Zhen (Leo) Liu

Multiphysics
in Porous
Materials
Multiphysics in Porous Materials
Zhen (Leo) Liu

Multiphysics in Porous
Materials
Zhen (Leo) Liu
Michigan Technological University
Houghton, MI, USA

ISBN 978-3-319-93027-5 ISBN 978-3-319-93028-2 (eBook)

https://doi.org/10.1007/978-3-319-93028-2

Library of Congress Control Number: 2018944447

© Springer International Publishing AG, part of Springer Nature 2018

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

Printed on acid-free paper

This Springer imprint is published by the registered company Springer International Publishing AG part of
Springer Nature.
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
I am dedicating this book to my wife,
Ye (Sarah), and my son, Brent. Their
understanding and support enabled this work
The support from the United States National
Science Foundation (NSF Grants 1562522
and 0700524) is acknowledged. The support
helped me step in and stay active in
multiphysics and research on porous
materials
Preface

This book started from a course that was first offered in the spring of 2015 at
Michigan Technological University. This course was possibly the first one on
multiphysics in porous materials worldwide and even likely among the first in the
general area of multiphysics. Back at that time, I could not find an appropriate
textbook for multiphysics, let alone one with a focus on porous materials. Therefore,
I had to develop all the lecture notes and computer labs from scratch. This book was
created based on this educational endeavor. Many of the book materials have been
tested multiple times via both teaching and learning in that course. In addition, the
book has been enriched with materials from several completed and active research
projects including cutting-edge studies supported by the US National Science
Foundation. In short, the book represents a very implementable and practical
learning guide for multiphysics and a path from the basics of multiphysics to
advanced research topics.
This book is intended to be a general reference for multiphysics. By now, more
than a dozen of books with a title including “multiphysics” can be found. However,
they are mostly conference proceedings, introductions to the development of specific
multiphysics applications, or summaries on multiphysics in specific areas. A relevant
fact is that multiphysics researchers usually tend to focus on the multiphysics
applications in their own areas, no matter in the general multiphysics context or in
porous materials. There is no effective reference for people to systematically learn
about multiphysics, especially to grab a big picture and the basics. This book is
presented to fill this gap.
This book is also designed to be a comprehensive introduction to the area of
multiphysics in porous materials. Porous materials are an ideal breeding ground of
multiphysics due to their multiphase and multicomponent nature. As a result, porous
materials are related to most common types of multiphysical phenomena. For
example, the coexistence of thermal, hydrological, and mechanical processes is a
major interest in geotechnical engineering, hydrogeology, soil science, and petro-
leum engineering. Particle transport and chemical reactions together with heat

vii
viii Preface

transfer and water movement have been extensively studied in environmental

engineering, agricultural engineering, and forestry. Electromagnetics, vibration,
and acoustics are the primary interest of geophysicists, earthquake engineers,
mechanical engineers, and exploratory geologists. Therefore, this book will be a
handy tool to researchers and practitioners in these areas involving porous materials,
no matter they are new to multiphysics or they want to step into a multiphysics topic
out of their home disciplines.
This book is organized in a way to serve readers with different backgrounds and
with different levels of knowledge accumulation in multiphysics. For the purpose,
this book provides materials that would be essential to obtain a general understand-
ing of the concepts, theories, and implementations of multiphysics. It is believed that
multiphysics is an “art” of mathematics, physics, applications, and numerical anal-
ysis. Accordingly, the book is organized into five parts: introduction, mathematics,
monolithic physics, multiphysics, and numerical analysis. In Part I, a general
introduction to the history, state, essence of multiphysics, and its applications in
porous materials is offered. The definition, scope, and classification of multiphysics
will be given based on the discussions on the existing studies. In Part II, necessary
mathematical backgrounds such as tensor and fields, tensor analysis, partial differ-
ential equations (PDE), and numerical solution of PDEs are provided. Eight most
representative monolithic physical fields are introduced in Part III, each of which is
presented as a chapter. Then in Part IV, typical multiphysical processes are intro-
duced. The introduction is made based on the mathematics in Part II and the
monolithic physics in Part III. Each multiphysical topic is introduced in the order
of background, theory, implementation, and applications. The theory will be
discussed in a way that mathematics and physics are integrated. Implementations
are demonstrated with a simple but representative example, which is solved using a
PDE solver. In Part V, a brief tour will be given to show how to implement a
multiphysics model with three typical numerical simulation methods, i.e., finite
difference method, finite volume method, and finite element method, using self-
developed computer code.
The use of physical fields extends over most science and engineering disciplines,
while a discipline or sub-discipline usually primarily deals with one or a few
physical fields. This fact makes multiphysics an extremely interdisciplinary topic
and poses challenges in enumerating the fields. The eight most common monolithic
physical fields introduced in this book are heat transfer (thermo-), pore water
movement (hydro-), concentration field (concentro or diffuso/convecto/advecto),
stress and strain analysis (mechano-), dynamics (dyno-), chemical reactions
(chemo- or chemico-), electrostatics (electro-), and magnetostatics (magneto-). For
multiphysics, the following nine most representative topics in porous materials will
be discussed: thermomechanics, hydromechanics (static poroelasticity),
thermohydromechanics, electrokinetics, electromagnetics, fluid dynamics,
hydrodynomechanics (e.g., fluid-structure interaction), thermoelectromagnetics
(including Joule heating, dielectric heating, and induction heating), and
electromagnetomechanics (e.g., piezoelectricity).
Preface ix

It is my hope that this book can significantly flatten and shorten the learning curve
of multiphysics. Usually, it takes many years to pick up the basic concepts, prepa-
ration knowledge, and software skills before we can excel in or even get a big picture
of multiphysics. Such a multiyear knowledge preparation includes math courses
such as partial differential equations and tensor analysis; specialty courses such as
mechanics, heat transfer, and electromagnetics; and numerical analysis courses such
as finite difference method, finite volume method, and finite element method. As a
result, the learning curve for multiphysics is both steep and long. Commercial
software such as COMSOL made a great effort at lowering the bar for implementing
multiphysics and expediting the process for research and design with multiphysics.
However, more advanced use especially that requires an in-depth understanding of
multiphysics and bottom-level theoretical and numerical operations still calls for a
framework and summary of multiphysics. Based on my 10 years of multiphysics
implementations and even more time in knowledge preparation, I want to share the
opinion that not all the knowledge in the above courses/subjects are equally needed
for understanding and practicing multiphysics. Moreover, it is possible to organize
the most relevant and significant knowledge into a book by which people can obtain
a smooth and pleasant ride to the kingdom of multiphysics especially for the
applications in porous materials. The book is presented and will be continuously
improved to show and prove this opinion.
A continuously updated website, multiphysics.us, is available to complement this
book. This website provides updated technical details, applications from communi-
ties, and networking information for multiphysics and its applications in porous
materials.

Houghton, MI, USA Zhen (Leo) Liu

Contents

Part I Introduction
1 History and Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Multiphysics Is There . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Multiphysics Is Coming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 What Is Multiphysics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Definitions and Scopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
From Monolithic Physics to Multiphysics: Couplings . . . . . . . . . . . . . 13
Physical Fields and Types of Multiphysics . . . . . . . . . . . . . . . . . . . . . 16
3 How to Do Multiphysics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Mathematical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Solution and Postprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4 Multiphysics in Porous Materials . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5 How to Use This Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Organization of the Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Suggestions to Readers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

xi
xii Contents

Part II Mathematical Background

6 Tensor and Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
From Scalar, Vector to Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Physical Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
7 Tensor Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Vector Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Vector Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Basics of Tensor Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Important Identities in Vector Calculus . . . . . . . . . . . . . . . . . . . . . . . . 57
8 Index Notation and Tensor Notation . . . . . . . . . . . . . . . . . . . . . . . . 61
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Rules of Index Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Tensor Algebra in Index Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Calculus Using Index Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Examples of Algebraic Manipulations Using Index Notation . . . . . . . . 64
9 Partial Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Numerical Simulation and PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Classification of Partial Differential Equations . . . . . . . . . . . . . . . . . . . 69
Common PDEs in Engineering Applications . . . . . . . . . . . . . . . . . . . . 71
Boundary Conditions and Initial Conditions . . . . . . . . . . . . . . . . . . . . 73
10 Numerical Solution of PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Solve PDEs with MATLAB PDE Toolbox . . . . . . . . . . . . . . . . . . . . . 80
A PDE Implementation Tool: FlexPDE . . . . . . . . . . . . . . . . . . . . . . . 84

Part III Monolithic Physics

11 Thermo: Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
General Background for Continuum Mechanics . . . . . . . . . . . . . . . . . 93
Heat Transfer in Continua . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
Heat Transfer in Porous Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
12 Hydro: Pore Water Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Groundwater Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Water Flow in Unsaturated Zones . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Contents xiii

Mixed Saturated-Unsaturated Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

Surface Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
13 Concentrato: Transport of Dispersed Mass . . . . . . . . . . . . . . . . . . . 123
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Definitions and Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Advection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
14 Mechano: Stress and Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
Constitutive Relationship: Linear Elastic Hooke’s Law . . . . . . . . . . . . 145
Governing Equation: Navier’s Equation . . . . . . . . . . . . . . . . . . . . . . . 152
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
15 Dyno: Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
Introduction: Meaning of Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 157
Elastodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
Fluid Dynamics and Acoustics in Fluids . . . . . . . . . . . . . . . . . . . . . . . 161
Acoustic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
16 Chemico: Chemical Reaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Uniqueness of Chemical Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Understanding of Chemical Reactions . . . . . . . . . . . . . . . . . . . . . . . . 174
Chemical Kinetics and Its Mathematical Description . . . . . . . . . . . . . . 176
Reaction Rate Coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
17 Electro: Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Coulomb’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
Electric Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Gauss’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Governing Equation and Electric Potential . . . . . . . . . . . . . . . . . . . . . 186
Electrostatic Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
Electric Displacement Field and Dielectric Materials . . . . . . . . . . . . . . 190
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
18 Magneto: Magnetostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
Introduction: Magnetostatics and Magnetic Field . . . . . . . . . . . . . . . . . 195
Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
xiv Contents

Gauss’s Law for Magnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

Governing Equation and Auxiliary Equations in Terms
of Vector Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
Magnetostatic Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

Part IV Multiphysics
19 Thermomechanics: Non-isothermal Mechanics . . . . . . . . . . . . . . . . 209
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Mathematical Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
Coefficient of Thermal Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
20 Hydromechanics: Poroelasticity as a Simple Case . . . . . . . . . . . . . . 219
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Definitions of Phase Composition, Stress, and Strain . . . . . . . . . . . . . . 221
Static Poroelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
Dynamic Poroelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
21 Thermohydromechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Thermal Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
Hydraulic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
Mechanical Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
General Boundary Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
22 Electrokinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Double Layer Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
Electroosmosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
Electromigration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
Electrophoresis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
Mathematical Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
23 Electromagnetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
Conventional Maxwell’s Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Physical Understanding of Maxwell’s Equations . . . . . . . . . . . . . . . . . 279
Equation System for Numerical Simulations . . . . . . . . . . . . . . . . . . . . 281
Electromagnetic Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
Formulation Using Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
Phasor Formulation (Frequency Domain) and Waveguide . . . . . . . . . . 288
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
Contents xv

24 Fluid Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
Flow Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Navier-Stokes Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Navier-Stokes Equations for Different Fluids . . . . . . . . . . . . . . . . . . . 306
Turbulent Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
25 Hydrodynomechanics: Fluid-Structure Interaction . . . . . . . . . . . . . 319
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
Dynamic Poroelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
Complete Fluid-Structure Interaction . . . . . . . . . . . . . . . . . . . . . . . . . 322
Acoustic-Structure Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
26 Thermoelectromagnetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Joule Heating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Dielectric Heating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
Induction Heating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
27 Electromagnetomechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
Magnetomechanics: Magnetostatic Force . . . . . . . . . . . . . . . . . . . . . . 351
Electromechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360

Part V Implementation Methods

28 Finite Difference Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
Introduction: Numerical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
Spatial Finite Difference and Derivative . . . . . . . . . . . . . . . . . . . . . . . 370
Temporal Finite Difference and Schemes . . . . . . . . . . . . . . . . . . . . . . 373
Boundary and Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
Error Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
Extension to 2D and 3D Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . 381
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
29 Finite Volume Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
Integral Form of the Governing Equation . . . . . . . . . . . . . . . . . . . . . . 386
Boundary and Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
Multidimensional Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
xvi Contents

30 Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
Galerkin Method and Weak Form . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
Example of Discretization: Thermomechanics . . . . . . . . . . . . . . . . . . . 400
Appendix: Typical Finite Elements (Q4 and Q9) . . . . . . . . . . . . . . . . . 405
Practice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
 Part I
Introduction
Chapter 1
History and Future

Introduction

The chapter starts the introduction with the ubiquitousness of multiphysics. Exam-
ples are given to demonstrate the wide existence of multiphysical phenomena in both
naturally occurring processes and man-made applications. These examples traced
back to ancient activities of human beings, showing that multiphysics and their
applications are neither new nor rare. We will also show that the understanding of
such phenomena via computers, mostly in terms of multiphysics simulation, has also
been around for decades. Brief introductions will also be made to reveal the relation-
ships between multiphysics and the trends of interdisciplinary research, industrial
collaboration, computer engineering, software engineering, rapidly increasing com-
plexity of new scientific problems, and engineering design and policy-making.
These will help readers understand the clear uprising trend of multiphysics and
foresee its future developments.

Multiphysics Is There

We live in a multiphysics world. Real-life processes are inherently multiphysics.

From atoms to galaxies and from amino acids to living organisms, naturally built
systems involve various types of physical phenomena operating at different spatial
and temporal scales. Loads and deformations on solids, complex flows, fluid-
structure interactions, plasma and chemical processes, and thermomechanical and
electromagnetic systems are just a few representative examples in fundamental and
applied sciences (Krzhizhanovskaya 2007). Take a cell phone in Fig. 1.1, for
example – the antenna receives electromagnetic waves, the touchscreen or buttons
handle the interaction between mechanical and electric components, and the battery
involves chemical reactions and electric currents [Multiphysics Cyclopedia].

© Springer International Publishing AG, part of Springer Nature 2018 3

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_1
4 1 History and Future

Fig. 1.1 An example of

multiphysics

Therefore, multiphysics is neither a research concept far from daily life nor a theory
or technique developed recently.
We have been dealing with this multiphysics world even before the civilization of
human beings. For example, a nature fire from lightning is a typical electrostatic,
thermal, and chemical phenomenon. Fire starting with a hand drill involves both
mechanical and thermal processes. Flood protection and irrigation required to
understand the interaction between hydrodynamics and groundwater movement.
Construction and earthwork called for the understanding of a coupled mechanical
and hydrologic process. The modern industrialization even stemmed from
multiphysical processes such as steam engines, which is an art of thermal, mechan-
ical, and dynamics. What would be more surprising is not that the modern lives are
full of multiphysics, but instead, we have been taking advantage of most of these
multiphysical processes in such an adept way.
Surprisingly, while multiphysics is part of nature and history of lives, the study of
multiphysics is not new either. We can describe what happens in the world using sets
of physical laws. Originally, due to the limited understanding of complex physical
processes and computing resources, physical effects were observed and described in
isolation. For example, traditional courses in engineering disciplines were usually
built upon the understanding of one type of physical process. For example, various
courses in civil engineering were established on the basis of mechanics while many
in electrical engineering were on electrostatics. However, there are still many
multiphysical phenomena, which are very fundamental and “multiphysics,” such
as electromagnetics and elastodynamics. These multiphysics topics were studied and
matured much earlier than other multiphysical phenomena, even before the advent
and prevalence of the word “multiphysics.” Since the 1940s, we have been using
computers to understand these physical phenomena. Thus, regardless of the trade-
mark “multiphysics,” simulations that incorporate multiple physical phenomena are
as old as simulations themselves (Keyes et al. 2013).
Despite the studies which are by nature multiphysics but were not traditionally
counted, multiphysics research has been on the playground for a couple of decades.
On one hand, many physical problems of interest these days are complex in nature
Multiphysics Is Coming 5

and consist of separate physical processes that each contributes to the problem as a
whole. Also, some problems which were historically investigated as an isolated
physical process are now needed to be considered as a holistic process. This fact
requires researchers to consider multiple physical processes when constructing
numerical models (Groen et al. 2014). On the other hand, the uprising computing
power in virtue of the innovations in computer engineering and science has been
lending us the power to deal with more complicated problems, which may be
insurmountable in the old days due to complicated couplings between fields. The
higher speed of computing unit, higher storage, and especially the popularization of
computer clusters enable the consideration of the whole process in a holistic way,
giving rise to various multiphysics techniques. Multiphysics studies under the name
of multiphysics have been conducted and widespread at least for two decades.
In fact, multiphysics has been developed beyond a research concept. For example, in
industrial design and product testing, engineers used to solve one physical phenomenon
at a time, such as structural integrity, and then import the results into another system to
solve for the aerodynamic behavior. Multiphysics simulation tools now allow them to
simulate and analyze both these behaviors and many others at the same time. Similar to
those physical processes discussed in research papers, in reality, product performance
usually depends on several physical phenomena interacting with each other:
multiphysics. To acquire dependable and realistic simulation results that allow better
design decisions, coupled simulations with many physical effects involved become
necessary and essential. Also, in light of the advance and promotion of multiphysics
computer-aided design tools such as COMSOL, multiphysics has become an accepted
consensus, trend, and even standard practice in many industries. With a multiphysics
capable simulation tool, designers and engineers now can better capture the important
aspects of the behavior of the product.

Multiphysics Is Coming

Multiphysics has rapidly developed into a research and application area across many
science and engineering disciplines. There is a clear trend that more and more chal-
lenging problems we are faced with do not involve physical processes covered by a
single traditional discipline in science or engineering disciplines. This trend requires us
to extend our analysis capacity to solve more complicated and more multidisciplinary
problems, advance the front edge of engineering and applied sciences by exploring
these problems, contribute back to the software development and computing tech-
niques, and establish and explore innovative but urgent application areas.
Modern academic communities are confronted with problems of rapidly increas-
ing complexity, which straddle across the traditional disciplinary boundaries
between physics, chemistry, material science, and biology. To respond to this
challenge, many disciplines have spontaneously initiated efforts at reaching out to
other disciplines to target at emerging issues which appear as a traditional topic in
those disciplines while also extend beyond the scope. At the same time,
6 1 History and Future

computational science, which is more focused on techniques rather than problems,

has been receiving a steadfast development. Innovative modeling techniques have
been investigated for the purpose of offering an optimal handling of the information
transfer which connects different aspects, components, scales, and levels involved in
the quantitative description of those multidisciplinary phenomena. This trend is
entailing the seamless coupling between different mathematical representations of
distinct physical natures across a wide spectrum of scales from geologic scales and
scale of lab specimen to atomistic trajectories.
Multiphysics has also become a frontier in industrial applications. Simulation pro-
grams have been evolving into a tool familiar to many engineers in design, product
development, and quality control. This change is reshaping traditional creation pro-
cesses: from design blueprints to prototypes to testing to validation to production,
which was once simple, to extremely complex ones involving engineers from several
disciplines. During these creation processes, engineers now must think in areas outside
of their training, even with the assistance of the simulation tools. As commented by
Courtney (2015), it is more and more necessary for the modern engineers to know and
grasp the concept of what is known deep inside the engineering world as
“multiphysics.” Take automotive industry, for example; traditionally, there are differ-
ent people focusing on the structures, fluids, electromagnets, and the other individual
aspects separately. Now, the intersection of aspects, which may represent two physics
topics and once was a gray area, can be the essential link in the life cycle of the product.
Such needs will enable multiphysics applications to attract a much wider engineering
user base and allow more designers to come together. As commented by Nilsson
(Thilmany 2010), “Design engineers are running more and more multiphysics simula-
tions every day because they need to add reality into their models.”
Multiphysics will continue to benefit from the progress in computer engineering
and science. The advent of commodity computing has been leading to profound
changes in numerical analysis of continuum mechanics problems. In the meanwhile,
the computing power, in terms of both the maximum computing flop rate and energy
efficiency, has been increasing exponentially as shown in Fig. 1.2. Besides, teraflop
rate computing facilities, which were available only to a few until recently, are
becoming commonplace in an industrial setting (Jasak 2006). This just reflects the
fact that the range and complexity of problems in numerical analysis are expanding,
leading to increasing interest in coupled problems which was computationally
prohibitive. Besides, spectacular advances in computer performance and emerging
technologies of parallel distributed grid computing are offering tools to break the
barriers and bring simulation to a higher level of detail and accuracy. In the
meanwhile, the advances in hardware and computing techniques also call for new
efficient numerical algorithms specific to the field where coupling different models
or scales within one simulation is essential (Krzhizhanovskaya 2007).
The developments in multiphysics have been in turn contributing to the software
engineering. The need and purpose of multiphysics software are to bring different
strands of engineering under one umbrella. However, this is not an easy and
straightforward task as it sounds. This is because research on complicated
multiphysical phenomena or computer-aided design can include issues as diverse
Multiphysics Is Coming 7

1E8

1E7
1000
1000000

MFlops per Watt

100000
GFlops

10000 100

1000

100
10
10

0.1 1
1995 2000 2005 2010 2008 2009 2010 2011 2012 2013
Year Year
Most Powerful Computer Most Power Efficient Computer
Computer Ranked #500 Computer Ranked #500

Fig. 1.2 Development of computing (a) power and (b) efficiency. (After top500.org)

as predictions of structural strength, electromagnetic behavior, and fluid flow.

Fortunately, the needs of the academic and industrial communities have generated
considerable momentum in the software market to transfer or extend from traditional
software packages with separate physics modules to tools with multiphysics genes
which can promote better communication and interaction during the project life
cycle. A general rule is that the more multiphysics phenomena, the more accurate the
final model will be, though the cost of building and solving more complicated model
usually discourage such effort. This would lead to a compromise between accuracy
and effort. Despite the progress, multiphysics is still a field in need of integration
(Thilmany 2010). An easy way is to “stitch” different packages together using file
imports, which is still common today, and continuous effort is being made to render
it smoother. More recent techniques aim at building up all the applications on the
same platform all the way from the bottom.
Multiphysics is not only reformatting traditional applications but also giving rise
to new applications in scientific prediction, engineering design, and policy-making.
In the automotive industry, multiphysics can be employed for traditional testing such
as forces of rotating wheels, environmental stresses, and shock and vibration, under
the conditions of different geometries, materials, and load scenarios. In the mean-
while, new application areas are under development in virtue of multiphysics. For
example, multiphysics is a significant impetus for the research and application of
cyber-physical systems, which combine electronic control with physical problems. It
is challenging but essential to understand the multi-domain problems underlying
these systems, particularly for those humanoid robots equipped with multiple sys-
tems with more complex electrical and mechanical elements. For example, engineer
Joel Gibbard’s Open Hand Project (Web 2013) explored the idea of how an amputee
can create his or her prosthetic hand using a 3D printer. This project involves the
design using National Instruments LabVIEW software and data acquisition hard-
ware and the measurement of the electrical activity of muscles at rest and during
contraction to understand how the nerves send electrical signals.
Chapter 2
What Is Multiphysics

Introduction

As introduced, multiphysics has been studied since the beginning of numerical

simulation and at least two decades under the name of “multiphysics.” However,
there has been no consensus on the definition of multiphysics. First of all, this word
might be a little misleading or controversial if used in areas such as physics or in a
context when the concept is not well known. Second, even in numerical simulation,
the word could still refer to significantly different things.
In this chapter, the possible meanings of multiphysics will be discussed based on
four definitions that can be found in the literature. Based on that, a clear definition of
multiphysics will be reached to lay down the scope of this book. With this definition,
more details will be shown about how to move from monolithic physics to
multiphysics. The purpose is to bridge this common knowledge gap to help some
readers who are familiar with traditional monolithic physics simulations move
toward multiphysics. Detailed discussions will be made regarding the couplings
and methods for developing multiphysics models. Finally, common types of
multiphysics, as well as the essence and scope of “(physical) fields,” will be
discussed to outline a big picture for multiphysics. This chapter aims to present
the first comprehensive discussion on “what is multiphysics” to help readers under-
stand and compromise the widely existing discrepancies in the definition, scope, and
use of terminology in multiphysics.

Definitions and Scopes

Definitions of multiphysics can be found in some literature which can be categorized

according to the broadness of the scope. The typical definitions are listed and
explained as follows in the sequence of scope broadness, i.e., from broad to narrow.

© Springer International Publishing AG, part of Springer Nature 2018 9

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_2
10 2 What Is Multiphysics

Definition 1

Multiphysics treats simulations that involve multiple physical models or multiple

simultaneous physical phenomena [Wikipedia]. While the use of “physical phenom-
ena” semantically rephrases “multiphysics,” the inclusion of “multiple physical
models” makes this definition a very broad and general concept. For example,
combining chemical kinetics and fluid mechanics is a multiphysical process
according to “physical phenomena” in the definition. Likewise, combining finite
elements with molecular dynamics is then also a type of multiphysics as it includes
“multiple physical models.” This definition is a little bit self-contradictory as the
implication of physical models may include that of physical phenomena. In fact, the
definition of physical models is possibly more appropriate for multimodel. The
multimodel concept may include multiscale, multirate, multilevel, and multilevel
problems (Keyes et al. 2013). In this definition, multiphysics typically involves the
solution of coupled systems of partial differential equations.

Definition 2

In a relatively narrow sense, multiphysics includes (1) coupled physical phenomena

in computer simulation and (2) the study of multiple interacting physical properties
[COMSOL]. This definition narrows the scope of Definition 1 to physical phenom-
ena and the materials associated with them. Thus a coupling between either physical
phenomena of different types or materials properties associated with these different
physical phenomena is an essential characteristic. The use of physical phenomena
implies that multiple physical processes, each of which should be a distinct physical
field, such as thermal field, can be described by a partial differential equation
(s) based on underlying physical mechanisms such as conduction. The use of
“physical properties” in the definition makes it more closely related to applications
because in many phenomena of practical interest, such as piezo resistance, these
physical properties instead of the coupled process are the major interest. However, it
is worthwhile to mention that the physical properties should be more than two
properties and the properties are associated with different fields. Otherwise, density
changes due to variations in water content in porous materials can be accounted as a
multiphysical phenomenon.

Definition 3

Semantically, a multiphysics system consists of more than one component governed

by its own principle(s) for evolution or equilibrium, typically conservation or
Definitions and Scopes 11

constitutive laws (Krzhizhanovskaya 2007; Groen et al. 2014). This definition is

very close to Definition 3 except for that it does not emphasize physical properties.

Definition 4

In a more strict way, multiphysics can be defined as those processes including

closely coupled interactions among separate continuum physics phenomena (Cross
& Slone 2005). In this definition, two-way exchange of information between
physical fields, which could involve implicit convergence within a time step, is the
essential feature. Accordingly, those processes only involve using data generated by
one code as input into another is not qualified as multiphysics but, instead, as
multidisciplinary.

Definition in This Book

The book defines multiphysics in the following way. First, the above definitions all
point out several features and leave out some others. Among these features, multiple
physical processes which are governed by their own laws of evolution or equilibrium
are necessary. It seems that this is a feature directly reflecting the name and
characteristics of most multiphysical processes. Second, the multimodel definition
is abandoned to differentiate this topic from multiscale and multilevel problems.
Third, couplings are not adopted in the definition as those are more an implemen-
tation technique than a fundamental characteristic.
Besides, a few neglected details need to be further discussed. First, “physics” in
“multiphysics” means a physical field, which is a physical state variable varying with
respect to space and/or time according to physical laws for its evolution or equilib-
rium. Thus, a field is either a time-dependent spatial distribution of a state variable in
a transient problem or a spatial distribution of a state variable in an equilibrium
problem. The word physical process/phenomenon is not that clear and closely
related to mathematics. Second, the simultaneous occurrence of these physical fields
seems to be more a rule than an exception. Though sequential occurrence is also
possible, it is far less common and leads to the direct coupling of physical processes
involved. Therefore, simultaneous occurrence will be emphasized to leave out this
possibility. Third, all the definitions implicitly imply a deterministic process where
stochastic factors do not seem to be present. However, if “physical field” is used,
then it will be unnecessary to emphasize this feature. Fourth, the above definitions
were made from the perspective of processes, phenomena, or system. All of these
make multiphysics equal to the problem under investigation. But in a more common
context, we also frequently use multiphysics to refer to the studies of and/or
knowledge pool for these problems. Finally, a consensus on the above definitions
is that multiphysics is about computer simulations. However, the implementation of
12 2 What Is Multiphysics

multiphysics is not necessary via computer simulations and is also relatively inde-
pendent of the methods of discretization and solution. Finally, it is believed that a
major pool of knowledge is about the establishment of the mathematical model
based on physical laws, which will also be the focus of this book. Thus, computer
simulations will not be emphasized in the definition.
Based on the above discussion, in this book, multiphysics is defined as the
coupled processes or systems involving more than one simultaneously occurring
physical fields and also the studies of and knowledge about these processes and
systems. For example, objects moving due to Newton’s second law will be viewed as
an example of multiphysics as long as velocity will be calculated at every point of the
domain, that is, to obtain a field. However, a model of a star cluster that resolves
Newtonian gravitational interactions or molecular dynamics, which is built upon
Newton’s second law, is not multiphysics according to this definition. From a
mathematical perspective, systems consisting of partial differential equations
(PDEs) of different types (e.g., elliptic-parabolic, elliptic-hyperbolic, or parabolic-
hyperbolic) may be classified as multiphysics because each of the classical PDE
archetypes possibly represents a different physical phenomenon (Keyes et al. 2013).
Multiphysics problems vary widely in their nature, both in the physics involved
and in the manner of coupling. To further clarify the scope, we will first classify
multiphysics into two major categories based on the way in which different physical
processes are coupled. In the first category, the coupling occurs in the bulk through
source terms or constitutive relations that are active in the overlapping domains of
the individual components. In the second category, the coupling occurs over an
idealized interface or a narrow buffer zone through boundary conditions that trans-
mit fluxes, pressures, or displacements (Dennis et al. 2015). Typical examples of
bulk-coupled multiphysics systems are radiation with hydrodynamics in astrophys-
ics (radiation hydrodynamics), electricity and magnetism with hydrodynamics in
plasma physics (magnetohydrodynamics), and chemical reaction with transport in
combustion or subsurface flows (reactive transport). Typical examples of interface-
coupled multiphysics systems are ocean-atmosphere dynamics in geophysics, fluid-
structure dynamics in aeroelasticity, and core-edge coupling in tokamaks.
Multiscale is frequently mentioned with multiphysics. This is possibly because
multiscale and multiphysics are different sides of a physical process. Multiphysics,
which lies in fundamental physical laws, comes into being as a result of interactions
originating from atomic scales and below and upscaling level by level to the scale of
interest. Here we use the term multiscale modeling to refer to both multiscale
modeling and multiscale simulation of physical problems, and the term multiscale
application to refer to the program used to do the modeling. Groen et al. (2014)
differentiated multiphysics and multiscale using the concept of submodel. Multiscale
and multiphysics are therefore two distinct concepts, but they are common in that
both of them consist of a number of submodels which are combined or coupled.
Therefore, both multiphysics and multiscale face the same challenge of coupling
these submodels such that the overall model is both accurate enough to be scientif-
ically relevant and reproducible and efficient enough to be executed conveniently by
modern computational resources (Groen et al. 2014).
From Monolithic Physics to Multiphysics: Couplings 13

From Monolithic Physics to Multiphysics: Couplings

Though multiphysics is not necessarily obtained from monolithic physics, an intu-

itive and widely accepted idea is to reach multiphysics by means of monolithic
physics. For such an idea, the first step to establish a general multiphysics platform is
to provide the required models in a “single-physics” mode (Jasak 2006). This is
possibly related to the historical development of numerical simulation. Monolithic
physics simulations are the traditional commonplace and still predominant, and
accordingly, numerous commercial software programs exist. It is thus intuitive and
possibly more reasonable to attempt to establish multiphysics frameworks based on
the current knowledge and tools for monolithic physics.
The relationship between monolithic physics and multiphysics can be treated in
two ways. A common and straightforward way is to approach multiphysics bottom-
up as the assembly of individual physical fields. As illustrated in Fig. 2.1, a bottom-
up approach, taking the thermo-hydro-mechanical models for example, is
established based on the understanding of thermal, hydraulic/hydrologic, and
mechanical processes and necessary couplings to stitch them together. The other
way is to view the multiphysical process from a complementary perspective: that
problems are intrinsically coupled, while a monolithic physics application is merely
an idealization made in asymptotic limits (Keyes et al. 2013). This approach is
exemplified by the non-isothermal poroelasticity, which can be simplified into
isothermal poroelasticity and thermomechanics for porous materials. The basic
physical fields used in the above example for the two methods are actually similar,
though the terminology is different.
The strength of couplings between the physical fields can be used to determine
which way to take. The strength of a coupling here is measured by its influence on
the overall process. For example, in most cases, a flow of pore water within a porous
material could change the porous skeleton, which in turn changes the flow regime. If
this flow occurs in stiff rocks, the effect of this coupling to the flow or the fluid flow-
rock deformation process is likely to be small. When this small effect is not of
concern, a feasible way to consider this problem is to establish a model based on pore
water flow and deformation of solids, for which monolithic models are available.
Then the major effort to be made is to find a way to formulate this deformation and, if

Fig. 2.1 Two ways for developing multiphysics model

14 2 What Is Multiphysics

needed, changes in permeability, both of which are caused by the flow. However, if
this process occurs in a soft clay, then there will be a process called consolidation,
which can take place over years. The couplings between water flow and solid
deformation are very strong, and a bottom-up coupling may not be able to provide
the desired accuracy. This is the reason why van Terzaghi’s theory of consolidation
and Biot’s theory of poroelasticity came into play. In this book, the introduction to
multiphysics will be made in a bottom-up way. That is, monolithic physics will be
introduced first to prepare the ingredients for cooking multiphysics. However,
whenever possible, multiphysical processes will be described from a complementary
perspective.
Couplings between different physical fields can be classified as one-way and
two-way couplings according to the way in which two fields interact and as explicit
and implicit according to solution techniques. The definitions of one-way and
two-way coupling are pretty straightforward. A one-way coupling indicates that,
for a specific type of interaction, one field affects the other field but not in the
opposite way, while for two-way coupling, two fields will influence each other via
the same or relevant physical mechanism(s). Further examinations of coupled-field
analysis reveal that two types of coupling generally exist for multiphysics problems
including implicit (directly) coupling and explicit (sequential) coupling. In an
implicitly (directly) coupled system as shown in Fig. 2.2, an algebraic matrix system
of equations based on all of the relevant physics is assembled and then solved. One
drawback of direct-coupled systems is that finding a solution can be costly in terms
of required processing power and computer memory. Alternatively, an explicitly
(sequentially) coupled system adopts a segregated solver, where the solution to the
first set of governing equations is passed to the second set of governing equations,
which is solved and then passed to a third set of equations, etc. This segregated
process is repeated until a final solution is obtained. Many commercial multiphysics
finite element analysis packages now are able to automatically choose an appropriate
default coupling depending on the physics involved or let the user customize these
settings, as needed, based on solution time, available computing power, and numer-
ical convergence of the problem at hand (Dede et al. 2014).
The two classification methods can then be combined to categorize couplings
based on the level of coupling. The first category is the one-way explicit coupling,

Fig. 2.2 Explicit

(sequential, left) and
implicit (direct, right)
couplings
From Monolithic Physics to Multiphysics: Couplings 15

which is usually shorted as one-way interaction. This is the simplest way to couple
two fields. One-way implicit coupling has rarely been mentioned possibly because
the incorporation of a one-way coupling implicitly into the stiffness matrix may not
make a significant difference. The next level is the explicit two-way coupling, which
is typical when multiple software or discretization methods are involved. In this
coupling, simulations run side-by-side and exchange coupling information during
the run. This approach formally operates in Picard iterations in computational fluid
mechanics and fails even in modestly interacting problems. The level with even
stronger coupling is implicit two-way coupling. This level provides closer model-to-
model coupling that involves matrix-level interaction, where two physical models
and model-to-model coupling terms are discretized separately and combined into a
single linear system before the solution. This is an implicit variant of the above, with
additional stability provided by the linear solver. Here, we can see the benefit of
shared matrix and solver modules between multiple discretization schemes. The
highest level of coupling is the equation-level coupling, which is however only
available for special cases. This type of coupling can be achieved based on the fact
that continuum mechanics models originate from mass, momentum, and energy
balance models. Thus, a fluid-structure system may be considered as a single
continuum governed by a single PDE, with different constitutive relations in various
regions. Representing the strongest coupling, this level of interaction, in fact,
represents a coupled physical model and may require special numerical techniques.
Such techniques are particularly suitable for strong coupling or cases involving
phase change and transition regions, e.g., melting and solidification.
Some facts have not been explicitly discussed in the literature but may deserve
our attention in the future. The first fact is that most existing couplings were
understood in the context of binary field systems. It is thus hypothesized that
physical fields interact in a pair-wise way. Therefore, the possibility that three or
more physical fields work together with multichannel entangled couplings is totally
ruled out. It is believed to be necessary not only to mention this possibility here but
also to point out that the couplings introduced in this book are pair-wise couplings if
not mentioned otherwise. The second fact is that it is usually unclear regarding how
to count couplings. One way is to define the interaction(s) between two different
physical fields as “a” coupling. The other way is to define a specific interaction as a
coupling, which is identified according to nature of this interaction including its
direction and physical mechanism. This book adopts the latter way. Again, take the
interactions between water flow and contaminant transport, for example – advection
and dispersion will be treated as two different couplings because their underlying
mechanisms are different. Similarly, the water flow due to particle movement in
electrophoresis and particle movement due to water flow are believed to be two
couplings because they have different directions. However, when coupling is used as
an uncountable noun, coupling also means any or the overall interaction between
two fields.
Establishing multiphysics based on monolithic physics is not a “one plus one
equals two” game (Keyes et al. 2013). The job of coupling individual simulations
may introduce limitations on stability, accuracy, or robustness that are severer than
16 2 What Is Multiphysics

the limitations imposed by the individual components. Furthermore, the data motion
and data structure conversions between independent simulations for each component
may be more costly in latency and electrical power than those of the individually
tuned components. As a result, “one plus one” may cost significantly more than
“two” and may be less amenable to scalable execution (Keyes et al. 2013). There-
fore, though the involved single physics, e.g., fluid flow, structural mechanics,
electromagnetic fields, and acoustics, are well-understood, coupling them into one
multiphysics simulation is not trivial.
The “plus” in this one-plus-one procedure is the main difficulty due to several
reasons. First, discretizing the whole set of equations for a multiphysics problem is
prone to ill-conditioned system matrices that are hard to solve with sufficient
accuracy. Also, the implementation of a new code for every possible (and required)
combination of single-physics phenomena would be an immense effort. Thirdly,
reuse of existing single-physics codes by just gluing them together requires a lot of
numerics such as data mapping between non-matching grids and numerical iteration
schemes to regain the solution of the fully coupled system and technical solutions for
code-to-code communication. Fourth, the high accuracy of multiphysics models can
only be exploited with a very high resolution of the underlying computational grids,
necessitating the use of massively parallel supercomputers.
Perhaps one of the greatest challenges for multiphysics simulation is to under-
stand all of the relevant physics involved and to set up a geometrically accurate
model with appropriate loads and boundary conditions (Dede et al. 2014). The
purpose of this book is to provide such a framework to multiphysics especially
those in porous materials. There is currently no such a comprehensive framework.
Therefore, whenever a designer would like to conduct a multiphysics analysis for
which he or she does not have such background, it is intended that this book can be
either a comprehensive introduction or a quick reference. However, we do not
consider multiphysics applications from algorithmic and architectural perspectives,
where “algorithmic” includes both mathematical analysis and computational com-
plexity and “architectural” includes both software and hardware environments. In
fact, many multiphysics applications can degenerate into an algebraic paradigm via
linearization in which individual components, i.e., monolithic physics, are
represented by diagonal blocks and the multiphysics coupling between them, as
off-diagonal blocks (Keyes et al. 2013).

