+σ,∞ (Ω)3 ), (5.9.51)
k→0
lim pk = p weakly in L 2 (0, T ; W σ,∞ (Ω)), (5.9.52)
k→0
lim z k = z weakly in L ∞ (0, T ; L r (Ω)3 ). (5.9.53)
k→0
Moreover,
lim wk − uk L 2 (0,T ;C1,σ (Ω)3 ) = 0,

k→0
(5.9.54)
lim ζ k − z k L 2 (Ω×]0,T [) = lim λk − z k L 2 (Ω×]0,T [) = 0,
k→0 k→0
and
lim wk = u strongly in L 2 (0, T ; C1 (Ω)3 ), (5.9.55)

k→0
lim ζ k = lim λk = z weakly in L 2 (Ω×]0, T [)3 . (5.9.56)
k→0 k→0
Finally,
α ∇ u L ∞ (Ω) ≤ μ − δ, (5.9.57)
where μ − δ with 0 < δ < μ, is the constant of (5.9.45).

Proof We use the fact that for any nonnegative integer m and real number σ with
0 < σ < 1, Cm,σ (Ω) coincides algebraically and topologically with W m+σ,∞ (Ω).
Thus Theorem 5.9.6 establishes that uk and pk are uniformly bounded, respectively, in
H 1 (0, T ; W 1+σ,∞ (Ω)3 ) and L 2 (0, T ; W σ,∞ (Ω)), as well as z k in L ∞ (0, T ; L r (Ω)3 ).
This accounts for the limits in (5.9.51)–(5.9.53). The first limit ensues from the fact
that for all 0 ≤ τ ≤ t ≤ T ,
uk (t) − uk (τ )C1,σ (Ω) ≤ (t − τ ) 2 uk L 2 (0,T ;C1,σ (Ω)3 ) .

1
As a consequence, the sequence {uk } is equicontinuous and bounded in the space

C0 (0, T ; C1 (Ω)3 ) for instance, and the strong convergence follows from Ascoli–
Arzela’s Lemma. The upper bound (5.9.57) stems from this strong convergence,
Lemma 5.9.5 and the assumption (5.9.46).
Regarding (5.9.54), the second line is the same as (5.7.65), while the first line is
a consequence of (5.9.47) and the following immediate generalization of (5.7.64),
valid for any norm X :
k2 1 n
N
wk − uk 2X ≤ u − un−1 2X ,
3 n=1 k
applied with X = C1,σ (Ω)3 . Finally, (5.9.56) is a consequence of (5.9.53) and

(5.9.54).
It remains to pass to the limit in (5.9.48) and (5.9.49). The next theorem is the
analogue of Theorem 5.7.22.
Theorem 5.9.8 Under the assumptions of Theorem 5.9.6, problem (5.9.2), (5.2.5),
(5.9.3) has at least one solution u ∈ H 1 (0, T ; C1,σ (Ω)3 ), p ∈ L 2 (0, T ; C0,σ (Ω) ∩
L 2m (Ω)) and z ∈ L ∞ (0, T ; L r (Ω)3 ). This solution satisfies the same bounds as in
Theorem 5.9.6, namely
u H 1 (0,T ;C1,σ (Ω)3 ) + p L 2 (0,T ;C0,σ (Ω)) + z L 2 (0,T ;L r (Ω)3 ) ≤ C,
where C depends only on the data. In particular u satisfies (5.9.57).
Proof Passing to the limit proceeds as in the proof of Theorem 5.7.22, with the
exception of the additional nonlinear transport term. This term is easily handled,
owing to the strong convergences of Lemma 5.9.7. For instance, we can write
T
T
([∇ wk ]ζ k , θ )ψ(t) dt = [∇(wk − uk )]ζ k , θ ψ(t) dt
0 0
T T

+ [∇(uk − u)]ζ k , θ ψ(t) dt + ([∇ u]ζ k , θ )ψ(t) dt.
0 0
For the last term one has

T T
lim ([∇ u]ζ k , θ )ψ(t) dt = ([∇ u]z, θ )ψ(t) dt,
k→0 0 0
because ζ k → z weakly in L 2 (Ω×]0, T [)3 and ψ [∇ u]θ is bounded in

L 2 (Ω×]0, T [)3 . The first two terms tend to zero because
T

([∇(wk − uk )]ζ k , θ )ψ(t) dt ≤∇(wk − uk ) L 2 (0,T ;L ∞ (Ω)3×3 )
0
× ζ k L ∞ (0,T ;L 2 (Ω)3 ) θ ψ L 2 (Ω×]0,T [) ,
T

([∇(uk − u)]ζ k , θ )ψ(t) dt ≤∇(uk − u) L ∞ (Ω×]0,T [)
0
× ζ k L 2 (Ω×]0,T [) θ ψ L 2 (Ω×]0,T [) ,
whence the desired limit.
We close this section with a remark on uniqueness. Since the split formulation
is equivalent to the original formulation in a bounded, connected, Lipschitz domain
(see Proposition 5.9.2), and the assumptions on the original formulation follow from
those on the split formulation, we can merge the existence Theorem 5.9.8 with the
uniqueness Theorem 5.2.17.
Theorem 5.9.9 We suppose that the assumptions of Theorem 5.9.8 hold and that the
solution (u, p, z) constructed above is such that u belongs to L 1 (0, T ; W 2,3 (Ω)3 ).
Then this is the only solution of Problem (5.9.2), (5.2.5), (5.9.3).
5.10 Grade-Two Fluids: Some Open Problems
We briefly discuss here a few open problems with regard to grade-two fluids. As
mentioned in the introduction, many aspects of grade-two fluids are still unresolved.
Among these issues, we broach below the problems posed by boundary conditions,
by rough forcing terms, by the analysis of numerical schemes for approximating the
solution, and by the limit as the α parameter tends to zero.
5.10.1 Other Boundary Conditions
From a physical point of view, the most important issue is the problem raised by
boundary conditions. Some of these conditions are presented in Section 2.5, and the
difficulties raised by the important class of free surface flows have been broached
in Section 2.10.3. In this work, we have only been able to analyze homogeneous or
tangential Dirichlet boundary conditions, with substantial technical difficulties in the
latter case, while the fully nonhomogeneous problem remains open. The Navier slip
(2.5.1) has been investigated by Tani and Leroux in [262], but there are not many other
references of work in this direction. Instead, over the past decade, several authors
have either studied the grade-two fluid in the whole space IR 2 or IR 3 or the free-slip
boundary condition (see for instance Busuioc and Ratiu [58] and the references in
Busuioc [56]) on the entire boundary, i.e.,
5.10 Grade-Two Fluids: Some Open Problems 309
u · n = 0 and T T n · τ = 0, (5.10.1)
which, compared with (2.5.1), i.e.,
u · n = 0 and u · τ = −kT T n · τ ,
is a limited example of a Navier slip.

In a sense a free-slip condition is less stringent that a Dirichlet boundary condition
and some properties are more readily established with a free-slip because free-slip
permits some mathematical manipulations that are not possible in the presence of a
Dirichlet boundary condition. This is also true if no boundary condition is prescribed.
From the stand point of mathematics, the corresponding results are interesting, but
they seem to have no practical application.
In contrast to free-slip, the stick-and-slip condition is physically significant, but
its analysis is fairly complex. We refer to the work of Hron et al [134] on the flow of
incompressible fluids with the simplified form of stick-and-slip boundary condition
given by (2.5.2) and (2.5.3),
|T T n · τ | ≤ g ⇒ v · τ = 0,
v·τ
|T T n · τ | > g ⇒ T T n · τ = − .
|v · τ |
5.10.2 Analysis of Numerical Schemes
A variety of numerical schemes are proposed and analyzed for the two-dimensional
grade-two fluid, steady or time dependent, with tangential Dirichlet boundary con-
ditions, in the reference by Girault and Hecht [112]. Their numerical analysis, based
on stability estimates, is possible because in two dimensions, sufficient a priori esti-
mates can be derived without requiring a Lipschitz velocity. This property plays a
fundamental part in the results established in Sections 5.6 and 5.7. This is not the
case in three dimensions, as can be observed in Sections 5.8 and 5.9. Surprisingly, the
difficulty does not lie in establishing uniform bounds for the gradient of the discrete
velocity in the maximum norm. These can be derived from the work of Guzmán
and Leykekhman [128] and Girault et al. [114]. Instead the elusive part of the proof
is concentrated on the derivation of uniform L p estimates for a discrete transport
equation. The reader will find in the material of Section 7.2.2, that in the exact case,
suitable a priori estimates are formally derived by testing the transport equation with
functions of the form |z| p−2 z. Clearly, this is not possible in nearly all discrete cases
because functions of this form do not belong to the discrete space. Thus, to this
date, discrete L p estimates for transport equations are an open problem for almost
all discrete schemes. Consequently, in three dimensions, the numerical analysis of
schemes solving grade-two fluids remains a challenging open problem.
5.10.3 Rough Force
The arguments of the preceding sections rely on some regularity of the curl of
f . Therefore, it does not extend to f in L p (Ω)d . Nevertheless, solutions have been
constructed by Bresch and Lemoine in [48], with f in L p (Ω)d , using another splitting
where the auxiliary variable is u − α Δ u instead of curl(u − α Δ u). The drawback
of this choice of auxiliary variable is that the equation satisfied by this new variable is
more complex that the transport equation (5.2.10), and, in particular, does not permit
to obtain in two dimensions existence of solutions for all data.
5.10.4 Limit as α Tends to Zero
To this date, passing to the limit as α tends to zero in the grade-two fluid system
with Dirichlet boundary conditions is an open problem, except in the steady two-
dimensional situation, as done in Section 5.6.6. This is made possible by the fact
that the principal estimates are independent of negative powers of α, in contrast
to the steady system in three dimensions or the time-dependent problem. Even the
time-dependent two-dimensional system with homogeneous Dirichlet boundary con-
ditions is baffling. Indeed, a uniform in α estimate for z is not known and in turn this
jeopardizes an estimate for the pressure. The Dirichlet boundary condition is largely
responsible for this difficulty. When the problem is set into the whole space IR d , this
limit is established by Iftimie in [139], but the extension of this work’s material to
Dirichlet boundary conditions is problematic due again to the presence of the pres-
sure. Indeed, when such conditions are prescribed, the formula above (12) in this
reference, which is used to eliminate the pressure from the time derivative, does not
seem to be valid.
Chapter 6
Short Survey on the Theory
of Grade-Three Fluids
6.1 Introduction
There is no space here for a detailed study of grade-three fluids. Consequently, in this
chapter we briefly present a few salient results on the mathematical analysis of the
equations modeling the flow of grade-three fluids. As is the case for the grade-two
model, the results are more favorable in two dimensions, but for the sake of brevity,
we restrict the presentation to the three-dimensional case. Moreover, for the sake of
simplicity, we only consider purely homogeneous Dirichlet boundary conditions.
The first successful analysis of the grade-three model was done in 1986 by
Amrouche in his thesis [5]; it appeared as a technical report by Amrouche and
Cioranescu [7] in 1988 and was published in 1997 by the same authors in [8].
Again, a large number of mathematicians have worked on the theoretical analy-
sis of both steady and unsteady grade-three fluids; the reader will find an extensive
list of references in Videman’s thesis [279], and recent references in Busuioc’s habil-
itation [56]. The results presented here are taken mostly from the above-cited contri-
butions of Amrouche and Cioranescu, the publications of Bernard [30] and Bernard
and Ouazar [33], and a very interesting recent work on the steady-state problem by
Busuioc and Iftimie in [57].
6.1.1 Formulations
In this section, we present two equivalent formulations for the equation of motion of
grade-three fluids.
We have seen in Section 2.4 that the Cauchy stress for a thermodynamically
compatible grade-three fluid is given by (2.4.20), i.e.,
T = − p I + μ A1 + α1 A2 + α2 A21 + β3 (tr A21 ) A1 ,

DOI 10.1007/978-3-319-39330-8_6
312 6 Short Survey on the Theory of Grade-Three Fluids
where the symmetry of A1 implies
tr A21 = A1 : A1 = | A1 |2 .
By expanding formula (2.3.26) for A2 ,
d ∂ ∂
3
A2 = A1 + A1 L + L T A1 = A1 + ui A1 + A1 L + L T A1 ,
dt ∂t i=1
∂ x i
and substituting into (2.4.20), this reads
∂ 3
∂
T = − p I + μ A1 + α1 A1 + ui A1 + A1 L + L T A1
∂t i=1
∂ xi
+ α2 A21 + β3 | A1 |2 A1 . (6.1.1)
As in Section 3.3, (6.1.1) can also be written in terms of the spin tensor W and it has
the equivalent formula
∂ 3
∂
T = − p I + μ A1 + α1 A1 + ui A1 + A1 W − W A1
∂t i=1
∂ xi
+ (α1 + α2 ) A21 + β3 | A1 |2 A1 . (6.1.2)
Recall the balance of linear momentum (5.1.2) where is the density and f a forcing
term,
d
u = div T + f .
dt
The divergence of T is deduced from an alternate expression of (5.1.4),
1
div( A21 ) = A1 Δ u + 2 div(L L T ) + ∇ | A1 |2 , (6.1.3)
4
and from the following analog of (5.1.3):

3
∂
div ui A1 + A1 L + L T A1 = curl(Δu) × u + A1 Δ u
i=1
∂ xi
1
+ 2 div(L L T ) + ∇ u · Δ u + (| A1 |2 . (6.1.4)
4
6.1 Introduction 313
By substituting (6.1.1) with (6.1.3) and (6.1.4) into this equation, dividing by the
density and denoting the parameters by the same symbols, we obtain the following
equation of motion:
∂
(u − α1 Δ u) − μ Δ u + curl(u − α1 Δ u) × u
∂t
− (α1 + α2 ) A1 Δ u + 2 div(L L T ) − β3 div | A1 |2 A1 + ∇ p = f , (6.1.5)
where p collects all the gradient terms,
1 1 1
p := p − α1 [∇ u]u − (2 α1 + α2 )| A1 |2 + |u|2 .
4 2
If we use instead (6.1.2), then (6.1.5) is replaced by the equivalent form (2.4.26)
∂
∂t
− (α1 + α2 ) Δ([∇ u]u) − 2[∇(Δ u)]u − β3 div(| A1 |2 A1 ) + ∇ p = f ,
where the new “pressure” is
1 1 1
p := p − (2 α1 + α2 ) u · Δ u + | A1 |2 + |u|2 .
4 2
These equations are complemented by the incompressibility condition, an initial

condition at time t = 0, and, as indicated above, a homogeneous Dirichlet boundary
condition.
Let Ω be a bounded domain of IR 3 with a Lipschitz-continuous boundary ∂Ω and
let ]0, T [, T > 0, be a time interval. If we use (6.1.5), the problem reads: Given
a vector-valued function f defined on Ω×]0, T [, and material constants α1 , α2 , μ,
β3 , find a vector-valued function u = (u 1 , u 2 , u 3 ) and a scalar function p defined on
Ω×]0, T [, satisfying in Ω×]0, T [,
∂
(u − α1 Δ u) − μ Δ u + curl(u − α1 Δ u) × u
∂t
− (α1 + α2 ) A1 (u)Δ u + 2div(L(u)L(u)T )

− β3 div | A1 (u)|2 A1 (u) + ∇ p = f , (6.1.6)
and
div u = 0 in Ω×]0, T [, (6.1.7)
with homogeneous Dirichlet boundary conditions,
u = 0 on ∂Ω×]0, T [, (6.1.8)
and initial data,

u(0) = ut 0 in Ω, (6.1.9)
satisfying the compatibility conditions
div ut 0 = 0 in Ω and ut 0 = 0 on ∂Ω. (6.1.10)
If we use instead (2.4.26), then (6.1.6) is replaced by
∂
∂t
− (α1 + α2 ) Δ([∇ u]u) − 2[∇(Δ u)]u
− β3 div(| A1 (u)|2 A1 (u)) + ∇ p = f , (6.1.11)
while (6.1.7)–(6.1.10) are unchanged. Of course, the steady formulation is obtained

by suppressing the partial time derivative and the initial condition.
In all cases, following (2.4.21), we restrict the parameters so that the fluid is
thermodynamically compatible according to Fosdick and Rajagopal [101],

μ ≥ 0 , α1 ≥ 0 , β3 ≥ 0 , |α1 + α2 | ≤ 24μβ3 .
Although the additional nonlinear terms in the grade-three model look formidable,
the flow has more stability than that of a grade-two model. This is not surprising,
because the factor of β3 represents a higher order viscosity that dominates the other
terms, if the last relation above is satisfied. For this reason, unless otherwise specified,
we restrict further the parameters and prescribe the following strict inequalities:

μ > 0 , α1 > 0 , β3 > 0 , |α1 + α2 | < 24μβ3 . (6.1.12)
6.1.2 Strategies
The methods of Amrouche, Cioranescu, Bernard, and Ouazar for constructing solu-
tions are essentially the same as those used here in the case of grade-two fluids. To sim-
plify, suppose that the domain Ω is simply-connected, with a boundary that is either
smooth or is a convex polyhedron. First, the original equation (6.1.6) (or (6.1.11))
gives rise to a valuable unconditional a priori estimate for u in L ∞ (0, T ; H 1 (Ω)3 )
and in L 4 (0, T ; W 1,4 (Ω)3 ). This stronger stability is due to the effect of β3 . Next,
these authors take the curl of the original equation (6.1.6), or (6.1.11), and introduce
the auxiliary variable z = curl(u − α1 Δ u) that lends more regularity to u, see
Lemma 5.2.5. This new equation yields suitable a priori estimates for z, through a
Riccati type differential inequality, that in turn permits to derive an a priori estimate
6.1 Introduction 315
for the time derivative of u. Again, the construction per se is achieved by discretiz-
ing (6.1.6), or (6.1.11), in the basis of eigenfunctions defined by (5.2.27), (with α1
instead of α),
∀v ∈ V α1 , ((w j , v)) = λ j {(w j , v) + α1 (∇ w j , ∇ v)},
that allows us to recover the above-mentioned a priori estimates for z without taking
explicitly the curl of (6.1.6), or (6.1.11), which would not be possible in a Galerkin
formulation. The discrete solution satisfies the same a priori estimates, uniform with
respect to the discretization parameter, that permit to pass to the limit and prove local
in time existence of a solution of the original equation.
In this procedure, the last part of (6.1.12) is employed to bound some terms in
the Riccati inequality relative to z. If it does not hold, the nonlinear terms of the
Riccati inequality are stronger, but nevertheless it still has a local-in-time solution
(on a shorter interval). Thus, Bernard in [30] still establishes existence of solutions,
while only assuming
μ > 0 , α1 > 0 , β3 > 0.
But of course, the interval of existence is reduced.

The analysis by Bernard and Ouazar in [33] of the steady grade-three model
proceeds along much the same lines: They discretize the equation of motion by
Galerkin’s method with the special basis (5.2.27) that permits to recover estimates
for the auxiliary variable z. Existence of discrete solutions follows by Brouwer’s
fixed point and makes use of the dominating character of β3 div(| A1 |2 A1 ) owing
to (6.1.12). Since there is no time derivative, a Riccati differential inequality is no
longer associated with the curl of the equation of motion. It is replaced by a continuity
property of the velocity solution that relies on the monotonicity of β3 div(| A1 |2 A1 ),
and again crucially depends on (6.1.12).
Finally, by restricting further the last part of (6.1.12), namely
3α12 + 4(α1 + α2 )2 ≤ 24 μ β, (6.1.13)
Busuioc and Iftimie in [57] made recently a remarkable contribution to the steady
three-dimensional grade-three model. More precisely, they establish existence of a
velocity solution u ∈ W 1,4 (Ω)3 corresponding to a forcing term f in W −1,4 (Ω)3 ,
with no restriction on the size of f . Their crucial idea is that the nonlinear operator
R(u) := −μ Δ u − α1 div( A1 L + L T A1 ) − α2 div( A21 ) − β3 div(| A1 |2 A1 )
is monotone. In addition to the basic estimate derived from (6.1.5), this property lends
sufficient stability without requiring the auxiliary function z, i.e., without requiring
the curl of the equation of motion, whence the low regularity requirement on the
forcing term. Then their construction proceeds also via a Galerkin method, but now
the special basis (5.2.27) is no longer necessary and the basis of eigenfunctions of
the Stokes operator is sufficient.
6.2 Variational Formulation. The Time-Dependent

Problem
In this section, we present essentially the work of Amrouche and Cioranescu. Their
initial assumptions are: Ω simply-connected with a smooth enough connected bound-
ary ∂Ω, f in H 1 (Ω)3 and ut 0 ∈ V α1 , where V α is defined in (5.2.12),
V α = {v ∈ V ; α curl Δ v ∈ L 2 (Ω)3 }.
While, in this work, this regularity of the initial data cannot be relaxed, it is obvious
that it suffices that f belongs to H (curl, Ω) instead of H 1 (Ω)3 . Furthermore, in
order to state the problem, it is sufficient that Ω meet the regularity requirements
of Lemma 5.2.5, i.e., either it is a convex polyhedron or its boundary is C 1,1 . The
connectedness of the boundary is not necessary, and finally the simple connectedness
of the domain can also be relaxed, as is done by Bernard and Ouazar in [33]. The
chances are that Lemma 5.3.2, which leads to simpler results, can also be applied here
for handling multiply connected domains, but this has not been done yet. However,
to simplify, we shall restrict the discussion to simply-connected domains. Thus, in
addition to (6.1.12), the assumptions are as follows:
Ω a convex polyhedron or simply-connected with a C 1,1 boundary, (6.2.1)
f ∈ H (curl, Ω), ut 0 ∈ V α1 . (6.2.2)
Amrouche and Cioranescu use the form (6.1.6) for the equation of motion. Thus,
their variational formulation is deduced by taking the scalar product of (6.1.6) with
a test function v in W01,4 (Ω)3 , and applying the following formulas:
1
∀u ∈ W 1,4 (Ω)3 , div | A1 (u)|2 A1 (u) , v = − | A1 (u)|2 A1 (u), A1 (v) ,
2
3
∂u ∂u (6.2.3)
∀u ∈ V ∩ H 2 (Ω)3 , 2 div(L(u)L(u)T ), v) = 2 c ; ,v ,
k=1
∂ xk ∂ xk
where the trilinear form c is defined in (5.2.56),
3
3
∂v j
c(u; v, w) = [∇ v]u, w = ui w j d x.
i=1 j=1 Ω ∂ xi
This leads to the following equation: Find u ∈ L ∞ (0, T ; V α1 ) ∩ H 1 (0, T ; V ) and p

4
in L 2 (0, T ; L m3 (Ω)) such that for all v ∈ W01,4 (Ω)3 ,
6.2 Variational Formulation. The Time-Dependent Problem 317
(u , v) + α1 (∇ u , ∇ v) + μ (∇ u, ∇ v) + (z × u, v)
3
∂ u ∂ u
− (α1 + α2 ) c(Δ u; u, v) + c(v; u, Δ u) + 2 c ; ,v

k=1
∂ xk ∂ xk
β3
+ | A1 (u)|2 A1 (u), A1 (v) − ( p, div v) = ( f , v), (6.2.4)
2
with the initial condition (6.1.9)
u(0) = ut 0 in Ω,
where ut 0 satisfies the compatibility conditions (6.1.10) and z and u are related by
z = curl(u − α1 Δ u). (6.2.5)
Note that the right-hand side of (6.2.4) is well-defined for f in L 2 (0, T ; W −1, 3 (Ω)3 )
4
and all terms in its left-hand side make sense when u belongs to L ∞ (0, T ; V α1 ), thus
implying that z belongs to L ∞ (0, T ; L 2 (Ω)3 ). Nevertheless, this regularity does not
follow directly from (6.2.4), but can be derived by taking the curl of (6.1.6).
Let us take the curl of both sides of (6.1.6). To simplify, we set ω = curl u. As
previously, an application of (5.2.9) gives immediately,
curl(z × u) = [∇ z]u − [∇ u]z.
Next, considering that div u = 0, we have

div | A1 (u)|2 A1 (u) = | A1 (u)|2 Δ u + A1 (u)∇ | A1 (u)|2 .
Hence

curl div | A1 (u)|2 A1 (u) = | A1 (u)|2 Δ ω + ∇ | A1 (u)|2 × Δ u

+ curl A1 (u)∇ | A1 (u)|2 .
Denoting the jth column of A1 (u) by A1 (u)·, j yields
3
∂ ∂ω
curl A1 (u)∇ | A1 (u)|2 = | A1 (u)|2 ·
j=1
∂x j ∂x j

3
∂
+ ∇ | A1 (u)|2 × A1 (u)·, j .
j=1
∂x j
Thus,

curl div | A1 (u)|2 A1 (u) = | A1 (u)|2 Δ ω − ∇ | A1 (u)|2 × curl ω
3
∂ ∂ω ∂
+ | A1 (u)|2 · +∇ | A1 (u)|2 × A1 (u)·, j .
j=1
∂x j ∂x j ∂x j
Note that the first term in the right-hand side can be expressed as a function of z,
1
| A1 (u)|2 Δ ω = − | A1 (u)|2 z − ω .
α1
Therefore
1
curl div | A1 (u)|2 A1 (u) = − | A1 (u)|2 z − ω − ∇ | A1 (u)|2 × curl ω
α1
3
∂ ∂ω ∂
+ | A1 (u)|2 · +∇ | A1 (u)|2 × A1 (u)·, j .
j=1
Finally,
3
curl A1 (u)Δ u = ∇(Δ u j ) × A1 (u)·, j + [∇ ω]Δ u,
j=1
and

curl div L(u)L(u)T = curl div ∇ u ∇ u T
3
∂2u
3
∂u j ∂ 2ω ∂u j
= +∇ × .
k=1 j=1
∂ xk ∂ x j ∂ xk ∂ xk ∂ x j ∂ xk
Hence, by substituting these equalities into the curl of (6.1.6), we obtain
∂ μ β3
z + z + [∇ z]u − [∇ u]z + | A1 (u)|2 z + β3 ∇ | A1 (u)|2 × curl ω
∂t α1 α1
∂
3 ∂ω ∂
− | A1 (u)|2 · +∇ | A1 (u)|2 × A1 (u)·, j
j=1

3
− (α1 + α2 ) [∇ ω]Δ u + ∇(Δ u j ) × A1 (u)·, j
j=1
3
∂ 2 u
3
∂u j ∂ 2 ω ∂u j
+2 +∇ ×
k=1 j=1
∂ xk ∂ x j ∂ xk ∂ xk ∂ x j ∂ xk
μ β3
= curl f + ω + | A1 (u)|2 ω. (6.2.6)
α1 α1
6.2 Variational Formulation. The Time-Dependent Problem 319
For any sufficiently smooth u, (6.2.6) can be viewed as a time-dependent transport

system in z of the form (7.3.23) with
μ β3
β =1, C = 1 + | A1 (u)|2 I − ∇ u,
α1 μ
and
μ β3
F = curl f + 1 + | A1 (u)|2 ω − β3 ∇ | A1 (u)|2 × curl ω
α1 μ
3
∂ ∂ω ∂
− | A1 (u)|2 · +∇ | A1 (u)|2 × A1 (u)·, j
j=1

