Kim. Equations of Motion For Incompressible Viscous Fluids With Mixed Boundary Conditions

Advances in Mathematical Fluid Mechanics
Tujin Kim
Daomin Cao
Equations
of Motion
for Incompressible
Viscous Fluids
With Mixed Boundary Conditions
Series Editors
Giovanni P. Galdi, University of Pittsburgh, Pittsburgh, USA
John G. Heywood, University of British Columbia, Vancouver, Canada
Rolf Rannacher, Heidelberg University, Heidelberg, Germany
The Advances in Mathematical Fluid Mechanics series is a forum for the publication
of high-quality, peer-reviewed research monographs and edited collections on the
mathematical theory of fluid mechanics, with special regards to the Navier-Stokes
equations and other significant viscous and inviscid fluid models. Titles in this series
consider theoretical, numerical, and computational methods, as well as applications
to science and engineering. Works in related areas of mathematics that have a direct
bearing on fluid mechanics are also welcome. All manuscripts are peer-reviewed to
meet the highest standards of scientific literature.
More information about this series at http://www.springer.com/series/5032

Tujin Kim Daomin Cao
•
Equations of Motion
for Incompressible Viscous
Fluids
With Mixed Boundary Conditions
Tujin Kim Daomin Cao
Institute of Mathematics Guangzhou University and Institute
State Academy of Sciences of Applied Mathematics
Pyongyang Chinese Academy of Sciences
Democratic People’s Republic of Korea Guanzhou, Beijing, China
ISSN 2297-0320 ISSN 2297-0339 (electronic)

ISBN 978-3-030-78658-8 ISBN 978-3-030-78659-5 (eBook)
https://doi.org/10.1007/978-3-030-78659-5
Mathematics Subject Classification: 35Q30, 76D03, 76D05, 49J40, 80A20, 47J20, 35A02, 35A15,
76D07, 35R35, 35J87
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2021
This work is subject to copyright. All rights are solely and exclusively licensed 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, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.
This book is published under the imprint Birkhäuser, www.birkhauser-science.com by the registered
company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
The study of the equations appearing in fluid mechanics is a core topic in the fields
of partial differential equations, fluid mechanics and applied sciences. Lots of real
phenomena of fluid are described by those equations with different kinds of
boundary conditions. We are concerned with various boundary conditions, and
nowadays for mathematical analysis of phenomena of fluid, the fluid equations with
new boundary conditions are studied. In practice, we are concerned with mixture of
various boundary conditions. Now there are many papers dealing with fluid
equations with mixed boundary conditions; however, almost all monographs deal
with the equations with Dirichlet boundary condition and a few monographs deal
with mixture of Dirichlet and stress or Dirichlet and a kind of friction boundary
conditions.
The purpose of this book is to introduce the recent results and research methods
in the field of fluid equations with mixed boundary conditions via one book. For
readers broadly ranging from students to engineers and mathematicians involved in
the fluid equations, first the fluid equations and the boundary conditions for the
Navier-Stokes equations are outlined in the level of senior university students.
Then, the recent results for the Navier-Stokes equations and the equations of motion
for fluid conducting heat with the mixed boundary conditions originated mainly
from authors’ results and the first author’s lectures for postgraduate students in the
Institute of Mathematics, State Academy of Sciences, DPR Korea are described in
the level of postgraduate students.
The style of this book is rather different from other mathematical monographs
for the Navier-Stokes equations in the following sense:
1. Many monographs for the fluid equations concentrate on mathematical prop-
erties of the equations, whereas this book first intends to show how the fluid
equations appearing in the journals are obtained and what kinds of boundary
conditions are suitable in various practical situations.
2. In the existing monographs, according to the bilinear forms used for variational
formulations of the problems, category of mixture of boundary conditions for
fluid is determined, whereas this book almost freely deals with mixture of the
v
vi Preface
boundary conditions for fluid relying on some properties on the boundaries of

vector fields given on domains, which is the main point to make this book
different from others.
3. Many monographs concentrate on mathematical investigation of the formulated
variational problems, whereas this book gives attention to equivalence between
the original partial differential equations with mixture of various boundary
conditions and the corresponding variational problems, especially variational
inequalities with one unknown, which is not obvious due to mixture of com-
plicated boundary conditions.
The book consists of 10 chapters.
Trying to make our book self-contained, in Chap. 1, we give some basic
knowledge and known results in analysis used in the book. We do not give the best
result for every subject, but sometimes we give results more than just required for
subsequent parts of the book to help readers to understand other literature. Readers
need not read the whole chapter, and can use the chapter as data to consult when
something is quoted in the book. The theory of pseudo-monotone operator in Sect.
1.5 and convex functional in Sect. 1.6 is needed only in Chaps. 7, 9 and 10.
In Chap. 2, we first show how the Navier-Stokes equations and the equations of
motion for fluid under consideration of heat are derived from the fundamental
physical laws. Though the aim of this book is to study the equations of motion for
the incompressible fluids, the equations for compressible fluids are discussed
together to widen reader’s understanding. Next, we outline some boundary con-
ditions for the Navier-Stokes equations, which would be helpful for readers to
understand the meaning of the boundary conditions in this book and to improve the
ability to apply these in practice. Then, we consider the widely used existing three
kinds of variational formulations for the Navier-Stokes problems with mixed
boundary conditions. Also, equivalence between the variational formulations and
the original PDE problems is discussed, which is a preparation to understand
equivalence between PDE problems with more complicated boundary conditions
and their variational formulations in the subsequent chapters. This chapter can serve
as an introduction for the students and postgraduate students in the fields of
equations describing fluids, mathematical modeling and numerical simulation.
In Chap. 3, we first study some properties on the boundary surfaces of vector
fields given on domains. To this end, a little knowledge of differential geometry is
used, which is contained in the chapter. This is preparation to embed wide various
mixed boundary conditions together into variational formulations and the founda-
tion for the whole book, which is the peculiar features of this book different from
others. Based on the results, we can consider mixtures of boundary conditions
which were not studied in other literature, for example, mixture of the boundary
conditions for stress and pressure which is important in practice. Thus, we study the
existence and uniqueness of solutions to the steady Navier-Stokes system with
mixed boundary conditions which may include seven kinds of boundary conditions
together. According to whether the static pressure or total pressure is included in the
Preface vii
boundary conditions, the research methods and results are different, and so these
cases are distinguished in this and next chapters.
In Chap. 4, we study the existence and uniqueness of a solution to the
non-steadyNavier-Stokes system with mixed boundary conditions corresponding to
the steady problem in Chap. 3.
In Chap. 5, we study the steady Navier-Stokes system with mixed boundary
conditions including friction-type conditions, which may include 11 kinds of
boundary conditions together. Unlike Chaps. 3 and 4, due to the friction boundary
conditions, the problems are described by the variational inequalities. Sometimes
variational inequalities are used as approximate mathematical models for real
phenomena; however, in this part equivalence between the variational inequalities
with one unknown and the original boundary value problem of PDE with 11 kinds
of boundary conditions is studied concretely. Relying on the results for the varia-
tional inequalities, we study the existence and uniqueness of solutions to the
problems under consideration.
In Chap. 6, we study the non-steadyNavier-Stokes system with mixed boundary
conditions corresponding to the steady problem of Chap. 5. The problems are
reduced to the non-steady variational inequalities.
Chapter 7 is devoted to the study of the steady Boussinesq system for heat
convection of fluid under mixed boundary conditions including friction conditions.
The boundary conditions for fluid are the same as the one in Chap. 5 and the
boundary conditions for temperature may include Dirichlet, Neumann and Robin
conditions together. The problem is formulated by a simultaneous system of a
steady variational inequality for velocity of fluid and a variational equation for
temperature.
Chapter 8 is devoted to the study of the non-steady problem corresponding to the
steady problem in Chap. 7.
In Chap. 9, we are concerned with the equation for steady flow of heat con-
ducting incompressible Newtonian fluids with dissipative heating by Joule effect
under mixed boundary conditions. The boundary conditions are the same as in
Chap. 7.
Chapter 10 is devoted to the non-steady problem corresponding to Chap. 9, but
the boundary conditions for fluid are restricted to the case of total pressure.
At the end of every chapter from Chap. 2, the bibliographical remark for the
problem in the chapter is given, which is to provide more information on the
research of the problem studied in the chapter.
We hope the book is helpful for readers to study the research methods and results
for the equations for fluid with mixed boundary conditions and to improve the
ability to apply the knowledge.
Pyongyang, DPR Korea Tujin Kim

Guangzhou, P.R. China Daomin Cao
May 2020
Contents
1 Miscellanea of Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Banach Space, Fixed Point and Basics of Mapping . . . . . . . . . . 1
1.1.1 Banach Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Fixed-Point Theorems . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.3 Basics of Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Lebesgue Space, Convergence . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Lebesgue Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 Convergence of Sequences of Functions . . . . . . . . . . . 8
1.3 Sobolev Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.1 Definition of Sobolev Space . . . . . . . . . . . . . . . . . . . . 12
1.3.2 Density and Continuation . . . . . . . . . . . . . . . . . . . . . . 14
1.3.3 Imbedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.4 Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.5 Some Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4 Space of Abstract Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4.1 Abstract Functions and Its Derivatives . . . . . . . . . . . . . 20
1.4.2 Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.5 Operator Equations and Operator-Differential Equations . . . . . . 28
1.5.1 Monotone Operator Equation . . . . . . . . . . . . . . . . . . . 28
1.5.2 Pseudo-Monotone Operator Equation . . . . . . . . . . . . . . 30
1.5.3 Operator-Differential Equations . . . . . . . . . . . . . . . . . . 31
1.6 Convex Functional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.7 Some Elementary Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . 37
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2 Fluid Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 41
2.1 Derivation of Equations for Fluid Motion . . . . . . . . . . . . . . . .. 41
2.1.1 Navier-Stokes Equations . . . . . . . . . . . . . . . . . . . . . .. 41
2.1.2 Equations of Motion for Fluid Under Consideration
of Heat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 49
ix
x Contents
2.2 Boundary Conditions for the Navier-Stokes Equations . . . . . . . 55

2.2.1 Boundary Conditions on the Walls . . . . . . . . . . . . . . . 58
2.2.2 Boundary Conditions on Symmetric Planes . . . . . . . . . 61
2.2.3 Boundary Conditions on Inlets and Outlets . . . . . . . . . 62
2.2.4 Outflow Boundary Conditions on Imaginary
Boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.2.5 Boundary Conditions on Free Surfaces . . . . . . . . . . . . 64
2.3 Bilinear Forms for Hydrodynamics . . . . . . . . . . . . . . . . . . . . . 66
2.3.1 Bilinear Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.3.2 Variational Formulations for Mixed Boundary
Value Problems of the Navier-Stokes Equations . ..... 71
2.4 Bibliographical Remarks . . . . . . . . . . . . . . . . . . . . . . . . ..... 76
2.4.1 Fluid Equations . . . . . . . . . . . . . . . . . . . . . . . . ..... 76
2.4.2 Boundary Conditions of the Navier-Stokes
Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 76
2.4.3 Bilinear Forms for Hydrodynamics . . . . . . . . . . ..... 77
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 77
3 The Steady Navier-Stokes System . . . . . . . . . . . . . . . . . . . . . . . . .. 83
3.1 Properties on the Boundary Surfaces of Vector Fields . . . . . . .. 83
3.1.1 The Second Fundamental Form and Shape Operator
of Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 84
3.1.2 Properties on the Boundary Surface of Vector
Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.2 Variational Formulations of the Steady Problems . . . . . . . . . . . 94
3.3 Existence of Solutions to the Steady Problems . . . . . . . . . . . . . 100
3.4 Bibliographical Remark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4 The Non-steady Navier-Stokes System . . . . . . . . . . . . . . . . . . . . . . 109
4.1 Existence of a Solution: The Case of Total Pressure . . . . . . . . . 109
4.1.1 Problem and Variational Formulation . . . . . . . . . . . . . 110
4.1.2 An Auxiliary Problem by Elliptic Regularization . . . . . 113
4.1.3 Proof of the Existence of a Solution . . . . . . . . . . . . . . 118
4.1.4 The Stokes Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.2 Existence and Uniqueness of Solutions: The Case of Static
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.2.1 Existence and Uniqueness of Solutions
to Problem I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
to Problem II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
for Perturbed Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Contents xi
4.3 Bibliographical Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
5 The Steady Navier-Stokes System with Friction Boundary
Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
5.1 Variational Formulations of Problems . . . . . . . . . . . . . . . . . . . 150
5.1.1 Variational Formulation: The Case of Static
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.1.2 Variational Formulation: The Case of Total
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
5.1.3 Variational Formulation: The Stokes Problem . . . . . . . 166
5.2 Solutions to Variational Inequalities . . . . . . . . . . . . . . . . . . . . . 167
5.3 Existence and Uniqueness of Solutions to the Steady
Navier-Stokes Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
6 The Non-steady Navier-Stokes System with Friction Boundary
Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
6.1 Variational Formulations of Problems . . . . . . . . . . . . . . . . . . . 185
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
6.1.3 Variational Formulation: The Stokes Problem . . . . . . . 193
6.2 The Existence and Uniqueness of Solutions to Variational
Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
6.3 Solutions to the Non-steady Navier-Stokes Problems . . . . . . . . 215
6.3.1 Existence of a Solution: The Case of Total
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
6.3.2 Existence of a Unique Solution: The Case of Static
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
6.3.3 Existence of a Unique Solution: The Stokes
Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
7 The Steady Boussinesq System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
7.1 Problems and Variational Formulations . . . . . . . . . . . . . . . . . . 227
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
xii Contents
7.2.1 Existence of a Solution to an Auxiliary Problem . . . . . 235

7.2.2 Existence and Estimates of Solutions to the
Approximate Problem . . . . . . . . . . . . . . . . . . . . . . . . . 240
7.2.3 Existence and Uniqueness of a Solution . . . . . . . . . . . 243
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
8 The Non-steady Boussinesq System . . . . . . . . . . . . . . . . . . . . . . . . . 251
8.1 Problems and Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
8.2 Variational Formulations for Problems . . . . . . . . . . . . . . . . . . . 253
8.2.1 Variational Formulations: The Case of Static
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
8.2.2 Variational Formulations: The Case of Total
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
8.3.1 Existence and Estimation of Solutions
to an Approximate Problem . . . . . . . . . . . . . . . . . . . . 260
8.3.2 Existence and Uniqueness of a Solution . . . . . . . . . . . 269
8.4.1 Existence of a Solution to an Approximate
Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
8.4.2 Existence of a Solution . . . . . . . . . . . . . . . . . . . . . . . . 278
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
9 The Steady Equations for Heat-Conducting Fluids . . . . . . . . . . . . . 285
9.1 Problems and Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
9.2 Variational Formulations for Problems . . . . . . . . . . . . . . . . . . . 286
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
9.3.1 Existence of a Solution to an Auxiliary Problem . . . . . 291
9.3.2 A Priori Estimates of Solutions to the Auxiliary
Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
9.3.3 Passing to Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
Contents xiii
10 The Non-steady Equations for Heat-Conducting Fluids . . . . . . . . . 321

10.1 Problem and Variational Formulation . . . . . . . . . . . . . . . . . . . . 321
10.1.1 Problem and Assumption . . . . . . . . . . . . . . . . . . . . . . 321
10.1.2 Variational Formulation for Problem . . . . . . . . . . . . . . 323
10.2 Existence of a Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
10.2.1 Existence of a Solution to an Approximate
Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
10.2.2 Estimates of Solutions to the Approximate
Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
10.2.3 Passing to the Limit . . . . . . . . . . . . . . . . . . . . . . . . . . 351
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
Chapter 1
Miscellanea of Analysis
In this chapter, we outline some knowledge of analysis: Banach space, fixed point,
Lebesgue and Sobolev spaces, operator and operator-differential equations and con-
vex functional, which will be used in the main part of this book. We do not describe
the best results, but to help readers’ understanding sometimes we say more than
necessary. The readers who are already acquainted with the elements of functional
analysis can skip this chapter and may consult the necessary parts when reading main
part of this book.
1.1 Banach Space, Fixed Point and Basics of Mapping
1.1.1 Banach Space
Let X be a normed space. A sequence {xn } in X is called a Cauchy sequence iff

for every ε > 0 there exists an integer N such that xn − xm X < ε holds whenever
m, n > N . A sequence {xn } in X is convergent to the element x0 if and only if
limn→∞ xn − x0 X = 0. A space X is said to be complete if every Cauchy sequence
in X converges to an element in X . A complete normed space is called a Banach
space. A subset S of a normed space X is said to be dense in M ⊆ X if each x ∈ M
is the limit of a sequence of elements of S. The normed space X is called separable
if it has a countable dense subset.
Denote by X ∗ the space of linear continuous functionals on a normed space X ,
which is called the dual of X . A norm on the dual X ∗ of a normed space X can be
defined by setting
x ∗ X ∗ = sup x ∗ , x ,
x X =1
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 1

T. Kim and D. Cao, Equations of Motion for Incompressible Viscous Fluids,
Advances in Mathematical Fluid Mechanics,
https://doi.org/10.1007/978-3-030-78659-5_1
2 1 Miscellanea of Analysis
where and in what follows x ∗ , x means the value of a functional x ∗ ∈ X ∗ at an

element x ∈ X . Since the fields of real and complex numbers R and C are complete,
with the topology induced by this norm X ∗ is a Banach space (no matter whether X
is complete).
A natural linear injection of a normed space X into its second dual space X ∗∗ =
∗ ∗
(X ) is provided by the mapping J whose value J x at x ∈ X is given by
J x, y = y, x ∀y ∈ X ∗ ,
J is an isometric isomorphism of X into X ∗∗ , and so X ⊂ X ∗∗ .

If the range of the isomorphism J is the entire space X ∗∗ , the space X is called
reflexive. A reflexive space must be complete and hence a Banach space.
For a sequence {xn } of a normed space X if
f, xn − x0 → 0 ∀ f ∈ X ∗ as n → ∞,
then {xn } is said to be weakly convergent to x0 . We denote norm convergence of a

sequence {xn } to x0 by xn → x0 , and weak convergence by xn x0 .
A sequence {xn } in a normed space X is called a weak Cauchy sequence iff for
every ε > 0 and f ∈ X ∗ there exists an integer N such that | f, xn − xm | < ε holds
whenever m, n > N . The space X is called weakly complete if every weak Cauchy
sequence in X weakly converges to an element in X .
For a sequence { f n } of a normed space X ∗
g, f n − f 0 → 0 ∀g ∈ X ∗∗ as n → ∞
means that { f n } weakly converges to f 0 in X ∗ . For a sequence { f n } of a normed space

X ∗ if
x, f n − f 0 → 0 ∀x ∈ X as n → ∞,
∗
then { f n } is said to be ∗-weakly convergent to f 0 and it is denoted by f n f .
Theorem 1.1 Let M be a convex closed subset of a Banach space X and a sequence
{xn } ⊂ M weakly converges to a limit x0 . Then x0 belongs to M, i.e. M is weakly
closed.
(See Theorem 3.7 of [1].)
Theorem 1.2 If a sequence {xn } in a Banach space X weakly converges to an element
x, then
x X ≤ lim inf xn X . (1.1)
n→∞
(See Theorem 1, Sect. 1, Chap. V of [2].)

A weak Cauchy sequence of a Banach space may not have a weak limit in the
space, and the Banach space may not be weakly complete. But the following theorem
holds.
1.1 Banach Space, Fixed Point and Basics of Mapping 3
Theorem 1.3 Reflexive spaces are weakly complete. In every bounded subset of a
reflexive Banach space there exists a weakly convergent sequence.
(See Theorem 7, Sect. 1 and Theorem 1, Sect. 2, Chap. V of [2].)
Theorem 1.4 If all weakly convergent subsequences of a sequence {xn } of a reflexive

space converge to the same element x, then the sequence {xn } weakly converges to
x.
(See Lemma 5.4, Chap. 1 of [3].)

Imbedding of a normed space X into a linear normed space Y means that set-
theoretic inclusion X ⊂ Y , for which the following inequality is valid
x X ≤ CxY ∃C, ∀x ∈ X.
We denote by X → Y the imbedding. If X is dense in Y , then f ∈ Y ∗ may be

identified with an element in X ∗ , and so we have Y ∗ → X ∗ . If X is reflexive, then
this imbedding is dense (see Remark 5.14, Chap.1 of [3]).
1.1.2 Fixed-Point Theorems
Theorem 1.5 (Banach fixed-point theorem) Let (X, d) be a complete metric space,
M a closed nonempty set of X , and T : M ⊂ X → M be a map such that
d(T x, T y) ≤ kd(x, y) ∃k(0 ≤ k < 1), ∀x, y ∈ M.
Then, there exists a unique point x ∈ M such that T x = x.
Let X be a topological space and M ⊆ X . The set M is called compact iff every
open covering of M contains a finite subcover, i.e. finitely many of these open sets
already cover M. The set M is called relatively compact iff its closure M is compact.
Let X be a complete metric space and M ⊆ X . Then M is relatively compact if and
only if for every ε > 0 there is a finite ε-net for M. Let X be a metric space. Then
M ⊆ X is relatively compact if and only if for any sequence {xn }∞ n=1 ⊂ M there is a
convergent subsequence (see Proposition 1.2.1 of [4]).
Definition 1.1 Let X, Y be normed spaces, and T : M ⊆ X → Y an operator. T is

called compact if it maps every bounded subset of M into a relatively compact set
of Y . If, in addition, T is continuous on M, then it is called completely continuous
on this set.
When X, Y are normed spaces, for linear operators T the concepts of a compact
and a completely continuous operators are the same since every compact linear
operator is continuous.
Remark 1.1 Note that in some literatures the completely continuous operator is
called compact.
Theorem 1.6 (The Schauder fixed-point theorem) If a completely continuous oper-
ator T maps a bounded closed convex set M of a normed linear space X into itself,
then there exists at least one point x ∈ M such that T x = x.
(See Theorem 5.1.11 of [4].)
1.1.3 Basics of Mappings
Let T be a single-valued mapping from a set M to N . Mapping T : M → N is said

to be surjective if each element of N is the image of a certain element of M; injective
if different elements of M are mapped into different elements of N ; bijective if the
mapping is both surjective and injective simultaneously. If T : M → N is bijective,
then the mapping is an one-to-one map between M and N .
If T : M → N is continuous and bijective, and if its inverse map T −1 : N → M
is also continuous, then T is called a homeomorphism and M and N are said to be
homeomorphic.
Definition 1.2 Let X, Y be Banach spaces and f : U (x) ⊆ X → Y be a map, where
U (x) is a neighborhood of x.
(1) The map f is said to be Gâteaux differentiable at x iff there exists a linear
continuous operator T (x) : X → Y such that
f (x + th) − f (x) = t T (x)h + o(t) (t → 0) (1.2)
for all h with h X = 1 and all real numbers t in some neighborhood of zero. If
it exists, this T (x) is called the Gâteaux derivative at x. Put T (x) = f (x). The
Gâteaux differential at x is defined by dG f (x; h) = f (x)h.
(2) The map f is said to be Fréchet differentiable at x iff there exists a linear
continuous operator T (x) : X → Y such that
f (x + h) − f (x) = T (x)h + o(h X ) (h → 0 X ). (1.3)
T (x) is called the Fréchet derivative at x and T (x)h is called the Fréchet differential
at x.
The Gâteaux derivative f (x) of f at x can be defined equivalently by
f (x + th) − f (x)
f (x)h = lim .
t→0 t
Definition 1.3 Let X, Y be Banach spaces, f : U ⊂ X → Y and r ∈ N (the set of

natural numbers).
1.1 Banach Space, Fixed Point and Basics of Mapping 5
(1) If U is open, then f is called a C r -map iff f has continuous Fréchet derivatives
up to order r on U .
(2) If U is arbitrary, then f is called a C r -map iff for each x ∈ U there exists an
open neighborhood V (x) such that f can be extended to a C r -map on V (x).
Definition 1.4 Let M and N be arbitrary sets in the Banach spaces X and Y , respec-
tively. Let 0 ≤ r ≤ ∞.
(1) The mapping f : M → N is called a C r -diffeomorphism iff f is bijective and
both f and f −1 are C r -maps.
(2) A local C r -diffeomorphism at x0 is a C r -diffeomorphism from some neigh-
borhood U (x0 ) in X onto some neighborhood U ( f (x0 )) in Y .
Theorem 1.7 (Local inverse mapping theorem) (Theorem 4.F of [5]) Let f : U (x0 )
⊆ X → Y is a C 1 -mapping, where both X and Y are Banach spaces. Then f is a
local C 1 -diffeomorphism at x0 iff f (x0 ) : X → Y is bijective.
Theorem 1.8 (Theorem 3.4 of [6]) Let X, Y be Banach spaces, T0 be a one-to-one
linear continuous operator from X onto Y , Tc be a compact linear operator from X
into Y . Then T0 + Tc is a one-to-one operator from X into Y if and only if it is an
operator from X onto Y .
1.2 Lebesgue Space, Convergence
In this section we consider real-valued functions defined on a (Lebesgue) measurable

subset Ω ⊂ Rl , where Rl is l-dimensional Euclidean space.
1.2.1 Lebesgue Space
A function u : Ω → R ∪ {±∞} is itself called (Lebesgue) measurable if the set

{x : u(x) > a} is measurable for every real a.
A function u : Ω → R ∪ {±∞} is said to be countably valued if it assumes at
most a countable set of values, assuming each value is on a measurable subset. A
countably valued function is written as u(x) = i ai χ Ai , where χ Ai is the charac-
teristic function of the measurable set Ai .
A function u(x) is measurable if and only if it is uniform limit almost everywhere
of a sequence of countably valued functions. (See Theorem 2, Sect. 5, Chap. V of [7]
or p. 30 of Sect. 2.0 of [8].)
A countably valued function u(x) = i ai χ Ai is integrable if i ai mes (Ai ) <
∞, where mes (Ai ) is (Lebesgue) measure of Ai , and the integral of u(x) is defined
by
I (u) = u(x) d x = ai mes (Ai ).
Ω i
A measurable function u(x) is said to be (Lebesgue) integrable if there exists a

sequence {u n (x)} of integrable countably valued functions such that u n (x) → u(x)
uniformly outside of a null-measure set and

|u n (x) − u m (x)| d x → 0 as m, n → ∞.
Ω
We define (Lebesgue) integral by

u(x) d x = lim u n (x) d x.
Ω n→∞ Ω
The definition of integral (or of integrability) does not depend on the particular
sequence {u n } chosen.
The class of integrable functions on Ω is denoted by L 1 (Ω).
Theorem 1.9 (Lebesgue) Let Ω ⊂ Rl be measurable and let {u k } be a sequence

of measurable functions converging to u a.e in x ∈ Ω. If there exists a function
v(x) ∈ L 1 (Ω) such that
|u k (x)| ≤ v(x), k = 1, 2, · · · ,
then u is integrable and

u(x) d x = lim u k (x) d x.
Ω k→∞ Ω
Definition 1.5 Let Ω be an open subset of Rl and let p be a positive real number.
The space L p (Ω) consists of all (Lebesgue) measurable real functions u(x) defined
on Ω such that
|u(x)| p d x < ∞.
Ω
The space L ∞ (Ω) consists of all measurable functions u(x) defined on Ω such that
|u(x)| ≤ M ∃M > 0, for a.e. x ∈ Ω
L p (Ω), (1 ≤ p < ∞) and L ∞ (Ω) are Banach spaces, respectively, with norms
1/ p
u p = |u(x)| p d x and u∞ = ess sup |u(x)|.
Ω x∈Ω

1.2 Lebesgue Space, Convergence 7
Let Ω be an open subset of Rl . The support of a continuous function u is defined

by
supp u = {x; u(x) = 0} ∩ Ω.
The space C0∞ (Ω) consists of all infinitely differentiable functions that have
compact support in Ω. If 1 ≤ p < ∞, then C0∞ (Ω) is dense in L p (Ω) and L p (Ω)
is separable.
Theorem 1.10 (The

pRiesz
∗ representation theorem for L p (Ω)) Let 1 ≤ p < ∞.
Then for any f ∈ L (Ω) there exists v ∈ L (Ω), 1/ p + 1/q = 1, such that
q

f, u = v(x)u(x) d x ∀u ∈ L p (Ω)
Ω
∗
and f (L p (Ω))∗ = v L q (Ω) . The correspondence between f ∈ L p (Ω) and v ∈
L q (Ω) is linear.
L p (Ω) is reflexive if and only if 1 < p < ∞. (See Sect. 4.3 of [1].)
Theorem 1.11 (Hölder’s inequality) Let 1 < p < ∞ and let q denote the conjugate
exponent defined by
1 1
+ = 1.
p q
If u ∈ L p (Ω) and v ∈ L q (Ω), then uv ∈ L 1 (Ω), and

|u(x)v(x)| d x ≤ u p vq .
Ω
Equality holds if and only if |u(x)| p and |u(x)|q are proportional a.e. in Ω.
Theorem 1.12 (cf. Theorem 2.11 of [9]) (complex interpolation of Lebesgue spaces)
Let 1 ≤ p < q < r , so that
1 1−θ θ
= +
q r p
for some θ satisfying 0 < θ < 1. If u ∈ L p (Ω) ∩ L r (Ω), then u ∈ L q (Ω) and
u(L r ,L q )[θ ] ≡ uq ≤ ur1−θ uθp .
Here the functor ( , )[θ] means the complex interpolation (see Definition 2.1.3 of [10]
or Sect. 4.1 of [11]).
1.2.2 Convergence of Sequences of Functions
An open and connected subset of Rl is called a domain. Let Ω be a domain.
Lemma 1.1 (Lemma 1.19, Chap. 2 of [3]) Let 1 ≤ p < ∞ and a sequence {u n } ⊂
L p (Ω) be such that
u n u in L p (Ω),
u n (x) → v(x) a.e. in Ω.
Then u = v.
Lemma 1.2 Let vn and v be functions of L q (Ω), 1 < q < ∞, such that
vn L q (Ω) ≤ C, vn (x) → v(x) a.e. in Ω.
Then, vn v in L q (Ω).
Proof Since L q (Ω), 1 < q < ∞, is reflexive, by Theorem 1.3 there exists a sub-
sequence {vk } ⊂ {vn } such that vk u in L q (Ω). By Lemma 1.1 the limits of all
subsequences of {vn } coincide with v. Thus, by Theorem 1.4 we come to the conclu-
sion.
A set of functions F ⊂ L 1 (Ω) is called uniformly integrable if

lim sup | f | dμ = 0.
C→+∞ f ∈F | f |>C
A set consisting of a single integrable function is uniformly integrable since the

Lebesgue integral is absolutely continuous. Hence, for any integrable function f 0 ,
the set of all measurable functions f with | f | ≤ | f 0 | is uniformly integrable.
The sequence of functions { f n } is said to converge in measure to a measurable
function f if, for every c > 0, one has
lim mes{x : | f (x) − f n (x)| ≥ c} = 0.

n→∞
In Lebesgue space Vitali’s convergence theorem (see Theorem 4.5.4 of [12]) is

written as follows.
Theorem 1.13 Suppose that f is a measurable function and { f n } is a sequence of
integrable functions. Then, the following assertions are equivalent:
(1) the sequence { f n } converges to f in measure and is uniformly integrable;
(2) the function f is integrable and the sequence { f n } converges to f in the space
L 1 (Ω).
Lemma 1.3 Let Ω be a bounded domain of Rl . If {u n } is such that
{u n } is bounded in L ∞ (Ω), u n (x) → u(x) a.e. in Ω,
then
u n → u in L p (Ω) ∀ p, 1 ≤ p < ∞.
Proof Sine Ω is bounded, u n is integrable. By the conditions, {u k } converges to u

in measure and is uniformly integrable. By Theorem 1.13,

|u n − u| p d x ≤ M |u n − u| d x → 0.
Ω Ω
Corollary 1.1 Let Ω be a bounded domain of Rl and μ(ξ ), ξ ∈ R, be a bounded

continuous function. If {u n } and {vn } are such that
u n (x) → u(x) a.e. in Ω, vn v in L p (Ω), 1 < p < ∞,
then
μ(u n )vn μ(u)v in L p (Ω).
Proof By Lemma 1.3,
1 1
μ(u n ) → μ(u) in L q (Ω), + = 1.
p q
Thus, μ(u n )vn μ(u)v in L 1 (Ω), that is,

μ(u n (x))vn (x)φ(x) d x → μ(u(x))v(x)φ(x) d x ∀φ ∈ C0∞ (Ω).
Ω Ω
Since {μ(u n )vn L p } is bounded and C0∞ (Ω) is dense in L p (Ω), from above we get
the asserted conclusion (see Theorem 3, Chap. 5 of [2]).
Corollary 1.2 Let Ω be a bounded domain of Rl and μ(ξ ), ξ ∈ R, be a positive

bounded continuous function. If {u n } and {vn } are such that
u n (x) → u(x) a.e. in Ω, vn v in L 2 (Ω),
then
μ(u(x))v(x), v(x) ≤ lim inf μ(u n (x))vn (x), vn (x) .
n→∞
Proof By Corollary 1.1
μ(u n )vn μ(u)v in L 2 (Ω).

Then,
μ(u)v L 2 (Ω)
≤ lim inf μ(u n )vn L 2 (Ω)
,
n→∞
which implies the asserted conclusion.
Lemma 1.4 Let Ω be a bounded domain of Rl . If
vε v in L 2 (Ω) as ε → 0,
then vε
v in L 2 (Ω).
1 + εvε2
Proof Since

1 2
− 1 dx ≤ 2 + εvε2 − 2 1 + εvε2 d x ≤ ε vε2 d x,
Ω 1 + εvε2 Ω Ω
we have
1
→ 1 in L 2 (Ω). (1.4)
1 + εvε2

The sequence √ vε is bounded and C0∞ (Ω) is dense in L 2 (Ω), and to get the
1+εvε2
conclusion, it is enough to prove that

vε
− v ϕ d x → 0 ∀ϕ ∈ C0∞ (Ω) as ε → 0.
Ω 1 + εvε2
Using (1.4), we have

vε 1
− v ϕ dx = vε ϕ − 1 d x + (vε − v)ϕ d x → 0
Ω 1 + εvε2 Ω 1 + εvε2 Ω
∀ϕ ∈ C0∞ (Ω) as ε → 0,
which implies the asserted conclusion.
Lemma 1.5 Let Ω be a domain of Rl . If for a p > 1
vε (x) → v(x) ∈ L p (Ω) a.e. in Ω as ε → 0,
then
1
1 in L r (Ω) ∀r (1 < r < ∞).
1 + εvε2
Proof Since { 1+εv

1
2 } is bounded, and there exists a subsequence weakly convergent
ε
to a limit. On the other hand,
1
1 a.e. in Ω.
1 + εvε2
Thus, by Lemma 1.1 and Theorem 1.4 we come to the conclusion.
1.3 Sobolev Space
To simplify the discussion, in this section, we also work with real-valued functions.
Let Ω be an open subset of Rl with boundary ∂Ω and {ρα } be a family continuous
functions defined on Ω. The family of supports {supp ρα } is said to be locally finite
on Ω if any compact subset of Ω intersects only finitely many elements of {supp ρα }.
Let α = (α1 , · · · , αl ), αi —nonnegative integer, |α| = α1 + · · · + αl , and
∂ |α|
Dα = .
∂ x1α1 · · · ∂ xlαl
Define the family of semi-norm on C0∞ (Ω)

pρ (ϕ) = sup |ρα (x)D α ϕ(x)|, ϕ ∈ C0∞ (Ω), (1.5)
α x∈Ω
where ρ = {ρα } are taken all families of continuous function supports of which are
locally finite. By the family of semi-norms, C0∞ (Ω) becomes a locally convex space,
which is denoted by D(Ω).
Let D ∗ (Ω) denote the space of linear continuous functional defined on D(Ω),
often called the space of distributions on Ω. We denote by ·, · the duality pair-
ing between D ∗ (Ω) and D(Ω) and we remark that when f is a locally integrable
function, then f can be identified with a distribution by

f, φ = f (x)φ(x) d x ∀φ ∈ D(Ω).
Ω
D ∗ (Ω) is given the ∗-weak topology as the dual of D(Ω), i.e. the fact that f n → f
in D ∗ (Ω) means
f n − f, ϕ → 0 ∀ϕ ∈ D(Ω).
For f ∈ D ∗ (Ω), we define D α f ∈ D ∗ (Ω) by
D α f, ϕ = (−1)|α| f, D α ϕ ∀ϕ ∈ D(Ω).
If f n → f in D ∗ (Ω), then for ϕ ∈ D(Ω)
lim D α f n − D α f, ϕ = lim (−1)|α| f n − f, D α ϕ = 0 ∀ϕ ∈ D(Ω).

n→∞ n→∞
Thus, mapping f → D α f is continuous in D ∗ (Ω).
1.3.1 Definition of Sobolev Space
Definition 1.6 Let Ω be an open subset of Rl , s—nonnegative real number and

1 ≤ p ≤ ∞. We define the Sobolev space W s, p (Ω) as follows.
(i) s = k (a nonnegative integer), 1 ≤ p ≤ ∞.
W k, p (Ω) = {u ∈ L p (Ω) : D α u ∈ L p (Ω) ∀|α| ≤ k},

1/ p
D α u L p
p
uW k, p (Ω) = (1 ≤ p < ∞),
|α|≤k

uW k,∞ (Ω) = ess sup |D α u| .
|α|≤k x∈Ω
(ii) s = k + σ, 0 < σ < 1, k—nonnegative integer, 1 ≤ p < ∞.

|D α u(x) − D α u(y)| p
W s, p (Ω) = u ∈ W k, p (Ω) : d xd y < ∞, |α| = k ,
Ω×Ω |x − y|n+σ p
⎛ ⎞1/ p
|D α u(x) − D α u(y)| p
uW s, p (Ω) = ⎝u k, p d xd y ⎠
p
+ .
W (Ω) Ω×Ω |x − y|n+σ p
|α|=k
(iii) s = k + σ, 0 < σ < 1, k—nonnegative integer, p = ∞.

|D α u(x) − D α u(y)|
W s,∞ (Ω) = u ∈ L ∞ : u ∈ W k,∞ (Ω), max ess sup < ∞ ,
|α|=k x,y∈Ω, x = y |x − y|σ

p |D α u(x) − D α u(y)|
uW s,∞ (Ω) = u k, p + max ess sup .
W (Ω) |α|=k x,y∈Ω, x = y |x − y|σ
The Sobolev spaces are Banach spaces for the norms.

For s > 0, we denote by W0 (Ω) the closure of C0∞ (Ω) in W s, p (Ω) and by
s, p
−s,q
(Ω) the dual space of W0 (Ω), where p −1 + q −1 = 1.
s, p
W
If a function f is k times continuously differentiable and its derivatives of order

k are Lipschitz continuous, then we say it belongs to the class C k,1 .
Definition 1.7 (cf. Definition 1.2.1.1 of [13]) Let Ω be a bounded open subset of Rl .
We say that its boundary ∂Ω is continuous (resp. Lipschitz continuous, of class C k , of
1.3 Sobolev Space 13
class C k,1 for some integer k ≥ 0) if for every x ∈ ∂Ω there exists a neighborhood O
of x in Rl and new orthogonal coordinates y = (y , yn ) where y = (y1 , · · · , yn−1 ),
such that
(i) O is a hypercube in the new coordinates:
O = {y : |yi | < ai , 1 ≤ i ≤ n};
(ii) there exists a continuous (resp. Lipschitz continuous, of class C k , of class C k,1 )
function φ defined in
O = {y : |yi | < ai , 1 ≤ i ≤ n − 1}
that satisfies
an
|φ(y )| ≤ ∀y ∈ O ,
2
Ω ∩ O = {y : yn < φ(y )},
∂Ω ∩ O = {y : yn = φ(y )}.
Remark 1.2 Differently from Definition 1.18, Chap. 2 of [3], in Definition 1.2.1.1
of [13] and Definition 1.1 of [14] they did not demand boundedness of Ω. It is not
enough for Lipschitz continuity and class C k,1 of ∂Ω. If Ω is bounded, there exist
finite family of neighborhood covering ∂Ω, and from local properties we can get
Lipschitz continuity and C k,1 of ∂Ω.
To examine the boundary values of functions in W s, p (Ω), we consider the spaces

of functions defined on open subset Γ of boundary ∂Ω. We assume that Ω is a
bounded open subset of Rl with a boundary ∂Ω that is at least Lipschitz continuous.
We can view Γ locally as an n − 1-dimensional submanifold of Rl by means of the
mapping
Φ(y ) = (y , φ(y )) (1.6)
from O onto Γ ∩ O, where φ, O and O are the ones in Definition 1.7.

We assume that ψ is at least a bi-Lipschitz mapping from Ω1 onto Ω2 where
Ω1 and Ω2 are bounded open subsets of Rl . This hypothesis ensures that Lebesgue
measure is mapped by ψ or ψ −1 to an equivalent image measure.
Let u ∈ W s, p (Ω2 ); assume that ψ and ψ −1 are of class C k,1 , where k is an integer
≥ s − 1. Then u ◦ Φ ∈ W s, p (Ω1 ) (Lemma 1.3.3.1 of [13]). It is a justification for
the following definition.
Definition 1.8 (Definition 1.3.3.2 of [13]) Let Ω be a bounded open subset of Rl
with a boundary ∂Ω of class C k,1 , where k is a nonnegative integer. Let Γ be an
open subset of ∂Ω. A distribution u on Γ belongs to W s, p (Γ ) with |s| ≤ k + 1 if
u ◦ Φ ∈ W s, p O ∩ Φ −1 (Γ ∩ O) for all possible O and φ fulfilling the assumptions
in Definition 1.7.
Let Φ j is defined by (1.6) with φ j instead of φ. Let (O j , Φ j ) Jj=1 be any atlas of

Γ such that each couple (O j , Φ j ) satisfies the assumptions of the above definition.
Then one possible Banach norm for W s, p (Γ ) is the functional:

J 1/ p
u → u W s, p (Γ ) = u ◦ Φ j W s, p V ∩Φ −1 (Γ ∩V ) . (1.7)
j j j
j=1
In the particular case when s ∈ (0, 1), one can easily check that any of the norms
defined above is equivalent to
1/ p
|u(x) − u(y)| p
|u| dσ +
p
dσ (x)dσ (y) .
Γ Γ ×Γ |x − y|n−1+sp
1.3.2 Density and Continuation
Unless otherwise stated we assume that 1 < p < ∞ in the following subsections of
this section.
Let Ω be open subset of Rl and
D(Ω) = {φ|Ω : φ ∈ D(Rl )}.
Theorem 1.14 (Theorem 1.4.2.1 of [13]) If the boundary of Ω is continuous, then

D(Ω) is dense in W s, p (Ω) for all s > 0.
When s is a nonnegative integer and the boundary is Lipschitz continuous, this
theorem is valid for p = 1 (see Theorem 1.2, Chap. 1 of [14]).
Theorem 1.15 (Theorem 1.4.2.4 of [13]) Let Ω be a bounded open subset of Rl
with a Lipschitz boundary. Then D(Ω) is dense in W s, p (Ω) for 0 < s ≤ 1/ p.
For every s > 0, we denote by W s, p (Ω) the space of all distributions in Ω which
are restrictions of elements of W s, p (Rl ) (Definition 1.3.2.4 of [13]).
In other words, W s, p (Ω) is the space of all u|Ω where u ∈ W s, p (Rl ) and u|Ω is
defined by u|Ω , ϕ = u, ϕ̃ for all ϕ ∈ D(Ω), where ϕ̃ is the continuation of ϕ by
zero, outside Ω. A Banach norm on W s, p (Ω) is defined by
uW s, p (Ω) = inf U W s, p (Rl ) . (1.8)

U ∈W s, p (Rl )
U |Ω =u
As obvious consequences of the definition, we have the following inclusions:
W s, p (Ω) ⊂ W s, p (Ω)
((1.3.2.5) of [13]). The following theorem shows a case that W s, p (Ω) = W s, p (Ω).
Theorem 1.16 (Theorem 1.4.3.1 of [13]) Let Ω be a bounded open subset of Rl with
a Lipschitz boundary. Then for every s > 0 there exists a continuous linear operator
Ps from W s, p (Ω) into W s, p (Rl ) such that
Ps u|Ω = u.
For every positive s, we denote by W s, p (Ω) the space of all u ∈ W s, p (Ω) such
that ũ ∈ W (R ), where ũ is the extension of u by zero outside Ω.
s, p l
Theorem 1.17 (Corollary 1.4.4.5 of [13]) Let Ω be a bounded open subset of Rl

with a Lipschitz boundary. Then when s − 1/ p is not integer, we have
s, p (Ω) = W0s, p (Ω),

W (1.9)
and furthermore, when 0 < s < 1/ p we have
s, p (Ω) = W0s, p (Ω) = W s, p (Ω).

W (1.10)

with a Lipschitz boundary. Then for all s > 0, we have

s, p (Ω) = u ∈ W0s, p (Ω) : ρ −σ D α u ∈ L p (Ω), |α| = m
W
, where ρ(x) is the distance from x to the boundary ∂Ω of Ω and s = m + σ , m

integer, σ ∈ [0, 1).
1/2 1/2
Following Sect. 11.5, Chap. 1 of [15], we denote by H00 (Ω), H0 (Ω) and
H (Ω), respectively, the spaces W
1/2 1/2,2 (Ω), W01/2,2 (Ω) and W 1/2,2 (Ω). By Theo-
rem 1.15
1/2
H0 (Ω) ≡ H 1/2 (Ω). (1.11)
1/2 1/2
The space H00 (Ω) is strictly contained in H0 (Ω) with a strictly finer topology
1/2
u H 1/2 (Ω) = u2H 1/2 (Ω) + ρ −1/2 u2L 2 (Ω) .
00
Since C0∞ (Ω) ⊂ H00 (Ω), H00 (Ω) is dense in H 1/2 (Ω).
1/2 1/2
Remark 1.3 Let s ∈ R and 1 < p ≤ ∞. We denote by H ps (Rl ) the space of all
distributions in Rl such that

F −1 (1 + |ξ |2 )s/2 F u ∈ L p (Rl ),
where F is the Fourier transform operator defined by


1
(F u)(ξ ) = e−i xξ u(x) d x.
(2π )n/2 Rl
The space H ps is called with different names, i.e. the Bessel potential space [9],
the fractional order Sobolev space [9], the generalized Sobolev space [11], the inho-
mogeneous Sobolev space [16], and in some literature the notation for Sobolev is
used for that.
It is known that

H ps (Rl ) s : all real number, p = 2,
W (R ) =
s, p l
(1.12)
H ps (Rl ) s : integer, 1 < p < ∞.
Furthermore, it is proved that for any real s
W s, p (Rl ) ⊆ H ps (Rl ) 1 < p ≤ 2,

H ps (Rl ) ⊆ W s, p (Rl ) 2 ≤ p < ∞.
(See Sect. 1.3.1 of [13].)

Define the space H ps (Ω) by
H ps (Ω) = {u|Ω : u ∈ H ps (Rl )}. (1.13)
Then by Theorem 1.16 and (1.12), for a bounded open subset Ω with a Lipschitz
boundary, we have

H ps (Ω) s : all real number, p = 2,
W s, p
(Ω) = (1.14)
H ps (Ω) s : integer, 1 < p < ∞.
Instead of (1.13) in some literatures H ps (Ω) is defined by complex interpolation of

L (Ω) and W m, p (Ω). Similarly to Theorem 1.16, under some condition for boundary
p
of Ω, there exists a continuous linear operator Ps from H ps (Ω) into H ps (Rl ). (Noting
H ps (Ω) = F s. p,2 (Ω), refer to Sect. 7.69 of [9].) Thus under such definition (1.14) is
valid.
1.3.3 Imbedding
If imbedding operator I of a linear normed space into another linear normed space
is compact, then the imbedding is called compact.
Theorem 1.19 (Theorem 1.4.4.1 of [13]) The following imbeddings hold:
W s, p (Rl ) → W t,q (Rl ) (1.15)

for t ≤ s and q ≥ p such that s − n/ p = t − n/q; and
W s, p (Rl ) → C k,α (Rl ) (1.16)
for k < s − n/ p < k + 1, where α = s − k − n/ p, k a nonnegative integer.
Taking into account (1.8), as a consequence we have the conclusions W s, p (Ω) →

W (Ω) and W s, p (Ω) → C k,α (Ω) instead of (1.15) and (1.16), where C k,α (Ω) is
t,q
the space of functions which are k times continuously differentiable and whose
derivatives of order k fulfill a uniform Hölder condition of order α throughout Ω.
In other words, (1.15) is valid when 1p − s−t
n
= q1 . On the other hand, we see that
for a bounded open subset Ω
W t,q (Ω) → W t,q1 (Ω) for 1 < q1 < q.
Then taking into account the fact above, by Theorem 1.16, we have
Theorem 1.20 (Sobolev’s imbedding theorem) Let Ω be a bounded open subset of
Rl with a Lipschitz boundary and 0 ≤ t < s, 1 < q < ∞.
If 1p − s−t
n
≤ q1 , then
W s, p (Ω) → W t,q (Ω). (1.17)
If 1
p
− s−k
n
< 0, where k is a nonnegative integer, then
W s, p (Ω) → C k (Ω), (1.18)
and if, in addition, s − n

p
< k + 1, then
W s, p (Ω) → C k,α (Ω), α = s − k − n/ p. (1.19)
Remark 1.4 When q = ∞, (1.17) is not valid (cf. Remark 9.2, Chap. 1 of [15]).
Theorem 4.47, Corollary 4.53 and Theorem 4.57 of [17] assert that when Ω is Rl or
a unbounded open set Ω with a Lipschitz boundary and 1p − ns = 0, W s, p (Ω) →
L q (Ω) for every q < ∞. Under the condition p ≤ q < ∞ it is right (see Sect. 4.12
of [9]).
Theorem 1.21 (Theorem 1.4.3.2 of [13]) Let s1 > s2 ≥ 0 and assume that Ω is
a bounded open subset of Rl with a Lipschitz boundary. Then the imbedding of
W s1 , p (Ω) in W s2 , p (Ω) is compact.
Theorem 1.22 Let Ω be a bounded open subset of Rl with a Lipschitz boundary.

Let 0 < s and 1 < p < ∞.
If 1p − ns < q1 , then the imbedding W s, p (Ω) → L q (Ω) is compact.

Theorem 1.23 (Theorem 1.4.4.2 of [13]) Let s1 ≥ s and s2 ≥ s be such that either

1 1 1 1 1
s1 + s 2 − s ≥ n + − ≥ 0, s j − s > n − , j = 1, 2
p1 p2 p pj p
or

1 1 1 1 1
s1 + s2 − s > n + − ≥ 0, s j − s ≥ n − , j = 1, 2,
p1 p2 p pj p
then (u, v) → u · v is a continuous bilinear map from W s1 , p1 (Rl ) × W s2 , p2 (Rl ) into

W s, p (Rl ).
A similar statement holds for Sobolev spaces defined on a bounded open subset of
Rl with a Lipschitz boundary (see Sect. 1.4.4 of [13]).
with a Lipschitz boundary. Then D α , |α| = 1, is a linear continuous operator from
W s, p (Ω) into W s−1, p (Ω) unless s = 1/ p. (s-real number)
1.3.4 Trace
When u is a continuous function on Ω, let γ be the operator such that γ u = u|∂Ω .

The following trace theorem extends the operator γ to functions in W s, p (Ω) (cf.
Definition, Sect. 5.5 of [18]).
Theorem 1.25 (Theorem 1.5.1.2 of [13]) Let Ω be a bounded open subset of Rl with
a C k,1 boundary. Assume that s − 1/ p is not an integer, s ≤ k + 1, s − 1p = l + σ ,
0 < σ < 1, l an integer ≥ 0. Then mapping

∂ν ∂l u
u → γ u, γ , · · · , γ l
∂u ∂ν
which is defined for u ∈ C k,1 (Ω) has a unique continuous extension

k
W s− j− p , p (∂Ω).
1
T r : from W s, p
(Ω) onto
j=0

with a C k,1 boundary. Assume that s ≤ k + 1 and s − 1/ p is not an integer. Let
s, p
s − 1/ p = l + σ , 0 < σ < 1, l an integer ≥ 0. Then u ∈ W0 (Ω) if and only if
u ∈ W s, p (Ω) and
∂u ∂l u
γu = γ = · · · = γ l = 0.
∂ν ∂ν
By Theorem 1.26, we see that

s, p
K er (T r ) = W0 (Ω).
Thus, by the open-mapping theorem (see Sect. 5, Chap. 2 of [2]) there exists a con-
tinuous linear operator

k
W s− j− p , p (∂Ω) onto W s, p (Ω) W0 (Ω) (quotient space).
1
T r −1 : from
s, p
j=0
(1.20)

with a Lipschitz boundary. Then there exists a constant K such that

|u| p dσ ≤ K ε1−1/ p |∇u| p d x + ε−1/ p |u| p d x ∀u ∈ W 1, p (Ω), ∀ε ∈ (0, 1).
∂Ω Ω Ω
1.3.5 Some Inequalities
Theorem 1.28 (Friedrichs’ inequality) Let Ω be a bounded domain of Rl with a

Lipschitz boundary and Γ ⊂ ∂Ω, mes(Γ ) = 0. Then for u ∈ W 1, p (Ω) ( p ≥ 1)

n
∂u 2 p/2 p
|u| d x ≤ C
p
dx + u dσ .
Ω Ω i=1
∂ xi Γ
(See Theorem 1.32 of [19] or Lemma 1.36, Chap. 2 of [3].)

∂u i ∂u j
Let u = (u 1 , · · · , u n ) ∈ W1, p (Ω) ≡ W 1, p (Ω)n and εi j (u) = 1
2 ∂x j
+ ∂ xi
.
Theorem 1.29 (The first Korn’s inequality) Let Ω be a bounded domain of Rl

with a Lipschitz boundary and Γ be a nonempty open subset of ∂Ω. Then for u ∈
1, p
WΓ (Ω) := {v ∈ W1, p (Ω) : v|Γ = 0},
p p
εi j (u) L p (Ω) ≥ CuW1, p (Ω) .
i, j
(See Theorem 6.3.4 of [20] for the case that n = 3, p = 2, and Corollary 4.1 of [21]
for the general case.)
1.4 Space of Abstract Functions
1.4.1 Abstract Functions and Its Derivatives
Let X be a Banach space, S be an interval on R and u : S → X .

A function u : S → X is said to be finitely valued if it is constant on each of a
finite number of disjoint (Lebesgue) measurable sets B j and equal to 0 X on S \ ∪ j B j .
A function u : S → X is said to be simple if it is finitely valued and if the set for
which u(s) > 0 is of finite measure. u : S → X is said to be countably valued if
it assumes at most a countable set of values in X , assuming each value is different
from 0 X on a measurable subset.
Definition 1.9 A function u(s) is said to be (Bochner) strongly measurable (simply

measurable) if there exists a sequence of simple functions strongly convergent to
u(s) at a.e. s ∈ S
u : S → X is called weakly measurable if, for any f ∈ X ∗ , the numerical function
f, u(s) is (Lebesgue) measurable.
v : S → X ∗ is called ∗-weakly measurable if, for any g ∈ X , the numerical func-
tion g, v(s) is (Lebesgue) measurable.
Remark 1.5 There are some different definitions for strongly measurable function,
and some of them are listed below:
(i) A function u(s) is said to be strongly measurable if it is the strong limit of a

sequence of finitely valued functions almost everywhere. (p. 1681 of [22])
(ii) A function u(s) is said to be strongly measurable if there exists a sequence of
countably valued functions converging almost everywhere. (Definition 3.5.4 of
[23])
(iii) A function u(s) is called measurable function if there exists a sequence {u n (s)}
of countably valued functions such that u n (s) → u(s) uniformly outside a null
set (i.e. the set with null measure). (p. 174 of [8])
Definition (i) above is a special case of (ii).

u(s) is called almost separably valued if there exists a measurable set B0 of
measure zero such that {u(s); s ∈ S − B0 } is separable.
By Theorem (Pettis) of Sect. 4, Chap. 5 in [2], u(s) is strongly measurable in
sense of Definition 1.9 iff it is weakly measurable and almost separably valued. By
Theorem 3.5.3 of [23], for u(s) to be strongly measurable in the sense of Definition
(ii), so is it. Thus, Definition 1.9 and Definition (ii) are equivalent.
By Corollary 1 of Theorem 3.5.3 of [23], Definitions (ii) and (iii) above are equiv-
alent.
Therefore, all definitions in Remark 1.5 are equivalent to Definition 1.9.
Theorem 1.30 (Theorem 5.0.11 of [8]) Let X be separable. Then u : S → X is

weakly measurable if and only if it is strongly measurable.
1.4 Space of Abstract Functions 21
If X ∗ is separable, then strong measurability of X ∗ -valued function v : S → X ∗

is equivalent to ∗-weak measurability,
Remark 1.6 An X ∗ - valued function v(s) is said to be weakly measurable if

f, v(s) is measurable for all f ∈ X ∗∗ . The weak measurability of X ∗ -valued func-
tion is in general stronger than ∗-weak measurability.
For an example of ∗-weakly measurable but not measurable X ∗ -valued function
we refer to Example 5.0.10 of [8].
Theorem 1.30 is valid when X is a normed space.
Let u(s) : S → X be a finitely valued function. Let u(s) = xi on Bi , where Bi

are disjoint and mes(Bi ) < ∞ for!i = 1, · · · , n and u(s) = 0 on
Sn − ∪i Bi . Then we
can define the (Bochner) integral S u(s) ds of u(s) over S by i=1 xi mes(Bi ).
Define the integral of more general functions.
Definition 1.10 A function u(s) : S → X is said to be Bochner integrable, if there
exists a sequence of finitely valued functions {u n (s)} which converges to u(s) a.e. in
S and
lim u(s) − u n (s) ds = 0.
n→∞ S
For any measurable set B ⊂ S, the Bochner integral of u(s) over B is defined by

u(s) ds = lim χ B (s)u n (s) ds,
B n→∞ S
where χ B is the characteristic function of the set B.

The definition is justified since the limit on the right above exists and the value of this
limit is independent of the approximating sequence of functions {u n } (see Sect. 5,
Chap. 5 of [2]).
If u(·) is strongly measurable, then u(·) is measurable, and u(·) is Bochner
integrable if and only if u(·) is Lebesgue integrable, both integrals standing in the
relation
u(s) ds ≤ u(s) ds.
S S
Theorem 1.31 (Theorem 1.8, Chap. 4 of [3]) Let u : S → X be Bochner integrable

and B ⊂ S be a Lebesgue measurable set. Then for any f ∈ X ∗ ,

f, u(s) ds = f, u(s) ds .
B B
(See Theorem 8, Appendix E of [18] or Corollary 2, Sect. 5 of Chap. V in [2].)

Differently from real functions, not every Banach space valued absolutely contin-
uous functions is the indefinite integral of an integrable function (see Example 5.0.13
of [8]). But the next holds.
Theorem 1.32 If u(·) : S → X is Bochner integrable in S, then the function v :

S → X defined by t
v(t) = u(s) ds, t0 ∈ S
t0
is differentiable almost everywhere on S and
v (t) = u(t) for a.e. t ∈ S.
(See p. 176 of [8] or Theorem 1.7, Chap. 4 of [3].)

Remark 1.7 A Banach space X is called a Gelfand space if every X -valued abso-
lutely continuous function is differentiable almost everywhere. Let X be either (a)
reflexive or (b) separable and the dual of another Banach space. Then X is a Gelfand
space (Theorem 5.0.14 of [8]).
Definition 1.11 The space L p (S; X ), (1 ≤ p < ∞) consists of all X -valued
strongly measurable functions u(·) defined in S and such that

u(s) p ds < ∞.
S
The space L ∞ (S; X ) consists of all X -valued strongly measurable functions u(·)
defined in S and such that
u(s) ≤ M ∃M > 0, a.e. s ∈ S.
L p (S; X ), (1 ≤ p < ∞) and L ∞ (S; X ) are Banach spaces, respectively, with

norms
1/ p
u L p (S;X ) = u(s) p ds and u L ∞ (0,T ;X ) = esssup u(s).
S s∈S
Remark 1.8 Let 1 ≤ p < ∞. If Y is a dense subspace of X , then the set {ϕ =

φ(t)u : φ ∈ D(0, T ), u ∈ Y } is dense in L p (0, T ; X ) (see Theorem 5.0.27 of [8]).
Denote by C k ([0, T ]; X ) the space of k-times continuously differentiable
functions
on [0, T ], 0 < T < ∞. Then, the set of polynomials { p : p(s) = m k=0 ak s , s ∈
k
[0, T ], ak ∈ X, m = 0, 1, · · · } is dense in the set C([0, T ]; X ) (Weierstrass’ approx-

imation theorem, see Theorem 1.3, Sect. 1, Chap. 4 of [3]). Therefore, the set {ϕ =
φ(t)u : φ ∈ C 1 [0, T ], u ∈ X } is dense in L p (0, T ; X ). Also, we can see that the
set {ϕ = φ(t)u : φ ∈ C 1 [0, T ], φ(T ) = 0, u ∈ X } is dense in {ϕ ∈ C 1 ([0, T ]; X ) :
ϕ(T ) = 0}.
Theorem 1.33 Let X 0 , X 1 be Banach spaces such that X 0 → X 1 , 1 ≤ p0 < ∞ and
0 < θ < 1. Then,
L p0 (S; X 0 ) ∩ L ∞ (S; X 1 ) → L p (S; (X 0 , X 1 )[θ] ) (1.21)

with p = p0 /(1 − θ ), where the functor ( , )[θ] means the complex interpolation (see
Definition 2.1.3 of [10]).
Proof Since X 0 → X 1 , u ∈ L p (S; X 0 ) ∩ L ∞ (S; X 1 ) is (X 0 , X 1 )[θ] -value strongly

measurable and u(t) ∈ (X 0 , X 1 )[θ] at a.e. t ∈ S. Then,
θ
u(t)(X 0 ,X 1 )[θ ] ≤ u(t)1−θ
X 0 u(t) X 1
(see Corollary 2.1.8 of [10]), which yields (1.21).
Complex interpolation between simple Sobolev spaces is as follows.

Let Ω be a bounded domain with Lipschitz boundary, s0 = s1 and 0 < θ < 1.
Then, s0 ,2 ∗
W (Ω), W s1 ,2 (Ω) [θ] = W s ,2 (Ω), (1.22)
where
s ∗ = (1 − θ )s0 + θ s1 .
(cf. Theorem 6.4.5 of [11]. We used that when p = 2, Sobolev space W s, p equals to
Bessel potential space H ps (see (1.12)). All spaces in [11] are defined on R3 ; however,
the results are valid for the spaces continuously expandable to the space defined on
R3 , for example, the space defined on a bounded domain with Lipschitz boundary,
since from the interpolation results for the spaces defined on R3 we can reduce the
results for the spaces defined on the restricted domain. We refer to Sect. 1.3.2.)
Remark 1.9 In [11, 24] a space L ∞ (S; X ) is defined as follows:

If χ B is the characteristic function of the measurable set B ⊂ S, then L ∞ (S; X )
is defined as a completion of

N
v(x) = a j χ B j (s),
j=1
where B j are disjoint sets, in the norm v∞ = ess supx∈S v(x) X . If above only sets
with mes(B j ) < ∞ are admitted, then the completion of those elements is denoted
by L 0∞ (S; X ).
Then, Remark 3 of Theorem 1.18.4 of [24] and Theorem 5.1.2 of [11] assert with-
out detailed proof the following.
Theorem 1.34 Assume that X 0 and X 1 are Banach spaces and that 1 ≤ p0 < ∞,
0 < θ < 1. Then
(L p0 (S; X 0 ), L 0∞ (S; X 1 ))[θ] = L p (S; (X 0 , X 1 )[θ] )
with p = p0 /(1 − θ ).
Let us consider whether the space L ∞ (S; X ) in [11, 24] is the same as the one in
Definition 1.11 and L 0∞ (S; X ) coincides with L ∞ (S; X ) when S = [0, T ], T < ∞.
The following counterexample shows that the answer is negative.
Counterexample (the case that S = [0, T ], T < ∞)
Let l 2 be the space of sequences (a1 , a2 , · · · , ai , · · · ) such that i ai2 < ∞ and
ei = (0, · · · , ai , · · · ) with ai = 1. Let Si = [ (2 2i−1
−1)T (2 −1)T
i−1 i
, 2i ), i = 1, 2, · · · . Define
a function u : S → l by 2
u(s) = en ∀s ∈ Sn , u(T ) = e1 .
It is easy to verify that when X = l 2 , this function belongs to L ∞ ([0, T ]; X ) in the

sense of Definition 1.11.
Let us consider a sequence { f n } of finite-value functions

kn
f n (s) = xi(n) χ B (n) (s) xi(n) ∈ X, Bi(n) − disjoint measurable subsets of [0, T ].
i
i=1
For every n there exists an i n ≤ kn such that Bi(n)

n
intersects with infinite set of Skin
on measurable set with positive measure. Thus,
∀N , ∃m N ≥ N mes(Sm N ∩ Bi(n)
n
) > 0. (1.23)
On the other hand, since xi(n)

n
∈ l 2,
∞

xi(n)
n
= αl(in ) el , lim αl(in ) = 0.
l→∞
l=1
Thus,
1
∃N , ∀l ≥ N |αl(in ) | < . (1.24)
2
Now let m N of (1.23) be the number corresponding to N of (1.24) and E m(n)N ≡

Sm N ∩ Bi(n)
n
. Then, at s ∈ E m(n)N we have
1
u(s) − f n (s)l 2 = αl(in ) el + (1 − αm(inN) )em N l 2 ≥ .
l=m N
2
By (1.23) mes(E m(n)N ) > 0, and so any sequence { f n } cannot converge to u in the
sense of essential supremum norm. Thus, L 0∞ ([0, T ]; X ) L ∞ ([0, T ]; X ).
Let us show that C([0, T ]; X 1 ) ⊂ L 0∞ ([0, T ]; X 1 ).

If T < ∞, then the set of polynomials { p : p(s) = m k=0 ak s , s ∈ [0, T ], ak ∈
k
X 1 } is dense in the set C([0, T ]; X 1

), where X 1 is a Banach space. Thus, for any ε > 0
there exists a polynomial p(s) = m k=0 ak s such that sups∈[0,T ] u(s) − p(s) X 1 ≤
ε k
ε. We can make a partition Bi = [si , si+1 ] of [0, T ] such that

ε
max max |sik − si+1
k
| ≤ m ε .
k∈{0,1,··· ,m ε } i k=0 ak X 1
m ε
Let us consider a simple function i bi χ Bi (s), where bi = k=0 ak sik . Then, we
have

mε

u(s) − bi χ Bi (s) = u(s) − ak sik χ Bi (s)
i k=0 i

mε
mε
≤ u(s) − p(s) + ak s −
k
ak sik χ Bi (s)
k=0 k=0 i

mε
≤ε+ ak (s k − sik χ Bi (s)) ≤ 2ε,
k=0 i
which shows the asserted conclusion.

Let X be a real Banach space. By Theorem 12.9.2 and (12.9.29) of [8] we get
Theorem 1.35
p ∗
Let 1 < p < ∞, X ∗ a separable Banach space. Then to every f ∈
L (0, T ; X ) corresponds an unique v f ∈ L q (0, T ; X ∗ ), 1/ p + 1/q = 1, such that

u, f = u(s), v f (s) ds, ∀u ∈ L p (0, T ; X )
S
and this correspondence is algebraically and metrically isomorphic.

By Theorem 12.2.11 and (12.2.13) of [8] we get
∗
Theorem 1.36 Let X ∗ be separable Banach space. Then to every f ∈ L 1 (0, T ; X )
corresponds an unique v f ∈ L ∞ (0, T ; X ∗ ) such that

u, f = u(s), v f (s) ds, ∀u ∈ L 1 (0, T ; X )
S
and this correspondence is algebraically and metrically isomorphic.

Remark 1.10 The space X = L 1 (0, 1) is separable, but its dual space X ∗ = L ∞ (0, 1)
is not separable (Example 5.0.31 of [8]). But if the dual X ∗ of a Banach space X is
separable, then X is separable (Lemma of Sect. 2, Chap. 5 of [2]). Thus, if X be a
separable reflective Banach space, then from the fact that X ∗∗ = X is separable it
follows that X ∗ is separable. Therefore, Theorems 1.35, 1.36 are valid when X is a
separable reflective Banach space. When X ∗ is a Gelfand space or a Banach space,
for generalizations of Theorems 1.35, 1.36 we refer to Chap. 12 of [8].
Let S be an open interval, D(S) be the local convex space made from C0∞ (S) by
introducing topology with the semi-norms (1.5) and X w denote a Banach space X
with weak topology.
Definition 1.12 The set L (D(S), X w ) of linear continuous operators from D(S) to
X w is denoted by D ∗ (S; X ) and its element is called a distribution on S with value
in X .
Definition 1.13 The distribution f ∈ D ∗ (S; X ) defined by
f (ϕ) = − f (ϕ ) ∀ϕ ∈ D(S)
is called derivative of f ∈ D ∗ (S; X ).
If f n → f in D ∗ (S; X ), then for y ∈ X ∗ and ϕ ∈ D(S)
lim y, f n (ϕ) − f (ϕ) = lim y, f (ϕ ) − f n (ϕ ) = 0.

n→∞ n→∞
Thus, mapping f → f is continuous in D ∗ (S; X ).
Let V be a real separable reflective Banach space and H be a Hilbert space such
that V → H and V is dense in it. Identifying H and H ∗ , we have
V → H → V ∗ .
Let 1 < p ≤ p0 < ∞ and
X = L p (S; V ) ∩ L p0 (S; H ), u X = u L p (S;V ) + u L p0 (S;H ) . (1.25)
Then,
1 1 1 1
X ∗ = L q (S; V ) + L q0 (S; H ), + = 1, + = 1,
p q p0 q0
f X ∗ = inf max{ f 1 L q (S;V ∗ ) , f 2 L q0 (S;H ) }.
f 1 ∈L q (S;V ∗ )
f 2 ∈L q0 (S;H )
f1 + f2 = f
Define a space W by
W = {u : u ∈ X, u ∈ X ∗ }, uW = u X + u X ∗ . (1.26)
Theorem 1.37 W ⊂ C(S; H ). If S is a compact interval, then the imbedding of W to C(S; H )

is continuous. The following integration by parts holds. For any u, v ∈ W ,
t t

u(t), v(t)) H − u(s), v(s) H = u (τ ), v(τ ) dτ + u(τ ), v (τ ) dτ s, t ∈ S.
s s
From the theorem above we have that when v = u,

1 t
u(t)2H − u(s)2H = u (τ ), u(τ ) dτ. (1.27)
2 s
Remark 1.11 If u = f 1 + f 2 , f 1 ∈ L q (S; V ), f 2 ∈ L q0 (S; H ), then

t t t
u (τ ), v(τ ) dτ = f 1 (τ ), v(τ ) V dτ + ( f 2 (τ ), v(τ )) H dτ.
s s s
(See Lemma 7.3 of [19] for the case of p = p0 , and Theorem 1.17, Chap. 4 of [3] for
the general case.)
1.4.2 Compactness
Let B be a Banach space. For a positive integer m and 1 ≤ p ≤ ∞ denote

∂m f
W m, p
(0, T ; B) = f : f, · · · , m ∈ L (0, T ; B) .
p
∂t
For 0 < σ < 1 and 1 ≤ p ≤ ∞ denote

p
σ, p
T T
f (t) − f (s) B
W (0, T ; B) = f : f ∈ L (0, T ; B),
p
dtds < ∞
0 0 |t − s|σ p+1
with the usual change if p = ∞ (see (iii) of Definition 1.6). For a positive integer m
and 0 < σ < 1 denote

∂ m−1 f ∂m f σ, p
W m+σ, p
(0, T ; B) = f : f, · · · , m−1 ∈ L (0, T ; B),
p
∈ W (0, T ; B) .
∂t ∂t m
If f is defined on [0, T ], then the translated function (τh f )(t) = f (t + h) for h >
0 is defined on [−h, T − h], where 0 < T < ∞.
Let X, B, Y be Banach spaces such that X → B → Y and the imbedding X → B
is compact.
Theorem 1.38 (Theorem 5 of [25]) Let 1 ≤ p ≤ ∞ and F be a bounded set of

L p (0, T ; X ). If
τh f − f L p (0,T −h;Y ) → 0 as h → 0, uniformly for f ∈ F,
then F is relatively compact in L p (0, T ; B) (and in C(0, T ; B) if p = ∞).

Theorem 1.39 (Corollary 5 of [25]) Let 1 ≤ p ≤ ∞, 1 ≤ r ≤ ∞ and F be a

bounded set of L p (0, T ; X ) ∩ W s,r (0, T ; Y ), where s > 0 if r ≥ p and where
s > 1/r − 1/ p if r ≤ p.
Then F is relatively compact in L p (0, T ; B) (and in C(0, T ; B) if p = ∞).
1.5 Operator Equations and Operator-Differential

Equations
1.5.1 Monotone Operator Equation
Let X be a separable real reflective Banach space such that X → H , where H is a

real separable Hilbert space..
Definition 1.14 The operator A : X → X ∗ is said to be radial continuous if s →
A(u + sv), v is continuous on [0, 1] for any fixed u, v ∈ X , hemi-continuous if s →
A(u + sv), h is continuous on [0, 1] for any fixed u, v, h ∈ X , demi-continuous if
from u n → u in X it follows that Au n Au in X ∗ , and Lipschitz continuous if
Au − Av X ∗ ≤ Mu − v X ∃M > 0, ∀u, v ∈ X.
Definition 1.15 Let u, v be any elements of X . The operator A : X → X ∗ is said to

be monotone if
Au − Av, u − v ≥ 0,
strictly monotone if
Au − Av, u − v > 0 for u = v,
strongly monotone if
Au − Av, u − v ≥ αu − v2X α > 0
and coercive if
Au, u
→ +∞ as u X → ∞.
u X
Definition 1.16 The operator A : X → X ∗ is called bounded iff the image of every
bounded set of X is bounded in X ∗ .
Theorem 1.40 Let A : X → X ∗ be radially continuous, monotone and coercive.
Then for any f ∈ X ∗ the set of solutions to
Au = f
is nonempty, weakly closed and convex.

1.5 Operator Equations and Operator-Differential Equations 29
Theorem 1.41 Let A : X → X ∗ be radially continuous, strictly monotone and coer-

cive. Then there exists A−1 : X ∗ → X which is strictly monotone, bounded and demi-
continuous.
Theorem 1.42 Let A : X → X ∗ be radially continuous, strongly monotone. Then

there exists A−1 : X ∗ → X which is Lipschitz continuous. If, in addition, A is Lips-
chitz continuous, then A−1 is strongly monotone.
(For Theorems 1.40–1.42 we refer to Theorems 2.1, 2.2 and Corollary 2.3 of Chap. 3
of [3] or Theorem 26.A of [26]. If A is continuous and strictly monotone, then for
Theorem 1.40 we refer to Theorem 5.16 of [1].)
Remark 1.12 If A is linear, strongly monotone and Lipschitz continuous on Hilbert

space X , Theorem 1.42 means the Lax-Milgram lemma (Theorem 1, Sect. 6.2 of
[18]).
Let X be real Hilbert space and a nonlinear functional be defined by
a( · ; · , · ) : (w, u, v) ∈ X × X × X → a(w; u, v) ∈ R,
where, for w ∈ X , the mapping (u, v) → a(w; u, v) is a bilinear continuous form

on X × X .
Let us consider the following problem:
Find u ∈ X such that
a(u; u, v) = f, v ∀v ∈ X, where f ∈ X ∗ . (1.28)
Theorems 1.2 and 1.3 of Chap. 4 of [14] are rewritten as follows.

Theorem 1.43 Assume that the following hypotheses hold:
(i) there exists a constant α > 0 such that
a(v; v, v) ≥ αv2X ∀v ∈ X.
(ii) the space X is separable and, for all v ∈ X , the mapping
u → a(u; u, v)
is sequentially weakly continuous on X , i.e.
weak − lim u m = u in X implies lim a(u m ; u m , v) = a(u; u, v) ∀v ∈ X.

m→∞ m→∞
Then, problem (1.28) has at least one solution u ∈ X .

Theorem 1.44 Assume that the following hypotheses hold:

(i) there exists a constant α > 0 such that
a(w; v, v) ≥ αv2X ∀v, w ∈ X.
(ii) there exists a continuous and monotonically increasing function L : R+ → R+

such that for all μ > 0
|a(w1 ; u, v) − a(w2 ; u, v)| ≤ L(μ)u X v X w1 − w2 X

∀u, v ∈ X, ∀w1 , w2 ∈ Oμ (0 X ),
where Oμ (0 X ) is μ-neighborhood of 0 X in X . Then, under the condition
f X ∗
L f X ∗ /α < 1
α 2
problem (1.28) has a unique solution u ∈ X .
1.5.2 Pseudo-Monotone Operator Equation
Theorem 1.45 Let A : X → X ∗ be an operator on the real, separable and reflexive

Banach space X . Let A be coercive and bounded. If for every sequence {xn } such
that
xn x in X,
lim supAxn , xn − x ≤ 0 (1.29)
n→∞
there exists a subsequence {xk } such that
lim inf Axk , xk − v ≥ Ax, x − v ∀v ∈ X, (1.30)

k→∞
then for any f ∈ X ∗ there exists a solution to
Au = f.
Remark 1.13 If (1.29) implies (not for a subsequence)
lim inf Axn , xn − v ≥ Ax, x − v ∀v ∈ X,

n→∞
then A is called pseudo-monotone. For the coercive, bounded and pseudo-monotone

operators A, the asserted conclusion was proved (see Theorem 27.A of [26] or Theo-
rem 2.7, Chap. 2 of [27]). However, proofs of the facts that A has property (M) (Propo-
sition 2.5, Chap. 2 in [27]) and A is demi-continuous (footnote of (2.27), Chap. 2 of

[27]), which guarantee the existence of a solution to Au = f , hold with (1.30) for
subsequence, and so we have the conclusion.
1.5.3 Operator-Differential Equations
Definition 1.17 Let X 1 , X 2 be spaces and S = [0, T ], T > 0. Mapping
D(F) ⊂ (S → X 1 ),
F : D(F) → (S → X 2 ),
where (S → X i ) is a set of functions from S to X i , is called Volterra operator iff
u(s) = v(s) a.e. s ∈ [0, t], t ∈ S implies (Fu)(t) = (Fv)(t) a.e. t ∈ S.
Let V be a real separable reflective Banach space and H be a Hilbert space such
that V → H and V is dense in it. Now, let the spaces X and W be the ones in (1.25)
and (1.26).
Theorem 1.46 Let A : X → X ∗ be a radial continuous, monotone, coercive and
Volterra operator. Then, for any f ∈ X ∗ and a ∈ H , there exists a unique solution
u ∈ W ⊂ C([0, T ]; H ) to
u + A u = f, u(0) = a,
and the mapping a → u is continuous from H to C([0, T ]; H ).

(See Theorem 1.1, Chap. 6 of [3] or Sect. 8.1 of [19].)
∗
Definition 1.18 Let A be multivalued operator from D(A) ⊂ X to 2 X (the family
of subsets of X ∗ ) and its graph G(A) = {(x, y) ∈ X × X ∗ ; y ∈ Ax}. The operator
is said to be monotone if
y1 − y2 , x1 − x2 ≥ 0 ∀(x1 , y1 ), (x2 , y2 ) ∈ G(A).
A monotone operator A is said to be maximal monotone if its graph is not properly

∗
contained in graph of any other monotone operator from X to 2 X .
Note that if A is a single-valued operator such that D(A) ⊆ X → X ∗ it is monotone
if
Au − Av, u − v ≥ 0 ∀u, v ∈ D(A)
and it is maximal monotone if, in addition,
f − Av, u − v ≥ 0 ∀v ∈ D(A) implies u ∈ D(A) and Au = f.

Example 1.1
Let V be a real separable reflective Banach space and H be a Hilbert space such that
V → H and V is dense in H . Identifying H and H ∗ , we have
V → H → V ∗ .
Let 1 < p < ∞ and
X = L p (0, T ; V ), u X = u L p (0,T ;V ) .
Then,
1 1
X ∗ = L q (0, T ; V ∗ ), + = 1.
p q
Put
W = {u : u ∈ X, u ∈ X ∗ }, uW = u X + u X ∗ .
Then, the operator L : D(L) ⊂ X → X ∗ defined by
d
L= , D(L) = {u ∈ W : u(0) = 0}
dt
is a maximal monotone linear operator (see Sect. 2.1 of Chap. 3 in [27]).
Theorem 1.47 Let X be a real reflexive Banach space and norms of X and X ∗ be
strictly convex. Let a linear operator L, where D(L) is a dense linear subspace of
X , be maximal monotone. Let the operator A : D(L) → X ∗ satisfy the following:
(i) For every sequence such that xn x in X , xn , x ∈ D(L), L xn L x in X ∗ and
lim supn→∞ Axn , xn − x ≤ 0, there exists a subsequence xk such that
lim inf Axk , xk − v ≥ Ax, x − v ∀v ∈ X ; (1.31)

k→∞
(ii) There exists a function ψ : R+ ≡ [0, ∞) → R+ , which is bounded on any com-

pact interval, and a number θ, 0 ≤ θ < 1, such that
A(x) X ∗ ≤ ψ(x X ) + θ L x X ∗ ∀x ∈ D(L); (1.32)
(iii)
A(x), x
→ ∞, as x X → ∞. (1.33)
x X
Then, for any f ∈ X ∗ there exists a solution to

L x + Ax = f.
Remark 1.14 When
lim inf Axn , xn − v ≥ Ax, x − v ∀v ∈ X,

n→∞
instead of (1.31) holds, the asserted conclusion was proved in [28]. In Theorem 1.2,
Chap. 3 of [27] it is proved by using Theorem 2.7, Chap. 2 of [27]. However, existence
of solutions to the equation holds with (1.31) for subsequence, and so we have the
conclusion (cf. Remark 1.13).
1.6 Convex Functional
Throughout this section, X is a real Banach space with dual X ∗ .
Definition 1.19 A functional f : X → R ≡ R ∪ +∞ is said to be proper if it is not

identically equal to ∞. If f (x) ∈ (−∞, +∞) ∀x ∈ X , then it is said to be finite.
If a function ϕ : X → R satisfies the inequality

ϕ (1 − λ)x + λy ≤ (1 − λ)ϕ(x) + λϕ(y) ∀x, y ∈ X, ∀λ ∈ [0, 1]
then it is said to be convex.

The functional ϕ : X → R is said to be lower semi-continuous (l.s.c.) on X if
lim inf ϕ(u) ≥ ϕ(v) ∀x ∈ X

u→v
or, equivalently, every level subset {x ∈ X : ϕ(x) ≤ λ} is closed.

The functional ϕ : X → R is said to be weakly lower semi-continuous if it is
lower semi-continuous on the space X endowed with weak topology.
Because every level set of a convex functional is convex and every closed convex
set is weakly closed (see Theorem 1.1), we may therefore conclude that a proper
convex functional is lower semi-continuous if and only if it is weakly lower semi-
continuous (see Sect. 1.2 of [29]).
As a special case of Definition 1.2 we have

Definition 1.20 Given a functional f : X → R,
f (x + λh) − f (x)
D f (x, h) = lim x, h ∈ X (1.34)
λ↓0 λ
(if it exists) is called the directional derivative of f at x in direction h.

If D f (x, h) is linear and continuous with respect to h, that is, there exists ∇ f (x) ∈
X ∗ such that
D f (x, h) = ∇ f (x), h ∀h ∈ X,
the functional f : X → R is said to be Gâteaux differentiable at x ∈ X and ∇ f (x) ∈

X ∗ is called the Gâteaux derivative at x.
If D f (x, h) in (1.34) is linear and continuous with respect to h and convergence
is uniform in h on bounded subsets, then f is said to be Fréchet differentiable and
∇ f (x) is called the Fréchet derivative of f at x.
Remark 1.15 For a finite functional f : X → R, if there exists the Gâteaux deriva-
tive at all x ∈ X , then the operator ∇ f : X → X ∗ defined by
x → ∇ f (x)
is called gradient of f . When a finite functional f has its gradient ∇ f, the functional
f is convex if and only if ∇ f : X → X ∗ is monotone (see Theorem 4.4 of [19] or
Lemma 4.10, Chap. 3 of [3]). Every monotone gradient operator of a finite functional
is demi-continuous (see the footnote9 in p. 113 of [19] or Lemma 4.12, Chap. 3 of
[3]).
Definition 1.21 Given a l.s.c., convex, proper function ϕ : X → R, the mapping
∗
∂ϕ : X → 2 X defined by
∂ϕ(x) = {x ∗ ∈ X ∗ : ϕ(y) − ϕ(x) ≥ x ∗ , y − x ∀y ∈ X }
is called the subdifferential of ϕ.

In general, ∂ϕ is a multivalued operator not everywhere defined. We denote by
D(∂ϕ) the set of all x ∈ X for which ∂ϕ(x) = ∅. If a functional ϕ is convex and
there exists Gâteaux derivative ∇ϕ(x) at x, then ∂ϕ(x) = ∇ϕ(x) (see Sect. 1.2 of
[29]). By the definition of ∂ϕ(x) we see that ϕ(x) = inf{ϕ(y); y ∈ X } if and only if
0 X ∗ ∈ ∂ϕ(x).
Example 1.2
Let Ω be a bounded domain, K (Ω) be a nonempty convex closed subset of a Sobolev

space H 1 (Ω), Γ be a open subset of ∂Ω and g(x) ∈ L 2 (Γ ). Let a functional Φ(u) :
H 1 (Ω) → R be defined by
⎧
⎨ g(x)u d x ∀u ∈ K (Ω),
Φ(u) = Γ
⎩
+∞ ∀u ∈
/ K (Ω).
Then the functional Φ is proper, convex and weakly lower semi-continuous.

1.6 Convex Functional 35
Theorem 1.48 (Theorem 2.8 of [29]) Let X be a real Banach space and let Φ : X →
R be a l.s.c. proper convex function. Then ∂Φ is a maximal monotone operator such
∗
that X → 2 X .
In what follows of this section for simplicity let X be a real Hilbert space and X ∗
be identified with X . Let A be a maximal monotone operator in X . For every x ∈ X
the equation
0 X ∈ (xε − x) + ε Axε
has a unique solution xε . Thus
xε = (I + ε A)−1 x,
where I is the unit operator on X . The operator Aε : X → X defined by
1 1
Aε x = (x − xε ) ≡ I − (I + ε A)−1 x
ε ε
is called the Yosida approximation of A.
Let the functional Φ : X → R be proper, convex and weakly lower semi-
continuous (l.s.c.). For every 0 < ε < 1, define a functional Φε by

y − u2X
Φε (y) = inf + Φ(u); u ∈ X , y ∈ X, (1.35)
2ε
which is called the Moreau-Yosida approximation (or regularization) of Φ.

When ∂Φ : X → 2 X is the subdifferential of Φ in the Hilbert space X , taking
into account Theorem 1.48 let
Jε = (I + ε∂Φ)−1 .
Then the Yosida approximation of ∂Φ is as follows:
(∂Φ)ε = ε−1 (I − Jε ) ∀ε > 0.
From Theorem 2.9 of [29], we have

Theorem 1.49 Let X be a real Hilbert space and let Φ : X → R be a l.s.c. proper
convex functional. Then the functional Φε is convex, continuous, Fréchet differen-
tiable, and ∇Φε = (∂Φ)ε . Moreover
y − Jε y2X
Φε (y) = + Φ(Jε y) ∀y ∈ X, (1.36)
2ε
lim Φε (y) = Φ(y), Φ(Jε y) ≤ Φε (y) ≤ Φ(y) ∀y ∈ X. (1.37)

ε→0
Remark 1.16 The operator ∇Φε is Lipschitz continuous with the constant 1/ε (see
Theorem 2.9 and Proposition 2.3 of [29]).
The Moreau-Yosida approximation Φε of a l.s.c. proper convex functional Φ is
finite, convex, Fréchet differentiable, and so ∇Φε is monotone (see Remark 1.15).
If a function u(t) : [0, T ] → X is differentiable at t ∈ [0, T ], then since ∇Φε is
Lipschitz continuous and monotone (for notation we refer to Remark 1.17),

∇Φε u(t + h) − ∇Φε u(t) u(t + h) − u(t)
(∇Φε (u(t)) , u(t) = lim , lim
h→0 h h→0 h
1
= lim 2 ∇Φε u(t + h) − ∇Φε u(t) , u(t + h) − u(t)
h→0 h
≥ 0.
(1.38)
(cf. (5.41), Chap. 1 of [30] or p. 116 of [31])
Remark 1.17 Let V, H be real Hilbert spaces such that V → H and V is dense
in H . Then, we have that H ∗ → V ∗ (see Sect. 1.1.1). Denote by f, v the value of
f ∈ V ∗ at v ∈ V . Identifying H and H ∗ by Riesz’s theorem, we have
V → H → V ∗ . (1.39)
In this case, H → V ∗ means that an element u of H is regarded as an element of

V ∗ in the sense that
u, v = (u, v) H ∀v ∈ V. (1.40)
On the other hand, by Riesz’s theorem, there exists an element Ru = gu ∈ V

corresponding to u ∈ V ∗ such that
u, v = (Ru, v)V = (gu , v)V ∀v ∈ V. (1.41)
According to inner products of V , the functions gu are different, however always
(gu , v)V = (u, v) H ∀v ∈ V.
Sometimes, identifying V and V ∗ by Riesz’s theorem, one finds a continuous

linear functional f on V as an element of V, and so
f, v = ( f, v)V ∀v ∈ V. (1.42)
If f ∈ V is not a functional and is an element of V , then it is also an element of H ,

and is an element of V ∗ by (1.39). In this case, the role of f ∈ V ∗ is represented as
f, v = ( f, v) H ∀v ∈ V, (1.43)
1.6 Convex Functional 37
which is different from (1.42). To avoid confusion, one can write the left-hand side
of (1.42) as R −1 f, v . However, if it is obvious that f represents a functional on V
and is not the functional via (1.39), then for simplicity we use (1.42).
1.7 Some Elementary Inequalities
Theorem 1.50 (Young’s inequality) Suppose that a function f (x) is continuous,

strictly monotone increasing in x ≥ 0, and f (0) = 0. Denote the inverse function of
f by f −1 . For a, b > 0, we have
a b
ab ≤ f (x) d x + f −1 (x) d x
0 0
and the equality holds only if b = f (a). (See Sect. 8.3 of [9].)
In particular, for f (x) = x p−1 (1 < p < ∞), we have that for s, t ≥ 0
tp sq 1 1
st ≤ + , + = 1.
p q p q
By this inequality, we have that for u ∈ L p (Ω) and v ∈ L q (Ω)

ε 1
|u(x)v(x)| d x ≤ u pp + q−1 vqq ∀ε > 0.
Ω p qε
Theorem 1.51 (Bellman-Gronwall’s inequality) Let u(t) and n(t) be real continu-
ous functions on J = [0, T ], f (t) ≥ 0 be integrable on J , and
t
u(t) ≤ n(t) + f (s)u(s) ds.
α
Then for all t ∈ J

t % t &
u(t) ≤ n(t) + f (s)n(s) exp f (r ) dr ds. (1.44)
0 s
In addition, if n(t) is nondecreasing, then

t
u(t) ≤ n(t) exp f (s) ds, (1.45)
0
which is called Gronwall’s inequality

Remark 1.18 For the proof of (1.45) in the case that f ≡ c ≥ 0 and n(t) is a real
and nonnegative function, we refer to Lemma 1.3, Chap. 5 of [3]. For the proof of
(1.45) in the case that u(t), f (t) are continuous and nonnegative functions and n(t)
is a continuous, positive and nondecreasing function, we refer to Theorem 1.1.4 of
[32]. Following the proof of Theorem 1.2.1 of [32], we can prove the general case.
References
1. H. Brézis, Functional Analysis, Sobolev Spaces and Partial Differential Equations (Springer,
2011)
2. K. Yosida, Functional Analysis (Springer, 1995)
3. H. Gajewski, K. Gröger, K. Zacharias, Nichtlineare Operatorgleichungen und Operatordif-
ferentialgleichungen (Academic Press, Berlin, 1974) (Russian 1978)
4. P. Drábek, J. Milota, Methods of Nonlinear Analysis (Birkhäser, 2007)
5. E. Zeidler, Nonlinear Functional Analysis and Its Applications I (Springer, 1986)
6. P. Kučera, Basic properties of solution of the non-steady Navier-Stokes equations with mixed
boundary conditions in a bounded domain. Ann. Univ. Ferrara 55, 289–308 (2009)
7. A.N. Kolmogorov, S.V. Fomin, Element of Theory of Functions and Functional Analysis
(Moscow “Science”, Russian, 1981)
8. H.O. Fattorini, Infinite Dimensional Optimization and Control Theory (Cambridge University
Press, Cambridge, 1999)
9. R.A. Adams, J.J.F. Fournier, Sobolev Spaces (Academic Press, 2005)
10. A. Lunardi, Interpolation Theory, 2nd edn. (Edizioni della Normale, Pisa, 2009)
11. J. Bergh, J. Löfström, Interpolation Spaces (Springer, 1976)
12. V.I. Bogachev, Measure Theory I, II (Springer, 2007)
13. P. Grisvard, Elliptic Problems in Nonsmooth Domains (SIAM, Philadelphia, 1985)
14. V. Girault, P.A. Raviart, Finite Element Methods for Navier-Stokes Equations (Springer, 1986)
15. J.L. Lions, E. Magenes, Non-Homogeneous Boundary Value Problems and Applications I
(Springer, Berlin, 1972)
16. L. Grafakos, Modern Fourier Analysis (Springer, 2009)
17. F. Demengel, G. Demengel, Functional Spaces for the Theory of Elliptic Partial Differential
Equations (Springer, 2012)
18. L.C. Evans, Partial Differential Equations (American Mathematical Society, 2010)
19. T. Roubíček, Nonlinear Partial Differential Equations with Applications (Birkhäuser Verlag,
2005)
20. P.G. Ciarlet, Mathematical Elasticity I: Three-Dimensional Elasticity (North-Holland, 1988)
21. W. Pompe, Korn’s first inequality with variable coefficients and its generalization. Comment.
Math. Univ. Carol. 44, 57–70 (2003)
22. The Mathematical Society of Japan, Encyclopedic Dictionary of Mathematics, 2edn. (The
MIT Press, Cambridge, 1993)
23. E. Hille, R.S. Phillips, Functional Analysis and Semi-Groups (American Mathematical Soci-
ety Providence, Rhode Island, 1957)
24. H. Triebel, Interpolation Theory, Function Spaces, Differential Operators (Johann Ambrosius
Barth Verlag, 1995)
25. J. Simon, Compact sets in the space L p (0, T ; B). Ann. Math. Pure Appl. 146(4), 65–96
(1987)
26. E. Zeidler, Nonlinear Functional Analysis and Its Applications II/B (Springer, 1990)
27. J.L. Lions, Quelques Méthodes de Résolution des Problèmes aux limites Non linéaires (Dunod,
Gauthier-Villars, Paris, 1969) (Russian 1972)
28. H. Brézis, Perturbation non lineaire d’opérateurs maximaux montones. C. R. Acad. Sci. Paris
Ser. I(269), 566–569 (1968)
References 39
29. V. Barbu, Nonlinear Differential Equations of Monotone Types in Banach Spaces (Springer,
2010)
30. G. Duvaut, J.L. Lions, Inequalities in Mechanics and Physics (Springer, Berlin, 1976)
31. J. Naumann, On evolution inequalities of Navier-Stokes type in three dimensions. Ann. Math.
Pure Appl. 124(4), 107–125 (1980)
32. Y. Qin, Integral and Discrete Inequalities and Their Applications I (Birkhäuser, 2016)
Chapter 2
Fluid Equations
In this chapter, we first show how the Navier-Stokes equations and the equations
of motion for fluid under consideration of heat are derived. Next, we outline some
boundary conditions for the Navier-Stokes equations, mainly being concerned with
the ones dealt with in this book. Last, we consider three kind of bilinear forms for the
Stokes and Navier-Stokes equations, variational formulations for the Navier-Stokes
problems with mixed boundary conditions and establish the equivalence between the
variational formulations and the original PDE problems.
2.1 Derivation of Equations for Fluid Motion
In this section, on the basis of conservation laws of mass, momentum and energy in
classical mechanics and physics, we derive the equations of motion for fluid. The
purpose of this section is for readers to be familiar with different kinds of equations
of motion for incompressible fluids as well as for compressible fluids.
2.1.1 Navier-Stokes Equations
2.1.1.1 Acceleration of Fluid Particle
Let (x, t) be coordinates in Rl × (−∞, +∞), l = 2, 3. There are two ways of

describing a fluid motion: Lagrangian (material) description and Eulerian (spatial)
description.

https://doi.org/10.1007/978-3-030-78659-5_2
42 2 Fluid Equations
In Lagrangian description, a kinematic equation of fluid particles x = Φ(x0 , t) =

φ1 (x0 , t), · · · , φl (x0 , t) is given, which represents the location of a particle at t
whose position was x0 at the initial instant t = 0. The velocity and acceleration of a
fluid particle in the Lagrangian description are given, respectively, by
∂Φ(x0 , t) du(x, t) ∂ 2 Φ(x0 , t)

u(x, t) = and = .
∂t dt ∂t 2
In Eulerian description, velocity of fluid particle passing a spatial point x at an
instant t, that is, velocity field v(x, t) = (v1 (x, t), · · · , vl (x, t)) is given.
In the two descriptions, velocities of fluid particles passing x at instant t are the
same, that is, u(x, t) = v(x, t), and acceleration is calculated by
∂ 2 Φ(x0 , t) du(x, t) dv(Φ(x0 , t), t)

= =
∂t 2 dt dt
∂v(x, t) ∂v(x, t) ∂φi (x0 , t)
= + ·
∂t i
∂ xi ∂t
∂v(x, t)
= + (v · ∇)v(x, t).
∂t
The total derivative
dv ∂v
= + (v · ∇)v (2.1)
dt ∂t
is called material derivative of v to emphasize the fact that the derivative is taken
following a fluid element. Thus, acceleration of a fluid particle equals to material
derivative of velocity field v(x, t). Similarly, we can consider material derivative of
other fields, for example, density field ρ(x, t), which is the rate of change of the
quantity following a fluid element.
The instantaneous curves that are everywhere tangent to the velocity vector are
called the streamlines of flow. For unsteady flows the streamline pattern changes with
time. If ∂v
∂t
= 0, then the flow is called steady flow. The path line is the trajectory of
a fluid particle of fixed identity over a period of time. Path lines and streamlines are
identical in a steady flow, but not in an unsteady flow.
The Eulerian description is used in most problems of fluid flows, but the
Lagrangian description is used occasionally, i.e. when one is interested in finding
path lines of fixed particles. In this book we use Eulerian description.
If Φ(x0 , t) of the kinematic equation is continuous with respect to their indepen-
dent variables, then particles filling a domain Ωt at instant t fill a new domain Ωt1
at instant t1 . A domain moving with fluid as this is called a material domain and
volume of material domain is called material volume.
For material domain the following transport theorem is valid.
Theorem 2.1 (Sect. 2.1 of [1]) Let Ωt be a material domain moving with fluid of
velocity field v(x, t) and f (x, t) be scalar or vector function. Then
2.1 Derivation of Equations for Fluid Motion 43

d ∂
f (x, t) d x = f (x, t) + v · ∇ f (x, t) + f (x, t) div v(x, t) d x.
dt Ωt Ωt ∂t
(2.2)
2.1.1.2 Conservation of Mass
Let us fix a subdomain W of a domain Ω occupied by a fluid. Let m(W, t) be the

mass of fluid in W at instant t and ρ(x, t)—density of fluid at (x, t).
In this section, we assume that ρ(x, t) and v(x, t) are differentiable with respect
to x and t. Then, the rate of change of fluid mass in the fixed subdomain W is as
follows.
d d ∂ρ
m(W, t) = ρ(x, t) d x = (x, t) d x.
dt dt W W ∂t
According to the conservation law of mass, the rate of change of fluid mass in the
fixed subdomain W at t equals to the total sum of mass passing through the boundary
of W in duration of unit time, and

∂ρ
(x, t) d x = − ρ(x, t) v(x, t) · n ds,
W ∂t ∂W
where n is outward unit vector normal to the boundary and ds is surface element of
the boundary. Thus, we have

∂ρ(x, t)
+ div ρ(x, t)v(x, t) d x = 0.
W ∂t
Since W is arbitrary, from above we get the differential form of law of conservation
of mass
∂ρ(x, t)
+ div ρ(x, t)v(x, t) = 0, (2.3)
∂t
which is called the continuity equation. In view of material derivative, it can be
written as
dρ(x, t)
+ ρ(x, t) div v(x, t) = 0. (2.4)
dt
If volume of any material domain is constant at all time t, the fluid is called
incompressible. The rate of change of volume equals to the total sum of volume of
fluid passing the boundary of subdomain W in duration of unit time. Since the rate
of change of any volume Ωt of incompressible fluid equals to zero, we have

d
dx = v(x, t) · n ds = div v(x, t) d x = 0.
dt Ωt ∂Ωt Ωt
Thus, incompressibility is equivalent to

div v(x, t) = 0. (2.5)
The continuity equation (2.3) for incompressible fluid is rewritten
dρ(x, t)
= 0,
dt
which means that the density of particles of incompressible fluid is not changed for
duration of motion. Thus, if the density of an incompressible fluid at initial instant
equals over whole domain, that is, ρ(x, 0) = ρ0 = const (homogeneous), then for
any t ≥ 0 the density of fluid ρ(x, t) equals to ρ0 . The problem
⎧
⎨ dρ(x, t) = 0,
dt
⎩
ρ(x, 0) = const
is trivial, and for homogeneous incompressible fluids continuity equation (2.4) is

equivalent to (2.5).
Note that incompressibility of a fluid does not mean the fluid is homogeneous
and for inhomogeneous fluid, incompressibility condition (2.5) is not equivalent to
continuity equation (2.3).
2.1.1.3 Conservation of Momentum
To derive equations of motion for fluid relying on the conservation of momentum,

let us clarify the forces acting on fluid elements. The forces acting on a fluid element
can be divided conveniently into two classes, namely, body forces and surface forces.
Body forces are those that arise from “action at a distance”, without physical
contact. They result from the medium being placed in a certain force field, which
can be gravitational, magnetic, electrostatic or electromagnetic in origin. They are
distributed throughout the mass of the fluid and are proportional to the mass. Body
forces are expressed per unit mass, which is denoted by f .
Surface forces are those that are exerted on an area element by the surroundings
through direct contact. The surface force acting on a small element of area in a
continuous medium depends not only on the magnitude of the area but also on the
orientation of the area. The stress, which is force per unit area on a plane, depends
not only on the magnitude of the force and orientation of the plane but also on the
direction of the force. Thus, specification of stress at a point requires two vectors, one
perpendicular to the plane on which the force is acting and the other in the direction
of the force. Then state of stress at a point is, in fact, specified by a stress tensor,
which has nine components. Let {si j (x)} be stress tensor and n be the outward unit
normal vector at point x of the boundary surface of a fluid element. Then the stress
on the area element of boundary including x is given by
σ (x) = S · n,
which can be decomposed into normal and tangent components.