Physical Fields and Types of Multiphysics

The part “physics” in “multiphysics” denotes “physical field.” There, multiphysics

means the coexistence of multiple physical fields in a process or a system. In physics,
a field is a physical quantity that has a value for each point in space and time. For
example, on a weather map, a vector at each point of the map is assigned to represent
the surface wind velocity, including both speed and direction of the movement of air
at that point. In another classic example, an electric field can be viewed as a
Physical Fields and Types of Multiphysics 17

“condition in space” emanating from an electric charge and extending throughout the
whole space. If a test electric charge is placed in this electric field, the particle will
accelerate due to the force associated with the “condition in space.” Possibly
attributed to the latter example, physicists tend to think of the notion of a field as
the cause of a force.
The origin of “field” is also related to force. In the eighteenth century, a new
concept was proposed to simplify all these gravitational forces. This quantity,
namely, the gravitational field, was believed to exist at each point in space, whereby
the total gravitational force can be felt by an object with a unit mass at that point.
This concept just provided another way to interpret the force rather than change the
physics. With the field concept, two treatments will be equivalent: one calculates all
the gravitational forces on an object individually and then adds them together, and
one first adds all the contributions together as a gravitational field and then applies it
to an object (Weinberg 1977). Then in the nineteenth century, the concept of a field
received substantive developments due to the development of the theory of electro-
magnetism. In the early stages, Newton-style laws were employed to express the
forces between pairs of electric charges or electric currents. Later the field concept
became much more predominant for expressing these laws in terms of electric and
magnetic fields. It is believed that Michael Faraday was the first to coin the term
“field” in 1849.
The scope of field may be wider than what is needed in multiphysics. In this book,
the fields we are dealing with are constrained to classical fields. Therefore, the
quantum field will not be considered. Actually, the discussion of the quantum field
in the context of multiphysics is not only interesting but also necessary. First, the
quantum field has seldom been discussed in the multiphysics literature. This has
been at least a convention, though it will not necessarily be in this way forever as the
scope of multiphysics is expanding. The quantum field will not be counted or
discussed in the book. However, it is necessary to point out that the physical fields
we are dealing with occur in virtue of quantum mechanics. This is because common
physical processes occur due to the basic forces. All of these basic forces except
gravity, i.e., strong interaction, electromagnetic force, and weak interaction, can find
their origins in the quantum field. As a result, multiphysics will be constrained above
the scale of atoms, though it is not necessarily a solely macroscopic theory. For
example, hydrodynamics can span over scales from geologic scale down to Kolmo-
gorov microscale, and chemical field can reach down to the molecular scale. Another
presumption is that the fields in multiphysics are non-relativistic. Descriptions of
physical fields were given before the advent of relativity theory and then revised in
light of this theory. Consequently, classical field theories are usually categorized as
non-relativistic and relativistic. Relativistic field theory can explain gravity. But we
do not need to worry too much about the origins of the basic forces because we just
deal with the non-relativistic fields and have physical laws above the quantum level.
The discussion in the above paragraph essentially determines the mathematical
languages to be used in multiphysics. A field can be classified as a scalar field, a
vector field, a spinor field, or a tensor field according to the type of physical quantity
they represent: a scalar, a vector, a spinor, or a tensor. A field has a unique tensorial
18 2 What Is Multiphysics

character at every point where it is defined. That is, a field can only exclusively be a
scalar field or a vector field. Take the Newtonian gravitational field, for example – it
requires three numbers to specify the components along the three spatial dimensions
at any point, leading to a vector field. However, within each category (scalar, vector,
tensor), a field can be either a classical field or a quantum field, depending on
whether it is characterized by numbers or quantum operators, respectively. As we
do not deal with the quantum mechanics here, so we just to use the normal operators
in calculus analysis.
The use of physical fields extends all over most science and engineering disci-
plines. However, a discipline or subdiscipline only primarily deals with one or a few
physical fields. This fact makes multiphysics an extremely interdisciplinary topic
and also poses challenges in enumerating the fields. However, the following fields
can be identified by overviewing the core courses in the disciplines and publications
on multiphysics: heat transfer (thermo-), pore water movement (hydro-), concentra-
tion field (concentro or diffuso/convecto/advecto), stress and strain (mechano-),
dynamics (dyno-), chemical reactions (chemo- or chemico-), electrostatics (electro-
), and magnetostatics (magneto-). In multiphysics, a multiphysical process is usually
titled using compound words such as “thermo-hydro-mechanical.” These words now
appear as the birthmarks or tattoos of multiphysics. The prefixes, roots, or suffixes in
the parentheses after each physical field are the ingredients for cooking multiphysics.
Each physical field is called a monolithic physical field, uniphysics, or single
physics.
Combinations of the above monolithic physics can lead to 247 possible types of
multiphysics. However, the world of multiphysics is not only a game of combination
in statistics. But instead, we coin a name and investigate a type of multiphysical
phenomenon under this name based on the observations in nature and sciences and
the needs from practice. Based on a review of multiphysics with an emphasis in
porous materials, the most representative multiphysical processes are believed to be
the following:
• Thermomechanics
• Hydromechanics
• Thermohydromechanics
• Electrokinetics
• Electromagnetics
• Elastodynamics
• Fluid dynamics
• Hydrodynomechanics
• Thermoelectricity
• Thermoelectromagnetics
• Electromagnetomechanics
Another point deserving attention is that the orders of the roots in the name.
Generally, it would be desirable to put the major process, cause or process of primary
interest in the front. However, there is no widely accepted rule. For a process such as
Physical Fields and Types of Multiphysics 19

Fig. 2.3 Holistic multiphysics framework

poroelasticity, it may even be a difficult job to judge which process is dominant, i.e.,
water movement and solid skeleton deformation.
Figure 2.3 shows an example of a comprehensive multiphysics framework which
was developed based on the compositions and physical mechanisms in porous
materials. Any multiphysical process can be reinterpreted using four fundamental
mechanisms in the real phases: mass conservation, balance (static), oscillation
(dynamic), and energy conservation, and mechanisms in two “latent” phases: polar-
ization (electric and magnetic) and transformation (phase change and chemical
reactions). Typical multiphysical phenomena are combinations of different compo-
nents (right). Also, classic and highly active research topics in geotechnical engi-
neering can be reinterpreted using the framework (left). As we move from classic
topics to upcoming ones, the level of difficulty raises in the context of multiphysics.
There is more than one way to categorize and name multiphysics. For example,
the multiphysics software COMSOL categorizes multiphysical phenomena based on
applications. The most common multiphysics applications were divided into four
major categories: electrical, mechanical, fluid, and chemical. Electrical applications
include Joule heating, induction heating, microwave heating, piezoelectric effect,
piezoresistive effect, and electromechanical effects. The mechanical section consists
of thermal expansion, thermal stress, and acoustic-structure interaction. The fluid
section contains Navier-Stokes equation, Boussinesq approximation, non-isothermal
flow, fluid-structure interaction, poroelasticity, and squeezed and sliding films. The
chemical section comprises convection and diffusion. The advantage of this classi-
fication is that it is conducted based on applications, so it uses terms closer to
applications. But the disadvantages are also very clear. The adopted terms do not
effectively reveal the underlying mechanisms and cannot really reflect the real
hierarchy and evolution of multiphysics and their relationships to monolithic phys-
ics. On the contrary, a classification purely based on the combinations of monolithic
physics is not able to reflect the applications in the real world. For example, many
20 2 What Is Multiphysics

combinations of monolithic physics may not have any existing applications at all.
The classification system used in this book is a compromise between theories and
applications. So you can see it still keeps the birthmarks or tattoos but, at the same
time, adopts terms that are widely accepted but actually do not reflect the involved
monolithic physics, such as hydrodynamics and poroelasticity.
Chapter 3
How to Do Multiphysics

Introduction

The implementation of multiphysics includes several essential sequential compo-

nents: identifying a multiphysical process/system, developing a mathematical
description of this process/system, discretizing this mathematical model into an
algebraic system, and finally, solving this algebraic equation system and
postprocessing the data to obtain information of interest. The abstraction of a
multiphysical problem from a complex phenomenon and the description of such a
problem are very critical to the success of the multiphysics analysis. This requires to
identify the system to be analyzed, including geometry, materials, and dominant
mechanisms, which will be interpreted using mathematical languages (function,
tensor, differential equation) as computational domain, boundary conditions, auxil-
iary equations, and governing equations. Mostly, discretization, solution, and
postprocessing are carried out using computers. Therefore, the above procedure is
very similar to those in general numerical simulation based on partial differential
equations.
This chapter is intended to give out a general but definite answer to the question:
“How to do multiphysics?” First, the abovementioned essential sequential compo-
nents of the implementation of multiphysics will be introduced. Emphasis will be
placed on both the abstraction of a multiphysical problem from a complex phenom-
enon with physical and mathematical languages. Finally, the discretization, solution,
and postprocessing using computers will be explained. The understanding under-
neath the above introduction is that multiphysics can be viewed as an art of physics,
mathematics, applications, and numerical analysis.

© Springer International Publishing AG, part of Springer Nature 2018 21

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_3
22 3 How to Do Multiphysics

Mathematical Model

A mathematical model boils down to a bunch of equations. According to the nature

and intended role, the equations can be divided into three categories. Couplings
between fields can be achieved in each category. As shown in Fig. 3.1, the mathe-
matical model which is defined in a continuous domain will be discretized into an
algebraic model defined on a meshed domain for solution.
The first category is governing equations. A governing equation describes the
major physical mechanisms and process without further revealing the change and
nonlinearity of the material properties. For example, in a heat transfer problem, the
governing equation could describe a process in which the thermal energy
(represented using temperature or enthalpy) at an infinitesimal point or a represen-
tative element volume is changed due to energy transferred from surrounding points
via conduction, advection, radiation, and internal heat sources or any combinations
of these four heat transfer mechanisms as the following equation:

∂u
þ∇  ðuvÞ ∇  ðK— uÞ ∇  ðD∇uÞ ¼ Q ð3:1Þ
∂t
|{z} |fflfflfflfflfflffl{zfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |{z}
Advection Diffusion ðConductionÞ Dispersion Source
Accumulation

According to the definition of multiphysics in this book, one physical field of one
component in one region only has one governing equation. Therefore, the mathe-
matical model for any multiphysical process should contain at least two equations.
Likewise, a mathematical model for a physical process containing two or more
governing equations indicates multiphysics according to this criterion. Couplings
can be embedded into governing equations. For example, in the thermal field, heat
advection as a separate term describes the coupling from the fluid movement to the
thermal field.
A governing equation is mostly a partial differential equation, that is, a differen-
tial equation with two or more independent variables. There could be two or more
spatial coordinates (x, y, . . .) in a 2D or 3D equilibrium problem or one temporal
coordinate (t) with one or more spatial coordinates (x,. . .) in any time-dependent

Fig. 3.1 Mathematical model and its discretization

Mathematical Model 23

(transient) problems. The dependent variable, i.e., the state variable of the field such
as temperature in the thermal field, is a function of these independent variables.
However, there are two exceptions when a governing equation is an ordinary
differential equation instead of a partial differential equation. This first exception
is 1D equilibrium problems, in which a spatial coordinate is the only independent
variable. The other is the chemical field. Technically speaking, chemical reactions
are not a part of physics, but in multiphysics, it is a convention to count it as a
physical field. This treatment also brings difficulties in the establishment of the
mathematical model because a chemical field is usually described using an ordinary
equation. However, it can be viewed as a “field” mathematically if we associated
every point or REV in the computational domain with such an ordinary equation.
Every ordinary differential equation describes a chemical reaction at the point it is
associated with and is independent of the others. However, connections can be made
via other fields such as mass transport.
The second category of equations is auxiliary equations. These auxiliary equa-
tions describe (1) the relationships between state variables and the relationships
between material properties and (2) other state variables or/and material properties.
These two types of auxiliary relationships represent two different ways that coupling
can occur or can be implemented. Two typical examples are shown in Fig. 3.2. The
first type exists because there are many cases in which there are more state variables
than the number of governing equations. As the number of independent variables
and number of governing equations should be equal to ensure a unique solution, only
some of the state variables can be selected as dependent variables. Then the
unselected state variables then can be related to the selected ones via constitutive
relationships. This treatment is quite common in both monolithic physics and
multiphysics, such as stress and strain in the mechanical field, suction and water
content in unsaturated flow, and pressure, density, and temperature in compressible
fluid mechanics. The second type is the dependence of a material property on state
variables or/and other material properties. For example, it is not difficult to imagine
that the heat capacity of a porous material would be a function of water content and

1
0.5
Volumetric water content

0.8
Relative permeability

0.4 Base Base

Subgrade Subgrade
0.6
0.3

0.4
0.2

0.1 0.2

0 0
0 0.5 1 1.5 2 2.5 3 3.5 0 100 200 300 400
log(h) (unit of h: kPa) Suction (h, kPa)

Fig. 3.2 Relationship between state variables (SWCC between water content and suction) and
between state variable and material properties (relative permeability and suction) in different porous
materials (base material and subgrade soil under pavement). (After Liu and Yu 2011)
24 3 How to Do Multiphysics

heat capacity of the constituents of the porous material. In reality, these two types
can possibly be mixed up, leading to even more complicated auxiliary relationships.
The auxiliary relationship is also a material property, such as the stress-strain
relationship, which defines the stiffness of an elastic object. It is not unusual that
the stiffness is a function of other state variables such as temperature and material
properties such as moduli of the constituents.
The third category of equations is the boundary and initial conditions. One PDE
or a PDE system may have an infinite number of solutions. This is very similar to the
solution to ordinary differential equations. When inadequate constraints are pro-
vided, that is, lack of enough boundary/initial conditions, we can only obtain general
solutions. These general solutions contain arbitrary constants and represent an
infinite number of possibilities. However, boundary and initial conditions as addi-
tional constraints can turn the general solutions into a particular solution. Boundary
conditions correspond to spatial coordinates, whereas initial conditions correspond
to the temporal coordinate. Therefore, no initial condition is needed or allowed in an
equilibrium problem. The existence of the boundary and initial conditions also has
practical implications. This is because, in numerical simulations, it is infeasible to
simulate an infinitely large domain or a process lasting forever. In fact, the
discretization size, algorithm, available computing resources, and the nonlinearity
of the problem determine the upper limit of the geometrical size and duration of the
process that can be simulated.
Multiphysics is an art of mathematics, physics, and applications. Just as shown in
Fig. 3.3, math, physics, and application provide three independent angles to perceive
multiphysics. As a result, a complete description of a mathematical problem needs
input from all of the three aspects. In detail, the mathematical model should be
established based on the real physical conditions. However, there are distinct
differences between the real physical processes and the conceptual mathematical
models. These differences result in differences in the sequences and languages
adopted in each and the differences between the real physical process and the
solution to the mathematical model.
Assuming there is a physical process/system to be simulated, from the perspective
of multiphysics, the first thing to think about is “whether this is a multiphysical

Fig. 3.3 Three aspects of

multiphysics
Discretization 25

process, if it is, then how many different physical fields are involved?” In most cases,
the answer is not that clear as multiphysics is more a rule than an exception, and we
can possibly find the trace of most physical processes if we zoom in enough. So the
question then turns into which physical fields are the primary ones. The answers can
also change with the expectation of the accuracy and effort for the analysis. Then the
question that needs to be considered is what object will be simulated. In physics, this
question includes what are the objects/areas of interested, what are the interactions
between these objects/areas, and how these objects interact with the remainder of the
world. After being translated into mathematical languages, the questions turn into
how many regions in the computational domain and what are the boundary condi-
tions on boundaries between regions and on the external boundaries of the whole
domain. Boundary conditions need to be defined for each physical field, so do initial
conditions. For multiregion problems, different physical fields and axillary relation-
ships may also be assigned to different regions.
In most commercial numerical analysis software, a user interface is usually
provided to help users to build up mathematical models. To effectively use this
user interface, we still need to understand the physics and determine the physical
fields and physical objects to be simulated. Once we have them in mind, it is usually
fairly easy to choose the physics, to build up the computational domain and to set up
boundary and initial conditions. The computational domain could be a line in 1D, an
area in 2D, or a volume in 3D. It can also turn out to be very complicated 2D or 3D
geometries with complex inclusive regions. Many programs offer a powerful graph-
ical user interface to help generate these geometries or provide an interface to import
geometries from other computer-aided design software. But repairing and
defeaturing of the geometry are often required to fix the parts of the geometry that
are not “watertight” and remove slender surfaces or merge unnecessary small edges.

Discretization

Discretization is the step right following the establishment of a mathematical model.

This is because the mathematical model is a bunch of equations and usually we do
not solve these equations directly using computers. It is possible to solve some
mathematical models by hand or by computer programs such as Mathematica or
Maple using symbolic computation. But for most mathematical models, it is possibly
easier to use numerical computation. Numerical computation requires an algebraic
equation system such as

AX¼B, ð3:2Þ

where A is the stiffness matrix, X is the matrix of unknowns, and B is the force
matrix.
Discretization is the step to transfer a mathematical model into a numerical model
or an algebraic equation system as shown above. In mathematics, discretization
26 3 How to Do Multiphysics

concerns the process of transferring continuous functions, models, and equations

into discrete counterparts. This process is usually carried out as a step toward making
them suitable for numerical evaluation and implementation on digital computers.
There are many options for discretization, among which finite difference method,
finite volume method, and finite element method are the most popular ones. These
numerical methods employ significantly different ways for discretization. The finite
difference method approximates the differential equations with difference equations,
in which finite differences approximate the derivatives. The finite volume method
employs volume integrals in a partial differential equation, in which a divergence
term is converted to a surface integral using the divergence theorem. The finite
difference method and finite volume method are similar to each other, and both work
on values calculated at discrete places on a meshed geometry. The finite element
method also works on a meshed geometry, but it connects many simple element
equations over many small subdomains, named finite elements, to approximate a
more complex equation over a larger domain. The method uses variational methods
from the calculus of variations to solve the problem by minimizing an associated
error function. In summary, these methods involve discretization of both the com-
putational domain and the equation system.
The discretization of the computational domain is called meshing in numerical
analysis, which is considered to be one of the most difficult tasks of preprocessing in
traditional numerical finite volume and finite element analysis. The difficulties
involve both the effort to obtain good quality grids/meshes automatically and also
mesh settings complying with the nature of the physical problem. For example, in
wave propagation problems, such as elastic waves in structural mechanics and
electromagnetic waves in radio frequency analysis, the size of the largest element
has to be substantially smaller than the wavelength to ensure a solution without
losing significant physical details. In fluid flow, boundary layer meshes are neces-
sary to simulate boundary layers, for which the cell Reynolds number may determine
the element size in the bulk of the fluid. In the modern numerical analysis, adaptive
meshing has become a widely accepted technique, in which an initial mesh may be
automatically altered during the solution process for the purpose of reducing the
error or changing the geometry.

Solution and Postprocessing

Compared to meshing, selecting and setting the solvers, and obtaining a solution to
the equations, constituting the numerical model within a reasonable computational
time is possibly an even more difficult task. The difficulties have different sources.
The first source is the size of the discretized model, which can be very large in
terms of algebraic equations. The degrees of freedom of numerical simulation can
easily reach several millions. Also, nonlinear, parametric, eigenvalue, and time-
dependent problems are usually transferred into a system of linear equations and
solved using iterative methods. Therefore, the challenge in numerical simulation
Solution and Postprocessing 27

ends up at the solution of a large linear system of algebraic equations. However, it is

usually not easy to solve large linear systems. General purpose solvers are available
yet usually not the optimal choice considering the high computational cost due to
lack of optimization for realistic problems. For example, direct solvers based on
lower-upper factorization methods or general purpose iterative methods usually
approach the solution without a clear consideration of the structure of the system.
Moreover, the structure of a multiphysics problem may be difficult to identify or
even does not exist. One way to help address this problem starts from the structure of
the included monolithic physics. For iterative methods, algebraic (Wesseling and
Oosterlee 2001) or geometric multigrid methods (Ruge and Stuben 1987) can also be
helpful.
The second source is the nonlinearity of the numerical models. Multiphysics
problems usually involve extreme nonlinearity due to the strongly coupled
governing equations, complex nonlinear material properties in terms of auxiliary
equations, and more complicated initial and boundary conditions. The nonlinearity
will lead to complex morphology in the structure of the solution. This is analogous to
a terrestrial region full of complex topographic features such as mountains and
canyons. In iterative methods, the search of the solution is like looking for a point
in the region, usually the highest or the lowest point. The complex topographic
features, or nonlinearity in this case, will tremendously increase the difficulty of a
successful search. Even without an iterative method, the numerical solution also
requires the use of approximation functions to approach the true solution. Whenever
the solution changes fast over a small region or a small amount of time, the effort for
an acceptable approximation will multiply. To solve this problem, different simpli-
fications or relaxations of the problem are usually adopted. For example, less
significant nonlinearities may be removed to obtain a linear problem that can be
solved easily. This type of solver is the segregated and continuation solver (Heil
et al. 2008).
Difficulties in solution can also come from the numerical model or numerical
schemes. For time-dependent problems, the stability of the numerical scheme needs
to be considered. Inappropriate settings of time steps and mesh size can lead to
instability, which can make the numerical solution deviate from the true solution or
even fail to converge. Sometimes, extremely small time steps or mesh sizes are
needed in order to obtain a reasonable solution, leading to a high requirement of
computational resources. The solution to such problems is much more difficult than
for a more benign and well-behaved numerical model. For such a condition, it is
usually helpful to investigate the numerical model and the schemes for discretization
and their relationship to the structure of the problem and the true solution. Tuning
solver settings may help identify the problem but may not be the most effective cure.
Postprocessing is the last link of the numerical solution. It usually involves
visualizing and investigating 2D/3D plots of the modeled fields and variation of
quantities with respect to time or spatial coordinates. Old-fashioned packages require
that the plots and derived values to be analyzed had to be defined before the solution
process. Omission of some key definitions means that the solution has to be carried
out again from the beginning. The definitions of these needed quantities and/or
28 3 How to Do Multiphysics

derived values are, therefore, deemed as part of preprocessing. More recent software
usually tends to allow any derived quantities and derived values to be calculated after
the solution process. Error estimation is also an essential part of the numerical
solution. It is usually helpful to try the numerical model with different mesh sizes
in order to estimate the convergence of the numerical solution. Besides, sensitivity
analysis, which involves the solution of the same model with different input, such as
material properties, initial conditions, and boundary conditions, tests the influence of
some parameters and the overall performance of the numerical model. This can help
understand the structure of the problem, hidden bugs in the code, and errors in the
model.
Chapter 4
Multiphysics in Porous Materials

Introduction

This chapter is intended to explain the strong connections between multiphysics and
porous materials by overviewing the history and applications of multiphysics in
porous materials. We will start with the relationship between multiphysics and
porous materials and then explain the reason for the existence of monolithic physics
as the mainstream in traditional porous studies and the transition of these studies to
multiphysics for pressing and upcoming issues. Finally, we will explain the relation-
ships between the major multiphysics topics and common porous material studies by
listing and elaborating well-known existing applications of multiphysics in porous
materials.

History

Porous materials are susceptible to multiphysical phenomena due to their nature as a

composition material consisting of a porous skeleton, pore fluids, and solutes. This is
because this distinct nature provides favorable conditions for energy transfer, water
migration, geomechanical responses, contaminant transport, and electrokinetic phe-
nomena. Many porous materials have complex constituents in both solids and
liquids, imperfections at different scales, and amorphous structures. For example,
soils as a natural product could contain hundreds of different minerals, structural
deficiencies from nanoscale partial substitutions to macroscale ink holes, and amor-
phous structure as a particulate material. This makes any process in it hardly a
monolithically physical one. Almost most common multiphysics problems can find
their trace or applications in porous materials. That is also the reason why it is
necessary to cover the whole multiphysics when introducing the multiphysics in
porous materials.

© Springer International Publishing AG, part of Springer Nature 2018 29

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_4
30 4 Multiphysics in Porous Materials

Porous materials are so closely related to multiphysics; no wonder multiphysics

has its footprints in the whole history of porous material studies. A major discipline
dealing with porous material, i.e., geotechnical engineering, was believed to be
developed into an independent discipline when von Terzaghi published his work
on the one-dimensional consolidation theory in the 1920s. This theory, however, is
about the coupled process between pore water flow and deformation of the solid
skeleton. The later rapid development of geotechnical engineering in the 1980s was
triggered by the earthquake investigations and geoenvironmental applications,
which boil down to considerations of more physical processes such as dynamics,
contaminant transport, and chemical reactions. In other disciplines related to porous
materials, the story is similar. For example, petroleum engineering featured the
multiphase and multicomponent flow, the influence of many chemicals and magnetic
waves on the flow, and also the explorations of the reserve with various electrical,
seismic, and magnetic approaches.
As physical processes in porous materials are, more or less, coupled in nature, it is
very difficult to separate one from the others. However, a single physical phenom-
enon can be analyzed individually without significantly affecting the analysis results
when the couplings between different physical fields are weak enough. This is, in
fact, the mainstream in the traditional studies of porous materials, though
multiphysics has always accompanied. Multiphysics was considered only in the
cases in which the couplings between some physical processes are too strong.
Typical examples include consolidation and porous materials under frost action.
However, multiphysics is turning to be more and more of a necessity as we are faced
with more and more challenging problems such as climate effects and more and
more stringent requirements on the accuracy of the analysis. Examples of such
applications include shale gas, carbon sequestration, geothermal energy, climate
effect on porous materials, the behavior of porous materials under extreme condi-
tions, soil erosion, extraterrestrial soils, synthetic of new materials, exploration of
traditionally unexplored areas, design and fabrication of actuators and sensors,
biomechanics, and microfluidics. More concerns have been shifted from individual
physical phenomena to coupled multiphysical processes due to emerging needs in
porous material-related disciplines and advances in multiphysics research.
Figure 4.1 schematically shows the structure of a typical thermo-hydromechan-
ical process in porous materials. Many multiphysical phenomena are triggered by a
change in the thermal field. This change serves as an excitation, similar to an external
force in the mechanical field, and gives rise to changes to the whole system in the
form of a multiphysical process. For example, thermal changes can alter the com-
pressibility and shear strength (mechanical), induce water migration, and change
hydraulic properties (hydraulic). These are the typical couplings from the thermal
field to mechanical and hydraulic fields, respectively. These changes in mechanical
and hydraulic fields will in return exert “reacting forces” to the thermal field. For
example, heat transfer can be promoted by energy convected by migratory water,
and thermal properties such as heat capacity and thermal conductivity can be
changed due to the variations in the structure and composition of the material. The
“reacting forces” represent the couplings from the other two physical processes to
Applications 31

Fig. 4.1 Couplings between physical fields in soils

the thermal field. It is worthwhile to point out that the couplings from the thermal
field to other fields are usually strong couplings. Among these couplings, the
thermally induced water flux is especially significant and thus of special interest.
This process of water transfer due to temperature gradients is critical to many
multiphysical processes in porous materials. When the phase change of water
(freezing or thawing) gets involved, the multiphysical phenomenon could be more
complicated, and the significance of the thermally induced water flux can be
seriously exaggerated. One typical example is frost heave, in which water is sucked
from deeper positions to the advancing frost front due to temperature changes on the
ground surface.

Applications

Thermomechanics is the coupling between the thermal field and mechanical field.
Typical applications can be dived into two categories. The first category includes
temperature-induced mechanical responses, that is, the thermomechanometry of
porous materials and its consequent physical processes. Thermomechanometry is
the measurement of a change of a dimension or a mechanical property of a sample
subjected to a temperature regime. It is also responsible for the mechanical behavior
32 4 Multiphysics in Porous Materials

of various materials under non-isothermal conditions. Typical examples include

thermo-cracking of pavement, energy piles, microstructural changes of porous
materials such as thermal barriers or heat exchangers, and effect of climate change
on geomaterials. The second category involves an opposite process: heat generated
by mechanical movements. This category attracts much less interest but is still
responsible for some very basic phenomena such as frictional heat.
Hydromechanics is more of an extension of poroelasticity. Poroelasticity con-
siders the coupled processes between the pore water movement and deformation of
porous materials. A typical application is the soil consolidation, which is a funda-
mental process in geotechnical engineering and petroleum engineering. Hydrome-
chanics is more general as it also includes nonelastic material properties. Several hot
applications are primarily related to multiphysical processes such as shale gas
recovery, carbon sequestration, and hydraulic fracturing.
Thermohydromechanics is possibly the most widely existing multiphysical phe-
nomenon in porous materials. It is more general than thermomechanics and hydro-
mechanics because (1) non-isothermal conditions or heat transfer is usually
unavoidable, (2) temperature gradients can cause water movement in porous mate-
rials, and (3) migratory water can advect heat. Typical applications include frost
heave, gas hydrate dissociation, cement hydration, geothermal reservoir, and
enhanced oil recovery using hot water flushing. It is also common that more fields
can be added. For example, chemical reactions and particle transport can be incor-
porated so that phenomena such as a bioreactor, sanitary landfills, ground remedi-
ation, ground improvement using chemical or biologic methods, and erosion can be
included. If the electric field is included, it then can be used to consider fuel cells.
Electrokinetic phenomena are a family of several different effects that occur in
heterogeneous fluids, in porous bodies filled with fluid, or in a fast flow over a flat
surface. According to Lyklema [ref], the complete family of electrokinetic phenom-
ena includes electrophoresis (motion of particles under influence of electric field),
electroosmosis (motion of liquid in porous body under influence of electric field),
diffusiophoresis (motion of particles under influence of a chemical potential gradi-
ent), capillary osmosis (motion of liquid in porous body under influence of the
chemical potential gradient), sedimentation potential (electric field generated by
sedimenting colloid particles), streaming potential/current (either electric potential
or current generated by fluid moving through porous body or relative to flat surface),
colloid vibration current (electric current generated by particles moving in fluid
under influence of ultrasound), and electric sonic amplitude (ultrasound generated
by colloidal particles in oscillating electric field). Typical applications are ground
remediation, electrokinetic chromatography, and electrokinetics in microdevices for
biotechnology.
Electromagnetics has a very long history and serves as the background of
electrical engineering. In porous materials, its significance is primarily attributable
to the application of various electromagnetic technologies in porous materials, such
as sensors based on electromagnetics (ground penetration radar, time domain reflec-
tometry, and capacitive-type water content sensor), microwave and induction
Applications 33

heating of porous materials for agricultural and petroleum applications, and novel
materials treating using electromagnetics.
Elastodynamics is the study of elastic waves and involves linear elasticity with
variation in time. An elastic wave is a type of mechanical wave that propagates in
elastic or viscoelastic materials. The elasticity of the material provides the restoring
force of the wave. When they occur in the Earth as the result of an earthquake or
other disturbance, elastic waves are usually called seismic waves. Therefore this
multiphysical phenomenon is one of the primary mechanisms underlying earth-
quakes. It also determines the applications of seismic waves in exploratory geology
and the applications of any acoustic methods in porous materials.
Fluid dynamics deals with fluid flow in motion. It usually becomes a concern in
porous materials in two ways. The first is that the flow of surface water interacts with
porous materials. This is very common in soil erosion, bridge scour, and piping. The
other way is obvious when pore water moves at very high velocity, such as those
microfluidics applications (e.g., inkjet print heads, micro-propulsion) and wicking in
porous materials. This is also of concern whenever pore water flow is connected to a
large body of water such as the interaction between groundwater and surface water
and natural convection affected by groundwater.
Hydrodynomechanics refers to the involvement of both hydrodynamics and
mechanical responses. In fact, this word has seldom been used in the literature.
However, an equivalent term, fluid-structure interaction (FSI) is used to represent the
interaction of some movable or deformable structure with an internal or surrounding
fluid flow. Fluid-structure interactions can be either stable or oscillatory. In oscilla-
tory interactions, the strain induced in the solid structure causes the solid to move
such that the source of strain is reduced. Sometimes the structure returns to its former
state so that the process can be repeated. This multiphysical process will possibly
attract more and more attention as we move from lands to seas and oceans.
Thermoelectricity has two major categories of applications: thermoelectric effect
and Joule heating. The thermoelectric effect is the direct conversion of temperature
differences to electric voltage and vice versa. A thermoelectric device creates a
voltage when there are different temperatures on two sides. Conversely, when a
voltage is applied to it, it creates a temperature difference. At the atomic scale, an
applied temperature gradient causes charge carriers in the material to diffuse from
the hot side to the cold side or vice versa. This is actually how thermocouples work
in porous materials. Joule heating provides a theoretical basis for the measurement of
thermal properties of porous materials. It is also of critical significance to many
electrical devices consisting of porous materials.
Thermoelectromagnetics includes microwave heating and induction heating.
They are of concern whenever electromagnetic waves with high energy exist in
porous materials. This would especially be an issue as the world is filled with more
and more types of microwaves. An extremely significant issue is the influence of
electromagnetic waves on biomaterials, which are another big category of porous
materials. In addition, this term also refers to the magnetism in non-isothermal
conditions and newly found phenomena such as thermoelectromagnetic convection
in liquid steel.
34 4 Multiphysics in Porous Materials

Magnetomechanics, electromechanics, and electromagnetomechanics are closely

related. These three types of multiphysical phenomena represent the interaction
between magnetostatics, electrostatics, and electromagnetics with mechanical
responses. These three mechanisms are the fundament of microelectromechanical
systems. They also provide a theoretical basis for the development of most electronic
sensors and smart materials and systems. Porous materials are both the raw materials
and the areas of application of these sensors and systems.
Chapter 5
How to Use This Book

Introduction

This chapter provides an overview of the organization of the book and suggestions to
readers with different backgrounds and interests regarding how to use the book. We
will first summarize the existing literature on multiphysics, the normal ways to learn
about multiphysics, and especially, the lack of a general book on multiphysics. In
this way, the users can better understand the purpose and organization of this book.
The book is organized to reflect the understanding of multiphysics as a fusion of
mathematics, physics, applications, and numerical analysis, which led to the major
parts of the book: introduction, mathematics, monolithic physics, multiphysics, and
numerical implementation. This book is designed for people with diversified back-
grounds, from college-level students, engineers, and numerical simulation practi-
tioners to experts in specific multiphysics applications, as a textbook, a reference
book, or a technical book. Accordingly, detailed suggestions will be provided for the
use of the book for different purposes.