3
+ (α1 + α2 ) [∇ ω]Δ u + ∇(Δ u j ) × A1 (u)·, j
j=1
3
∂ 2 u
3
∂u j ∂ 2 ω ∂u j
+2 +∇ × .
k=1 j=1
∂ xk ∂ x j ∂ xk ∂ xk ∂ x j ∂ xk
The matrix C meets the requirements of Proposition 7.3.9 provided u belongs

to the space L ∞ (0, T ; V ∩ W 1,∞ (Ω)3 ). But this is not enough, because for F to
be in L 1 (0, T ; L 2 (Ω)3 ) (see the last inequality in (7.3.32)), we need in addition u
in L 2 (0, T ; H 3 (Ω)3 ). In a domain with corners, this regularity cannot be deduced
from the fact that u is in L ∞ (0, T ; V α1 ). Indeed, according to Remark 7.3.32, u in
V α1 implies u in H 3 (Ω)3 if the class of ∂Ω is C 2,1 . Therefore, by this approach, the
analysis of the time-dependent grade-three fluid is done under the assumption that
∂Ω is of class C 2,1 . Then the following stability holds a.e. in ]0, T [, with a constant
C independent of u and t,
u(t) H 3 (Ω) ≤ Cz(t) L 2 (Ω) . (6.2.7)
6.3 A Priori Estimates
As expected, a basic a priori estimate is derived by testing (6.2.4) with v = u. In

the resulting expression, the factor of α1 + α2 simplifies by virtue of the following
formula that holds for all u ∈ V ∩ H 2 (Ω)3 :
3
∂u ∂u
c(Δ u; u, u) + c(u; u, Δ u) + 2 c ; ,u
k=1
∂ xk ∂ xk

1
=− tr A1 (u)3 . (6.3.1)
2 Ω
Furthermore, all u in V satisfy

1
∇u:∇u= A1 (u) : A1 (u).
Ω 2 Ω
Then (6.2.4) yields the energy equality

d
u(t)2α1 + μ| A1 (u(t))|2 + (α1 + α2 )tr A1 (u(t))3 + β3 | A1 (u(t))|4
dt Ω
= 2 f (t), u(t) , (6.3.2)
where · α is defined in (5.2.37), i.e., v2α = v2L 2 (Ω) + αv2H 1 (Ω) , and the duality

0
pairing in the right-hand side is taken between W −1, 3 (Ω)3 and W01,4 (Ω)3 .
4
The last part of the restriction (6.1.12) is used to bound the factor of α1 + α2 .
Indeed, we use the following extensions of Lemma 2.4.3; the proof of the first part
can be found in [101] and the proof of the second part in [33].
Lemma 6.3.1 Let A be any traceless symmetric matrix in IR 3×3 , let α, μ > 0 and
β > 0 be three real numbers. Then,
1
μtr( A2 ) + αtr( A3 ) + β| A|4 ≥ 0 ⇐⇒ |α| ≤ (24μ β) 2 . (6.3.3)
1
If in addition |α| < (24μ β) 2 , then
24μ β − α 2
μtr( A2 ) + αtr( A3 ) + β| A|4 ≥ tr( A2 ),
24β
1
(24μ β) 2 − |α| 3 24μ β − α 2
μtr( A2 ) + αtr( A3 ) + β| A|4 ≥ √ | A| ≥ √ | A|3 .
6 24 μ β
By applying (6.3.2) and Lemma 6.3.1, we easily derive the following a priori estimate.
Proposition 6.3.2 Let f be given in L 2 (Ω×]0, T [)3 and ut 0 in V α1 . If (6.1.12)
holds, then any solution u ∈ L ∞ (0, T ; V α1 ) of (6.2.4), (6.1.9) satisfies the following
bounds a.e. in ]0, T [:

u(t)2α1 ≤ et ut 0 2α1 + f 2L 2 (Ω×]0,t[) ,
12 β3
u(t)2α1 ≤ ut 0 2α1 + f 2L 2 (0,t;H −1 (Ω)3 ) ,
24μ β3 − (α1 + α2 )2
12 β3 S0,2
2
u(t)2α1 ≤ ut 0 2α1 + f 2L 2 (Ω×]0,t[) .
24μ β3 − (α1 + α2 )2
Furthermore, there exists a real number 0 < ε < β3 such that

1
|α1 + α2 | ≤ 24μ(β3 − ε) 2 ,
6.3 A Priori Estimates 321
and

u(t)2α1 + ε A(u)4L 4 (Ω×]0,t[) ≤ et ut 0 2α1 + f 2L 2 (Ω×]0,t[) , (6.3.4)
i.e., u is bounded in L 4 (0, T ; W 1,4 (Ω)3 ).
This last conclusion is an immediate consequence of Korn’s inequality (3.2.19) with

r = 4.
Now, we turn to estimates for z, assuming that ∂Ω is of class C 2,1 . Again, we set
ζ = z2L 2 (Ω) ,
and a formal argument readily yields that ζ solves the following Riccati differential
inequality:
μ β3 2
ζ (t) + ζ (t) − c1 ζ 2 (t) − c2 ζ (t) + | A1 (u)(t)|z(t) L 2 (Ω)
1
α1 α1
α1 μ β3
≤ 2 curl f (t) L 2 (Ω) + 2 curl u(t)2L 2 (Ω) + c0 A1 (u)(t)4L 4 (Ω) .
2
μ α1 α1
The argument becomes rigorous when applied to the Galerkin method on the spe-
cial basis of eigenfunctions defined by (5.2.27). It stems from an easy variant of
the material of Section 7.4, that, on its interval of existence, say [0, T ], ζ is
bounded by the data, more precisely by curl(ut 0 − α1 Δ ut 0 ) in L 2 (Ω)3 and f in
L 2 (0, T ; H (curl, Ω)). This implies the same bound for z.
Regarding the time derivative, by testing (6.2.4) with u , the time derivative of u,
we use the estimate for z and u to deduce an a priori estimate for u . For this, we use
the following equalities. The first one is straightforward, and the second one stems
from Green’s formula and the divergence-free constraint,
1 1 ∂ 1 d
(| A1 (u)|2 A1 (u), A1 (u )) = (| A1 (u)|2 , | A1 (u)|2 ) = A1 (u)4L 4 (Ω) ,
2 4 ∂t 8 dt
and

3
∂ u ∂ u
− c(Δ u; u, u ) + c(u ; u, Δ u) + 2 c ; ,u
∂ xk ∂ xk
k=1
3
∂ u ∂ u ∂ u ∂ u ∂ u ∂ u
= [∇ u] , + [∇ u] , + [∇ u ] ,
∂ xk ∂ xk ∂ xk ∂ xk ∂ xk ∂ xk
k=1
3
∂ u ∂ u ∂ u ∂ u
= ( A1 (u) , + [∇ u ] , .
k=1
Thus, we obtain
μ d β3 d
u (t)2α1 + u(t)2H 1 (Ω) + μ A1 (u(t))4L 4 (Ω) = −(z(t) × u(t), u (t))
2 dt 0 8 dt
3
∂u ∂ u ∂u ∂u
− (α1 + α2 ) {( A1 (u(t)) (t), (t)) + ([∇ u (t)] (t), (t))}
k=1
∂ x k ∂ x k ∂ x k ∂ xk
+ ( f (t), u (t)). (6.3.5)
As u is bounded in L ∞ (0, T ; V α1 ), whence in L ∞ (0, T ; W 1,4 (Ω 3 ), and z is

bounded in L ∞ (0, T ∗ ; L 2 (Ω)3 ), by integrating (6.3.5) over ]0, T ∗ [ we easily derive
that u is bounded in L 2 (0, T ∗ ; H 1 (Ω)3 ), in terms of ut 0 V α1 and f in L 2
(0, T ; H (curl, Ω)).
Finally, the isomorphism of Theorem 3.2.10 with r = 4 and the above bounds
4
for u, z and u , imply that p is bounded in L 2 (0, T ∗ ; L m3 (Ω)). The next theorem
summarizes these a priori bounds.
Theorem 6.3.3 Let Ω be a bounded, simply-connected open set of IR 3 with a bound-
ary ∂Ω of class C 2,1 and let f and ut 0 be given respectively in L 2 (0, T ; H (curl, Ω))
and V α1 . If (6.1.12) holds, then any solution (u, p) of (6.2.4)–(6.1.9) that exists on
the interval [0, T ], 0 < T ≤ T , satisfies
u L ∞ (0,T ∗ ;V α1 ) + p 4 ≤ C,
L 2 (0,T ∗ ;L m3 (Ω))
with a constant C that depends only on Ω, T , f L 2 (0,T ;H (curl,Ω)) , ut 0 V α1 , μ, α1 ,

α2 , and β3 .
6.4 Construction of Solutions by Galerkin’s Method
6.4.1 Existence
As in the case of the grade-two fluid, solutions can be constructed by discretizing

(6.2.4) by Galerkin’s method in the space Vm spanned by the first m eigenfunctions
that solve (5.2.27). This process leads to a square system of m nonlinear first-order
ordinary differential equations with constant
coefficients, with the exception of the
forcing term, that is L 2 in time: Find um = mj=1 c j,m (t) w j such that, for 1 ≤ j ≤ m,
(um (t), w j ) + α1 (∇ um (t), ∇ w j ) + μ(∇ um (t), ∇ w j )

+ (curl(um (t) − α Δ um (t)) × um (t), w j )

− (α1 + α2 ) (Δ um (t); um (t), w j ) + c(w j ; um (t), Δ um (t))
6.4 Construction of Solutions by Galerkin’s Method 323
3
∂ ∂
+ 2 (α1 + α2 ) c um (t); um (t), w j
k=1
∂ xk ∂ xk
β3
+ | A1 (um (t))|2 A1 (um (t)), A1 (w j ) = ( f (t), w j ), (6.4.1)
2
and
um (0) = Pm (ut 0 ), (6.4.2)
where again Pm denotes the orthogonal projection operator onto Vm for the scalar
product of V α1 . This system has a unique solution on a time interval, say Tm > 0;
this solution satisfies the same uniform a priori estimates as in Proposition 6.3.2,
sup um (t)2α1 ≤ C1 , A1 (um )4L 4 (Ω×]0,Tm∗ [) ≤ C2 , (6.4.3)

0≤t≤Tm∗
with constants C1 and C2 independent of m and Tm . These are obtained by multiplying

the jth equation of (6.4.1) with c j,m (t) and summing over j.
Next, by proceeding as in the proof of Theorem 5.2.9 (see Amrouche and Cio-
ranescu [8]), by setting
z m = curl(um − α1 Δ um ), ζm = z m 2L 2 (Ω) ,
it can be shown that the pair (um , z m ) satisfies the inequality (6.3.4)
μ β3 2
ζm (t) + ζm (t) − c1 ζm (t) − c2 ζm (t) + | A1 (um )(t)|z m (t) L 2 (Ω)
α1 α1
α1 μ
≤2 curl f (t) L 2 (Ω) + 2 curl um (t)2L 2 (Ω)
2
μ α1
β3
+ c0 A1 (um )(t)4L 4 (Ω) .
α1
Considering the regularity of the eigenfunctions established in Lemma 5.2.8, the

proof of this inequality is rigorous. According to (6.4.3), its right-hand side is
bounded, independently of m, by a function that belongs to L 1 (0, T ). Therefore,
in view of Theorem 7.4.1, the solution um exists on an interval, say [0, T ∗ ], with
T > 0 independent of m. Finally, a bound for the time derivative um can be derived
by multiplying the jth equation of (6.4.1) with cj,m (t) and summing over j.
These bounds are collected in the next theorem.
Theorem 6.4.1 Under the assumptions of Theorem 6.3.3, there exists a time T ∗ with
0 < T ∗ ≤ T , and constants C1 , C2 , C3 , depending only on T , f L 2 (0,T ;H (curl,Ω)) ,
ut 0 V α1 , μ, α1 , α2 , and β3 such that for all m, the Galerkin solution um of (6.4.1)–
(6.4.2) exists on the time interval [0, T ∗ ] and satisfies the uniform bounds,
um L ∞ (0,T ∗ ;V α1 ) ≤ C1 , z m L ∞ (0,T ∗ ;L 2 (Ω)3 ) ≤ C2 , um L 2 (0,T ∗ ;H 1 (Ω)3 ) ≤ C3 ;

(6.4.4)
as usual, z m = curl(um − α1 Δ um ).
Owing to the uniform bounds of (6.4.4), there exists a function
u ∈ L ∞ (0, T ∗ ; V α1 ) ∩ H 1 (0, T ∗ ; V )
such that, up to subsequences
lim um = u weakly * in L ∞ (0, T ∗ ; V α1 ),

m→∞
lim um = u weakly in L 2 (0, T ∗ ; V ).
m→∞
The first limit implies in particular that
lim um = u weakly in L 4 (0, T ∗ ; V α1 ).

m→∞
As V α1 is continuously imbedded into H 2 (Ω)3 (see Lemma 5.2.5), then its imbed-
ding is compact into W 1,4 (Ω)3 ∩ V , for instance. Therefore Theorem 3.3.14 applied
with X = V α1 , E = W 1,4 (Ω)3 ∩ V , Y = V , and q = 4 implies
lim um = u strongly in L 4 (0, T ∗ ; W 1,4 (Ω)3 ∩ V ).

m→∞
Hence, fixing j = j0 in (6.4.1) we can pass first to the limit as m tends to infinity and
next use the density of the basis in V α1 to recover (6.2.4), with test function v ∈ V α1 .
Again, by density the range of the test functions can be extended to V ∩ W 1,4 (Ω)3 .
In addition, (6.4.2) and the continuity of u in time imply that u satisfies the initial
condition (6.1.9).
As with the grade-two fluid, the pressure is recovered by Babuška–Brezzi’s The-
orem 3.2.12 part 2. Indeed, by examining (6.1.5) and using (6.2.3), we note that the
expression,

3
∂u ∂u

u − α1 Δ u − μ Δ u + z × u − (α1 + α2 ) A1 (u)Δ u + 2 [∇ ]
k=1
∂ xk ∂ xk

− β3 div | A1 (u)|2 A1 (u) − f ,
belongs to L 2 (0, T ∗ ; W −1, 3 (Ω)3 ), where W −1, 3 (Ω) is the dual space of W01,4 (Ω).
4 4
Hence we infer from Theorem 3.2.10 with r = 4 and the extension of the Babuška–
Brezzi theorem to reflexive Banach spaces, see [116, Chap. I, Remark 4.2], that there
exists a function p in L 2 (0, T ∗ ; L 3 (Ω)) such that the pair (u, p) verifies (6.2.4) for
4
1,4
all v in W0 (Ω)3 . This complements Theorem 6.3.3.
Theorem 6.4.2 Let Ω be a bounded, simply-connected open set of IR 3 with a bound-
ary ∂Ω of class C 2,1 and let f and ut 0 be given respectively in L 2 (0, T ; H (curl, Ω))
and V α1 . If (6.1.12) holds, then there exists an interval [0, T ], 0 < T ≤ T and a
solution (u, p) of (6.2.4)–(6.1.9) on this interval. It satisfies the bounds of Theorem
6.3.3,
u L ∞ (0,T ∗ ;V α1 ) + p 2 ∗ 43 ≤ C,
L (0,T ;L m (Ω))
with C depending only on Ω, T , f L 2 (0,T ;H (curl,Ω)) , ut 0 V α1 , μ, α1 , α2 , β3 .

Remark 6.4.3 As u ∈ L ∞ (0, T ∗ ; H 3 (Ω)3 ), considering the C 2,1 regularity of
∂Ω, it is easy to check that in fact, we can recover the pressure p in L 2 (0, T ∗ ;
L 2m (Ω)).
6.4.2 Uniqueness
For establishing uniqueness, it is convenient to use the equivalent equation (6.1.11),
∂
∂t

− (α1 + α2 ) Δ([∇ u]u) − 2[∇(Δ u)]u − β3 div(| A1 (u)|2 A1 (u)) + ∇ p = f .
In variational form, when multiplied with a smooth enough test function v in V , the
factor of α1 + α2 reads

− Δ([∇ u]u), v + 2 [∇(Δ u)]u, v = − Δ([∇ u]u), v + 2 c(u; Δ u, v).
By applying twice Green’s formula and the antisymmetry of c, the first term has the
following expression:

− Δ([∇ u]u), v = −c(u; u, Δ v) = c(u; Δ v, u),
whence the equivalent variational formulation for all smooth enough v in V ,
(u , v) + α1 (∇ u , ∇ v) + μ (∇ u, ∇ v) + (curl(u − (2α1 + α2 )Δ u) × u, v)
β
3
+ (α1 + α2 ) c(u; Δ v, u) + 2 c(u; Δ u, v) + | A1 (u)|2 A1 (u), A1 (v)
2
− ( p, div v) = ( f , v). (6.4.5)
As in the grade-two case, we consider any two solutions (u1 , p 1 ) and (u2 , p 2 ) of
(6.4.5)–(6.1.9) on an interval [0, T ], 0 < T ≤ T , set w = u1 − u2 , and test with
w the difference of the equations satisfied by these solutions. On one hand, (5.2.57)
immediately implies that

curl(w − (2α1 + α2 )Δ w) × u1 , w = (curl w × u1 , w)

− (2α1 + α2 ) c(u1 ; Δ w, w) − c(w; Δ w, u1 ) . (6.4.6)
On the other hand, we express the factor of α1 + α2 as follows:
c(u1 ;Δ w, u1 ) − c(u2 ; Δ w, u2 ) + 2 c(u1 ; Δ u1 , w) − 2 c(u2 ; Δ u2 , w)

= c(w; Δ w, u1 ) + c(u2 ; Δ w, w) + 2 c(u1 ; Δ w, w) + 2 c(w; Δ u2 , w)
= c(w; Δ w, u1 ) + c(u2 ; Δ w, w) + 2 c(u1 ; Δ w, w) − 2 c(w; w, Δ u2 ).
Thus, by combining this equality with (6.4.6) and substituting into (6.4.5) tested with
w, we obtain
1 d
w2α1 + μ |w|2H 1 (Ω) + (curl w × u1 , w) + (3α1 + 2α2 )c(w; Δ w, u1 )
2 dt
+ c α2 u1 + (α1 + α2 )u2 ; Δ w, w − 2(α1 + α2 )c(w; w, Δ u2 )
β3
+ | A1 (u1 )|2 A1 (u1 ) − | A1 (u2 )|2 A1 (u2 ), A1 (w) = 0. (6.4.7)
2
Here again, the nonlinear terms containing Δ w must be rewritten so that their upper
bounds involve only the H 1 norm of w. This is achieved by formulas (5.2.59) and
(5.2.60),
3
∂w ∂w ∂ u ∂w
c(w; Δ w, u) = c ; u, + c w; , ,
k=1
3
∂u ∂w
c(u; Δ w, w) = c ; w, .
k=1
∂ xk ∂ xk
By substituting these expressions into (6.4.7), we derive
1 d
w2α1 + μ w2H 1 (Ω) + (curl w × u1 , w)
2 dt 0
∂
3
∂w
+ c α2 u1 + (α1 + α2 )u2 ; w,
k=1
∂ xk ∂ xk
3
∂w 1 ∂w ∂ u2 ∂w
+ (3α1 + 2α2 ) c ;u , + c w; ,
k=1
β3
+ | A1 (u1 )|2 A1 (u1 ) − | A1 (u2 )|2 A1 (u2 ), A1 (w)
2
− 2(α1 + α2 )c(w; w, Δ u2 ) = 0, (6.4.8)
where we have used the antisymmetry (5.2.58) in the last term of the first line, so
that the second derivative acts only on u2 .
Finally, the last term in (6.4.8) is processed by monotony. Indeed, we use the fact
that any vectors U and V of IR satisfy
2 1 2 1
|U| U −|V |2 V · U − V = |U|2 −|V |2 + |U − V |2 |U|2 +|V |2 . (6.4.9)
2 2
Thus, we obtain the following uniqueness result.
Theorem 6.4.4 We retain the notation and assumptions of Theorem 6.4.2
and assume that problem (6.2.4)–(6.1.9) has a solution (u, p) with u in L 1 (0, T ∗ ;
W 2,3 (Ω)3 ). Then this problem has no other solution on [0, T ∗ ].
Proof From (6.4.8) and (6.4.9), we easily infer that
1 d
w2α1 + μ w2H 1 (Ω) ≤ w(t)2H 1 (Ω) S0,3 S0,6 u1 (t) H01 (Ω)
2 dt 0 0

+ |3α1 + 2α2 |∇ u (t) L (Ω) + ∇ α2 u1 (t) + (α1 + α2 )u2 (t) L ∞ (Ω)
1
∞
√
+ S0,6 |3α1 + 2α2 | α1 + 2 3(α1 + α2 ) |u2 (t)|W 2,3 (Ω) .
Then Gronwall’s lemma implies that w = 0.
6.5 The Steady-State Problem
The analysis we present here was developed by Busuioc and Iftimie in [57]. Their
assumptions are: Ω simply-connected with a smooth boundary; but we shall specify
further on the regularity that is required. When reaching a steady state, the grade-three
fluid’s equation of motion (6.1.5) reduces to

−μ Δ u + [∇ u]u − α2 div( A21 ) − α1 div [∇ A1 ]u + A1 L + L T A1
− β3 div(| A1 |2 A1 ) + ∇ p = f . (6.5.1)
6.5.1 A Remarkable Monotonicity
As stated in the introduction, the success of the analysis in [57] is based on the
monotonicity not only of the factor of β3 , but also of the full operator R(u) defined
as
R(u) := −μ Δ u − α1 div( A1 L + L T A1 ) − α2 div( A21 ) − β3 div(| A1 |2 A1 ).
Note that if u belongs to W 1,4 (Ω)3 , then R(u) is in W −1, 3 (Ω)3 , more precisely
4
R(u)W −1, 43 (Ω) ≤ μ |u|W 1, 43 (Ω) + 4(α1 + |α2 |)u H01 (Ω) |u|W 1,4 (Ω)
+ 8β3 |u|3W 1,4 (Ω) . (6.5.2)
We state below the monotonicity result in three dimensions.

Lemma 6.5.1 In addition to (6.1.12), let the coefficients of (6.5.1) satisfy
3α12 + 4(α1 + α2 )2 ≤ 24 μ β3 . (6.5.3)
Then the operator R(u) is monotone, i.e.,
∀u ∈ V ∩ W 1,4 (Ω)3 , ∀v ∈ V ∩ W 1,4 (Ω)3 , R(u) − R(v), u − v ≥ 0, (6.5.4)
where ·, · denotes the duality pairing between W01,4 (Ω)3 and W −1, 3 (Ω)3 .
4
Proof As the proof is long and technical, we only present its main steps. The reader
will find a complete proof in [57]. To simplify, we denote A1 (u) by A, A1 (v) by Ã,
and use a similarly simplification for L(u) and L(v).
1. First, by Green’s formula, the symmetry of A and Ã, and (6.4.9), we derive

2 R(u) − R(v), u − v = μ | A − Ã|2 + 2α1 ( AL − Ã L̃) : ( A − Ã)
Ω
2 β3 2 2
+ α2 A2 − Ã : ( A − Ã) + | A| − | Ã|2
2
β3 2
+ | A − Ã| | A| + | Ã| .
2 2
(6.5.5)
2
2. We introduce the matrices M = L − L̃, B = A + Ã, C = A − Ã = M + M T ,

D = M − M T , E = C M T + M C. After some manipulations (6.5.5) simplifies to

β3 1
2 R(u) − R(v), u − v = μ |C|2 + (C : B)2 + |C|2 (|C|2 + |B|2 )
Ω 2 2
α1 α1
+ ( + α2 )(C : B) + (E : B) .
2
(6.5.6)
2 2
3. Keeping C and M fixed, the integrand in (6.5.6) becomes a (real valued)
function of B only. We denote this functional by F(B),
β3 1
F(B) = μ |C|2 + (C : B)2 + |C|2 (|C|2 + |B|2 )
2α 2
α1
1
+ + α2 (C 2 : B) + (E : B). (6.5.7)
2 2
Now, we compute the minimum of F(B) with respect to B. This is the most
technical part of the proof. Note that B is symmetric and tr(B) = 0, but since the
minimum will be attained by a symmetric matrix, we only retain the constraint on
the trace and consider IR 9 with this constraint. As F(B) tends to infinity when |B|
tends to infinity, F(B) has a global minimum; let B 0 be such a minimum point. The
Lagrange multiplier method and some calculations yield,
6.5 The Steady-State Problem 329
4(α1 + α2 )tr(C 3 ) α1 + 2α2 2 α1 2(α1 + α2 )

B0 = C− C − E+ I, (6.5.8)
3β3 |C|4 β3 |C|2 β3 |C|2 3β3
which is indeed a symmetric matrix since C and E are symmetric. Moreover,

tr(B 0 ) = 0, considering that tr(C) = 0 and tr(E) = tr(C 2 ) = |C|2 . Then some
more manipulations yield the value of F(B 0 ),
(α1 + α2 )2
α 2 |C D − DC|2 β3
F(B 0 ) = μ − |C|2 − 1 + |C|4
6β3 16β3 |C|2 4

3 2
2(α1 + α2 ) tr(C )
2
+ . (6.5.9)
3β3 |C|4
4. Next, we deal with the second term in (6.5.9) and show that
|C D − DC|2 ≤ 2|C|2 | D|2 .
This is achieved by diagonalization and explicit computation. Then, by substituting

this inequality into (6.5.9), we infer that
(α1 + α2 )2
2 α2 β3
F(B 0 ) ≥ μ − |C| − 1 | D|2 + |C|4
6β3 8β3 4
2
2(α1 + α2 ) tr(C 3 )
2
+ . (6.5.10)
3β3 |C|4
Finally, since u and ũ vanish on the boundary and are both divergence-free
D L 2 (Ω) = C L 2 (Ω) .
Thus a combination of this equality with (6.5.6) and (6.5.10) yields
2 R(u) − R(v), u − v
2
(α1 + α2 )2 α12 2 β3 2(α1 + α2 )2 tr(C 3 )
≥ μ− − |C| + |C| +
4
,
Ω 6β3 8β3 4 3β3 |C|4
which, together with the assumptions (6.5.3), imply (6.5.4).
6.5.2 Existence of a Solution

6.5.2.1 Galerkin’s Method
Let {ϕ n }n≥1 be the basis of the eigenfunctions of the Stokes operator, i.e., ϕ n ∈ V
− Δ ϕ n + ∇ pn = λn ϕ n in Ω. (6.5.11)
It is well-known that this system defines a unique set of positive eigenvalues, and
the set of eigenfunctions {ϕ n }n≥1 is an orthogonal basis of V and an orthonormal
basis of H , where
H = {v ∈ Hτ (div, Ω); div v = 0}.
Corollary 3.2.16 and Theorem 3.2.14 guarantee that in a convex polyhedron (or a
3
domain with a smooth boundary), the functions ϕ n belong to H 2 (Ω) ∩ W 1,∞ (Ω) .
We discretize problem (6.5.1) by setting it into a variational formulation in the
space spanned by the first m elements of this basis
Vm = Vect[ϕ 1 , ϕ 2 , . . . ϕ m ].
Then Galerkin’s method reads: Find um ∈ Vm such that for all vm ∈ Vm

c(um ; um , vm ) + α1 um · ∇ A1 (um ), ∇ vm + (R(um ), vm ) = f , vm . (6.5.12)
3
As um and vm belong to H 2 (Ω) ∩ W 1,∞ (Ω) , all the terms in (6.5.12) make sense;
in particular R(um ) belongs to L 2 (Ω)3 .
Since (6.5.1) and (6.1.5) have the same space operator, by disregarding the time
derivative in the energy equality (6.3.2), we derive for all functions u in V ∩W 1,4 (Ω)3

1
R(u), u = μ| A1 (u)|2 + (α1 + α2 )tr A1 (u)3 + β3 | A1 (u)|4 . (6.5.13)
2 Ω
β3 β3 3β3
Thus, splitting 2
into 8
and 8
, we write, in view of (6.3.3) and assumption (6.5.3)

3β3
R(u), u ≥ | A1 (u)|4 , (6.5.14)
8 Ω
for all u in V ∩ W 1,4 (Ω)3 and this includes all vm in Vm . Therefore, owing to (6.5.14),
the antisymmetry of c, and the fact that
(um · ∇ A1 (um ), ∇ um ) = (um · ∇ A1 (um ), A1 (um )) = 0, (6.5.15)
it is easy to prove by Brouwer’s Fixed Point Theorem, that problem (6.5.12) has at
least one solution.
Now, consider any solution um of (6.5.12). The choice vm = um in (6.5.12) and
the above arguments yield

R(um ), um = f , um . (6.5.16)
Therefore by using Korn’s inequality (3.2.19) in W01,4 (Ω)3

1
|v|W 1,4 (Ω) ≤ C4,κ A1 (v) L 4 (Ω) ,
2
we derive that any solution um of (6.5.12) is uniformly bounded in this space,
C 13
4,κ
|um |W 1,4 (Ω) ≤ f W −1, 43 (Ω) . (6.5.17)
6β3
In turn, with (6.5.2), this implies that R(um ) is uniformly bounded in W −1, 3 (Ω)3 :
4
There exists a constant C, independent of m, such that
R(um )W −1, 43 (Ω) ≤ C. (6.5.18)
In view of (6.5.17) and (6.5.18), there exist a function u in W01,4 (Ω)3 ∩ V and a
distribution ζ in W −1, 3 (Ω)3 , such that, up to subsequences,
4
lim um = u weakly in W 1,4 (Ω)3 ,

m→∞
weakly in W −1, 3 (Ω)3 .