Stress depends on deformation of a fluid element, which is described in terms of
the strain rate tensor or the deformation rate tensor
⎛ ∂v1
1 ∂v1 ⎞
1 ∂v1 ∂v2 ∂v3
+ +
⎜ ∂ x1 2 ∂ x2 ∂ x1 2 ∂ x3 ∂ x1
∂v3 ⎟
E(v(x)) = εi j (v) = ⎝ 21 ∂∂vx21 + ∂v1 ∂v2 1 ∂v2
+ ∂ x2 ⎠ . (2.6)

1 ∂v3
∂ x2
∂v1
∂ x2
1 ∂v3 ∂v2
2 ∂ x3∂v3
2 ∂ x1
+ ∂ x3 2 ∂ x2
+ ∂ x3 ∂ x3
The relation between the stress and deformation in a continuum is called a constitutive
equation. When the constitutive equation is the first-order equation, that is, the stress
tensor S(v, p) is given by
si j = μ2εi j (v) + Cδi j ,
where δi j is the Kronecker symbol and μ is the dynamic viscosity coefficient, the
fluid is called Newtonian fluid. For thermodynamical reasons, C is of the form C =
λ div v − p, where p denotes the pressure. Therefore, the stress tensor of Newtonian
fluid S(v) is given by
si j = 2μεi j (v) + λ div v δi j − pδi j . (2.7)
(cf. Sect. 5.3 of [2]). By the second law of thermodynamics it is known that μ ≥
0, 2μ + 3λ ≥ 0 (see Exercise 2.11 of [1]). A fluid is called perfect if μ = λ = 0.
(λ + 23 μ is called the coefficient of bulk viscosity. For many applications, the Stokes
assumption
2
λ+ μ=0
3
is found to be sufficiently accurate, and can also be supported from the kinetic theory
of monatomic gases. In [3], the fluid such that λ = − 23 μ holds is called the Newto-
nian fluid.)
The surface force acting on a fluid element W by the surroundings is

S(v, p) · n ds
∂W
and the conservation law of momentum on fluid element W is

∂v(x, t)
ρ + (v · ∇)v(x, t) d x = ρ f (x, t) d x + S(v, p) · n ds.
W ∂t W ∂W
(2.8)
From (2.8) we get the differential equation of conservation law of momentum

∂v(x, t)
ρ + (v · ∇)v(x, t) = ρ f (x, t) + div S(v, p), (2.9)
∂t
T
where div S(v, p) is the vector div S1 (v, p), div S2 (v, p), div S3 (v, p) ,

Si (v, p) = si1 (v, p), si2 (v, p), si3 (v, p) , i = 1, 2, 3.
By (2.7) for Newtonian fluid we have

∂v
ρ + (v · ∇)v = ρ f (x, t) + div 2μE(v) + λ∇(div v) − ∇ p. (2.10)
∂t
Since for vectors w, u

⎛ ⎞
w1 u 1 w1 u 2 w1 u 3
w ⊗ u = ⎝w2 u 1 w2 u 2 w2 u 3 ⎠ ,
w3 u 1 w3 u 2 w3 u 3
we have
div (w ⊗ u) = w div u + (u · ∇)w.
Then,
div (ρv ⊗ v) = div (v ⊗ ρv) = v div (ρv) + ρ(v · ∇)v. (2.11)
In view of (2.11) and continuity equation (2.3), Eq. (2.10) is rewritten as follows:
∂(ρv)
+ div (ρv) ⊗ v = ρ f (x, t) + div 2μE(v) + λ∇(div v) − ∇ p. (2.12)
∂t
Since for Ei (v) = (εi1 (v), εi2 (v), εi3 (v)) (i = 1, 2, 3)
∂
div 2Ei (v) = Δvi + div v (2.13)
∂ xi
(see (2.77)), in the case that μ = const (2.10) and (2.12) become, respectively,

∂v
ρ + (v · ∇)v = ρ f (x, t) + μΔv + (μ + λ)∇(div v) − ∇ p (2.14)
∂t
and
∂(ρv)
+ div (ρv) ⊗ v = ρ f (x, t) + μΔv + (μ + λ)∇(div v) − ∇ p. (2.15)
∂t
2.1.1.4 The Navier-Stokes Equations
Now, from (2.14), (2.15) and (2.3) we get the system of equations of motion for the
Newtonian fluid
⎧
⎪ ∂v
⎪
⎨ρ + (v · ∇)v = ρ f (x, t) + div 2μE(v) + λ∇(div v) − ∇ p,
∂t
(2.16)
⎪
⎩ ∂ρ + div (ρv) = 0
⎪
∂t
or
⎧ ∂(ρv)
⎪
⎨ + div (ρv) ⊗ v = ρ f (x, t) + μΔv + (μ + λ)∇(div v) − ∇ p,
∂t (2.17)
⎩ ∂ρ + div (ρv) = 0.
⎪
∂t
From (2.16) we have the system of equations of motion for the incompressible New-
tonian fluids
⎧
⎪ ∂v
⎪
⎪ρ + (v · ∇)v = ρ f (x, t) + div 2μE(v) − ∇ p,
⎪
⎪ ∂t
⎨
div v = 0, (2.18)
⎪
⎪
⎪
⎪ ∂ρ(x, t)
⎪
⎩ + ∇ρ(x, t) · v(x, t) = 0.
∂t
For the incompressible fluid with a constant density, continuity equation (2.3) is
equivalent to (2.5), and from (2.18) we have
⎧
⎨ ∂v − div 2νE(v) + (v · ∇)v + 1 ∇ p = f (x, t),
∂t ρ (2.19)
⎩
div v = 0,
where ν = μ/ρ is called kinematic viscosity coefficient. In view of (2.13), we have

the equations of motion for incompressible Newtonian fluids with constant kinematic
viscosity and constant density, the Navier-Stokes equations
⎧
⎨ ∂v − νΔv + (v · ∇)v + 1 ∇ p = f (x, t),
∂t ρ (2.20)
⎩
div v = 0.
On the other hand, in the case that l = 2, for convenience, vector v = (v1 (x1 , x2 ),
v2 (x1 , x2 )) is identified with v̄ = (v1 (x1 , x2 ), v2 (x1 , x2 ), 0). Then, we have

∂v2 ∂v1
rot v = rot v̄ ≡ 0, 0, − ,
∂ x1 ∂ x2
and in the case that l = 2, 3
− Δv = rot rot v − ∇ (div v),

1 (2.21)
(v · ∇)v = rot v × v + ∇ |v|2 .
2
By using the relation above, (2.20) is rewritten as
⎧
⎨ ∂v + ν rot rot v + (v · ∇)v + 1 ∇ p = f (x, t),
∂t ρ
⎩
div v = 0,
⎧
⎨ ∂v − νΔv + rot v × v + 1 ∇q = f (x, t),
∂t ρ (2.22)
⎩
div v = 0,
where q = 21 ρ|v|2 + p is called total pressure or Bernoulli’s pressure. To avoid

confusion, sometimes p is called the static pressure.
Remark 2.1 Due to the conservation law of momentum we get a dynamic equation
(2.9). There is another fundamental law of dynamics in classical mechanics, the law
of conservation of angular momentum. For an arbitrary continuous medium obeying
the continuity equation (2.3) and the dynamical equation (2.9), the symmetry of
stress tensor is equivalent to that the equation by conservation of angular momentum
holds (see Theorem 2.1 of [1]). Thus, for Newtonian fluid with symmetry stress
tensor, a dynamical equation by conservation of angular momentum is equivalent to
(2.9). Symmetry of stress tensor is true, however, under a crucial assumption that
all internal momenta of the continuum are neglected which might not be sufficient
for some materials. Therefore, for modeling of liquid crystals, micropolar fluid and
ferrofluid an extra equation for conservation of angular momentum is used (see [4–6]
and references therein).
Remark 2.2 Any displacement of a rigid body can be expressed by composition of

a translation and a rotation. A velocity field v can be regarded as a velocity field of
a rigid body motion if and only if E(v(x, t)) = 0 (see Theorem 5.1 of [2]).
Remark 2.3 Laws of conservation of momentum and angular momentum are uni-
versal, but constitutive laws (equations) depend on physical properties of materials
under consideration. Whenever different nonlinear constitutive equations are taken
for materials, then, correspondingly, different equations for motion of non-Newtonian
fluids are obtained.
2.1.2 Equations of Motion for Fluid Under Consideration of

Heat
2.1.2.1 Conservation of Energy
Let f (x, t) be a scalar function and ρ(x, t) a density field. Then by the transport
theorem we have

d ∂
ρ f dx = (ρ f )(x, t) + div (ρ f v)(x, t) d x
dt Ωt Ω ∂t
t
∂ρ ∂f
= f +ρ + div (ρv) f + ρv · ∇ f d x
Ω ∂t ∂t
t
∂ρ ∂ f
= + div (ρv) f + ρ + v · ∇ f d x.
Ωt ∂t ∂t
Hence, in view of the continuity equation, we get

d ∂ f d
ρ f dx = ρ + v · ∇ f dx = ρ f d x. (2.23)
dt Ωt Ωt ∂t Ωt dt
Let e(x, t) be internal energy of unit mass of fluid at (x, t). The total energy of
a material domain
Ωt at time t is the sum of its kinetic energy and of its internal
energy: Ωt ρ e + 21 |v|2 d x. By (2.23) the rate of change of the total energy of Ωt
is as follows:

d 1 2 d 1
ρ e + |v| d x = ρ e + |v|2 d x.
dt Ωt 2 Ωt dt 2
This equals to sum of the work done by body force and by surface force, heat flux
through the boundary and heat radiated by the inside heat source for the duration
of unit time. Let g(x, t) be mass density of inside heat source and χ be a surface
density of heat. It is known that χ is expressed by
χ = −q · n,
where n is the unit outward normal on the boundary and q is the heat flux vector at
(x, t).
Therefore, taking into account symmetry of S, we have

d 1
ρ e + |v|2 d x
Ωt dt 2

= ρ f · v dx + (S(v)n) · v ds − q · n ds + ρg d x
Ω ∂Ω ∂Ωt Ωt
t t
= ρ f · v dx + div (S1 · v, · · · , Sl · v) d x − div q d x + ρg d x.
Ωt Ωt Ωt Ωt
(2.24)
Taking into account (2.1) and making the Rl -inner product of (2.9) with v, we
have
1 d 2
ρ |v| = ρ f (x, t) · v + div S(v) · v. (2.25)
2 dt
Substituting (2.25) into (2.24), we have

d
ρ e dx = − div S(v) · v d x
Ωt dt Ωt

+ div (S1 · v, · · · , Sl · v) d x − div q d x + ρg d x.
Ωt Ωt Ωt
(2.26)
By symmetry of (2.6),
∂vi
div (S1 · v, · · · , Sl · v) = div S(v) · v + si j = div S(v) · v + si j εi j (v),
ij
∂x j ij
(2.27)
and from (2.26) and (2.27) we get a differential equation for internal energy

∂e
ρ + (v · ∇)e = si j εi j (v) − div q + ρg. (2.28)
∂t ij
In view of (2.7), we have from (2.28) a differential equation for internal energy of
Newtonian fluid

∂e
ρ + (v · ∇)e = 2μE(v) : E(v) + λ(div v)2 − p div v − div q + ρg, (2.29)
∂t

where E(v) : E(v) = i j εi j εi j . The first three terms on the right-hand side express
change of internal energy per unit time by deformation of fluid.
Relying on the rate of change of total energy on a material domain Ωt , we got

(2.24). Now using a fixed element ω, not a material domain, we can get another
equation. For this we additionally need to take care of the flow of total energy
through the boundary of fluid element ∂ω. Then, we have

d 1 1
ρ e + |v|2 d x = ρ e + |v|2 (v · n) ds + ρ f · v dx
dt ω 2 ∂ω 2 ω

+ (S(v)n) · v ds − q · n ds + ρg d x
∂ω ∂Ωt ω

1
= div ρv e + |v|2 d x + ρ f · v dx
ω 2 ω

+ div (S1 · v, · · · , Sn · v) d x − div q d x + ρg d x,
ω ω ω
which yields the differential equation
∂ 1 1
ρ e + |v|2 + div ρ e + |v|2 v
∂t 2 2 (2.30)
= ρ f · v + div (S1 · v, · · · , Sn · v) − div q + ρg.
In view of (2.27), we have from (2.30) the differential equation for total energy
∂ 1 1
ρ e + |v|2 + div ρ e + |v|2 v
∂t 2 2
(2.31)
= ρ f · v + div S(v) · v + si j εi j (v) − div q + ρg.
ij
For Newtonian fluid, taking into account (2.7), we have from (2.31)
∂ 1 2 1
ρ e+ |v| + div ρv e + |v|2 v
∂t 2 2 (2.32)
= ρ f · v + div S(v) · v + 2μE(v) : E(v) + λ(div v)2 − p div v − div q + ρg.
Using (2.7), we can rewrite (2.32) as
∂ 1 1
ρ e + |v|2 + div ρv e + |v|2 v
∂t 2 2
= ρ f · v + div (2μE(v)) · v − ∇ p · v + λ∇(div v) · v
+ 2μE(v) : E(v) + λ(div v)2 − p div v − div q + ρg,
which yields the differential equation for total energy of Newtonian fluid
∂ 1 2 1
ρ e+ |v| + div ρv e + |v|2 + pv
∂t 2 2
= ρ f · v + div (2μE(v)) · v + λ∇(div v) · v (2.33)
+ 2μE(v) : E(v) + λ(div v)2 − div q + ρg.
Let us show that (2.33) is equivalent to (2.29). To this end, we rewrite the terms
on the left-hand side of (2.33).
∂ 1 ∂ρ ∂e ∂ρ |v|2 ∂v
ρ e + |v|2 = e+ρ + + ρv ,
∂t 2 ∂t ∂t ∂t 2 ∂t
1 2 1 (2.34)
div ρv e + |v| + pv = div (ρv)e + ρ(v · ∇)e + div (ρv) |v|2
2 2
+ ρv · (v · ∇)v + ∇ p · v + pdiv v.
Taking into account (2.3) and
dv ∂v
= + (v · ∇)v,
dt ∂t
de ∂e
= + (v · ∇)e
dt ∂t
(see (2.2)) and adding two equations of (2.34), we have
∂ 1 2 1
ρ e+ |v| + div ρv e + |v|2 + pv
∂t 2 2
de dv ∂e 1 d|v|2
=ρ + ρv = ρ( + (v · ∇)e) + ρ + ∇ p · v + p div v.
dt dt ∂t 2 dt
(2.35)
By (2.25) and (2.7), we have
1 d 2
ρ |v| = ρ f (x, t) · v + div S(v) · v
2 dt
= ρ f · v + div (2μE(v)) · v − ∇ p · v + λ∇(div v) · v.
Thus, we get (2.29) from (2.33) taking into account the above and (2.35).
The relation between the heat flux vector and temperature in thermodynamics is
also called a constitutive equation. We take Fourier’s law as the constitutive equation
q = −k∇θ, (2.36)
where k is thermal conductivity and θ (x, t) is temperature of fluid. Denoting the

specific heat of fluid by γ and assuming the specific internal energy of the fluid is
proportional to its temperature, we have
e = γ θ. (2.37)
Specific heat, thermal conductivity and viscosity of fluid, in general, depend on the
temperature of fluid.
Taking into account (2.37), for the incompressible Newtonian fluid with constant
density we have from (2.29)
∂
γ (θ )θ (x, t) − div κ(θ )∇θ + v · ∇(γ (θ )θ ) − 2ν(θ )E(v) : E(v) = g,
∂t
(2.38)
where κ = ρk . In the left-hand side of (2.38) the second term expresses heat flow by
conducting, the third one expresses heat transport by fluid movement (convection)
and the fourth one represents dissipation of kinetic energy by viscous friction of fluid
(the Joule effect).
As in [3], in this book we shall reserve the term convection to describe heat trans-
port by fluid movements, and so (v · ∇)v describing movement of fluid from place
to place is called “advection” (see pp. 55–56 of [3]). Such terminology is suitable
with the terms: the convective heat transfer, convection diffusion, heat convection
and so on.
2.1.2.2 Equations of Motion for Fluid Under Consideration of Heat
From (2.16), (2.17) and (2.29) we get the system of equations of Newtonian fluid
under consideration of internal energy
⎧

⎪ ∂v
⎪
⎪ ρ + (v · ∇)v = ρ f (x, t) + div 2μE(v) + λ∇(div v) − ∇ p,
⎪
⎪ ∂t
⎪
⎨
∂ρ
+ div (ρv) = 0,
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ ∂e
⎩ρ + (v · ∇)e = 2μE(v) : E(v) + λ(div v)2 − p div v − div q + ρg
∂t
(2.39)
or, equivalently,
⎧
⎪ ∂(ρv)
⎪
⎪ + div (ρv) ⊗ v = ρ f (x, t) + μΔv + (μ + λ)∇(div v) − ∇ p,
⎪
⎪ ∂t
⎪
⎨ ∂ρ
+ div (ρv) = 0,
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ ∂e
⎩ρ + (v · ∇)e = 2μE(v) : E(v) + λ(div v)2 − p div v − div q + ρg.
∂t
(2.40)
Remark 2.4 If we use the equation of total energy instead of the equation of inter-
nal energy, we get the system of equations of motion of Newtonian fluids under
consideration of total energy
⎧
⎪ ∂(ρv)
⎪
⎪ + div (ρv) ⊗ v = ρ f (x, t) + μΔv + (μ + λ)∇(divv) − ∇ p,
⎪
⎪ ∂t
⎪
⎪
⎪ ∂ρ
⎨ + div (ρv) = 0,
∂t
⎪
⎪ ∂ 1 2 1 2
⎪
⎪
⎪ ∂t ρ e + 2 |v| + div ρv e + 2 |v| + pv = ρ f · v + div (2μE(v)) · v
⎪
⎪
⎪
⎩
+ λ∇(div v) · v + 2μE(v) : E(v) + λ(div v)2 − div q + ρg.
(2.41)
The total number of equations constituting (2.39) (or (2.40)), (2.36) and (2.37)
is less than the total number of unknown functions, and so one usually needs, in
addition, a relation in terms of pressure, density and temperature. Then, the left-hand
side of the new system obtained from (2.41) is written as
∂u
+ div F(u),
∂t
which is convenient for study of equations for compressible fluid.
When the density of fluid is constant and it is independent of temperature, by
(2.19) and (2.38), we have the system of equations of motion of incompressible
Newtonian fluids under consideration of heat
⎧ ∂v 1
⎪
⎪ − div 2ν(θ )E(v) + (v · ∇)v + ∇ p = f (x, t),
⎪
⎨ ∂t
⎪ ρ
div v = 0, (2.42)
⎪
⎪
⎪
⎪ ∂
⎩ γ (θ )θ (x, t) − div (κ(θ )∇θ ) + (v · ∇(γ (θ )θ )) − 2ν(θ )E(v) : E(v) = g(x, t).
∂t
If a fluid is in the gravitational field and density is not constant, then there is
buoyancy. For slightly compressible homogeneous fluids in gravitational field, we
simplify the equations for the mass and momentum conservation laws as follows:
(i) ρ is assumed to be constant (= ρ0 ) everywhere in the equations, except in the
gravity term;
(ii) In the term of gravity, ρ is replaced by a linear function of θ
ρ = ρ0 − α(θ − θ0 ),
where ρ0 and θ0 are the average value of the density and temperature.
If there is not any body force other than gravity, then f is the gravitational acceleration
gg . Then density of buoyancy is −α(θ − θ0 )gg , and we have an approximation of
the conservation of momentum
∂v 1 1
+ (v · ∇)v − νΔv + ∇ p = [ρ0 − α(θ − θ0 )]gg .
∂t ρ0 ρ0
Replacing the first equation of (2.42) with the above and neglecting energy dissipation
by viscous friction of fluid, we have
⎧ ∂v 1
⎪
⎪ + (v · ∇)v − νΔv + ∇ p = [1 − α1 (θ − θ0 )]gg ,
⎪
⎪ ∂t ρ
⎨ 0
div v = 0, (2.43)
⎪
⎪
⎪
⎩ ∂ γ θ (x, t) − div (κ∇θ ) + (v · ∇(γ θ )) = g,
⎪
∂t
which is used for heat convection in gravitational field and is called Boussinesq
equations.
Putting
[1 − α1 (θ − θ0 )]gg = f (1 − α0 θ )
and adding the work α1 θ f · v done by the force α1 θ f due to heat expansion to the
heat equation, we have
⎧ ∂v 1
⎪
⎪ − div 2ν(θ )E(v) + (v · ∇)v + ∇ p = f (1 − α0 θ ),
⎪
⎪
⎨ ∂t ρ0
div v = 0,
⎪
⎪
⎪
⎩ ∂ γ θ (x, t) − div (κ∇θ ) + (v · ∇(γ θ )) − 2νE(v) : E(v) = α θ f · v + g,
⎪
1
∂t
(2.44)
which is a system of equations of motion for incompressible Newtonian fluid under
consideration of dissipation of energy in gravitational field [7].
2.2 Boundary Conditions for the Navier-Stokes Equations
When solving the Navier-Stokes equations, appropriate initial and boundary con-
ditions need to be applied. The derivation of suitable boundary conditions for flow
problems is not obvious and depends on the physics which is to be modeled.
In this section, without deriving boundary conditions from point of view of
physics, we outline those concerned in this book for the Stokes equations
∂v
− μΔv + ∇ p = f, div v = 0 in Ω ⊂ Rl , l = 2, 3, (2.45)
∂t
and the Navier-Stokes equations
∂v
− μΔv + (v · ∇)v + ∇ p = f, div v = 0 in Ω. (2.46)
∂t
Note that Eq. (2.46) is the one with ρ = 1 in (2.20), and p + 21 |v|2 is the total pressure
instead of p + ρ 21 |v|2 .
Let v = (v1 , · · · , vl ) be a solution to (2.45) or (2.46). The strain tensor E(v) is
the matrix with components
1
εi j (v) = (∂x vi + ∂xi v j )
2 j
and the stress tensor S(v, p) is the one with components
si j = − pδi j + 2μεi j (v).
In the present book the boundary surfaces concerned by us are pieces of boundary
of 3-D or 2-D bounded connected domains, and so we can assume the surfaces are
oriented. Thus, let n be an outward unit normal vector on a boundary and τ be unit
vector tangent to the boundary. Then the stress vector on the boundary is
σ (v, p) = S · n ≡ − pn + 2μE(v) · n, (2.47)
the value of normal component of the stress vector is
σn (v, p) = σ · n
and its tangent component is

στ = σ − σn · n.
Replacing the static pressure p in the stress tensor by the total pressure (Bernoulli’s
pressure) p + 21 |v|2 , we have the total stress tensor S t (v, p) with components
1
sit j = −( p + |v|2 )δi j + 2μεi j (v). (2.48)
2
Then the total stress vector on the boundary is
σ t (v, p) = S t · n,
the value of normal component of the total stress is
σnt (v, p) = σ t · n
and its tangent component is

στt = σ t − σnt · n.
2.2 Boundary Conditions for the Navier-Stokes Equations 57
Note that the tangent components of stress and total stress are equal, i.e. στ = στt .
Denote the value of normal component v · n of velocity v by vn , and the tangent
component of v is vτ = v − vn n.
In some literatures, the terms “stress” and “total stress”, respectively, are used for
(2.6) and (2.47), but in this book we will use the term total stress for (2.48).
From mathematical point of view, there are four kinds of boundary conditions:
(i) Dirichlet boundary condition assigning variable value (velocity, pressure).
(ii) Neumann boundary condition assigning (normal) gradient of variable.
(iii) Combination of Dirichlet and Neumann boundary condition.
(iv) Periodic boundary conditions assuming equality between values of a variable
on distant parallel two planes.
From physical point of view, there are two kinds of boundary conditions:
(i) Natural boundary conditions.
These are boundary conditions on the real boundary of a domain occupied by
fluid, for example, no-slip condition on the wall, threshold slip condition on the
wall and so on.
(ii) Artificial boundary conditions.
These are boundary conditions on imaginary boundaries of fluid domain, for
example, a boundary condition on symmetric planes usually taken to reduce
computational burden, Navier slip-with-friction boundary condition on a sur-
face close to a real rough wall and so on.
According to relationship between fluid domain under consideration and sur-
roundings, the boundaries of fluid domain may be classified as follows:
(i) Wall,
(ii) Symmetric plane,
(iii) Inlet and outlet,
(iv) Free surface.
Boundary conditions first of all must be able to reflect to a mathematical for-
mulation of problem and must be consistent with real phenomena. Some boundary
condition involved to a mathematical formulation has not a correct physical mean-
ing, which is used as an artificial boundary condition for numerical practice. In such
a case for mathematical result to approximate the real phenomenon it is important
where the artificial boundary is drawn.
We outline what kinds of boundary conditions are used on every class of bound-
aries above when the domain of fluid is not variable along with time t.
2.2.1 Boundary Conditions on the Walls
In practice, some domain of fluid has a solid wall boundary and changes of momen-
tum, velocity and scalar quantities near the wall are serious. Thus, boundary condition
on the wall must be given in accordance with physical situations.
2.2.1.1 Stick Boundary Condition on the Walls
Most solid surfaces are impermeable to fluid and the fluid sticks to the surfaces. Thus,
there is no slip and no penetration, and the fluid particles on the wall move with the
velocity of wall w:
v = w.
If the walls do not move, then stick condition on the wall Γ D ⊂ ∂Ω is the homoge-
neous Dirichlet condition
v = 0 on Γ D . (2.49)
Pressure on the wall is seriously variable according to the flow situation, and usually
the pressure boundary condition on the wall is not given.
2.2.1.2 Slip Boundary Conditions on the Walls
When a wall is impermeable to fluid and fluid is not stick, a slip boundary condition
may be used. Accurate experiments clearly demonstrate that the no-slip boundary
condition postulated in most hydrodynamics investigations may be far to be fulfilled
(see [8]).
If a fluid slips without friction on the wall Γ , then Navier slip boundary condition
vn = 0, στ (v) = 0 on Γ (2.50)
is used. Sometimes free slip on a surface Γ is expressed by a vorticity boundary

condition
vn = 0, rot v × n = 0 on Γ. (2.51)
In some papers the boundary condition (2.51) is called Navier slip-type condition
[9–13], slip condition [14] or Hodge boundary condition [15, 16]. On flat portions
of the boundary, the vorticity boundary condition (2.51) coincides with the classical
Navier slip boundary condition (2.50), but usually these are different as a quantity
depending on the shape of boundary surface (cf. Sect. 6 of [10], Remark 1.1 of [14],
Remark 3.1 of [17]). The vorticity boundary condition (2.51) is also used on a portion
of boundary of a tank containing fluid, which is closed by a membrane (see [18]).
C.-L. Navier claimed that the tangential component of the stress at the boundary
should be proportional to the tangential velocity. This condition is expressed by
vn = 0, στ (v) + αv = 0, α : constant or function on Γ, (2.52)
which is called Navier slip-with-friction boundary condition (see [19–21]). The

Navier slip-with-friction boundary condition was rigorously justified as a homoge-
nization of the no-slip condition on a rough boundary and might be more realistic
than the classical no-slip boundary condition when the wall is rough. The condition
is used for simulations of flows near rough boundaries, such as in aerodynamics,
in weather forecast, in haemodynamics, etc. When an artificial boundary condition
near the real boundary with periodic micro-roughness is used, usually α is a matrix,
which maps each vector tangent to boundary into such a vector. From now on, when
we are concern with (2.52) with matrix α, we will always regard the matrix as being
such a one. If the shape operator of the boundary Γ (see Sect. 3.1) is used instead
α/2 in (2.52), then the condition (2.52) becomes (2.51) (see [13] and Sect. 3.1).
2.2.1.3 Threshold Slip Boundary Conditions on the Walls
According to Tresca friction law between two solids, there is not any slip on the
surface when the magnitude of tangent stress is less than a threshold, but when the
magnitude of tangent stress reaches the threshold, slipping can occur, the magnitude
of tangent stress is not greater than the threshold and the direction of velocity is oppo-
site to the tangent stress. Such phenomena between walls and fluids are expressed
by
vn = 0, |στ (v)| ≤ gτ , στ (v) · vτ + gτ |vτ | = 0 on Γ, (2.53)
where gτ is a threshold of tangent stress for slip. This condition is called Tresca fric-
tion boundary condition or threshold slip boundary condition. Using subdifferential
operator, we can rewrite (2.53) as
vn = 0, −στ (v) ∈ gτ ∂|vτ |, (2.54)
where
z
|z|
z = 0, z ∈ Rl ,
∂|z| =
{w ∈ R , |w| ≤ 1} z = 0, z ∈ Rl .
l
Physical and experimental backgrounds of such boundary conditions are mentioned

in several papers [22–24].
2.2.1.4 Leak Boundary Conditions on the Walls
Similarly to the threshold slip boundary condition, leak boundary conditions of fric-
tion type are used when one wants to model a flow problem involving a leak of flow
through the boundary.
Assume that the tangent component of velocity on a potion of boundary vanishes,
there is not any leak through the surface when the magnitude of normal stress on a
portion of boundary surface is less than a given threshold, leak through the surface
can occur if the magnitude of normal stress reaches the threshold and the magnitude
of normal stress is not greater than the threshold. Such phenomena on the boundary
surface are expressed by
vτ = 0, |σn (v)| ≤ gn , σn (v)vn + gn |vn | = 0 on Γ, (2.55)
where gn (> 0) is a threshold of normal stress to show leak [24–27]. Using subdif-
ferential operator, we can rewrite (2.55) as
vτ = 0, −σn (v) ∈ gn ∂|vn |. (2.56)
For physical and experimental backgrounds of such boundary conditions, we refer to

[24, 27]. The boundary condition (2.55) or (2.56) is called threshold leak boundary
condition.
Another threshold leak boundary condition, which is based on the normal total
stress,
vτ = 0, |σnt (v)| ≤ gn , σnt (v)vn + gn |vn | = 0 on Γ (2.57)
may be also used. (Note that since στ (v, p) = στt (v, p) = 2μεnτ (v), there is not
another slip boundary condition based on στt (v, p).)
The boundary conditions (2.55) and (2.57) mean that according to direction of
normal stress, fluid may penetrate out or into through boundary. The stress on a
boundary surface is density of force applied to the boundary from surroundings.
Note n is outward unit vector on a boundary. Therefore, if σn (v) > 0, which means
that direction of force acting on the boundary surface of domain of fluid by surround-
ing is outward, then the value of density of force reacting on the boundary by fluid
equals to −σn (v) < 0, and (2.55) implies vn < 0, which means that fluid penetrates
into through the boundary. If σn (v) < 0, which means that direction of force acting
on the boundary surface of domain of fluid by surrounding is inward, then the value
of density of force reacting on the boundary by fluid equals to −σn (v) > 0, which
means that the reacting force is outward. On the other hand (2.55) implies vn > 0,
which means that fluid penetrates out through the boundary.
In practice, we are encountered with one-sided leak of fluid, for example, sand
layers, a semipermeable membranes and so on.
Assume that the tangent component of velocity on a portion of boundary vanishes, the
fluid can only leak out through boundary, there is not any flow through the boundary
when −σn (v) is less than a threshold g+n (> 0), a leak out can occur if −σn (v)
reaches g+n (> 0) and −σn (v) cannot be greater than the threshold. We can describe
such a phenomenon by
vτ = 0, vn ≥ 0, σn (v) + g+n ≥ 0, (σn (v) + g+n )vn = 0, (2.58)
which is called threshold leak out boundary condition [28, 29].

In contrast, assume that a fluid can only leak into through boundary, there is not
any flow through the boundary when −σn (v) is greater than −g−n (for a threshold
g−n > 0), but there can be a leak into if −σn (v) is same as −g−n , and −σn (v) cannot
be less than −g−n . Such a phenomenon on the boundary surface is expressed by
vτ = 0, vn ≤ 0, σn (v) − g−n ≤ 0, (σn (v) − g−n )vn = 0, (2.59)
which is called threshold leak into boundary condition.

Other threshold one-sided leak boundary conditions
vτ = 0, vn ≥ 0, σnt (v) + g+n ≥ 0, (σnt (v) + g+n )vn = 0,

(2.60)
vτ = 0, vn ≤ 0, σnt (v) − g+n ≤ 0, (σnt (v) − g+n )vn = 0
based on the normal total stress may be also used.

The threshold leak into or out boundary condition is called the one-sided leak
boundary conditions. The threshold slip, leak and one-sided leak boundary conditions
are called the boundary conditions of friction type.
2.2.2 Boundary Conditions on Symmetric Planes
If flow field and geometry of domain of fluid is symmetric with respect to a plane,
then to reduce the amount of computation for the problem a half of domain with the
symmetry plane as an imaginary boundary may be under consideration. The normal
velocity and normal gradients of all variables on a symmetry plane are zero, and
tangent stress on the symmetric plane vanishes. Therefore, on the symmetric plane,
the boundary conditions
vn = 0, στ (v) = 0 (2.61)
or
vn = 0, rot v × n = 0 (2.62)
can be used.
2.2.3 Boundary Conditions on Inlets and Outlets
If there are inlets and outlets on the boundary of fluid domain, then Dirichlet bound-
ary conditions for velocity may be used. But pressure boundary condition is more
common because it is difficult to know velocity profile except special cases. In some
real-life situations, it is natural to prescribe the value of the pressure on some part of
the boundary, as, for instance, in case of pipelines, blood vessels, different hydraulic
systems involving pumps, etc. Thus, it seems perfectly reasonable to impose the
pressure boundary condition on inlets and outlets for the Navier-Stokes system [30].
In practice according to measurement instruments, we can obtain the static pres-
sure p or total pressure 21 |v|2 + p (Bernoulli’s pressure), and both the static pressure
[30, 31] and total pressure [32, 33] may be used for boundary conditions on the real
inlets and outlets.
When pressure boundary conditions are used, direction of fluid flow (usually flow
is orthogonal to the boundary) is given together. But it is known that the static pressure
boundary condition and direction of flow are not enough to determine a velocity field
of the Navier-Stokes equations. Moreover, when one of static pressure (instead of
total pressure), stress (instead of total stress) or the outflow boundary conditions
(see the next subsection) is given on a portion of boundary, for the initial boundary
value problems of the Navier-Stokes equations only existence of a unique local-in-
time solution and a unique solution on a given interval for small given data (in what
follows we call it a solution for small data) are proved. From the mathematical point
of view, the main difficulty of such problems results from the fact that in the process
of a priori estimation of velocity, the term

(v · ∇)v, v Ω
arising from the nonlinear (advection) term (v · ∇)v is not canceled since the normal
component of velocity on the portion of boundary does not vanish. From the mechan-
ical point ofview, it is explained by the fact that the kinetic energy of fluid (with
density 1) 21 Ω |v(t, x)|2 d x is not controlled by the data of problem and uncontrolled
“backward flow” can take place at the portion of boundary (see Preface in [34]).
If the total pressure condition is given, then mathematical treatment is more easy
because by
1
(v · ∇)v = rot v × v + ∇ |v|2
2
the Navier-Stokes equations (2.46) are rewritten as
∂v 1
− μΔv + rot v × v + ∇( p + |v|2 ) = f, ∇ · v = 0 in Ω, (2.63)
∂t 2
and in the process of a priori estimation of solutions the term

rot v × v, v Ω
arising from the nonlinear term rot v × v vanishes by a property of the mixed product
of vectors.
As a boundary condition on in/out-stream surfaces, a combination of the tangent
component of the velocity and the normal component of stress
vτ = 0,
(2.64)
σn (v, p) = f
is also used [35].

If the velocity field v of the Navier-Stokes equations on a domain is obtained,
then by de Rham Theorem (Proposition 2.3), pressure p is determined with different
constants. If, moreover, pressure condition on a portion of boundary is given, then
the constant is determined and pressure is uniquely determined when the solution
is smooth enough for existence of traces of velocity and pressure on the portion of
boundary. Also, if a boundary condition including pressure p as addend (stress, total
stress, normal stress) on a portion of boundary is given, then the pressure on the
domain is determined uniquely when the solution is smooth enough (see Theorem
4.1 of [36] and Theorems 5.1, 5.3).
2.2.4 Outflow Boundary Conditions on Imaginary Boundary
To reduce amount of computation, imaginary boundaries may be drawn inside flow.

On the portions of imaginary boundary where flow is out, velocity and pressure
boundary conditions are not used since the velocity and pressure on the potion of
imaginary boundary are not known prior to prescription of solution.
In such portions of imaginary boundary, the free outflow boundary conditions
∂v
μ − pn = 0 (2.65)
∂n
[34, 37–47], which is called “do nothing” boundary condition, and
μE(v)n − pn = 0 (2.66)
[47, 48] are used. “Do nothing” boundary condition (2.65) results from variational
formulation based on Dirichlet bilinear form (∇v, ∇u) obtained by integrating by
parts (Δv, u) (see (2.86)) and does not have a real physical meaning, but is rather
used in truncating large physical domains to get smaller computational domains. The
condition is appropriate where the exit flow is close to a fully developed one and
the normal gradient for velocity is close to zero. Thus, the artificial boundary for the
outflow boundary condition must be placed downstream fully and stream line similar
to parallel (see [37]).
2.2.5 Boundary Conditions on Free Surfaces
Sometimes, we are encountered with a boundary between two fluids, which is called
free surface. A common example occurs when a liquid film flows down an inclined
plane. The surface of the liquid film in contact with the surrounding gas is a fluid-
fluid interface. Other examples include the interface between a liquid drop and the
surrounding continuous phase or that between two liquid layers. Free surfaces are
unknowns to be determined and require two boundary conditions to be applied.
(i) A kinematic condition: This condition relates the motion of the free interface
to the fluid velocities at the free surface.
Let v = (v1 , v2 , v3 ) be the velocity field of one fluid and the position of a free
surface be written in an implicit form f (x, t) = 0. Fluid particles on the free surface
do not move across the free surface, therefore we can write
∂f ∂f ∂f
+ vi = + v · ∇ f = 0, (2.67)
∂t i
∂ xi ∂t
which is a kinematic condition. When a free surface is written in the form x3 =

h(x1 , x2 , t), the kinematic boundary condition is
∂f ∂f ∂f
v3 = + v1 + v2 .
∂t ∂ x1 ∂ x2
∂f
For steady problems, we have ∂t
= 0, and the kinematic condition is written as
v · n = 0,
where n is normal unit vector on the surface, since ∇ f is a normal vector on the
surface given by f (x) = 0. This condition means that there is no flow through the
free surface. Note that there can be a flow tangent to the free surface.
(ii) A dynamic condition: This condition is concerned with force balance on the
free surface.
When one is concerned with a free surface, if the capillary (surface) force is
neglected, then the fluid is called a heavy fluid and if the capillary force is essential,
then the fluid is called a capillary fluid.
For a heavy fluid, the traction exerted by fluid (1) onto fluid (2) is equal and
opposite to the traction exerted by fluid (2) onto fluid (1). Since the outward normal
vectors on the boundary between two domains of fluids are opposite, we have the
dynamic boundary condition
σ (1) = −σ (2) .
If n is the normal unit vector oriented from fluid (1) to fluid (2) on the surface, then
by (2.47) we have
σ (1) ≡ S (1) · n = −σ (2) ≡ S (2) · n, (2.68)
where S (i) is the stress tensor of fluid (i). If fluid (2) is air, then the dynamic condition
on the free surface of fluid (1) is rewritten as (see (1.16), Sect. 3.1 of [49])
σ ≡ − pn + 2μE(v) · n = − pa n, (2.69)
where pa is the air pressure.

For free surface of capillary fluids, surface tensions must be taken into account.
The pressure jump induced by surface tension is given by
sk,
where s is the surface tension and k is twice the mean curvature of the free surface
(see Sect. 3.1.1). Surface tension acts like a tensioned membrane at the free surface
and tries to minimize the surface area. Hence the pressure inside of fluid including
the center of curvature tends to be higher than other. Thus, the dynamic boundary
condition for the capillary fluid on free surfaces is
S (1) · n + skn = S (2) · n, (2.70)
where k > 0 if the centers of curvature lie inside fluid (1) when the free surface is a
sphere or a circle. Moreover, if a free surface is convex (concave) (see Definition 3.2.)
with respect to the normal unit vector from fluid (1) to fluid (2) on the surface, then
k ≥ 0 (k ≤ 0) (see Chap. 4 of [3] and Sect. 3.1.1).
In what follows let us further consider heavy fluids. If a free surface to be deter-
mined is steady and a fluid may move freely along the free surface, then we have a
free boundary condition
v · n = 0,
(2.71)
σ (v, p) = 0,
which is useful in many physical problems, especially in mathematical modeling of

coating flows, flows of melted semiconductors (see [50] and references therein).
When one fluid is air, sometimes free surfaces are regarded as fixed surfaces.
Then, the kinematic boundary condition vanishes and the boundary condition on the
fixed surface is (2.69). Neglecting the press of air, we get
σ (v, p) = 0,
which in some papers is called Neumann boundary condition (see [51] and references
therein, [15]).
If there is no flow across a fixed surface and the fluid flows along the surface, then
a boundary condition
v · n = 0,
(2.72)
στ (v, p) = f τ
may be used.
In practice we deal with mixture of some kinds of boundary conditions. For a
channel flow a mixture of Dirichlet condition v = 0 on the wall and “do nothing”
condition on the outlet is used. But for a channel flow with a rough boundary surface a
mixture of Dirichlet condition, the Navier slip-with-friction boundary condition and
“do nothing” condition may be used. For a flow in a vessel with in/outlet a mixture
of Dirichlet condition v = 0 on the wall and pressure conditions on the inlet/outlet is
used. But for the flow in a vessel with in/outlet and a free surface a mixture of Dirichlet
condition, a Neumann condition 2νε(v)n − pn = 0 and pressure conditions may be
used.
There is a vast body of literature for the Stokes and Navier-Stokes problems with
mixed boundary conditions, and several variational formulations have been used for
them.
2.3 Bilinear Forms for Hydrodynamics
In this section, we consider three types of bilinear form reduced from (Δv, u) by
integration by parts and outline the variational formulations of the Navier-Stokes
problems with mixed boundary conditions.
From now on, in this book we use the following notations.
Let Ω be an open subset of Rl , l = 2, 3.
When X is a Banach space, X = X l and X∗ is the dual of X. Let W k,α (Ω)
be Sobolev spaces, H k (Ω) = W k,2 (Ω), and so Hk (Ω) = {H k (Ω)}l , H0k (Ω) =
{W0k,2 (Ω)}l .
An inner product and norm in the space L2 (Ω) or L 2 (Ω) are, respectively, denoted
by (· , ·) and · . · , · means the duality pairing between a Sobolev space X and
its dual one.
Also, (· , ·)Γi is an inner product in the L2 (Γi ) or L 2 (Γi ), and · , ·Γi means the
duality pairing between H 2 (Γi ) and H− 2 (Γi ) or between H 2 (Γi ) and H − 2 (Γi ) (with
1 1 1 1
1
an exception in Remark 5.1 concerned with the duality pairing between H002 (Γi ) and
1
(H002 (Γi ))∗ ).
The inner product and norms in Rl , respectively, are denoted by (· , ·)Rl and | · |.
Sometimes a · b or ab is used for inner product in Rl between a and b. For simplicity,
when A is a matrix and n is vector, sometimes we write A · n by An.
2.3 Bilinear Forms for Hydrodynamics 67
2.3.1 Bilinear Forms
2.3.1.1 Dirichlet Bilinear Form
Unless otherwise specified, let Ω be a bounded domain with Lipschitz boundary

of Rl , l = 2, 3, and φ ∈ H 1/2 (∂Ω). Then there exists a continuous linear operator
H 1/2 (∂Ω) → H 1 (Ω) (lift operator), and for every φ ∈ H 1/2 (∂Ω) there exists a
w ∈ H 1 (Ω) with
w|∂Ω = φ, w H 1 (Ω) ≤ c φ H 1/2 (∂Ω) , (2.73)
where c is independent of φ (Theorem 1.5.1.2 of [52] or (1.3.15)).

Let us introduce the space

H (div; Ω) = v ∈ L2 (Ω); div v ∈ L 2 (Ω) ,
which is a Hilbert space for the inner product
(v, u) H (div;Ω) = (v, u) + (div v, div u).
If Ω is an open subset with Lipschitz boundary of Rl (not necessarily bounded), then

the space D(Ω̄)l is dense in H (div; Ω) (see Theorem 2.4, Chap. 1 of [53]).
For u ∈ H (div; Ω), let us consider the functional on φ ∈ H 1/2 (∂Ω) defined by
Fu (φ) = (div u, w) + (u, ∇w).
By (2.73) we have

|Fu (φ)| ≤ c u L2 + div u L2 φ H 1/2 (∂Ω) ,
which shows that the functional Fu is linear and continuous on H 1/2 (∂Ω). Since
(div u, w) + (u, ∇w) = (u · n, φ)∂Ω for u ∈ C∞ (Ω),
where n is the outward unit normal vector on ∂Ω, we can define a trace of u · n
γn u ∈ H −1/2 (∂Ω)
as a generalization of u · n of smooth functions u. Thus, we have

Proposition 2.1 Let Ω be a bounded domain with Lipschitz boundary of Rl . The
mapping γn : v → v · n|∂Ω defined on D(Ω̄)l can be extended by continuity to a
linear and continuous mapping, from H (div; Ω) into H −1/2 (∂Ω).
Moreover, the following Green’s formula holds:
(div u, w) + (u, ∇w) = γn u, w∂Ω ∀u ∈ H (div; Ω), ∀w ∈ H 1 (Ω). (2.74)

(cf. Theorem 2.5 and (2.17), Chap. 1 of [53]).
By Proposition 2.1, for v ∈ H1 (Ω) with Δv ∈ L2 (Ω) and w ∈ H1 (Ω), we have

− (Δv, w) = (∇vi , ∇wi ) − ∂n vi , wi ∂Ω , (2.75)
i i
∂vi

where ∂n vi = γn (∇vi ). Having in mind the fact that γn (∇vi ) = ∂n ∂Ω
for smooth
functions vi , in what follows we write (2.75) as
− (Δv, w) = (∇v, ∇w) − ∂n v, w∂Ω . (2.76)
We call (∇v, ∇w) Dirichlet bilinear form.
2.3.1.2 Strain Bilinear Form
When v ∈ H1 (Ω), let us prove
∂
div 2Ei (v) = Δvi + div v in H −1 (Ω), (2.77)
∂ xi
where Ei (v) = (εi1 (v), εi2 (v), εi3 (v)), i = 1, 2, 3. If q ∈ C0∞ (Ω), then we have

∂vi ∂v j ∂q
2 (div Ei (v))q dω = −2 Ei (v) · ∇q dω = − + dω
Ω Ω ∂x j
Ω j
∂ xi ∂ x j

∂v ∂v
=− ∇vi · ∇q dω − · ∇q dω = Δvi · q dω + div · q dω.
Ω Ω ∂ xi Ω Ω ∂ xi
Since C0∞ (Ω) is dense in H01 (Ω), from above we get (2.77).
Let v ∈ H2 (Ω) and u ∈ H1 (Ω). By (2.74) and (2.77), we have
∂v
− (Δvi , u i ) = 2(Ei (v), ∇u i ) − 2(Ei (v) · n, u i )∂Ω + (div , u i ). (2.78)
∂ xi
Since the tensor E(v) is symmetric, we have
∂u i
E(v), ∇u = Ei (v), ∇u i = εi j (v),
i i j
∂x j
(2.79)
1 ∂u i ∂u j
= εi j (v), + = E(v), E(u) .
i j
2 ∂x j ∂ xi
Taking into account (2.79), we have from (2.78)

−(Δv, u)Ω = 2(E(v), E(u)) + (∇(div v), u) − 2(E(v)n, u)∂Ω

= 2(E(v), E(u)) − (div v, div u) − 2(E(v)n, u)∂Ω + (div v, u · n)∂Ω .
(2.80)
If v ∈ H1 (Ω), Δv ∈ L2 (Ω) and div v = 0, then by (2.77) div Ei (v) ∈ L 2 (Ω), and
there exists γn (Ei (v)) ∈ H −1/2 (∂Ω).
Therefore for v ∈ H1 (Ω) with Δv ∈ L2 (Ω), div v = 0, and u ∈ H1 (Ω), we have
− (Δv, u)Ω = 2(E(v), E(u)) − 2(E(v)n, u)∂Ω , (2.81)
where E(v)n|∂Ω = (γn (E1 (v)), γn (E2 (v)), γn (E3 (v)) ∈ H−1/2 (∂Ω).
We call (E(v), E(u)) the strain bilinear form.
2.3.1.3 Vorticity Bilinear Form
Let us introduce the space

H (rot; Ω) = v ∈ L2 (Ω); rot v ∈ L2 (Ω) .
When Ω is a domain with Lipschitz boundary of Rl (not necessarily bounded),

D(Ω̄)l is dense in H (rot; Ω) (Theorem 2.10, Chap. 1 of [53]).
Proposition 2.2 Let Ω be a domain with Lipschitz boundary of R3 (not necessarily

bounded). The mapping γτ : u → u × n|∂Ω defined on D(Ω̄)3 can be extended by
continuity to a linear and continuous mapping, from H (rot; Ω) into H−1/2 (∂Ω).
Moreover, the following Green’s formula holds:
(rot u, w)Ω − (u, rot w)Ω = − γτ u, w∂Ω ∀w ∈ H1 (Ω). (2.82)
(cf. Theorem 2.11, Chap. 1 of [53]). In what follows for convenience we write γτ u
by u × n.
Whenever we are concerned with vorticity of vector fields with l = 2, for

convenience as before, vector u = (u 1 (x1 , x2 ), u 2 (x1 , x2 )) is identified with ū =
(u 1 (x1 , x2 ), u 2 (x1 , x2 ), 0).
Thus, for u = (u 1 (x1 , x2 ), u 2 (x1 , x2 )), v = (v1 (x1 , x2 ), v2 (x1 , x2 )) and n = (n 1 , n 2 ),
we think that

∂u 2 ∂u 1
rot u = rot ū ≡ 0, 0, − ,
∂ x1 ∂ x2
u × n = ū × n̄ = (0, 0, u 1 n 2 − u 2 n 1 ),

∂v2 ∂v1 ∂v2 ∂v1
rot v × n = rot v̄ × n̄ = −n 2 − , n1 − ,0 ,
∂ x1 ∂ x2 ∂ x1 ∂ x2

(rot v, rot u)Ω = rot v̄ · rot ū d x,
Ω

(rot v × n, u)∂Ω = (rot v̄ × n̄) · ū dσ.
∂Ω
Corollary 2.1 Let Ω be a bounded domain with Lipschitz boundary of Rl , l = 2, 3.

If v ∈ H1 (Ω), Δv ∈ L2 (Ω) and div v = 0, then there exists γτ (rot v) ≡ rot v × n ∈
H−1/2 (∂Ω) and the following holds:
−(Δv, u)Ω = (rot v, rot u)Ω − rot v × n, u∂Ω . (2.83)
Proof Since
−Δv = rot rot v − grad(div v)
(see (2.21)) and div v = 0, we have rot rot v ∈ L2 (Ω).

If l = 3, then by Proposition 2.2 we come to the conclusion.
If l = 2. Let n be the outward unit normal on ∂Ω and n̄ = (n 1 , n 2 , 0). On the
domain Ω̃ = Ω × (0, 1) ⊂ R3 let us consider vector fields v̄, ū. Then by Proposi-
tion 2.2 we have
−(Δv̄, ū)Ω̃ = (rot v̄, rot ū)Ω̃ − (rot v̄ × ñ, ū)∂ Ω̃ , (2.84)
where

∂ Ω̃ = ∂Ω × (0, 1) Ω × {0} Ω × {1}
outward normal on ∂ Ω̃. Since both

and ñ is the vectors rot v̄ and ñ are orthogonal to
Ω × {0} Ω × {1}, rot v̄ × ñ = 0 on Ω × {0} Ω × {1}. Then, noting that ñ = n̄
on ∂Ω × (0, 1), we have

(rot v̄ × ñ, ū)∂ Ω̃ = (rot v̄ × n̄) · ū dσ = (rot v̄ × n̄) · ū dσ = (rot v × n, u)∂Ω .
∂Ω×(0,1) ∂Ω
Also,

(Δv̄, ū)Ω̃ = Δv̄ · ū d x = Δv̄ · ū d x = (Δv, u)Ω ,
Ω̃ Ω

(rot v̄, rot ū)Ω̃ = rot v̄ · rot ū d x = rot v̄ · rot ū d x = (rot v, rot u)Ω .
Ω̃ Ω
Therefore, from (2.84) we get (2.83) for l = 2.
We call (rot v, rot u) the vorticity bilinear form.
2.3.2 Variational Formulations for Mixed Boundary Value

Problems of the Navier-Stokes Equations
Now, we will briefly outline the main points of variational formulations for the Stokes
and Navier-Stokes problems with mixed boundary conditions.
In this subsection from now on, let Ω be a bounded domain with Lipschitz bound-
ary of Rl , l = 2, 3. Let us introduce the following spaces of divergence-free func-
tions:
V = φ ∈ D(Ω)l ; div φ = 0 , V = v ∈ H10 (Ω); div v = 0 .
The space V is dense in V for the norm of H1 (Ω) (Corollary 2.5, Chap. 1 of [53]).
The following proposition which is a simplified version of de Rham’s theorem
(see Proposition 1.1, Chap. 1 of [54]) holds.
Proposition 2.3 (Theorem 2.3, Chap. 1 of [53]) If f ∈ H−1 (Ω) satisfies
f, φ = 0 ∀φ ∈ V ,
then there exists p ∈ L 2 (Ω) such that
f = ∇ p.
If Ω is connected, then p is unique up to an additive constant.
2.3.2.1 Variational Formulation Based on Dirichlet Bilinear Form
Let ∂Ω = Γ D ∪ Γ N , where Γ D , Γ N are open subsets of ∂Ω such that Γ D ∩ Γ N = ∅,

and
V D = v ∈ H1 (Ω); div v = 0, v|Γ D = 0 .
Let us consider the problem

⎧
⎪
⎪ − μΔv + (v · ∇)v + ∇ p = f in Ω,
⎪
⎪
⎪
⎨ div v = 0 in Ω,
⎪ v|Γ D = 0, (2.85)
⎪
⎪
⎪
⎩ μ ∂v − pn |Γ N = g.
⎪
∂n
Assuming that (v, p) is smooth and making duality product with u ∈ V D in the first
equation, by (2.76) we get
μ(∇v, ∇u) + (v · ∇)v, u = f, u + g, uΓ N ∀u ∈ V D . (2.86)
Thus, when f ∈ V∗D , g ∈ H−1/2 (Γ N ) are given, a function v ∈ V D satisfying (2.86)

is called a solution to problem (2.85).
To show that the concept of solution is well defined, we will prove that if a solution
v is smooth (v ∈ H2 (Ω), f ∈ L2 (Ω)), then there exists a p such that (v, p) satisfies
(2.85). By (2.76) we get from (2.86)
− μ(Δv, u) + (v · ∇)v, u + μ ∂n v, uΓ N = f, u + g, uΓ N ∀u ∈ V D .

(2.87)
Taking any u ∈ V , we have from (2.87)
(−μΔv + (v · ∇)v − f, u) = 0.

By Proposition 2.3, there exists a unique function P ∈ L 2 (Ω) such that Ω P dx = 0
and
− μΔv + (v · ∇)v − f = −∇ P. (2.88)
Since v ∈ H2 (Ω), f ∈ L2 (Ω) and (v · ∇)v ∈ L2 (Ω), we have that ∇ P ∈ L2 (Ω),

which shows that P ∈ H 1 (Ω). Substituting (2.88) into (2.87) and using integration
by parts, we have
μ∂n v − Pn − g, uΓ N = 0. (2.89)
1/2
For any tangent vector field φ ∈ H00 (Γ N ) on Γ N there exists its continuation φ̃ ∈
H1/2 (∂Ω) such that φ̃ = 0 on ∂Ω\Γ N (see Theorem 1.18). There exists a solution
u ∈ H1 (Ω) to the Stokes problem
⎧
⎨ − Δu + ∇ p = 0,
⎪
div u = 0,
⎪
⎩
u |∂Ω = φ̃
(see Theorem IV.1.1 of [55]), and u ∈ V D . Taking such u in (2.89), we see that the
tangent components of μ∂ n v − Pn − g vanish. Similarly, for any normal vector field
1/2
φ ∈ H00 (Γ N ) such that Γ N φ · n ds = 0 there exists a u ∈ V D such that u|Γ N = φ.
Using such u in (2.89), we can verify that the values of the normal components of
μ∂n v − Pn − g must be the same. Therefore, we have
∂v
μ − Pn − g = cn on Γ N , (2.90)
∂n
where c is some constant. Putting p = P + c, we get the boundary condition on Γ N .

Thus, taking into account v ∈ V D , we know that (v, p) satisfies (2.85).
If Γ D = ∅, then by Friedrichs’ inequality the norm v V D is equivalent to ∇v L2
and
(∇ v, ∇ v) ≥ c v 2H1 (Ω) , (2.91)
which is fundamental for study of the problem (2.86).
2.3.2.2 Variational Formulation Based on Strain Bilinear Form

3
Let ∂Ω = Γ D ∪ i=1 Γ i , where Γ D , Γi are open subsets of ∂Ω such that Γ D ∩
Γi = ∅, Γi ∩ Γ j = ∅ for i = j and

V D = v ∈ H1 (Ω); div v = 0, v|Γ D = 0, vτ |Γ1 = 0, vn |Γ2 = 0 .

⎧
⎪ − μΔv + (v · ∇)v + ∇ p = f in Ω,
⎪
⎪
⎪
⎪ div v = 0 in Ω,
⎪
⎪
⎪
⎨ v| = 0,
ΓD
(2.92)
⎪
⎪ vτ |Γ1 = 0, (− p + 2μεnn (v))|Γ1 = φ1 ,
⎪
⎪
⎪
⎪ vn |Γ2 = 0, 2(μεnτ (v) + αvτ )|Γ2 = φ2 ,
⎪
⎪
⎩
(− pn + 2μεn (v))|Γ3 = φ3 ,
where εn (v) = E(v) · n, εnn (v) = (E(v) · n, n)Rl , εnτ (v) = E(v) · n − εnn (v)n.
Assuming that (v, p) is smooth and making duality product with u ∈ V D in the
first equation, by (2.81) we get
2μ(E(v), E(u)) + (v · ∇), u + 2(αvτ , u)Γ2

(2.93)
= f, u + φ1 , u n Γ1 + φ2 , uΓ2 + φ3 , uΓ3 ∀u ∈ V D .
Thus, when f ∈ V∗D , φ1 ∈ H −1/2 (Γ1 ), φ2 ∈ H−1/2 (Γ2 ), φ3 ∈ H−1/2 (Γ3 ) are given,
a function v ∈ V D satisfying (2.93) is called a solution to problem (2.92).
If we are concerned with
⎧
⎪
⎪ − μΔv + (v · ∇)v + ∇ p = f in Ω,
⎪
⎪
⎪
⎪ div v = 0
⎪
⎪
in Ω,
⎪
⎪ v| =
⎪
⎪ Γ 0,
⎨ D
1 2 (2.94)
⎪ vτ | Γ = 0, − p + |v| + 2με nn (v) = φ1 ,
⎪
⎪
1
2 Γ1
⎪
⎪
⎪
⎪ vn |Γ2 = 0, 2(μεnτ (v) + αvτ )|Γ2 = φ2 ,
⎪
⎪
⎪
⎪ − p + 1 |v|2 n + 2με (v) = φ ,
⎪
⎩ n 3
2 Γ3
which includes total stress boundary conditions, then by
1
(v · ∇)v = rot v × v + ∇ |v|2
2
we can rewrite the Navier-Stokes equations as
∂v 1
− μΔv + rot v × v + ∇( p + |v|2 ) = f, div v = 0 in Ω, (2.95)
∂t 2
and using (2.81) for a smooth solution we get
2μ(E(v), E(u)) + rot v × v, u + 2(αvτ , u)Γ2

(2.96)
= f, u + φ1 , u n Γ1 + φ2 , uΓ2 + φ3 , uΓ3 ∀u ∈ V D .
Thus, when f ∈ V∗D , φ1 ∈ H −1/2 (Γ1 ), φ2 ∈ H−1/2 (Γ2 ), φ3 ∈ H−1/2 (Γ3 ) are given,
a function v ∈ V D satisfying (2.96) is called a solution to problem (2.94).
To show that the concepts of solution are well defined, as above we can prove that
if a solution v is smooth (v ∈ H2 (Ω), f ∈ L2 (Ω)), then there exists a p such that
(v, p) satisfies (2.92) or (2.94). For more general cases, we will prove such a fact in
Chap. 5.
If Γ D = ∅, then by Korn’s and Friedrichs’ inequalities the norm v V D is equiv-
alent to E(v) L2 and
(E(v), E(v)) ≥ c v 2H1 (Ω) , (2.97)
which is fundamental for study of the problems (2.93) and (2.96).
2.3.2.3 Variational Formulation Based on Vorticity Bilinear Form

2
Let ∂Ω = Γ D ∪ i=1 Γ i , where Γ D , Γi , i = 1, 2, are open subsets of ∂Ω such
that Γ D ∩ Γi = ∅, Γi ∩ Γ j = ∅ for i = j and

V D = v ∈ H1 (Ω); div v = 0, v|Γ D = 0, vτ |Γ1 = 0, vn |Γ2 = 0 .

⎧
⎪
⎪ − μΔv + (v · ∇)v + ∇ p = f in Ω,
⎪
⎪
⎪
⎨ div v = 0
⎪ in Ω,
v|Γ D = 0, (2.98)
⎪
⎪
⎪ vτ |Γ1 = 0, − p|Γ1 = φ1 ,
⎪
⎪
⎪
⎩
vn |Γ2 = 0, (rot v × n)|Γ2 = φ2 /μ.
Assuming that (v, p) is smooth and making duality product with u ∈ V D in the first
equation, by (2.83) we get
μ(rot v, rot u) + (v · ∇)v, u = f, u + φ1 , u n Γ1 + φ2 , uΓ1 ∀u ∈ V D ,

(2.99)
where (rot v × n, u)|Γ1 = 0 was used.
Thus, when f ∈ V∗D , φ1 ∈ H −1/2 (Γ1 ) and φ2 ∈ H−1/2 (Γ2 ) are given, a function
v ∈ V D satisfying (2.99) is called a solution to problem (2.98).
Like (2.96), we can get a variational formulation for the problem with the total
pressure boundary condition instead of the static pressure boundary condition.
For the discussion of that the concepts of solutions are well defined, we refer to
Chap. 5.
Let Ω ⊂ R3 , ∂Ω ∈ C 1,1 or Ω be a convex polyhedron, Γ D = ∅ and Γ¯1 ∩ Γ¯2 = ∅.
Then, by Lemma 4.1 of [56]
(rot v, rot v) ≥ c v 2
H1 (Ω)
, (2.100)
which is fundamental for study of the problem (2.99). When Ω ⊂ R3 , ∂Ω ∈ C 2 and

Γ D = ∅, (2.100) is also valid without the condition Γ¯1 ∩ Γ¯2 = ∅ (see Lemma 2 of
[57]).
As shown above, relying on the Dirichlet bilinear form
a(v, u) = (∇v, ∇u), (2.101)
we can get a variational formulation for the Navier-Stokes problem with Dirichlet
and the outlet boundary conditions. Relying on strain bilinear form
a(v, u) = 2i, j (εi j (v), εi j (u)), (2.102)
we can get a variational formulation for the problem with Dirichlet and the stress
(total stress) boundary conditions, whereas relying on vorticity bilinear form
a(v, u) = (rot v, rot u) (2.103)

we get one for the problem with Dirichlet, static pressure (total pressure) and vorticity
boundary conditions.
2.4 Bibliographical Remarks
2.4.1 Fluid Equations
For Eulerian and Lagrangian descriptions of movement of particles, one can see
Chap. 3 of [3] or Chap. 1 of [2]. For proof of the transport theorem, we refer to
Chap. 2 of [1] or Chap. 5 of [58], and for derivation of equations using conservation
laws we refer to Chap. 4 of [3], Chap. 1 of [59] or [60].
For papers studying (2.16) and (2.17), respectively, we refer to [61–64]. For papers
dealing with (2.39) and (2.40), respectively, see [65–68]. For papers discussing (2.41)
we refer to [69–71] and refer to [72] for the incompressible fluid. In [73] Eq. (2.19) is
considered and most papers for incompressible Newtonian fluid are concerned with
Eq. (2.20).
For papers investigating (2.42), we refer to [72, 74–81]. There are many papers
dealing with (2.43), for example, [82–84]. For papers dealing with (2.44), we refer
to [7, 85].
2.4.2 Boundary Conditions of the Navier-Stokes Equations
We refer to Chap. 4 of [3] and Chap. 2 of [59] for explanation of some boundary
conditions.
The stick boundary condition was formulated by G. Stokes in 1845 and the Navier
slip-with-friction boundary condition was suggested by C.-L. Navier in 1823. (cf.
Introduction of [86]). In [87] the Navier slip-with-friction boundary condition has
been derived rigorously from the boundary condition at the kinetic level (Boltz-
mann equation) for compressible fluids. For papers dealing with vorticity boundary
condition see [9–11, 13, 14, 18, 88, 89].
By introducing the concept of energy-preserving boundary conditions, in [90] a
rational derivation of a large class of nonstandard boundary conditions containing
several different artificial boundary conditions is given.
In 1990s, the leak and slip boundary conditions with threshold have been intro-
duced by H. Fujita (see Introduction of [25]). Physical and experimental backgrounds
of the threshold slip boundary conditions are mentioned in several papers (see [8,
22–24]). For physical backgrounds of the threshold leak boundary conditions, we
refer to [24, 27, 91]. For non-Newtonian fluid equations with friction slip boundary
conditions, we refer to [92–95].
2.4 Bibliographical Remarks 77
In [96], the Navier-Stokes problem with one-sided threshold leak boundary con-
dition based on the total stress was considered as an example of application of a
variational inequality. The one-sided threshold leak boundary conditions based on
the stress for the Navier-Stokes equations were considered first in [28]. For similar
one-sided boundary conditions of elasticity, we refer to [97], Sect. 5.4.1 of Chap. 3
in [98]. “Do nothing” boundary condition was introduced in [99].
2.4.3 Bilinear Forms for Hydrodynamics
For papers applying the Dirichlet bilinear form (2.101), we refer to [43, 48, 91, 100,
101].
For papers utilizing the strain bilinear form (2.102), we refer to [24–27, 35, 73, 91,
102–109]. In [73], another equivalent variational formulation, where strain, pressure,
velocity and vorticity are unknown functions, also is given.
For the papers relying on the vorticity bilinear form (2.103), see [18, 30, 32, 33,
56, 57, 110–116]. In Sect. 1 of [114], the Dirichlet bilinear form (2.101) instead
of the vorticity bilinear form (2.103) is used since two bilinear forms (2.101) and
(2.103) for polygon or polyhedral domain under some boundary conditions are equal
(see [117]). When one deals with the boundary condition for pressure or vorticity on
a portion of boundary, there are other variational formulations using three unknown
functions v, p and ω, where ω = rot v, (see [31, 118–120]) for the two-dimensional
case and v, p and a vector potential for the three-dimensional case (see [121]).
References
1. G. Łukaszewicz, P. Kalita, Navier-Stokes Equations An Introduction with Applications

(Springer, 2016)
2. R. Temam, A. Miranvillie, Mathematical Modeling in Continuum Mechanics (Cambridge
University Press, 2005)
3. P.K. Kundu, I.M. Cohen, Fluid Mechanics, 4th edn. (Elsevier, 2008)
4. Y. Amirat, K. Hamdache, Unique solvability of equations of motion for ferrofluids. Nonlinear
Anal. 73, 471–494 (2010)
5. D. Bian, Y. Xiao, Global solution to the nematic liquid crystal flows with heat effect. J. Diff.
Equ. 263, 5298–5329 (2017)
6. L. Ma, On two-dimensional incompressible magneto-micropolar system with mixed partial
viscosity. Nonlinear Anal.?: RWA 40, 95–139 (2018)
7. Y. Kagei, M. Růžicka, G. Thater, Natural convection with dissipative heating. Commun. Math.
Phys. 214, 287–313 (2000)
8. H. Hervet, L. Léger, Flow with slip at the wall: from simple to complex fluids. C. R. Phys. 4,
241–249 (2003)
9. H. Beirão da Veiga, F. Crispo, Sharp inviscid limit results under Navier type boundary con-
ditions, an Lp theory. J. Math. Fluid Mech. 12(3), 397–411 (2010)
10. H. Beirão da Veiga, F. Crispo, A missed persistence property for the Euler equations, and its
effect on inviscid limits. Nonlinearity 25(6), 1661–1669 (2012)
11. H. Bellout, J. Neustupa, P. Penel, On viscosity-continuous solutions of the Euler and Navier-
Stokes equations with a Navier-type boundary condition. C. R. Math. Acad. Sci. Paris Ser. I
347(19–20), 1141–1146 (2009)
12. L.C. Berselli, S. Spirito, On the vanishing viscosity limit of 3D Navier-Stokes equations under
slip boundary conditions in general domains. Commun. Math. Phys. 316, 171–198 (2012)
13. G.-M. Gie, J.P. Kelliher, Boundary layer analysis of the Navier-Stokes equations with gener-
alized Navier boundary conditions. J. Diff. Equ. 253(6), 1862–1892 (2012)
14. H. Beirão da Veiga, F. Crispo, The 3-D inviscid limit result under slip boundary conditions.
A negative answer. J. Math. Fluid Mech. 14(1), 55–59 (2012)
15. S. Monniaux, Various boundary conditions for Navier-Stokes equations in bounded Lipschitz
domains. Discret. Contin. Dyn. Syst. Ser. S 6, 1355–1369 (2013)
16. S. Monniaux, Z. Shen, Stokes problems in irregular domains with various boundary conditions,
in Handbook of Mathematical Analysis in Mechanics of Viscous Fluids, ed. by Y. Giga, A.
Novotny (Springer, 2016), https://doi.org/10.1007/978-3-319-10151-4-4
17. T. Kim, D. Cao, Some properties on the surfaces of vector fields and its application to the
Stokes and Navier-Stokes problems with mixed boundary conditions. Nonlinear Anal. 113,
94–114 (2015). Erratum, ibid 135, 249–250 (2016)
18. C. Bernardi, F. Hecht, R. Verfürth, A finite element discretization of the three-dimensional
Navier-Stokes equations with mixed boundary conditions. ESAIM: Math. Model. Numer.
Anal. 43, 1185–1201 (2009)
19. D. Iftimie, G. Planas, Inviscid limits for the Navier-Stokes equations with Navier friction
boundary conditions. Nonlinearity 19, 899–918 (2006)
20. D. Iftimie, F. Sueur, Viscous boundary layers for the Navier-Stokes equations with the Navier
slip conditions. Arch. Rat. Mech. Anal. 199, 145–175 (2011)
21. N. Masmoudi, F. Rousset, Uniform regularity for the Navier-Stokes equation with Navier
boundary condition. Arch. Rat. Mech. Anal. 203, 529–575 (2012)
22. M. Ayadi, M.K. Gdoura, T. Sassi, Mixed formulation for Stokes problem with Tresca friction.
C. R. Acad. Sci. Paris Ser. I 348, 1069–1072 (2010)
23. G. Bayada, M. Boukrouche, On a free boundary problem for the Reynolds equation derived
from the Stokes system with Tresca boundary conditions. J. Math. Anal. Appl. 282, 212–231
(2003)
24. H. Fujita, A mathematical analysis of motions of viscous incompressible fluid under leak or
slip boundary conditions. RIMS Kokyuroku 888, 199–216 (1994)
25. H. Fujita, Non-stationary Stokes flows under leak boundary conditions of friction type. J.
Comput. Math. 19, 1–8 (2001)
26. H. Fujita, Variational inequalities and nonlinear semi-groups applied to certain nonlinear
problems for the Stokes equation. RIMS Kokyuroku 1234, 70–85 (2001)
27. H. Fujita, A coherent analysis of Stokes flows under boundary conditions of friction type. J.
Comput. Appl. Math. 149, 57–69 (2002)
28. T. Kim, D. Cao, The steady Navier-Stokes and Stokes systems with mixed boundary conditions
including one-sided leaks and pressure. Methods Appl. Anal. 23, 329–364 (2016)
29. T. Kim, D. Cao, Non-stationary Navier-Stokes equations with mixed boundary conditions. J.
Math. Sci. Univ. Tokyo 24, 159–194 (2017)
30. S. Marušić, On the Navier-Stokes system with pressure boundary condition. Ann. Univ. Fer-
rara 53, 319–331 (2007)
31. M. Amara, D. Capatina-Papaghiuc, D. Trujillo, Stabilized finite element method for Navier-
Stokes equations with physical boundary conditions. Math. Comput. 76(259), 1195–1217
(2007)
32. J.M. Bernard, Non-standard Stokes and Navier-Stokes problems: existence and regularity in
stationary case. Math. Methods Appl. Sci. 25, 627–661 (2002)
33. J.M. Bernard, Time-dependent Stokes and Navier-Stokes problems with boundary conditions
involving pressure: existence and regularity. Nonlinear Anal.?: RWA 4, 805–839 (2003)
34. M. Beneš, Mixed initial-boundary value problem for the three-dimensional Navier-Stokes
equations in polyhedral domains. Discret. Contin. Dyn. Syst. Suppl. 02, 135–144 (2011)
References 79
35. V. Maz’ya, J. Rossmann, Lp estimates of solutions to mixed boundary value problems for the
Stokes system in polyhedral domains. Math. Nachr. 280(7), 751–793 (2007)
36. T. Kashiwabara, On a strong solution of the non-stationary Navier-Stokes equations under
slip or leak boundary conditions of friction. J. Differ. Equ. 254, 756–778 (2013)
37. M. Beneš, Solutions to the mixed problem of viscous incompressible flows in a channel. Arch.
Math. 93, 287–297 (2009)
38. M. Beneš, Strong solutions to non-stationary channel flows of heat-conducting viscous incom-
pressible fluids with dissipative heating. Acta Appl. Math. 116, 237–254 (2011)
39. M. Beneš, Solutions of the Navier-Stokes equations with various types of boundary conditions.
Arch. Math. 98, 487–497 (2012)
40. M. Beneš, A note on the regularity of thermally coupled viscous flows with critical growth in
nonsmooth domains. Math. Methods Appl. Sci. 36, 1290–1300 (2013)
41. M. Beneš, A note on regularity and uniqueness of natural convection with effects of viscous
dissipation in 3D open channels. Z. Angew. Math. Phys. 65, 961–975 (2014)
42. M. Beneš, The “do nothing” problem for Boussinesq fluids. Appl. Math. Lett. 31, 25–28
(2014)
43. M. Beneš, P. Kučera, Non-steady Navier-Stokes equations with homogeneous mixed boundary
conditions and arbitrarily large initial condition. Carpathian J. Math. 23(1–2), 32–40 (2007)
44. M. Beneš, P. Kučera, On the Navier-Stokes flows for heat-conducting fluids with mixed
boundary conditions. J. Math. Anal. Appl. 389, 769–780 (2012)
45. M. Beneš, P. Kučera, Solutions to the Navier-Stokes equations with mixed boundary conditions
in two-dimensional bounded domains. Math. Nachr. 289(2–3), 194–212 (2016)
46. S. Kračmar, J. Neustupa, A weak solvability of a steady variational inequality of the Navier-
Stokes type with mixed boundary conditions. Nonlinear Anal. 47, 4169–4180 (2001)
47. P. Kučera, The time periodic solutions of the Navier-Stokes equations with mixed boundary
conditions. Discret. Contin. Dyn. Syst. Ser. S 3(2), 325–337 (2010)
49. N.D. Kopachevsky, S.G. Krein, Operator Approach to Linear Problems of Hydrodynamics 1,
2 (Springer Basel AG 2001, 2003)
50. E. Bänsch, K. Deckelnick, Optimal error estimates for the Stokes and Navier-Stokes equations
with slip-boundary condition. Math. Model. Numer. Anal. 33(5), 923–938 (1999)
51. R. Brown, I. Mitrea, M. Mitrea, M. Wright, Mixed boundary value problems for the Stokes
system. Trans. Am. Math. Soc. 362(3), 1211–1230 (2010)
52. P. Grisvard, Elliptic Problems in Nonsmooth Domains (SIAM, Philadelphia, 1985)
53. V. Girault, P.A. Raviart, Finite Element Methods for Navier-Stokes Equations (Springer, 1986)
54. R. Temam, Navier-Stokes Equations (North-Holland, 1985)
55. G.P. Galdi, An Introduction to the Mathematical Theory of the Navier-Stokes Equations
(Springer, 2011)
56. C. Conca, F. Murat, O. Pironneau, The Stokes and Navier-Stokes equations with boundary
conditions involving the pressure. Jpn. J. Math. 20(2), 279–318 (1994)
57. A.A. Illarionov, A.Yu. Chebotarev, Solvability of a mixed boundary value problem for the
stationary Navier-Stokes equations. Differ. Equ. 37(5), 724–731 (2001)
58. J.N. Reddy, An Introduction to Continuum Mechanics with Applications (Cambridge Univer-
sity Press, 2008)
59. M. Köhne, L p -Theory for Incompressible Newtonian Flows, (Springer Spektrum, 2013)
60. J. Málek, V. Průša, Derivation of equations for continuum mechanics and thermodynamics
of fluids, in Handbook of Mathematical Analysis in Mechanics of Viscous Fluids, ed. by Y.
Giga, A. Novotny (Springer, 2016), https://doi.org/10.1007/978-3-319-10151-4-1
61. R. Duan, H. Liu, S. Ukai, T. Yang, Optimal L p − L q convergence rates for the compressible
Navier-Stokes equations with potential force. J. Diff. Equ. 238, 220–233 (2007)
62. Y. Kagei, K. Tsuda, Existence and stability of time periodic solution to the compressible
Navier-Stokes equation for time periodic external force with symmetry. J. Diff. Equ. 258,
399–444 (2015)
63. Z. Guo, Z. Li, Global existence of weak solution to the free boundary problem for compressible
Navier-Stokes. Kinet. Relat. Model 9, 75–103 (2016)
64. J. Qian, J. Zhao, Global existence of classical solutions for isentropic compressible Navier-
Stokes equations with small initial density. J. Diff. Equ. 259, 6830–6850 (2015)
65. H. Ma, S. Ukai, T. Yang, Time periodic solutions of compressible Navier-Stokes equations.
J. Diff. Equ. 248, 2275–2293 (2010)
66. J. Qian, H. Yin, On the stationary solutions of the full compressible Navier-Stokes equations
and its stability with respect to initial disturbance. J. Diff. Equ. 237, 225–256 (2007)
67. C. Doua, S. Jiang, Y. Ou, Low Mach number limit of full Navier-Stokes equations in a 3D
bounded domain. J. Diff. Equ. 258, 379–398 (2015)
68. S. Jiang, Y. Ou, Incompressible limit of the non-isentropic Navier-Stokes equations with well-
prepared initial data in three-dimensional bounded domains. J. Math. Pures Appl. 96, 1–28
(2011)
69. R. Danchin, Global existence in critical spaces for flows of compressible viscous and heat-
conductive gases. Arch. Rat. Mech. Anal. 160, 1–39 (2001)
70. R. Danchin, Local theory in critical spaces for compressible viscous and heat-conductive
gases. Commun. Part. Differ. Equ. 26(7 & 8), 1183–1233 (2001)
71. X. Zhao, L. Yao, Global classical solutions of the full compressible Navier-Stokes equations
with cylindrical or spherical symmetry. Nonlinear Anal.?: RWA 33, 139–167 (2017)
72. M. Bulíček, E. Feireisl, J. Málek, A Navier-Stokes-Fourier system for incompressible flu-
ids with temperature dependent material coefficients. Nonlinear Anal.?: RWA 10, 992–1015
(2009)
73. M. Farhloul, S. Nicaise, L. Paquet, A refined mixed finite-element method for the stationary
Navier-Stokes equations with mixed boundary conditions. IMA J. Numer. Anal. 28, 25–45
(2008)
74. L. Consiglieri, Regularity for the Navier-Stokes-Fourier system. Differ. Equ. Appl. 1, 583–604
(2009)
75. L. Consiglieri, Existence of proper weak solutions to the Navier-Stokes-Fourier system.
Funkc. Ekvacioj 55, 347–365 (2012)
76. H. Inoue, On heat convection equations with dissipative terms in time dependent domains.
Nonlinear Anal. 30, 4441–4448 (1997)
77. H. Inoue, Strong solutions of two dimensional heat convection equations with dissipating
terms. Tokyo J. Math. 22, 445–472 (1999)
78. H. Inoue, Global and periodic solutions of heat convection equations with internal friction
terms. Nonlinear Anal. 47, 2881–2886 (2001)
79. J. Naumann, Existence of weak solutions to the equations of stationary motion of heat-
conducting incompressible viscous fluids, Progress in Nonlinear Differential Equations and
Their Applications, vol. 66 (Birkhäuser, 2005), pp. 373–390
80. J. Naumann, On the existence of weak solutions to the equations of non-stationary motion
of heat-conducting incompressible viscous fluids. Math. Methods Appl. Sci. 29, 1883–1906
(2006)
81. J. Naumann, An existence theorem for weak solutions to the equations of non-stationary
motion of heat-conducting incompressible viscous fluids. J. Nonlinear Convex Anal. 7(3),
483–497 (2006)
82. H. Morimoto, On non-stationary Boussinesq equations. Proc. Jpn. Acad. Ser. A 67 159–161
(1991)
83. H. Morimoto, Non-stationary Boussinesq equations. J. Fac. Sci. Univ. Tokyo Sect. IA Math.
39, 61–75 (1992)
84. H. Morimoto, Heat convection equation with nonhomogeneous boundary conditions. Funkc.
Ekvacioj 53, 213–229 (2010)
85. J. Naumann, J. Wolf, Existence of weak solutions to the equations of natural convection
with dissipative heating, Advances in Mathematical Fluid Mechanics (Springer, 2010), pp.
367–384
References 81
86. C. Amrouche, P. Penel, N. Seloula, Some remarks on the boundary conditions in the theory
of Navier-Stokes equations. Ann. Math. Blaise Pascal 20, 37–73 (2013)
87. F. Coron, Derivation of slip boundary conditions for the Navier-Stokes system from the
Boltzmann equation. J. Stat. Phys. 54, 829–857 (1989)
88. H. Bellout, J. Neustupa, A Navier-Stokes approximation of the 3D Euler equation with the
zero flux on the boundary. J. Math. Fluid Mech. 10(4), 531–553 (2008)
89. H. Bellout, J. Neustupa, P. Penel, On the Navier-Stokes equation with boundary conditions
based on vorticity. Math. Nachr. 269(270), 59–72 (2004)
90. D. Bothe, M. Köhne, J. Prüss, On a class of energy preserving boundary conditions for
incompressible Newtonian flows. SIAM J. Math. Anal. 45, 3768–3822 (2013)
91. R. An, Y. Li, K. Li, Solvability of Navier-Stokes equations with leak boundary conditions.
Acta Math. Appl. Sin. Engl. Ser. 25, 225–234 (2009)
92. M. Boukrouche, I. Boussetouan, L. Paoli, Non-isothermal Navier-Stokes system with mixed
boundary conditions and friction law: uniqueness and regularity properties. Nonlinear Anal.
102, 168–185 (2014)
93. M. Boukrouche, I. Boussentoua, L. Paoli, Existence for non-isothermal fluid flows with
Tresca’s friction and Cattaneo’s heat law. J. Math. Anal. Appl. 427, 499–514 (2015)
94. M. Boukrouche, G. Łukaszewicz, On global in time dynamics of a planar Bingham flow
subject to a subdifferential boundary conditions. Discret. Contin. Dyn. Syst. 34, 3969–3983
(2014)
95. J.K. Djoko, P.A. Razafimandimby, Analysis of the Brinkman-Forchheimer equations with
slip boundary conditions. Appl. Anal. 93, 1477–1494 (2014)
Pure Appl. 124(4), 107–125 (1980)
97. A. Khludnev, G. Leugering, Unilateral contact problems for two perpendicular elastic struc-
tures. J. Anal. Appl. 27, 157–177 (2008)
98. G. Duvaut, J.L. Lions, Inequalities in Mechanics and Physics (Springer, Berlin, 1976)
99. P.M. Gresho, Some current CFD issues relevant to the incompressible Navier-Stokes equa-
tions. Comput. Methods Appl. Mech. Eng. 87, 201–252 (1991)
100. K. Li, R. An, On the rotating Navier-Stokes equations with mixed boundary conditions. Acta
Math. Sin. 24(4), 577–598 (2008)
101. Z. Skalák, P. Kučera, An existence theorem for the Boussinesq equations with non-Dirichlet
boundary conditions. Appl. Math. 45(2), 81–98 (2000)
102. R. An, K. Li, Variational inequality for the rotating Navier-Stokes equations with subdiffer-
ential boundary conditions. Comput. Math. Appl. 55, 581–587 (2008)
103. R. An, K. Li, Existence of weak solution to nonhomogeneous steady Navier-Stokes equations
with mixed boundary conditions (Chinese). Acta Math. Appl. Sin. 32(4), 664–672 (2009)
104. L. Formaggia, A. Moura, F. Nobile, Coupling 3D and 1D fluid-structure interaction models
for blood flow simulations. Proc. Appl. Math. Mech. 6, 27–30 (2006)
105. L. Formaggia, A. Moura, F. Nobile, On the stability of the coupling of 3D and 1D fluid-
structure interaction models for blood flow simulations. ESAIM: Math. Model. Numer. Anal.
41(4), 743–769 (2007)
106. Z. Gao, Y. Ma, Shape gradient of the dissipated energy functional in shape optimization for
the viscous incompressible flow. Appl. Numer. Math. 58, 1720–1741 (2008)
107. P. Kučera, Z. Skalák, Local solutions to the Navier-Stokes equations with mixed boundary
conditions. Acta Appl. Math. 54, 275–288 (1998)
108. V. Maz’ya, J. Rossmann, Mixed boundary value problems for the Navier-Stokes system in
polyhedral domains. Arch. Rat. Mech. Anal. 194, 669–712 (2009)
109. J. Rossmann, Mixed boundary value problems for Stokes and Navier-Stokes systems in poly-
hedral domains. Oper. Theory: Adv. Appl. 193, 269–280 (2009)
110. G.V. Alekseev, A.B. Smishliaev, Solvability of the boundary value problems for Boussinesq
equations with inhomogeneous boundary conditions. J. Math. Fluid Mech. 3(1), 18–39 (2001)
111. K. Amedodji, G. Bayada, M. Chambat, On the unsteady Navier-Stokes equations in a time-
moving domain with velocity-pressure boundary conditions. Nonlinear Anal. 49(4), 565–587
(2002)
112. C. Begue, C. Conca, F. Murat, O. Pironneau, A nouveau sur les equations de Stokes et de
Navier-Stokes avec des conditions aux limites sur la pression. C. R. Acad. Sci. Paris Ser. I
304, 23–28 (1987)
113. C. Begue, C. Conca, F. Murat, O. Pironneau, Les equations de Stokes et de Navier-Stokes avec
des condition sur la pression. Nonlinear Part. Diff. Equ. Appl. College de France, Seminar
IX, 179–384 (1988)
114. P. Ciarlet Jr., System de Stokes avec flux de vitesse et pression imposes. C. R. Acad. Sci. Paris
Ser. I 337, 119–124 (2003)
115. C. Conca, C. Pares, O. Pironneau, M. Thiriet, Navier-Stokes equations with imposed pressure
and velocity fluxes. Int. J. Numer. Methods Fluids 20, 267–287 (1995)
116. G. Łukaszewicz, On the Navier-Stokes equations in time dependent domains and with bound-
ary conditions involving the pressure. J. Math. Sci. Univ. Tokyo 4(3), 529–550 (1997)
117. M. Costabel, A coercive bilinear form for Maxwell’s equations. J. Math. Anal. Appl. 157,
527–541 (1991)
118. M. Amara, D. Capatina-Papaghiuc, E.C. Vera, D. Trujillo, A three field stabilized finite ele-
ment method for the Stokes equations. C. R. Acad. Sci. Paris Ser. I 334, 603–608 (2002)
119. M. Amara, D. Capatina-Papaghiuc, E. Chacón-Vera, D. Trujillo, Vorticity-velocity-pressure
formulation for the Stokes problem. Math. Comput. 73, 1673–1697 (2003)
120. M. Amara, D. Capatina-Papaghiuc, E. Chacón-Vera, D. Trujillo, Vorticity-velocity-pressure
formulation for the Stokes equations. Comput. Vis. Sci. 6, 47–52 (2004)
121. V. Girault, Incompressible finite element methods for Navier-Stokes equations with nonstan-
dard boundary conditions in R 3 . Math. Comput. 51(183), 55–74 (1988)
Chapter 3
The Steady Navier-Stokes System
In this chapter, we are concerned with the steady Navier-Stokes systems with mixed
boundary conditions involving Dirichlet, pressure, vorticity, stress and normal deriva-
tive of velocity together. As we have seen in Sect. 2.3.2, according to what kinds
of bilinear forms for variational formulation are used, types of boundary conditions
under consideration together are different. The variational formulations in Sect. 2.3.2
do not reflect, for example, the boundary conditions for stress and pressure together,
but this case is important in practice. To include wider boundary conditions together,
we first study the relations among strain, vorticity, normal derivative of velocity, and
shape of boundary surfaces. Using the relations on the boundary surfaces, we reflect
all boundary conditions into variational formulations of the problems. Then we prove
the existence and uniqueness of solutions to the problems.
3.1 Properties on the Boundary Surfaces of Vector Fields
In this section, we first recall ways describing the shape of surface and study the
relations among strain, vorticity, normal derivative of vector fields and shape of a
surface when the vector fields near a surface are tangent or normal on the surface.

https://doi.org/10.1007/978-3-030-78659-5_3
84 3 The Steady Navier-Stokes System
3.1.1 The Second Fundamental Form and Shape Operator of

Surface
3.1.1.1 The Second Fundamental Form of a Surface
When γ : (α, β) → Rl is a parametrized curve, if its derivative γ (t) at every

t ∈ (α, β) is a unit vector, then γ is said to be a unit-speed curve. If the natural
parameter (length) of curve is taken as the parameter t, then the curve is a unit-speed
curve.
A subset M of R3 is called a surface if, for every point p ∈ M, there is an open set
U ⊂ R2 and an open set W ⊂ R3 containing p such that M ∩ W is homeomorphic
to U . A subset of a surface M of form M ∩ W , where W is an open subset of R3 , is
called an open subset of M. A homeomorphism σ : U → M ∩ W as in this definition
is called a surface patch for parametrization of the open subset M ∩ W of M.
If function σ : U → R3 for a surface M is in class C k (or C k,1 ), then the surface
is called class C k −surface (or C k,1 -surface), and a class C 1 -surface is called smooth.
If a vector is tangent to a curve in M passing through p, then it is called a tangent
vector to the surface M at the point p ∈ M.
For a surface patch σ (ξ, ζ ) for a surface M,
∂σ ∂σ
σξ (ξ0 , ζ0 ) = and σζ (ξ0 , ζ0 ) =
∂ξ (ξ0 ,ζ0 ) ∂ζ (ξ0 ,ζ0 )
are tangent at the point p(ξ0 , ζ0 ) ∈ M to the surface M. If, moreover, ξ and ζ are,
respectively, the lengths of curves σ (ξ, ζ0 ) and σ (ξ0 , ζ ) (natural parameters), then
those are tangent units at the point p(ξ0 , ζ0 ) ∈ M.
The set of all vectors tangent to M at p is called the tangent space T p M of M at p.
The tangent space T p M of a surface M at a point p ∈ M is completely determined
by giving a unit vector orthogonal to it at p, called a unit normal to M at p. There
are, of course, two unit normal vectors, but choosing a surface patch σ : U → R3
containing point p leads to a definite choice, namely,
σξ × σζ
nσ = .
σξ × σζ
This is called the standard unit normal of the surface patch σ at point p. To be
specific, unless otherwise stated we always take the standard unit normal.
If in a neighborhood of (ξ0 , ζ0 ) ∈ U a surface patch σ : U → R3 is in class C 1 ,
then a unit normal field n of class C always exists in a neighborhood of σ (ξ0 , ζ0 ) ∈ M.
But it may not be possible to extend n to all of M, e.g. Möbius band. If a smooth
unit normal field n defined on all of M is chosen, then the surface M is said to be
oriented.
3.1 Properties on the Boundary Surfaces of Vector Fields 85
Now we are interested in studying the shape of surfaces in R3 . The objects describ-
ing the shape of a surface M ⊂ R3 are the second fundamental form and the shape
operator (Weingarten map) of the surface.
Suppose that σ is a surface patch for M ⊂ R3 with the standard unit normal n. As
the parameters (ξ, ζ ) of σ change to (ξ + Δξ, ζ + Δζ ), the surface moves away from
the tangent plane through σ (ξ, ζ ). Making inner product in R3 between the vector
σ (ξ + Δξ, ζ + Δζ ) − σ (ξ, ζ ) and the normal vector n(ξ, ζ ) at point σ (ξ, ζ ) ∈ M,
we get the deviation of σ along n from its tangent plane through σ (ξ, ζ ) ∈ M:

σ (ξ + Δξ, ζ + Δζ ) − σ (ξ, ζ ) · n(ξ, ζ ).
By the two variable form of Taylor’s theorem,
σ (ξ + Δξ, ζ + Δζ ) − σ (ξ, ζ )
1
= σξ Δξ + σζ Δζ + σξ ξ (Δξ )2 + 2σξ ζ Δξ Δζ + σζ ζ (Δζ )2 + o (Δξ )2 + (Δζ )2 ,
2

where o (Δξ )2 + (Δζ )2 )/((Δξ )2 + (Δζ )2 tends to zero as (Δξ )2 + (Δζ )2 → 0.
Note that σξ and σζ are tangent to the surface, hence perpendicular to n, and so
the deviation of σ in the direction of the unit vector n from its tangent plane is
1
L(Δξ )2 + 2K Δξ Δζ + N (Δζ )2 + o (Δξ )2 + (Δζ )2 , (3.1)
2
where
L = σξ ξ (ξ, ζ ) · n(ξ, ζ ), K = σξ ζ (ξ, ζ ) · n(ξ, ζ ), N = σζ ζ (ξ, ζ ) · n(ξ, ζ ).

(3.2)
At point σ (ξ, ζ ) ∈ M
σξ (ξ, ζ ) · n(ξ, ζ ) = 0, σζ (ξ, ζ ) · n(ξ, ζ ) = 0,
which yields by differentiation that
L = σξ ξ (ξ, ζ ) · n(ξ, ζ ) = −σξ (ξ, ζ ) · n ξ (ξ, ζ ),

K = σξ ζ (ξ, ζ ) · n(ξ, ζ ) = −σξ (ξ, ζ ) · n ζ (ξ, ζ ) = −σζ (ξ, ζ ) · n ξ (ξ, ζ ), (3.3)
N = σζ ζ (ξ, ζ ) · n(ξ, ζ ) = −σζ (ξ, ζ ) · n ζ (ξ, ζ ).
The expression
L(Δξ )2 + 2K Δξ Δζ + N (Δζ )2 (3.4)
is called the second fundamental form of the surface patch σ .

Remark 3.1 If in a neighborhood of a point p ∈ M the surface is deviated from the
tangent plane to the side opposite to the unit normal, then the quadratic form (3.4)
at p ∈ M is negative.
Since all vectors w ∈ T p M are expressed as
w = λσξ + μσζ ,
denote by
dξ(w) = λ, dζ (w) = μ for w ∈ T p M.
Let us define a symmetric bilinear form on T p M by

v, w p,M := Ldξ(v)dξ(w) + K dξ(v)dζ (w) + dξ(w)dζ (v) + N dζ (v)dζ (w)
∀v, w ∈ T p M.
(3.5)
Let a surface patch σ̃ (ξ̃ , ζ̃ ) be a reparametrization of a surface patch σ (ξ, ζ )
with reparametrization map (ξ, ζ ) = Φ(ξ̃ , ζ̃ ). Denote the Jacobian matrix of the
reparametrization map (ξ̃ , ζ̃ ) → (ξ, ζ )
∂ξ ∂ξ
∂ ξ̃ ∂ ζ̃
∂ζ ∂ζ
∂ ξ̃ ∂ ζ̃
by J (Φ). Then we can prove that
L̃ K̃ L K
= ±J (Φ)t J (Φ) (3.6)
K̃ Ñ K N
with the plus sign if det(J (Φ)) > 0 and the minus sign if det(J (Φ)) < 0.
From (3.6), we can see that the second fundamental form of a surface patch
is unchanged by a reparametrization of the patch preserving its orientation (which
means det(J (Φ)) > 0) (see Exercise 6.1.4 of [1]).
3.1.1.2 The Shape Operator of Surface
Let M be a smooth surface and f : M → R be a smooth function on M. This means

that f = f ◦ σ is smooth for all patches σ : U → M.
For p ∈ M and X ∈ T p M, the directional derivative of f at p in the direction X ,
denoted by ∇ X f , is defined as follows. Let δ : (−ε, ε) → M ⊂ R3 be any smooth
curve in M such that δ(0) = p and δ (0) = X . Then,
d
∇X f = f ◦ δ(t)|t=0 .
dt
The directional derivative is well defined, i.e. independent of the particular choice
of δ, and linear.
A vector field along a surface M is a rule which assigns a vector (need not be
tangent to M) to each point of M,
p ∈ M → Y ( p) ∈ T p R3 ,
where T p R3 means
the space of 3-D vectors with the origin at p. Then when X ∈ T p M
and Y ( p) = Y1 ( p), Y2 ( p), Y3 ( p) ∈ T p M, we define an operator
X ∈ T p M → ∇ X Y ( p) ∈ T p R3
by
X ∈ T p M → ∇ X Y1 ( p), ∇ X Y1 ( p), ∇ X Y1 ( p) ∈ T p R3 .
Let M be a surface, p ∈ M and n be a smooth unit normal vector field defined

along a neighborhood W ⊂ M of p. Let us consider the operator
X ∈ T p M → ∇ X n( p).

Since n( p), n( p) = 1, we know that
∇ X (n( p), n( p)) = 0,
and
(∇ X n( p), n( p)) + (n( p), ∇ X n( p)) = 0,
which implies
2(∇ X n( p), n( p)) = 0.
Therefore,
∇ X n( p) ⊥ n( p) and ∇ X n ∈ T p M.
Definition 3.1 Let M be a surface, p ∈ M and n be a smooth unit normal defined

along a neighborhood W ⊂ M of p. The map S : X ∈ T p M → T p M defined by
S(X ) = ∇ X n( p)
is called the shape operator of M at p (Weingarten map).

Remark 3.2 In some books, the operator −S is called the shape operator of M at
p.
Let us introduce a bilinear form L on T p M

L(v, w) = − S(v), w = − ∇v n( p), w ∀v, w ∈ T p M. (3.7)
Proposition 3.1 (Proposition 7.2.2 of [1]) Let p be a point of a surface M, let σ (ξ, ζ )
be a surface patch of M with p in its image and let v, w p,M be the bilinear form
defined by (3.5). Then, for any v, w ∈ T p M,
L(v, w) = v, w p,M , (3.8)
where L , K , N are as in (3.3).

Proof Since both sides of (3.8) define bilinear forms on T p M and
v = dξ(v)σξ + dζ (v)σζ ,
w = dξ(w)σξ + dζ (w)σζ ,
it suffices to verify that (3.8) is valid when v and w are σξ or σζ . Recalling that
dξ(σξ ) = dζ (σζ ) = 1 and dξ(σζ ) = dζ (σξ ) = 0, we only need to show that
L(σξ , σξ ) = L , L(σξ , σζ ) = L(σζ , σξ ) = K , L(σζ , σζ ) = N . (3.9)
Taking into account (3.3), we have
L(σξ , σξ ) = −(∇σξ n( p), σξ ) = −n ξ · σξ = L .
In the same way as above, the other relations in (3.9) can be proved.
Corollary 3.1 (Corollary 7.2.4 of [1]) The shape operator is self-adjoint.
In what follows let ξ, ζ be natural parameters. The curves σ (ξ, ζ0 ) and σ (ξ0 , ζ )
intersect at point p(ξ0 , ζ0 ), and σξ (ξ, ζ0 ) = e1 and σζ (ξ0 , ζ ) = e2 are the tangent
unit vectors at point p(ξ0 , ζ0 ). Thus the shape operator is expressed by the following
matrix:
−L −K
S= , (3.10)
−K −N
where
∂n ∂n ∂n ∂n
−L = e1 · , −K = e2 · = e1 · , −N = e2 · .
∂e1 ∂e1 ∂e2 ∂e2
Since the shape operator S is self-adjoint, there exists an orthonormal basis {e1 , e2 }
of T p M and real numbers λ1 , λ2 such that
Se1 = λ1 e1 , Se2 = λ2 e2 ,
and the matrix for S at p ∈ M is
λ1 0
S= .
0 λ2
The trace tr (S) of matrix S is invariable with respect to orthogonal transformation. On

the other hand, if σ (ξ, ζ ) is a surface patch of M with p = σ (ξ0 , ζ0 ) ∈ M, σξ | p = e1
and σζ | p = e2 , then by (3.3)
−σξ ξ · n = σξ · n ξ = λ1 , −σζ ζ · n = σζ · n ζ = λ2 .
Since σξ ξ and σζ ζ are, respectively, curvature vectors at the p of the curves σ (ξ, ζ0 )
and σ (ξ0 , ζ ), σξ ξ · n and σζ ζ · n are the normal curvatures. Therefore, for any orthog-
onal basis 21 tr (S)( p) is called the mean curvature at p ∈ M.
Let us consider plane curves and their curvatures. If γ (t) is a unit-speed plane
curve with parameter t, then its curvature k(t) at the point γ (t) is defined to be
|γ (t)|. Let n be the normal on γ . Since γ (t) · n = 0,
dn
γ (t) · n = −γ (t) · n (t) ≡ −e · , (3.11)
de
where e is the tangent unit vector. Taking into account above, we call e · dn
de
the signed
curvature of γ .
In this book, the surfaces under consideration are pieces of boundary of 3-D or
2-D bounded connected domains, and so the surfaces are oriented and the outsides
of surfaces are fixed.
Definition 3.2 If a piece of surface Γ on a neighborhood of x ∈ Γ is on the opposite
(same) side of the outward normal vector with respect to tangent plane (line for
l = 2) at x or coincides with the tangent plane, then the surface is said to be convex
(concave) at x. If the surface is convex (concave) at all x ∈ Γ , then Γ is said to be
convex (concave).
Lemma 3.1 If Γ = j Γ j , Γ j are convex (concave), S p is the shape operator of

Γ at p ∈ Γ and k( p) is twice the mean curvature at p ∈ Γ , then for v ∈ T p Γ the
quadratic forms (S p v, v)Rl−1 and (k( p)v, v)Rl−1 are positive (negative).
Proof Let us consider only the case of convexity, since the case of concavity can be
proved in the same way. Let l = 3. Since surface Γ j is convex, then at every point
of Γ j the quadratic form
L(Δξ )2 + 2K Δξ Δζ + N (Δζ )2
is negative (cf. Remark 3.1), and so the matrix S is positive at every point of Γ j . By
the argument as above the case that l = 2 is proved.
Since k( p) = tr (S), we have the second conclusion.
3.1.2 Properties on the Boundary Surface of Vector Fields
Let Γ be a surface (curve for l = 2) of C 2 and v be a vector field of C 2 on a domain

of Rl near Γ . Let n be the outward unit normal on Γ .
Theorem 3.1 Suppose that v · n|Γ = 0. Then, at every point of the surface Γ the
following hold:
1
(E(v)n, τ )Rl = (rot v × n, τ )Rl − (S ṽ, τ̃ )Rl−1 , (3.12)
2
∂v
(rot v × n, τ )Rl = ,τ + (S ṽ, τ̃ )Rl−1 , (3.13)
∂n Rl
1 ∂v 1
(E(v)n, τ )Rl = ,τ − (S ṽ, τ̃ )Rl−1 , (3.14)
2 ∂n Rl 2
∂v
where E(v) denotes the matrix with the components εi j (v) = 21 ( ∂∂vx ij + ∂ xij ), τ is tan-
gent vectors on Γ , S is the shape operator of the surface Γ (the matrix (3.10)) for
l = 3 and the signed curvature of Γ for l = 2, and ṽ, τ̃ are expressions of the vectors
v, τ in a local orthogonal curvilinear coordinates on Γ .
Proof Let l = 3. For vector field v(x) = (v1 (x), v2 (x), v3 (x)) denoted in orthogonal
coordinates x, let ⎛ ∂v1 ⎞
∂ x1
· · · ∂∂vx31
J (v) = ⎝ · · · · · · · · · ⎠ .
∂v1
∂ x3
· · · ∂∂vx33
Since
1 1
E(v)n = J (v)T + J (v) n = J (v)T − J (v) n + J (v)n, (3.15)
2 2
J (v)T − J (v) = rot v × n, (3.16)
we have
1
E(v)n = rot v × n + J (v)n. (3.17)
2
On the other hand, n(x) ∈ C 1 since Γ ∈ C 2 . Let us make a vector field on a small
enough neighborhood of p ∈ Γ in R3 as follows. Consider a family of curves orthog-
onal to Γ none of which intersect with all the others. Assign tangent unit vector to
every point of the lines. Then, the vector coincides with the unit normal n(x) at
x ∈ Γ . Denote the vector field again by n(x). Since v(x) · n(x)|Γ = 0, the surface
Γ is a contour of the scalar function v · n, and so ∇(v · n) is orthogonal on Γ . Thus,
τ · ∇(v · n)|Γ = 0. (3.18)
∇(v · n) = J (v)n + J (n)v, and from (3.18) we get
τ · J (v)n = −τ · J (n)v on Γ. (3.19)
Since
τ · J (n)v = J (n)T τ · v = ∇τ n · v = (S τ̃ , ṽ)Rl−1 = (S ṽ, τ̃ )Rl−1 on Γ, (3.20)
where the fact that the shape operator is self-adjoint (Corollary 3.1) was used, from
(3.17), (3.19) and (3.20) we have (3.12).
Since
∂v
(J (v)T − J (v))n = rot v × n, J (v)T n = ,
∂n
we get
∂v
(rot v × n, τ ) = ,τ − (J (v)n, τ ), (3.21)
∂n
which, together with (3.19) and (3.20), implies (3.13). Formulas (3.12) and (3.13)
imply (3.14).
When l = 2, putting v = (v1 (x1 , x2 ), v2 (x1 , x2 ), 0) and considering a cylindrical
surface with the line Γ as a section, we get the conclusion.
Remark 3.3 The bilinear form (S ṽ, ũ)Rl−1 for vectors u, v tangent to the surface
is independent of choice of orthogonal curvilinear coordinate system on the surface
(see (3.6)).
Theorem 3.2 On the surface Γ the following holds:
∂v
(E(v)n, n)Rl = ,n . (3.22)
∂n Rl
If v · τ |Γ = 0, then
∂v
(E(v)n, n)Rl = ,n = −(k(x)v, n)Rl + div v, (3.23)
∂n Rl
where k(x) = 2 · mean curvature.