Organization of the Book

This book is intended to be a general reference for multiphysics. By now, a dozen of

books can be found whose title includes multiphysics. However, they are mostly
conference proceedings, introductions to the development of a specific multiphysics
application, or summary on multiphysics in a specific area. A relevant fact is that
multiphysics researchers usually tend to focus on the multiphysics applications in
their own areas, no matter in a general multiphysics context or in porous materials.
There is no such an effective reference when people would like to know about
multiphysics, especially to get a comprehensive understanding and a big picture of
this topic of increasing interest. While an overall understanding is usually very

© Springer International Publishing AG, part of Springer Nature 2018 35

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_5
36 5 How to Use This Book

difficult, a common way to get familiar with the topic is to check papers and online
materials in the area related to the intended multiphysics applications. Therefore,
knowledge is disseminated in a very ineffective way due to the lack of a compre-
hensive multiphysics book. Some commercial software provides possibly the most
comprehensive introductions as the developers have a much broader interest than
researchers specialized in any single area. However, such interest tends to lean more
toward industrial needs and possibly lags a little behind the front edge in many
research areas. In addition, the documentation is also not that comprehensive as most
of it is designed for users rather than developers or researchers.
This book is designed to be a comprehensive introduction to the area of
multiphysics in porous materials. As introduced, porous materials are an ideal
breeding ground of multiphysics due to their multiphase, multicomponent, and
possibly amorphous structures. Their ubiquitous existence also makes the
multiphysics in them closely related to human lives. It is also necessary to point
out that porous materials are related to most multiphysical phenomena. For example,
the coexistence of thermal, hydrological, and mechanical processes is often of the
most interest in geotechnical engineering, hydrogeology, hydrological engineering,
soil science, and petroleum engineering as well as many other energy application
domains. Particle transport and chemical reactions together with heat transfer and
water movement have been extensively studied in environmental engineering,
agricultural engineering, and forestry. Electromagnetics, vibration, and acoustics
are the primary interest of geophysicists, earthquake engineers, structural engineers,
and exploratory geologists. Therefore, this book will be a handy tool to researchers
and practitioners in these areas of porous materials, no matter they are new to
multiphysics or they want to step into a multiphysics topic out of their area of
specialty or an interdisciplinary one. More emphasis is placed on porous materials to
reach the depth of research for many applications.
This book is organized in a way to serve readers with different backgrounds and
with different levels of knowledge accumulation in multiphysics. For the purpose,
this book provides materials that, the author believes, would be essential to obtain a
general understanding of the concepts, theories, and implementations of
multiphysics. It is believed that multiphysics is a fusion of mathematics, physics,
applications, and numerical analysis. So the book is roughly organized in this way.
First, a general introduction to the history, state of the art, the idea of multiphysics,
and its applications in porous materials are offered in the first part. Right after this
part, necessary mathematical backgrounds such as tensor and fields, tensor analysis,
partial differential equations (PDEs), and numerical solutions of PDEs are provided
in the second part. For physics and applications, a bottom-up way is taken for the
introduction. The most representative monolithic physics is introduced in Part III, for
which each monolithic physical phenomenon is presented as a chapter. Then in Part
IV, typical multiphysical processes are introduced. The introduction is made based
on the mathematics in Part II and the monolithic physics in Part III. Each
multiphysical topic is introduced in the order of background, theory, implementa-
tion, and applications. The theory will be discussed in a way that mathematics and
physics are integrated. Implementations are demonstrated with a simplified but
Suggestions to Readers 37

typical example, which is solved using a PDE solver. In Part V, a brief introduction
will be made to show how to implement a multiphysics model using three typical
numerical simulation methods: finite difference method, finite volume method, and
finite element method.
The focus of the book is “physics,” which is currently discussed a little, rather
than mathematics and numerical methods, which have been more thoroughly
discussed in many other monographs. Therefore, Part II for mathematics and Part
V for numerical methods will only provide the most fundamental and necessary
components. For example, the mathematics will be restrained to Cartesian coordi-
nates which are mostly encountered in applications. In numerical methods, accuracy
and error analysis are only briefly or even not discussed, and only the simplest
schemes and element types will be introduced. For the introduction to these numer-
ical methods, there are several ways. But in this book, only the simplest and most
straightforward way in the author’s opinion is taken. The purpose is to provide a
broad but focused picture with minimal effort.

Suggestions to Readers

As indicated, this book is prepared for readers with different backgrounds. Here are a
few suggestions to potential users:
1. The necessary background for this book is college-level mathematics on calculus
and linear algebra and a little understanding of computer simulations. If some of
these are missing, it is suggested that the readers take some time to read or review
these parts. For calculus and linear algebra, most textbooks should be able to fill
the gaps. For computer simulations, many introductory materials such as
Wikipedia may help fill the vacancy.
2. Given that you have the above knowledge but do not know much of multiphysics,
then possibly reading the books sequentially is a good choice. If you would like to
know all about multiphysics, then try to read the chapters one by one. However,
this is possibly unnecessary for most people. This is because multiphysics is a
very interdisciplinary area and it has not been developed into an independent
discipline. Thus most people from a specific area are interested in some applica-
tions that are closely related to their area. Cracking a book with equal effort at all
the chapters may turn out to be unnecessary.
3. If you are a researcher from a specific area, you possibly know the general idea of
multiphysics and have the background in mathematics. Then you possibly would
like to know more about a specific application or several applications. Then it is
advised that you quickly go through Part I if necessary, skip Part II if possible,
and then directly jump into those chapters for monolithic physics and
multiphysics related to your applications. The introduction to these specific topics
is made in the premise of striking a balance between generality, depth, ease of
understanding, and constraints of length. As a result, there may be a distance from
38 5 How to Use This Book

the presented materials and your topic or the front edge of the research. However,
the distance should be able to be shortened with a limited amount of work on
recent publications specified for the topic.
4. If you are an engineer, who is interested in some applications. Depending on
whether you know the concept of multiphysics and the required physics, you may
decide whether and how you should read Part I and Part II. For applications, it is
suggested that you start from the example provided at the end of the chapters
corresponding to the application that you are interested in after reading the
theories. You should be able to consider more complicated or specified cases
by further integrating and/or expanding the code.
5. If you are a numerical analysis practitioner or a software developer, you may want
to go directly to chapters in Part III and Part IV. The theories in these chapters are
ready for implementation. However, more and accurate axillary relationships may
be needed if down-to-earth applications or computer programs are targeted at.
6. If you are an expert in multiphysics, you possibly agree that it would be enjoyable
to have such a reference in the playground, though the book is still far from
adequate due to the author’s limited vision and capacity. The author had been
waiting for such a book for several years and, at the same time, could not stop
reviewing his own notes. He finally decided to make one himself and share the
knowledge with anyone who may have the same needs. You possibly once had
the same feeling or thought. So it would be a great encouragement if you can pay
a smile and, if possible, leave your comments. The authors strongly desire to
improve this book with you.
7. All in all, the author would like to prompt the spirit of sharing. Some materials are
excluded from the book, such as problems for learners, due to a variety of
concerns. But please feel free to contact the author for these materials. Also,
please do not hesitate to contact the author if you have anything to share or
anything you would like to put into the next version of the book. Any feedback on
the grammatical, editorial, technical aspects of the book will be sincerely
welcomed.
 Part II
Mathematical Background
Chapter 6
Tensor and Field

Introduction

As introduced in Part I, the physics that we are dealing with in multiphysics is

classical physical fields that can be described using mathematical languages such as
tensors and partial differential equations. Part II of this book is intended to present
such necessary mathematical languages to enable later introductions to monolithic
physics and multiphysics.
In this chapter for tensors and fields, we will first explain scalars, vectors, and
tensors from both the physical and mathematical aspects. The explanation of the
tensor will start with the definition and understanding of this concept with the
assistance of daily examples and end with the operations and changes of tensors.
Next, we will discuss the physical field including its definition; relevant concepts
such as coordinates, domain, and grid; and the materialization and storage of the
physical field in the computer.

From Scalar, Vector to Tensor

A scalar represents a physical quantity that can be completely described using a real
number. As shown in Fig. 6.1, the expressions of its component are independent on
the choice of the coordinate system. Examples include temperature, mass, density,
potential, and so on. In contrast, a vector can be used to describe a physical quantity
that has both a direction and a length. The expression of its components is dependent
on the choice of the coordinate system. Examples include displacement, velocity,
force, heat flow, and so on. Tensors are a generalization of scalars and vectors: a
scalar is a zeroth-order tensor, while a vector is a first -order tensor. While orders of
tensors can vary from 0 to infinite, the orders of the tensors that we encounter in
multiphysics usually do not exceed 4. The tensors that we mostly see are

© Springer International Publishing AG, part of Springer Nature 2018 41

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_6
42 6 Tensor and Field

Fig. 6.1 Understanding and

structure of scalar, vector,
and second-order tensor

second-order tensors that define an operation for transforming a vector to another

vector. A second-order tensor can be visualized as a two-dimensional array (matrix),
while tensors of the third or higher order are difficult to visualize. In a narrow sense,
tensors only refer to those with an order of two or higher.
The concept of field or physical field is frequently used with scalars, vectors, and
tensors. In general, a scalar field associates a scalar value to every point in a space
(and time if it is involved). The scalar at a point is a real number quantifying the state
or property at that point. Examples of scalar fields in applications include the
temperature distribution throughout a space and the concentration of a solute in a
liquid. These fields are the subject of scalar field theory. A vector field is an
assignment of a vector to each point in a subset of a space. A vector field in a
plane, i.e., 2D, can be visualized as a collection of arrows with a given magnitude
and direction assigned at every point in the plane. Vector fields are often used to
describe the dynamics of a fluid moving throughout a space under the influence of
some forces such as the magnetic or gravitational force. The magnitude and direction
of the vector quantity may vary from point to point.
The operations and changes of tensors including scalars and vectors are usually of
interest as they the mathematical languages to describe the operations on and the
physical mechanisms underlying a multiphysical process. The operations and
changes of scalars are studied in the context of arithmetic and calculus, which are
usually introduced in fundamental and advanced mathematics, respectively, and will
not be discussed in this book. In the following chapters of Part II, the operations and
changes of vectors will first be introduced sequentially as vector algebra and vector
calculus. Tensor analysis including its algebra and calculus will be briefly introduced
later as an extension to vector analysis. We will stay with the fundamental but most
useful contents to multiphysics for the sake of simplicity and clarity. As a result,
concepts such as covariance and contravariance, metric coefficients, and curvilinear
coordinate systems, which are more advanced and less common in applications of
multiphysics, will be excluded. A useful notation, index notation, which is different
from the vector notation used in the first few sections of this chapter, will also be
introduced. In addition, knowledge of partial differential equations needed for
multiphysics studies will be presented. Finally, tools for the numerical solution of
PDEs will be introduced.
Physical Field 43

Physical Field

Similar to tensors, the introduction to physical fields should also include both
physical and mathematical aspects. The physical implications of physical fields
have been discussed in Part I. The general physical meaning of a physical field is
the values of a state variable associated with spatial coordinates, which can change
over time in a transient problem or be stable in an equilibrium problem. The state
variable can be a scalar, a vector, or higher-order tensors. The spatial coordinates can
be a scalar (1D), a vector with two (2D) or three numbers (3D) at any point within the
computational domain. As the computational domain is usually discretized
(meshed), the physical fields can be visualized as the associated presentation of
state variables and its corresponding spatial coordinates. The spatial coordinates are
determined by means of the discretization methods, which are usually defined at the
grid points, vertexes of elements, centroids of volume surfaces, or even points of
Gauss integration. Usually, grid points are used for FDM and FVM, and vertexes are
used in FEM. The result of a discretization is a finite number of points/elements and
consequently a collection of a finite number of coordinates.
Both the state variables and associated coordinates can be depicted using arrays or
matrices. However, the definitions of arrays and matrices could be different
depending on the context. Both terms are commonly used to refer to
two-dimensional datasets. But different conventions are adopted by people from
different areas. For example, people working with images tend to call
two-dimensional datasets arrays, while mathematicians tend to use matrices. While
these two terms are interchangeable, the different conventions may still lead to
confusion.
It is helpful to understand the relationship between the mathematical definitions
of matrix, data structure, and computer storage. Take a two-dimensional dataset with
dimensions m and n, for example – this dataset is stored in a one-dimensional set of
memory addresses in a computer. It is the indices used by the software to access the
data in memory that make the data “two-dimensional.” Further understanding of the
process of mapping multidimensional data into one-dimensional computer memory
needs more in-depth and specified materials. Strictly, a matrix can be defined as a
rectangular array: a two-dimensional array of numbers, symbols, or expressions,
arranged in rows and columns, which is treated in certain prescribed ways. One such
way is to state the order of the matrix. For example, the order of the matrix below is
2  3 because there are two rows and three columns. The individual items in a matrix
are called its elements or entries.
 
5 2 1
:
13 14 0

If two matrices have the same size or order, in other words, have the same number
of rows and the same number of columns, then the two matrices can be added or
44 6 Tensor and Field

u1x x1 y1
u1 (x1, y1) u1 (x2, y2)
u1y x2 y2

u2x x3 y3

u2y

Fig. 6.2 Storage of matrices in multiphysics

subtracted element by element. However, it would be better to use arrays for cases
where there are more than two dimensions.
The description of a physical field is straightforward but could vary in different
programs. First, the dimension of arrays is related to the order of vectors. For
example, in a 2D water flow problem, the state variable, velocity, is a vector
containing two elements. Therefore, the enumeration of the values of state variables
at all points, say N, could be N by 2 or 2 by N matrix (or array) as shown in Fig. 6.2.
However, in many cases, a 2 N by 1 or a 1 by 2 N matrix is used instead.
Accordingly, the coordinates at every point also involve two numbers. Hence, the
enumeration of the coordinates can be described in the same way. If the same type of
arrangement is used for the variables and coordinates, then they can be associated in
an easy way. But if they are described in distinct ways, rules to correlate them need
to be defined so that computer programs can establish the mapping from one to the
other. It is less common but also possible that variable values and coordinates can be
combined and indexed based on the point number. All in all, the word associated in
the definition of the physical field is very loosely defined.
Chapter 7
Tensor Analysis

Introduction

The chapter presents the basic mathematical knowledge for tensors, including both
vectors and higher-order tensors. Tensor analysis is frequently called tensor calculus.
But it actually includes both algebra, which involves the basic operations, and
calculus, which involves more advanced differential and calculus. This chapter
first gives out the notation for the description of tensors of different orders using
different symbols. Considering that vectors are a special case of tensors and are the
ones that we are more familiar with, we will start from the introduction to the algebra
and calculus of vectors. Extensions to higher-order tensors will be made based on
that. This extension covers both differential and integration of tensors. Important
identities in vector calculus will also be given out to facilitate the introductions to
monolithic physics and multiphysics theories in later chapters.

Notation

A notation defines the way to describe a mathematical term using symbols. Thus we
first need to address this most fundamental element of the mathematical language for
tensors. Using vectors as an example, we here define three notations that will be used
throughout this book and in most multiphysics literature: tensor notation, full
notation, and array notation. One more notation, i.e., index notation, will be intro-
duced in the last chapter of this part of the book. The tensor notation uses distinct
symbols for different types of tensors. In this book, a scalar is written in italics, e.g.,
α; a vector is usually represented by a bold English letter, e.g., u; and a tensor can be
a bold Greek letter or a bold capital English letter, e.g., σ and A. In handwriting, it is
!
recommended to use italics, italics with an arrow, and italics with a bar, i.e., u, u , and
u
, for scalars, vectors, and higher-order tensors, respectively.

© Springer International Publishing AG, part of Springer Nature 2018 45

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_7
46 7 Tensor Analysis

We can conveniently write down physical quantities with different natures and
physical mechanisms/processes involving these quantities using the tensor notation.
This is in fact what we usually do. However, such practice requires the users to have
a good understanding of the meanings of these symbols and their operation. Unfor-
tunately, we cannot directly read the meanings from the symbols without training.
To better illustrate what tensors and their operations stand for, we can employ the
full notation and the array notation.
In the full notation, a scalar is a real number while a vector is the sum of the
products of a number and a basis vector. Considering a three-dimensional Euclidean
space, then there will be an arbitrary basis (e1,e2,e2) associated with the three
dimensions. In general, the basis vectors are unnecessarily unit vectors nor mutually
orthogonal. However, they must be linearly independent. Then a vector u can be
expressed as

u ¼ u1 e 1 þ u 2 e 2 þ u3 e 3 : ð7:1Þ

The Einstein summation convention is usually used with the full notation to achieve
notational brevity. A term with an index variable appearing twice implies summation
of that term over all the values of the index. Accordingly, the above vector can be
written as

u ¼ ui e i : ð7:2Þ

In a three Cartesian coordinate system, i will enumerate from the index of the first
axis, 1, to that of the third axis, 3.
The array notation is also helpful for understanding the meaning of tensors and
their operations, especially for visualization. For example, the above vector can be
represented using an array such as

u ¼ ½ u1 ; u2 ; u3  ð7:3Þ

or
2 3
u1
u ¼ 4 u2 5: ð7:4Þ
u3

Either a row or a column array works. However, operations with the tensor need to
be defined accordingly to ensure consistency.

Vector Algebra

The basic algebraic operations, i.e., non-differential, in vector calculus are referred to
as vector algebra. These operations are defined for a vector space and globally
applied to a vector field associated with space. Basic algebraic operations consist of:
Vector Algebra 47

1. Vector addition: addition of two vector fields, yielding a vector field: u + v

2. Scalar multiplication: multiplication of a scalar field α and a vector field u,
yielding a vector field: αu
3. Dot product: multiplication of two vector fields, yielding a scalar field: u  v
4. Cross product: multiplication of two vector fields, yielding a vector field: u  v
5. Scalar triple product: multiplication of three vectors with two different multipli-
cation operators, yielding a scalar field: u  (v  w)
6. Vector triple product: multiplication of three vectors with two cross product
operators, yielding a vector field: u  (v  w)

Vector Additions and Scalar Multiplication

Vector addition can be understood via the resultant of two forces. When two vectors
are placed head to tail, their sum is the diagonal pointing from the free tail to the free
head. As shown in Fig. 7.1, the subtraction operation is similar. Therefore, the
operation can be represented using the full notation and array notation in the
following way:

u þ v ¼ ui ei þ vi ei ¼ ðui þ vi Þei , ð7:5Þ

u þ v ¼ ½u1 þ v1 ; u2 þ v2 ; u3 þ v3 : ð7:6Þ

Basic laws of algebra, i.e., associative, commutative, and distributive laws, are
valid for scalar multiplication and vector addition whenever applies. The above
equations show that the vector addition is the addition of corresponding components.
Since the additions of the corresponding components obey the commutative law, the
commutative law holds for vector addition:

u þ v ¼ v þ u: ð7:7Þ

Similarly, we can easily obtain the associative law

ð u þ vÞ þ w ¼ u þ ð v þ w Þ ð7:8Þ

is valid. As a result, the parentheses in the above equation can be omitted without
any ambiguities. The distributive law does not apply to the vector addition.
In scalar multiplication, a scalar will be multiplied with each component of the
vector. That is, αu or uα is defined as a vector which has a length of |α||u| and a
direction identical to that of u when α is positive and opposite to that of u when α is

Fig. 7.1 Schematic of

vector addition and
subtraction
48 7 Tensor Analysis

negative. Due to the reason,u and u are vectors equal in the magnitude but opposite
in the direction. Scalar multiplication obeys all of the three rules:

αu ¼ uα ð7:9Þ

αðβuÞ ¼ ðαβÞu ð7:10Þ

ðα þ βÞu ¼ αu þ βu ð7:11Þ

αðu þ vÞ ¼ αu þ αv: ð7:12Þ

As can be seen, these laws are identical to those in ordinary algebra.

Products of Vectors

The multiplication of vectors is different from that of scalars. Especially, two types
of products are defined: dot product and cross product.
The dot or scalar product of two vectors u and v, written as u  v, produces a real
number |u|  |v| cos (u, v):

u  v ¼ juj  jvj cos ðu; vÞ, ð7:13Þ

where (u, v) denotes the angle between the directions of u and v. As shown in
Fig. 7.2, the dot product of two vectors thus can be understood as the elongated
(according to the magnitude of the vector along the projection direction) projection
of one vector on the direction of the other.
If u and v are at right angles, then u  v ¼ 0; and if neither u nor v is a zero vector,
then a zero dot product indicates that the two vectors are perpendicular. If u ¼ v then
cos(u  v) ¼ 1, and u  u ¼ |u|2 gives the square of the length of u. It is also called
scalar product in the context of tensor analysis.
The cross product can also be formulated using the full notation as follows:
   
u  v ¼ ðui ei Þ  v j e j ¼ ui v j ei  e j : ð7:14Þ

In Cartesian coordinate systems, the axes are perpendicular to each other; therefore,
the dot product of unit factors, ei  ej, is 0 when they are different vectors and is
1 when they are identical. The Kronecker delta is defined as

Fig. 7.2 Schematic of dot

and cross products
Vector Algebra 49


0 i f i 6¼ j
δij ¼ ei  e j ¼ ð7:15Þ
1 if i ¼ j

The vector form of the Kronecker delta is the identity matrix, which has ones on the
main diagonal and zeros elsewhere. The symbols δ and I are reserved for this matrix
in this book.
Using the Kronecker delta, the dot product can be represented as
 
u  v ¼ ui v j δij : ð7:16Þ

In the above summation, any product with i 6¼ j will be equal to 0; therefore, the
above equation can be further simplified into

u  v ¼ ui vi : ð7:17Þ

This way of simplification to remove δij is very useful and widely used in deductions
in multiphysics. The above equation, in fact, gives out the definition of dot product
using the array notation:
2 3
v1 X
u  v ¼ ½ u1 u2 u3  4 v 2 5 ¼ ui v i : ð7:18Þ
v3

It is identical to Eq. 7.17 considering Einstein’s summation convention used with the
full notation.
The commutative and distributive laws of elementary algebra are valid for the dot
multiplication of vectors. The associative law does not apply here because the dot
product of two vectors is a scalar, which is not able to form a dot product with
another vector.
The cross or vector product of two vectors u and v, written as u  v, is defined as

u  v ¼ njuj  jvj sin ðu; vÞ, ð7:19Þ

where n is a vector of a unit length and a direction perpendicular to the plane of u and
v. Therefore, the cross product of two vectors generates another vector (Fig. 7.2).
The direction of this vector can be obtained using the right-hand rule in which a
right-handed screw that rotates from u toward v will advance in the direction of n. If
u and v are parallel, then u  v ¼ 0. The magnitude of u  v is identical to the area of
the parallelogram which has u and v as two adjacent sides.
The cross product can be defined using the full notation as
   
u  v ¼ ðui ei Þ  v j e j ¼ ui v j ei  e j : ð7:20Þ

In a Cartesian coordinate system, the special nature of ei  ej will lead to the

definition of the Levi-Civita symbol, εijk:
50 7 Tensor Analysis

8
< 0 if i ¼ j, or j ¼ k, or k ¼ i
εijk ¼ þ1 if ði; j; k Þ is ði; j; k Þ, ði; j; kÞ, or ði; j; kÞ ð7:21Þ
:
1 if ði; j; k Þ is ði; j; k Þ, ði; j; kÞ, or ði; j; kÞ

Then the cross product can be simplified into

u  v ¼ ui v j ek εijk : ð7:22Þ

Using the array notation, we will obtain the following definition of the cross product:
 
 ei e j ek 
 
u  v ¼  ui u j uk : ð7:23Þ
 vi v j vk 

The associative law does not hold for the cross product operation. Instead, we can
use Lagrange’s formula and the Jacobi identity, which will be given out in the next
section for the significant identities in tensor analysis. In fact, the cross product of
three vectors is usually treated as a separate operation, which will be introduced later
in this section. Although the associative law does not apply, the following equation
is valid:

ðAuÞ  v ¼ Aðu  vÞ, ð7:24Þ

where A can be a tensor of any order: Aa or A(a  b) is a scalar product when A is a

scalar and an outer product when A is a vector and higher-order tensors.
The distributive law is valid for cross products:

u  ðv þ wÞ ¼ u  v þ u  w: ð7:25Þ

However, the cross product is not commutative. This is because the rotation from
u to v is opposite to that from v to u:

u  v ¼ v  u: ð7:26Þ

Hence, a counter-commutative law is instead valid for the cross product.

Vector Calculus

Vector calculus is the mathematical study of the change of vectors. Similar to the
calculus of scalars, vector calculus includes differential calculus and integral calcu-
lus. Changes of the vector, both differential and integral, are formulated through
several vector operators. Vector operators are defined in terms of del. Del is a vector
differential operator, usually represented by the nabla symbol — . In a three-
dimensional Cartesian coordinate system with coordinates (x1, x2, x3) and a standard
basis (e1, e2, e3), del is written as
Vector Calculus 51

 
∂ ∂ ∂ ∂ ∂ ∂ ∂
∇¼ ; ; ¼ e1 þ e2 þ e3 ¼ ei , ð7:27Þ
∂x1 ∂x2 ∂x3 ∂x1 ∂x2 ∂x3 ∂xi

where ei is the unit vector along the direction of the coordinate axes i and i¼1, 2, or 3.
Other common vector operators include gradient, divergence, and curl, which are
defined using del in the following.
1. Gradient is a generalization of the concept of derivative of a function in one
dimension into a function in several dimensions. The operator maps a scalar field
to a vector field:

grad ¼ ∇
∂u ∂u ∂u ∂u
∇u ¼ e1 þ e2 þ e3 ¼ ei
∂x1 ∂x2 ∂x3 ∂xi ð7:28Þ
 
∂u ∂u ∂u ∂ u je j ∂u j
∇u ¼ e1 þ e2 þ e3 ¼ ei ¼ ei e j
∂x1 ∂x2 ∂x3 ∂xi ∂xi

2. Divergence is a vector operator that measures the magnitude of the source or sink
in a vector field at any given point, in terms of a signed scalar. The operator maps
a vector field to a scalar field:

div ¼ ∇  
∂ u je j ∂u j ∂u j ∂ui ð7:29Þ
∇u¼  ei ¼ ei  e j ¼ δij ¼
∂xi ∂xi ∂xi ∂xi

3. Curl is a vector operator that describes the infinitesimal rotation of a three-

dimensional vector field. The operator maps a vector field to another vector field:

curl ¼ ∇
 
 ei ej ek 
 
 ∂ ∂ ∂  ∂ui ð7:30Þ
∇  u ¼  ¼ ek εijk
 ∂xi ∂x j ∂xk  ∂x j
 u uj uk 
i

4. Laplacian measures the difference between the value of the scalar/vector field and
its average on infinitesimal balls. The operator maps between scalar/vector fields:

∇2 ¼ div grad ¼ ∇  ∇ ¼ Δ ð7:31Þ


∂u
∂ ∂x j
ej 2
∂ ui
Δu ¼  ei ¼ ð7:32Þ
∂xi ∂x2i
52 7 Tensor Analysis

Therefore, these operators could change the order of the vectors that they act on. The
order of tensors is a very useful tool to examine the consistency of governing
equations and physical terms in multiphysics in a way similar to units.
Vector operators need to be placed right in front of a scalar or vector to take effect.
Therefore — f represents the gradient of f while f— just produces another vector
operator, which does not act on anything. Also, as just mentioned, a vector operator
can operate on another vector operator to produce a new compound vector operator.
Finally, the curl and divergence differ because the former uses a cross product while
the latter uses a dot product.

Theorems

Three common theorems in calculus, i.e., Green’s theorem, Stokes’ theorem, and
Gauss (divergence) theorem, are frequently encountered in the derivation of the
equations in multiphysics. These theorems for the calculus of scalars can be gener-
alized to higher-order tensors.
Green’s theorem states that the integral of the scalar curl of a vector field over
some regions in a plane equals the line integral of the vector field over the closed
curve bounding the region oriented in the counterclockwise direction. This theorem
can be formulated using vector operators as
Z I
∇  FdS ¼ F  ndL: ð7:33Þ
S ∂S

In Stokes’ theorem, the integral of the curl of a vector field over a surface in R3
equals the line integral of the vector field over the closed curve bounding the surface.
The corresponding mathematical formulation is
ZZ I
∇  F  dσ ¼ F  dr: ð7:34Þ
P
σR 3
∂

The divergence theorem says that the integral of the divergence of a vector field
over a volume equals the integral of the flux through the closed surface bounding the
volume, which is formulated as
ZZZ ZZ
ð∇  FÞdV ¼  F  dS: ð7:35Þ
VR3 ∂V

It is worthwhile to point out that the above three theorems can be formulated
using a general form:
Basics of Tensor Analysis 53

Z Z Z
ð∇∗f ÞdΩ ¼ ðn∗ f ÞdΩ ¼ dΩ∗ f, ð7:36Þ
Ω ∂Ω ∂Ω

where ∗ is used to represent any tensor product, i.e., inner, outer, and cross products.

Basics of Tensor Analysis

Tensor analysis can be understood as the extension or generalization of the vector

calculus to higher-order tensors. Therefore, it is concerned with relations or laws that
remain valid regardless of the coordinate system for describing the quantities. In fact,
tensor analysis includes the definition, formulation, operation, differentiation, and
integration of tensors from scalars to higher-order tensors in various coordinate
systems in a general sense. Multiphysics deals with physical fields, which involve
the changes of tensors of different orders. Therefore, tensor analysis is an essential
tool and necessary language in multiphysics. Tensor analysis is widely used in most
science and engineering topics that require a continuum description. Due to time
constraints, only the most relevant and critical content is included here. More details
will be provided in later chapters whenever necessary.

Definition

Tensors are a generalization of scalars and vectors. Many different definitions from
different viewpoints are available, such as multilinear algebra and differential
geometry. But in a simple way, tensors can be just viewed as arrays of numbers or
functions, which transform according to certain rules under a change of coordinates.
Tensors can also be understood as geometric objects that describe linear relations
between vectors, scalars, and other tensors. Examples of such relations include the
dot product, the cross product, and the linear maps. In physics, tensors are used to
characterize the properties or describe the state of a physical system.
For a simple and smooth introduction, let us first get back to the definition of first-
order tensors, i.e., vectors. In a three-dimensional Euclidean space with an arbitrary
basis (e1, e2, e3), a vector u can be expressed as

u ¼ u1 e 1 þ u2 e 2 þ u3 e 3 ¼ ui e i : ð7:37Þ

Similarly, for a second-order tensor, two sets of bases will be needed to reflect the
order of “two.” If we use mi and nj for the two sets of basis, a general second-order
tensor A can be written as

A ¼ aij mi n j : ð7:38Þ
54 7 Tensor Analysis

Similar to vectors, we can also use two-dimensional arrays (matrices) to represent

and visualize tensors. For example, a second-order tensor in this case can be
visualized as a two-dimensional (or second rank) matrix:
2 3
a11 a12 ... a1n
6 a21 a22 ... a2n 7
A¼6
4 ...
7:
5 ð7:39Þ
...
am1 am2 . . . amn

The above notation only includes the components while omitting the basis.
Therefore, the implication of the matrix notation is that each element will go with
an invisible basis vector. An element is the projection of the tensor along the
direction of its associated basis vector.
When the order of a tensor gets beyond 2, it is usually not easy to visualize tensors
in terms of matrix anymore. But for scalars, vectors, and second-order tensors, which
are extensively used in FDEs, matrices are still a good tool to help us understand the
structures of tensors and the operations between them.

Tensor Product

The tensor product is a type of operation between tensors. This operation also applies
to vectors, though we skipped it in the section for vectors. In linear algebra, the term
outer product is typically used to refer to the tensor product of two vectors. In the
dyadic context, dyadic product, outer product, and tensor product all share the same
definition and meaning and thus are used synonymously. However, the tensor
product is the most general and abstract term among them.
There are several equivalent terms and notations for this product:
1. The dyadic product of two vectors u and v is denoted by the juxtaposition of them.
2. The outer product of two column vectors u and v is denoted and defined as u  v
or uvT, where T means transpose.
3. The tensor product of two vectors u and vis denoted by u  v.
The above usages can be proven to be equivalent. Consider a three-dimensional
Euclidean space, in which we have the two following vectors:

u ¼ u1 e 1 þ u2 e 2 þ u3 e 3 ð7:40Þ

v ¼ v1 e1 þ v2 e2 þ v3 e3 , ð7:41Þ

where e1, e2, and e3 are the standard basis vectors in this vector space. Then the
dyadic product of u and v can be represented as a sum:
Basics of Tensor Analysis 55

uv ¼ u1 v1 e1 e1 þ u1 v2 e1 e2 þ u1 v3 e1 e3
þ u2 v 1 e 2 e 1 þ u2 v 2 e 2 e 2 þ u2 v 3 e 2 e 3 ð7:42Þ
þ u3 v 1 e 3 e 1 þ u3 v 2 e 3 e 2 þ u3 v 3 e 3 e 3 :

While using row and column vectors, the result of the outer product or tensor product
of u and v is a 3  3 matrix:
1 0 0 1
u1 u1 v 1 u1 v 2 u1 v 3
uv ¼ u  v ¼ uvT ¼ @ u2 Að v1 v2 v 3 Þ ¼ @ u2 v 1 u2 v 2 u2 v 3 A : ð7:43Þ
u3 u3 v 1 u3 v 2 u3 v 3

These two operations are thus essentially equivalent.

Unlike tensor products and outer products, which apply to the products of two
tensors of any order, the dyad product is usually confined to the product of two
vectors. In the dyad product, a dyad, e.g., u1v1, is a component of the dyadic uv. The
dyad thus can be viewed as a monomial of the sum, equivalent entry of the matrix, or
the juxtaposition of a pair of basis vectors scalar multiplied by a number.

Identities and Algebra

Tensor product is associative and distributive but not commutative.

As for the associative law, we have

ðuvÞw ¼ uðvwÞ ¼ uðvwÞ: ð7:44Þ

The associative low is compatible with scalar multiplication for any scalar α:

ðαuÞv ¼¼ αðuvÞ ¼ uðαvÞ: ð7:45Þ

The term dyadic algebra is introduced as an extension of vector algebra to include

the dyadic product of vectors. In this case, dyadic algebra, dot product, and cross
product may coexist. When working with dot and cross products, four operations can
be defined on a vector and dyadic with the three forms of vector products. As shown
in the following table, the associative law still applies for the coexistence of these
products.

Left Right
Dot product c  ab ¼ (c  a)b ab  c ¼ a(b  c)
Cross product c  ab ¼ (c  a)b ab  c ¼ a(b  c)

When dealing with higher-order tensors, we can still take advantage of the above
rules by treating a higher-order tensor as the juxtaposition of multiple vectors. A
third-order tensor, A ¼ Aijkeiejek, for example, can be treated as the juxtaposition, or
tensor product, of three vectors:
56 7 Tensor Analysis

 
A ¼ Aijk ei e j ek ¼ uvw ¼ ðui ei Þ v j e j ðwk ek Þ, ð7:46Þ

in which Aijk ¼ uivjwk.