4
lim R(um ) = ζ
m→∞
By the compact imbedding in (3.1.3), we also have that, up to another subsequence,
lim um = u strongly in L ∞ (Ω)3 .

m→∞
6.5.2.2 Passing to the Limit
Passing to the limit in (6.5.12) is difficult because the above convergences are not
sufficient to take the limit of the nonlinear terms in R(um ). As a first step, we can
derive the following intermediate result:
Proposition 6.5.2 In a convex polyhedron, or a connected domain with a smooth
boundary, there exists a distribution p in H −1 (Ω)/IR such that the triple (u, ζ , p)
satisfies in H −2 (Ω)3

[∇ u]u − α1 div u · ∇ A1 (u) + ζ − ∇ p = f . (6.5.19)
Proof By choosing an arbitrary integer j0 ≤ m, we can let m tend to infinity in

(6.5.12). On one hand,
lim c(um ; um , v j0 ) = c(u; u, v j0 ).

m→∞
On the other hand, considering that
3
∂
um · ∇ A1 (um ), ∇ v j0 = − A1 (um ), u km ∇ v j0 ,
k=1
∂ x k
since um belongs to V , the weak convergence of A1 (um ) to A1 (u) in L 4 (Ω)3×3 , and

the strong convergence of um to u in L ∞ (Ω)3 imply that
3
∂
lim um · ∇ A1 (um ), ∇ v j0 = − A1 (u), uk ∇ v j0
m→∞
k=1
∂ xk
3
∂
= A1 (u), u k ∇ v j0 .
k=1
∂ x k
Therefore, the limit of (6.5.12) reads, for all j ≥ 1,
3
∂
c(u; u, v j ) + α1 A1 (u), u k ∇ v j + ζ , v j = f , v j .
k=1
∂ xk
By density, this implies that for all v in H 2 (Ω)3 ∩ V ,
3
∂
c(u; u, v) + α1 A1 (u), u k ∇ v + ζ , v = f , v .
k=1
∂ xk
In particular, for all v in H02 (Ω)3 ∩ V

c(u; u, v) − α1 div(u · ∇ A1 (u)), v + ζ , v = f , v . (6.5.20)
In (6.5.20), the dominating distribution belongs at least to H −2 (Ω)3 . Therefore, the

quantity
[∇ u]u − α1 div(u · ∇ A1 (u)) + ζ − f
belongs to H −2 (Ω)3 and vanishes when applied to functions of H 2 (Ω)3 ∩V , i.e., with
the notation of (3.2.25), it belongs to P V2,2 (Ω). Hence, we deduce form Corollary
3.2.20 applied with m = 1 and r = 2, that there exists p in H −1 (Ω)/IR such that
(u, ζ , − p) verifies (6.5.19).
Now, the difficulty lies in proving that ζ = R(u). This is done in [57] by a
monotonicity argument in which the most delicate part consists in showing the fol-
lowing limit that one expects for (6.5.16).
Lemma 6.5.3 In addition to the assumptions of Proposition 6.5.2, we suppose that

Ω is simply-connected. Then the limit functions ζ and u satisfy
ζ , u = f , u . (6.5.21)
Proof The proof is long and technical because the duality pairing of (6.5.19) with
u is not defined. This situation is reminiscent of the transport equation treated in
the Appendix, Section 7.2.1, but considering the highly irregular factor of α1 (which
is not integrable), it is not clear that the analysis developed there directly applies.
The strategy of [57] consists in taking the duality product of (6.5.19) with a good
regularization uε ∈ V of u. Thus the proof proceeds in four steps.
1. The idea is to construct uε by truncation and convolution with a classical
mollifier; but we have seen in Sections 5.5.1 and 5.6.2 that the zero divergence can
only be preserved by truncating a vector potential ψ ∈ W02,4 (Ω)3 ,
u = curl ψ in Ω. (6.5.22)
In a simply-connected Lipschitz domain, such a vector potential can be con-

structed by a variant of the material of Section 5.2.2.
Let us sketch the proof. First, u is extended by zero to IR 3 ; the extended function,
say ū belongs to W 1,4 (IR 3 )3 , has compact support, and is divergence-free. Thus,
a preliminary divergence-free vector potential ψ̄ can be constructed by solving a
Laplace equation in IR 3 ; this gives
ū = curl ψ̄ in IR 3 ,
with ψ̄ ∈ W 2,4 (B)3 in any large ball B such that Ω B. As Ω is simply-connected,

B \ Ω is also simply-connected, and hence, since
curl ψ̄ = 0 in B \ Ω,
there exists a function q ∈ W 3,4 (B \ Ω) such that
ψ̄ = ∇ q in B \ Ω.
This function q has a continuous extension in W 3,4 (Ω), so that the extended function
q̄ belongs to W 3,4 (B). By construction, the function
ψ = ψ̄ − ∇ q̄,
solves (6.5.22) and the global regularity of ψ̄ and q̄ in B guarantees that ψ belongs
to W02,4 (Ω)3 .
2. For truncating ψ, we introduce tubular neighborhoods of ∂Ω for η sufficiently

small, as follows
Ωη = {x ∈ Ω; Δ(x) ≤ η}, (6.5.23)
based on Stein’s regularized distance function Δ(x) ∈ C ∞ (Ω) that satisfies (5.5.10)
and (5.5.11). Then, for any small enough ε > 0, ψ is multiplied by a cutoff function
h ε ∈ D(Ω), 0 ≤ h(x) ≤ 1, such that
h ε (x) = 0 in Ω2ε , h ε (x) = 1 in Ω \ Ω3ε .
Owing to (5.5.10) and (5.5.11), h ε satisfies for any derivative D k
Ck
D k h ε L ∞ (Ω) ≤ . (6.5.24)
ε|k|
Finally, the product h ε ψ is regularized by convolution with the standard mollifier ρε

defined by
1 x
ρε (x) = 3 ρ ,
ε ε
as usual ρ ∈ D(IR ), ρ ≥ 0 is supported by B(0; 1), and ρ is normalized so

3
where
that IR 3 ρ = 1. Then uε is defined by

uε = curl ρε (h ε ψ) . (6.5.25)
3. Now we can take the duality product of both sides of (6.5.19) with uε . By virtue
of the regularity of uε , this yields

3
∂
c(u; u, uε ) + α1 uk A1 (u), ∇ uε + ζ , uε = f , uε . (6.5.26)
k=1
∂ xk
The limit as ε tends to zero of all terms except the factor of α1 is straightforward,
and all the efforts in this proof are now concentrated on this last limit.
4. First, we observe that

3
∂ 1 3
k ∂
uk A1 (u), ∇ uε = u A1 (u), A1 (uε ) . (6.5.27)
k=1
∂ xk 2 k=1 ∂ xk
Therefore, if u k ∂∂xk A1 (u) were in a suitable L p space, the limit would follow from an
easy variant of the material of Section 7.2.1. But such regularity cannot be inferred
from (6.5.19). This difficulty is mastered in [57] by taking full advantage of the
properties of the truncation and regularization operators. Here is a sketch of the
proof; the full details can be found in [57]. By expanding (6.5.25) and using (6.5.22),

A1 uε = ρε h ε A1 (u) + ρε u ⊗ ∇ h ε + ∇ h ε ⊗ u + ρε B,
where
∂ ∂ψ ∂ ∂ψ
Bi, j = (∇ hε ) × ψ + ∇ hε × + (∇ hε ) × ψ + ∇ hε × .
∂x j ∂x j i ∂ xi ∂ xi j
All terms in this expansion, except the first one, involve derivatives of the cutoff
function h ε . By using their support, (6.5.24), Green’s formula, Hardy’s inequality
(3.1.11), and arguing as in Proposition 5.6.3, it can be readily proved that they are
bounded by a factor involving a positive power of ε. Therefore their contributions
tend to zero. It remains to investigate the limit of the first term, in (6.5.27). To simplify,
we set Z = A1 (u) ∈ L 4 (Ω)3×3 . Then it is easily seen that

3
∂ 1 k ∂3

uk
A1 (u), ∇ uε = u Z, ρε (h ε Z) = Iε .
k=1
∂ xk 2 k=1 ∂ xk
Without changing notation, let u stand for the extension of u by zero outside Ω, and
let Z̃ denote the extension of Z by zero outside Ω. Note that we still have Z̃ = A1 (u).
It is easy to check that Iε also has the expressions
1 k ∂
3
Iε = u Z̃, ρε (h ε Z̃)
2 k=1 ∂ xk
3
1 ∂
= ρε u k Z̃ : h ε Z̃,
2 k=1 IR 3 ∂ xk
where now the dualities are taken in IR 3 . Then we write

1 ∂ ∂
3
Iε = uk ρε Z̃ : h ε Z̃ + ρε u k Z̃ : h ε Z̃
2 k=1 ∂ xk ∂ xk
IR 3
∂
− uk ρε Z̃ : h ε Z̃
∂ xk
3
1 k ∂ ∂
= u ρε (h ε Z̃) : Z̃ + u k h ε ρε ) Z̃ : Z̃
2 k=1 IR 3 ∂ xk ∂ xk
∂ ∂ ∂
− uk ρε (h ε Z̃) : Z̃ + ρε u k Z̃ : h ε Z̃ − u k ρε Z̃ : h ε Z̃ .
∂ xk ∂ xk ∂ xk
Interestingly, the argument for treating the last difference above is a variant of that
of Corollary 7.2.4, adapted to other exponents. Considering that u is in W 1,4 (IR 3 )3
and Z̃ in L 4 (IR 3 )3 , it can be proved that
∂ ∂
lim ρε u k Z̃ − u k ρε Z̃ L 2 (IR 3 ) = 0. (6.5.28)
ε→0 ∂ xk ∂ xk
Next, by taking into account the support of h ε , the second difference can also be
bounded by a positive power of ε. Therefore we are left with the first term, which
reduces to an integral in Ω,

1 ∂ 1 k ∂
3 3
uk ρε (h ε Z̃) : Z̃ = − u A1 (u), ρε (h ε A1 (u) = −Iε .
2 k=1 IR 3 ∂ xk 2 k=1 ∂ xk
From here we easily obtain that limε→0 Iε = 0, whence (6.5.21).
Remark 6.5.4 The above proof also establishes that
3
∂
lim u k A1 (u), ∇ uε = 0. (6.5.29)
ε=0
k=1
∂ xk

It remains to prove that ζ = R(u).
Lemma 6.5.5 Under the assumptions and notation of Lemma 6.5.3, we have
ζ = R(u). (6.5.30)
Proof The proof proceeds in two steps.

1. Let us show the following preliminary inequality:

ζ − R(ϕ), u − ϕ ≥ 0, ∀ϕ ∈ V ∩ W 1,4 (Ω)3 , (6.5.31)
where ·, · stands for the duality pairing between W −1, 3 (Ω)3 and W01,4 (Ω)3 . For
4
any integer m ≥ 1, the expression in the left-hand side can be split into
ζ − R(ϕ), u − ϕ = R(um ) − R(ϕ), um − ϕ + R(um ) − ζ , ϕ

+ R(ϕ), um − u + ζ , u − R(um ), um ,
where um is a solution of (6.5.12). Note that the last term in the above right-hand
side coincides with f , um according to (6.5.16); the first term is positive owing
to the monotonicity of R; the second, third, and last terms tend to zero as m tends
to infinity in view of the weak convergences of R(um ) in W −1, 3 (Ω)3 and um in
4
1,4
W0 (Ω)3 , and in view of (6.5.21). This implies (6.5.31).
2. The second step is a simplified Minty argument when dealing with monotone
operators, see for instance [162] and [191]. In (6.5.31), we choose ϕ = u + λψ, for
any λ > 0 and any ψ in V ∩ W 1,4 (Ω)3 . This gives
ζ − R(u + λψ), ψ ≤ 0. (6.5.32)
For fixed u and ψ, the mapping λ → R(u + λψ) is a cubic polynomial and hence
is continuous. Thus, letting λ tend to zero in (6.5.32) yields
∀ψ ∈ V ∩ W 1,4 (Ω)3 , ζ − R(u), ψ ≤ 0.
Therefore (6.5.30) follows by changing ψ into −ψ.
The next existence theorem collects and complements these results.
Theorem 6.5.6 Let Ω be a convex polyhedron or a simply-connected domain with a

smooth boundary, and suppose that the coefficients of (6.5.1) satisfy μ > 0, β > 0,
and (6.5.3). Then for any data f in W −1, 3 (Ω)3 , problem (6.5.1) has at least one
4
solution (u, p) and all solutions satisfies the energy equality,

μ | A1 (u)|2 + (α1 + α2 )tr( A1 (u))3 + β3 | A1 (u)|4 = 2 f , u , (6.5.33)

Ω
where ·, · stands for the duality pairing between W −1, 3 (Ω)3 and W01,4 (Ω)3 .
4
Proof Existence follows by substituting (6.5.30) into (6.5.19). The energy equality
is an immediate consequence of (6.5.30), (6.5.21), and (6.5.13).
6.5.3 Uniqueness
Under fairly mild conditions, uniqueness is established in [57]. The reason why these
conditions are invoked is readily seen by considering the equation satisfied by the
difference of two solutions. Indeed, let (u1 , p 1 ) and (u2 , p 2 ) be two solutions of
(6.5.1), and let
w = u1 − u2 , q = p 1 − p 2 .
These functions satisfy in H −2 (Ω)3 ,

−α1 div u1 · ∇ A1 (w) − α1 div w · ∇ A1 (u2 ) + [∇ w]u1 + [∇ u2 ]u
+ R(u1 ) − R(u2 ) + ∇ q = 0. (6.5.34)
When tested with w, the first term in (6.5.3) will vanish as expected, but the trouble
arises from the second term that cannot be bounded if u2 has no more regularity
4
than W 1,4 (Ω). We shall see below that it is sufficient that u2 belong to W 2, 3 (Ω) ∩
3
W 1,4 (Ω) .
4 3
(6.5.1) has a solution u in V ∩ W 2, 3 (Ω) ∩ W 1,4 (Ω) with sufficiently small norm,
uW 1,4 (Ω) + uW 2, 43 (Ω) ≤ M, (6.5.35)
where M is a constant that depends only on Ω and the parameters of (6.5.1). Then
(6.5.1) has no other solution in V ∩ W 1,4 (Ω)3 .
4
Proof With the above notation, let u2 be a solution of (6.5.1) in V ∩ W 2, 3 (Ω) ∩
3
W 1,4 (Ω) . For small enough ε, let us take the duality pairing of both sides of (6.5.3)
with wε , the regularization of w defined by (6.5.25),

3
∂ 3
∂
α1 (u1 )k A1 (w), ∇ wε + α1 wk A1 (u2 ), ∇ wε
k=1
∂ xk k=1
∂ xk
+ c(u1 ; w, wε ) + c(w; u2 , wε ) + R(u1 ) − R(u2 ), wε = 0. (6.5.36)
As stated in Remark 6.5.4, the first term in (6.5.36) tends to zero with ε. By assump-
tion, each coefficient of the product wk ∂∂xk A1 (u2 ) belongs to L 3 (Ω). As all coeffi-
4
cients of the other factor converges in L 4 (Ω), it follows that

3
∂ 3
k ∂
lim wk A1 (u2 ), ∇ wε = w A1 (u2 ), ∇ w .
ε→0
k=1
∂ xk k=1
∂ xk
The other limits are straightforward. Hence, by taking the limit of both sides of
(6.5.36), we obtain

3
∂
R(u1 ) − R(u2 ), w = −α1 wk A1 (u2 ), ∇ w − c(w; u2 , w).
k=1
∂ xk
Then the uniqueness follows from the monotonicity of R and the assumptions (see
4 3
(6.5.14)) on the parameters, provided u2 belongs to W 2, 3 (Ω) ∩ W 1,4 (Ω) with a
sufficiently small norm.
Chapter 7
Appendix
This last chapter is devoted to the proofs of auxiliary results, with particular emphasis
on properties of simple linear transport equations that lie at the core of many complex
fluids.
7.1 Korn’s Inequality in Lr
The following proof of Korn’s inequality in L r (Ω) is based on a well-known result

of Nečas [194] (see Bramble [47], Duvaut and Lions [91], and Wang [281] for
similar results). Theorem 7.1.1 is established by Nečas in a Lipschitz domain, with
a direct proof. However, it also follows from Theorem 3.2.10 and the extension of
Theorem 3.2.12 to reflexive Banach spaces.
Theorem 7.1.1 Let Ω be a bounded, connected domain that is a finite union of open
sets with Lipschitz-continuous boundaries. Then, for any real number r, 1 < r <
+∞, there exists a constant cr∗ depending only on r and Ω such that

∀p ∈ L r (Ω), pLr (Ω) ≤ cr∗ pW −1,r (Ω) + ∇ pW −1,r (Ω) . (7.1.1)
r
Proof When p belongs to Lm (Ω), it follows from (3.2.17) that
1
pLr (Ω) ≤ ∇ pW −1,r (Ω) . (7.1.2)
Kr
Otherwise, we use the expansion

1
p(x) = p̃(x) + p(y) dy,
|Ω| Ω
r
so that p̃ belongs to Lm (Ω).

DOI 10.1007/978-3-319-39330-8_7
340 7 Appendix
We must now prove that there exists a constant γr such that

1
p(y) dy ≤ γr pW −1,r (Ω) + ∇ pW −1,r (Ω) , (7.1.3)
|Ω| Ω
and to established it we will argue by contradiction. If (7.1.3) does not hold, we can
construct a sequence (pn )n in L r (Ω) such that
lim pn W −1,r (Ω) = 0, lim ∇ pn W −1,r (Ω) = 0, (7.1.4)

n→∞ n→∞
and
1
pn (y) dy = 1. (7.1.5)
|Ω| Ω

Then, setting p̃n (x) = pn (x) − |Ω|
1
Ω pn (y) dy, we deduce from (7.1.2) and the
second part of (7.1.4) that the sequence (p̃n )n converges to zero in L r (Ω). In turn,
the first part of (7.1.4) yields

1
lim pn (y) dy = 0,
n→∞ |Ω| Ω
thus contradicting (7.1.5).
In view of this result, the proof of Korn’s inequality is an easy consequence of the
following identity, see [91]:
∂ 2 vi ∂ ∂ ∂
= εi,k (v) + ε i,j (v) − ε j,k (v). (7.1.6)
∂xj ∂xk ∂xj ∂xk ∂xi
Corollary 7.1.2 Let Ω be a bounded, connected domain that is a finite union of

open sets with Lipschitz-continuous boundaries. Then, for any real number r, 1 <
r < +∞, there exists a constant Cr,κ , depending only on r and Ω, such that (3.2.19)
holds
∀v ∈ W01,r (Ω)d , |v|W 1,r (Ω) ≤ Cr,κ ε(v)Lr (Ω) .
∂vi
Proof Let v belong to W 1,r (Ω)d , set w = ∂xj
, and apply (7.1.1) to w,

wLr (Ω) ≤ C1 wW −1,r (Ω) + ∇ wW −1,r (Ω) ,
i.e.,
∂vi
r ≤ C2 vi Lr (Ω) + ∇ wW −1,r (Ω) .
∂xj L (Ω)
7.1 Korn’s Inequality in L r 341
By substituting (7.1.6) into the last term above and collecting terms we obtain the
intermediate inequality valid for all v in W 1,r (Ω)d

|v|W 1,r (Ω) ≤ C3 vLr (Ω) + ε(v)Lr (Ω) . (7.1.7)
When v belongs to W01,r (Ω)d , (3.2.19) follows readily from (7.1.7) by a contradiction
and a compactness argument.
The argument in the above proof is the same as in [281], but the assumptions here
are weaker because the assumptions of Theorem 7.1.1 are weaker than in [281].
7.2 The Steady Transport Equation
7.2.1 Some L2 Results on the Scalar Transport Equation for

Steady Flow
The material in this section is taken essentially from [117]. We propose to solve the
scalar transport equation for steady flow in a bounded Lipschitz-continuous domain
Ω in arbitrary dimension d: for a given parameter α = 0, for f given in L 2 (Ω) and
u given in W (see (3.1.18)), find z in L 2 (Ω) solution of
z + α u · ∇ z = f in Ω , (7.2.1)
where u · ∇ z is defined by (5.6.5),

d
∂z
u·∇z = ui ,
i=1
∂xi
The restriction α = 0 is imposed because otherwise (7.2.1) is trivial. As f belongs

to L 2 (Ω), any solution z ∈ L 2 (Ω) of (7.2.1) satisfies u · ∇ z ∈ L 2 (Ω). This suggests
introducing the space, for fixed u in W ,
Xu = {z ∈ L 2 (Ω) ; u · ∇ z ∈ L 2 (Ω)} , (7.2.2)
which is a Hilbert space equipped with the norm

21
zu = z2L2 (Ω) + u · ∇ z2L2 (Ω) . (7.2.3)
An easy argument based on a Galerkin approximation shows that problem (7.2.1)

has at least one solution. It relies on Green’s formula (5.2.43).
342 7 Appendix
Proposition 7.2.1 Let Ω ⊂ IRd be bounded and Lipschitz-continuous. For all u in

W , all f in L 2 (Ω) and all real numbers α = 0, the transport equation (7.2.1) has at
least one solution z in Xu and this solution z satisfies
zL2 (Ω) ≤ f L2 (Ω) . (7.2.4)
approximate z by a Galerkin method in a basis (wi )i≥1 of H (Ω) for

Proof Let us
> 2 : zm = m
d
i=1 ζi wi , such that
∀1 ≤ i ≤ m , (zm , wi ) + α(u · ∇ zm , wi ) = (f , wi ). (7.2.5)
Owing to the regularity of the test functions wi , we can use (5.2.43) to show that
(7.2.5) has a unique solution zm and
(f , zm ) = zm 2L2 (Ω) , (7.2.6)
thus yielding immediately that zm satisfies (7.2.4). Therefore, we can extract from
{zm } a subsequence (still denoted by the index m) that converges weakly in L 2 (Ω)
and, passing to the limit in (7.2.5), it is easy to check that the limit function satisfies
(7.2.1).
This existence result is well-known and holds in more general situations, for example,
when u belongs to H(div, Ω), div u = 0, and u · n = 0. The key requirement is the
integrability of the product z(u · ∇ϕ) for appropriate test functions ϕ.
Whereas constructing a solution of (7.2.1) is easy, establishing uniqueness of the
solution when the boundary of the domain is not smooth and the driving velocity
u has little regularity is far from trivial. Indeed, as (7.2.1) is a linear equation, it is
tempting to take the product of both sides of (7.2.1) with z when f = 0, integrate
in space and apply Green’s formula (5.2.43) to the integral of (u · ∇ z)z. However,
this formula is valid if z belongs to H 1 (Ω) and in order to extend it to z ∈ Xu , we
must prove that H 1 (Ω) is dense in Xu . This is a difficult result for which we have
no direct proof. Instead we propose the following strategy developed by Girault and
Scott in [117] for a scalar transport equation:
1. First, we prove that smooth functions are dense in the space
{z ∈ L 2 (Ω) ; u · ∇ z ∈ L 1 (Ω)} ,
where u is a fixed, arbitrary function in H 1 (Ω)d .