Proof Let n(x) be the vector field expanded on a domain near Γ as in the proof of
Theorem 3.1. Since
1 1 1 ∂v 1
E(v)n = J (v)T n + J (v)n = + J (v)n,
2 2 2 ∂n 2

l
l
∂v j ∂v
(J (v)n, n) Rl = n j ni = ,n ,
j=1 i=1
∂ xi ∂n Rl
we have (3.22).
In a neighborhood W ⊂ R3 of a point x0 ∈ Γ let us introduce local curvilin-
ear coordinates (y1 (x1 , x2 , x3 ), y2 (x1 , x2 , x3 ), y3 (x1 , x2 , x3 )) (if l = 2 then omit y3 )
such that the coordinate lines (y1 ), (y2 ) and (y3 ) are orthogonal at all points to each
other, the linear (y3 ) is the outward unit normal n on Γ and the surface y3 = 0 coin-
cides with Γ , where x = (x1 , x2 , x3 ) is the original orthogonal coordinates. Denote
the unit vector of ∇x yk by ek , then e1 , e2 are tangent on Γ and e3 = n. At x0 ∈ Γ let
us calculate div v(x0 ). Denote vτ = v − (v · n)n. Then
∂v ∂v ∂v
div v(x0 ) = (x0 ) · e1 + (x0 ) · e2 + (x0 ) · n
∂e1 ∂e2 ∂n
∂vτ ∂vn n ∂vτ ∂vn n
= (x0 ) · e1 + (x0 ) · e1 + (x0 ) · e2 + (x0 ) · e2 (3.24)
∂e1 ∂e1 ∂e2 ∂e2
∂v
+ (x0 ) · n.
∂n
Since v · n = 0,
∂vτ ∂vτ
(x0 ) · e1 + (x0 ) · e2 = 0. (3.25)
∂e1 ∂e2
Also,
∂vn n ∂vn n ∂vn ∂n

(x0 ) · e1 + (x0 ) · e2 = (x0 )n · e1 + vn (x0 ) · e1
∂e1 ∂e2 ∂e1 ∂e1
∂vn ∂n
+ (x0 )n · e2 + vn (x0 ) · e2
∂e2 ∂e2
∂n ∂n
= · e1 + · e2 vn
∂e1 ∂e2
= tr(S)v · n = 2(mean curvature of Γ ) v · n.
(3.26)
(see Lemma 7 of [2]). Taking into account (3.25) and (3.26), we have from (3.24)
∂v
,n = −(k(x)v, n)Rl + div v on Γ, (3.27)
∂n Rl
which implies (3.23).
Remark 3.4 If Γ ∈ C 2,1 , then elements of the matrix S belong to C 1 (Γ ) and so

does k(x).
For elements in the divergence-free spaces, Dirichlet, strain and vorticity bilinear
forms are not equal in usual. Below we show that for more general bilinear forms in
space H1 (Ω) when the bilinear forms are equal to each other.
3
Let Ω be a bounded domain such that ∂Ω ∈ C 0,1 , ∂Ω = i=1 Γ i , Γi = j Γi j ,
Γi j ∈ C and
2

HΓ1 1 (Ω) = u ∈ H1 (Ω) : u |Γ1 = 0, u · τ |Γ2 = 0, u · n |Γ3 = 0 .
Corollary 3.2 Assume that Γi j for i = 2, 3 are pieces of plane (straight segments
for 2-D). Then,
(∇v, ∇u) = (rot v, rot u) + (div v, div u) = 2(E(v), E(u)) − (div v, div u)
(3.28)
∀v, u ∈ HΓ1 1 (Ω).
Proof By density of smooth functions, it suffices to prove for v, u ∈ C 2 (Ω̄) ∩

HΓ1 1 (Ω).
By integration by parts, we get
∂v
−(Δv, u)Ω = (∇v, ∇u) − ,u . (3.29)
∂n Γ2 ∪Γ3
On the other hand, using the facts that
−Δv = rot rot v − grad(div v),
(rot v, u) − (v, rot u) = −(v × n, u)∂Ω
(see (2.82)), we get
−(Δv, u)Ω = (rot v, rot u)Ω + (div v, div u) − (rot v × n, u)Γ2 ∪Γ3 − (div v, u · n)Γ2 ∪Γ3
(3.30)
= (rot v, rot u)Ω + (div v, div u) − (rot v × n, u)Γ3 − (div v, u · n)Γ2 ,
where u · τ |Γ2 = 0 and u · n |Γ3 = 0 were used. Since S ≡ 0 on Γ3 , by (3.13) we

have
∂v
(rot v × n, u)Γ3 = ,u . (3.31)
∂n Γ3
Since k(x) = tr (S) ≡ 0 on Γ2 and v · τ |Γ2 = 0, multiplying (3.27) by (u · n), we get
∂v
(div v, u · n)Γ2 = ,u . (3.32)
∂n Γ2
From (3.29)–(3.32), the first equality of (3.28) follows.

Also, using
∂v
Δvi = 2div Ei (v) − div , where Ei (v) = (εi1 (v), · · · , εil (v)),
∂ xi
∂v
(Δvi , u i ) = −2(Ei (v), ∇u i ) + 2(Ei (v) · n, u i )∂Ω − (div , ui )
∂ xi
((2.77) and (2.78)), by the fact that the tensor E(v) is symmetric we get
−(Δv, u)Ω = 2(E(v), E(u)) + (∇(div v), u) − 2(E(v)n, u)Γ2 ∪Γ3

= 2(E(v), E(u)) − (div v, div u) − 2(E(v)n, u)Γ2 ∪Γ3 + (div v, u · n)Γ2 .
(3.33)
By (3.14) and (3.22), we have
∂v
−2(E(v)n, u)Γ3 = − ,u (3.34)
∂n Γ3
and
∂v
−2(E(v)n, u)Γ2 = −2 ,u . (3.35)
∂n Γ2
From (3.32)–(3.35), we obtain
∂v
−(Δv, u)Ω = 2(E(v), E(u)) − (div v, div u) − ,u . (3.36)
∂n Γ2 ∪Γ3
Thus, (3.36) and (3.29) imply the second equality of (3.28).
Remark 3.5 For polygon or polyhedron, the first equality of (3.28) follows from
Theorem 4.1 of [3].
3.2 Variational Formulations of the Steady Problems
Let
(i) Ω be a bounded domain of Rl , l = 2, 3;
7
(ii) ∂Ω ∈ C 0,1 , ∂Ω = i=1 Γ i , Γi ∩ Γ j = ∅ for i = j;
(iii) Γi = j Γi j , where Γi j are connected open subsets of ∂Ω and Γi j ∈ C 2 , i =
2, 3, 4, 5, 7.
We consider the Stokes equations
−μΔv + ∇ p = f, div v = 0 in Ω (3.37)
and the Navier-Stokes equations

3.2 Variational Formulations of the Steady Problems 95
− μΔv + (v · ∇)v + ∇ p = f, div v = 0 in Ω (3.38)
with the following boundary conditions:
(1) v|Γ1 = h 1 ,
(2) vτ |Γ2 = 0, − p|Γ2 = φ2 ,
(3) vn |Γ3 = 0, rot v × n|Γ3 = φ3 /μ,
(4) vτ |Γ4 = h 4 , (− p + 2μεnn (v))|Γ4 = φ4 ,
(3.39)
(5) vn |Γ5 = h 5 , 2(μεnτ (v) + αvτ )|Γ5 = φ5 , α : a matrix,
(6) (− pn + 2μεn (v))|Γ6 = φ6 ,
∂v
(7) vτ |Γ7 = 0, − p + μ · n = φ7 ,
∂n Γ7
or
(1) v|Γ1 = h 1 ,
1
(2) vτ |Γ2 = 0, −( p + |v|2 )|Γ2 = φ2 ,
2
(3) vn |Γ3 = 0, rot v × n|Γ3 = φ3 /μ,

1
(4) vτ |Γ4 = h 4 , − p − |v|2 + 2μεnn (v) = φ4 , (3.40)
2 Γ4

1
(6) − pn − |v|2 n + 2μεn (v) = φ6 ,
2 Γ6
1 2 ∂v
(7) vτ |Γ7 = 0, − p − |v| + μ · n = φ7 ,
2 ∂n Γ7
where
(i) vn = v · n, vτ = v − (v · n)n;
(ii) εn (v) = E(v) · n, εnn (v) = (E(v) · n, n)Rl , εnτ (v) = E(v) · n − εnn (v)n;
(iii) h i , φi , αi j (components of matrix α) are given functions or vectors of functions.
The boundary condition (3.40) is obtained from (3.39) by replacing the static
pressure p with the total pressure p + 1/2|v|2 . Taking into account (v · ∇)v =
rot v × v + ∇ 21 |v|2 and rewriting the Navier-Stokes equation as
∂v 1
− μΔv + rot v × v + ∇( p + |v|2 ) = f, div v = 0 in Ω,
∂t 2
we are concerned with the total pressure and total stress. Then we study the Stokes
equation with boundary conditions (3.39), whereas we study the Navier-Stokes equa-
tions with boundary condition (3.39) or (3.40).
For convenience in what follows, the Navier-Stokes problems with boundary

conditions (3.39) and (3.40) are, respectively, called the case of static pressure and
the case of total pressure.
Let
V = {u ∈ H1 (Ω) : div u = 0, u |Γ1 = 0, u τ |(Γ2 ∪Γ4 ∪Γ7 ) = 0, u n |(Γ3 ∪Γ5 ) = 0},

VΓ 237 = {u ∈ H1 (Ω) : div u = 0, u τ |(Γ2 ∪Γ7 ) = 0, u n |Γ3 = 0}.
Let us consider variational formulations based on the bilinear form (E(v), E(u)).
Taking into account (2.81) and applying Theorems 3.1 and 3.2 on Γi (i = 2, 3, 7),
for v ∈ H2 (Ω) ∩ VΓ 237 and u ∈ V, we have
−(Δv, u) = 2(E(v), E(u)) − 2(E(v) · n, u)∪i=2

7
Γi
= 2(E(v), E(u)) + 2(k(x)v, u)Γ2 − (rot v × n, u)Γ3

+ 2(S ṽ, ũ)Γ3 − 2(εn (v), u)∪i=4
7
Γi
= 2(E(v), E(u)) + 2(k(x)v, u)Γ2 − (rot v × n, u)Γ3 + 2(S ṽ, ũ)Γ3

− 2(εnn (v), u · n)Γ4 − 2(εnτ (v), u)Γ5 − 2(εn (v), u)Γ6
∂v
− ,u + (k(x)v, u)Γ7 .
∂n Γ7
(3.41)
Also, for p ∈ H 1 (Ω) and u ∈ V we have
(∇ p, u) = ( p, u · n)∪i=2
7
Γi
(3.42)
= ( p, u n )Γ2 + ( p, u n )Γ4 + ( pn, u)Γ6 + ( p, u n )Γ7 ,
where u · n |Γ3 ∪Γ5 = 0 was used. By (3.41) and (3.42), we have
−μ(Δv, u) + (∇ p, u)
= 2μ(E(v), E(u)) + 2μ(k(x)v, u)Γ2 + 2μ(S ṽ, ũ)Γ3 + μ(k(x)v, u)Γ7

+ ( p, u · n)Γ2 − μ(rot v × u, u)Γ3 − (− p + 2μεnn (v)), u · n Γ
4
∂v
− 2μ(εnτ (v), u)Γ5 − (− pn + 2μ(εn (v)), u)Γ6 − − p + μ · n , un .
∂n Γ7
(3.43)
Assume that the following holds.

Assumption 3.1 There exists a function U ∈ H1 (Ω) such that
div U = 0, U |Γ1 = h 1 , Uτ |(Γ2 ∪Γ7 ) = 0, Un |Γ3 = 0, Uτ |Γ4 = h 4 , Un |Γ5 = h 5 .
1 1
Also, f ∈ V∗ , φi ∈ H − 2 (Γi ), i = 2, 4, 7, φi ∈ H− 2 (Γi ), i = 3, 5, 6, αi j ∈ L ∞ (Γ5 ), and
Γ1 = ∅.
Then, in view of (3.43), we get a variational formulation for the Stokes problem
(3.37), (3.39):
Problem 3.1 Find v such that
v − U ∈ V,
2μ(E(v), E(u)) + 2μ(k(x)v, u)Γ2 + 2μ(S ṽ, ũ)Γ3 + 2(α(x)vτ , u)Γ5 + μ(k(x)v, u)Γ7

= f, u + φi , u n Γi + φi , u Γi ∀u ∈ V.
i=2,4,7 i=3,5,6
(3.44)
Also, we get a variational formulation for the Navier-Stokes problem of the case
of static pressure:
v − U ∈ V,
2μ(E(v), E(u)) + (v · ∇)v, u + 2μ(k(x)v, u)Γ2 + 2μ(S ṽ, ũ)Γ3
+ 2(α(x)vτ , u)Γ5 + μ(k(x)v, u)Γ7 (3.45)

i=2,4,7 i=3,5,6
On the other hand, taking (v · ∇)v = rot v × v + 21 grad|v|2 into account, we get
a variational formulation for the Navier-Stokes problem of the case of total pressure:
v − U ∈ V,
2μ(E(v), E(u)) + rot v × v, u + 2μ(k(x)v, u)Γ2 + 2μ(S ṽ, ũ)Γ3
+ 2(α(x)vτ , u)Γ5 + μ(k(x)v, u)Γ7 (3.46)

i=2,4,7 i=3,5,6
To show that the formulations above are reasonable, we can prove that if solutions
v are smooth (v ∈ H2 (Ω), f ∈ L2 (Ω)), then there exists p such that (v, p) satisfies
the original problems. We will do it for more general cases including the boundary
conditions of friction type in Chap. 5.
Next, let us consider variational formulations using the bilinear form (∇v, ∇u).
Assume that Γ6 = ∅ and the boundary conditions are as follows:
(1) v|Γ1 = h 1 ,
(2) vτ |Γ2 = 0, − p|Γ2 = φ2 ,
(3) vn |Γ3 = 0, rot v × n|Γ3 = φ3 /μ,
(4 ) vτ |Γ4 = 0, (− p + μεnn (v))|Γ4 = φ4 , (3.47)
(5) vn |Γ5 = 0, 2(μεnτ (v) + αvτ )|Γ5 = φ5 , α : a matrix,
∂v
(7 ) (− pn + μ )|Γ7 = φ7 , φ7 : vector.
∂n
Let
VΓ 1−5 = {u ∈ H1 (Ω) : div u = 0, u |Γ1 = 0, u τ |(Γ2 ∪Γ4 ) = 0, u · n |(Γ3 ∪Γ5 ) = 0},

VΓ 2−5 = {u ∈ H1 (Ω) : div u = 0, u τ |(Γ2 ∪Γ4 ) = 0, u · n |(Γ3 ∪Γ5 ) = 0}.
Applying Theorems 3.1 and 3.2 on Γi (i = 2, 3, 4, 5), for v ∈ H2 (Ω) ∩ VΓ 2−5 and
u ∈ VΓ 1−5 , we have
∂v
−(Δv, u) = (∇v, ∇u) − ,u
∂n ∂Ω
= (∇v, ∇u) + (k(x)v, u)Γ2 − (rot v × n, u)Γ3 + (S ṽ, ũ)Γ3
∂v
− (εn (v), u)Γ4 − (εn (v), u)Γ5 + (S ṽ, ũ)Γ5 − ,u
∂n Γ7
= (∇v, ∇u) + (k(x)v, u)Γ2 − (rot v × n, u)Γ3 + (S ṽ, ũ)Γ3
∂v
− (εnn (v), u · n)Γ4 − 2(εnτ (v), u)Γ5 − (S ṽ, ũ)|Γ5 − ,u .
∂n Γ7
(3.48)
Using (3.48) and (3.42) yields
−μ(Δv, u) + (∇ p, u)
= μ(∇v, ∇u) + μ(k(x)v, u)Γ2 + μ(S ṽ, ũ)Γ3 + μ(k(x)v, u)Γ7

+ ( p, u · n)Γ2 − (rot v × u, u)Γ3 − (− p + μεnn (v)), u · n Γ4
∂v
− 2μ(εnτ (v), u)Γ5 − − pn + μ , un .
∂n Γ7
(3.49)
In view of (3.49), we get a variational formulation for the Stokes problem (3.37),
(3.47):
Problem 3.4 Find v ∈ VΓ 2−5 such that

v|Γ1 = h 1 ,
μ(∇v, ∇u) + μ(k(x)vτ , u)Γ2 + μ(S ṽ, ũ)Γ3 + 2(α(x)vτ , u)Γ5 − μ(S ṽ, ũ)Γ5

= f, u + φi , u n Γi + φi , u Γi ∀u ∈ VΓ 1−5 .
i=2,4,7 i=3,5
(3.50)
For existence of solutions to the problems of (3.44), (3.45) and (3.46), coercivity
of the quadratic form corresponding to the bilinear form
2μ(E(v), E(u)) + 2μ(k(x)v, u)Γ2 + 2μ(S ṽ, ũ)Γ3 + μ(k(x)v, u)Γ7
is important (see the next section), and so is coercivity of the quadratic form corre-
sponding to
μ(∇v, ∇u) + μ(k(x)v, u)Γ2 + μ(S ṽ, ũ)Γ3 − μ(S ṽ, ũ)Γ5
for problem (3.50). Thus, from the point of view of coercivity, two bilinear forms
2(E(v), E(u)) and (∇v, ∇u) seem to be similar. However, relying on the bilinear
form (∇v, ∇u), we cannot reflect the boundary conditions (6) in (3.39), (3.40) into
variational formulations, because we know the relation between strain and normal
derivative of vector fields on the boundary only when the vector fields are tangent or
orthogonal to the boundary. The conditions (4 ), (7 ) of (3.47) are slightly different
from (4), (7) of (3.39).
Remark 3.6 Since S ≡ 0 on the flat surface Γ , the vorticity boundary condition
vn |Γ = 0, rot v × n|Γ = φ/μ
and the Navier slip condition vn |Γ = 0, μεnτ (v)|Γ = φ are equivalent.
Remark 3.7 Condition (7’) of (3.47) (with φ7 = 0) is “do nothing” condition, but
(7) of (3.39) (with φ7 = 0) is rather different from “do nothing” condition and we
cannot replace (7) with (7’).
Let us consider why (7) of (3.39) cannot be replaced by (7’) of (3.47). Above
relying on the bilinear form 2(E(v), E(u)) and integrating by parts (−μΔv + ∇ p, u),
we get boundary integral (−2μ(E(v)n, u)∂Ω + ( p, u · n)∂Ω . Then, in order to reflect
the boundary conditions into Problems 3.1 and 3.2, using vτ = 0 or vn = 0 and
applying Theorems 3.1 or 3.2, we transform the boundary integrals on Γi , i = 2, 3, 7.
(cf. (3.43)). Concretely, under condition vτ |Γ7 = 0 we have
∂v
− pn + μ ,u ∀u with u τ = 0. (3.51)
∂n Γ7
∂v
Usually, vτ = 0 does not imply ∂n
· τ = 0, but by virtue of the conditions u τ = 0
and (7) of (3.39) we have
∂v ∂v
− pn + μ , u Γ7 = − p + μ n, u n Γ7 = φ7 , u n Γ7 ∀u with u τ = 0.
∂n ∂n
(3.52)
∂v
Thus, replacing − p + μ ∂n · n, u n Γ7 by φ7 , u n Γ7 , we reflect the boundary condi-
tion (7) of (3.39) into Problems 3.1 and 3.2.
∂v
Let us replace (7) of (3.39) by (− pn + μ ∂n )|Γ7 = φ7 with a vector φ7 . Trans-
∂v

forming as above and replacing − pn + μ ∂n , u Γ7 by φ7 , u Γ7 , we can come to a
formal variational formulation. But when a solution v is smooth enough, going back
from the formal variational formulation to the original problem, we come to
∂v
(− pn + μ , u)Γ7 = φ7 , u Γ7 ∀u with u τ = 0. (3.53)
∂n
If we have (3.53) without u τ = 0, then from (3.53) we can get
∂v
− pn + μ = φ7 on Γ7 .
∂n
But due to u τ = 0, we get only
∂v
(− pn + μ , n)Γ7 =< φ7 , n >Γ7 .
∂n
This shows that the formal variational formulation is not equivalent to the original
∂v
condition on Γ7 and equivalent to (− p + μ ∂n n)|Γ7 = φ7 · n.
Remark 3.8 “Do nothing” boundary condition (7 ) with φ7 = 0 is used in truncating
large physical domains to smaller computational domains by assuming parallel flow.
If the flow is parallel near the boundary, then (7) of (3.39) is the same as (7’). In the
outlet the boundary condition μE(v)n − pn = 0 is also used. But, to our knowledge
it seems not known whether this condition and “do nothing” condition are equivalent
or not. As “do nothing” boundary condition, the boundary condition
(4 ) vτ = 0, μE(v)n − pn = φ4
also results from the variational formulation based on (∇v, ∇u). By Theorem 3.2,
we know that for divergence-free flows orthogonal to the boundary, two conditions
(4 ) and (7 ) are equivalent in variational formulations above.
3.3 Existence of Solutions to the Steady Problems
When V is a Banach space, the zero element of V is denoted by 0V and O M (0V )

means M-neighborhood of 0V .
For the Stokes problem we have
3.3 Existence of Solutions to the Steady Problems 101
Theorem 3.3 Assume that the surfaces Γ2 j , Γ3 j , Γ7 j are convex and α is a positive
matrix. Then, under Assumption 3.1 there exists a unique solution to Problem 3.1
for the steady Stokes system with mixed boundary condition (3.39) for any f and
φi , i = 2, · · · , 7.
Proof Having in mind Assumption 3.1 and putting v = w + U , we get a new prob-
lem equivalent to Problem 3.1:
Find w ∈ V such that
2μ(E(w), E(u)) + 2μ(k(x)w, u)Γ2 + 2μ(S w̃, ũ)Γ3 + 2(α(x)w, u)Γ5 + μ(k(x)w, u)Γ7
= −2μ(E(U ), E(u)) − 2μ(k(x)Uτ , u)Γ2 − 2μ(SŨ , ũ)Γ3 − 2(α(x)Uτ , u)Γ5

− μ(k(x)U, u)Γ7 + f, u + φi , u · n Γi + φi , u Γi ∀u ∈ V.
i=2,4,7 i=3,5,6
(3.54)
Define a linear operator A : V → V∗ by
Aw, u =2μ(E(w), E(u)) + 2μ(k(x)w, u)Γ2 + 2μ(S w̃, ũ)Γ3

(3.55)
+ 2(α(x)w, u)Γ5 + μ(k(x)w, u)Γ7 ∀w, u ∈ V.
Then, by Remark 3.4 and Assumption 3.1, we have
|Aw, u | ≤ K wV · uV ∀w, u ∈ V. (3.56)
By Korn’s inequality
2μ(E(w), E(w)) ≥ δw2V ∃δ > 0
(see Theorem 1.29), Assumption 3.1 and Lemma 3.1, we have
Aw, w ≥ β1 w2V ∃β1 > 0. (3.57)
Define an element F ∈ V∗ by
F, u = − 2μ(E(U ), E(u)) − 2μ(k(x)U, u)Γ2 − 2μ(SŨ , ũ)Γ3 − 2(α(x)Uτ , u)Γ5

− μ(k(x)U, u)Γ7 + f, u + φi , u · n Γi + φi , u Γi
i=2,4,7 i=3,5,6
∀u ∈ V.
(3.58)
Thus, in view of (3.56)–(3.58), by the Lax-Milgram lemma we come to our
assertion.
Remark 3.9 In (3.54), we used (α(x)w, u)Γ5 instead of (α(x)wτ , u)Γ5 since w =
wτ on Γ5 , and for simplicity from now on we will use such expressions.
1
Remark 3.10 Since n(x) ∈ C1 (Γ i ), u · n ∈ H 2 (Γi ) (see Theorem 1.23). Also,
1
H (Γi ) = H0 (Γi ) (see (1.11)), and so for φi ∈ H−1/2 (Γi ), i = 3, 5, 6, and φi ∈
1
2
2
H −1/2 (Γi ), i = 2, 4, 7, respectively, φi , u Γi and φi , u n Γi are meaningful.

When the surfaces Γ2 j , Γ3 j and Γ7 j are not convex, if the absolute value of twice
the mean curvature k(x) and the norm of the shape operator S are small enough, then
by Korn’s inequality estimate (3.57) is valid, and so is the assertion.
For the Navier-Stokes problem of the case of total pressure, we have

Theorem 3.4 Suppose that Assumption 3.1 holds and that the surfaces Γ2 j , Γ3 j ,
Γ7 j are convex, α is a positive matrix and U L3 (Ω) is small enough. Then, there
exists a solution to Problem 3.3 for the steady Navier-Stokes system (the case of total
pressure) for any f and φi , i = 2 − 7. If U = 0, f V∗ , φi H − 21 (Γ ) , i = 2, 4, 7,
i
and φi H− 21 (Γ ) , i = 3, 5, 6, are small enough, then the solution is unique.
i
Proof In the same way as in the proof of Theorem 3.3, we get a new problem
equivalent to Problem 3.3:
2μ(E(w), E(u)) + rot w × w, u + rot U × w, u + rot w × U, u

+ 2μ(k(x)w, u)Γ2 + 2μ(S w̃, ũ)Γ3 + 2(α(x)w, u)Γ5 + μ(k(x)w, u)Γ7
= −2μ(E(U ), E(u)) − rot U × U, u − 2μ(k(x)U, u)Γ2 − 2μ(SŨ , ũ)Γ3

− 2(α(x)Uτ , u)Γ5 − μ(k(x)U, u)Γ7 + f, u + φi , u n Γi + φi , u Γi
i=2,4,7 i=3,5,6
∀u ∈ V.
(3.59)
Define a a(w; v, u) : V × V × V → R by
a(w; v, u) =2μ(E(v), E(u)) + rot w × v, u + rot U × v, u + rot w × U, u

+ 2μ(k(x)v, u)Γ2 + 2μ(S ṽ, ũ)Γ3 + 2(α(x)v, u)Γ5 + μ(k(x)v, u)Γ7
∀w, v, u ∈ V.
(3.60)
On the other hand, for any w ∈ V we have
rot w × w, w = 0, rot U × w, w = 0,

(3.61)
|rot w × U, w | ≤ |(rot w × U ) · w| d x ≤ γ w2V · U L3 .
Ω
Also, by Korn’s inequality,
2μ(E(v), E(v)) ≥ δv2V . (3.62)
Therefore, if δ − γ U L3 (Ω) = β2 > 0, then by Assumption 3.1, Lemma 3.1,

(3.61) and (3.62), we have
a(v; v, v) ≥ β2 v2V β2 > 0. (3.63)
Define an element F ∈ V∗ by
F, u = − 2μ(E(U ), E(u)) − rot U × U, u − 2μ(k(x)U, u)Γ2

− 2μ(SŨ , ũ)Γ3 − 2(α(x)Uτ , u)Γ5 − μ(k(x)U, u)Γ7 + f, u
(3.64)
+ φi , u n Γi + φi , u Γi ∀u ∈ V.
i=2,4,7 i=3.5,6
Now, let us prove that
if w k w in V as k → ∞, then a(wk ; w k , u) → a(w; w, u) ∀u ∈ V.

(3.65)
First, let us prove that
rot wk × w k , u → rot w × w, u ∀u ∈ V as k → ∞. (3.66)
Indeed,
rot wk × w k , u − rot w × w, u
(3.67)
= rot w k × (w k − w), u + rot (w k − w) × w, u .
Let us estimate the first term on the right-hand side of (3.67).

rot wk × (w k − w), u ≤ γ1 |rot w k | · |(w k − w)| · |u| d x
Ω (3.68)
≤ γ2 rot w k L2 · (w k − w)L3 · uL6 .
Since wk → w in L3 (Ω) and rot w k is bounded in L2 (Ω), by virtue of (3.68) the

first term on the right-hand side of (3.67) converges to zero as k → ∞.
By the imbedding of the space V into L6 (Ω), we have that wi u j ∈ L 2 (Ω), i, j =
1, · · · , l, for any w, v ∈ V. Also, since wk w in V as k → ∞, it follows that
rot wk rot w in L2 (Ω). Then, the second term on the right-hand side of (3.67)
converges to zero as k → ∞. Thus, we get (3.66).
All terms except rot w × w, u in a(w; w, u) are linear with respect to w, and so
it is easy to check their convergence as k → ∞.
Therefore, we have proved (3.65).
By (3.63) and (3.65), there exists a solution to (3.59) (see Theorem 1.43), and so
(3.46) has a solution.
Let us now prove uniqueness. Taking into account U = 0 and rot w × v, v = 0,
we get from (3.60)
a(w; v, v) ≥ βv2V β > 0 ∀w, v ∈ V.

We can get the following estimates:
|rot w1 × u, v − rot w2 × u, v | ≤ γ3 rot w1 − rot w2 L2 · uL3 · vL6

≤ γ4 w1 − w2 V · uV · vV ∀u, v, w1 , w2 ∈ V,

FV∗ ≤ M1 f V∗ + φi H − 21 (Γ ) + φi H− 21 (Γ ) ,
i i
i=2,4,7 i=3,5,6
where M1 depends on mean curvature, shape operator, μ and α.

Thus, if F
β2
V∗
γ4 < 1, then by virtue of Theorem 1.44 the solution is unique.
For the Navier-Stokes problem of the case of static pressure, we have

Theorem 3.5 Suppose that Assumption 3.1 holds and α is a positive matrix. Let
Γ2 = Γ4 = Γ6 = Γ7 = ∅, the surfaces Γ3 j are convex and U L3 (Ω) is small enough.
Then, there exists a solution to Problem 3.2 for the steady Navier-Stokes system for
any f and φi , i = 3, 5. If, moreover, U H1 , f V∗ , φi H− 21 (Γ ) , i = 3, 5, are small
i
enough, then the solution is unique.
Proof In the same way as in the proof of Theorem 3.3, we get a new problem
2μ(E(w), E(u)) + (w · ∇)w, u + (U · ∇)w, u

+ (w · ∇)U, u + 2μ(S w̃, ũ)Γ3 + 2(α(x)w, u)Γ5
= −2μ(E(U ), E(u)) − (U · ∇)U, u − 2μ(SŨ , ũ)Γ3 (3.69)

− 2(α(x)Uτ , u)Γ5 + f, u + φi , u Γi ∀u ∈ V.
i=3,5
Define a nonlinear operator A : V → V∗ by
Aw, u =2μ(E(w), E(u)) + (w · ∇)w, u + (U · ∇)w, u

+ (w · ∇)U, u + 2μ(S w̃, ũ)Γ3 + 2(α(x)w, u)Γ5 ∀w, u ∈ V.
(3.70)
Using the fact that wn = 0, vn = 0 on any parts of the boundary, we can prove that
(w · ∇)u, v = −(w · ∇)v, u ∀w, v ∈ V, u ∈ H1 (Ω),
and so we have
(w · ∇)w, w = 0, (w · ∇)U, w = −(w · ∇)w, U .
Then,
|(U · ∇)w, w + (w · ∇)U, w | ≤ γ1 w2V · U L3 . (3.71)
If δ − γ1 U L3 = β3 > 0, then by Korn’s inequality (3.62), Assumption 3.1, Lemma

3.1, (3.70) and (3.71) we have
Aw, w ≥ β3 w2V , β3 > 0. (3.72)
Define F ∈ V∗ by
F, u = − 2μ(E(U ), E(u)) − (U · ∇)U, u − 2μ(SŨ , ũ)Γ3

− 2(α(x)Uτ , u)Γ5 + f, u + φi , u Γi ∀u ∈ V. (3.73)
i=3,5
Now, by virtue of (3.72), (3.73) in the same way as in the proof of Theorem 3.4
we have the existence of a solution. Uniqueness can also be proved as in the proof
of Theorem 3.4.
For the Navier-Stokes problem of the case of static pressure, we have
Theorem 3.6 Suppose that Assumption 3.1 holds, α is a positive matrix, the surfaces
Γ2 j , Γ3 j , Γ7 j are convex and U H1 (Ω) is small enough. Then, when f and φi , i =
2, · · · , 7, are small enough, there exists a unique solution to Problem 3.2 for the
steady Navier-Stokes system in a neighborhood of U in H1 (Ω).
Proof In the same way as in the proof of Theorem 3.5, we get the following new
problem equivalent to Problem 3.2:
2μ(E(w), E(u)) + (w · ∇)w, u + (U · ∇)w, u + (w · ∇)U, u

+ 2μ(k(x)w, u)Γ2 + 2μ(S w̃, ũ)Γ3 + 2(α(x)w, u)Γ5 + μ(k(x)w, u)Γ7
= −2μ(E(U ), E(u)) − (U · ∇)U, u − 2μ(k(x)U, u)Γ2 − 2μ(SŨ , ũ)Γ3

− 2(α(x)Uτ , u)Γ5 − μ(k(x)U, u)Γ7 + f, u + φi , u n Γi
i=2,4,7

+ φi , u Γi ∀u ∈ V.
i=3,5,6
(3.74)
Define operators A, B : V → V∗ and an element F ∈ V∗ , respectively, by
Aw, u = 2μ(E(w), E(u)) + (U · ∇)w, u + (w · ∇)U, u + 2μ(k(x)w, u)Γ2

+ 2μ(S w̃, ũ)Γ3 + 2(α(x)w, u)Γ5 + μ(k(x)w, u)Γ7 ∀w, u ∈ V,
Bw, u = (w · ∇)w, u ∀w, u ∈ V,
F, u = −2μ(E(U ), E(u)) − (U · ∇)U, u − 2μ(k(x)U, u)Γ2 − 2μ(SŨ , ũ)Γ3

− 2(α(x)Uτ , u)Γ5 − μ(k(x)U, u)Γ7 + f, u + φi , u n Γi
i=2,4,7

+ φi , u Γi ∀u ∈ V.
i=3,5,6
Then, we can rewrite (3.74) as follows:
Aw, u = F, u − Bw, u ∀u ∈ V. (3.75)
We can see that
|Aw, u | ≤ mwV · uV ,

(3.76)
Aw, w ≥ β4 w2V , ∃β4 > 0, ∀w, u ∈ V.
Also
|Bw1 − Bw2 ,u | ≤ δ Mw1 − w2 V · uV

(3.77)
∀wi ∈ O M (0V ) ≡ {v ∈ V : vV ≤ M}, ∀u ∈ V,

FV∗ ≤ M1 U 2H1 + f V∗ + φi H − 21 (Γ ) + φi H− 21 (Γ ) ,
i i
i=2,4,7 i=3,5,6
(3.78)
where M1 depends on the mean curvature, shape operator, μ and α.
By the Lax-Milgram lemma and (3.76), for any fixed z ∈ O M (0V ) there exists a
unique solution to the problem
Aw = F − Bz, (3.79)
and by (3.76) and (3.77) the solution w to (3.79) is estimated as follows:
1 1
wV ≤ (FV∗ + BzV∗ ) ≤ (FV∗ + δ M 2 ).
β4 β4
Thus, if FV∗ and M are small enough, then the map z → w is strict contract in
O M (0V ), and so there exists a unique solution to (3.75) in O M (0V ), which shows our
assertion.
In the same way as in Theorem 3.3, for the the Stokes system with boundary
condition (3.47), we have
Theorem 3.7 Assume that Γ6 = ∅, the surfaces Γ2 j , Γ3 j are convex, Γ5 j is concave
and α is a positive matrix. Then, under Assumption 3.1 without the condition Uτ |Γ7 =
0 there exists a unique solution to Problem 3.4 for the stationary Stokes system with
mixed boundary condition (3.47) for any f and φi , i = 2, · · · , 5, 7.
Remark 3.11 Assuming that the surfaces Γ2 j , Γ3 j are convex and Γ5 j are concave,
we can obtain existence and uniqueness of solution to the Navier-Stokes problem
with boundary condition (3.47) formulated similarly to Problem 3.4.
3.4 Bibliographical Remark 107
3.4 Bibliographical Remark
Let us point out that explanation for the second fundamental form of surface and
shape operator comes mainly from [1], but the shape operator in this book has the
sign opposite to the one in [1] and so are the signed curvature and mean curvature.
The remaining content of Sect. 3.1 is taken from Sect. 2 of [4]. Formula (3.12)
is a generalization of Lemma 2.1 for 2-D in [5] and for three-dimensional case it is
known in the form
∂n
2 (E(v)n, τ )R3 = (rot v × n, τ )R3 − 2v ·
∂τ
in [6, 7]. For a direct elementary proof of (3.12) not using the knowledge of differ-
ential geometry in Sect. 3.1, we refer to Theorem 2.1 and Lemma A.1 of [4]. The
proof of (3.23) follows from Lemma 7 of [2].
The content of Sect. 3.2 is from Sect. 3 of [4] and the content of Sect. 3.3 is taken
from Sect. 4 of [4]. Some mistakes in [4] are corrected here.
Based on the strain bilinear form, the Stokes and Navier-Stokes problems with
the mixture of Dirichlet and stress boundary conditions were studied. For such a
monograph, we refer to [8]. In a polyhedral domain with Γi = ∅, i = 2, 3, 7,, the
Stokes problem in [9, 10] and the Navier-Stokes problem in [10, 11] were studied.
In [12], the Navier-Stokes problem with Γi = ∅, i = 2, 3, 4, 5, 7, was studied.
In [13], by the potential theory, a mixed boundary value problem for the Navier-
Stokes equations in a bounded Lipschitz two-dimensional domain with Γi = ∅, i =
2, 3, 4, 6, 7, was studied. Reducing the original problem to a boundary integral
equation, [14] studied the Stokes problem in a class of Lipschitz domains with
Γi = ∅, i = 2, 3, 4, 7. In [15], smoothness of a solution to the Navier-Stokes prob-
lems with Γi = ∅, i = 2, 4, 5, 6, 7, or Γi = ∅, i = 2, 3, 4, 6, 7, was studied.
Applying the vorticity bilinear form, someone studied the Stokes and Navier-
Stokes problems with the mixture of Dirichlet, pressure and vorticity boundary
conditions. The Stokes and Navier-Stokes problems with Γi = ∅, i = 4, 5, 6, 7 in
[16–19] and [20], where for the Navier-Stokes problem the total pressure condition
is used on Γ2 . In [21, 22], the Navier-Stokes problems with Γi = ∅, i = 2, 4, 5, 6, 7,
were studied.
Relying on the Dirichlet bilinear form, someone studied the Navier-Stokes prob-
lems with the mixture of Dirichlet, “do nothing” condition (see [23]).
Before [4] no one has considered mixed boundary problems for the Stokes and
Navier-Stokes equations with both stress and pressure boundary conditions.
References
1. A. Pressley, Elementary Differential Geometry (Springer, 2010)

2. U. Kangro, R. Nicoaides, Divergence boundary conditions for vector Helmholtz equations
with divergence constraints. Math. Model. Numer. Anal. 33(3), 479–492 (1999)
3. M. Costabel, A coecive bilinear form for Maxwell’s equations. J. Math. Anal. Appl. 157,
527–541 (1991)
94–114 (2015). Erratum, ibid 135, 249–250 (2016)
5. T. Clopeau, A. Mikelić, R. Robert, On the vanishing viscosity limit for the 2D incompressible
Navier-Stokes equations with the friction type boundary conditions. Nonlinearity 11, 1625–
1636 (1998)
6. H. Beirão da Veiga, F. Crispo, Sharp inviscid limit results under Navier type boundary con-
ditions, An Lp theory. J. Math. Fluid Mech. 12(3), 397–411 (2010)
7. H. Beirão da Veiga, F. Crispo, A missed persistence property for the Euler equations, and its
effect on inviscid limits. Nonlinearity 25(6), 1661–1669 (2012)
8. N.D. Kopachevsky, S.G. Krein, Operator Approach to Linear Problems of Hydrodynamics 1,
2 (Springer Basel AG 2001, 2003)
9. V. Maz’ya, J. Rossmann, Lp estimates of solutions to mixed boundary value problems for the
Stokes system in polyhedral domains. Math. Nachr. 280(7), 751–793 (2007)
10. J. Rossmann, Mixed boundary value problems for Stokes and Navier-Stokes systems in poly-
hedral domains. Oper. Theory Adv. Appl. 193, 269–280 (2009)
11. V. Maz’ya, J. Rossmann, Mixed boundary value problems for the Navier-Stokes system in
polyhedral domains. Arch. Rational Mech. Anal. 194, 669–712 (2009)
12. M. Orlt, A.-M. Sändrig, Regularity of viscous Navier-Stokes flows in nonsmooth domains,
Boundary value problems and integral equations in nonsmooth domains. Lect. Note Pure
Appl. Math. 167, 185–201 (1995)
13. A. Russo, G. Starita, A mixed problem for the steady Navier-Stokes equations. Math. Comput.
Model. 49, 681–688 (2009)
14. R. Brown, I. Mitrea, M. Mitrea, M. Wright, Mixed boundary value problems for the Stokes
system. Trans. Am. Math. Soc. 362(3), 1211–1230 (2010)
15. C. Ebmeyer, J. Frehse, Steady Navier-Stokes equations with mixed boundary value conditions
in three-dimensional Lipschitzian domains. Mathematische Annalen 319, 349–381 (2001)
16. C. Begue, C. Conca, F. Murat, O. Pironneau, A nouveau sur les equations de Stokes et de
Navier-Stokes avec des conditions aux limites sur la pression. C. R. Acad. Sci. Paris Ser.
I(304), 23–28 (1987)
17. C. Begue, C. Conca, F. Murat, O. Pironneau, Les Equations de Stokes et de Navier-Stokes
Avec Des Condition Sur La Pression. Nonlinear Partial Diff. Equat. and Their Appl. College
de France, Seminar IX, 179–384 (1988)
18. C. Conca, F. Murat, O. Pironneau, The Stokes and Navier-Stokes equations with boundary
conditions involving the pressure. Japan J. Math. 20(2), 279–318 (1994)
19. C. Conca, C. Pares, O. Pironneau, M. Thiriet, Navier-Stokes equations with imposed pressure
and velocity fluxes. Inter. J. Numer. Meth. Fluids 20, 267–287 (1995)
20. A.A. Illarionov, A.Yu. Chebotarev, Solvability of a mixed boundary value problem for the
stationary Navier-Stokes equations. Diff. Equat. 37(5), 724–731 (2001)
21. C. Bernardi, F. Hecht, R. Verfürth, A finite element discretization of the three-dimensional
Navier-Stokes equations with mixed boundary conditions. ESAIM: Math. Model. Numer.
Anal. 43, 1185–1201 (2009)
22. V. Girault, Incompressible finite element methods for Navier-Stokes equations with nonstan-
dard boundary conditions in R 3 . Math. Comput. 51(183), 55–74 (1988)
23. M. Beneš, Solutions to the mixed problem of viscous incompressible flows in a channel. Arch.
Math. 93, 287–297 (2009)
Chapter 4
The Non-steady Navier-Stokes System
In this chapter we are concerned with the non-steady Navier-Stokes equations and
Stokes equations with mixed boundary conditions including conditions for velocity,
pressure, stress, vorticity and Navier slip condition together. As in Sect. 3.2, relying
on the result in Sect. 3.1, we embed all these boundary conditions into variational
formulations of problems.
For the problem with boundary conditions involving the total pressure and total
stress, by a transformation of unknown functions and a penalty method we can turn
the problem into an elliptic operator equation for functions defined in the time-spatial
domain. In this case we need not assume that the given data are small enough.
For the problem with boundary conditions involving the static pressure and the
stress, we prove the existence of a unique solution for small data under a compatibility
condition at initial time. We can also prove that if a smooth solution is given, then
under the compatibility condition for the small perturbed data there exists a unique
solution.
Section 4.1 is devoted to the problems with boundary conditions involving the
total pressure and the total stress, and Sect. 4.2 is concerned with the problems with
boundary conditions involving the static pressure and the stress.
4.1 Existence of a Solution: The Case of Total Pressure
Let Ω satisfy the following conditions:

(i) Ω be a bounded domain
7 of R , l = 2, 3,
l
(ii) ∂Ω ∈ C , ∂Ω = i=1 Γ i , Γi ∩ Γ j = ∅ for i = j,

0,1
https://doi.org/10.1007/978-3-030-78659-5_4
110 4 The Non-steady Navier-Stokes System

(iii) Γi = j Γi j , where Γi j are connected open subsets of ∂Ω and Γi j ∈ C 2,1 , i =
2 − 5, 7,
and Q = Ω × (0, T ), Σi = Γi × (0, T ), 0 < T < ∞.
4.1.1 Problem and Variational Formulation
In this section we study the Navier-Stokes problem

⎧ ∂v
⎪
⎨ ∂t − μΔv + (v · ∇)v + ∇ p = f,
⎪
div v = 0, (4.1)
⎪
⎪
⎩
v(0) = v0
with boundary conditions including the total pressure
(1) v|Γ1 = h 1 ,
1
(2) vτ |Γ2 = 0, −( p + |v|2 )|Γ2 = φ2 ,
2
(3) vn |Γ3 = 0, rot v × n|Γ3 = φ3 /μ,
1
(4) vτ |Γ4 = h 4 , − p − |v|2 + 2μεnn (v) = φ4 , (4.2)
2 Γ4
1
(6) − pn − |v|2 n + 2μεn (v) = φ6 ,
2 Γ6
1 ∂v
(7) vτ |Γ7 = 0, − p − |v|2 + μ ·n = φ7 ,
2 ∂n Γ7
where vn = v · n, vτ = v − (v · n)n, εn (v) = E(v)n, εnn (v) = (E(v)n, n)Rl , εnτ (v)
= ε(v)n − εnn (v)n and h i , φi , α jk (components of matrix α) are given functions or
vectors of functions of x, t on Σi .
Define
V = {u ∈ H1 (Ω) : div u = 0, u |Γ1 = 0, u τ |(Γ2 ∪Γ4 ∪Γ7 ) = 0, u n |(Γ3 ∪Γ5 ) = 0},

(Q) = {u ∈ C2 ( Q̄) : div u = 0, u |Σ1 = 0, u τ |(Σ2 ∪Σ4 ∪Σ7 ) = 0, u n |(Σ3 ∪Σ5 ) = 0},
V(Q) = L 2 (0, T ; V),
W(Q) = {the completion of (Q) in the space H1 (Q)}.
Assume that the following holds:

4.1 Existence of a Solution: The Case of Total Pressure 111
Assumption 4.1 There exists a function U ∈ H1 (Q) ∩ L∞ (Q) such that
div U = 0, U |Σ1 = h 1 , Uτ |(Σ2 ∪Σ7 ) = 0, Un |Σ3 = 0, Uτ |Σ4 = h 4 , Un |Σ5 = h 5 .
1 1
Also, f ∈ L 2 (0, T ; V∗ ), φi ∈ L 2 (0, T ; H − 2 (Γi )), i = 2, 4, 7 φi ∈ L 2 (0, T ; H− 2 (Γi )),
i = 3, 5, 6, αi j ∈ L ∞ (0, T ; L ∞ (Γ5 )), v0 − U (x, 0) ∈ H , where H is closure of V in L2 (Ω),
and Γ1 = ∅.
Applying Theorems 3.1, 3.2 on Γi j (i = 2, 3, 7) and taking (v · ∇)v = rot v ×

v + 21 grad|v|2 into account (cf. (3.46)), we get a variational formulation for problem
(4.1), (4.2):
v − U ∈ V(Q),
∂u T T
− v, dt + 2μ(E(v), E(u)) dt + rot v × v, u dt
Q ∂t 0 0
T T
+ 2μ(k(x)v, u)Γ2 dt + 2μ(S ṽ, ũ)Γ3 dt
0 0
T T
+ 2(α(t, x)vτ , u)Γ5 dt + μ(k(x)v, u)Γ7 dt
0 0
T T T
= (v0 , u(0)) + f, u dt + φi , u n Γi dt + φi , u Γi dt
0 0 i=2,4,7 0 i=3,5,6
∀u ∈ (Q) with u(x, T ) = 0.

(4.3)
Theorem 4.1 Under Assumption 4.1 for any U , f and φi , i = 2, · · · , 7, there exists
a solution to Problem 4.1 such that
ess sup v ≤ c.
t∈(0,T )
For the proof of Theorem 4.1, we transform problem (4.3) into an equivalent one.
Taking into account Assumption 4.1 and putting v = z + U , from Problem 4.1 we
get an equivalent new problem:
Find z ∈ V(Q) such that
∂u T T
− z, dt + 2μ(E(z), E(u)) dt + rot z × z, u dt
Q ∂t 0 0
T T
+ (rot z × U + rot U × z, u) dt + 2μ(k(x)z, u)Γ2 dt
0 0
T T T
+ 2μ(S z̃, ũ)Γ3 dt + 2(α(t, x)z, u)Γ5 dt + μ(k(x)z, u)Γ7 dt
0 0 0
T
∂u T T
= U, dt − 2μ(E(U ), E(u)) dt − (rot U × U, u) dt
0 ∂t 0 0 (4.4)
T T
− 2μ(k(x)U, u)Γ2 dt − 2μ(SŨ , ũ)Γ3 dt
0 0
T T T
− 2(α(x)Uτ , u)Γ5 dt − μ(k(x)U, u)Γ7 + f, u dt
0 0 0
T T
+ φi , u n Γi dt + φi , u Γi dt + (v0 , u(0))
0 i=2,4,7 0 i=3,5,6
∀u ∈ (Q) with u(x, T ) = 0.
In (4.4) let us make again a change of the unknown function by w = ek1 t z where
k1 is a constant to be determined in Lemma 4.1 later. Then we have
∂u ∂u ∂ û
− d xdt = −
z e−k1 t w d xdt = − w d xdt − k1 w û d xdt,
Q ∂t Q ∂t Q ∂t Q
∂u ∂u ∂ û
U d xdt = U ek1 t e−k1 t d xdt = U e k1 t + k1 û d xdt,
Q ∂t Q ∂t Q ∂t
where û = e−k1 t u.
Substituting these into (4.4), we see that the problem to find a solution to (4.3) is
equivalent to the following problem.
Find w ∈ V(Q) such that
T ∂ û T T
− w, dt + 2μ (E(w), E(û)) dt + e−k1 t (rot w × w, û) dt
0 ∂t 0 0
T T T
+ rot w × U + rot U × w, û dt + 2μ(k(x)w, u)Γ2 dt + ˜ Γ dt
2μ(S w̃, û) 3
0 0 0
T T T
+ 2(α(t, x)w, û)Γ5 dt + μ(k(x)w, û)Γ7 dt − k1 (w, û) dt
0 0 0
∂ û T T
= Ū + k1 û d xdt − 2μ (E(Ū ), E(û)) dt − (rot Ū × U, û) dt
Q ∂t 0 0
T T
− 2μ(k(x)Ū , û)Γ2 dt − ˜ û)
2μ(SŪ, ˜ Γ dt
3
0 0
T T T
− 2(α(t, x)Ūτ , û)Γ5 dt − μ(k(x)U, û)Γ7 + f¯, û dt
0 0 0
T T
+ φ̄i , û n Γi dt + φ̄i , û Γi dt + (v0 , u(x, 0))
0 i=2,4,7 0 i=3,5,6
∀û ∈ (Q), with û(x, T ) = 0,
(4.5)
where γ̄ = ek1 t γ for any γ .

Therefore, for the proof of Theorem 4.1 it suffices to prove the existence of a
solution w to problem (4.5) and
ess sup w ≤ c. (4.6)

t∈(0,T )
4.1.2 An Auxiliary Problem by Elliptic Regularization
First we will consider the following auxiliary problem by elliptic regularization.

For any positive integer m, we find the functions w m ∈ W(Q) satisfying the
following:

1 ∂w m ∂u T
−w m
· d xdt + 2μ (E(w m ), E(u)) dt
Q m ∂t ∂t 0
T T
+ e−k1 t (rot w m × w m , u) dt + rot w m × U + rot U × w m , u dt
0 0
T T T
+ 2μ(k(x)w m , u)Γ2 dt + 2μ(S w˜m , ũ)Γ3 dt + 2(α(t, x)w m , u)Γ5 dt
0 0 0
T T
+ μ(k(x)w m , u)Γ7 dt − k1 (w m , u) dt + (w m (T ), u(T ))
0 0
∂u T T
= Ū + k1 u d xdt − 2μ (E(Ū ), E(u)) dt − (rot Ū × U, u)Ω(t) dt
Q ∂t 0 0
T T T
− 2μ(k(x)Ū , u)Γ2 dt − ˜ ũ) dt −
2μ(SŪ, 2(α(t, x)Ūτ , u)Γ5 dt
Γ3
0 0 0
T T T
− μ(k(x)Ū , u)Γ7 dt + f¯, u dt + φ̄i , u n Γi dt
0 0 0 i=2,4,7
T
+ φ̄i , u Γi dt + (v0 , u(x, 0))
0 i=3,5,6
∀u ∈ W(Q).
(4.7)
For (4.7) we have the following result on the existence and uniqueness of the
solution:
Lemma 4.1 Under Assumptions 4.1, there exists a constant k1 independent of m
such that there exists a unique solution to problem (4.7).
Proof Define an operator Am from W(Q) into its dual space by

T 1 ∂z ∂u T ∂u
Am z, u = , dt − z, dt
0 m ∂t ∂t 0 ∂t
T T
+ 2μ (E(z), E(u)) dt + e−k1 t (rot z × z, u) dt
0 0
T T
+ (rot z × U + rot U × z, u) dt + 2μ(k(x)z, u)Γ2 dt
0 0 (4.8)
T T
+ 2μ(S z̃, ũ)Γ3 dt + 2(α(t, x)z, u)Γ5 dt
0 0
T T
+ μ(k(x)z, u)Γ7 dt − k1 (z, u) dt + (z(T ), u(T ))
0 0
∀z, u ∈ W(Q).
And also define an element F ∈ W(Q)∗ by

∂u T
F, u = Ū + k1 u d xdt − 2μ (E(Ū ), E(u)) dt
Q ∂t 0
T T T
− (rot Ū × U, u) dt − 2μ(k(x)Ū , u)Γ2 dt − ˜ ũ) dt
2μ(SŪ, Γ3
0 0 0
T T T
− 2(α(t, x)Ūτ , u)Γ5 dt − μ(k(x)Ū , u)Γ7 dt + f¯, u dt
0 0 0
T T
+ φ̄i , u n Γi dt + φ̄i , u Γi dt + (v0 , u(x, 0))
0 i=2,4,7 0 i=3,5,6
∀u ∈ W(Q).
(4.9)
Now, let us consider the existence of a solution to the following problem:
Am w m = F, (4.10)
which is equivalent to the existence of a solution to the auxiliary problem (4.7).

For all z ∈ W(Q) we have
Am z, z =
T
1 ∂z ∂z T ∂z T
, dt − z, dt + 2μ (E(z), E(z)) dt
0 m ∂t ∂t 0 ∂t 0
T T T
+ (rot z × U, z) dt + 2μ(k(x)z, z)Γ2 dt + 2μ(S z̃, z̃)Γ3 dt
0 0 0
T T T
+ 2(α(t, x)z, z)Γ5 dt + μ(k(x)z, z)Γ7 dt − k1 z(t) 2
dt + z(T ) 2 ,
0 0 0
(4.11)
where (rot z × z, z) = 0 and (rot U × z, z) = 0 were used.
Integrating by parts yields
∂z 1
− z d xdt = z(0) 2
− z(T ) 2
for z ∈ W(Q). (4.12)
Q ∂t 2
By (4.11) and (4.12), for all z ∈ W(Q) we have
Am z, z =
T
1
∂z 2
T T
dt + 2μ Σi j εi j (z) 2
dt + (rot z × U, z) dt
0 m ∂t 0 0
T T T (4.13)
+ 2μ(k(x)z, z)Γ2 dt + 2μ(S z̃, z̃)Γ3 dt + 2(α(t, x)z, z)Γ5 dt
0 0 0
T T
1
+ μ(k(x)z, z)Γ7 dt − k1 z(t) 2 dt + z(0) 2
+ z(T ) 2
.
0 0 2
By Korn’s inequality,
T T
2 Σi j εi j (z) 2
dt ≥ c1 z 2
H1 (Ω) dt, c1 > 0. (4.14)
0 0
By Young’s inequality and Assumption 4.1,

T
μc1 T T
(rot z × U, z) dt ≤ z 2
H1 (Ω) dt + c2 z 2
dt. (4.15)
0 2 0 0
On the other hand, by virtue of Remark 3.4 and Assumption 4.1 there exists a
constant M such that
S(x) ∞, k(x) ∞, α L ∞ (0,T ;L ∞ (Γ5 )) ≤ M.
Therefore, there exists a constant c3 such that

T T T
2(α(t, x)z, z)Γ5 dt + 2μ(k(x)z, z)Γ2 dt + 2μ(S z̃, z̃)Γ3 dt
0 0 0
T T μc
1
+ μ(k(x)z, z)Γ7 dt ≤ z 2
H1 (Ω) + c3 z 2
dt ∀z ∈ W(Q)
0 0 4
(4.16)
(see Theorem 1.27).
Taking −k1 large enough in (4.13) independently of m, from (4.13)–(4.16) we
have, therefore
Am z, z ≥ c4 z 2
W(Q) , ∃c4 > 0, ∀z ∈ W(Q), (4.17)
where c4 depends on m.
Now, let us prove that if z k z in W(Q) as k → ∞, then
Am z k , u → Am z, u ∀u ∈ W(Q). (4.18)
First, let us prove that
T T
e−k1 t (rot z k × z k , u) dt → e−k1 t (rot z × z, u) dt ∀u ∈ W(Q) as k → ∞.
0 0
(4.19)
Indeed, straightforward calculation gives
T T
e−k1 t (rot z k × z k , u) dt − e−k1 t (rot z × z, u) dt =
0 0
T T
e−k1 t (rot z k × (z k − z), u) dt + e−k1 t (rot (z k − z) × z), u) dt.
0 0
(4.20)
By the embedding of H 1 (Q) in L 4 (Q) we have e−k1 t zu ∈ L2 (Q) and rot (z k −
z) 0 in L2 (Q) as z k z in W(Q). Thus, the second integral on the right-hand
side of (4.20) converges to zero when k → ∞. Let us consider the first integral on
the right-hand side of (4.20). For any ε ≥ 0 we can choose u ε ∈ (Q) such that
u − u ε W(Q) ≤ ε. Then,
T
e−k1 t (rot z k × (z k − z), u) dt
0
T T
= e−k1 t (rot z k × (z k − z), u ε ) dt + e−k1 t (rot z k × (z k − z), u − u ε ) dt.
0 0
(4.21)
Since z k → z in L2 (Q) as k → ∞,
T
e−k1 t (rot z k × (z k − z), u ε ) dt
0 (4.22)
≤ C ∇z k L2 (Q) zk − z L2 (Q) uε L∞ (Q) → 0 as k → ∞.
Also, since {z k − z} is bounded in W(Q), we have

T
e−k1 t (rot z k × (z k − z), u − u ε ) dt
0
(4.23)
≤ C ∇z k L2 (Q) z k − z L4 (Q) u − u ε L4 (Q)
≤ C z k W(Q) z k − z W(Q) u − u ε W(Q) ≤ Cε.
From (4.21)–(4.23), we see that the first integral on the right-hand side of (4.20)
goes to zero when k → ∞, and so we get (4.19).
It is easy to check that other terms in Am z k , u converge when k → ∞. This fact
together with (4.19) implies (4.18).
By (4.17) and (4.18), there exists a solution to (4.10) (see Theorem 1.43), and
therefore the assertion is proved.
4.1.3 Proof of the Existence of a Solution
Now, we prove Theorem 4.1. To do this we need to establish two lemmas first.
Lemma 4.2 If w m ∈ W(Q) are solutions to problem (4.7) with the k1 in Lemma
4.1, then
1 ∂w m ∂u
d xdt → 0 ∀u ∈ W(Q) as m → ∞. (4.24)
Q m ∂t ∂t
Proof By (4.10), (4.13)–(4.16), we have

2
1 ∂w m μc1 T
1
d xdt + wm 2
H1 (Ω) dt + w m (x, 0) 2
+ w m (x, T ) 2
Q m ∂t 4 0 2

≤ F, w m .
(4.25)
On the other hand,
∂w m ∂ Ū m
Ū d xdt = (Ū , w m (x, T )) − (Ū , w m (x, 0)) − w d xdt
Q ∂t Q ∂t (4.26)
∀w ∈ W(Q).
m
Taking (4.26) into account in (4.9) and applying Young’s inequality to the right-
hand side of (4.25), we have
T 1
∂wm 2 1 m
dt + μc1
T
wm 2H1 (Ω) dt + w (x, 0) 2 + wm (x, T ) 2 ≤ c,
m ∂t 8 0 8
0
(4.27)
where c is independent of m and depends on Ū , φ̄i , v0 , f¯, S, k and k1 .
m 2
Using Q m1 ∂w ∂t
d xdt ≤ c, we can get (4.24). Indeed, by Hölder’s inequality
21
2 2
1 ∂w m ∂u 1 1 ∂w m ∂u
d xdt ≤ √ √ d xdt · d xdt ,
Q m ∂t ∂t m Q m ∂t Q ∂t
which shows (4.24).
Lemma 4.3 If wm ∈ W(Q) are solutions to problem (4.7) guaranteed by Lemma

4.1, then {wm } are relatively compact in L2 (Q).
Proof Putting w̄ m (x, t) = 0 on (Ω × (−T, 2T )) \ Q, let us make w̄m (x, t) an exten-

sion of wm . Thus, by virtue of (4.27) we get
1 ∂ w̄ m 2
d xdt ≤ c,
Ω×(0,T ) m ∂t
T
w̄ m 2 1 dt ≤ c, (4.28)
H ()
0
w̄ m (x, 0) ≤ c,
w̄ m (x, T ) ≤ c.
For 0 < h < T let

t+h
1
whm (x, t) = w̄ m (x, s) ds.
h t
Then, whm | Q ∈ W (Q) and
∂whm (x, t) 1 m
(whm )t := = w̄ (x, t + h) − w̄ m (x, t) .
∂t h
Replacing u by whm in (4.7), we have
T 1 ∂w m (x, t) T T
, (whm )t dt − wm , (whm )t dt + 2μ E(wm ), E(whm ) dt
0 m ∂t 0 0
T
+ e−k1 t rot wm × wm + rot wm × U + rot U × w m , whm dt
0
T T
+ 2μ k(x)wm , whm Γ dt + 2μ S w˜m , w˜hm Γ dt
2 3
0 0
T T T
+ 2 α(t, x)wm , whm Γ dt + μ k(x)wm , whm Γ dt − k1 (wm , whm ) dt
5 7
0 0 0
T
= F(t), whm (t) d xdt,
0
(4.29)
where whm (T ) = 0 was used and F is the one in (4.9).

Assuming w̄(x, t) ∈ C1 (Ω̄ × [0, T ]), we estimate
2
1 1
[w̄(x, t + h) − w̄(x, t)] d xdt.
m h
Ω×(0,T )
Applying Hölder’s inequality, we have

1 1∂ t+h 2
w̄(x, s) ds d xdt
m Ω×(0,T −h) t ∂s
h2
1 1 t
∂ 2
+ w̄(x, T ) + w̄(x, s) ds d xdt
m Ω×(T −h,T ) h 2 T ∂s
1 1 t+h
∂ w̄(x, s) 2 1 2
≤ ds d xdt + w̄(x, T ) 2
m Ω×(0,T −h) h t ∂s mh
1 2 T
∂ 2
(4.30)
+ w̄(x, s) ds · h d xdt
m h2 Ω×(T −h,T ) t ∂s
1 1 ∂ 2
≤ (T − h) w̄(x, t) d xdt + 2 w̄(x, T ) 2
hm Ω×(0,T ) ∂t
∂ 2
+ 2h w̄(x, t) d xdt
Ω×(0,T ) ∂t
1 1 ∂ 2
≤ (T + h) w̄(x, t) d xdt + 2 w̄(x, T ) 2 .
hm Ω×(0,T ) ∂t
Since C1 (Ω̄ × [0, T ]) is dense in H1 (Ω × (0, T )), by (4.28) and (4.30) for any
w̄ m ∈ H1 (Ω × (0, T )) we have
1 1 m 21 c
2
w̄ (x, t) − w̄ m (x, t − h) d xdt ≤√ . (4.31)
m Ω×(0,T ) h h
By Hölder’s inequality, we get from (4.27) and (4.31)
1 ∂w m (x, t) 1 m
· w̄ (x, t) − w̄ m (x, t − h) d xdt
Q m ∂t h
1 ∂w m 2 21 1 1 m 2 21
≤ d xdt w̄ (x, t) − w̄ m (x, t − h) d xdt
Q m ∂t m Q h
c
≤√ .
h
(4.32)
Using (4.28), we have
T T 1 t+h

2μ (E(w m ), E(whm )) dt ≤ c wm H1 () w̄ m (x, s) ds dt
0 0 h t H1 ()
T
1 t+h 1
2
√
≤c wm H1 () √ w̄ m (x, s) 2
H1 () ds dt ≤ c/ h.
0 h t
T (4.33)
Now, let us estimate 0 e−k1 t rot w m × w m , whm dt.
Since
t+h √ t+h 21 √
wm L3 (Ω) ds ≤ h wm 2
L3 () ds ≤ c h w̄ m L 2 (0,T ;H1 (Ω)) ,
2
t t
by Hölder’s inequality and (4.28),
T c T t+h
e−k1 t rot wm × wm , whm dt ≤ rot wm · wm L6 wm L3 ds dt
0 h 0 t
c T c
≤ √ wm 2H1 () dt ≤ √ .
h 0 h
(4.34)
In the same way, we get
T √
rot w m × U + rot U × w m , whm dt ≤ c/ h ,
0
(4.35)
T √
− k1 (w m
, whm ) dt ≤ c/ h.
0
Taking into account
∂whm ∂ Ū m
Ū d xdt = Ū (T ), whm (T ) − Ū (0), whm (0) − wh d xdt
Q ∂t Q ∂t
T
∂ Ū m
= − Ū (0), whm (0) − , wh dt,
0 ∂t
we have
T ∂ Ū T
F, whm = − Ū (0), whm (0) − , whm dt + (Ū k1 whm ) dt
0 ∂t 0
T T T
− 2μ E(Ū ), E(whm ) dt − (rot Ū × U, whm ) dt − 2μ(k(x)Ū , whm )Γ2 dt
0 0 0
T T T
− ˜ w̃ m ) dt −
2μ(SŪ, 2(α(t, x)Ū , whm )Γ5 dt − μ(k(x)Ū , whm )Γ7 dt
h Γ3
0 0 0
T T T
+ f¯, whm dt + φ̄i , whm · n Γi dt + φ̄i , whm Γi dt + (v0 , whm (0)).
0 0 i=2,4,7 0 i=3,5,6
Similarly, we have
c
F, whm ≤√ . (4.36)
h
Let us estimate
1 m
− wm · w̄ (x, t + h) − w̄ m (x, t) d xdt.
Q h
Using −ab = − 21 [(a + b)2 − a 2 − b2 ], we have
T
1
− w̄ m (t), w̄ m (x, t + h) − w̄ m (x, t) Ω
dt
h 0
1 m
T T
= w (x, t)2 dt − 1 w̄ m (x, t + h)2 dt
2h 0 2h 0
T
1 w̄ m (x, t + h) − w̄ m (x, t)2 dt
+
2h 0
T T
1 w m (x, t)2 dt − 1 w̄ m (x, t)2 dt
=
2h 0 2h h
T
(4.37)
1
+ w̄ m (x, t + h) − w̄ m (x, t)2 dt
2h 0
T T
1 w m (x, t)2 dt − 1 w m (x, t)2 dt
=
2h 0 2h h
T
1 w̄ m (x, t + h) − w̄ m (x, t)2 dt
+
2h 0
T
1 w̄ m (x, t + h) − w̄ m (x, t)2 dt.
≥
2h 0
Formulas (4.29) and (4.32)–(4.37) imply

T m √
w̄ (x, t + h) − w̄ m (x, t)2 dt ≤ c[ h + ω(h) 23 ] for T > h > 0.
0
Therefore, by Theorem 1.38 we have the conclusion.
Proof of Theorem 4.1.

Let {wm (Q)} be the sequence of solutions to (4.7) guaranteed by Lemma 4.1. By
(4.27) {w m (Q)} is bounded in V(Q). Then by Lemma 4.3, we can choose its subse-
quence {w k (Q)} such that w k (Q) → w ∈ V(Q) in L2 (Q).
First, let us prove that for u ∈ (Q)
T T
e−k1 t rot w k × w k , u dt → e−k1 t (rot w × w, u) dt as k → ∞.
0 0
(4.38)
We can write
T T
e−k1 t rot w k × w k , u dt − e−k1 t (rot w × w, u) dt
0 0
T T
−k1 t
= e rot w × (w − w), u dt +
k k
e−k1 t rot (w k − w) × w, u dt.
0 0
(4.39)
Since {rot wk } is bounded in L2 (Q), w k → w in L2 (Q) and u ∈ L∞ (Q), the first
integral on the right-hand side of (4.39) converges to zero as k → ∞. Meanwhile,
since e−k1 t wu ∈ L2 (Q) and wk w in V(Q), the second integral on the right-hand
side of (4.39) converges to zero. Thus, (4.38) holds.
It is easy to verify the convergence of other terms in (4.7) when k → ∞. Thus,
passing to limit as k → ∞ in (4.7) with k instead of m, by Lemma 4.2 we have (4.5).
The estimate (4.6) can be obtained in the same way as in the proof of (4.2) in [1].
4.1.4 The Stokes Problem
Let us consider the Stokes problem

⎧
⎨ ∂v − μΔv + ∇ p = f, div v = 0,
∂t (4.40)
⎩
v(0) = v0
with boundary condition
(1) v|Γ1 = h 1 ,
(2) vτ |Γ2 = 0, − p|Γ2 = φ2 ,
(3) vn |Γ3 = 0, rot v × n|Γ3 = φ3 /μ,
(4) vτ |Γ4 = h 4 , (− p + 2μεnn (v))|Γ4 = φ4 ,
(4.41)
(5) vn |Γ5 = h 5 , 2(μεnτ (v) + αvτ )|Γ5 = φ5 ,
(6) (− pn + 2μεn (v))|Γ6 = φ6 ,
∂v
(7) vτ |Γ7 = 0, (− p + μ · n)|Γ7 = φ7 .
∂n
Applying Theorems 3.1 and 3.2 on Γi j (i = 2, 3, 7) (and so using (3.41) and

(3.42)), we get a variational formulation for problem (4.40), (4.41):

v − U ∈ V(Q),
T
∂u T T
− v, dt + 2μ (E(v), E(u)) dt + 2μ (k(x)v, u)Γ2 dt
0 ∂t 0 0
T T T
+ 2μ (S ṽ, ũ)Γ3 dt + 2 (α(t, x)vτ , u)Γ5 dt + μ(k(x)v, u)Γ7
0 0 0
T T T
= (v0 , u(0)) + f, u dt + φi , u n Γi dt + φi , u Γi dt
0 0 i=2,4,7 0 i=3,5,6
∀u ∈ (Q) with u(x, T ) = 0.

(4.42)
Theorem 4.2 Under Assumption 4.1 there exists a unique solution to Problem 4.2
for the non-stationary Stokes system with mixed boundary condition (4.41) and the
solution belongs to C([0, T ]; L2 (Ω)).
Proof Put v = z + U , w = ek1 t z, where k1 is taken as in Lemma 4.1 under consid-

eration of the fact that there is not any nonlinear term. Then, we get the new problem
Find w ∈ L 2 (0, T ; V) such that
T ∂ û T T
− w, dt + 2μ (E(w), E(û)) dt + 2μ(k(x)w, û)Γ2 dt
0 ∂t 0 0
T T
+ ˜ Γ dt +
2μ(S w̃, û) 2(α(t, x)w, û)Γ5 dt
3
0 0
T T
+ μ(k(x)w, û)Γ7 dt − k1 (w, û) dt
0 0
∂ Ū T
= k1 Ū û d xdt − û d xdt − 2μ (E(Ū ), E(û)) dt
Q Q ∂t 0
T T T
− 2μ(k(x)Ū , û)Γ2 dt − ˜ û)
2μ(SŪ, ˜ Γ dt − 2(α(t, x)Ūτ , û)Γ5 dt
3
0 0 0
T T T
− μ(k(x)Ū , u)Γ7 + f¯, û dt + φ̄i , û n Γi dt
0 0 0 i=2,4,7
T
+ φ̄i , û Γi dt + v0 − U (x, 0), u(x, 0) ∀û ∈ (Q), û(x, T ) = 0,
0 i=3,5,6
(4.43)
where γ̄ = ek1 t γ for γ .
Existence of a solution to (4.43) is proved in much the same way as Theorem 4.1
without applying the condition U ∈ L∞ (Q) and Lemma 4.3, which are needed for
the nonlinear term. To complete the proof, by Assumption 4.1 it is enough to prove
that the solutions to (4.43) belong to C([0, T ]; H ) and are unique.
When w ∈ L 2 (0, T ; V) is a solution to (4.43), estimating
T
I ≡ ˜ Γ
− 2μ(E(w), E(û)) − 2μ(k(x)w, u)Γ2 − 2μ(S w̃, û) 3
0

∂ Ū
− 2(α(t, x)w, û)Γ5 − μ(k(x)w, û)Γ7 − k1 (w, û) + k1 (Ū , û) − , û
∂t
− 2μ(E(Ū ), E(û)) − 2μ(k(x)Ū , û)Γ2 − 2μ(SŪ, ˜ û)
˜ Γ − 2(α(t, x)Ūτ , û)Γ

3 5
¯
− μ(k(x)U, u)Γ7 + f , û + φ̄i , û n Γi + φ̄i , û Γi dt
i=2,4,7 i=3,5,6
∀û ∈ L 2 (0, T ; V),
we get
|I | ≤ K û L 2 (0,T ;V) ∀û ∈ L 2 (0, T ; V).
This means that I is a continuous linear functional on L 2 (0, T ; V). Thus, there exists
a F ∈ L 2 (0, T ; V∗ ) such that
T
I = F, û dt. (4.44)
0
Taking any φ ∈ D(0, T ) and u ∈ V ∩ C 2 (Ω) and putting û = φu, by definition of

w ∈ D ∗ (0, T ; V ∗ ) (see Definition 1.13), (4.43), (4.44) and Theorem 1.31 we have
T T T
w (φ), u = − wφ dt, u = − w, φ u dt = F, φu dt
0 0 0
T
= Fφ dt, u ,
0
which means w = F ∈ L 2 (0, T ; V∗ ).

Therefore, w ∈ C([0, T ]; H ) and
T
1
w, w dt = w(T ) 2
− w(0) 2
. (4.45)
0 2
Let w1 , w2 be solutions to (4.43) corresponding to the same given data and w =

w1 − w2 . Then, by (4.43) we have
T T T
w , û dt + 2μ (E(w), E(û)) dt + 2μ(k(x)w, u)Γ2 dt dt
0 0 0
T T
+ ˜ Γ +
2μ(S w̃, û) 2(α(t, x)w, û)Γ5 dt (4.46)
3
0 0
T T
+ μ(k(x)w, û)Γ7 dt − k1 (w, û) dt = 0 for û = φ · u.
0 0
Since the set {û = φu : φ ∈ D(0, T ), u ∈ V ∩ C 2 (Ω)} is dense in L 2 (0, T ; V)

(see Remark 1.8), (4.46) is valid for û = w. Thus, from (4.45) and (4.46) it follows
that
T
w(T ) 2
+β w 2
V dt ≤ 0, β > 0
0
(cf. (4.14), (4.16)), which shows that w ≡ 0, that is, the solution to (4.43) is
unique.
4.2 Existence and Uniqueness of Solutions: The Case of

Static Pressure
In this section we study the Navier-Stokes problem

⎧ ∂v
⎪
⎨ ∂t − μΔv + (v · ∇)v + ∇ p = f,
⎪
div v = 0, (4.47)
⎪
⎪
⎩
v(0) = v0
with boundary conditions including the static pressure. We are concerned with the
Problems I and II, which are distinguished according to boundary conditions. Prob-
lem I is the one with the boundary conditions
(1) v|Γ1 = h 1 ,
(2) vτ |Γ2 = 0, − p|Γ2 = φ2 ,
(3) vn |Γ3 = 0, rot v × n|Γ3 = φ3 /μ,
(4) vτ |Γ4 = h 4 , (− p + 2μεnn (v))|Γ4 = φ4 ,
(4.48)
(5) vn |Γ5 = h 5 , 2(μεnτ (v) + αvτ )|Γ5 = φ5 ,
(6) (− pn + 2μεn (v))|Γ6 = φ6 ,
∂v
(7) vτ |Γ7 = 0, − p + μ ·n = φ7 ,
∂n Γ7
and Problem II is the one with the conditions

4.2 Existence and Uniqueness of Solutions: The Case of Static Pressure 127
(1) v|Γ1 = h 1 ,
(2) vτ |Γ2 = 0, − p|Γ2 = φ2 ,
(3) vn |Γ3 = 0, rot v × n|Γ3 = φ3 /μ,
(4) vτ |Γ4 = h 4 , (− p + μεnn (v))|Γ4 = φ4 , (4.49)
(5) vn |Γ5 = h 5 , 2(μεnτ (v) + αvτ )|Γ5 = φ5 ,
∂v
(6) − pn + μ = φ7
∂n Γ7
together Γ6 = ∅. Condition (6) of (4.49) is “do nothing” condition, but (7) of (4.48)
is rather different from “do nothing” condition (see Remark 3.7).
For Problems I and II stated above (for the corresponding perturbation problems
in Sect. 4.2.3), respectively, conditions Γi j ∈ C 2,1 , i = 2, 3, 7, and Γi j ∈ C 2,1 , i =
2, · · · , 5, are used since Theorems 3.1, 3.2 are applied for variational formulations
for Problem I and Problem II, respectively, on Γi j ∈ C 2,1 (i = 2, 3, 7) and Γi j ∈
C 2,1 (i = 2, · · · , 5).
For the proof of the existence of a unique solution to Problems I and II we use the
Local inverse mapping theorem (see Theorem 1.7).
4.2.1 Existence and Uniqueness of Solutions to Problem I
We use the following notation:

V = u ∈ H1 (Ω) : div u = 0, u|Γ1 = 0, u τ |Γ2 ∪Γ4 ∪Γ7 = 0, u n |Γ3 ∪Γ5 = 0
and
VΓ 237 (Ω) = u ∈ H1 (Ω) : div u = 0, u τ |Γ2 ∪Γ7 = 0, u n |Γ3 = 0 .
Denote by H the completion of V in the space L2 (Ω). Throughout this section

V = {u ∈ H1 (Ω) : div u = 0}.
Let
X = {w ∈ L 2 (0, T ; V); w ∈ L 2 (0, T ; V), w ∈ L 2 (0, T ; V∗ )},

w X = w L 2 (0,T ;V) + w L 2 (0,T ;V) + w L 2 (0,T ;V∗ ) ,
∗ ∗
Y = {w ∈ L (0, T ; V ); w ∈ L (0, T ; V )},
2 2
w Y = w L 2 (0,T ;V∗ ) + w L 2 (0,T ;V∗ ) ,

∗
W = {w ∈ L 2 (0, T ; V); w ∈ L 2 (0, T ; V), w ∈ L 2 (0, T ; V )},
w W = w L 2 (0,T ;V) + w L 2 (0,T ;V) + w L 2 (0,T ;V∗ ) .
Here and in what follows w means the derivative with respect to t of w(t).
For Problem I, we will use the following assumptions:

Assumption 4.2 f, f ∈ L 2 (0, T ; V∗ ), φi , φi ∈ L 2 (0, T ; H − 2 (Γi )), i = 2, 4, 7,
1
φi , φi ∈ L 2 (0, T ; H− 2 (Γi )), i = 3, 5, 6, αi j ∈ L ∞ (Γ5 ), where αi j are components

1
of the matrix α, and Γ1 = ∅.
Assumption 4.3 There exists a function U ∈ W such that
div U = 0, U |Γ1 = h 1 , Uτ |Γ2 ∪Γ7 = 0, Un |Γ3 = 0, Uτ |Γ4 = h 4 , Un |Γ5 = h 5 .
Also, U (0, x) − v0 ∈ V.
1
Remark 4.1 For the particular situation such that h 4 , h 5 = 0, h 1 (t, x) ∈ H002 (Γ1 ),
1
Γ1 h 1 (t, x) · n d x = 0 for every fixed t, and h 1 (t, x) ∈ C (0, T ; H (Γ1 )), applying
2 2
the existence and estimate of a solution to the steady Stokes problem with non-
homogeneous Dirichlet boundary condition (see Theorem IV.1.1 [2]), we can prove
the existence of a function U above.
Taking (3.43) into account, we get the following variational formulation for Prob-
lem I (with boundary conditions (4.48)):
v − U ∈ L 2 (0, T ; V),
v(0) = v0 ,
v , u + 2μ(E(v), E(u)) + (v · ∇)v, u + 2μ(k(x)v, u)Γ2
(4.50)
+ 2μ(S ṽ, ũ)Γ3 + 2(α(x)vτ , u)Γ5 + μ(k(x)v, u)Γ7
= f, u + φi , u n Γi + φi , u Γi for all u ∈ V.
i=2,4,7 i=3,5,6
Using Assumption 4.3 and taking v = z + U , we deduce the following problem

Find z such that
z ∈ L 2 (0, T ; V),
z(0) = z 0 ≡ v0 − U (0) ∈ V,
z , u + 2μ(E(z), E(u)) + (z · ∇)z, u + (U · ∇)z, u + (z · ∇)U, u
+ 2μ(k(x)z, u)Γ2 + 2μ(S z̃, ũ)Γ3 + 2(α(x)z, u)Γ5 + μ(k(x)z, u)Γ7
(4.51)
= −(U , u) − 2μ(E(U ), E(u)) − (U · ∇)U, u − 2μ(k(x)U, u)Γ2
+ φi , u n Γi + φi , u Γi for all u ∈ V.
i=2,4,7 i=3,5,6
Now, define an operator A0 : V → V∗ by
A0 y, u =2μ(E(y), E(u)) + 2μ(k(x)y, u)Γ2 + 2μ(S ỹ, ũ)Γ3

(4.52)
+ 2(α(x)y, u)Γ5 + μ(k(x)y, u)Γ7 for all y, u ∈ V.
Lemma 4.4 ∃δ > 0, ∃k0 ≥ 0; A0 u, u ≥ δ u 2

V − k0 u 2
for all u ∈ V.
Proof By Korn’s inequality we have
2μ(E(u), E(u)) ≥ β u 2
V ∃β > 0, for all u ∈ V. (4.53)
By Remark 3.4 and Assumption 4.2, there exists a constant M such that
S(x) ∞, k(x) ∞, α(x) ∞ ≤ M,
and so there exists a constant c0 (depending on β) such that
2μ(k(x)z, z)Γ2 + 2μ(S z̃, z̃)Γ3 + μ(k(x)z, z)Γ7 + 2(α(x)y, u)Γ5

β (4.54)
≤ z 2H1 (Ω) + c0 z 2 dt for all z ∈ V
2
β
(see Theorem 1.27). Set δ = 2
and k0 = c0 . Then, by (4.53) and (4.54) we come to
the asserted conclusion.
Remark 4.2 In the process of proving Lemma 4.4, we see that if Γi = ∅, i = 2, 3, 7,

or these are unions of pieces of planes (segments in case of 2-D) and Γ5 = ∅ or
α(x) = 0, then we can take k0 = 0.
When k0 > 0, if k0 is not small enough, then the operator defined by (4.52) is not
positive, and so let us transform the unknown function to get a positive operator A
in (4.56) below. Now, let k0 be the constant in Lemma 4.4 and put z = e−k0 t z. Then,
since e−k0 t z = z + k0 z, we get the following problem equivalent to problem (4.51):
Find z such that
z ∈ L 2 (0, T ; V),
z(0) = v0 − U (0) ∈ V,
z (t), u + 2μ(E(z(t)), E(u)) + ek0 t (z(t) · ∇)z(t), u + (U (t) · ∇)z(t), u
+ (z(t) · ∇)U (t), u + k0 (z(t), u) + 2μ(k(x)z(t), u)Γ2
+ 2μ(S z̃(t), ũ)Γ3 + 2(α(x)z(t), u)Γ5 + μ(k(x)z(t), u)Γ7

= e−k0 t − (U (t), u) − 2μ(E(U (t)), E(u)) − (U (t) · ∇)U (t), u
− 2μ(k(x)U (t), u)Γ2 − 2μ(SŨ (t), ũ)Γ3 − 2(α(x)U (t)τ , u)Γ5
− μ(k(x)U (t), u)Γ7 + f (t), u + φi (t), u n Γi
i=2,4,7

+ φi (t), u Γi for all u ∈ V.
i=3,5,6
(4.55)
Define operators A, AU (t) : V → V∗ , respectively, by
Av, u = A0 v, u + (k0 v, u) for all v, u ∈ V, (4.56)
and
AU (t)v, u = (U (t, x) · ∇)v, u + (v · ∇)U (t, x), u for all v, u ∈ V,

(4.57)
where A0 is the operator by (4.52) and k0 is the one in Lemma 4.4. Since U ∈ W , we
have U ∈ C([0, T ]; H1 (Ω)) and so such a definition makes sense. Then, the operator
A is positive definite, and this fact is used in (4.67).
Define an operator B(t) : V → V∗ and an element F(t) ∈ V ∗ , respectively, by
B(t)v, u = ek0 t (v · ∇)v, u for all v, u ∈ V, (4.58)
and

F(t), u =e−k0 t − (U (t), u) − 2μ(E(U (t)), E(u)) − (U (t) · ∇)U (t), u
− 2μ(k(x)U (t), u)Γ2 − 2μ(SŨ (t), ũ)Γ3 − 2(α(x)U (t)τ , u)Γ5

− μ(k(x)U (t), u)Γ7 + f (t), u + φi (t), u n Γi + φi (t), u Γi
i=2,4,7 i=3,5,6
for all u ∈ V.
(4.59)
Then, (4.55) is written as

z ∈ L 2 (0, T ; V),
z(0) = v0 − U (0) ∈ V, (4.60)
z (t) + (A + AU (t) + B(t)) z(t) = F(t).
Now, define operators L , AU , L U , B : X → Y , C : X × X → Y and an ele-

ment F ∈ Y by
(Lz)(t), u = z (t), u + Az(t), u for all z ∈ X , for all u ∈ V,

( AU z)(t), u = AU (t)z(t), u for all z ∈ X , for all u ∈ V,
(L U z)(t), u = z (t), u + (A + AU (t))z(t), u for all z ∈ X , for all u ∈ V,
( Bz)(t), u = B(t)z(t), u for all z ∈ X , for all u ∈ V,
C(w, z)(t), u = e k0 t
(w(t) · ∇)z(t), u + ek0 t (z(t) · ∇)w(t), u
for all w, z ∈ X , for all u ∈ V,
(F)(t) = F(t).
(4.61)
For AU and F, we have
Lemma 4.5 C is a bilinear continuous operator such that X × X → Y . Under
Assumptions 4.2 and 4.3, AU is a linear continuous operator such that X → Y
and F ∈ Y .
Proof Obviously, C is bilinear. When w ∈ X ,

w ∈ L ∞ (0, T ; V), w L ∞ (0,T ;V) ≤c w L 2 (0,T ;V) + w L 2 (0,T ;V)
and by virtue of Hölder’s inequality and Sobolev’s imbedding theorem (Theorem

1.20)
ek0 t (w · ∇)z, u + ek0 t (z · ∇)w, u

≤ c( w L3 ∇z L2 u L6 + z L3 ∇w L2 u L6 ) ≤c w V z V u V
for all w, z, u ∈ V.
Thus,
C(w, z) L 2 (0,T ;V∗ ) ≤c w L ∞ (0,T ;V) z L 2 (0,T ;V) ≤c w X · z X . (4.62)
Also, since
| C(w, z) (t), u | =ek0 t k0 (w · ∇)z, u + k0 (z · ∇)w, u + (w · ∇)z, u

+ (w · ∇)z , u + (z · ∇)w, u + (z · ∇)w , u ,
taking (4.62) into account, we have

C(w, z) L 2 (0,T ;V∗ ) ≤ c C(w, z) L 2 (0,T ;V∗ )

+c w L 2 (0,T ;V) + w L ∞ (0,T ;V) )( z L 2 (0,T ;V) + z L ∞ (0,T ;V)
≤c w X · z X .
(4.63)
(4.62) and (4.63) imply
C(w, z) Y ≤c w X · z X . (4.64)
By the same argument as above, we have
AU z Y ≤c U W · z X . (4.65)
By Assumption 4.2, Remark 3.4 and the trace theorem, we can see that F ∈ Y .
Lemma 4.6 The operator L defined by Lz = (z (0), Lz) for z ∈ X is a linear

continuous and one-to-one operator from X onto H × Y .
Proof The linearity of L is obvious. The fact z ∈ X implies that z ∈ C([0, T ]; H ),

z C([0,T ];H ) ≤ c z X , and so we see that a map z ∈ X → z (0) ∈ H is continu-
ous.
Clearly, z Y ≤ c z X . Also, by Assumption 4.2, Remark 3.4 and the trace
theorem,
| Av, u | ≤ c v V · u V for all v, u ∈ V. (4.66)
Equation (4.66) implies that the mapping z ∈ X → Az ∈ Y is continuous. There-

fore, L is continuous.
Next, we show that L is a one-to-one and surjective operator from X onto H × Y .
First, let us prove that this operator is injective. For this, it is enough to prove that
the inverse image of (0 H , 0Y ) ∈ H × Y by the operator L is 0X . By Lemma 4.4
and (4.56), we can find δ > 0 such that
Av, v ≥ δ v 2
V for all v ∈ V. (4.67)
By (4.66) and (4.67), for any q ∈ V∗ there exists a unique solution y ∈ V to the
following problem:
Ay = q. (4.68)
Let z ∈ X be the inverse image of (0 H , 0Y ) ∈ H × Y by L. Then since z (0) = 0 H ,

putting t = 0 in the first equation of (4.61) we get
Az(0), u = 0 for all u ∈ V,
where z(0) = z(0, x). This means that z(0) is a unique solution to (4.68) for q =
0V∗ , i.e. z(0) = 0V . Putting w = z , we get w(0) = z (0) = 0 H . Taking Lz = 0 into
account and differentiating the first equation of (4.61), we have
w (t), u + Aw(t), u = 0 for all u ∈ V. (4.69)
The operator A defined by
(Aw)(t) = Aw(t) ∀w ∈ L 2 (0, T ; V)
satisfies all conditions of Theorem 1.46. Thus, for problem (4.69) with an initial con-
dition w(0) ∈ H there exists a unique solution w such that w ∈ L 2 (0, T ; V), w ∈
L 2 (0, T ; V∗ ). Since w(0) = 0 H , we have w = 0, which means z = 0X since
z(0) = 0V .
Second, let us prove that L is surjective. Let (w0 , g) ∈ H × Y . Since g ∈ Y , we
have g(0) ∈ V∗ . Then, by (4.66) and (4.67), there exists a unique solution z 0 ∈ V to
problem
Az 0 = g(0) − w0 . (4.70)
Let us consider problem !

w + Aw = g ,
(4.71)
w(0) = w0 .
By Theorem 1.46, there exists a unique solution w such that w ∈ L 2 (0, T ; V), w ∈
L 2 (0, T ; V∗ ) to problem (4.71). Now, put
t
z = z0 + w(s) ds, (4.72)
0
where z 0 is the solution to (4.70). Then, z = w and z ∈ X . Integrating both sides

of the first one in (4.71) from 0 to t and using (4.72), we have
w(t), u + Az(t), u − [ w0 , u + Az 0 , u ] = g(t), u − g(0), u for all u ∈ V. (4.73)
Taking (4.70), (4.72) into account, from (4.73) we get
z (t), u + Az(t), u = g(t), u for all u ∈ V. (4.74)
This means that z ∈ X defined by (4.72) is the inverse image of (w0 , g) ∈ H × Y

by the operator L, i.e. L is surjective.
Lemma 4.7 Under Assumption 4.3, let U (0, x) V be small enough. The operator
L U defined by L U z = (z (0), L U z) for z ∈ X is a linear continuous one-to-one
operator from X onto H × Y .
Proof If z ∈ X , then z ∈ C([0, T ]; V) and

z C([0,T ];V) ≤c z L 2 (0,T ;V) + z L 2 (0,T ;V) .
By virtue of this fact and Lemma 4.5, the operator AU ∈ (X → H × Y ) defined

by AU z = (0 H , AU z) is continuous. Thus, the operator L U defined on X is linear
continuous.
As in Lemma 3.5 of [3] it is proved that the operator AU ∈ (X → Y ) is compact.
Thus, AU ∈ (X → H × Y ) is also compact. Since L U = L + AU , by virtue of
Theorem 1.8 and Lemma 4.6 we see that in order to prove that the operator L U is
one-to-one from X onto H × Y , it is enough to prove that L U is one-to-one from
X into H × Y . To prove the last fact it is enough to show that the inverse image of
(0 H , 0Y ) by L U is 0X . By Hölder’s inequality and Sobolev’s imbedding theorem

(U (t, x) · ∇)v, v + (v · ∇)U (t, x), v ≤ K 0 v V U (t, x) H1 v V. (4.75)
δ
Thus, if U (0, x) V is so small that U (0, x) H1 ≤ 2K 0
, then (4.66), (4.67) and
(4.75) imply

A + AU (0) v, u ≤ c v V · u V for all v, u ∈ V,
δ (4.76)
A + AU (0) v, v ≥ v 2V for all v ∈ V.
2
By (4.76), for any q ∈ V∗ there exists a unique solution y ∈ V to
(A + AU (0))y = q. (4.77)
Let z ∈ X be the inverse image of (0 H , 0Y ) by L. Then z (0) = 0 H , and putting

t = 0 from the third one in (4.61) we get

A + AU (0) z(0), u = 0 for all u ∈ V,
where z(0) = z(0, x). This means that z(0) is the unique solution to (4.77) with
q = 0V∗ , i.e. z(0) = 0V . Therefore, z ∈ X satisfies
!
z (t) + A + AU (t) z(t) = 0,
(4.78)
z(0) = 0V .
Now, making duality pairing with z(t) on both sides of
z (t) + Az(t) = −AU (t)z(t)
and taking (4.67) into account and using Gronwall’s inequality, we can prove z = 0X
as in Lemma 3.8 of [3]. The proof is thus completed.
Lemma 4.8 Under Assumption 4.3 the operator T defined by T z = z (0), (L U

+ B)z for z ∈ X is continuously differentiable, T (0X ) = (0 H , 0Y ) and the Fréchet
derivative of T at 0X is L U .
Proof It is easy to verify that T (0X ) = (0 H , 0Y ). Since the operator L U is linear, its
Fréchet derivative is the same as itself. Therefore, if B is continuously differentiable,
then so is T .
For any w, z ∈ X ,
B(w + z) − Bw (t) = ek0 t w(t) · ∇ z(t) + ek0 t z(t) · ∇ w(t) + ( Bz)(t).
By (4.64), we get
2
Bz Y c z X
lim ≤ lim = 0.
z X →0 z X z X →0 z X
Then, put
C(w, z)(t) ≡ ek0 t w(t) · ∇ z(t) + ek0 t z(t) · ∇ w(t) = ( Bw z)(t).
By Lemma 4.5 Bw : X → Y is continuous, and it is the Fréchet derivative of B

at w and also continuous with respect to w. Thus, T is continuously differentiable.
Also from the formula above we can see that the Fréchet derivative of B at 0X is
zero. Therefore, the Fréchet derivative of T at 0X is L U .
Let us consider problem
A + AU (0) + B(0) u = q. (4.79)
Lemma 4.9 Assume that U (0, x) V is small enough. If the norm of q ∈ V ∗ is small
enough, then there exists a unique solution to (4.79) in some O M (0V ).
Proof Since U (0, x) V is small enough, by (4.76), for any fixed z ∈ V there exists
a unique solution to problem
A + AU (0) w = q − B(0)z. (4.80)
On the other hand,
| B(0)w1 − B(0)w2 , u | ≤K M w1 − w2 V · u V for all wi ∈ O M (0V ), for all u ∈ V.

(4.81)
Owing to (4.76) the solution w to (4.80) is estimated as follows:
2 2
w V ≤ q V∗ + B(0)z V∗ ≤ q V∗ + K M2 .
δ δ
Thus, if q V∗ and M are small enough, then the operator (z → w) maps O M (0V )
into itself and by (4.81) this operator is strictly contract. Therefore, in O M (0V ) there
exists a unique solution to (4.80). Thus, we come to the asserted conclusion.
One of main results of this section is the following:

Theorem 4.3 Suppose that Assumptions 4.2 and 4.3 hold. Assume that U W and
the norms of f, f , φi , φi in the spaces where they belong to are small enough.
If
w0 ≡ F(0) − (A + AU (0) + B(0))z 0 ∈ H (4.82)
(compatibility condition at the initial time), where z 0 = v0 − U (0, ·), and w0 H is

small enough, then there exists a unique solution to Problem 4.3 in the space W .
Proof First, let us prove the existence of a solution.

If U W and the norms of f, f , φi , φi in the spaces that they belong to are small
enough, then F Y is also small enough. By virtue of Lemmas 4.7, 4.8 and the Local
inverse mapping theorem (Theorem 1.7), for any R1 > 0 small enough if F Y , R
are small enough and w1 ∈ O R (0 H ), then there exists a unique z ∈ O R1 (0X ) such
that
z (t) + A + AU (t) + B(t) z(t) = F(t),
(4.83)
z (0) = w1 ∈ O R (0 H ).
Putting t = 0 in (4.83), we get
F(0) − A + AU (0) + B(0) z(0) = w1 ∈ O R (0 H ).
On the other hand, if U W is small enough, then so is U (0, x) V . Thus, when

F(0) − w1 V ∗ is small enough, by Lemma 4.9 there exists a unique solution z 0 ∈
O R2 (0V ) for some R2 > 0 to
A + AU (0) + B(0) z 0 = F(0) − w1 . (4.84)
Since z(0) V ≤ c z X , we can choose R1 such that z(0) ∈ O R2 (0V ), and we have
z(0) = z 0 . Therefore, if F Y is small enough, F(0) − A + AU (0) + B(0) z 0
belongs to H and its norm is small enough, then z ∈ X , the solution to (4.83),
is a solution to problem
!
z (t) + A + AU (t) + B(t) z(t) = F(t),
(4.85)
z(0) = z 0 .
By definitions of A, AU (t), B(t), F, the solution z of (4.85) is also a solution to

(4.55) which is equivalent to (4.51). Thus, ek0 t z + U ∈ W is a solution to (4.50).
Second, let us prove uniqueness. We argue by contradiction.
Let v1 , v2 be two solutions to (4.49) corresponding to the same data. Putting
w = v1 − v2 , we have
w ∈ L 2 (0, T ; V),
w(0) = 0,
w , u + 2μ(E(w), E(u)) + (v1 · ∇)w, u + (w · ∇)v2 , u + 2μ(k(x)w, u)Γ2
+ 2μ(S w̃, ũ)Γ3 + 2(α(x)w, u)Γ5 + μ(k(x)w, u)Γ7 = 0 for all u ∈ V.
(4.86)
Putting w = e−k0 t w, where k0 is the constant in Lemma 4.4, we get e−k0 t w = w +

k0 w. Then, we have
w ∈ L 2 (0, T ; V),
w(0) = 0,
w , u + 2μ(E(w), E(u)) + (v1 · ∇)w, u + (w · ∇)v2 , u (4.87)
+ k0 (w, u) + 2μ(k(x)w, u)Γ2 + 2μ(S w̃, ũ)Γ3
+ 2(α(x)w, u)Γ5 + μ(k(x)w, u)Γ7 = 0 ∀u ∈ V,
which is equivalent to (4.86). By Lemma 4.4,
2μ(E(w), E(w)) + k0 (w, w) + 2μ(k(x)w, w)Γ2 + 2μ(S w̃, w̃)Γ3

+ 2(α(x)w, w)Γ5 + μ(k(x)w, w)Γ7 ≥ δ w V.
2
Taking this into account, we can prove w = 0X as in Lemma 3.8 of [3]. Thus,
uniqueness of solutions is proved, and so the proof is completed.
Remark 4.3 Let us consider more precisely the condition that F(0) − A + AU (0)
+ B(0) z 0 belongs to H and its norm is small enough. By (4.56)–(4.61), we have

F(0) − A + AU (0) + B(0) z 0 , u =

− (U (0, x), u) − 2μ(E(U (0, x)), E(u)) − (U (0, x) · ∇)U (0, x), u
− 2μ(k(x)U (0, x), u)Γ2 − 2μ(SŨ (0), ũ)Γ3 − 2(α(x)U (0, x)τ , u)Γ5

− μ(k(x)U (0, x), u)Γ7 + f (0), u + φi (t), u n Γi + φi (t), u Γi
i=2,4,7 i=3,5,6

− 2μ(E(z 0 ), E(u)) + 2μ(k(x)z 0 , u)Γ2 + 2μ(S z˜0 , ũ)Γ3

+ 2(α(x)z 0 , u)Γ5 + μ(k(x)z 0 , u)Γ7

− (U (0, x) · ∇)z 0 , u + (z 0 · ∇)U (0, x), u − (z 0 , ∇)z 0 , u for all u ∈ V.
(4.88)
Taking into account the fact that U (0, x) + z 0 = v0 , U (0, x) ∈ L2 (Ω) and its norm
is small enough, from (4.88) we can see that the condition mentioned above is
equivalent to the condition w0 ∈ O R (0 H ) for R > 0 small enough, where w 0 is
defined by
w 0 , u ≡ f (0), u + φi (0, x), u n Γi + φi (0, x), u Γi

i=2,4,7 i=3,5,6

− 2μ(E(v0 ), E(u)) + 2μ(k(x)v0 , u)Γ2 + 2μ(S ṽ0 , ũ)Γ3 + 2(α(x)v0 , u)Γ5

+ μ(k(x)v0 , u)Γ7 + (v0 · ∇)v0 , u + k0 (v0 , u) for all u ∈ V.
(4.89)
Remark 4.4 If Γi = ∅, i = 2, · · · , 5, 7, then the problem is reduced to the one in

[4] where a local-in-time solution was studied. In this case k0 = 0 (see Remark 4.1),
and condition (4.82) is the same as (25) in [4]. And our condition for U is also the
same as the one in [4].
4.2.2 Existence and Uniqueness of Solutions to Problem II
Let

V1 = u ∈ H1 (Ω) : div u = 0, u|Γ1 = 0, u τ |(Γ2 ∪Γ4 ) = 0, u n |(Γ3 ∪Γ5 ) = 0 ,

VΓ 2−5 = u ∈ H1 (Ω) : div u = 0, u τ |(Γ2 ∪Γ4 ) = 0, u n |(Γ3 ∪Γ5 ) = 0 .
Denote by H1 the completion of V1 in the space L2 (Ω).

By Theorems 3.1 and 3.2, for v ∈ H2 (Ω) ∩ VΓ 2−5 (Ω), p ∈ H 1 (Ω) and u ∈ V1
we have
−μ(Δv, u) + (∇ p, u)
= μ(∇v, ∇u) + μ(k(x)v, u)Γ2 + μ(S ṽ, ũ)Γ3 + μ(k(x)v, u)Γ7
+ ( p, u · n)Γ2 − (rot v × u, u)Γ3 − (− p + μεnn (v)), u · n Γ4
∂v
− 2μ(εnτ (v), u)Γ5 − − pn + , un .
∂n Γ7
(4.90)
Let
X1 = {w ∈ L 2 (0, T ; V1 ); w ∈ L 2 (0, T ; V1 ), w ∈ L 2 (0, T ; V∗1 )},

w X1 = w L 2 (0,T ;V1 ) + w L 2 (0,T ;V1 ) + w L 2 (0,T ;V∗1 ) ,
Y1 = {w ∈ L 2
(0, T ; V∗1 ); w ∈L 2
(0, T ; V∗1 )},
w Y1 = w L 2 (0,T ;V∗1 ) + w L 2 (0,T ;V∗1 ) .
Unlike Problem I, for Problem II we do not require the condition vτ |Γ7 = 0, and so
instead of Assumptions 4.2 and 4.3, we use the following assumptions:
Assumption 4.4 Assumption 4.2 holds with φ7 , φ7 ∈ L 2 (0, T ; H− 2 (Γ7 )) instead
1
of φ7 , φ7 ∈ L 2 (0, T ; H − 2 (Γ7 )).