The tensor product operation obeys the distributive law:

uðv þ wÞ ¼ uv þ uw: ð7:47Þ

It is noted that the tensor product operation is not commutative, that is,

uv 6¼ vu: ð7:48Þ

Contraction and Inner Product

The inner product is another operator in need of further explanation for tensor
algebra. The inner product is used interchangeably with the dot product in many
occasions. Their difference lies in that the inner product generalizes the dot product
to abstract vector spaces over a field of scalars, either real or complex numbers. The
definition of inner product in this way is usually written as ha, bi. In mathematics,
the dot product (or called scalar product), or the more general inner product in the
context of Euclidean space, is an algebraic operation that multiplies two equal-size
arrays of numbers, e.g., coordinate vectors, and returns a single number. The dot
product of two vectors

u ¼ ð u1 ; u2 ; . . . ; un Þ ð7:49Þ

and

v ¼ ð v1 ; v2 ; . . . ; vn Þ ð7:50Þ

is defined as

X
n
u  v ¼ u1 v 1 þ u2 v 2 þ . . . þ un v n ¼ ui v i : ð7:51Þ
i¼1

It is seen from the symbols that the major difference between outer product and
inner product is the symbol “.” which is called contraction or tension contraction.
The contract can be defined as follows:

ei  e j ¼ δij : ð7:52Þ

In a simple way, we can understand the contraction operation as relating two

vectors on two sides of the operator using the same Cartesian coordinate system.
Therefore, the two sets of bases associated with the two vectors degenerate into the
Kronecker delta. So you can see that this operation reduces the order of a tensor by
two. That is why the dot product makes the combination of two first-order tensors
Important Identities in Vector Calculus 57

(vectors, 1 + 1) into a scalar (1 + 1–2 ¼ 0). Besides dot product, two types of double-
dot products are also useful in manipulating dyads: the vertical and horizontal
double-dot products (Taber 2004):
   
σ:τ ¼ σ ij ei e j :ðτkl ek el Þ ¼ σ ij τkl ðei  ek Þ e j  el ¼ σ ij τkl δik δ jl ¼ σ il τil ð7:53Þ
 
σ  τ ¼ σ ij ei e j  ðτkl ek el Þ
ð7:54Þ
¼ σ ij τkl ðei  el Þ e j  ek ¼ σ ij τkl δil δ jk ¼ σ ik τki :

These two operators are not commutative. The rule can be seen in the following
operation:

abc : def ¼ aðbc : deÞf ¼ ðb  dÞðc  eÞaf: ð7:55Þ

Important Identities in Vector Calculus

Operator and Order of Tensor

First, a few typical examples are given in the following to show the changes that
different operators can cause to the order of the tensors that the operators act on.
div(u) ¼ —  u is a scalar
div(σ) ¼ —  σ is a vector
curl(u) ¼ —  u is a vector
grad(φ) ¼ — φ is a vector
grad(u) ¼ — u is a second-order tensor

Special Combination of Multiple Operators

∇  ð∇φÞ ¼ 0 ð7:56Þ

∇  ð∇  v Þ ¼ 0 ð7:57Þ

Distributivity over Vector Addition

∇  ð v1 þ v2 Þ ¼ ∇  v1 þ ∇  v2 ð7:58Þ

∇  ð v1 þ v2 Þ ¼ ∇  v1 þ ∇  v2 ð7:59Þ
58 7 Tensor Analysis

Product of Scalar and Vector

∇ðψφÞ ¼ φ∇ψ þ ψ∇φ ð7:60Þ

∇  ðφvÞ ¼ ð∇φÞ  v þ φð∇  vÞ ð7:61Þ

∇  ðφvÞ ¼ ð∇φÞ  v þ φ∇  v ð7:62Þ

Scalar Triple Product

The scalar triple product (also called the mixed or box product) is defined as the dot
product of one vector and the cross product of the other two:

u  ðv  wÞ:

As for the scalar triple product, there exists an important relationship:

u  ðv  wÞ ¼ v  ðw  uÞ ¼ w  ðu  vÞ: ð7:63Þ

The above relationship leads to another important deduction:

u  ðu  vÞ ¼ u  ðv  uÞ ¼ u  ðv  vÞ ¼ u  ðu  uÞ ¼ 0: ð7:64Þ

The gradient operator, — , can be treated as a normal vector in the scalar triple
product; thus

∇  ðu  vÞ ¼ u  ðv  ∇Þ ¼ v  ð∇  uÞ: ð7:65Þ

Vector Triple Product

While the scalar triple product produces a scalar, the vector triple product will lead to
a vector. The vector triple product is defined as the cross product of one vector and
the cross product of the other two. The following relationship holds:

u  ðv  wÞ ¼ vðu  wÞ  wðu  vÞ: ð7:66Þ

This is known as the triple product expansion, or Lagrange’s formula, although the
latter name is also used for several other formulae. Its right-hand side can be
remembered by using the mnemonic “BAC-CAB,” provided one keeps in mind
which vectors are dotted together.
Important Identities in Vector Calculus 59

Similar to the scalar triple product, the gradient vector can be treated as a normal
vector in the vector triple product. A vector triple product with two gradient vectors
in it leads to

∇  ð∇  uÞ ¼ ∇ð∇  uÞ  ∇2 u: ð7:67Þ

If we replace only one of the vectors with the gradient, then we can obtain

∇  ðu  vÞ ¼ uð∇  vÞ  vð∇  uÞ ð7:68Þ

∇ðu  vÞ ¼ u  ð∇  vÞ þ vð∇  uÞ: ð7:69Þ

In addition, Lagrange’s formula can lead to another important relationship, the

Jacobi identity:

u  ðv  wÞ þ v  ðw  uÞ þ w  ðu  vÞ ¼ 0: ð7:70Þ
Chapter 8
Index Notation and Tensor Notation

Introduction

The full notation and array notation are very helpful when introducing the operations
and rules in tensor analysis. However, tensor notation and index notation are more
commonly used in the context of partial differential equations and tensor analysis.
The tensor notation just requires the utilization of different symbols for tensors of
different orders and the use of appropriate symbols as operators connecting these
tensors. The tensor notation thus enables us to write PDEs in a concise way, which is
also independent of the adopted coordinate system. But in many cases, the index
notation is preferred as it is proven to be much more powerful for occasions such as
derivations. In this chapter, we will start from the basic rules of the index notation,
then move to the use of the index notation for tensor algebra, and finally reach the
calculus in terms of the index notation. At the end of the chapter, two examples will
be given to show the algebraic manipulations, i.e., derivations, using the index
notation.

Rules of Index Notation

In the index notation, indices are categorized into two groups: free indices and
dummy indices. A free index means an “independent dimension” or an order of
the tensor whereas a dummy index means summation. The following three basic
rules must be met for the index notation:
1. The same index (subscript) may not appear more than twice in a product of two
(or more) vectors or tensors. Thus

© Springer International Publishing AG, part of Springer Nature 2018 61

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_8
62 8 Index Notation and Tensor Notation

Aik uk , Aik Bkj , Aij B jk Cnk

are valid, but

Akk uk , Aik Bkk , Aij Bik C ik

are meaningless.
2. Free indices on each term of an equation must agree. Thus

ui ¼ vi þ wi ð8:1Þ

ui ¼ Aik Bkj v j þ C ik wk ð8:2Þ

are valid, but

ui ¼ Aij ð8:3Þ

u j ¼ Aik uk ð8:4Þ

ui ¼ Aik vk þ w j ð8:5Þ

are meaningless.
3. Free and dummy indices may be changed without altering the meaning of an
expression under the condition that Rules 1 and 2 are not violated. Thus

u j ¼ A jk vk , ui ¼ Aik vk , u j ¼ A ji vi :

Tensor Algebra in Index Notation

First, according to the meaning of free dimension, a vector u and a second-order

tensor A should be written as ui and Aij, respectively. The free indices can be
changed to other symbols. The basic operations in tensor algebra can be expressed
using the index notation as follows:
Addition ui ¼ vi + wi , u ¼ v + w
Dot product λ ¼ u  v, λ ¼ uivi
Vector product u ¼ v  w , ui ¼ εijkvjwk
Dyadic product A ¼ uv , Aij ¼ uivj
Addition C ¼ A + B , Cij ¼ Aij + Bij
Transpose A ¼ BT , Aij ¼ Bji
Scalar products λ ¼ A : B , λ ¼ AijBij λ ¼ A   B , λ ¼ AijBji
Calculus Using Index Notation 63

Inner product of a
tensor and a vector u ¼ Av, ui ¼ Aijvj
Inner product of two
tensors C ¼ A  B, Cij ¼ AikBkj
Determinant λ ¼ detA , λ ¼ 16 εijk εlmn Ali Amj Ank ¼ εijk Al1 A j2 Ak3,εlmn
λ ¼ εijkAliAmjAnk ¼ εijkAilAjmAkn
Inverse 1
A1
ji ¼ εipq ε jkl Apk Aql
2detðAÞ

Calculus Using Index Notation

The derivative with respect to one coordinate in the index notation can be
represented by putting the corresponding symbols for the coordinate after a
comma in the indices. The gradient of a scalar u thus can be written as

∇u ¼ u, i ð8:6Þ

where i is the free index. Therefore, u,i is still a vector since it has one free index
like ui. Accordingly, the gradient of a vector u is written as

∇u ¼ ui, j , ð8:7Þ

ui, j has two free indices; hence, it is a second-order tensor.

The divergence of a vector, which can be viewed as the combination of the
gradient vector and contraction, requires the contraction of two indices into one via
Kronecker delta. Therefore, we will have this operation in the index notation as

∇  u ¼ ui, i : ð8:8Þ

The divergence of a second-order tensor A is

∇  A ¼ Aij, j : ð8:9Þ

Likewise, the curl of a vector u is

∇  u ¼ εijk u j, k : ð8:10Þ

The Laplacian of a quantity involves both the gradient and the divergence. For a
scalar u, we apply the gradient and divergence sequentially, accordingly,

Δu ¼ u, ii : ð8:11Þ

For a vector u, we will have

64 8 Index Notation and Tensor Notation

Δu ¼ ui, jj : ð8:12Þ

The derivative ∂xi/∂xj can be deduced by noting that ∂xi/∂xj ¼ 1when i ¼ j and
∂xi
∂xi/∂xj ¼ 0 when i 6¼ j. Therefore ∂x j
¼ δij .
The same argument can be used for higher-order tensors

∂Aij
¼ δik δ jl : ð8:13Þ
∂Akl

Examples of Algebraic Manipulations Using Index Notation

1. Let a, b, c,d be vectors. Prove that

ða  bÞ  ðc  dÞ ¼ ða  cÞðb  dÞ  ðb  cÞða  dÞ: ð8:14Þ

Solution
Let us first express the left-hand side of the equation using index notation. Please
check the rules for cross products and dot products of vectors to see how this is done.

ða  bÞ  ðc  dÞ ¼ εijk a j bk εimn cm d n ð8:15Þ

Recall the identity

εijk εimn ¼ δ jm δkn  δ jn δmk : ð8:16Þ

Then we will have

εijk a j bk εimn cm d n ¼ δ jm δkn  δ jn δmk a j bk cm dn : ð8:17Þ

Multiply out, and note that

δ jm a j ¼ am

and

δkn ak ¼ an :

That is, multiplying a Kronecker delta has the effect of switching indices; hence


δ jm δkn  δ jn δmk a j bk cm dn ¼ am bn cm d n  an bm cm d n ð8:18Þ
Examples of Algebraic Manipulations Using Index Notation 65

Finally, note that amcm ¼ a  c and similarly for other products with the same
index so that

am bn c m d n  an bm c m d n ¼ am c m bn d n  bm c m an d n
ð8:19Þ
¼ ða  cÞðb  dÞ  ðb  cÞða  dÞ

2. The stress-strain relation for linear elasticity may be expressed as

E  v 
σ ij ¼ εij þ εkk δij ð8:20Þ
1þv 1  2v
where σ ij and εij are the components of the stress and strain tensor and E and v are
Young’s modulus and Poisson’s ratio, respectively. Find an expression for strain in
terms of stress.
Solution
Set i ¼ j to see that

E  v 
σ ii ¼ εii þ εkk δii : ð8:21Þ
1þv 1  2v
Recall that σ ii ¼ 3, and notice that we can replace the remaining ii by kk, so that

E  v  v
σ kk ¼ εkk þ 3εkk ¼ εkk ð8:22Þ
1þv 1  2v 1  2v
and

1  2v
εkk ¼ σ kk : ð8:23Þ
E
Now, substitute for εkk in the given stress-strain relation and we obtain

E  v 
σ ij ¼ εij þ σ kk δij ð8:24Þ
1þv E
and
 
1þv v
εij ¼ σ ij  σ kk δij : ð8:25Þ
E 1þv
Chapter 9
Partial Differential Equations

Introduction

Partial differential equations (PDEs) are another mathematical language required for
expressing multiphysics in addition to tensors. The chapter starts the introduction to
PDE with the role of PDE in the numerical simulation. Basic definitions including
dependent variables, coordinates, domain, boundary, and numerical methods will be
briefly introduced to lay down the background for the use of PDE in multiphysics.
The common classification of PDEs will be discussed next. Later, the PDEs that we
would possibly encounter in science and engineering applications, including linear,
nonlinear, and PDE systems, will be presented. Finally, boundary conditions, which
are needed for the solution of PDEs, will be introduced. Effort will be made to detail
the common types of boundary conditions, confusion about boundary conditions,
boundary conditions in 3D problems and computer treatments, and initial conditions.

Numerical Simulation and PDEs

A computer simulation is a numerical experiment to reproduce the behavior of a

system. Computer simulations can be categorized into stochastic and deterministic
simulations. Stochastic simulations adopt a collection of random variables to repre-
sent the evolution of some systems of random values over time, which will not be
discussed in this book. By contrast, a deterministic algorithm is an algorithm which,
given a particular input, will always produce the same output, which is multiphysics
simulation in porous materials subordinate to. A deterministic algorithm aims at
obtaining a mathematical function,

u ¼ uðt; x1 ; x2 ; . . . ; xN Þ, ð9:1Þ

© Springer International Publishing AG, part of Springer Nature 2018 67

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_9
68 9 Partial Differential Equations

which has a unique value for any input in its domain. The algorithm gives out a way
that produces this particular value as output (or solution). u is the unknown(s) and it
is the function of both temporal coordinate t and spatial coordinate xi. N is the
number of dimensions (spatial) associated with the problem of interest. Because an
analytical solution, which is a continuous function, is usually difficult to obtain or
even unavailable, numerical methods are usually adopted to seek a discrete form of
the solution. In other words, we can get the values of u at some specific positions and
times, which depend on pre-defined grid/mesh sizes and time steps. For the points
between the grid points or time steps, interpolations are usually employed to estimate
the function values.
In order to obtain a discrete solution (function), a collection of equations needs to
be solved numerically. There are a variety of numerical methods, within which the
finite difference method (FDM), finite volume method (FVM), and finite element
method (FEM) are the most popular ones and thus will be introduced in this course.
As introduced in the Introduction part, the equations to be solved can be catego-
rized based on their role in the description of the problem as governing equations,
auxiliary equations, and boundary/initial conditions. A typical example is the
Cauchy problems for the Laplace equation,
2 2
∂ u ∂ u
þ ¼ A, ð9:2Þ
∂x2 ∂y2
with boundary conditions,

uðx; 0Þ ¼ 0, ð9:3Þ

∂u sin ðnxÞ
ðx; 0Þ ¼ , ð9:4Þ
∂x n
and an auxiliary equation,

A ¼ x2 þ y2 , ð9:5Þ

where n is a real number defining the Cauchy problem.

This example is not time-dependent as time, t, is not involved, so we often call it a
boundary value problem (BVP) in mathematics. But in engineering applications,
when the equations are endowed with physical meanings, we usually call it a steady-
state problem. When time is involved, we call it an Initial-Boundary Value Problem
(IBVP) and a time-dependent problem in mathematics and engineering applications,
respectively. Initial value problems, in which no spatial coordinates are involved, are
not quite common as what we target at in engineering is usually a domain, 1D, 2D, or
3D, rather than a point.
The governing equations are usually partial differential equations, which describe
the major governing mechanisms underlying a physical process. When there is a
single type of phenomenon for a single type of material/object, only one governing
Classification of Partial Differential Equations 69

equation is necessary. Otherwise, multiple equations will be introduced. Initial

conditions provide some information for the function at the initial time point,
while boundary conditions lay down constraints on the boundaries. Auxiliary equa-
tions could be the relationships between any two or more of the materials properties,
dependent variables, and independent variables. In this book, we will primarily focus
on governing equations, especially partial differential equations. Initial and bound-
ary conditions are relatively easy to deal with. Auxiliary relationships are more
closely related to the research in individual disciplines. In any of the aforementioned
numerical methods, the whole equation system will be discretized over a computa-
tional domain, Ω, leading to an algebraic equation system,

K  u ¼ b, ð9:6Þ

where K is the stiffness matrix, u is the matrix of unknowns, and b is the matrix of
forces.
In PDEs, it is also common to denote partial derivatives using subscripts, that is,

∂u
ux ¼ ð9:7Þ
∂x
and
2

∂ u ∂u ∂u
uxy ¼ ¼ : ð9:8Þ
∂x∂y ∂y ∂x

Especially in physics, del (— ) is often used for spatial derivatives and u_ and €u the
first and second temporal derivatives, respectively.

Classification of Partial Differential Equations

... partial differential equations are the basis of all physical theorems.
------Bernhard Riemann (1826–1866)

A partial differential equation (PDE) is an equation consisting of a multivariable

function(s) and partial derivatives of the function(s). Similar to ordinary differential
equations (ODE), PDEs are also a sub-category of differential equations (DE). The
difference between ODEs and PDEs lies in the fact that ODEs contain a single
variable and its derivatives while PDEs deal with multiple dependent variables and
their partial derivatives. A general form of a second-order PDE for the function u(x1,
x2,   , xn) is
70 9 Partial Differential Equations

!
2 2 2
∂ u ∂ u ∂ u ∂u ∂u
F ;...; ;...; ; ;...; ; x1 ; . . . xn ¼ 0, ð9:9Þ
∂x1 ∂x1 ∂x1 ∂xn ∂xn ∂xn ∂x1 ∂xn

where xi’s are general coordinates including both spatial and temporal coordinates
here. It is noted that xi’s are general coordinates in the context of mathematics but are
spatial coordinates in the context of physics.
If F is a linear function of u and its derivatives, then the PDE is called linear.
Otherwise, it is nonlinear. In mathematics, nonlinear PDEs can be further catego-
rized according to the relationships between u and its derivatives, especially the
highest order of the derivatives. However, this detailed categorization will not be
discussed here due to its less significant role in numerical simulations. The highest
order of derivatives determines the order of the PDE. In numerical simulations, the
second-order PDEs are of the major interest.
∂u2 ∂u2
Assuming ∂x∂y ¼ ∂y∂x , the general form of a second-order PDE with two inde-
pendent variables are as follows:

Auxx þ 2Buxy þ Cuyy þ . . . ðlower order termsÞ ¼ 0, ð9:10Þ

where the coefficients A, B, and C may depend upon x and y. If A2 + B2 + C2 > 0 over
a region of the xy plane, the PDE is second-order in that region. This form is
analogous to the equation for a conic section by replacing ∂x with x and likewise
for other variables:

Ax2 þ 2Bxy þ Cy2 þ . . . ¼ 0: ð9:11Þ

Similar to the way of classifying conic sections and quadratic forms into para-
bolic, hyperbolic, and elliptic functions based on the discriminant B2  AC, we can
categorize a second-order PDE into the three types using the discriminant as well:
1. Elliptic PDEs: B2  AC < 0:

2 2
∂ u ∂ u
e:g:, þ ¼ 0, ð9:12Þ
∂x2 ∂y2
in which A ¼ 1, B ¼ 0, C ¼ 1, and B2  AC ¼  1
2. Parabolic PDEs: B2  AC ¼ 0

2
∂u ∂ u
e:g:,  ¼ 0, ð9:13Þ
∂t ∂x2
in which, A ¼ 1, B ¼ 0, C ¼ 0, and B2  AC ¼0
Common PDEs in Engineering Applications 71

3. Hyperbolic PDEs: B2  AC > 0

2 2
∂ u ∂ u
e:g:,  ¼ 0, ð9:14Þ
∂t 2 ∂x2
in which A ¼ 1, B ¼ 0, C ¼ 1, and B2  AC ¼1
The above classification method may not be directly used when there are more
than two independent variables. In this case, we first need to reduce the quadratic
form of the second-order PDE, i.e.,

X ∂ u
2
aij þ lower order terms ¼ 0 ð9:15Þ
∂xi ∂x j

into the canonical form

X 2
∂ u
ci þ lower order terms ¼ 0 ð9:16Þ
∂x2i

by means of an appropriate linear nondegenerate transformation (Polyanin 1997).

The following criterion can be used:
1. The PDE is elliptic if all the ci’s are nonzero and have the same sign.
2. The PDE is parabolic if at least one of the ci’s is zero.
3. The PDE is hyperbolic if all the ci’s are nonzero and have different signs.

Common PDEs in Engineering Applications

Linear Equations

1. Laplace’s equations: Δu ¼ 0. This equation is widely used to describe the

electric, gravitational, and fluid potentials in the equilibrium state.
2. Helmholtz’s equation (involves eigenvalues):Δu ¼ λu. This equation a time-
independent form of the wave equation, resulting from applying the technique of
separation of variables to reduce the complexity of the analysis. It is widely used
for applications such as electromagnetic radiation, seismology, and acoustics.
3. First-order linear transport equation: ut + c∇u ¼ 0. This is the general time-
dependent transport equation, e.g., for mass, momentum, and energy, with a
diffusive term.
4. Heat or diffusion equation: ut  Δu ¼ 0. This is a simplified version of the
above linear transport equation.
5. Schrodinger’s equation: iut + Δu ¼ 0. This is the time-dependent version of the
Schrodinger’s equation for quantum mechanical systems.
6. Wave equation: utt  c2Δu ¼ 0. The wave equation can be used to describe both
mechanical waves and electromagnetic waves.
72 9 Partial Differential Equations

7. Telegraph equation: utt + dut  uxx ¼ 0. The telegraph equation is proposed to

formulate changes of the voltage and current on an electrical transmission line
with distance and time.

Nonlinear Equations

1. Eikonal equation: |— u|  f(x) ¼ 1. This equation is proposed to solve problems of

wave propagation when the wave equation is approximated using the WKB
(Wentzel-Kramers-Brillouin) theory. Derivable from Maxwell’s equations, the
equation provides a link between physical (wave) optics and geometric (ray)
optics.
2. Nonlinear Poisson equation: — u ¼ f(u). This equation is similar to the
Laplace equation but with a source term that relies on the dependent variable u.
It turns into the linear Poisson equation if the function f is not dependent on u.
3. Burgers’ equation:ut + uux ¼ 0. This equation provides a mathematical descrip-
tion for some very special problems
 in fluid dynamics.
∇u
4. Minimal surface equation: ∇  ¼ 0. This equation is the key equa-
ð1þj∇uj2 Þ
1=2

tion in the minimal surface theory. The theory is used to solve the variational
problem of finding the surface of least area stretched across a given closed
contour.
5. Monge-Ampère equation: det(∇∇u) ¼ f. The Monge-Ampère equation arises in
several problems in Riemannian geometry and conformal geometry. One of the
simplest of applications is the prescribed Gauss curvature.
6. Korteweg-de Vries equation (KdV):ut + f(u)ux + uxxx ¼ 0. This equation presents
a mathematical model of waves on shallow water surfaces.
7. Reaction-diffusion equation: ut  Δu ¼ f(u). This is a transport equation with a
diffusive term and nonlinear source term.

System of Partial Differential Equations

1. Evolution equation of linear elasticity.

2. utt  μΔu  (λ + μ)— (—  u) ¼ 0. This is the governing equation of the linear
stress-strain problems.
3. System of conservation laws: ut + —  F(u) ¼ 0. This is the general form of the
conservation equation with multiple scalar quantities of the dependent variables.
4. Maxwell’s equations in vacuum:

∇  E ¼ Bt
∇  B ¼ μ0 ε0 Et
Boundary Conditions and Initial Conditions 73

∇B¼0
∇E¼0

Maxwell’s equation for describing electromagnetic fields in vacuum.

5. Reaction-diffusion system:

vt  Δv ¼ f ðvÞ
ut þ u  ∇u ¼ ∇p
∇u¼0

The equation systems consider both the fluid movement and the solute balance
involving both transient transport and chemical reactions in the fluid.

6. Euler’s equations for incompressible and inviscid fluid:

ut þ u  ∇u ¼ ∇p
∇u¼0

A simplified version of the Navier-Stokes equations for impressible and inviscid

fluids.

7. Navier-Stokes equations for incompressible viscous fluid:

ut þ u  ∇u ¼ ∇p þ Δu
∇u¼0

The Navier-Stokes equations for incompressible viscous fluids.

The equations in bold are those we encounter more often in engineering applications.
It is interesting to mention that the above equations can be summarized into a generic
form. For example, the generic PDE used by COMSOL is a good generalization of
common PDEs:
2

∂ u ∂u ∂ ∂u ∂u
eα þ d a  c þ αu  γ þβ þ au ¼ 0: ð9:17Þ
∂t 2 ∂t ∂xi ∂xi ∂xi

Boundary Conditions and Initial Conditions

Common Types of Boundary Conditions

Boundary conditions, which exist in the form of mathematical equations, exert a set
of additional constraints to the problem on specified boundaries. The concept of
74 9 Partial Differential Equations

boundary conditions applies to both ordinary and partial differential equations.

There are five types of boundary conditions: Dirichlet, Neumann, Robin, Mixed,
and Cauchy, within which Dirichlet and Neumann are predominant. To understand
the difference, let us take a look at an ordinary differential equation:

du
þ u ¼ 0 in the domain ½a; b
dx
We have the Dirichlet boundary condition when the boundary prescribes a value
to the dependent variable(s). A Dirichlet boundary condition for the above ODE
looks like

yð aÞ ¼ A ð9:18Þ
yðbÞ ¼ B: ð9:19Þ

For example, in a 1D heat transfer problem, when both ends of a wire are maintained
in a water bath with constant temperatures, the above boundary condition will be
appropriate.
In the Neumann boundary condition, the derivative of the dependent variable is
known in all parts of the boundary:
0
y ð aÞ ¼ α ð9:20Þ

and
0
y ðbÞ ¼ β: ð9:21Þ

In the above heat transfer example, if heaters exist at both ends of the wire, via which
energy would be added at a constant rate, the Neumann boundary condition will
apply.
The Robin boundary condition is a weighted combination of the Dirichlet bound-
ary and the Neumann boundary condition in all the parts of the boundary:
0
χ 1  yð aÞ þ χ 2  y ð aÞ ¼ A α ð9:22Þ

0
χ 1  y ð bÞ þ χ 2  y ð bÞ ¼ B β ð9:23Þ

where χ i’s are constants representing the weights. The Dirichlet, Neumann, and
Robin are also called the first-type, second-type, and third-type boundary condition,
respectively.
The mixed boundary condition refers to the cases in which Dirichlet boundary
conditions are prescribed in some parts of the boundary while Neumann boundary
conditions exist in the others. Accordingly, for the above ODE, the following is a
typical mixed boundary condition:
Boundary Conditions and Initial Conditions 75

yð aÞ ¼ A ð9:24Þ

0
y ðbÞ ¼ β: ð9:25Þ

In the above 1D heat transfer problem, this corresponds to the condition that one end
of the wire is placed in a water bath while the other end is connected to a heater with
constant heat transfer rate.
While being less common, Cauchy boundary conditions are also used in second-
order differential equations, in which one may specify the value of the function y and
the value of the derivative y' at a given point:

uð a Þ ¼ A ð9:26Þ

0
u ðaÞ ¼ α: ð9:27Þ

Therefore, the Cauchy boundary conditions correspond to imposing Dirichlet and

Neumann boundary conditions simultaneously.

Confusion About Boundary Conditions

In the above introduction to boundary conditions, we determine the boundary type

by assessing the constraints applied on the whole boundary of a computational
domain, which may consist of multiple segments. A rectangular (2D) domain, for
example, has four boundary segments. In many cases, we do not call the top side of
the rectangular area “top boundary segment,” but instead, we call it “top boundary”
directly. This may cause confusion. In many software packages, we need to assign a
boundary condition to this “top boundary,” e.g., Dirichlet and Neumann boundary
conditions. In this case, the Dirichlet and Neumann boundary conditions just
correspond to the assignment of value to the dependent variable and its derivative,
respectively, on the “top boundary.” As a result, we may have different boundary
conditions on different boundary segments. In this use, we mostly talk about
Dirichlet, Neumann, and Robin or other physical boundary conditions rather than
mixed boundary conditions, which need to be defined with respect to the whole
boundary. The use of boundary conditions in this way is somewhat in conflict with
the boundary condition categorization in the previous section. However, we possibly
need to understand and keep both of them in mind due to their wide use.
Another common confusion is associated with the level of boundary conditions.
The definition and categorization of boundary conditions introduced in the previous
section were laid down on the mathematical level. That is, we define and classify
boundary conditions by evaluating whether the dependent variable or its derivative
is equal to a prescribed value. However, boundary conditions can also be defined and
categorized on other levels, such as geometric and physical levels. Common
76 9 Partial Differential Equations

examples of boundary condition defined on the geometric level are symmetric

boundary conditions and periodic boundary conditions. Taking monolithic transport
in the 2D Cartesian coordinate system, for example, a symmetric boundary implies
that the derivative along the direction normal to the boundary is zero, which can be
explained using a Neumann boundary condition defined on the mathematical level.
However, symmetric boundary conditions for more complicated physical fields may
imply more complicated mathematical boundary conditions. In the FVM platform
OpenFOAM, the boundary conditions defined on the geometric level is called
“primitive” boundary conditions. In addition, boundary conditions can also be
defined on the physical level. That is, they are defined from the perspective of
physical meanings. Typical examples are those used in computational fluid dynam-
ics packages: slip, nonslip, outflow, and so on. These boundary conditions are
defined based on the description of the physical processes on the boundaries, and
they can be formulated using the more fundamental mathematical level boundary
conditions. The definitions and categorization of physical boundary conditions
depend on the targeted physical fields. In addition, different software packages
may provide different physical boundary conditions, though some physical bound-
ary conditions have been accepted across packages. These physical boundary con-
ditions serve as preset combinations of mathematical boundary conditions for
boundary segments to save users’ efforts at repeatedly implementing these combi-
nations themselves. Due to the above reason, this book will mostly focus on
mathematical boundary conditions. The physical boundary conditions will be intro-
duced for specific physical fields if necessary.

Boundary Conditions in 3D Domains

When moving from 1D to 2D or 3D problems, the Dirichlet boundary condition

requires no extra treatment as the dependent variable or its components can be
assigned a value directly. Some quantities such as velocity that appear as a scalar
in 1D application will retrieve its vector nature in 2D or 3D. The treatment of a
Dirichlet boundary condition of this type of quantity just changes from assigning one
component in 1D to two and three components in 2D and 3D, respectively.
However, the Neumann boundary condition will call for additional treatment.
This is because Neumann boundary conditions deal with a derivative. In 2D and 3D,
there is more than one direction, and the derivative is generalized into the gradient as
introduced in previous sections. Therefore, theoretically, we now have the freedom
to assign a value to each component of the gradient. This also happens in real
simulations. However, a more common scenario in multiphysics simulation is that
we introduce the directional derivative, which defines the projection of the gradient
in a prescribed direction:

∂u
¼ ð∇uÞ  n, ð9:28Þ
∂n
Boundary Conditions and Initial Conditions 77

where u is the dependent variable which can be a tensor of any order and n is the
prescribed direction. In general, this boundary condition is utilized to represent the
flux across the boundary. The flux can usually be correlated to the gradient of the
dependent variable. Therefore, n is usually the outward normal direction of the
boundary. It is noticed that n is not necessarily unchanged from point to point.
Detailed examples will be given in the chapters for monolithic physics.

Boundary Condition Treatments in Computer Programs

The treatment of the Neumann boundary condition in FEM software packages

desires special attention, such as MATLAB PDE Toolbox and FlexPDE used in
this book. The FEM usually starts the consideration of the Neumann boundary
condition in the derivation of the weak form of the governing equation. In this
process, terms with second or higher derivatives of the dependent variable(s) will be
integrated by parts so that the order of the derivatives can be lowered. Taking the
diffusion term, —  (D— u), for example, the integration by parts produces
(D— u)  n. As J ¼  D— u is the flux, (D— u)  n defines the normal outflux
on the boundary. Accordingly, when we define the Neumann boundary condition by
specifying a value β, we, in fact, assign this value to the normal outflux

n  ðD∇uÞ ¼ β: ð9:29Þ

The software package will substitute the value into the stiffness matrix for
calculation. Likewise, if the second or higher derivative exists in the curl operator
as —  (D— u), the software package will carry out the following replacement

n  ðD∇uÞ ¼ β: ð9:30Þ

There is a significant ramification of this treatment. That is, other parameters in

these terms containing second or higher derivatives will also be included in the
definition. In FlexPDE, any parameters, regardless of its position relative to the
differential operator, will be included. That is, if we multiply the two sides of the
PDE with a constant, such as 2, then the Neumann boundary condition, taking
diffusion for example, will change accordingly,

n  ð2D∇uÞ ¼ β: ð9:31Þ

This means, if we still assign the same β value to the boundary, then the outflux
will be one half of the original case without multiplying the PDE by 2 as

β
nJ¼ : ð9:32Þ
2
78 9 Partial Differential Equations

Initial Condition

The treatment of the initial condition is usually straightforward. If the initial condi-
tion for a field is a constant, then the dependent variable will be equal to this constant
at all the points of discretization in the domain at the beginning of the transient
process to be simulated. Otherwise, a function may be given for the dependent
variable to define its initial distribution. In that case, we just need to discretize the
function to find out the values of the dependent variables at all the points of
discretization. All the other parameters containing the dependent variable will be
calculated with the dependent variable values calculated above.
Chapter 10
Numerical Solution of PDEs

Introduction

As stated before, multiphysics can be viewed as a fusion of mathematics, physics,

applications, and numerical analysis. For a normal multiphysics end user, researcher,
or developer, it is suggested to cover necessary mathematics, understand the physics
on an application basis, and touch upon numerical analysis to an appropriate depth,
just like the arrangement of this book. The content of numerical analysis, especially
the discretization methods and computing techniques, could be overwhelming.
However, what many multiphysics users and researchers would be interested in is
not very likely how to develop a new numerical method or develop an iterative
solver for a problem but instead to mathematically describe the problem using
multiphysics and then to solve it using existing numerical methods and numerical
solvers.
Following the above viewpoint, a handy tool would be a computer program
containing preprocessing for geometry development, numerical solver, and
postprocessing as well as a necessary interface to assist the setup of the model. If a
multiphysics description of a problem has already been well established, then the
remaining task is to create and solve the problem using the way prescribed in the
problem. Currently, there are many commercial and open-source programs that can
satisfy this need, such as COMSOL, Elmer, MOOSE, FiPy, OpenFOAM, FEniCS,
and FEATool. This book adopts the MATLAB PDE Toolbox and FlexPDE which
are very friendly to starters and helpful in exhibiting and implementing the content of
this book. The MATLAB PDE Toolbox will be introduced and used in the first few
chapters considering that it has a convenient graphical user interface to create
models. FlexPDE will be used as the major tool considering its simplicity, close
relationship to the essence of multiphysics, capacity, and complimentary licenses for
education and evaluation purposes.
In the following sections, we will focus on two tools that will be used in the later
implementations of monolithic physics and multiphysics problems in this book:

© Springer International Publishing AG, part of Springer Nature 2018 79

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_10
80 10 Numerical Solution of PDEs

MATLAB PDE Toolbox and FlexPDE. Step-by-step directions will be given for the
use of these two tools to solve a simple PDE with typical boundary conditions. It is
expected that readers will get a real feeling of numerical implementations with these
two tools with distinctly different styles: one with a graphical user interface for pre-
and postprocessing and the other with script files for model setup. Advanced
numerical implementations of multiphysics with self-developed code using the finite
difference method, finite volume method, and finite element method will be intro-
duced in the last three chapters of this book.

Solve PDEs with MATLAB PDE Toolbox

The MATLAB PDE Toolbox is a tool for solving 2D partial differential equations
with the finite element method. An example is provided in this section to guide you
through the use of the graphical user interface for interactive use. Here we use
MATLAB R2014b to demonstrate the use of the PDE Toolbox.
At the MATLAB prompt, type “pdetool,” and you should see a window like the
one below popping out (Fig. 10.1).
We will show the process of solving Laplace’s equation with Dirichlet boundary
conditions using the PDE Toolbox. The solutions to Laplace’s equation are har-
monic functions, which describe situations in which forces are balanced, for exam-
ple, in electrostatics or thermal equilibrium. This is an example which might be first

Fig. 10.1 Main window of PDE Toolbox

Solve PDEs with MATLAB PDE Toolbox 81

Fig. 10.2 Setup of canvas

encountered and be solved using Fourier series methods by hand in an elementary

PDE course. You will find that you obtain the solution faster with the PDE Toolbox
than the Fourier series methods. Moreover, the solution can be visually represented
for a better understanding of the problem and the solution. We here show an example
which might not be easily solved by hand.
1. Draw the region of interest: suppose we want the rectangle (x, y) ¼ [0, 1]  [0, 1].
• In the Options menu, select Grid and Snap; then set the Axes’ Limits to [0.2
1.2] and [0.2 1.2]. When finished, click Apply and Close the Axes’ Limits
window (Fig. 10.2).
• In the Draw menu, select Rectangle/Square (not centered).
• In the drawing region, click and hold with the cursor at the grid point (0, 0),
and drag it to the point with coordinates (1, 1). The region will be shaded and
marked SQ1 (Fig. 10.3).
2. Set up the boundary conditions: the dependent variable u is 1 on the top and
bottom and + 1 on the sides.
• In the Boundary menu, select Boundary Mode. The window changes to this
(Fig. 10.4):
• In the drawing region, double-click on the left edge of the computational
domain (SQ1). The boundary condition window pops up. The default Dirichlet
condition, h*u ¼ r, is OK; the default h ¼ 1 is OK; change the value of r to +1
(Fig. 10.5).
82 10 Numerical Solution of PDEs

Fig. 10.3 Development of geometry (computational domain)

Fig. 10.4 Boundary mode

Solve PDEs with MATLAB PDE Toolbox 83

• Do the same thing with the right edge: double-click and set the value. On the
top and bottom edges, set r to 1. Be patient and careful. When the boundary
conditions are correctly set, you should be able to click on any edge, and see
the correct value for r.
3. Specify the particular PDE of interest: Δu ¼ 0.
• In the PDE menu, select PDE specification. The PDE specification window
pops up; we are doing the default type of problem, elliptic, of the form –div
(c*grad(u)) + a*u ¼ f, where a ¼ 0, f ¼ 0, and c ¼ 1. Change the values as
needed, and click on OK (Fig. 10.6).
4. Generate the mesh with the automatic mesh generator:
• In the Mesh menu, select Initialize Mesh. This time we will use the default
mesh (Fig. 10.7).

Fig. 10.5 Setup of boundary conditions

Fig. 10.6 Setup of governing equation

84 10 Numerical Solution of PDEs

Fig. 10.7 Meshing

5. Solve the Finite Element problem using default solution parameters.

• In the Solve menu, select Solve PDE. The figure in the main drawing area
changes (Fig. 10.8).
6. Generate other plots:
• In the Plot menu, select Parameters. The plot parameter window pops up; click
on the checkboxes for color, contour, and height and then on the Plot button
(Fig. 10.9).
A new figure window pops up, with the appropriate plot (Fig. 10.10).