2. Next, we prove uniqueness of the solution of (7.2.1) in L 1 (Ω) by means of the
Di Perna and Lions renormalizing argument, see Theorem II.2 in [82] (cf. also
Desjardins [81]).
3. Then, using this uniqueness, we establish Green’s formula for all z in Xu , with
u ∈ W.
4. Finally, we prove that D(Ω) is dense in Xu , for u ∈ W .
7.2 The Steady Transport Equation 343
The proof of our first density result is delicate because ∂Ω is not smooth and u does
not vanish on ∂Ω. Indeed, a usual density proof relies on extending functions outside
the domain, while retaining the structure of the space, and regularizing the extended
functions by convolution with mollifiers. However, a standard extension does not
guarantee that the extended functions belong to Xu in IR. Therefore, following Puel
and Roptin [220], we shall not extend functions outside the domain, and instead use
convolution in the domain with a suitable family of mollifiers. Beforehand, we need
the following lemma.
Lemma 7.2.2 Let Ω be a bounded Lipschitz-continuous domain of IRd ; then Ω has

a finite open covering,
N
Ω⊂ Or ,
r=1
with the following property: for each r with 1 ≤ r ≤ N, there exists a nonzero vector
t r of IRd and a number δr > 0 such that, for all 0 < ε ≤ 1 and for all x in Ω ∩ Or ,
B(x; εδr ) + εt r ⊂ Ω ,
where B(x; σ ) denotes the ball with center x and radius σ .
Proof The proof follows readily from the uniform cone property, which holds
because Ω is a Lipschitz-continuous domain (cf. Adams [3]). Indeed, the uniform
cone property states that Ω has a finite open covering as above with which we can
associate a cone Cr , such that for every x in Ω ∩ Or , the cone Cr with apex x is
contained in Ω. Thus, for fixed r with 1 ≤ r ≤ N, we denote by αr the angle of the
cone Cr , by er the direction of its axis and by hr its height. Then, from the equation
of the sphere inscribed in Cr , it is easy to check that we can take
sin( α2r ) hr
δr = hr , tr = er ,
1 + sin( α2r ) 1 + sin( α2r )
and this completes the proof.
For the sequel, it is convenient to extend z by zero outside Ω, so that the extended
function (still denoted z) belongs to L 2 (IRd ) without changing its norm. Similarly,
we extend u continuously outside Ω so that the extended function (still denoted u)
belongs to H 1 (IRd )d . We retain the notation of Lemma 7.2.2. For each index r, with
1 ≤ r ≤ N we set Ωr = Ω ∩ Or , and we choose a mollifier ρr ∈ D(IRd ) such
that 0 ≤ ρr (x) ≤ 1 for all x ∈ IRd , supp ρr ⊂ B(0; δr ), and IRd ρr (x) dx = 1.
Following [220], we define for all ε ∈]0, 1],
1 x
∀x ∈ Or , ρε,r (x) = ρr + tr . (7.2.7)

ε d ε
344 7 Appendix
The use of the mollifier ρε,r is justified by the fact that for all x ∈ Ωr and for all
y ∈ B(0; δr ), x − ε(y − t r ) belongs to Ω. This allows one to define formally the
convolution z ρε,r in Ωr for any function z defined in Ω and we have

∀x ∈ Ωr , z ρε,r (x) = (−1)d z(x − ε(y − t r ))ρr (y) dy. (7.2.8)
B(0;δr )
Thus the convolution with ρε,r regularizes z in Ωr without using values of z outside
Ωr . Furthermore, if z ∈ L 1 (Ω), then z ρε,r ∈ L 1 (Ωr ) and
z ρε,r L1 (Ωr ) ≤ zL1 (Ω) . (7.2.9)
The next lemma proves a “commuting” property of convolution with ρε,r .
Lemma 7.2.3 Let Ω be a bounded Lipschitz-continuous domain of IRd , let u be

given in H 1 (Ω)d , and let z be given in L 2 (Ω). With the notation of Lemma 7.2.2,
there exists a constant C independent of z and u such that for all r with 1 ≤ r ≤ N
and all ε ∈]0, 1], we have
u · ∇(z ρε,r ) − (u · ∇ z) ρε,r L1 (Ωr ) ≤ C∇ uL2 (Ω) zL2 (Ω) . (7.2.10)
Proof The argument adapts the proof written in [220] to a domain with a Lipschitz-
continuous boundary and a function u that only belongs to H 1 . For x ∈ Ωr , let us
evaluate the quantity

Aε,r (x) = u · ∇(z ρε,r ) − (u · ∇ z) ρε,r (x).
Expanding, we have

Aε,r (x) = u(x) · ∇x z(y)ρε,r (x − y) dy − u(y) · ∇y z(y), ρε,r (x − y)D(IRd ) .
IRd
But
u(y) · ∇y z(y),ρε,r (x − y)D(IRd ) = ∇y z(y), u(y)ρε,r (x − y)H 1 (IRd )

=− z(y)divy (u(y)ρε,r (x − y)) dy
d
IR
=− z(y)u(y) · ∇y ρε,r (x − y) dy − z(y)ρε,r (x − y)divy u(y) dy.
IRd IRd
Hence

Aε,r (x) = z(y)[u(x) · ∇x ρε,r (x − y) + u(y) · ∇y ρε,r (x − y)] dy
d
IR
+ z(y)ρε,r (x − y)divy u(y) dy
IRd

1 x−y
= d+1 z(y)(u(x) − u(y)) · ∇ ρr ( + t r ) dy
ε IRd ε

+ z(y)ρε,r (x − y)divy u(y) dy. (7.2.11)
IRd
Thus
Aε,r (x) = a + b ,
where

1
a = (−1)d z(x − ε(y − t r )) [u(x) − u(x − ε(y − t r ))] · ∇ ρr (y) dy ,
B(0;δr ) ε

b= z(y)ρε,r (x − y)divy u(y) dy.
IRd
Observe that
1
1
[u(x) − u(x − ε(y − t r ))] = ∇ u(x − (1 − θ )ε(y − t r )) · (y − t r ) dθ.
ε 0
Substituting into (7.2.11), taking the absolute value, integrating with respect to x in
Ωr , and applying Fubini’s theorem, we can find a constant C, independent of r, u
and z, such that
aL1 (Ωr ) ≤ C∇ uL2 (Ω) zL2 (Ω) . (7.2.12)
Similarly, (7.2.9) yields
bL1 (Ωr ) ≤ z(div u)L1 (Ω) ≤ div uL2 (Ω) zL2 (Ω) ,
and (7.2.10) follows from (7.2.12) and this inequality.
The next corollary is an extension of Friedrichs’ Lemma (cf. Hörmander [133]).
Corollary 7.2.4 Under the assumptions of Lemma 7.2.3, for all r with 1 ≤ r ≤ N,
lim u · ∇(z ρε,r ) − (u · ∇ z) ρε,r L1 (Ωr ) = 0. (7.2.13)

ε→0
346 7 Appendix
Proof Let {z } ⊂ D(Ω) be a sequence converging to z in L 2 (Ω). Let us write
u · ∇(z ρε,r )− (u · ∇ z) ρε,r L1 (Ωr ) ≤ u · ∇(z ρε,r ) − (u · ∇ z ) ρε,r L1 (Ωr )
+ u · ∇((z − z ) ρε,r ) − (u · ∇(z − z )) ρε,r L1 (Ωr ) .
The first term tends to zero with ε because z is smooth and the second term tends
to zero by virtue of (7.2.10) that holds uniformly with respect to r and ε.
Now, we are in a position to prove that smooth functions are dense in
{z ∈ L 2 (Ω) ; u · ∇ z ∈ L 1 (Ω)}.
Theorem 7.2.5 Let Ω be a bounded Lipschitz-continuous domain of IRd and let u

be given in H 1 (Ω)d . Then for each z in L 2 (Ω) such that u · ∇ z belongs to L 1 (Ω)
(e.g., if z ∈ Xu ), there exists a sequence {zk }k≥1 of functions zk ∈ D(Ω) such that
lim zk = z in L 2 (Ω) , lim u · ∇ zk = u · ∇ z in L 1 (Ω).

k→∞ k→∞
Proof Let {ψr }Nr=1 be a partition of unity on Ω, subordinated to the covering {Or }Nr=1 .
Then setting zr = zψr , we have that zr belongs to L 2 (Ω) and u · ∇ zr belongs to
L 1 (Ω). Thus, it suffices to examine zr in Ωr and to simplify, we drop the index r.
Let us choose ε = 1k and zk = z ρε,r , with ρε,r defined by (7.2.7). One has
lim z ρε,r − zL2 (Ωr ) = 0 ,

ε→0
lim (u · ∇ z) ρε,r − u · ∇ zL1 (Ωr ) = 0.
ε→0
Then the theorem follows from these two limits and Corollary 7.2.4.
Proving uniqueness of the solution of (7.2.1) with only this density result is not
straightforward. The proof of the proposition below relies on the renormalizing argu-
ment of Theorem II.2 in [82].
Proposition 7.2.6 Let Ω ⊂ IRd be bounded and Lipschitz-continuous. For all u in

W , all f in L 2 (Ω) and all real numbers α = 0, the transport equation (7.2.1) has
one and only one solution z in Xu and this solution satisfies (7.2.4).
Proof Let z ∈ L 2 (Ω) be a solution of
z+αu·∇z = 0 in Ω.
We must show that necessarily z = 0. Since z belongs to Xu , there exists a sequence

{zk }k≥1 of functions in D(Ω) satisfying the statement of Theorem 7.2.5. Then
zk + α u · ∇ zk = zk − z + α u · ∇(zk − z) = rk , (7.2.14)
where, owing to Theorem 7.2.5,
lim rk = 0 strongly in L 1 (Ω).

k→∞
On multiplying both sides of (7.2.14) by the function sign(zk ), this becomes
|zk | + α u · ∇(|zk |) = rk sign(zk ) in Ω. (7.2.15)
But |zk | belongs to H 1 (Ω) and u belongs to W ; therefore Green’s formula, which is
valid in this case, yields
u · ∇(|zk |) dx = 0.
Ω
Hence, integrating both sides of (7.2.15) in Ω, we obtain

|zk | dx = rk sign(zk ) dx ,
Ω Ω
and we let k tend to infinity in this equation. This gives zL1 (Ω) = 0 , thus proving
uniqueness. The bound (7.2.4) is a consequence of the uniqueness and of Proposi-
tion 7.2.1.
Proposition 7.2.7 Let Ω ⊂ IRd be Lipschitz-continuous and let u be given in W .

Then (5.2.43) extends to all z in Xu with
∀z ∈ Xu , (u · ∇ z, z) = 0. (7.2.16)
Proof Let z be in Xu and define h in L 2 (Ω) by
z + u · ∇ z = h in Ω. (7.2.17)
Then z is the unique solution of the transport equation (7.2.1) with α = 1 and data
h. The uniqueness of z implies that it is the limit of the sequence zm defined with
these data by (7.2.5). Now, taking the product of both sides of (7.2.17) with zm and
integrating in space, we obtain
(z, zm ) + (u · ∇ z, zm ) = (h, zm ) = zm 2L2 (Ω) . (7.2.18)
Thus passing to the limit in this equation, and using the lower semi-continuity of the
norm for the weak topology, we derive
z2L2 (Ω) + (u · ∇ z, z) = lim (zm 2L2 (Ω) ) ≥ z2L2 (Ω) .

m→∞
348 7 Appendix
Hence
(u · ∇ z, z) ≥ 0.
Similarly, changing u into −u and observing that z belongs also to X−u , we derive
the opposite inequality
(u · ∇ z, z) ≤ 0.
Therefore (u · ∇ z, z) = 0.
Proposition 7.2.7 has important corollaries. First, we have strong convergence of zm .
Corollary 7.2.8 In the setting of Proposition 7.2.7, the whole Galerkin sequence
{zm } defined by (7.2.5), with α = 1 and data h, converges strongly in L 2 (Ω) to the
solution z of the transport equation (7.2.17).
Proof In view of (7.2.18), we can write
z − zm 2 = (z − zm , z) − (z − zm , zm ) = (z − zm , z) − [(z, zm ) − zm 2 ]
= (z − zm , z) + (u · ∇ z, zm ).
In the last right-hand side, the first term obviously tends to zero and the second term
tends to (u · ∇ z, z) = 0. This proves the strong convergence of zm and since the
solution z is unique, the entire sequence converges to z.
Next, we have the following Green formula:

Corollary 7.2.9 Let Ω ⊂ IRd be Lipschitz-continuous and let u be given in W . Then

∀v ∈ Xu , ∀w ∈ Xu , (u · ∇ v)w dx + (u · ∇ w)v dx = 0. (7.2.19)
Ω Ω
Proof The proof follows immediately by applying (7.2.16) to v + w, which belongs

to the vector space Xu .
We also have the following orthogonal decomposition of L 2 (Ω):

Corollary 7.2.10 Let Ω ⊂ IRd be Lipschitz-continuous and let u be given in W .
Then any h in L 2 (Ω) has the orthogonal decomposition
h = z + u · ∇ z in Ω ,
where z belongs to Xu , and
z2u = z2L2 (Ω) + u · ∇ z2L2 (Ω) = h2L2 (Ω) . (7.2.20)

Proposition 7.2.7 and Corollary 7.2.9 have important consequences. The first one
is the following crucial density result:
Theorem 7.2.11 Let Ω ⊂ IRd be Lipschitz-continuous and let u be given in W . Then
D(Ω) is dense in Xu .
Proof Since Xu is a Hilbert space, we shall establish this density by proving that if
z in Xu satisfies
∀ϕ ∈ D(Ω) , (z, ϕ) + (u · ∇ z, u · ∇ ϕ) = 0 , (7.2.21)
then necessarily z = 0. Thus, let z be a solution of (7.2.21) and set w = u · ∇ z. As

div u = 0, we have u · ∇ ϕ = div(ϕ u) and since w is in L 2 (Ω), we can write
∀ϕ ∈ D(Ω) , (z, ϕ) − u · ∇ w, ϕD(Ω) = 0.
Therefore, in the sense of distributions, we have
z = u·∇ w, (7.2.22)
and as z belongs to L 2 (Ω), this implies that w belongs to Xu . Hence, we can take the
scalar product of both sides of (7.2.22) with z and apply (7.2.19) to get

z2L2 (Ω) = (u · ∇ w)z dx = − (u · ∇ z)w dx = −u · ∇ z2L2 (Ω) .
Ω Ω
Thus,
z2L2 (Ω) + u · ∇ z2L2 (Ω) = 0 ,
and z = 0.
Another consequence is the following corollary on the extension by zero of func-
tions of Xu . It shows that when u is in W , then extending z by zero has essentially the
same effect as extending u · ∇ z by zero. Exceptionally, the extension of z by zero is
denoted here with a special symbol.
Corollary 7.2.12 Let Ω ⊂ IRd be Lipschitz-continuous, let u be any function in W
and let z belong to Xu . Let ũ be any extension of u in H 1 (IRd )d and let z̃ ∈ L 2 (IRd )
denote the extension of z by zero outside Ω. Then ũ · ∇ z̃ belongs to L 2 (IRd ) and

∀ϕ ∈ D(IRd ) , (ũ · ∇ z̃)ϕ dx = (u · ∇ z)ϕ dx = c(u; z, ϕ) , (7.2.23)
IRd Ω
i.e.,
ũ · ∇ z̃ = u
·∇z,
where
· denotes the extension by zero to the whole of IRd .
350 7 Appendix
Proof For all functions ϕ of D(IRd ), we have
ũ · ∇ z̃, ϕD(IRd ) = ∇ z̃, ũϕH 1 (IRd ) = −z̃, div(ũϕ)L2 (IRd )

=− z u · ∇ ϕ dx ,
Ω
since div u = 0 in Ω. But ϕ belongs to Xu ; therefore Green’s formula (7.2.16) gives

ũ · ∇ z̃, ϕD(IRd ) = (u · ∇ z)ϕ dx = (u
· ∇ z)ϕ dx ,
Ω IRd
i.e., in the sense of distributions in IRd ,
ũ · ∇ z̃ = u
· ∇ z.
As u
· ∇ z belongs to L 2 (IRd ), this implies that ũ · ∇ z̃ belongs also to L 2 (IRd ).
Finally, we have the following truncation result:

Theorem 7.2.13 Let Ω ⊂ IRd be Lipschitz-continuous and let u be any function
in W . Then for any function v in W 1,p (Ω) with p > d, there exists a sequence of
functions ϕk ∈ D(Ω) such that
lim ϕk = v strongly in L 2 (Ω) , lim u · ∇ ϕk = u · ∇ v strongly in L 2 (Ω).

k→∞ k→∞
Proof The proof is an immediate consequence of the fact that v belongs to Xu and
Theorem 7.2.11.
Theorem 7.2.13 is remarkable, because D(Ω) is not dense in W 1,p (Ω); that is, its
statement is false if u · n does not vanish on the boundary.
7.2.2 Extension to a Steady Transport System in Lp
The material of this section is taken from Girault and Tartar [121]. We retain the
setting of the previous section and consider the following system of steady transport
equations: Find z ∈ L p (Ω)d satisfying
C z + α [∇ z]u = f in Ω , (7.2.24)
with the data: u ∈ W , α = 0, f ∈ L p (Ω)d , and C ∈ L ∞ (Ω)d×d , uniformly positive

definite, i.e., there exists c0 > 0 such that
a.e. in Ω, ∀w ∈ IRd , C w · w ≥ c0 |w|2 , (7.2.25)

recall that | · | denotes the Euclidean norm for vectors or tensors.

Owing to (7.2.25), any solution z ∈ L 2 (Ω)d of (7.2.24) satisfies zi ∈ Xu , for
1 ≤ i ≤ d. Moreover, it is easily shown by Galerkin’s method that (7.2.24) has at
least one solution z ∈ L 2 (Ω)d . Indeed, by using (7.2.25) and the fact that u belongs
to W , the argument is the same as in proving Proposition 7.2.1. Uniqueness is an
immediate consequence of (7.2.25), and Green’s formula (7.2.16) applied to each
component zi of z. Then (7.2.25) and (7.2.16) imply
1
zL2 (Ω) ≤ f L2 (Ω) . (7.2.26)
c0
Now, we restrict the dimension d to 2 ≤ d ≤ 4. For 2 < p < ∞ and f ∈ L p (Ω)d ,

we shall prove that this solution is in L p (Ω)d and (7.2.26) holds with p instead of 2.
This is not straightforward. Indeed, if we already know that z is in L p (Ω)d , then the
analogue of (7.2.26) in L p is derived by taking the scalar product of (7.2.24) with
test function |z|p−2 z. This is a classical approach, but the difficulty lies in proving
first that z is in L p (Ω)d . We shall achieve this by two regularizations: first a standard
elliptic regularization and next a Yosida regularization.
7.2.2.1 Elliptic Regularization
Consider the following regularization of (7.2.24). For each ε > 0, find zε ∈ H01 (Ω)d
solution of the problem
− ε Δ zε + C zε + α [∇ zε ]u = f in Ω. (7.2.27)
Owing to (7.2.25) and (7.2.16), the bilinear form associated with the left-hand side
of (7.2.27),
aε (z, ζ ) = ε (∇ z, ∇ ζ ) + C z, ζ + α [∇ z]u, ζ ,
is elliptic and considering the restriction d ≤ 4 on the dimension, it is continuous

on H01 (Ω)d × H01 (Ω)d . Indeed, since d ≤ 4, H 1 (Ω) ⊂ L 4 (Ω) so that the product
([∇ v]u)·w belongs to L 1 (Ω) for all u, v, w in H 1 (Ω)d . Therefore, by Lax–Milgram’s
Lemma, (7.2.27) defines a unique function zε ∈ H01 (Ω)d , and
1 √ ε 1
zε L2 (Ω) ≤ f L2 (Ω) , εz H01 (Ω) ≤ √ f L2 (Ω) . (7.2.28)
c0 2 c0
We shall extend the first part of (7.2.28) to a uniform L p bound by a Yosida reg-
ularization. Note that we cannot test directly (7.2.27) with |zε |p−2 zε , because in a
Lipschitz domain, this function may not be smooth enough.
352 7 Appendix
7.2.2.2 Yosida Regularization and Lp Estimate
For fixed p with 2 < p < ∞, consider the following Yosida approximation of zε .
For each δ > 0, find σ δ in L p (Ω)d such that
σ δ + δ |σ δ |p−2 σ δ = zε in Ω. (7.2.29)
The regularizing effect of σ δ is derived in the following proposition:

Proposition 7.2.14 If zε is given in L p (Ω)d , 1/p + 1/p = 1, Problem (7.2.29) has
a unique solution σ δ ∈ L p (Ω)d and if zε ∈ H01 (Ω)d , then both σ δ and |σ δ |p−2 σ δ
belong to H01 (Ω)d .
Proof For any σ ∈ L p (Ω)d , let A(σ ) = σ + δ |σ |p−2 σ . Then A maps L p (Ω)d into

L p (Ω)d ; it is coercive, hemicontinuous, and monotone in L p (Ω)d , since it satisfies

∀σ , τ ∈ IRd , A(σ ) − A(τ ) · σ − τ ≥ |σ − τ |2

+δ |σ |p−1 − |τ |p−1 |σ | − |τ | ≥ |σ − τ |2 .
Hence (7.2.29) has a unique solution σ δ in L p (Ω)d and σ δ = ϕ(δ |zε |p−2 ) zε , where
ϕ : IR+ →]0, 1] satisfies

∀t ∈ IR+ , ϕ(t) 1 + t ϕ(t)p−2 ) = 1 , (7.2.30)
whence ϕ(0) = 1. Moreover, ϕ is Lipschitz-continuous in IR+ and differentiating

(7.2.30), we easily derive that
ϕ(t)p−1
ϕ (t) = − .
1 + (p − 1)tϕ(t)p−2
So ϕ ∈ C 1 (IR+ ) and for all t ∈ IR+ , we have −1 ≤ ϕ (t) < 0. Now, assume that
zε ∈ H01 (Ω)d and let S ε = {x ∈ Ω ; zε (x) = 0}. Then we have for 1 ≤ j ≤ d,
∂ ε p−2 ∂zε
a.e. in S ε , |z | = (p − 2)|zε |p−4 zε · .

∂ xj ∂ xj
Therefore
∂σ α
∂zε ∂zε
ε
a.e. in S ε , = ϕ δ |zε |p−2 + ϕ δ |zε |p−2 δ (p − 2)|zε |p−4 zε · z.
∂ xj ∂ xj ∂ xj
As p > 2, this formula is valid a.e. in Ω and using the previous bounds for ϕ and ϕ ,
we obtain ∂σ
δ ∂zε
a.e. in Ω , ≤ δ (p − 2)|zε |p−2 + 1 .
∂ xj ∂ xj
Thus ∇ σ δ is defined a.e. in Ω and by differentiating both sides of (7.2.29), we readily

derive
∂zε ∂σ δ ∂σ δ 2 ∂σ δ
2
· = 1 + δ |σ δ |p−2 + δ (p − 2) |σ δ |p−4 σ δ · .
∂ xj ∂ xj ∂ xj ∂ xj
Hence,
1
a.e. in Ω , |∇ σ δ | ≤ |∇ zε | ≤ |∇ zε |.
1 + δ|σ δ |p−2
This implies that σ δ ∈ H 1 (Ω)d ; then (7.2.29) yields immediately that |σ δ |p−2 σ δ
belongs also to H 1 (Ω)d and all components of both functions vanish on ∂Ω.
Therefore, we can test (7.2.27) with |σ δ |p−2 σ δ , and it gives

aε zε , |σ δ |p−2 σ δ = f , |σ δ |p−2 σ δ . (7.2.31)
Then with (7.2.31), we can obtain a uniform bound for σ δ in L p (Ω).

Proposition 7.2.15 Let zε be the solution of (7.2.27) with f given in L p (Ω)d . Then
the solution σ δ of (7.2.29) satisfies the uniform bound
1
σ δ Lp (Ω) ≤ f Lp (Ω) . (7.2.32)
c0
Proof Let us substitute the expression (7.2.29) for zε into (7.2.31). As |σ δ |p−2 σ δ
belongs to H01 (Ω)d , the ellipticity of aε (·, ·) yields

aε |σ δ |p−2 σ δ , |σ δ |p−2 σ δ ≥ 0 ,
and we are left with

ε ∇ σ δ , ∇(|σ δ |p−2 σ δ ) + α [∇ σ δ ]u, |σ δ |p−2 σ δ
p−1
+ C σ δ , |σ δ |p−2 σ δ ≤ f Lp (Ω) σ δ Lp (Ω) .
On one hand, since
∂ ∂ σδ ∂ σδ
|σ δ |p−2 σ δ = |σ δ |p−2 + (p − 2)σ δ |σ δ |p−4 σ δ · ,

∂ xj ∂ xj ∂ xj
we have
d
∂ σ δ
2
∇ σ δ : ∇ |σ δ |p−2 σ δ = |σ δ |p−2 |∇ σ δ |2 + (p − 2)|σ δ |p−4 σδ · ≥ 0.
j=1
∂ xj
354 7 Appendix
On the other hand, we can write
1
[∇ σ δ ]u · |σ δ |p−2 σ δ = u · ∇(|σ δ |p ).
p
Therefore, since d ≤ 4, u · ∇(|σ δ |p ) belongs to L 1 (Ω) and similarly, |σ δ |p belongs

to L 2 (Ω). According to Theorem 7.2.5, D(Ω) is dense in the space ζ ∈ {L 2 (Ω) ; u ·
∇ ζ ∈ L 1 (Ω)}; therefore Green’s formula yields

u · ∇(|σ δ |p ) = 0.
Ω
Hence there remains

p p−1
c0 σ δ Lp (Ω) ≤ C σ δ , |σ δ |p−2 σ δ ≤ f Lp (Ω) σ δ Lp (Ω) ,
whence (7.2.32).
The uniform bound (7.2.32) allows to prove our main result.
Theorem 7.2.16 Let p ≥ 2 be a real number, and let Ω, u, α, f and C be as above.

Then the unique solution z of (7.2.24) belongs to L p (Ω)d and
1
zLp (Ω) ≤ f Lp (Ω) . (7.2.33)
c0
Proof The uniform bound (7.2.32) shows that up to subsequences, as δ tends to

zero, σ δ converges weakly in L p (Ω)d to some function σ . In addition it implies that

|σ δ |p−2 σ δ is also uniformly bounded in L p (Ω)d . Therefore, by setting (7.2.29) into
an equivalent variational form, testing it with a smooth function, and passing to the
limit as δ tends to zero, we easily derive that σ = zε . Then in view of (7.2.32),
1
zε Lp (Ω) ≤ f Lp (Ω) .
c0
Hence, up to subsequences, zε converges weakly, as ε tends to zero, to a function z̄

in L p (Ω)d , and
1
z̄Lp (Ω) ≤ f Lp (Ω) .
c0
Then the two parts of (7.2.28) allow to pass to the limit in the variational formulation
of (7.2.27) and we see that z̄ solves (7.2.24). Uniqueness of the solution of (7.2.24)
completes the proof.
Remark 7.2.17 Of course, Theorem 7.2.16 applies to a single transport equation

(7.2.1):
z + α u · ∇ z = f in Ω ,
but in this case, the L p regularity of z can be established without restriction on the
dimension d. Indeed, the next theorem is proved in [118]. The proof relies on an
analogous result established by Ortega [202] (see also [97]) in a domain with smooth
boundary, with a transport velocity u in W 1,∞ (Ω)d , vanishing on ∂Ω.
Theorem 7.2.18 Let the dimension d ≥ 2, let Ω ⊂ IRd be bounded and Lipschitz-
continuous and let p ≥ 2 be a real number. For all u in W , all f in L p (Ω), and all
real numbers α, the transport equation (7.2.1) has one and only one solution z in
L p (Ω), with u · ∇ z in L p (Ω). This solution satisfies
zLp (Ω) ≤ f Lp (Ω) .
The proof proceeds by imbedding Ω into a smooth ball B ⊂ Ω, extending (7.2.1) to

B and regularizing u in order to apply the result of Ortega. The details of the proof
can be found in [118].
7.2.2.3 Application to the W 1,p Regularity of a Scalar Transport

Equation
In the same domain Ω, we consider the slight variant of (7.2.1): Find z ∈ L 2 (Ω)
satisfying
μ z + α u · ∇ z = f in Ω , (7.2.34)
where μ is a given positive constant, α = 0, f is given in L 2 (Ω) and u in W . This

problem has a unique solution. Several authors (for instance cf. Beirao Da Veiga [23]
and references therein) have established that if, in addition, ∂Ω is sufficiently smooth,
if f ∈ H 1 (Ω), and if u is in W ∩ W 1,∞ (Ω)d with
|α| ∇ uL∞ (Ω) < μ , (7.2.35)
then z belongs to H 1 (Ω) and
1
∇ zL2 (Ω) ≤ ∇ f L2 (Ω) . (7.2.36)
μ − |α| ∇ uL∞ (Ω)
There are several proofs of this result, but all either require a smooth boundary
or rely on the H 2 regularity of the Laplace equation with homogeneous Dirichlet or
Neumann boundary conditions. This regularity holds either if the boundary is smooth
or if the domain is a convex polygon or polyhedron.
Let p > 2 be a real number and let f be given in W 1,p (Ω). We shall prove that,
under the above assumptions, z ∈ W 1,p (Ω). Since z belongs to H 1 (Ω), the gradient
of each term in (7.2.34) is well defined in the sense of distributions and ∇ z solves:
Find w in L 2 (Ω)d such that
356 7 Appendix
T
μ w + α [∇ w]u + α ∇ u w = ∇ f . (7.2.37)
T
It is a particular case of (7.2.24) with C = μ I + α ∇ u . The fact that u belongs
to W 1,∞ (Ω)d implies that C is uniformly bounded in Ω and owing to (7.2.35),
C satisfies (7.2.25) with c0 = μ − |α| ∇ uL∞ (Ω) . Hence Theorem 7.2.16 implies
immediately the next result.
Theorem 7.2.19 Let the dimension satisfy 2 ≤ d ≤ 4. Assume that Ω has a smooth
boundary, or is a convex polygon or polyhedron. Let p > 2 be a real number, let
μ > 0, α = 0, f ∈ W 1,p (Ω), and u ∈ W ∩ W 1,∞ (Ω)d satisfying (7.2.35). Then the
unique solution z of (7.2.34) belongs to W 1,p (Ω) and
1
∇ zLp (Ω) ≤ ∇ f Lp (Ω) . (7.2.38)
μ − |α| ∇ uL∞ (Ω)
Remark 7.2.20 The statement of Theorem 7.2.19 is valid on a bounded Lipschitz

domain in the case when u vanishes on ∂Ω. Indeed, it suffices to fix a smooth
large ball B containing Ω, extend u by zero outside Ω and extend f continuously
in W 1,p (B). Let f˜ and ũ denote the extended functions. As u vanishes on ∂Ω, ũ is
divergence free in B, it belongs to W 1,∞ (B)d and has the same W 1,∞ norm as u.
Therefore, it satisfies (7.2.35) in B. Then consider the scalar transport equation :
Find Z ∈ L 2 (B) such that
μ Z + α ũ · ∇ Z = f˜ in B.
The assumptions of Theorem 7.2.19 hold in B and thus Z belongs to W 1,p (B). Let z
denote the restriction of Z to Ω, and observe that

ũ · ∇ Z |Ω = u · ∇ z .
Hence z is the unique solution of (7.2.34) and so z = z . Thus z ∈ W 1,p (Ω) and the
bound (7.2.38) follows immediately by applying Theorem 7.2.16 to (7.2.37).
7.3 The Time-Dependent Transport Equation
7.3.1 A Scalar Equation: Existence of Solutions
The material of this section is a variant of [119].