1
Assumption 4.5 There exists a function U ∈ W such that

div U = 0, U |Γ1 = h 1 , Uτ |Γ2 = 0, Un |Γ3 = 0, Uτ |Γ4 = h 4 , Un |Γ5 = h 5 ,
where W is the same as in the previous section. Also, U (0, x) − v0 ∈ V1 .
Applying (4.90), we get the following variational formulation for Problem II (with
boundary conditions (4.49)):
v − U ∈ L 2 (0, T ; V1 ),
v(0) = v0 ,

v , u + μ(∇v, ∇u) + (v · ∇)v, u + μ(k(x)v, u)Γ2
(4.91)
+ μ(S ṽ, ũ)Γ3 + 2(α(x)vτ , u)Γ5 − μ(S ṽ, ũ)Γ5
= f, u + φi , u n Γi + φi , u Γi for all u ∈ V1 .
i=2,4 i=3,5,7
Taking into account Assumption 4.4 and putting v = z + U , we get the following
problem equivalent to Problem 4.4:
Find z such that
z ∈ L 2 (0, T ; V1 ),
z(0) ≡ v0 − U (0) ∈ V1 ,
z , u + μ(∇z, ∇u) + (z · ∇)z, u + (U · ∇)z, u + (z · ∇)U, u
+ μ(k(x)z, u)Γ2 + μ(S z̃, ũ)Γ3 + 2(α(x)z, u)Γ5 − μ(S z̃, ũ)Γ5
= − U , u − μ(∇U, ∇u) − (U · ∇)U, u − μ(k(x)U, u)Γ2 − μ(SŨ , ũ)Γ3
− 2(α(x)Uτ , u)Γ5 + μ(SŨ , ũ)Γ5 + f, u + φi , u n Γi + φi , u Γi
i=2,4 i=3,5,7
for all u ∈ V1 .
(4.92)
Define an operator A01 : V1 → V∗1 by
A01 y, u = μ(∇ y, ∇u) + μ(k(x)y, u)Γ2 + μ(S ỹ, ũ)Γ3 + 2(α(x)y, u)Γ5 − μ(S ỹ, ũ)Γ5
for all y, u ∈ V1 .
(4.93)
By virtue of the same argument as in the proof of Lemma 4.4 we get
Lemma 4.10 ∃δ > 0, ∃k1 ≥ 0:
A01 u, u ≥ δ u 2
V1 − k1 u 2
for all u ∈ V1 .
Putting z = e−k1 t z, where k1 is the constant in Lemma 4.10, and using the fact
that e−k1 t z = z + k1 z, we get the following problem equivalent to (4.92):
Find z such that
z ∈ L 2 (0, T ; V1 ),
z(0) = z 0 ≡ v0 − U (0) ∈ V1 ,
z (t), u + μ(∇z, ∇u) + ek1 t (z(t) · ∇)z(t), u + (U (t) · ∇)z(t), u
+ (z(t) · ∇)U (t), u + k1 (z(t), u) + μ(k(x)z(t), u)Γ2 + μ(S z̃(t), ũ)Γ3
+ 2(α(x)z(t), u)Γ5 − μ(S z̃(t), ũ)Γ5

= e−k1 t − (U (t), u) − μ(∇U, ∇u) − (U (t) · ∇)U (t), u
− μ(k(x)U (t), u)Γ2 − μ(SŨ (t), ũ)Γ3 − 2(α(x)U (t), u)Γ5 − μ(SŨ (t)τ , ũ)Γ5

+ f (t), u + φi (t), u n Γi + φi (t), u Γi for all u ∈ V1 .
i=2,4 i=3,5,7
(4.94)
Define operators A1 , A1U (t) by
A1 v, u = A01 v, u + (k1 v, u) for all v, u ∈ V1 , (4.95)
A1U (t)v, u = (U (t, x) · ∇)v, u + (v · ∇)U (t, x), u for all v, u ∈ V1 ,

(4.96)
where A01 is the one defined in (4.93). U ∈ W implies U ∈ C([0, T ]; H1 (Ω)), and
such definitions make sense. Also, define an operator B1 (t) : V1 → V∗1 by
B1 (t)v, u = ek1 t (v · ∇)v, u for all v, u ∈ V1 . (4.97)
Define an element F1 ∈ Y1 by

F1 (t), u = e−k1 t − U (t), u − μ(∇U (t), ∇u) − (U (t) · ∇)U (t), u
− μ(k(x)U (t), u)Γ2 − μ(SŨ (t), ũ)Γ3 − 2(α(x)U (t)τ , u)Γ5 + μ(SŨ (t), ũ)Γ5

+ f, u + φi , u n Γi + φi , u Γi for all u ∈ V1 .
i=2,4 i=3,5,7
(4.98)
Now, in the same way as in the proof of Theorem 4.3 we can prove the following
theorem which is one of the main results of this section.
Theorem 4.4 Suppose that Assumptions 4.4 and 4.5 hold. Assume that U W and
the norms of f, f , φi , φi in the spaces where they belong to are small enough.
If (compatibility condition at the initial instance)
w1 ≡ F1 (0) − (A1 + A1U (0) + B1 (0))z 0 ∈ H1 , (4.99)

where z 0 = v0 − U (0, ·), and w1 H1 is small enough, then in the space W there
exists a unique solution to Problem 4.4.
Remark 4.5 By the same argument as in Remark 4.3, we can see that the condition
(4.99) is equivalent to the condition w 1 ∈ H1 , where w 1 ∈ V1∗ is defined by
w 1 , u = f (0), u + φi (0, x), u n Γi + φi (0, x), u Γi

i=2,4 i=3,5,7

− μ(∇v0 , ∇u) + μ(k(x)v0 , u)Γ2 + μ(S ṽ0 , ũ)Γ3 + 2(α(x)v0 , u)Γ5

− μ(S ṽ0 , ũ)Γ5 + (v0 · ∇)v0 , u + k1 (v0 , u) for all u ∈ V1 ,
(4.100)
with k1 in Lemma 4.10.
Remark 4.6 If U ≡ 0 and Γi = ∅, i = 2, · · · , 5, then problem (4.91) is reduced to

one in [3]. In this case k1 = 0 (see Remark 4.2). If v0 ∈ Hl/2 (Ω), then (v0 · ∇)v0 ∈
L2 (Ω). Thus, the condition above for w 1 being in H1 is the same as one of conditions
of Theorems 3.5–3.8 of [3], but we do not demand v0 ∈ Hr0 (Ω), r0 > 2l .
4.2.3 Existence and Uniqueness of Solutions for Perturbed

Data
In [3] it is proved that if a solution satisfying smoothness and a compatibility condition

is given, then there exists a unique solution for small perturbed data satisfying the
compatibility condition. In this section we get such results for the Problems I and
II. In our results the conditions for a given solution is essentially the same as the
one in [3], but the smoothness condition for the initial functions in the compatibility
condition for small perturbed data is weaker than the one in [3] (see Remark 4.8).
r
Let V 0 = {u ∈ Hr0 (Ω) : div u = 0}, r0 > l/2, and

∗ r
W = w ∈ L 2 (0, T ; V); w ∈ L 2 (0, T ; V), w ∈ L 2 (0, T ; V ), w(0) ∈ V 0 ,
w W = w L 2 (0,T ;V) + w L 2 (0,T ;V) + w ∗
L 2 (0,T ;V ) + w(0) V0.
r
Let us first consider Problem I.

Let W (x, t) ∈ W be a given solution to Problem I. Let v be the solution for the
data perturbed except h i , and put v = z + W . Then, we get a problem for z:
Find z such that
z ∈ L 2 (0, T ; V),
z(0) = z 0 ≡ v0 − W (0, x) ∈ V,
z , u + 2μ(E(z), E(u)) + (z · ∇)z, u + (W · ∇)z, u + (z · ∇)W, u
+ 2μ(k(x)z, u)Γ2 + 2μ(S z̃, ũ)Γ3 + 2(α(x)z, u)Γ5 + μ(k(x)z, u)Γ7
= f, u + φi , u n Γi + φi , u Γi for all u ∈ V,
i=2,4,7 i=3,5,6
(4.101)
where z 0 , f, φi are perturbations of corresponding data.
Remark 4.7 The proofs of this section are similar to the ones in the preceding
subsections. The main difference is that unlike U (0, x) in the preceding subsections
we do not assume smallness of W (0, x).
Define an operator A02 : V → V∗ by

+ 2(α(x)y, u)Γ5 + μ(k(x)y, u)Γ7 + (W (0, x) · ∇)y, u (4.102)
+ (y · ∇)W (0, x), u for all y, u ∈ V.
For A02 we have

Lemma 4.11 There exists δ > 0 and k2 ≥ 0 such that
A02 u, u ≥ δ u 2
V − k2 u 2
for all u ∈ V.
Proof By Korn’s inequality, there exists β > 0 such that
2μ(E(u), E(u)) ≥ β u 2
V for all u ∈ V. (4.103)
By Remark 3.4, there exists a constant M such that
S(x) ∞, k(x) ∞, α(x) ∞ ≤ M.
Then, there exists a constant c0 (depending on β) such that
2μ(k(x)u, u)Γ2 + 2μ(S ũ, ũ)Γ3 + μ(k(x)u, u)Γ7 + 2(α(x)u, u)Γ5

β (4.104)
≤ u 2H1 (Ω) + c0 u 2 for all u ∈ V
4
(see Theorem 1.27).
Let us estimate (W (0, x) · ∇)u, u + (u · ∇)W (0, x), u . Since W (0, x) ∈ C(Ω),
β
(W (0, x) · ∇)u, u ≤ u 2
H1 (Ω) + c1 u 2 . (4.105)
8
Taking div u = 0 into account, we get
∂ W (0, x)
(u · ∇)W (0, x), u = uj u dx
j Ω ∂x j
∂u
= (u · n)(W (0, x) · u) dΓ − uj W (0, x) d x.
∂Ω j Ω ∂x j
Estimating the first term in the right-hand side of the equality above as in (4.104)
and applying Hölder’s inequality in the second term, we have
β
(u · ∇)W (0, x), u ≤ u 2
H1 (Ω) + c2 u 2 . (4.106)
8
Taking δ = β2 , k2 = c0 + c1 + c2 , from (4.103)–(4.106) we get the asserted conclu-

sion.
Put z = e−k2 t z, where k2 is a constant in Lemma 4.11. Then e−k2 t z = z + k2 z,

and we have the following problem equivalent to (4.101):
Find z such that
z ∈ L 2 (0, T ; V),
z(0) = z 0 = v0 − W (0) ∈ V,
z (t), u + 2μ(E(z(t)), E(u)) + ek2 t (z(t) · ∇)z(t), u + (W (t) · ∇)z(t), u
+ (z(t) · ∇)W (t), u + k2 (z(t), u) + 2μ(k(x)z(t), u)Γ2 + 2μ(S z̃(t), ũ)Γ3
+ 2(α(x)z(t), u)Γ5 + μ(k(x)z(t), u)Γ7

−k2 t
=e f (t), u + φi (t), u n Γi + φi (t), u Γi for all u ∈ V.
i=2,4,7 i=3,5,6
(4.107)
Define operators A2 , A W (t) : V → V∗ , respectively, by

+ 2(α(x)y, u)Γ5 + μ(k(x)y, u)Γ7 + k2 (y, u) for all y, u ∈ V,
(4.108)
and
A W (t)v, u = (W (t, x) · ∇)v, u + (v · ∇)W (t, x), u

for all v, u ∈ V,
(4.109)
where k2 is the constant in Lemma 4.11. W ∈ W implies W ∈ C([0, T ]; H1 (Ω)),
and such definitions are well.
In the proof of Lemma 4.11 it is clear that
3β
A2 u, u ≥ u V.
2
(4.110)
4
Also, by Lemma 4.11 we have
β
A2 + A W (0) u, u ≥ u V.
2
(4.111)
4
Define an operator B2 (t) : V → V∗ by
B2 (t)v, u = ek2 t (v · ∇)v, u for all v, u ∈ V. (4.112)
Define operators L 2 , A W , L 2W , B2 : X → Y , C2 : X × X → Y and an ele-

ment F2 ∈ Y by
(L 2 z)(t), u = z (t), u + A2 z(t), u ∀z ∈ X , ∀u ∈ V,

( A W z)(t), u = A W (t)z(t), u ∀z ∈ X , ∀u ∈ V,

(L 2W z)(t), u = z (t), u + A2 + A W (t) z(t), u ∀z ∈ X , ∀u ∈ V,
( B2 z)(t), u = B2 (t)z(t), u ∀z ∈ X , ∀u ∈ V,
C2 (w, z)(t), u = e (w · ∇)z, u + ek2 t (z · ∇)w, u
k2 t
∀z ∈ X , ∀u ∈ V,

(F2 )(t), u = e−k2 t f (t), u + φi (t), u n Γi + φi (t), u Γi ∀u ∈ V.
i=2,4,7 i=3,5,6
(4.113)
By the same argument as in Lemma 4.5 we get
Lemma 4.12 C2 is a bilinear continuous operator such that X × X → Y . Under
Assumption 4.2 A W is a linear continuous operator such that X → Y and F2 ∈ Y .
Using (4.110) instead of (4.67), as Lemma 4.6 we get
Lemma 4.13 The operator L 2 defined by L 2 z = (z (0), L 2 z) for z ∈ X is a linear
continuous one-to-one operator from X onto H × Y .
Now, using (4.111) without assuming the fact that W (0, x) V is small enough,
as in Lemma 4.7 we prove the following:
Lemma 4.14 The operator L 2W defined by L 2W z = (z (0), L 2W z) for z ∈ X is a
linear continuous one-to-one operator from X onto H × Y .
Proof As in Lemma 3.5 of [3] the operator A W : X → Y is compact. Thus, A W :

X → H × Y defined by A W z = {0 H , A W z} is also compact. Since L 2W = L 2 +
A W , in order to get the asserted conclusion by virtue of Theorem 1.8 it suffices to
prove that L 2W is one-to-one from X into H × Y .
To prove the last statement it is enough to show that the inverse image of (0 H , 0Y )
by L 2W is 0X . It is easy to verify that

A2 + A W (0) v, u ≤ c v V · u V for all v, u ∈ V. (4.114)
By (4.111) and (4.114), for any q ∈ V∗ there exists a unique solution y ∈ V to

(A2 + A W (0))y = q. (4.115)
Let z ∈ X be the inverse image of (0 H , 0Y ) by L. Then, z (0) = 0 H , and putting

t = 0 from the third one of (4.113) we get

A2 + A W (0) z(0), u = 0 for all u ∈ V,
where z(0) = z(0, x). This means that z(0) is the unique solution to (4.115) with
q = 0V∗ , i.e. z(0) = 0V . Therefore, z ∈ X satisfies
!
z (t) + A2 + A W (t) z(t) = 0,
(4.116)
z(0) = 0V .
Now, using (4.116) and Gronwall’s inequality, as in Lemma 3.8 of [3] we can prove
z = 0X . The proof is thus completed.
By the same argument as in Lemma 4.8 we get

Lemma 4.15 The operator T2 defined by T2 z = z (0), (L 2W + B2 )z for z ∈ X is
continuously differentiable, T2 0X = (0 H , 0Y ) and the Fréchet derivative of T2 at
0X is L 2W .
Let us now turn to the following problem:
A2 + A W (0) + B2 (0) u = q. (4.117)
Using (4.111) without assuming the fact that W (0, x) V is small enough, as in
Lemma 4.9 we can prove
Lemma 4.16 If the norm of q ∈ V∗ is small enough, then there exists a unique
solution to (4.117) in some O M (0V ).
Using Lemmas 4.12–4.15 and Theorem 1.7 (Local inverse mapping theorem), in
the same way as in Theorem 4.3 we get
Theorem 4.5 Suppose that Assumption 4.2 holds and the norms of f, f , φi , φi in
the spaces that they belong to are small enough.
If
w2 ≡ F2 (0) − (A2 + A2W (0) + B2 (0))z 0 ∈ H, (4.118)
where z 0 = v0 − U (0, ·), and w2 H is small enough, then there exists a unique
solution to (4.101) in the space W .
Remark 4.8 By the same argument as in Remark 4.3, we can see that the condition
(4.118) is equivalent to the condition w 2 ∈ H1 , where w 2 ∈ V1∗ is defined by
w 2 , u = f (0), u + φi (0, x), u n Γi + φi (0, x), u Γi

i=2,4 i=3,5,7

− 2μ(E(z 0 ), E(u)) + 2μ(k(x)z 0 , u)Γ2 + 2μ(S z̃ 0 , ũ)Γ3
+ 2(α(x)z 0 , u)Γ5 + μ(k(x)z 0 , u)Γ7 + (W (0, x) · ∇)z 0 , u

+ (z 0 · ∇)W (0, x), u + k2 (z 0 , u) + (z 0 · ∇)z 0 , u for all u ∈ V
(4.119)
with k2 in Lemma 4.11.
Now let us consider Problem II.
Let W (x, t) ∈ W be a given solution to Problem II. Let v be the solution for the
data perturbed except h i and put v = z + W . Then, we get a problem for z as follows:
Find z such that
z ∈ L 2 (0, T ; V1 ),
z(0) = z 0 ≡ v0 − W (0, x) ∈ V1 ,
z , u + μ(∇z, ∇u) + (z · ∇)z, u + (W · ∇)z, u + (z · ∇)W, u
(4.120)
+ μ(k(x)z, u)Γ2 + μ(S z̃, ũ)Γ3 + 2(α(x)z, u)Γ5 − μ(S z̃, ũ)Γ5
= f, u + φi , u n Γi + φi , u Γi for all u ∈ V1 ,
i=2,4 i=3,5,7
where z 0 , f, φi are perturbations of corresponding data.

By the same argument as in Theorem 4.5 we have
Theorem 4.6 Suppose that Assumption 4.2 holds and the norms of f, f , φi , φi in
the spaces that they belong to are small enough. Define an element w3 ∈ V1∗ by
w3 , u = f (0), u + φi (0, x), u n Γi + φi (0, x), u Γi

i=2,4 i=3,5,7

− μ(∇z 0 , ∇u) + μ(k(x)z 0 , u)Γ2 + μ(S z̃ 0 , ũ)Γ3 + 2(α(x)z 0 , u)Γ5
− μ(S z̃ 0 , ũ)Γ5 + (W (0, x) · ∇)z 0 , u + (z 0 · ∇)W (0, x), u

+ k3 (z 0 , u) + (z 0 · ∇)z 0 , u for all u ∈ V1 ,
(4.121)
where k3 is a constant determined as in Lemma 4.11.
If w3 ∈ O R (0 H1 ) for R > 0 small enough, then there exists a unique solution to
(4.120) in the space W .
Remark 4.9 If Γi = ∅, i = 2, · · · , 5, then problem (4.120) is reduced to the one
in [3]. If z 0 ∈ Hl/2 (Ω), then (z 0 · ∇)z 0 , (W (0, x) · ∇)z 0 , (z 0 · ∇)W (0, x) ∈ L2 (Ω)
and k3 z 0 ∈ L2 (Ω). Thus, the last four terms on the right-hand side of (4.121) do not
give any effect to the condition for w3 being in H1 , and so the conditions in Theorem
4.6 are the same as one of conditions of Theorems 3.5–3.8 of [3]. Thus, Theorem 4.6
guarantees existence of a unique solution under a condition weaker than the one in
[3]. Note that taking W (t, x) ≡ 0 in Theorems 4.5 and 4.6, we can not get Theorems
4.3 and 4.4, since there h i = 0.
Remark 4.10 Let Hk (Ω) = (W k,2 (Ω))l be a Sobolev spaces on Ω with dimension
l, V be a divergence-free subspace of H1 satisfying appropriate boundary conditions,
H —the closure of V in (L 2 (Ω))l , V r0 (Ω) = V ∩ Hr0 (Ω), where r0 > l/2,
X = {w ∈ L 2 (0, T ; V ); w ∈ L 2 (0, T ; V ), w ∈ L 2 (0, T ; V ∗ )},
Y = {w ∈ L 2 (0, T ; V ∗ ); w ∈ L 2 (0, T ; V ∗ )}
and A : V → V ∗ —the Stokes operator. Concerned with the linearized problem of

!
u (t) + (A + B)u(t) = f (t),
u(0) = ϕ,
in [3] it is proved that a map u → {u(0), Lu ≡ u + Au} is linear continuous one-

to-one from X = {u ∈ X : u(0) ∈ V r0 (Ω)} onto Y = {[ϕ, h] : ϕ ∈ V r0 (Ω), h ∈
Y , h(0) − Aϕ ∈ H } and the compatibility condition
f (0) − Aϕ ∈ H, ϕ ∈ V r0 (Ω)
for the existence of a unique solution to the nonlinear problem is obtained.

However, in Sect. 4.2 it is proved that for a modified operator A a map u →
{u (0), Lu ≡ u + Au} is linear continuous one-to-one from X onto H × Y
(Lemma 4.6). Then, it is obtained the compatibility condition
f (0) − Aϕ − B(0)ϕ ∈ H
without ϕ ∈ V r0 (Ω) for the existence of unique solution to a nonlinear problems.

Since B(0)ϕ = (ϕ · ∇)ϕ, for ϕ ∈ V l/2 (Ω) we get B(0)ϕ ∈ H , and so from point of
view of smoothness of the initial functions, the compatibility conditions in Sect. 4.2
are weaker than the one in [3].
The content of Sect. 4.1 is the one of Sect. 5 of [5] and the content of Sect. 4.2 is
taken from [6].
The Navier-Stokes equations on a domain decreasing in time with mixture of
Dirichlet condition, the total pressure and vorticity conditions were studied in [1].
For a more general problem we refer to [7]. For the 2-D Navier-Stokes equations
with mixture of the Dirichlet and pressure boundary conditions, existence of a unique
solution for small data was proved in [8].
The Navier-Stokes equations with mixture of Dirichlet condition and stress con-
dition were studied. In [4] under smoothness condition and a compatibility condition
of data at the initial time the existence of a unique local-in-time solution to the 3-D
Navier-Stokes equations was discussed. In [9] for the Navier-Stokes equations on
the polyhedral domain with mixture of Dirichlet condition, Navier slip condition
and stress condition a local-in-time solution was considered. Here smoothness of
solutions to the corresponding steady Stokes problem was used essentially.
The Navier-Stokes equations with mixture of Dirichlet condition and the outflow
boundary condition were studied. For 2-D Navier-Stokes equations a local-in-time
solution in [10] and a solution for small data were obtained in [11]. Here also smooth-
ness of solutions to the corresponding 2-D steady linear problem is important. In [3]
it was proved that if under a compatibility condition at initial time there exists a
unique solution, then so does for small perturbed data. This result shows that under
the compatibility condition there exists a unique solution for small data. Smooth-
ness of initial function in the compatibility condition of [3] is stronger than the one
in [4, 12].
The Navier-Stokes equations with mixture of Dirichlet condition and the static
pressure boundary condition were studied. In [8] for the 2-D Navier-Stokes problem
existence of a unique solution for small data was established.
References
1. G. Łukaszewicz, On the Navier-Stokes equations in time dependent domains and with bound-
ary conditions involving the pressure. J. Math. Sci. Univ. Tokyo 4(3), 529–550 (1997)
4. P. Kučera, Z. Skalák, Local solutions to the Navier-Stokes equations with mixed boundary
conditions. Acta Appl. Math. 54, 275–288 (1998)
94–114 (2015). Erratum, ibid 135, 249–250 (2016)
6. T. Kim, D. Cao, Existence of solutions to the heat convection equations in a time-dependent
domain with mixed boundary conditions. J. Math. Sci. Univ. Tokyo 22, 531–568 (2015)
7. T. Kim, D. Cao, A non-steady system with friction boundary conditions for flow of heat-
conducting incompressible viscous fluids. J. Math. Anal. Appl. 484, 123676 (2020)
8. S. Marušić, On the Navier-Stokes system with pressure boundary condition. Ann. Univ. Fer-
rara 53, 319–331 (2007)
9. M. Beneš, Mixed initial-boundary value problem for the three-dimensional Navier-Stokes
equations in polyhedral domains. Discret. Contin. Dyn. Syst. Suppl. 02, 135–144 (2011)
10. M. Beneš, P. Kučera, Non-steady Navier-Stokes equations with homogeneous mixed boundary
conditions and arbitrarily large initial condition. Carpathian J. Math. 23(1–2), 32–40 (2007)
11. M. Beneš, P. Kučera, Solutions to the Navier-Stokes equations with mixed boundary conditions
in two-dimensional bounded domains. Math. Nachr. 289(2–3), 194–212 (2016)
Chapter 5
The Steady Navier-Stokes System with
Friction Boundary Conditions
In this chapter we are concerned with the steady Navier-Stokes systems with mixed
boundary conditions which may include Tresca slip condition, leak boundary con-
dition, one-sided leak boundary conditions, velocity, pressure, vorticity, stress and
normal derivative of velocity together. Relying on the results in Sect. 3.1 and using
the strain bilinear form, we embed all these boundary conditions into variational
formulations of corresponding problems.
In Sect. 5.1 for every problem, which is distinguished according to boundary
conditions, we first get the variational formulation which consists of five formulas
with five unknown functions, that is, using velocity, tangent stress on slip surface,
normal stress on leak surface, normal stresses on one-sided leak surfaces together as
unknown functions. To show that such formulations are well defined, we note that
when the solution is smooth enough, these variational formulations are equivalent to
the original PDE problems. It is shown that if the boundary condition for pressure
or stress on a portion of boundary where there is flux is given, then the pressure
is determined uniquely. Then, we get variational inequalities with one unknown
function (velocity) equivalent to the variational formulations with five unknown
functions, by which the Navier-Stokes problem with 11 kinds of boundary conditions
is reduced to one variational inequality.
In Sect. 5.2 we study three kinds of variational inequality obtained in Sect. 5.1.
In Sect. 5.3, using the results in Sect. 5.2, we study the existence, uniqueness
and estimates of solutions to the Navier-Stokes problems with 11 kinds of boundary
conditions. For the problem with boundary conditions involving the static pressure
and stress, the existence of a unique solution is proved when the data are small
enough. For the problem with boundary conditions involving the total pressure and
total stress, the existence and estimate of solutions are proved without smallness of
the data of problem.
https://doi.org/10.1007/978-3-030-78659-5_5
150 5 The Steady Navier-Stokes System with Friction Boundary Conditions
From now on throughout this book the following notations are used. When f ∈
H −1/2 (Γi ), if f, wΓi ≥ 0 (≤ 0) ∀w ∈ C0∞ (Γi ) with w ≥ 0, then we denoted by
f ≥ 0 (≤ 0).
5.1 Variational Formulations of Problems
Let Ω be a bounded
domain of R , l = 2, 3. ∂Ω ∈ C , ∂Ω = ∪i=1 Γ i , Γi ∩ Γ j = ∅2
l 0,1 11
for i = j, Γi = j Γi j , where Γi j are connected open subsets of ∂Ω and Γi j ∈ C

for i = 2, 3, 7 and Γi j ∈ C 1 for others (see Remark 5.2).
We are concerned with the Problems I and II for the Navier-Stokes equations
− μΔv + (v · ∇)v + ∇ p = f, div v = 0, (5.1)
which are distinguished according to boundary conditions. Problem I is one with the
following boundary conditions:
(1) v|Γ1 = h 1 ,
(2) vτ |Γ2 = 0, − p|Γ2 = φ2 ,
(3) vn |Γ3 = 0, rot v × n|Γ3 = φ3 /μ,
(4) vτ |Γ4 = h 4 , (− p + 2μεnn (v))|Γ4 = φ4 ,
(6) (− pn + 2μεn (v))|Γ6 = φ6 ,
(5.2)
∂v
(7) vτ |Γ7 = 0, (− p + μ · n)|Γ7 = φ7 ,
∂n
(8) vn = h 8 , |στ (v)| ≤ gτ , στ (v) · vτ + gτ |vτ | = 0 on Γ8 ,
(9) vτ = h 9 , |σn (v)| ≤ gn , σn (v)vn + gn |vn | = 0 on Γ9 ,
(10) vτ = 0, vn ≥ 0, σn (v) + g+n ≥ 0, (σn (v) + g+n )vn = 0 on Γ10 ,
(11) vτ = 0, vn ≤ 0, σn (v) − g−n ≤ 0, (σn (v) − g−n )vn = 0 on Γ11 ,
and Problem II is one with the following boundary conditions:

5.1 Variational Formulations of Problems 151
(1) v|Γ1 = h 1 ,
(2) vτ |Γ2 = 0, −( p + 1/2|v|2 )|Γ2 = φ2 ,
(3) vn |Γ3 = 0, rot v × n|Γ3 = φ3 /μ,
(4) vτ |Γ4 = h 4 , (− p − 1/2|v|2 + 2μεnn (v))|Γ4 = φ4 ,
(6) (− pn − 1/2|v|2 n + 2μεn (v))|Γ6 = φ6 , (5.3)
∂v
(7) vτ |Γ7 = 0, (− p − 1/2|v|2 + μ · n)|Γ7 = φ7 ,
∂n
(8) vn = h 8 , |στt (v)| ≤ gτ , στt (v) · vτ + gτ |vτ | = 0 on Γ8 ,
(9) vτ = h 9 , |σnt (v)| ≤ gn , σnt (v)vn + gn |vn | = 0 on Γ9 ,
(10) vτ = 0, vn ≥ 0, σnt (v) + g+n ≥ 0, (σnt (v) + g+n )vn = 0 on Γ10 ,
(11) vτ = 0, vn ≤ 0, σnt (v) − g−n ≤ 0, (σnt (v) − g−n )vn = 0 on Γ11 ,
where and in what follows εn (v) = E(v)n, εnn (v) = (E(v)n, n)Rl , εnτ (v) = E(v)n −
εnn (v)n and h i , φi , αkl (components of matrix α) are given functions or vec-
tors of functions. σn and σnt are, respectively, the normal components of stress
and total stress on a boundary surface, that is, σn = σ · n and σnt = σ t · n. Also,
στ (v, p) = σ (v, p) − σn (v, p)n, στt (v, p) = σ t (v, p) − σnt (v, p)n.
Throughout this book we always assume that gτ ∈ L 2 (Γ8 ), gn ∈ L 2 (Γ9 ), g+n ∈

L (Γ10 ), g−n ∈ L 2 (Γ11 ) and that gτ > 0, gn > 0, g+n > 0, g−n > 0, for a.e. x of the
2
portions of boundary.
For Problem II the static pressure p and the stress in the boundary conditions
for Problem I are, respectively, changed with the total pressure and the total stress.
Thus, as in Chaps. 3, 4 for convenience in what follows, the Navier-Stokes problems
with boundary conditions (5.2) and (5.3) are, respectively, called the case of static
pressure and the case of total pressure.
We also consider the Stokes equations
−μΔv + ∇ p = f, div v = 0 (5.4)
with the boundary conditions (5.2), which is Problem III.
Assuming that f ∈ L2 (Ω), φi ∈ L 2 (Γi ), i = 2, 4, 7, and φi ∈ L2 (Γi ), i = 3,

5, 6, formally we introduce the following:
Definition 5.1 A function (v, p) ∈ H2 (Ω) × H 1 (Ω) is called a solution to Problem
I (or Problem II ) if (5.1) holds in L2 (Ω) and each of (5.2) (or (5.3)) holds in L 2 (Γi )
or L2 (Γi ). A function (v, p) ∈ H2 (Ω) × H 1 (Ω) is called a solution to Problem III
if (5.4) holds in L2 (Ω) and each of (5.2) holds in L 2 (Γi ) or L2 (Γi ).
Usually, it is difficult (or impossible) to find solutions in the sense of the definition
above, and we will consider variational formulations of the problems.
5.1.1 Variational Formulation: The Case of Static Pressure
Let

V = u ∈ H1 (Ω) : div u = 0, u|Γ1 = 0, u τ |(∪i=2,4,7,9,10,11 Γi ) = 0, u n |(∪i=3,5,8 Γi ) = 0 ,

VΓ 237 (Ω) = u ∈ H1 (Ω) : div u = 0, u τ |(Γ2 ∪Γ7 ) = 0, u n |Γ3 = 0 ,

K (Ω) = u ∈ V : u n |Γ10 ≥ 0, u n |Γ11 ≤ 0 .
(5.5)
Taking into account (2.81) and applying Theorems 3.1 and 3.2 on Γi j (i = 2, 3, 7),
for v ∈ H2 (Ω) ∩ VΓ 237 (Ω) and u ∈ V we have
−(Δv, u) = 2(E(v), E(u)) − 2(E(v)n, u)∪i=2

11
Γi
= 2(E(v), E(u)) + 2(k(x)v, u)Γ2 − (rot v × n, u)Γ3 + 2(S ṽ, ũ)Γ3

− 2(εnn (v), u n )Γ4 − 2(εnτ (v), u)Γ5 − 2(εn (v), u)Γ6

∂v
− ,u + (k(x)v, u)Γ7 − 2(εnτ (v), u)Γ8
∂n Γ7
− 2(εnn (v), u n )Γ9 − 2(εnn (v), u n )Γ10 − 2(εnn (v), u n )Γ11 .
(5.6)
Also, for p ∈ H 1 (Ω) and u ∈ V we have
(∇ p, u) = ( p, u n )∪i=2
11
Γi = ( p, u n )Γ2 + ( p, u n )Γ4 ∪Γ7 ∪Γ9 ∪Γ10 ∪Γ11 + ( pn, u)Γ6 ,
(5.7)
where u n |Γ3 ∪Γ5 ∪Γ8 = 0 was used.
We assume that the following holds:

Assumption 5.1 (1) There exists a function U ∈ H1 (Ω) such that
div U = 0, U |Γ1 = h 1 , Uτ |(Γ2 ∪Γ7 ) = 0, Un |Γ3 = 0, Uτ |Γ4 = h 4 ,

Un |Γ5 = h 5 , U |Γ8 = h 8 n, U |Γ9 = h 9 , U |Γ10 ∪Γ11 = 0.
f ∈ V∗ , φi ∈ H − 2 (Γi ), i = 2, 4, 7, φi ∈ H− 2 (Γi ), i = 3, 5, 6, αi j ∈
1 1
(2)
∞
L (Γ5 ), and Γ1 = ∅.
(3) If Γi , where i is 10 or 11, is nonempty, then at least one of {Γ j : j ∈
{2, 4, 7, 9 − 11}\{i}} is nonempty and there exists a diffeomorphism in C 1 between
Γi and Γ j .
Having in mind Assumption 5.1 and putting v = w + U , by (5.6) and (5.7) we

can see that solutions (v, p) of the problem (5.1), (5.2) in the sense of Definition 5.1
satisfy the following:
⎧
⎪ v − U = w ∈ K (Ω),
⎪
⎪
⎪
⎪ 2μ(E(w), E(u)) + (w · ∇)w, u + (U · ∇)w, u + (w · ∇)U, u
⎪
⎪
⎪
⎪
⎪
⎪ + 2μ(k(x)w, u)Γ2 + 2μ(S w̃, ũ)Γ3 + 2(α(x)w, u)Γ5 + μ(k(x)w, u)Γ7
⎪
⎪
⎪
⎪ − 2μ(εnτ (w + U ), u)Γ8 + ( p − 2μεnn (w + U ), u n )Γ9 ∪Γ10 ∪Γ11
⎪
⎪
⎪
⎪
⎪
⎪ = −2μ(E(U ), E(u)) − (U · ∇)U, u − 2μ(k(x)U, u)Γ2
⎪
⎪
⎨
⎪
⎪
⎪
⎪ + φi , u n Γi + φi , uΓi ∀u ∈ V,
⎪
⎪
⎪
⎪ i=2,4,7 i=3,5,6
⎪
⎪
⎪
⎪ |στ (v)| ≤ gτ , στ (v) · vτ + gτ |vτ | = 0 on Γ8 ,
⎪
⎪
⎪
⎪ |σn (v)| ≤ gn , σn (v)vn + gn |vn | = 0 on Γ9 ,
⎪
⎪
⎪
⎪
⎪
⎪ σn (v) + g+n ≥ 0, (σn (v) + g+n )vn = 0 on Γ10 ,
⎪
⎩
σn (v) − g−n ≤ 0, (σn (v) − g−n )vn = 0 on Γ11 .
(5.8)
Define a01 (·, ·), a11 (·, ·, ·) and F1 ∈ V ∗ by
a01 (w, u) = 2μ(E(w), E(u)) + (U · ∇)w, u + (w · ∇)U, u

+ 2μ(k(x)w, u)Γ2 + 2μ(S w̃, ũ)Γ3 + 2(α(x)w, u)Γ5
+ μ(k(x)w, u)Γ7 ∀w, u ∈ V,
a11 (w, u, v) = (w · ∇)u, v ∀w, u, v ∈ V,
(5.9)
F1 , u = −2μ(E(U ), E(u)) − (U · ∇)U, u − 2μ(k(x)U, u)Γ2
+ φi , u n Γi + φi , uΓi ∀u ∈ V.
i=2,4,7 i=3,5,6
Then, taking into account
στ (v) = 2μεnτ (v),

σn (v) = − p + 2μεnn (v)
and (5.8), we introduce the following variational formulation for problem (5.1), (5.2).
Problem I-VE. Find (v, στ , σn , σ+n , σ−n ) ∈ U + K (Ω) × L2τ (Γ8 ) × L 2 (Γ9 ) ×
H −1/2 (Γ10 ) × H −1/2 (Γ11 ) such that
⎧
⎪ v − U = w ∈ K (Ω),
⎪
⎪
⎪
⎪
⎪
⎪ a01 (w, u) + a11 (w, w, u) − (στ , u τ )Γ8 − (σn , u n )Γ9
⎪
⎪
⎪
⎪ − σ+n , u n Γ10 − σ−n , u n Γ11 = F1 , u ∀u ∈ V,
⎨
|στ | ≤ gτ , στ · vτ + gτ |vτ | = 0 on Γ8 , (5.10)
⎪
⎪
⎪
⎪ |σn | ≤ gn , σn vn + gn |vn | = 0 on Γ9 ,
⎪
⎪
⎪
⎪
⎪
⎪ σ+n + g+n ≥ 0, σ+n + g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎩
σ−n − g−n ≤ 0, σ−n − g−n , vn Γ11 = 0 on Γ11 ,
where L2τ (Γ8 ) is the subspace of L2 (Γ8 ) consisting of functions such that[
(u, n)L2 (Γ8 ) = 0.
Remark 5.1 If u ∈ H 1 (Ω) and u|∂Ω = 0 on O(Γi )\Γ i , where O(Γi ) is an open
1
subset of ∂Ω such that Γ i ⊂ O(Γi ), then u|Γi ∈ H002 (Γi ) (see Sect. 1.3.2). Since
1 1 1
1
H002 (Γi ) → H02 (Γi ) and H 2 (Γi ) = H02 (Γi ) (see (1.11)),
1 1
H002 (Γi ) → H 2 (Γi ) → H − 2 (Γi ) → (H002 (Γi ))∗ .
1 1
1
Thus, under condition u|∂Ω = 0 on O(Γi )\Γ i , for φi ∈ (H002 (Γi )) a dual prod-
uct φi , uΓi is well defined. However, if φi ∈ H − 2 (Γi ), then without knowing that
1
u|∂Ω = 0 on O(Γi )\Γ i the dual product φi , uΓi is meaningful. Therefore, under
(2) of Assumption 5.1 the dual products on Γi in (5.8) is meaningful.
The following theorem shows that the variational formulation above is well
defined.
Theorem 5.1 Let Assumption 5.1 hold. If (v, p) is a solution in the sense of Defini-
tion 5.1 of the problem (5.1), (5.2), then (v, στ |Γ8 , σn |Γ9 , σn |Γ10 , σn |Γ11 ) is a solution
to Problem I-VE.
Conversely, if f ∈ L2 (Ω) and Problem I-VE has a smooth solution (v, στ , σn , σ+ ,
σ− ) such that v ∈ H2 (Ω), σ+ ∈ L 2 (Γ10 ) and σ− ∈ L 2 (Γ11 ), then there exists p ∈
H 1 (Ω) such that (v, p) is a solution to the problem (5.1), (5.2). Moreover, if at least
one of the sets Γi , i = 2, 4, 6, 7, is nonempty, then p is unique.
Proof It suffices to prove the conversion from Problem I-VE to the problem (5.1)–
(5.2).
Let v be a solution smooth enough as above to Problem I-VE. From (5.9) and
(5.10) we have
2μ(E(v), E(u)) + (v · ∇)v, u + 2μ(k(x)v, u)Γ2 + 2μ(S ṽ, ũ)Γ3

+ 2(α(x)vτ , u)Γ5 + μ(k(x)v, u)Γ7 − (στ , u τ )Γ8 − (σn , u n )Γ9
− σ+n , u n Γ10 − σ−n , u n Γ11 − φi , u n Γi − φi , uΓi (5.11)
i=2,4,7 i=3,5,6
= f, u ∀u ∈ V.
From (5.6) we get
2μ(E(v), E(u)) = − μ(Δv, u) − 2μ(k(x)v, u)Γ2 + μ(rot v × n, u)Γ3 − 2μ(S ṽ, ũ)Γ3
+ 2μ(εnn (v), u · n)Γ4 + 2μ(εnτ (v), u)Γ5 + 2μ(εn (v), u)Γ6

∂v (5.12)
+μ ,u − μ(k(x)v, u)Γ7 + 2μ(εnτ (v), u)Γ8 + 2μ(εnn (v), u)Γ9
∂n Γ7
+ 2μ(εnn (v), u)Γ10 + 2μ(εnn (v), u)Γ11 .
Combining (5.11) with (5.12) yields
(−μΔv + (v · ∇)v − f, u) + μ(rot v × n, u)Γ3 + 2μ(εnn (v), u · n)Γ4

∂v
+ 2μ(εnτ (v), u)Γ5 + 2(α(x)vτ , u)Γ5 + 2μ(εn (v), u)Γ6 + μ ,u
∂n Γ7
+ 2μ(εnτ (v), u)Γ8 + 2μ(εnn (v), u n )Γ9 + 2μ(εnn (v), u n )Γ10 + 2μ(εnn (v), u n )Γ11
− (στ , u τ )Γ8 − (σn , u n )Γ9 − σ+n , u n Γ10 − σ−n , u n Γ11
− φi , u n Γi − φi , uΓi = 0.
i=2,4,7 i=3,5,6
(5.13)
Taking any u ∈ C0∞ (Ω) with div u = 0, we have
(−μΔv + (v · ∇)v − f, u) = 0,

and by Proposition 2.3 there exists a unique P ∈ L 2 (Ω) such that Ω P d x = 0 and
− μΔv + (v · ∇)v − f = −∇ P. (5.14)
Since v ∈ H2 (Ω), f ∈ L2 (Ω) and (v · ∇)v ∈ L2 (Ω), we have that ∇ P ∈ L2 (Ω),

which shows that P ∈ H 1 (Ω).
Substituting (5.14) into (5.13), integrating by parts and taking into account (5.7),
we have (see Remark 1.17)
(−P − φ2 , u n )Γ2 + μ(rot v × n − φ3 /μ, u)Γ3 + (−P + 2μεnn (v) − φ4 , u n )Γ4

+ (2μεnτ (v) + α(x)vτ − φ5 , u)Γ5 + (−Pn + 2μεn (v) − φ6 , u)Γ6
∂v
+ − P +μ · n − φ7 , u n Γ7 + (2μεnτ (v) − στ , u)Γ8
∂n
+ (−P + 2μεnn (v) − σn , u n )Γ9 + (−P + 2μεnn (v) − σ+n , u n )Γ10
+ (−P + 2μεnn (v) − σ−n , u n )Γ11 = 0,
(5.15)
∂v ∂v
where (v, u)Γ5 = (vτ , u)Γ5 and (μ ∂n , u Γ7 = (μ ∂n · n, u n Γ7 were used.
For every i = 3, 5, 8, respectively, let us take any u ∈ V such that u n |Γi = 0 and
u|(∂Ω\Γi ) = 0. The set of traces of such functions on Γi includes any tangent vector
1/2
fields ϕ ∈ H00 (Γi ) (see Sect. 2.3.2.1), and so from (5.15) we get
rot v × n = φ3 /μ on Γ3 ,
2μεnτ (v) + α(x)vτ − φ5 = 0 on Γ5 , (5.16)
2μεnτ (v) − στ = 0 on Γ8 .
If for all i = 2, 4, 6, 7, 9, 10, 11, Γi = ∅, then putting p = P + c, where c is any

constant, we get a solution (v, p) to problem (5.1), (5.2).
Assume that among Γi , i = 2, 4, 6, 7, 9, 10, 11, at least one is nonempty. For
every i = 2, 4, 7, 9, 10, 11, respectively, let us take any u ∈ V such that u τ |Γi = 0
and u|(∂Ω\Γi ) = 0. The set of traces of such functions on Γi include any normal vector
1/2
fields ϕ ∈ H00 (Γi ) such that Γi ϕ d x = 0 (see Theorem IV.1.1 of [1]), and so from
(5.15) we have that for some constants ci , i = 2, 4, 7, 9, 10, 11, respectively,
− P − φ2 = c2 on Γ2 ,
− P + 2μεnn (v) − φ4 = c4 on Γ4 ,
∂v
− P +μ · n − φ7 = c7 on Γ7 ,
∂n (5.17)
− P + 2μεnn (v) − σn = c9 on Γ9 ,
− P + 2μεnn (v) − σ+n = c10 on Γ10 ,
− P + 2μεnn (v) − σ−n = c11 on Γ11 .
(See the proof of (2.90).) Taking any u ∈ V such that u|(∂Ω\Γ6 ) = 0 and arguing as
in the proof of (2.90), we have from (5.15) that for a constant c6
− Pn + 2μεn (v) − φ6 = c6 n on Γ6 . (5.18)
Let us prove that indeed, all ci are equal to one constant c. For example, assume that
Γ2 and Γ4 are nonempty. Taking any u ∈ V such that u|∂Ω = 0 on ∂Ω \ (Γ2 ∪ Γ4 ),
we get from (5.15)
c2 u n d x + c4 u n d x = 0,
Γ2 Γ4

which implies c2 = c4 = c since Γ2 u n d x = − Γ4 u n d x. Thus, from (5.14), (5.17)
and (5.18) we see that p = P + c satisfies
−μΔv + (v · ∇) + ∇ p = f
and all the following boundary conditions:

− p = φ2 on Γ2 ,
− p + 2μεnn (v) = φ4 on Γ4 ,
− pn + 2μεn (v) = φ6 on Γ6 ,
∂v
− p+μ · n = φ7 on Γ7 , (5.19)
∂n
− p + 2μεnn (v) = σn on Γ9 ,
− p + 2μεnn (v) = σ+n on Γ10 ,
− p + 2μεnn (v) = σ−n on Γ11 .
By virtue of (5.10), (5.16) and (5.19), all conditions in (5.2) are satisfied. Therefore,
(v, p) is a solution to problem (5.1), (5.2). If at least one of the sets Γi , i = 2, 4, 6, 7,
is nonempty, then the equality with the given φi above on Γi holds, and p is unique.

We will find a variational inequality equivalent to Problem I-VE.
Let (v, στ , σn , σ+n , σ−n ) be a solution of Problem I-VE. Subtracting the second
formula of (5.10) with u = w from the second one of (5.10), we get
a01 (w, u − w) + a11 (w, w, u − w) − (στ , u τ − wτ )Γ8 − (σn , u n − wn )Γ9

− σ+n , u n − wn Γ10 − σ−n , u n − wn Γ11 = F1 , u − w ∀u ∈ V.
(5.20)
Define the functionals φτ , φn , φ+ , φ− , respectively, by

φτ (η) = gτ |η| d x ∀η ∈ L2τ (Γ8 ),
Γ8

φn (η) = gn |η| d x ∀η ∈ L 2 (Γ9 ),
Γ9
(5.21)
φ+ (η) = g+n η d x ∀η ∈ L 2 (Γ10 ),
Γ10

φ− (η) = − g−n η d x ∀η ∈ L 2 (Γ11 ).
Γ11
Since if u ∈ K (Ω), then u|Γ8 ∈ L2τ (Γ8 ), u n |Γ9 ∈ L 2 (Γ9 ), u n |Γ10 ∈ L 2 (Γ10 ) and
u n |Γ11 ∈ L 2 (Γ11 ), in what follows for convenience we use the notation
φτ (u) = φτ (u|Γ8 ),
φn (u) = φn (u n |Γ9 ),
φ+ (u) = φ+ (u n |Γ10 ),
φ− (u) = φ− (u n |Γ11 ) ∀u ∈ K (Ω).
Define a functional Φ : V → R by

φτ (u) + φn (u) + φ+ (u) + φ− (u) ∀u ∈ K (Ω),
Φ(u) = (5.22)
+∞ ∀u ∈
/ K (Ω).
Then, Φ is proper, convex and weakly lower semi-continuous (cf. Example 1.2).
Note Φ ≥ 0 since u n |Γ10 ≥ 0 and u n |Γ11 ≤ 0 for u ∈ K (Ω).
By Assumption 5.1, wτ = vτ on Γ8 and wn = vn on ∪i=9 11
Γi . Taking into account
the fact that gτ |vτ | + στ · vτ = 0 and |στ | ≤ gτ on Γ8 , we have that
φτ (u) − φτ (w) + (στ , u τ )Γ8 − (στ , wτ )Γ8

= (gτ |u τ | + στ · u τ ) d x − (gτ |wτ | + στ · wτ ) d x
Γ8 Γ8

= (gτ |u τ | + στ · u τ ) d x − (gτ |vτ | + στ · vτ ) d x ≥ 0 ∀u ∈ K (Ω).
Γ8 Γ8
Taking into account the fact that gn |vn | + σn · vn = 0 and |σn | ≤ gn on Γ9 , in the
same way we have
φn (u) − φn (w) + (σn , u n )Γ9 − (σn , wn )Γ9 ≥ 0. (5.23)
Also,
φ+ (u) − φ+ (w) + σ+n , u n Γ10 − σ+n , wn Γ10
(5.24)
= g+n + σ+n , u n Γ10 − g+n + σ+n , wn Γ10 ≥ 0,
where the facts that u n ≥ 0, σ+n + g+n ≥ 0 and σ+n + g+n , vn Γ10 = 0, wn = vn
on Γ10 were used. In the same way, we have
φ− (u) − φ− (w) + σ−n , u n Γ11 − σ−n , wn Γ11 ≥ 0. (5.25)
By virtue of (5.22)–(5.25), we have
Φ(u) − Φ(w) ≥ − (στ , u τ − wτ )Γ8 − (σn , u n − wn )Γ9

(5.26)
− σ+n , u n − wn Γ10 − σ−n , u n − wn Γ11 ∀u ∈ V.
Therefore, we get from (5.20) and (5.26)
a01 (w, u − w) + a11 (w, w, u − w) + Φ(u) − Φ(w) ≥ F1 , u − w ∀u ∈ V.

(5.27)
Thus, we come to the following formulation associated with Problem I (the case of
static pressure) by a variational inequality.
Problem I-VI Find v = w + U such that
a01 (w, u − w) + a11 (w, w, u − w) + Φ(u) − Φ(w) ≥ F1 , u − w ∀u ∈ V,

(5.28)
where a01 , a11 , F1 are the same as in (5.9), U is the same as in Assumption 5.1 and
Φ is as in (5.22).
To prove equivalence of Problem I-VI and Problem I-VE we need first to establish
the following result:
Lemma 5.1 For ψ ∈ C0∞ (Γi ), i = 10, 11, there exists a function u ∈ V such that
u n |Γi = ψ, uV ≤ Ci ψ H 1/2 (Γi ) ,
where Ci are independent of ψ.
Proof By (3) of Assumption 5.1 if Γ10 ∪ Γ11 = ∅, then, for example, Γ2 = ∅ and
there exists a diffeomorphism y = f i (x) ∈ C 1 from Γi onto Γ2 . Define ϕ(y) at point
y ∈ Γ2 corresponding to point x ∈ Γi by ϕ(y) = D f1(x) ψ( f i−1 (y)), where D f i (x) is
Jacobian of the transformation f i . Then,

1
ϕ(y) dy = ψ( f i−1 (y))D f i (x) d x = ψ(x) d x, (5.29)
Γ2 Γi D f i (x) Γi
and

1
ϕ(y) H 21 (Γ ) ≤
D f (x) ψ(x) H 21 (Γ ) ≤ ci ψ(x) H 21 (Γ ) . (5.30)
2 i i
i C(Γi )
When ψ ∈ C0∞ (Γi ), define a function φ ∈ H1/2 (∂Ω) on ∂Ω as follows:
φ × n|Γ2 ∪Γi = 0, φ n |Γ2 = −ϕ, φ n |Γ10 = ψ, φ|(∪i=1,3−9,11 Γi ) = 0.

Thus, by (5.29) ∂Ω φ n ds = 0. Then, there exists a solution u ∈ W1,2 (Ω) to the
Stokes problem ⎧
⎨ − Δu + ∇ p = 0,
⎪
div u = 0,
⎪
⎩
u|∂Ω = φ
and
uV ≤ cφH1/2 (∂Ω) .
(cf. Theorem IV.1.1 of [1]). Taking into account (5.30), we come to the asserted
estimation with Ci = 1 + ci . Thus u is the asserted function.
Problem I-VE and Problem I-VI are equivalent in the following sense:
Theorem 5.2 If (v, στ , σn , σ+n , σ−n ) is a solution to Problem I-VE, then v is a

solution to Problem I-VI.
Conversely, if v is a solution to Problem I-VI, then there exist στ , σn , σ+n , σ−n

such that (v, στ , σn , σ+n , σ−n ) is a solution to Problem I-VE.
Proof We already showed that if (v, στ , σn , σ+n , σ−n ) is a solution to Problem I-VE,
then v is a solution to Problem I-VI. Thus, it remains to prove the second assertion
of theorem.
Since the functional Φ is proper, we have from (5.28)
v − U = w ∈ K (Ω) (5.31)
because if w ∈ / K (Ω), then the left hand side of (5.28) is −∞, which is a contra-
diction to the fact that the right-hand side is finite.
Let ψ ∈ V8−11 ≡ {u ∈ V : u|Γ8 ∪Γ9 ∪Γ10 ∪Γ11 = 0} (⊂ K (Ω)). Putting u = w + ψ,
u = w − ψ and taking into account
φτ (w) = φτ (w + ψ),
φn (w) = φn (w + ψ),
φ+ (w) = φ+ (w + ψ),
φ− (w) = φ− (w + ψ),
we get from (5.22) and (5.28)
a01 (w, ψ) + a11 (w, w, ψ) ≥ F1 , ψ,

a01 (w, −ψ) + a11 (w, w, −ψ) ≥ F1 , −ψ ∀ψ ∈ V8−11 ,
which imply
a01 (w, ψ) + a11 (w, w, ψ) = F1 , ψ ∀ψ ∈ V8−11 . (5.32)
When u ∈ V10−11 ≡ {u ∈ V : u|Γ10 ∪Γ11 = 0} (⊂ K (Ω)), the set {(u|Γ8 , u n |Γ9 )} is

a subspace of L2τ (Γ8 ) × L 2 (Γ9 ), where u n |Γ9 is u|Γ9 · n.
Define a functional σ ∗ on the set by

σ ∗ , (u|Γ8 , u n |Γ9 ) = a01 (w, u) + a11 (w, w, u) − F1 , u ∀u ∈ V10−11 . (5.33)
This functional is well defined. Because if u, u 1 ∈ V10−11 are such that (u|Γ8 , u|Γ9 ) =
(u 1 |Γ8 , u 1 |Γ9 ), then since u − u 1 ∈ V8−11 , by (5.32)
a01 (w, u − u 1 ) + a11 (w, w, u − u 1 ) − F1 , u − u 1 = 0,
that is,
a01 (w, u) + a11 (w, w, u) − F1 , u = a01 (w, u 1 ) + a11 (w, w, u 1 ) − F1 , u 1 ,
and so by (5.33) ∗
σ , (u|Γ8 , u n |Γ9 ) = σ ∗ , (u 1 |Γ8 , u 1n |Γ9 ) .
This function is linear.

Putting u = w + ψ, where ψ ∈ V10−11 , and taking into account
φ+ (w + ψ) = φ+ (w),
φ− (w + ψ) = φ− (w),
we have from (5.33) and (5.28)

− σ ∗ , (ψ|Γ8 , ψn |Γ9 ) = − [a01 (w, ψ) + a11 (w, w, ψ) − F1 , ψ]
≤ Φ(w + ψ) − Φ(w)
= φτ (w + ψ) − φτ (w) + φn (w + ψ) − φn (w) (5.34)

≤ gτ |ψ|Γ8 d x + gn |ψ|Γ9 d x ∀ψ ∈ V10−11 .
Γ8 Γ9
Putting u = w − ψ, in the same way we have

∗
σ , (ψ|Γ8 , ψn |Γ9 ) = [a01 (w, ψ) + a11 (w, w, ψ) − F1 , ψ]
≤ φτ (w − ψ) − φτ (w) + φn (w − ψ) − φn (w)
(5.35)
≤ gτ |ψ|Γ8 d x + gn |ψ|Γ9 d x ∀ψ ∈ V10−11 .
Γ8 Γ9
By (5.34) and (5.35), we can see that σ ∗ is a bounded linear functional with a norm
not greater than 1 on a subspace of L1gτ (Γ8 ) × L 1gn (Γ9 ), where L1gτ (Γ8 ), L 1gn (Γ9 )
are, respectively, the spaces of functions integrable with weights gτ , gn on Γ8 and
Γ9 . By the Hahn-Banach theorem the functional can be extended as a functional
on L1gτ (Γ8 ) × L 1gn (Γ9 ) norms of which is not greater than 1. Therefore, there exist
elements στ ∈ L∞1 (Γ8 ) with στ L∞1 (Γ8 ) ≤ 1 and σn ∈ L ∞1 (Γ9 ) with σn L ∞1 (Γ9 ) ≤
gτ gτ gn gn
1, which imply
|στ | ≤ gτ , |σn | ≤ gn (5.36)
and

σ ∗ , (u|Γ8 , u n |Γ9 ) = στ , u|Γ8 Γ8
+ σn , u n |Γ9 Γ9
∀u ∈ V10−11 . (5.37)
1 1
When u ∈ V, the set {(u n |Γ10 , u n |Γ11 )} is a subspace of H 2 (Γ10 ) × H 2 (Γ11 ).
Define a functional σ1∗ on the set V by

σ1∗ , (u n |Γ10 , u n |Γ11 ) =
a01 (w, u) + a11 (w, w, u) − στ , u|Γ8 Γ8
− σn , u|Γ9
− F1 , u ∀u ∈ V.
Γ9
(5.38)
This functional is also well defined. Because if u, u 1 ∈ V are such that (u Γ10 , u|Γ11 ) =
(u 1 |Γ10 , u 1 |Γ11 ), then since u − u 1 ∈ V10−11 (Ω), by (5.33) and (5.37)

a01 (w, u − u 1 ) + a11 (w, w, u − u 1 ) − στ , (u − u 1 )|Γ8
Γ8

− σn , (u − u 1 )|Γ9 − F1 , u − u 1
Γ9

∗
= σ , ((u − u )|Γ8 , (u − u 1 )|Γ9 ) − στ , (u − u 1 )|Γ8
1
− σn , (u − u 1 )|Γ9 = 0,
Γ8 Γ9
and so by (5.38)
∗
σ1 , (u n |Γ10 , u n |Γ11 ) = σ1∗ , (u 1n |Γ10 , u 1n |Γ11 ) .
The functional σ1∗ is linear. Let us prove next its continuity.

Let u be the function corresponding to ψ ∈ C0∞ (Γ10 ) by Lemma 5.1. Then, by
Lemma 5.1 we have from (5.38)

| σ1∗ , (ψ, 0) |

≤ C wV uV + w2V uV + (στ L 2 (Γ8 ) + στ L 2 (Γ9 ) )u + F1 V∗ uV

≤ C wV + w2V + (στ L 2 (Γ8 ) + στ L 2 (Γ9 ) ) + F1 V∗ · ψ H 21 (Γ ) .
10
(5.39)
Also assuming that u is the function corresponding to ψ ∈ C0∞ (Γ11 ) by Lemma 5.1,
we have

| σ1∗ , (0, ψ) |

≤ C wV uV + w2V uV + (στ L 2 (Γ8 ) + στ L 2 (Γ9 ) )u + F1 V∗ uV

≤ C wV + w2V + (στ L 2 (Γ8 ) + στ L 2 (Γ9 ) ) + F1 V∗ · ψ H 21 (Γ ) .
11
(5.40)
1/2 1
Since H0 (Γi ) = H 2 (Γi ), i = 10, 11, (cf. (1.11)), (5.39) and (5.40) show that the
functional σ1∗ is continuous on the subspace of H 2 (Γ10 ) × H 2 (Γ11 ) mentioned
1 1
above. Thus, by the Hahn-Banach theorem the functional is extended as a functional

1 1
on H 2 (Γ10 ) × H 2 (Γ11 ).
Therefore, there exists an element (σ+n , σ−n ) ∈ H −1/2 (Γ10 ) × H −1/2 (Γ11 ) such
that
∗
σ1 , (u|Γ10 , u|Γ11 ) = σ+n , u|Γ10 Γ10 + σ−n , u|Γ11 Γ11 ∀u ∈ V. (5.41)
When ψ ≥ 0 is such that ψ ∈ C0∞ (Γ10 ), let u ∈ K (Ω) be the function asserted
in Lemma 5.1. Putting u = w + u, by (5.28) we have
a01 (w, u) + a11 (w, w, u) + Φ(w + u) − Φ(w) − F1 , u ≥ 0. (5.42)
On the other hand, by (5.38), (5.41) and property of u,
a01 (w, u) + a11 (w, w, u) − F1 , u = σ+n , ψΓ10
and so from (5.42) we have that
σ+n , ψΓ10 + Φ(w + u) − Φ(w) ≥ 0. (5.43)
By (5.21), (5.22) and the property of u, we get
Φ(w + u) − Φ(w) = g+n , ψΓ10 ,
which combined with (5.43) gives
σ+n , ψΓ10 + (g+n , ψ)Γ10 ≥ 0,
that is,
σ+n + g+n ≥ 0. (5.44)
When ψ ≤ 0 is such that ψ ∈ C0∞ (Γ11 ), let u ∈ K (Ω) be the function asserted
in Lemma 5.1. Then, in the same way we have that
σ−n , −ψΓ11 − (g−n , −ψ)Γ11 ≥ 0,
that is,
σ−n − g−n ≤ 0. (5.45)
From (5.38) and (5.41), we have
a01 (w, u) + a11 (w, w, u) − (στ , u τ )Γ8 − (σn , u)Γ9 − σ+n , uΓ10 − σ−n , uΓ11
= F1 , u ∀u ∈ V.
(5.46)
Putting u = 0 in (5.28) and taking into account (5.46) with u = w, we have
(στ , w)Γ8 + (σn , w)Γ9 + σ+n , wn Γ10 + σ−n , wn Γ11

+ φτ (w) + φn (w) + φ+ (w) + φ− (w) ≤ 0,
that is,

(στ wτ + gτ |wτ |) ds + (σn wn + gn |wn |) ds
Γ8 Γ9 (5.47)
+ σ+n + g+n , wn Γ10 + σ−n − g−n , wn Γ11 ≤ 0.
Since on Γ8 , Γ9 , Γ10 and Γ11 , respectively, wτ = vτ , wn = vn , wn = vn ≥ 0 and

wn = vn ≤ 0, taking into account (5.36), (5.44), (5.45), by (5.47) we have
στ vτ + gτ |vτ | = 0,
σn vn + gn |vn | = 0,
(5.48)
σ+n + g+n , vn = 0,
σ−n − g−n , vn = 0.
Therefore, by virtue of (5.31), (5.36), (5.44)–(5.46) and (5.48), we come to the

conclusion.
5.1.2 Variational Formulation: The Case of Total Pressure
Taking (v · ∇)v = rot v × v + 21 grad|v|2 into account and putting v = w + U , by

(5.6), (5.7) and Assumption 5.1 we can see that solutions (v, p) of the problem (5.1),
(5.3) in the sense of Definition 5.1 satisfy the following:
⎧
⎪ v − U = w ∈ K (Ω),
⎪
⎪
⎪
⎪ 2μ(E(w), E(u)) + rot w × w, u + rot U × w, u
⎪
⎪
⎪
⎪
⎪
⎪ + rot w × U, u + 2μ(k(x)w, u)Γ2 + 2μ(S w̃, ũ)Γ3
⎪
⎪
⎪
⎪ + 2(α(x)w, u)Γ5 + μ(k(x)w, u)Γ7 − 2μ(εnτ (w + U ), u)Γ8
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪ + p + |v|2 − 2μεnn (w + U ), u n Γ9 ∪Γ10 ∪Γ11
⎪
⎪ 2
⎪
⎪
⎪
⎪ = −2μ(E(U ), E(u)) − rot U × U, u − 2μ(k(x)U, u)Γ2
⎪
⎨
− 2μ(SŨ , ũ)Γ3 − 2(α(x)Uτ , u)Γ5 − μ(k(x)U, u)Γ7 (5.49)
⎪
⎪
⎪
⎪ + f, u + φi , u n Γi + φi , uΓi
⎪
⎪
⎪
⎪
⎪
⎪ i=2,4,7 i=3,5,6
⎪
⎪
⎪
⎪ ∀u ∈ V,
⎪
⎪
⎪
⎪ |σ t
(v)| ≤ , σ t
· v + |v | = Γ ,
⎪
⎪ τ g τ τ τ gτ τ 0 on 8
⎪
⎪
⎪
⎪ |σ t
(v)| ≤ g , σ t
(v)v + g |v | = 0 on Γ 9,
⎪
⎪
n n n n n n
⎪ σ t (v) + g ≥ 0, (σ t (v) + g )v = 0 on Γ ,
⎪
⎪
⎪ n +n n +n n 10
⎩ t
t
Define a02 (·, ·), a12 (·, ·, ·) and F2 ∈ V∗ , respectively, by

a02 (w, u) = 2μ(E(w), E(u)) + rot U × w, u + rot w × U, u

+ μ(k(x)w, u)Γ7 ∀w, u ∈ V,
a12 (w, u, v) = rot w × u, v ∀w, u, v ∈ V,
(5.50)
F2 , u = −2μ(E(U ), E(u)) − rot U × U, u − 2μ(k(x)U, u)Γ2
− 2μ(SŨ , ũ)Γ3 − 2(α(x)Uτ , u)Γ5 − μ(k(x)U, u)Γ7
+ f, u + φi , u n Γi + φi , uΓi ∀u ∈ V.
i=2,4,7 i=3,5,6
στt (v) = 2μεnτ (v),

1
σnt (v) = −( p + |v|2 ) + 2μεnn (v)
2
and (5.49), we introduce the following variational formulation for the problem (5.1),
(5.3).
Problem II-VE. Find (v, στt , σnt , σ+n

t
, σ−n
t
) ∈ U + K (Ω) × L2τ (Γ8 ) × L 2 (Γ9 ) ×
− 21 − 21
H (Γ10 ) × H (Γ11 ) such that
⎧
⎪
⎪ v − U = w ∈ K (Ω),
⎪
⎪
⎪
⎪ a02 (w, u) + a12 (w, w, u) − (στt , u τ )Γ8 − (σnt , u n )Γ9
⎪
⎪ t t
⎪
⎪
⎪
⎪ − σ+n , u n Γ10 − σ−n , u n Γ11 = F2 , u ∀u ∈ V,
⎨
|στt | ≤ gτ , στt · vτ + gτ |vτ | = 0 on Γ8 , (5.51)
⎪
⎪
⎪ |σnt | ≤ gn , σnt vn + gn |vn | = 0 on Γ9 ,
⎪
⎪
⎪ t
⎪
⎪
⎪
⎪ σ+n
t
+ g+n ≥ 0, σ+n + g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎪ t
⎩ t
σ−n − g−n ≤ 0, σ−n − g−n , vn Γ11 = 0 on Γ11 .
In the same way as in Theorem 5.1 we have

tion 5.1 of the problem (5.1), (5.3), then (v, στt |Γ8 , σnt |Γ9 , σnt |Γ10 , σnt |Γ11 ) is a solution
to Problem II-VE.
Conversely, if f ∈ L2 (Ω) and Problem II-VE has a smooth solution (v, στt , σnt ,
σ+ , σ−t ) such that v ∈ H2 (Ω), σ+t ∈ L 2 (Γ10 ) and σ−t ∈ L 2 (Γ11 ), then there exists
t
p ∈ H 1 (Ω) such that (v, p) is a solution to the problem (5.1), (5.3). Moreover, if at
least one of the sets Γi , i = 2, 4, 6, 7, is nonempty, then p is unique.
Then, in the same way as in Problem I we get Problem II-VI (the case of total
pressure) formulated by a variational inequality and can prove that the problem is
equivalent to Problem II-VE.
Problem II-VI. Find v = w + U such that
a02 (w, u − w) + a12 (w, w, u − w) + Φ(u) − Φ(w) ≥ F2 , u − w ∀u ∈ V,

(5.52)
where a02 , a12 , F2 are in (5.50) and Φ is defined by (5.22).
Theorem 5.4 If (v, στt , σnt , σ+n t
, σ−n
t
) is a solution to Problem II-VE, then v is a
solution to Problem II-VI.
Conversely, if v is a solution to Problem II-VI, then there exist στt , σnt , σ+n
t
, σ−n
t
such that (v, στ , σn , σ+n , σ−n ) is a solution to Problem II-VE.

t t t t
5.1.3 Variational Formulation: The Stokes Problem
In the same way as in Problem I we get the following equivalent formulations of

Problem III for the Stokes equations with boundary condition (5.2).
Problem III-VE. Find (v, στ , σn , σ+n , σ−n ) ∈ U + K (Ω) × L2τ (Γ8 ) × L 2 (Γ9 ) ×
H − 2 (Γ10 ) × H − 2 (Γ11 ) such that
1 1
⎧
⎪
⎪ v − U = w ∈ K (Ω),
⎪
⎪
⎪
⎪ a03 (w, u) − (στ , u τ )Γ8 − (σn , u n )Γ9 − σ+n , u n Γ10 − σ−n , u n Γ11 = F3 , u
⎪
⎪
⎪
⎪
⎪
⎨ ∀u ∈ V,
|στ | ≤ gτ , στ · vτ + gτ |vτ | = 0 on Γ8 ,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ σ+n + g+n ≥ 0, σ+n + g+n , vn = 0 on Γ10 ,
⎪
⎩
σ−n − g−n ≤ 0, σ−n − g−n , vn = 0 on Γ11 ,
(5.53)
where
a03 (w, u) = 2μ(E(w), E(u)) + 2μ(k(x)w, u)Γ2

+ 2μ(S w̃, ũ)Γ3 + 2(α(x)w, u)Γ5 + μ(k(x)w, u)Γ7 ∀w, u ∈ V,
F3 , u = −2μ(E(U ), E(u)) − 2μ(k(x)U, u)Γ2 − 2μ(SŨ , ũ)Γ3 − 2(α(x)Uτ , u)Γ5
− μ(k(x)U, u)Γ7 + f, u + φi , u n Γi + φi , uΓi ∀u ∈ V.
i=2,4,7 i=3,5,6
(5.54)
Problem III-VI Find v such that
v − U = w ∈ K (Ω),
(5.55)
a03 (w, u − w) + Φ(u) − Φ(w) ≥ F3 , u − w ∀u ∈ V,
where the functionals Φ is defined by (5.22).

Remark 5.2 Using the strain bilinear form and applying Theorems 3.1 and 3.2 on
Γi j , i = 2, 3, 7, we embedded all the boundary conditions into variational formula-
tions of problems. For this, we used condition Γi j ∈ C 2 , i = 2, 3, 7. The condition
Γi j ∈ C 1 is for the existence of a diffeomorphism in C 1 between Γi and Γ j in (3) of
Assumption 5.1.
5.2 Solutions to Variational Inequalities
In this section we study some variational inequalities obtained in Sect. 5.1. We denote
by X → → Y compact imbedding of a space X into Y .
Theorem 5.5 Let X, X 1 be real separable Hilbert spaces such that X → → X 1 ,
and X ∗ be dual space of X . Assume the followings:
(1) Φ : X → [0, +∞] is a proper lower semi-continuous convex functional such
that Φ(0 X ) = 0.
(2) a0 (·, ·) ∈ (X × X → R) is a bilinear form such that
|a0 (u, v)| ≤ K u X v X ∀u, v ∈ X,

a0 (u, u) ≥ αu2X ∃α > 0, ∀u ∈ X.
(3) a1 (·, ·, ·) ∈ (X 1 × X × X → R) is a triple linear functional such that
a1 (w, u, u) = 0 ∀w ∈ X 1 , ∀u ∈ X,
|a1 (w, u, v)| ≤ K w X 1 u X v X , ∀w ∈ X 1 , ∀u, v ∈ X.
Then for each f ∈ X ∗ there exists a solution to the variational inequality
a0 (v, u − v) + a1 (v, v, u − v) + Φ(u) − Φ(v) ≥ f, u − v ∀u ∈ X (5.56)
and all solutions v satisfy the estimate
1
v X ≤ f X ∗ . (5.57)
α
If moreover
Kc
f X ∗ < 1, (5.58)
α2
then solution is unique, where c is the constant in · X 1 ≤ c · X .
Proof Fix w ∈ X 1 . Let us consider the following variational inequality:
a0 (v, u − v) + a1 (w, v, u − v) + Φ(u) − Φ(v) ≥ f, u − v ∀u ∈ X, (5.59)

where f ∈ X ∗ . From existing results there exists a unique solution to (5.59) (see
Theorem 10.5 of [2]). Let v1 , v2 be the solutions corresponding to f 1 , f 2 instead of
f . Then, under consideration of condition (2) it is easy to verify that
1
v1 − v2 X ≤ f1 − f2 X ∗ . (5.60)
α
Now, let us consider the operator which maps w to the solution v of (5.59)
T ∈ (X 1 → X ) : w → T (w) = v.
Taking into account condition (1), we can easily verify that the solution corresponding
to f = 0 X ∗ is 0 X . Thus, we have from (5.60)
1
v X ≤ f X ∗ ∀w ∈ X 1 . (5.61)
α
Note that this estimate is independent of w.
Denote by v1 and v2 , respectively, the solutions to (5.59) corresponding to w1 and
w2 . Then
a0 (v1 , u − v1 ) + a1 (w1 , v1 , u − v1 ) + Φ(u) − Φ(v1 ) ≥ f, u − v1 ∀u ∈ X,

a0 (v2 , u − v2 ) + a1 (w2 , v2 , u − v2 ) + Φ(u) − Φ(v2 ) ≥ f, u − v2 ∀u ∈ X.
(5.62)
Putting u = v2 and u = v1 , respectively, in the first formula and the second one of
(5.62), and adding two formulas, we get
a0 (v1 − v2 , v2 − v1 ) + a1 (w1 , v1 , v2 − v1 ) + a1 (w2 , v2 , v1 − v2 ) ≥ 0. (5.63)
From (5.63), conditions (2), (3) of Theorem and (5.61), we get
1
v2 − v1 2X ≤ |a1 (w1 , v1 , v2 − v1 ) − a1 (w2 , v1 , v2 − v1 )
α
+ a1 (w2 , v1 , v2 − v1 ) − a1 (w2 , v2 , v2 − v1 )|
1 1
≤ |a1 (w1 − w2 , v1 , v2 − v1 )| + |a1 (w2 , v2 − v1 , v2 − v1 )|
α α
K
≤ w1 − w2 X 1 v1 X v2 − v1 X
α
K f X ∗
≤ w1 − w2 X 1 v2 − v1 X ∀w1 , w2 ∈ X 1 ,
α2
which implies
K f X ∗
v2 − v1 X ≤ w1 − w2 X 1 ∀w1 , w2 ∈ X 1 . (5.64)
α2
5.2 Solutions to Variational Inequalities 169
By (5.61), (5.64) and Schauder fixed-point theorem (Theorem 1.6), there exists a
solution to (5.56). And any solution is a fixed point of operator T, and by (5.61) all
solutions satisfy the estimate (5.57).
If (5.58) holds, then the operator T : w ∈ X → v ∈ X is contract, and so we come
to the last conclusion.
Let us next study variational inequalities when condition (3) of the above
theorem is weakened.
Theorem 5.6 Let X be a real separable Hilbert space. Assume the followings:
(1) Condition (1) of Theorem 5.5 holds.
(2) Condition (2) of Theorem 5.5 holds.
(3) a1 (·, ·, ·) ∈ (X × X × X → R) is a triple linear functional such that
|a1 (w, u, v)| ≤ K w X u X v X , ∀w, u, v ∈ X.
If f is small enough, then in O M (0 X ), where M is determined in (5.73), there

exists a unique solution to the variational inequality
a0 (v, u − v) + a1 (v, v, u − v) + Φ(u) − Φ(v) ≥ f, u − v ∀u ∈ X. (5.65)
Proof Fix w ∈ X . Let us consider a variational inequality
a0 (v, u − v) + a1 (w, w, u − v) + Φ(u) − Φ(v) ≥ f, u − v ∀u ∈ X, (5.66)
where f ∈ X ∗ . Defining an element a1 (w) ∈ X ∗ by
a1 (w), u = a1 (w, w, u) ∀u ∈ X,
by condition (3) we have
a1 (w) X ∗ ≤ K w2X ∀w ∈ X. (5.67)
Then, (5.66) can be rewritten as follows:
a0 (v, u − v) + Φ(u) − Φ(v) ≥ f − a1 (w), u − v ∀u ∈ X. (5.68)
By the same argument as in the proof of Theorem 5.5, there exists a unique
solution vw to (5.68) and
1 1
vw ≤ ( f X ∗ + a1 (w) X ∗ ) ≤ ( f X ∗ + K w2X ), (5.69)
α α
where (5.67) was used and α is the one of Theorem 5.5.
Now, let us consider the operator which maps w to the solution of (5.68)
T ∈ (X → X ) : w → T (w) = v.
Denote by v1 and v2 , respectively, the solutions to (5.66) corresponding to w1 , w2 ∈

O M (0 X ), where M is to be determined below. Then
1
v1 − v2 X ≤ a1 (w1 ) − a1 (w2 ) X ∗ . (5.70)
α
By condition (3)
a1 (w1 ) − a1 (w2 ) X ∗ ≤ K (w2 − w1 X w2 X + w1 X w1 − w2 X ) .

(5.71)
Thus, by (5.70) and (5.71),
K
v1 − v2 X ≤ (w2 − w1 X w2 X + w1 X w1 − w2 X )
α (5.72)
2K M
≤ w2 − w1 X ∀w1 , w2 ∈ O M (0 X ).
α
Therefore, if M is taken such that
⎧ 1
⎪
⎨ M = ( f X ∗ + K M 2 ),
α (5.73)
⎪
⎩ 2K M < 1
α
(If α is large and f X ∗ is small enough, then such a choice is possible.), then by
(5.69) and (5.72) the operator T on O M (0 X ) is contract, and so there exists a unique
solution to (5.65).
Theorem 5.7 Let X be a real separable Hilbert space and X ∗ be its dual space.
Assume that
(1) Φ : X → R is a finite weak continuous convex functional, Φε : X → R is
convex such that
Φε (v) → Φ(v) uniformly on X as ε → 0,
Gâteaux derivative DΦε ≡ Aε ∈ (X → X ∗ ) is weak continuousand Aε (0 X ) =

0 X [∗ .
(2) a(·, ·, ·) ∈ (X × X × X → R) is a form such that
when w ∈ X, (u, v) → a(w; u, v) is bilinear on X × X,

a(v, v, v) ≥ αv2X ∃α > 0, ∀v ∈ X and
when vm v in X,
a(vm , vm , u) → a(v, v, u) ∀u ∈ X and lim inf a(vm , vm , vm ) ≥ a(v, v, v).
m→∞
Then, for f ∈ X ∗ there exists a solution to a variational inequality
a(v, v, u − v) + Φ(u) − Φ(v) ≥ f, u − v ∀u ∈ X (5.74)
satisfying the estimate

1
v X ≤ f X ∗ . (5.75)
α
Proof First let us prove the existence of a solution to a variational equation
a(v, v, u) + Aε (v), u = f, u ∀u ∈ X. (5.76)
We will rely on Theorem 1.43. Let {wn } be a basis of X and denote by X m the
subspace of X spannedby w1 , · · · , wm .
m
We will find vm = i=1 μi wi ∈ X m satisfying
a(vm , vm , u) + Aε (vm ), u = f, u ∀u ∈ X m . (5.77)
Define Fm ∈ (X m → X m ) by
(Fm (v), wi ) = a(v, v, wi ) + Aε (v), wi − f, wi , 1 ≤ i ≤ m. (5.78)
Since Gâteaux derivative of finite convex functional is monotone (see Remark 1.16)
and Aε (0 X ) = 0 X ∗ , we get
Aε (u) − Aε (0 X ), u − 0 X = Aε (u), u ≥ 0 ∀u ∈ X.
Thus,
a(u, u, u) + Aε (u), u ≥ αu2X ∀u ∈ X. (5.79)
Combining (5.78) with (5.79) yields
(Fm (v), v) ≥ (αv X − f X ∗ )v X ∀v ∈ X m . (5.80)
Therefore,
f X ∗
(Fm (v), v) ≥ 0 ∀v ∈ X with v X = .
α
And by virtue of assumption (2), Fm is continuous in X m . Thus, there exists a solution
vεm to problem (5.77). By (5.80), we have that
0 = (Fm (vεm ), vεm ) ≥ (αvεm X − f X ∗ )vεm X
for all solution vεm to (5.77), which implies

1
vεm X ≤ f X ∗ . (5.81)
α
Note this estimation is independent of ε, m. Thus, from {vεm } we can extract a
subsequence {vεm p } such that
vεm p vε in X as p → +∞.
By the assumptions of theorem,
a(vεm p , vεm p , u) + Aε (vεm p ), u → a(vε , vε , u) + Aε (vε ), u ∀u ∈ X. (5.82)
By virtue of (5.77), (5.82) and (5.81), we see that vε is a solution to (5.76) and
satisfies
1
vε X ≤ f X ∗ . (5.83)
α
Subtracting the following two formulas which are got from (5.76)
a(vε , vε , u) + Aε (vε ), u = f, u ∀u ∈ X,

a(vε , vε , vε ) + Aε (vε ), vε = f, vε
and taking into account that
Φε (u) − Φε (vε ) ≥ Aε (vε ), u − vε
which is due to convexity of Φε , we come to the following inequality:
a(vε , vε , u − vε ) + Φε (u) − Φε (vε ) ≥ f, u − vε ∀u ∈ X. (5.84)
By (5.83) we can choose {vεk } such that
vεk v ∗ in X as εk → 0. (5.85)
By virtue of assumption (1),
|Φεk (vεk ) − Φ(v ∗ )| ≤ |Φεk (vεk ) − Φ(vεk )| + |Φ(vεk ) − Φ(v ∗ )| → 0 as εk → 0,
and so
Φεk (vεk ) → Φ(v ∗ ) as εk → 0. (5.86)
Also
Φεk (u) → Φ(u) ∀u ∈ X as εk → 0. (5.87)
By virtue of assumption (2),

a(vεk , vεk , u) → a(v ∗ , v ∗ , u) ∀u ∈ X as εk → 0,

(5.88)
lim inf a(vεk , vεk , vεk ) ≥ a(v ∗ , v ∗ , v ∗ ).
k→∞
Taking into account (5.86)–(5.88), we get from (5.84)
a(v ∗ , v ∗ , u − v ∗ ) + Φ(u) − Φ(v ∗ ) ≥ f, u − v ∗ ∀u ∈ X.
By (5.83) we have
1
v ∗ X ≤ f X ∗ . (5.89)
α
Therefore, we finish our proof.
Remark 5.3 The estimate of solutions in Theorem 5.5 is for all solutions of the
problem, but the one in Theorem 5.7 is for the solution whose existence is guaranteed
by the theorem.
5.3 Existence and Uniqueness of Solutions to the Steady

Navier-Stokes Problems
In this section, on the basis of the results in Sect. 5.2, we study the problems formu-
lated in Sect. 5.1. Our first result is the following:
Theorem 5.8 Let Assumption 5.1 hold, the surfaces Γ2 j , Γ3 j , Γ7 j be convex (see
Definition 3.2), α positive and U H1 (Ω) small enough. Then, when f and φi , i =
2, · · · , 7, are small enough, in a neighborhood of U in H1 (Ω) there exists a unique
solution to Problem I-VI (for the steady Navier-Stokes problem of the case of static
pressure).
Proof Define a functional Φ by (5.22). Trace operator is continuous and sum of con-
vex functions is also convex. Thus, the functional satisfies condition (1) of Theorem
5.6.
Set w = v − U , where U is the function in Assumption 5.1. Let a01 (·, ·), a11 (·, ·, ·)
and F1 ∈ V∗ be the same as in (5.9).
2μ(E(w), E(w)) ≥ δw2V , δ > 0. (5.90)
On the other hand, applying Hölder’s inequality, for w ∈ V we have
|(U · ∇)w, w + (w · ∇)U, w| ≤ γ w2V · U H1 (Ω) . (5.91)
Therefore, if δ − γ U H1 (Ω) = β1 > 0, then by (5.90), (5.91), Assumption 5.1

and Lemma 3.1 we have
a01 (u, u) ≥ β1 u2V ∀u ∈ V. (5.92)
It is easy to verify that
|a01 (u, v)| ≤ cuV vV ∀u, v ∈ V. (5.93)
By (5.92) and (5.93), a0 (u, v) satisfies condition (2) of Theorem 5.6.

By Hölder’s inequality we can see
|a11 (w, u, v)| ≤ cwV uV vV ∀w, u, v ∈ V. (5.94)
which means that a11 (w, u, v) satisfies condition (3) of Theorem 5.6.
Also

F1 V∗ ≤ M1 U H1 + U 2H1 + f V∗ + φi − 21 + φi − 21 ,
H (Γi ) H (Γi )
i=2,4,7 i=3,5,6
(5.95)
where M1 depends on mean curvature of Γ7 , shape operator of Γ3 , μ and α.
By Theorem 5.6, if U H1 , f V∗ , φi H − 21 (Γ ) , i = 2, 4, 7, and φi H− 21 (Γ ) , i =
i i
3, 5, 6, are small enough, then there exists a unique solution w ∈ K (Ω) to
a01 (w, u − w) + a11 (w, w, u − w) + Φ(u) − Φ(w) ≥ F1 , u − w ∀u ∈ K (Ω). (5.96)
Since v = w + U is solution, we come to the asserted conclusion.
Theorem 5.9 Let Assumption 5.1 hold, the surfaces Γ2 j , Γ3 j , Γ7 j be convex, α

positive and U H1 (Ω) small enough. Then, for any f φi , i = 2 ∼ 7, there exists a
solution v to Problem II-VI (the steady Navier-Stokes problem for the case of total
pressure) in a neighborhood of U in H1 (Ω) and all solutions satisfy
M1
v − U H1 ≤ U H1 + U 2H1 + f V∗
δ − γ U H1
(5.97)
+ φi H − 21 (Γ ) + φi H− 21 (Γ ) ,
i i
i=2,4,7 i=3,5,6
where δ, γ , M1 are as in (5.98), (5.99), (5.107).

If U H1 , f V∗ , φi H − 21 (Γ ) , i = 2, 4, 7, and φi H− 21 (Γ ) , i = 3, 5, 6, are small
i i
Proof Define a functional Φ(u) by (5.22). Then, Φ satisfies condition (1) of Theorem
5.5.
Let a02 (·, ·), a12 (·, ·, ·) and F2 ∈ V∗ be as in (5.50).

5.3 Existence and Uniqueness of Solutions to the Steady Navier-Stokes Problems 175
2μ(E(w), E(w)) ≥ δw2V , δ > 0. (5.98)
On the other hand, for any w ∈ V we have
rot U × w, w = 0,
(5.99)
|rot w × U, w| ≤ γ w2V · U H1 (Ω) .
Therefore, if δ − γ U H1 (Ω) = β1 > 0, then by (5.98), (5.99), Assumption 5.1 and
Lemma 3.1 we have
a02 (u, u) ≥ β1 u2V ∀u ∈ V. (5.100)
It is easy to verify
|a02 (u, v)| ≤ cuV vV ∀u, v ∈ V. (5.101)
Then, (5.100) and (5.101) imply that a02 (u, v) satisfies condition (2) of Theorem
5.5.
By a property of mixed product of vectors,
2
a12 (w, u, u) = rot w × u, u = 0 ∀w ∈ V 3 (Ω), ∀u ∈ V, (5.102)
where
2
2

V 3 (Ω) = u ∈ H 3 (Ω) : div u = 0, u|Γ1 = 0, u τ |(∪i=2,4,7,9 Γi ) = 0, u · n|(∪i=3,5,8 Γi ) = 0 .
On the other hand, by density argument we get
a12 (w, u, v) = rot w × u, v = −rot w, v × u. (5.103)

1
When u, v ∈ V, v × u ∈ H 2 (Ω) and
v × uH 13 (Ω) ≤ c1 v × uH 21 (Ω) ≤ cvV uV (5.104)
(cf. Theorem 1.23). Also, if w ∈ V 3 (Ω), then rot w ∈ H− 3 (Ω) and

2 1
rot wH− 13 (Ω) ≤ cwH 23 (Ω) (5.105)
1
1
(see Theorem 1.24). Since H03 (Ω) = H 3 (Ω) (see Theorem 1.15), by (5.103)–
(5.105) we get
2
|a12 (w, u, v)| ≤ K wV 23 (Ω) uV vV ∀w ∈ V 3 (Ω) ∀u, v ∈ V. (5.106)
2 2
Since V → → V 3 (Ω), setting X = V, X 1 = V 3 (Ω), we can see that by (5.102)
and (5.106) a11 (w, u, v) satisfies condition (3) of Theorem 5.5.
Also, we have

F2 V∗ ≤ M1 U H1 + U 2H1 + f V∗ + φi − 21 + φi − 21 ,
H (Γi ) H (Γi )
i=2,4,7 i=3,5,6
(5.107)
where M1 depends on the mean curvature, shape operator, μ and α.
Therefore, by Theorem 5.5, we have the existence and estimate of solutions to
a02 (w, u − w) + a12 (w, w, u − w) + Φ(u) − Φ(w) ≥ F2 , u − w ∀u ∈ V.
Since v = w + U is a solution to the given problem, we have the existence of solu-

tions and the estimate (5.97).
If U H1 , f V∗ , φi H − 21 (Γ ) , i = 2, 4, 7, and φi H− 21 (Γ ) , i = 3, 5, 6, are small
i i
Let us consider now a special case of the Navier-Stokes problem with boundary
condition (5.2) in which there is no flux across boundary except Γ1 , Γ8 .
Theorem 5.10 Let Assumption 5.1 hold, Γi = ∅ (i = 2, 4, 6, 7, 9, 10, 11), the sur-
faces Γ3 j be convex, α positive and U H1 (Ω) small enough. Then, for any f and
φi , i = 3, 5, there exists a solution v to Problem I-VI (the steady Navier-Stokes
problem for the case of static pressure) and all solutions satisfy
M1
v − U H1 ≤ U 2H1 + f V∗ + φi H− 21 (Γ ) , (5.108)
δ − γ U H1 i=3,5
i
where δ, γ and M1 are as in (5.90), (5.91) and (5.95), respectively.

In addition, if f V∗ , φi H− 21 (Γ ) , i = 3, 5, are small enough, then the solution
i
is unique.
Proof Define a functional Φ(u) = φτ (u) by (5.21) and (5.22). Then, the functional
satisfies condition (1) of Theorem 5.6.
Let w = v − U , U be a function in Assumption 5.1 and a01 (·, ·), a11 (·, ·, ·) and
F1 ∈ V∗ be as in (5.9).
We can see that condition (2) in Theorem 5.5 is satisfied (see proof of Theorem
5.8).
By the condition of theorem,
2
a11 (w, u, u) = (w · ∇)u, u = 0 ∀w ∈ V 3 (Ω), ∀u ∈ V. (5.109)
By Hölder’s inequality we can see

2
|a11 (w, u, v)| ≤ K wV 23 (Ω) uV vV ∀w ∈ V 3 (Ω), ∀u, v ∈ V. (5.110)
By (5.109) and (5.110), a11 (w, u, v) satisfies condition (3) of Theorem 5.5.
Applying Theorem 5.5 to
a01 (w, u − w) + a11 (w, w, u − w) + Φ(u) − Φ(w) ≥ F1 , u − w ∀u ∈ K (Ω),
we come to the asserted conclusion.
Remark 5.4 Assumption Γi = ∅, i = 2, 4, 6, 7, 9 − 11, is only used to get (5.109).
Applying Theorem 5.7, next we revisit the problem concerned within Theorem
5.10. We will get a generalization of methods used in papers based on smooth approx-
imation of functional in variational inequalities (see [3]).
Lemma 5.2 Let X, Y be reflexive Banach spaces, an operator i : X → Y be com-

pletely linear continuous, j : Y → R be convex and Gâteaux derivative D j (y) =
a(y) for y ∈ Y . Then, Φ(v) ≡ j (iv) : X → R is convex, DΦ(v) ≡ A(v) = i ∗ a(iv),
where i ∗ is the operator adjoint to i, and A : X → X ∗ is weak continuous.
Proof It is easy to verify the convexity of Φ.
Φ(v + tu) − Φ(u) j (i(v + tu)) − j (iu)

A(v), u X = lim = lim
t→0 t t→0 t
= a(iv), iuY = i ∗ a(iv), u X ∀v, u ∈ X,
which means A(v) = i ∗ a(iv).

Let vn v in X . Since gradient of a finite convex functional is monotone and
demi-continuous (see Remark 1.15) and ivn → iv in Y ,
A(vn ), u X = i ∗ a(ivn ), u X = a(ivn ), iuY → a(iv), iuY = i ∗ a(iv), u X ∀u ∈ X,
that is, DΦ = A : X → X ∗ is weak continuous.
Theorem 5.11 Let Assumption 5.1 hold, Γi = ∅ (i = 2, 4, 6, 7, 9, 10, 11), the sur-
faces Γ3 j be convex, α positive and U H1 (Ω) small enough. Then, for any f and
φi , i = 3, 5, there exists a solution v to Problem I-VI (the steady Navier-Stokes prob-
lem for the case of static pressure) and the solution satisfies the estimate (5.108).
Proof Define an operator i : V → L2τ (Γ8 ) by iu = u|Γ8 and a functional Φ : V → R

1
by Φ(v) ≡ φτ (iv), where φτ is as in (5.21). Since the trace operator V → H 2 (∂Ω)
1
is continuous and H 2 (∂Ω) → → L (∂Ω), the operator i is compact, and by Lemma
2
5.2 Φ : V → R is weak continuous and convex.
Define a functional Φε : V → R by
Φε (v) = φτ ε (iv),

φτ ε (η) = gτ ρε (η) ds,
Γ8 (5.111)

|η| − ε/2 |η| > ε,
ρε (η) =
|η|2 /2ε |η| ≤ ε .
Since ε
|φτ ε (η) − φτ (η)| ≤ |gτ | ∀η ∈ L2τ (Γ8 )
2
(cf. Lemma 2.1 of [3]), we have
ε
|Φε (v) − Φ(v)| ≤ |gτ | ∀v ∈ V. (5.112)
2
Also, φτ ε is convex, and so its Gâteaux derivative is demi-continuous. Thus, by
Lemma 5.2 DΦε ≡ Aε ∈ (V → V∗ ) is weak continuous. By this fact together
(5.112), condition (1) of Theorem 5.7 is satisfied.
Under the assumption of theorem a01 (·, ·), a11 (·, ·, ·) and F1 ∈ V∗ of (5.9) are as
follows:
a01 (u, v) = 2μ(E(u), E(v)) + (U · ∇)u, v + (u · ∇)U, v

+ 2μ(S ũ, ṽ)Γ3 + 2(α(x)u, v)Γ5 ∀u, v ∈ V,
a11 (w, u, v) = (w · ∇)u, v ∀w, u, v ∈ V,
(5.113)
F1 , u = −2μ(E(U ), E(u)) − (U · ∇)U, u − 2μ(SŨ , ũ)Γ3
− 2(α(x)Uτ , u)Γ5 + f, u + φi , uΓi ∀u ∈ V.
i=3,5
2μ(E(u), E(u)) ≥ δu2V , δ > 0. (5.114)
On the other hand, for any u ∈ V we have
|(U · ∇)u, u + (u · ∇)U, u| ≤ γ u2V · U H1 (Ω) . (5.115)
Therefore, if δ − γ U H1 (Ω) = β1 > 0, then by (5.114), (5.115), Assumption 5.1

and Lemma 3.1 we have
a01 (u, u) ≥ β1 u2V ∀u ∈ V. (5.116)
Under condition Γi = ∅, i = 2, 4, 6, 7, 9, 10, 11, it is easy to verify that
a11 (v, v, v) = 0 ∀v ∈ V. (5.117)

Let
a(w, u, v) = a01 (u, v) + a11 (w, u, v).
Then, by (5.116) and (5.117) we have
a(v, v, v) ≥ β1 u2V ∀v ∈ V. (5.118)
Let us prove that when vm v in V, there exists a subsequence {vm p } such that
a(vm p , vm p , u) → a(v; v, u) ∀u ∈ V as p → ∞. (5.119)
To this end, first let us prove that when vm v in V, there exists a subsequence
{vm p } such that
a01 (vm p , u) → a01 (v, u) ∀u ∈ V as p → ∞. (5.120)
Since Ui u j ∈ L 2 (Ω), i, j = 1, 2, 3, and ∂i vm ∂i v in L 2 (Ω), we have
(U · ∇)vm , u → (U · ∇)v, u as m → ∞. (5.121)
By Hölder’s inequality,
|((vm − v) · ∇)U, u| ≤ cvm − vL3 (Ω) ∇U L2 (Ω) uL6 (Ω) .
Since H 1 (Ω) → → L 3 (Ω), we can choose a subsequence {vm p } such that vmp → v
in L3 (Ω). Then, we have
(vm p · ∇)U, u → (v · ∇)U, u as m p → ∞. (5.122)
It is easy to verify the convergence of other terms. Thus, using (5.121) and (5.122),
we have (5.120).
Using Hölder’s inequality and a11 (v, u, w) = −a11 (v, w, u), we have
|a11 (vm , vm , u) − a11 (v, v, u)|

≤ |a11 (vm , vm , u) − a11 (v, vm , u)| + |a11 (v, vm , u) − a11 (v, v, u)|
≤ c vm − vL3 (Ω) ∇vm L2 (Ω) uL6 (Ω) + vL6 (Ω) ∇uL2 (Ω) vm − vL3 (Ω) ∀u ∈ V.
Thus, we have
a11 (vm p , vm p , u) → a11 (v, v, u) ∀u ∈ V as m p → ∞. (5.123)
Combining (5.120) and (5.123) yields (5.119).

Let us prove that
lim inf a(vm p , vm p , vm p ) ≥ a(v, v, v). (5.124)

m→∞
By lower semi-continuity of norm
lim inf 2μ(E(vm ), E(vm )) ≥ 2μ(E(v), E(v)) as vm v in V. (5.125)

m→∞
It is easy to prove that
2μ(S ṽm , ũ)Γ3 + 2(α(x)vm , u)Γ5 → 2μ(S ṽ, ũ)Γ3 + 2(α(x)v, u)Γ5 ∀u ∈ V.
(5.126)
Using Hölder’s inequality and a11 (v, vm , u) = −a11 (v, u, vm ), we have
|a11 (vm , vm , vm ) − a11 (v, v, v)|

≤ |a11 (vm , vm , vm ) − a11 (v, vm , vm )| + |a11 (v, vm , vm ) − a11 (v, vm , v)|
+ |a11 (v, vm , v) − a11 (v, v, v)|
≤ c vm − vL3 (Ω) ∇vm L2 (Ω) vm L6 (Ω) + vL6 (Ω) ∇vm L2 (Ω) vm − vL3 (Ω)
+ vL6 (Ω) ∇vL2 (Ω) vm − vL3 (Ω) ,
which implies
a11 (vm p , vm p , vm p ) → a11 (v, v, v) as m p → ∞. (5.127)
From (5.124)–(5.127), we have (5.124).

By virtue of (5.118), (5.119) and (5.124), condition (2) of Theorem 5.7 is satisfied.
Therefore, by Theorem 5.7 we have the existence and estimate of a solution w ∈ V
to
a01 (w, u − w) + a11 (w, w, u − w) + φτ (u) − φτ (w) ≥ F1 , u − w ∀u ∈ V.

(5.128)
Since v = w + U is a solution, we come to the asserted conclusion.
Remark 5.5 The estimate of solution of Theorem 5.11 may not be true for all
solutions, and so Theorem 5.11 is weaker than Theorem 5.10.
Let us consider Problem III for the Stokes system.
Theorem 5.12 Let Assumption 5.1 hold, the surfaces Γ2 j , Γ3 j , Γ7 j be convex and
α positive. Then, for any f φi , i = 2, · · · , 7, there exists a unique solution v to
Problem III-VI for the steady Stokes problem with mixed boundary condition (5.2)
and the following estimate holds:
M1
v − U H1 ≤ U H1 + f V∗ + φi H − 21 (Γ ) + φi H− 21 (Γ ) ,
δ i=2,4,7
i
i=3,5,6
i
(5.129)
where δ and M1 are, respectively, as in (5.98) and (5.107) (for F3 instead of F2 ).

If v1 , v2 are solutions, respectively, to Problem-III-VI with gτ 1 , gn1 , g+n1 , g−n1 , f 1 ,
h i1 , φi1 and gτ 2 , gn2 , g+n2 , g−n2 , f 2 , h i1 , φi2 , then
M1
v1 − v2 H1 ≤ U1 − U2 H1 + f 1 − f 2 V∗ + gτ 1 − gτ 2 Lτ2 (Γ8 )
δ
+ gn1 − gn2 L 2 (Γ9 ) + g+n1 − g+n2 L 2 (Γ10 ) + g−n1 − g−n2 L 2 (Γ11 )

+ φi1 − φi2 − 1 + φi1 − φi2 − 1 + U1 − U2 H1 ,
H 2 (Γi ) H 2 (Γi )
i=2,4,7 i=3,5,6
(5.130)
j
where U j , j = 1, 2, are the functions in Assumption 5.1 with h i instead h i .
Proof By an argument similar to the proof of Theorem 5.6 we can apply the well
known result for variational inequality (see Theorem 10.5 of [2])
a03 (w, u − w) + Φ(u) − Φ(w) ≥ F3 , u − w ∀u ∈ X, (5.131)
where Φ(u) is defined by (5.22) and a03 (v, u), F3 are as in (5.54).
Thus, we have the existence of a unique solution and estimate (5.129).
If v1 = w1 + U1 , v2 = w2 + U2 are solutions corresponding to the given data, we
get
a03 (w1 , u − w1 ) + Φ1 (u) − Φ1 (w1 ) ≥ F31 , u − w1 ,

(5.132)
a03 (w2 , u − w2 ) + Φ2 (u) − Φ2 (w2 ) ≥ F32 , u − w2 ∀u ∈ V,
j
where Φ j (u), F3 , j = 1, 2, are the one corresponding to U j , gτ j , gn j , g+n j , g−n j ,
j j
f j , h i , φi . Putting u = w2 , u = w1 , respectively, in the first and second inequality
in (5.132) and adding those obtained, we have
a03 (w1 − w2 , w2 − w1 ) + Φ1 (w2 )−Φ1 (w1 ) + Φ2 (w1 ) − Φ2 (w2 )

(5.133)
≥ F31 − F32 , w2 − w1 .
By Korn’s inequality and Lemma 3.1 we have
a03 (w1 − w2 , w1 − w2 ) ≥ δw1 − w2 2V . (5.134)
From (5.133) and (5.134) we have
1 1
w1 − w2 2V ≤ |F3 − F32 , w2 − w1 | + |Φ1 (w2 ) − Φ1 (w1 ) + Φ2 (w1 ) − Φ2 (w2 )| .
δ
(5.135)
Since w1 , w2 ∈ K (Ω),

Φ1 (w2 ) − Φ1 (w1 ) = gτ 1 (|w2τ | − |w1τ |) ds + gn1 (|w2n | − |w1n |) ds
Γ8 Γ9

+ g+n1 (w2n − w1n ) ds − g−n1 (w2n − w1n ) ds,
Γ Γ11
10
Γ Γ
8 9
+ g+n2 (w2n − w1n ) ds − g−n2 (w2n − w1n ) ds.
Γ10 Γ11
(5.136)
Subtracting two formulas of (5.136), we have
|Φ1 (w2 ) − Φ1 (w1 ) + Φ2 (w1 ) − Φ2 (w2 )|

≤ gτ 1 − gτ 2 L2 (Γ8 ) w2τ − w1τ L2 (Γ8 ) + gn1 − gn2 L 2 (Γ9 ) w2n − w1n L 2 (Γ9 )
τ τ
+ g+n1 − g+n2 L 2 (Γ10 ) wn2 − wn1 L 2 (Γ10 ) + g−n1 − g−n2 L 2 (Γ11 ) w2n − w1n L 2 (Γ11 )
≤ M gτ 1 − gτ 2 L2 (Γ8 ) + gn1 − gn2 L 2 (Γ9 ) + g+n1 − g+n2 L 2 (Γ10 )
τ
+ g−n1 − g−n2 L 2 (Γ11 ) w2 − w1 V .

(5.137)
By (5.135) and (5.137), we have
M 1
w1 − w2 V ≤ F3 − F32 V∗ + gτ 1 − gτ 2 L2 (Γ8 ) + gn1 − gn2 L 2 (Γ9 )
δ τ

+ g+n1 − g+n2 L 2 (Γ10 ) + g−n1 − g−n2 L 2 (Γ11 ) ,
from which we get (5.130).
Remark 5.6 The estimates of solutions to the problems (5.97), (5.108) and (5.129)
are independent of the thresholds gτ , gn , g+n , g−n . (See (8) of [4] and (25) of [3].)
The content of Chap. 5 is taken from [5].