A PDE Implementation Tool: FlexPDE

This section introduces the use of FlexPDE for solving partial differential equations.
We will show how to solve the problem that was solved with MATLAB PDE
Toolbox in the previous section using the script files of FlexPDE.
FlexPDE can solve mathematical systems consisting of first- or second-order
partial differential equations in 1D, 2D, or 3D Cartesian geometry, in 1D spherical or
cylindrical geometry, or in axisymmetric 2D geometry. Other geometries can be
supported by including the proper terms in the PDE. The system may be steady-state
A PDE Implementation Tool: FlexPDE 85

Fig. 10.8 Solution and results

Fig. 10.9 Setup for postprocessing

86 10 Numerical Solution of PDEs

Fig. 10.10 Visualization of results

or time-dependent or eigenvalue problems. Steady-state and time-dependent equa-

tions can be mixed in a single problem. Any number of simultaneous equations can
be solved, subject to the limitations of the computer on which FlexPDE is run.
The education/student version can handle up to five governing equations. The
equations can be linear or nonlinear. FlexPDE automatically applies a modified
Newton-Raphson iteration process in nonlinear systems. Any number of regions of
different material properties may be defined. Modeled variables are assumed to be
continuous across material interfaces. Discontinuous variables or jumps in the across
boundaries can be handled with special boundary conditions. FlexPDE has a
deeper learning curve than MATLAB FDE Toolbox as it uses a script file instead
of a graphical user interface for pre- and postprocessing. However, it is still one of
the most convenient PDE solution tools. Despite the deeper learning curve, it pro-
vides more flexibility in modeling and better tailoring of the mathematical descrip-
tion. By full use of its power, FlexPDE can be used for very complex numerical
analysis.
A problem description script is a readable text file. The contents of the file consist
of a number of sections, each identified by a header. The fundamental sections are
listed in the following.
A PDE Implementation Tool: FlexPDE 87

TITLE A descriptive label for the output

SELECT User controls that override the default behavior of FlexPDE
VARIABLES Here the dependent variables are named
DEFINITIONS Useful parameters, relationships, or functions are defined
EQUATIONS Each variable is associated with a partial differential equation
BOUNDARIES The geometry is described by walking the perimeter of the domain, stringing
together line or arc segments to bound the figure
MONITORS and Desired graphical output is listed, including any combination of CON-
PLTS TOUR, SURFACE, ELEVATION, or VECTOR plots
END Complete the script

There are several other optional sections for describing special aspects of the
problem. Some of these will be introduced later in this document. Detailed rules for
all sections are presented in the FlexPDE Problem Descriptor Reference chapter
“The Sections of a Descriptor.” Comments can be placed anywhere in a script to
describe or clarify the work. Two forms of comment are available:
• {Anything inside curly brackets is a comment.}
• ! from an exclamation to the end of the line is a comment
The script file is not case sensitive. But some people prefer to use capital letters
for the keywords reserved by FlexPDE.
To demonstrate the use of FlexPDE, here we will solve the same problem that we
just solved using PDE Toolbox with FlexPDE 6.32s (a free student version). The
complete script file is listed in the following.
Example
A simple diffusion equation on a square might look like this:

TITLE '2D Laplace Demo' { the problem identification }

! COORDINATES cartesian2 { coordinate system, 1D,2D,3D, etc; the
default is cartesian2: 2D Cartesian}
VARIABLES { system variables }
u { choose your own names }
! SELECT { method controls }
! DEFINITIONS { parameter definitions }
! INITIAL VALUES
EQUATIONS { PDE's, one for each variable }
DIV(GRAD(u))=0 { one possibility }
! CONSTRAINTS { Integral constraints }
BOUNDARIES { The domain definition }
REGION 1 { For each material region }
START(0,0) { Walk the domain boundary }
VALUE(u)=-1
LINE TO (1,0)
VALUE(u)=+1
LINE TO (1,1)
VALUE(u)=-1
LINE TO (0,1)
VALUE(u)=+1
LINE TO CLOSE
88 10 Numerical Solution of PDEs

! TIME 0 TO 1 { if time-dependent }
MONITORS { show progress }
PLOTS { save result displays }
CONTOUR(u)
SURFACE(u)
END

In the script file, VARIABLEs and EQUATIONS define the PDEs and BOUND-
ARIES give out the computational domain and boundary conditions. In the simplest
scenario, as shown above, a domain is plotted by assigning the start and end points of
a line and then connecting these lines to form a closed 2D domain. Boundary
conditions need to be specified for each boundary segment, with VALUE reserved
for Dirichlet boundary conditions and NATURAL reserved for Neumann boundary
conditions. Auxiliary relationships can be given in the DEFINITIONS section.
MONITORS and PLOTS are designed for showing the process and final results,
respectively. Common plot types include CONTOUR (2D contour), SURFACE
(3D contour), ELEVATION (curves for spatial distribution), and HISTORY (tem-
poral variation). The keyword END is needed to close a script file. Many sections
have default settings, for example, a Neumann boundary condition will be used if no
boundary conditions are assigned to a boundary segment.
To solve the above Laplace’s equation example, we just need to click the “New
Script” tab in the Menu drop-down list. Then we paste the script file into the editable
area. After saving the file before the first time we run it, we can obtain the following
results by clicking the “run” button in the quick access tool area in the editor mode.
After clicking the “run” button, the program will automatically enter the “Plot” mode
to show the solution process and results (Figs. 10.11 and 10.12).

Fig. 10.11 Simulation result 1: contour

A PDE Implementation Tool: FlexPDE 89

Fig. 10.12 Simulation result 2: 3D contour

 Part III
Monolithic Physics
Chapter 11
Thermo: Heat Transfer

Introduction

Heat transfer is a great example to show how a fundamental monolithic physical

phenomenon can be mathematically formulated. In this chapter, the fundamentals of
heat transfer for continua and mixtures including porous materials will be described
consecutively. We will start with the concept of the representative elementary
volume (REV) and conservation laws. Based on that, the general governing equation
of conservation problems will be derived. Following that, we will derive the
governing equation of the heat transfer in continuous materials. The validity of the
REV assumption and conservation laws will help us extend the basics of continuum
mechanics to the theories in porous materials. Emphasis will be placed on the heat
transfer in porous materials which represents a much more complicated and general
framework for heat transfer. Boundary conditions for heat transfer will be discussed
from both a mathematical and a physical perspective. Finally, a practice problem will
be solved using the MATLAB PDE Toolbox and compared against the analytical
solution.

General Background for Continuum Mechanics

There are two types of methods to implement multiphysics: continuous methods and
particle methods. Continuous methods directly deal with the PDEs which describe
the physical processes in the continuous materials. There are rare examples of perfect
continuous materials as imperfections widely exist in natural materials. However, if
we assume that the scale of these imperfections, such as vacancies, dislocations,
cracks, surfaces, and pores, is much smaller than that of the domain of interest, in
which the processes occur, then we can still use the knowledge based on the
continuum mechanics. The assumption of REV will be introduced in the following

© Springer International Publishing AG, part of Springer Nature 2018 93

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_11
94 11 Thermo: Heat Transfer

paragraphs. Porous materials are an extreme example of imperfect materials, in

which imperfections are more of normality than an exception. But the knowledge
of continuum mechanics still applies as long as the assumption of REV can be
ensured. This type of method includes most of the numerical simulation methods
that we are familiar with such as FEM, FVM, FDM, element free Galerkin, boundary
element method, discontinuous Galerkin method, and collocation method, which is
thus the interest of this book. The second type of method deals with multiphysics
from a discontinuous perspective, in which the material is usually assumed to be an
ensemble of particles. Typical examples include the discrete element method,
smoothed particle hydrodynamics, lattice Boltzmann method, and molecular dynam-
ics, which will not be discussed in this book. It is noted that methods such as
smoothed particle hydrodynamics are more of a hybrid of the above two groups
than any of them because of the employment of both PDEs from continuum
mechanics and particles in the method.

Representative Elementary Volume

The representative elementary volume, which can be abbreviated as REM, is also

called representative volume element (RVE). In continuum mechanics for a hetero-
geneous material, as shown in Fig. 11.1, an RVE can be considered as a volume Ω
that represents a composite statistically. The volume should effectively include a
sampling of all microstructural heterogeneities occurring in the composite, such as
grains, inclusions, voids, and fibers (Kanit et al. 2003). This volume, in the mean-
while, must stay small enough to be considered as a volume element of the bulk
continuum. Several types of boundary conditions can be prescribed on the boundary
of the volume, ∂Ω, such as a given mean strain or mean stress in the mechanical
field. The word “volume” and the symbol Ω here are general terms which apply to
1D, 2D, and 3D problems. The REV has a volume of V in 3D, a surface area of S in
2D, and a length of L in 1D.
In the theory of composite materials, the REV is believed to be the smallest
volume over which a value representative of the whole can be obtained. Porous
materials can be treated as a special and very complicated case of composite
materials. Let us revisit the sampling of materials to understand the REV in porous
materials. We usually collect and measure samples of a porous medium to acquire

Fig. 11.1 Representative

elementary volume
General Background for Continuum Mechanics 95

the properties of the material. If the sample is too small, the readings tend to oscillate.
As we increase the sample size, the oscillations begin to dampen out. Eventually, the
sample size will become large enough that we begin to get consistent readings. This
sample size is referred to as the representative elementary volume. If we continue to
increase the sample size, the measurement will remain stable until it becomes
comparable to the size of the domain of interest. However, the sample needs to be
much smaller than the size of the domain (whole sample); otherwise, sampling will
be meaningless. The validity of the REV assumption is thus the foundation based
upon which you can apply the PDEs developed for continuous materials to “imper-
fect” materials such as porous materials.

Conservation Law

The conservation laws, i.e., mass, momentum, or energy, define the way that how a
physical process of a conservation nature is formulated with a PDE(s). Conservation,
in fact, accounts for a major part of physical processes that we encounter in
multiphysics, including heat transfer. The derivation of the PDE for conservation
problems starts from the global or integral balance laws as the following equation:
Z I Z
d
udx þ J  ndx ¼ qdx, ð11:1Þ
dt
Ω ∂Ω Ω

where Ω corresponds to the REV in continuum mechanics, u is the generic con-

served quantity defined by unit mass, x is the coordinate vector, J is the flux of this
conserved quantity, and q is the source term. Ω is a more general representation of
the computational domain. Material points occupying Ω(0) may move under the
influence of inertia and applied loads, so that at any time, t, they may occupy a
volume Ω(t) which is different from Ω(0). The function Ω(t) is a material volume
which is defined by a set of material points rather than by any explicit spatial
coordinates.
After applying the Gauss theorem, Reynolds transport theorem, and DuBois-
Reymond lemma theorem (Allen 1988), we can obtain the local balance equation:

∂u
þ ∇  ðuvÞ þ ∇  J ¼ q, ð11:2Þ
∂t
where the flux, J, can be induced by different mechanisms such as diffusion,
dispersion, electromagnetic field, and so on. This flux term can be formulated as
Eq. 11.3:

J ¼ f ðt; x; u; ∇uÞ ¼ K∇u  D∇u þ . . . ð11:3Þ

96 11 Thermo: Heat Transfer

By substituting this flux term into the local balance equation, we then obtain the
general conservation equation for transport phenomena (Eq. 11.4). More mecha-
nisms responsible for flux across the boundary can be added by extending the flux
term:

∂u
þ∇  ðuvÞ ∇  ðK∇uÞ ∇  ðD∇uÞ ¼ q : ð11:4Þ
∂t
|{z} |fflfflfflfflfflffl{zfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |{z}
Advection Diffusion ðConductionÞ Dispersion Source
Accumulation

The conservation equation states that the accumulation of conserved quantity is

balanced with the conserved quantity advected into the control volume via mobile
phases and diffused directly as diffusion and indirectly as dispersion (Deen 1998)
into the control volume via the aqueous and gas phases, plus the mass generation
associated with sources.

Heat Transfer in Continua

The governing equation for heat transfer can be easily obtained based on the
conservation equation introduced below:

∂e
þ ∇  ðevÞ þ ∇  JT ¼ qT , ð11:5Þ
∂t
where the subscript, T, denotes the conservation of energy, which is the extended
definition of thermal fields. The conservation of quantities and the corresponding
physical fields are used interchangeably throughout this book.
The above governing equation can be further transformed to obtain the solution.
Generally, terms such as energy density, heat flux, and source can be formulated as
functions of the dependent variables and material properties. In such a way, the
equation only contains the dependent variables, materials properties depending on
the dependent variables, and constants. One governing equation may contain mul-
tiple dependent variables, but the total numbers of dependent variables and that of
the governing equations should be equal to guarantee a unique solution to the
equation system. Auxiliary relationships, which are necessary for the internal energy
and heat flux, mainly refer to the heat capacity and thermal conductivity and will be
discussed in next section. The auxiliary relationships for the source term will be
considered in the chapter for the chemical field (Chap. 14). In order to get a heat
transfer equation that we are more familiar with, let us start from the formation of the
energy density term as Eq. 11.6:
Heat Transfer in Porous Materials 97

0 1
ZT  
B C 1
e¼@ ρC v T þ U 0A þ ρv  v þ ðρgzÞ , ð11:6Þ
2 |ffl{zffl}
T0 |fflfflfflfflfflffl{zfflfflfflfflfflffl} Potential Energy
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} Kinetic Engergy
Internal Energy

where Cv is the gravimetric heat capacity at a constant volume and U0 is the internal
energy per unit volume at the reference temperature T0.
The internal energy is part of the accumulation term in addition to the kinetic
energy and potential energy. But in general, the internal energy is dominant over the
other energy types. As a result, the energy density is frequently represented by only
the internal energy density as in Eq. 11.7, in which the pressure term vanishes for
solids:

ZT ZT Z
e¼ ρC v T þ U 0 ¼ ρCp T  pdV þ U 0 , ð11:7Þ
T0 T0

where Cp is the heat capacity at a constant pressure.

Next, the flux term is linked to the temperature by Fourier’s law of heat conduc-
tion as

JT ¼ λ∇T, ð11:8Þ

where λ is the thermal conductivity. Substituting the above equations for energy
density and heat flux into the energy balance equation, we can obtain the heat
equation including advection and source terms as:

∂ðρC v T Þ
þ ∇  ðρCv TvÞ  ∇  ðλ∇T Þ ¼ qT : ð11:9Þ
∂t
As can be seen, this equation is a comprehensive version of the Fourier’s equation
for heat transfer. The above deduction shows how to obtain the equation from the
basic heat conservation law.

Heat Transfer in Porous Materials

Energy balance is mostly considered for the whole system instead of individual
phases. This treatment is based on the assumption that thermodynamic equilibrium is
maintained between different phases. Therefore, most existing models for porous
materials employed one governing equation for heat transfer for the whole system.
Governing equations are established for individual phases/components only in a few
cases. The general form of the governing equation for heat transfer of the whole
system is as follows:
98 11 Thermo: Heat Transfer

P
∂ð ρi θi C v, i T Þ X
þ ∇  ðρi θi C v, i Tvi Þ  ∇  ðλ∇T Þ ¼ qT : ð11:10Þ
∂t
But in most cases, researchers prefer simple treatment in which phases and
components are not differentiated. In such a case, the material properties are the
overall properties of the porous materials instead of that of any individual phase. In
even more simplified treatments, these material properties are moved out of the
operators, leading to Eq. 11.11:

∂T
ρC v þ ρCv ∇  ðTvÞ  λ∇  ð∇T Þ ¼ qT : ð11:11Þ
∂t
However, a considerable error may be involved when these materials’ properties
have spatial or temporal variabilities as they are functions of dependent/independent
variables or other materials properties.

Heat Capacity

Heat capacities of individual components are mostly used to directly construct the
accumulation term in Eq. 11.11 based on the formulation of the internal energy
(Eq. 11.7). This formulation, which is the weighted average of the heat capacities of
the components with respect to mass, has long been used for soils (de Vries 1963;
Campbell 1985). This mass-weighted average method was also used in many other
mixtures as
X X
Cv ¼ ðρi θi C v, i Þ= ðρi θi Þ: ð11:12Þ

The above formulation is based on the definition of the thermal energy and can be
derived based on thermodynamics.
A constant value is also frequently used for the average heat capacity of porous
materials for simplicity. There are also more complex relationships that allow more
effects on the heat capacity to be considered. For example, the capacity of gas can be
described as a function of temperature:

Cg ¼ C0 þ C1  T þ C2  T 2 þ C3  T 3 , ð11:13Þ

where C0 is 1238.79, C1 is 3.1303, C2 is 7.905  104, C3 is 6.858  107, and Ci

is the fitting constants throughout this section. When the phase changes of water
occur, especially for freezing and thawing, the latent heat is typically considered
with the concept of apparent heat capacity. This term helps ensure the computational
stability (Anderson et al. 1973). In this treatment, the released or absorbed energy is
incorporated into the heat capacity term using Eq. 11.14:
Heat Transfer in Porous Materials 99

dθi
Ca ¼ C þ L f , ð11:14Þ
dT
where Ca is the apparent gravimetric heat capacity and Lf is the latent heat of phase
transition.

Thermal Conductivity

The formulation of the thermal conductivity is not as straightforward as that of the

heat capacity. The average thermal conductivity of a composite is usually used. In
many simulation models, this average thermal conductivity was assumed to be a
constant. This treatment was adopted to reduce the nonlinearities of the models or to
enable analytical solutions. However, a volume-weighted value of thermal conduc-
tivities was more frequently employed in the existing numerical models. The
mathematical formulation of this volume-weighted average value is as follows:
X X
λ¼ θ i λi = θi : ð11:15Þ

The formulation in Eq. 11.15, which has been extensively used, is essentially an
empirical equation. There are still other ways for the calculation of the average
thermal conductivity, such as a geometric average as the following:
Y 1= P θi
λ¼ λi θi : ð11:16Þ

Russell’s equation (Perry and Chilton 1973) has also been applied to the formu-
lation of the thermal conductivity by Scott et al. (2006):
h i
2=3
X ½θi =ð1  θm Þ ð1  θm Þ ðλi =λm  1Þ þ 1
λ ¼ λm h i , ð11:17Þ
ð1  θm Þ2=3  ð1  θm Þ ðλi =λm  1Þ þ 1

where m indicates solid matrix and i denotes the phases other than the solid matrix.

Thermal Diffusivity

In many documents about PDE, especially those used in mathematics, the thermal
diffusivity is frequently used. The adoption of the thermal diffusivity was also
common in early simulation models. As thermal diffusivity is the combination of
the heat capacity and thermal conductivity, the purpose for using this term is to
simplify the governing equation and the solution process. So in these methods,
100 11 Thermo: Heat Transfer

a constant value of the thermal diffusivity is usually employed. More recent research
such as Rempel and Buffett (1997) claimed that the effective thermal diffusivity
should include dispersive effects.

Boundary Conditions

The boundary conditions for heat transfer is just materialization of the aforemen-
tioned basic types of boundary conditions for general PDEs. But as we are moving
into real physical problems, the basic boundary condition types will be detailed and
extended to allow for physical mechanisms. The following boundary conditions can
be specified on outward and inner boundaries of a region.
The Dirichlet boundary condition is used to handle the condition when the
temperature on a boundary is known. For the purpose, a known value of temperature
Tenv is assigned to the vertex or the edge of the computational domain. This is
common in chemical reactions when we have a water bath with a constant temper-
ature in contact with the boundary. The Tenv value at the edge can be specified as a
linear function of coordinates. The function parameters can vary from one edge to
another but have to be adjusted to avoid discontinuities at edges’ junction points.
This boundary condition sometimes is called the boundary condition of the first kind.
The Neumann boundary condition is used to consider the condition when the heat
flux across a boundary is prescribed. This type of boundary condition can be applied
to external boundaries or internal boundaries between regions (Fig. 11.2):
n  J ¼  Fn at external boundaries,
n+  J+  n  J ¼ Fn at internal boundaries.
For external boundaries, Fn is the outward normal component of heat flux. For the
internal boundary, “+” and “–” superscripts denote quantities to the left and to the
right side of the boundary by assuming the axis is pointing to the right. The left
region is “+” because it has a boundary outward normal vector pointing in the
positive direction. On the internal boundary, Fn is the generated power per unit
area or equivalently the surface source.
If Fn on the boundary is zero, the boundary condition is called homogeneous. The
homogeneous condition at the outward boundary indicates vanishing of the heat flux
across the surface. This kind of boundary condition is used when the boundary is
thermally insulated or on a symmetry plane when heat transfer across the plane is
obviously absent. In a few cases, a heat source in terms of heat flux can exist on the

Fig. 11.2 External and n

internal boundaries
y n n- n+

x External BC Internal BC
Practice Problem 101

symmetry plane; then this boundary can be simulated as an external boundary, but
the surface power has to be divided based on the symmetry. A homogeneous
condition for an internal boundary indicates that there is no source on the boundary,
and the flux across the internal boundary is continuous at least along the normal
direction of the boundary.
Two other physical boundary conditions can be derived from the classic Neu-
mann boundary condition by incorporating the real heat transfer mechanisms across
boundaries: convection boundary condition and radiation boundary condition. Tech-
nically, they are still the Neumann boundary condition, but the heat flux density is
related to temperature and material properties by considering the physical
mechanisms.
The convection boundary condition describes convective heat transfer. The
convective heat transfer can be described using Newton’s law of cooling, in which
the heat flux is proportional to the difference between the temperature of the material
adjacent to the boundary and the ambient temperature. This type of boundary
condition including Newton’s law of cooling is defined as

F n ¼ hðT  T env Þ, ð11:18Þ

where h is the coefficient of convective heat transfer and Tenv is the temperature of
contacting fluid medium. Parameters h and Tenv may differ from part to part of the
boundary.
The Neumann boundary condition can also be extended to consider radiative heat
transfer across boundaries:
 
F n ¼ βk SB T 4  T 4env , ð11:19Þ

where kSB is a Stephan-Boltzmann constant (5.67032108 W/m2/K4) and β is an

emissivity coefficient. Parameters β and T0 may differ from part to part of the
boundary.

Practice Problem

Problem Description

In rectangular Cartesian coordinates (Fig. 11.3), the 2D sourceless heat equation can
be simplified into the following form when the density, heat capacity, and thermal
conductivity are constants:
!
2 2
∂w ∂ w ∂ w
¼a þ , ð11:20Þ
∂t ∂x2 ∂y2
102 11 Thermo: Heat Transfer

Fig. 11.3 Computational 1m

domain

T0=273.15

1m
Water
y

T=0 on ∂Ω
x

where w is the dependent variable, which is the temperature in this case; and a is the
thermal diffusivity. The material filling the domain is water.
1. The problem to be solved corresponds to a 2D square region (1 m by 1 m). The
initial temperature is 273.15 K throughout the interior of the region. The temper-
ature on all the boundaries is fixed as 0 K. The question is how the temperature in
the domain changes in the first hour. Please describe the problem using mathe-
matical terms we learned in the past chapters.
2. Please conduct simulations using the PDE toolbox in MATLAB. Explain the
process step by step (print your screen and add some text to explain what you do).
3. Compare the numerical results with the analytical solution as follows:
( " #)
16w0 X 1
1 ð2n þ 1Þπx π2 ð2n þ 1Þ2 at
w¼ 2 sin exp 
π 2n þ 1 l1 l1 2
( n¼0 " #) ð11:21Þ
X1
1 ð2m þ 1Þπy π2 ð2m þ 1Þ2 at
 sin exp  :
n¼0
2m þ 1 l2 l2 2

To save your time, a MATLAB program has been developed for implementing
the analytical solution. But you still need to extract values from both numerical and
analytical solutions so that you can do a quantitative evaluation. For example, you
can try to compare the temperature distributions along y ¼ 0 at t ¼ 3600. For the
numerical solution, you will need to export mesh and solution from the toolbox GUI
first. Then can you use a MATLAB command tri2grid to get the solution on a
straight line, i.e., tri2grid(p,e,u(:,3600),-0.5:0.01:0.5,0). As for the distributions in
the analytical solution, you can extract the values using a couple of commands as
long as you know the structure of the solution (spend a few minutes to understand the
code in the appendix).
4. Refine the mesh in the PDE toolbox and run the simulation again. Please show
whether the accuracy of the solution is improved.
Practice Problem 103

250

Numerical
200 Analytic
Temperature (K)

150

100

50

0
0 20 40 60 80 100
x coordinate

Fig. 11.4 Comparison between numerical and analytical results

Solution

Typical results are shown in Fig. 11.4.

Appendix: MATLAB Code

L1=1;
L2=1;
t=0:1:3600;
a=1e-5;
u0=273.15;
x=0:0.01:1;
y=0:0.01:1;
[xx,yy]=meshgrid(x,y);
solution=zeros(length(x),length(y),length(t));
A_temp=zeros(length(x),length(y));
A=zeros(length(x),length(y));
B_temp=zeros(length(x),length(y));
B=zeros(length(x),length(y));
time_step=0;
for temp=t
time_step=time_step+1
A=zeros(length(x),length(y));
B=zeros(length(x),length(y));
for n=0:100
104 11 Thermo: Heat Transfer

A_temp=1/(2*n+1)*sin((2*n+1)*pi*xx/L1).*exp(-pi^2*a*temp*
(2*n+1).^2/L1^2);
A=A+A_temp;
B_temp=1/(2*n+1)*sin((2*n+1)*pi*yy/L2).*exp(-pi^2*a*temp*
(2*n+1).^2/L2^2);
B=B+B_temp;
end
solution(:,:,time_step)=16*u0/pi^2*A.*B;
figure(1)
surf(xx,yy,solution(:,:,time_step))
end
%surf(xx,yy,solution(:,:,time_step))
Chapter 12
Hydro: Pore Water Movement

Introduction

Groundwater is the water located beneath the Earth’s surface in soil pore spaces and
in the fractures of rock formations. Usually, the topsoil or rock layer is unsaturated
unless it is right close or beneath the surface water. However, a groundwater table
may be encountered if we go deeper. The boundary between the unsaturated layer
and the completely saturated layer is called the water table. A rock body or an
unconsolidated deposit is called an aquifer when it can yield a usable quantity of
water. Groundwater is closely related to surface water which it is recharged from and
eventually flows to. Natural discharge often occurs at springs and seeps, possibly
forming oases or wetlands. Groundwater is also often withdrawn for agricultural,
municipal, and industrial use via extraction wells. The study of the distribution and
movement of groundwater is hydrogeology, also called groundwater hydrology.
Besides serving as an essential resource, groundwater can also be a harmful factor
in various natural hazards such as landslides and earthwork construction. Therefore,
groundwater movement is also of great interest in water resources and geotechnical
engineering. Hydro is used to represent water in the general context and water
movement in the context of multiphysics.
Water moves in porous materials in terms of saturated flow or unsaturated flow. In
the hydrologic cycle, as illustrated in Fig. 12.1, water flows both in saturated and
unsaturated porous media (soils, rock formations). The vadose zone, also termed
unsaturated zone or zone of aeration elsewhere, is the Earth portion connecting the
atmosphere and the phreatic zone where the groundwater presents. The vadose zone
contains, at least sometimes, air and water in the pores, leading to an unsaturated
condition. The thickness of vadose zone ranges from zero, as when a lake or marsh is
at the surface, to hundreds of meters, as is common in arid regions. The upper part of
the vadose zone commonly includes the plant root zone. The phreatic zone, also
called saturated zone, is the part below the groundwater table. This part is where we
can locate aquifers. It is worthwhile to mention the water flows in the unsaturated

© Springer International Publishing AG, part of Springer Nature 2018 105

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_12
106 12 Hydro: Pore Water Movement

Fig. 12.1 Schematic of vadose zone and the hydrologic cycle

and saturated zones occur according to significantly different mechanisms and thus
are described in distinct ways. The water in the unsaturated zone is mostly called
water flows in the vadose zone or just unsaturated flow and is formulated using the
Richards equation. The water in the saturated zone is usually called groundwater
flow and is formulated using the groundwater flow equation.
In the following sections, the groundwater flow will be introduced first, including
mass conservation and Darcy’s law. The introduction to these laws and their
corresponding mathematical formulations will be conducted for both the saturated
flow and unsaturated flow. Next, brief information regarding surface water will also
be provided to lay down a complete framework for the water movement related to
porous materials. Finally, a practice problem for coupled saturated-unsaturated flow
will be presented, which can be solved with FlexPDE and validated against
documented testing results.

Groundwater Movement

The groundwater can be described using the groundwater flow equation. This
equation is extensively used in hydrogeology as the mathematical formulation for
the flow of groundwater through an aquifer. The transient flow corresponds to a flow
Groundwater Movement 107

field changing with time. That is, the water head/pressure at any point could change
with time. If dye is placed in the porous media, then the flow net generated by the
migratory dye with water movement will also change with time. A parabolic PDE
will be used for this transient field. This PDE is also used for many other types of
transient conservation problems, such as the heat conduction and the particle diffu-
sion. For an equilibrium (or steady-state) problem, the time-dependent term or
accumulation term drops out, and the governing equation will reduce into an elliptic
PDE, i.e., Laplace’s equation.
The groundwater flow equation can be derived using the general REV concept.
The assumption of REV still needs to be ensured in which the properties of the
medium are assumed to be constants. In this chapter, we will introduce a mass
balance between the water flowing in and out of a small control volume. The flux
terms in the balance are usually expressed using head via a constitutive equation
called Darcy’s law, whose validity requires that the flow is slow.

Mass Balance

The scientific fact underlying the groundwater equation is the mass balance. This
balance is analogous to the energy balance used in heat transfer, which leads to the
heat equation. The mass balance in a porous material states that, for a given control
volume, aside from sources or sinks, mass cannot be created or destroyed. Alterna-
tively, the conservation of mass defines that, for a given increment of time Δt, the
difference between the mass flowing in across the boundaries, the mass flowing out
across the boundaries, and the sources within the volume is the change in storage.
Mathematically, we would have

ΔM stor M in M out M gen

¼   : ð12:1Þ
Δt Δt Δt Δt
Applying the above equation to a REV, we obtain
Z I Z
1d
mdV ¼  ρJH  ndS þ ρqH dV: ð12:2Þ
V dt
Ω ∂Ω Ω

Applying the divergence theorem and turning the integral form into the differential
form, we will obtain

1 ∂V
¼ ∇  JH þ qH : ð12:3Þ
V ∂t
108 12 Hydro: Pore Water Movement

Darcy’s Law

Narrow Definition for Special Case

Darcy’s law at a constant elevation is a simple proportional relationship between the

instantaneous discharge rate through a porous medium, the viscosity of the fluid, and
the pressure drop over a given distance:

kA ðpb  pa Þ
Q¼ : ð12:4Þ
μ L
The total discharge, Q (units of volume per time, e.g., m3/s), is equal to the product
of the intrinsic permeability of the medium, k (m2); the cross-sectional area to flow,
A (units of area, e.g., m2); and the total pressure drop ( pb  pa), (Pa), all divided by
the viscosity, μ (Pas), and the length over which the pressure drop is taking place
(m). The negative sign is needed because fluid flows from a location with a high
pressure to that with a low pressure. It is noted that the elevation head must be taken
into account if the inlet and outlet are at different elevations. If the change in pressure
is negative (where pa > pb), then the flow will be in the positive direction. Dividing
both sides of the equation by the area and using more general notation leads to

k
JH ¼  ∇p, ð12:5Þ
μ
where JH is the flux (discharge per unit area per unit time, m/s) and — p is the
pressure-gradient vector (Pa/m). It is also very common to formulate Darcy’s law
using hydraulic conductivity and water head:

JH ¼ K∇h ð12:6Þ

where

ρgk
K¼ , ð12:7Þ
μ
p ¼ ρgh: ð12:8Þ

Equation 12.7 implies the intrinsic permeability k is a property of the solid matrix
while hydraulic conductivity K is a property of the whole porous material. The
misuse of these two terms is a common mistake in technical literature for the pore
water movement.

General Definition for Porous Materials

A more general Darcy’s law requires a better understanding of the energy state of
pore water. The water head h is a measure of the energy state of pore water in terms
Groundwater Movement 109

of the height of an equivalent water column. In porous materials, the total energy of
the pore water has three components: pressure, elevation, and velocity. Accordingly,
the water head also has these three components. In normal conditions, the water
velocity in porous materials is low due to the small pore size. Therefore, the total
head usually contains pressure head and elevation head:

h  hp þ hz , ð12:9Þ

where hp and hz are pressure and elevation head, respectively. However, the follow-
ing symbols will be used to render a consistent symbol system for the flow in both
saturated and unsaturated porous materials:

h ¼ ψ þ z, ð12:10Þ

where ψ is the pressure head, and z is the elevation head as the coordinate z is usually
reserved for the gravitational direction.
The flux JH, which is often referred to as the Darcy flux or velocity, is not the
velocity at which the fluid travels through the pores v. However, this true fluid
velocity v is related to the Darcy flux JH by the porosity ϕ. The flux is divided by
porosity to account for the fact that only a fraction of the total formation volume is
available for the fluid flow. The true fluid velocity would be the velocity that a
conservative tracer would experience when it is carried by the moving fluid through
the porous material, that is,

JH
v¼ , ð12:11Þ
ϕeff

where ϕeff is the effective porosity, which is defined as the ratio between the flow
area and the total sectional area available to the flow. It is noted that the porosity of
the porous material is defined as the ratio between the pore volume and the total
volume. Therefore, the effective porosity and porosity are defined using area and
volume, respectively. Their values may be close despite their different definitions.
The Darcy’s law holds when the velocity of pore water is slow, which is the case
for most cases in porous materials. The Darcy’s law can be derived from the Navier-
Stokes equations under the condition of very slow liquid flow velocity, i.e.,
creep flow.

Derivation of Governing Equation

Mass is the product of the density and volume. Under creep flow conditions in
porous materials, water can be considered incompressible: the density does not
depend on the pressure. The mass fluxes across the boundaries then become volume
fluxes as are used in Darcy’s law. Using Taylor series to represent the in- and outflux
terms across the boundaries of the control volume, and using the divergence theorem
110 12 Hydro: Pore Water Movement

to turn the flux across the boundary into a flux over the entire volume, the final form
of the groundwater flow equation (in differential form) is

1 ∂V
¼ ∇  JH þ qH : ð12:12Þ
V dt
In reality, we define a material property as a specific storage, which is the volume of
water released from storage per unit decline in the hydraulic head:

∂h
Ss ¼ : ð12:13Þ
∂V=V

Then we will have

∂h
Ss ¼ ∇  JH þ qH : ð12:14Þ
∂t
This parabolic PDE states that the change in the hydraulic head with time (left-hand
side) equals the negative divergence of the flux JH and the source term qH. This
equation contains two unknowns, the head h and flux JH, while qH is usually known
for a given problem. As Darcy’s law relates the flux to the head, we can substitute it
into the above equation to eliminate the flux JH, leading to

∂h
Ss ¼ ∇  ðK∇hÞ  qH : ð12:15Þ
∂t
Now if the hydraulic conductivity K is spatially uniform and isotropic can be
formulated using a scalar rather than a tensor or a function, then this term can be
taken out of the spatial derivative, further simplifying the above equation into

∂h
Ss ¼ K∇2 h  qH : ð12:16Þ
∂t
In some cases, the above equation is further simplified by dividing both sides of
the equation by the specific storage Ss, leading to a new term, hydraulic diffusivity
α ¼ K/Ss on the right-hand side. The hydraulic diffusivity is proportional to the speed
at which a finite pressure pulse will propagate through the system. The groundwater
flow equation for transient flow then becomes

∂h q
¼ α∇2 h  H , ð12:17Þ
∂t Ss
where the sink/source term, qH, is divided by the appropriate storage term.
The steady-state equation for groundwater movement is obtained by eliminating
the transient term in the above equation, which has the form of Poisson’s equation:
Water Flow in Unsaturated Zones 111

∇2 h ¼ qH =Ss : ð12:18Þ

As the source terms usually do not exist, the above equation is reduced to
Laplace’s equation, which serves as the theoretical basis of equilibrium groundwater
problems and engineering methods for solving such problems such as the flow net
method:

∇2 h ¼ 0: ð12:19Þ

The boundary conditions of either the transient or steady-state governing equa-

tions can be easily dealt with. For transient flow, either Dirichlet or Neumann
boundary conditions can be applied. For the steady-state, it seems that Neumann
boundary condition may not be included because an equilibrium state may be
impossible unless with exactly equal influx and outflux. The Dirichlet boundary
condition assigns the value of the head on the boundary, while the Neumann
boundary condition assigns the derivative of the head to define the flux through
the boundary via Darcy’s law:

ðK∇hÞ  ðnÞ ¼ influx, ð12:20Þ

where n is the unit outward normal vector of the boundary.

Water Flow in Unsaturated Zones

The Richards equation represents the movement of water in unsaturated soils and
was formulated by Lorenzo A. Richards in 1931. It is a nonlinear PDE, which is
often difficult to approximate since it does not have a closed-form analytical
solution.