Let d ≥ 2 be the dimension; let Ω be a bounded domain in IRd , with a Lipschitz-
continuous boundary ∂Ω and exterior unit normal n, let u be given in L 2 (0, T ; W ),
and let β > 0, μ ∈ IR, and α = 0 be any given constants. Consider the following
scalar time-dependent transport equation:
7.3 The Time-Dependent Transport Equation 357
Find z ∈ L ∞ (0, T ; L 2 (Ω)) satisfying
∂z
β + μ z + α u · ∇ z = f in Ω×]0, T [ , (7.3.1)
∂t
z(0) = z0 in Ω , (7.3.2)
with the given data f in L 2 (Ω×]0, T [) and z0 in L 2 (Ω).

Remark 7.3.1 Note that any solution z ∈ L ∞ (0, T ; L 2 (Ω)) of (7.3.1)–(7.3.2)
belongs to H 1 (0, T ; W −1,q (Ω)), where
d
q= if d ≥ 3, q < 2 if d = 2.
d−1
To check this, we examine the behavior of u · ∇ z in space. Since div u = 0, we

have u · ∇ z = div(z u). By Sobolev’s imbedding, u ∈ L 2 (0, T ; L p (Ω)d ) with p ≤
2d/(d − 2) if d ≥ 3 and p < ∞ when d = 2. Therefore z u ∈ L 2 (0, T ; L q (Ω)d )
with q ≤ d/(d − 1) if d ≥ 3 and q < 2 when d = 2. As f ∈ L 2 (Ω×]0, T [), this
implies that
∂z
∈ L 2 (0, T ; W −1,q (Ω)),
∂t
and consequently, the initial condition (7.3.2) makes sense.
Establishing existence of a solution of (7.3.1)–(7.3.2) is straightforward. It can
be done for instance by semi-discretization in time : Starting from z0 = z0 , find a
sequence (zn )1≤n≤N such that for 1 ≤ n ≤ N,
1
β (zn − zn−1 ) + μ zn + α un · ∇ zn = f n , (7.3.3)
k
where k is the time step, T = N k, t n = n k for 0 ≤ n ≤ N,

tn
1
f n (x) = f (x, t) dt in Ω , 1 ≤ n ≤ N ,
k t n−1
tn
1
un (x) = u(x, t) dt in Ω, 1 ≤ n ≤ N.
k t n−1
Clearly for all n ≥ 1, un ∈ W . Hence (7.3.3) reduces to a steady transport equation

and the theory developed in Section 7.2.1 under the assumption (7.3.4) below when
μ ≤ 0, shows that, starting from z0 , (7.3.3) defines a unique sequence (zn )1≤n≤N that
satisfies a suitable uniform bound depending on the sign of μ. If μ ≤ 0, assuming
1
k < k0 , where k0 = , (7.3.4)
1+ 2|μ|
β
358 7 Appendix
setting
2|μ|
k̃ = 1 + k,
β
taking the scalar product of (7.3.3) with zn , summing over n, we get for 1 ≤ n ≤ N,
β
n−1
βzn 2L2 (Ω) + zi+1 − zi 2L2 (Ω)
1 − k̃ i=0

tn
1 1 2|μ|
≤ β z0 L2 (Ω) + f L2 (Ω×]0,t n [) exp

2 2
1+ . (7.3.5)
1 − k̃ β 1 − k̃ β
If μ > 0, the exponential factor can be avoided and we have for 1 ≤ n ≤ N,

n−1
1
βzn 2L2 (Ω) + β zi+1 − zi 2L2 (Ω) ≤ β z0 2L2 (Ω) + f 2L2 (Ω×]0,t n [) ,
i=0
2μ
and

n−1
βzn 2L2 (Ω) + β zi+1 − zi 2L2 (Ω)
i=0

n
1
+μ kzi 2L2 (Ω) ≤ β z0 2L2 (Ω) + f 2L2 (Ω×]0,t n [) .
i=1
μ
Moreover, when μ is a small positive number, division by μ can be avoided by taking

again the scalar product of (7.3.3) with zn , but canceling by zn as follows:
β zn L2 (Ω) + k μ zn L2 (Ω) ≤ β zn−1 L2 (Ω) + k f n L2 (Ω) ,
if zn = 0; if zn = 0, this inequality is trivially true. By summing over n, this yields

n
βzn L2 (Ω) + μ k zi L2 (Ω) ≤ β z0 L2 (Ω) + f L1 (0,t n ;L2 (Ω)) .
i=1
Then standard arguments permit to pass to the limit as N tends to infinity and the
limiting function solves (7.3.1)–(7.3.2), see for instance the part of the proof of
Theorem 5.7.22 relative to the transport equation (5.7.67). Thus we have the following
existence result:
Theorem 7.3.2 For any u in L 2 (0, T ; W ), f in L 2 (Ω×]0, T [), z0 in L 2 (Ω), β > 0,

μ ∈ IR, α = 0, problem (7.3.1)–(7.3.2) has at least one solution z in L ∞ (0, T ; L 2 (Ω)),
and this solution satisfies the following bounds for almost any t in [0, T ]:
If μ ≤ 0,
1
2|μ|
β z(t)2L2 (Ω) ≤ β z0 2L2 (Ω) + f 2L2 (Ω×]0,t]) exp t 1 + . (7.3.6)

β β
If μ > 0,
1
β z(t)2L2 (Ω) ≤ β z0 2L2 (Ω) + f 2L2 (Ω×]0,t]) ,
2μ
1
β z(t)2L2 (Ω) + μz2L2 (Ω×]0,t]) ≤ β z0 2L2 (Ω) + f 2L2 (Ω×]0,t]) , (7.3.7)
μ
β z(t)L2 (Ω) ≤ β z0 L2 (Ω) + f L1 (0,t;L2 (Ω)) .
Proving uniqueness is not straightforward, considering the low regularity of the

domain and of the driving velocity. We propose to adapt the regularization technique
introduced in Section 7.2.1 for the steady equation.
7.3.2 A Scalar Equation: Uniqueness
Here we consider an arbitrary solution z ∈ L ∞ (0, T ; L 2 (Ω)) of
∂z
β + μ z + α u · ∇ z = 0 in Ω×]0, T [ , (7.3.8)
∂t
z(0) = 0 in Ω. (7.3.9)
Again, it is convenient to extend z by zero outside Ω, so that the extended function

(still denoted z) belongs to L ∞ (0, T ; L 2 (IRd )) without changing its norm. Similarly,
we extend u continuously outside Ω so that the extended function (still denoted u)
belongs to L 2 (0, T ; H 1 (IRd )d ).
7.3.2.1 Local Regularization
We propose to regularize z by convolution as in Section 7.2.1. We refer to Section 7.2.1

for the notation: Or , Ωr , δr , ρr , t r . For all ε ∈]0, 1], we associate with Ωr the special
mollifier defined in (7.2.7),
1 x
∀x ∈ Or , ρε,r (x) = ρr ( + t r ).
εd ε
360 7 Appendix
To begin with, we make no assumption on the divergence of u and we suppose that z is

only in L 2 (Ω×]0, T [). The formula u·∇ z = div(z u)−z div u implies that the product
u · ∇ z is the sum of a term in L 1 (0, T ; W −1,q (IRd )) and a term in L 1 (IRd ×]0, T [); thus
its convolution product with ρε,r is well defined almost everywhere in Ωr ×]0, T [ and
is independent of the extension of u and z
(u · ∇ z)ρε,r (x, t) = u(y, t) · ∇ z(y, t), ρε,r (x − y)D(IRd )

= (−1)d (u · ∇ z)(x − ε(y − t r ), t), ρε,r (y)D(B(0;δr ) in Ωr ×]0, T [.
The same conclusion holds for u · ∇(z ρε,r ), and we have the same “commuting "
property of convolution with ρε,r as in Lemma 7.2.3. Again, its statement makes no
assumption on u · ∇ z nor on the divergence or boundary values of u.
Lemma 7.3.3 Let Ω be a bounded Lipschitz domain of IRd , let z be given in
L 2 (Ω×]0, T [) and let u be given in L 2 (0, T ; H 1 (Ω)d ). With the notation of
Lemma 7.2.2, there exists a constant C independent of z and u such that for all r
with 1 ≤ r ≤ R and all ε ∈]0, 1], we have
u · ∇(z ρε,r ) − (u · ∇ z) ρε,r )L1 (Ωr ×]0,T [)

≤ C∇ uL2 (Ω×]0,T [) zL2 (Ω×]0,T [) . (7.3.10)
Proof The proof is an easy adaptation of that of Lemma 7.2.3. Set
Aε,r (x, t) = [u · ∇(z ρε,r ) − (u · ∇ z) ρε,r ](x, t) in Ωr ×]0, T [.
Expanding, we find a.e. in Ωr ×]0, T [

Aε,r (x, t) = z(y, t)[u(x, t) · ∇x ρε,r (x − y) + u(y, t) · ∇y ρε,r (x − y)] dy
d
IR
+ z(y, t)ρε,r (x − y)divy u(y, t) dy
IRd
x − y
1
= d+1 z(y, t)(u(x, t) − u(y, t)) · ∇ ρr + t r dy
ε IRd ε

+ z(y, t)divy u(y, t)ρε,r (x − y) dy. (7.3.11)
IRd
On one hand, taking the absolute value, applying Fubini’s theorem, and integrating
with respect to x in Ωr , we find

T
z(y, t)divy u(y, t)ρε,r (x − y) dy dx dt ≤ z div uL1 (Ω×]0,T [) . (7.3.12)
0 Ωr IRd
On the other hand, we have a.e. in ]0, T [

1 x−y
I (x, t) = z(y, t)(u(x, t) − u(y, t)) · ∇ ρr ( + t r ) dy
εd+1 IRd ε
1
= (−1)d z(x − ε(y − t r ), t) ∇ u(x−(1 − θ )ε(y − t r ), t) · (y − t r ) dθ ·∇ρr (y) dy.
0
B(0;δr )
By proceeding as above, we obtain

T T
I(x, t) dx dt ≤ Cr z(·, t)L2 (Ω) ∇ u(·, t)L2 (Ω) dt , (7.3.13)
0 Ωr 0
where the constant Cr depends only on Ωr . Then (7.3.10) follows by substituting

(7.3.12) and (7.3.13) into (7.3.11).
The quantity Aε,r (x, t) will be part of a residue in the uniqueness argument further
on. The next corollary states that this residue tends to zero with ε. Its proof is the
same as that of Corollary 7.2.4. Note again that it makes no assumption on u · ∇ z
nor on the divergence or boundary values of u.
Corollary 7.3.4 Under the assumptions of Lemma 7.3.3, for all r with 1 ≤ r ≤ R,
lim u · ∇(z ρε,r ) − (u · ∇ z) ρε,r L1 (Ωr ×]0,T [) = 0. (7.3.14)

ε→0
7.3.2.2 Global Regularization and Uniqueness
Let z ∈ L ∞ (0, T ; L 2 (Ω)) be any solution of (7.3.8)–(7.3.9), extended by zero outside

Ω. Let {ψr }Rr=1 be a partition of unity in Ω, subordinated to the covering {Or }Rr=1 and
let zr = zψr . Observe that the support of zr is contained in Ωr and more precisely,
it is contained in the intersection of the support of ψr and Ω. Furthermore, as far as
regularity is concerned, zr behaves like z. Now, take ε of the form 1k , for any integer
k ≥ 1, and set
R

zk = zr ρ 1k ,r . (7.3.15)
r=1
Then zk regularizes z and is a suitable approximation of z in L 1 (Ω×]0, T [).

Lemma 7.3.5 There exists an integer k0 such that for all f in L 1 (Ω×]0, T [) extended
outside Ω, for all k ≥ k0 , and for all r, 1 ≤ r ≤ R, the support of the function
by zero
f ψr ρ 1k ,r is contained in Ωr . Furthermore
R

lim f ψr ρ 1k ,r − f L1 (Ω×]0,T [) = 0. (7.3.16)
k→∞
r=1
362 7 Appendix
Proof Clearly, f ψr is supported by Ωr . Let us prove that this is also true for (f ψr )
ρ 1k ,r when k is large enough: There exists k0 such that
∀k ≥ k0 , (f ψr ) ρ 1k ,r = 0 in (Ω \ Ωr )×]0, T [.
r be the intersection of the support of ψr and Ω. There exists η > 0

Indeed, let Ω
such that for all r, 1 ≤ r ≤ R,
r ) > η.
∀x ∈ Ω \ Ωr , d(x, Ω
Choose an integer k0 such that
δr + |t r |
k0 ≥ sup .
1≤r≤R η
It is easy to check that
∀k ≥ k0 , ∀1 ≤ r ≤ R, ∀x ∈ Ω \ Ωr , ∀y ∈ B(0; δr ) ,
1
x − (y − t r ) is not in Ω r ,
k

hence (f ψr ) ρ 1k ,r (x) = 0.
Now, we take k ≥ k0 . As {ψr }Rr=1 is a partition of unity in Ω, we can write
R

R

f ψr ρ 1k ,r − f = f ψr ρ 1k ,r − f ψr in Ω×]0, T [.
r=1 r=1
Therefore

R

R

f ψr ρ 1k ,r − f L1 (Ω×]0,T [) ≤ (f ψr ρ 1k ,r − f ψr L1 (Ω×]0,T [) .
r=1 r=1
From the properties of ρ 1k ,r , we have for each r, 1 ≤ r ≤ R,

lim (f ψr ρ 1k ,r − f ψr L1 (Ωr ×]0,T [) = 0. (7.3.17)
k→∞

But, as f ψr ρ 1k ,r is supported by Ωr , we have for all r, 1 ≤ r ≤ R,

(f ψr ρ 1k ,r − f ψr L1 (Ω×]0,T [) = (f ψr ρ 1k ,r − f ψr L1 (Ωr ×]0,T [) ,
and (7.3.16) follows from (7.3.17).
The next lemma describes the effect of the transport operator on zk .

Lemma 7.3.6 With the above notation, zk satisfies a.e. in Ω×]0, T [
∂zk R
β + μ zk + α u · ∇ zk =α u · ∇(zr ρ 1k ,r ) − (u · ∇ zr ) ρ 1k ,r
∂t r=1

+ (u · ∇ ψr )z ρ 1k ,r − (u · ∇ ψr )z . (7.3.18)
Proof As ψr and ρ 1k ,r are both independent of time, we can write a.e. in Ω×]0, T [
∂zk ∂z
R
β + μ zk + α u · ∇ zk = β ψr ρ 1k ,r + μ zr ρ 1k ,r + α u · ∇(zr ρ 1k ,r ) .
∂t r=1
∂t
Then (7.3.18) follows readily from the following facts: (7.3.8) gives a.e. in Ω×]0, T [
∂z
β = −μ z − α u · ∇ z ,
∂t
the following formula holds
∀ϕ ∈ H02 (Ω) , (u · ∇ z)ψr , ϕ = −(u · ∇ ψr )z, ϕ + (u · ∇(z ψr )), ϕ ,
and a.e. in Ω×]0, T [

R
(u · ∇ ψr )z = 0 ,
r=1
because {ψr }Rr=1 is a partition of unity in Ω.

The right-hand side of (7.3.18) is the sum of two residual terms (up to the factor
α),
ak,r = u · ∇(zr ρ 1k ,r ) − (u · ∇ zr ) ρ 1k ,r ,
(7.3.19)
bk,r = (u · ∇ ψr )z ρ 1k ,r − (u · ∇ ψr )z.
The first part of Lemma 7.3.5 guarantees that for all k ≥ k0 and for each r, 1 ≤ r ≤ R,
ak,r L1 (Ω×]0,T [) = ak,r L1 (Ωr ×]0,T [) ,
bk,r L1 (Ω×]0,T [) = bk,r L1 (Ωr ×]0,T [) .
Then, Corollary 7.3.4 and Lemma 7.3.5 yield immediately the following result:
Proposition 7.3.7 The residue in (7.3.18) satisfies
R

lim ak,r + bk,r L1 (Ω×]0,T [) = 0. (7.3.20)
k→∞
r=1
364 7 Appendix
With this, the renormalizing argument of Theorem II.2 in [82] implies the main result
of this section.
Theorem 7.3.8 Let Ω ⊂ IRd be Lipschitz-continuous. For any u in L 2 (0, T ; W ), f
in L 2 (Ω×]0, T [), z0 in L 2 (Ω), β > 0, μ ∈ IR, and α = 0, problem (7.3.1)–(7.3.2)
has exactly one solution z in L ∞ (0, T ; L 2 (Ω)).
This solution, for almost all t in [0, T ], satisfies the estimates from Theorem 7.3.2,
i.e., (7.3.6) if μ ≤ 0, respectively (7.3.7) if μ > 0.
Proof Let z ∈ L ∞ (0, T ; L 2 (Ω)) be any solution of (7.3.8)–(7.3.9), extended by zero

outside Ω. We must show that necessarily z = 0. Let zk be defined by (7.3.15), and
with the notation of Proposition 7.3.7, let

R

rk = α ak,r + bk,r ,
r=1
denote the right-hand side of (7.3.18). Note that ∂z∂tk belongs to L 1 (Ω×]0, T [) because
zk , u · ∇ zk , and rk belong to L 1 (Ω×]0, T [). Therefore, we can multiply both sides
of (7.3.18) with sgn(zk ) since it is in L ∞ (Ω×]0, T [). This gives
∂zk
β sgn(zk ) + μ zk sgn(zk ) + α(u · ∇ zk ) sgn(zk ) = rk sgn(zk ) in Ω×]0, T [ ,
∂t
i.e.,
∂|zk |
β + μ |zk | + α u · ∇(|zk |) = rk sgn(zk ) in Ω×]0, T [. (7.3.21)
∂t
∂|zk |
Note that |zk | belongs to L ∞ (0, T ; H 1 (Ω)) and belongs at least to
∂t
L 1 (Ω×]0, T [); in particular, |zk | belongs to C([0, T ]; L 1 (Ω)). Therefore, we can
integrate both sides of (7.3.21) in space and time. Considering that on one hand

∂|zk | d
dx = zk L1 (Ω) in ]0, T [ ,
Ω ∂t dt
and on the other hand

u · ∇(|zk |) dx = 0 in ]0, T [ ,
Ω
since u ∈ L (0, T ; W ) and |zk | ∈ L ∞ (0, T ; H 1 (Ω)), there remains
2
t t
d
∀t ∈]0, T [ , β zk L1 (Ω) dt + μ zk L1 (Ω×]0,t[) = rk sgn(zk ) dx dt.
0 dt 0 Ω
As zk (0) = 0, this implies a.e. in ]0, T [
β zk (t)L1 (Ω) + μ zk L1 (Ω×]0,t[) ≤ rk L1 (Ω×]0,t[) .
When μ ≥ 0, the limit (7.3.20) and the fact that β > 0 yield
lim zk L1 (Ω×]0,T [) = 0. (7.3.22)

k→∞
When μ < 0, we write
β zk (t)L1 (Ω) ≤ rk L1 (Ω×]0,t[) + |μ| zk L1 (Ω×]0,t[) .
As zk (t)L1 (Ω) is continuous on [0, T ], we can apply Gronwall’s Lemma 3.3.13 to
get
|μ|
β zk (t)L1 (Ω) ≤ rk L1 (Ω×]0,t[) exp( t) ,
β
and we recover (7.3.22). But Lemma 7.3.5 implies that
lim zk − zL1 (Ω×]0,T [) = 0 ,

k→∞
thus proving uniqueness.
7.3.3 The Case of a System with a Lipschitz Velocity
We consider here the time-dependent version of (7.2.24): Find z ∈ L ∞ (0, T ; L 2 (Ω)d )

solution of
∂z
β + C z + α [∇ z]u = f in Ω×]0, T [ ,
∂t (7.3.23)
z(0) = z0 in Ω ,
with the data: u ∈ L 2 (0, T ; W ∩ W 1,∞ (Ω)d ), α = 0, β > 0, f ∈ L 2 (Ω×]0, T [)d ,

and C ∈ L ∞ (Ω×]0, T [)d×d .
By combining the techniques of Section 7.3.1 and the beginning of Section 7.2.2,
it is easy to prove that (7.3.23) has at least one solution. Indeed, consider the semi-
discretization: Find a sequence (zn )n=1 , z ∈ L 2 (Ω)d , solving
n=N n
1
β (zn − zn−1 ) + C n zn + α [∇ zn ]un = f n in Ω ,
k (7.3.24)
z0 = z0 in Ω ,
366 7 Appendix
with f n ∈ L 2 (Ω)d , C n ∈ L ∞ (Ω)d×d , un ∈ W ∩ W 1,∞ (Ω)d defined by

tn
1
f (x) =
n
f (x, t) dt in Ω , 1 ≤ n ≤ N ,
k t n−1
tn
1
C n (x) = C(x, t) dt in Ω , 1 ≤ n ≤ N ,
k t n−1
tn
1
u (x) =
n
u(x, t) dt in Ω, 1 ≤ n ≤ N.
k t n−1
For each n, system (7.3.24) is a particular case of (7.2.24). Hence, under the condition
(7.3.25), when C is arbitrary, it defines a unique sequence (zn )n=1
n=N
. Regarding a priori
estimates, assuming
1
k < k0 where k0 = , (7.3.25)
1+ 2
β
CL∞ (Ω×]0,T [)
and setting
2
k̃ = k 1 + CL∞ (Ω×]0,T [) ,
β
we have the analogue of (7.3.5) for 1 ≤ n ≤ N,
β
n−1
1
βzn 2L2 (Ω) + zi+1 − zi 2L2 (Ω) ≤ βz0 2L2 (Ω)
1 − k̃ i=0 1 − k̃
1 tn 2
+ f 2L2 (Ω×]0,t n [) exp 1 + CL∞ (Ω×]0,T [) .

β 1 − k̃ β
If in addition we assume that C satisfies (7.2.25): There exists c0 > 0 such that
a.e. in Ω×]0, T [, ∀w ∈ IRd , C w · w ≥ c0 |w|2 , (7.3.26)
then for each n ≥ 1, C n verifies
a.e. in Ω, ∀w ∈ IRd , C n (x) w · w ≥ c0 |w|2 , (7.3.27)
and the sequence (zn ) satisfies the following a priori estimates for 1 ≤ n ≤ N:

n−1
1
βzn 2L2 (Ω) + β zi+1 − zi 2L2 (Ω) ≤ βz0 2L2 (Ω) + f 2L2 (Ω×]0,t n [) , (7.3.28)
i=0
2c0

n−1
n
βzn 2L2 (Ω) + β zi+1 − zi 2L2 (Ω) + c0 kzi 2L2 (Ω)
i=0 i=1
1
≤ βz0 2L2 (Ω) + f 2L2 (Ω×]0,t n [) , (7.3.29)
c0
and

n
βzn L2 (Ω) + c0 k zi L2 (Ω) ≤ βz0 L2 (Ω) + f L1 (0,t n ;L2 (Ω)d ) (7.3.30)
i=1
Again, passing to the limit is standard and yields the following existence result.
Its proof does not require the higher regularity of u and therefore it is stated for u in
L 2 (0, T ; W ).
Proposition 7.3.9 For any u in L 2 (0, T ; W ), f in L 2 (Ω×]0, T [)d , z0 in L 2 (Ω)d ,

β > 0, C ∈ L ∞ (Ω×]0, T [)d×d , problem (7.3.23) has at least one solution z in
L ∞ (0, T ; L 2 (Ω)d ). Moreover, the following bound holds for almost any t in [0, T ]:
1 2
β z(t)2L2 (Ω) ≤ β z0 2L2 (Ω) + f 2L2 (Ω×]0,t[) exp t(1 + CL∞ (Ω×]0,T [) ) .
β β
(7.3.31)
If in addition (7.3.26) holds, then
1
β z(t)2L2 (Ω) ≤ β z0 2L2 (Ω) + f 2L2 (Ω×]0,t]) ,
2c0
1
β z(t)2L2 (Ω) + c0 z2L2 (Ω×]0,t[) ≤ β z0 2L2 (Ω) + f 2L2 (Ω×]0,t]) , (7.3.32)
c0
β z(t)L2 (Ω) ≤ β z0 L2 (Ω) + f L1 (0,t;L2 (Ω)d ) .
If μ > 0,
1
β z(t)2L2 (Ω) ≤ β z0 2L2 (Ω) + f 2L2 (Ω×]0,t]) .
2μ
7.3.3.1 The Case of a System: Uniqueness
Consider one line of (7.3.23) with zero data,
∂zi
d
β + Ci,j zj + α u · ∇ zi = 0 in Ω×]0, T [, zi (0) = 0 in Ω.
∂t j=1
First, we observe that the higher regularity of u allows to sharpen Lemma 7.3.3 and
its corollary. We have instead,
368 7 Appendix
Lemma 7.3.10 Let Ω be a bounded Lipschitz domain of IRd , let z and u be given,
respectively, in L ∞ (0, T ; L 2 (Ω)) and L 2 (0, T ; W 1,∞ (Ω)d ). With the notation of
Lemma 7.3.3, there exists a constant C independent of z and u such that for all
r with 1 ≤ r ≤ R and all ε ∈]0, 1], we have
u·∇(zρε,r )−(u·∇ z)ρε,r )L2 (Ωr ×]0,T [) ≤ C∇ uL2 (0,T ;L∞ (Ω)d×d ) zL∞ (0,T ;L2 (Ω)) .
Corollary 7.3.11 Under the assumptions of Lemma 7.3.10, we have for all r with
1 ≤ r ≤ R,
lim u · ∇(z ρε,r ) − (u · ∇ z) ρε,r L2 (Ωr ×]0,T [) = 0. (7.3.33)

ε→0
We skip the proofs because they are an easy exercise considering that (7.3.12) and
(7.3.13) are replaced by
T
2
z(y, t)divy u(y, t)ρε,r (x − y) dy dx dt
0 Ωr IRd
≤ div uL2 (0,T ;L∞ (Ω)) zL∞ (0,T ;L2 (Ω)) ,
respectively,

T 1 x−y 2
z(y, t)(u(x, t) − u(y, t)) · ∇ ρr ( + t r ) dy dx dt
Ωr ε ε
d+1
0 IRd
T
≤ Cr z(·, t)L2 (Ω) ∇ u(·, t)L∞ (Ω) dt.
0
Similarly, instead of Lemma 7.3.5, we have

Lemma 7.3.12 There exists an integer k0 such that for all f in L 2 (Ω×]0, T [)
extended by zero outside
Ω, for all k ≥ k0 , and for all r, 1 ≤ r ≤ R, the sup-
port of the function f ψr ρ 1k ,r is contained in Ωr . Furthermore

R

lim f ψr ρ 1k ,r − f L2 (Ω×]0,T [) = 0. (7.3.34)
k→∞
r=1
d
Therefore, it remains to examine the regularizing effect of ρ 1k ,r on j=1 Ci,j zj .
To this end, we revert to the proof of Lemma 7.3.6. Considering that
∂zi d
β =− Ci,j zj − α u · ∇ zi ,
∂t j=1
we can write
∂ d
β (zi )k + Ci,j (zj )k + α u ·∇ (zi )k
∂t j=1
R