Till now, for the Stokes and Navier-Stokes problems with friction type boundary
conditions rather simple cases were studied. More clearly, problems with the Dirichlet
boundary condition on a portion of boundary and either Tresca slip condition or the
leak condition on the other portion have been dealt with.
In [6] the existence of solutions to the steady Stokes and Navier-stokes equations
with the homogeneous Dirichlet boundary condition on a portion of boundary and
leak or threshold slip boundary condition on the other portion was studied. Also, [7–
9] concerned with the steady or non-steady Stokes equations with the homogeneous
Dirichlet boundary condition and leak boundary condition.
When a portion of boundary with Dirichlet boundary condition and another mov-
ing portion where nonlinear slip occurs are separated, the existence, uniqueness and
continuous dependence on the data were studied for the steady Stokes equations
in [9]. In [10] when a portion of boundary with Dirichlet boundary condition and
another portion with slip condition are separated, the existence of strong solution
to the steady Stokes equations was established. In [11] when a portion with homo-
geneous Dirichlet boundary condition and another portion with nonlinear boundary
condition are separated, for the steady Stokes equations a relation between a regu-
larized problem and the original problem, the regularity of solution were discussed.
In [12] for the steady Navier-Stokes equations, the existence, uniqueness and
continuous dependence on the data were considered when a portion of boundary with
Dirichlet boundary condition and another moving portion where nonlinear slip occurs
are separated. In [13] the existence and uniqueness of a local solution to the steady
Navier-Stokes problem with homogeneous Dirichlet boundary condition and one of
friction boundary conditions was investigated. In [4] the existence and uniqueness of
solution to the steady rotating Navier-Stokes equations are studied when boundary
consists of a portion with homogeneous Dirichlet boundary condition and other
portions with a threshold slip. In [3] under similar boundary condition the steady
Navier-Stokes problem is studied. With exception of [13] in all above-mentioned
papers dealing the Navier-Stokes problem with friction boundary conditions, ones
approximated the functionals in the considering variational inequalities with smooth
one resulting to study of operator equation and it’s convergence.
Numerical solution methods were studied for the Stokes and Navier-Stokes prob-
lems with friction boundary conditions. For the 2-D steady Stokes problems we refer
to [14–18] and for the 3-D steady Stokes problems see [19]. For the 2-D steady
Navier-Stokes problem we refer to [20–23].
References

2. C. Baiocchi, A. Capelo, Variational and Quasivariational Inequalities: Applications to Free-
Boundary Problems (Wiley, Chichester, 1984) (Russian, 1988)
3. Y. Li, K. Li, Existence of the solution to stationary Navier-Stokes equations with nonlinear slip
4. R. An, K. Li, Variational inequality for the rotating Navier-Stokes equations with subdifferential
boundary conditions. Comput. Math. Appl. 55, 581–587 (2008)
5. T. Kim, D. Cao, The steady Navier-Stokes and Stokes systems with mixed boundary conditions
including one-sided leaks and pressure. Methods Appl. Anal. 23, 329–364 (2016)
6. H. Fujita, A mathematical analysis of motions of viscous incompressible fluid under leak or
slip boundary conditions. RIMS Kokyuroku 888, 199–216 (1994)
Comput. Math. 19, 1–8 (2001)
Comput. Appl. Math. 149, 57–69 (2002)
9. C.L. Roux, Steady Stokes flows with threshold slip boundary conditions. Math. Model. Methods
Appl. Sci. 15, 1141–1168 (2005)
10. N. Saito, On the Stokes equations with the leak and slip boundary conditions of friction type:
regularity of solutions. Publ. RIMS (Kyoto University) 40, 345–383 (2004)
11. N. Saito, H. Fujita, Regularity of solutions to the Stokes equation under a certain nonlinear
boundary condition. Lect. Notes Pure Appl. Math. 223, 73–86 (2001)
12. C.L. Roux, A. Tani, Steady solutions of the Navier-Stokes equations with threshold slip bound-
ary conditions. Math. Methods Appl. Sci. 30, 595–624 (2007)
13. F. Saidi, On the Navier-Stokes equations with the slip boundary conditions of friction type:
regularity of solution. Math. Model. Anal. 12, 389–398 (2007)
14. M. Ayadi, M.K. Gdoura, T. Sassi, Mixed formulation for Stokes problem with Tresca friction.
C. R. Acad. Sci. Paris Ser. I 348, 1069–1072 (2010)
15. T. Kashiwabara, On a finite element approximation of the Stokes problem under a leak boundary
condition of friction type. Jpn. J. Ind. Appl. Math. 30, 227–261 (2013)
16. Y. Li, K. Li, Penalty finite element method for Stokes problem with nonlinear slip boundary
conditions. Appl. Math. Comput. 204, 216–226 (2008)
17. Y. Li, K. Li, Locally stabilized finite element method for Stokes problem with nonlinear slip
boundary conditions. J. Comput. Math. 28, 826–836 (2010)
18. Y. Li, K. Li, Uzawa iteration method for Stokes type variational inequality of the second kind.
19. T. Kashiwabara, Finite element method for Stokes equations under leak boundary condition of
friction type. SIAM J. Numer. Anal. 52, 2448–2469 (2013)
20. R. An, Comparisons of Stokes/Oseen/Newton iteration methods for Navier-Stokes equations
with friction boundary conditions. Appl. Math. Model. 38, 5535–5544 (2014)
21. Y. Li, R. An, Penalty finite element method for Navier-Stokes equations with nonlinear slip
boundary conditions. Int. J. Numer. Methods Fluids 69, 550–566 (2012)
22. Y. Li, R. An, Two-level pressure projection finite element methods for Navier-Stokes equations
with nonlinear slip boundary conditions. Appl. Numer. Math. 61, 285–297 (2011)
23. Y. Li, R. An, Two-level variational multiscale finite element methods for Navier-Stokes type
variational inequality problem. J. Comput. Appl. Math. 290, 656–669 (2015)
Chapter 6
The Non-steady Navier-Stokes System
with Friction Boundary Conditions
In this chapter we are concerned with the non-steady Navier-Stokes and Stokes
problems corresponding to the steady problems in Chap. 5. In Sect. 6.1 relying on
the results of Sect. 3.1, we embed all boundary conditions to variational formulations.
We get variational inequalities with one unknown which are equivalent to the original
PDE problems for the smooth solutions. In Sect. 6.2 we study the existence and
uniqueness of solutions to the variational inequalities obtained in Sect. 6.1. In Sect.
6.3 using the results of Sect. 6.2, we get the existence, uniqueness and estimates of
solutions to the Navier-Stokes and Stokes problems with the boundary conditions.
For the problem with boundary conditions involving the total pressure and total
stress, the existence of a solution without restriction of data of the problem is proved.
For the problem with boundary conditions involving the static pressure and stress,
under a compatibility condition at the initial time for the small data it is proved that
there exists a unique solution on the given interval of time.
6.1 Variational Formulations of Problems
Let Ω be a bounded
domain of R , l = 2, 3. ∂Ω ∈ C , ∂Ω = ∪i=1 Γ i , Γi ∩ Γ j =2,1
l 0,1 11
∅
for i = j, Γi = j Γi j , where Γi j are connected open subsets of ∂Ω and Γi j ∈ C
for i = 2, 3, 7 and Γi j ∈ C 1 for others. Also, let Q = Ω × (0, T ), Σi = Γi × (0, T ),
0 < T < ∞.
https://doi.org/10.1007/978-3-030-78659-5_6
186 6 The Non-steady Navier-Stokes System with Friction Boundary Conditions
Remark 6.1 In Chap. 5 for the boundary of Ω it is assumed that Γi j ∈ C 2 , i =

2, 3, 7, and those are convex in Theorems 5.8–5.12, whereas in this chapter it is
assumed that Γi j ∈ C 2,1 , i = 2, 3, 7, and convexity of Γi j , i = 2, 3, 7, is not assumed
in the following theorems. In the following chapters for the steady problems the
assumptions as in Chap. 5 are used, whereas for the non-steady problems the assump-
tions as in this chapter are used (cf. Remark 5.2).
We are concerned with Problems I and II for the non-steady Navier-Stokes prob-
lem ⎧ ∂v
⎪
⎨ ∂t − μΔv + (v · ∇)v + ∇ p = f,
⎪
div v = 0, (6.1)
⎪
⎪
⎩
v(0) = v0 ,
which are distinguished according to boundary conditions. Problem I is the one with
the boundary conditions (5.3) (the case of total pressure) and Problem II is the one
with the boundary conditions (5.2) (the case of static pressure).
We also consider the Stokes problem
⎧ ∂v
⎪
⎨ ∂t − μΔv + ∇ p = f,
⎪
div v = 0, (6.2)
⎪
⎪
⎩
v(0) = v0
with the boundary conditions (5.2), which is Problem III.
Assuming that f ∈ L 2 (0, T ; L2 (Ω)), φi ∈ L 2 (0, T ; L 2 (Γi )), i = 2, 4, 7, and

φi ∈ L 2 (0, T ; L2 (Γi )), i = 3, 5, 6, formally we introduce the following:
Definition 6.1 A function (v, p) ∈ L 2 (0, T ; H2 (Ω)) × L 1 (0, T ; H 1 (Ω)) is called
a solution to Problem I (or Problem II ) if the first two equations of (6.1) hold
in L 1 (0, T ; L2 (Ω)), v(0) = v0 in L2 (Ω) and each of (5.3) (or (5.2)) holds in
L 1 (0, T ; L 2 (Γi )) or L 1 (0, T ; L2 (Γi )).
A function (v, p) ∈ L 2 (0, T ; H2 (Ω)) × L 2 (0, T ; H 1 (Ω)) is called a solution to
Problem III if the first two equations of (6.2) hold in L 2 (0, T ; L2 (Ω)), v(0) = v0 in
L2 (Ω) and each of (5.2) holds in L 2 (0, T ; L 2 (Γi )) or L 2 (0, T ; L2 (Γi )).
Remark 6.2 Under the conditions for (v, p) and f , we have that v ∈ C([0, T ];
L2 (Ω)) (see Lemma 1.1 of Chap. III of [1]), and so the condition that v(0) = v0 in
L2 (Ω) is meaningful.
Let V, K (Ω) be the same as in (5.5) and
H : the closer of V in L2 (Ω),

V(Q) = L 2 (0, T ; V)
VΓ 237 (Ω) = {u ∈ H1 (Ω) : div u = 0, u τ |(Γ2 ∪Γ7 ) = 0, u n |Γ3 = 0},
K (Q) = {u ∈ L 2 (0, T ; V) : u ∈ L 1 (0, T ; V∗ ); u n |Γ10 ≥ 0, u n |Γ11 ≤ 0},
(Q) = {u ∈ C2 ( Q̄) : div u = 0, u |Σ1 = 0, u τ |(Σ2 ∪Σ4 ∪Σ7 ) = 0, u n |(Σ3 ∪Σ5 ) = 0}.
We will use the following assumption:
Assumption 6.1 The followings hold:

(1) There exists a function U ∈ W 1,2 (0, T ; H1 ) ∩ L∞ (Q) such that
div U = 0, U |Σ1 = h 1 , Uτ |(Σ2 ∪Σ7 ) = 0, Un |Σ3 = 0, Uτ |Σ4 = h 4 ,

Un |Σ5 = h 5 , U |Σ8 = h 8 n, U |Σ9 = h 9 , U |Σ10 ∪Σ11 = 0,
where Σi = Γi × (0, T ).
(2) f ∈ L 2 (0, T ; V∗ ), φi ∈ L 2 (0, T ; H − 2 (Γi )), i = 2, 4, 7, φi ∈ L 2 (0, T ;
1
H− 2 (Γi )), i = 3, 5, 6, αi j ∈ L ∞ (Γ5 ), v0 − U (x, 0) ∈ H , and Γ1 = ∅.

1
(3) of Assumption 5.1 holds.
Remark 6.3 On the Γ10 (Γ11 ) outflow (inflow) only is possible. Thus, to guarantee
div u = 0 we assume that if Γ10 = ∅ (Γ11 = ∅), then for example Γ2 = ∅. In Sect. 5.1
for proof of equivalence of variational formulations between variational inequalities,
(3) of Assumption 5.1 was used via Lemma 5.1 to get existence of σ+n , σ−n such that
σ+n + g+n ≥ 0, σ−n − g−n ≤ 0 from variational inequalities, which is also necessary
for Theorems 6.2 and 6.4.
Taking (v · ∇)v = rot v × v + 21 grad|v|2 into account and putting v = w + U ,

by (5.6), (5.7) and Assumption 6.1 we can see that solutions (v, p) of the problem
(6.1), (5.3) in the sense of Definition 6.1 satisfy the following:
⎧
⎪
⎪ v − U = w ∈ K (Q), v0 − U (x, 0) = w(0),
⎪
⎪
⎪
⎪ ∂w
⎪
⎪ , u + 2μ(E(w), E(u)) + rot w × w, u + rot U × w, u
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ + rot w × U, u + 2μ(k(x)w, u)Γ2 + 2μ(S w̃, ũ)Γ3
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪ + p + |v|2 − 2μεnn (w + U ), u n Γ9 ∪Γ10 ∪Γ11
⎪
⎪
⎪
⎪ 2
⎪
⎨ ∂U
=− , u − 2μ(E(U ), E(u)) − rot U × U, u − 2μ(k(x)U, u)Γ2 (6.3)
⎪ ∂t
⎪
⎪
⎪
⎪ − 2μ(SŨ , ũ)Γ3 − 2(α(x)Uτ , u)Γ5 − μ(k(x)U, u)Γ7
⎪
⎪
⎪
⎪
⎪
⎪ + f, u + φi , u n Γi + φi , uΓi ∀u ∈ V,
⎪
⎪
⎪
⎪ i=2,4,7 i=3,5,6
⎪
⎪
⎪
⎪ |στt (v)| ≤ gτ , στt · vτ + gτ |vτ | = 0 on Γ8 ,
⎪
⎪
⎪
⎪
⎪
⎪ |σnt (v)| ≤ gn , σnt (v)vn + gn |vn | = 0 on Γ9 ,
⎪
⎪
⎪
⎪
t t
⎪
⎩ t
σn (v) − g−n ≤ 0, (σnt (v) − g−n )vn = 0 on Γ11 .
Define a01 (t, ·, ·), a11 (·, ·, ·) and F1 (t) ∈ V∗ by
a01 (t, w, u) = 2μ(E(w), E(u)) + rot U (t) × w, u + rot w × U (t), u

+ μ(k(x)w, u)Γ7 ∀w, u ∈ V,
a11 (w, u, v) = rot w × u, v ∀w, u, v ∈ V,
∂U (t)
F1 (t), u = − , u − 2μ(E(U (t)), E(u)) − rot U (t) × U (t), u (6.4)
∂t
− 2μ(k(x)U (t), u)Γ2 − 2μ(SU ˜(t), ũ)Γ3
− 2(α(x)U (t)τ , u)Γ5 − μ(k(x)U (t), u)Γ7 + f (t), u
+ φi (t), u n Γi + φi (t), uΓi ∀u ∈ V.
i=2,4,7 i=3,5,6
1
στt (v) = 2μεnτ (v), σnt (v) = −( p + |v|2 ) + 2μεnn (v)
2
and (6.3), we introduce the following variational formulation for problem (6.1), (5.3).
Problem I-VE. Find v=U +w ∈ (U + K (Q)) and (στt , σnt , σ+n t

, σ−n
t
) ∈ L2τ (Γ8 ) ×
− 21 − 21
L (Γ9 ) × H (Γ10 ) × H (Γ11 ) in a.e. t ∈ (0, T ) such that w(0) = v0 − U (0) and
2
⎧
⎪ ∂w
⎪
⎪ , u + a01 (t, w, u) + a11 (w, w, u) − (στt , u τ )Γ8 − (σnt , u n )Γ9
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ − σ+nt
, u n Γ10 − σ−n
t
, u n Γ11 = F1 , u
⎪
⎪
⎪
⎪
⎪
⎨ ∀u ∈ L 2 (0, T ; V),
⎪
⎪
⎪
⎪
⎪
⎪ |σn | ≤ gn ,
t
σnt vn + gn |vn | = 0 on Γ9 ,
⎪
⎪
⎪
⎪ σ+n
t
+ g+n ≥ 0, σ+n
t
+ g+n , vn = 0 on Γ10 ,
⎪
⎪ Γ10
⎪
⎪
⎩σt − g ≤ 0, σ−n
t
− g−n , vn = 0 on Γ11 ,
−n −n Γ11
where and in what follows L2τ (Γ8 ) is the subspace of L2 (Γ8 ) consisting of functions
such that (u, n)L2 (Γ8 ) = 0.
Remark 6.4 Under (2) of Assumption 6.1 the dual products on Γi in (6.4) and (6.5)
is meaningful. For more detail we refer to Remark 5.1.
In the same way as Theorem 5.1, we have
tion 6.1 of the problem (6.1), (5.3), then (v, στt |Γ8 , σnt |Γ9 , σnt |Γ10 , σnt |Γ11 ) is a solution
to Problem I-VE.
Conversely, if f ∈ L 2 (0, T ; L2 (Ω)) and Problem I-VE has a smooth solution
(v, στt , σnt , σ+t , σ−t ) such that v ∈ L 2 (0, T ; H2 (Ω)), στt ∈ L 2 (0, T ; L2 (Γ8 )), σnt ∈
L 2 (0, T ; L 2 (Γ9 )), σ+t ∈ L 2 (0, T ; L 2 (Γ10 )) and σ−t ∈ L 2 (0, T ; L 2 (Γ11 )), then there
exists p ∈ L 1 (0, T ; H 1 (Ω)) such that (v, p) is a solution to the problem (6.1), (5.3).
Moreover, if at least one of the sets Γi , i = 2, 4, 6, 7, is nonempty, then p is unique.
We will find a variational inequality corresponding to Problem I-VE.

Let (v, στ , σn , σ+n , σ−n ) be a solution of Problem I-VE. Subtracting the first
formula of (6.5) with u = w from the first formula of (6.5), we get
∂w
, u − w + a01 (t, w, u − w) + a11 (w, w, u − w) − (στt , u τ − wτ )Γ8
∂t
− (σnt , u n − wn )Γ9 − σ+n
t
, u n − wn − σ−n
t
, u n − wn (6.6)
Γ10 Γ11
= F1 , u − w.
Let Φ : V → R be the functional defined by (5.22). Define a functional Ψ (u) by

⎧ T
⎨ Φ(u(t)) dt i f Φ(u(t)) ∈ L 1 (0, T ),
Ψ (u) = (6.7)
⎩ 0
+∞ other wise.
In the same way as in Problem I of Chap. 5, we get from (6.6)

∂w
, u − w + a01 (t, w, u − w) + a11 (w, w, u − w) + Φ(u) − Φ(w)
∂t (6.8)
≥ F1 , u − w.
If u − w ∈ (Q), then
T ∂w T ∂(w − u) T ∂u
, u − w dt = , w − u dt + , u − w dt
0 ∂t 0 ∂t 0 ∂t
T ∂u (6.9)
1
= , u − w dt + (w0 − u(0)2 − w(T ) − u(T )2 ).
0 ∂t 2
Define operators A1 (t) : V → V∗ and B1 : V × V → V∗ by
A1 (t)w, u = a01 (t, w, u) ∀w, u ∈ V,

(6.10)
B1 (w, u), v = a11 (w, u, v) ∀w, u, v ∈ V.
If w ∈ L 2 (0, T, V) ∩ L ∞ (0, T ; L2 (Ω)), then w ∈ L 4 (0, T, L3 (Ω)) (see (1.21))

and B1 (w, w) ∈ L 3 (0, T, V∗ ). Thus, when u ∈ L 4 (0, T ; V), the integral
4
T
0 B1 (w, w), u − w dt is well defined since B1 (w, w), w = 0.
Thus, neglecting 21 (w(T ) − u(T )2 ), from (6.8) and (6.9) we come to a varia-
tional inequality corresponding to Problem I (the case of total pressure).
Problem I-VI. Find v such that v − U ≡ w ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; H ) and

T
u + A1 (t)w(t) + B1 (w(t), w(t)) − F1 (t), u(t) − w(t) dt + Ψ (u) − Ψ (w)
0
1
≥ − w0 − u(0)2 ∀u ∈ L 4 (0, T ; V) with u ∈ L 2 (0, T ; V∗ ).
2
(6.11)
If the solutions to Problem I-VI is smooth as much as v − U ≡ w ∈ L 2 (0, T ; V),
w ∈ L 2 (0, T ; V∗ ), then we can see that the solutions satisfy
T
w (t) + A1 (t)w(t) + B1 (w(t), w(t)) − F1 (t), u(t) − w(t) dt + Ψ (u) − Ψ (w) ≥ 0
0
∀u ∈ L 4 (0, T ; V).
(6.12)
Especially, for the cases of 2-D and small data we will study such solutions to
Problem I.
Thus, we come to another formulation associated with Problem I by a variational
inequality.
Problem I-VI’. Find v such that v − U ≡ w ∈ L 2 (0, T ; V), w ∈ L 2 (0, T ; V∗ ),

w(0, x) = v0 − U (0, x) and (6.12) is satisfied.
Remark 6.5 (Remark, p. 114 of [2]) The inequality (6.12) is equivalent to
w (t) + A1 (t)w(t) + B1 (w(t), w(t)) − F1 (t), u − w(t) + Φ(u) − Φ(w(t)) ≥ 0

for a.e. t ∈ [0, T ], ∀u ∈ K (Ω).
(6.13)
Further, the function w in (6.12) also satisfies the inequality (6.11).
In (6.13) putting F, u − w = − ∂w ∂t
+ F1 , u − w, by the argument in the proof
of Theorem 5.2 we can obtain existence of (στt , σnt , σ+n t
, σ−n
t
) ∈ L2τ (Γ8 ) × L 2 (Γ9 ) ×
−1/2 −1/2
H (Γ10 ) × H (Γ11 ) at a.e. t ∈ (0, T ) such that (v, στ , σnt , σ+n
t t
, σ−n
t
) is a solu-
tion to Problem I-VE. Thus, we have
Theorem 6.2 If (v, στt , σnt , σ+n
t
, σ−n
t
) is a solution to Problem I-VE such that v −
∗
U ∈ L (0, T ; V ), then v is a solution to Problem I-VI’.
2
Conversely, if v is a solution to Problem I-VI’, then there exist στt , σnt , σ+nt
, σ−n
t
for a.e. t ∈ (0, T ) such that (v, στ , σn , σ+n , σ−n ) is a solution to Problem I-VE.
t t t t
In the same way as in Problem I, we will get the variational formulations of Problem
II for the Navier-Stokes equations with boundary condition (5.2).
Having in mind Assumption 6.1 and putting v = w + U , by (5.6) and (5.7) we
can see that solutions (v, p) of problem (6.1), (5.2) in the sense of Definition 6.1
satisfy the following:
⎧
⎪ v − U = w ∈ K (Q),
⎪
⎪
⎪
⎪ ∂w
⎪
⎪ , u + 2μ(E(w), E(u)) + (w · ∇)w, u + (U · ∇)w, u
⎪
⎪ ∂t
⎪
⎪
⎪
⎪
⎪
⎪ + (w · ∇)U, u + 2μ(k(x)w, u)Γ2 + 2μ(S w̃, ũ)Γ3
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ + ( p − 2μεnn (w + U ), u n )Γ9 ∪Γ10 ∪Γ11
⎪
⎪
⎪
⎪ ∂U
⎨ =− , u − 2μ(E(U ), E(u)) − (U · ∇)U, u − 2μ(k(x)U, u)Γ2
∂t
⎪
⎪ − 2μ(SŨ , ũ)Γ3 − 2(α(x)Uτ , u)Γ5 − μ(k(x)U, u)Γ7
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ + f, u + φi , u n Γi + φi , uΓi ∀u ∈ V,
⎪
⎪
⎪
⎪ i=2,4,7 i=3,5,6
⎪
⎪
⎪
⎪ |στ (v)| ≤ gτ , στ (v) · vτ + gτ |vτ | = 0 on Γ8 ,
⎪
⎪
⎪
⎪ |σn (v)| ≤ gn , σn (v)vn + gn |vn | = 0 on Γ9 ,
⎪
⎪
⎪
⎪
⎪
⎪
⎩
(6.14)
Define a02 (t, ·, ·), a12 (·, ·, ·) and F2 (t) ∈ V∗ by
a02 (t, w, u) = 2μ(E(w), E(u)) + (U (t) · ∇)w, u + (w · ∇)U (t), u

+ μ(k(x)w, u)Γ7 ∀w, u ∈ V,
a12 (w, u, v) = (w · ∇)u, v ∀w, u, v ∈ V,
∂U (t)
F2 (t), u = − , u − 2μ(E(U (t)), E(u)) − (U (t) · ∇)U (t), u (6.15)
∂t
− 2μ(k(x)U (t), u)Γ2 − 2μ(SU ˜(t), ũ)Γ3
− 2(α(x)U (t)τ , u)Γ5 − μ(k(x)U (t), u)Γ7 + f (t), u
+ φi (t), u n Γi + φi (t), uΓi ∀u ∈ V.
i=2,4,7 i=3,5,6
στ (v) = 2μεnτ (v), σn (v) = − p + 2μεnn (v)
and (6.14), we introduce the following variational formulation for problem (6.1),
(5.2).

Problem II-VE. Find v = U +w ∈ U + K (Q) and (στ , σn , σ+n , σ−n ) ∈ L2τ (Γ8 ) ×
L 2 (Γ9 ) × H −1/2 (Γ10 ) × H −1/2 (Γ11 ) in a.e. t ∈ (0, T ) such that
w(0) = v0 − U (0)
and
⎧
⎪ ∂w
⎪
⎪ , u + a02 (t, w, u) + a12 (w, w, u) − (στ , u τ )Γ8 − (σn , u n )Γ9
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ − σ+n , u n Γ10 − σ−n , u n Γ11 = F2 , u ∀u ∈ L 2 (0, T ; V),
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎪ σ+n + g+n ≥ 0, σ+n + g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎪
⎩
σ−n − g−n ≤ 0, σ−n − g−n , vn Γ11 = 0 on Γ11 .
In the same way as in Theorem 5.1 we have

tion 6.1 of the problem (6.1), (5.2), then (v, στ |Γ8 , σn |Γ9 , σn |Γ10 , σn |Γ11 ) is a solution
to Problem II-VE.
Conversely, if f ∈ L 2 (0, T ; L2 (Ω)) and Problem II-VE has a smooth solution

(v, στ , σn , σ+ , σ− ) such that v ∈ L 2 (0, T ; H2 (Ω)), στ ∈ L 2 (0, T ; L2 (Γ8 )), σn ∈
L 2 (0, T ; L 2 (Γ9 )), σ+ ∈ L 2 (0, T ; L 2 (Γ10 )) and σ− ∈ L 2 (0, T ; L 2 (Γ11 )), then there
exists p ∈ L 1 (0, T ; H 1 (Ω)) such that (v, p) is a solution to the problem (6.1), (5.2).
We will find a variational inequality corresponding to Problem II-VE.
Define operators A2 (t) : V → V∗ and B2 : V × V → V∗ by
A2 (t)w, u = a02 (t, w, u) ∀w, u ∈ V,

(6.17)
B2 (w, u), v = a12 (w, u, v) ∀w, u, v ∈ V.
Unlike Problem I, in this problem the property B2 (w, w), w = 0 fails and for
T
w ∈ L 2 (0, T ; V) integral 0 B2 (w, w), w ds is not meaningful. Thus, we will find
more smooth solution w ∈ L 4 (0, T ; V), such that w ∈ L 2 (0, T ; V∗ ).
Thus, as Problem I-VI’ we come to the following variational inequality.
Problem II-VI. (the case of static pressure) Find v such that w ≡ v − U ∈ L 4

(0, T ; V), ∂w
∂t
∈ L 2 (0, T ; V∗ ), w(0) = v0 − U (0, x) and
T ∂w
+ A2 (t)w(t) + B2 (w(t), w(t)) − F2 (t), u(t) − w(t) dt + Ψ (u) − Ψ (w) ≥ 0
0 ∂t
∀u ∈ L 4 (0, T ; V),
(6.18)
where Ψ is defined by (6.7).
As Theorem 6.2 we have the following theorem:
Theorem 6.4 If (v, στ , σn , σ+n , σ−n ) is a solution to Problem II-VE such that v −
U ∈ L 2 (0, T ; V∗ ), then v is a solution to Problem II-VI.
Conversely, if v is a solution to Problem II-VI, then there exist στ , σn , σ+n , σ−n
for a.e. t ∈ (0, T ) such that (v, στ , σn , σ+n , σ−n ) is a solution to Problem II-VE.
6.1.3 Variational Formulation: The Stokes Problem
In the same way as Problem I we get the equivalent formulations of Problem III for
the Stokes equation with boundary condition (5.2).

Problem III-VE. Find v = U + w ∈ U + K (Q) and (στ , σn , σ+n , σ−n ) ∈
L2τ (Γ8 ) × L 2 (Γ9 ) × H − 2 (Γ10 ) × H − 2 (Γ11 ) in a.e. t such that w(0) = v0 − U (0)
1 1
and
⎧
⎪ ∂w
⎪
⎪ , u + a03 (w, u) − (στ , u τ )Γ8 − (σn , u n )Γ9 − σ+n , u n Γ10 − σ−n , u n Γ11
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ = F3 , u ∀u ∈ L 2 (0, T ; V),
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎪ σ+n + g+n ≥ 0, σ+n + g+n , vn = 0 on Γ10 ,
⎪
⎪
⎩
σ−n − g−n ≤ 0, σ−n − g−n , vn = 0 on Γ11 ,
(6.19)
where
a03 (w, u) = 2μ(E(w), E(u)) + 2μ(k(x)w, u)Γ2 + 2μ(S w̃, ũ)Γ3

+ 2(α(x)w, u)Γ5 + μ(k(x)w, u)Γ7 ∀w, u ∈ V,
∂U
F3 , u = − , u − 2μ(E(U ), E(u)) − 2μ(k(x)U, u)Γ2
∂t (6.20)
+ φi , u n Γi + φi , uΓi ∀u ∈ V.
i=2,4,7 i=3,5,6
∂w
Problem III-VI. Find v such that w ≡ v − U ∈ L 2 (0, T ; V), ∂t
∈ L 2 (0, T ; V∗ ),
w(0) = v0 − U (0) and
T
∂w
+ A3 w(t) − F3 (t), u(t) − w(t) dt + Ψ (u) − Ψ (w) ≥ 0
0 ∂t (6.21)
∀u ∈ L 2 (0, T ; V),
where A3 : V → V∗ is defined by
A3 w, u = a03 (w, u) ∀w, u ∈ V. (6.22)
As Theorem 6.2 we have

Theorem 6.5 If (v, στ , σn , σ+n , σ−n ) is a solution to Problem III-VE such that v −
U ∈ L 2 (0, T ; V∗ ), then v is a solution to Problem III-VI. Conversely, if v is a
solution to Problem III-VI, then there exist στ , σn , σ+n , σ−n at a.e. t ∈ (0, T ) such
that (v, στ , σn , σ+n , σ−n ) is a solution to Problem III-VE.
6.2 The Existence and Uniqueness of Solutions to

Variational Inequalities
In this section we study the variational inequalities obtained in Sect. 6.1.

6.2 The Existence and Uniqueness of Solutions to Variational Inequalities 195
Let us start with the existence of a solution w ∈ L 2 (0, T ; V) to the following

problem:
T
u (t) + A(t)w(t) + B(w(t), w(t)) − F(t), u(t) − w(t) dt + Ψ (u) − Ψ (w)
0
1
≥ − w0 − u(0)2 ∀u ∈ L 4 (0, T ; V) with u ∈ L 2 (0, T ; V∗ ).
2
(6.23)
Theorem 6.6 Assume that
(1) For any u, v ∈ V, A(t)u, v is continuous with respect to t and
A(t)u, u ≥ c1 u2V − c2 u2 ∃c1 > 0, c2 ≥ 0, ∀u ∈ V,

| A(t)u, v | ≤ c3 uV vV ∃c3 > 0, ∀u, v ∈ V,
lim inf A(t)u m , u m ≥ A(t)u, u as u m u in V;
m→∞
(2) B(u, v) is bilinear continuous from V × V into V∗ and
1 1
| B(w, v), z | ≤ K wH1 v 2 vH2 1 zH1 ,
B(w, v), v = 0;
(3) F ∈ L 2 (0, T ; V∗ );
(4) The functional Φ : V → R is proper, convex, weakly lower semi-continuous and
Φ(u) ≥ 0 ∀u ∈ V, Φ(0V ) = 0.
Then, for any initial function w0 ∈ H there exists a solution to (6.23) satisfying
w(t)2 ≤ K 1 (w0 2 + F2L 2 (0,T ;V∗ ) ),

(6.24)
w2L 2 (0,T ;V) ≤ K 1 (w0 2 + F2L 2 (0,T ;V∗ ) ),
where K 1 is independent of Φ.
Proof For every ε > 0, let Φε be the Moreau-Yosida approximation of Φ and ∇Φε
be Fréchet derivative of Φε (see Sect. 1.6).
mLet {v j , j = 1, 2, · · · } be a basis of the space V. We find a solution wm =

j=1 g jm (t)v j to problem
⎧
⎨ wm (t), v j + A(t)wm (t), v j + B(wm (t), wm (t)), v j + ∇Φε (wm (t)), v j
⎪
= F(t), v j ,
⎪
⎩
wm (0) = w0m ,
(6.25)
m
where wm (t) = dtd wm (t) and w0m is such that w0m = i=1 k j v j → w0 in H as
m → ∞. The solutions to (6.25) depend on ε, but for convenience of notation here
and in what follows we use subindex m instead of subindex m, ε. For a tm there exists
an absolute continuous function g jm (t) on [0, tm ) and by the first a priori estimate
of (6.30) continuation of the function beyond tm is possible. Thus, we can see that
tm = T.
Multiplying (6.25) by g jm and summing over j = 1, · · · , m, yield
1 dwm (t)2
+ A(t)wm (t) + B(wm (t), wm (t)) + ∇Φε (wm (t)), wm (t) = F(t), wm (t) .
2 dt
(6.26)
By (1.37) and condition (4) we know that Φε (0V ) = 0. Also, since Φε is convex,
continuous and Fréchet differentiable, we have
Φε (v) − Φε (wm (t)) ≥ ∇Φε (wm (t)), v − wm (t) ∀v ∈ V, (6.27)
and so by Φε (0V ) = 0,
0 ≤ Φε (wm (t)) ≤ ∇Φε (wm (t)), wm (t) . (6.28)
By virtue of conditions (1), (2) and (6.28), we have from (6.26)
1 dwm (t)2
+ c1 wm (t)2V + Φε (wm (t)) ≤ F(t), wm (t) + c2 wm (t)2 ,
2 dt
d 1
wm (t)2 + 2c1 wm (t)2V + 2Φε (wm (t)) ≤ F2V∗ + c1 wm (t)2V + 2c2 wm (t)2 ,
dt c1
and
d 1
wm (t)2 + c1 wm (t)2V + 2Φε (wm (t)) ≤ F2V∗ + 2c2 wm (t)2 . (6.29)
dt c1
We have from (6.29)
1
wm (t)2 ≤ (w0 2 + F2L 2 (0,T ;V∗ ) )e2c2 T ≤ C1 (w0 2 + F2L 2 (0,T ;V∗ ) ),
c1
2c2 1 1
wm (t)2L 2 (0,T ;V) ≤ (w0 2 + F2L 2 (0,T ;V∗ ) )e2c2 T + 2 F2L 2 (0,T ;V∗ )
c1 c1 c1
≤ C2 (w0 2 + F2L 2 (0,T ;V∗ ) )
(6.30)
and T
Φε (wm (t)) dt ≤ C3 , (6.31)
0
where Ci , i = 1, 2, 3, are independent of m, ε and Φ. From (1.36), (1.37) and (6.31)

we have T
wm (t) − Jε (wm (t))2V dt ≤ c7 ε (6.32)
0
ε.
with c7 independent of m and
Multiplying (6.25) by g jm (t) − g jm (s) , s ∈ (0, T ), and summing over j =
1, · · · , m, we have
1 dwm (t) − wm (s)2

+ A(t)wm (t) + B(wm (t), wm (t)) − F(t), wm (t) − wm (s)
2 dt
= ∇Φε (wm (t)), wm (s) − wm (t)
≤ Φε (wm (s)) − Φε (wm (t))
≤ Φε (wm (s)).
By conditions (1) and (2), from the above it follows that
dwm (t) − wm (s)2

≤Φε (wm (s)) + A(t)wm (t), wm (s) + B(wm (t), wm (t)), wm (s)
dt
+ F(t), wm (t) − wm (s) + c2 wm (t)2 .
(6.33)
Let us integrate every terms in (6.33) first with respect to t from s to s + h and
then with respect to s from 0 to T , where wm (t) = 0 when t ∈ (T, T + h). We get

T s+h
dwm (t) − wm (s)2 T
dtds = wm (s + h) − wm (s)2 ds. (6.34)
0 s dt 0
By (6.31),
T s+h T
Φε (wm (s)) dtds ≤ h Φε (wm (s)) ds ≤ c6 h. (6.35)
0 s 0
By conditions (1) and (6.30) we have

T s+h √ T
A(t)wm (t), wm (s) dtds ≤ c8 h wm (s)2V ds. (6.36)
0 s 0
By conditions (2) and (6.30) we have

T s+h
B(wm (t), wm (t))wm (s) dtds
0 s
T s+h 3
(6.37)
1 1
≤K wm (t)V2 wm (t) 2 wm (s)V dtds ≤ c9 h 4 .
0 s
Using (6.30) we have

T s+h T t
F(t), wm (t) dtds ≤ | F(t), wm (t)| ds dt ≤ K h,
0 s 0 t−h
(6.38)
T s+h T s+h √
F(t), −wm (s) dtds ≤ K wm (s)V F(t)V∗ dtds ≤ c9 h.
0 s 0 s
By virtue of (6.34)–(6.38), uniformly with respect to m and ε

T
1
wm (s + h) − wm (s)2 ds ≤ O(h 4 ), (6.39)
0
which implies that the set {wm } is relatively compact in L 2 (0, T ; H ) (see Theorem
1.38). Therefore, by (6.30) and (6.39), there exists w and a subsequence {wm k } such
that
∗
wm k w in L ∞ (0, T ; H ),
wm k → w in L 2 (0, T ; H ), (6.40)
wm k w in L (0, T ; V) 2
when m k → ∞ and ε → 0 (note that for convenience we used subindex m instead

of m, ε).
On the other hand, putting v = M j=1 k j (t)v j , where k j (t) ∈ C [0, T ] and M-
1
positive integer, multiply (6.25) by k j (t) and sum for j = 1, · · · , M. Then, multiply
(6.25) by g jm (t) and sum for j = 1, · · · , m. Substituting the resulting equations, we
have
wm (t) + A(t)wm (t) + B(wm (t), wm (t)) + ∇Φε (wm ), v(t) − wm (t)
(6.41)
= F(t), v(t) − wm (t) .
Since
T
wm (t), v(t) − wm (t) dt =
0
T
1 1
v (t), v(t) − wm (t) dt − wm (T ) − v(T )2 + wm (0) − v(0)2 ,
0 2 2
we have
T
v (t) + A(t)wm (t) + B(wm (t), wm (t)) + ∇Φε (wm (t)) − F(t), v(t) − wm (t) dt
0 (6.42)
1 1
= − wm (0) − v(0)2 + wm (T ) − v(T )2 .
2 2

T
v (t) + A(t)wm (t) + B(wm (t), wm (t)) − F(t), v(t) − wm (t) dt
0
T (6.43)
1
+ Φε (v(t)) − Φε (wm (t)) dt ≥ − wm (0) − v(0)2 .
0 2
Since Φ(v) ≥ Φε (v) and Φ(Jε wm (t)) ≤ Φε (wm (t)) (see (1.37)), we have from
(6.43)
T
v (t) + A(t)wm (t) + B(wm (t), wm (t)) − F(t), v(t) − wm (t) dt
0
T (6.44)
1
+ Ψ (v) − Φ(Jε wm (t)) dt ≥ − wm (0) − v(0)2 .
0 2
By (6.32) it follows that Jε wm k w in L 2 (0, T ; V) as m k → ∞ and ε → 0, and

T T
lim inf Φ(Jε wm k (t)) dt ≥ Φ(w(t)) dt = Ψ (w). (6.45)
m k →∞,ε→0 0 0
It is not difficult to prove that

T T
B(wm k (t), wm k (t)), v(t) dt → B(w(t), w(t)), v(t) dt (6.46)
0 0
as m k → ∞ and ε → 0. We give a proof below for convenience of the readers.
T
( B(wm k (t), wm k (t)), v(t) − B(w(t), w(t)), v(t)) dt
0
T T
= B(wm k (t), wm k (t) − w(t)), v(t) dt + B(wm k (t) − w(t), w(t)), v(t) dt
0 0
≡ I1 + I2 .
(6.47)
By condition (2), Hölder’s inequality with exponents 2, 4, 4 and (6.40) we have
T 1 1
|I1 | ≤ K sup v(t)H1 wm k (t)H1 wm k (t) − w(t)H2 1 wm k (t) − w(t) 2 dt
t∈[0,T ] 0
1 1
≤ K sup v(t)H1 wm k L 2 (0,T ;H1 ) wm k − w L2 2 (0,T ;H1 ) wm k − w L2 2 (0,T ;L2 ) → 0.
t∈[0,T ]
(6.48)
By condition (2),

T 1 1
B(z(t), w(t)), v(t) dt ≤ K sup v(t)H1 z(t) L 2 (0,T ;H1 ) w L2 2 (0,T ;H1 ) w L2 2 (0,T ;L2 )
0 t∈[0,T ]
≤ K 1 z(t) L 2 (0,T ;H1 ) ,
which means that

T
z(t) ∈ L 2 (0, T ; V) → B(z(t), w(t)), v(t) dt
0
is a continuous linear on L 2 (0, T ; V), that is, there exists a f ∈ L 2 (0, T ; V)∗ such
that T
B(z(t), w(t)), v(t) dt = z(t), f (t)
0
and so by (6.40) it follows that
|I2 | → 0 as m k → ∞ and ε → 0. (6.49)
By (6.47)–(6.49) we get (6.46).

Since lim inf m k →∞,ε→0 A(t)wm k (t), wm k (t) ≥ A(t)w(t), w(t) , by (6.45) and
(6.46) we have from (6.44)
T
v (t) + A(t)w(t) + B(w(t), w(t)) − F(t), v(t) − w(t) dt + Ψ (v) − Ψ (w)
0
1
≥ − w(0) − v(0)2 .
2
(6.50)
3
1
On the other hand, B(w(t), w(t)) ≤ K w(t)V w(t) (cf. condition(2))
V∗
2
2
and w(t) ∈ L ∞ (0, T ; L2 (Ω)), and we can get that B(w(t), w(t)) ∗
4
M ∈ L (0, T ; V ).
3
Therefore, taking into account the fact that the set {v = i=1 k j (t)v j ; k j (t) ∈
C 1 [0, T ]} is dense in {L 4 (0, T ; V) : u ∈ L 2 (0, T ; V∗ )} (see Remark 1.8), we see
that (6.50) is valid for all v ∈ {L 4 (0, T ; V) : u ∈ L 2 (0, T ; V∗ )}.
From (6.30) we get estimates (6.24).
Remark 6.6 The operator A1 (t) in Problem I-VI is not symmetric and depends on
time t. Thus, unlike Theorem 1 of [2] we can not use eigenfunctions of a symmetric
operator as the basis for Galërkin method.
Next, let us study the following problem:
⎧ T
⎪
⎪ w (t) + A(t)w(t) + B(w(t), w(t)) − F(t), u(t) − w(t) dt + Ψ (u) − Ψ (w) ≥ 0
⎪
⎨ 0
⎪
⎪ ∀u ∈ L 4 (0, T ; V),
⎪
⎩
w(0) = w0 .
(6.51)
Condition (2) for B(w, w) in the following theorem is for 2-D problem corre-
sponding to Problem I (the case of total pressure).
(1) Let l = 2. For any u, v ∈ V, A(t)u, v and A (t)u, v are continuous with
respect to t and
A(t)u, u ≥ c1 u2V − c2 u2 ∃c1 > 0, c2 ≥ 0, ∀u ∈ V,

| A(t)u, v | ≤ c3 uV vV ∃c3 > 0, ∀u, v ∈ V, (6.52)
| A (t)u, v | ≤ c4 uV vV ∃c4 ≥ 0, ∀u, v ∈ V;
1 2
| B(w, v), z | ≤ K wH1 vL4 z 3 zH3 1 ,
B(w, v), v = 0;
(3) F, F ∈ L 2 (0, T ; V∗ );
(4) The functional Φ : V → R is proper, weakly lower semi-continuous and
Φ(u) ≥ 0 ∀u ∈ V, Φ(0V ) = 0 ;
(5) w0 ∈ V, Φ(w0 ) = 0 and (A(0)w0 + B(w0 , w0 ) − F(0)) ∈ H (a compatibility

condition at initial time).
Then, there exists a unique solution w to (6.51) such that
w ∈ C([0, T ]; V), w ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; H ), w(0) = w0 , (6.53)
and the estimates (6.24) and the following hold:
w (t)2 ≤ K 1 , w 2L 2 (0,T ;V) ≤ K 1 , (6.54)
where K 1 depends on A(0)w0 + B(w0 , w0 ) − F(0), w0 , F1 2L 2 (0,T ;V∗ ) ,
F1 2L 2 (0,T ;V∗ ) and is independent of Φ.
Proof Let {v j , j = 1, 2, · · · } be a basis of the space
V. Without loss of generality,
m
we assume that v1 = w0 . We find a solution wm = i=1 gim (t)vi to problem

wm (t), v j + A(t)wm (t), v j + B(wm (t), wm (t)), v j + ∇Φε (wm (t)), v j = F, v j ,
wm (0) = w0 .
(6.55)
By the same argument as in the proof of Theorem 6.6, there exists continuously
differentiable function gm on the interval [0, T ]. Also, as (6.30) in the proof of
Theorem 6.6 we have
wm (t)2 ≤ C1 (w0 2 + F2L 2 (0,T ;V∗ ) ),

(6.56)
wm 2L 2 (0,T ;V) ≤ C2 (w0 2 + F2L 2 (0,T ;V∗ ) ),
where C1 , C2 are independent of m, ε.

We are going to derive a priori estimates for functions wm . In the same way as in the
proof of Theorem 6.6 we see that there exists a solution gim (t) on the interval [0, T ].
Owing to F ∈ L 2 (0, T ; V∗ ), the function F, v j is absolute continuous with respect
to t. Therefore, by the Lipschitz continuity of ∇Φε and continuity of A (t)u, v with
respect to t, gim (t) is in fact absolute continuous.
Setting t = 0 in (6.55), multiplying by gim (0) and summing over i = 1, · · · , m,

we get
wm (0)2 + A(0)wm (0), wm (0) + B(wm (0), wm (0)), wm (0) + ∇Φε (w0 ), wm (0)
(6.57)
= F(0), wm (0) .
By conditions (4) and (5), for any u ∈ V we have Φ(u) ≥ Φ(w0 ), and so ∇Φε (w0 ) =
0. Thus, we have from (6.57)
wm (0) ≤ A(0)w0 + B(w0 , w0 ) − F(0). (6.58)
Differentiating (6.55) yields
wm (t), v j + A (t)wm (t), v j + A(t)wm (t), v j + B(wm (t), wm (t)), v j

(6.59)
+ B(wm (t), wm (t)), v j + (∇Φε (wm )) , v j = F , v j ,
where (∇Φε (wm )) = d

dt
(∇Φε (wm (t))). Multiplying (6.59) by gim (t) and summing
for i, we have
wm (t), wm (t) + A (t)wm (t), wm (t) + A(t)wm (t), wm (t)

(6.60)
+ B(wm (t), wm (t)), wm (t) + (∇Φε (wm )) , wm (t) = F , wm (t) ,
where the fact that B(wm , wm ), wm = 0 was used. Using monotonicity of ∇Φε , we
know that (∇Φε (wm )) , wm ≥ 0 (see (1.38)). By condition (2), (6.56) and Young’s
inequality with exponents 6/5, 6, we get
1 2
2| B(wm , wm ), wm | ≤ 2K wm V wm L4 wm 3 wm V3
5/3 1
≤ 2K wm V wm L4 wm 3
c1
≤ wm 2V + K 1 wm 6L4 wm 2 .
4
Thus, we have from (6.60)
d c4
wm (t)2 + c1 wm 2V + c1 wm 2V − c1 wm 2V − wm 2V
dt c1
c1 4 c1
≤ wm V + F1 V∗ + 2c2 wm + wm V + K 2 wm 2 wm 6L4 ,
2 2 2 2
4 c1 4
and so
d c1 c4 4
wm (t)2 + wm 2V ≤ wm 2V + F 2V∗ + (2c2 + K 2 wm 6L4 )wm 2 .
dt 2 c1 c1
(6.61)
Integrating (6.61) yields
t c1 c4 4
wm (t)2 + wm (s)2V ds ≤wm (0)2 + wm 2L 2 (0,T ;V) + F1 2L 2 (0,T ;V∗ )
0 2 c1 c1
t (6.62)

+ 2c2 + K 2 wm (s)6L4 wm (s)2 ds.
0
By (6.56),
wm ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; L2 (Ω))
and {wm } is bounded in the space above. Since V → Lq (Ω) for any 1 < q < ∞
when l = 2, by virtue of Theorems 1.12 and 1.33 we have that
wm ∈ L 6 (0, T ; L4 (Ω))
and {wm } is bounded in L 6 (0, T ; L4 (Ω)). Taking into account (6.56), (6.58) and
Gronwall’s inequality (see Theorem 1.51), we have from (6.62)
t
2c2 +K 2 wm (s)6L4 ds
wm (t)2 ≤ C4 e 0 , (6.63)
where C4 depends on Aw0 + B(w0 , w0 ) − F(0), w0 , F1 2L 2 (0,T ;V∗ ) ,
F1 2L 2 (0,T ;V∗ ) and is independent of ε, m, Φ.
From (6.56), (6.62) and (6.63) we have
T
wm (s)2V ds ≤ C5 , (6.64)
0
where C5 is similar to C4 .
Therefore, by (6.56), (6.63) and (6.64)
wm (t) ≤ const ∀t ∈ [0, T ], ∀m, ∀ε > 0;

wm L 2 (0,T ;V) ≤ const ∀m, ∀ε > 0;
(6.65)
wm (t) ≤ const ∀t ∈ [0, T ], ∀m, ∀ε > 0;
wm L 2 (0,T ;V) ≤ const ∀m, ∀ε > 0.
Extracting a subsequence and passing to limit in (6.55) as m → ∞, and then

letting ε → 0 and using (6.27), (6.45), similarly to the proof of Theorem 6.6 (from
(6.41)) we can prove the existence of a solution w ∈ L 2 (0, T ; V) to (6.51) satisfying
that w ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; H ).
From (6.56), (6.63) and (6.64), we get the estimates (6.24) and (6.54).
Let us prove uniqueness next.
Let w1 , w2 be two solutions asserted in Theorem 6.7. Then, taking into account (6.13)
and Remark 6.5, we have from (6.51)
w1 (t) + A(t)w1 (t) + B(w1 (t), w1 (t)) − F(t), w2 (t) − w1 (t)

+ Φ(w2 (t)) − Φ(w1 (t)) ≥ 0,
w2 (t) + A(t)w2 (t) + B(w2 (t), w2 (t)) − F(t), w1 (t) − w2 (t)
+ Φ(w1 (t)) − Φ(w2 (t)) ≥ 0,
which imply
w1 (t) − w2 (t), w1 (t) − w2 (t) + A(t)(w1 (t) − w2 (t)), w1 (t) − w2 (t) ≤

| B(w1 (t), w1 (t)) − B(w2 (t), w2 (t)), w1 (t) − w2 (t) |.
(6.66)
By virtue of conditions (1) and (2) we have
1 ∂(w1 (t) − w2 (t)

+ A(t)(w1 (t) − w2 (t)), w1 (t) − w2 (t)
2 ∂t
≤ |B(w1 (t), w1 (t)) − B(w2 (t), w2 (t)), w1 (t) − w2 (t)|,
1 ∂w1 (t) − w2 (t)2

+ c1 w1 (t) − w2 (t)2V
2 ∂t
5 1
≤ c2 w1 (t) − w2 (t)2 + w2 (t)L4 w1 (t) − w2 V3 w1 (t) − w2 3
≤ c2 (w1 (t) − w2 (t)2 + c1 w1 (t) − w2 2V + K w2 (t)6L4 w1 (t) − w2 2 .
(6.67)
We have from (6.67)
t
(w1 (t) − w2 (t) ≤ const
2
(c2 + K w2 (s)6L4 )w1 (s) − w2 (s)2 ds. (6.68)
0
Since w2 ∈ C([0, T ]; V), by Gronwall’s inequality we have that w1 (t) = w2 (t) for
all t ∈ [0, T ].
The following theorem is for small data in (6.51) corresponding to Problem I (the
case of total pressure).
(1) For any u, v ∈ V, A(t)u, v and A (t)u, v are continuous with respect to t and
A(t)u, u ≥ c1 u2V − c2 u2 , ∃c1 > 0, c2 ≥ 0, ∀u ∈ V,

| A(t)u, v | ≤ c3 uV vV ∃c3 > 0, ∀u, v ∈ V,
| A (t)u, v | ≤ c4 uV vV ∃c4 > 0, ∀u, v ∈ V;
1 1
B(w, v), u ≤ c5 wV vV u 2 uV2 ,
B(w, v), v = 0;
(3) F, F ∈ L 2 (0, T ; V∗ );
(4) The functional Φ : V → R is proper, weakly lower semi-continuous and
Φ(u) ≥ 0 ∀u ∈ V, Φ(0V ) = 0 ;
(5) w0 ∈ V, Φ(w0 ) = 0 and (A(0)w0 + B(w0 , w0 ) − F(0)) ∈ H (a compatibility

condition at the initial time).
If w0 V , ( A(0)w0 + B(w0 , w0 ) − F(0)) H and norms of F, F in the spaces
where they belong to are small enough, then there exists a unique solution w to (6.51)
such that
w ∈ C([0, T ]; V), w ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; H ), (6.69)
and (6.24), (6.54) hold.

Proof First, let us prove the existence of the asserted solution.
Let {v j , j = 1, 2, · · · } be a basis of the space
V. Without loss of generality, we
m
assume that v1 = w0 . We find a solution wm = i=1 gim (t)vi to problem

wm (t), v j + A(t)wm (t), v j + B(wm , wm ), v j + ∇Φε (wm ), v j = F, v j ,
wm (0) = w0 .
(6.70)
As (6.30) in the proof of Theorem 6.6 we get
wm (t)2 ≤ C1 (w0 2 + F2L 2 (0,T ;V∗ ) ),

(6.71)
wm 2L 2 (0,T ;V) ≤ C2 (w0 2 + F2L 2 (0,T ;V∗ ) ),
where C1 , C2 are independent of m and ε.

For a t˜m there exists a differentiable function gim (t) on [0, t˜m ] such that gim (t) is
absolute continuous (cf. proof of Theorem 6.7).
Multiplying (6.70) by gim (t) and summing over i = 1, · · · , m, yield
⎧
⎨ wm (t), wm (t) + A(t)wm (t), wm (t) + B(wm (t), wm (t)), wm (t)
⎪
+ ∇Φε (wm ), wm (t) = F, wm (t) , (6.72)
⎪
⎩
wm (0) = w0 .
We are going to derive a priori estimates for wm .

Setting t = 0 in (6.70), multiplying the resulting equation by gim (0) and adding on
i = 1, · · · , m, we get
wm (0)2 + A(0)wm (0), wm (0) + B(wm (0), wm (0)), wm (0)

(6.73)
+ ∇Φε (w0 ), wm (0) = F(0), wm (0) .
By conditions (4), (5) for any u ∈ V, Φ(u) ≥ Φ(w0 ), and so ∇Φε (w0 ) = 0. Thus,
we have from (6.73)
wm (0) ≤ A(0)w0 + B(w0 , w0 ) − F(0). (6.74)
On the other hand, taking into account (6.28), we have from (6.72)
c1 wm (t)2V ≤FV∗ wm (t)V +c2 wm (t)2 − (wm (t), wm (t)),
and so
c1 wm (t)V ≤ F L ∞ (0,T ;V∗ ) + α1 wm (t) + c2 α1 wm (t), (6.75)
where and in what follows of this section α1 is the one in · ≤ α1 · V .

wm (t), v j + A(t) wm (t), v j + A(t)wm (t), v j + B(wm (t), wm (t)), v j

(6.76)
+ B(wm (t), wm (t)), v j + (∇Φε (wm )) , v j = F , v j .
Multiplying (6.76) by gim (t), summing for i and observing the fact that (∇Φε (wm )) ,
wm ≥ 0 due to monotonicity of ∇Φε (see (1.38)), by conditions (2), (3) we have
d √ c1 2c2
wm (t)2 + (c1 − 2c5 α1 wm (t)V )wm 2V + c1 wm 2V − wm 2V − 4 wm 2V
dt 2 c1
c1 2
≤ wm V + F V∗ + 2c2 wm ,
2 2 2
2 c1
√
where | B(wm , wm ), wm | ≤ c5 α1 wm V wm 2V , B(wm , wm ), wm = 0 and
c1 2c2
2| A(t) wm (t), wm (t) | ≤ wm 2V + 4 wm 2V
2 c1
were used. Thus,
d √ 2 2c2
wm (t)2 + (c1 − 2c5 α1 wm (t)V )wm 2V ≤ F 2V∗ + 2c2 wm 2 + 4 wm 2V .
dt c1 c1
(6.77)
If wm (0)V = w0 V < c√

1
4c5 α1
, then
√ c1
c1 − 2c5 α1 w0 V ≥ , (6.78)
2
√
and there exists a tm such that c1 − 2c5 α1 wm (t)V > 0 for t ∈ [0, tm ]. Therefore,
we have from (6.74) and (6.77) that
wm (t)2 ≤Aw0 + B(w0 , w0 ) − F(0)2

t
2 T 2c2 T
+ F 2V∗ dt + 4 wm (s)2V ds + 2c2 wm (s)2 ds
c1 0 c1 0 0
for t ∈ [0, tm ], from which by Gronwall’s inequality and (6.71) it follows that
2c42 2
wm (t)2 ≤ Aw0 + B(w0 , w0 ) − F(0)2 + wm 2L 2 (0,T ;V) + F 2L 2 (0,T ;V∗ ) e2c2 T
c1 c1
2c2 C2
≤ Aw0 + B(w0 , w0 ) − F(0)2 + 4 w0 2 + F2L 2 (0,T ;V∗ )
c1
2
+ F 2L 2 (0,T ;V∗ ) e2c2 T .
c1
(6.79)
Using (6.71) and (6.79), we have from (6.75)
1
wm (t)V ≤ F L ∞ (0,T ;V∗ ) + (α1 wm (t) + c2 α1 wm (t))
c1

1 2
≤ F L ∞ (0,T ;V∗ ) + α1 Aw0 + B(w0 , w0 ) − F(0) + α1 F L 2 (0,T ;V∗ ) ec2 T
c1 c1

2c42 C2 c2 T
+ c2 α1 C1 + α1 e (w0 + F L 2 (0,T ;V∗ ) ) .
c1
(6.80)
Putting

1 2
β≡ F L ∞ (0,T ;V∗ ) + α1 Aw0 + B(w0 , w0 ) − F(0) + α1 F L 2 (0,T ;V∗ ) ec2 T
c1 c1

2c42 C2 c2 T
+ c2 α1 C1 + α1 e (w0 + F L 2 (0,T ;V∗ ) ) ,
c1
(6.81)
we have from (6.80)
wm (t)V ≤ β ∀t ∈ [0, tm ). (6.82)
By the assumption of theorem, we can assume that β is small so that

√ c1
c1 − 2c5 α1 β ≥ . (6.83)
2
Now, let us prove that if w(0)V ≤ c√

1
4c5 α1
and β satisfies (6.83), then for any m
√
c1 − 2c5 α1 wm (t)V > 0 ∀t ∈ [0, T ]. (6.84)
To this end, let us assume that there exists an m and t m (≤ T ) such that
√
c1 − 2c5 α1 wm (t)V > 0 ∀t ∈ [0, t m ),
√ (6.85)
c1 − 2c5 α1 wm (t m )V = 0.
Then, by (6.82) and (6.83), we have

√ c1
c1 − 2c5 α1 wm (t)V > ∀t ∈ [0, t m ), (6.86)
2
which is a contradiction to the second formula of (6.85), and so we get (6.84). From
(6.83) we have (6.82) with tm = T .
Therefore, by (6.79), (6.83), (6.77) and (6.86) we have that
wm (t)V ≤ const ∀t ∈ [0, T ], ∀m, ∀ε > 0,

wm (t) ≤ const ∀t ∈ [0, T ], ∀m, ∀ε > 0, (6.87)
wm L 2 (0,T ;V) ≤ const ∀m, ∀ε > 0.
Extracting a subsequence and passing to limit in (6.70) as m → ∞, and then ε → 0

and using (6.27), (6.45), in a similar way as in the proof of Theorem 6.6 (from (6.41))
we can prove the existence of a solution w to (6.51) such that w ∈ L 2 (0, T ; V) ∩
L ∞ (0, T ; H ), and so w ∈ C([0, T ]; V).
By the same argument as in Theorem 6.7 we get the estimates for solutions.
It only remains to prove the uniqueness of solution to the problem.
Let w1 , w2 be two solutions asserted in Theorem 6.7. Then, as (6.67) we have from
(6.51)
1 ∂w1 (t) − w2 (t)2

+ c1 w1 (t) − w2 (t)2V
2 ∂t
3 1
≤ c2 w1 (t) − w2 (t)2 + w2 (t)V w1 (t) − w2 V2 w1 (t) − w2 2
≤ c2 w1 (t) − w2 (t)2 + c1 w1 (t) − w2 2V + K w2 (t)4V w1 (t) − w2 2 .
(6.88)
Equation (6.88) implies
t
w1 (t) − w2 (t) ≤ const
2
(c2 + K w2 (s)4V )w1 (s) − w2 (s)2 ds,
0
which by Gronwall’s inequality deduces w1 (t) = w2 (t) for all t ∈ [0, T ].
Remark 6.7 Following Theorem 3.7 of [1], to study existence of a unique solution
to a variational inequality under smallness condition of data in Theorem 2 of [2] the
author assumed that F ∈ L ∞ (0, T ; V∗ ) ∩ L 1 (0, T ; H ), F ∈ L 1 (0, T ; H ).
The right-hand side of (6.15) for the definition of F2 includes some boundary
integrals:
− 2μ(k(x)U, u)Γ2 − 2μ(SŨ , ũ)Γ3 − 2(α(x)Uτ , u)Γ5

− μ(k(x)U, u)Γ7 + φi , u n Γi + φi , uΓi ∀u ∈ V.
i=2,4,7 i=3,5,6
1
For fixed t let us assume that φi (t) ∈ H 2 (Γi ), i = 2, 4, 7. Due to u τ |Γi = 0,
u · n ∈ H − 2 (Γi ), u · n H − 21 (Γ ) ≤ K u H
1
(see Proposition 2.1), and so
| φi , u n Γi | ≤ φi H 21 (Γ ) u n H − 21 (Γ ) ≤ K φi H 21 (Γ ) u H ≤ K 1 u H .

i i i
Then, there exist elements f i ∈ H such that φi , u n Γi = ( f i , u). Thus, for φi ∈
1
L 2 (0, T ; H 2 (Γi )), i = 2, 4, 7, there exist f i ∈ L 2 (0, T ; H ) such that
φi , u n Γi = f i , u H ∀u ∈ V.
i=2,4,7 i=2,4,7
But, for other boundary integrals we can not find such functions belonging to
L 2 (0, T ; H ).
Therefore, the assumption such that F ∈ L 1 (0, T ; H ), F ∈ L 1 (0, T ; H ) is not

suitable for our problems with mixed boundary conditions, and in Theorems 6.8 and
6.9 we assume that F ∈ L 2 (0, T ; V∗ ), F ∈ L 2 (0, T ; V∗ ).
The following theorem is for small data in (6.51) corresponding to Problem II

(the case of static pressure).
(1) For any u, v ∈ V, A(t)u, v, A (t)u, v are continuous with respect to t and
A(t)u, u ≥ c1 u2V − c2 u2 ∃c1 > 0, c2 ≥ 0, ∀u ∈ V,

| A(t)u, v | ≤ c3 uV vV ∃c3 > 0, ∀u, v ∈ V,
| A (t)u, v | ≤ c1 α2 uV vV ∃α2 (0 ≤ α2 < 1), ∀u, v ∈ V;
c5 1 1
B(w, v), u ≤ √ wV vV u 2 uV2 ;
α1
(3) F, F ∈ L 2 (0, T ; V∗ );
(4) The functional Φ : V → R is proper lower weak semi-continuous and
Φ(u) ≥ 0 ∀u ∈ V, Φ(0V ) = 0 ;
(5) w0 ∈ V, Φ(w0 ) = 0 and (A(0)w0 + B(w0 , w0 ) − F(0)) ∈ H (a compatibil-

ity condition at the initial time).
If w0 V , ( A(0)w0 + B(w0 , w0 ) − F(0)) H and norms of F, F in the spaces
that they belong to are small enough, then there exists a unique solution v to problem
(6.51) such that
w ∈ C([0, T ]; V), w ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; H ). (6.89)
Proof First, let us prove the existence of the asserted solution.
Let {v j , j = 1, 2, · · · } be a basis
of V. Without loss of generality, we assume that
m
v1 = w0 . We find a solution wm = i=1 gim (t)vi to problem

wm (t), v j + A(t)wm (t), v j + B(wm , wm ), v j + ∇Φε (wm ), v j = F, v j ,
wm (0) = w0 ,
(6.90)
which gives us an ordinary differential system for gim (t), i = 1, · · · , m. In the same
way as in Theorem 6.7 we see that for a t˜m > 0 there exists a solution gim (t) on the
interval [0, t˜m ] and gim (t) is absolute continuous. If wm (t) is bounded, then gim (t)
can be extended over t˜m . Under smallness of data we will find estimate for wm (t)
below, by which we can say that t˜m = T.
Multiplying (6.90) by gim (t) and adding for i = 1, · · · , m, yield
⎧
⎨ wm (t), wm (t) + A(t)wm (t), wm (t) + B(wm , wm ), wm (t) + ∇Φε (wm ), wm (t)
⎪
= F, wm (t) ,
⎪
⎩
wm (0) = w0 .
(6.91)
We will find a priori estimates for wm (t)2 + wm (t)2 .

By virtue of (6.28), we have from (6.91)
d
wm (t)2 + 2c1 wm (t)2V − 2c5 wm (t)3V
dt
1
≤ F2V∗ + c1 (1 − α2 )wm (t)2V + 2c2 wm (t)2 ,
c1 (1 − α2 )
and so
d
wm (t)2 + c1 − 2c5 wm (t)V wm (t)2V + c1 α2 wm (t)2V
dt
1
≤ F2 2V∗ + 2c2 wm (t)2 .
c1 (1 − α2 )
(6.92)
Setting t = 0 in (6.90), multiplying the resulting equation by gim (0) and adding for
i = 1, · · · , m, we get
wm (0)2 + A(0)wm (0), wm (0) + B(wm (0), wm (0)), wm (0) + ∇Φε (w0 ), wm (0)
(6.93)
= F(0), wm (0) .
By conditions (4), (5) we have that for any u ∈ V, Φ(u) ≥ Φ(w0 ), and so ∇Φε (w0 ) =
0. Thus, we have from (6.91)
wm (0) ≤ A(0)w0 + B(w0 , w0 ) − F(0). (6.94)
On the other hand, taking into account (6.28), we have from (6.91)
c1 wm (t)2V ≤ F(t)V∗ wm (t)V + c5 wm (t)3V + c2 wm (t)2 − (wm (t), wm (t)),
and so
c1 wm (t)V ≤ F(t)V∗ + c5 wm (t)2V + (α1 wm (t) + c2 α1 wm (t)), (6.95)
where α1 is in · ≤ α1 · V .
wm (t), v j + A (t)wm (t), v j + A(t)wm (t), v j + B(wm (t), wm (t)), v j

(6.96)
+ B(wm (t), wm (t)), v j + (∇Φε (wm )) , v j = F , v j .
Multiplying (6.96) by gim (t), summing for i and observing the fact that (∇Φε (wm )) ,
wm ≥ 0 (see (1.38)), by conditions (2), (3) we have
d
w (t)2 + (c1 − 4c5 wm (t)V )wm (t)2V + c1 wm (t)2V
dt m
− c1 α2 wm (t)2V − c1 α2 wm (t)2V
1
≤ c1 (1 − α2 )wm (t)2V + F (t)2V∗ + 2c2 wm (t)2 ,
c1 (1 − α2 )
where | B(wm , wm ), wm + B(wm , wm ), wm | ≤ 2c5 wm V wm 2V was used. Thus,
d
w (t)2 + (c1 −4c5 wm (t)V )wm (t)2V − c1 α2 wm (t)2V
dt m
(6.97)
1
≤ F (t)2V∗ + 2c2 wm (t)2 .
c1 (1 − α2 )
Adding (6.92) and (6.97) yields
d
wm (t)2 + wm (t)2 + (c1 − 4c5 wm (t)V ) wm (t)2V + wm (t)2V
dt
1
≤ F(t)2V∗ + F (t)2V∗ + 2c2 wm (t)2 + wm (t)2 .
c1 (1 − α2 )
(6.98)
If wm (0)V = w0 V < 4cc15 , then there exists a tm such that c1 − 4c5 wm (t)V ≥
0 for t ∈ [0, tm ]. Therefore, we from (6.94) and (6.98) have

wm (t)2 + wm (t)2 ≤ α1 w0 2 + Aw0 + B(w0 , w0 ) − F(0)2
T
1
+ F2V∗ + F 2V∗ dt
c1 (1 − α2 ) 0
t

+ 2c2 wm (s)2 + wm (s)2 ds
0
for t ∈ [0, tm ], from which by Gronwall’s inequality we get

wm (t)2 + wm (t)2 ≤ α1 w0 2V + Aw0 + B(w0 , w0 ) − F(0)2
1
+ F2L 2 (0,T ;V∗ ) + F 2L 2 (0,T ;V∗ ) e2c2 T .
c1 (1 − α2 )
(6.99)
Using the estimate, we will obtain a quadratic inequality satisfied by wm (t)V .
Put
√
β ≡F L ∞ (0,T ;V∗ ) + max{α1 , c2 α1 } 2 α1 w0 2V + Aw0 + B(w0 , w0 ) − F(0)2
1 1
2
+ F2L 2 (0,T ;V∗ ) + F 2L 2 (0,T ;V∗ ) ec2 T .
c1 (1 − α2 )
(6.100)
Taking into account
√ 1
wm (t) + wm (t) ≤ 2 wm (t)2 + wm (t)2 2 ,
we have from (6.99)

√
wm (t) + wm (t) ≤ 2 α1 w0 2V + Aw0 + B(w0 , w0 ) − F(0)2
1 1/2
+ F2L 2 (0,T ;V∗ ) + F 2L 2 (0,T ;V∗ ) ec2 T .
c1 (1 − α2 )
(6.101)
Taking into account (6.101) and (6.100), we have from (6.95) a quadratic inequality
for wm (t)V
0 ≤ β − c1 wm (t)V + c5 wm (t)2V ∀t ∈ [0, tm ], (6.102)
which is the one we want.

By the assumption of theorem, we can assume that β is small so that
4c12
c12 − 4c5 β > . (6.103)
9
√
c1 − c12 −4c5 β
Now, let us prove that if w(0)V ≤ 2c5
(< 6cc15 ), then for any m, ε
c1
c1 − 4c5 wm (t)V > ∀t ∈ [0, T ]. (6.104)
4
Taking into account
c1
c1 − 4c5 wm (0)V > ,
3
let us assume that there exists an m and t m (≤ T ) such that

c1 c1
c1 − 4c5 wm (t)V > ∀t ∈ [0, t m ) and c1 − 4c5 wm (t m )V = . (6.105)
4 4
Setting y = wm (t)V in (6.102), we get
0 ≤ β − c1 y + c5 y 2 ∀t ∈ [0, t m ]. (6.106)
By virtue of (6.103), there exist two real roots of quadratic polynomial β − c1 y +

c5 y 2
c1 − c12 − 4c5 β c1 + c12 − 4c5 β
y1 = and y2 = ,
2c5 2c5
and on the intervals [0, y1 ] and [y2 , +∞) (6.106) holds. Thus, by the continuity
of wm (t)V with respect to t we have from w(0)V ∈ [0, y1 ] that wm (t)V ∈
[0, y1 ] ∀t ∈ [0, t m ), that is,

c1 − c12 − 4c5 β c1
wm (t)V ≤ < ∀t ∈ [0, t m ].
2c5 6c5
Thus,
c1
c1 − 4c5 wm (t)V > ∀t ∈ [0, t m ], (6.107)
3
which is a contradiction to (6.105), and so we get (6.104).
Therefore, by (6.98), (6.99) and (6.104) we have
wm (t) ≤ const ∀t ∈ [0, T ], ∀m, ∀ε > 0;

wm L 2 (0,T ;V) ≤ const ∀m, ∀ε > 0;
(6.108)
wm (t) ≤ const ∀t ∈ [0, T ], ∀m, ∀ε > 0;
wm L 2 (0,T ;V) ≤ const ∀m, ∀ε > 0.
Extracting a subsequence, passing to limit in (6.41) as m → ∞ and then letting ε → 0

and using (6.27) and (6.45), by a way similar to the proof of Theorem 6.6 we can prove
the existence of a solution w to (6.51) such that w ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; H ),
and so w ∈ C([0, T ]; V).
As in Theorem 6.7 we can prove uniqueness.
Remark 6.8 In Sect. 4.2 the non-steady Navier-Stokes equations with boundary
condition (5.2) with Γi = ∅, i = 8, 9, 10, 11, is considered. For the proof of exis-
tence of a unique solution, the following property of the linear problem obtained by
neglecting the nonlinear term B(w, w) from

w (t) + Aw + B(w, w) = f (t),
u(0) = ϕ,
was essential.
Property: (Lemma 4.6) When X = {w ∈ L 2 (0, T ; V); w ∈ L 2 (0, T ; V), w ∈
L 2 (0, T ; V∗ )} and Y = {w ∈ L 2 (0, T ; V∗ ); w ∈ L 2 (0, T ; V∗ )}, a map u → {u (0),
Lu ≡ u + Au} is linear continuous one-to-one from X onto H × Y .
But, in the case of Problem II in this chapter, owing to the boundary conditions
on Γi (= ∅), i = 8, 9, 10, 11, neglecting the nonlinear term B(w, w), we only get a
nonlinear problem
⎧ ∂w
⎪
⎨ ∂t , u − w + Aw, u − w + Φ(u) − Φ(w) ≥ f, u − w
⎪
⎪
⎪ for a.e. t ∈ (0, T ) ∀u ∈ V,
⎩
w(0) = w0 ∈ V,
which seems not to have the property above (cf. Proposition 6.1), and so we can not
use the method in Sect. 4.2.
6.3 Solutions to the Non-steady Navier-Stokes Problems
6.3.1 Existence of a Solution: The Case of Total Pressure
Let us study Problem I-VI.

Theorem 6.10 Under Assumption 6.1 there exists a solution to Problem I-VI (the
non-steady Navier-Stokes problem for the case of total pressure) for any f, φi , i =
2 ∼ 7 and initial function v0 such that v0 − U (0) ∈ H . Moreover the following
estimate for the solution holds:
v(t) − U (t) ≤ K ∀t ∈ [0, T ], v − U L 2 (0,T ;V) ≤ K , (6.109)
where
∂U
K = v0 − U (0) + 2 ∗ + U L 2 (0,T ;H1 ) + U L 2 (0,T ;H1 ) U L ∞ (0,T ;H1 )
∂t L (0,T ;V )
+ f L 2 (0,T ;V∗ ) + φi 1 + φi 1 .
L 2 (0,T ;H − 2 (Γi )) L 2 (0,T ;H− 2 (Γi ))
i=2,4,7 i=3,5,6
Proof By Assumption 6.1 we can prove that F1 ∈ L 2 (0, T ; V∗ ) and F1 ∈ L 2

(0, T ; V∗ ).
Define an operator A1 (t) : V → V∗ by
A1 (t)v, u = a01 (t, v, u) ∀v, u ∈ V, (6.110)
where a01 (t, v, u) is the one in (6.4). Then, A1 (t)v, u is continuous with respect to
t.
2Σi j εi j (z)2 ≥ C1 z2H1 (Ω) , C1 > 0. (6.111)
Since rot U (t) × w, w = 0, by Young’s inequality and Assumption 6.1, we have
μC1
| rot U (t) × w, w + rot w × U (t), w| ≤ w2H1 (Ω) + C2 w2 ,
C2 > 0
4
(6.112)
for any t ∈ [0, T ]. On the other hand, by virtue of Remark 3.4 and Assumption 6.1
there exists a constant M such that
S(x)∞ , k(x)∞ , α L ∞ (Γ5 ) ≤ M.
Therefore, there exists C3 such that

μ(k(x)z, z)Γ + 2μ(S z̃, z̃)Γ + (α(x)z, z)Γ + μ(k(x)z, z)Γ ≤ μC1 z2 1 + C3 z2
2 3 5 7
4 H (Ω)
∀z ∈ V
(6.113)
(see Theorem 1.27). Setting c2 = C2 + C3 , we have therefore from (6.110)–(6.113)

μC1
A1 (t)z, z ≥ c1 z2V − c2 z2 , c1 = > 0, ∀z ∈ V. (6.114)
2
By Hölder’s inequality and Assumption 6.1, we have
| rot U (t) × u, v + rot u × U (t), v| ≤ K uV vV ∀u, v ∈ V (6.115)
for any t ∈ [0, T ]. Also, we can prove that
| A1 (t)u, v | ≤ K uV vV ∀u, v ∈ V. (6.116)
Let us prove that
lim inf A1 (t)u m , u m ≥ A1 (t)u, u as u m u in V. (6.117)

m→∞
By (6.4) we have
A1 (t)u m , u m ≡ a01 (t, u m , u m )

= 2μ(E (u m ), E (u m )) + rot U (t) × u m , u m + rot u m × U (t), u m
+ 2μ(k(x)u m , u m )Γ2 + 2μ(S ũ m , ũ m )Γ3 + 2(α(x)u m , u m )Γ5 + μ(k(x)u m , u m )Γ7 .
(6.118)
Taking into account
lim inf (E(u m ), E(u m )) ≥ (E(u), E(u)) as u m u in V

m→∞
6.3 Solutions to the Non-steady Navier-Stokes Problems 217
(see (1.1)), from (6.118) we get (6.117).
Define a bilinear operator B1 : V × V → V∗ by
B1 (z, v), u = a11 (z, v, u) ∀z, v, u ∈ V, (6.119)
where a11 (t, v, u) is the one in (6.4). Then, by a property of mixed production of
vectors,
B1 (w, v), v = rot w × v, v = 0. (6.120)
Using Hölder’s inequality, imbedding theorems and interpolation inequality vL3 ≤

1 1
K vL2 2 vL2 6 , we have
| B1 (w, v), z | = | rot w × v, z| ≤ K rot wL2 vL3 zL6

1 1 (6.121)
≤ K wV v 2 vV2 zV .
By the definition of the functional Φ, we know Φ(u) ≥ 0 and Φ(0V ) = 0.

Let F1 be the one in (6.4). Then, by Assumption 6.1 we can prove
∂U
F1 L 2 (0,T ;V∗ ) ≤M1 2 ∗ + U L 2 (0,T ;H1 ) + U L 2 (0,T ;H1 ) |U L ∞ (0,T ;H1 )
∂t L (0,T ;V )
+ f L 2 (0,T ;V∗ ) + φi 1 + φi 1 .
L 2 (0,T ;H − 2 (Γi )) L 2 (0,T ;H− 2 (Γi ))
i=2,4,7 i=3,5,6
(6.122)
By virtue of (6.114), (6.116) and (6.120)–(6.122), we can verify that all conditions
of Theorem 6.6 are satisfied, and we come to the conclusion.
(1) In addition to Assumption 6.1, ∂t∂ U ∈ W 1,2 (0, T ; H1 (Ω));

f, f ∈ L 2 (0, T ; V∗ ), φi ∈W 1,2 (0, T ; H − 2 (Γi )), i = 2, 4, 7,
1
(2) and
φi ∈ W 1,2 (0, T ; H− 2 (Γi )), i = 3, 5, 6;
1
(3) w0 ≡ v0 − U ∈ V, w0 |∪i=8−11 Γi = 0 and
(A1 (0)w0 + B1 (w0 , w0 ) − F1 (0)) ∈ H, (6.123)
where F1 is the one in (6.4) and A1 , B1 are in (6.110), (6.119), respectively.

If l = 2, then there exists a unique solution to Problem I-VI’ satisfying (6.109)
and
v − U ≡ w ∈ C([0, T ]; V), w ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; H ). (6.124)
In the case of l = 3, if w0 V , A1 w0 + B1 (w0 , w0 ) − F1 (0), U W 1,2 (0,T ;H1 ) ,

∂t∂ 2 U L 2 (0,T ;(H1 )∗ ) and norms of f, f , φi in the spaces of condition (2) are small
2
enough, then there exists a unique solution v to Problem I-VI’ satisfying (6.124) and
(6.109).
Proof We know that the first two of condition (6.52) of Theorem 6.7 are satisfied
(see (6.114), (6.116)). By (6.110), we know that
∂a01 (t, v, u) ∂U (t) ∂U (t)

A1 (t)v, u = = rot × v, u + rot v × ,u ,
∂t ∂t ∂t
which show that A1 (t)v, u is Lipschitz continuous with respect to t. Also,
∂U (t)
| A1 (t)v, u | ≤ rot 2 vV uV + rot vL2 ∂U (t) 1 uV
∂t L ∂t H
and the third one of condition (6.52) of Theorem 6.7 is satisfied since
∂ ∂
U (t) 1 ≤ K
U (t) 1,2 .
∂t C([0,T ];H ) ∂t W (0,T ;H1 )
By Hölder’s inequality and interpolation inequality, we have

1 2 (6.125)
≤ K wV vL4 z 3 zV3
for l = 2 and

1 1 (6.126)
≤ K wV vV z 2 zV2
for l = 3. Since
∂U
F1 , u = − , u − 2μ(E(U ), E(u)) − rot U × U, u
∂t
− 2μ(k(x)U, u)Γ2 − 2μ(SŨ , ũ)Γ3 − 2(α(x)Uτ , u)Γ5
− μ(k(x)U, u)Γ7 + f, u + φi , u n Γi + φi , uΓi ,
i=2,4,7 i=3,5,6
by the conditions we have

F1 , F1 ∈ L 2 (0, T ; V∗ ).
It is easy to verify that condition (4) of Theorem 6.7 is satisfied. By the condition
w0 |∪i=8−11 Γi = 0 and (5.22), we know Φ(w0 ) = 0, which shows that condition (5) of
Theorem 6.7 is satisfied.
Thus, in the case of 2-D by Theorem 6.7 we come to the conclusion.
If ∂U
∂t L 2 (0,T ;(H )∗ )
1 , U L 2 (0,T ;H1 ) are small and U ∈ C(0, T ; H1 ), then under
smallness conditions of the theorem for f, φi we can know that F1 L 2 (0,T ;V∗ ) is small.
Also if ∂∂tU2 L 2 (0,T ;(H1 )∗ ) , ∂t∂ U L 2 (0,T ;H1 ) are small and U ∈ C(0, T ; H1 ), then
2
under smallness conditions of the theorem for f, φi we can know that F1 L 2 (0,T ;V∗ )
is small. Thus, if U W 1,2 (0,T ;H1 ) , ∂t∂ 2 U L 2 (0,T ;(H1 )∗ ) and norms of f, f , φi in
2
the spaces of condition (2) are small enough, then norms of F1 , F1 in the space
L 2 (0, T ; V∗ ) are small. Therefore, in case of 3-D by Theorem 6.8 we come to the
conclusion.
Remark 6.9 From (6.109) we know that the estimates for solutions are independent
of the thresholds of friction conditions gτ , gn , g+n , g−n .
6.3.2 Existence of a Unique Solution: The Case of Static

Pressure
Let us study now Problem II-VI.

(1) Assumption 6.1 and dU dt

∈ W 1,2 (0, T ; H1 (Ω)) hold;
(2) f, f ∈ L 2 (0, T ; V∗ ), φi ∈ W 1,2 (0, T ; H − 2 (Γi )), i = 2, 4, 7, and φi ∈
1
W 1,2 (0, T ; H− 2 (Γi )), i = 3, 5, 6;

1
(3) w0 ≡ v0 − U ∈ V and w0 |∪i=8−11 Γi = 0 and
(A2 (0)w0 + B2 (w0 , w0 ) − F2 (0)) ∈ H, (6.127)
where A2 , B2 and F2 are the one, respectively, (6.17) and (6.15).

If w0 V , (A2 (0)w0 + B2 (w0 , w0 ) − F2 (0)) , U W 1,2 (0,T ;H1 ) , U W 1,2 (0,T ;H1 )
and norms of f, f , φi in the spaces that they belong to are small enough, then there
exists a unique solution v to Problem II-VI such that
v − U ≡ w ∈ C([0, T ]; V), w ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; H ). (6.128)
Proof First, let us prove the existence of the desired solution.

The operator A2 (t) : V → V∗ was defined in (6.17) by
A2 (t)v, u = a02 (t, v, u) ∀v, u ∈ V. (6.129)
Then, A2 (t)v, u is continuous with respect to t. Note the operator A2 (t) is not
symmetric. By Korn’s inequality
2Σi j εi j (z)2 ≥ C1 z2H1 (Ω) , C1 > 0. (6.130)
If U C([0,T ;H1 ) is small enough, then

μC1
| (U (t) · ∇)w, u + (w · ∇)U (t), u| ≤ wH1 (Ω) uH1 (Ω) . (6.131)
4
By (6.130), (6.131) and (6.113), we have
μC1
A2 (t)z, z ≥ c1 z2V − c2 z2 c1 = > 0, c2 > 0 ∀z ∈ V. (6.132)
2
Using (6.130), we can prove that
| A2 (t)u, v | ≤ c3 uV vV ∀u, v ∈ V. (6.133)
By (6.129), we know
∂U (t) ∂U (t)
A2 (t)w, u = · ∇ w, u + (w · ∇) ,u ,
∂t ∂t
which shows that A2 (t)w, u is continuous with respect to t.

Also, U (t)C([0,T ;H1 ) is small enough, then

∂U (t) ∂U (t)
| A2 (t)w, u | = · ∇ w, u + (w · ∇) ,u
∂t ∂t (6.134)
≤ c1 αwV uV , α < 1.
The bilinear operator B2 : V × V → V∗ was defined in (6.17) by
B2 (w, v), u = a12 (z, v, u) ∀w, v, u ∈ V. (6.135)
Then, by virtue of Hölder’s inequality, imbedding theorems and interpolation inequal-

1
1
ity uL3 ≤ K u 2 uL2 6 we have
| B2 (w, v), u | = | (w · ∇v), u| ≤ wL6 ∇vL2 uL3

1 1 (6.136)
≤ K wV vV u 2 uV2 .
Since
∂U
F2 , u = − , u − 2μ(E(U ), E(u)) − rot U × U, u
∂t
− 2μ(k(x)U, u)Γ2 − 2μ(SŨ , ũ)Γ3 − 2(α(x)U, u)Γ5
− μ(k(x)U, u)Γ7 + f, u + φi , u n Γi + φi , uΓi ,
i=2,4,7 i=3,5,6
by Conditions (1), (2) we have

F2 , F2 ∈ L 2 (0, T ; V∗ ). (6.137)
As in the proof of Theorem 6.11 we know that under conditions of the theorem
F2 L 2 (0,T ;V∗ ) and F2 L 2 (0,T ;V∗ ) are also small enough.
As in Theorem 6.11, we can verify that other conditions of Theorem 6.9 are
satisfied as well. Therefore, by Theorem 6.9 we come to the asserted conclusion.
6.3.3 Existence of a Unique Solution: The Stokes Problem
To study Problem III-VI, we use the following known result. To state it we need to
introduce a problem first.
Let A be a linear continuous and symmetric operator from V to V∗ satisfying the
coercivity condition
Au, u + αu ≥ ωu2V ω > 0, α ∈ R.
We find w ∈ L 2 (0, T ; V) ∩ C([0, T ]; H ) ∩ W 1,2 (0, T ; V∗ ) such that

⎧ ∂w
⎪
⎪ , u − w + Aw, u − w + Φ(u) − Φ(w) ≥ f, u − w
⎨ ∂t
for a.e. t ∈ (0, T ) ∀u ∈ V, (6.138)
⎪
⎪
⎩
w(0) = w0 ∈ V,
where f ∈ L 2 (0, T ; V∗ ).
Proposition 6.1 (Theorem 5.1 of [3]) Let f ∈ L 2 (0, T ; V∗ ) and w0 ∈ V be such
that
{Aw0 + ∂Φ(w0 ) − f (0)} ∩ H = ∅. (6.139)
Then, problem (6.138) has a unique solution w ∈ W 1,2 ([0, T ]; V)∩W 1,∞ ([0, T ]; H )
and the map (w0 , f ) → w is Lipschitz from H × L 2 (0, T ; V∗ ) to C([0, T ]; H ) ∩
L 2 (0, T ; V). If f ∈ W 1,2 ([0, T ]; V∗ ) and Φ(w0 ) < ∞, then problem (6.138) has a
unique solution w ∈ W 1,2 ([0, T ]; H ) ∩ Cw ([0, T ]; V).

(1) U ∈ L 2 (0, T ; H1 ), ∂U
∂t
∈ L 2 (0, T ; (H1 )∗ );
∗ 1
(2) f ∈ L (0, T ; V ), φi ∈ L 2 (0, T ; H 2 (Γi )), i = 2, 4, 7, and φi ∈ L 2 (0, T ;
2
− 21
H (Γi )), i = 3, 5, 6;
(3) w0 ≡ v0 − U ∈ V and w0 |∪i=8−11 Γi = 0 and
(A3 w0 − F3 (0)) ∈ H, (6.140)
where A3 and F3 are, respectively, as in (6.22) and (6.20).

Then, there exists a unique solution v to Problem III-VI for the non-steady Stokes
problem with mixed boundary condition (5.2) and
v − U C([0,T ];L2 )∩L 2 (0,T ;H1 )

∂U
≤ K w0 + U L 2 (0,T ;H1 ) + 2 1 ∗ + f L 2 (0,T ;V∗ )
∂t L (0,T ;(H ) )
+ φi L 2 (0,T ;H − 21 (Γ ) + φi L 2 (0,T ;H− 21 (Γ ) .
i i
i=2,4,7 i=3,5,6
(6.141)
If v1 , v2 are solutions, respectively, to Problem-III-VI corresponding to data
v01 , gτ 1 , gn1 , g+n1 , g−n1 , f 1 , h i1 , φi1 and v02 , gτ 2 , gn2 , g+n2 , g−n2 , f 2 , h i2 , φi2 , then
v1 −v2 C([0,T ];L2 )∩L 2 (0,T ;H1 )

≤ K v01 − v02 V + U1 − U2 C([0,T ];L2 ) + U1 − U2 L 2 (0,T ;H1 )
∂U1 − U2
+ L 2 (0,T ;(H1 )∗ ) + f 1 − f 2 L 2 (0,T ;V∗ ) + gτ 1 − gτ 2 Lτ2 (Γ8 )
∂t
+ gn1 − gn2 L 2 (Γ9 ) + g+n1 − g+n2 L 2 (Γ10 ) + g−n1 − g−n2 L 2 (Γ10 )
+ φi1 − φi2 L 2 (0,T ;H − 21 (Γ )) + φi1 − φi2 L 2 (0,T ;H− 21 (Γ )) ,
i i
i=2,4,7 i=3,5,6
(6.142)
j
where U j , j = 1, 2, are the functions in Assumption 6.1 with h i instead of h i .
Proof By (6.111) and (6.113), we have
A3 w, w ≥ c1 w2V − c2 w2 , c1 > 0, c2 > 0. (6.143)
Also, we can easily prove
| A3 w, u| ≤ c3 wV uV . (6.144)
By definition of F3 in (6.20), we know
∂U
F3 L 2 (0,T ;V∗ ) ≤c 2 1 ∗ + U L 2 (0,T ;(H1 )∗ ) + f L 2 (0,T ;V∗ )
∂t L (0,T ;(H ) )
+ φi L 2 (0,T ;H − 21 (Γ )) + φi L 2 (0,T ;H− 21 (Γ )) .
i i
i=2,4,7 i=3,5,6
(6.145)
By the condition w0 |∪i=8−11 Γi = 0, we know Φ(w0 ) = 0. Since Φ(u) ≥ 0, we have
that ∂Φ(w0 ) = 0. Thus, taking into account Remark 6.5, by Proposition 6.1 we have
existence of a unique solution and estimate (6.141).
If v1 = w1 + U1 , v2 = w2 + U2 are solutions corresponding to the given data, we
get
∂w
1
+ A3 w1 , u − w1 + Φ1 (u) − Φ1 (w1 ) ≥ F31 , u − w1 ,
∂t
∂w (6.146)
2
+ A3 w2 , u − w2 + Φ2 (u) − Φ2 (w2 ) ≥ F32 , u − w2 ,
∂t
where Φ j (u), F3 j , j = 1, 2, are one corresponding to U j , gτ j , gn j , g+n j , g−n j , f j ,

j j
h i , φi . Putting u = w2 and u = w1 , respectively, in the first and second one of
(6.146), and adding those, we have
∂(w − w )
1 2
+ A3 (w1 − w2 ), w2 − w1 + Φ1 (w2 ) − Φ1 (w1 ) + Φ2 (w1 ) − Φ2 (w2 )
∂t (6.147)
≥ F31 − F32 , w2 − w1 .
∂w1 − w2 2
+ 2c1 w1 − w2 2V ≤ F31 − F32 2V∗ + w2 − w1 2V
∂t
+ 2|Φ1 (w2 ) − Φ1 (w1 ) + Φ2 (w1 ) − Φ2 (w2 )| + 2c2 w1 − w2 2 .
(6.148)
Since w1 , w2 ∈ K (Ω),

Γ8 Γ9

+ g+n1 (w2n − w1n ) ds − g−n1 (w2n − w1n ) ds,
Γ10 Γ11

Γ8 Γ9

+ g+n2 (w2n − w1n ) ds − g−n2 (w2n − w1n ) ds.
Γ10 Γ11
(6.149)
Subtracting two formulas in (6.149) yields
c1
|Φ1 (w2 ) − Φ1 (w1 ) + Φ2 (w1 ) − Φ2 (w2 )| ≤ G 12 + w2 − w1 2V , (6.150)
2
where
2
G 12 = gτ 1 − gτ 2 2L2 (Γ8 ) + gn1 − gn2 2L 2 (Γ9 )
c1 τ
+ g+n1 − g+n2 2L 2 (Γ10 ) + g−n1 − g−n2 2L 2 (Γ11 ) .
By (6.148) and (6.150) we have


c1 t
w1 (t) − w2 (t)2 + w1 − w2 2V ≤
2 0
t t
w10 − w20 2 + F31 (s) − F32 (s)2V∗ ds + G 12 t + 2c2 w1 (s) − w2 (s)2 ds.
0 0
(6.151)
By Gronwall’s inequality, we have from (6.151)
T
w1 (t) − w2 (t)2 ≤ w10 − w20 2 + F31 (s) − F32 (s)2V∗ ds + G 12 T e2c2 T ,
0
T
w1 (t) − w2 (t)2L 2 (0,T ;V) ≤ K w10 − w20 2 + F31 (s) − F32 (s)2V∗ ds + G 12 .
0
(6.152)
Also F31 − F32 V∗ is estimated as (6.145), and so from (6.152) we get estimate
(6.142).
The content of Chap. 6 is taken from [4]. One mistake in [4] is corrected here.
In [5–7] the non-steady Stokes equations with the homogeneous Dirichlet bound-
ary condition and leak boundary condition were studied.
In [8] the existence of a weak solution and a local-in-time strong solution to the
non-steady Navier-Stokes problem was studied when boundary consists of a portion
with homogeneous Dirichlet boundary condition and another portion with the leak
condition based on the total stress. In [2] a variational inequality for the Navier-
Stokes problem with homogeneous Dirichlet boundary condition and one-sided leak
condition based on the total stress was considered. Also, in [9] a variational inequality
for the Navier-Stokes problems with homogeneous Dirichlet boundary condition and
one-sided boundary conditions based on total pressure was studied.
When boundary consists of separated portions with homogeneous Dirichlet
boundary condition and with nonlinear slip condition or leak condition, in [10] exis-
tence of a unique strong solution to the non-steady Navier-Stokes problem was stud-
ied. In the case of nonlinear slip condition, existence of a unique strong solution to
2-D problem and a unique local-in-time strong solution to 3-D problem was proved.
In the case of leak condition, existence of a unique local-in-time strong solution to
3-D problem was proved.
The semi-discrete finite element approximation to the time-dependent 2-D Navier-
Stokes equations with mixture of homogeneous Dirichlet boundary condition and
Tresca slip boundary conditions was discussed in [11].
For other kinds of non-steady fluid equations with mixture of friction slip boundary
conditions and Dirichlet condition, we refer to [12–15].
References 225
References
1. R. Temam, Navier-Stokes Equations (North-Holland, Amsterdam, 1985)

Pure Appl. 124(4), 107–125 (1980)
3. V. Barbu, Nonlinear Differential Equations of Monotone Types in Banach Spaces (Springer,
Berlin, 2010)
4. T. Kim, F. Huang, The non-steady Navier-Stokes systems with mixed boundary conditions
including friction conditions. Methods Appl. Anal. 25, 13–50 (2018)
Comput. Math. 19, 1–8 (2001)
6. H. Fujita, Variational inequalities and nonlinear semi-groups applied to certain nonlinear
problems for the Stokes equation. RIMS Kokyuroku 1234, 70–85 (2001)
Comput. Appl. Math. 149, 57–69 (2002)
8. R. An, Y. Li, K. Li, Solvability of Navier-Stokes equations with leak boundary conditions.
9. D.S. Konovalova, Subdifferential boundary value problems for the nonstationary Navier-
Stokes and Stokes equations. Differ. Equ. 36, 875–885 (2000)
10. T. Kashiwabara, On a strong solution of the non-stationary Navier-Stokes equations under
slip or leak boundary conditions of friction. J. Differ. Equ. 254, 756–778 (2013)
11. Y. Li, R. An, Semi-discrete stabilized finite element methods for Navier-Stokes equations
with nonlinear slip boundary conditions based on regularization procedure. Numer. Math.
117, 1–36 (2011)
12. M. Boukrouche, I. Boussetouan, L. Paoli, Non-isothermal Navier-Stokes system with mixed
boundary conditions and friction law: uniqueness and regularity properties. Nonlinear Anal.
102, 168–185 (2014)
13. M. Boukrouche, I. Boussentoua, L. Paoli, Existence for non-isothermal fluid flows with
Tresca’s friction and Cattaneo’s heat law. J. Math. Anal. Appl. 427, 499–514 (2015)
14. M. Boukrouche, G. Łukaszewicz, On global in time dynamics of a planar Bingham flow
subject to a subdifferential boundary conditions. Discret. Contin. Dyn. Syst. 34, 3969–3983
(2014)
15. J.K. Djoko, P.A. Razafimandimby, Analysis of the Brinkman-Forchheimer equations with
slip boundary conditions. Appl. Anal. 93, 1477–1494 (2014)
Chapter 7
The Steady Boussinesq System
In this chapter we are concerned with the steady Boussinesq system with mixed
boundary conditions. The boundary conditions for fluid may include Tresca slip, leak
condition, one-sided leak conditions, velocity, pressure, vorticity, stress together and
the conditions for temperature may include Dirichlet, Neumann and Robin conditions
together. We will get variational formulations consisting of a variational inequality
for velocity and a variational equation for temperature, which are equivalent to the
original PDE problems for smooth solutions.
Then, we will study the existence of solutions to the variational problems. For
the problem with boundary conditions involving the static pressure and stress, it
is proved that if the data of problem are small enough, then there exists a unique
solution. For the problem with boundary conditions involving the total pressure and
total stress, the existence of a solution is proved as well without smallness of the
data.
7.1 Problems and Variational Formulations
Let us consider the steady Boussinesq system

⎧
⎨ − 2∇ · μ(θ )E(v) + (v · ∇)v + ∇ p = (1 − α0 θ ) f,
⎪
div v = 0, (7.1)
⎪
⎩
− ∇ · κ(θ )∇θ + v · ∇(γ (θ )θ ) = g
https://doi.org/10.1007/978-3-030-78659-5_7
228 7 The Steady Boussinesq System
under mixed boundary conditions. Here v, p and θ are, respectively, velocity, pressure
and temperature, and α0 —parameter for buoyancy effect, f —body force, g—heat
source, κ(θ )—thermal conductivity, γ (θ )—specific heat of the fluid. The viscosity,
thermal conductivity and specific heat of fluid depend on the temperature.
Let Ω be a bounded domain of Rl , l = 2, 3. ∂Ω ∈ C 0,1 ,
∂Ω = ∪i=1
11
Γi = ΓD ∪Γ R

and Γ D ∩ Γ R = ∅, Γi ∩ Γ j = ∅ for i = j, Γi = j Γi j , where Γi j are connected
open subsets of ∂Ω, Γi j ∈ C 2 for i = 2, 3, 7 and Γi j ∈ C 1 for others. (See Remark
6.1.)
For temperature we are concerned with the boundary conditions
(1) θ |Γ D = 0,
∂θ
(2) κ(θ ) + β(x)θ = g R (x), β(x), g R (x) − given functions on Γ R .
ΓR
∂n
(7.2)
According to boundary conditions for fluid, the Problems I and II are distin-
guished. Problem I is the one with the following boundary conditions (the case of
static pressure)
(1) v|Γ1 = 0,
(2) vτ |Γ2 = 0, − p|Γ2 = φ2 ,
(3) vn |Γ3 = 0, rot v × n|Γ3 = φ3 /μ(θ ),
(4) vτ |Γ4 = 0, (− p + 2μ(θ )εnn (v))|Γ4 = φ4 ,
(5) vn |Γ5 = 0, 2(μ(θ)εnτ (v) + αvτ )|Γ5 = φ5 , α : a matrix,
(6) (− pn + 2μ(θ )εn (v))|Γ6 = φ6 ,
∂v
(7) vτ |Γ7 = 0, (− p + μ(θ ) · n)|Γ7 = φ7 ,
∂n
(8) vn = 0, |στ (θ, v)| ≤ gτ , στ (θ, v) · vτ + gτ |vτ | = 0 on Γ8 ,
(9) vτ = 0, |σn (θ, v, p)| ≤ gn , σn (θ, v, p)vn + gn |vn | = 0 on Γ9 ,
(10) vτ = 0, vn ≥ 0, σn (θ, v, p) + g+n ≥ 0, (σn (θ, v, p) + g+n )vn = 0 on Γ10 ,
(11) vτ = 0, vn ≤ 0, σn (θ, v, p) − g−n ≤ 0, (σn (θ, v, p) − g−n )vn = 0 on Γ11 ,
(7.3)
and Problem II is the one with the following boundary conditions (the case of total
pressure)
7.1 Problems and Variational Formulations 229
(1) v|Γ1 = 0,
(2) vτ |Γ2 = 0, −( p + 1/2|v|2 )|Γ2 = φ2 ,
(3) vn |Γ3 = 0, rot v × n|Γ3 = φ3 /μ(θ),
(4) vτ |Γ4 = 0, (− p − 1/2|v|2 + 2μ(θ)εnn (v))|Γ4 = φ4 ,
(5) vn |Γ5 = 0, 2(μ(θ)εnτ (v) + αvτ )|Γ5 = φ5 , α : a matrix,
(6) (− pn − 1/2|v|2 n + 2μ(θ)εn (v))|Γ6 = φ6 , (7.4)
∂v
(7) vτ |Γ7 = 0, (− p − 1/2|v| + μ(θ)
2
· n)|Γ7 = φ7 ,
∂n
(8) vn = 0, |στt (θ, v)| ≤ gτ , στt (θ, v) · vτ + gτ |vτ | = 0 on Γ8 ,
(9) vτ = 0, |σnt (θ, v, p)| ≤ gn , σnt (θ, v, p)vn + gn |vn | = 0 on Γ9 ,
(10) vτ = 0, vn ≥ 0, σnt (θ, v, p) + g+n ≥ 0, (σnt (θ, v, p) + g+n )vn = 0 on Γ10 ,
(11) vτ = 0, vn ≤ 0, σnt (θ, v, p) − g−n ≤ 0, (σnt (θ, v, p) − g−n )vn = 0 on Γ11 .
Difference between (7.3) and (5.2), or (7.4) and (5.3), is that μ in (7.3) and (7.4)
depends on θ .
We use the following assumption.
Assumption 7.1 Assume the followings.
(1) Γ2 j , Γ3 j and Γ7 j are convex. Γ1 = ∅, Γ D = ∅ and

Γ R ⊂ ∪i=1,3,5,8 Γi . (7.5)
(2) (3) of Assumption 5.1 holds.

(3) For the functions of (7.1) f ∈ L3 (Ω), g ∈ L 6/5 (Ω) and
μ ∈ C(R), 0 < μ0 ≤ μ(ξ ) ≤ μ1 < ∞ ∀ξ ∈ R;

κ ∈ C(R), 0 < κ0 ≤ κ(ξ ) ≤ κ1 < ∞ ∀ξ ∈ R; (7.6)
γ ∈ C(R), |γ (ξ )| ≤ γ0 ∀ξ ∈ R.
(4) For the functions of (7.2), (7.3) and (7.4),
g R ∈ L 4/3 (Γ R );
β0 ≥ β(x) ≥ 0, β0 − a constant, β(x) − measurable;
(7.7)
φi ∈ H − 2 (Γi ), i = 2, 4, 7, φi ∈ H− 2 (Γi ), i = 3, 5, 6;
1 1
the matrix α is positive, αi j ∈ L ∞ (Γ5 ).
Remark 7.1 For the first part of (2) in Assumption 7.1 we refer to Remark 6.3.
In this chapter this assumption is also necessary to guarantee equivalence between
Problem I-VE and Problem I-VI, and between Problem II-VE and Problem II-VI.
Assuming that f ∈ L2 (Ω), g ∈ L 2 (Ω), g R ∈ L 2 (Γ R ), φi ∈ L 2 (Γi ), i = 2, 4, 7,

and φi ∈ L2 (Γi ), i = 3, 5, 6, formally we introduce the following
Definition 7.1 A function (v, p, θ ) ∈ H2 (Ω) × H 1 (Ω) × H 2 (Ω) is called a solu-

tion to Problem I (or Problem II) if the first two equations and the last equation of
(7.1), respectively, hold in L2 (Ω) and L 2 (Ω), and each of (7.2) and (7.3) (or (7.4)),
respectively, holds in L 2 (Γ D ), L 2 (Γ R ), L 2 (Γi ) and L2 (Γi ).
WΓ1,2
D
(Ω) = y ∈ W 1,2 (Ω) : y|Γ D = 0 .
Since Γ1 = ∅ and Γ D = ∅, by Korn’s and Poincaré’s inequalities we use
(v, u)V = (E(v), E(u)), (y, z)WΓ1,2 (Ω) = (∇ y, ∇ z). (7.8)

D
As in Chap. 5, we will get variational formulations for Problems I, II.

Integrating by parts, taking into account (2.79), and applying Theorems 3.1 and
3.2 on Γi j (i = 2, 3, 7), for v ∈ H2 (Ω) ∩ V, θ ∈ H 2 (Ω) and u ∈ V we have
−2 ∇ · (μ(θ)E (v)), u = 2(μ(θ)E (v), E (u)) − 2(μ(θ)E (v)n, u)∪11

i=2 Γi
= 2(μ(θ)E (v), E (u)) + 2(μ(θ)k(x)v, u)Γ2 − (μ(θ)rot v × n, u)Γ3

+ 2(μ(θ)S ṽ, ũ)Γ3 − 2(μ(θ)εnn (v), u n )Γ4 − 2(μ(θ)εnτ (v), u)Γ5
∂v
− 2(μ(θ)εn (v), u)Γ6 − μ(θ) , u + (μ(θ)k(x)v, u)Γ7 − 2(μ(θ)εnτ (v), u)Γ8
∂n Γ7
− 2(μ(θ)εnn (v), u n )Γ9 − 2(μ(θ)εnn (v), u n )Γ10 − 2(μ(θ)εnn (v), u n )Γ11 .
(7.9)
For p ∈ H 1 (Ω) and u ∈ V we have
(∇ p, u) = ( p, u n )∪i=2
11
Γi = ( p, u n )Γ2 + ( p, u n )Γ4 ∪Γ7 ∪Γ9 ∪Γ10 ∪Γ11 + ( pn, u)Γ6 ,
(7.10)
where u n |Γ3 ∪Γ5 ∪Γ8 = 0 was used. For θ ∈ H 2 (Ω) and ϕ ∈ WΓ1,2 D
(Ω), by (7.2) we
have
∂θ
−∇ · (κ(θ)∇θ), ϕ = (κ(θ)∇θ, ∇ϕ) − κ(θ) , ϕ Γ = (κ(θ)∇θ, ∇ϕ) + (βθ − g R , ϕ)Γ R .
∂n R
(7.11)
By (7.5), vn = 0 on Γ R , and so for v ∈ V, θ ∈ H 1 (Ω) and ϕ ∈ WΓ1,2
D
(Ω) we have
v · ∇(γ (θ )θ ), ϕ = (vn γ (θ )θ, ϕ)Γ R − (γ (θ )θ v, ∇ϕ) = −(γ (θ )θ v, ∇ϕ).