Derivation

We can derivate the governing equation for water flow in unsaturated zones in a way
similar to that is used for saturated flow. The major difference lies in both the water
accumulation and Darcy’s law. For the water accumulation, the difference is that the
water volume is no more majorly attributed to the volume expansion, δV/V, caused
by the change in volume pressure. This is because, as the porous material is
unsaturated, the change in the water content δθ can now be well accommodated
by the partially filled pore space. Therefore, the governing equation can be written as
112 12 Hydro: Pore Water Movement

∂θ
¼ ∇  JH þ qH : ð12:21Þ
∂t
Darcy’s law still holds in unsaturated flow as the velocity of water movement is
still proportional to the energy gradient:

JH ¼ KΔh: ð12:22Þ

However, in unsaturated porous materials, the water pressure head now turns into a
negative value, which indicates that its energy is lower than the reference energy,
i.e., the pressure head of bulk water under standard conditions (1 atmosphere).
Substituting JH into the equation above, we get

∂θ
¼ ∇  ðK∇hÞ þ qH : ð12:23Þ
∂t
Substituting h ¼ ψ + z into the above equation, then we have

∂θ
¼ ∇  ½K ð∇ψ þ ∇zÞ ¼ ∇  ðK∇ψ Þ þ ∇  ðKek Þ, ð12:24Þ
∂t
where ek is the unit vector along the gravitational (z) direction. The above equation
contains two unknowns: water content and pressure head. Due to the reason, it is
called the mixed form of the Richards equation.

Formulations

The Richards equation is used extensively in hydrology, hydrological engineering,

and geotechnical engineering. It also appears in pure mathematical journals due to
the fact that it has nontrivial solutions. Three forms of the equation can be found in
the literature. The mixed form contains the pressure and the degree of saturation as
introduced above. It can also appear in two other forms: the head-based formulation
and saturation-based formulation.

Head Based

∂ψ
C ð hÞ ¼ ∇  ½K ðhÞ∇h, ð12:25Þ
∂t
where C(h) [1/L] is a function describing the rate of the change of saturation with the
change of the pressure (suction) head:
Water Flow in Unsaturated Zones 113

∂θ
C ð hÞ ¼ : ð12:26Þ
∂ψ
This function is called “specific moisture capacity” in the literature and could be
determined for different soil types using curve fitting and laboratory experiments
measuring the rate of infiltration of water into a soil column as described in van
Genuchten (1980).

Saturation Based

∂θ
¼ ∇  ½DðθÞ∇θ, ð12:27Þ
∂t
where D(θ) [L2/T] is “the soil-water diffusivity”:

K ðθ Þ ∂h
DðθÞ ¼ ¼ K ðθ Þ : ð12:28Þ
C ðθ Þ ∂θ

A concept called the soil-water characteristic curve needs to be introduced before

we further investigate C and K. This SWCC (mostly used in geotechnical engineer-
ing), which is also called water retention curve (in soil science) and saturation-
capillary pressure curve (in petroleum engineering), is a fundamental constitutive
relationship in porous materials. The SWCC defines a unique relationship between
the water content, or degree of saturation, and the suction (pressure head or potential)
if hysteresis is neglected. Both C and K are functions of the SWCC in unsaturated
porous materials. The hydraulic conductivity can be estimated by integration of the
SWCC (Mualem 1976; Fredlund and Xing 1994). Both van Genuchten’s equation
(van Genuchten 1980) and Fredlund and Xing’s equation (1994) are common for
formulating the SWCC. Different approaches were then adopted to predict the
hydraulic conductivity based on SWCC.

θ  θr m
Se ¼ ¼ ð1 þ jαψ jn Þ , ð12:29Þ
θs  θr
where Se is the effective saturation, θs and θr are the saturated and residual water
content, respectively; and α, m, and n are empirical parameters.
The hydraulic conductivity in unsaturated porous materials can be viewed as the
product of the saturated hydraulic conductivity and the relative hydraulic conduc-
tivity. The relative conductivity Kr quantifies how the hydraulic conductivity in
unsaturated porous materials changes with the water content or the degree of
saturation. One widely adopted relationship is proposed by van Genuchten (1980):
114 12 Hydro: Pore Water Movement

h   i2
1 m
K ðSe Þ ¼ K s K r ¼ K s Se 1  1  Se m : ð12:30Þ

The treatment of the boundary condition for the Richards equation is also
straightforward, similar to those in the groundwater movement equation.

Mixed Saturated-Unsaturated Flow

In the real world, water moves in both saturated and unsaturated zones which may
coexist as contagious saturated and unsaturated zones, whose boundary may change
over time. So the numerical simulation of such a case requires the simultaneous
implementations of the groundwater flow equation and the Richards equation. A
standard formulation of the Richards equation that also includes groundwater (sat-
urated zone) flow (Richards 1931) is

Ss θðhÞ ∂h ∂θ
þ ¼ ∇  ½K s K r ∇ðh þ zÞ þ qH , ð12:31Þ
ϕ ∂t ∂t
where h is the pressure head, z is the vertical coordinate, Ks is the saturated hydraulic
conductivity tensor, Kr is the relative hydraulic conductivity, Ss is the specific storage
coefficient, and ϕ is the porosity. The detailed implementation of the equation will
be practiced in a later practice problem.

Surface Flow

Surface flow is much different from the flow in porous materials; the latter can be
termed subsurface flow in hydrogeology. However, it is very common that these two
types of flow interfere with each other. In the water cycle, when precipitation falls on
the earth’s land, some of the water flows on the surface forming streams and rivers.
The remaining water, through infiltration, penetrates the soil traveling underground,
hydrating the vadose zone soil, recharging aquifers, with the excess flowing in the
subsurface runoff. Water flows from areas where the water table is higher to areas
where it is lower. This flow can be either surface runoff in rivers or streams or
subsurface runoff infiltrating rocks and soils. The amount of runoff reaching surface
and groundwater can vary significantly, depending on rainfall, soil moisture, per-
meability, groundwater storage, evaporation, upstream use, and frost depth.
Subsurface water may return to the surface as groundwater flow, such as from a
spring, seep, or a water well, rivers, and oceans. Water returns to the land surface at a
lower elevation than the place where infiltration occurs, under the force of gravity or
gravity-induced pressures. Groundwater tends to move slowly and be replenished
Practice Problem 115

slowly, so it can remain in aquifers for thousands of years. Mainly, water flows
through the ground and eventually to the ocean where the cycle begins again.
Despite the difference, the mathematical description of the surface flow resembles
that of the flow in porous materials. Mass conservation for surface flow is

∂h
¼ ∇  ðvhÞ þ qe þ qr , ð12:32Þ
∂t
where qe is the exchange flux between surface and subsurface domains, and qr is a
general source/sink term. Manning’s equation is often used to establish a flow depth-
discharge relationship, where the velocity vector v in may be written as follows:
pffiffiffiffiffi
S f 2=3
v¼ h , ð12:33Þ
n
where Sf [L] is the friction slope and n [TL1/3] is Manning’s coefficient. Other
approaches are available to take the place of Manning’s equation (Nepf 1999).

Practice Problem

Experiment Description

A lab test for mixed saturated-unsaturated flow was conducted and documented by
Luthin et al. (1975). Figure 12.2 shows the schematic of a tank used in their test.
Initially, the sector tank was filled with the sand. A gravel-filled reservoir was placed
at the wide end of the tank to simulate a constant horizontal replenishment. Please
simulate the experiment and validate your simulation against the lab test.

Fig. 12.2 Schematic of the sector tank used and the locations of pressure measuring points
116 12 Hydro: Pore Water Movement

Model Development and Implementation

This is a typical mixed saturated-unsaturated flow problem, in which the boundary

between the two regions may change with time. To describe the flow, we need to
implement both the groundwater flow equation and the Richards equation, which
can be formulated with either the total head or the pressure head:
Total head:

Ss θðhÞ ∂h ∂ðh  zÞ
þ C ð hÞ ¼ ∇  ½K s K r ∇h ð12:34Þ
ϕ ∂t ∂t
Pressure head:

Ss θðψ þ zÞ ∂ðψ þ zÞ ∂ψ
þ C ðψ Þ ¼ ∇  ½K s K r ∇ðψ þ zÞ ð12:35Þ
ϕ ∂t ∂t
where Ss is the specific storage, θ is the water content, h is the total water head, Ks is
the saturated hydraulic conductivity, ϕ is the porosity, and Kr is the relative hydraulic
conductivity. It is worthwhile to mention that the boundary conditions and auxiliary
relationships such as SWCC and unsaturated (relative) hydraulic conductivity will
take different forms depending on the choice of the dependent variable.
The specific storage is assumed to be 105 as a dense sand was used in the test.
C(h) is the derivative of the function of water content with respect to the water head.
The saturated hydraulic diffusivity is 0.946 cm/min. The relative hydraulic conduc-
tivity and SWCC were obtained via curve fitting using the experimental data in
Fig. 12.3. The porosity is 0.6 and the specific gravity is 2.7.

0.30 1.0
Water content
Relative hydraulic conductivity

0.25 Boltzmann fit of SWCC

0.8
Relative K (Kr)
BiDoseResp fit of Kr
0.20
Water content

0.6
0.15
0.4
0.10

0.2
0.05

0.00 0.0
-0.8 -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0
Water head (m)

Fig. 12.3 Moisture content and hydraulic conductivity vs soil moisture tension
Practice Problem 117

The computational domain takes a cylindrical coordinate system (r and z). The
governing equation needs to be formatted accordingly if we do not use the nabla
symbol.

S=phi∗ dt ðH Þ ¼ K=r ∗ dr ðH þ zÞ þ dr ðK ∗ dr ðH þ zÞÞ þ dzðK ∗ dzðH þ zÞÞ

Alternatively, we can use the nabla symbol

S=phi∗ dt ðH Þ ¼ K ∗ divðgradðH þ zÞÞ

where S is the water content, H is the pressure head, and K is the general hydraulic
conductivity for both saturated and unsaturated soils. The “IF. . .THEN. . .” com-
mand needs to be used to define K, H, and the coexistence of the two accumulation
terms on the left-hand side of the governing equation.
Boundary conditions are key factors influencing the behavior of simulations. In
this case, the boundary conditions were established to approximate the experimental
conditions:
1. Bottom of the tank: no flux
2. Left (reservoir): constant total head
3. Top of the tank: no flux
4. Right (well): constant total head from the bottom to the free water surface; no flux
in the rest of the boundary
Here is the mesh of the computational domain (Fig. 12.4).
After the model is established, cases with different heights of water in the well are
simulated with different time steps. The definition of each parameter is illustrated in
the codes in the Appendix.

Typical Numerical Results

The major purpose of this practice problem is to guide the readers through the
development of numerical models using FlexPDE and the validation of the model
against experimental results. In the previous chapter, we used one practice problem
to show how to create a model using MATLAB PDE Toolbox and validate it against
analytical solutions. Figure 12.5 plots the water table (boundary between saturated
and unsaturated zones) and pressure head distribution at the end of 5 min in the test,
which are comparable to the experimental results in Fig. 12.6.
118 12 Hydro: Pore Water Movement

Fig. 12.4 Meshing with default settings

Appendix: FlexPDE Script File

{ POROUS.PDE
This problem describes the flow through an a soil tank, Luthin
and Orhun, Water Resrouces Research, Vol. 11, No. 6, 1975
}
TITLE 'Coupled saturated and unsaturated flow'
SELECT
CHANGELIM=200
fixdt = OFF
ngrid=10
regrid=ON
order=3
COORDINATES YCYLINDER
VARIABLES
H !Pressure Head
DEFINITIONS
K0=0.946/60/100
Ss=0.22473/0.0547*(1+exp((H+0.35549)/0.0547))^(-2)*exp((H
+0.35549)/0.0547)
Practice Problem 119

Fig. 12.5 Numerical solutions at 5 min for water table (upper) and pressure head (lower)
120 12 Hydro: Pore Water Movement

Fig. 12.6 Experimental results at 5 min (Luthin et al. 1975)

Kr=0.01628+1.09848*0.70914/(1+10^((-0.17064-H)/4.03296))
+1.09848*(1-0.70914)/(1+10^((-0.24585-H)/50.58359))
K= if H>=0 then K0 else K0*Kr
S= if H>=0 then 1e-8 else Ss
GW=if H>=0 then 1 else 0
Discharge=LINE_INTEGRAL(NORMAL(-k*grad(H)),'outlet')
phi=0.6
EQUATIONS
H : S/phi*dt(H)=K/r*dr(H+z)+dr(K*dr(H+z))+dz(K*dz(H+z))
INITIAL VALUES
H = 1.8288-z
BOUNDARIES
REGION 1
START 'outlet' (0,0) NATURAL(H)=0
LINE TO (7.0104,0) VALUE(H)=0.031-z
LINE TO (7.0104,0.031) NATURAL(H)=0
LINE TO (7.0104,1.8288) NATURAL(H)=0
LINE TO (0,1.8288) VALUE(H)=1.8288-z
LINE TO CLOSE
TIME 0 TO 300 BY 0.001
MONITORS
FOR cycle=10
CONTOUR(GW)
CONTOUR(H)
PLOTS
FOR TIME=10 BY 10 TO endtime
CONTOUR(GW)
CONTOUR(H)
Practice Problem 121

!surface(H)
ELEVATION (H) FROM (0,0) TO (7.0104,0)
ELEVATION (H) FROM (0,1.8288) TO (7.0104,1.8288)
HISTORY (H) AT (7,1.8)
HISTORY (Discharge)
END
Chapter 13
Concentrato: Transport of Dispersed Mass

Introduction

This chapter first clarifies the different definitions of concentrations. Following the
clarification on the basic terms, the formulation of the general governing equation of
the concentration field will be given out. Additional effort will be made to introduce
the three major dispersed mass transport mechanisms in detail, i.e., advection,
diffusion, and dispersion. The introduction will cover the definitions, origins in
physics, mathematical formulations, material properties, and applications in porous
materials. A practice problem developed from a documented test will be presented at
the end to exemplify the implementation of the concentration field.

Definitions and Mechanisms

The root “concentrato” is newly coined in this book to represent any process
involving the evolution or/and equilibrium of the concentration field. In other
words, it is used as a generic term for various mechanisms underlying mass transport
of dispersed phases in the context of multiphysics. Though the mass transport of
dispersed phases in a mixture (Fig. 13.1), i.e., solution, colloid, and suspension, is a
very basic monolithic physical process, a representative root such as concentrato has
seldom been used. This fact is possibly due to a few inconspicuous reasons related to
the nature and historical development of this topic.
First, mass transport includes three mechanisms: diffusion, dispersion, and advec-
tion. One unique feature of concentrato is that these three mechanisms are mutually
independent but not exclusive of each other. This feature enables any of them to
constitute a governing equation for the mass transport in a system in which the
corresponding mechanism is the only (considered) mechanism or the dominant one
therein, such as diffusion:

© Springer International Publishing AG, part of Springer Nature 2018 123

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_13
124 13 Concentrato: Transport of Dispersed Mass

Fig. 13.1 Dispersed phases

in different types of
mixtures

∂c
¼ ∇  ðD∇cÞ, ð13:1Þ
∂t
where c is the concentration and D is the diffusion coefficient. Also, as they are not
exclusive, they can work together to consider any combined mass transport mech-
anisms, such as combined diffusion and advection:

∂c
¼ ∇  ðD∇cÞ  v∇c, ð13:2Þ
∂t
where v is the velocity, which is associated with the dispersed phase or the bulk
phase (or dispersion phase) that the dispersed phase is distributed (dispersed) in. It
seems that, in applications, these three individual mechanisms have been more of
interest than the ensemble of the three. Accordingly, roots such as “diffuso” and
“advecto” are still present in the literature though not common, while “concentrato”
is rarely seen.
Second, mass transport is a topic in need of further integration. Mass transport
could refer to two apparently different types of processes. One is the mass transport
of a continuum, which includes an accumulation term and an advection term. The
equation describes that the mass accumulation of a material in a REV, usually a fluid,
is caused by the mass advected by the flux of the fluid into the REV. This mass
transport of the fluid occurs due to its own inertial, viscous, gravitational, and other
types of forces. In the meanwhile, we are also familiar with the second type of mass
transport such as the contaminant transport. The distinction between these two
makes mass transport a loose and confusing term. The solution is to treat the
aforementioned single-phase case as a special case of the mixture and resort to the
definition of concentration.
Concentration can be described in four different ways. The mass concentration ρi
is defined as the mass of a constituent mi divided by the volume of the mixture V:
mi
ρi ¼ : ð13:3Þ
V
The SI unit is kg/m3 (equal to g/L). This use is common in environmental engineer-
ing such as the study of sulfur dioxide in the air and that of dissolved oxygen in the
water.
Definitions and Mechanisms 125

The molar concentration ci is defined as the amount of a constituent ni (in mole)

divided by the volume of the mixture, V:
ni
ci ¼ : ð13:4Þ
V
The SI unit is mol/m3, while the use of mol/L (mol/dm3) is also common. This
expression is convenient when water is the dispersion fluid and chemical reactions
occur in the fluid.
The number concentration Ci is defined as the number of entities of a constituent
Ni in a mixture divided by the volume of the mixture V:

Ni
Ci ¼ : ð13:5Þ
V
The SI unit is 1/m3. This expression is helpful when dealing with large objects
contained in a bulk phase such as suspensions.
The volume concentration θi is defined as the volume of a constituent Vi divided
by the volume of the mixture V:

Vi
θi ¼ : ð13:6Þ
V
Being dimensionless, the volume concentration is expressed as a number, e.g., 0.18
or 18%. This expression is frequently adopted in multiphase and multicomponent
problems, such as the water content concept in soil science. The mass transport of a
single continuum can be described using the first definition of concentration.
Third, the terms of transport, e.g., diffusion and dispersion, have different mean-
ings in different disciplines; and even within multiphysics, they could mean different
things. In multiphysics, transport and diffusion could mean processes involving
mass, energy, or momentum. Therefore, in a broad sense, diffusion includes both
heat diffusion, mass diffusion, and momentum diffusion. Besides, terms such as
dispersion are widely used in chemistry, geology, materials science, and physics for
totally different things. This fact impedes the wide adoption of these terms in
multiphysics and prevents the use of these terms as a purely mass transport concept.
It is proposed in this book that these terms are reserved for mass transport only.
Accordingly, heat diffusion will be replaced with conduction, and momentum
diffusion mainly is referred to as momentum transfer via the Reynolds stress.
Finally, the physical nature of the three mechanisms blurs the distinctions
between them. Transport mechanisms need to be discussed with respect to the
mixture type. Mixtures can be roughly divided into suspension, colloid, and solu-
tion, which correspond to suspended particles (>1 μm), colloids (between 1 nm and
1 μm), and solutes (dissolved molecule, atoms, or ions). The scale boundaries are in
fact not as clear as noted. We know that diffusion includes both molecular diffusion
and turbulent diffusion which occur due to different reasons and at different scales.
Usually, molecular diffusion occurs in solutions and colloids; however, turbulent
126 13 Concentrato: Transport of Dispersed Mass

diffusion can be associated with all three types of mixtures. In general, molecular
diffusion is far better known than turbulent diffusion. This explains why in many
places diffusion is introduced as a microscopic phenomenon, which possibly means
molecular diffusion only. Another example is dispersion, which has a very similar
mathematical formation to diffusion. However, dispersion is a macroscopic phe-
nomenon caused by changes in velocities, which is totally different from the causes
and mechanisms of diffusion.
A general equation for the dispersed mass transport or concentration field,
including diffusion, dispersion, and advection, is formulated as Eq. 13.7. A combi-
nation of any two of them, such as diffusion and advection, is named with these two
mechanisms, that is, diffusion-advection equation:

∂c  
¼ ∇  ðDdiff ∇cÞ þ ∇  Ddisp ∇c  v  ∇c , ð13:7Þ
∂t |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflffl{zfflffl}
diffusion dispersion advection

where c can be replaced by the other definitions of the concentration. But such a
replacement requires to change the coefficients for diffusion Ddiff and dispersion
Ddisp accordingly.
In porous materials, the fact that only part of the volume is occupied by fluids
needs to be considered. Thus porosity and tortuosity need to be incorporated into the
formulation of the material properties or even the definition of the concentration.
Detailed mathematical formations for advection, diffusion, and advection will be
discussed in the following sections.

Advection
Definition

In a broader sense, advection is a transport mechanism of a substance or a conserved

property by a fluid due to the bulk motion of the fluid. An example of advection is
the transport of pollutants and sediments by the bulk water flow. Here the fluid may
be any material that contains thermal energy, such as water or air. In general, any
substance or conserved quantity can be advected by a fluid that can carry the
substance or quantity.
In advection, a fluid transports some conserved quantities or dispersed materials
in it via its bulk motion. The motion of the fluid is described mathematically as a
vector field, whereas the transported materials are described by scalar fields showing
the distribution of the materials over space, i.e., concentration fields. Therefore,
advection requires the presence of currents; and as a result, it cannot happen in rigid
solids. The term advection sometimes serves as a synonym for convection. However,
technically, convection covers the transport attributed to both diffusion and
advection.
Advection 127

Mathematical Description

Advection is the transport of mass due to the bulk fluid motion. A closer experi-
mental investigation can easily observe that the convective flux vector Jadv is
proportional to the fluid velocity v, so it has the same direction as the velocity.
The motion of a bulk fluid, therefore, contributes to the flux of its dispersed phases,
in addition to the flux due to diffusion. This flux can be formulated as follows:

Jadv ¼ v  ∇c, ð13:8Þ

where Jadv is the advective flux, v is the velocity field, and c is the concentration of
the dispersed material. One assumption underlying this equation is that the move-
ment of the dispersed phase is identical to that of the bulk phase carrying it. The
advection equation can be obtained with this flux, which is a special case of the
comprehensive mass transport equation:

∂c
¼ v  ∇c, ð13:9Þ
∂t
where v  — is the advection operator, and in Cartesian coordinates the advection
operator is

∂ ∂ ∂
v  ∇ ¼ vx þ vy þ vz , ð13:10Þ
∂x ∂y ∂z
where v ¼ (vx, vy, vz) is the velocity field in Cartesian coordinates. Both the advection
operator and the above advection equation have a much more general sense than
mass transport. The advection equation for a conserved quantity described by a
scalar field ψ can be expressed mathematically using the following general continu-
ity equation:

∂ψ
þ ∇  ðψvÞ ¼ 0 ð13:11Þ
∂t
Frequently, it is assumed that the flow is incompressible; as a result, the velocity
field satisfies the following equation to ensure continuity (mass balance):

∇  v ¼ 0, ð13:12Þ

and v is said to be solenoidal. If this is the fact, the above equation can be rewritten as

∂ψ
þ v∇  ψ ¼ 0: ð13:13Þ
∂t
The above is the general convection equation, a time-dependent, first-order partial
differential equation. The concept of the solenoidal field will be encountered in the
future chapters.
128 13 Concentrato: Transport of Dispersed Mass

In particular, if the flow is steady, then

v∇  ψ ¼ 0, ð13:14Þ

which shows that ψ is a constant along a streamline. Hence, ∂ψ/∂t ¼ 0, so ψ does

not vary in time.
If a vector quantity u, such as a magnetic field, is advected by a solenoidal
velocity field v, the above advection equation becomes

∂u
þ v∇  u ¼ 0: ð13:15Þ
∂t

Advection in Porous Materials

The advection in porous materials is much different from that in bulk fluids as only
part of the volume is occupied by the pore fluids. Taking this fact into consideration,
the advection equation can be modified into

∂c
ϕ ¼ vDarcy  ∇c, ð13:16Þ
∂t
where ϕ is the porosity and vDarcy is the Darcy velocity in Darcy’s law. vDarcy is
related to v via the effective porosity ϕeff as vDarcy ¼ vϕeff. ϕeff is defined using areas
instead of volume as used for ϕ. ϕeff is close to ϕ in many cases; therefore, the above
equation for porous materials is also close to the equation for bulk fluids.

Physical Implications of the Mathematical Description

The velocity in the above equation appears straightforward but could be hard to deal
with in real applications. The velocity of a molecule undergoing mass transfer
incorporates both an advective and a diffusive component. For a dilute species
(qualitative definition, species with low concentrations), the understanding is quite
clear as the dilute species is the diffusive component and the solvent is the advective
component. Correspondingly, the velocity is the velocity of the solvent or carrier
material. However, there are a few special cases in which the identification of the
velocity is not that easy. First, if a flow involves multiple phases, which occupy
comparable fractions, and all are discontinuous to some extent, then it may be hard to
identify an advective phase and its velocity. Second, for the transport of concentrated
species (qualitative definition, species with high concentrations), a more careful
definition of the bulk fluid velocity is required. For example, if the dispersed
phase is large and heavy conglomerates, then its velocity may not be identical to
that of the advective phase.
Diffusion 129

The boundary and relation between advection and diffusion also deserve consid-
eration. We understand advection as the mass transport due to the average velocity of
all molecules and diffusion as mass transport due to the instantaneously varying and
randomized velocity of individual molecules with reference to the average velocity
of the whole fluid. Hence, the choice of the reference velocity will affect the
definition of both the diffusive flux and the convective flux. For the transport of
dilute species, where one component dominates the momentum of the system, such
as a solvent or carrier gas, it is normal to take the velocity of the dominant
component as the reference velocity.
Natural convection, which is also called free convection or simply buoyant
convection, also deserves attention. It happens in the case where density variations
are caused by temperature and salinity differences. Without a forced flow, a
buoyancy-driven flow may still arise in the presence of temperature gradients
because of the density differences. For fluid parcels with dimensions larger than a
few milliliters in normal ambient conditions, convection currents usually cannot
achieve a steady state. This implies that this convection will contribute to mass
transport even when no forced convection is applied. A typical example is the
convection caused by gradients in salinity in oceans, which is usually referred to
as gravitational convection.

Diffusion

Diffusion is the process by which a substance is moved from one place to another
under the action of random fluctuations. On the molecular level, the cause is the
perpetual agitation of molecules; while on a higher level, its occurrence is attributed
to the turbulent eddies of the carrying fluid. Both types of diffusion are determined
by concentrations, leading to a proportionality between the mass flux and the
concentration gradient. This proportionality is characterized as a material property
and a key quantity in diffusion problems. Both types of diffusion can be formulated
using Fick’s laws. In the following paragraphs, the general formulation of Fick’s
laws will be introduced first. Then special cases including the dependence of the
diffusion coefficient on various factors, diffusion in porous materials,
multicomponent diffusion, and turbulent diffusion will be introduced consecutively.
The driving force for molecular diffusion is the dynamic oscillation of molecules,
which also contributes to the macroscopic thermal state. As we know, molecules stay
in motion at temperatures above absolute zero. This motion leads to their kinetic
energy. The motion and collision of the particles are random at the microscopic scale,
however, follow thermodynamic rules at the macroscopic scale. These collisions are
common even in the air at an atmospheric pressure which can be hardly viewed as a
dense fluid. If the concentration of a species is initially not uniform, e.g., the
concentration is greater in one region of a fluid than another, then, diffusion will
enable mass transfer in the direction toward a more uniform concentration in the fluid.
Turbulent diffusion is a special category of diffusion pertaining to turbulent fluid
systems. In turbulent diffusion, the variation of the concentration occurs when the
130 13 Concentrato: Transport of Dispersed Mass

turbulent fluid system reaches critical conditions in response to shear flow. Such a
condition is usually caused by high velocities while it can also result from steep
concentration gradients, steep density gradients, or their combinations. The rate of
turbulent diffusion is usually much higher than molecular diffusion. Therefore, it is
necessary to take turbulent diffusion into consideration for diffusion in fluids where
turbulent flow patterns are dominant.

Mathematical Description: Fick’s Laws

Fick’s laws provide a simple and universal mathematical description of diffusion.

These physical laws were reported by Adolf Fick in the nineteenth century. Two
different laws exist:
1. The first Fick’s law states that the molar flux due to diffusion is proportional to the
concentration gradient.
2. The second law states that the rate of the change in the concentration at any point
is proportional to the second spatial derivative of the concentration at that point.

Fick’s First Law of Diffusion

Fick’s first law of diffusion can be written in a modern mathematical form as

Jdiff ¼ Ddiff ∇c, ð13:17Þ

where Jdiff is the mass flux, Ddiff is the diffusion coefficient, and c is the concentra-
tion. Ddiff is a vector here to allow for anisotropy. Other definitions of the concen-
tration can also be used, but consistent units need to be used for the diffusion
and flux.

Fick’s Second Law of Diffusion

From the continuity equation for mass balance, we have

∂c
þ ∇  Jdiff ¼ 0: ð13:18Þ
∂t
Then we can derive Fick’s second law directly:

∂c
¼ ∇  ðDdiff ∇cÞ ¼ Ddiff ∇2 c, ð13:19Þ
∂t
Diffusion 131

where Ddiff is moved out of the Laplacian operator if Ddiff is a constant along
different directions. Usually, the assumption is also a good approximation for the
diffusion in solids, diffusion of chemicals in a liquid solvent, and diffusion of dilute
species in the gas phase, such as carbon dioxide in the air. However, this may not be
the case in porous materials, where anisotropy is common.

Understanding of Diffusion Coefficient

In dilute species transport, the flux due to diffusion, as formulated by Fick’s first law,
only depends on the diffusion coefficient, which quantifies the interaction between
the solute and the solvent. The diffusion coefficient is the magnitude of the molar
flux through a unit surface area under a unit concentration gradient. Therefore, it is
analogous to the thermal diffusivity in heat transfer, hydraulic diffusivity in ground-
water flow, and electric conductivity in electric currents in conductors. When
moving to the mixtures with multiple dispersed phases/species, the above Fick’s
law will also be valid:

Jdiff , i ¼ Ddiff , i ∇ci : ð13:20Þ

Or in 1D,

J diff , i ¼ Ddiff , i ∇ci , ð13:21Þ

so

jJ diff , i j
Ddiff , i ¼ : ð13:22Þ
j∇ci j

However, in multicomponent diffusion, the diffusion coefficient is possibly not a

constant or composition-independent anymore. Especially, the intermolecular depen-
dencies cannot be overlooked as the interactivity of the molecules of different species
becomes prevalent. To allow for the interactivity, it is more accurate to formulate the
diffusion coefficient using a tensor. Also, the equation for diffusion needs to be
altered to relate the mass flux of one chemical species to the concentration gradients
of all chemical species. Such a mechanism can be formulated using the Maxwell-
Stefan theory of diffusion (Krishna and Wesselingh 1997). The Maxwell-Stefan
theory is a more comprehensive version of the “classical” Fick’s diffusion theory.
Fick’s laws also apply to turbulent diffusion. The diffusion coefficient associated
with turbulent diffusion is called the eddy diffusion coefficient or eddy diffusivity.
The magnitude of the diffusion coefficient in molecular diffusion varies signifi-
cantly from mixture to mixture. In gaseous solutions, a typical diffusion coefficient
for a molecule lies in the range of 106 to 105 m2/s. By contrast, the diffusion
coefficient for molecules dissolved in liquids is much slower. In aqueous solution
such as water, typical diffusion coefficients range from 1010 to 109 m2/s. For
132 13 Concentrato: Transport of Dispersed Mass

example, the diffusion coefficient for salt (NaCl) diffusing into water is about
1.5  109 m2/s. Accordingly, it could take more than 1 month for sugar in a cup
with a height of 5 cm to become uniformly distributed. By contrast, the turbulent
diffusion coefficient in typical rivers could be 0.01 m2/s along the horizontal
direction and 0.1 m2/s along the vertical direction (Roberts and Webster 2002)
The molecular diffusion coefficient can be predicted in some simple cases. In an
ideal gas, the following relationship can be derived from the Maxwell-Boltzmann
distribution by considering the mean free path and average velocity of the molecule:
3
T2
Ddiff / , ð13:23Þ
p
where T and p are the temperature and pressure, respectively. According to this
equation, diffusion would be faster in hotter and more rarefied gasses. In a viscous
fluid, the diffusion coefficient of particles or large molecules obeys the Stokes-
Einstein equation (Logan 2012):

kT
Ddiff ¼ , ð13:24Þ
6πμr

where k is the Boltzmann constant, μ is the solvent viscosity, and r is the radius of the
diffusing particle. The equation is obtained by calculating the drag force from
Stokes’ law.

Diffusion Coefficient in Porous Materials

In a porous material, the fluid phase must travel a tortuous path that winds through
the pores perforating the solid skeleton. Therefore, the effective diffusion coefficient
alters from the theoretical diffusion coefficient for a bulk solution due to various
effects (Fig. 13.2). For example, the available cross section for diffusion in the
porous material is smaller than for the bulk solution. This effect can be considered
using the effective porosity defined in terms of areas, which can be approximated
using the porosity of the porous material defined in terms of volumes. Additionally,
the distance that a molecule must travel from one point to another in the porous
material is greater than the distance between these points because the molecule must
navigate between the solid skeleton. As a result, the real concentration gradient is
less than the apparent concentration gradient.
This effect is considered by multiplying the diffusion coefficient by the tortuosity
(τ > 1). With the above considerations, Fick’s first law can be modified into

Jdiff ¼ Ddiff
eff
∇c, ð13:25Þ
eff
where Ddiff is the effective diffusion coefficient. This effective diffusion coefficient is
smaller than that in a continuous liquid phase due to consideration of the porosity ϕ
and tortuosity τ
Dispersion 133

Fig. 13.2 Flow and particle

transport in porous materials
affected by (a) velocity
variation, (b) particle size
distribution, and (c)
tortuosity

ϕ
eff
Ddiff ¼ Ddiff , ð13:26Þ
τ
where tortuosity is defined as the ratio of the length of the real traveling path to the
distance between the two ends of the path.
Here the flux Jdiff and concentration gradient — c are still defined with respect to
the solution where diffusion actually occurs instead of the whole porous material. It
is worthwhile to mention that the concentration and flux defined with respect to the
whole porous material are also used for porous materials.
The tortuosity is easy to obtain. Some correlations relating the tortuosity to the
porosity are available for rough calculations. Typical correlations are the Millington-
Quirk equation (Campbell 1974)

τ ¼ ϕ 3 ,
1
ð13:27Þ

and the Bruggeman equation (Brakel and Heertjes 1974)

τ ¼ ϕ 2 :
1
ð13:28Þ

It is also very common that the effective diffusion coefficient is anisotropic in the
porous material. In this case, the rate of diffusion depends on the direction of the
concentration gradient. The corresponding mathematical treatment is to formulate
the diffusivity using a vector.

Dispersion

Diffusion can be significant or even dominant in the vapor transport in unsaturated

zones of porous materials. However, diffusion is usually not a significant mass
transport mechanism in saturated pore liquid flow where dispersion could be the
dominant one. Dispersion, or dispersive mass transfer, is also the spreading of
dispersed mass from highly concentrated areas to less concentrated areas. However,
this dispersed mass transport mechanism is caused by non-ideal flow patterns like
turbulent diffusion. The difference is that the turbulent diffusion is still one type of
random mixing which is related to the turbulence (perturbation) in the dispersion
liquid while dispersion relies heavily on the average velocity. Another major
134 13 Concentrato: Transport of Dispersed Mass

Fig. 13.3 Dispersion and

non-ideal flow conditions

difference is that dispersion is a macroscopic phenomenon while diffusion occurs at

much smaller scales because diffusion is related to the vibration of molecules, i.e.,
Brownian motion.
The non-ideal flow patterns refer to the deviations from a “plug flow” in a
conceptual pipe housing the flow, in which the velocity of the fluid is assumed to
be a constant across any cross section perpendicular to the axis of the pipe. The
following example can be used to understand the deviation from such ideal flow
patterns (Deen 1998). If the flow velocity is uniform everywhere, i.e., plug flow
(Fig. 13.3), then molecules at different places in the flow will move with the same
convective velocity. In that case, only the diffusion rates differentiate between them.
However, in a non-ideal flow, velocity gradients exist along the radial direction of
the pipe – the fluid at the center of the pipe moves faster than that adjacent to the
walls. As a result, different streamlines have different concentrations. Molecules
then tend to go across streamlines, leading to diffusion in the radial directions. This
special transport mechanism, which results from this smearing out or enhanced mass
transfer due to the combined effect of radial diffusion and velocity differences, is
interpreted as dispersion. Surprisingly, dispersion can decrease with an increase in
diffusion coefficients. This is because the molecules will not get enough time to be
transported far along each streamline when the diffusion is too strong along the
radial direction.
Based on the above description, dispersive mass flux is analogous to diffusion,
and it can also be described using Fick’s first law:

Jdisp ¼ Ddisp ∇c, ð13:29Þ

where c is the mass concentration of the dispersed species and Ddisp is the dispersion
coefficient. Dispersion is usually more significant than diffusion in typical
convection-diffusion problems.
The dispersion coefficient is frequently modeled as the product of a characteristic
length scale, α, and the fluid velocity, v:

Ddisp ¼ αv: ð13:30Þ

Dispersion in porous materials is usually appreciable because of the complicated

structure of the solid skeleton which can significantly deviate the flow patterns from
Practice Problem 135

the ideal flow situation and thus enhance dispersion therein significantly. The
dispersion coefficient or the characteristic length scale can be obtained directly via
experiments (Fischer 1966; Irwin et al. 1996).

Practice Problem

Problem Description

An experiment and its analytical solution for typical concentration fields were
reported by Massabo et al. (2011). As shown in Fig. 13.4, the experiment was
conducted in a Perspex box, which was 0.28 m long, 0.2 m wide, and 0.01 m
thick. The box was filled with transparent glass beads to simulate a porous material.
There is a constant flow from the left to the right with a velocity of 2.9  104 m/s
along the length direction (horizontal). The combined diffusion and dispersion
coefficients considering the porosity and tortuosity of the porous material are
3.322  107 m2/s and 2.672  107 m2/s along the horizontal direction and vertical
direction, respectively. Dye was injected into the box at the location that is 0.05 m
from the left side and 0.1 m from the top. Unfortunately, the accurate formulation of
the pulse source is hard to determine based on the description in the paper. Here we
can assume it is a square pulse with a magnitude of 0.1 between t ¼ 0 s and t ¼ 1 s in
a circular area with a radius of 0.001 m. The initial concentration is zero across the
domain. All the boundaries have no mass influx or outflux. Please simulate
this experiment, and compare the simulation result against the analytical solution
at t  v/L ¼ 1/50 (18.96 s) and t  v/L ¼ 1/10 (t ¼ 94.8 s).
For simplicity, the effective porosity is assumed to be the same as the porosity.