=α − (u ·∇ zi )ψr ρ 1k ,r + u ·∇ zi ψr ρ 1k ,r
r=1
d

R

+ − (Ci,j zj ψr ) ρ 1k ,r + Ci,j (zj ψr ) ρ 1k ,r .
r=1 j=1
It suffices to study the second sum in the right-hand side of this identity since the
first sum is the same as in Lemma 7.3.6. Set
R

Ji,j = − (Ci,j zj ψr ) ρ 1k ,r + Ci,j (zj ψr ) ρ 1k ,r .
r=1
We have
R

Ji,j ≤ (Ci,j zj ψr ) ρ 1k ,r − Ci,j zj
L 2 (Ω×]0,T [)
r=1

R

+ Ci,j L∞ (Ω×]0,T [) zj − (zj ψr ) ρ 1k ,r L2 (Ω×]0,T [) .
r=1
Therefore, Lemma 7.3.12 implies that
R

lim − (Ci,j zj ψr ) ρ 1k ,r + Ci,j (zj ψr ) ρ 1k ,r L2 (Ω×]0,T [) = 0.
k→∞
r=1
These results yield the next theorem.
Theorem 7.3.13 Let Ω ⊂ IRd be bounded and Lipschitz-continuous. For any u

in L 2 (0, T ; W ∩ W 1,∞ (Ω)d ), α = 0, β > 0, f ∈ L 2 (Ω×]0, T [)d , and C ∈
L ∞ (Ω×]0, T [)d×d , (7.3.23) has exactly one solution z in L ∞ (0, T ; L 2 (Ω)d ). This
solution satisfies the bound (7.3.31) for almost all t in [0, T ],
1
2
β z(t)2L2 (Ω) ≤ β z0 2L2 (Ω) + f 2L2 (Ω×]0,t[) exp t(1 + CL∞ (Ω×]0,T [) ) .
β β
If in addition (7.3.26) holds, then the solution satisfies the bounds (7.3.32).
370 7 Appendix
7.3.4 Extension to a System in Lp
We study again the time-dependent system (7.3.23) in the setting of Section 7.3.3,
but with data f in L 2 (0, T ; L p (Ω)d ) and z0 in L p (Ω)d for some real number p > 2.
As p > 2, we know from Theorem 7.3.13 that (7.3.23) has a unique solution
in L ∞ (0, T ; L 2 (Ω)d ), and we propose to establish that this solution belongs to
L ∞ (0, T ; L p (Ω)d ), assuming that 2 ≤ d ≤ 4. For this, we can use the semi-discrete
formulation (7.3.24). Indeed, considering that f n belongs to L p (Ω)d for all integers
1 ≤ n ≤ N, and applying Theorem 7.2.16, a proof by induction on n shows that zn
belongs to L p (Ω)d . Moreover, zn is bounded as follows.
Lemma 7.3.14 Let Ω be a bounded Lipschitz domain of IRd , 2 ≤ d ≤ 4. Assume

that f ∈ L 2 (0, T ; L p (Ω)d ) and z0 ∈ L p (Ω)d for some p > 2. Define k0 by
β
k0 = . (7.3.35)
CL∞ (Ω×]0,T [)
Assume that the time step k < k0 , and set
k
k̃ = CL∞ (Ω×]0,T [) .
β
Then we have for 1 ≤ n ≤ N,
1
tn
β zn Lp (Ω) ≤ β z0 Lp (Ω) +f L1 (0,t n ;Lp (Ω)d ) exp k0 . (7.3.36)
1 − k̃ 1 − k̃
If in addition, C satisfies (7.3.26), then the bound for zn is the following extension of
(7.3.30):

n
β z
n
L p (Ω) + c0 kzi Lp (Ω) ≤ β z0 Lp (Ω) + f L1 (0,t n ;Lp (Ω)d ) . (7.3.37)
i=1
Proof We argue by induction on n. By assumption z0 belongs to L p (Ω)d . Then, for

some integer n with 1 ≤ n ≤ N, we assume that zn−1 ∈ L p (Ω)d . It follows that
(7.3.24) is a steady system of the form (7.2.24) with right-hand side
β n−1
g = fn + z ∈ L p (Ω)d ,
k
and matrix
β
D= I + C n ∈ L ∞ (Ω)d×d .
k
Therefore, Theorem 7.2.16 implies that zn belongs to L p (Ω)d and by induction, this
is valid for all n. To derive the bound (7.3.36), we rewrite (7.3.24) as
β n β
z + α [∇ zn ]un = f n + zn−1 + C n zn ,
k k
β
i.e., with matrix I, and bound the above right-hand side by
k
β n−1 β
f n + z + C n zn Lp (Ω) ≤ f n Lp (Ω) + zn−1 Lp (Ω) + CL∞ (Ω×]0,T [) zn Lp (Ω) .
k k
β
Therefore, (7.2.33) with c0 = yields
k
β n β
z Lp (Ω) ≤ f n Lp (Ω) + zn−1 Lp (Ω) + CL∞ (Ω×]0,T [) zn Lp (Ω) . (7.3.38)
k k
From here, (7.3.36) follows easily by summing over n, assuming that k < k0 with
k0 defined by (7.3.35), dividing by 1 − k̃, applying a discrete Gronwall’s inequality,
and observing that
n
k f i Lp (Ω) = f L1 (0,t n ;Lp (Ω)d ) .
i=1
When C satisfies (7.3.26), then C n satisfies (7.3.27) for all n, and D satisfies (7.2.25)
with constant
β
d0 = + c0 .
k
Therefore, (7.3.38) is replaced by
β n β
z Lp (Ω) + c0 zn Lp (Ω) ≤ f n Lp (Ω) + zn−1 Lp (Ω) .
k k
This readily implies (7.3.37).
From the previous section, we know that zn tends weakly star in L ∞ (0, T ; L 2 (Ω)d )
to the unique solution z of (7.3.23). The estimate (7.3.36), uniform for all k < k0 ,
proves that zn also converges weakly star in L ∞ (0, T ; L p (Ω)d ). Then, the uniqueness
of the limit implies that z is in L ∞ (0, T ; L p (Ω)d ) and we have the following extension
of Theorem 7.3.13:
Theorem 7.3.15 Let 2 ≤ d ≤ 4 and Ω ⊂ IRd be bounded and Lipschitz-continuous.

For any u in L 2 (0, T ; W ∩ W 1,∞ (Ω)d ), α = 0, β > 0, f ∈ L 2 (0, T ; L p (Ω)d ), and
C ∈ L ∞ (Ω×]0, T [)d×d , (7.3.23) has exactly one solution z ∈ L ∞ (0, T ; L p (Ω)d ).
This solution satisfies the following bound for almost all t in [0, T ]:
372 7 Appendix

t
β z(t)Lp (Ω) ≤ β z0 Lp (Ω) + f L1 (0,t;Lp (Ω)d ) exp CL∞ (Ω×]0,T [) . (7.3.39)
β
If in addition, C satisfies (7.3.26), then for almost all t in [0, T ],
β z(t)Lp (Ω) ≤ β z0 Lp (Ω) + f L1 (0,t;Lp (Ω)d ) . (7.3.40)
7.4 A Family of Riccati Integro-Differential Inequalities
In this short section we briefly recall the derivation of bounds for the positive solutions
of a family of Riccati integro-differential inequalities.
Let T > 0, κ > 0, α > 0, β > 0, γ ≥ 0, μ > 0, and s ≥ 1 be given real numbers.
We consider the following integro-differential inequality a.e. in ]0, T [:
t
αζ (t) + μζ (t) − βζ (t)s − γ e−μκ(t−τ ) ζ (τ )dτ ≤ G1 (t) + G2 (t) , (7.4.1)
0
where G1 ∈ L 1 (0, T ) and G2 ∈ L ∞ (0, T ) are two given functions. Let ζ ∈ W 1,1 (0, T̃ )
be any nonnegative solution of (7.4.1) in an interval ]0, T [, 0 <
T ≤ T , with a given
nonnegative starting value ζ0 ,
ζ (0) = ζ0 . (7.4.2)
First, through a division by α and the change in variables

μ
ξ(t) = e α t ζ (t) ,
(7.4.1) is transformed into the equivalent integro-differential inequality

β − μ t(s−1) γ t −μ(κ− 1 )(t−τ )
ξ (t) − e α ξ(t) −
s
e α ξ(τ ) dτ
α α 0
1 μ
≤ e α t G1 (t) + G2 (t) , (7.4.3)

α
with the same starting value ξ(0) = ζ0 . Under the above assumptions, ξ ∈ W 1,1 (0,
T)
is a nonnegative solution of (7.4.3). The next theorem gives an upper bound for ξ .
Theorem 7.4.1 Let ακ > 1. For any positive number M there exists a positive time
T ≤ T depending only on T , α, β, γ , κ, μ, s, G1 L1 (0,T ) , and G2 L∞ (0,T ) , but not
on
T , with
∀t ∈ [0, T ] , ξ(t) < ζ0 + M. (7.4.4)
7.4 A Family of Riccati Integro-Differential Inequalities 373
Proof Let us choose a real number M > 0. Since ξ(0) = ζ0 and M > 0, (7.4.4)
is satisfied at initial time, and as ξ is continuous in time, there exists a nontrivial
interval where (7.4.4) still holds. Let us prove that the length of this interval depends
only on the parameters of (7.4.1) and not on T . To this end, let T > 0 satisfy
T s−1 γ 1
ζ0 + M β ζ0 + M +
α μ κ − α1
μ
1 1
+ eαT G1 L1 (0,T ) + G2 L∞ (0,T ) ≤ M. (7.4.5)

α μ
There is no loss of generality in assuming that T ≤ T . On one hand, this inequality

holds by choosing T sufficiently small, and on the other hand, T is independent of
T̃ . We prove by contradiction that on this interval [0, T ], ξ satisfies (7.4.4). Indeed,
suppose that (7.4.4) holds for all t < T , but ξ(T ) = ζ0 + M. Then, by integrating
(7.4.3) over [0, T ] and applying (7.4.4) on t < T , we obtain the sequence of
inequalities, in view of (7.4.5):
T t
β s μ γ T 1
ξ(T ) − ζ0 < ζ0 + M e− α t(s−1) dt +
ζ0 + M e−μ(κ− α )(t−τ ) dτ dt
α 0 α 0 0
T
1 μ T μ

+ e α G1 L1 (0,T ) + G2 L∞ (0,T ) e α t dt

α 0
β s
≤ ζ0 + M T
α
γ 1 1 1 μ 1 μ
+ ζ0 + M T + e α T G1 L1 (0,T ) + G2 L∞ (0,T ) e α T − 1
α μκ− α 1 α μ
T γ 1
s−1
< ζ0 + M β ζ0 + M +
α μ κ − α1

μ 1 1
+ eαT G1 L1 (0,T ) + G2 L∞ (0,T ) ≤ M ,

α μ
whence the strict inequality at time T , thus proving (7.4.4).
An easy variant treats the case ακ ≤ 1. Finally, by reverting to ζ , we obtain

μ
∀t ∈ [0, T ] , ζ (t) < e− α t ζ0 + M). (7.4.6)
List of Notations
a acceleration, a = v̇ 25
A angular velocity tensor 25
An nth Rivlin–Ericksen tensor, (2.3.22) 22
∇
A upper-convected Oldroyd derivative, (2.3.27) 23
α1 , α2 material moduli 181
α α = α1 = −α2 31
b specific body force, (2.3.42) 27
β3 material modulus for grade-three fluid, (2.4.20) 35
B(0; R) ball centered at the origin with radius R 240
C, B Cauchy–Green tensors, (2.3.16), (2.3.17) 21
C t , Bt (x, τ ) relative Cauchy–Green tensors, (2.3.20), (2.3.21) 22
C m,σ (Ω) Hölder spaces of continuous functions 95
vC m,σ (Ω) norm of C m,σ (Ω) 144
[g]3/2 , [g]1,σ broken norms, (5.4.5) 209
C −1,σ (Ω) subspace of the dual of C 1,σ (Ω) satisfying (3.1.2) 230
curl v 2D curl, ∂v 2
∂x1
∂v1
− ∂x2
23
C∞ constant of (5.2.8) 183
d(x) distance from x to ∂Ω 223
D symmetric velocity gradient, (2.3.23) 22
D(Ω) C ∞ (Ω) with compact support in Ω 93
D (Ω) distributions in Ω 93
div v divergence of v, (2.3.9) 20
∂ϕ
∂t
local time derivative, (2.3.5) 20
ϕ̇, dϕ
dt
material time derivative, (2.3.6) 20
∂vi
∇v gradient tensor of a vector (∇ v)ij = ∂x j
, see L 98
Δ Laplace operator, Δ = grad div 101
Δp p-Laplacian, (4.2.15) 121

DOI 10.1007/978-3-319-39330-8
376 List of Notations
Δ(x) regularized distance, (5.5.10) 222

∂Ω boundary of the domain Ω 49
e Almansi–Hamel strain tensor, (2.3.19) 21
E Green–Saint Venant strain tensor, (2.3.18) 21
|·| Euclidian or Frobenius norm 41
ε(v) deformation rate tensor 106
FκR deformation gradient, (2.3.3) 19
Ft (x, τ ) relative deformation gradient, (2.3.14) 21
F:G product of tensors tr(FGT ) 93
F(D(u)) constitutive equation, F(D(u)) = μ(u) + gΨ (u) 150
γm−1 trace operator of order m − 1 103
γσ,α constant of (5.2.21) 187
γ̃σ,α constant of (5.3.2) 203
Γi connected components of ∂Ω 230
gc plasticity or stress threshold in Bingham fluids, (4.3.17) 163
H m (Ω) Sobolev space W m,2 (Ω) 94
vα equivalent to H 1 (Ω)d norm, (5.2.37) 192
((u, v))H m (Ω) scalar product of H m (Ω) 94
(u, v) scalar product of L 2 (Ω) 94
H01 (Ω) H 1 (Ω) with zero boundary condition 97
vH01 (Ω) equivalent norm of H01 (Ω), = |v|H 1 (Ω) , (3.1.19) 98
H −1 (Ω) dual space of H01 (Ω) 97
H −1 (Ω) norm of H −1 (Ω), (3.1.10) 97
Hτ1 (Ω) H 1 (Ω)d with zero normal boundary condition on ∂Ω, (3.1.12) 97
H(div, Ω) vectors in L 2 (Ω)d with divergence in L 2 (Ω) 98
Hτ (div, Ω) H(div, Ω) with zero normal component on ∂Ω 98
H(curl, Ω) vector in L 2 (Ω)3 with vector curl in L 2 (Ω)3 234
H(curl, Ω) vector in L 2 (Ω)2 with scalar curl in L 2 (Ω), (3.1.21) 98
H closure of V in L 2 (Ω)d , (4.2.20) 122
H 1 (a, b; X) H 1 in t with values in X 99
f H 1 (a,b;X) norm of H 1 (a, b; X) 99
I identity tensor 21
I1 , I 2 , I 3 principal invariants of A1 31
I1 , . . . , I9 invariants
of A1 , A2 51
j(v) Ω (D2 (v)) 1/2
dx, functional Bingham fluids, (4.2.91) 155
κR (B) reference configuration of body B 19
κt (B) current (at time t) configuration of body B 19
Kr constant of inf-sup condition, (3.2.10) 105
K̃ constant of inf-sup condition, (3.2.12) 105
K constant defined in (5.2.39) 193
λ, μ bulk and shear viscosities 32
L velocity gradient grad v, (2.3.8), see ∇ v 20
Lmr
(Ω) L r (Ω) with zero mean value, (2.3.24) 22
List of Notations 377
L r (a, b; X) L r in t with values in X 99

f Lr (a,b;X) norm of L r (a, b; X) 99
n unit exterior normal vector to ∂Ω 41
Oi open covering of Ω 100
Ωi Oi ∩ Ω 203
Ωε tubular ε neighborhood of ∂Ω, (5.5.14) in 3D 223
Ωε tubular ε neighborhood of ∂Ω, (5.6.16) in 2D 234
PV polar of V in X 105
Pm projection operator from H to Vm , (4.2.4) 118
ψ specific Helmholtz potential 37
Ψε G-differentiable approximation of Ψ , (4.2.33) 127
{wj }j≥0 special basis for Galerkin’s discretization, (4.2.44) 131
R rotation tensor, (2.3.15) 21
Re Reynolds number 53
R(u) monotone operator for grade-three fluid 328
κR density in configuration κR (B) 27
current density in current configuration κt (B) 27
ρ usual mollifier 222
ρε,r special mollifier, (7.2.7) 343
S extra stress tensor 28
S tempered distributions 94
S0,p Sobolev’s imbedding constant of H01 (Ω) in L p (Ω), (3.1.5) 96
S0,∞ Sobolev’s imbedding constant of W01,r (Ω) in L p (Ω), (3.1.6) 96
Sp Sobolev’s imbedding constant of H 1 (Ω) in L p (Ω), (3.1.7) 96
S∞ Sobolev’s imbedding constant of W 1,r (Ω) in L ∞ (Ω), (3.1.8) 96
S̃p Sobolev’s imbedding constant of Hτ1 (Ω) in L p (Ω)d , (3.1.13) 97
S̃∞ Sobolev’s constant of W 1,r (Ω)d ∩ Hτ1 ⊂ L ∞ (Ω)d , (3.1.14) 97
Uad space of admissible functions, (4.2.3) 118
ug lifting of boundary value g, (5.4.4) 209
ug,ε lifting of boundary value g supported by Ωε , (5.5.16) in 3D 223
ug,ε lifting of boundary value g supported by Ωε , (5.6.21) in 2D 234
um solutions of Galerkin’s system (4.2.27) 125
ut 0 initial velocity, (5.2.4) 182
T Cauchy stress tensor, (2.3.42) 27
t surface traction vector 41
τ unit tangent vector to ∂Ω 49
θε basic truncation function, (5.5.7) 22
U, V stretch tensors, (2.3.15) 21
v velocity, (2.3.2) 19
ω vorticity, curl v, (2.3.29) 23
V D(Ω)d with zero divergence, (3.1.15) 96
Vs closure of V in H s (Ω)d , (4.2.27) 125
V H01 (Ω)d with zero divergence, (3.1.16) 98
V⊥ orthogonal of V in H01 (Ω)d , (3.1.17) 98
378 List of Notations
V dual space of V 95
V norm of V , (3.1.19) 98
Vk,r (Ω) W0k,r (Ω)d with zero divergence, (3.2.24) 108
Vα V with α curl Δ v in L 2 (Ω)d , (5.2.12) 184
vV α graph norm of V α , (5.2.13) 184
((u, v)) scalar product of V α , (5.2.14) 185
Vm m-dimensional subspace of H spanned by w1 , . . . , wm 131
W spin tensor, (2.3.24) 22
W Hτ1 (Ω) with zero divergence, (3.1.18) 98
W m,r (Ω) Sobolev space 94
|v|W m,r (Ω) seminorm of Sobolev space 26
v (Ω)
W
m,r norm of Sobolev space 26
v 1,p
equivalent norm of W0 (Ω)d , (4.2.28) 125
p
W m+s,r (Ω) fractional Sobolev space, 0 < s < 1 11
vW m+s,r (Ω) norm of fractional Sobolev space 11
((u, v))H m+s (Ω) scalar product of H m+s (Ω), (3.1.1) 95
W01,r (Ω) W 1,r (Ω) with zero boundary condition 12

W −1,r (Ω) dual space of W01,r (Ω), 1/r + 1/r = 1 95
Wp closure of V in W 1,p (Ω)d , (4.2.27) 125
W (0, T ; X) (3.3.8) 114
Wα W with αcurl Δ v in L 2 (Ω)d , (5.4.2) 208
vW α norm of W α , (5.4.3) 209
zu graph norm of Xu , (7.2.3) 341
χε effective truncation function, (5.5.12) 223
χ κR motion of a body, (2.3.1) 19
χt relative deformation, (2.3.13) 21
ξ rate of dissipation, (2.4.23) 27
Xu L 2 (Ω) with u · ∇ z in L 2 (Ω), (7.2.2) 341
z auxiliary variable curl(u − α Δ u), (5.2.7) 183
References
1. Oxford English Dictionary. Clarendon Press, Oxford, 2000. Prepared by J. A. Simpson and
E. S. C. Weiner.
2. H. Abboud and T. Sayah. Upwind discretization of a time-dependent two-dimensional grade-
two fluid model. Comp. Math. with Appl., 57(8):1249–1264, 2009.
3. R. A. Adams and J. J. Fournier. Sobolev Spaces, volume 140 of Pure and Applied Mathematics.
Academic Press, New-York, London, 2003.
4. M. Amara, C. Bernardi, V. Girault, and F. Hecht. Regularized finite element discretizations
of a grade-two fluid model. Internat. J. Numer. Meth. Fluids, 48(12):1375–1414, 2005.
5. C. Amrouche. Étude Globale des Fluides de Troisième Grade. PhD thesis, Université Pierre
et Marie Curie, Paris VI, France, 1986.
6. C. Amrouche, C. Bernardi, M. Dauge, and V. Girault. Vector potentials in three-dimensional
non-smooth domains. Math. Methods Appl. Sci., 21(9):823–864, 1998.
7. C. Amrouche and D. Cioranescu. On a class of fluids of grade three. Technical Report
RR88006, Laboratoire d’Analyse Numérique, Université Pierre et Marie Curie, Paris VI,
1988.
8. C. Amrouche and D. Cioranescu. On a class of fluids of grade three. Internat. J. Non-Linear
Mech., 32(1):73–88, 1997.
9. C. Amrouche and V. Girault. Decomposition of vector spaces and application to the Stokes
problem in arbitrary dimensions. Czech. Math. Journal, 44(1):109–140, 1994.
10. C. Amrouche, V. Girault, and J. Giroire. Weighted Sobolev spaces for Laplace’s equation in
Rn . J. Math. Pures Appl., 73(6):579–606, 1994.
11. E. C. Andrade. Viscosity of liquids. Nature, 125:309–310, 1930.
12. D. N. Arnold, L. R. Scott, and M. Vogelius. Regular inversion of the divergence operator
with Dirichlet boundary conditions on a polygon. Ann. Scuola Norm. Sup. Pisa, Cl. Sci. IV,
15(2):169–192, 1988.
13. J. P. Aubin. Un théorème de compacité. C.R. Acad. Sci. Paris Sér. I Math., 256:5042–5044,
1963.
14. K. Aziz and G. W. Govier. The Flow of Complex Mixtures in Pipes. Technology and Engi-
neering, 792. Van Nostrand Reinhold, New York, NY, 1972.
15. I. Babuška. The finite element method with Lagrangian multipliers. Numer. Math., 20:179–
192, 1972/73.
16. R. L. Bagley. Applications of Fractional Derivatives to Viscoelasticity. PhD thesis, Air Force
Inst. of Tech., USA, 1979.

DOI 10.1007/978-3-319-39330-8
380 References
17. R. L. Bagley and P. J. Torvik. On the fractional calculus model of viscoelastic behavior. J.
Rheology, 30(1):133–155, 1986.
18. S. Bair and P. Kottke. Pressure-viscosity relationships for elastohydrodynamics. Tribology
Trans., 46(3):289–295, 2003.
19. S. Bair and W. O. Winer. The high pressure high shear stress rheology of liquid lubricants. J.
Tribology, 114(1):1–9, 1992.
20. H. A. Barnes. Thixotropy – a review. J. Non-Newtonian Fluid Mech., 70:1–33, 1997.
21. C. Barus. Isothermals, isopiestics and isometrics relative to viscosity. American J. Sci.,
45(255):87–96, 1983.
22. D. Begis. Analyse Numérique de l’Écoulement d’un Fluide de Bingham. PhD thesis, Université
Pierre et Marie Curie, Paris VI, France, 1972.
23. H. Beirao Da Veiga. Existence results in Sobolev spaces for a stationary transport equation.
Ricerche Mat., Suppl., 36:173–184, 1987.
24. J. F. Bell. The experimental Foundations of Solid Mechanics. In C. Truesdell, editor, Mechan-
ics of Solids, Vol. I, Handbuk der Physik. Springer-Verlag, Berlin, 1973.
25. J. T. Bendler, J. J. Fontanella, and M. F. Schlesinger. A new Vogel-like law: ionic conduc-
tivity, dielectric relaxation and viscosity near the glass transition. Technical report, Physics
Department, U.S. Naval Academy, Annapolis, MD 21402–5026, USA, 2001.
26. A. N. Beris and B. J. Edwards. Thermodynamics of Flowing Systems with Internal Microstruc-
ture, volume 36 of Oxford Engineering Science. Oxford University Press, New York, NY,
1994.
27. R. Berker. Contrainte sur une paroi en contact avec un fluide visqueux classique, un fluide de
Stokes, un fluide de Coleman-Noll. C.R. Acad. Sci. Paris Sér. I, 258:5144–5147, 1964.
28. J. M. Bernard. Fluides de Second et Troisième Grade en Dimension trois: Solution Globale
et Régularité. PhD thesis, Université Pierre et Marie Curie, Paris VI, France, 1998.
29. J. M. Bernard. Stationary problem of second-grade fluids in three dimensions: existence,
uniqueness and regularity. Math. Methods Appl. Sci., 22(8):655–687, 1999.
30. J. M. Bernard. Weak and classical solutions of equations of motion for third grade fluids.
M2AN Math. Mod. Numer. Anal., 33(6):1091–1120, 1999.
31. J. M. Bernard. Solutions W 2,p , p > 3, for second grade fluid equations with a boundary of
class C 1,1 . Comm. Appl. Nonlinear Anal., 9(1):1–29, 2002.
32. J. M. Bernard. Problem of second grade fluids in convex polyhedrons. SIAM J. Math. Anal.,
44(3):2018–2038, 2012.
33. J. M. Bernard and E. H. Ouazar. Stationary problem of third grade fluids in two and three
dimensions: existence and uniqueness. Internat. J. Non-linear Mech., 40(5):603–620, 2005.
34. C. Bernardi, M. Dauge, and Y. Maday. Polynomials in the Sobolev world. Technical report,
Laboratoire Jacques-Louis Lions, Paris VI, France, 2007.
35. B. Bernstein and R. L. Fosdick. Non uniqueness of second-order fluids under steady radial
flow in annuli. Internat. J. Engrg. Sci., 7(6):555–569, 1969.
36. B. Bernstein, E. A. Kearsley, and L. J. Zapas. A study of stress-relaxation with finite strain.
Trans. Soc. Rheology, 7(1):391–410, 1963.
37. E. C. Bingham. Fluidity and Plasticity. McGraw-Hill, New York, NY, 1922.
38. E. C. Bingham and R. A. Stephens. The alternating stress method for the measurement of the
consistency of very stiff materials. J. Appl. Phys., 5(8):217–220, 1934.
39. R. B. Bird, R. C. Armstrong, and O. Hassager. Dynamics of Polymeric Liquids, 2nd Edition,
volume 1 Fluid Mechanics. John Wiley, New York, NY, 1987.
40. B. Birkhoff and E. H. Zarantanello. Jets, Wakes and Cavities, volume 2 of Applied Mathematics
and Mechanics. Academic Press, New York, NY, 1957.
41. G. W. S. Blair, B. C. Veinoglou, and J. E. Caffyn. Limitations of the Newtonian time scale in
relation to non-equilibrium rheological states and the theory of quasi-properties. Proc. Roy.
Soc. London Series A, 189(1016):69–87, 1947.
42. M. E. Bogovskii. Solution of some vector analysis problems connected with operators div
and grad. In Trudy Seminar, N. L. Sobolev, volume 80, pages 5–40. Akademia Nauk SSSR,
Sibirskoe Otdelenie Matematiki, Novosibirsk, 1979. in Russian.
References 381
43. M. E. Bogovskii. Solution of the first boundary value problem for the equation of continuity
of an incompressible medium. Soviet Math. Dokl., 20(5):1094–1098, 1979.
44. L. Boltzmann. Zut Theorie der elastichen Nachwirkung. Sitzgsber. Akad. Wiss. Wien, 70:275–
306, 1874.
45. S. Boujena. Étude d’une Classe de Fluides Non Newtoniens, les Fluides Newtoniens Général-
isés. PhD thesis, Université Pierre et Marie Curie, Paris VI, France, 1986.
46. P. Bourgin and J. A. Tichy. The effect of an additional boundary condition on the plane
creeping flow of a second-order fluid. Internat. J. Non-linear Mech., 24(6):561–569, 1989.
47. J. H. Bramble. A proof of the inf-sup condition for the Stokes equations on Lipschitz domains.
Math. Models Methods Appl. Sci., 13(3):361–371, 2003.
48. D. Bresch and J. Lemoine. Stationary solutions for second-grade fluids equations. Math.
Models Methods Appl. Sci., 8(5):737–748, 1998.
49. H. Brezis. Inéquations variationnelles relatives à l’opérateur de Navier-Stokes. J. Math. Anal.
Appl., 19:159–165, 1972.
50. F. Brezzi. On the existence, uniqueness and approximation of saddle-point problems arising
from Lagrangian multipliers. Rev. Française Automat. Informat. Recherche Opérationelle Sér.
Rouge, 8(R2):129–151, 1974.
51. P. W. Bridgman. The effect of pressure on the viscosity of forty-three pure liquids. Proc. Am.
Acad. Arts Sci., 61(3):57–99, 1926.
52. P. W. Bridgman. Physics of High Pressure. Dover Publications, New York, NY, 1931.
53. M. Bulíček, P. Gwiazda, J. Málek, K. R. Rajagopal, and A. Świerczewska-Gwiazda. On flows
of fluids described by an implicit constitutive equation characterized by a maximal monotone
graph. In Mathematical aspects of fluid mechanics, volume 402 of London Math. Soc. Lecture
Note Ser., pages 23–51. Cambridge Univ. Press, Cambridge, 2012.
54. M. Bulíček, P. Gwiazda, J. Málek, and A. Świerczewska-Gwiazda. On unsteady flows of
implicitly constituted incompressible fluids. SIAM J. Math. Anal., 44(4):2756–2801, 2012.
55. J. M. Burgers. Mechanical Consideration – Model Systems – Phenomenological Theories of
Relaxation and Viscosity. First Report on Viscosity and Plasticity. Nordemann, 2nd ed., New
York, NY, 1939.
56. A. V. Busuioc. Un Regard Mathématique sur les Fluides Non Newtoniens. Habilitation à
Diriger des Recherches. Université de Saint-Etienne, 2014.
57. A. V. Busuioc and D. Iftimie. Global existence and uniqueness of solutions for the equations
of third grade fluids. Internat. J. Non-linear Mech., 39(1):1–12, 2004.
58. A. V. Busuioc and T S. Ratiu. The second grade fluid and averaged Euler equations with
Navier-slip boundary conditions. Nonlinearity, 16(3):1119–1149, 2003.
59. L. Cattabriga. Su un problema al contorno relativo al sistema di equazioni di Stokes. Rend.
Sem. Mat. Univ. Padova, 31:308–340, 1961.
60. W. D. Chang, N. D. Kazarinoff, and C. Q. Lu. A new family of explicit solutions for the
similarity equations modelling flow of a non-Newtonian fluid over a stretching sheet. Arch.
Rational Mech. Anal., 113(2):191–195, 1990.
61. P. G. Ciarlet and P. A. Raviart. A mixed finite element method for the biharmonic equation. In
C. DeBoor, editor, Mathematical Aspects of Finite Elements in Partial Differential Equations,
pages 125–145, New York, NY, 1974. Academic Press.
62. D. Cioranescu. Sur une classe de fluides non-newtoniens. Appl. Math. Optim., 3(2–3):263–
282, 1977.
63. D. Cioranescu and V. Girault. Weak and classical solutions of a family of second grade fluids.
Internat. J. Non-Linear Mech., 32(2):317–335, 1997.
64. D. Cioranescu and E. H. Ouazar. Existence and uniqueness for fluids of second grade. In
Nonlinear Partial Differential Equations and their Applications, Collège de France Seminar
Vol. VI, volume 109, pages 178–197. Pitman, Boston, MA, 1984.
65. D. Cioranescu and E. H. Ouazar. Existence et unicité pour les fluides de second grade. C.R.
Acad. Sci. Paris Sér. I Math., 298(13):285–287, 1984.
66. E. A. Coddington and N. Levinson. Theory of Ordinary Differential Equations. McGraw-Hill,
New York, NY, 1955.
382 References
67. B. D. Coleman, H. Markovitz, and W. Noll. Viscometric Flows of Non-Newtonian Fluids:

Theory and Experiment, volume 5 of Springer Tracts in Natural Philosophy. Springer, Berlin,
1966.
68. B. D. Coleman and W. Noll. An approximation theorem for functionals, with applications in
continuum mechanics. Arch. Rational Mech. Anal., 6:355–370, 1960.
69. B. D. Coleman and W. Noll. The thermodynamics of elastic materials with heat conduction
and viscosity. Arch. Rational Mech. Anal., 13:167–178, 1963.
70. M. Costabel and M. Dauge. On some regularity results in corner domains. Private communi-
cation.
71. W. D. Criminale, J. L. Ericksen, and G. L. Fibley. Steady shear flow of non-Newtonian fluids.
Arch. Rational Mech. Anal., 1:410–417, 1958.
72. W. G. Cutler, R. H. McMickle, W. Webb, and R. W. Schiessler. Study of the compression of
several high molecular weight hydrocarbons. J. Chem. Phys., 29(4):727–740, 1958.
73. M. Dauge. Stationary Stokes and Navier-Stokes systems on two or three-dimensional domains
with corners. I. Linearized equations. SIAM J. Math. Anal., 20(1):74–97, 1989.
74. M. Dauge. Neumann and mixed problems on curvilinear polyhedra. Integral Equ. Operator
Theory, 15(2):227–261, 1992.
75. P. R. de Souza-Mendes. Modeling the thixotropic behavior of structured fluids. J. Non-
Newtonian Fluid Mech., 164(1):66–75, 2009.
76. P. R. de Souza-Mendes. Thixotropy elasto-viscoplastic model for structured fluids. Soft Mat-
ter, 7(6):2471–2483, 2011.
77. P. R. de Souza-Mendes, K. R. Rajagopal, and R. L. Thomson. A thermodynamic framework
to model thixotropic materials. Internat. J. Non-Linear Mech., 55:48–54, 2013.
78. P. R. de Souza-Mendes and R. L. Thomson. A critical overview of elasto-viscoplastic
thixotropic modeling. J. Non-Newtonian Fluid Mech., 187–188:8–15, 2012.
79. P. R. de Souza-Mendes and R. L. Thomson. A unified approach to model elasto-viscoplastic
thixotropic yield stress materials and apparent yield stress fluids. Theological Acta, 52(7):673–
694, 2013.
80. Deborah. The Soncino Edition of the Pentateuch and Haftorahs. J. H. Hertz, London, UK,
1960.
81. B. Desjardins. A few remarks on ordinary differential equations. Comm. Partial Diff. Equ.,
21(11–12):1667–1703, 1996.
82. R. J. Di Perna and P. L. Lions. Ordinary differential equations, transport theory and Sobolev
spaces. Invent. Math., 98(3):511–547, 1989.
83. M. Doi and S. F. Edwards. The Theory of Polymer Dynamics, volume 73. Oxford University
Press, Oxford, UK, 1988.
84. D. Dowson and G. R. Higginson. Elastohydrodynamic Lubrication: the Fundamentals of
Roller and Gear Lubrication. Pergamon Press, London, UK, 1966.
85. P. L. G. Du Buat. Principes d’hydraulique et de pyrodynamique vérifiés par un grand nombre
d’expériences faites par ordre du gouvernement; ouvrage en trois volumes. Didot, Paris,
France, 1816.
86. R. Dugas. A History of Mechanics. Dover Publications, New York, 1988.
87. J. E. Dunn and R. L. Fosdick. Thermodynamics, stability, and boundedness of fluids of
complexity 2 and fluids of second grade. Arch. Rational Mech. Anal., 56:191–252, 1974.
88. J. E. Dunn and K. R. Rajagopal. Fluids of differential type: Critical review and thermodynamic
analysis. Internat. J. Engrg. Sci., 33(5):689–729, 1995.
89. T. F. Dupont and I. Mogultay. A symmetric error estimate for Galerkin approximations of
time-dependent Navier-Stokes equations in two-dimensions. Math. of Comp., 78(268):1919–
1927, 2009.
90. R. Durán and A. Muschietti. An explicit right inverse of the divergence operator which is
continuous in weighted norms. Studia Math., 148(3):207–219, 2001.
91. G. Duvaut and J. L. Lions. Les Inéquations en Mécanique et en Physique. Travaux et
Recherches Mathématiques, 21. Dunod, Paris, France, 1972.
References 383
92. C. Eckart. The thermodynamics of irreversible processes IV. The theory of elasticity and
anelasticity. Phys. Rev., 73(4):373–382, 1948.
93. R. Edgeworth, B. J. Dalton, and T. Parnell. The pitch drop experiment. Eur. J. Phys., 5:198–
200, 1984.
94. J. L. Ericksen. Overdetermination of the speed in rectilinear motion of non-Newtonian fluids.
Quart. Appl. Math., 14:318–321, 1956.
95. L. Euler. De progressionibus transcentibus sev, quarum termini algebraice davi nequeunt.
Comment. Acad. Sci. Imperialis Petropolitanae, 5:38–57, 1730.
96. V. M. Falkneb and S. W. Skan. Solutions of the boundary-layer equations. Phil. Magazine,
12(80):865–896, 1931.
97. E. Fernández-Cara, F. Guillén-González, and R. R. Ortega. Mathematical Modeling and
Analysis of Viscoelastic Fluids of the Oldroyd Kind. In P. G. Ciarlet and J. L. Lions, editors,
Handbook of Numerical Analysis, Vol. VIII, pages 543–661. North-Holland, Amsterdam, The
Netherlands, 2002.
98. P. J. Flory and M. Volkenstein. Statistical Mechanics of Chain Molecules. Biopolymers. Wiley,
New York, NY, 1969.
99. R. L. Fosdick and K. R. Rajagopal. Uniqueness and drag for fluids of second grade in steady
motion. Internat. J. Non-Linear Mech., 13(3):131–137, 1978.
100. R. L. Fosdick and K. R. Rajagopal. Anomalous features in the model of “Second order fluids".
Arch. Rational Mech. Anal., 70(2):145–152, 1979.
101. R. L. Fosdick and K. R. Rajagopal. Thermodynamics and stability of fluids of third grade. In
Proc. Roy. Soc. London Series A: Math. Phys. Engrg. Sci., volume 369, pages 351–377, 1980.
102. R. L. Fosdick and J. Serrin. Rectilinear steady flows of simple fluids. In Proc. Roy. Soc.
London Series A: Math. Phys. Engrg. Sci., volume 332, pages 311–333, 1973.
103. R. L. Fosdick and C. Truesdell. Universal flows in the simplest theories of fluids. Ann. Scuola
Norm. Sup. Pisa, Cl. Sci. IV, 4(2):323–341, 1977.
104. J. Frehse and J. Málek. Problems due to the no-slip boundary condition in the incompressible
fluid dynamics. In S. Hildebrandt and H. Karcher, editors, Geometric Analysis and Nonlinear
Partial Differential Equations, pages 559–571. Springer-Verlag, Berlin, 2003.
105. A. Friedman. Foundations of Modern Analysis. Holt, Rinehart and Winston, New York NY,
1970.
106. H. Frohlick and R. Sack. Theory of the rheological properties of dispersion. In Proc. Roy.
Soc. London Series A: Math. Phys. Engrg. Sci., volume 185, pages 415–430, 1946.
107. G. P. Galdi. An Introduction to the Mathematical Theory of the Navier-Stokes Equations I:
Linearized Steady Problem, volume 28 of Springer Tracts in Natural Philosophy. Springer-
Verlag, New York, NY, 1994.
108. G. P. Galdi, M. Padula, and K. R. Rajagopal. On the conditional stability of the rest state of a
fluid of second grade in unbounded domains. Arch. Rational Mech. Anal., 109(2):173–182,
1990.
109. G. P. Galdi and K. R. Rajagopal. Slow motion of a body in a fluid of second grade. Internat.
J. Engrg. Sci., 35(1):33–54, 1997.
110. C. F. Gauss. On a New General Principle of Mechanics. Phil. Magazine, 8(44):117–120, 1830.
translation of Üeber ein neues allegemeines Grundgesetz der Mechanik, Journal für die reine
und angewandte Mathematik 1829 4 (2009), pp. 232–235.
111. A. Gemant. A method of analyzing experimental results obtained from elasto-viscous bodies.
J. Applied Phys., 7(8):311–317, 1936.
112. V. Girault and F. Hecht. Numerical methods for grade-two fluid models: finite-element dis-
cretizations and algorithms. In R. Glowinski and J. Xu, editors, Numerical Methods for Non-
Newtonian Fluids. Handbook of Numerical Analysis, XVI, pages 5–209. North-Holland, Am-
sterdam, The Netherlands, 2011.
113. V. Girault and J. L. Lions. Two-grid finite-element schemes for the transient Navier-Stokes
problem. M2AN Math. Model. Numer. Anal., 35(5):945–980, 2001.
114. V. Girault, R. H. Nochetto, and L. R. Scott. Maximum-norm stability of the finite element
Stokes projection. J. Math. Pures Appl., 84(3):279–330, 2005.
384 References
115. V. Girault and P. A. Raviart. Finite Element Approximation of the Navier-Stokes Equations,
volume 749 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 1979.
116. V. Girault and P. A. Raviart. Finite Element Methods for Navier-Stokes Equations: Theory
and Algorithms, volume 5 of Springer Series in Computational Mathematics. Springer-Verlag,
Berlin, Germany, 1986.
117. V. Girault and L. R. Scott. Analysis of a two-dimensional grade-two fluid model with a
tangential boundary condition. J. Math. Pures Appl., 78(10):981–1011, 1999.
118. V. Girault and L. R. Scott. Finite-element discretizations of a two-dimensional grade-two fluid
model. M2AN Math. Model. Numer. Anal., 35(6):1007–1053, 2001.
119. V. Girault and L. R. Scott. On a time-dependent transport equation in a Lipschitz domain.
SIAM J. Math. Anal., 42(4):1721–1731, 2010.
120. V. Girault and A. Sequeira. A well-posed problem for the exterior Stokes equations in two
and three dimensions. Arch. Rational Mech. Anal., 114(4):313–333, 1991.
121. V. Girault and L. Tartar. L p and W 1,p regularity of the solution of a steady transport equation.
C.R. Acad. Sci. Paris Sér. I Math., 348(15–16):885–890, 2010.
122. J. Giroire. Étude de Quelques Problèmes aux Limites Extérieurs et Résolution par Équations
Intégrales. PhD thesis, Université Pierre et Marie Curie, Paris VI, France, 1987.
123. S. Goldstein. Modern Development in Fluid Dynamics: an Account of Theory and Experiment
Relating to Boundary Layers, Turbulent Motions and Wakes. Dover Publications, New York,
NY, 1965.
124. D. L. Goodstein. States of Matter. Dover Books on Physics. Dover Phoenix Editions, New
York, NY, 1985.
125. A. E. Green and R. S. Rivlin. The mechanics of non-linear materials with memory. III. Arch.
Rational Mech. Anal., 4:387–404, 1960.
126. E. M. Griest, W. Webb, and R. W. Schiessler. Effect of pressure on viscosity of higher hydro-
carbons and their mixtures. J. Chem. Phys., 29(4):711–720, 1958.
127. P. Grisvard. Elliptic Problems inNonsmooth Domains, volume 24 of Monographs and Studies
in Mathematics. Pitman, Boston, MA, 1985.
128. J. Guzmán and D. Leykekhman. Pointwise error estimates of finite element approximations
to the Stokes problem on convex polyhedra. Math. Comp., 81(280):1879–1902, 2012.
129. J. S. Hadamard. Mémoire sur le problème d’analyse relatif à l’équilibre des plaques élastiques
encastrées. Mémoires des Savants Étrangers, 33(4):515–629, 1908.
130. G. Hamel. Spiralförmige Bewegungen zäher Flüssigkeiten. Jahresbericht der Deutschen
Mathematiker-Vereinigung, 25:34–60, 1917.
131. S. G. Hatzikiriakos and J. M. Dealy. Wall slip of molten high density polyethylene I. Sliding
plate rheometer studies. J. Rheology, 35(4):497–523, 1991.
132. E. Hopf. Über die Anfangswertaufgabe für die hydrodynamischen Grundgleichungen. Math.
Nachr., 4:213–231, 1951.
133. L. Hörmander. The Analysis of Linear Partial Differential Operators. III. Pseudodifferential
Operators. Springer-Verlag, Berlin, Germany, 1985.
134. J. Hron, C. Le Roux, J. Màlek, and K. R. Rajagopal. Flows of incompressible fluids subject
to Navier’s slip on the boundary. Comput. Math. Appl., 56(8):2128–2143, 2008.
135. W. J. Hrusa, J. A. Nohel, and M. Renardy. Mathematical Problems in Viscoelasticity. Pitman
Monographs and Surveys in Pure and Applied Mathematics, 35. Wiley, New York, NY, 1987.
136. Y. N. Huang and K. R. Rajagopal. On necessary conditions for the secondary flow of non
Newtonian fluids in straight tubes. Internat. J. Engrg. Sci., 32(8):1277–1281, 1994.
137. R. R. Huilgol. On uniqueness and non-uniqueness in the plane creeping flows of second order
fluids. SIAM J. Appl. Math., 24(2):226–233, 1973.
138. R. R. Huilgol. Continuum Mechanics of Viscoelatic Liquids. Wiley, New York, NY, 1975.
139. D. Iftimie. Remarques sur la limite α → 0 pour les fluides de grade deux. C.R. Acad. Sci.
Paris Sér. I Math., 334(1):83–86, 2002.
140. J. B. Irving and A. J. Barlow. An automatic high pressure viscometer. J. Phys. E: Scientific
Instruments, 4(3):232–236, 1971.
References 385
141. W. Jaunzemis. Continuum Mechanics. MacMillan Series in Continuum Mechanics. MacMil-

lan, New York, NY, 1967.
142. G. B. Jeffery. The two-dimensional steady motion of a viscous fluid. The London, Edinburgh,
and Dublin Phil. Mag. and J. Sci., 29(172):455–465, 1915.
143. K. L. Johnson and R. Cameron. Fourth Paper: Shear behavior of elastohydrodynamic oil films
at high rolling contact pressures. In Proc. Instit. Mech. Engrs., volume 182, pages 307–330,
1967.
144. K. L. Johnson and J. A. Greenwood. Thermal analysis of an Eyring fluid in elastohydrody-
namic traction. Wear, 61(2):355–374, 1980.
145. K. L. Johnson and J. L. Tevaarwerk. Shear behavior of elastohydrodynamic oil films. In Proc.
Roy. Soc. London Series A: Math. Phys. Engrg. Sci., volume 356, pages 215–236, 1977.
146. M. W. Johnson and D. A. Segalman. A model for viscoelastic fluid behavior which allows
non-affine deformation. J. Non-Newtonian Fluid Mech., 2(3):255–270, 1977.
147. A. S. Kaye. Non-newtonian flow in incompressible fluids, Part 1: A general rheological
equation of state; Part 2: Some problems in steady flow, Note 134. Technical report, College
of Aeronautics, Cranfield, UK, 1962.
148. R. B. Kellogg and J. E. Osborn. A regularity result for the Stokes problem in a convex polygon.
J. Functional Anal., 21:397–431, 1976.
149. W. Thomson Lord Kelvin and P. G. Tait. Treatise on Natural Philosophy, volume I. Cambridge
University Press, Cambridge, UK, 1867.
150. R. C. Koeller. Applications of fractional calculus to the theory of viscoelasticity. Trans. ASME
J. Appl. Mech., 51(2):299–307, 1984.
151. R. W. Kolkka, D. S. Malkus, M. G. Hansen, and G. R. Ierley. Spurt phenomena of the Johnson-
Segalman fluid and related models. J. Non-Newtonian Fluid Mech., 29:303–335, 1988.
152. A. M. Kraynik and W. R. Schowalter. Slip at the wall and extrudate roughness with aqueous
solutions of polyvinyl alcohol and sodium borate. J. Rheology, 25(1):95–114, 1981.
153. J. Murali Krishnan and K. R. Rajagopal. Review of the uses and modeling of bitumen from
ancient to modern times. Appl. Mech. Rev., 56(2):149–214, 2003.
154. J. Murali Krishnan and K. R. Rajagopal. Thermodynamic framework for the constitutive mod-
eling of asphalt concrete: Theory and applications. J. Materials in Civil Engrg., 16(2):155–
166, 2004.
155. O. A. Ladyzhenskaya. The Mathematical Theory of Viscous Incompressible Flow. Gordon
Breach, 1969.
156. P. D. Lax and A. N. Milgram. Parabolic equations. In Contributions to the Theory of Partial
Differential Equations, Annals of Mathematical Studies, pages 167–190, Princeton, NJ, 1954.
Princeton University Press.
157. C. Le Roux. Steady Stokes flows with threshold slip boundary conditions. Math. Models
Methods Appl. Sci., 15(8):1141–1168, 2005.
158. C. Le Roux, J. Málek, and K. R. Rajagopal. A review of boundary conditions. to be submitted
for publication, 2015.
159. G. W. Leibniz. Letter from Hanover, Germany, September 30, 1695 to G. A. L’Hospital. In
Leibnizen Mathematische Schriften, volume 2, pages 301–302. Olms Verlag, Hildescheim,
Germany, 1962. first published in 1849.
160. J. Leray. étude de diverses équations intégrales nonlinéaires et de quelques problèmes que
pose l’hydrodynamique. J. Math. Pures Appl., 12:1–82, 1933.
161. J. L. Lions. Equations Différentielles Opérationnelles et Problèmes aux Limites. Die
Grundlehren der Mathematischen Wissenschaften, Band 111. Springer-Verlag, Berlin, 1961.
162. J. L. Lions. Quelques Méthodes de Résolution des Problèmes aux Limites Non Linéaires.
Dunod, Paris, France, 1969.
163. J. L. Lions and E. Magenes. Problemi ai limiti non omogenei, V. Ann. Scuola Norm. Sup.
Pisa, 16:1–44, 1962.
164. J. L. Lions and E. Magenes. Problèmes aux Limites non Homogènes et Applications, I. Travaux
et Recherches Mathématiques, 17. Dunod, Paris, France, 1968.
386 References
165. J. Liouville. Mémoire sur quelques questions de géométrie et de mécanique, et sur un nouveau
genre de Calcul pour résoudre ces Questions. J. École Polytechnique, Paris, 13:1–69, 1832.
166. I. S. Liu. Method of Lagrange multipliers for exploitation of the entropy principle. Arch.
Rational Mech. Anal., 46:131–148, 1972.
167. A. S. Lodge. Elastic Liquids: an Introductory Vector Treatment of Finite-strain Polymer
Rheology. Academic Press, London, UK, 1964.
168. J. Málek, J. Nečas, M. Rokyta, and M. Ruzicka. Weak and Measure-Valued Solutions to Evolu-
tionary Partial Differential Equations, volume 13 of Applied Mathematics and Mathematical
Computation. Chapman and Hall, London, UK, 1996.
169. J. Málek and K. R. Rajagopal. Mathematical issues concerning the Navier-Stokes equations
and some of its generalizations. In C. M. Dafermos and D. Feireisl, editors, Handbook of
Differential Equations, Evolutionary Equations, Volume II, pages 371–459. Elsevier/North-
Holland, Amsterdam, The Netherlands, 2005.
170. J. Málek and K. R. Rajagopal. Mathematical properties of the solutions to the equations gov-
erning the flow of fluids with pressure and shear rate dependent viscosities. In S.J. Friedlander
and D. Serre, editors, Handbook of Mathematical Fluid Dynamics, Volume 4, pages 407–444.
Elsevier, Amsterdam, The Netherlands, 2007.
171. J. Málek, K. R. Rajagopal, and M. Ruzicka. Existence and regularity of solutions and the
stability of the rest state for fluids with shear dependent viscosities. Math. Models Methods
Appl. Sci., 5(6):789–812, 1995.
172. D. S. Malkus, J. A. Nohel, and B. J. Plohr. Dynamics of shear flow of a non-Newtonian fluid.
J. Comp. Phys., 87(2):464–487, 1990.
173. D. S. Malkus, J. A. Nohel, and B. J. Plohr. Analysis of new phenomenon in shear flow of
non-Newtonian fluids. SIAM J. Appl. Math., 51(4):899–929, 1991.
174. L. E. Malvern. Introduction to the Mechanics of a Continuous Medium. Prentice-Hall Inc.,
Englewood Cliffs, NJ, 1969.
175. D. Mansutti, G. Pontrelli, and K. R. Rajagopal. Non-similar flow of a non-Newtonian fluid
past a wedge. Internat. J. Engrg. Sci., 31(4):637–647, 1993.
176. D. Mansutti and K. R. Rajagopal. Flow of a shear-thinning fluid between intersecting planes.
Internat. J. Non-linear Mech., 26(5):769–775, 1991.
177. A. W. Marris. On steady three-dimensional motions. Arch. Rational Mech. Anal., 35:122–168,
1969.
178. A. W. Marris. Steady non-rectilinear complex-lamellar universal motions of a Navier-Stokes
fluid. Arch. Rational Mech. Anal., 41(5):354–362, 1971.
179. A. W. Marris. Steady rectilinear universal motions of a Navier-Stokes fluid. Arch. Rational
Mech. Anal., 48(5):379–396, 1972.
180. A. W. Marris. Steady universal motions of a Navier-Stokes fluid: the case when the velocity
magnitude is constant on a Lamb surface. Arch. Rational Mech. Anal., 50(2):99–110, 1972.
181. A. W. Marris and M. G. Aswani. On the general impossibility of controllable axi-symmetric
Navier-Stokes motions. Arch. Rational Mech. Anal., 63(2):107–153, 1977.
182. J. C. Maxwell. II. Illustrations of the dynamical theory of gases. The London, Edinburgh, and
Dublin Phil. Magazine and J. Sci., 20(130):21–37, 1860.
183. J. C. Maxwell. V. Illustrations of the dynamical theory of gases. Part I. On the motions and
collisions of perfectly elastic spheres. The London, Edinburgh, and Dublin Phil. Magazine
and J. Sci., 19(124):19–32, 1860.
184. J. C. Maxwell. On the dynamical theory of gases. Phil. Trans. Roy. Soc. London, 157:49–88,
1867.
185. V. G. Maz’ya and B. A. Plamenevskii. Estimates of Green’s function and Schauder estimates of
solutions of elliptic boundary value problems in a dihedral angle. Siberian Math. J., 19(5):752–
764, 1978.
186. V. G. Maz’ya and B. A. Plamenevskii. The first boundary value problem for classical equations
of mathematical physics in domains with piecewise smooth boundaries, Part 1. Z. Anal.
Anwendungen, 2(4):335–359, 1983. In Russian.
References 387
187. V. G. Maz’ya and J. Rossmann. Schauder estimates for solutions to a mixed boundary value
problem for the Stokes system in polyhedral domains. Math. Methods Appl. Sci., 29(9):965–
1017, 2006.
188. V. G. Maz’ya and J. Rossmann. Elliptic Equations in Polyhedral Domains, volume 162 of
Mathematical Surveys and Monographs. A. M. S., 2010.
189. J. B. McLeod and K. R. Rajagopal. On the uniqueness of flow of a Navier-Stokes fluid due
to a stretching boundary. Arch. Rational Mech. Anal., 98(4):385–393, 1987.
190. J. Mewis and N. J. Wagner. Thixotropy. Adv. Colloid Interface Sci., 147–148:214–227, 2009.
191. G. Minty. On a “monotonicity" method for the solution of nonlinear equations in Banach
spaces. Proc. Natl. Acad. Sci. USA, 50(6):1038–1041, 1963.
192. F. Mollica and K. R. Rajagopal. Secondary flows due to axial shearing of a third grade fluid
between two eccentrically placed cylinders. Internat. J. Engrg. Sci., 37(4):411–429, 1999.
193. C. L. M. H. Navier. Mémoire sur les lois du mouvement des fluides. Mém. Acad. R. Sci. Paris,
6:389–440, 1822.
(k)
194. J. Nečas. Sur les Normes équivalentes dans Wp (Ω) et sur la Coercivité des Formes Formelle-
ment Positives. In Équations aux Dérivées Partielles. Presses de l’Université de Montréal,
1966.
195. J. Nečas. Les Méthodes Directes en Théorie des Équations Elliptiques. Masson, Paris, France,
1967.
196. I. Newton. Philosophiae Naturalis Principia Mathematica. Prometheus Books, New York,
NY, 1687. Translation by Andrew Motte, 1995.
197. W. Noll. A mathematical theory of the mechanical behavior of continuous media. Arch.
Rational Mech. Anal., 2(1):197–226, 1958.
198. W. Noll. A new mathematical theory of simple materials. Arch. Rational Mech. Anal., 48(1):1–
50, 1972.
199. K. B. Oldham and J. Spanier. The Fractional Calculus: Theory and Applications of Differen-
tiation and Integration to Arbitrary Order. Academic press, New York, NY, 1974.
200. J. G. Oldroyd. On the formulation of rheological equations of state. In Proc. Roy. Soc. London
Series A: Math. Phys. Engrg. Sci., volume 200, pages 523–541, 1950.
201. L. Onsager. Reciprocal relations in irreversible thermodynamics. Phys. Rev., 37:405–426,
1931.
202. R. R. Ortega. Contribución al Estudio Teórico de Algunas E.D.P. No Lineales Relacionadas
con Fluidos No Newtonianos. PhD thesis, Universidad de Sevilla, Sevilla, Spain, 1995.
203. C. W. Oseen. Neuere Methoden und Ergebnisse in der Hydrodynamik. Akademische Verlags-
gesellschaft, Leipzig, Germany, 1927.
204. A. P. Oskolkov. Solvability in the large of the first boundary-value problem for a quasilinear
third-order system Pertaining to the Motion of a Viscous Fluid. Zap. Nauchn. Semin. LOMI,
V.A. Steklova Akad. Nauk SSSR, 27:145–160, 1972.
205. E. H. Ouazar. Sur les Fluides de Second Grade. PhD thesis, Université Pierre et Marie Curie,
Paris VI, France, 1981.
206. E. A. Overman, W. C. Troy, G. B. Ermentrout, and J. P. Keener. Uniqueness of flow of a
second-order fluid past a stretching sheet. Quart. Appl. Math., 44(4):753–755, 1987.
207. L. I. Palade, P. Attané, R. R. Huilgol, and B. Mena. Anomalous stability behavior of a properly
invariant constitutive equation which generalizes fractional derivative models. Internat. J.
Engrg. Sci., 37(3):315–329, 1999.
208. L. I. Palade and J. A. De Santo. Dispersion equations which model high-frequency linear
viscoelastic behavior as described by fractional derivative models. Internat. J. Non-linear
Mech., 36(1):13–24, 2001.
209. M. Paluch, Z. Dendzik, and S. J. Rzoska. Scaling of high pressure viscosity data in low
molecular-weight glass-forming liquids. Phys. Rev., 60:2979–2982, 1999.
210. J. Peetre. Espaces d’interpolation et théorème de Sobolev. Ann. Inst. Fourier, 16(1):279–317,
1966.
211. T. Pérlacová and V. Prŭša. Tensorial implicit constitutive relations in mechanics of incom-
pressible non-Newtonian fluids. J. Non-Newtonian Fluid Mech., 216:13–21, 2015.
388 References
212. A. C. Pipkin and R. S. Rivlin. Normal stresses in flows through tubes of non-circular cross-
section. Zeitschrift für angewandte Mathematik und Physik ZAMP, 14(6):738–742, 1963.
213. S. D. Poisson. Mémoire sur les équations générales de l’équilibre et du mouvement des corps
solides élastiques et des fluides. J. de l’Ecole Polytechnique, Paris, 13:1–174, 1831.
214. J. H. Poynting. Radiation-pressure. Phil. Magazine, 9(52):393–406, 1905.
215. W. Prager. Introduction to Mechanics of Continua. Ginn and Company, Boston, 1961.
216. L. Prandtl. Über Flüssigkeitshewegung bei selv Kleiner Reibung, verh. III. In Math. Kongr.
Heidelberg, pages 484–491, Leipzig,Germany, 1904. Teubner.
217. L. Prandtl. The mechanics of viscous fluids. In W. F. Durand, editor, Division G, Aerodynamic
Theory, volume 3, pages 34–208, Berlin, Germany, 1935. Springer-Verlag.
218. I. Prigogine. Introduction to Thermodynamics of Irreversible Processes, 3rd ed. Wiley, Hobo-
ken, 1967.
219. V. Prŭša and K. R. Rajagopal. On implicit constitutive relations for materials with fading
memory. J. Non-Newtonian Fluid Mech., 181:22–29, 2012.
220. J. P. Puel and M. C. Roptin. Lemme de Friedrichs. Théorème de densité résultant du lemme
de Friedrichs. Diplôme d’etudes approfondies, University of Rennes, Université de Rennes,
France, 1967.
221. K. R. Rajagopal. A note on the drag for fluids of grade three. Internat. J. Non-linear Mech.,
14(5–6):361–364, 1979.
222. K. R. Rajagopal. Thermodynamics and stability of non-Newtonian fluids. Developments in
Mech., 10:149–154, 1979.
223. K. R. Rajagopal. Boundedness and uniqueness of fluids of the differential type. Acta Ciencia
Indica, 18:1–26, 1982.
224. K. R. Rajagopal. On the creeping flow of the second-order fluid. J. Non-Newtonian Fluid
Mech., 15(2):239–246, 1984.
225. K. R. Rajagopal. Multiple configurations in continuum mechanics. Report 6, Institute for
Computational and Applied Mechanics, University of Pittsburgh, Pittsburgh, PA, 1995.
226. K. R. Rajagopal. On boundary conditions for fluids of the differential type. In Navier-Stokes
Equations and Related Problems, pages 273–278. Plenum Press, 1995.
227. K. R. Rajagopal. On implicit constitutive theories. Application of Math., 48(4):279–319, 2003.
228. K. R. Rajagopal. On the cavalier attitude towards referencing. Internat. J. Engrg. Sci., 51:338–
343, 2012.
229. K. R. Rajagopal. A new development and interpretation of the Navier-Stokes fluid which
reveals why the “Stokes Assumption" is inapt. Internat. J. Non-linear Mech., 50:141–151,
2013.
230. K. R. Rajagopal and A. S. Gupta. An exact solution for the flow of a non-Newtonian fluid
past an infinite porous plate. Meccanica, 19:158–160, 1984.
231. K. R. Rajagopal, A. S. Gupta, and A. S. Wineman. On a boundary-layer theory for non-
Newtonian fluids. Internat. J. Engrg. Sci., 18(6):875–883, 1980.
232. K. R. Rajagopal and P. N. Kaloni. Some Remarks on Boundary Conditions for the Flow of
Fluids of the Differential Type. In Continuum Mechanics and its Applications. Hemisphere
Press, 1989.
233. K. R. Rajagopal, T. Y. Na, and A. S. Gupta. A non-similar boundary layer on a stretching
sheet in a non-Newtonian fluid with uniform free stream. J. Math. Phys. Sci., 21(2):189–200,
1987.
234. K. R. Rajagopal and A. R. Srinivasa, On the development of fluid models of the differential
type within a thermodynamic framework, Mech. Res. Comm., 35:483–489, 2008.
235. K. R. Rajagopal and A. R. Srinivasa. Modeling anisotropic fluids within the framework of
bodies with multiple natural configurations. J. Non-Newtonian Fluid Mech., 99(2):109–124,
2001.
236. K. R. Rajagopal and A. R. Srinivasa. On the thermodynamics of fluids defined by implicit
constitutive relations. Zeitschrift für Angewandte Mathematik und Physik ZAMP, 59:715–729,
2008.
References 389
237. K. R. Rajagopal and A. R. Srinivasa. A Gibbs-potential-based formulation for obtaining the