(7.12)
By (7.9)–(7.12), we can see that solutions in the sense of Definition 7.1 (v, p, θ ) of
the problem (7.1), (7.2), (7.3) satisfy the following (cf. Remark 3.9).
⎧
⎪
⎪ 2(μ(θ)E(v), E(u)) + (v · ∇)v, u + 2(μ(θ)k(x)v, u)Γ2
⎪
⎪
⎪
⎪ + 2(μ(θ)S ṽ, ũ)Γ3 + 2(α(x)v, u)Γ5 + (μ(θ)k(x)v, u)Γ7
⎪
⎪
⎪
⎪
⎪
⎪ − 2(μ(θ)εnτ (v), u)Γ8 + ( p − 2μ(θ )εnn (v), u n )Γ9 ∪Γ10 ∪Γ11
⎪
⎪
⎪
⎪
⎪
⎪ = (1 − α0 θ ) f, u + φi , u n Γi + φi , uΓi ∀u ∈ V,
⎪
⎪
⎨ i=2,4,7 i=3,5,6
⎪
⎪
⎪ (κ(θ)∇θ, ∇ϕ) − (γ (θ)θv, ∇ϕ) + (βθ, ϕ)Γ R = g R , ϕΓ R + g, ϕ ∀ϕ ∈ WΓ1,2 (Ω),
⎪
⎪
D
⎪
⎪ |στ (θ, v)| ≤ gτ , στ (θ, v) · vτ + gτ |vτ | = 0 on Γ8 ,
⎪
⎪
⎪
⎪
⎪
⎪ |σn (θ, v, p)| ≤ gn , σn (θ, v, p)vn + gn |vn | = 0 on Γ9 ,
⎪
⎪
⎪
⎪
⎪
⎪ σ n (θ, v, p) + g+n ≥ 0, (σn (θ, v, p) + g+n )vn = 0 on Γ10 ,
⎪
⎩
σn (θ, v, p) − g−n ≤ 0, (σn (θ, v, p) − g−n )vn = 0 on Γ11 .
(7.13)
Define a0 (θ ; ·, ·), a1 (·, ·, ·) and f 1 ∈ V∗ by
a0 (θ ; w, u) = 2(μ(θ )E(w), E(u)) + 2(μ(θ )k(x)w, u)Γ2 + 2(μ(θ )S w̃, ũ)Γ3

+ 2(α(x)w, u)Γ5 + (μ(θ )k(x)w, u)Γ7 ∀w, u ∈ V, θ ∈ W 1,2 (Ω),
a1 (v, w, u) = (v · ∇)w, u ∀v, w, u ∈ V,

f 1 , u = φi , u n Γi + φi , uΓi ∀u ∈ V.
i=2,4,7 i=3,5,6
(7.14)
Define b0 (θ̃; ·, ·) and g1 ∈ (WΓ1,2
D
(Ω))∗ by
b0 (θ̃; θ, ϕ) = (κ(θ̃)∇θ, ∇ϕ) + (β(x)θ, ϕ)Γ R ∀θ̃ , θ ∈ W 1,2 (Ω), ∀ϕ ∈ WΓ1,2

D
(Ω),
g1 , ϕ = g R , ϕΓ R + g, ϕ ∀ϕ ∈ WΓ1,2
D
(Ω).
(7.15)
Then, taking into account στ (θ, v) = 2μ(θ )εnτ (v), σn (θ, v, p) = − p + 2μ(θ )
εnn (v) and (7.13), we introduce the following variational formulation for the problem
(7.1)–(7.3).
Problem I-VE. Find (v, θ, στ , σn , σ+n , σ−n ) ∈ K (Ω) × WΓ1,2
D
(Ω) × L2τ (Γ8 )×
−1/2 −1/2
L (Γ9 ) × H
2
(Γ10 ) × H (Γ11 ) such that
⎧
⎪ a0 (θ ; v, u) + a1 (v, v, u) − (στ , u τ )Γ8 − (σn , u n )Γ9
⎪
⎪
⎪
⎪ − σ+n , u n Γ10 − σ−n , u n Γ11 − f − α0 θ f, u = f 1 , u ∀u ∈ V,
⎪
⎪
⎪
⎪
⎪ 1,2
⎨ b0 (θ ; θ, ϕ) − γ (θ )θ v, ∇ϕ = g1 , ϕ ∀ϕ ∈ WΓ D (Ω),
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ σ+n + g+n ≥ 0, σ+n + g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎩
σ−n − g−n ≤ 0, σ−n − g−n , vn Γ11 = 0 on Γ11 ,
where L2τ (Γ8 ) is the subspace of L2 (Γ8 ) consisting of functions such that
(u, n)L2 (Γ8 ) = 0.
Remark 7.2 Under (4) of Assumption 7.1 the duality products f 1 , u of (7.14) is
meaningful (see Remark 5.1).
Theorem 7.1 Let Assumption 7.1 hold. If (v, p, θ ) is a solution in the sense of
Definition 7.1 of the problem (7.1)–(7.3), then (v, θ, στ |Γ8 , σn |Γ9 , σn |Γ10 , σn |Γ11 ) is a
solution to Problem I-VE.
Conversely, if f ∈ L2 (Ω) and Problem I-VE has a smooth solution (v, θ, στ , σn ,
σ+ , σ− ) such that v ∈ H2 (Ω), θ ∈ H 2 (Ω), σ+ ∈ L 2 (Γ10 ) and σ− ∈ L 2 (Γ11 ), then
there exists p ∈ H 1 (Ω) such that (v, p, θ ) is a solution to the problem (7.1)–(7.3).
Proof From the problem (7.1)–(7.3) we deduced Problem I-VE, and thus it is suf-
fices to prove conversion from Problem I-VE to the problem (7.1)–(7.3). By Theorem
5.1 there exists a p such that (v, p) satisfies (7.1) and the boundary condition (7.3),
and p is unique under the additional condition above. In a routine way (see Sect.
1, Ch. 2 of [1]) we can prove that θ satisfies (7.1) and the boundary condition (7.2).
We will find another variational formulation consisting of a variational inequality

and a variational equation, which is equivalent to Problem I-VE.
Let (v, θ, στ , σn , σ+n , σ−n ) be a solution of Problem I-VE. Subtracting the first
formula of (7.16) with u = v from the first formula of (7.16), we get
a0 (θ; v, u − v) + a1 (v, v, u − v) − (στ , u τ − vτ )Γ8 − (σn , u n − vn )Γ9 − σ+n , u n − vn Γ10

− σ−n , u n − vn Γ11 − f − α0 θ f, u − v = f 1 , u − v ∀u ∈ V.
(7.17)
Let Φ : V → R be the functional defined by (5.22).Then, Φ is proper, convex,

weakly lower semi-continuous and nonnegative. By Theorem 5.2, under Assumption
7.1 for fixed θ the problem
⎧
⎪
⎪
⎪
⎪
⎪ − σ+n , u n Γ10 − σ−n , u n Γ11 − f − α0 θ f, u = f 1 , u,
⎪
⎨
(7.18)
⎪
⎪
⎪
⎪
⎪ σ+n + g+n ≥ 0, σ+n + g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎩
σ−n − g−n ≤ 0, σ−n − g−n , vn Γ11 = 0 on Γ11
is equivalent to the following variational inequality.

Find v ∈ V such that
a0 (θ ; v, u − v) + a1 (v, v, u − v)+Φ(u) − Φ(v) − f − α0 θ f, u − v
(7.19)
≥ f 1 , u − v ∀u ∈ V.
Therefore, we have the following variational formulation equivalent to Problem I-VE
which consists of a variational inequality for velocity and a variational equation for
temperature.
Problem I-VI. Find (v, θ ) ∈ V × WΓ1,2

D
(Ω) such that
⎧
⎪
⎨ a0 (θ ; v, u − v) + a1 (v, v, u − v) + Φ(u) − Φ(v) − f − α0 θ f, u − v
≥ f 1 , u − v ∀u ∈ V, (7.20)
⎪
⎩ b (θ ; θ, ϕ) − γ (θ )θ v, ∇ϕ = g , ϕ ∀ϕ ∈ W 1,2 (Ω).
0 1 ΓD
Taking (v · ∇)v = rot v × v + 21 grad|v|2 into account, by (7.9)–(7.12) we can see

that solutions in the sense of Definition 7.1 (v, p, θ ) of problem (7.1), (7.2), (7.4)
satisfy the following.
⎧
⎪ 2(μ(θ )E(v), E(u)) + rotv × v, u + 2(μ(θ )k(x)v, u)Γ2
⎪
⎪
⎪
⎪
⎪
⎪ + 2(μ(θ )S ṽ, ũ)Γ3 + 2(α(x)v, u)Γ5 + (μ(θ )k(x)v, u)Γ7
⎪
⎪
⎪
⎪
⎪
⎪
1
− 2(μ(θ )εnτ (v), u)Γ8 + ( p + |v|2 − 2μ(θ )εnn (v), u n )Γ9 ∪Γ10 ∪Γ11
⎪
⎪
⎪
⎪ 2
⎪
⎪
⎪
⎪ = (1 − α0 θ ) f, u + φi , u n Γi + φi , uΓi ∀u ∈ V,
⎪
⎪
⎨ i=2,4,7 i=3,5,6
(7.21)
⎪
⎪ (κ(θ )∇θ, ∇ϕ) − (γ (θ )θ v, ∇ϕ) + (βθ, ϕ)Γ R = g R , ϕΓ R + g, ϕ ∀ϕ ∈ WΓ1,2 (Ω),
⎪
⎪
⎪
⎪
D
⎪ t
⎪
⎪
⎪ |στ (θ, v)| ≤ gτ , στt (θ, v) · vτ + gτ |vτ | = 0 on Γ8 ,
⎪
⎪
⎪
⎪
⎪
⎪ |σnt (θ, v, p)| ≤ gn , σnt (θ, v, p)vn + gn |vn | = 0 on Γ9 ,
⎪
⎪
⎪ σ t (θ, v, p) + g ≥ 0, (σ t (θ, v, p) + g )v = 0 on Γ ,
⎪
⎪
⎪ n +n n +n n 10
⎪
⎩ t
σn (θ, v, p) − g−n ≤ 0, (σnt (θ, v, p) − g−n )vn = 0 on Γ11 .
Define a2 (·, ·, ·) by
a2 (v, u, w) = rot v × u, w ∀v, u, w ∈ V. (7.22)
1
στt (θ, v) = 2μ(θ )εnτ (v), σnt (θ, v, p) = −( p + |v|2 ) + 2μ(θ )εnn (v)
2
(7.2) (7.4).
Problem II-VE. Find (v, θ, στt , σnt , σ+n t
, σ−n
t
) ∈ K (Ω) × WΓ1,2
D
(Ω) × L2τ (Γ8 )×
L 2 (Γ9 ) × H −1/2 (Γ10 ) × H −1/2 (Γ11 ) such that
⎧
⎪ a0 (θ ; v, u) + a2 (v, v, u) − (στt , u τ )Γ8 − (σnt , u n )Γ9
⎪
⎪
⎪
⎪ − σ+n t
, u n Γ10 − σ−n
t
, u n Γ11 − f − α0 θ f, u = f 1 , u ∀u ∈ V,
⎪
⎪
⎪
⎪
⎪
⎪ b0 (θ ; θ, ϕ) − γ (θ )θ v, ∇ϕ = g1 , ϕ ∀ϕ ∈ WΓ1,2 (Ω),
⎪
⎨ D

⎪
⎪
⎪
⎪ |σn | ≤ gn ,
⎪ σnt vn + gn |vn | = 0 on Γ9 ,
t
⎪
⎪
⎪
⎪ σ+n
t
+ g+n ≥ 0, σ+n
t
+ g+n , vn = 0 on Γ10 ,
⎪
⎪ Γ10
⎪
⎩ t
σ−n − g−n ≤ 0, σ−n
t
− g−n , vn Γ11
= 0 on Γ11 .
Relying on Theorem 5.3, in the same way as in Theorem 7.1 we have

Theorem 7.2 Let Assumption 7.1 hold. If (v, p, θ ) is a solution in the sense of
Definition 7.1 of the problem (7.1), (7.2), (7.4), then (v, θ, στt |Γ8 , σnt |Γ9 , σnt |Γ10 , σnt |Γ11 )
is a solution to Problem I-VE.
Conversely, if f ∈ L2 (Ω) and Problem I-VE has a smooth solution (v, θ, στt , σnt ,
σ+ , σ−t ) such that v ∈ H2 (Ω), θ ∈ H 2 (Ω), σ+t ∈ L 2 (Γ10 ) and σ−t ∈ L 2 (Γ11 ), then
t
there exists p ∈ H 1 (Ω) such that (v, p, θ ) is a solution to the problem (7.1), (7.1),
(7.4). Moreover, if at least one of the sets Γi , i = 2, 4, 6, 7, is nonempty, then p is
unique.
Then, using Theorem 5.4, in the same way as used for Problem I-VI we get
Problem II-VI equivalent to Problem II-VE which consists of a variational inequality
for velocity and a variational equation for temperature.
Problem II-VI. Find (v, θ ) ∈ V × WΓ1,2

D
(Ω) such that
⎧
⎨ a0 (θ ; v, u − v) + a2 (v, v, u − v) + Φ(u) − Φ(v) − f − α0 θ f, u − v
⎪
≥ f 1 , u − v ∀u ∈ V, (7.24)
⎪
⎩ 1,2
b0 (θ ; θ, ϕ) − γ (θ )θ v, ∇ϕ = g1 , ϕ ∀ϕ ∈ WΓ D (Ω).
where a2 (·, ·, ·) is the one in (7.22).

In view of the results above, we will study Problem I-VI and Problem II-VI in the
following sections.

Static Pressure
The aim of this section is to prove the following

Theorem 7.3 Under Assumption 7.1 assume that f, φi , i = 2, · · · , 7, g, g R are
small enough (depending on α0 ) in the spaces in (3), (4) of Assumption 7.1 (see
(7.51)). Then, there exists a solution (v, θ ) to Problem I-VI such that
μ0
vV ≤ ,
K (7.25)
θ W 1,2 (Ω) ≤ c g R L 4/3 (Γ R ) + g L 6/5 (Ω) ,
where K is the one in (7.26) below.

If μ(θ ), κ(θ ) and γ (θ ) are independent of θ and f L3 is small enough, then
the solution satisfying vV ≤ c, θ WΓ1,2 (Ω) ≤ c for a constant c small enough is
D
unique.
7.2.1 Existence of a Solution to an Auxiliary Problem
Since
|a1 (v, v, u)| = | (v · ∇)v, u | ≤ K v2V uV ∀v, u ∈ V, (7.26)
define a 1 (v) ∈ V∗ by
a 1 (v), u = a1 (v, v, u) ∀v, u ∈ V.
Define γε (t) by
γ (t)t
γε (t) := t ∈ R, ε > 0.
(1 + ε|γ (t)|)(1 + ε|t|)
Then,
1
|γε (t)| ≤ , |γε (t)| ≤ |γ (t)||t| ≤ γ0 |t|, γε (t) → γ (t)t as ε → 0. (7.27)
ε2
For every ε > 0, let Φε be the Moreau-Yosida approximation of Φ and ∇Φε be

Fréchet derivative of Φε .
We first consider an auxiliary problem involving two parameters δ, ζ concerned
with the norm of velocity (which is useful when there is fluid flux across a portion
of boundary), one parameter λ concerned with the norm of temperature (which is
useful to deal with buoyancy effect) and a parameter ε for approximation.
Problem I-VIA. Let δ > 0, ζ > 0, λ > 0 and ε > 0. Find (v, θ ) ∈ V × WΓ1,2 D
(Ω)
such that
⎧ δ
⎪
⎪ a0 (θ ; v, u) + max{δ, a (v) ∗ } a1 (v, v, u) + ∇Φε (v), u
⎪
⎪
⎪
⎪ 1 V
⎨ λ
− 1− α0 θ f, u = f 1 , u ∀u ∈ V, (7.28)
⎪
⎪ max{λ, θ }
⎪
⎪
⎪
⎪ ζ
⎩ b0 (θ ; θ, ϕ) − γε (θ )v, ∇ϕ = g1 , ϕ ∀ϕ ∈ WΓ1,2 (Ω).
max{ζ, vV } D
Theorem 7.4 There exists a solution (vε , θε ) ∈ V × WΓ1,2

D
(Ω) to Problem I-VIA.
Proof Let H = V × WΓ1,2

D
(Ω). Define an operator A : H → H ∗ by
δ
A (v, θ ), (u, φ) =a0 (θ ; v, u) + a1 (v, v, u) + ∇Φε (v), u
max{δ, a 1 (v)V∗ }
α0 λ
− 1− θ f, u + b0 (θ ; θ, φ)
max{λ, θ }
ζ
− γε (θ )v, ∇φ ∀(v, θ ), (u, φ) ∈ H .
max{ζ, vV }
(7.29)
Let us first check that operator A is well-defined. By the definition of Φε , Φε (0V ) = 0
and ∇Φε (0V ) = 0. Since ∇Φε is Lipschitz continuous with the constant ε−1 (see
Remark 1.16),
∇Φε (v), u = ∇Φε (v) − ∇Φε (0V ), u ≤ ε−1 vV uV . (7.30)
By (7.26) we have
δ
a1 (v, v, u) ≤ δuV . (7.31)
max{δ, a 1 (v)V∗ }
Also, by Hölder’s inequality we have

α0 λ
θ f, u ≤ cα0 λ f L3 uV . (7.32)
max{λ, θ }
By (7.27) we have
ζ cζ
γε (θ )v, ∇φ ≤ 2 φW 1,2 . (7.33)
max{ζ, vV } ε
Estimation of other terms of (7.29) is easy, and so operator A is well defined.

Then, existence of a solution to Problem I-VIA is equivalent to existence of a
solution to
f
A (v, θ ) = F , F = 1 .
g1
We will use Theorem 1.45 to prove the existence of a solution to the equation above.
To this end, we need to show that A satisfies the requirement in Theorem 1.45.
(i) Let us prove that A is coercive, i.e.
1
A (v, θ ), (v, θ ) → ∞ as (v, θ )H → ∞.
(v, θ )H
Since Γ2 j , Γ3 j , Γ7 j are convex (see Lemma 3.1) and the matrix α is positive, we have
from (7.14)
a0 (θ ; v, v) ≥ 2μ0 v2V . (7.34)
Since the operator ∇Φε is monotone and ∇Φε (0V ) = 0,
∇Φε (v), v = ∇Φε (v) − ∇Φε (0V ), v − 0V ≥ 0. (7.35)
Taking into account (7.31)–(7.35), we have from (7.29)
δ
A (v, θ), (v, θ) = a0 (θ; v, v) + a1 (v, v, v) + ∇Φε (v), v
max{δ, a 1 (v)V∗ }
α0 λ ζ
− 1− θ f, v + b0 (θ; θ, θ) − γε (θ)v, ∇θ
max{λ, θ} max{ζ, vV }
cζ
≥ min{2μ0 , κ0 } v2V + θ2 1,2 − δv − f V∗ vV − cλ f L3 vV − 2 θW 1,2
WΓ
D ε ΓD
∀(v, θ) ∈ H ,
(7.36)
which implies coercive property of A .
(ii) Taking into account (7.30)–(7.33), we have from (7.29)
A (v, θ )H ∗ = sup A (v, θ ), (u, φ)

(u,φ)H =1
1 ζ
≤ c vV + δ + vV + f L3 + λ f L3 + θ WΓ1,2 + 2
ε D ε
∀(v, θ ) ∈ H ,
(7.37)
which shows that A maps bounded sets of H into bounded sets of H ∗ .
(iii) Let {(vk , ηk )} be a sequence such that
(vk , ηk ) (v, η) in H ,
lim sup A (vk , ηk ), (vk , ηk ) − (v, η) ≤ 0.
k→∞
By taking a subsequence and denoting with the same subindex if necessary, we may
assume
vk → v in Ls (Ω)(1 ≤ s < 6) and a.e. in Ω,

(7.38)
ηk → η in L s (Ω)(1 ≤ s < 6) and a.e. in Ω as k → ∞.
Since
a0 (ηk ; vk − v, vk − v) = a0 (ηk ; vk , vk − v) − a0 (ηk ; v, vk − v),

b0 (ηk ; ηk − η, ηk − η) = b0 (ηk ; ηk , ηk − η) − b0 (ηk ; η, ηk − η),
by (7.29) we have

min{μ0 ,κ0 } vk − v2V + ηk − ηW 1,2 ≤ A (vk , ηk ), (vk , ηk ) − (v, η)
ΓD
− a0 (ηk ; v, vk − v) − b0 (ηk ; η, ηk − η)
δ
− a1 (vk , vk , vk − v) − ∇Φε (vk ), vk − v
max{δ, a 1 (vk )V∗ }
α0 λ ζ
+ 1− ηk f, vk − v + γε (ηk )vk , ∇(ηk − η).
max{λ, ηk } max{ζ, vk V }
(7.39)
By Corollary 1.1 we have that
a0 (ηk ; v, vk − v) = 2(μ(ηk )E (v), E (vk − v)) + 2(μ(ηk )k(x)v, vk − v)Γ2

+ 2(μ(ηk )S ṽ, ṽk − ṽ)Γ3 + 2(α(x)v, vk − v)Γ5 + (μ(ηk )k(x)v, vk − v)Γ7 → 0, (7.40)

b0 (ηk ; η, ηk − η) = κ(ηk )∇(η), ∇(ηk − η) + β(x)η, (ηk − η) Γ → 0
R
as k → ∞. Also,
δ
a1 (vk , vk , vk − v) ≤ vk L4 ∇vk L2 vk − vL4 → 0 as k → ∞.
(7.41)
Since ∇Φε is monotone,
− ∇Φε (vk ), vk − v = − ∇Φε (vk ) − ∇Φε (v), vk − v − ∇Φε (v), vk − v

≤ − ∇Φε (v), vk − v → 0 as k → ∞.
(7.42)
By (7.38), the followings hold.
α0 λ
ηk f, vk − v ≤ cηk L3 f L3 vk − vL3 → 0,
max{λ, ηk }
ζ
| γε (ηk )vk , ∇(ηk − η)| ≤ | γε (ηk ) − γε (η) vk , ∇(ηk − η)|
max{ζ, vk V }
+ | γε (η)(vk − v), ∇(ηk − η)| + | γε (η)v, ∇(ηk − η)|
≤ cγε (ηk ) − γε (η) L 3 vk V ηk − ηW 1,2
ΓD
+ cγε (η) L 6 (vk − v)L3 ηk − ηW 1,2

ΓD
+ γε (η)v, ∇(ηk − η) → 0

(7.43)
as k → ∞, where the fact that by Lemma 1.3 γε (ηk ) → γε (η) in L 3 (Ω) as k → ∞
was used. It is easy to prove convergence of other terms on the right hand side of
(7.39). Thus, by (7.39)–(7.43) we have

lim sup min{μ0 , κ0 }| vk − v2V + ηk − ηWΓ1,2
k→∞ D
≤ lim sup A (vk , ηk ), (vk , ηk ) − (v, η) ≤ 0,

k→∞
which implies
(vk , ηk ) → (v, η) in H as k → ∞,
(7.44)
vk → v, ηk → η a.e. in Ω as k → ∞.
By definition of A ,
A (vk , ηk ),(vk , ηk ) − (u, φ) = a0 (ηk ; vk , vk − u)

δ
+ a1 (vk , vk , vk − u) + ∇Φε (vk ), vk − u
α0 λ (7.45)
− 1− ηk f, vk − u + b0 (ηk ; ηk , ηk − φ)
max{λ, ηk }
ζ
− γ (ηk )vk , ∇(ηk − φ) ∀(u, φ) ∈ H .
max{ζ, vk V }
Taking into account (7.44), by Corollary 1.2 we have
lim inf a0 (ηk ; vk , vk ) ≥ a0 (η; v, v),

k→∞
(7.46)
lim inf b0 (ηk ; ηk , ηk ) ≥ b0 (η; η, η).
k→∞
αλ
α0 λ
The sequence {max{λ,η η k} converges a.e. in Ω to α0 λ
η and 0
η k
2≤
k } max{λ,η} max{λ,ηk } L
α0 λ
α0 λ, and so this sequence weakly converges to max{λ,η} η in L (see Lemma 1.1).
2
Since f ∈ L3 and (v − u) ∈ L6 , we have f · (v − u) ∈ L 2 . Thus, taking into account

the first formula of (7.43), we have
α0 λ
lim ηk f, vk − u
k→∞ max{λ, ηk }
α0 λ α0 λ
= lim ηk f, vk − v + lim ηk f, v − u
k→∞ max{λ, ηk } k→∞ max{λ, ηk }
α0 λ
= η f, v − u . (7.47)
max{λ, η}
Using the second formula in (7.43), in the same way we get

ζ ζ
lim γε (ηk )vk , ∇(ηk − φ) = γε (η)v, ∇(η − φ) .
k→∞ max{ζ, vk V } max{ζ, vV }
(7.48)
Similarly, we have
δ δ
lim a1 (vk , vk , vk − u)= a1 (v, v, v − u).
k→∞ max{δ, a 1 (vk )V }
∗ max{δ, a 1 (v)V∗ }
(7.49)
Since ∇Φε is monotone and
∇Φε (vk ),vk − u = ∇Φε (vk ), vk − v + ∇Φε (vk ), v − u

≥ ∇Φε (vk ) − ∇Φε (v), vk − v + ∇Φε (v), vk − v + ∇Φε (vk ), v − u ,
we have
lim ∇Φε (vk ), vk − u ≥ ∇Φε (v), v − u . (7.50)
k→∞
Thus, by (7.46)–(7.50) we have existence of a subsequence {(vk , ηk )} such that
lim inf A (vk , ηk ), (vk , ηk ) − (u, φ) ≥ A (v, η), (v, η) − (u, φ).
k→∞
Therefore, by virtue of Theorem 1.45 we come to the conclusion.
7.2.2 Existence and Estimates of Solutions to the

Approximate Problem
In this section the following theorem is proved.

Theorem 7.5 If
μ20
g1 (WΓ1,2 )∗ f L3 + f + f 1 V∗ ≤ , (7.51)
D K cα0
where K is the one in (7.26) and cα0 is the one in (7.63) below, then there exists a
solution (vε , θε ) ∈ V × WΓ1,2
D
(Ω) to the following problem

a0 (θε ; vε , u) + a1 (vε , vε , u) + ∇Φε (vε ), u − (1 − α0 θε ) f, u = f 1 , u ∀u ∈ V,
b0 (θε ; θε , ϕ) − γε (θε )vε , ∇ϕ = g1 , ϕ ∀ϕ ∈ WΓ1,2 (Ω),
D
(7.52)
and the solution satisfies:
μ0
vε V ≤ ,
K (7.53)
θε WΓ1,2 (Ω) ≤ cg1 (WΓ1,2 )∗ .
D D
Proof Let (vε , θε ) be solutions to (7.28). Putting ϕ = θε in the second formula of

(7.28), we have
ζ
(κ(θε )∇θε , ∇θε ) + (β(x)θε , θε )Γ R − γε (θε )vε , ∇θε = g1 , θε .
max{ζ, vV }
(7.54)
Let us first prove
γε (θε )vε , ∇θε = 0. (7.55)
To this end, define t

(t) := γε (s) ds, t ∈ R.
0
Then, ∈ C 1 (R) and
∇(θ ) = γε (θ )∇θ, (θ ) ∈ W 1,2 (Ω) ∀θ ∈ W 1,2 (Ω),

(7.56)
(θ )|Γ D = 0 ∀θ ∈ WΓ1,2
D
(Ω).
Taking into account vε · n|Γ R = 0, by (7.56) we have

γε (θ )vε , ∇θε = γε (θε )vε · ∇θε d x = vε · ∇(θε ) = 0,
Ω Ω
which means (7.55).
Next, we estimate other terms in (7.54).

κ0
g1 , θε ≤ θε 2H 1 + cg1 2(W 1,2 )∗ ,
4 ΓD (7.57)
(β(x)θε , θε )Γ R ≥ 0.
By (7.54), (7.54) and (7.57), we have
2c
θε 2W 1,2 ≤ g1 2(W 1,2 )∗ , (7.58)
ΓD κ0 ΓD
which implies
θε ≤ c1 g1 (WΓ1,2 )∗ . (7.59)
D
Putting
λ = c1 g1 (WΓ1,2 )∗
D
and taking into account (7.59), we have
λ
= 1, (7.60)
max{λ, θε }
α0 λ
θε f, u ≤ c1 α0 g1 (WΓ1,2 )∗ f L3 uV . (7.61)
max{λ, θε } D
Putting u = vε in the first equation of (7.28), we have
δ
a0 (θε ; vε , vε ) + a1 (vε , vε , vε ) + ∇Φε (vε ), vε
max{δ, a 1 (vε )V∗ }
α0 λ (7.62)
− 1− θε f, vε = f 1 , vε .
max{λ, θε }
Thus taking into account (7.26), (7.34), (7.35) and (7.61), we have from (7.62)
2μ0 vε 2V ≤ a0 (θε ; vε , vε )

δ
≤ |a1 (vε , vε , vε )| + |α0 θε f, vε | + | f + f 1 , vε |

≤ K vε 3V + cα0 g1 (WΓ1,2 )∗ f L3 + f + f 1 V∗ vε V .
D
(7.63)
Note that the estimate above is independent of δ, but cα0 depends on α0 . This implies

0 ≤ K vε 2V − 2μ0 vε V + cα0 g1 (WΓ1,2 )∗ f L3 + f + f 1 V∗ .
D
Let us consider a quadratic polynomial for x > 0 related to the inequality above
K x 2 − 2μ0 x + a.
μ0
If 0 ≤ K a ≤ μ20 , then there exists a nonnegative minimum root x1 (≤ K
) and a
maximum root x2 . Thus, we can see that if
μ20
g1 (WΓ1,2 )∗ f L3 + f + f 1 V∗ ≤ , (7.64)
D K cα0
then
μ0
vε V ≤ or vε V ≥ x2 . (7.65)
K
On the other hand, we have from (7.63) another estimation under consideration
of δ

2μ0 vε 2V ≤ a0 (θε ; vε , vε ) ≤ δvε V +cα0 g1 (WΓ1,2 )∗ f L3 + f + f 1 V∗ vε V ,
D
which implies
1
vε V ≤ δ + cα0 g1 (WΓ1,2 )∗ f L3 + f + f 1 V∗ . (7.66)
2μ0 D
2 μ2
In view of (7.65), let us take δ = K μK0 = K0 .
Thus, in view of (7.51), we have from (7.66)
δ 1 μ20 μ0
vε V ≤ + = . (7.67)
2μ0 2μ0 K K
μ2
By (7.67) under the condition (7.51) we have that a 1 (vε )V∗ ≤ K vε 2V ≤ K0 (see
(7.26)), and so we get
δ
= 1. (7.68)
μ0
Taking ζ = K
, by (7.67) we get
ζ
= 1. (7.69)
max{ζ, vε V }
By (7.60), (7.68) and (7.69), we see that under condition (7.64) (vε , θε ) satisfies
(7.52). By virtue of (7.67) and (7.58) we get (7.53).
7.2.3 Existence and Uniqueness of a Solution
First, by passing to the limit of solutions in Theorem 7.5, we will prove existence of
a solution to Problem I-VI. Owing to (7.53) we can extract subsequences, which are
denoted as before, such that
vε v in V,
vε → v in Lq , 1 ≤ q < 6,
(7.70)
θε θ in H 1 (Ω),
θε → θ in L q (Ω), 1 ≤ q < 6,
as ε → 0.
Subtracting the first formula of (7.52) with u = vε from the first formula of (7.52),
we have
a0 (θε ; vε , u − vε ) + a1 (vε , vε , u − vε ) + ∇Φε (vε ), u − vε

(7.71)
− (1 − α0 θε ) f, u − vε = f 1 , u − vε ∀u ∈ V.
By Corollaries 1.1 and 1.2, we have
a0 (θε ; vε , u) → a0 (θ ; v, u) as ε → 0,
lim inf a0 (θε ; vε , vε ) ≥ a0 (θ ; v, v),
ε→0
which imply that
lim sup a0 (θε ; vε , u − vε ) ≤ a0 (θ ; v, u − v). (7.72)

ε→0
It is easy to prove
a1 (vε , vε , u − vε ) → a1 (v, v, u − v) as ε → 0. (7.73)
Since Φε is convex, continuous and Fréchet differentiable, we have
Φε (u) − Φε (vε ) ≥ ∇Φε (vε ), u − vε ∀u ∈ V, (7.74)
which, together with (1.37), implies
Φε (u) − Φ(Jε vε ) ≥ ∇Φε (vε ), u − vε ∀u ∈ V. (7.75)
Since Φ(0V ) = 0, by (1.37) we get Φε (0V ) = 0, and so we have from (7.74)
Φε (vε ) ≤ ∇Φε (vε ), vε . (7.76)
On the other hand, putting u = vε in the first formula of (7.52), we have
a0 (θε ; vε , vε ) + a1 (vε , vε , vε ) + ∇Φε (vε ), vε = (1 − α0 θε ) f, vε + f 1 , vε .

(7.77)
a0 (θε ; vε , vε ) + a1 (vε , vε , vε ) + Φε (vε ) ≤ (1 − α0 θε ) f, vε + f 1 , vε ,
from which we get


|Φε (vε )| ≤ c g1 (WΓ1,2 )∗ f L3 + f L3 + f 1 V∗ vε V + |a1 (vε , vε , vε )|.
D
(7.78)
By virtue of (1.36), (7.26), (7.66) and (7.78), we have
μ0 μ3
vε − Jε vε 2V ≤ c g1 (WΓ1,2 )∗ f L3 + f L3 + f 1 V∗ vε V + 02 2ε,
D K K
which shows that since vε v in V,
Jε vε v in V as ε → 0.
Then, by virtue of lower weak semi-continuity of Φ(v),
lim inf Φ(Jε vε ) ≥ Φ(v). (7.79)

ε→0
By (1.37) we have
Φε (u) → Φ(u) as ε → 0. (7.80)
Taking into account (7.79) and (7.80), we have from (7.75)
Φ(u) − Φ(v) ≥ lim sup ∇Φε (vε ), u − vε ∀u ∈ V. (7.81)

ε→0
Using
| θε f, vε − θ f, v | ≤ | θε f, vε − θ f, vε | + | θ f, vε − θ f, v |
(7.82)
≤ θε − θ L 3 f L2 vε L6 + θ L 6 f L2 vε − vL3 ,
we can prove
(1 − α0 θε ) f, u − vε → (1 − α0 θ ) f, u − v as ε → 0. (7.83)
It is easy to prove
f 1 , u − vε → f 1 , u − v as ε → 0. (7.84)
By virtue of (7.72), (7.73), (7.81), (7.83) and (7.84), from (7.71) we get
a0 (θ; v, u − v) + a1 (v, v, u − v) + Φ(u) − Φ(v) − (1 − α0 θ) f, u − v ≥ f 1 , u − v

∀u ∈ V,
which is the first formula in (7.20).

We will get the second equation in (7.20). By Corollary 1.1, we have
b0 (θε ; θε , ϕ) → b0 (θ ; θ, ϕ) ∀ϕ ∈ WΓ1,2
D
(Ω) as ε → 0. (7.85)
Let us prove
γε (θε )vε , ∇ϕ → γ (θ )θ v, ∇ϕ ∀ϕ ∈ WΓ1,2

D
(Ω) as ε → 0. (7.86)
By Hölder’s inequality we have
| γε (θε )vε , ∇ϕ − γ (θ )θ v, ∇ϕ|

≤ | γε (θε )vε , ∇ϕ − γ (θ )θ vε , ∇ϕ| + | γ (θ )θ vε , ∇ϕ − γ (θ )θ v, ∇ϕ|
≤ γε (θε ) − γ (θ )θ L 3 vε L6 ∇ϕL2 + γ (θ )θ L 4 vε − vL4 ∇ϕL2 .
(7.87)
By the definition of γε (t) we have

γ (θε )θε
γε (θε ) − γ (θ)θ L 3 ≤
− γ (θ)θ
3
(1 + ε|γ (θε )|)(1 + ε|θε |) L

≤ γ (θε )θε − γ (θ)θ L 3 + ε γ (θ)θ |γ (θε )| + |θε | + ε|γ (θε )||θε | L3
.
(7.88)
By virtue of Lemma 1.3 we know that γ (θε ) converges to γ (θ ) in space L p (Ω)
(∀ p, 1 < p < ∞, ) as ε goes to zero. Thus, from (7.87) and (7.88) we get (7.86).
By virtue of (7.85) and (7.86), from the second formula in (7.52) we get the
second formula in (7.20).
Estimates (7.25) follow from (7.53).
Next, let us prove uniqueness of solutions. Suppose that there are two solutions
(v1 , θ1 ) and (v2 , θ2 ). Since μ is independent of θ by the condition, denoting a0 (·; v, u)
by a0 (v, u) yields
a0 (v1 , v2 − v1 ) + a1 (v1 , v1 , v2 − v1 ) + Φ(v2 ) − Φ(v1 ) − f − α0 θ1 f, v2 − v1

≥ f 1 , v2 − v1 ,
a0 (v2 , v1 − v2 ) + a1 (v2 , v2 , v1 − v2 ) + Φ(v1 ) − Φ(v2 ) − f − α0 θ2 f, v1 − v2
≥ f 1 , v1 − v2 ,
which imply

a0 v1 − v2 , v1 − v2 ≤ |α0 ||(θ1 − θ2 ) f, v1 − v2 )|
(7.89)
+ |a1 (v1 , v1 , v1 − v2 ) − a1 (v2 , v2 , v1 − v2 )|.
By virtue of (7.89), we have
2μv1 − v2 2V
μ
≤ v1 − v2 2V + c f 2L3 θ1 − θ2 2
2
+ |a1 (v1 − v2 , v1 , v1 − v2 ) + a1 (v2 , v1 − v2 , v1 − v2 )|
μ
≤ v1 − v2 2V + c f 2L3 θ1 − θ2 2 + c1 (v1 V + v2 V )v1 − v2 2V
2
and so
3μ
v1 − v2 2V ≤ c f 2L3 θ1 − θ2 2 + c1 (v1 V + v2 V )v1 − v2 2V . (7.90)
2
Since κ(θ ), γ (θ ) are independent of θ , put κ(θ ) = κ, γ (θ ) = cv . Then, from
(κ∇θ1 , ∇ϕ) + (β(x)θ1 , ϕ)Γ R − cv v1 θ1 , ∇ϕ1 = g1 , ϕ,

(κ∇θ2 , ∇ϕ) + (β(x)θ2 , ϕ)Γ R − cv v2 θ2 , ∇ϕ = g1 , ϕ
we have
κ(∇θ1 − ∇θ2 , ∇θ1 − ∇θ2 ) + (β(x)(θ1 − θ2 ), θ1 − θ2 )Γ R

(7.91)
− cv v1 (θ1 − θ2 ), ∇(θ1 − θ2 ) − cv (v1 − v2 )θ2 , ∇(θ1 − θ2 ) = 0.
Using cv v1 (θ1 − θ2 ), ∇(θ1 − θ2 ) = 0 (see (7.12) with γ (θ )=const), we obtain from

(7.91)
cv c κ
κ∇θ1 − ∇θ2 2 ≤ v1 − v2 2V θ2 2W 1,2 + ∇θ1 − ∇θ2 2 ,
κ ΓD 2
and so
κ cv c
θ1 − θ2 2W 1,2 ≤ v1 − v2 2V θ2 2W 1,2 . (7.92)
2 κ ΓD
Therefore adding (7.90) and (7.92), we get
3μ κ
min{ , } v1 − v2 2V + θ1 − θ2 2W 1,2
2 2
cv c
≤ c f 2L3 θ1 − θ2 2 + c1 (v1 V + v2 V )v1 − v2 2V + v1 − v2 2V θ2 2W 1,2 .
κ ΓD
(7.93)
Thus, if vi V , θ2 WΓ1,2 and f L3 are small, then we have from the above that
D
v1 = v2 and θ1 = θ2 .
Theorem 7.6 Under Assumption 7.1 there exists a solution (v, θ ) to Problem II-VI
such that

vV ≤ c f L3 + φi Γi + g R L 4/3 (Γ R ) + g L 6/5 (Ω) ,
i=2−7 (7.94)

θ W 1,2 (Ω) ≤ c g R L 4/3 (Γ R ) + g L 6/5 (Ω) .
First, we look for solutions to an auxiliary problem:
Problem II-VIA. Let ζ > 0, λ > 0 and ε > 0. Find (v, θ ) ∈ V × W 1,2 (Ω) such that
η = θ − θ D ∈ WΓ1,2
D
(Ω) and
⎧
⎪ α0 λ
⎪
⎪ a (θ ; v, u) + a (v, v, u) + ∇Φ (v), u − 1 − θ f, u
⎪
⎪
0 2 ε
max{λ, θ }
⎨
= f 1 , u ∀u ∈ V, (7.95)
⎪
⎪
⎪
⎪ ζ
⎪
⎩ b0 (θ ; θ, ϕ) − γε (θ )v, ∇ϕ = g1 , ϕ ∀ϕ ∈ WΓ1,2 (Ω).
max{ζ, vε V } D
Theorem 7.7 There exists a solution (vε , θε ) ∈ V × W 1,2 (Ω) to Problem II-VIA.

D
A (v, η), (u, φ) =a0 (η; v, u) + a2 (v, v, u) + ∇Φε (v), u

α0 λ
− 1− θ f, u + b0 (θ ; θ, ϕ)
max{λ, θ }
ζ
− γε (η + θ D )v, ∇ϕ ∀(v, η), (u, φ) ∈ H .
max{ζ, vV }
Using
a2 (v, v, v) = 0,
|a2 (v, v, u)| ≤ K v2V uV ,
|a2 (vε , vε , vε − u)| ≤ c∇vε L2 vε L4 vε − uL4 ,
respectively, in (7.36), (7.37) and (7.41), we can verify that the proof of Theorem
7.4 for Problem I-VIA is valid for Problem II-VIA. Thus, we come to the asserted
conclusion.
Theorem 7.8 There exists a solution (vε , θε ) ∈ V × W 1,2 (Ω) to the following prob-
lem

a0 (θε ; vε , u) + a2 (vε , vε , u) + ∇Φε (vε ), u − (1 − α0 θε ) f, u = f 1 , u ∀u ∈ V,
b0 (θε ; θε , ϕ) − γε (θε )vε , ∇ϕ = g1 , ϕ ∀ϕ ∈ WΓ1,2 (Ω),
D
(7.96)
c
vε V ≤ g1 (WΓ1,2 )∗ f L3 + f V∗ + f 1 V∗ ,
2μ0 D
(7.97)
θε WΓ (Ω) ≤ cg1 (WΓ )∗ .
1,2 1,2
D D
Proof Let (vε , θε ) be solutions to (7.95). In the same way as in (7.54)–(7.58) we

have
2c
θε 2W 1,2 ≤ g1 2(W 1,2 )∗ , (7.98)
ΓD κ0 ΓD
which implies
θε (x) ≤ c1 g1 (WΓ1,2 )∗ . (7.99)
D
Putting λ = c1 g1 (WΓ1,2 )∗ , we have from the first equation of (7.95)

D

a0 (θε ; vε , u) + a2 (vε , vε , u) + ∇Φε (vε ), u − 1 − α0 θε f, vε = f 1 , u ∀u ∈ V,
(7.100)
which is the first formula of (7.96).
Putting u = vε in (7.100), we have

a0 (θε ; vε , vε ) + a2 (vε , vε , vε ) + ∇Φε (vε ), vε − 1 − α0 θε f, vε = f 1 , vε .
(7.101)
Taking into account a2 (vε , vε , vε ) = 0, (7.34), (7.35) and (7.99), we have from
(7.101)

2μ0 vε 2V ≤ a0 (θ ; v, v) ≤ c α0 c1 g1 (WΓ1,2 )∗ f L3 + f V∗ + f 1 V∗ vε V ,
D
which implies
c
vε V ≤ α0 c1 g1 (WΓ1,2 )∗ f L3 + f V∗ + f 1 V∗ . (7.102)
2μ0 D
Taking the right hand side of (7.102) as ζ in (7.95), we get the second equation of
(7.96). By (7.98) and (7.102), we get (7.97).
Now repeating the arguments in Sect. 7.2.3 with the solutions of Theorem 7.8, we
complete proof of Theorem 7.6.
Remark 7.3 The Eq. (9.1) of Chap. 9 is more general than (7.1), and from the results
of Chap. 9 some results for (7.1) with the boundary conditions (7.2), (7.3) or (7.2),
(7.4) can be obtained. However the result for the case of total pressure demands that
the parameter for buoyancy effect α0 is small enough in accordance with the data of
problem (see (9.138), and the result for the case of static pressure demands that the
data of problem satisfy two smallness conditions together (see (9.50) and (9.97)).

Several papers are concerned with (7.1). In [3, 4] under homogeneous Dirich-
let boundary condition for velocity and mixture of non-homogeneous Dirichlet and
Neumann conditions for temperature existence of a solution to (7.1) was studied.
In [5] under non-homogeneous Dirichlet boundary condition for velocity and mix-
ture of non-homogeneous Dirichlet and Neumann conditions for temperature, where
smoothness of boundary data is weaker than [3, 4], existence of a solution to (7.1)
was obtained.
In [6] under mixture of non-homogeneous Dirichlet, total pressure and vorticity
boundary conditions for fluid and mixture of non-homogeneous Dirichlet, Neumann
and Robin conditions for temperature the existence of a solution was established. In
[7] variational inequalities for Navier-Stokes type operators were studied, which can
describe (7.1) with one-sided flow boundary conditions for fluid and heat on a portion
of boundary. In [8] under homogeneous Dirichlet boundary condition for velocity
and mixture of non-homogeneous Dirichlet and homogeneous Neumann conditions
for temperature the existence and uniqueness and smoothness of a weak solution
were discussed. In [9] when the boundary consists of several connected components,
boundary value problem of (7.1) with non-homogeneous Dirichlet boundary condi-
tion was studied. In [10] when the boundary consists of several connected compo-
nents, Dirichlet problem of (7.1) under a weaker condition than [9] was investigated.
In [11] Dirichlet problem of (7.1) for arbitrarily large and very weak boundary
data was studied.
References
1. H. Gajewski, K. Gröger, K. Zacharias, Nichtlineare Operatorgleichungen und Operatordiffer-

entialgleichungen (Academic-Verlag Berlin, 1974) (Russian 1978)
2. T. Kim, The steady Boussinesq system with mixed boundary conditions including conditions
of friction type. Applications of Mathematics, to appear
3. H. Morimoto, On the existence of weak solutions of equations of natural convection. J. Fac.
Sci. Univ. Tokyo, Sect. IA 36, 87–102 (1989)
4. H. Morimoto, On the existence and uniqueness of the stationary solution to the equation of
natural convection. Tokyo J. Math. 14, 217–226 (1991)
5. E.J. Villamizar-Roa, M.A. Rodríguez-Bellido, M.A. Rojas-Medar, The Boussinesq system with
mixed nonsmooth boundary data. C. R. Acad. Sci. Paris, Ser. I 343, 191–196 (2006)
6. G.V. Alekseev, A.B. Smishliaev, Solvability of the boundary value problems for Boussinesq
equations with inhomogeneous boundary conditions. J. Math. Fluid Mech. 3(1), 18–39 (2001)
7. A.Yu. Chebotarev, Variational inequalities for Navier-Stokes type operators and one-sided
problems for equations of viscous heat-conducting fluids. Math. Notes 70, 264–274 (2001)
8. D.A. Kovtunov, Solvability of the stationary heat convection problem for a high-viscosity fluid.
Diff. Equat. 45(1), 73–85 (2009)
9. H. Morimoto, Heat convection equation with nonhomogeneous boundary conditions. Funk-
cialai Ekvacioj 53, 213–229 (2010)
10. P. Acevedo, C. Amrouche, C. Conca, Boussinesq system with non-homogeneous boundary
conditions. Appl. Math. Lett. 53, 39–44 (2016)
11. H. Kim, The existence and uniqueness of very weak solutions of the stationary Boussinesq
system. Nonlinear Anal. 75, 317–330 (2012)
Chapter 8
The Non-steady Boussinesq System
In this chapter we are concerned with the non-steady Boussinesq problem corre-
sponding to the steady problem in Chap. 7. The formulations consist of a non-steady
variational inequality for velocity and a non-steady variational equation for temper-
ature.
For the problem with boundary conditions involving the static pressure and stress,
it is proved that if the data of problem are small enough and compatibility conditions
at the initial time for velocity and temperature are satisfied, then there exists a unique
solution on the given interval. For the problem with boundary conditions involving the
total pressure and total stress, the existence of a solution is proved without restriction
on the data of problem.
8.1 Problems and Assumptions
The non-steady Boussinesq system with initial condition for heat convection is as
follows.
⎧
⎪ ∂v
⎪
⎪ − 2∇ · μ(θ )E(v) + (v · ∇)v + ∇ p = (1 − α0 θ ) f,
⎪
⎪ ∂t
⎪
⎪
⎨ div v = 0,
(8.1)
⎪
⎪ ∂θ
⎪
⎪ − ∇ · κ(θ )∇θ + v · ∇θ = g,
⎪
⎪ ∂t
⎪
⎩
v(0) = v0 , θ (0) = θ0 .
Let Ω be as in Chap. 7 with Γi j ∈ C 2,1 for i = 2, 3, 7 instead of Γi j ∈ C 2 . (See

Remark 6.1) Let Q = Ω × (0, T ), 0 < T < ∞.
https://doi.org/10.1007/978-3-030-78659-5_8
252 8 The Non-steady Boussinesq System
The boundary conditions for temperature are as follows:
(1) θ |Γ D = 0,
∂θ
(2) κ(θ ) + β(x)θ Γ R = g R (t, x), β(x), g R (t, x) − given functions.
∂n
(8.2)
Problems I and II are distinguished according to boundary conditions for fluids.
Problem I is the one with boundary condition (7.3) assumed that μ and κ are inde-
pendent of θ (the case of static pressure) and Problem II is the one with boundary
condition (7.4) (the case of total pressure).
H : completion in L2 (Ω) of V,
HK : closure in L2 (Ω) of K (Ω),
WΓ1,2
D
(Ω) = {y ∈ W 1,2 (Ω) : y|Γ D = 0}.
We use the inner products in (7.8) for V and WΓ1,2

D
(Ω).
We will use the following two kinds of assumptions according to different bound-
ary conditions. The assumption for Problem I is stronger than the one for Problem II,
and consequently smoothness of solution with respect to t for Problem I is stronger
than the one in Problem II.
Assumption 8.1 (for the case of static pressure) We assume the followings.
(1) Γ1 = ∅, Γ D = ∅ and

Γ R ⊂ ∪i=1,3,5,8 Γi . (8.3)

(3) For the functions of (8.1)
f ∈ W 1,∞ (0, T ; L3 (Ω)),

g ∈ W 1,2 0, T ; (WΓ1,2
D
)∗ , (8.4)
μ and κ are independent of θ.
(4) For the functions of (8.2) and (7.3),
g R ∈ W 1,2 (0, T ; L 4/3 (Γ R )),

β1 ≥ β(x) ≥ 0, β1 − a constant, β(x) − measurable,
φi ∈ W 1,∞ (0, T ; H −1/2 (Γi )), i = 2, 4, 7, (8.5)
−1/2
φi ∈ W (0, T ; H
1,∞
(Γi )), i = 3, 5, 6,
αi j ∈ L ∞ (Γ5 ).
8.1 Problems and Assumptions 253
Assumption 8.2 (for the case of total pressure) Let (1) and (2) of Assumption 8.1
hold and suppose the followings.
(3’) For the functions of (8.1),
f ∈ L ∞ (0, T ; L3 (Ω)),

g ∈ L 2 0, T ; (WΓ1,2 )∗ ),
D
(8.6)
μ ∈ C(R), 0 < μ0 ≤ μ(ξ ) ≤ μ1 < ∞ ∀ξ ∈ R,
κ ∈ C(R), 0 < κ0 ≤ κ(ξ ) ≤ κ1 < ∞ ∀ξ ∈ R.
(4’) For the functions of (8.2) and (7.4),
g R ∈ L 2 (0, T ; L 4/3 (Γ R ));

φi ∈ L 2 (0, T ; H − 2 (Γi )), i = 2, 4, 7,
1
(8.7)
− 21
φi ∈ L 2 (0, T ; H (Γi )), i = 3, 5, 6;
αi j ∈ L ∞ (Γ5 ).
Remark 8.1 For (2) of Assumption 8.1 we refer to Remark 6.3. In this chapter this
assumption is also necessary to guarantee equivalence between Problem I-VE and
Problem I-VI (in Remark 8.4), and between Problem II-VE and Problem II-VI.
8.2 Variational Formulations for Problems
In this section we first give variational formulations for the problems above. Taking
into account (8.3) and vn |Γ3 ∪Γ5 ∪Γ8 = 0, for v ∈ V, θ ∈ W 1,2 (Ω) and ϕ ∈ WΓ1,2
D
(Ω),
we have
(v · ∇θ, ϕ) = (vn θ, ϕ)Γ R − (θ v, ∇ϕ) = −(θ v, ∇ϕ). (8.8)
8.2.1 Variational Formulations: The Case of Static Pressure
By (7.9)–(7.11) and (8.8), as in Sect. 7.1 we can see that smooth solutions (v, p, θ )
of problem (8.1), (8.2), (7.3) satisfy the following system.
⎧
⎪ ∂v
⎪
⎪ ( , u) + 2(μE (v), E (u)) + (v · ∇)v, u + 2(μk(x)v, u)Γ2 + 2(μS ṽ, ũ)Γ3
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ + 2(α(x)v, u)Γ5 + (μk(x)v, u)Γ7 − 2(μεnτ (v), u)Γ8 + ( p − 2μεnn (v), u n )Γ9 ∪Γ10 ∪Γ11
⎪
⎪
⎪
⎪
⎪
⎪ = (1 − α0 θ) f, u + φi , u n Γi + φi , uΓi ∀u ∈ V,
⎪
⎪
⎪
⎪
⎪
⎨ i=2,4,7 i=3,5,6
∂θ
⎪
⎪ ( , ϕ) + (κ∇θ, ∇ϕ) − (θv, ∇ϕ) + (βθ, ϕ)Γ R = g R , ϕΓ R + g, ϕ ∀ϕ ∈ WΓ1,2 (Ω),
⎪ ∂t
⎪
D
⎪
⎪
⎪
⎪
⎪
⎪
⎪ |σn (v, p)| ≤ gn , σn (v, p)vn + gn |vn | = 0 on Γ9 ,
⎪
⎪
⎪
⎪
⎪
⎪ σn (v, p) + g+n ≥ 0, (σn (v, p) + g+n )vn = 0 on Γ10 ,
⎪
⎪
⎩
σn (v, p) − g−n ≤ 0, (σn (v, p) − g−n )vn = 0 on Γ11 .
(8.9)
Define a01 (·, ·), a11 (·, ·, ·) and f 1 (t) ∈ V∗ by
a01 (w, u) = 2(μE(w), E(u)) + 2(μk(x)w, u)Γ2 + 2(μS w̃, ũ)Γ3

+ 2(α(x)w, u)Γ5 + (μk(x)w, u)Γ7 ∀w, u ∈ V,
a11 (v, u, w) = (v · ∇)u, w ∀v, u, w ∈ V, (8.10)
f 1 (t), u = φi (t), u n Γi + φi (t), uΓi ∀u ∈ V.
i=2,4,7 i=3,5,6
Define b1 (·, ·) and g1 (t) ∈ (WΓ1,2

D
(Ω))∗ by
b1 (θ, ϕ) = (κ∇θ, ∇ϕ) + (β(x)θ, ϕ)Γ R ∀θ ∈ W 1,2 (Ω), ϕ ∈ WΓ1,2

D
(Ω),
(8.11)
g1 (t), ϕ = g R (t), ϕΓ R + g(t), ϕ ∀ϕ ∈ WΓ1,2
D
(Ω).
Remark 8.2 Under (4) of Assumption 8.1 the duality product f 1 , u of (8.10) is
meaningful (see Remark 5.1). By (8.4) and (8.5),
f 1 ∈ W 1,∞ (0, T ; V∗ ), g1 ∈ W 1,2 (0, T ; (WΓ1,2

D
)∗ ). (8.12)
στ (θ, v) = 2μεnτ (v), σn (θ, v, p) = − p + 2μεnn (v)
and (8.9), we introduce the following variational formulation for problem (8.1), (8.2),
(7.3).
Problem I-VE. Find v ∈ K (Q), θ ∈ L 2 (0, T ; WΓ1,2 D

(Ω)) and (στ , σn , σ+n , σ−n ) ∈
L2τ (Γ8 ) × L 2 (Γ9 ) × H − 2 (Γ10 ) × H − 2 (Γ11 ) for a.e. t ∈ (0, T ) such that v(0) = v0 ,
1 1
θ (0) = θ0 and
8.2 Variational Formulations for Problems 255
⎧
⎪ ∂v
⎪
⎪ , u + a01 (v, u) + a11 (v, v, u) − (στ , u τ )Γ8 − (σn , u n )Γ9
⎪ ∂t
⎪
⎪
⎪
⎪
⎪ − σ+n , u n Γ10 − σ−n , u n Γ11 − f − α0 θ f, u = f 1 , u ∀u ∈ L 2 (0, T ; V),
⎪
⎪
⎪
⎪
⎪
⎪ ∂θ , ϕ + b (θ, ϕ) − θv, ∇ϕ = g , ϕ ∀ϕ ∈ L 2 (0, T ; W 1,2 (Ω)),
⎨ 1 1 ΓD
∂t (8.13)
⎪
⎪ |στ | ≤ gτ , στ · vτ + gτ |vτ | = 0 on Γ8 ,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ σ+n + g+n ≥ 0, σ+n + g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎩
σ−n − g−n ≤ 0, σ−n − g−n , vn Γ11 = 0 on Γ11 ,
(u, n)L2 (Γ8 ) = 0.
Remark 8.3 We showed that smooth solutions to the problem (8.1), (8.2), (7.3)
are solutions
to Problem I-VE. Assume that a solution to Problem I-VE is smooth
enough v ∈ L 2 (0, T ; H2 (Ω)), v ∈ L 1 (0, T ; L2 (Ω)), θ ∈ L 2 (0, T ; H 2 (Ω)), θ ∈
L 1 (0, T ; L 2 (Ω)), στ ∈ L 2 (0, T ; L2 (Γ 8 )), σn ∈ L 2 (0, T ; L 2 (Γ9 )), σ+ ∈ L 2 (0, T ;
L 2 (Γ10 )) and σ− ∈ L 2 (0, T ; L 2 (Γ11 )) and f ∈ L 2 (0, T ; L2 (Ω)). Then, for a.e. t ∈
(0, T ) there exists p(t) ∈ H 1 (Ω) (see Theorem 5.3) such that p ∈ L 1 (0, T ; H 1 (Ω))
and (v, p, θ ) satisfies the first two equations of (8.1) and the boundary condition (7.3)
in suitable spaces. By a standard way, it is proved that (v, θ ) satisfies the third equa-
tion of (8.1), the boundary condition (8.2) and the initial conditions.
In this sense, Problem I-VE is equivalent to the problem (8.1), (8.2), (7.3).
and a variational equation, which is equivalent to Problem I-VE if the solution is
smooth enough (see Remark 8.4).
For fixed θ , let us consider the problem
⎧
⎪ ∂v
⎪
⎪ , u + a01 (v, u) + a11 (v, v, u) − (στ , u τ )Γ8 − (σn , u n )Γ9 − σ+n , u n Γ10
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ − σ−n , u n Γ11 − ( f − α0 θ f, u = f 1 , u, ∀u ∈ L 2 (0, T ; V),
⎪
⎨
⎪
⎪
⎪
⎪
⎪
⎪
⎪ σ+n + g+n ≥ 0, σ+n + g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎪
⎩
σ−n − g−n ≤ 0, σ−n − g−n , vn Γ11 = 0 on Γ11 .
(8.14)
Subtracting the first formula of (8.14) with u = v from the first one of (8.14), we get
∂v
, u − v + a01 (v, u − v) + a11 (v, v, u − v) − (στ , u τ − vτ )Γ8 − (σn , u n − vn )Γ9
∂t
− σ+n , u n − vn Γ10 − σ−n , u n − vn Γ11 − f − α0 θ f, u − v (8.15)
= f 1 , u − v ∀u ∈ V.
Let Φ : V → R be the functional defined by (5.22). Then, the functional Φ is

proper, convex, lower weak semi-continuous and nonnegative.
Define a functional Ψ (u) by
⎧ T
⎨ Φ(u(t)) dt if Φ(u(t)) ∈ L 1 (0, T ),
Ψ (u) = (8.16)
⎩ 0
+∞ otherwise.
In the same way as Problem I of Sect. 5.1, from (8.15) we get
∂v
, u − v + a01 (v, u − v) + a11 (v, v, u − v) + Φ(u) − Φ(v)
∂t (8.17)
≥ (1 − α0 θ ) f, u − v + f 1 , u − v.
Define operators A1 : V → V∗ , B1 : V × V → V∗ and C1 : WΓ1,2

D
(Ω) →
1,2 ∗
(WΓ D (Ω)) , respectively, by
A1 v, u = a01 (v, u) ∀v, u ∈ V,

B1 (v, u), w = a11 (v, u, w) ∀v, u, w ∈ V, (8.18)
C1 θ, ϕ = b1 (θ, ϕ) ∀θ, ϕ ∈ WΓ1,2
D
(Ω).
If v is a solutions to (8.14), then we can see that the solution satisfies (see (6.18))
T
v (t) + A1 v(t) + B1 (v(t), v(t)) − (1 − α0 )θ (t) f (t) − f 1 (t), u(t) − v(t) dt
0
+ Ψ (u) − Ψ (v) ≥ 0 ∀u ∈ L 4 (0, T ; V).
(8.19)
Therefore, we have the following variational formulation of the problem (8.1),

(8.2), (7.3) which consists of a variational inequality for velocity and a variational
equation for temperature.

Problem I-VI. Find (v, θ ) ∈ L ∞ (0, T ; H ) ∩ L 2 (0, T ; V) × L ∞ (0, T ; L 2 (Ω)) ∩

L 2 (0, T ; WΓ1,2
D
(Ω)) such that
⎧ T
⎪
⎪ v + A1 v(t) + B1 (v(t), v(t)) − (1 − α0 θ) f − f 1 , u(t) − v(t) dt
⎪
⎪
⎪
⎪ 0
⎪
⎪
⎨ + Ψ (u) − Ψ (v) ≥ 0 ∀u ∈ L 4 (0, T ; V),
(8.20)
⎪
⎪ T ∂θ
⎪
⎪ , ϕ + C1 θ(t), ϕ − θv, ∇ϕ − g1 , ϕ dt = 0 ∀ϕ ∈ L 2 (0, T ; WΓ1,2 ),
⎪
⎪ ∂t
⎪
⎪ 0
D
⎩
v(0) = v0 , θ(0) = θ0 .
Remark 8.4 If the solution to Problem I-VI is smooth as much as v ∈ L 2 (0, T ; V),
v ∈ L 2 (0, T ; V∗ ), then the first one of (8.20) is equivalent to
v (t) + A1 v(t) + B1 (v(t), v(t)) − (1 − α0 θ ) f − f 1 , u − v(t) + Φ(u) − Φ(v(t)) ≥ 0

for a.e. t ∈ [0, T ], ∀u ∈ K (Ω) (8.21)
(Remark, p. 114 of [1]). In (8.17) putting F1 , u − v = − ∂v ∂t

− (1 − α0 θ ) f −
f 1 , u − v, by Theorem 5.4 we can see the existence of (στ , σn , σ+n ,
σ−n ) ∈ L2τ (Γ8 ) × L 2 (Γ9 ) × H −1/2 (Γ10 ) × H −1/2 (Γ11 ) for a.e. t ∈ (0, T ) such that
(v, θ, στ , σn , σ+n , σ−n ) is a solution to Problem I-VE.
8.2.2 Variational Formulations: The Case of Total Pressure
Since (v · ∇)v = rot v × v + 21 grad|v|2 , by (7.9)–(7.11) and (8.8), we can see that
smooth solutions (v, p, θ ) of problem (8.1), (8.2), (7.4) satisfy the following.
⎧ ∂v
⎪
⎪ , u + 2(μ(θ)E (v), E (u)) + rotv × v, u + 2(μ(θ)k(x)v, u)Γ2
⎪
⎪ ∂t
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪ − 2(μ(θ)εnτ (v), u)Γ8 + ( p + |v|2 − 2μ(θ)εnn (v), u n )Γ9 ∪Γ10 ∪Γ11
⎪
⎪
⎪
⎪ 2
⎪
⎪
⎪
⎪ = (1 − α0 θ) f, u + φi , u n Γi + φi , uΓi ∀u ∈ V,
⎨
i=2,4,7 i=3,5,6
⎪
⎪ ∂θ
⎪
⎪ , ϕ + (κ(θ)∇θ, ∇ϕ) − (θv, ∇ϕ) + (βθ, ϕ)Γ R = g R , ϕΓ R + g, ϕ ∀ϕ ∈ WΓ1,2 (Ω),
⎪
⎪
⎪
⎪ ∂t D
⎪ t
⎪
⎪
t
⎪
⎪
⎪
⎪
⎪
⎪ |σ n (θ, v, p)| ≤ gn , σn (θ, v, p)vn + gn |vn | = 0 on Γ9 ,
t t
⎪
⎪
⎪
⎪ σn (θ, v, p) + g+n ≥ 0, (σnt (θ, v, p) + g+n )vn = 0 on Γ10 ,
t
⎪
⎪
⎩ t
(8.22)
Define a02 (θ ; ·, ·), a12 (·, ·, ·) and f 2 ∈ V ∗ by
a02 (θ̃ ; w, u) = 2(μ(θ̃ )E(w), E(u)) + 2(μ(θ̃)k(x)w, u)Γ2 + 2(μ(θ̃ )S w̃, ũ)Γ3
+ 2(α(x)w, u)Γ5 + (μ(θ̃)k(x)w, u)Γ7 ∀w, u ∈ V, ∀θ̃ ∈ WΓ1,2
D
(Ω),
a12 (v, u, w) = rot v × u, w ∀v, u, w ∈ V,
i=2,4,7 i=3,5,6
(8.23)
Define b2 (θ ; ·, ·) and g2 ∈ (WΓ1,2
D
(Ω))∗ by
b2 (θ̃; θ, ϕ) = (κ(θ̃)∇θ, ∇ϕ) + (β(x)θ, ϕ)Γ R ∀ θ̃ , θ, ϕ ∈ WΓ1,2

D
(Ω),
(8.24)
g2 , ϕ = g R , ϕΓ R + g, ϕ ∀ϕ ∈ WΓ1,2
D
(Ω).
By (8.6) and (8.7),
f 2 ∈ L 2 (0, T ; V∗ ), g2 ∈ L 2 (0, T ; (WΓ1,2

D
)∗ ). (8.25)
1
2
and (8.22), we are led to the following variational formulation for problem (8.1),
(8.2) (7.4).
Problem II-VE. Find v ∈ K (Q), θ ∈ L ∞ (0, T ; L 2 (Ω)) ∩ L 2 (0, T ; WΓ1,2

D
(Ω)) and
− 21 − 21
(στ , σn , σ+n , σ−n ) ∈ Lτ (Γ8 ) × L (Γ9 ) × H (Γ10 ) × H (Γ11 ) for a.e. t ∈ (0, T )
t t t t 2 2
such that v(0) = v0 , θ (0) = θ0 and

⎧
⎪ ∂v
⎪
⎪ , u + a02 (θ ; v, u) + a12 (v, v, u) − (στt , u τ )Γ8 − (σnt , u n )Γ9
⎪
⎪ ∂t
⎪
⎪ t t
⎪
⎪
⎪
⎪
⎪
⎪
⎪ ∂θ
⎨ , ϕ + b2 (θ ; θ, ϕ) − θ v, ∇ϕ = g2 , ϕ ∀ϕ ∈ WΓ1,2 (Ω),
∂t D
(8.26)
⎪
⎪ |στt | ≤ gτ , στt · vτ + gτ |vτ | = 0 on Γ8 ,
⎪
⎪
⎪
⎪
⎪
⎪
⎪ t
⎪
⎪ σ+n
t
+ g+n ≥ 0, σ+n + g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎪
⎪
⎩ t t
σ−n − g−n ≤ 0, σ−n − g−n , vn Γ11 = 0 on Γ11 ,
(u, n)L2 (Γ8 ) = 0.
Remark 8.5 Similarly to Remark 8.3, Problem II-VE is equivalent to the problem
(8.1), (8.2), (7.4).
Define operators A2 (θ̃ ) : V → V∗ and B2 : V × V → V∗ , respectively, by
A2 (θ̃)v, u = a02 (θ̃; v, u) ∀v, u ∈ V, θ̃ ∈ WΓ1,2

D
(Ω),
(8.27)
B2 (v, u), w = a12 (v, u, v) ∀v, u, w ∈ V.
Let functional Ψ be defined by (8.16). Then, in the same way as in Problem I-VI of
Sect. 6.1 we find a variational inequality for velocity. Then we have another varia-
tional formulation consisting of a variational inequality for velocity and a variational
equation for temperature, which is equivalent to Problem II-VE if the solution is
smooth enough.

Problem II-VI. Find (v, θ ) ∈ L ∞ (0, T ; H ) ∩ L 2 (0, T ; V) × L ∞ (0, T ; L 2

(Ω)) ∩ L 2 (0, T ; WΓ1,2
D
(Ω)) such that
⎧ T
⎪
⎪
⎪
⎪ u + A2 (θ)v(t) + B2 (v(t), v(t)) − (1 − α0 θ) f − f 2 , u(t) − v(t) dt
⎪
⎪
⎪
⎪
0
⎪
⎪ 1
⎨ + Ψ (u) − Ψ (v) ≥ − v0 − u(0)2 ∀u ∈ L 4 (0, T ; V) with u ∈ L 2 (0, T ; V∗ ),
2 (8.28)
⎪
⎪
⎪
⎪
T ∂ϕ
⎪
⎪ − θ, + b2 (θ; θ, ϕ) − θv, ∇ϕ − g2 , ϕ dt
⎪
⎪ ∂t
⎪
⎪
0
⎩
= θ0 (x), ϕ(x, 0) ∀ϕ ∈ C 1 ([0, T ]; WΓ1,2
D
(Ω)) with ϕ(·, T ) = 0.

Static Pressure
The aim of this section is to prove the following

Theorem 8.1 (the case of static pressure) Let Assumption 8.1 be satisfied. Suppose
that
(1) The norms of f, φi , i = 2, · · · , 6, g, g R in the spaces that they belong to are
small enough (depending on α0 ) ;
(2) v0 ∈ V, Φ(v0 ) = 0;
(3) θ0 ∈ WΓ1,2 (Ω);
D
(4) A1 v0 + B1 (v0 , v0 ) − f 1 (0) ∈ H (compatibility condition at the initial time
for velocity);
0 V , A1 v0 + B1 (v0 , v0 ) − (1 − α0 θ0 ) f (0) − f 1 (0) are small enough;
(5) v
(6) C1 θ0 + v0 · ∇θ0 − g1 (0) ∈ L 2 (Ω) (compatibility condition at the initial
time for temperature);
(7) θ0 WΓ1,2 (Ω) and C1 θ0 + v0 · ∇θ0 − g1 (0) L 2 (Ω) are small enough.
D
Then there exists a solution (v, θ ) to (8.20) such that
v ∈ C([0, T ]; V), v ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; H ),

(8.29)
θ ∈ C([0, T ]; WΓ1,2
D
), θ ∈ L 2 (0, T ; WΓ1,2
D
(Ω)) ∩ L ∞ (0, T ; L 2 (Ω)).
Furthermore, the solution satisfying vV ≤ c, θ WΓ1,2 ≤ c for a constant c > 0

D
small enough is unique.
Theorem 8.1 will be proved in the next two subsections.
8.3.1 Existence and Estimation of Solutions to an

Approximate Problem
We first consider an approximate problem for (8.20).

Fréchet derivative of Φε (see Sect. 1.6).
By the fact that Γ2 j , Γ3 j , Γ7 j are in C 2.1 (Γi j ) and 4) of Assumption 8.1, there
exists a constant M such that
S(x)∞ , k(x)∞ , α L ∞ (Γ5 ) ≤ M.
Thus, there exists c∗ such that
μ
2 μ(k(x)z, z)Γ2 + 2μ(S z̃, z̃)Γ3 + (α(x)z, z)Γ5 + μ(k(x)z, z)Γ7 ≤ z2V + c∗ z2
4 (8.30)
∀z ∈ V
(see Theorem 1.27). Thus, we have
7μ
A1 u, u ≥ u2V − c∗ u2 ∀u ∈ V,
4 (8.31)
|A1 u, v| ≤ c1 uV vV ∃c1 > 0, ∀u, v ∈ V
and
|B1 (v, u), w| ≤ c2 vV uV wV , (8.32)
where the operators A1 , B1 are the ones in (8.18).

Let {u j , j = 1, 2, · · · } and {ϕ j , j = 1, 2, · · · } be, respectively, bases of the space
V and WΓ1,2 D
(Ω). Without loss of generality,
we assume that u 1 = v0 , ϕ1 = θ0 as in
[1]. We are to find a solution vm = mj=1 g jm (t)u j , θm = mj=1 r jm (t)ϕ j to problem
⎧
⎪ ∂vm
⎪
⎪ , u j + 2 μE (vm ), E (u j ) + (vm · ∇)vm , u j + 2(μk(x)vm , u j )Γ2
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ + 2(μS ṽm , ũ j )Γ3 + 2(α(x)vm , u j )Γ5 + (μk(x)vm , u j )Γ7 + ∇Φε (vm (t)), u j
⎪
⎨
= (1 − α0 θm ) f, u j + f 1 , u j ,
⎪
⎪
⎪
⎪ ∂θ
⎪
⎪
m
, ϕ j + (κ∇θm , ∇ϕ j ) + (β(x)θm , ϕ j )Γ R − vm θm , ∇ϕ j = g1 , ϕ j ,
⎪
⎪ ∂t
⎪
⎪
⎩
vm (0) = v0 , θm (0) = θ0 .
(8.33)
which gives us an ordinary differential system for g jm (t), r jm (t), j = 1, · · · , m.
The solutions to (8.33) depend on ε, but for convenience of notation hereafter
we use subindex m instead of subindex m, ε. For a tm there exist absolute con-
tinuous functions g jm (t) and r jm (t) on [0, tm ). Since f ∈ W 1,∞ (0, T ; L3 (Ω)),
f 1 ∈ W 1,∞ (0, T ; V∗ ), g1 ∈ W 1,2 (0, T ; (WΓ1,2

D
)∗ ) and ∇Φε is Lipschitz continuous,

g jm (t) and r jm (t) are in fact absolute continuous. If vm (t), θm (t) are bounded
and vm (t), θm (t) are integrable, then g jm (t), r jm (t) are prolonged over tm . Under
smallness of the data of problem and compatibility condition of the data at the initial
instant we will find estimates for vm (t) and θm (t) below, by which we obtain
(8.82) and see that tm = T.
Multiplying the first and second equation of (8.33), respectively, by g jm (t), ϕ jm (t)
and adding for i = 1, · · · , m, we get
⎧
⎪ ∂vm
⎪
⎪ , vm + 2 μE (vm ), E (vm ) + (vm · ∇)vm , vm + 2(μk(x)vm , vm )Γ2
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ + 2(μS ṽm , ṽm )Γ3 + 2(α(x)vm , vm )Γ5 + (μk(x)vm , vm )Γ7 + ∇Φε (vm (t)), vm
⎪
⎨
= (1 − α0 θm ) f, vm + f 1 , vm ,
⎪
⎪
⎪
⎪ ∂θ
⎪
⎪
m
, θm + (κ∇θm , ∇θm ) + (β(x)θm , θm )Γ R − vm θm , ∇θm = g1 , θm ,
⎪
⎪ ∂t
⎪
⎪
⎩
vm (0) = v0 , θm (0) = θ0 .
(8.34)
We will find a priori estimate for
I (t) := vm (t)2 + vm (t)2 + θm (t)2 + θm (t)2 .
Φε (u) − Φε (vm (t)) ≥ ∇Φε (vm (t)), u − vm (t) ∀u ∈ V. (8.35)
On the other hand, by condition (2) of the theorem and (1.37), we have that Φε (0V ) =
0. Thus
0 ≤ Φε (vm (t)) ≤ ∇Φε (vm (t)), vm (t). (8.36)
Also,
μ
2|−α0 θm (t) f, vm (t)| ≤ c|α0 |θm (t)2W 1,2 f 2L3 + vm (t)2V . (8.37)
ΓD 4
By virtue of (8.31), (8.32), (8.36) and (8.37), we have from the first equation of
(8.34)
d 7μ
vm (t)2 + vm (t)2V − 2c2 vm (t)3V + 2Φε (vm (t))
dt 2
μ
≤ c|α0 |θm (t)2 f 2L3 + c f 2L3 + c f 1 2V∗ + vm (t)2V + 2c∗ vm (t)2 ,
WΓ1,2 2
D
where c∗ and c2 are, respectively, the one in (8.30) and (8.32), and so
d
vm (t)2 + 3μ − 2c2 vm (t)V vm (t)2V + 2Φε (vm (t))
dt
≤ c|α0 |θm (t)2W 1,2 f 2L3 + c f 2L3 + c f 1 2V∗ + 2c∗ vm (t)2 .
ΓD
(8.38)
Here and the rest of this section constants independent of the data of problem are
written by c with the exceptions of c∗ , c2 .
Setting t = 0 in the first equation of (8.33), multiplying the resulting equation by
g jm (0) and adding for j = 1, · · · , m, we get
vm (0)2 + A1 vm (0), vm (0) + B1 (vm (0), vm (0)), vm (0) + ∇Φε (v0 ), vm (0)
= (1 − α0 θ0 ) f (0), vm (0) + f 1 (0), vm (0).
(8.39)
By condition (2) of theorem, for any u ∈ V we have Φ(u) ≥ Φ(v0 ) = 0, which by
(1.37) implies Φε (v0 ) = 0 and ∇Φε (v0 ) = 0. Then, we have from (8.39)
vm (0) ≤ A1 v0 + B1 (v0 , v0 ) − (1 − α0 θ0 ) f (0) − f 1 (0), (8.40)
which is valid by the compatibility condition at the initial time for velocity (condition
(4)) and the conditions for θ0 , f . On the other hand, taking into account (8.31), (8.32)
and (8.36), we have from the first equation of (8.34)
7μ
vm (t)2V ≤2c2 vm 3V + 2(1 − α0 θm (t)) f (t), vm (t)
2
+ 2 f 1 (t), vm (t) + 2c∗ vm 2 − 2(vm (t), vm (t)),
and so
3μvm (t)V ≤2c2 vm (t)2V + c f (t)L3 + c|α0 |θm (t) f (t)L3
(8.41)
+ c f 1 (t)V∗ + (2δvm (t) + 2c∗ δvm (t)),
where δ is such that · ≤ δ · V .

Since vm θm , ∇θm = 0 by (8.8), we get from the second equation of (8.34)
d
θm (t)2 + 2κθm (t)2W 1,2 + 2(β(x)θm , θm )Γ R = 2 g1 , θm (t) . (8.42)
dt Γ D
We have from (8.42)
d 1
θm (t)2 + κθm (t)2W 1,2 + (β(x)θm , θm )Γ R ≤ g1 2(W 1,2 )∗ (8.43)
dt Γ D κ ΓD
and
t t
1
θm (t)2 + κ|∇θm (s)|2 d xds ≤ θ0 2 + g1 (s)2(W 1,2 )∗ ds. (8.44)
0 Ω κ 0 ΓD
Setting t = 0 in the second equation of (8.33), multiplying the resulting equation

by r jm (0) and adding for j = 1, · · · , m, we get
θm (0)2 + b1 (θ0 , θm (0)) + (v0 · ∇θ0 , θm (0)) = g1 (0), θm (0), (8.45)
where −(v0 θ0 , ∇θm (0)) = (v0 · ∇θ0 , θm (0)) was used. We have from (8.45)
θm (0) ≤ C1 θ0 + v0 · ∇θ0 − g1 (0), (8.46)
which is valid by the compatibility condition at the initial time for temperature
(condition (6)).
On the other hand, taking into account vm θm , ∇θm = 0, we have from the second
equation of (8.34)
κθm (t)2W 1,2 ≤ g1 (WΓ1,2 )∗ θm (t)WΓ1,2 + δ1 θm (t)WΓ1,2 θm (t),
ΓD D D D
where δ1 is such that · ≤ δ1 · WΓ1,2 , and so

D
1
θm (t)WΓ1,2 ≤ g1 (WΓ1,2 )∗ + δ1 θm (t) . (8.47)
D κ D
substituting (8.44) into (8.41), we have
3μvm (t)V ≤c2 vm (t)2V + c f L ∞ (0,T ;L3 )

1/2
1 T
+ c|α0 | θ0 2 + g1 2(W 1,2 )∗ ds f L ∞ (0,T ;L3 ) (8.48)
κ 0 ΓD
+ f 1 L ∞ (0,T ;V∗ ) + max{2δ, 2c∗ δ}(vm (t) + vm (t)).
Differentiating the first equality of (8.33) with respect to t yields

vm (t), v j + A1 vm (t), v j + (B1 (vm (t), vm (t)) , v j + (∇Φε (vm )) , v j
(8.49)
= − α0 θm (t) f, v j − (1 − α0 θm (t)) f , v j + f 1 , v j .
Multiplying (8.49) by g jm (t) and summing for j yield

vm (t), vm (t) + A1 vm (t), vm (t) + (B1 vm (t), vm (t)) , vm

(t) + (∇Φε (vm )) , vm

(t)
= −α0 θm (t) f, vm

(t) − (1 − α0 θm (t)) f , vm

(t) + f 1 , vm

(t).
(8.50)
Calculating (B((vm (t), vm (t))) , we have
|(B1 vm (t), vm (t)) , vm (t)| = |B1 (vm , vm ), vm + B1 (vm , vm ), vm |
(8.51)
≤ 2c2 vm V vm 2V ,
where c2 is the one in (8.32). Also, by Hölder’s and Young’s inequalities we have
μ
2|α0 θm (t) f, vm (t)| ≤ c|α0 |θm (t)2W 1,2 f 2L3 + v (t)2V ,
ΓD 8 m
μ
2|α0 θm (t) f , vm (t)| ≤ c|α0 |θm (t)2W 1,2 f 2L3 + vm (t)2V ,
ΓD 8 (8.52)
μ
2| f (t), vm (t)|
≤ c f
+ vm (t)2V ,
(t)2L3
8
μ
2| f 1 (t), vm (t)| ≤ c f 1 (t)V∗ + vm (t)2V .
2
8
Taking into account (8.31), (8.51), (8.52) and the fact that (∇Φε (vm )) , vm ≥ 0
(see (1.38)), we have from (8.50)
d μ
vm (t)2 + (3μ − 4c2 vm (t)V )vm (t)2V + vm (t)2V
dt 2
≤ cα0 θm (t)2W 1,2 f 2L3 + c|α0 |θm (t)2W 1,2 f 2L3
ΓD ΓD
μ
+ c f 2L3 + c f 1 2V∗ + 2c∗ vm (t)2 + v (t)2V ,
2 m
that is,
d
v (t)2 + (3μ − 4c2 vm (t)V )vm (t)2V
dt m
≤ c|α0 |θm (t)2W 1,2 f 2L3 + c|α0 |θm (t)2W 1,2 f 2L3 (8.53)
ΓD ΓD
+ c f 2L3 + c f 1 2V∗ + 2c∗ vm (t)2 .
Differentiating the second equality of (8.33) with respect to t, we have

θm (t), ϕ j + C1 θm (t), ϕ j − vm θ, ∇ϕ j − vm θ , ∇ϕ j = g1 , ϕ j . (8.54)
Multiplying (8.54) by r jm (t) and adding for j, we have
(t), θ (t) + C θ (t), θ (t) − v θ , ∇θ (t) − v θ , ∇θ (t) = g , θ (t).

θm m 1 m m m m m m m 1 m
(8.55)
On the other hand, we get
c 2
2|vm θ, ∇θm (t)| ≤ v θ 2W 1,2 + κθm (t)2W 1,2 . (8.56)
κ m V ΓD ΓD
Taking into account vm θ , ∇θm (t) = 0 (see (8.8)) and (8.56), we have from (8.55)
d
θ (t)2 + 2κθm (t)2W 1,2
dt m ΓD
c 2 κ
≤ vm V θ 2W 1,2 + κθm (t)2W 1,2 + cg1 2(W 1,2 )∗ + θm (t)2W 1,2 .
κ ΓD ΓD ΓD 4 ΓD
(8.57)
Rewriting (8.57) yields
d 3κ c
θm (t)2 + θm (t)2W 1,2 ≤ vm 2V θ 2W 1,2 + cg1 2(W 1,2 )∗ . (8.58)
dt 4 ΓD κ ΓD ΓD
Adding (8.38), (8.53), (8.43) and (8.58), we have
d c

I (t) + (2μ − 4c2 vm (t)V ) vm (t)2V + vm (t)2V + μ − θm 2 1,2 vm (t)2V
dt κ WΓ
D
3κ
+ (κ − c|α0 | f 2L3 − c|α0 | f 2L3 )θm 2 1,2 + − c|α0 | f 2L3 θm 2 1,2
WΓ 4 WΓ
D D
≤ c( f 2L3 + f 2L3 ) + c( f 1 2V∗ + f 1 2V∗ ) + c(g1 2 + g1 2 )

(WΓ1,2 )∗ (WΓ1,2 )∗
D D

+ 2c∗ (vm (t)2 + vm (t)2 ).
(8.59)
Integrating (8.59) yields
t c

I (t) + (2μ − 4c2 vm (s)V ) vm (t)2V + vm (s)2V + μ − θm 2 1,2 vm (s)2V
0 κ W ΓD
3κ
+ (κ − c|α0 | f 2L3 − c|α0 | f 2L3 )θm 2 1,2 + − c|α0 | f 2L3 θm 2 1,2 ds
WΓ 4 W ΓD
D
t

≤ I (0) + F(t) + 2c∗ (vm (s)2 + vm (s)2 ) ds,
0
(8.60)
where
F(t) :=ct ( f 2W 1,∞ (0,T ;L3 ) + f 1 2W 1,∞ (0,T ;V∗ ) ) + cg1 2W 1,2 (0,t;(W 1,2 )∗ ) . (8.61)
ΓD
By (8.40) and (8.46) we have
I (0) ≤v0 2 + A1 v0 + B1 (v0 , v0 ) − (1 − α0 θ0 ) f (0) − f 1 (0)2

(8.62)
+ θ0 2 + C1 θ0 + v0 · ∇θ0 − g1 (0)2 .
By the condition of theorem, we can assume f W 1,∞ (0,T ;L3 ) to be so small that
κ − c|α0 | f (t)2L3 − c|α0 | f (t)2L3 ≥ 0 at a.e. t ∈ [0, T ],

3κ (8.63)
− c|α0 | f (t)2L3 ≥ 0 at a.e. t ∈ [0, T ].
4
If
μ
vm (0)V = v0 V < (8.64)
2c2
and μκ
θm (0)WΓ1,2 = θ0 WΓ1,2 < (8.65)
D D c
are valid, then there exists a tm such that on [0, tm ]
2μ − 4c2 vm (t)V ≥ 0,

c (8.66)
μ − θm (t)WΓ1,2 ≥ 0.
κ D
Therefore, taking into account (8.62), by Gronwall’s inequality we have

I (t) ≤ v0 2 + A1 v0 + B1 (v0 , v0 ) − (1 − α0 θ0 ) f (0) − f 1 (0)2
(8.67)
+ θ0 2 + C1 θ0 + v0 · ∇θ0 − g1 (0)2 + F(T ) e2c∗ t
on the all intervals of t satisfying (8.66).

Using this estimate, we will obtain a quadratic inequality satisfied by vm (t)V .
Put

β := v0 2 + A1 v0 + B1 (v0 , v0 ) − (1 − α0 θ0 ) f (0) − f 1 (0)2
(8.68)
+ θ0 2 + C1 θ0 + v0 · ∇θ0 − g1 (0)2 + F(T ) e2c∗ T .
Obviously, β depends only on the data of problem.

Then, when f satisfies (8.63), we can get from (8.67)
√ 1
vm (t) + vm (t) ≤ 2 vm (t)2 + vm (t)2 2 ≤ 2β,
(8.69)
θm (t) ≤ β
on [0, tm ] where (8.66) holds. Let the data of problem be so small that
c μ
(g1 W 1,2 (0,T ;(WΓ1,2 )∗ ) + δ1 β) ≤ . (8.70)
κ 2 D 2
By (8.47) and (8.69), for the small data of problem we have on [0, tm ]
c c μ
θm (t)WΓ1,2 ≤ 2 (g1 W 1,2 (0,T ;(WΓ1,2 ))∗ + δ1 β) ≤ ,
κ D κ D 2
which implies
c μ
μ− θm (t)WΓ1,2 ≥ ∀t ∈ [0, tm ]. (8.71)
κ D 2
Therefore, for such small data of problem that (8.70) is valid, if
2μ − 4c2 vm (t)V ≥ 0 ∀t ∈ [0, tm + γ ], γ > 0, tm + γ ≤ T,
then owing to (8.71), step by step, we have

c μ
μ− θm (t)WΓ1,2 ≥ ∀t ∈ [0, tm + γ ]. (8.72)
κ D 2
From above we see that for the small data of problem satisfying (8.63)–(8.65) and
(8.70),
c μ
μ − θm (t)WΓ1,2 ≥ (8.73)
κ D 2
is valid on the interval where the first inequality of (8.66) holds.
Put
1 T 1/2
γ := f (t) L ∞ (0,T ;L3 ) + c|α0 | θ0 2 + g1 2(W 1,2 )∗ ds f (t) L ∞ (0,T ;L3 )
κ 0 ΓD

+ f 1 (t) L ∞ (0,T ;V ∗ ) + max{2δ, 2c∗ δ} 2β.
(8.74)
By (8.44), (8.69) and (8.41), for the small data satisfying (8.63)–(8.65) and (8.70)
we have a quadratic inequality for vm (t)V , which is the one we want,
0 ≤ γ − 3μvm (t)V + 2c2 vm (t)2V (8.75)
on the intervals where the first inequality of (8.66) is satisfied.

By the conditions of theorem, we can assume that the data of problem are so small
that (8.63)–(8.65) and (8.70) are valid and γ satisfies the following inequality
9μ2 − 8c2 γ > 4μ2 . (8.76)
Now, let us prove that if

3μ − 9μ2 − 8c2 γ μ
v0 V ≤ < , (8.77)
4c2 4c2
then for any m

2μ − 4c2 vm (t)V ≥ μ ∀t ∈ [0, T ]. (8.78)
Since 2μ − 4c2 v0 V > μ, on a interval [0, tm ]
2μ − 4c2 vm (t)V ≥ μ.

Let us prove that if the first inequality of (8.66) is valid on an interval [0, t m ], then
more strong estimate
2μ − 4c2 vm (t)V ≥ μ ∀t ∈ [0, t m ] (8.79)
is valid. To this end, putting y = vm (t)V in (8.75) (which is valid on the interval
where the first inequality of (8.66) holds when (8.63)–(8.65) and(8.70) are valid),
we get
0 ≤ γ − 3μy + 2c2 y 2 on [0, t m ]. (8.80)
By virtue of (8.76), there exist two real roots of polynomial γ − 3μy + c2 y 2

3μ − 9μ2 − 8c2 γ 3μ + 9μ2 − 8c2 γ
y1 = and y2 = ,
4c2 4c2
and on the intervals [0, y1 ] and [y2 , +∞) (8.80) holds. Thus, by continuity
of vm (t)V with respect to t we have from v(0)V ∈ [0, y1 ] that vm (t)V ∈ [0, y1 ]
∀t ∈ [0, t m ], that is,

3μ − 9μ2 − 8c2 γ μ
vm (t)V ≤ < ∀t ∈ [0, t m ].
4c2 4c2
Thus,
2μ − 4c2 vm (t)V > μ ∀t ∈ [0, t m ], (8.81)
which shows (8.79). Thus, step by step we see that the first inequality of (8.66) is
valid on [0, T ] and so is (8.78).
If (8.77) is valid, then so is (8.64). Therefore, for the small data satisfying (8.63),
(8.65), (8.70), (8.76) and (8.77), we have also (8.73) on [0, T ]. By (8.78) and (8.73),
we have
μ
vm (t)V ≤ ∀t ∈ [0, T ], ∀m, ∀ε > 0,
4c2
c μ (8.82)
θm (t)WΓ1,2 ≤ ∀t ∈ [0, T ], ∀m, ∀ε > 0.
κ D 2
Note that β, γ depend on α0 , and the restriction of data depends on α0 .
Then, by (8.67) and (8.60), we have
vm (t) ≤ const ∀t ∈ [0, T ], ∀m, ∀ε > 0,

vm L 2 (0,T ;V) ≤ const ∀m, ∀ε > 0,
(8.83)
θm (t) ≤ const ∀t ∈ [0, T ], ∀m, ∀ε > 0,
θm L 2 (0,T ;WΓ1,2 ) ≤ const ∀m, ∀ε > 0.
D
By (8.82),
T
Φε (vm (t)) dt ≤ const ∀m, ∀ε > 0, (8.84)
0
and so by (1.36) and (1.37),

T
vm (t) − Jε (vm (t))2V dt ≤ cε ∀m, ∀ε > 0. (8.85)
0
8.3.2 Existence and Uniqueness of a Solution
In this subsection we complete proof of Theorem 8.1. Let us first prove existence of
a solution. Due to (8.82) and (8.83), we can extract a subsequence from {(vm , θm )}
obtained in preceding subsection, which is still denoted with the subindex as before,
such that
vm → v in C([0, T ]; V),
vm v in L 2 (0, T ; V),
∗
vm v in L ∞ (0, T ; H ),
(8.86)
θm → θ in C([0, T ]; WΓ1,2
D
),
θm θ in L 2 (0, T ; WΓ1,2
D
),
∗
θm θ in L ∞ (0, T ; L 2 (Ω))
when m → ∞ and ε → 0 (note that for convenience in the preceding subsection we

m instead of m, ε).
used subindex
Put u = M j=1 k j (t)u j , where k j (t) ∈ C [0, T ], M is a positive integer. Multiply
1
the first equation of (8.33) by k j (t) and sum for j = 1, · · · , M. Then, multiply
the first equation of (8.33) by g jm (t) and sum for j = 1, · · · , m. Subtracting the
resulting equations yields
vm (t) + A1 vm (t) + B1 (vm (t), vm (t)) + ∇Φε (vm ), u(t) − vm (t)
(8.87)
= (1 − α0 θ ) f + f 1 , u(t) − vm (t).
Φε (u(t)) − Φε (vm (t)) ≥ ∇Φε (vm (t)), u − vm (t). (8.88)

T
vm (t) + A1 vm (t) + B1 (vm (t), vm (t)) − (1 − α0 θm ) f − f 1 , u(t) − vm (t) dt
0
T
+ Φε (u(t)) − Φε (vm (t)) dt ≥ 0.
0
(8.89)
Since Φ(u) ≥ Φε (u) and Φ(Jε wm (t)) ≤ Φε (wm (t)) (see (1.37)), we have from
(8.89)
T
vm (t) + A1 vm (t) + B1 (vm (t), vm (t)) − (1 − α0 θm ) f − f 1 , u(t) − vm (t) dt
0
T
+ Ψ (u(t)) − Φ(Jε vm (t)) dt ≥ 0. (8.90)
0
By (8.85), we have that Jε vm → v in L 2 (0, T ; V) as m → ∞ and ε → 0, and so by

lower semi-continuity of Φ we have
T T
lim Φ(Jε vm (t)) dt ≥ Φ(v(t)) dt. (8.91)
m→∞,ε→0 0 0
By standard way we can prove that

T T
B1 (vm (t), vm (t)), v(t) dt → B1 (v(t), v(t)), v(t) dt (8.92)
0 0
as m → ∞ and ε → 0.
Since limm k →∞,ε→0 A1 vm (t), vm (t) = A1 v(t), v(t), by (8.91) and (8.92), we
have from (8.90)
T
v (t) + A1 v(t) + B1 (v(t), v(t)) − (1 − α0 θ ) f − f 1 , u(t) − v(t) dt
0
+ Ψ (u) − Ψ (v) ≥ 0. (8.93)
Since vm (0) = v0 , by (8.86) it is obvious that v(0) = v0 .

B1 (v(t), v(t)) ∈ L ∞ (0, T ; V∗ ) and the set {u = M j=1 k j (t)u j ;
k j (t) ∈ C [0, T ], M : positive integer} is dense in L (0, T ; V), and so (8.93) is valid
1 4
for all u ∈ L 4 (0, T ; V).

By (8.86), from the second equation of (8.33) we can get
T ∂θm T
, ϕ + (κ∇θm , ∇ϕ) + (β(x)θm , ϕ)Γ R −vm θm , ∇ϕ dt = g1 , ϕ dt,
0 ∂t 0
ϕ ∈ L 2 (0, T ; WΓ1,2
D
). (8.94)
It is easy to see that

(κ∇θm , ∇ϕ) → (κ∇θ, ∇ϕ) for a.e. t ∈ [0, T ].
Also,
T
vm θm , ∇ϕ − vθ, ∇ϕ dt
0
T T
≤ vm − vL6 θm L 3 ∇ϕL2 dt + v(θm − θ ), ∇ϕ dt
0 0
T
≤ vm − v L ∞ (0,T ;V) θm L 2 (0;T ;L 3 ) ϕ L 2 (0,T ;WΓ1,2 ) + |v(θm − θ ), ∇ϕ| dt.
D
0
(8.95)
Since v∇ϕ ∈ L 2 (0, T ; L 6/5 (Ω)), by (8.86) we have that as m → ∞ and ε → 0,
T
0 |v(θm − θ ), ∇ϕ| dt → 0. Thus,
T T
vm θm , ∇ϕ dt → vθ, ∇ϕ dt.
0 0

T T
(β(x)θm , ϕ)Γ R dt → (β(x)θ, ϕ)Γ R dt.
0 0
Therefore, we have from (8.94)

T ∂θ T
, ϕ + (κ∇θ, ∇ϕ) + (β(x)θ, ϕ)Γ R − vθ, ∇ϕ dt = g1 , ϕ dt.
0 ∂t 0
(8.96)
Since θm (0) = θ0 , by (8.86) it is obvious that θ (0) = θ0 .
Therefore, we have proved the existence of a solution.
Next, let us prove uniqueness of a solution. Let (v1 , θ1 ), (v2 , θ2 ) be two solutions
to Problem I-VI satisfying inequality (8.82) instead of approximate solutions. Then,
taking (8.29) into account (Remark 8.4), we have from (8.21)
v1 (t) + A1 v1 (t) + B1 (v1 (t), v1 (t)) − (1 − α0 θ1 ) f − f 1 , v2 (t) − v1 (t)

+ Φ(v2 (t)) − Φ(v1 (t)) ≥ 0,
v2 (t) + A1 v2 (t) + B1 (v2 (t), v2 (t)) − (1 − α0 θ2 ) f − f 1 , v1 (t) − v2 (t)
+ Φ(v1 (t)) − Φ(v2 (t)) ≥ 0,
which imply
v1 (t) − v2 (t), v1 (t) − v2 (t) + A1 (v1 (t) − v2 (t)), v1 (t) − v2 (t)
≤ |α0 ||(θ1 − θ2 ) f, v1 (t) − v2 (t))| + |B1 (v1 (t), v1 (t)) − B1 (v2 (t), v2 (t)), v1 (t) − v2 (t)|.
(8.97)
By virtue of (8.31) and (8.32), we have
d 7μ
((v1 (t) − v2 (t)2 ) + v1 (t) − v2 (t)2V
dt 2
μ
≤ 2c∗ v1 (t) − v2 (t)2 + v1 (t) − v2 (t)2V + c f 2L3 θ1 − θ2 2
2
+ 2|B(v1 (t) − v2 (t), v1 (t)), v1 (t) − v2 (t)|
+ 2|B(v2 (t), v1 (t) − v2 (t)), v1 (t) − v2 (t)|
μ
≤ 2c∗ v1 (t) − v2 (t)2 + v1 (t) − v2 (t)2V + c f 2L3 θ1 − θ2 2
2
+ 2c2 (v1 (t)V + v2 (t)V )v1 (t) − v2 (t)2V ,
where c2 is the one in (8.32). By (8.82),
2c2 (v1 (t)V + v2 (t)V ) ≤ μ,
and so we have
dv1 (t) − v2 (t)2

+ 2μv1 (t) − v2 (t)2V ≤ 2c∗ v1 (t) − v2 (t)2 + c f 2L3 θ1 − θ2 2 .
dt
(8.98)
Also, from
∂θ1
, ϕ + (κ∇θ1 , ∇ϕ) + (β(x)θ1 , ϕ)Γ R − v1 θ1 , ∇ϕ1 = g1 , ϕ,
∂t
∂θ2
, ϕ + (κ∇θ2 , ∇ϕ) + (β(x)θ2 , ϕ)Γ R − v2 θ2 , ∇ϕ = g1 , ϕ
∂t
we have
∂θ1 − θ2
, θ1 − θ2 + κ(∇θ1 − ∇θ2 , ∇θ1 − ∇θ2 ) + (β(x)(θ1 − θ2 ), θ1 − θ2 )Γ R
∂t
− v1 (θ1 − θ2 ), ∇(θ1 − θ2 ) − (v1 − v2 )θ2 , ∇(θ1 − θ2 ) = 0.
(8.99)
Taking into account v1 (θ1 − θ2 ), ∇(θ1 − θ2 ) = 0 (see (8.8)), by (8.56) we have
d c
θ1 − θ2 2 + 2κ∇θ1 − ∇θ2 2 ≤ v1 (t) − v2 (t)2V θ2 (t)2 1,2 + κ∇θ1 − ∇θ2 2 ,
dt κ WΓ
D
and so
d c
(θ1 (t) − θ2 (t)2 ) ≤ v1 (t) − v2 (t)2V θ2 (t)2W 1,2 . (8.100)
dt κ ΓD
By (8.82),
c μ
θ2 (t)2W 1,2 ≤ .
κ ΓD 2
Therefore adding (8.98) and (8.100), we get
d
(v1 (t) − v2 (t)2 + θ1 (t) − θ2 (t)2 )
dt
≤ 2c∗ + c f 2L3 v1 (t) − v2 (t)2 + θ1 (t) − θ2 (t)2 . (8.101)
We have from (8.101)
v1 (t) − v2 (t)2 + θ1 (t) − θ2 (t)2

t (8.102)
≤ 2c∗ + 2c f 2L3 v1 (s) − v2 (s)2 + θ1 (s) − θ2 (s)2 ds,
0
which implies v1 (t) = v2 (t), θ1 (t) = θ2 (t) for all t ∈ [0, T ], and the proof is com-
plete.
Theorem 8.2 (the case of total pressure) Let Assumption 8.2 be satisfied, v0 ∈ HK
and θ0 ∈ L 2 (Ω). Then there exists a solution (v, θ )∈ L ∞ (0, T ; H ) ∩L 2 (0, T ; V) ×
∞
L (0, T ; L 2 (Ω)) ∩ L 2 (0, T ; WΓ1,2
D
(Ω)) to (8.28).
As in the preceding section, we will show Theorem 8.2 in the next two subsections.
8.4.1 Existence of a Solution to an Approximate Problem
We first consider an approximate problem for (8.28). For every 0 < ε < 1, let a
functional Φε be Moreau-Yosida approximation of Φ.
Let {u j , j = 1, 2, · · · } and {ϕ j , j = 1, 2, · · · } be, respectively, bases of the space
V and WΓ1,2 D
(Ω). Without
loss of generality,we assume that u 1 = v0 , ϕ1 = θ0 . We
find a solution vm = mj=1 g jm (t)u j , θm = mj=1 r jm (t)ϕ j to the following problem
⎧
⎪ ∂vm
⎪
⎪ , u j + 2 μ(θm )E (vm ), E (u j ) + rotvm × vm , u j + 2(μ(θm )k(x)vm , u j )Γ2
⎪
⎪ ∂t
⎪
⎪
⎪
⎪
⎨ + 2(μ(θm )S ṽm , ũ j )Γ3 + 2(α(x)vm , u j )Γ5 + (μ(θm )k(x)vm , u j )Γ7 + ∇Φε (vm (t)), u j
⎪
= (1 − α0 θm ) f, u j + f 2 , u j ,
⎪
⎪
⎪
⎪ ∂θm
⎪
⎪
⎪ ∂t , ϕ j + (κ(θm )∇θm , ∇ϕ j ) + (β(x)θm , ϕ j )Γ R − vm θm , ∇ϕ j = g2 , ϕ j ,
⎪
⎪
⎪
⎩
vm (0) = v0 , θm (0) = θ0 .
(8.103)
which gives us a system for g jm (t), r jm (t), j = 1, · · · , m. The solutions to (8.103)
depend on ε, but for convenience of notation here and in what follows of this section
we use subindex m. For a tm there exist absolute continuous functions g jm (t) and
r jm (t) on [0, tm ). If vm (t), θm (t) are bounded and vm (t), θm (t) are integrable,
then g jm (t), r jm (t) are prolonged over tm . We will find estimates (8.118) below, by
which we see that tm = T.
Multiplying the first and second equations of (8.103), respectively, by g jm (t) and
ϕ jm (t) and adding for i = 1, · · · , m, we get
⎧
⎪ ∂vm
⎪
⎪ , vm + 2 μ(θm )E (vm ), E (vm ) + rot vm × vm , vm + 2(μ(θm )k(x)vm , vm )Γ2
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ + 2(μ(θm )S ṽm , ṽm )Γ3 + 2(α(x, t)vm , vm )Γ5 + (μ(θm )k(x)vm , vm )Γ7
⎪
⎨
+ ∇Φε (vm (t)), vm + α0 θm f, vm = f, vm + f 2 , vm , (8.104)
⎪
⎪
⎪
⎪ ∂θm
⎪
⎪ , θm + (κ(θm )∇θm , ∇θm ) + (β(x)θm , θm )Γ R − vm θm , ∇θm = g2 , θm ,
⎪
⎪ ∂t
⎪
⎪
⎩
vm (0) = v0 , θm (0) = θ0 .
Let us estimate terms on the left hand side above. It is easy to see
2 μ(θm )E(vm ), E(vm ) dt ≥ 2μ0 vm 2V . (8.105)
By the fact that Γ2 j , Γ3 j , Γ7 j are in C 2.1 (Γi j ) and Assumption 8.2, there exists a
constant M such that
S(x)∞ , k(x)∞ , αL∞ (Γ5 ) ≤ M. (8.106)
Thus,

2(μ(θm )k(x)v, v)Γ2 + 2(μ(θm )S ṽm , ṽm )Γ3 + 2(α(x)vm , vm )Γ5 + (μ(θm )k(x)vm , vm )Γ7
μ0
≤ vm 2V + k11 vm 2
2
(8.107)
(see Theorem 1.27). Obviously,
rot vm × vm , vm = 0. (8.108)
Φε (y) − Φε (x) ≥ ∇Φε (x), y − x ∀x, y ∈ V. (8.109)
Thus,
∇Φε (vm (t)), 0V − vm (t) ≤ Φε (0V ) − Φε (vm (t)) ≤ −Φε (vm (t)),
and
∇Φε (vm (t)), vm (t) ≥ Φε (vm (t)). (8.110)
Also, by the Hölder inequality we have

κ0

α0 θm f, vm dt ≤ k12 vm (t)2 + θm (t)2W 1,2 , (8.111)
4 ΓD
where k12 = c|α0 | f 2L ∞ (0,T ;L3 ) .
μ0
| f, wm + f 2 , wm | ≤ wm 2V + c( f 2L3 + f 2 2V∗ ). (8.112)
2
Also, we have
κ(θm )∇θm , ∇θm ) ≥ κ0 θm 2W 1,2 ,
ΓD
(8.113)
(β(x)θm , θm )Γ R ≥ 0.
By (8.8) we have
wm θm , ∇θm = 0. (8.114)
Also,
κ0
|g2 , θm | ≤ θm (t)2W 1,2 + cg2 2(W 1,2 )∗ . (8.115)
4 ΓD ΓD
Taking
k1 = k11 + k12 (8.116)
and using (8.105)–(8.115), we have from (8.104)
d d
vm (t)2 + θm (t)2 + 2μ0 vm 2V + κ0 θm 2W 1,2 + Φε (vm (t))
dt dt ΓD

≤ c f (t)L3 + f 2 (t)V∗ + g2 (t)2(W 1,2 )∗ + 2k1 vm (t)2 .
2 2
ΓD
(8.117)
Applying Gronwall’s inequality, we have from (8.117)
vm (t)2 + θm (t)2

t
≤ v0 2 + θ0 2 + f (s)2L3 + f 2 (s)2V∗ + g2 (s)2 ds e2k1 t ,
0 (WΓ1,2 )∗
D
vm 2L 2 (0,T ;V) + θm 2

2
L (0,t;WΓ1,2 )
D
T
≤ c v0 + θ0 +
2 2
f (t)2L3 + f 2 (t)2V∗ + g2 (t)2 dt ,
0 (WΓ1,2 )∗
D
T T
Φε (vm (t)) dt ≤ c v0 2 + θ0 2 + f (t)2L3 + f 2 (t)2V∗ + g2 (t)2 dt .
0 0 (WΓ1,2 )∗
D
(8.118)
Note that c in (8.118) depends on T and f (via k12 ), but independent of m and ε.
Since Φ is nonnegative, by (1.36) and the third inequality of (8.118), we have
T
vm (t) − Jε (vm (t))2V dt ≤ cε (8.119)
0
with c independent of ε.
Multiplying the first equation of (8.103) by g jm (t) − g jm (s), summing for j and
taking into account (8.109), we have
1 dvm (t) − vm (s)2

2 dt
+ A2 (θm )vm (t) + B2 (vm (t), vm (t)) + α0 θm (t) f (t) − f (t) − f 2 (t), vm (t) − vm (s)
= ∇Φε (vm (t)), vm (s) − vm (t) ≤ Φε (vm (s)) − Φε (vm (t)) ≤ Φε (vm (s)),
(8.120)
where the operators A2 (θm ), B2 are the ones in (8.27).
By (8.107) and (8.106), we have
3μ0
A2 (θm )vm (t), vm (t) ≥ vm (t)2V − k11 vm (t)2 ,
2 (8.121)
|A2 (θm )vm (t), vm (s)| ≤ cvm (t)V vm (s)V ,
where k11 is the one in (8.107). Taking into account (8.121) and the fact that
B2 (vm (t),
vm (t)), vm (t) = 0, we have from (8.120)
1 dvm (t) − vm (s)2

≤Φε (vm (s)) + A2 (θm )vm (t), vm (s) + B2 (vm (t), vm (t)), vm (s)
2 dt
+ −α0 θm (t) f (t) + f (t) + f 2 (t), vm (t) − vm (s) + k11 vm (t)2 .
(8.122)
Let us integrate every term of (8.122) first with respect to t from s to s + h and
then with respect to s from 0 to T , where vm (t) = 0 when t ∈ (T, T + h). It is clear
that
T s+h
dvm (t) − vm (s)2 T
dtds = vm (s + h) − vm (s)2 ds. (8.123)
0 s dt 0
By the third inequality of (8.118),

T s+h T
Φε (vm (s)) dtds ≤ h Φε (vm (s)) ds ≤ c1 h. (8.124)
0 s 0
By (8.118) and (8.122), we have

T s+h T s+h

A2 (θm )vm (t), vm (s) dtds ≤ c vm (s)V vm (t)V dt ds
0 s 0 s
T √ √
≤c vm (s)V ( hvm L 2 (0,T ;V) ) ds ≤ c2 h.
0
(8.125)
1 1
Since wL3 ≤ K wL2 2 wL2 6 (see Theorem 1.12), we have
|B2 (v, w), z| = |rot v × w, z| ≤ K rot vL2 wL3 zL6
1 1 (8.126)
≤ K vV w 2 wV2 zV ,
and so by (8.118) we have

T s+h

B2 (vm (t), vm (t))vm (s) dtds
0 s
T s+h 3 1
≤K vm (t)V2 vm (t) 2 vm (s)V dtds
0 s
T s+h 43 s+h 14
≤K vm (s)V vm (t)2V dt vm (t)2 dt ds
0 s s
1
≤ c3 h 4 .
(8.127)
Also, by (8.118) we have
T s+h T t

( f + f 2 )(t), vm (t) dtds ≤ |( f + f 2 )(t), vm (t)| ds dt ≤ c4 h,
0 s 0 t−h
T s+h T s+h

( f + f 2 )(t), −vm (s) dtds ≤ K vm (s)V ( f + f 2 )(t)V∗ dtds
0 s 0 s
√
≤ c5 h. (8.128)
In the same way, we get

T s+h √

α0 θm f, vm (t) − vm (s) dtds ≤ c6 h + c7 h,
0 s
T s+h (8.129)

k11 vm (t) dtds ≤ c8 h.
2
0 s
Note that constants ci , i = 1, · · · , 8, are independent of m and ε. By virtue of

(8.123)–(8.129), uniformly with respect to m, ε
T
1
vm (s + h) − vm (s)2 ds ≤ O(h 4 ), (8.130)
0
and the set {vm } is relatively compact in L 2 (0, T ; W 10 ,2 (Ω)) (see Theorem 1.38).
9
Also, we have
|(κ(θm )∇θm , ∇ϕ)| ≤ κ1 ∇θm L2 ϕWΓ1,2 (Ω) ,

D
|(β(x)θm , ϕ)Γ R | ≤ cθm WΓ1,2 (Ω) ϕWΓ1,2 (Ω) , (8.131)

D D
|vm θm , ∇ϕ| ≤ cvm V θm L 3 ϕWΓ1,2 (Ω) .