Results

Direct comparisons between the simulation results and the analytical (Fig. 13.5) and
experimental results from Massabo et al. (2011) are difficult due to the unknown
source function. In Fig. 13.6, we normalized the concentration for a quantitative
comparison for the concentration distributions along x symmetry axis at two differ-
ent times (Figs. 13.5 and 13.6).

Fig. 13.4 Schematic of the

experiment
136 13 Concentrato: Transport of Dispersed Mass

Fig. 13.5 The concentration distributions at 18.96 s (upper) and 94.8 s (lower)
Practice Problem 137

Fig. 13.6 Comparison

between simulation and 1.0
analytical results

Normalized concentration
0.8

0.6 Current Simulation

Current Simulation
Analytical
0.4
Analytical

0.2

0.0

0.0 0.2 0.4 0.6 0.8 1.0

Normalized distance from source

Appendix: FlexPDE Script File

{ Concentration field with diffusion, advection, and a pulse source }
TITLE 'Concentration'
VARIABLES
u(1e-12)
DEFINITIONS
D_x=3.322e-7 { Diffusivity in x/horizontal/longitudinal direction }
D_y=2.672e-7 { Diffusivity in y/vertical direction }
vx=2.9e-4
M=0* UPULSE(t-0, t-01) { Rectangular pulse between t=0 and t=1 }
INITIAL VALUES
u = 0.0
EQUATIONS
Dx(D_x*Dx(u)) + Dy(D_y*Dy(u)) - vx*Dx(u)+M= dt(u)
BOUNDARIES
REGION 1
M=0
START(0,-0.1)
NATURAL(u) = 0
LINE TO (0.28,-0.1)
NATURAL(u) = 0
LINE TO (0.28,0.1)
NATURAL(u) = 0
LINE TO (0,0.1)
NATURAL(u) = 0
LINE TO CLOSE
REGION 2
M=0.1*UPULSE(t-0, t-1) { Dye was injected into Region 2 betweent=
0~1 as a pulse }
START(0.051,0)
ARC (CENTER=0.05,0)
138 13 Concentrato: Transport of Dispersed Mass

ANGLE=360
TO CLOSE
TIME 0 TO 425 BY 1
PLOTS
FOR t=0 BY 0.001 TO 0.01 BY 0.1 TO 3 BY 1 TO ENDTIME
CONTOUR(u)
ELEVATION(u) FROM (0.05,0) TO (0.28,0)
FOR t=94.8/5
ELEVATION(u) FROM (0.05,0) TO (0.28,0)
FOR t=94.8
ELEVATION(u) FROM (0.05,0) TO (0.28,0)
END
Chapter 14
Mechano: Stress and Strain

Introduction

This chapter discusses the mechanical response. In the context of multiphysics, the
word “mechanical field” is usually used to describe physical processes involving this
aspect of the material behavior, in which quantities such as stress, strain, and
displacement change with reference to a geometrical description of an object. A
more widely used term for the topic is stress-strain analysis or stress analysis. Stress
analysis is an engineering practice to determine the stresses and strains in materials
and structures subjected to forces or loads. Stress analysis is a primary interest of
civil, mechanical, and aerospace engineers. The practice lays down a theoretical
basis for the design of structures and components, such as buildings, bridges,
aircrafts, mechanical parts, and earth structures. Stress analysis is also used in the
maintenance and retrofit of such objects for the investigations into the causes of their
failures.
The input data for stress analysis are a geometrical description of the object,
material properties, and a loading condition. The output data are typically a quan-
titative description of the stress as well as strain and deformation across the compu-
tational domain under the given loading condition. Stress analysis for porous
materials is also built on the assumption of REV and the basic theories of continuum
mechanics. Therefore, all material properties are homogeneous at scales smaller than
that of the REV. The assumption of REV needs to be ensured to make sure that the
mathematical description extended from continuum mechanics can surrender satis-
factory results for problems in porous materials.
The scope of mechanical response needs to be more clearly defined to differen-
tiate the mechanical field from other relevant topics and also to clarify what will be
discussed in the rest of this chapter.
First, stress and strain analysis is restricted to solids. The study of stresses in
liquids and gasses is the subject of fluid mechanics, which is a multiphysical

© Springer International Publishing AG, part of Springer Nature 2018 139

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_14
140 14 Mechano: Stress and Strain

phenomenon and thus will not be counted as the mechanical field which is a
monolithic physical phenomenon.
Second, the introduction to the mechanical field in this book will be limited to
homogeneous linear elastic materials and the small strain theory. The relationships
between several common quantities, i.e., stress, strain, and displacement, and the
choice of the dependent variable(s) from these terms make the mechanical field
appear much more complicated than other physical fields such as heat transfer. The
explanation of these relationships is thus the major content of this chapter. An
intuitive goal of stress analysis is to determine the distribution of the internal stresses
throughout the system under given external forces. In principle, this implies the
determination of, either implicitly or explicitly, the Cauchy stress tensor at every
point. However, the stress is usually not the dependent variable. The stress needs to
be connected to the displacement, which is preferred as the dependent variable. For
the purpose, the stress needs to be linked to the strain and then to the displacement.
For the purpose, the relationship between stress and strain, i.e., constitutive relation-
ship, is required as predetermined material properties. Among different types of
constitutive relationships, e.g., elastic, elastoplastic, and viscoplastic, the linear
elastic relationship is one of the simplest and most common ones and thus will be
introduced in this chapter. The second step for the purpose is to relate the strain to the
displacement via strain theories, i.e., small strain theory (infinitesimal strain theory)
and large strain theory (finite strain theory). The former will be discussed in this
book considering its predominant use in numerical simulation.
Third, the external forces (loading) may be body forces such as gravity or
magnetic attraction, which act throughout the volume of a material, or concentrated
loads such as friction between parts. These forces are usually given or can be easily
identified for real problems. Therefore, they will not be extensively discussed.
Furthermore, in stress analysis, the physical causes of external forces and the precise
nature of the materials, i.e., other aspects of the materials apart from the mechanical
behavior, are usually not of interest.
In the following sections, we will first introduce the stress concept including its
physical meaning and mathematical formulations. Then the strain and its relation-
ship to the displacement including the small strain theory will be explained.
Followed is the introduction to the constitutive relationship for linear elasticity.
The stress, strain, and constitutive relationship will be integrated into the mechanical
balance equation to obtain the governing equation for the mechanical field. Finally,
an example will be given to demonstrate the implementation of the mechanical field.

Stress

The most common mathematical description of stress is the Cauchy stress tensor. In
continuum mechanics, the Cauchy stress tensor σ, or called the true stress tensor or
simply the stress tensor, is a second-order tensor. The Cauchy stress tensor can be
Stress 141

Fig. 14.1 Stress tensor and

components

used to formulate the stress state at any point inside a material. In a general 3D space,
the Cauchy stress tensor has nine components σ ij:
2 3 2 3 2 3
σ 11 σ 12 σ 13 σ xx σ xy σ xz σx τxy τxz
σ ¼ 4 σ 21 σ 22 σ 23 5 ¼ 4 σ yx σ yy σ yz 5 ¼ 4 τyx σy τyz 5: ð14:1Þ
σ 31 σ 32 σ 33 σ zx σ zy σ zz τzx τzy σz

The three stress components on the diagonal of the matrix are normal stresses, which
act perpendicular to the plane, while those of the diagonal are called shear stresses as
they are parallel to the plane. Equation 14.1 shows three common math representa-
tions of the stress tensor. The rightmost one is frequently used in soil mechanics to
better differentiate normal stresses from shear stresses.
The use of the Cauchy stress tensor for formulating the stress state at a point can
be understood in the following way. If we “zoom in,” any infinitesimal point can be
viewed as a cube as shown in Fig. 14.1. The stress state at the point can be
represented by the stresses on the surfaces of the cube. Among the six surfaces of
the cube, three surfaces have identical stresses to the other three. Therefore, the stress
tensor can be made with the nine stresses on these three surfaces. This is what was
obtained in Eq. 14.1.
We can further define the stress vector T(n) to represent the stresses on a plane
whose outward normal vector is n. With the stress vector, the Cauchy stress tensor
can be written in the following form:
2 3 2 3
Tð1Þ σ 11 σ 12 σ 13
σ ¼ σ ij ¼ 4 Tð2Þ 5 ¼ 4 σ 21 σ 22 σ 23 5: ð14:2Þ
Tð3Þ σ 31 σ 32 σ 33

However, there is an infinite number of ways to align the cube. If the cube is aligned
along different directions, the stresses on its surfaces, i.e., the components of the
Cauchy stress tensor, will also be different. This leads to a contradiction: the stress
state for one point is the same regardless of the alignment of the cube; however, the
142 14 Mechano: Stress and Strain

components of the stress tensor are different for different alignments. This contra-
diction can be compromised by the Cauchy stress theorem (or called the Cauchy
fundamental theorem). This theorem states that the stress vector on any other plane
passing through a point can be found through coordinate transformation equations.
Equivalently, we can understand the three stress vectors as the projections of the
stress tensor onto three directions such as three axes in the Cartesian coordinate
system. If we rotate the axes, the projections will be different; however, the tensor
stays the same. This understanding leads to the following equation:

TðnÞ ¼ n  σ, ð14:3Þ

or

ðnÞ
T j ¼ σ ij ni : ð14:4Þ

This equation implies that the stress vector T(n) on a plane with a normal unit vector
n is the projection of the stress tensor σ on that plane. In a 3D space, there are three
independent stress vectors or projections for each stress tensor. Likewise, a stress
tensor for a 2D case consists of two independent stress vectors.
In this chapter and, in fact, in most of stress analyses, the Cauchy stress tensor is
usually introduced and adopted acquiescently. However, please note that the Cauchy
stress tensor is used for stress analysis of material bodies experiencing small
deformations. It is thus a central concept in the linear theory of elasticity. For large
deformations, which are also called finite deformations, other measures of stress are
required, such as the Piola-Kirchhoff stress tensor, Biot stress tensor, and Kirchhoff
stress tensor (Hackett 2016).

Strain

The definition of strain usually requires the introduction of two other closely related
terms: displacement and deformation. Displacement can be defined as the movement
of individual material points in an object due to external loads. This movement of the
point with respect to a reference system of axes can be described using a vector:
displacement vector. Displacements can be classified into the following four types as
shown in Fig. 14.2:
1. Translation of points
2. Rotation of lines
3. Change in length, i.e., elongation and contraction
4. Distortion, i.e., angle change between lines

Fig. 14.2 Schematic of translation, rotation, length change, and distortion

Strain 143

Fig. 14.3 Displacement and deformation

The first two types of displacements occur because of the movement of the entire
body, i.e., rigid body movement, while the latter two are associated with the local
deformation of the body. In the latter two, individual points of the body move
relative to each another as displacements induced by applied loads alter the size
and/or the shape of the body. The change in any dimension associated with these two
types of relative displacements is defined as deformation. Deformation provides a
global description of the object. However, the understanding of the internal of the
object requires local descriptions for all the points inside the object. This calls for
stresses and strains. Strain can be roughly understood as deformation normalized
with respect to the original or current geometry of the object or other things,
depending on the choice of strain theories. The strain theory that we usually use is
the small strain theory, in which the original geometry is adopted for the normali-
zation. For example, the strain of the rod subjected to axial loading (1D) can be
defined as the deformation per unit length of a body.
Similar to stresses, strains may also be classified into the normal strain and shear
strain, i.e., acting perpendicular to and parallel to the plane (surface of a material
element), respectively. In order to derive the mathematical description of the normal
strain, let us consider a 2D infinitesimal rectangular material element with dimen-
sions dx  dy as illustrated in Fig. 14.3. After deformation, the rectangular becomes
a rhombus. Taking the side with the length of dx, for example, the length of this side
in the deformed rectangular, i.e., the rhombus, will be:
144 14 Mechano: Stress and Strain

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
   ffi
2 2
∂u
Deformed length ¼ dx þ ∂u ∂x
x
dx þ ∂xy dx
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2
∂ux ∂ux ∂uy
¼ dx 1 þ 2 þ þ
∂x ∂x ∂x

As mentioned, this book restricts the introduction to the small strain theory. In the
case of small strains, the squares of the derivatives are negligible as the derivatives
are small, thus we have

∂ux
Deformed length  dx þ dx
∂x
The normal strain in the x direction of the rectangular element is defined by

deformed length ∂ux

εx ¼  : ð14:5Þ
original length ∂x

Similarly, the normal strain in the y direction and z direction, becomes

∂uy
εy ¼ ð14:6Þ
∂y
∂uz
εz ¼ : ð14:7Þ
∂z
Likewise, we can obtain the formulation of the shear strain in the small strain
theory:

∂uy ∂ux
γ xy ¼ α þ β ¼ þ : ð14:8Þ
∂x ∂y
Combining the formations of the normal and shear strain, we can obtain the
infinitesimal strain tensor as

1 
ε¼ ∇u þ ð∇uÞT , ð14:9Þ
2
or in the index notation as

1
εij ¼ ui, j þ u j, i , ð14:10Þ
2
or in the matrix notation as
Constitutive Relationship: Linear Elastic Hooke’s Law 145

2    3
∂ux 1 ∂ux ∂uy 1 ∂ux ∂uz
6 þ þ
2 3 6 ∂x 2 ∂y ∂x 2 ∂z ∂x 7
7
σ xx σ xy σ xz 6    7
6 1 ∂uy ∂ux ∂uy 1 ∂uy ∂uz 7
4 σ yx σ yy σ yz 5 ¼ 6 þ þ 7:
6 2 ∂x ∂y ∂y 2 ∂z ∂y 7
σ zx σ zy σ zz 6 7
6     7
4 1 ∂uz ∂ux 1 ∂uz ∂uy ∂uz 5
þ þ
2 ∂x ∂z 2 ∂y ∂z ∂z
ð14:11Þ

Constitutive Relationship: Linear Elastic Hooke’s Law

The general equation for Hooke’s law is

σ ¼ C : ε, ð14:12Þ

where σ is the Cauchy stress tensor, ε is the infinitesimal strain tensor, and C is the
fourth-order stiffness tensor. However, in engineering applications, we usually do
not deal with the above equation in which the stress and strain are second-order
tensors, while the stiffness is a fourth-order tensor. Instead, we usually adopt the
Voigt notation which reduces the orders of the above tensors by means of specified
rules and various types of symmetry.

Voigt Notation

The Voigt notation reduces the order of the stress and strain tensor from 2 to
1. Accordingly, the fourth-order stiffness tensor C (Cijkl) degenerates into a
second-order tensor C(Cαβ). To achieve the goal, we first take advantage of the
symmetry of the stress and strain tensors: σ ij ¼ σ ji and εij ¼ εji. Therefore, there are
only six independent components in the following stress tensor:
2 3
σ 11 σ 12 σ 13
σ ¼ 4 σ 12 σ 22 σ 23 5: ð14:13Þ
σ 13 σ 23 σ 33

So we can use a vector (or 1D array) including the six independent components to
represent the above second-order tensor (or 2D array):
146 14 Mechano: Stress and Strain

2 3
σ 11
6 σ 22 7
6 7
6 σ 33 7
6
σ¼6 7: ð14:14Þ
7
6 σ 12 7
4 σ 13 5
σ 23

Similarly, the strain tensor can be formulated as

2 3 2 3
ε11 ε11
6 ε22 7 6 ε22 7
6 7 6 7
6 ε33 7 6 ε33 7
ε¼6 7¼6 7
6 2ε12 7 6 γ 12 7: ð14:15Þ
6 7 6 7
4 2ε13 5 4 γ 13 5
2ε23 γ 23

To comply with the above changes, the fourth-order stiffness tensor is reduced to a
second-order tensor as follows:
2 3
C 1111 C 1122 C1133 C 1112 C1113 C 1123
6 C 2211 C 2222 C2233 C 2212 C2213 C 2223 7
6 7
6 C 3311 C 3322 C3333 C 3312 C3313 C 3323 7
C¼6
6 C 1211
7: ð14:16Þ
6 C 1222 C1233 C 1212 C1213 C 1223 7
7
4 C 1311 C 1322 C1333 C 1312 C1313 C 1323 5
C 2311 C 2322 C2333 C 2312 C2313 C 2323

Besides symmetries, Voigt notation also adopts the following mapping rules for
tensor indices:

ij or kl ¼ 11 22 33 12, 21 13, 31 23, 32

+ + + + + + +
α or β ¼ 1 2 3 4 5 6

With the mapping, the stress tensor will be simplified into

2 3 2 3
σ 11 σ1
6 σ 22 7 6 σ 2 7
6 7 6 7
6 σ 33 7 6 σ 3 7
6
σ¼6 7 ¼ 6 7; ð14:17Þ
7 6 7
6 σ 12 7 6 σ 4 7
4 σ 13 5 4 σ 5 5
σ 23 σ6

and the strain tensor is transformed in a similar way:

Constitutive Relationship: Linear Elastic Hooke’s Law 147

2 3 2 3 2 3 2 3
ε11 ε1 ε11 ε1
6 ε22 7 6 ε2 7 6 ε22 7 6 ε2 7
6 7 6 7 6 7 6 7
6 ε33 7 6 ε3 7 6 ε33 7 6 ε3 7
6
ε¼6 7 6 7 6 7 ¼ 6 7:
7¼6 7¼6 7 6 7 ð14:18Þ
6 2ε12 7 6 2ε4 7 6 γ 12 7 6 γ 4 7
4 2ε13 5 4 2ε5 5 4 γ 13 5 4 γ 5 5
2ε23 2ε6 γ 23 γ6

With Cijkl ) Cαβ, the degenerated stiffness tensor C can be expressed with the new
indices as:
2 3
C11 C 12 C13 C 14 C 15 C16
6 C22 C 22 C23 C 24 C 25 C26 7
6 7
6 C33 C 33 C33 C 34 C 35 C36 7
C¼6
6 C41
7: ð14:19Þ
6 C 42 C43 C 44 C 45 C46 7
7
4 C51 C 52 C53 C 54 C 55 C56 5
C61 C 62 C63 C 64 C 65 C66

The matrix Cαβ can be proven to be symmetric. This symmetry is attributable to

the existence of a strain energy density function which satisfies σ ij ¼ ∂W
∂εij
. Hence,
there are at most 21 independent elements in the matrix of Cαβ.
In the case of orthotropic materials, which have material properties that differ
along three mutually orthogonal twofold axes of rotational symmetry, the above
stiffness matrix further reduces to the following form with nine independent
elements:
2 3
C11 C 12 C 13 0 0 0
6 C12 C 22 C 23 0 0 0 7
6 7
6 C13 C 23 C 33 0 0 0 7
C αβ ¼6
6 0
7: ð14:20Þ
6 0 0 C44 0 0 77
4 0 0 0 0 C 55 0 5
0 0 0 0 0 C66

The most common materials or assumptions that we encounter in applications are

homogeneous isotropic linear elastic materials. For such materials, the constitutive
relationship can be formulated using an even simpler stiffness matrix. One of the
most common forms of the matrix is
2 3 2 32 3
σ 11 λ þ 2μ λ λ 0 0 0 ε11
6 σ 22 7 6 λ λ þ λ 0 07 6 7
6 7 6 2μ 0 76 ε22 7
6 σ 33 7 6 λ λ λ þ 2μ 0 0 07 6 ε33 7
6 7 6 76 7,
6 σ 12 7 ¼ 6 0 0 0 μ 0 07 6 7 ð14:21Þ
6 7 6 76 2ε12 7
4 σ 13 5 4 0 0 0 0 μ 0 54 2ε13 5
σ 23 0 0 0 0 0 μ 2ε23
148 14 Mechano: Stress and Strain

where λ and μ are the Lamé constants. μ also has the physical meaning of shear
modulus and is frequently represented using G in engineering.
Using the index notation, we can write the above equation as

σ ij ¼ λεkk δij þ 2μεij : ð14:22Þ

Elastic Modulus

The constitutive relationship of any homogeneous isotropic linear elastic material

can be formulated by two independent elastic moduli, such as the Lamé constants,
i.e., μ and λ. Common elastic moduli also include Young’s modulus E, Poisson’s
ratio ν, and bulk modulus K. As shown in the following table, any two out of the five
can be represented by any other two moduli.

Eν Eμ λ, μ K, μ Kν
K E
2ð12νÞ
Eμ
3ð3μE Þ
λþ 3μ
2 K K
λ Eν
ð1þνÞð12νÞ
μðE2μÞ λ K  23 μ 3Kν
1þν
3μE
μ E
2ð1þνÞ
μ μ μ 3K ð12νÞ
2ð1þνÞ
E E E μð3λþ2μÞ 9Kμ 3K(1  2ν)
λþμ 3Kþμ
ν ν E
1 λ
2ðλþμÞ
3K2μ ν
2μ 2ð3KþμÞ

Among the five quantities, the combination of E and ν is possibly the most popular
one in engineering applications. The constitutive relationship for isotropic linear
elastic materials is formulated using E and ν as follows:
2 3
1ν ν ν 0 0 0
2 3 2 3
σ 11 6 ν 1ν ν 0 7
6 0 0 7 ε11
6 σ 22 7 6 ν ν 1ν 0 7 6 7
6 7 6 0
1  2ν
0 76 ε22 7
6 σ 33 7 E 6 0 7
0 76 33 7
6 ε
6 7¼ 6 0 0 0 7
6 σ 12 7 ð1 þ νÞð1  2νÞ 6 2 76 2ε12 7:
6 7 6 1  2ν 76 7
4 σ 13 5 6 0 0 0 0 0 7 4 2ε13 5
6 2 7
σ 23 4 1  2ν 5 2ε23
0 0 0 0 0
2
ð14:23Þ

If we inverse the stiffness matrix for the material, then we can obtain the following
equation in terms of ε ¼ C1σ, in which C1 is the compliance matrix:
Constitutive Relationship: Linear Elastic Hooke’s Law 149

2 3 2 32 3
ε11 1 ν ν 0 0 0 σ 11
6 ε22 7 6 ν ν 0 7 6 7
6 7 6 1 0 0 76 σ 22 7
6 ε33 7 1 6 ν ν 1 0 0 0 7 6 σ 33 7
6 7¼ 6 76 7: ð14:24Þ
6 2ε12 7 E 6 0 2 þ 2ν 0 7 6 7
6 7 6 0 0 0 76 σ 12 7
4 2ε13 5 4 0 0 0 0 2 þ 2ν 0 5 4 σ 13 5
2ε23 0 0 0 0 0 2 þ 2ν σ 23

Given the formation of C or C1, we can obtain the other using any symbolic
calculation programs such as Mathcad and MATLAB.
Please note that these stiffness and compliance matrices are the matrices for a
material point rather than the whole object. Thus, they are local matrices. In
numerical simulation, we also commonly encounter another type of stiffness matrix
or compliance matrix, which is the global stiffness or compliance matrix for the
object.

Plane Strain and Plane Stress

The degeneration from 3D to 2D in the mechanical field is different from and more
complicated than other physical fields: We usually need to make it clear whether it is
a plane strain or a plane stress condition.

Plane Strain

In the plane strain condition, the strain component along the direction perpendicular
to the plane is zero. Accordingly, we only have non-zero strains along the two
directions in that plane. This is usually a good approximation to the mechanical field
in thick objects undergoing a distributed load on its side surfaces along the longitu-
dinal direction, for example, an earth dam under a constant surcharge pressure on its
top. The constitutive relationship for the plane strain case can be obtained by turning
the off-plane strain components into zero in the following equation:
2 3
1ν ν ν 0 0 0
2 3 2 3
σ 11 6 ν 1ν ν 0 7
6 0 0 7 ε11
6 σ 22 7 6 ν ν 1ν 0 7 6 7
6 7 6 0
1  2ν
0 76 ε22 7
6 σ 33 7 E 6 0 7
0 76 33 7
6 ε
6 7¼ 6 0 0 0 7
6 σ 12 7 ð1 þ νÞð1  2νÞ 6 2 76 2ε12 7:
6 7 6 1  2ν 76 7
4 σ 13 5 6 0 0 0 0 0 7 4 2ε13 5
6 2 7
σ 23 4 1  2ν 5 2ε23
0 0 0 0 0
2
ð14:25Þ
150 14 Mechano: Stress and Strain

For example, if the non-zero strain plane is parallel to the surface consisting of axes
1 and 2 (normal to axis 3), then all the strain terms with the subscript “3” will
diminish, leading to
2 3
1ν ν ν 0 0 0
2 3 2 3
σ 11 6 ν 1ν ν 0 7
6 0 0 7 ε11
6 σ 22 7 6 ν ν 1ν 0 7 6 7
6 7 6 0
1  2ν
0 76 ε22 7
6 σ 33 7 E 6 0 76 0 7
6 7¼ 6 0 0 0 0 76 7
6 σ 12 7 ð1 þ νÞð1  2νÞ 6 2 76 2ε12 7:
6 7 6 1  2ν 76 7
4 σ 13 5 6 0 0 0 0 0 7 4 0 5
6 2 7
σ 23 4 1  2ν 5 0
0 0 0 0 0
2
ð14:26Þ

The above equation can be further reduced into:

2 3 2 32 3
σ 11 1ν ν 0 ε11
4 σ 22 5 ¼ E 6 ν ν 0 74 5
ð1 þ νÞð1  2νÞ
4 1  2ν 5 ε11 : ð14:27Þ
σ 12 0 0 2ε12
2
Inversing the above equation, we can obtain the compliance matrix for the plane
strain condition:
2 3 2 32 3
ε11 1ν ν 0 σ 11
4 ε11 5 ¼ 1 þ ν 4 ν 1ν 0 54 σ 22 5: ð14:28Þ
E
2ε12 0 0 1 σ 12

Plane Stress

The plane stress conditions apply to thin objects such as membranes which only have
non-zero stress components along the directions within the membrane but no stresses
on the plane whose unit normal vector is perpendicular to the membrane. Following
a similar procedure as above, we can start with the following equation:
2 3 2 32 3
ε11 1 ν ν 0 0 0 σ 11
6 ε22 7 6 ν ν 0 7 6 7
6 7 6 1 0 0 76 σ 22 7
6 ε33 7 1 6 ν ν 1 0 0 7
0 76 σ 33 7
6
6 7¼ 6 7: ð14:29Þ
6 2ε12 7 E 6 0 2 þ 2ν 0 7 6 7
6 7 6 0 0 0 76 σ 12 7
4 2ε13 5 4 0 0 0 0 2 þ 2ν 0 5 4 σ 13 5
2ε23 0 0 0 0 0 2 þ 2ν σ 23
Constitutive Relationship: Linear Elastic Hooke’s Law 151

Likewise, assuming the non-zero strain plane is parallel to the surface consisting of
axes 1 and 2 (normal to axis 3), all the strain terms with the subscript “3” will be
equal to 0, leading to
2 3 2 32 3
ε11 1 ν ν 0 0 0 σ 11
6 ε22 7 6 ν ν 0 7 6 7
6 7 6 1 0 0 76 σ 22 7
6 ε33 7 1 6 ν ν 1 0 0 0 7 6 0 7
6 7¼ 6 76 7: ð14:30Þ
6 2ε12 7 E 6 0 2 þ 2ν 0 7 6 7
6 7 6 0 0 0 76 σ 12 7
4 2ε13 5 4 0 0 0 0 2 þ 2ν 0 54 0 5
2ε23 0 0 0 0 0 2 þ 2ν 0

The above equation can be further reduced into

2 3 2 32 3
ε11 1 ν 0 σ 11
4 ε11 5 ¼ 1 4 ν 1 0 54 σ 22 5: ð14:31Þ
E
2ε12 0 0 2 þ 2ν σ 12

The matrix between stress and strain is the compliance matrix for the plane strain
condition. Inversing the above equation, we can obtain the stiffness matrix:
2 3 2 32 3
σ 11 1 ν 0 ε11
4 σ 22 5 ¼ E 6 4ν 1 0 74 5
1  ν2 0 1  ν 5 ε11 : ð14:32Þ
σ 12 0 2ε12
2
Stiffness and compliance matrices in terms of the other elastic moduli can be
derived following the above procedure. Taking the plane strain condition, for
example, the stiffness matrix can be obtained from
2 3 2 32 3
σ 11 λ þ 2μ λ λ 0 0 0 ε11
6 σ 22 7 6 λ λ þ 2μ λ 0 07 6 7
6 7 6 0 76 ε22 7
6 σ 33 7 6 λ λ λ þ 2μ 0 7
0 0 76 0 7
6
6 7 6 7,
6 σ 23 7 ¼ 6 0 0 0 μ 0 07 6 7 ð14:33Þ
6 7 6 76 0 7
4 σ 13 5 4 0 0 0 0 μ 0 5 4 0 5
σ 12 0 0 0 0 0 μ 2ε12

which is reduced to:

2 3 2 32 3
σ 11 λ þ 2μ λ 0 ε11
4 σ 22 5 ¼ 4 λ λ þ 2μ 0 54 ε11 5: ð14:34Þ
σ 12 0 0 μ 2ε12
152 14 Mechano: Stress and Strain

Governing Equation: Navier’s Equation

The governing equation for the mechanical field can be obtained based on the
balance law: The Cauchy equation of motion. According to the principle of conser-
vation of linear momentum, if the continuum body is at mechanical equilibrium, it
can be demonstrated that the components of the Cauchy stress tensor at every
material point in the body satisfy the equilibrium equations. Let us consider a
continuum body occupying a volume V, having a surface area S, with defined
ðnÞ
traction or surface forces T i per unit area acting on every point of the body surface
and body forces Fi per unit of volume on every point within the volume V. If the
body is at mechanical equilibrium, the resultant force acting on the volume should be
zero; mathematically, this is:
Z Z
ðnÞ
T i dS þ F i dV ¼ 0: ð14:35Þ
S V

ðnÞ
Then recall the definition of the stress vector, T j ¼ σ ji n j , yielding:
Z Z
σ ji n j dS þ F i dV ¼ 0: ð14:36Þ
S V

Using Gauss’s divergence theorem to convert a surface integral to a volume

integral gives
Z Z
σ ji, j dV þ F i dV ¼ 0, ð14:37Þ
V V

or
Z
σ ji, j þ F i dV ¼ 0: ð14:38Þ
V

For an arbitrary volume, the integral vanishes, so we have the following differ-
ential form of the equilibrium equation:

σ ji, j þ F i ¼ 0: ð14:39Þ

But as mentioned above, we usually deal with displacement directly, which

serves as the dependent variable. This leads to the formation of Navier’s equation.
The displacement is also a physical field as it exists everywhere across the
domain. To turn the displacement into the dependent variable, we need to connect
the current unknown, i.e., stress, to the displacement. This goal will be achieved
following two steps. First, the strain-displacement equations are substituted into the
Practice Problem 153

constitutive equations (Hooke’s law), yielding the following expression of stress in

terms of displacement:

σ ij ¼ λεkk δij þ 2μεij ¼ λδij εk, k þ μ ui, j þ u j, i : ð14:40Þ

Differentiating the stress gives:

σ ij, j ¼ λεk, ki þ μ ui, jj þ u j, ij : ð14:41Þ

Second, substituting the above gradient of the stress into the equilibrium equation
yields

λuk, ki þ μ ui, jj þ u j, ij þ F i ¼ 0, ð14:42Þ

or equivalently,

μεi, jj þ ðλ þ μÞu j, ji þ F i ¼ 0: ð14:43Þ

This governing is also called the Navier-Cauchy equation, Navier’s equation, or

elastostatic equation. The equation is more frequently formulated using the tensor
notation as:

μ∇2 u þ ðλ þ μÞ∇ð∇  uÞ þ F ¼ 0: ð14:44Þ

Practice Problem

This section shows a 2D example. 3D examples with an analytical solution can be

found in the Physics of Continuous Matter (Lautrup 2005).

Problem Description

A distributed load of 1 kPa is applied on the right side of a square element. As shown
in Fig. 14.4, the element has a side length of 1 m. The material consisting the element
has Young’s modulus of 10 GPa and Poisson’s ratio of 0.2. The left boundary is
fixed, while the upper and lower boundary was constrained using “rollers”: displace-
ment is allowed along the boundary surface while not allowed along the normal
direction. A plane strain condition is assumed for this 2D problem. Please calculate
the stress distribution inside of the element.
154 14 Mechano: Stress and Strain

Fig. 14.4 Computation

domain and boundary
conditions

Governing Equation Natural Boundary Condition

The governing equation can be formulated in several different equivalent ways. In

FlexPDE, we can implement using 1. Navier’s equation with vector operators (nabla
symbol) (Method 1), 2. the equilibrium equation with Cauchy stress (Method 2), and
3. Navier’s equation with the derivatives of each stress component (no nabla symbol)
(Method3).
Method 1

u : mu∗ DIVðGRADðuÞÞ þ ðlambda þ muÞ∗GRADðDIVðuÞÞ ¼ 0

Method 2

u : DIVðsigma tensorÞ ¼ VECTORð0; 0Þ

Method 3

ux : mu∗ ðDxxðuxÞ þ DyyðuxÞÞ þ ðlambda þ muÞ∗ðDxxðuxÞ þ DyxðuyÞÞ ¼ 0

uy : mu∗ ðDyyðuyÞ þ DxxðuyÞÞ þ ðlambda þ muÞ∗ðDxyðuxÞ þ DyyðuyÞÞ ¼ 0

The Dirichlet boundary condition can be implemented by assigning a displace-

ment value to the corresponding boundary segment, while the Neumann boundary
condition needs special attention. Details regarding the Neumann boundary condi-
tion for FEM packages such as FlexPDE have been given in Chap. 9. According to
the previous introduction, the use of Navier’s equation defines that the Neumann
boundary condition for any boundary segment gives out the value of σ  n on that
boundary segment. In the mechanical field, this value corresponds to the magnitude
of the traction or pressure on the boundary. Therefore, a distributed load of 1 kPa on
the right boundary can be applied using the command “NATURAL (ux) ¼ 1000” or
“LOAD (ux) ¼ 1000.” The two commands are equivalent in FlexPDE.
Practice Problem 155

Results

The results of the above simulation are obvious. A tensile stress (positive) of 1 kPa is
generated along the horizontal direction.