response functions for a class of viscoelastic materials. In Proc. Roy. Soc. London Series A:
Math. Phys. Engrg. Sci., volume 467, pages 39–58, 2011.
238. K. R. Rajagopal and A. Z. Szeri. On an inconsistency in the derivation of the equations of
elastohydrodynamic lubrication. In Proc. Roy. Soc. London Series A: Math. Phys. Engrg. Sci.,
volume 459, pages 2771–2786, 2003.
239. K. R. Rajagopal and L. Tao. Mechanics of Mixtures. Series on Advances in Mathematics for
Applied Sciences, 35. World Scientific, Singapore, 1995.
240. A. V. Ramamurthy. Wall slip in viscous fluids and influence of materials of construction. J.
Rheology, 30(2):337–357, 1986.
241. I. J. Rao and K. R. Rajagopal. The status of the K-BKZ model within the framework of
materials with multiple natural configurations. J. Non-Newtonian Fluid Mech., 141(2–3):79–
84, 2007.
242. M. Reiner. A mathematical theory of dilatancy. Am. J. Math., 67:350–362, 1945.
243. O. Reynolds. On the theory of lubrication and its application to Mr. Beauchamp Tower’s
experiments, including an experimental determination of the viscosity of olive oil. Phil. Trans.
Roy. Soc. London, 177:157–234, 1886.
244. B. Riemann. Versuch Einer Allgemeinen Auffassung der integration und Differentiation. In
H. Weber, editor, The Collected Works of Bernhard Riemann. Dover Publications, Second
Edition, New York, NY, 1953.
245. R. S. Rivlin. The relation between the flow of non-Newtonian fluids and turbulent Newtonian
fluids. Quarterly of App. Math., 15(2):212–214, 1957.
246. R. S. Rivlin and J. L. Ericksen. Stress-deformation relations for isotropic materials. J. Rational
Mech. Anal., 4:323–425, 1955.
247. R. T. Rockafellar. Convex Analysis. Princeton Mathematical Series. Princeton University
Press, Princeton, NJ, 1970.
248. C.J. A. Roelands. Correlational Aspects of the Viscosity-Temperature-Pressure Relationship
of Lubricating Oils. PhD thesis, TU Delft, Delft University of Technology, Delft, The Nether-
lands, 1966.
249. Y. A. Rossikin and M. V. Shitikova. Applications of fractional calculus to dynamic problems
of linear and nonlinear hereditary mechanics of solids. Appl. Mech. Rev., 50(1):15–67, 1997.
250. R. N. J. Saal and G. Koens. Investigation into the plastic properties of asphaltic bitumen.
Physics, 7:408–412, 1933.
251. D. Serre. Équations de Navier-Stokes stationnaires avec données peu régulières. Ann. Scuola
Norm. Sup. Pisa, Cl. Sci. IV, 10(4):543–559, 1983.
252. J. Simon. Compact sets in the space L p (0, t; B). Annali di Mat. Pura ed Applicata, 146(1):65–
96, 1986.
253. J. Simon. Equations de Navier-Stokes. Cours de Diplôme d’Etudes Approfondies, Université
de Clermont-Ferrand, France, 2003.
254. J. Simon. On the identification H = H in the Lions theorem and a related inaccuracy. Ricerche
Mat., 59:245–255, 2010.
255. P. E. Sobolevsky. Investigation of the asymptotically newtonian model of fluids. Soviet Math.
Dokl., 1:95–98, 1983.
256. A. J. M. Spencer. Part III. Theory of invariants. In A. C. Eringen, editor, Continuum Physics,
volume 1, pages 239–353. Academic Press, New York, NY, 1971.
257. J. Stastna, L. Zanzotto, and K. Ho. Fractional complex modulus manifested in asphalts. Rhe-
ologica Acta, 33(4):344–354, 1994.
258. E. M. Stein. Singular Integrals and Differentiability Properties of Functions, volume 2. Prince-
ton University Press, Princeton, NJ, 1970.
259. G. G. Stokes. On the theories of the internal friction of fluids in motion, and of the equilibrium
and motion of elastic solids. Trans. Cambridge Phil. Soc., 8:287–305, 1845.
260. G. G. Stokes. On the effect of the internal friction of fluids on the motion of pendulums. Trans.
Cambridge Phil. Soc., 9:8–106, 1851.
390 References
261. A. Z. Szeri. Fluid Film Lubrication: Theory and Design. Cambridge University Press, Cam-
bridge, UK, 1998.
262. A. Tani and C. Le Roux. Steady-state solutions to the equations of motion of second-grade
fluids with general Navier type slip boundary conditions in Hölder spaces. J. Math. Sciences,
130(4):4899–4909, 2005.
263. R. I. Tanner. Plane creeping flows of incompressible second-order fluids. Physics of Fluids,
9(6):1246–1247, 1966.
264. L. Tartar. Topics in Nonlinear Analysis. Publications Mathématiques d’Orsay, Université
Paris-Sud, Orsay, France, 1978.
265. R. Temam. Theory and Numerical Analysis of the Navier-Stokes Equations. North-Holland,
Amsterdam, The Netherlands, 1977.
266. R. Temam. Infinite-Dimensional Dynamical Systems in Mechanics and Physics. In Applied
Mathematical Sciences, volume 68. Springer-Verlag, Berlin, Germany, 1997.
267. G. B. Thurston. Non-Newtonian viscosity of human blood: flow-induced changes in mi-
crostructure. Biorheology, 31(2):179–192, 1993.
268. T. W. Ting. Certain non-steady flows of second-order fluids. Arch. Rational Mech. Anal.,
14(1):1–26, 1963.
269. G. A. Tokaty. A History and Philosophy of Fluid Mechanics. Dover Publications, New York,
NY, 1994.
270. C. Truesdell. A new definition of a fluid: the Stokesian fluid, Part I. J. Math. Pures Appl.,
29:215–244, 1950.
271. C. Truesdell. On the viscosity of fluids according to the kinetic theory. Zeitschrift für Physik
A Hadrons and Nuclei, 131(3):273–289, 1952.
272. C. Truesdell. A program towards rediscovering the rational mechanics of the age of reason.
Arch. History of Exact Sci., 1(1):1–36, 1960.
273. C. Truesdell. The natural time of a viscoelastic fluid; its significance and measurement. Physics
of Fluids, 7(8):1134–1142, 1964.
274. C. Truesdell. The meaning of viscometry in fluid dynamics. Annual Rev. Fluid Mech.,
6(1):111–146, 1974.
275. C. Truesdell. A First Course in Rational Continuum Mechanics, volume 1. Academic Press,
New York, NY, 1992.
276. C. Truesdell and W. Noll. The Non-Linear Field Theories of Mechanics. In Handbuch der
Physik, volume III. Springer-Verlag, Berlin, Germany, 1965.
277. C. Truesdell and K. R. Rajagopal. An Introduction to the Mechanics of Fluids. In Modeling
and Simulation in Science, Engineering and Technology. Birkhäuser, Basel, CH, 2000.
278. W. E. Van Arsdale. Fractional rates of deformation. J. Rheology, 29(6):851–857, 1985.
279. J. H. Videman. Mathematical Analysis of Visco-Elastic non Newtonian Fluids. PhD thesis,
Universidade de Lisboa, Portugal, 1997.
280. M. H. Wagner. Analysis of stress-growth data for simple extension of a low-density branched
polyethylene melt. Rheologica Acta, 15(2):133–135, 1976.
281. L. H. Wang. On Korn’s inequality. J. Computational Math., 21(3):321–324, 2003.
282. K. Weissenberg. A continuum theory of rheological phenomena. Nature, 159(4035):310–311,
1947.
283. W. L. Wilkinson. Non-Newtonian Fluids. Fluid Mechanics, Mixing and Heat Transfer. Perg-
amon Press, London, UK, 1960.
284. J. Wloka. Partial Differential Equations. Cambridge University Press, Cambridge, UK, 1987.
Index
A natural, 69, 70
Admissible test functions, 129 reference, 19
Aubin-Lions-Simon Lemma, 113 Constraint response stress, 30
Convex operator, 109
Convexity inequality, 100
B Creep, 7, 16, 54, 56, 60
Balance
of angular momentum, 27
of energy, 27 D
of linear momentum, 10, 27, 38 DEBORAH number, 8
of mass, 27 Deck structure, 59, 61
Bingham constitutive equation, 159 Derivative
Biregularization, 131 fractional, 88
Boundary condition, 12, 75, 77 upper-convected Oldroyd, 23, 26
free-slip, 48 Die-swell, 17
free surface, 64–66, 90 Dilatancy, 31
no-slip, 12, 48 Dimensionless, 52, 55, 59
shark skin effect, 13, 90 Distribution, 93
slip, 48 tempered, 94
stick-slip, 12, 309 Drag, 55, 57, 58
stress-free, 91
threshold-type slip, 48
Boundary layer, 23, 59 E
approximation, 13 Eigenfunctions, 131
confinement of vorticity, 13 Elastic-plastic, 70
inertial, 63 Elastohydrodynamic, 10, 34
Reynolds number, 13 Energy equality, 42, 183, 320, 337
Brouwer Fixed-Point Theorem, 113 Entropy production, 36
C F
Clausius–Duhem inequality, 11, 12, 27 Fluid
Commuting convolution, 344, 360 anisotropic, 30
Compactness, 281 Bingham, 115, 158
Configuration Bingham plastic, 18
current, 19 Carreau, 168
DOI 10.1007/978-3-319-39330-8
392 Index
Cross, 168 velocity, 20

differential type, 9 Granular, 18, 31
dilatant, 120, 152 Green formula, 114, 194, 238, 348
Eyring–Prandtl, 168 Gronwall Lemma, 113
generalized Newtonian, 32, 167
grade three, 35, 38, 67, 313
grade two, 35, 38, 50, 67, 181 H
homogeneous, 33 Hardy inequality, 97
incompressible, 29, 33 Helmholtz potential, 37
incompressible Reiner, 117 Hemicontinuity, 169
incompressible simple, 28 History, 28
integral type, 9 Hölder space, 95
isotropic, 30, 35
Navier–Stokes, 33, 59, 66, 68
Newtonian, 120 I
non-Newtonian, 8 Implicit constitutive relation, 28, 29, 34
of complexity n, 30 Inf-sup condition, 105
of complexity one, 31, 34 Instantaneous elastic response, 16, 69
of complexity two, 31, 35 Integro-differential inequality, 211, 227, 372
power-law, 32 Interpolation inequality, 100
pseudoplastic, 121 Invariants, 35, 72, 86
rate type, 9, 28 principal, 31
Reiner, 115 Irrotational flow, 23
Reiner-Rivlin, 31, 32, 115, 117 Isochoric, 86
shear-thickening, 14
shear-thinning, 14, 75
simple, 14, 38 J
Jeffrey–Hamel flow, 60
steady grade two, 231, 273
Stokesian, 29, 31
variational grade three, 316
K
variational grade two, 189
Korn inequality, 106, 340
Williamson, 168
Formulation
strong, 130
L
weak, 129 Ladyzhenskaya
Frame, 24 example, 151
change, 24 inequality, 99
frame-indifferent, 23, 24, 30, 37, 89 Lagrange multipliers, 155
Friedrichs Lemma, 345 Laplace equation, 101
Frobenius norm, 93 Leray–Hopf Lemma, 220
Lipschitz domain, 93
Lower semi-continuous (l.s.c.), 111
G
Galerkin
decomposition, 132 M
discretization, 131, 191, 213, 240, 322, Material modulus, 32
330 Mean value theorem, 110
projection operator, 132 Mechanically isolated, 30, 42
G-differentiable, 109 Memory, 9
Glass transition, 10, 34 fading, 29
Gradient kernel, 89
deformation, 19, 25 Minty’s method, 135, 177, 337
relative deformation, 21 Model
Index 393
Burgers, 67, 78, 79 Regularization, 333

Johnson–Segalman, 81, 82 elliptic, 351
K-BKZ, 86 interior, 344, 359
linear integral, 86 Yosida, 352
Lodge, 86 Regularized distance, 223
Maxwell, 67, 77 Reiner constitutive equation, 116, 117, 119
Oldroyd 8 constant, 68 Relative deformation, 20
Oldroyd-B, 67, 74, 77 Relaxation
upper-convected Maxwell, 72 modulus, 86
upper-convected Oldroyd-B, 69 time, 8, 69, 74
Mollifier, 222 Renormalization, 346
Monotone, 110, 326, 328, 352 Retardation Theorem, 36
Monotonicity, 170 Retardation time, 69
Monotonous flow, 38 Reynolds number, 9, 52, 56, 59, 60
Motion, 19 Rivlin-Ericksen representation theorem, 117
retarded, 29 Rod-climbing, 17, 31
N S
Navier–Stokes Schauder Fixed-Point Theorem, 113
incompressible, 7 Secondary flow, 17
system, 120, 150 Semi-discretization, 259, 291
turbulence, 39 Shear-thickening, 7
Nonlinear form C, 120 Shear-thinning, 7
Normal stress difference, 8, 17, 31
Simple shear, 13, 40
Sisko equation, 121
Sobolev imbedding, 96
O
Sobolev space, 94
Objective, 24
fractional, 94
Special basis, 184, 190, 315
P Specification
Partition of unity, 204, 346 Eulerian, 20
Perturbation Lagrangian, 20
formal, 63 Spurt, 82
singular, 59, 63 Steady flow, 20
Pitch drop experiment, 5 Stokes
P-Laplacian, 121 approximation, 55, 58
Plasticity threshold, 163 assumption, 33
Poincaré system, 54, 104
constant, 43, 96 Stress integral representation, 68
inequality, 43, 96 Stress power, 41
Porous boundary, 50–52, 75 Stress relaxation, 7, 15
Potential Structure parameter, 14
Gibbs, 83 Subdifferential, 111, 118
Helmholtz, 37 constitutive equation, 118
Power Theorem, 41 Suction, 76
Poynting effect, 18 Surface traction, 41, 49
Pressure, 153, 155, 157
T
R Tensor
Radiation, 11 Almansi–Hamel strain, 21
Rate of dissipation, 36, 37, 70, 72, 74, 80, 84 Cauchy–Green, 21, 26, 71
394 Index
conformation, 74 V
extra stress, 28, 39, 75 Variational inequality, 112, 119, 121, 129,
Green–Saint Venant strain, 21 155, 161
relative rotation, 21, 22 Vector potential, 185, 222, 333
relative stretch, 21, 22, 26 Velocity, 19
Rivlin–Ericksen, 22, 26, 88 angular, 25
rotation, 21, 23 Viscometric
spin, 22, 23, 26 flow, 39
stretch, 21, 24 function, 14, 39, 40
Thermodynamically compatible, 27, 35, 37, Viscosity, 13
314 bulk, 32
Thixotropy, 7, 14 constant, 7, 10
Time derivative generalized, 14
local, 20
pressure dependent, 7, 9, 10, 34, 35
material, 20, 22
shear, 32
Time semi-discretization, 357
shear rate dependent, 14
Traction-free, 64
Vorticity, 23
Transport
steady scalar, 341 confinement, 60
steady system, 350
time-dependent scalar, 356
time-dependent system, 365 W
Trilinear form c, 122 Weissenberg effect, 18
Truncation, 222, 334
U Y
Uniform cone property, 343 Yield, 8
Uniqueness theorems, 146, 148 condition, 18
Universal flow, 66 stress, 18, 163

(D. Cioranescu, V. Girault, K.R. Rajagopal (Auth.) (B-Ok - Xyz)

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

(D. Cioranescu, V. Girault, K.R. Rajagopal (Auth.) (B-Ok - Xyz)

Încărcat de

Drepturi de autor:

Formate disponibile

Advances in Mechanics and Mathematics 35

More information about this series at http://www.springer.com/series/5613

Mechanics and Mathematics

ISSN 1571-8689 ISSN 1876-9896 (electronic)

© Springer International Publishing Switzerland 2016

Printed on acid-free paper

This Springer imprint is published by Springer Nature

2.7.3 Inertial Boundary Layers. . . . . . . . . . . . . . . . . . . . . . . 63

5 Grade-Two Fluids: Some Theoretical Results . . . . . . . . . . . . . . . . . 179

5.8.3 A Fixed-Point Argument. . . . . . . . . . . . . . . . . . . . . . . 282

List of Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375

Many real fluids exhibit response characteristics that cannot be satisfactorily

is devoted to the mathematical analysis of grade 2 and to a lesser extent of grade-three

In this chapter, the reader will be introduced to a variety of non-Newtonian phenom-

2.1 Introductory Remarks

© Springer International Publishing Switzerland 2016 5

5. develop normal stress differences,

to a time interval associated with the process. In considering problems involving

Later, according to Bridgman [52], Andrade proposed the following relationship

where 0 is the reference density (density where the pressure is zero).

The variation of the viscosity with pressure is of great significance in a variety of

However, controversy surfaced concerning the consequences of using Clausius–

2.2 Non-Newtonian Behavior

2.2.1 Shear-Thinning and Shear-Thickening

Fig. 2.1 Shear dependent

shear rate (κ̇)

In a general simple fluid (cf. Noll [197]), it can be shown that

Tx y = τ (κ) := [μ̃(κ)]κ, (2.2.2)

Thixotropy is not a very well-understood phenomenon and is usually confused with

2.2.3 Stress Relaxation

Fig. 2.2 Stress relaxation in viscoelastic bodies

Many materials which stress-relax also exhibit instantaneous “elastic” response.

they typically respond in the manner shown in Figure 2.3.

2.2.5 Normal Stress Differences

Fig. 2.3 Nonlinear creep

We will see later that in many fluids of differential type, we have

Fig. 2.4 Flow in a pipe of

Fig. 2.5 Rod-climbing due

The development of normal stresses are common to many types of nonlinear

Fig. 2.6 A material with a

Now, for any function ϕ defined on B × RI , we can define a function ϕ̃ on κ R (B) × R

ϕ( P, t) = ϕ̃(X, t) = ϕ̂(x, t). (2.3.4)

The specification in terms of (X, t) is usually referred to as Lagrangian specification

∂v1 ∂v2 ∂v3

Div v = tr(Grad v). (2.3.10)

We shall assume that F κ R is non-singular. For convenience, we shall suppress the

Let ξ denote the position occupied by X at time τ . Then

ξ = χ (X, τ ) = χ (χ −1 (x, t), τ ) := χ t (x, τ ). (2.3.12)

where R ∈ O, the group of orthogonal transformations6 and U, V are positive

F t (x, τ ) = Rt (x, τ )U t (x, τ ) = V t (x, τ )Rt (x, τ ),

C t (x, τ ) := [F t (x, τ )]T [F t (x, τ )], (2.3.20)

It immediately follows that

D(x, t) = U̇ t (x, t) = V̇ t (x, t),

Let us define L n through

where the superscript n denotes n material time derivatives. It immediately follows

It is possible to express the Rivlin–Ericksen tensors An by

and thus for all vectors a

2.3.2 Frames, Frame-Indifference and Restrictions

is said to be Frame-Indifferent, and a second-order tensor T that transforms according

v∗ = ċ + Q̇(t)[x − x 0 ] + Q(t)v, (2.3.36)

and thus, we immediately recognize that velocity is not frame-indifferent. Equation

v∗ − Q(t)v = ċ + A[x ∗ − c], (2.3.37)

a∗ − Q(t)a = c̈ + 2 Q̇(t)v + Q̈(t)[x − x 0 ],