D
By (8.131), we have from the second equation of (8.103)

∂θ
m
, ϕ ≤ c θm WΓ1,2 (Ω) + vm 2V + θm 2W 1,2 (Ω) + g2 (WΓ1,2 )∗ ϕWΓ1,2
∂t D ΓD D D
1,2
∀ϕ ∈ WΓ D (Ω).
Hence, by (8.118) we see that

θm ∈ L 1 0, T ; (WΓ1,τ
D
)∗ ,
(8.132)
θ
m L 1 0,T ;(W )∗
1,τ
≤ c,
ΓD
where c is independent of m and ε. Thus, the set {θε } is relatively compact in

L 2 (0, T ; W 10 ,2 (Ω)) (see Theorem 1.39).
9
8.4.2 Existence of a Solution
Having obtained solutions to (8.103), which are actually a sequence of approximating

solutions to (8.28), we can extract subsequences, which are still denoted as before,
such that as m → ∞, ε → 0,
vm v in L 2 (0, T ; V),
∗
vm v in L ∞ (0, T ; H ),
vm → v in L 2 (0, T ; W 10 ,2 (Ω)),
9
(8.133)
θm θ in L 2 (0, T ; W D1,2 (Ω)),
∗
θm θ in L ∞ (0, T ; L 2 (Ω)),
θm → θ in L 2 (0, T ; W 10 ,2 (Ω)).
9

On the other hand, putting u = M j=1 k j (t)u j , where k j (t) ∈ C [0, T ] and M is
1
a positive integer, let us multiply the first equation of (8.103) by k j (t) and sum for
j = 1, · · · , M. Then, let us multiply the first equation of (8.103) by g jm (t) and sum
for j = 1, · · · , m. Substituting the resulting equations yields
T ∂vm
+ A2 (θm )vm (t) + B2 (vm (t), vm (t)) + ∇Φε (vm (t)), u(t) − vm (t) dt
0 ∂t
T
= −α0 θm f + f + f 2 , u(t) − vm (t) dt.
0
(8.134)
Since
T
vm (t), u(t) − vm (t) dt
0
T
1 1
= u (t), u(t) − vm (t) dt − vm (T ) − u(T )2 + vm (0) − u(0)2 ,
0 2 2
taking into account (8.109), we have from (8.134)

T
u (t) + A2 (θm )vm (t) + B2 (vm (t), vm (t)), u(t) − vm (t) dt
0
T T
− −α0 θm f + f + f 2 , u(t) − vm (t) dt + Φε (u(t)) − Φε (vm (t)) dt
0 0
1
≥ − vm (0) − u(0)2 .
2
(8.135)
Since Φε (u) ≤ Φ(u) and Φ(Jε vm (t)) ≤ Φε (vm (t)) (see (1.37)), we have from
(8.135)
T
u (t) + A2 (θm )vm (t) + B2 (vm (t), vm (t)), u(t) − vm (t) dt
0
T T
− −α0 θm f + f + f 2 , u(t) − vm (t) dt + Ψ (u) − Φ(Jε vm (t)) dt
0 0
1
≥ − vm (0) − u(0)2 .
2
(8.136)
By (8.133) and Corollary 1.1, we get
T T T
A2 (θm )vm (t), u(t) dt ≡ a02 (θm (t); vm (t), u(t)) dt → A2 (θ)v(t), u(t) dt.
0 0 0
(8.137)
Owing to (8.133),
vm → v in L 2 (0, T ; L2 (∂Ω)).
Thus taking a subsequence if necessary, by Lemma 1.3 we deduce
2(μ(θm )k(x)vm , vm )Γ2 + 2(μ(θm )S ṽm , ṽm )Γ3 + 2(α(x)vm , vm )Γ5 + (μ(θm )k(x)vm , vm )Γ7
→ 2(μ(θ)k(x)v, v)Γ2 + 2(μ(θ)S ṽ, ṽ)Γ3 + 2(α(x)v, v)Γ5 + (μ(θ)k(x)v, v)Γ7 .
Therefore, taking into account
lim inf 2(μ(θm )E(vm ), E(vm )) ≥ 2(μ(θ )E(v), E(v))
(see Corollary 1.2), we have
lim infA2 (θm )vm (t), vm (t) ≥ A2 (θ )v(t), v(t). (8.138)
From (8.137) and (8.138), it follows

T T
lim inf A2 (θm )vm (t), u(t) − vm (t) dt ≤ A2 (θ )v(t), u(t) − v(t) dt.
m→∞
ε→0 0 0
(8.139)
By (8.119) and (8.133), we have that Jε (vm ) v in L 2 (0, T ; V) as m → ∞,
ε → 0. Since the functional Φ : V → R is weakly lower semi-continuous, we have
T T
lim inf Φ(Jε vm (t)) dt ≥ Φ(v(t)) dt ≡ Ψ (v). (8.140)
m→∞
ε→0 0 0
It is not difficult to prove that

T T
B2 (vm (t), vm (t)), u(t) dt → B2 (v(t), v(t)), u(t) dt (8.141)
0 0
as m → ∞, ε → 0 (see (6.46)–(6.49)).
Let us prove that
T T
α0 θm f, u(t) − vm (t) dt → α0 θ f, u(t) − v(t) dt → 0 as m → ∞, ε → 0.
0 0
(8.142)
By (8.133), we have
T
(α0 θm − α0 θ ) f, u(t) dt → 0 (8.143)
0
and
T
α0 θm f, vm (t) − α0 θ f, v(t) dt
0
T T
(8.144)
= α0 (θm − θ ) f, vm (t) dt + α0 θ f, vm (t) − v(t) dt → 0.
0 0
By (8.143) and (8.144) we get (8.142). It is obvious that B2 (v(t), v(t)), v(t) = 0.
Therefore, by (8.139)–(8.141) and (8.142), we have from (8.136)
T
u (t) + A2 (θ )v(t) + B2 (v(t), v(t)) − (1 − α0 θ ) f − f 2 , u(t) − v(t) dt
0
1
+ Ψ (u) − Ψ (v) ≥ − v0 − u(0)2 . (8.145)
2
3
Since B2 (v(t), v(t))V∗ ≤ K v(t)V2 v(t) 2 (see (8.126)) and v ∈ L ∞ (0, T ;
1
we see that B2 (v, v) ∈ L 3 (0, T ; V∗ ). Therefore, by density of the set {u =

4
(Ω)),
2
L
M ∗
j=1 k j (t)u j , k j (t) ∈ C [0, T ], M=1, 2, · · · } in {L (0, T ; V) : u ∈L (0, T ; V )}
1 4 2
(see Remark 1.8), the inequality (8.145) is valid for all u ∈ {L 4 (0, T ; V) : u ∈
L 2 (0, T ; V∗ )}.
Thus, the first formula
of (8.28) holds.
Putting ϕ(t) = M j=1 j (t)ϕ j , where k j (t) ∈ C [0, T ], k j (T ) = 0, multiplying
k 1
the second equation of (8.103) by k j (t) and summing up for j = 1, · · · , M, we

obtain
t ∂ϕ t t
θm (t), ϕ(t) − θm , ds + (κ(θm ∇θm , ∇ϕ) ds + (β(x)θm , ϕ)Γ R ds
0 ∂t 0 0
t t
− vm θm , ∇ϕ ds = θm (0), ϕ(0) + g2 , ϕ ds ∀t ∈ [0, T ].
0 0
(8.146)
By Corollary 1.1 we have
t t
(κ(θm )∇θm , ∇ϕ) ds → (κ(θ )∇θ, ∇ϕ) ds as m → ∞, ε → 0. (8.147)
0 0
Since W 10 ,2 (Ω) ⊂ L 4 (Ω) (see Theorem 1.20), we have

9
vm → v in L 2 (0, T ; L4 (Ω)), θm → θ in L 2 (0, T ; L 4 (Ω)). (8.148)
By (8.148), we have
t
|vm θm , ∇ϕ − vθ, ∇ϕ| ds
0
t t
≤ |(vm − v)θm , ∇ϕ| ds + |v(θm − θ), ∇ϕ| ds
0 0

≤ vm − v L 2 (0,T ;L4 ) θm L 2 (0,T ;L 4 ) + v L 2 (0,T ;L4 ) θm − θ L 2 (0,T ;L 4 ) ∇ϕ L ∞ (0,T ;L2 ) → 0,
which implies
t t
vm θm , ∇ϕ ds → vθ, ∇ϕ ds as m → ∞, ε → 0. (8.149)
0 0
Therefore taking into account (8.147) and (8.149), we have from (8.146)
T ∂ϕ T T
− θ, dt + (κ(θ )∇θ, ∇ϕ) dt + (β(x)θ, ϕ)Γ R dt
0 ∂t 0 0
(8.150)
T t
− vθ, ∇ϕ dt = θ (0), ϕ(0) + g2 , ϕ ds.
0 0
M
Since the set {ϕ(t) = j=1 k j (t)ϕ j : k j (t) ∈ C 1 [0, T ], k j (T ) = 0, M = 1, 2, · · · }
is dense in {ϕ ∈ C 1 ([0, T ]; WΓ1,2
D
) : ϕ(T ) = 0} (see Remark 1.8), we deduce from
(8.150) the second equation of (8.28).
Remark 8.6 The Eq. (10.1) of Chap. 10 is more general than (8.1), and from the
result of Chap. 10 a result for (8.1) with the boundary conditions (8.2), (7.4) can be
got, however the result demands that the parameter for buoyancy effect α0 is small
enough in accordance with the data of problem and the solution includes “defect
measure” as in [2].

In [4, 5] existence and uniqueness (for 2-D) of a solution to the problem were
shown under homogeneous Dirichlet boundary condition for velocity and mixture
of non-homogeneous Dirichlet and Neumann boundary conditions for temperature.
In [6] for the problem with non-homogeneous Dirichlet boundary conditions for
velocity and temperature the existence of time periodic solution was shown (see
[7]). In [8–11] problem (8.1) on the time dependent domain was studied under non-
homogeneous Dirichlet boundary condition for velocity and temperature. In [12, 13]
the problem on exterior domains with homogeneous Dirichlet boundary condition
for velocity and non-homogeneous Dirichlet boundary condition for temperature was
discussed. In [14] problem (8.1) was studied under mixture of non-homogeneous
Dirichlet and stress boundary conditions for fluid and mixture of non-homogeneous
Dirichlet, Neumann and Robin boundary conditions for temperature. They proved
existence of a unique local-in-time solution under a compatibility condition at initial
time (see (27) and (31) of [14]). In [15] problem (8.1) in cylindrical pipe with
inflow and outflow was studied under slip boundary conditions for velocity and the
Neumann conditions for temperature. There it was proved the existence of a solution
on the given interval when norms of derivatives in the direction along the cylinder
of the initial velocity, initial temperature and the external force are small enough. In
[16] the existence of a solution to problem (8.1) on the time dependent domain was
obtained under mixture of Dirichlet condition of velocity, total pressure and vorticity
boundary conditions for fluid and mixture of Dirichlet, Neumann and Robin boundary
conditions for temperature.
In [17] on 3-D channel under mixture of homogeneous Dirichlet boundary condi-
tion and “do nothing” condition for fluid and mixture of homogeneous Dirichlet and
Neumann boundary conditions for temperature, local existence and global unique-
ness of strong solutions were studied.
References

Pure Appl. 124(4), 107–125 (1980)
2. J. Naumann, J. Wolf, Existence of weak solutions to the equations of natural convection with
dissipative heating. Advances in Mathematical Fluid Mechanics, pp. 367–384 (Springer, 2010)
3. T. Kim, The non-steady Boussinesq system with mixed boundary conditions including condi-
tions of friction type. Inter. J. Diff. Equat. 2020 Article ID 6096531 (2020)
4. H. Morimoto, On non-stationary Boussinesq equations. Proc. Japan Acad. 67 Ser. A, 159–161
(1991)
5. H. Morimoto, Non-stationary Boussinesq equations. J. Fac. Sci. Univ. Tokyo Sect. IA Math.
39, 61–75 (1992)
6. H. Morimoto, Heat convection equation with nonhomogeneous boundary conditions. Funk-
cialai Ekvacioj 53, 213–229 (2010)
7. H. Morimoto, Survey on time periodic problem for fluid flow under inhomogeneous boundary
condition. Discrete Cont. Dyn. Syst. Ser. S 5, 631–639 (2012)
8. H. Inoue, M. Ôtani, Strong solutions of initial boundary value problems for heat convection
equations in noncylindrical domains. Nonlinear Anal. 24, 1061–1090 (1995)
9. H. Inoue, M. Ôtani, Periodic problems for heat convection equations in noncylindrical domains.
Funkcialaj Ekvacioj 40, 19–39 (1997)
10. K. Oeda, On the initial value problem for the heat convection equation of Boussinesq approx-
imation in a time-dependent domain. Proc. Japan Acad. 64 Ser. A, 143–146 (1988)
11. K. Oeda, Remarks on the periodic solution of the heat convection equation in a perturbed
annulus domain. Proc. Japan Acad. 73 Set. A, 21–25 (1997)
12. K. Oeda, Periodic solutions of the heat convection equations in exterior domains. Proc. Japan
Acad. 73 Set. A, 49–54 (1997)
13. K. Oeda, N. Matsuda, Initial value problems for the heat convection equations in exterior
domains. Tokyo J. Math. 21, 359–375 (1998)
15. P. Kacprzyk, Long-time existence of solutions to the Navier-Stokes equations with inflow-
outflow and heat convection. Math. Meth. Appl. Sci. 35, 1000–1013 (2012)
16. T. Kim, D. Cao, Existence of solutions to the heat convection equations in a time-dependent
domain with mixed boundary conditions. J. Math. Sci. Univ. Tokyo 22, 531–568 (2015)
17. M. Beneš, The “do nothing” problem for Boussinesq fluids. Appl. Math. Lett. 31, 25–28 (2014)
Chapter 9
The Steady Equations for
Heat-Conducting Fluids
In this chapter we are concerned with the equation for steady flow of heat-conducting
incompressible Newtonian fluids with dissipative heating under mixed boundary con-
ditions. The boundary conditions for fluid may include Tresca slip, leak condition,
one-sided leak conditions, velocity, pressure, vorticity, stress together and the condi-
tions for temperature may include Dirichlet, Neumann and Robin conditions together.
On the basis of results of Sect. 3.1, we get variational formulations consisting of a
variational inequality for velocity and a variational equation for temperature, which
are equivalent to the original PDE problems for smooth solutions.
Then, we study the existence of solutions to the variational problems. To this end, we
first study the existence of solutions to auxiliary problems including one parameter for
approximation and two or three parameters concerned with the norms of velocity and
temperature. Then we determine the parameters concerned with the norms of velocity
and temperature in accordance with the data of problems, and we get the existence of
solutions by passing to limits as the parameter for approximation goes to zero.
For the problem of the case of static pressure it is proved that if the body force and
boundary data for fluid are small enough and buoyancy effect and energy dissipation
effect due to expansion are small enough, then there exists a solution. However, for
the problem of the case of total pressure it is proved that if buoyancy effect and energy
dissipation effect due to expansion are small enough, then there exists a solution.
9.1 Problems and Assumptions
The system studied in this chapter is as follows:

⎧
⎨ − 2∇ · μ(θ)E (v) + (v · ∇)v + ∇ p = (1 − α0 θ) f,
⎪
div v = 0, (9.1)
⎪
⎩
− ∇ · (κ(θ)∇θ) + v · ∇(γ (θ)θ) − α2 μ(θ)E (v) : E (v) = α1 θ f · v + g.
https://doi.org/10.1007/978-3-030-78659-5_9
286 9 The Steady Equations for Heat-Conducting Fluids
Here α1 is parameter for dissipation of energy due to expansion, α2 is a nonnega-

tive real number and others are the same as in Chap. 7. The term μ(θ )E(v) : E(v)
represents the dissipation of energy due to viscosity (the Joule effect).
Let Ω be as in Chap. 7. (See Remark 6.1.)
For temperature we consider the following boundary conditions
(1) θ |Γ D = θ D |Γ D , θ D − a given function on Ω,

∂θ
(2) κ(θ ) + β(x)θ Γ R = g R (x), β(x), g R (x) − given functions on Γ R .
∂n
(9.2)
As before, Problems I and II are distinguished according to boundary conditions
for fluid. Problem I is the one with the boundary conditions (7.3) (the case of static
pressure) and Problem II is the one with boundary conditions (7.4) (the case of total
pressure).
Assumption 9.1 We assume the followings.
(1) Γ2 j , Γ3 j and Γ7 j are convex. Γ1 = ∅, Γ D = ∅ and

Γ R ⊂ ∪i=1,3,5,8 Γi . (9.3)

(3) For the functions of (9.1) f ∈ Lt (Ω), t > 3, g ∈ L 6/5 (Ω) and
μ ∈ C(R), 0 < μ0 ≤ μ(ξ ) ≤ μ1 < ∞ ∀ξ ∈ R;

κ ∈ C(R), 0 < κ0 ≤ κ(ξ ) ≤ κ1 < ∞ ∀ξ ∈ R; (9.4)
γ ∈ C(R), |γ (ξ )| ≤ γ0 ∀ξ ∈ R.
(4) For the functions of (9.2), (7.3) and (7.4),
θ D ∈ W 1,2 (Ω), ∇θ D ∈ L ∞ (Ω), θ D ≥ 0, g R ∈ L 4/3 (Γ R );

(9.5)
φi ∈ H − 2 (Γi ), i = 2, 4, 7, φi ∈ H− 2 (Γi ), i = 3, 5, 6;
1 1
the matrix α is positive, α jk ∈ L ∞ (Γ5 ).
Remark 9.1 For (2) of Assumption 9.1 we refer to Remark 7.1.
9.2 Variational Formulations for Problems
In this section for every problem above we first obtain a variational formulation which
consists of six formulas with six unknown functions. Then, for every problem we get
another variational formulation equivalent to the variational formulation obtained,
which consists of one variational inequality for velocity and a variational equation
for temperature.
1, p
WΓ D (Ω) = {y ∈ W 1, p (Ω) : y|Γ D = 0}.
We will use inner products in (7.8) for V and WΓ1,2

D
(Ω).
By (7.9)–(7.12), we can see that smooth solutions (v, p, θ ) of problem (9.1), (9.2),
(7.3) satisfy the following (cf. Remark 3.9).
⎧
⎪ 2(μ(θ)E(v), E(u)) + (v · ∇)v, u + 2(μ(θ)k(x)v, u)Γ2
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ − 2(μ(θ)εnτ (v), u)Γ8 + ( p − 2μ(θ )εnn (v), u n )Γ9 ∪Γ10 ∪Γ11
⎪
⎪
⎪
⎪
⎪
⎪ = (1 − α0 θ ) f, u + φi , u n Γi + φi , u Γi ∀u ∈ V,
⎪
⎪
⎪
⎪
⎪
⎪
i=2,4,7 i=3,5,6
⎪
⎪ (κ(θ)∇θ, ∇ϕ) − (γ (θ)θv, ∇ϕ) − (α μ(θ )|E(v)|2 , ϕ) + (βθ, ϕ) − (α θ f · v, ϕ)
⎨ 2 ΓR 1
⎪
⎪ = g R , ϕ Γ R + g, ϕ ∀ϕ ∈ WΓ1,∞ (Ω),
⎪
⎪
⎪
⎪
D
⎪
⎪ |σ τ (θ, v)| ≤ gτ , σ τ (θ, v) · vτ + gτ |vτ | = 0 on Γ8 ,
⎪
⎪
⎪
⎪
⎪
⎪ |σ n (θ, v, p)| ≤ g n , σn (θ, v, p)vn + g n n | = 0 on Γ9 ,
|v
⎪
⎪
⎪
⎪ σn (θ, v, p) + g+n ≥ 0, (σn (θ, v, p) + g+n )vn = 0 on Γ10 ,
⎪
⎪
⎪
⎪
⎪ σ
⎪ n (θ, v, p) − g−n ≤ 0, (σn (θ, v, p) − g−n )vn = 0 on Γ11 ,
⎪
⎩
θ |Γ D = θ D |Γ D on Γ D .
(9.6)

+ 2(α(x)w, u)Γ5 + (μ(θ )k(x)w, u)Γ7 ∀w, u ∈ V, ∀θ ∈ W 1,2 (Ω),
a1 (v, w, u) = (v · ∇)w, u ∀v, w, u ∈ V,
f1 , u = φi , u n Γi + φi , u Γi ∀u ∈ V.
i=2,4,7 i=3,5,6
(9.7)
Define b0 (θ̃; ·, ·) and f 2 ∈ (WΓ1,2
D
(Ω))∗ by
b0 (θ̃; θ, ϕ) = (κ(θ̃)∇θ, ∇ϕ) + (β(x)θ, ϕ)Γ R ∀θ̃ , θ ∈ W 1,2 (Ω), ϕ ∈ WΓ1,2

D
(Ω),
f2 , ϕ = g R , ϕ ΓR + g, ϕ ∀ϕ ∈ WΓ1,2
D
(Ω).
(9.8)
στ (θ, v) = 2μ(θ )εnτ (v), σn (θ, v, p) = − p + 2μ(θ )εnn (v)
and (9.6), we introduce the following variational formulation for problem (9.1), (9.2),
(7.3).
Problem I-VE. Find (v, θ, στ , σn , σ+n , σ−n ) ∈ K (Ω) × 1,r (Ω) × L2 (Γ )

1≤r < 32 W τ 8
× L 2 (Γ9 ) × H −1/2 (Γ10 ) × H −1/2 (Γ11 ) such that θ |Γ D = θ D |Γ D and
⎧
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ b0 (θ ; θ, ϕ) − γ (θ )θ v, ∇ϕ − α2 μ(θ )|E(v)|2 , ϕ − α1 θ f · v, ϕ = f 2 , ϕ
⎪
⎪
⎨ ∀ϕ ∈ W 1,∞ (Ω),ΓD
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ σ+n + g+n ≥ 0, σ+n + g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎪
⎩
σ−n − g−n ≤ 0, σ−n − g−n , vn Γ11 = 0 on Γ11 ,
(9.9)
where L2τ (Γ8 ) is the subspace of L2 (Γ8 ) consisting of functions such that (u, n)L2 (Γ8 ) =
0.
Remark 9.2 As in Definition 7.1, except that the third equality of (9.1) holds in
L 1 (Ω), define a solution to the problem (9.1), (9.2), (7.3). Then, as in Theorem
7.1, Problem I-VE is equivalent to the problem (9.1), (9.2), (7.3), which shows that
Problem I-VE is well-defined.

and a variational equation, which is equivalent to Problem I-VE.
Let (v, θ, στ , σn , σ+n , σ−n ) be a solution of Problem I-VE. Subtracting the first
formula of (9.9) with u = v from the first formula of (9.9), we get
a0 (θ ; v, u − v) + a1 (v, v, u − v) − (στ , u τ − vτ )Γ8 − (σn , u n − vn )Γ9 − σ+n , u n − vn Γ10

− σ−n , u n − vn Γ11 − f − α0 θ f, u − v = f 1 , u − v ∀u ∈ V.
(9.10)
Let Φ : V → R be the functional defined by (5.22). Then, Φ is proper, convex

lower weak semi-continuous and nonnegative.
By Theorem 5.2, under Assumption 9.1 for a fixed θ , the problem
⎧
⎪
⎪
⎪
⎪ − σ+n , u n Γ10 − σ−n , u n Γ11 − f − α0 θ f, u = f 1 , u ,
⎪
⎪
⎪
⎨ |σ | ≤ g , σ · v + g |v | = 0 on Γ ,
τ τ τ τ τ τ 8
(9.11)
⎪
⎪ |σ | ≤ g , σ v + g |v | = 0 on Γ ,
⎪
⎪
n n n n n n 9
⎪
⎪ σ + ≥ σ + , v = 0 on Γ10 ,
⎪
⎪ +n g+n 0, +n g +n n Γ
⎩
10
σ−n − g−n ≤ 0, σ−n − g−n , vn Γ11 = 0 on Γ11
is equivalent to the following variational inequality.

Find v ∈ V such that
a0 (θ ; v, u − v) + a1 (v, v, u − v) + Φ(u) − Φ(v) − f − α0 θ f, u − v

(9.12)
≥ f1, u − v ∀u ∈ V.
Therefore, we have the following variational formulation equivalent to Problem I-

VE, which consists of a variational inequality for velocity and a variational equation
for temperature.
Problem I-VI. Find (v, θ ) ∈ V × 1≤r < 23 W 1,r (Ω) such that θ |Γ D = θ D |Γ D and
⎧
⎪ a0 (θ ; v, u − v) + a1 (v, v, u − v) + Φ(u) − Φ(v) − f − α0 θ f, u − v ≥ f 1 , u − v
⎪
⎪
⎪
⎪
⎨ ∀u ∈ V,
(9.13)
⎪
⎪ b0 (θ ; θ, ϕ) − γ (θ )θ v, ∇ϕ − α2 μ(θ )|E(u)|2 , ϕ − α1 θ f · v, ϕ = f 2 , ϕ
⎪
⎪
⎪
⎩ ∀ϕ ∈ WΓ1,∞ (Ω).
D
Since (v · ∇)v = rot v × v + 21 grad|v|2 , by (7.9)–(7.12) we can verify that smooth

solutions (v, p, θ ) of problem (9.1), (9.2), (7.4) satisfy the following.
⎧
⎪
⎪ 2(μ(θ)E(v), E(u)) + rotv × v, u + 2(μ(θ)k(x)v, u)Γ2
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪ − 2(μ(θ)εnτ (v), u)Γ8 + ( p + |v|2 − 2μ(θ )εnn (v), u n )Γ9 ∪Γ10 ∪Γ11
⎪
⎪ 2
⎪
⎪
⎪
⎪ = (1 − α θ ) + φ ,
⎪
⎪ 0 f, u i u n Γi + φi , u Γi ∀u ∈ V,
⎪
⎪
⎪
⎪ i=2,4,7 i=3,5,6
⎪
⎪
⎨ (κ(θ)∇θ, ∇ϕ) − (γ (θ)θv, ∇ϕ) − (α μ(θ )|E(v)|2 , ϕ) + (βθ, ϕ) − (α θ f · v, ϕ)
2 ΓR 1
⎪
⎪ = g R , ϕ Γ R + g, ϕ ∀ϕ ∈ WΓ1,∞ (Ω),
⎪
⎪
⎪
⎪
D
⎪
⎪ t (θ, v)| ≤ g , σ t (θ, v) · v + g |v | = 0 on Γ ,
⎪
⎪ |σ τ τ τ τ τ τ 8
⎪
⎪
⎪
⎪ t t
|σn (θ, v, p)| ≤ gn , σn (θ, v, p)vn + gn |vn | = 0 on Γ9 ,
⎪
⎪
⎪
⎪
⎪ σ t (θ, v, p) + g+n ≥ 0, (σ t (θ, v, p) + g+n )vn = 0 on Γ10 ,
⎪
⎪
⎪ n n
⎪
⎪
⎪ σnt (θ, v, p) − g−n ≤ 0, (σnt (θ, v, p) − g−n )vn = 0 on Γ11 ,
⎪
⎪
⎪
⎩
θ | Γ D = θ D |Γ D .
(9.14)
Let us define a2 (·, ·, ·) by
a2 (v, u, w) = rot v × u, w ∀v, u, w ∈ V. (9.15)
Then, taking into account (9.14) and
1
στt (θ, v) = 2μ(θ )εnτ (v), σnt (θ, v, p) = −( p + |v|2 ) + 2μ(θ )εnn (v),
2
we get the following variational formulation for problem (9.1), (9.2), (7.4).
Problem II-VE. Find (v, θ, στt , σnt , σ+n

t
, σ−n
t
) ∈ K (Ω) × 1≤r < 23 W 1,r (Ω)
× L2τ (Γ8 ) × L 2 (Γ9 ) × H −1/2 (Γ10 ) × H −1/2 (Γ11 ) such that θ |Γ D = θ D |Γ D and
⎧
⎪ a0 (θ ; v, u) + a2 (v, v, u) − (στt , u τ )Γ8 − (σnt , u n )Γ9 − σ+n
t
, un
⎪
⎪
Γ10
⎪
⎪
⎪
⎪ − σ−n t
, u n Γ11 − f − α0 θ f, u = f 1 , u ∀u ∈ V,
⎪
⎪
⎪
⎪
⎪
⎪ b0 (θ ; θ, ϕ) − γ (θ )θ v, ∇ϕ − α2 μ(θ )|E(v)|2 , ϕ − α1 θ f · v, ϕ
⎪
⎪
⎨ = f 2 , ϕ ∀ϕ ∈ WΓ1,∞ (Ω),
D
(9.16)
⎪
t t
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ σ+n
t
+ g+n ≥ 0, σ+n t
+ g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎪
⎪
⎩ t
σ−n − g−n ≤ 0, σ−n t
− g−n , vn Γ11 = 0 on Γ11 .
Remark 9.3 Similarly to Remark 9.2, Problem II-VE is equivalent to problem (9.1),
(9.2), (7.4).
Then, using Theorem 5.4, in the same way as Problem I-VI we get Problem II-VI
equivalent to Problem II-VE which consists of a variational inequality for velocity
and a variational equation for temperature.
Problem II-VI. Find (v, θ ) ∈ V × 1≤r < 3 W
1,r
(Ω) such that θ |Γ D = θ D |Γ D and
2
⎧
⎪ a0 (θ ; v, u − v) + a2 (v, v, u − v) + Φ(u) − Φ(v) − f − α0 θ f, u − v
⎪
⎪
⎪
⎨ ≥ f 1 , u − v ∀u ∈ V,
(9.17)
⎪
⎪ b0 (θ ; θ, ϕ) − γ (θ )θ v, ∇ϕ − α2 μ(θ )|E(v)|2 , ϕ − α1 θ f · v, ϕ
⎪
⎪
⎩
= f 2 , ϕ ∀ϕ ∈ WΓ1,∞ D
(Ω),
where a2 (·, ·, ·) is the one defined in (9.15) and Φ is defined by (5.22).

Static Pressure
In this section we prove the main result for the case of static pressure.
Theorem 9.1 Let Assumption 9.1 hold and assume that

(1) f, φi , i = 2, · · · , 7, are small enough in the spaces in (3), (4) of Assumption
9.1 (see (9.50)),
(2) max{|α0 |, |α1 |} is small enough in accordance with f, φi (i = 2, · · · , 7),
g, g R , θ D (see (9.97)).
Then, there exists a solution (v, θ ) to Problem I-VI such that
μ0
vV ≤ ,
K
μ0
θ − W 1,2 (Ω) ≤ c f Lt + g R L 4/3 (Γ R ) + g L 6/5 (Ω) ,
K

|∇θ |r d x ≤ cK σr/(1−σ ) 1 + θ D 2W 1,2 ∀r (1 < r < 3/2),
Ω
where θ − (x) = min{θ (x), 0}, K is the constant in (9.18) below, K σ is the one in
(9.90) and σ = 3−2r
3−r
.
9.3.1 Existence of a Solution to an Auxiliary Problem
Since
|a1 (v, v, u)| = | (v · ∇)v, u | ≤ K v2V uV ∀v, u ∈ V, (9.18)
define a 1 (v) ∈ V∗ by a 1 (v), u = a1 (v, v, u) ∀v, u ∈ V.

Define γε (t) by
γ (t)t
γε (t) := t ∈ R, ε > 0.
(1 + ε|γ (t)|)(1 + ε|t|)
Then,
1
|γε (t)| ≤ , |γε (t)| ≤ |γ (t)||t| ≤ γ0 |t|, γε (t) → γ (t)t as ε → 0. (9.19)
ε2
Fréchet derivative of Φε .
Similarly to Sect. 7.2.1, we first consider an auxiliary problem involving two
parameters δ, ζ concerned with the norm of velocity (which is useful when there is
fluid flux across a portion of boundary), one parameter λ concerned with the norm
of temperature (which is useful to deal with buoyancy effect and energy dissipation
effect due to expansion) and a parameter ε for approximation.
Problem I-VIA. Let δ > 0, ζ > 0, λ > 0, ε > 0 and q ∈ ( 12

5
, 6). Find (v, θ ) ∈ V ×
1,2
W (Ω) such that η = θ − θ D ∈ WΓ D (Ω) and
1,2
⎧ δ
⎪
⎪ a0 (θ ; v, u) + a1 (v, v, u) + ∇Φε (v), u
⎪
⎪ max{δ, a 1 (v)V∗ }
⎪
⎪
⎪
⎪ α0 λ
⎪
⎪ − − θ , = f 1 , u ∀u ∈ V,
⎪
⎨ 1 f ε u
max{λ, θ 2L q }
⎪ ζ |E(v)|2 (9.20)
⎪
⎪ (θ ; θ, ϕ) − γ (θ )v, ∇ϕ − α μ(θ ) , ϕ
⎪ 0
⎪ b ε 2
⎪
⎪ max{ζ, vV } 1 + ε|E(v)|2
⎪
⎪
⎪
⎪ α1 λ
⎩ − θ f ε · v, ϕ = f 2 , ϕ ∀ϕ ∈ WΓ1,2 (Ω),
max{λ, θ 2L q } D
where f ε ∈ L∞ (Ω) is such that f − f ε Lt ≤ ε.
Theorem 9.2 There exists a solution (vε , θε ) ∈ V × W 1,2 (Ω) to Problem I-VIA.
Proof Let r be such that

2 1
+ = 1. (9.21)
q r
Since q ∈ ( 12
5
, 6), we see that 3
2
< r < 6. Let H = V × WΓ1,2
D
(Ω). Define an oper-
ator A : H → H ∗ by
δ
A (v, η), (u, φ) = a0 (η + θ D ; v, u) + a1 (v, v, u) + ∇Φε (v), u
max{δ, a 1 (v)V∗ }
α0 λ
− 1− 2
(η + θ D ) f ε , u + b0 (η + θ D ; η + θ D , φ)
max{λ, η + θ D L q }
(9.22)
ζ |E(v)|2
− γε (η + θ D )v, ∇φ − α2 μ(η + θ D ) ,φ
max{ζ, vV } 1 + ε|E(v)|2
α1 λ
− 2
(η + θ D ) f ε · v, φ ∀(v, η), (u, φ) ∈ H .
Let us show that the operator A is well-defined. By definition of Φε it follows that

Φε (0V ) = 0 and ∇Φε (0V ) = 0. Since ∇Φε is Lipschitz continuous with the constant
ε−1 (see Remark 1.16),

∇Φε (v), u = ∇Φε (v) − ∇Φε (0V ), u ≤ ε−1 vV uV . (9.23)
By (9.18) we have
δ

a1 (v, v, u) ≤ δuV . (9.24)
max{δ, a 1 (v)V∗ }
On the other hand, it is easily checked that by (9.19)

ζ cζ

γε (η + θ D )v, ∇φ ≤ 2 φW 1,2 . (9.25)
max{ζ, vV } ε
The trivial inequality ab ≤ max{a 2 , b2 } implies that

√
λ
η + θ D L q ≤ 1. (9.26)
max{λ, η + θ D 2L q }
Taking into account (9.21) and (9.26), we have

α0 λ α0 λ
(η + θ D ) f ε , u ≤ η + θ D L q f ε Lq uLr
max{λ, η + θ D 2L q } max{λ, η + θ D 2L q }
√
≤ cα0 λ f ε Lq uLr ,
α1 λ

(η + θ D ) f ε · v, φ
α1 λ √
≤ η + θ D L q f ε L∞ vLr φ L q ≤ cα1 λ f ε L∞ vLr φ L q .
(9.27)
Estimation of other terms is easy, and so operator A is well defined.

Then, the existence of a solution to Problem I-VIA is equivalent to the one of a
solution to

f1
A (v, η) = F , F = .
f2
By applying Theorem 1.45, we will prove the existence of a solution to the equation
above. To this end, we verify that A satisfies assumptions of Theorem 1.45.
(i) Let us first prove that A is coercive, i.e.,
1
A (v, η), (v, η) → ∞ as (v, η)H → ∞.
(v, η)H
Since Γ2 j , Γ3 j , Γ7 j are convex and the matrix α is positive, it follows from (9.7) (see
Lemma 3.1) that
a0 (η + θ D ; v, v) ≥ 2μ0 v2V . (9.28)
Taking into account (9.24), (9.25), (9.28) and the first formula of (9.27), we have
δ
A (v, η), (v, η) = a0 (η + θ D ; v, v) + a1 (v, v, v)
max{δ, a 1 (v)V∗ }
α0 λ
+ ∇Φε (v), v − 1 − 2
(η + θ D ) f ε , v
ζ
+ b0 (η + θ D ; η, η) + b0 (η + θ D ; θ D , η) − γε (η + θ D )v, ∇η
max{ζ, vV }
|E(v)|2 α1 λ
− α2 μ(η + θ D ) ,η − (η + θ D ) f ε · v, η
1 + ε|E(v)|2 max{λ, η + θ D 2L q }
3 1
≥ min{2μ0 , κ0 } v2V + η2 1,2 − δv − c f ε 2Lq + θ D 2 2 + 2 + ∇Φε (v), v
4 WΓ
D
L ε
cζ α1 λ
− 2 η 1,2 − (η + θ D ) f ε · v, η ∀(v, η) ∈ H .
ε W ΓD
2
(9.29)
Since the operator ∇Φε is monotone and ∇Φε (0V ) = 0, we have

∇Φε (v), v = ∇Φε (v) − ∇Φε (0V ), v − 0V ≥ 0. (9.30)
Since (η + θ D ) f · v, η = (η + θ D ) f · v, (η + θ D ) − (η + θ D ) f · v, θ D , by (9.21)
and (9.26) we have
α1 λ

(η + θ D ) f ε · v, η
λη + θ D 2L q λη + θ D L q (9.31)
≤c f ε L∞ vLr + c f ε L∞ vLr θ D L q
max{λ, η + θ D 2L q } max{λ, η + θ D 2L q }
√
≤ cλ f ε L∞ vLr + c λ f ε L∞ vLr θ D L q .
By virtue of (9.29)–(9.31), we conclude that

1 cζ
A (v, η), (v, η) ≥ min{2μ0 , κ0 } v2V + η2 1,2 − δv − 2 η 1,2
4 WΓ
D
ε WΓ
D
√ 1
− cλ f ε L∞ vV − c λ f ε L∞ vV θ D L q − c f ε Lt + θ D 2 2 + 2
2
L ε
∀(v, η) ∈ H ,
which implies coercivity of A .

(ii) Taking into account (9.23)–(9.27), we have from (9.22)

A (v, η)H ∗ = sup A (v, η), (u, φ)
(u,φ)H =1
cζ 1 √
≤ c vV + δ + 2 + vV + f ε L2 + λ f ε Lq + η + θ D 1,2 (9.32)
ε ε WΓ
D
1 √
+ vL2 + + λ f ε L∞ vV ∀(v, η), (u, φ) ∈ H ,
ε
which shows that A maps bounded sets of H into bounded sets of H ∗ .

(iii) Let {(vk , ηk )} be a sequence such that
(vk , ηk ) (v, η) in H ,
lim sup A (vk , ηk ), (vk , ηk ) − (v, η) ≤ 0.
k→∞
By taking a subsequence and denoting with same subindex if necessary, we may

assume
vk → v in Ls (Ω) (1 ≤ s < 6) and a.e. in Ω,

(9.33)
ηk → η in L s (Ω) (1 ≤ s < 6) and a.e. in Ω as k → ∞.
Since
a0 (ηk + θ D ; vk − v, vk − v) = a0 (ηk + θ D ; vk , vk − v) − a0 (ηk + θ D ; v, vk − v),

b0 (ηk + θ D ; ηk − η, ηk − η) = b0 (ηk + θ D ; ηk + θ D , ηk − η)
− b0 (ηk + θ D ; η + θ D , ηk − η),
by (9.22) we have

min{μ0 , κ0 } vk − v2V + ηk − ηWΓ1,2 ≤ A (vk , ηk ), (vk , ηk ) − (v, η)
D
− a0 (ηk + θ D ; v, vk − v) − b0 (ηk + θ D ; η + θ D , ηk − η)
δ
− a1 (vk , vk , vk − v) − ∇Φε (vk ), vk − v
max{δ, a 1 (vk )V } ∗
α0 λ
+ 1− (ηk + θ D ) f ε , vk − v
max{λ, ηk + θ D L q }2
(9.34)
ζ
+ γε (ηk + θ D )vk , ∇(ηk − η)
max{ζ, vk V }
|E(vk )|2
+ α2 μ(θ ) , η k − η
1 + ε|E(vk )|2
α1 λ
+ (η k + θ D ) f ε · vk , η k − η .
max{λ, ηk + θ D 2L q }
Now let us estimate all terms except the first one on the right hand side of (9.34).
By Corollary 1.1 it follows that as k → ∞
a0 (ηk + θ D ; v, vk − v) = 2(μ(ηk + θ D )E(v), E(vk − v))

+ 2(μ(ηk + θ D )k(x)v, vk − v)Γ2 + 2(μ(ηk + θ D )S ṽ, ṽk − ṽ)Γ3
+ 2(α(x)v, vk − v)Γ5 + (μ(ηk + θ D )k(x)v, vk − v)Γ7 → 0,

b0 (ηk + θ D ; η + θ D , ηk − η) = κ(ηk + θ D )∇(η + θ D ), ∇(ηk − η)

+ β(x)(η + θ D ), (ηk − η) Γ → 0.
R
(9.35)
Also, as k → ∞
δ

a1 (vk , vk , vk − v)
(9.36)
δ
≤ vk L4 ∇vk L2 vk − vL4 → 0.
Since ∇Φε is monotone, as k → ∞

− ∇Φε (vk ), vk − v = − ∇Φε (vk ) − ∇Φε (v), vk − v − ∇Φε (v), vk − v

≤ − ∇Φε (v), vk − v → 0.
(9.37)
By (9.26) and (9.33), the followings hold.
α0 λ √

(ηk + θ D ) f ε , vk − v ≤ c λ f ε L∞ vk − vL3 → 0,
α1 λ √

(ηk + θ D ) f ε · vk , ηk − η ≤ c λ f ε L∞ vk L6 ηk − η L r → 0,
ζ
γε (ηk + θ D )vk , ∇(ηk − η)
max{ζ, vk V }

≤ | γε (ηk + θ D ) − γε (η + θ D ) vk , ∇(ηk − η) |
+ | γε (η + θ D )(vk − v), ∇(ηk − η) | + | γε (η + θ D )v, ∇(ηk − η) |
≤ cγε (ηk + θ D ) − γε (η + θ D ) L 3 vk V ηk − η
WΓ1,2
D
+ cγε (η + θ D ) L 6 (vk − v)L3 ηk − η
WΓ1,2
D
+ γε (η + θ D )v, ∇(ηk − η) → 0
(9.38)
as k → ∞, where the fact that by Lemma 1.3 γε (ηk + θ D ) → γε (η + θ D ) in L 3 (Ω)

as k → ∞ was used. It is easy to prove convergence to zero of other terms in the
right hand side of (9.34). Thus, by (9.33)–(9.38) we conclude that

lim sup min{μ0 , κ0 }| vk − v2V + ηk − ηWΓ1,2
k→∞ D
≤ lim sup A (vk , ηk ), (vk , ηk ) − (v, η) ≤ 0,

k→∞
which implies
(vk , ηk ) → (v, η) in H as k → ∞,
(9.39)
vk → v, ηk → η a.e. in Ω as k → ∞.
By the definition of A , for (u, φ) ∈ H
A (vk , ηk ), (vk , ηk ) − (u, φ)

δ
= a0 (ηk + θ D ; vk , vk − u) + a1 (vk , vk , vk − u)
α0 λ
+ ∇Φε (vk ), vk − u − 1 − 2
(ηk + θ D ) f, vk − u
max{λ, ηk + θ D L q } (9.40)
ζ
+ b0 (ηk + θ D ; ηk + θ D , ηk − φ) − γ (ηk + θ D )vk , ∇(ηk − φ)
max{ζ, vk V }
|E(vk )|2 α1 λ
− α2 μ(θk ) , ηk − φ − (η k + θ D ) f ε · vk , ηk − φ .
1 + ε|E(vk )|2 max{λ, ηk + θ D 2L q }
Taking into account (9.39), by Corollary 1.2 we have
lim inf a0 (ηk + θ D ; vk , vk ) ≥ a0 (η + θ D ; v, v),

k→∞
(9.41)
lim inf b0 (ηk + θ D ; ηk , ηk ) ≥ b0 (η + θ D ; η, η).
k→∞
The sequence { max{λ,ηα0 λ+θ 2 } (ηk + θ D )} converges a.e. in Ω to max{λ,η+θ

α0 λ
(η +
D Lq }
2
k D Lq
√
θ D ) and α0 λ
(ηk + θ D ) q ≤ α0 λ, and so this sequence weakly con-
max{λ,ηk +θ D L q }
2 L
verges in L q (see Lemma 1.2). Then by virtue of this fact and the first formula of
(9.38), we have
α0 λ
lim (η k + θ D ) f ε , vk − u
k→∞ max{λ, ηk + θ D 2L q }
α0 λ
= lim (η k + θ D ) f ε , v k − v
k→∞ max{λ, ηk + θ D 2 q }
L
α0 λ (9.42)
+ lim (η k + θ D ) f ε , v − u
k→∞ max{λ, ηk + θ D 2 q }
L
α0 λ
= (η + θ D ) f ε , v − u ,

where the fact that owing to (9.21) f ε · (v − u) ∈ L 3 ⊂ L q , q1 + 1
q
= 1, was used.
Using the second formula in (9.38), in the same way we get
α1 λ
lim (η k + θ D ) f ε · vk , η k − φ
k→∞ max{λ, ηk + θ D 2L q }
α1 λ (9.43)
= (η + θ D ) f ε · v, η − φ .
It is easy to prove convergence of other terms on the right hand side of (9.40). Thus,
by (9.41)–(9.43) we have existence of a subsequence {(vk , ηk )} such that
lim inf A (vk , ηk ), (vk , ηk ) − (u, φ) ≥ A (v, η), (v, η) − (u, φ) .

k→∞
Therefore, we can apply Theorem 1.45 to obtain the conclusion.
9.3.2 A Priori Estimates of Solutions to the Auxiliary

Problem
Let us choose q0 such that
1 1 1 12
+ + ≤ 1, < q0 < 3. (9.44)
q0 t 3 5
Since t > 3 (see (3) of Assumption 9.1), such a choice is possible. Then, q0 satisfies
the condition for q in Problem I-VIA.
√
Lemma 9.1 If v ∈ V and |α1 | λ ≤ 1, then for all θε satisfying the second formula
of (9.20) and the condition θε − θ D ∈ WΓ1,2
D
(Ω) the following estimate holds.
θε− WΓ1,2 (Ω) ≤ c f ε Lt vV + g R L 4/3 (Γ R ) + g L 6/5 (Ω) , (9.45)
D
where constant c is independent of ε.

Proof Since θε |Γ D = θ D |Γ D ≥ 0, a function ϕ(x) = θε− (x) ≡ min{θε (x), 0} is admis-
sible in the second formula of (9.20), and we have
ζ
(κ(θε )∇θε ,∇θε− ) + (β(x)θε , θε− )Γ R − γε (θε )v, ∇θε−
max{ζ, vV }
|E(v)|2 α1 λ (9.46)
− α2 μ(θ ) , θε− − θε f ε · v, θε− = f 2 , θε− .
1 + ε|E(v)|2 2
max{λ, θε L q0 }
Let us prove
γε (θε )v, ∇θε− = 0. (9.47)
To this end, define t

(t) := γε (s) ds, t ∈ R.
0
Then, ∈ C 1 (R) and
∇(θ ) = γε (θ )∇θ, (θ ) ∈ W 1,2 (Ω) ∀θ ∈ W 1,2 (Ω),

(9.48)
(θ )|Γ D = 0 ∀θ ∈ WΓ1,2
D
(Ω).
Since ∇θε− = 0 on Ω + = {x : θ (x) ≥ 0} and v · n|Γ R = 0, by (9.48) we have

γε (θ )v, ∇θε− = γε (θε− )v · ∇θε− dx = v · ∇(θε− ) = 0,
Ω Ω
which means (9.47).

Also,
|E(v)|2
−
− α2 μ(θ ) , θ ≥ 0,
1 + ε|E(v)|2 ε
α1 λ
−
θε f ε · v, θε ≤ c f ε Lt vV θε− WΓ1,2
max{λ, θε 2L q0 } D
κ0 − 2 (9.49)
≤ θε W 1,2 + c f ε 2Lt v2V ,
4 ΓD

f 2 , θ − ≤ κ0 θ − 2 1 + c f 2 2 1,2 ∗ ,
ε
4 ε H (WΓ )
D
−
(β(x)θε , θε )Γ R ≥ 0,
where to get the second inequality (9.26) and (9.44) were used. By (9.46)–(9.49) we
have
2c
θε− 2W 1,2 ≤ f ε 2Lt v2V + f 2 2(W 1,2 )∗ ,
ΓD κ0 ΓD

√
Lemma 9.2 If |α0 | λ ≤ 1 and
μ20
( f Lt + f 1 V∗ ) < , (9.50)
K c0
where K is the constant in (9.18) and c0 is the one in (9.55) below, then there exist
parameters δ and ζ such that
δ ζ
= 1, =1 (9.51)
max{δ, a 1 (vε )V∗ } max{ζ, vε V }
for all small ε and solutions of (9.20) vε . In addition,

μ0
vε V ≤ . (9.52)
K
√
Proof Since |α0 | λ ≤ 1, we have
√
α0 λ c λ
θε f ε , u ≤ θε L q0 f ε Lt uL6 ≤ c f ε Lt uV , (9.53)
max{λ, θε 2L q0 } max{λ, θε 2L q0 }
where c is independent of α0 . Putting u = vε in the first equation of (9.20) yields
δ
a0 (θε ; vε , vε ) + a1 (vε , vε , vε ) + ∇Φε (vε ), vε
α0 λ (9.54)
− 1− θε f ε , vε = f 1 , vε .
max{λ, θε 2L q0 }
There exists a constant c0 such that
f ε L 65 + c f ε Lt ≤ c0 f ε Lt ,
where c is the one in (9.53). Thus taking into account (9.18), (9.28), (9.30) and (9.53),
we have from (9.54)
2μ0 vε 2V ≤ a0 (θε ; vε , vε )

δ
≤ |a1 (vε , vε , vε )|
α0 λ

+ 1− θε f ε , vε + | f 1 , vε |
max{λ, θε L q0 }
2

≤ K vε 3V + c0 f ε Lt + f 1 V∗ vε V
≤ K vε 3V + c0 ( f ε Lt + f 1 V∗ )vε V ,
where
c0 = max{c0 , 1}. (9.55)
Note that the estimate above is independent of δ. This implies
0 ≤ K vε 2V − 2μ0 vε V + c0 ( f ε Lt + f 1 V∗ ).
Let us consider a quadratic polynomial for x > 0 concerned with the inequality above
K x 2 − 2μ0 x + a.
μ0
If 0 ≤ K a ≤ μ20 , then there exists a nonnegative minimum root x1 (≤ K
) and a
maximum root x2 . Thus, we can know that if
μ20
( f ε Lt + f 1 V∗ ) ≤ ,
K c0
then μ0
vε V ≤ or vε V ≥ x2 . (9.56)
K
On the other hand, we have from (9.54) another estimation under consideration
of δ
2μ0 vε 2V ≤ a0 (θε ; vε , vε ) ≤ δvε V + c0 ( f ε Lt + f 1 V∗ )vε V ,
which implies
1
vε V ≤ δ + c0 ( f ε Lt + f 1 V∗ ) . (9.57)
2μ0
μ0 2 μ20
In view of (9.56), we can take δ = K K
= K
.
μ20
If ( f Lt + f 1 V∗ ) < K c0
, then for all ε small enough
μ20
( f ε Lt + f 1 V∗ ) ≤ .
K c0
Thus, without loss of generality we have from (9.57)
δ 1 μ20 μ0
vε V ≤ + = . (9.58)
2μ0 2μ0 K K
μ2
By (9.58) under the condition (9.50) a 1 (vε )V∗ ≤ K vε 2V ≤ K0 (see (9.18)), and
so we get the first one in (9.51). Taking ζ = μK0 , we get the second one in (9.51).
√
Lemma 9.3 If max{|α0 |, |α1 |} λ ≤ 1 and (9.50) holds, then under the parameter
ζ by Lemma 9.2 there exists a λ1 independent of ε such that

θε L q0 ≤ λ1 . (9.59)
Moreover, if 1 < r < 23 , then

3 − 2r
|∇θε |r d x ≤ cK σr/(1−σ ) 1 + θ D 2W 1,2 , σ = , (9.60)
Ω 3−r
where K σ is the constant in (9.90) below.
Proof Using Lemmas 9.1 and 9.2, we will arrive at the conclusion. For simplicity
of notation, from now on in this lemma we denote vε , θε by v, θ. Set
d0 := θ D L ∞ (Ω) ≥ θ D L ∞ (∂Ω) . (9.61)
Then, since (θ − d0 )+ ≡ max{0, θ − d0 } = 0 on Γ D , the function
1
ϕ =1− , σ >0 (9.62)
(1 + (θ − d0 )+ )σ
belongs to WΓ1,2
D
(Ω) and 0 ≤ ϕ(x) ≤ 1 a.e. in Ω.
Taking ϕ of (9.62) and ζ satisfying (9.51), we have from the second equation of
(9.20)

|∇(θ − d0 )+ |2 ∇(θ − d0 )+
σ κ(θ ) d x + (βθ, ϕ)Γ − σ γε (θ )v · dx
Ω (1 + (θ − d0 )+ )1+σ R
Ω (1 + (θ − d0 )+ )1+σ

|E(v)|2 1
= α2 μ(θ ) 1− dx
Ω 1 + ε|E(v)| 2 (1 + (θ − d0 )+ )σ

α1 λ 1 1
+ θ fε · v 1 − + σ d x + f2 , 1 − + σ
2
max{λ, θ q } Ω (1 + (θ − d0 ) ) (1 + (θ − d0 ) )
L 0
≡ I1 + I2 + I3 .
(9.63)
Note that ϕ(x) = 0 at x such that θ (x) ≤ d0 , and (β(x)θ, ϕ)Γ R ≥ 0. Let us show the
third term on the left hand side of (9.63) vanishes. To this end, define
t
γε (s + d0 )
σ (t) := ds, t ≥ 0.
0 (1 + s)1+σ
Then, σ ∈ C 1 (R) and
σ ((θ − d0 )+ )|Γ D = 0,
∇(θ − d0 )+ (9.64)
∇σ ((θ − d0 )+ ) = γε (θ ) ,
(1 + (θ − d0 )+ )1+σ
where the fact that if θ (x) − d0 ≥ 0, then (θ (x) − d0 )+ + d0 = θ (x) was used. Tak-
ing into account the fact that vn = 0 on Γ R (see (9.3)) and the first equality of (9.64),
we have

∇(θ − d0 )+
γε (θ )v · + )1+σ
d x = v · ∇σ ((θ − d0 )+ ) = 0. (9.65)
Ω (1 + (θ − d0 ) Ω
It is easily seen that

|I1 | ≤ cv2V . (9.66)
√
Since max{|α0 |, |α1 |} λ ≤ 1, by (9.26) we have
√
λ
|I2 | ≤ θ q0 f ε Lt vL3 ≤ c f ε Lt vV . (9.67)
max{λ, θ 2L q0 }
Also,
√
|I3 | ≤ c mesΩ f 2 (WΓ1,2 )∗ ≤ c g R L 4/3 (Γ R ) + g L 6/5 (Ω) . (9.68)
D
By (9.65)–(9.68), we have from (9.63)

|∇θ |2
σ + 1+σ
d x ≤ c v2V + f ε Lt vV + g R L 4/3 (Γ ) + g L 6/5 (Ω) .
{x;θ≥d0 } (1 + (θ − d0 ) ) R
(9.69)
Next, taking ϕ = min{θ − θ D , d0 } admissible in the second formula of (9.20),

we have

κ0 |∇θ |2 d x
{x;θ−θ D <d0 }

≤ κ(x)∇θ · ∇θ D d x − β(x)θ (x) min{θ (x) − θ D (x), d0 } d x
{x;θ−θ D <d0 } ΓR

|E(v)|2 (9.70)
+ γε (θ )v · ∇(min{θ − θ D , d0 }) d x + α2 μ(θ ) min{θ − θ D , d0 } d x
Ω Ω 1 + ε|E(v)|2

α1 λ
+ θ f ε · v min{θ − θ D , d0 } d x + f 2 , min{θ − θ D , d0 }
max{λ, θ 2L q0 } Ω
≡ I 1 + I 2 + I 3 + I 4 + I 5 + I 6.
By Young’s inequality,

κ0 κ12
|I 1 | ≤ |∇θ |2 d x + |∇θ D |2 d x. (9.71)
4 {x;θ−θ D <d0 } κ0 Ω
Let us estimate I 2 . If θ (x) − θ D (x) ≥ d0 , then ϕ(x) = d0 and θ (x) ≥ θ D (x) +

d0 ≥ 0. Hence,

β(x)θ (x) min{θ (x) − θ D (x), d0 } d x ≥ 0. (9.72)
Γ R ∩{x;θ(x)−θ D (x)≥d0 }
On the other hand, if θ (x) − θ D (x) ≤ d0 , then ϕ(x) = θ (x) − θ D (x) and θ (x) ≤
θ D (x) + d0 . Thus

β(x)θ (x) min{θ (x) − θ D (x), d0 } d x
Γ R ∩{x;θ(x)−θ D (x)≤d0 }

= β(x)θ (x)θ (x) d x − β(x)θ (x)θ D (x) d x
Γ R ∩{x;θ(x)≤θ D (x)+d0 } Γ R ∩{x;θ(x)≤θ D (x)+d0 }

≥− β(x)θ (x)θ D (x) d x − β(x)θ (x)θ D (x) d x
Γ R ∩{x;0≤θ(x)≤θ D (x)+d0 } Γ R ∩{x;θ(x)≤0}

≥ −β0 2d02 d x − β0 d0 |θ − (x)| d x
ΓR ΓR

≥ −c 1 + θ − (x)WΓ1,2 (Ω) .
D
(9.73)
By (9.72) and (9.73), we get

I 2 ≤ c 1 + θ − (x)WΓ1,2 (Ω) . (9.74)
D
Taking into account (9.19), we have


I3 = γε (θ )v · ∇θ d x − γε (θ )v · ∇θ D d x
{x;θ−θ D ≤d0 } {x;θ−θ D ≤d0 }

κ0
≤ |∇θ |2 d x + cγ02 |θ |2 |v|2 d x + cγ02 |θ |2 |v|2 d x
4 {x;θ−θ D ≤d0 } {x;0≤θ≤d0 +θ D } {x;θ≤0}

+ γ0 |θ ||v · ∇θ D | d x + γ0 |θ ||v · ∇θ D | d x.
{x;0≤θ≤d0 +θ D } {x;θ≤0}
(9.75)
Let us estimate all terms except the first one on the right hand side of (9.75). Since
|θ | ≤ 2d0 on {x; 0 ≤ θ ≤ d0 + θ D },

θ 2 |v|2 d x ≤ 4d02 |v|2 d x ≤ cv2V . (9.76)
{x;0≤θ≤d0 +θ D } Ω
And, by Hölder’s inequality we obtain

1/2
1/2
θ 2 |v|2 d x ≤ θ4 dx |v|4 d x ≤ cθ − 2W 1,2 v2V ,
ΓD
{x;θ≤0} {x;θ≤0} {x;θ≤0}
(9.77)
|θ ||v · ∇θ D | d x ≤ 2d0 vL2 θ D W 1,2 ≤ cvV θ D W 1,2 , (9.78)
{x;0≤θ≤d0 +θ D }

|θ ||v · ∇θ D | d x ≤ cθ − WΓ1,2 vV θ D W 1,2 . (9.79)
{x;θ≤0} D
Combining (9.75)–(9.79) yields

κ0
I3 ≤ |∇θ |2 d x
4 {x;θ−θ D ≤d0 }

+ c v2V + θ − 2W 1,2 v2V + vV θ D W 1,2 + θ − WΓ1,2 vV θ D W 1,2 .
ΓD D
(9.80)
Next, we have

|E(v)|2 |E(v)|2
I4 ≤ α2 μ(θ ) d0 d x + α2 μ(θ ) 2d0 d x
{x;θ−θ D ≥d0 } 1 + ε|E(v)| 2
{x;0≤θ≤d0 +θ D } 1 + ε|E(v)|2

|E(v)|2 |E(v)|2
+ α2 μ(θ ) θ d x + α2 μ(θ ) d0 d x
{x;θ≤0} 1 + ε|E(v)|2
{x;θ≤0} 1 + ε|E(v)|2

≤ α2 μ(θ )|E(v)|2 2d0 d x ≤ cv2V .
Ω
(9.81)
Taking into account (9.44) and applying Hölder’s inequality, we have


|α1 |λ
I5 ≤ |θ f ε · v|d0 d x
max{λ, θ 2L q0 } {x;θ−θ D ≥d0 }

|α1 |λ
+ |θ f ε · v|2d0 d x
max{λ, θ 2L q0 } {x;0≤θ≤d0 +θ D }

α1 λ |α1 |λ (9.82)
+ θ f ε · vθ d x + |θ f ε · v|d0 d x
max{λ, θ 2L q0 } {x;θ≤0} max{λ, θ 2L q0 } {x;θ≤0}

|α1 |λ |α1 |λ
≤ 2d0 2
|θ f ε · v| d x + 2
|θ f ε · v||θ − | d x
max{λ, θ L q0 } Ω max{λ, θ L q0 } Ω
≤ c f ε Lt vV + c f ε Lt vV θ − W 1,2 ≤ c f ε Lt vV (1 + θ − W 1,2 ).
Let us estimate I 6 = (g R , ϕ)Γ R + (g, ϕ), where ϕ = min{θ (x) − θ D (x), d0 }. If

θ (x) − θ D (x) ≥ d0 , then ϕ(x) = d0 and

g R ϕ d x ≤ d0 |g R | d x. (9.83)
Γ R ∩{x;θ(x)−θ D (x)≥d0 } Γ R ∩{x;θ(x)−θ D (x)≥d0 }
If 0 ≤ θ (x) ≤ d0 + θ D (x), then |ϕ(x)| = |θ (x) − θ D (x)| ≤ 2d0 and

g R ϕ(x) d x ≤ 2d0 |g R | d x. (9.84)
Γ R ∩{x;0≤θ(x)≤d0 +θ D (x)} Γ R ∩{x;0≤θ(x)≤d0 +θ D (x)}
And

g R ϕ(x) d x = g R (θ − (x) − d0 ) d x
Γ R ∩{x;θ(x)≤0} Γ R ∩{x;θ(x)≤0}

≤ d0 |g R | d x + g R L 4/3 (Γ R ) θ − WΓ1,2 (Ω) .
Γ R ∩{x;θ(x)≤0} D
(9.85)
Also, we have

g, ϕ ≤ |g|d0 d x + |g|2d0 d x + g(θ − − d0 ) d x
{x;θ−θ D ≥d0 } {x;0≤θ≤d0 +θ D } {x;θ≤0}
≤ 2d0 g L 1 (Ω) + g L 6/5 (Ω) θ − WΓ1,2 (Ω)
D
≤ cg L 6/5 (Ω) (1 + θ − WΓ1,2 (Ω) ).

D
(9.86)
By (9.83)–(9.86), we get
I 6 ≤ c(g R L 4/3 (Γ R ) + g L 6/5 (Ω) )(1 + θ − WΓ1,2 (Ω) ). (9.87)

D
Therefore, by virtue of (9.70), (9.71), (9.74), (9.80)–(9.82) and (9.87) we have


|∇θ |2 d x ≤ c θ D 2W 1,2 + v2V + θ − 2W 1,2 v2V
ΓD
{x;θ−θ D <d0 }

+ c 1 + f ε Lt vV + g R L 4/3 (Γ R ) + g L 6/5 (Ω) 1 + θ − WΓ1,2 .
D
(9.88)
Combining (9.69), (9.88) and taking into account (9.45), we get

|∇θ |2 |∇θ |2
+
dx ≤ +
dx + |∇θ |2 d x
Ω (1 + (θ − d0 ) ) {x;θ≥d0 } (1 + (θ − d0 ) )
1+σ 1+σ
{x;θ≤d0 }
1
≤c 1+ 1 + v2V + f ε Lt vV + g R L 4/3 (Γ ) + g L 6/5 (Ω) 1 + θ − 1,2
σ R WΓ
D
+ c θ D 2 + v2V + θ − 2 v2V
W 1,2 WΓ1,2
D

1 2
≤c 1+ 1 + v2V + f ε Lt vV + g R L 4/3 (Γ ) + g L 6/5 (Ω)
σ R

2
+θ D 2 + v2V + f ε Lt vV + g R L 4/3 (Γ ) + g L 6/5 (Ω) v2V .
W 1,2 R
(9.89)
Since f ε → f in Lt (Ω), we may assume that f ε Lt ≤ 1 + f Lt for all ε. Thus
taking into account (9.58), we have from (9.89) that under (9.50)

|∇θ |2
dx
Ω (1 + (θ − d0 )+ )1+σ

1 μ0 2 μ0 2
≤c 1+ 1+ + (1 + f Lt ) + g R L 4/3 (Γ ) + g L 6/5 (Ω)
σ K K R (9.90)

μ0 2 μ0 2 μ0 2
+θ D 2 1,2 + + (1 + f Lt ) + g R L 4/3 (Γ ) + g L 6/5 (Ω)
W K K R K
≡ Kσ .
To get boundedness of θ ∈ L q0 (Ω) independent of ε, we will use the property

W 1,r0 (Ω) ⊂ L q0 , where r0 is such that r10 − 13 = q10 . Then, 43 < r0 < 23 . To this end,
take σ0 > 0 such that
r0 (1 + σ0 )
= q0 .
2 − r0
Then, σ0 = 3−2r
3−r0
0
, and 0 < σ0 < 15 . Putting σ = σ0 in (9.90), by virtue of Hölder’s
inequality with exponents r20 , 2−r
2
0
and the inequality |a + b| p ≤ 2 p (|a| p + |b| p ),
1 1
|a| + |b| ≤ (|a| p + |b| p ) p , p ∈ (1, ∞), we have

|∇θ |r0 d x
Ω
r0 /2 (2−r0 )/2
|∇θ |2 + (1+σ0 ) 2−r0 d x
r0
≤ d x 1 + (θ − d 0 ) (9.91)
Ω (1 + (θ − d0 )+ )(1+σ0 ) Ω

r /2 (2−r 0 )/2
≤ cK σ00 1+ |θ |3r0 /(3−r0 ) d x ,
Ω
where (1 + σ0 ) 2−r
r0
0
= 3r0 /(3 − r0 ) was used and K σ0 is the one with σ = σ0 in K σ
of (9.90). By Sobolev’s imbedding theorem and Friedrichs’ inequality,
(2−r0 )/2 r0
(1+σ0 )/2
(1+σ0 )
|θ |3r0 /(3−r0 ) d x ≤ cθ W 1,r0 2 ≤ c |∇θ |r0 d x + θ D rW0 1,r0
Ω Ω

(1+σ0 )/2
r (1+σ )/2
≤c |∇θ |r0 d x + θ D W0 1,r0 0 .
Ω
(9.92)
Substituting (9.92) into (9.91) and using Young’s inequality and Hölder’s inequality
with exponents 1−σ2
0
, 1+σ
2
0
on the right hand side, we have
(1+σ0 )/2 r (1+σ )/2
|∇θ |r0 d x ≤ cK σr00 /2 |∇θ |r0 d x + cK σr00 /2 1 + θ D W0 1,r0 0
Ω Ω

1
≤ |∇θ |r0 d x + cK σr00 /(1−σ0 ) + cK σr00 /(1−σ0 ) 1 + θ D rW0 1,r0 .
2 Ω
Thus, we get

|∇θ |r0 d x ≤ cK σr00 /(1−σ0 ) 1 + θ D rW0 1,r0 . (9.93)
Ω
By virtue of (9.93), Sobolev’s imbedding theorem and Friedrichs’ inequality, we

have 1/r0
θ L q0 ≤ cθ W 1,r0 ≤ c |∇θ |r0 d x + θ D rW0 1,r0
Ω
r0 /(1−σ0 )
1/r0
≤ c K σ0 (1 + θ D rW0 1,r0 ) + θ D rW0 1,r0 (9.94)
0)

≤ c K σ1/(1−σ (1 + θ D W 1,r0 ) + θ D W 1,r0
0
0)

≤ c K σ1/(1−σ
0
(1 + θ D W 1,2 ) + θ D W 1,2 ≡ K σ0 .
K σ0 is independent of λ, ε and depends on f Lt , θ D W 1,2 , g R L 4/3 (Γ R ) , g L 6/5 (Ω)

and φi , i = 2, · · · , 7, via f 1 in (9.7).
Putting
λ 1 = K σ0 , (9.95)
we have (9.59).
Now, for all r with 1 < r < 23 and all ε > 0 let us get an estimate of Ω |∇θ |r d x
independent of ε. Putting σ = 3−2r
3−r
, we have that 0 < σ < 21 . Repeating the argument
of (9.91)–(9.93), we have

|∇θ |r d x ≤ cK σr/(1−σ ) 1 + θ D rW 1,r ,
Ω

When λ = λ1 and
max{|α0 |, |α1 |} λ1 ≤ 1,
by (9.94) and (9.95) we have
λ1
= 1 ∀ε > 0. (9.96)
max{λ1 , θε 2L q0 }
Therefore, summarizing Lemmas 9.1–9.3, we have

Theorem 9.3 If
max{|α0 |, |α1 |} λ1 ≤ 1,
μ20 (9.97)
( f 2Lt + f 1 2V∗ ) < ,
K c0
then there exists a solution (vε , θε ) ∈ V × W 1,2 (Ω) to the following problem
⎧
⎪ a0 (θε ; vε , u) + a1 (vε , vε , u) + ∇Φε (vε ), u − (1 − α0 θε ) f ε , u = f 1 , u ∀u ∈ V,
⎪
⎪
⎪
⎨
|E(vε )|2
b0 (θε ; θε , ϕ) − γε (θε )vε , ∇ϕ − α2 μ(θε ) , ϕ − α1 θε f ε · vε , ϕ (9.98)
⎪
⎪ 1 + ε|E(vε )|2
⎪
⎪
⎩
= f 2 , ϕ ∀ϕ ∈ WΓ1,2 (Ω),
D
θε − θ D ∈ WΓ1,2
D
(Ω), (9.99)

μ0
vε V ≤ ,
K
θε− WΓ1,2 (Ω) ≤ c f Lt vε V + g R L 4/3 (Γ R ) + g L 6/5 (Ω) , (9.100)
D
3
|∇θε |r d x ≤ cK σr/(1−σ ) 1 + θ D 2W 1,2 ∀r, 1 < r < ,
Ω 2
where σ = 3−2r
3−r
.
9.3.3 Passing to Limits
By passing to the limit of solutions in Theorem 9.3, we will prove Theorem 9.1.
Owing to (9.100) we can extract subsequences, which are denoted as before, such
that as ε → 0,
vε v in V,
vε → v in Lq , 1 ≤ q < 6, and a.e. in Ω,
3 (9.101)
θε θ in W 1,r (Ω) ∀r, 1 ≤ r < ,
2
θε → θ in L s (Ω) ∀s, 1 ≤ s < 3, and a.e. in Ω.
By (9.99), we have that θ |Γ D = θ D |Γ D .

Subtracting the first formula of (9.98) with u = vε from the first formula of (9.98),
we have

a0 (θε ; vε , u − vε ) + a1 (vε , vε , u − vε ) + ∇Φε (vε ), u − vε − (1 − α0 θε ) f, u − vε
(9.102)
= f 1 , u − vε ∀u ∈ V.
By Corollaries 1.1 and 1.2 we have
a0 (θε ; vε , u) → a0 (θ ; v, u) as ε → 0,
lim inf a0 (θε ; vε , vε ) ≥ a0 (θ ; v, v),
ε→0
which imply that
lim sup a0 (θε ; vε , u − vε ) ≤ a0 (θ ; v, u − v). (9.103)

ε→0
It is easy to prove
a1 (vε , vε , u − vε ) → a1 (v, v, u − v) as ε → 0. (9.104)
Φε (u) − Φε (vε ) ≥ ∇Φε (vε ), u − vε ∀u ∈ V, (9.105)
which together with (1.37) implies
Φε (u) − Φ(Jε vε ) ≥ ∇Φε (vε ), u − vε ∀u ∈ V. (9.106)
Since Φ(0V ) = 0, by (1.37) we get Φε (0V ) = 0, and so we have from (9.105)
Φε (vε ) ≤ ∇Φε (vε ), vε . (9.107)
On the other hand, putting u = vε in the first formula of (9.98), we have

a0 (θε ; vε , vε ) + a1 (vε , vε , vε ) + ∇Φε (vε ), vε = (1 − α0 θε ) f ε , vε + f 1 , vε .
(9.108)

a0 (θε ; vε , vε ) + a1 (vε , vε , vε ) + Φε (vε ) ≤ (1 − α0 θε ) f ε , vε + f 1 , vε ,
from which we get

|Φε (vε )| ≤ c (1 + K σ0 ) f ε Lt + f 1 V∗ vε V + |a1 (vε , vε , vε )|, (9.109)
where (9.94) was used. By virtue of (1.36), (9.18), (9.58) and (9.109), we have
μ0 μ3
vε − Jε vε 2V ≤ c (1 + K σ0 ) f ε Lt + f 1 V∗ + 02 2ε,
K K
which shows that
Jε vε v in V as ε → 0.
Then, by virtue of weak lower semi-continuity of Φ(v),
lim inf Φ(Jε vε ) ≥ Φ(v). (9.110)

ε→0
By (1.37) we have
Φε (u) → Φ(u) as ε → 0. (9.111)
Taking into account (9.110) and (9.111), we have from (9.106)
Φ(u) − Φ(v) ≥ lim sup ∇Φε (vε ), u − vε ∀u ∈ V. (9.112)

ε→0
Using

| θε f ε , vε − θ f, v |

≤ | θε f ε , vε − θ f ε , vε | + | θ f ε , vε − θ f, vε | + | θ f, vε − θ f, v |
≤ θε − θ L 2 f ε L3 vε L6 + θ L 2 f ε − f L3 |vε L6 + θ L 15/7 f L3 |vε − vL5 ,
(9.113)
we can prove

(1 − α0 θε ) f, u − vε → (1 − α0 θ ) f, u − v as ε → 0. (9.114)
It is easy to prove
f 1 , u − vε → f 1 , u − v as ε → 0. (9.115)
By virtue of (9.103), (9.104), (9.112), (9.114) and (9.115), we get from (9.102)

a0 (θ ; v, u − v) + a1 (v, v, u − v) + Φ(u) − Φ(v) − (1 − α0 θ ) f, u − v ≥ f 1 , u − v
∀u ∈ V,
which is the first formula in (9.13). From the above discussion we know that v ∈
K (Ω), i.e. Φ(v) < +∞. Thus putting u = v, we have from (9.112)
0 ≥ lim sup ∇Φε (vε ), v − vε . (9.116)

ε→0
Now we will get the second equation of (9.13). By Corollary 1.1, we have
b0 (θε ; θε , ϕ) → b0 (θ ; θ, ϕ) ∀ϕ ∈ WΓ1,∞
D
(Ω) as ε → 0. (9.117)
Let us prove
γε (θε )vε , ∇φ → γ (θ )θ v, ∇ϕ ∀ϕ ∈ WΓ1,∞

D
(Ω) as ε → 0. (9.118)
By Hölder’s inequality we have
| γε (θε )vε ,∇ϕ − γ (θ )θ v, ∇ϕ |

≤ | γε (θε )vε , ∇ϕ − γ (θ )θ vε , ∇ϕ | + | γ (θ )θ vε , ∇ϕ − γ (θ )θ v, ∇ϕ | (9.119)
≤ γε (θε ) − γ (θ )θ L 6/5 vε L6 ∇ϕL∞ + γ (θ )θ L 4 vε − vL4 ∇ϕL∞ .
By the definition of γε (t) it follows that

γ (θε )θε

γε (θε ) − γ (θ )θ L 6/5 ≤ − γ (θ )θ
(1 + ε|γ (θε )|)(1 + ε|θε |) 6/5
L

≤ γ (θε )θε − γ (θ )θ L 6/5 + ε γ (θ )θ |γ (θε )| + |θε | + ε|γ (θε )||θε | L 6/5 .
(9.120)
Then, by virtue of Lemma 1.3 we know that γ (θε ) converges to γ (θ ) in L p (Ω)
(∀ p, 1 < p < ∞, ) as ε goes to zero. Thus, from (9.119) and (9.120) we get (9.118).
Let us consider
μ0 v − vε 2V ≤ a0 (θε ; v − vε , v − vε )
= a0 (θε ; v, v − vε ) + a1 (vε , vε , v − vε ) + ∇Φε (vε )(v − vε )

− (1 − α0 θε ) f, v − vε − f 1 , v − vε ,
(9.121)
which is obtained from the first formula of (9.98) by putting u = v − vε . By virtue
of (9.103), (9.104), (9.114) and (9.116), the right hand side of (9.121) converges to
0 as ε goes to 0. Thus, we have
vε → v in V as ε → 0. (9.122)
On the other hand,

|E(vε )|2
α2 μ(θε ) − α2 μ(θ )|E(v)|2
1 + ε|E(vε )|2

= α2 μ(θε )
|E(vε )|2
− α μ(θ )|E(v)|2 + α μ(θ ) − μ(θ )|E(v)|2
2 ε 2 ε
1 + ε|E(vε )|2
|E(vε )|2 − |E(v)|2 ε|E(vε )|2 |E(v)|2
= α2 μ(θε ) − α2 μ(θε ) + α2 μ(θε ) − μ(θ ) |E(v)|2
1 + ε|E(vε )|2 1 + ε|E(vε )|2

≡ Iε1 + Iε2 + α2 μ(θε ) − μ(θ ) |E(v)|2 .
By (9.122) we have
Iε1 L 1 (Ω) → 0 as ε → 0.
Since |Iε2 | ≤ c|E(v)|2 and (passing to a subsequence if necessary) Iε2 → 0 a.e. in Ω,

by virtue of Lebesgue’s theorem (see Theorem 1.9) we have
Iε2 L 1 (Ω) → 0 as ε → 0.
By Corollary 1.1, we have
μ(θε )εi j (v) μ(θ )εi j (v) in L 2 (Ω) as ε → 0,
and so

α2 μ(θε ) − μ(θ ) |E(v)|2 , ϕ → 0 ∀ϕ ∈ WΓ1,∞
D
(Ω) as ε → 0.
Thus, we have that as ε → 0

|E(vε )|2
α2 μ(θε ) − α2 μ(θ )|E(v)| 2
, ϕ → 0 ∀ϕ ∈ WΓ1,∞ (Ω). (9.123)
1 + ε|E(vε )|2 D
Taking into account
| α1 θε f ε · vε , ϕ − α1 θ f · v, ϕ |
≤ | α1 θε f ε · vε , ϕ − α1 θ f ε · vε , ϕ | + | α1 θ f ε · vε , ϕ − α1 θ f · vε , ϕ |
+ | α1 θ f · vε , ϕ − α1 θ f · v, ϕ |
≤ cθε − θ L 2 f ε L3 vε L6 ϕ L ∞ + cθ L 2 f ε − f L3 vL6 ϕ L ∞
+ cθ L 2 f L3 vε − vL6 ϕ L ∞ ,
we can prove
α1 θε f · vε , ϕ → α1 θ f · v, ϕ ∀ϕ ∈ WΓ1,∞
D
(Ω) as ε → 0. (9.124)
By virtue of (9.117), (9.118), (9.123) and (9.124), from the second formula in
(9.98) we get the second formula in (9.13).
The estimates of solutions follow from (9.100) by (9.101).
In this section we prove the main result for the case of total pressure.
Theorem 9.4 Let Assumption 9.1 hold and assume that max{|α0 |, |α1 |} is small
enough in accordance with f, φi , i = 2, · · · , 7, g, g R , θ D (see (9.138)). Then, there
exists a solution (v, θ ) to Problem II-VI such that

vV ≤ c 1 + f Lt + f 1 V∗ ,
θ − W 1,2 (Ω) ≤ c f Lt + f 2Lt + f 1 V∗ f Lt + g R L 4/3 (Γ R ) + g L 6/5 (Ω) ,

)

|∇θ |r d x ≤ cL r/(1−σ
σ 1 + θ D 2W 1,2 ∀r (1 < r < 3/2),
Ω
(9.125)
where L σ is the one in (9.135) below and σ = 3−2r
3−r
.
First, we look for solutions to the following auxiliary problem:
Problem II-VIA. Let ζ > 0, λ > 0, ε > 0 and q ∈ ( 12

5
, 6). Find (v, θ ) ∈ V ×
1,2
W (Ω) such that θ − θ D ∈ WΓ D (Ω) and
1,2
⎧
⎪ α0 λ
⎪
⎪ a (θ ; v, u) + a (v, v, u) + ∇Φ ε (v), u − 1 − θ f ε , u
⎪
⎪
0 2
max{λ, θ 2L q }
⎪
⎪
⎪
⎪
⎪
⎨ = f 1 , u ∀u ∈ V,
ζ |E(v)|2
⎪
⎪ b0 (θ ; θ, ϕ) − γε (θ )v, ∇ϕ − α2 μ(θ ) , ϕ
⎪
⎪ max{ζ, vε V } 1 + ε|E(v)|2
⎪
⎪
⎪
⎪ α λ
⎪
⎪ −
1
θ f ε · v, ϕ = f 2 , ϕ ∀ϕ ∈ WΓ1,2 (Ω),
⎩
max{λ, θ 2L q } D
(9.126)
where f ε ∈ L∞ (Ω) is such that f − f ε Lt ≤ ε.
Theorem 9.5 There exists a solution (vε , θε ) ∈ V × W 1,2 (Ω) to Problem II-VIA.
D

A (v, η), (u, φ) = a0 (η + θ D ; v, u) + a2 (v, v, u) + ∇Φε (v), u
α0 λ
− 1− 2
(η + θ D ) f ε , u + b0 (η + θ D ; η + θ D , ϕ)
ζ |E(v)|2
− γε (η + θ D )v, ∇ϕ − α2 μ(η + θ D ) ,ϕ
max{ζ, vε V } 1 + ε|E(v)|2
α1 λ
− 2
(η + θ D ) f ε · v, ϕ
∀(v, η), (u, φ) ∈ H .
Note that instead of max{δ,aδ1 (v)V∗ } a1 (v, v, u) in Problem I-VIA, for Problem II-VIA
the term a2 (v, v, u) is used.
Using
a2 (v, v, v) = 0,
|a2 (v, v, u)| ≤ K v2V uV ,
|a2 (vε , vε , vε − u)| ≤ c∇vε L2 vε L4 vε − uL4 ,
respectively, in (9.29), (9.32) and (9.36), we can see that the proof of Theorem
9.2 for Problem I-VIA is valid for Problem II-VIA. Thus, we come to the asserted
conclusion.
As (9.44) let us choose q0 such that
1 1 1 12
+ + ≤ 1, < q0 < 3.
q0 t 3 5
√
Lemma 9.4 If |α0 | λ ≤ 1, then there exists a parameter ζ such that
ζ
=1 (9.127)
max{ζ, vε V }
for all small ε and solutions (vε , θε ) of (9.126).

√
Proof Since |α0 | λ ≤ 1, we have
√
α0 λ c λ
θε f ε , u ≤ θε L q0 f ε Lt uL6 ≤ c f ε Lt uV∗ .
max{λ, θε 2L q0 } max{λ, θε 2L q0 }
(9.128)
Putting u = vε in the first equation of (9.126), we have
α0 λ
a0 (θε ; vε , vε ) + a2 (vε , vε , vε ) + ∇Φε (vε ), vε − 1 − θε f ε , vε
max{λ, θε 2L q0 }
= f 1 , vε .
(9.129)
Taking into account a2 (vε , vε , vε ) = 0, (9.30) and (9.128), we have from (9.129)
α0 λ
2μ0 vε 2V ≤ a0 (θε ; vε , vε ) ≤ 1− θε f ε , v + f 1 , vε
max{λ, θε L q0 }
2
≤ c( f ε L 65 + f ε Lt + f 1 V∗ )vε V ,
which implies
c
vε V ≤ ( f ε Lt + f 1 V∗ ).
2μ0
Since f ε → f in Lt (Ω), again we may assume that f ε Lt ≤ 1 + f Lt for all ε.
Therefore,
c
vε V ≤ (1 + f Lt + f 1 V∗ ). (9.130)
2μ0
Putting ζ = c
2μ0
(1 + f Lt + f 1 V∗ ), we come to the asserted conclusion.
√
Lemma 9.5 If max{|α0 |, |α1 |} λ ≤ 1, then under the parameter ζ by Lemma 9.4
there exists a λ2 independent of ε such that

θε L q0 ≤ λ2 . (9.131)
If moreover 1 < r < 23 , then

)
3 − 2r
|∇θε |r d x ≤ cL r/(1−σ
σ 1 + θ D 2W 1,2 , σ = , (9.132)
Ω 3−r
where L σ is the one in (9.135) below.
Proof By Lemma 9.1 (which is valid for the second formula of (9.126)), we have

θε− 2W 1,2 ≤ c f ε Lt vε V + g R L 4/3 (Γ R ) + g L 6/5 (Ω) . (9.133)
Using (9.133) and arguing as in (9.61)–(9.89) yield

|∇θε |2
+ 1+σ
dx
Ω (1 + (θ ε − d0 ) )

1 2
≤c 1+ 1 + vε 2V + f ε Lt vε V + g R L 4/3 (Γ ) + g L 6/5 (Ω)
σ R

2
+θ D 2 + vε 2V + f ε Lt vε V + g R L 4/3 (Γ ) + g L 6/5 (Ω) vε 2V .
W 1,2 R
(9.134)
Combining (9.134) with (9.130), we have

|∇θε |2
+ 1+σ
dx
Ω (1 + (θε − d0 ) )

1
≤c 1+ 1 + (1 + f Lt + f 1 V∗ )2 + f Lt (1 + f Lt + f 1 V∗ )
σ
2
+g R L 4/3 (Γ ) + g L 6/5 (Ω) + θ D 2 + (1 + f Lt + f 1 V∗ )2
R W 1,2

2 2
+ f Lt (1 + f Lt + f 1 V∗ ) + g R L 4/3 (Γ ) + g L 6/5 (Ω) 1 + f Lt + f 1 V∗
R
≡ Lσ .
(9.135)
Using (9.135), in the same way as in (9.94) we have

0)

θε L q0 ≤ c L 1/(1−σ
σ0 (1 + θ D W 1,2 ) + θ D W 1,2 ≡ L σ0 , (9.136)
where L σ0 is the one with σ0 instead of σ in L σ of (9.135). Putting


λ 2 = L σ0 , (9.137)
we get (9.131).
Now, for 1 < r < 23 putting σ = 3−2r
3−r
and repeating the arguments of (9.91)–
(9.93), we have (9.132).
Fixing λ = λ2 , under the condition

max{|α0 |, |α1 |} λ2 ≤ 1, (9.138)
by (9.136) and (9.137) we have
λ2
= 1 ∀ε > 0.
max{λ2 , θε 2L q0 }
Therefore, by virtue of Lemmas 9.4 and 9.5 we have

Theorem 9.6 If
max{|α0 |, |α1 |} λ2 ≤ 1,
then there exists a solution (vε , θε ) ∈ V × W 1,2 (Ω) to the following problem
⎧
⎪
⎪ a0 (θε ; vε , u) + a2 (vε , vε , u) + ∇Φε (vε ), u − (1 − α0 θε ) f ε , u = f 1 , u
⎪
⎪
⎪
⎪ ∀u ∈ V,
⎨
|E(vε )|2
⎪
⎪ b0 (θε ; θε , ϕ) − γε (θε )vε , ∇ϕ − α2 μ(θε ) , ϕ − α1 θε f ε · vε , ϕ
⎪
⎪ 1 + ε|E(vε )| 2
⎪
⎪
⎩
= f 2 , ϕ ∀ϕ ∈ WΓ1,2 D
(Ω),
θε − θ D ∈ WΓ1,2
D
(Ω),

c
vε V ≤ (1 + f Lt + f 1 V∗ ),
2μ0
θε− WΓ1,2 (Ω) ≤ c f Lt vε V + g R L 4/3 (Γ R ) + g L 6/5 (Ω) ,
D
)
3
|∇θ |r d x ≤ cL r/(1−σ
σ 1 + θ D 2W 1,2 ∀r, 1 < r < ,
Ω 2
where σ = 3−2r
3−r
.
Now repeating the arguments as in Sect. 9.3.3 with the solutions of Theorem 9.6, we
complete the proof of Theorem 9.4.

Several papers concerning (9.1) have appeared.
In [2] the problem with α0 = α1 = 0 was studied under homogeneous Dirichlet
boundary conditions for velocity and mixture of non-homogeneous Dirichlet condi-
tion and homogeneous Neumann condition for temperature.
In [3] when |α0 |, |α1 | are small enough or α0 = 0, αα01 > 1, the existence of a solu-
tion to the problems was shown under homogeneous Dirichlet boundary conditions
for velocity and mixture of non-homogeneous Dirichlet and homogeneous Neumann
conditions for temperature.
In [4] for problem on an open channel domain when α0 = α1 = 0, local solvability
was studied under mixture of Dirichlet boundary condition of velocity and the free
outflow condition for fluid and mixture of non-homogeneous Dirichlet and Neumann
boundary conditions for temperature.
In [5] a modified problem on an open channel domain, where the buoyancy term
(1 − α0 θ ) f is changed by ρ(θ ) f, 0 < ρ(θ ) < ρ1 (const), α1 = 0 and the viscosity,
specific heat and thermal conductivity are independent of temperature, was consid-
ered. For the problem under mixture of Dirichlet boundary condition of velocity and
the free outflow condition for fluid and mixture of non-homogeneous Dirichlet and
homogeneous Neumann boundary conditions for temperature it was proved that if
the data of problem are small enough, then there exists a unique strong solution.
In [6] when the viscosity, specific heat and thermal conductivity are indepen-
dent of temperature, the steady problem on 2-D bounded domain was considered.
When the body force is small enough, the existence of a strong solution was proved
under mixture of Dirichlet boundary condition of velocity, tangent stress and stress
condition for fluid and mixture of non-homogeneous Dirichlet and homogeneous
Neumann boundary conditions for temperature.
In several papers the non-Newtonian fluid were studied.
In [7] for non-Newtonian fluid with α0 = α1 = 0, under homogeneous Dirichlet
boundary condition for velocity and mixture of homogeneous Dirichlet and Robin
conditions for temperature existence of a solution was studied.
In [8] for the problems of a generalized Newtonian fluid with α0 = α1 = 0 regu-
larity of weak solutions was proved under homogeneous Dirichlet boundary condi-
tions of velocity and temperature. In [9] for the problems of a non-Newtonian fluid
with α0 = α1 = 0 existence of a solutions was proved under homogeneous Dirichlet
boundary conditions for velocity and temperature on a portion of boundary and a
generalized Navier slip and Robin conditions for velocity and temperature on another
portion of boundary.
In [10] for the problem of a non-Newtonian fluid with heat sources allowed in L 1
and even as measures, under homogeneous Dirichlet boundary conditions for velocity
and Robin condition for temperature the existence of a distributional solution was
shown for sufficiently small data.
But all results above exclude Newtonian fluid owing to conditions for nonlinear
terms for strain.
References
1. T. Kim, D. Cao, Mixed boundary value problems of the system for steady flow of heat-
conducting incompressible viscous fluids with dissipative heating. Meth. Appl. Anal. 27, 87–
124 (2020)
2. J. Naumann, Existence of weak solutions to the equations of stationary motion of heat-
conducting Incompressible viscous fluids. Progress in Nonlinear Differential Equations and
Their Applications, vol. 66 (Birkhäuser, 2005), pp. 373–390
3. J. Naumann, M. Pokorny, J. Wolf, On the existence of weak solutions to the equations of steady
flow of heat-conducting fluids with dissipative heating. Nonlinear Anal. RWA 13, 1600–1620
(2012)
4. M. Beneš, P. Kučera, On the Navier-Stokes flows for heat-conducting fluids with mixed bound-
ary conditions. J. Math. Anal. Appl. 389, 769–780 (2012)
5. M. Beneš, A note on regularity and uniqueness of natural convection with effects of viscous
dissipation in 3D open channels. Z. Angew. Math. Phys. 65, 961–975 (2014)
6. M. Beneš, A note on the regularity of thermally coupled viscous flows with critical growth in
nonsmooth domains. Math. Meth. Appl. Sci. 36, 1290–1300 (2013)
7. L. Consiglieri, Stationary weak solutions for a class of non-Newtonian fluids with energy
transfer. Int. J. Non-Linear Mech. 32, 961–972 (1997)
8. L. Consiglieri, T. Shilkin, Regularity of stationary weak solutions in the theory of generalized
Newtonian fluid with energy transfer. J. Math. Sci. 115, 2771–2788 (2003)
9. L. Consiglieri, A ( p − q) coupled system in elliptic nonlinear problems with nonstandard
10. T. Roubíček, Steady-state buoyancy-driven viscous flow with measure data. Mathematica
Bohemica 126, 493–504 (2001)
Chapter 10
The Non-steady Equations for
Heat-Conducting Fluids
In this chapter we are concerned with a non-steady system for motion of incom-
pressible Newtonian heat-conducting fluids with mixed boundary conditions. The
boundary condition for fluid is the case of total pressure and the boundary conditions
for temperature may include Dirichlet, Neumann and Robin conditions together. On
the basis of results of Sect. 3.1, we get a variational formulation for the problem.
The variational formulation consists of a time-dependent variational inequality for
velocity due to the boundary conditions of friction type and a variational equation
for temperature.
Then we prove the existence of a solution to the problem. It is proved that if the
buoyancy effect and energy dissipation effect due to expansion are small enough in
accordance with data of problem, then there exists a solution with “defect measure”.
10.1 Problem and Variational Formulation

10.1.1 Problem and Assumption
We are concerned with the following non-steady problem

⎧
⎪ ∂v
⎪
⎪ − 2∇ · μ(θ )E(v) + (v · ∇)v + ∇ p = (1 − α0 θ) f,
⎪
⎪ ∂t
⎪
⎪
⎨ div v = 0,
(10.1)
⎪
⎪ ∂θ
⎪
⎪ − ∇ · (κ(θ)∇θ ) + v · ∇θ − α2 μ(θ )E(v) : E(v) = α1 θ f · v + g,
⎪
⎪ ∂t
⎪
⎩
v(0) = v0 , θ(0) = θ0 .
https://doi.org/10.1007/978-3-030-78659-5_10
322 10 The Non-steady Equations for Heat-Conducting Fluids
Due to the dissipation of energy μ(θ )E(v) : E(v), it is more difficult to study the
system (10.1) than the Boussinesq system.
Let Ω be as in Chap. 8. (See Remark 6.1.)
For temperature we are concerned with the boundary conditions
(1) θ |Γ D = 0,
∂θ
(2) κ(θ ) + β(x)θ Γ R = g R (t, x), (10.2)
∂n
β(x), g R (t, x) − given functions on Γ R , (0, T ) × Γ R .
The boundary condition for fluid is as in (7.4).

H : completion in L2 (Ω) of V,
HK : closure in L2 (Ω) of K (Ω),
(Q) = {u ∈ C2 ( Q̄) : div u = 0, u| 1 = 0, u τ |( 2
∪ 4 ∪ 7 ) = 0, u · n|( 3∪ 5)
= 0},
1, p
WΓ (Ω) = {y ∈ W 1, p (Ω) : y|Γ D = 0},
D
CΓ1 D (Ω) = {y ∈ C 1 (Ω̄) : y|Γ D = 0}.
We use inner products in (7.8) for V and WΓ1,2

D
(Ω).
Assumption 10.1 Assume the followings.

(1) Γ1 = ∅, Γ D = ∅ and

Γ R ⊂ ∪i=1,3,5,8 Γi . (10.3)

(3) For the functions of (10.1) f ∈ L ∞ (Q), g ∈ L 1 0, T ; (WΓ1,2
D
)∗ and
μ ∈ C(R), 0 < μ0 ≤ μ(ξ ) ≤ μ1 < ∞ ∀ξ ∈ R,

κ ∈ C(R), 0 < κ0 ≤ κ(ξ ) ≤ κ1 < ∞ ∀ξ ∈ R, (10.4)
v0 ∈ HK , θ0 ∈ L (Ω). 1
(4) For the functions of (10.2), (7.4)

10.1 Problem and Variational Formulation 323
g R ∈ L 1 (0, T ; L 4/3 (Γ R )),

β1 ≥ β(x) ≥ 0, β1 − a constant, β(x) − measurable,
φi ∈ L 2 (0, T ; H − 2 (Γi )), i = 2, 4, 7, φi ∈ L 2 (0, T ; H− 2 (Γi )), i = 3, 5, 6,
1 1
αi j ∈ L ∞ (Γ5 ).
(10.5)
10.1.2 Variational Formulation for Problem
In this section we first give variational formulations for the problem above.
Taking (v · ∇)v = rot v × v + 21 grad|v|2 into account, by (7.9)–(7.11), (8.8), we
can see that smooth solutions (v, p, θ ) of problem (10.1), (10.2), (7.4) satisfy the
following.
⎧
⎪ ∂v
⎪
⎪ , u + 2(μ(θ )E(v), E(u)) + rotv × v, u + 2(μ(θ )k(x)v, u)Γ2
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ + 2(μ(θ )S ṽ, ũ)Γ3 + 2(α(x)v, u)Γ5 + (μ(θ )k(x)v, u)Γ7
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪ − 2(μ(θ )εnτ (v), u)Γ8 + ( p + |v|2 − 2μ(θ )εnn (v), u n )Γ9 ∪Γ10 ∪Γ11
⎪
⎪ 2
⎪
⎪
⎪
⎪ = (1 − α θ ) f, u + φ , u n Γi + φi , uΓi ∀u ∈ V,
⎪
⎪ 0 i
⎪
⎨ i=2,4,7 i=3,5,6
⎪ ∂θ
⎪
⎪ , ϕ + (κ(θ )∇θ, ∇ϕ) − (θ v, ∇ϕ) − (α2 μ(θ )|E(v)|2 , ϕ) + (βθ, ϕ)Γ R
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ − (α1 θ f · v, ϕ) = g R , ϕΓ R + g, ϕ ∀ϕ ∈ WΓ1,∞ (Ω),
⎪
⎪ D
⎪
⎪
⎪
t t
⎪
⎪
⎪
⎪ |σnt (θ, v, p)| ≤ gn , σnt (θ, v, p)vn + gn |vn | = 0 on Γ9 ,
⎪
⎪
⎪
⎪
⎪
⎪ σnt (θ, v, p) + g+n ≥ 0, (σnt (θ, v, p) + g+n )vn = 0 on Γ10 ,
⎪
⎩ t
(10.6)

+ 2(α(x)w, u)Γ5 + (μ(θ )k(x)w, u)Γ7 ∀w, u ∈ V, θ ∈ W 1,2 (Ω),
a1 (v, u, w) = rot v × u, w ∀v, u, w ∈ V,
i=2,4,7 i=3,5,6
(10.7)
Define b0 (θ̃; ·, ·) and f 2 ∈ (WΓ1,2
D
(Ω))∗ by
b0 (θ̃; θ, ϕ) = (κ(θ̃)∇θ, ∇ϕ) + (β(x)θ, ϕ)Γ R ∀θ, θ̃ ∈ W 1,2 (Ω), ϕ ∈ WΓ1,2

D
(Ω),
f 2 , ϕ = g R , ϕΓ R + g, ϕ ∀ϕ ∈ WΓ1,2
D
(Ω).
(10.8)
By (10.4) and (10.5),
f 1 ∈ L 2 (0, T ; V∗ ), f 2 ∈ L 1 (0, T ; (WΓ1,2

D
)∗ ). (10.9)
1
2
(10.2), (7.4).
Problem VE. Find v ∈ K (Q), θ ∈ L ∞ (0, T ; L 1 (Ω)) ∩ L r (0, T ; WΓ1,rD (Ω)) for all
) ∈ L2τ (Γ8 ) × L 2 (Γ9 ) × H − 2 (Γ10 ) × H − 2 (Γ11 )
1 1
r ∈ [1, 5/4) and (στt , σnt , σ+n
t
, σ−n
t
for a.e. t ∈ (0, T ) such that v(0) = v0 , θ (0) = θ0 and

⎧
⎪ ∂v
⎪
⎪ , u + a0 (θ ; v, u) + a1 (v, v, u) − (στt , u τ )Γ8 − (σnt , u n )Γ9
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ − σ+n
t
, u n Γ10 − σ−n
t
, u n Γ11 − f − α0 θ f, u = f 1 , u ∀u ∈ V,
⎪
⎪
⎪
⎪ ∂θ
⎪
⎪
⎪
⎪ , ϕ + b0 (θ ; θ, ϕ) − θ v, ∇ϕ − α2 μ(θ )|E(v)|2 , ϕ − α1 θ f · v, ϕ
⎨ ∂t
⎪ = f 2 , ϕ ∀ϕ ∈ WΓ1,∞ (Ω),
⎪
⎪
D
⎪
⎪
t t
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ σ+nt
+ g+n ≥ 0, σ+n t
+ g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎩ t
σ−n − g−n ≤ 0, σ−n t
− g−n , vn Γ11 = 0 on Γ11 ,
(10.10)
(u, n)L2 (Γ8 ) = 0.
Remark 10.1 As in Remark 8.3, for solutions smooth enough Problem VE is equiv-
alent to the problem (10.1), (10.2), (7.4).

and a variational equation, which is equivalent to Problem VE if the solution is
smooth enough (see Remark 10.2).
For fixed θ , let us consider the problem
⎧
⎪ ∂v
⎪
⎪ , u + a0 (θ ; v, u) + a1 (v, v, u) − (στ , u τ )Γ8 − (σn , u n )Γ9
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ − σ+n , u n Γ10 − σ−n , u n Γ11 − ( f − α0 θ f, u = f 1 , u,
⎪
⎨ t
|στ | ≤ gτ , στt · vτ + gτ |vτ | = 0 on Γ8 , (10.11)
⎪
⎪
⎪
⎪
⎪
⎪
⎪ σ t + g+n ≥ 0, σ+n t
+ g+n , vn Γ10 = 0 on Γ10 ,
⎪
⎪ +n
⎩
σ−n − g−n ≤ 0, σ−n − g−n , vn Γ11 = 0 on Γ11 .
t t
Subtracting the first formula of (10.11) with u = v from the first one of (10.11), we
get
∂v
, u − v + a0 (θ ; v, u − v) + a1 (v, v, u − v) − (στt , u τ − vτ )Γ8 − (σnt , u n − vn )Γ9
∂t (10.12)
t ,u − v
− σ+n t
n n Γ10 − σ−n , u n − vn Γ11 − f − α0 θ f, u − v = f 1 , u − v ∀u ∈ V.
Let Φ : V → R be the functional defined by (5.22). Then, the functional Φ is

proper, convex, lower weak semi-continuous and nonnegative.
Define a functional Ψ (u) by
⎧ T
⎨ Φ(u(t)) dt i f Φ(u(t)) ∈ L 1 (0, T ),
Ψ (u) = (10.13)
⎩ 0
+∞ other wise.
In the same way as in Problem I of Chap. 5, we get from (10.7) and (10.12)
∂v
, u − v + a0 (θ ; v, u − v)+a1 (v, v, u − v) + Φ(u) − Φ(v)
∂t (10.14)
≥ (1 − α0 θ ) f, u − v + f 1 , u − v.
If u − v ∈ (Q), then

T
∂v T
∂(v − u) T
∂u
, u − v dt = ,v − u dt + , u − v dt
0 ∂t 0 ∂t 0 ∂t
T
∂u
1
= , u − v dt +
(v0 − u(0)2 − v(T ) − u(T )2 ).
0 ∂t
2
(10.15)
Define operators A(θ ) : V → V∗ and B : V × V → V∗ , respectively, by
A(θ )v, u = a0 (θ ; v, u) ∀w, u ∈ V,

(10.16)
B(v, u), w = a1 (v, u, w) ∀v, u, w ∈ V.
If v ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; L2 ), then v ∈ L 4 (0, T ; L3 ) (see

T (1.21)) and B(v, v) ∈
L 3 (0, T ; V∗ ). Thus, when u ∈ L 4 (0, T ; V), the integral 0 B(v, v), u − v dt is
4
meaningful since B(v, v), v = 0.

Combining (10.14), (10.15), and neglecting 21 (w(T ) − u(T )2 ), we come to the
following variational inequality corresponding to problem (10.11) (see (6.11), or
(1.7) of [1], (2.1) of [2]).
Find v ∈ L 2 (0, T ; V) ∩ L ∞ (0, T ; H ) such that
T
u + A(θ )v(t) + B(v(t), v(t)) − (1 − α0 θ ) f − f 1 , u(t) − v(t) dt + Ψ (u) − Ψ (v)
0 (10.17)
1
≥ − v0 − u(0)2 ∀u ∈ L 4 (0, T ; V) with u ∈ L 2 (0, T ; V∗ ).
2
If (v, θ, στ , σn , σ+n , σ−n ) is a solution of Problem VE, then (v, θ ) satisfies the prob-
lem above.
Therefore, we have the following variational formulation of the problem which
consists of a variational inequality for velocity and a variational equation for tem-
perature.

Problem VI. Find (v, θ ) ∈ L ∞ (0, T ; H ) ∩ L 2 (0, T ; V) × L ∞ (0, T ; L 1 (Ω)) ∩

L r (0, T ; WΓ1,rD ) ∀r ∈ [1, 5/4) such that
⎧ T
⎪
⎪
⎪
⎪ u + A(θ )v(t) + B(v(t), v(t)) − (1 − α0 θ ) f − f 1 , u(t) − v(t) dt + Ψ (u) − Ψ (v)
⎪
⎪
⎪
⎪
0
⎪
⎪ 1
⎪
⎨ ≥ − v0 − u(0)2 ∀u ∈ L 4 (0, T ; V) with u ∈ L 2 (0, T ; V∗ ),
2

⎪ T
⎪
⎪ ∂ϕ
⎪
⎪ − θ, + b0 (θ ; θ, ϕ) − θ v, ∇ϕ − α2 μ(θ )|E(u)|2 , ϕ − α1 θ f · v, ϕ
⎪
⎪ 0 ∂t
⎪
⎪
⎪
⎪
⎩ − f 2 , ϕ dt = θ0 (x), ϕ(x, 0) ∀ϕ ∈ C 1 (0, T ; CΓ1 (Ω)) with ϕ(·, T ) = 0.
D
(10.18)
Remark 10.2 If the solutions to Problem VI is smooth such that v ∈ L 2 (0, T ; V),
v ∈ L 2 (0, T ; V∗ ), then v satisfies (10.14), which is equivalent to
v (t) + A(θ )v(t) + B(v(t), v(t)) − (1 − α0 θ ) f − f 1 , u − v(t) + Φ(u) − Φ(v(t)) ≥ 0

(10.19)
for a.e. t ∈ [0, T ], ∀u ∈ K (Ω)

(Remark, p. 114 of [1]). In (10.14) putting F1 , u − v = − ∂v ∂t
− (1 − α0 θ ) f −
f 1 , u − v, by Theorem 5.4 we can get the existence of (στ , σn , σ+n , σ−n ) ∈ L2τ (Γ8 ) ×
L 2 (Γ9 ) × H −1/2 (Γ10 ) × H −1/2 (Γ11 ) for a.e. t ∈ (0, T ) such that (v, θ, στ , σn , σ+n , σ−n )
satisfies (10.10).
It is desirable to get a solution satisfying (10.18), but we can only prove the
existence of (v, θ ) with a “defect measure” in the second equation of (10.18)
The main result of this chapter is the following
Theorem 10.1 Let Assumption 10.1 be satisfied. If |α0 | + |α1 | is small enough in
accordance with the date of the problem, then there exists (v, θ ) ∈ L ∞ (0, T ; H )

∩L 2 (0, T ; V) × L ∞ (0, T ; L 1 (Ω)) ∩ L r (0, T ; WΓ1,rD ) ∀r ∈ [1, 5/4) and a Radon
measure μ such that
⎧ T
⎪
⎪ u + A(θ )v(t) + B(v(t), v(t)) − (1 − α0 θ ) f − f 1 , u(t) − v(t) dt + Ψ (u) − Ψ (w)
⎪
⎪
⎪
⎪ 0
⎪
⎪
⎪
⎪ 1
⎪
⎪
⎪ ≥ − v0 − u(0)2 ∀u ∈ L 4 (0, T ; V) with u ∈ L 2 (0, T ; V∗ ),
⎪
⎪ 2
⎨
T ∂ϕ
⎪ − θ, + b0 (θ ; θ, ϕ) − θ v, ∇ϕ − α2 μ(θ )|E(u)|2 , ϕ − α1 θ f · v, ϕ
⎪
⎪ ∂t
⎪
⎪
0
T
⎪
⎪
⎪
⎪
⎪
⎪ − f 2 , ϕ dt = θ0 (x), ϕ(x, 0) + ϕ dμ
⎪
⎪
⎪
⎪
0
⎩
∀ϕ ∈ C 1 (0, T ; CΓ1 (Ω)) with ϕ(·, T ) = 0.
D
(10.20)
Moreover, (v, θ ) satisfies the estimate

|∇θ |2
sup v(t)2 + v2 2 + ess sup θ (t) L 1 (Ω) + |∇θ |r d xdt + d xdt
L (0,T ;V) (1 + |θ |)1+δ
t∈[0,T ] t∈[0,T ] Q Q
≤ C1 v0 , f ∞ , φi −1 , φi −1 ,

i=2,4,7 L 2 (0,T ;H 2 (Γi )) i=3,5,6 L 2 (0,T ;H 2 (Γi ))
δ, |α|, θ0 L 1 (Ω) , g R L 1 (0,T ;L 4/3 (Γ )) , g 1

R L (0,T ;(W 1,2 )∗ )
ΓD
5 5 − 4r
∀1 ≤ r < , ∀0 < δ < .
4 3
(10.21)
10.2 Existence of a Solution
10.2.1 Existence of a Solution to an Approximate Problem
We first consider a problem approximating (10.18).

For every 0 < ε < 1, let Φε be the Moreau-Yosida approximation of Φ and ∇Φε
be Fréchet derivative of Φε .
Let v0 ∈ W1,6 (Ω) ∩ K (Ω), θ0 ∈ WΓ1,6 D

(Ω) and 0 < ε ≤ 1. Also assume that
g R ∈ L (0, T ; L (Γ R )), g ∈ L (0, T ; (WΓ1,2
2 4/3 2
D
)∗ ), and so f 2 ∈ L 2 (0, T ; (WΓ1,2
D
)∗ ).
Problem VEA. Find v ∈ L 6 (0, T ; V) ∩ C([0, T ]; H ), θ ∈ L 2 (0, T ; WΓ1,2

D
) ∩ C([0, T ];
2
L (Ω)) such that
⎧
⎪ ∂v
⎪
⎪ , u + 2 [μ(θ ) + εv4V ]E(v), E(u) + rotv × v, u + 2(μ(θ )k(x)v, u)Γ2
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ + 2(μ(θ )S ṽ, ũ)Γ3 + 2(α(x)v, u)Γ5 + (μ(θ )k(x)v, u)Γ7 + ∇Φε (v(t)), u
⎪
⎪
⎪
⎪
⎪
⎪ α 0θ
⎪ ∀u ∈ L 6 (0, T ; V),
⎨ = (1 − 1 + εθ 2 ) f, u + f 1 , u
⎪
⎪
⎪ ∂θ |E(v)|2
⎪
⎪ ∂t , ϕ + (κ(θ )∇θ, ∇ϕ) + (β(x)θ, ϕ)Γ R − vθ, ∇ϕ − α2 μ(θ ) 1 + ε|E(v)|2 , ϕ
⎪
⎪
⎪
⎪
⎪
⎪ α1 θ
⎪
⎪
⎪
⎪ = f · v, ϕ + f 2 , ϕ ∀ϕ ∈ L 2 (0, T ; WΓ1,2 (Ω)),
⎪
⎪ 1 + εθ 2 D
⎪
⎩
v(0) = v0 , θ (0) = θ0 .
(10.22)
Let
V := L 6 (0, T ; V) × L 2 (0, T ; WΓ1,2
D
),
1/2
(·, ·)V = · 2L 6 (0,T ;V) + · 2L 2 (0,T ;W 1,2 ) .
ΓD
Then
V ∗ = L 6/5 (0, T ; V∗ ) × L 2 0, T ; (WΓ1,2
D
)∗ ,
1/2
(·, ·)V ∗ = · 2L 6/5 (0,T ;V∗ ) + · 2 2 .
L 0,T ;(WΓ1,2 )∗
D
Theorem 10.2 There exists a solution (vε , θε ) ∈ V to Problem VEA.
Proof In (10.22) let us make changes of the unknown functions by v = ek1 t ŵ, θ = θ̂,
where k1 is a constant to be determined later (in (10.37)). Owing to the changes,
differently from Ch. 9 we need not assume that Γ2 j , Γ3 j , Γ7 j are convex and the
matrix α is positive. Then we have the following problem equivalent to Problem
VEA.
Problem VEA’. Find ŵ ∈ L 6 (0, T ; V) ∩ C([0, T ]; H ), θ̂ ∈ L 2 (0, T ; WΓ1,2

D
)∩ C([0, T ]; L 2 (Ω))
such that
10.2 Existence of a Solution 329
⎧
⎪ ∂ ŵ
⎪
⎪ , u + 2 [μ(θ̂) + e4k1 t εŵ4V ]E(ŵ), E(u) + ek1 t rotŵ × ŵ, u
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ ˜ ũ)Γ
⎪
⎪ + k1 (ŵ, u) + 2(μ(θ̂ )k(x)ŵ, u)Γ2 + 2(μ(θ̂ )S ŵ,
⎪
⎪
3
⎪
⎪ −k
+ 2(α(x)ŵ, u)Γ5 + (μ(θ̂ )k(x)ŵ, u)Γ7 + e ∇Φε (ek1 t ŵ(t)), u
t
⎪
⎪
1
⎪
⎪
⎪
⎪ α0 θ̂
⎪
⎪ = e−k1 t (1 − f ), u + e−k1 t f 1 , u
⎪
⎪
⎪
⎪ 1 + ε θ̂ 2
⎪
⎪
⎪
⎪ ∀u ∈ L 6 (0, T ; V),
⎨
⎪ ∂ θ̂
⎪
⎪ , ϕ + (κ(θ̂)∇ θ̂ , ∇ϕ) + (β(x)θ̂ , ϕ)Γ R − ek1 t ŵ θ̂ , ∇ϕ
⎪
⎪ ∂t
⎪
⎪
⎪
⎪ |E(ŵ)|2
⎪
⎪ − α μ( θ̂) ,ϕ
⎪
⎪
2
e−2k1 t + ε|E(ŵ)|2
⎪
⎪
⎪
⎪
⎪
⎪ α1 θ̂
⎪
⎪ = e2k1 t f · ŵ, ϕ + f 2 , ϕ
⎪
⎪ + ε θ̂
⎪
⎪ 1 2
⎪
⎪
⎪
⎪ ∀ϕ ∈ L 2 (0, T ; WΓ1,2 (Ω)),
⎪
⎪ D
⎩
ŵ(0) = v0 , θ̂(0) = θ0 .
(10.23)
Define an operator L : D(L ) → V ∗ by

D(L ) := (v, θ ) ∈ V , (v , θ ) ∈ V ∗ , v(0) = 0, θ (0) = 0 ,
L (v, θ ) = (v , θ ),
(·, ·) D(L ) := (·, ·V + ·, ·V ∗ ).
Then, L is a linear, maximal monotone operator (see Example 1.1).

Putting
ŵ = w + v0 , θ̂ = θ + θ0 ,
we will consider an evolution problem with zero initial conditions.

Define an operator Aε : D(L ) → V ∗ by
T

Aε (w, θ ), (u, φ) = 2 [μ(θ̂ ) + e4k1 t εŵ4V ]E(ŵ), E(u)
0
+ ek1 t rotŵ × ŵ, u + k1 (ŵ, u) + 2(μ(θ̂ )k(x)ŵ, u)Γ2
˜ ũ)Γ + 2(α(x)ŵ, u)Γ + (μ(θ̂ )k(x)ŵ, u)Γ
+ 2(μ(θ̂ )S ŵ, 3 5 7
α0 θ̂
+ e−k1 t ∇Φε (ek1 t ŵ(t)), u + e−k1 t f, u (10.24)
1 + ε θ̂ 2
+ (κ(θ̂ )∇ θ̂, ∇φ) + (β(x)θ̂ , φ)Γ R − ek1 t ŵθ̂, ∇φ
|E(ŵ)|2 α1 θ̂
− α2 μ(θ̂ ) −2k t , φ − e2k1 t f · ŵ, φ dt
e 1 + ε|E(ŵ)|2 1 + ε θ̂ 2
∀(w, θ ) ∈ D(L ), (u, φ) ∈ V .
Estimate (10.47) to be given later shows that this operator is well-defined. Then, the
existence of a solution to Problem VEA’ is equivalent to one of a solution to
−k t
e 1 ( f + f1 )
L (w, θ ) + Aε (w, θ ) = F , F = .
f2
Applying Theorem 1.47, we will prove existence of a solution to the equation above.
To this end, we need to check that Aε satisfies assumptions of Theorem 1.47.
(i) Let us prove property (iii) of Theorem 1.47 for Aε (w, θ ):
1
Aε (w, θ ), (w, θ ) → ∞ as (w, θ )V → ∞. (10.25)
(w, θ )V
We have that
T

Aε (w, θ ), (w, θ ) = 2 [μ(θ̂ ) + e4k1 t εŵ4V ]E(ŵ), E(w)
0
+ ek1 t rotŵ × ŵ, w + k1 (ŵ, w) + 2(μ(θ̂ )k(x)ŵ, w)Γ2
˜ w̃)Γ + 2(α(x)ŵ, w)Γ + (μ(θ̂ )k(x)ŵ, w)Γ
+ 2(μ(θ̂ )S ŵ, 3 5 7
α0 θ̂ (10.26)
+ e−k1 t ∇Φε (ek1 t ŵ(t)), w + e−k1 t f, w
1 + ε θ̂ 2
+ (κ(θ̂)∇ θ̂, ∇θ ) + (β(x)θ̂ , θ )Γ R − ek1 t ŵθ̂ , ∇θ
|E(ŵ)|2 α1 θ̂
− α2 μ(θ̂ ) −2k t , θ − e2k1 t f · ŵ, θ dt.
e 1 + ε|E(ŵ)|2 1 + ε θ̂ 2
Let us now estimate each term on the right hand side above. By Hölder’s inequality
we have
T
2 [μ(θ̂ ) + e4k1 t εŵ4V ]E(ŵ), E(w) dt
0
T
≥ 2μ0 w2 2 + 2 μ(θ̂ )E(v̂0 ), E(w) dt
L (0,T ;V) 0
T T
+ε 2 e4k1 t ŵ4V E(ŵ), E(ŵ) dt − ε 2 e4k1 t ŵ4V E(ŵ), E(v0 ) dt
0 0
≥ μ0 w2 2 + 2εw + v0 6 6 − 2cεw + v0 5 6 v − kT v0 2
L (0,T ;V) L (0,T ;V) L (0,T ;V) 0 L 6 (0,T ;V)
≥ μ0 w2 2 + εw6 6 − K1,
L (0,T ;V) L (0,T ;V)
(10.27)
where K 1 depends on v0 V , ε, T . By the fact that Γ2 j , Γ3 j , Γ7 j are in C 2.1 and

Assumption 10.1, there exists a constant M such that
S(x)∞ , k(x)∞ , α L ∞ (Γ5 ) ≤ M. (10.28)

Thus,
T

2(μ(θ̂ )k(x)w, w Γ + 2 μ(θ̂ )S w̃, w̃ Γ + 2 α(x)w, w Γ + μ(θ̂ )k(x)w, w Γ dt
0 2 3 5 7
1
≤ μ0 w2 2 + k11 w2 2
8 L (0,T ;V) L (0,T ;H )
(10.29)
(see Theorem 1.27) and

T

2(μ(θ̂ )k(x)v0 , w)Γ2 + 2(μ(θ̂ )S ṽ0 , w̃)Γ3 + 2(α(x)v0 , w)Γ5 + (μ(θ̂ )k(x)v0 , w)Γ7 dt
0
1
≤ μ0 w2 2 + K2,
8 L (0,T ;V)
(10.30)
where K 2 depends on v0 V , μ0 , μ1 , T . Since rot w × v, v = 0,
T T

ek1 t rotŵ × ŵ, w dt = ek1 t rot w × v0 , w + rot v0 × v0 , w dt
0 0 (10.31)
1
≤ μ0 w2 2 + K 3 + k12 w2 2 ,
8 L (0,T ;V) L (0,T ;H )
where K 3 depends on v0 W1,6 (Ω) , μ0 , μ1 , T .

Let us estimate
T T
e−k1 t ∇Φε (ek1 t ŵ(t)), w dt = e−k1 t ∇Φε (ek1 t ŵ(t)), ek1 t ŵ − ek1 t v0 dt.
0 0
Since Φε is convex, continuous and Fréchet differentiable, by (1.37) we have
∇Φε (ek1 t ŵ(t)), ek1 t v0 − ek1 t ŵ ≤ Φε (ek1 t v0 ) − Φε (ek1 t ŵ(t)) ≤ Φ(ek1 t v0 ).
Thus,
T T
e−k1 t ∇Φε (ek1 t ŵ(t)), w dt ≥ − e−k1 t Φ(ek1 t v0 ) dt = −K 4 , K 4 ≥ 0.
0 0
(10.32)
Also, we have
T

(κ(θ )∇θ, ∇θ ) dt ≥ κ0 θ 2 2 ,
0 L (0,T ;W 1,2 ) ΓD
T
κ0
(κ(θ )∇θ0 , ∇θ ) dt ≤ θ 2 2 + K5,
0 6 L (0,T ;WΓ1,2 )
T
D (10.33)
(β(x)θ, θ )Γ R dt ≥ 0,
0
T κ
0
(β(x)θ0 , θ )Γ R dt ≤ θ 2 2 + K6,
0 6 L (0,T ;WΓ1,2 )
D
where K 5 , K 6 depend on θ0 W 1,6 (Ω) and T . Since wθ, ∇θ = 0 (see (8.8)),
T T

−ek1 t ŵ θ̂ , ∇θ dt = −ek1 t ŵθ0 , ∇θ dt
0 0
T
(10.34)
= −ek1 t wθ0 , ∇θ + v0 θ0 , ∇θ dt
0
κ0
≤ θ 2L 2 (0,T ;W 1,2 ) + K 7 + k13 w2L 2 (0,T ;H ) .
6 ΓD
It is easy to get
T
|E(ŵ)|2 c
α2 μ(θ̂) , θ dt ≤ θ L 2 (0,T ;L 2 (Ω)) . (10.35)
0 e−2k1 t + ε|E(ŵ)|2 ε
Since
ξ 1
≤ √ ∀ξ ∈ [0, ∞),
1 + εξ 2 2 ε
we have
T
α0 θ̂
c
e−k1 t f, w dt ≤ √ w L 2 (0,T ;H ) ,
0 1 + εθ̂ 2 ε

T
α1 θ̂ K8 k14 κ0
e2k1 t f · ŵ, θ dt ≤ + w2L 2 (0,T ;H ) + θ 2L 2 (0,T ;W 1,2 ) .
0 1 + εθ̂ 2 ε ε 6 ΓD
(10.36)
Taking
k14
k1 = k11 + k12 + k13 + , (10.37)
ε
by (10.27)–(10.36) we have from (10.26)

Aε (w, θ ), (w, θ )
μ κ0
0 −k1 T
≥ min e , ε, w2L 2 (0,T ;V) + w6L 6 (0,T ;V) + θ 2L 2 (0,T ;W 1,2 )
2 2 ΓD
7
K8 c c
− − √ w L 2 (0,T ;H ) − θ L 2 (0,T ;L 2 (Ω)) ∀(w, θ ) ∈ D(L ),
Ki −
i=1
ε ε ε
(10.38)
(ii) Let us prove property (ii) of Theorem 1.47 for Aε (w, θ ).
Taking into account (10.28) and applying an inequality |a + b| p ≤ 2 p (|a| p +
|b| p ), p ∈ (1, ∞), we have
T
1
2 [μ(θ̂) + e4k1 t εŵ4V ]E (ŵ), E (u) + k1 (ŵ, u) + 2(μ(θ̂)k(x)ŵ, u)Γ2
u L 6 (0,T ;V) 0
(10.39)
+ 2(μ(θ̂)S ŵ, ˜ ũ)Γ + 2(α(x)ŵ, u)Γ + (μ(θ̂)k(x)ŵ, u)Γ dt
3 5 7

≤ c ŵ L 2 (0,T ;V) + ŵ6L 6 (0,T ;V) ≤ c K + w L 2 (0,T ;V) + w6L 6 (0,T ;V) ,
where K depends on v0 .
By (1.37) and Φ(0V ) = 0, we know that Φε (0V ) = 0. Also, since Φε is convex,
continuous and Fréchet differentiable, we have ∇Φε (0V ) = 0V∗ . Then, since the
operator ∇Φε is Lipschitz continuous with the constant ε−1 (see Remark 1.16), we
have
T
1
e−k1 t ∇Φε (ek1 t ŵ(t)), u dt
u L 6 (0,T ;V) 0
T
1
= e−k1 t ∇Φε (ek1 t ŵ(t)) − ∇Φε (0V ), u dt
u L 6 (0,T ;V) 0 (10.40)
T
1
1
≤ ŵ(t)V uV dt
u L 6 (0,T ;V) 0 ε
c c
≤ ŵ L 2 (0,T ;V) ≤ K + w L 2 (0,T ;V) .
ε ε
Similarly, we have
T |E(ŵ)|2
1
(κ(θ̂)∇ θ̂ , ∇φ) + (β(x)θ̂ , φ)Γ R − α2 μ(θ̂ ) −2k t , φ dt
φ 2 1,2 0 e 1 + ε|E(ŵ)|2
L (0,T ;W ) ΓD
1
≤ c θ̂ +
L 2 (0,T ;WΓ1,2 ) ε
D
1
≤ c K + θ + ,
L 2 (0,T ;WΓ1,2 ) ε
D
(10.41)
where K depends on θ0 .
On the other hand, taking into account w(0) = 0, θ (0) = 0 and applying (1.27),
for (w, θ ) ∈ D(L ) we have
(w, θ ) ∈ C([0, T ]; H ) × C([0, T ]; L 2 (Ω)),

wC([0,T ];H ) ≤ cw L 6/5 (0,T ;V∗ ) w L 6 (0,T ;V) ,
1/2 1/2
(10.42)
1/2 θ 1/2
θ C([0,T ];H ) ≤ cθ 2 L (0,T ;WΓ1,2 )
.
L 2 0,T ;(WΓ1,2 )∗ D
D
Also, by Theorem 1.12

1/2 1/2
wL3 (Ω) ≤ cwL2 (Ω) ∇wL2 (Ω) ∀w ∈ V,
1/4 3/4 1, p
(10.43)
θ L 4 (Ω) ≤ cθ L 2 (Ω) ∇θ L2 (Ω) ∀θ ∈ WΓ D (Ω).
Applying (10.42) and (10.43), by Hölder’s inequality with exponents 4, 4/3 and
Young’s inequality we have
T T

ek1 t rotŵ × ŵ, u dt ≤ c rotŵL2 ŵL3 uL6 dt
0 0
T
1/2 3/2
≤ cŵC([0,T ];H ) ∇ ŵ uV dt
0 L2
≤ cw
1/4 1/4 3/2
ŵ 6 ŵ 6 u L 4/3 (0,T ;V)
L 6/5 (0,T ;V∗ ) L (0,T ;V) L (0,T ;V)
1/4
≤ cw 6/5
7/4
ŵ 6 u L 6 (0,T ;V)
L (0,T ;V∗ ) L (0,T ;V)
1 7/3
≤ w L 6/5 (0,T ;V∗ ) + K ŵ 6 u L 6 (0,T ;V) .
4 L (0,T ;V)
(10.44)
Applying Hölder’s inequality with exponents 8, 8/3, 2 and Young’s inequality with
exponents 4, 4/3, in the same way we have
T T

ek1 t ŵ θ̂ , ∇φ dt ≤ c ŵL4 θ̂ L4 ∇φL2 dt
0 0
T
1/4 1/4 3/4 3/4
≤ cŵC([0,T ];H ) θ̂ C([0,T ];L 2 ) ∇ ŵL2 ∇ θ̂ L2 φWΓ1,2 dt
D
0
1/8 1/8 1/8 1/8 3/4
≤ cŵ L 6/5 (0,T ;V∗ ) ŵ L 6 (0,T ;V) θ̂ L 2 (0,T ;(W 1,2 )∗ ) θ̂ L 2 (0,T ;W 1,2 ) ŵ L 6 (0,T ;V)
Γ D Γ D
3/4
× θ̂ L 2 (0,T ;W 1,2 ) φ L 2 (0,T ;WΓ1,2 )
Γ D D
≤ cŵ L 6/5 (0,T ;V∗ ) ŵ L 6 (0,T ;V) θ̂ L 2 (0,T ;(W 1,2 )∗ ) θ̂ L 2 (0,T ;W 1,2 ) φ L 2 (0,T ;WΓ1,2 )
1/8 7/8 1/8 7/8
ΓD ΓD D
δw L 6/5 (0,T ;V∗ ) δθ L 2 (0,T ;(W 1,2 )∗ )

1/2 7/6 1/2 7/6
≤ + K δ ŵ L 6 (0,T ;V) + K δ θ̂ L 2 (0,T ;W 1,2 )
Γ D Γ D
× φ L 2 (0,T ;WΓ1,2 )

D
1 2
δw L 6/5 (0,T ;V∗ ) + K δ ŵ L 6 (0,T ;V)
1/2 7/6
≤
2
2
+ δθ L 2 (0,T ;(W 1,2 )∗ ) + K δ θ̂ L 2 (0,T ;W 1,2 )
1/2 7/6
· φ L 2 (0,T ;WΓ1,2 )
ΓD ΓD D

≤ δ 2 w L 6/5 (0,T ;V∗ ) + K δ2 ŵ L 6 (0,T ;V) + δ 2 θ L 2 (0,T ;(WΓ1,2 )∗ ) + K δ2 θ̂ L 2 (0,T ;W 1,2 )
7/3 7/3
D ΓD
× φ L 2 (0,T ;WΓ1,2 ) .

D
(10.45)
Also,
T
α0 θ̂
c
e−k1 t f, u dt ≤ √ u L 2 (0,T ;H ) ,
0 1 + εθ̂ 2 ε
T
α1 θ̂ c
e2k1 t f · ŵ, φ dt ≤ √ (K + w L 2 (0,T ;H ) )φ L 2 (0,T ;WΓ1,2 ) .
0 1 + εθ̂ 2 ε D
(10.46)
Owing to (10.39)–(10.41), (10.44), (10.35), (10.46) and (10.45) with δ = 21 , we
have from (10.24)

Aε (w, θ )V ∗ = sup Aε (w, θ ), (u, φ)
(u,φ)V =1
K 1
≤c K+ + w L 2 (0,T ;V) + w L 2 (0,T ;V) + w6 6 + θ 2 1, p
ε ε L (0,T ;V) L (0,T ;WΓ )
D
7/3 7/3 1 7/3
+ K ŵ + K ŵ 6 + θ L 2 (0,T ;L 2 (Ω)) + K θ̂ 2
L 6 (0,T ;V) L (0,T ;V) ε L (0,T ;WΓ1,2 )
D
1
+ w L 6/5 (0,T ;V∗ ) + θ 2
2 L (0,T ;(WΓ1,2 )∗ )
D
K 1 1
≤c K+ + w L 2 (0,T ;V) + θ L 2 (0,T ;L 2 (Ω)) + w6 6 + θ 2 1, p
ε ε ε L (0,T ;V) L (0,T ;WΓ )
D
7/3 7/3 1
+ w + θ 2 + L (w, θ )V ∗ ,
L 6 (0,T ;V) L (0,T ;WΓ1,2 ) 2
D
(10.47)
which shows the property (ii) for Aε (w, θ ).

(iii) Let us prove property (i) of Theorem 1.47 for Aε (w, θ ).
Let {(wk , θk )} ⊂ D(L ) be a sequence such that
(wk , θk ) (w, θ ) ∈ D(L ) in V ,

L (wk , θk ) L (w, θ ) in V ∗ , (10.48)
lim supAε (wk , θk ), (wk , θk ) − (w, θ ) ≤ 0.
k→∞
By taking a subsequence if necessary, we may assume that
wk → w in L 6 (0, T ; Ls (Ω)) (1 ≤ s < 6) and a.e. in Q

(10.49)
θk → θ in L 2 (0, T ; L s (Ω)) (1 ≤ s < 6) and a.e. in Q
when k → ∞. Then, by (10.24) we have


c min{μ0 , κ0 } wk − w2V + θk − θ dt
Q WΓ1,2
D

= c min{μ0 , κ0 } ŵk − ŵ2V + θ̂k − θ̂ dt
Q WΓ1,2
D
T
≤ 2 [μ(θ̂k ) + e4k1 t εŵk 4V ]E(ŵk ), E(wk − w) dt
0
T T

− 2 μ(θ̂k )E(ŵ), E(wk − w) dt − 2e4k1 t εŵk 4V E(ŵk ), E(wk − w) dt
0 0
T T

+ κ(θ̂k )∇ θ̂k , ∇θk − ∇θ dt − κ(θ̂k )∇ θ̂, ∇θk − ∇θ dt
0 0
T

≤ Aε (ŵk , θ̂k ), (wk − w, θk − θ ) − 2 μ(θ̂k )E(ŵ), E(wk − w) dt
0
T T
+ 2e4k1 t ε ŵk 4V E(ŵ), E(w − wk ) dt − ek1 t rotŵk × ŵk , wk − w dt
0 0
T T
α0 θ̂k
− e−k1 t ∇Φε (ek1 t ŵk (t)), wk − w dt − e−k1 t f, wk − w dt
0 0 1 + ε θ̂k2
T T

− κ(θ̂k )∇ θ̂ , ∇θk − ∇θ dt − (k1 (ŵk , wk − w) + 2(μ(θ̂k )k(x)ŵk , wk − w)Γ2
0 0
+ 2(μ(θ̂k )S ŵ˜ k , w̃k − w̃)Γ3 + 2(α(x)ŵk , wk − w)Γ5
+ (μ(θ̂k )k(x)ŵk , wk − w)Γ7 + (β(x)θ̂k , θk − θ )Γ R − ek1 t ŵk θ̂k , ∇(θk − θ )
|E(ŵk )|2 α1 θ̂k
− α2 μ(θ̂k ) −2k t , θk − θ − e2k1 t f · ŵk , θk − θ dt,
e 1 + ε|E(ŵk )|2 1 + ε θ̂k2
(10.50)
where k1 is the one in (10.37) and c is a positive constant.

Now let us estimate all terms except the first one on the right hand side of (10.50).
T
lim sup 2 e4k1 t εŵk 4V E(ŵk ), E(w − wk ) dt
k→∞ 0
T
= lim sup 2 e4k1 t εŵk 4V E(ŵk ), E(ŵ − ŵk ) dt
k→∞ 0
T
≤ 2e4k1 t ε lim sup ŵk 4V · lim sup E(ŵk ), E(ŵ − ŵk ) dt (10.51)
0 k→∞ k→∞
T
≤ 2e4k1 t ε lim sup ŵk 4V · E(ŵ)2 − lim inf E(ŵk )2 dt
0 k→∞ k→∞
T
≤ 2e4k1 t ε lim sup ŵk 4V · E(ŵ)2 − E(ŵ)2 dt = 0.
0 k→∞
Taking into account Corollary 1.1, by (10.48) we know

T
2 μ(θ̂k )E(ŵ), E(wk − w) dt → 0,
0
(10.52)
T
κ(θ̂k )∇ θ̂ , ∇θk − ∇θ dt → 0.
0
Also, by (10.48) and (10.49) we have that

T
ek1 t rotŵk × ŵk , wk − w dt ≤ cŵk L4 (Q) ∇ ŵk L2 (Q) wk − wL4 (Q) → 0
0
(10.53)
as k → ∞. Since ∇Φε is monotone, we have
T
− e−k1 t ∇Φε (ek1 t ŵk (t)), wk − w dt
0
T
=− e−2k1 t ∇Φε (ek1 t ŵk (t)), ek1 t ŵk − ek1 t ŵ dt
0
T
=− e−2k1 t ∇Φε (ek1 t ŵk (t)) − ∇Φε (ek1 t ŵ), ek1 t ŵk − ek1 t ŵ dt
0
T
− e−2k1 t ∇Φε (ek1 t ŵ), ek1 t ŵk − ek1 t ŵ dt
0
T
≤− e−2k1 t ∇Φε (ek1 t ŵ), ek1 t wk − ek1 t w dt → 0 as k → ∞.
0
(10.54)
By (10.49) we have that
T
α0 θ̂k
e−k1 t f, wk − w dt → 0,
0 1 + εθ̂k2
(10.55)
T
α1 θ̂k
e 2k1 t
f · ŵk , θk − θ dt → 0
0 1 + εθ̂k2
as k → ∞. By taking a sequence if necessary, we get from (10.48)
wk → w in L 6 (0, T ; Wα,2 (Ω)) (2/3 < α < 1),

θk → θ in L 2 (0, T ; W α,2 (Ω)) (2/3 < α < 1).
Then, by the trace theorem and Remark 3.4 we have

T
− (k1 (ŵk , wk − w) + 2(μ(θ̂k )k(x)ŵk , wk − w)Γ2 + 2(μ(θ̂k )S ŵ˜ k , w̃k − w̃)Γ3
0
+ 2(α(x)ŵk , wk − w)Γ5 + (μ(θ̂k )k(x)ŵk , wk − w)Γ7 + (β(x)θ̂k , θk − θ )Γ R
|E(ŵk )|2
− α2 μ(θ̂k ) −2k t , θk − θ dt → 0 as k → ∞.
e 1 + ε|E(ŵk )|2
(10.56)
Now let us prove that by taking subsequences there holds
T T
e ŵk θ̂k , ∇(θk − θ ) dt =
k1 t
ek1 t (ŵk − ŵ)θ̂k , ∇(θk − θ ) dt
0 0
T T
+ e ŵ(θ̂k − θ̂ ), ∇(θk − θ ) dt +
k1 t
ek1 t ŵ θ̂, ∇(θk − θ ) dt → 0
0 0
as k → ∞.
(10.57)
Since {ŵk } is bounded in L 6 (0, T ; V), taking into account (10.48) and applying The-
orem 1.39, we have that the set is relatively compact in the space L 5 (0, T H9/10 (Ω))
⊂ L5 (Q), where imbedding H9/10 (Ω) → L5 (Ω) was used.
Since {θk } is bounded in L 2 (0, T ; W 1,2 (Ω)) ∩ L ∞ (0, T ; L 2 (Ω)), by Theorem 1.33
and (1.22) it is bounded in
L 10/3 (0, T ; W 3/5,2 (Ω)) ⊂ L 10/3 (Q)
as well and thus by Theorem 1.39 the set {θk } is also relatively compact in L 3 (Q).
Consequently, there exist the subsequences such that
ŵk → ŵ in L5 (Q),
{θk } is bounded in L 10/3 (Q),
θk → θ in L 3 (Q).
Applying Hölder’s inequalities with exponents 5, 10/3, 2 for the first term and with
exponents 6, 3, 2 for the second term of the right hand side of (10.57), we get con-
vergence to zero of the two terms. The convergence to zero of the last term of the
right hand side of (10.57) is obvious.
Taking into account (10.51)–(10.57) and Assumption (10.48), we have from
(10.50)

c min{μ0 , κ0 } lim sup wk − w2V + θk − θ WΓ1,2 dt
k→∞ Q D

≤ lim sup Aε (ŵk , θ̂k ), (wk − w, θk − θ ) ≤ 0,
k→∞
which implies that

wk → w in L 2 (0, T ; V),
θk → θ in L 2 (0, T ; WΓ1,2 ),
D
(10.58)
∇wk → ∇w a.e. in Q,
∇θk → ∇θ a.e. in Q.
When (u, φ) ∈ V , by definition of A

Aε (wk , θk ), (wk − u, θk − φ)
T
= 2 [μ(θ̂k ) + e4k1 t εŵ4V E(ŵk ), E(wk − u) + ek1 t rotŵk × ŵk , wk − u
0
+ k1 (ŵk , wk − u) + 2(μ(θ̂k )k(x)ŵk , wk − u)Γ2 + 2(μ(θ̂k )S ŵ˜ k , w̃k − ũ)Γ3

+ 2(α(x)ŵk , wk − u)Γ5 + (μ(θ̂k )k(x)ŵk , wk − u)Γ7
α0 θ̂k
+ e−k1 t ∇Φε (ek1 t ŵk (t)), wk − u − e−k1 t f, wk − u
1 + εθ̂k2
+ (κ(θ̂k )∇ θ̂k , ∇(θk − φ)) + (β(x)θ̂k , θk − φ)Γ R − ek1 t ŵk θ̂k , ∇(θk − φ)
|E(ŵk )|2 α1 θ̂k
− α2 μ(θ̂k ) −2k t θk − φ − e 2k1 t
f · ŵ k , θk − φ dt.
e 1 + ε|E(ŵk )|2 1 + εθ̂k2
(10.59)
By Corollary 1.1 and (10.58), we have

T T
lim 2 μ(θ̂k )E(ŵk ), E(wk − u) dt = 2 μ(θ̂ )E(ŵ), E(w − u) dt.
k→∞ 0 0
(10.60)
The sequence {e4k1 t ŵk 4V εi j (ŵk )} is bounded in L 6/5 (0, T ; L 2 (Ω)) and εi j (u) ∈
L 6 (0, T ; L 2 (Ω)), and by taking a subsequence and denoting with the same subindex,
we see that
T T
lim 2e4k1 t ŵk 4V E(ŵk ), E(û) dt → 2e4k1 t ŵ4V E(ŵ), E(û) dt.
k→∞ 0 0
(10.61)
Since lim inf k→∞ ŵk 4V E(ŵk ), E(ŵk ) ≥ ŵ4V E(ŵ), E(ŵ) (see Exercises 2,
(c), p. 173 of [3]), by (10.60) and (10.61) we have
T
lim inf 2 [μ(θ̂k )+e4k1 t εŵk 4V E(ŵk ), E(wk − u) dt
k→∞ 0
T
≥ 2 [μ(θ̂) + e4k1 t εŵ4V ]E(ŵ), E(w − u) dt.
0
(10.62)
Since ek1 t rotŵk × ŵk , wk − u = ek1 t rotŵk × ŵk , −v0 − u, by (10.58) we can see
that
T T
e rotŵk × ŵk , wk − u dt →
k1 t
ek1 t rotŵ × ŵ, w − u dt. (10.63)
0 0
Also, by (10.58) we can see that

T T
− e−k1 t ∇Φε (ek1 t ŵk (t)), wk − u dt → − e−k1 t ∇Φε (ek1 t ŵ(t)), w − u dt. (10.64)
0 0
Since
α0 θ̂k α0 θ̂ α0 θ̂k
→ a.e. in Q and is bounded in L 2 (0, T ; L 2 ),
1+ εθ̂k2 1+ εθ̂ 2 1 + εθ̂k2
we have
α0 θ̂k α0 θ̂
in L 2 (0, T ; L 2 ) (10.65)
1+ εθ̂k2 1 + εθ̂ 2
(see Lemma 1.2). Owing to (10.58) and (10.65), we have

T
α0 θ̂k
e−k1 t f, wk − u dt
0 1 + εθ̂k2
T T
−k1 t α0 θ̂k α0 θ̂k
= e f, wk − w dt + e−k1 t f, w − u dt
0 1 + εθ̂k 2
0 1 + εθ̂k2
T
α0 θ̂
→ e−k1 t f, w − u dt.
0 1 + εθ̂ 2
(10.66)
Using (10.58) and (10.65), we have
T T
α1 θ̂k α1 θ̂k
e2k1 t f · ŵk , θk − φ dt = e2k1 t f · ŵk , θk − θ dt
0 1 + ε θ̂k2 0 1 + ε θ̂k2
T T
α1 θ̂k α1 θ̂k
+ e2k1 t 2
f · ŵk − w, θ dt + e2k1 t f · ŵ, θ − φ dt (10.67)
0 1 + ε θ̂k 0 1 + ε θ̂k2
T
α1 θ̂
→ e2k1 t f · ŵ, θ − φ dt.
0 1 + ε θ̂ 2
By Corollary 1.1 and (10.58)
(κ(θ̂k )∇ θ̂k , ∇(θk − φ)) → (κ(θ̂)∇ θ̂ , ∇(θ − φ)). (10.68)
It is easy to prove convergence of other terms in the right hand side of (10.59).
Thus, by (10.60)–(10.64) and (10.66)–(10.68), we have the existence of a subse-
quence {(vk , θk )} such that

lim inf Aε (wk , θk ), (wk − u, θk − φ) ≥ Aε (w, θ ), (w − u, θ − φ) , (10.69)
k→∞
by which property (i) of Theorem 1.47 for Aε (w, θ ) is proved.
Therefore, by Theorem 1.47 there exists a solution to Problem VEA’, which gives
us the conclusion of Theorem 10.2.
10.2.2 Estimates of Solutions to the Approximate Problem
Let 0 < ε < 1. For v0 ∈ HK , θ0 ∈ L 1 (Ω) and f 2 ∈ L 1 (0, T ; (WΓ1,2 )∗ ), there exists
1,6 1,2 ∗
D
v0ε ∈ W (Ω) ∩ K (Ω), θ0ε ∈ WΓ D (Ω) and f 2ε ∈ L 0, T ; (WΓ D ) such that

1,6 2
v0 − v0ε H ≤ ε, θ0 − θ0ε L 1 (Ω) ≤ ε, f 2 − f 2ε L 1 (0,T ;(WΓ1,2 )∗ ) ≤ ε.

D
(10.70)
Let (vε (t), θε (t)) be a solution to Problem VEA with the initial function (v0ε , θ0ε )
and f 2ε ∈ L 2 (0, T ; (WΓ1,2
D
)∗ ) instead of f 2 .
Define Aε (θ ) : V → V∗ by

Aε (θ )v, w = A(θ )v, w + 2εv4V E(v), E(w) ∀v, w ∈ V,
where A(θ ) is the one in (10.16). Then, we get from (10.22)
∂vε (t)
, u(t) + Aε (θε )vε (t) + B(vε (t), vε (t)), u(t) + ∇Φε (vε (t)), u(t)
∂t
α0 θ
=− f, u(t) + f + f 1 , u(t) ∀u ∈ L 6 (0, T ; V).
1 + εθ 2
(10.71)
By (10.28) there exists c∗ such that

ν(k(x)z, z)Γ + 2ν(S z̃, z̃)Γ + (α(x)z, z)Γ + ν(k(x)z, z)Γ
2 3 5 7
(10.72)
≤ μ0 z2V + c∗ z2 ∀z ∈ V
(see Theorem 1.27).

Remark 10.3 Note that c∗ depends on the shape of Γi , i = 2, 3, 7, and the norm of
matrix α on Γ5 , and so for the fixed domain Ω it depends only on the norm of α.
By (10.5), (10.7), (10.15) and (10.72), we have
Aε (θε )z, z ≥ μ0 z2V + 2εz6V − c∗ z2 ∀z ∈ V,

(10.73)
|Aε (θε )z, w| ≤ c2 zV + εz5V wV ∀z, w ∈ V.
Putting u = vε in (10.71) and taking into account (10.73), we have from (10.71)
t
1 t
vε (t)2 + μ0 vε 2V + 2εvε 6V ds + ∇Φε (vε (s)), vε (s) ds
2 0 0
t t
1
≤ v0ε 2 + c∗ vε 2 ds + c ( f 2∞ + f 1 2V∗ ) ds (10.74)
2 0 0
t
+ |α0 | f ∞ |θε ||vε | d xds.
0 Ω
Φε (u) − Φε (vε (t)) ≥ ∇Φε (vε (t)), u − vε (t) ∀u ∈ V, (10.75)
and so by Φε (0V ) = 0
0 ≤ Φε (vε (t)) ≤ ∇Φε (vε (t)), vε (t). (10.76)

t t
vε (t) + 2
2
μ0 vε 2V + 2εvε 6V ds + 2Φε (vε (t)) ds
0 0
t
≤ v0ε + 2c f ∞ T + 2c f 1 L 2 (0,T ;V∗ ) + 2c∗
2 2 2
vε (s)2 ds
0
t
+ |α0 | f ∞ |θε ||vε | d xds ∀t ∈ [0, T ].
0 Ω
(10.77)
For fixed 0 < δ < 1, define
1
Φ(ξ ) := 1 − sign ξ, ξ ∈ R,
(1 + |ξ |)δ
ξ
Φ(τ ) dτ for ξ ≥ 0,
Ψ (ξ ) := 0 0
− ξ Φ(τ ) dτ for ξ < 0.
Then
δ
Φ (ξ ) = ∀ξ ∈ R,
(1 + |ξ |)1+δ
|Φ(ξ )| ≤ 1, 0 ≤ Φ (ξ ) ≤ δ ∀ξ ∈ R,
Ψ (ξ ) = Φ(ξ ),
(10.78)
1
Ψ (ξ ) = |ξ | + 1 − (1 + |ξ |)1−δ ∀ξ ∈ R,
1−δ
|ξ | 2(1−δ)/δ
− ≤ Ψ (ξ ) ≤ |ξ | ∀ξ ∈ R.
2 1−δ
From (10.22) we get
∂θε |E(vε )|2

, ϕ +(κ(θε )∇θε , ∇ϕ) + (β(x)θε , ϕ)Γ R − vε θε , ∇ϕ − α2 μ(θε ) ,ϕ
∂t 1 + ε|E(vε )|2
(10.79)
α1 θε
= f · vε , ϕ + f 2ε , ϕ ∀ϕ ∈ L 2 (0, T ; WΓ1,2 (Ω)).
1 + εθε2 D
Since θε ∈ L 2 (0, T ; WΓ1,2

D
(Ω)) ∩ C([0, T ]; L 2 (Ω)), we have
Φ(θε (t)) ∈ WΓ1,2

D
(Ω), Ψ (θε (t)) ∈ WΓ1,2
D
(Ω) for a.e. t ∈ [0, T ],
vε θε , ∇Φ(θε (t)) = −(vε · ∇θε ), Φ(θε (t))
(10.80)
= −vε , ∇Ψ (θε (t))
= div vε , Ψ (θε (t)) = 0,
where v · n|Γ R = 0 (see (10.3)) and Ψ (θε ) = 0 on Γ D were used.

Also
t
∂θε
, Φ(θε ) ds = Ψ (θε (t)) d x − Ψ (θ0ε ) d x ∀t ∈ [0, T ],
∂t
0
Ω Ω
(10.81)
(κ(θε )∇θε , ∇Φ(θε )) = κ(θε )|∇θε |2 Φ (θε ) d x,
Ω
(β(x)θε , Φ(θε ))Γ R ≥ 0.
Putting ϕ(t) = Φ(θε (t)) in (10.79) and using (10.80) and (10.81), we have
t t
Ψ (θε (t)) d x + κ(θε )|∇θε |2 Φ (θε ) d x + (β(x)θε , Φ(θε ))Γ R ds
Ω 0 Ω 0
t
|E(vε )|2
= Ψ (θ0ε ) d x + α2 μ(θε ) , Φ(θε ) ds
Ω 0 1 + ε|E(vε )|2
t t
α1 θε
+ f · v ε , Φ(θ ε ) ds + f 2ε , Φ(θε ) ds.
0 1 + εθε
2
0
(10.82)
By virtue of (10.78) and (10.81), we have from (10.82)
t
1 2(1−δ)/δ |∇θε |2
θε (t) L 1 (Ω) − meas Ω + δ κ0 d xds
2 1−δ 0 Ω (1 + |θε |)1+δ
t
≤ θ0ε L 1 (Ω) + α2 μ1 vε 2V ds
0
t t
+ |α1 | f ∞ |θε ||vε | d xds + c f 2ε (WΓ1,2 )∗ ds.
0 Ω 0 D
(10.83)
μ0
Multiplying (10.83) by α2 μ1
and adding the resulting formula to (10.77), we obtain
t

vε (t)2 + θε (t) L 1 (Ω) + μ0 vε 2V + 2εvε 6V ds
0
t t
|∇θε |2
+ Φε (vε (t)) ds + δ d xdt
Ω (1 + |θε |)
1+δ
0 0
t t
≤ cΛδ + c∗ vε (s)2 ds + c1 (|α0 | + |α1 |) f ∞ |θε ||vε | d xds),
0 0 Ω
0 < δ < 1,
(10.84)
where
Λδ = cδ + v0 2 + θ0 L 1 (Ω) + f 2∞ T + f 1 2L 2 (0,T ;V∗ ) + f 2 L 1 (0,T ;(WΓ1,2 )∗ ) .

D
(10.85)
By Hölder’s inequality and Young’s inequality,
t
c1 (|α0 | + |α1 |) f ∞ |θε ||vε | d xds
0 Ω
t 1/2
t 1/2
≤ c1 (|α0 | + |α1 |) f ∞ θε 2L 6/5 ds vε 2V ds
0 0
t
μ0 t
≤ c1 (μ0 )(|α0 | + |α1 |) f ∞ θε L 6/5 ds +
2
vε 2V ds.
0 2 0
(10.86)
By Theorem 1.12 with θ = 13 18
, we have
13/18 5/18
θε L 6/5 ≤ θε L 1 θε L 5/2 . (10.87)
Taking into account (10.86) and (10.87), we deduce from (10.84)

t
μ0
vε (t)2 + θε (t) L 1 (Ω) + vε 2V + 2εvε 6V ds
0 2
t t
|∇θε |2
+ Φε (vε (t)) ds + δ d xdt (10.88)
0 Ω (1 + |θε |)
1+δ
0
t t
13/9 5/9
≤ cΛδ + c∗ vε (s)2 ds + c1 (μ0 )(|α0 | + |α1 |) f ∞ θε (s) 1 θε (s) 5/2 ds.
0 0 L L
For 0 < δ < 1 define

|θε |
ηε := a.e. in Q.
(1 + |θε |)(1+δ)/2
Then,
|∇θε |
(1 + |θε |)(1−δ)/2 ≤ 1 + ηε , |∇ηε | ≤ a.e. in Q. (10.89)
(1 + |θε |)(1+δ)/2
Taking δ = 16 , we have from the first inequality of (10.89)
5/6
θε (s) L 5/2 ≤ c6 (1 + ηε 2L 6 ).
Thus, we have from (10.88)

t
μ0
0 2
t
1 t |∇θε |2
+ Φε (vε (t)) ds + d xdt
0 6 0 Ω (1 + |θε |)7/6
t (10.90)
≤ cΛ1/6 + c∗ vε (s)2 ds
0
t
13/9
+ c1 (μ0 )(|α0 | + |α1 |) f ∞ θε (s) L 1 (1 + ηε 2L 6 )2/3 ds.
0
By Gronwoll’s inequality, we have from (10.90)

t
13/9
vε (t)2 ≤ cΛ1/6 + c1 (|α0 | + |α1 |) f ∞ θε (s) (1 + ηε 2 6 )2/3 ds ec∗ t . (10.91)
0 L1 L
It follows from (10.90) and (10.91) that

t
μ0
0 2
t
1 t |∇θε |2
+ Φε (vε (t)) ds + d xdt
0 6 0 Ω (1 + |θε |)7/6
≤ cΛ1/6 (1 + c∗ ec∗ T T )
t
13/9
+ (1 + c∗ ec∗ T T )c1 (|α0 | + |α1 |) f ∞ θε (s) L 1 (1 + ηε (s)2L 6 )2/3 ds.
0
(10.92)
Put
φ(t) := θε C([0,t];L 1 ) + ηε 2L 2 (0,t;L 6 ) .
Taking into account (10.89) and using Hölder’s inequality with exponents 3/2, 3, we
have from (10.92)
t
|∇θε |2
φ(t) ≤ θε C([0,t];L 1 ) + c d xds
Ω (1 + |θε |)
7/6
0
≤ cΛ1/6 (1 + c∗ ec∗ T T )
13/9 2/3
+ (1 + c∗ ec∗ T T )c1 (|α0 | + |α1 |) f ∞ T 1/3 θε (s) t + ηε 2 2 ,
C[0,t];L 1 ) L (0,t;L 6 )
and so
19/9
t + φ(t) ≤ K 0 + (1 + c∗ ec∗ T T )c5 (|α0 | + |α1 |) f ∞ T 1/3 t + φ(t) ,
(10.93)
where K 0 is cΛ1/6 (1 + c∗ ec∗ T T ) + T . By (10.85), without loss of generality, we
may assume that φ(0) ≤ K 0 . If
1
(1 + c∗ ec∗ T T )c(|α0 | + |α1 |) f ∞ T 1/3 ≤ (2K 0 )(1−19/9) , (10.94)
3
then by Lemma 1 of Appendix of [4] we have from (10.93)
φ(t) ≤ 2cΛ1/6 (1 + c∗ ec∗ T T ) + T ∀t ∈ [0, T ]. (10.95)
By virtue of (10.92) and (10.95), using (10.94) we have

t t
μ0 |∇θε |2
vε (t)2 + θε (t) L 1 (Ω) + vε 2V + 2εvε 6V ds+ d xdt
0 Ω (1 + |θε |)
2 1+δ (10.96)
0
≤ C(Λ, c∗ )
and T
Φε (vε (t)) dt ≤ C(Λ, c∗ ), (10.97)
0
where C(Λ, c∗ ) is independent of ε, Φ. Since Φ is nonnegative, by (1.36) and (10.97)

we have T
vε (t) − Jε (vε (t))2V dt ≤ c7 ε (10.98)
0
of ε.
with c7 independent
Taking u(t) = vε (t) − vε (s) , s ∈ (0, T ), in (10.71), we then have
1 dvε (t) − vε (s)2

2 dt
α0 θε
+ Aε (θε )vε (t) + B(vε (t), vε (t)) + f − f − f 1 , vε (t) − vε (s)
1 + εθε2
= ∇Φε (vε (t)), vε (s) − vε (t) ≤ Φε (vε (s)) − Φε (vε (t)) ≤ Φε (vε (s)).
(10.99)
Then by virtue of (10.73), we have from (10.99)
1 dvε (t) − vε (s)2

≤Φε (vε (s)) + Aε vε (t), vε (s) + B(vε (t), vε (t)), vε (s)
2 dt
α0 θε
+ − f + f + f 1 , vε (t) − vε (s) + c∗ vε (t)2 ,
1 + εθε2
(10.100)
where B(vε (t), vε (t)), vε (t) = 0 was used.
Let us integrate each term of (10.100) respectively first with respect to t from s to
s + h and then with respect to s from 0 to T , where vε (t) = 0 when t ∈ (T, T + h).

T s+h
dvε (t) − vε (s)2 T
dtds = vε (s + h) − vε (s)2 ds. (10.101)
0 s dt 0
By (10.97)
T s+h T
Φε (vε (s)) dtds ≤ h Φε (vε (s)) ds ≤ c6 h. (10.102)
0 s 0
By (10.96) and (10.73), we have
T s+h
Aε (θ )vε (t), vε (s) dtds
0 s
T s+h

≤c vε (s)V vε (t)V + εvε (t)5V dt ds
0 s
T √ T
≤c vε (s)V ( hvε L 2 (0,T ;V) ) ds + c vε (s)V εh 1/6 vε 5 6 ds
0 0 L (0,T ;V)
≤ c8 h 1/6 ,
(10.103)
1 1
where c8 is independent of ε. Since wL3 ≤ K wL2 2 wL2 6 ,
|B(v, w), z| = |rot v × w, z| ≤ K rot vL2 wL3 zL6

1 1 (10.104)
≤ K vV w 2 wV2 zV ,
and so by (10.96) we have

T s+h
B(vε (t), vε (t))vε (s) dtds
0 s
T s+h 3 1
≤K vε (t)V2 vε (t) 2 vε (s)V dtds
0 s
T s+h 3 s+h 1
4 4
≤K vε (s)V vε (t)2V dt vε (t)2 dt ds
0 s s
1
≤ c9 h 4 ,
(10.105)
where vε (t)2 ≤ C(Λ, c∗ ) (see (10.96)) was used. Also, by (10.96) we have
T s+h T t

( f + f 1 )(t), vε (t) dtds ≤ |( f + f 1 )(t), vε (t)| ds dt ≤ K h,
0 s 0 t−h
T s+h T s+h
(10.106)
( f + f 1 )(t), −vε (s) dtds ≤ K vε (s)V ( f + f 1 )(t)V∗ dtds
0 s 0 s
√
≤ c9 h.
In the same way we get

√
T s+h
α0 θε
f, vε (t) − vε (s) dtds ≤ K h + c9 h, (10.107)
0 s 1 + εθε 2
T s+h
c∗ vε (t)2 dtds ≤ K h. (10.108)
0 s
By virtue of (10.100)–(10.108), uniformly with respect to ε

T
1
vε (s + h) − vε (s)2 ds ≤ O(h 6 ), (10.109)
0
which implies that the set {vε } is relatively compact in L 2 (0, T ; W 10 ,2 (Ω)) (see
9
Theorem 1.38).
Therefore, by (10.96), there exists v and a subsequence {vεk } such that as εk → 0,
∗
vεk v in L ∞ (0, T ; H ),
vεk → v in L 2 (0, T ; W 10 ,2 (Ω)),
9
(10.110)
vεk v in L 2 (0, T ; V).
We will show that θε is relative compact in L r (0, T ; W α,r ) ∀r (1 < r < 5/4),
∀α (0 < α < 1). To this end, we need to get an estimate on ∇θε . Let 1 ≤ r < 3/2. By
1 1
(10.96), inequalities |a + b| p ≤ 2 p (|a| p + |b| p ), |a| + |b| ≤ (|a| p + |b| p ) p , p ∈
(1, ∞), and Hölder’s inequality with exponents r , 2−r , we have
2 2
r/2 (2−r )/2

|∇θε |2 r (1+δ)/(2−r ) d xdt
|∇θε |r d xdt ≤ d xdt 1 + |θ ε |)
Q (1 + |θε |)
1+δ
Q Q

(2−r )/2
≤ 2r (1+δ)/2 C(Λ, c∗ )r/2 (mes Q)(2−r )/2 + |θε |r (1+δ)/(2−r ) d xdt ,
Q
(10.111)
where C(Λ, c∗ ) is the one in (10.96).

To use the property W 1,r (Ω) ⊂ L q (Ω), let us take q such that 1
r
− 1
3
= q1 , i.e.
q = 3−r
3r
. Let us take δ0 > 0 such that
1 < r (1 + δ0 )/(2 − r ) < q,

which holds if 0 < δ0 < (3 − 2r )/(3 − r ). Set s = r (1 + δ0 )/(2 − r ) and take λ

such that
1 λ 1−λ
= + .
s 1 q
Then, λ = q−s
s(q−1)
, 0 < λ < 1 and by Theorem 1.12
η L s ≤ cηλL 1 η1−λ λ

L q ≤ cη L 1 ∇ηLr
1−λ
∀η ∈ WΓ1.rD . (10.112)
Therefore, by (10.96) and (10.112) we have

T (1−λ)s/r
|θε |r (1+δ0 )/(2−r ) d xdt ≤ cC(Λ, c∗ )λs |∇θε |r d x dt.
Q 0 Ω
We will fix r and δ0 so that
5 5 − 4r
1≤r < , 0 < δ0 < ,
4 3
which ensure that 0 < δ0 < (3 − 2r )/(3 − r ).
r (1+δ0 )
q−s
Taking into account λ = s(q−1) , q = 3−r
3r
and s = 2−r
, we have

1 q(s − 1) 3(2r − 2 + r δ0 ) 3 2r − 2 + r 5−4r
(1 − λ)s/r = = < 3
= 1.
r q −1 (4r − 3)(2 − r ) (4r − 3)(2 − r )
Then, by Hölder’s inequality we have

T (1−λ)s/r
|θε |r (1+δ0 )/(2−r ) d xdt ≤ cC(Λ, c∗ )λs |∇θε |r d x dt ,
Q 0 Ω
(2−r )/2 (1−λ)s(2−r )/2r

|θε |r (1+δ0 )/(2−r ) d xdt ≤ cC(Λ, c∗ )λs(2−r )/2 |∇θε |r d xdt ,
Q Q
(10.113)
where (1−λ)s(2−r
2r
)
= 21 (1 − λ)(1 + δ0 ) < 1. Then, it follows from Young’s inequality
and (10.113) that
(2−r )/2
1
2r (1+δ)/2 C(Λ, c∗ )r/2 |θε |r (1+δ0 )/(2−r ) d xdt ≤ |∇θε |r d xdt + R(C(Λ, c∗ )),
Q 2 Q
(10.114)
where R(·) is a nonnegative continuous function on [0, ∞). Substituting (10.114)

into (10.111) with δ = δ0 , we have

5 5 − 4r
|∇θε |r d xdt ≤ cC(Λ, c∗ )r/2 + 2R(C(Λ, c∗ )), 1 < r < , 0 < δ0 < . (10.115)
Q 4 3
Taking r = 6/5, we have from (10.115)

6
θε L5 6/5 (0,T ;L 2 ) ≤ cC(Λ, c∗ )r/2 + 2R(C(Λ, c∗ )). (10.116)
From (10.79) we will get an estimate on θε . Let 1 ≤ r < 5

4
and r := r/(r − 1) >
5. By Hölder’s inequality,
|(κ(θε )∇θε , ∇ϕ)| ≤ κ1 ∇θε Lr ∇ϕLr ,

(10.117)
|(β(x)θε , ϕ)Γ R | ≤ β1 θε L 1 (Γ R ) ϕ L ∞ (Γ R ) ≤ cθε WΓ1,r (Ω) ϕW 1,r (Ω) .
D ΓD
In the last inequality a trace theorem (see Theorem 1.25) was used. As in [5] let us
estimate vε θε , ∇ϕ. Let r satisfy 20
21
< r < 45 . Then there exists 43 < σ < q such
that
q(σ − 1) 5r
= ,
σ (q − 1) 8
where q = 3r
3−r
. Define
4σ q −σ
ρ := , λ := .
3σ − 4 σ (q − 1)
Then,
1 1 1 1 λ 1−λ 5r
+ + = 1, = + , 1−λ= .
4 σ ρ σ 1 q 8
Taking into account
θε L σ ≤ cθε λL 1 θε 1−λ λ

L q ≤ cθε L 1 ∇θε Lr ,
1−λ
1/4 3/4
vε L4 (Ω) ≤ cvε L2 (Ω) ∇vε L2 (Ω) ,
we have
|vε θε , ∇ϕ| ≤ cvε L4 θε Lσ ∇ϕLρ

(10.118)
≤ cvε L ∞ (0,T ;H ) θε λL ∞ (0,T ;L 1 ) ∇vε L2 (Ω) ∇θε 1−λ
1/4 3/4
Lr ϕW 1,ρ ,
ΓD
which is the desired one. Also,

α2 μ(θε ) |E(vε )|
2
,ϕ ≤ cμ1 vε 2 max |ϕ| ≤ cvε 2 ϕ 1,r
1 + ε|E(v )|2 V
x∈Ω
V WΓ
(10.119)
ε D

∀ϕ ∈ WΓ1,rD (Ω),
α1 θε
| f · vε , ϕ| ≤ cθε f ∞ vε max |ϕ| ≤ cθε f ∞ vε ϕW 1,r
1 + εθε2 x∈Ω ΓD

∀ϕ ∈ WΓ1,rD (Ω).
(10.120)
Let τ := max{r , ρ}. Then, by (10.117)–(10.120), we have from (10.79)
∂θ
ε
, ϕ ≤c ∇θε Lr + vε L ∞ (0,T ;H ) θε λ ∞
1/4 3/4
∇vε 2 ∇θε 1−λ
Lr
+ vε 2V
∂t L (0,T ;L 1 ) L (Ω)
+ θε f ∞ vε + f 2ε ϕ ∀ϕ ∈ WΓ1,τ (Ω).

(WΓ1,2 )∗ WΓ1,τ D
D D
Hence, taking into account (10.115) and (10.116), we see that

θε ∈ L 1 0, T ; (WΓ1,τ
D
)∗
and
θε L 1 0,T ;(W 1,τ )∗

ΓD
≤ c θε L 1 0,T ;W 1,r + vε L ∞ (0,T ;H ) θε λL ∞ (0,T ;L 1 ) vε L 2 (0,T ;V) θε L8r (0,T ;W 1,r )
5r
1/4 3/4
ΓD ΓD
+ vε 2L 2 (0,T ;V) + θε L 1 (0,T ;L 2 ) f ∞ vε L ∞ (0,T ;H ) + f 2ε L 1 (0,T ;(WΓ1,2 )∗ ) ,
D
(10.121)
21
where 20 < r < 45 . Therefore, θε L 1 0,T ;(W 1,τ )∗ is bounded.
ΓD
Thus, by (10.115) {θε } is relatively compact in L r (0, T ; W α,r ) ∀r (1 < r <
5/4), ∀α (0 < α < 1) (see Theorem 1.39). Therefore, there exists θ and a subse-
quence {θεk } such that as εk → 0
∗
θεk θ in L ∞ (0, T ; L 1 (Ω)),
θεk θ in L r (0, T ; W D1,r (Ω)) ∀r (1 < r < 5/4),
1 (10.122)
θεk → θ in L r (0, T ; W α,r (Ω)) ∀r (1 < r < 5/4), ∀α ( < α < 1)
r
and a.e. in Q.
10.2.3 Passing to the Limit
By passing to limit of solutions in Theorem 10.2 we will prove Theorem 10.1. Owing
to (10.110) and (10.122), we can extract subsequences, which are denoted as before,
such that as εk → 0
vεk v in L 2 (0, T ; V),

∗
vεk v in L ∞ (0, T ; H ),
vεk → v in L 2 (0, T ; W 10 ,2 (Ω)),
9
(10.123)
θεk θ in L r (0, T ; W D1,r (Ω)) ∀r (1 < r < 5/4),
1
θεk → θ in L r (0, T ; W α,r (Ω)) ∀r (1 < r < 5/4), ∀α <α<1
r
and a.e. in Q.
Subtracting (10.71) with u = vεk from (10.71), we have

T ∂vεk
+ Aεk (θεk )vεk (t) + B(vεk (t), vεk (t)) + ∇Φεk (vεk ), u(t) − vεk (t) dt
0 ∂t
T
α0 θεk
= − f + f + f 1 , u(t) − vεk (t) dt ∀u ∈ L 6 (0, T ; V).
0 1 + εk θε2k
(10.124)
Since
T
vε k (t), u(t) − vεk (t) dt
0
T
1 1
= u (t), u(t) − vεk (t) dt − vεk (T ) − u(T )2 + vεk (0) − u(0)2
0 2 2
∀u ∈ L 6 (0, T ; V) with u ∈ L 2 (0, T ; V∗ ),
we have from (10.124)

T

u (t) + Aεk (θεk )vεk (t) + B(vεk (t), vεk (t)) + ∇Φεk (vεk ), u(t) − vεk (t) dt
0
T
α0 θεk 1 (10.125)
− − f + f + f 1 , u(t) − vεk (t) dt ≥ − vεk (0) − u(0)2
0 1 + εθε2k 2
∀u ∈ L 6 (0, T ; V) with u ∈ L 2 (0, T ; V∗ ).

Taking into account (10.75) and εk |vεk 4V E(vεk ), E(vεk ) ≥ 0, we have from
(10.125)
T
u (t) + A(θεk )vεk (t) + B(vεk (t), vεk (t), u(t) − vεk (t) dt
0
T T
α0 θεk
+ εk |vεk 4V E(vεk ), E(u) dt − − f + f + f 1 , u(t) − vεk (t) dt
0 0 1 + εθε2k
T
1
+ Φεk (u(t)) − Φεk (vεk (t)) dt ≥ − vεk (0) − u(0)2 .
0 2
(10.126)
Since Φ(u) ≥ Φεk (u) and Φ(Jεk vεk (t)) ≤ Φεk (vεk (t)) (see (1.37)), we have from
(10.126)
T
u (t) + A(θεk )vεk (t) + B(vεk (t), vεk (t)), u(t) − vεk (t) dt
0
T T
α0 θεk
+ εk |vεk 4V E(vεk ), E(u) dt − − f + f + f 1 , u(t) − vεk (t) dt
0 0 1 + εθε2k
T
1
+ Ψ (u) − Φ(Jεk vεk (t)) dt ≥ − vεk (0) − u(0)2 .
0 2
(10.127)
Taking into account (10.123), by Corollary 1.1 we get as εk → 0

T T T
A(θεk )vεk (t), u(t) dt ≡ a0 (θεk ; vεk , u) dt → A(θ )v(t), u(t) dt.
0 0 0
(10.128)
Owing to (10.123), as εk → 0
vεk → v in L 2 (0, T ; L2 (∂Ω)).
Thus, by Lemma 1.3 we have
2(μ(θεk )k(x)vεk , vεk )Γ2 + 2(μ(θεk )S ṽεk , ṽεk )Γ3 + 2(α(x)vεk , vεk )Γ5 + (μ(θεk )k(x)vεk , vεk )Γ7
→ 2(μ(θ )k(x)v, v)Γ2 + 2(μ(θ )S ṽ, ṽ)Γ3 + 2(α(x)v, v)Γ5 + (μ(θ )k(x)v, v)Γ7 .
Therefore, taking into account
lim inf 2(μ(θεk )E(vεk ), E(vεk )) ≥ 2(μ(θ )E(v), E(v)),

εk →0
we have
lim inf A(θεk )vεk (t), vεk (t) dt ≥ A(θ )v(t), v(t) dt. (10.129)
εk →0
By (10.128) and (10.129), we have

T T
lim inf A(θεk )vεk (t), u(t) − vεk (t) dt ≤ A(θ )v(t), u(t) − v(t) dt.
εk →0 0 0
(10.130)
By (10.96) we have
T
T

εk vεk 4V E(vεk ), E(u) dt ≤ cεk 1/6
εk 5/6 vεk 5V u(t)V dt
0 0
T T
5/6
5 6/5
5/6 1/6
≤ cεk 1/6
εk vεk V dt u(t)6V dt
0 0
≤ cεk 1/6 C(Λ, c∗ )5/6 u L 6 (0,T ;V) → 0 as εk → 0.
(10.131)
By (10.98), we have that Jεk vεk v in L 2 (0, T ; V) as εk → 0. Since the func-
tional Φ : V → R is weakly lower semi-continuous, we have
T T
lim inf Φ(Jεk vεk (t)) dt ≥ Φ(v(t)) dt ≡ Ψ (v). (10.132)
εk →0 0 0

T T
B(vεk (t), vεk (t)), u(t) dt → B(v(t), v(t)), u(t) dt (10.133)
0 0
as εk → 0 (see (6.46)–(6.49)).
Let us prove that

T
α0 θεk T
f, u(t) − vεk (t) dt → α0 θ f, u(t) − v(t) dt → 0 as εk → 0.
0 1 + εk θε2k 0
(10.134)
By (10.123) and Lemma 1.5, we have
T
α0 θεk
− α0 θ f, u(t) dt
0 1 + εk θε2k
T T
α0 (θεk − θ ) 1
= f, u(t) dt + − 1 α0 θ f, u(t) dt → 0
0 1 + εk θε2k 0 1 + εk θε2k
(10.135)
and
T T
α0 θεk α0 (θεk − θ )
f, vε (t) − α 0 θ f, v(t) dt = f, vεk (t) dt
0 1 + εk θε2k k
0 1 + εk θε2k
T T (10.136)
α0 θ 1
+ f, vεk (t) − v(t) dt + − 1 α0 θ f, v(t) dt → 0.
0 1 + εk θεk 2
0 1 + εk θε2k
By (10.135) and (10.136), we get (10.134).

Therefore, by (10.130)–(10.133) and (10.134), we have from (10.127)
T
u (t) + A(θ )v(t) + B(v(t), v(t)) − (1 − α0 θ ) f − f 1 , u(t) − v(t) dt
0
1
+ Ψ (u) − Ψ (v) ≥ − v0 − u(0)2 .
2
(10.137)
3
Since B(v(t), v(t))V∗ ≤ K v(t)V2 v(t) 2 (see (10.104)) and v ∈ L ∞ (0, T ; L 2 ),
1
we have B(v, v) ∈ L 3 (0, T ; V∗ ). Therefore, by density of the set {L 6 (0, T ; V) :

4
u ∈ L 2 (0, T ; V∗ )} in {L 4 (0, T ; V) : u ∈ L 2 (0, T ; V∗ )}, (10.137) is valid for all

u ∈ {L 4 (0, T ; V) : u ∈ L 2 (0, T ; V∗ )}.
Thus, the first formula of (10.20) is valid.
For any ϕ ∈ C 1 ([0, T ]; CΓ1 D (Ω)) with ϕ(T ) = 0, we have from (10.79)
t t t
∂ϕ
θεk (t),ϕ(t) − θεk , ds + (κ(θεk )∇θεk , ∇ϕ) ds + (β(x)θεk , ϕ)Γ R ds
0 ∂t 0 0
t t
|E(vεk )|2
− vεk θεk , ∇ϕ ds − α2 μ(θεk ) , ϕ ds (10.138)
0 0 1 + ε|E(vεk )|2
t t
α1 θεk
= θεk (0), ϕ(0) + f · vεk , ϕ ds + f 2εk , ϕ ds ∀t ∈ [0, T ].
0 1 + εθεk
2
0
By Corollary 1.1 we have

t t
(κ(θεk )∇θεk , ∇ϕ) ds → (κ(θ )∇θ, ∇ϕ) ds as εk → 0. (10.139)
0 0
Since vεk → v in L 2 (0, T ; W 10 ,2 (Ω)) and W 10 ,2 (Ω) ⊂ L5 (Ω) (see Theorem 1.20),
9 9
we have
vεk → v in L 2 (0, T ; L5 (Ω)). (10.140)

Since {θεk } ⊂ L 6/5 (0, T ; W 1,6/5 (Ω)) ∩ L ∞ (0, T ; L 1 (Ω)) and W 1,6/5 (Ω) ⊂ L 2 (Ω),
7
by complex interpolation with exponent 12 we have
24 5
{θεk } ⊂ L 2 (0, T ; L 19 (Ω)) ⊂ L 2 (0, T ; L 4 (Ω)) (10.141)
(see Theorem 1.33). Taking into account (10.140) and (10.141), we have
t t
vεk θεk , ∇ϕ ds → vθ, ∇ϕ ds as εk → 0. (10.142)
0 0
Since
t t √ √
μ(θεk )|E(vεk )|2 |E(vεk )| ϕ |E(vεk )| ϕ
, ϕ ds = μ(θεk ) ! ,! ds,
0 1 + εk |E(vεk )| 1 + εk |E(vεk )|2 1 + εk |E(vεk )|2
2
0
by Lemma 1.4 and Corollary 1.2, we have


μ(θεk )|E(vεk )|2 t
lim inf , ϕ ds ≥ μ(θ )|E(v)|2 , ϕ ds. (10.143)
εk →0 Ω 1 + εk |E(vεk )|2 0
In the same way as in (10.136) we have

t
α1 θεk t
f · vεk , ϕ ds → α1 θ f · v, ϕ ds. (10.144)
0 1 + εk θε2k 0
Therefore taking into account (10.139), (10.123) and (10.142)–(10.144), we have

from (10.138)
t t
∂ϕ
θ (t), ϕ(t) − θ, ds + (κ(θ )∇θ, ∇ϕ) ds
0 ∂t 0
t t
+ (β(x)θ, ϕ)Γ R ds − vθ, ∇ϕ ds
0 0
t t t

≥ α2 μ(θ )|E(v)|2 , ϕ ds + θ (0), ϕ(0) + α1 θ f · v, ϕ ds + f 2 , ϕ ds
0 0 0
for a.e. t ∈ [0, T ] ∀ϕ ∈ C 1 ([0, T ]; CΓ1 D (Ω)) with ϕ(T ) = 0, ϕ ≥ 0.
(10.145)
Now, we will prove that θ satisfies the second equation of (10.20) with “defect
measure”.
By (10.96), we have that α2 μ(θεk )|E(vεk )|2 ∈ L 1 (Q), and these give continuous
linear functionals on C(Q). Since Q is compact, every α2 μ(θεk )|E(v
εk )| is identified
2

with a Radon measure (see Theorem 7.10.4, p.111 of [6]). Since α2 μ(θεk )|E(vεk )|2
is bounded in L 1 (Q) and C(Q) is separable, there exists a ∗−weak convergent
subsequence and a positive Radon measure μ0 ∈ M ( Q̄) such that as εk → 0
T
α2 μ(θεk )|E(vεk )|2 , ϕ dt → ϕ dμ0 ∀ϕ ∈ C(Q)
0 Q̄
and
T T T T
∂ϕ
− dt +
θ, (κ(θ )∇θ, ∇ϕ) dt + (β(x)θ, ϕ)Γ R dt − vθ, ∇ϕ dt
0 ∂t 0 0 0
t (10.146)

= ϕ dμ0 + θ (0), ϕ(0) + f 2 , ϕ ds ∀ϕ ∈ C 1 ([0, T ]; CΓ1 (Ω)), ϕ(·, T ) = 0.
D
Q̄ 0
From (10.145) and (10.146), we have

T
ϕ dμ0 ≥ α2 μ(θ )|E(v)|2 , ϕ dt ∀ϕ ∈ C 1 ([0, T ]; CΓ1 (Ω)), ϕ ≥ 0, ϕ(·, T ) = 0.
D
Q̄ 0
Define
μ = μ0 − α2 μ(θ )|E(v)|2 λ4 |B (Q) ,
where λ4 is the Lebesgue measure in R4 and B(Q) is the σ −algebra of Borel subsets
of Q. Then, the second equation of (10.20) holds with a “defect measure” μ.
To get estimate (10.21), we use (10.96) and (10.115).
Let us prove

|∇θ |2 |∇θεk |2
d xdt ≤ lim inf d xdt 0 < δ < 1. (10.147)
Q (1 + |θ |)1+δ ε→0 Q (1 + |θεk |)1+δ
Put
∂θεk /∂ xi
z εk ,i := a.e. in Q (i = 1, 2, 3).
(1 + |θεk |)(1+δ)/2
By (10.96) {z εk } is bounded in L 2 (Q), and by choosing a subsequence if necessary,

we can assume as k → +∞ that
z εk ,i z i in L 2 (Q). (10.148)
To get (10.147), it is enough to prove that
∂θ/∂ xi
zi = . (10.149)
(1 + |θ |)(1+δ)/2
By (10.122) as k → +∞
5
∂θεk /∂ xi ∂θ/∂ xi in L r (Q), 1 < r < ,
4
1 1
(1+δ)/2
→ in L r (Q).
(1 + |θεk |) (1 + |θ |)(1+δ)/2
Then,
∂θεk /∂ xi ∂θ/∂ xi
→ in L 1 (Q),
(1 + |θεk |)1+δ (1 + |θ |)1+δ
from which we get
∂θεk /∂ xi ∂θ/∂ xi
→ a.e. in Q (10.150)
(1 + |θεk |) 1+δ (1 + |θ |)1+δ
for a subsequence if necessary. By (10.148) and (10.150), we get (10.149) (see

Lemma 1.1).
By virtue of (10.96), (10.115), (10.123) and (10.147), we get (10.21). The main
result is thus proved.
Remark 10.4 The existence of a solution to the non-steady problem (10.1) with
mixed boundary conditions including the static pressure is open yet.

The system (10.1) was studied in several papers. For the equation with α0 = α1 =
0, the existence of solution to the problem was studied in [5] under homogeneous
Dirichlet boundary conditions for velocity and temperature, while in [8] existence of
a solution under homogeneous Dirichlet boundary conditions for velocity and homo-
geneous Neumann condition for temperature was studied. In [4] when |α0 |, |α1 | are
small enough in accordance with data of the problem, the existence of a solution
was discussed under homogeneous Dirichlet boundary conditions for velocity and
mixture of homogeneous Dirichlet condition and homogeneous Neumann condition
for temperature. All papers mentioned above concern with homogeneous Dirichlet
boundary condition for velocity. The definitions of solutions in [4, 5, 8] for tem-
perature θ include the “defect measures” due to bad smoothness of the term for
dissipation of energy.
In [9] for the problem with α0 = α1 = 0, the existence of solution to the problem
was studied under Navier slip boundary conditions for velocity and homogeneous
Dirichlet condition for temperature. But, the definition of solution for temperature is
inequality instead of equality (see (11) in [9]), and so in essence the solution satisfies
an equality with a “defect measure” as [4, 5, 8, 10].
Dealing with the problem α0 = α1 = 0 and g = 0, in [11, 12] instead of the
equation for temperature of (10.1) the equation of total energy

∂ |v|2 |v|2
θ+ + div θ+ + p v − div κ(θ )∇θ = div μ(θ )E(v)v + f · v (10.151)
∂t 2 2
is used. Under Navier slip boundary conditions for velocity and homogeneous Neu-
mann condition for temperature in [11] and under periodic boundary conditions for
velocity and temperature in [12], the existence of solutions to the problems is proved,
but unlike [4, 5, 8, 10] the equation (10.151) is satisfied in the sense of distribution
without the defect measures. It is not only reason of success in [11] to use the equa-
tion (10.151) to avoid the term for dissipation of energy in temperature equation of
(10.1). Using smoothness of eigenvectors of a steady linear problem, they obtained
the existence and estimates of more smooth solutions to approximate equations,
which gives possibility to avoid the term for dissipation of energy.
In [13] for the 2-D problem with α0 = α1 = 0 existence of a strong solution
was proved under homogeneous Dirichlet boundary conditions for velocity and tem-
perature when the ratio between the derivatives of the viscosity and the thermal
conductivity functions and their lower bounds is small enough.
[14–17] were concerned with (10.1) with α1 = 0. In [14] the problem under non-
homogeneous Dirichlet boundary conditions for velocity and temperature in time
dependent domain was studied and existence of a local-in-time solution or existence
of solution on the given interval for small data was established. In [15] existence
of a strong solution and periodic solution for 2-D problem was studied under the
boundary conditions and domain as above. In [16] under homogeneous Dirichlet

boundary conditions for velocity and temperature existence of a strong solution
and periodic solution were studied when data of the problem are small enough.
[17] deals with the initial boundary value problem on a time-dependent domain
with homogeneous Dirichlet boundary condition for velocity and temperature. The
existence and uniqueness of local weak solutions, and as well as the existence of a
global weak solution for small initial data are proved.
The problems with mixture of Dirichlet boundary condition of velocity and out-
flow condition for fluid were studied in [18–20]. In [18] when the viscosity, spe-
cific heat and thermal conductivity are independent of temperature, for non-steady
problem on an open 2-D channel domain the local-in-time existence of a solution
was proved under mixed boundary conditions above for fluid and mixture of non-
homogeneous Dirichlet and homogeneous Neumann boundary conditions for tem-
perature. In [20] on exterior-like domains, which is surrounded by a solid surface
and a disjoint artificial boundary outside of the solid, a modified problem, where
the buoyancy term (1 − α0 θ ) f is changed by ρ(θ ), (0 < ρ(θ ) < ρ1 ), α1 = 0 and
specific heat is independent of temperature, was studied. For the problem the local-
in-time existence of a solution was proved under mixed boundary conditions above
for fluid and mixture of homogeneous Dirichlet and Neumann boundary condi-
tions for temperature. In [19] another modified problem on 3-D pipes, where the
buoyancy term and α1 θ f · v are replaced, respectively, by ρ(θ ) f and α1 ρ(θ ) f · v,
(0 < ρ0 < ρ(θ ) < ρ1 ), and viscosity, specific heat and thermal conductivity are inde-
pendent of temperature, was considered. When the data of problem are small enough,
the existence, regularity and uniqueness of a solution were proved under mixed
boundary conditions above for fluid and mixture of Robin and homogeneous Neu-
mann boundary conditions for temperature. In the papers with mixed boundary con-
ditions for fluid above the assumptions for shape of domain is essential because in
such cases the solutions to the corresponding steady Stokes problem belong to more
smooth space than W 1,2 (Ω) which is used for estimations of solutions.
For a system for non-homogeneous heat-conducting viscous incompressible fluid
we refer to [21, 22].
There are several papers concerning with the systems for heat conducting non-
Newtonian fluid corresponding to (10.1). In [23] when α0 = α1 = 0, under homoge-
neous Dirichlet boundary condition for velocity and mixture of homogeneous Dirich-
let and Robin conditions for temperature existence of a solution was obtained. In [24]
when α0 ≥ 0, α1 ≥ 0 are small enough in accordance with data of the problem, under
homogeneous Dirichlet boundary condition for velocity and homogeneous Neumann
boundary condition for temperature the existence of a solution was proved. In [25]
when α0 = α1 = 0, under mixture of Navier slip condition and homogeneous Dirich-
let boundary condition for velocity and mixture of a generalized Robin conditions
and the homogeneous Dirichlet boundary conditions for energy, the existence of a
solution to the problem was established. However, due to the conditions for nonlinear
terms of strain the results of [23–25] exclude Newtonian fluid.
References

Pure Appl. 124(4), 107–125 (1980)
2. D.S. Konovalova, Subdifferential boundary value problems for the nonstationary Navier-Stokes
and Stokes equations. Differ. Equ. 36, 875–885 (2000)
3. H. Amann, J. Escher, Analysis I (Birkhäuser, 2005)
4. J. Naumann, J. Wolf, Existence of weak solutions to the equations of natural convection with
dissipative heating, Advances in Mathematical Fluid Mechanics (Springer, 2010), pp. 367–384
5. J. Naumann, On the existence of weak solutions to the equations of non-stationary motion
of heat-conducting incompressible viscous fluids. Math. Methods Appl. Sci. 29, 1883–1906
(2006)
6. V.I. Bogachev, Measure Theory I, II (Springer, 2007)
7. T. Kim, D. Cao, A non-steady system with friction boundary conditions for flow of heat-
conducting incompressible viscous fluids. J. Math. Anal. Appl. 484, 123676 (2020)
8. J. Naumann, An existence theorem for weak solutions to the equations of non-stationary motion
of heat-conducting incompressible viscous fluids. J. Nonlinear Convex Anal. 7(3), 483–497
(2006)
9. L. Consiglieri, Existence of proper weak solutions to the Navier-Stokes-Fourier system. Funkc.
Ekvacioj 55, 347–365 (2012)
10. J. Naumann, M. Pokorny, J. Wolf, On the existence of weak solutions to the equations of steady
flow of heat-conducting fluids with dissipative heating. Nonlinear Anal.: RWA 13, 1600–1620
(2012)
11. M. Bulíček, E. Feireisl, J. Málek, A Navier-Stokes-Fourier system for incompressible fluids
with temperature dependent material coefficients. Nonlinear Anal.: RWA 10, 992–1015 (2009)
12. E. Feireisl, J. Málek, On the Navier-Stokes equations with temperature -dependent transport
coefficients. Differ. Equ. Nonlinear Mech. (2006) 14 pp. (electronic) Art. ID 90616
13. L. Consiglieri, Regularity for the Navier-Stokes-Fourier system. Differ. Equ. Appl. 1, 583–604
(2009)
14. H. Inoue, On heat convection equations with dissipative terms in time dependent domains.
Nonlinear Anal. 30, 4441–4448 (1997)
15. H. Inoue, Strong solutions of two dimensional heat convection equations with dissipating terms.
Tokyo J. Math. 22, 445–472 (1999)
16. H. Inoue, Global and periodic solutions of heat convection equations with internal friction
terms. Nonlinear Anal. 47, 2881–2886 (2001)
17. G. Łukaszewicz, P. Krzyżanowski, On the heat convection equations with dissipation term in
regions with moving boundaries. Math. Methods Appl. Sci. 20(4), 347–368 (1997)
18. M. Beneš, Strong solutions to non-stationary channel flows of heat-conducting viscous incom-
pressible fluids with dissipative heating. Acta Appl. Math. 116, 237–254 (2011)
19. M. Beneš, I. Pazaninb, On existence, regularity and uniqueness of thermally coupled incom-
pressible flows in a system of three dimensional pipes. Nonlinear Anal. 149, 56–80 (2017)
20. M. Beneš, J. Tichy, On coupled Navier-Stokes and energy equations in exterior-like domains.
Comput. Math. Appl. 70, 2867–2882 (2015)
21. Y. Liu, Global strong solutions for the incompressible nematic liquid crystal flows with density-
dependent viscosity coefficient. J. Math. Anal. Appl. 462, 1387–1397 (2018)
22. W. Wang, H. Yu, P. Zhang, Global strong solutions for 3D viscous incompressible heat con-
ducting Navier-Stokes flows with the general external force. Math. Methods Appl. Sci. 41,
4589–4601 (2018)
23. L. Consiglieri, Weak solutions for a class of non-Newtonian fluids with energy transfer. J.
Math. Fluid Mech. 2, 267–293 (2000)
24. J. Nečas, T. Roubíček, Buoyancy-driven viscous flow with L 1 -data. Nonlinear Anal. 46, 737–
755 (2001)
25. L. Consiglieri, Friction boundary conditions on thermal incompressible viscous flows. Ann.
Math. Pure Appl. 187, 647–665 (2008)
Index
Symbols σ t (v, p), 56

xn → x0 , 2 σnt (v, p), 56
xn x0 , 2 στt , 56
( , )[θ ] , 7 σn (v, p), 56
∗
2 X , 31 στ , 56
C −diffeomorphism, 5
r supp u, 7
C r −map, 5 εn (v), 95
C k,1 , 12 εi j (v), 19, 45
C k,1 −class boundary, 13 εnτ (v), 95
C0∞ (), 7 εnn (v), 95
s, p (), 15
W
H k (), 66 ∗
S(v, p), 45 fn f , 2
S t (v, p), 56 mes (Ai ), 5
W −s,q (), 12 vτ , 57
W m, p (0, T ; B), 27 vn , 57
W s, p (), 12 w ⊗ u, 46
W s, p (), 14 E i (v), 46, 68
s, p
W0 (), 12 E (v), 45
X →→ Y , 167 E (v) : E (v), 50
X → Y , 3 D ∗ (S; X ), 26
X ∗∗ , 2 D ∗ (), 11
χ Ai , 5 · , ·, 66
· , ·i , 66 C k,α (), 17
f, wi ≥ 0 (≤ 0), 149
R, 2
Rl , 5 A
Hk (), 66 Advection, 53
H0k (), 66
D (), 11
D (), 14 B
O M (0V ), 100 Backward flow, 62
R, 33 Bilinear form
∂ϕ, 34 Dirichlet, 68
σ (v, p), 56 strain, 69
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer 361
Nature Switzerland AG 2021
https://doi.org/10.1007/978-3-030-78659-5
362 Index
vorticity, 71 D
Body force, 44 Deformation rate tensor, 45
Boundary condition Dense, 1
do nothing, 63 De Rham’s theorem, 71
dynamic, 64 Derivative
free outflow, 63 directional, 33
free surface, 63 Fréchet, 4
friction type, 61 Gâteaux, 4
kinematic, 64 Dissipation of energy, 286
Navier slip, 58 Domain, 8
Navier slip-with-friction, 59
one-sided leak, 61
pressure, 62 E
stick, 58 Equation
threshold leak, 60 Boussinesq, 55
threshold leak into, 61 conservation law of momentum, 45
threshold leak out, 61 incompressible Newtonian fluid, 47
threshold slip, 59 incompressible Newtonian fluid under
Tresca friction, 59 heat, 54
vorticity, 58 internal energy, 50
Buoyancy, 228 internal energy of Newtonian fluid, 50
kinematic equation of fluid particles, 42
Newtonian fluid, 47
Newtonian fluid under consideration of
C
total energy, 53
Case
Newtonian fluid under internal energy,
static pressure, 96, 151
53
total pressure, 96, 151
total energy, 51
Compact, 3
total energy of Newtonian fluid, 51
Compatibility condition, 136, 140, 201, 205, Eulerian description, 42
210
for temperature, 259
for velocity, 259 F
Complex interpolation Fixed-point
Lebesgue spaces, 7 Banach, 3
Sobolev spaces, 23 Schauder, 4
Concave, 89 Fluid
Conservation capillary, 64
angular momentum, 48 heavy, 64
energy, 49 incompressible, 43
mass, 43 Newtonian, 45
momentum, 44 perfect, 45
Constitutive equation, 45, 52 Fourier’s law, 52
Continuity equation, 43 Free surface, 64
Continuous Functional
demi, 28 convex, 33
hemi, 28 finite, 33
Lipschitz, 28 lower semi-continuous(l.s.c.), 33
radial, 28 proper, 33
Convection, 53 weakly lower semi-continuous, 33
Convergence
∗−weak, 2
weak, 2 G
Convex, 89 Gradient, 34
Index 363
H Navier-Stokes equations, 47
Heat flux vector, 49
Homeomorphism, 4
O
Operator
I coercive, 28
Imbedding compact, 3
compact, 16 completely continuous, 3
normed space, 3 maximal monotone, 31
Sobolev’s, 17 monotone, 28, 31
Inequality pseudo-monotone, 30
Friedrichs’, 19 strictly monotone, 28
Gronwall’s, 37 strongly monotone, 28
Hölder’s, 7 Oriented, 84
the first Korn’s, 19
Young’s, 37
Integrable P
Bochner, 21 Parametrization, 84
Lebesgue, 6 Path line, 42
Integral Pressure
Bochner, 21 Bernoulli’s, 48
Lebesgue, 6 static, 48
total, 48
J
Joule effect, 53, 286 R
Relatively compact, 3
L
Lagrangian description, 42 S
Lax-Milgram lemma, 29 Second fundamental form, The, 85
Level subset, 33 Sequence
Cauchy, 1
weak Cauchy, 2
M Shape operator, 87
Mapping Signed curvature, 89
bijective, 4 Simple function, 20
injective, 4 Space
surjective, 4 Banach, 1
Material Bessel potential, 16
derivative, 42 dual, 1
domain, 42 reflexive, 2
volume, 42 separable, 1
Mean curvature, 89 Sobolev, 12
Measurable Steady flow, 42
∗−weakly, 20 Stokes assumption, 45
Bochner strongly, 20 Strain rate tensor, 45
Lebesgue, 5 Streamline, 42
weakly, 20 Stress tensor, 44
Moreau-Yosida approximation, 35 Stress vector, 56
Subdifferential, 34
Surface, 84
N Surface force, 44
Natural parameter, 84 Surface patch, 84
364 Index
T V
Tangent space, 84 Viscosity
Tangent vector, 84 bulk, 45
Theorem dynamic, 45
Lebesgue, 6 kinematic, 47
trace, 18 Vitali’s convergence theorem, 8
Total stress tensor, 56
Total stress vector, 56
Transport theorem, 42 W
Weingarten map, 87
U Y
Unit-speed curve, 84 Yosida approximation, 35

Kim. Equations of Motion For Incompressible Viscous Fluids With Mixed Boundary Conditions

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Kim. Equations of Motion For Incompressible Viscous Fluids With Mixed Boundary Conditions

Uploaded by

Copyright:

Available Formats

Advances in Mathematical Fluid Mechanics

More information about this series at http://www.springer.com/series/5032

ISSN 2297-0320 ISSN 2297-0339 (electronic)

boundary conditions for fluid relying on some properties on the boundaries of

Pyongyang, DPR Korea Tujin Kim

2.2 Boundary Conditions for the Navier-Stokes Equations . . . . . . . 55

4.3 Bibliographical Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

7.2.1 Existence of a Solution to an Auxiliary Problem . . . . . 235

10 The Non-steady Equations for Heat-Conducting Fluids . . . . . . . . . 321

1.1 Banach Space, Fixed Point and Basics of Mapping

1.1.1 Banach Space

Let X be a normed space. A sequence {xn } in X is called a Cauchy sequence iff

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 1

where and in what follows x ∗ , x means the value of a functional x ∗ ∈ X ∗ at an

J is an isometric isomorphism of X into X ∗∗ , and so X ⊂ X ∗∗ .

then {xn } is said to be weakly convergent to x0 . We denote norm convergence of a

means that { f n } weakly converges to f 0 in X ∗ . For a sequence { f n } of a normed space

(See Theorem 1, Sect. 1, Chap. V of [2].)

Theorem 1.4 If all weakly convergent subsequences of a sequence {xn } of a reflexive

(See Lemma 5.4, Chap. 1 of [3].)

x X ≤ CxY ∃C, ∀x ∈ X.

We denote by X → Y the imbedding. If X is dense in Y , then f ∈ Y ∗ may be

1.1.2 Fixed-Point Theorems

d(T x, T y) ≤ kd(x, y) ∃k(0 ≤ k < 1), ∀x, y ∈ M.

Then, there exists a unique point x ∈ M such that T x = x.

Definition 1.1 Let X, Y be normed spaces, and T : M ⊆ X → Y an operator. T is

1.1.3 Basics of Mappings

Let T be a single-valued mapping from a set M to N . Mapping T : M → N is said

f (x + th) − f (x) = t T (x)h + o(t) (t → 0) (1.2)

f (x + h) − f (x) = T (x)h + o(h X ) (h → 0 X ). (1.3)

Definition 1.3 Let X, Y be Banach spaces, f : U ⊂ X → Y and r ∈ N (the set of

1.2 Lebesgue Space, Convergence

In this section we consider real-valued functions defined on a (Lebesgue) measurable

1.2.1 Lebesgue Space

A function u : Ω → R ∪ {±∞} is itself called (Lebesgue) measurable if the set

A measurable function u(x) is said to be (Lebesgue) integrable if there exists a

We define (Lebesgue) integral by

Theorem 1.9 (Lebesgue) Let Ω ⊂ Rl be measurable and let {u k } be a sequence

then u is integrable and

|u(x)| ≤ M ∃M > 0, for a.e. x ∈ Ω

(See Theorem 4.8 of [1].)

Let Ω be an open subset of Rl . The support of a continuous function u is defined

Theorem 1.10 (The

If u ∈ L p (Ω) and v ∈ L q (Ω), then uv ∈ L 1 (Ω), and

u(L r ,L q )[θ ] ≡ uq ≤ ur1−θ uθp .

1.2.2 Convergence of Sequences of Functions

An open and connected subset of Rl is called a domain. Let Ω be a domain.

vn L q (Ω) ≤ C, vn (x) → v(x) a.e. in Ω.

A set of functions F ⊂ L 1 (Ω) is called uniformly integrable if

A set consisting of a single integrable function is uniformly integrable since the

lim mes{x : | f (x) − f n (x)| ≥ c} = 0.

In Lebesgue space Vitali’s convergence theorem (see Theorem 4.5.4 of [12]) is

Lemma 1.3 Let Ω be a bounded domain of Rl . If {u n } is such that

{u n } is bounded in L ∞ (Ω), u n (x) → u(x) a.e. in Ω,

Proof Sine Ω is bounded, u n is integrable. By the conditions, {u k } converges to u

Corollary 1.1 Let Ω be a bounded domain of Rl and μ(ξ ), ξ ∈ R, be a bounded

u n (x) → u(x) a.e. in Ω, vn v in L p (Ω), 1 < p < ∞,

Proof By Lemma 1.3,

Thus, μ(u n )vn μ(u)v in L 1 (Ω), that is,

Corollary 1.2 Let Ω be a bounded domain of Rl and μ(ξ ), ξ ∈ R, be a positive

u n (x) → u(x) a.e. in Ω, vn v in L 2 (Ω),

Proof By Corollary 1.1

μ(u n )vn μ(u)v in L 2 (Ω).

where and in what follows x ∗ , x means the value of a functional x ∗ ∈ X ∗ at an

lim D α f n − D α f, ϕ = lim (−1)|α| f n − f, D α ϕ = 0 ∀ϕ ∈ D(Ω).

from O onto Γ ∩ O, where φ, O and O are the ones in Definition 1.7.

s, p (Ω) = W0s, p (Ω),

s, p (Ω) = W0s, p (Ω) = W s, p (Ω).