Appendix: FlexPDE Script File

{ Basic linear elastic mechanical problems for 2D (plain strain)
homogenous cases}
TITLE '2D Plain Strain'
VARIABLES
u=VECTOR(ux,uy) { Methods 1&2 use the displacement vector as
dependent variable }
!ux,uy { Methods 3 uses components of the vector as
dependent variable }
DEFINITIONS
E=10e9
nu=0.2
mu=E/2/(1+nu)
lambda =E*nu/(1+nu)/(1-2*nu)
C=TENSOR((2*mu+lambda,lambda,0),(lambda,2*mu+lambda,0),(0,0,
mu)) ! Stiffness matrix
epsilon=VECTOR( dx(ux),dy(uy),0.5*(dx(uy)+dy(ux)) )
sigma=dot( C,epsilon)
{ The above C and epsilon are defined using the Voigt notation;
sigma_tensor uses the tensor notation }
sigma_tensor=TENSOR((XCOMP(sigma),ZCOMP(sigma)),(ZCOMP
(sigma),YCOMP(sigma)))
! INITIAL VALUES { Not time dependent }
EQUATIONS
{ Two equivalent ways to implement the governing equation }
u: mu*DIV(GRAD(u))+(lambda+mu)*GRAD(DIV(u))=0 { Method 1 }
!u: DIV(sigma_tensor)=VECTOR(0,0) { Method 2 } !
ux: mu*(Dxx(ux)+Dyy(ux))+(lambda+mu)*(Dxx(ux)+Dyx(uy))=0 {
Method 3 }
!uy: mu*(Dyy(uy)+Dxx(uy))+(lambda+mu)*(Dxy(ux)+Dyy(uy))=0 {
Method 3 } BOUNDARIES
REGION 1 { For each material region }
START(0,0)
VALUE(ux)=0
VALUE(uy)=0
LINE TO (0,1)
VALUE(uy)=0
NATURAL(ux)=0
LINE TO (1,1)
NATURAL(ux)=1000
VALUE(uy)=0
LINE TO (1,0)
NATURAL(ux)=0
VALUE(uy)=0
156 14 Mechano: Stress and Strain

LINE TO CLOSE
! TIME 0 TO 1 { Not time dependent }
MONITORS
PLOTS
VECTOR(u) !for Method 3: VECTOR(VECTOR(ux,uy))
CONTOUR(XCOMP(sigma))
ELEVATION(uy) FROM (0,0) TO (0,1)
ELEVATION(ux) FROM (1,0) TO (1,1)
END
Chapter 15
Dyno: Dynamics

Introduction: Meaning of Dynamics

Dynamics, which is abbreviated as dyno in this book, is a relatively ambiguous

concept in multiphysics. The word or the corresponding root is used in some
multiphysics applications such as aerodynamics and elastodynamics. However, it
is also used in many topics out of the traditional multiphysics area such as thermo-
dynamics and molecular dynamics. In addition, dynamics has been historically
studied with respect to material types, leading to two distinct areas: elastodynamics
for solids and aerodynamics (gas) and hydrodynamics (liquid) for fluids. Acoustics
consists of a small part from each of the above two areas but is more focused on the
vibration of the material points and the energy transfer associated with the vibration.
These facts prohibit the wide use of dynamics or dyno as a word for general
dynamics phenomena in multiphysics.
Broadly speaking, dynamics is the study of the motion of entities resulting from
forces. Traditionally, these entities are generally an assembly of solid inertial
elements and rigid particles. Dynamics is usually differentiated from kinematics
and statics. Kinematics is concerned with the motion of a material point, which
generalizes particles, bodies, conglomerates of such particles or systems, without
reference to the underlying cause, i.e., forces. The scope of statics is confined to
bodies at rest and forces at equilibrium – a system undergoes no displacements with
respect to a specified frame of reference. We also have a quasi-static state if such
equilibria can be assumed within a sufficiently long time.
Dynamics and dyno are used in this book as a generic term for the dynamic
phenomena in internal systems involving forces or momenta. As shown in Fig. 15.1,
internal systems include deformable solids and fluids, whereas external systems refer
to particles or rigid bodies, which correspond to multibody dynamics and rigid body
dynamics, respectively. Therefore, theories such as molecular dynamics, discrete
element method, and Lattice Boltzmann method are out of the scope of dynamics in
multiphysics in this book. An interesting example is the smoothed particle

© Springer International Publishing AG, part of Springer Nature 2018 157

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_15
158 15 Dyno: Dynamics

Fig. 15.1 From general mechanics areas to the dynamics in multiphysics

hydrodynamics (SPH). The particles used in the SPH are more like a mesh for
discretization than particles to which Newton’s second law is applied. Instead,
governing equations are still discretized and solved. Therefore, the SPH can still
be viewed as a discretization method for implementing dynamics in multiphysics.
The relationship between dynamics and the wave propagation theory needs to be
clarified. In deformable solids or fluids, the systems to be considered are not rigid.
This turns dynamics into the study of the correlation between forces, either surface
forces or body forces, and the motion of material points within the flexible body.
Further clarification needs to be made with respect to both the phase type, i.e., solids
and fluids, and the scale of deformation.
In solids, the above fact results in a knowledge pool consisting of both dynamics,
i.e., Newton’s second law, and the understanding of the constitutive relationships of
solids, e.g., elastic, plastic, elastoplastic, and viscoelastic. As a result, the dynamics
of solids is subdivided into areas such as elastodynamics and plastodynamics.
Among these areas, elastodynamics is the predominant one. The scale of the strain
amplitudes also determines the nature of dynamic phenomena in solids. If a study is
confined to small-amplitude oscillations in strains, then this problem can be regarded
as vibration. A close look at the governing equation of the elastodynamics, i.e.,
Navier’s equation with a momentum term, reveals that the equation has the form of a
wave equation. Therefore, the dynamics in solids, at least elastodynamics, can be
viewed as an application of wave theories in solids. However, the dynamic phenom-
ena in solids, especially those involving large strains, are a broader scope of studies
than merely wave propagation which focuses mostly on the oscillation and the
associated transfer of energy. In such dynamic phenomena with large strains, the
stress, strain, complicated constitutive relationships, and fatigue are also of great
interest.
Fluids are much different from solids especially in that the shear stresses in fluids
are related to the first-order time derivative of strain, i.e., strain rate or velocity
gradient, instead of the strain. Due to the reason, the fluids can have bulk mass
movement without altering the stresses, for example, at a constant strain rate.
Introduction: Meaning of Dynamics 159

The above difference leads to two noteworthy outcomes. First, fluids can hardly
transmit shearing waves (transverse waves) because a shearing strain does not lead to
a shearing stress. Consequently, the shearing motion and energy can be hardly
transmitted. The volumetric train is still proportional to the mean stress component
(pressure) of the stress tensor. Therefore, pressure (longitudinal) waves can still
propagate in fluids. Second, another type of dynamics exists in fluids in addition to
pressure waves. This is because, though shear strains do not cause stresses in fluids,
their time derivatives can still cause stresses and further contribute to the motion of
fluid parcels. As a result, this fluid motion exists as a form of dynamics in addition to
mechanical waves (or acoustics in fluids). This type of dynamics, which is the fluid
dynamics that we are familiar with, exists as the “background” on which the wave
propagation occurs. The Navier-Stokes equations are used to describe the first type
of dynamics, i.e., fluid dynamics, while the acoustics can be derived by investigating
the disturbance to the volumetric stresses terms and velocity. Acoustics excludes the
hydrostatic pressure and the fluid velocity associated with fluid dynamics. Therefore,
acoustic phenomena can occur in fluids at rest in the same way as in moving fluids.
Surface waves are excluded in the above discussion as we focus on physical
fields here.
Therefore, the acoustics in fluids can be viewed as a part or a derivative of the area
of fluid dynamics, but it is different from the “fluid dynamics” which is directly
described by the Navier-Stokes equations. It is noted that fluid dynamics involves
both mass and momentum transport, so it is treated as a multiphysical process and
will be discussed in the next part of this book. By contrast, acoustics will be treated
as a monolithic physical field in this chapter because only one dependent governing
equation is needed for the acoustics in one material.
However, the wave theory is more than acoustics. The wave equations in
elastodynamics and fluid dynamics share the same characteristics: a wave equation,
oscillation of physical properties (stress and strain), and transfer of energy. The wave
equation in this case depicts the process that a wave propagates through media, i.e.,
solids, gasses, or liquids. This type of wave is pertained to materials and mechanical
responses and thus termed the mechanical wave. The mechanical wave is frequently
encountered in earth materials in seismic engineering and in exploratory geophysics.
The wave is also called acoustic waves in acoustics applications. However, wave
phenomena, as oscillation accompanied by a transfer of energy, also widely exist in
other forms such as electromagnetic waves. The behavior of particles in quantum
mechanics can also be described by waves. In addition, a vibration or movement in
gravitational fields can lead to gravitational waves traveling through space.
In summary, acoustics as the study of mechanical waves can be classified as a
special case of dynamics, consisting of part of elastodynamics and part of fluid
dynamics. In the meanwhile, the wave theory discussed in acoustics is part of the
wave theory which also includes other types of waves.
Due to the above consideration, in this chapter, elastodynamics will be introduced
first, in which we will show that the governing equation of elastodynamics directly
describes the mechanical waves in solids. Then we will use two different ways to
show that the wave equation for the acoustics in fluids can be derived as a special
160 15 Dyno: Dynamics

case (or a derivative) of fluid dynamics via the Navier-Stokes equation. The wave
equation generalized from acoustics will be discussed next. In the end, a practice
problem will be provided to show the idea and implementation of elastodynamics.
Fluid dynamics, which is treated as a multiphysical field, will be introduced in the
next part of the book.

Elastodynamics
Governing Equation of Isotropic, Homogeneous, and Elastic
Material

Elastodynamics is the study of waves and time-dependent mechanical responses,

e.g., strain and stress, in linearly elastic materials. An elastic wave is a mechanical
wave that propagates in elastic or viscoelastic materials. Elastic waves occurring in
the Earth as the result of earthquakes or other disturbance are usually called seismic
waves. The disturbance, either in terms of stress or strain, produces the wave, while
the elasticity of the material provides the restoring force for the wave. Therefore,
elastodynamics covers the dynamics in materials with an elastic constitutive
relationship.
The wave equation of elastodynamics is simply the equilibrium equation of
elastostatics with an additional inertial term:

σ ji, j þ F i ¼ ρ€ui ¼ ρ∂tt ui : ð15:1Þ

Substituting the linearly elastic constitutive relationship into the above equation,
we can obtain the following governing equation by following a procedure similar to
that was introduced in the previous chapter:

μui, jj þ ðλ þ μÞμu j, ji þ F i ¼ ρ∂tt ui : ð15:2Þ

Using the tensor notation, the above governing equation of elastodynamics turns into
2
∂ u
μ∇2 u þ ðλ þ μÞ∇ð∇  uÞ þ F ¼ ρ : ð15:3Þ
∂t 2

Acoustics in Solids

To obtain a better understanding of the acoustics in solids, we can separate the two
major types of waves: pressure wave (P-wave, primary wave, or longitudinal wave)
from the shear wave (S-wave or transverse/transversal wave).
For the pressure wave, there are only volumetric changes but no shearing.
Mathematically, this type of physical field can be formulated as:
Fluid Dynamics and Acoustics in Fluids 161

∇  u 6¼ 0 and ∇  u ¼ 0:

The physical field with such a dependent variable is called an irrotational field, a
curl-free/curl-less field, or a longitudinal field.
For the shear wave, there are no volumetric changes but only shearing. Mathe-
matically, this type of field can be formulated as:

∇  u ¼ 0 and ∇  u 6¼ 0:

The physical field with such a dependent variable is called a solenoidal field, an
incompressible field, a divergence-free field, a transverse field, or an equivoluminal
field.
To separate the two types of waves, we can recall the following identity from the
tensor analysis:

∇  ð∇  uÞ ¼ ∇ð∇  uÞ  ∇2 u: ð15:4Þ

Then Navier’s equation can be reformulated into the following equation by

neglecting the body force:
2
∂ u
ðλ þ 2μÞ∇ð∇  uÞ  μ∇  ð∇  uÞ ¼ ρ : ð15:5Þ
∂t 2
Then in the case of pressure waves, we obtain:
2
∂ u
ðλ þ 2μÞ∇ð∇  uÞ ¼ ρ : ð15:6Þ
∂t 2

qffiffiffiffiffiffiffiffi is the square root of the ratio between the “modulus” and the
The wave velocity
density, i.e., λþ2μ
ρ .
In the case of shear waves, we obtain
2
∂ u
μ∇  ð∇  uÞ ¼ μ∇2 u ¼ ρ , ð15:7Þ
∂t 2
qffiffi
μ
where the shear wave velocity is ρ.

Fluid Dynamics and Acoustics in Fluids

In fluids such as water, it is more common to derive the propagation of sound waves
from the equation of continuity, i.e., conservation of mass and the equation of
motion, i.e., conservation of momentum, instead of Navier’s equation as for solids.
We start the derivation with a few simplifications such as a constant density and
162 15 Dyno: Dynamics

negligible viscous and gravitational force, which are reasonable considering the
magnitude and time scale of vibration. More importantly, we need to extract the
mass and momentum balance from the “background” fluid dynamics (mass and
momentum transfer) associated with the viscosity (shearing), gravity, and advection.
Then we will obtain the following equations

∂p
þ k∇  v ¼ 0 ðmass balanceÞ ð15:8Þ
∂t
∂v
ρ0 þ ∇p ¼ 0 ðmomentum balanceÞ ð15:9Þ
∂t
where p is the acoustic pressure, which can be viewed as the deviation from the
hydraulic pressure in fluid dynamics due to the existence of the wave, and v is the
flow velocity excluding the background velocity associated with the background
fluid dynamics. The use of the p and v from the fluid dynamics will not impact the
following derivation because all the terms in Eqs. 15.8 and 15.9 are about differences
instead of absolute values. However, when assigning boundary conditions, p and
v need to use the acoustic values instead of the hydraulic pressure and the real fluid
velocity that are used in background fluid dynamics.
A liquid can usually be assumed as elastic under small strains, which is most
likely to be true for the case of waves in a liquid. Accordingly, the bulk modulus and
the density can be related in the following way

k ¼ ρ0 c20 : ð15:10Þ

where c0 can be later proven to be the velocity of the acoustic wave.

As mentioned above, the resistance of fluids such as water to shearing deforma-
tion is negligible. Thus, we can infer that the flow velocity field in the fluids is
irrotational, —  v ¼ 0. We then can substitute this irrotational fluid equation and the
above constitutive relationship between the bulk modulus and wave velocity, i.e.,
Eq. 15.10, into the two governing equations (Eqs. 15.8 and 15.9). The following
governing equation for the elastic wave in fluids can be obtained
2
∂ v
 c20 ∇2 v ¼ 0, ð15:11Þ
∂t 2
or equivalently in terms of pressure
2
∂ p
 c20 ∇2 p ¼ 0: ð15:12Þ
∂t 2
In the above general form of the wave equation, c0 has the physical meaning of wave
velocity. The following identity in the tensor analysis was employed in the above
deduction:

∇2 v ¼ ∇ð∇  vÞ  ∇  ð∇  vÞ: ð15:13Þ

Fluid Dynamics and Acoustics in Fluids 163

It is also very common to express the acoustic wave equation using a scalar
potential ψ where v ¼ — ψ. In that case, the acoustic wave equation is written as:
2
∂ ψ
 c20 ∇2 ψ ¼ 0: ð15:14Þ
∂t 2

Comprehensive Derivation from Fluid Dynamics

The above derivation of the wave equation for the acoustics in fluids adopts simple
governing equations. However, the meanings of p and v require some effort to
comprehend. A more comprehensive version of the above derivation is conducted
with the Navier-Stokes equations with step-by-step assumptions. Also, the concept
of disturbance is utilized so that the physical meaning of p and v can be more easily
understood. Since the Naiver-Stokes equations have not been introduced yet, this
section is provided only as a reference. If necessary, please finish the chapter for fluid
dynamics first.

Conservation of Momentum

The equation for the momentum conservation of a fluid with a constant density is
 
∂v
ρ þ v  ∇v ¼ ∇p þ ∇  τ þ ρg, ð15:15Þ
∂t

where g is the body force per unit mass, p is the pressure, and τ is the deviatoric
stress. Using the Cauchy stress for τ, then we have

1
p ¼  traceðσÞ ð15:16Þ
3
τ ¼ σ þ pI ð15:17Þ

where I is the rank-2 identity tensor.

Several assumptions will be made step by step in the following subsections to
derive the momentum balance equation for acoustics in fluids.

Assumption 1: Newtonian Fluid

In acoustics, the fluid material can be assumed to be Newtonian. Accordingly, the

deviatoric stress tensor is related to the flow velocity by
164 15 Dyno: Dynamics

 
τ ¼ μ ∇v þ ð∇vÞT þ λð∇  vÞI, ð15:18Þ

where μ is the shear viscosity and λ is the bulk viscosity. The divergence of τ is
      2 2 2
∂ ∂ui ∂u j ∂ ∂uk ∂ ui ∂ ui ∂ uk
∇τ ¼μ þ þλ ¼μ þμ þλ
∂xi ∂x j ∂xi ∂xi ∂xk ∂xi ∂x j ∂xi ∂xi ∂xk ∂xi
2 2
∂ ui ∂ ui
¼ ðμ þ λÞ þμ
∂xi ∂x j ∂x j ∂x j
¼ ðμ þ λÞ∇ð∇  vÞ þ μ∇2 v:
ð15:19Þ

Employing the identity — 2v ¼ — (—  v)  —  —  v, the divergence of the

deviatoric stress then becomes

∇  τ ¼ ð2μ þ λÞ∇ð∇  vÞ  μ∇  ∇  v: ð15:20Þ

Substituting the above formulation of the deviatoric stress into the momentum
balance equation yields
 
∂v
ρ þ v  ∇v ¼ ∇p þ ð2μ þ λÞ∇ð∇  vÞ  ∇  ∇  v þ ρg: ð15:21Þ
∂t

Assumption 2: Irrotational Flow

The flow is irrotational in acoustics, hence

∇v¼0 ð15:22Þ

Substituting the above equation into the momentum equation, we obtain

 
∂v
ρ þ v  ∇v ¼ ∇p þ ð2μ þ λÞ∇ð∇  vÞ þ ρg: ð15:23Þ
∂t

Assumption 3: No Body Forces

The gravitational force as a body force is negligible in acoustic waves in fluids. This
further simplifies the momentum equation into
 
∂v
ρ þ v  ∇v ¼ ∇p þ ð2μ þ λÞ∇ð∇  vÞ: ð15:24Þ
∂t
Fluid Dynamics and Acoustics in Fluids 165

Assumption 4: No Viscous Forces

The viscous forces can be excluded as explained in the introduction section, leading
to the following momentum equation:
 
∂v
ρ þ v  ∇v ¼ ∇p: ð15:25Þ
∂t

Assumption 5: Small Disturbances

Instead of understanding the acoustic field as another field on top of the background
fluid dynamics field, we also introduce the concept of disturbance to explain the
pressure and fluid velocity associated with the acoustic field. The underlying
assumption of the disturbance concept is that the amplitude of the disturbance of
the field quantities is small compared with the magnitudes of these quantities. Then
we can express the variables as the sum of a time-averaged mean field () varying in
space and a small fluctuating field (~) varying in space and time:

ρ ; v ¼ v þ v~
p ¼ p þ p~; ρ ¼ ρ þ e

and

∂
p ∂ρ ∂
v
¼ 0; ¼ 0; ¼ 0:
∂t ∂t ∂t
Then the momentum equation is reformulated into
 
∂~
v
ρ þ ρe þ v þ v~  ∇ v þ v~ ¼ ∇ p þ p~ : ð15:26Þ
∂t

Considering the fluctuations are small, products of the fluctuation terms can be
neglected, yielding:

∂~
v
ρ ρ v  ∇
þ ρþe v þ ρ v~  ∇
v þ v  ∇~
v ¼ ∇ p þ p~ : ð15:27Þ
∂t

Assumption 6: Homogeneous Medium

Assuming the material is homogeneous, the time-averaged variables p and ρ have

zero gradients:

∇
p ¼ 0; ð15:28Þ
166 15 Dyno: Dynamics

and

∇ρ ¼ 0: ð15:29Þ

The momentum equation then reduces into:

∂~
v
ρ ρ v  ∇
þ ρþe v þ ρ v~  ∇
v þ v  ∇~
v ¼ ∇ p~ : ð15:30Þ
∂t

Assumption 7: Medium at Rest

The last step is to remove the background fluid dynamics field. For the purpose, we
just need to assume that the medium is at rest, i.e., v ¼ 0. Then the momentum
balance equation becomes:

∂~
v
ρ ¼ ∇ p~ : ð15:31Þ
∂t
Dropping the tildes and using ρ ¼ ρ0 , we obtain the commonly used form of the
acoustic momentum equation:

∂v
ρ0 þ ∇p ¼ 0: ð15:32Þ
∂t

Conservation of Mass

The mass conservation equation for acoustics in fluids can also be derived in a
manner similar to what is used for the conservation of momentum. The mass
conservation equation in a fluid volume without any sources/sinks is:

∂ρ
þ ∇  ðρvÞ ¼ 0: ð15:33Þ
∂t

Assumption 1: Small Disturbances

The assumption of small disturbances gives

∂e
ρ
ρ ∇  v þ v~ þ ∇ ρ þ e
þ ρþe ρ  v þ v~ ¼ 0: ð15:34Þ
∂t
Expanding the equation and neglecting the products of disturbance terms, the mass
balance equation becomes
Fluid Dynamics and Acoustics in Fluids 167

∂e
ρ
þ ρþe ρ  v þ ∇ρ  v~ ¼ 0:
ρ ∇  v þ ρ∇  v~ þ ∇ ρ þ e ð15:35Þ
∂t

Assumption 2: Homogeneous Medium

Adopting the homogeneous assumption, the mass balance equation becomes

∂e
ρ
ρ ∇  v þ ρ∇  v~ þ ∇e
þ ρþe ρ  v ¼ 0: ð15:36Þ
∂t

Assumption 3: Medium at Rest

Similar to the momentum balance, remove the background fluid dynamics by

assuming the fluid is at rest, i.e., v ¼ 0. Then the mass balance equation can be
simplified as

∂e
ρ
þ ρ∇  v~ ¼ 0: ð15:37Þ
∂t

Assumption 4: Ideal Gas, Adiabatic, and Reversible

An equation of state for relating the pressure and density is needed to close the
equation system. For the purpose, we assume that the fluid is an ideal gas and all
acoustic waves compress the medium in an adiabatic and reversible manner. Accord-
ingly, the following equations of the thermodynamic state are valid:

dp γp Cp γp
¼ ;γ ¼ ; and c2 ¼ ,
dρ ρ Cv ρ
where Cp is the specific heat at a constant pressure, Cv is the specific heat at a
constant volume, and c is the wave speed. The value of γ is 1.4 for air.
For small disturbances
dp p
dρ~
p 
p
; ρ ρ ; and c2  c20 ¼ γ p
eρ
ρ,
where c0 is the speed of sound in the fluid.Therefore, we have
168 15 Dyno: Dynamics

p~
ð15:38Þ
eρ ¼γp
ρ¼c20

∂~
p ∂e
ρ
þ c20 ¼ 0: ð15:39Þ
∂t ∂t
The balance of mass can then be formulated as

1 ∂~p
þ ρ∇  v~ ¼ 0: ð15:40Þ
c0 ∂t
2

Dropping the tildes and defining ρ ¼ ρ0 , we obtain the following mass balance
equation for acoustic waves in fluids:

∂p
þ ρ0 c20 ∇  v ¼ 0: ð15:41Þ
∂t
Combining the momentum and mass balance equation. We can obtain a wave
equation identical to what we obtained in the simple derivation.

Acoustic Field
Wave Equation

The acoustic field here is used to represent the propagation of mechanical waves.
Therefore, acoustics is an interdisciplinary area that deals with all mechanical waves
in gasses, liquids, and solids. It covers topics such as vibration, sound, ultrasound,
and infrasound. Acoustic applications widely exist in modern society, while the most
obvious examples are the audio and noise control industries.
It was demonstrated in previous sections that the governing equation of the
acoustic field can be directly obtained from the Navier’s equation in elastodynamics
for solids or derived from the Navier-Stokes equations for fluids. However, the wave
equation has a much more general sense than just a description of mechanical waves.
It is also the governing equation of electromagnetic waves, which can be derived
from Maxwell’s equations. In addition, the wave equation has applications out of the
scope of multiphysics: quantum mechanics, plasma physics, and general relativity.
The wave equation is a hyperbolic partial differential equation. It typically
concerns a time variable t, one or more spatial variables x1, x2, . . ., xn, and a scalar
function u ¼ u (x1, x2, . . ., xn; t), which can be any of these oscillating quantities
associated with the wave, e.g., the mechanical displacement of a wave. The general
form of the wave equation is
2
∂ u
¼ c2 ∇2 u, ð15:42Þ
∂t 2
Acoustic Field 169

where c is the velocity of wave propagation. As shown in the previous sections, this
velocity depends on materials for mechanical waves. However, it could also be
independent of material properties in other types of waves.
The wave equation alone does not specify a physical solution. Further constraints
or conditions on time and boundaries are needed to ensure a unique solution. These
constraints could be initial conditions, which prescribe the amplitude and phase of
the wave. Another important class of problems occurs in enclosed spaces specified
by boundary conditions. In these problems, the solutions represent standing waves or
harmonics, which are analogous to the harmonics of musical instruments.
To discuss the boundary conditions for the wave equation, let us use the pressure
as the dependent variable:
2
∂ p
¼ c2 ∇2 p: ð15:43Þ
∂t 2
The Dirichlet and Neumann boundary conditions now take the following form

Dirichlet : p ¼ q, ,
Neumann : n  ∇p ¼ ρ€u, ,

where n is the magnitude of traction on the surface (or boundary pressure) and €u is
the acceleration prescribed on the boundary. Special cases of the above equations
could generate geometric or physical boundary conditions that we frequently
encounter, for example, p ¼ 0 corresponds to the sound soft boundary, and n  — p ¼ 0
corresponds to the symmetry boundary condition.

Speed of Wave

In general, the speed of sound c is given by the Newton-Laplace equation:

sffiffiffiffiffiffi
Ks
c¼ , ð15:44Þ
ρ

where Ks is a coefficient of stiffness, e.g., the isentropic bulk modulus for longitu-
dinal waves, and ρ is the density. Therefore, the speed of sound increases with the
stiffness of the material while decreases with the density.
For general equations of state, the speed of sound c is given by
ffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s ffi
∂p
c¼ , ð15:45Þ
∂ρ s
170 15 Dyno: Dynamics

where p is the pressure and ρ is the density. The derivative is taken adiabatically, i.e.,
at a constant entropy s.
A common practice in acoustics is to decompose a wave into a pressure wave
(longitudinal or irrotational) and a shear wave (transverse or equivoluminal). This
has been shown in the elastodynamics section. In summary, a longitudinal plane
wave propagating in an infinite medium has no equivoluminal components but
solely a dilatational
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi component. This wave propagates at a phase speed of
ðλ þ 2μÞ=ρ – longitudinal wave speed. Similarly, a shear plane wave propagating
in an infinite medium has no dilatational components but only an equivoluminal
component
pffiffiffiffiffiffiffiffi and is thus named the shear wave. This wave propagates at a phase speed
of μ=ρ. For mechanical waves, the wave velocity has a general expression as the
square root of the ratio between the “modulus” and the density. The definition of the
“modulus” changes with the type of the mechanical wave.

Practice Problem

Problem Description

This problem is modified from the practice problem in the mechanical field chapter.
A square element with a side length of 1 m is composed of a material with Young’s
modulus of 10 GPa, Poisson’s ratio of 0.2, and a density of 1000. An electric wave is
generated by repeatedly compressing and stretching the upper boundary of the
element at a given frequency. Please simulate the wave inside the element to tell
what you observe at a frequency of 500 Hz and 5000 Hz and explain the reason for
the difference.
Please note that many numerical simulation programs including FlexPDE do not
handle the second-order temporal derivatives directly. Therefore, we need to split the
original governing into two by defining a new dependent variable for the first
derivative of the displacement. The boundary condition for this new dependent
variable will also need to be set up accordingly.

Results

Fig. 15.2 shows the displacement field at t ¼ 0.125 ms. The figure clearly shows that
the wave is starting to propagate into the element from the top.
Practice Problem 171

Fig. 15.2 Displacement in the domain (upper) and the horizontal displacement along x ¼ 0.5 m
(lower) at t ¼ 0.125 s
172 15 Dyno: Dynamics

Appendix: FlexPDE Script File

{ Elastodynamics in 2D (Plain Strain): Stretch a membrane }

TITLE 'Elastic Wave'
VARIABLES
u(1e-6)=VECTOR(ux,uy)
v(1e-6)=VECTOR(vx,vy)
DEFINITIONS
E=1e10
nu=0.2
mu=E/2/(1+nu)
lambda=E*nu/(1+nu)/(1-2*nu)
rho=1000
INITIAL VALUES
v=VECTOR(0,0)
u=VECTOR(0,0)
EQUATIONS
u: v=Dt(u)
v: mu*DIV(GRAD(u))+(lambda+mu)*GRAD(DIV(u))=rho*dt(v)
BOUNDARIES
REGION 1
START(0,0)
VALUE(ux)=0
VALUE(vx)=0
LOAD(uy)=0
LOAD(vy)=0
LINE TO (0,1)
LOAD(ux)=0
LOAD(vx)=0
VALUE(uy)=1e-3*sin(2*pi*5000*t)
VALUE(vy)=1e-3*2*pi*5000*cos(2*pi*5000*t)
LINE TO (1,1)
VALUE(ux)=0
VALUE(vx)=0
LOAD(uy)=0
LOAD(vy)=0
LINE TO (1,0)
VALUE(uy)=0
VALUE(vy)=0
VALUE(ux)=0
VALUE(vx)=0
LINE TO CLOSE
TIME 0 TO 10 BY 0.01 { Time dependent }
MONITORS
PLOTS
FOR cycle=1
SURFACE(uy)
VECTOR(u)
ELEVATION(uy) FROM (1,0) TO (1,1)
END
Chapter 16
Chemico: Chemical Reaction

Introduction

In this chapter, basics of chemical reactions will be presented first to provide

background knowledge for the chemical field. Then chemical kinetics will be
introduced as it defines the governing equation of the chemical field. An example
will be given to demonstrate the use of chemical kinetics for establishing the
governing equation. The reaction rate coefficient is the key auxiliary relationship
in the governing equation. The formulation of this coefficient and its dependence on
various environmental factors will be discussed at last.

Uniqueness of Chemical Field

The chemical field is unique in multiphysics. As described in Part I, it is still counted

as a physical field in multiphysics though chemical reactions are not counted as a
physical process. This treatment generates an aftereffect in its mathematical descrip-
tions, differentiating the chemical field from the others. One major difference is that
the chemical field adopts the concentration or quantities depending on it such as
reaction quotients as the state variables. However, the concentration is the dependent
variable of the concentration field. An understanding of this contradiction is that the
chemical field is more like an auxiliary relationship in mass transfer. This relation-
ship defines the source/sink term and thus describes how the mass of a phase/species
is consumed or produced. The relationship usually affects the mass balance of
several phases/species simultaneously since a chemical reaction usually involves
multiple materials. However, the chemical field still owns the characteristics of a
field, that is, a spatial distribution of the state variable which can change with time in
transient processes or stay unchanged at equilibrium.

© Springer International Publishing AG, part of Springer Nature 2018 173

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_16
174 16 Chemico: Chemical Reaction

The uniqueness of the chemical field in multiphysics also endows it with a special
treatment in numerical analysis. Since it is similar to auxiliary relationships, it is
possible to treat it as a normal auxiliary relationship during the discretization and
solution processes. Alternatively, to improve the accuracy, its treatment can be
similar to that of governing equations, for example, more synchronized with the
iteration of the governing equations. However, because it is a time-dependent
ordinary differential equation associated with every point or element instead of the
whole region, it cannot be discretized and placed into the matrix like other governing
equations. To reduce the computational load, it is also possible to divide the
computation regions into connected regions so that only one governing equation
of the chemical field is needed for a region rather than each element or grid point
within the region. But this simplification should comply with the real situation of the
chemical reactions because one pre-assumption of the chemical reaction is that the
reactants are perfectly mixed within each reactor, i.e., “regions” in the above
treatment.

Understanding of Chemical Reactions

A chemical reaction is a process within which one set of chemical substances is

transformed into another. This transformation only involves the relocations of
electrons in the forming and breaking of chemical bonds between atoms without
any change to the nuclei that determine the types of the elements. In much less
common cases, both electronic and nuclear changes exist in the chemical reactions of
unstable and radioactive elements, leading to nuclear chemistry, which is excluded
in this book.
The substances initially involved in a chemical reaction are reactants or reagents.
These reactants undergo the changes involving the forming and breaking of bonds
and produce other substance(s), i.e., product(s), which usually have properties
different from that of the reactants. Such a reaction usually occurs as a sequence
of individual sub-steps, i.e., elementary reactions, instead of one step. Many reac-
tions require no input of extra energy other than the thermal energy and thus termed
spontaneous. Non-spontaneous reactions usually happen too slowly at human time
scales. As a result, such reactions demand the input of some types of additional
energy such as heat, light, and electricity to proceed to completion, i.e., chemical
equilibrium, at a rate that we can observe.
Chemical reactions are usually characterized by a characteristic reaction rate
which is dependent on the given temperature and chemical concentration. The rate
at which a reaction takes place is studied in the scope of reaction kinetics. As a
branch of chemistry, chemical kinetics studies both the rate of a chemical reaction
and factors that affect the rate. The rate depends on various parameters such as
temperature, pressure, reactant concentrations, surface area, activation energy, and
the presence of a catalyst. Temperature facilitates chemical reactions because a
higher temperature provides more energy to the molecules, resulting in more
Understanding of Chemical Reactions 175

Fig. 16.1 Schematic of

basics chemical reaction
types

collisions per unit time. Pressure increases lead to decreases in the volume between
molecules and consequently increase in the frequency of collisions between the
molecules and the reaction rates. Reactant concentrations usually affect the reaction
rate as higher concentrations promote the collisions between particles. However,
there are also reactions whose rates are independent of reactant concentrations, i.e.,
zero order reactions. The surface area determines the area of sites which are available
for the reactions. Increasing the surface area generally increases the reaction rates.
This is particularly important in porous materials with complicated internal struc-
tures. The activation energy is defined as the amount of energy required to make the
reaction start and carry on spontaneously. Therefore, a higher activation energy
corresponds to a lower reaction rate. Catalysts are additives which remain unchanged
but are capable of changing the pathway or mechanism of a reaction. For example, a
catalyst can expedite a reaction by lowering the activation energy of the reaction.
The types of chemical reactions need to be briefly introduced to lay down a basis
for the following introduction to the mathematical description of the chemical
reactions. There are four basic types of chemical reactions shown in Fig. 16.1:
synthesis, decomposition, single replacement, and double replacement.
In a synthesis reaction, two or more simple substances combine to yield a more
complex substance. As shown in the above figure, these reactions have a general
form of A + B ! AB. One example is the burning of hydrogen gas, which combines
hydrogen and oxygen into water:

2H2 þ O2 ! 2H2 O: ð16:1Þ

In a decomposition reaction, a relatively complex substance breaks down into

multiple simple substances in the form of AB ! A + B. One example is the
electrolysis of water which produces oxygen and hydrogen gas:

2H2 O ! 2H2 þ O2 : ð16:2Þ

In a single replacement reaction, a single element replaces an element in another

compound. These reactions occur in the general form of A + BC ! AC + B. One
example is the replacement of magnesium for hydrogen in water to form magnesium
hydroxide and hydrogen gas:
176 16 Chemico: Chemical Reaction

Mg þ 2H2 O ! MgðOHÞ2 þ H2 : ð16:3Þ

A double replacement reaction involves the exchange of places for the anions and
cations in two compounds, which produces two entirely different compounds. These
reactions can be represented in the general form of AB + CD ! AD + CB. A typical
example is the reaction of lead nitrate with potassium iodide to form lead iodide and
potassium nitrate:

PbðNO3 Þ2 þ 2KI ! PbI2 þ 2KNO3 : ð16:4Þ

Chemical Kinetics and Its Mathematical Description

Chemical kinetics or reaction kinetics investigates the rates of chemical processes

and the influence of experimental conditions on the rates. For the chemical field in
numerical simulation, the information needed from chemical kinetics is the mathe-
matical models for describing the characteristics of a chemical reaction. The deter-
mination of reaction rates including both rate laws and rate constants can be fulfilled
by means of experiments, theories, or ab initio simulations.
The mechanisms are usually analyzed and summarized from a macroscopic
perspective. This is consistent with multiphysics, which relies on statistical mechan-
ics, molecular dynamics, and quantum mechanics but is established primarily with a
language of continuum mechanics. Understanding of reaction rates on the molecular
level falls into the realm of chemical dynamics or reaction dynamics. To understand
chemical kinetics, let us consider a typical chemical reaction:

aA þ bB ! cC þ dD, ð16:5Þ

where the lowercase letters (a, b, c, and d ) represent the stoichiometric coefficients
and the capital letters represent the reactants (A and B) and the products (C and D).
The reaction rate r for a chemical reaction occurring in a closed isochoric system
such as in Eq. 16.5 (no reaction intermediates) is defined as [IUPAC‘s Gold Book]

1 d ½A 1 d ½B 1 d ½C 1 d½D

r¼ ¼ ¼ ¼ , ð16:6Þ
a dt b dt c dt d dt
where [X] denotes the concentration of the substance X.
The rate of a reaction is always positive. A negative sign indicates the reactant
concentration is decreasing. As can be seen, the rates of reaction for different
reactants and products are related by the concentrations of the substances and the
stoichiometric numbers. The reaction rate usually has the unit of mol L1 s1. It is
noted that the above relationship is only valid for a single reaction in a closed system
with a constant volume.
Chemical Kinetics and Its Mathematical Description 177

Reaction Rate Equation

The relationship between the reaction rates of different substances is described

above. However, extra work is still needed to determine the absolute value of the
reaction rate r. A rate equation or a rate law is a mathematical expression used in
chemical kinetics to link the rate of a reaction to the concentration of each reactant.
In gas phase reactions, the rate is often alternatively related to the partial pressures.
For the chemical reaction aA + bB ! cC + dD, the rate equation is of the following
form:

r ¼ k½Aa ½Bb , ð16:7Þ

where k is the reaction rate coefficient or rate constant. However, this parameter is
not necessarily a constant. The reaction rate coefficient allows for the influences of
all the parameters except concentration. Among these factors being considered,
temperature is usually the most important one. The exponents a and b are the
reaction orders which can be determined based on the reaction mechanism. For
elementary or single-step reactions, the order of each reactant is equal to its stoi-
chiometric coefficient. However, for complex or multistep reactions, the rate equa-
tion needs to be determined based on the detailed step-by-step mechanisms. In the
above example, the reaction orders are a and b for A and B, respectively, while the
overall reaction order is a + b.
Combining Eqs. 16.6 and 16.7 for the reaction rate, we then can obtain the
following first-order ordinary differential equation:

d ½A
¼ ak ðT Þ½Aa ½Bb : ð16:8Þ
dt
This can be viewed as the governing equation for the chemical field, though a
comprehensive description of the field requires the mass transport as well. A
theoretical solution is possible for simple cases such as simple reactions. Taking
the reaction aA!cC + dD, for example, the reaction rate of this first-order reaction
can be obtained as

d½A
r¼ ¼ k½A: ð16:9Þ
dt
The integration of the above equation yields

½A ¼ ½A0  ekt ð16:10Þ

where [A]0 is the concentration of A at the beginning of the chemical reaction.

As can be seen, the rate of a first-order reaction depends only on the concentration
and the properties of the involved substance. More than one time constant is needed
when describing reactions of a higher order. Taking the reaction of hydrogen and
nitric oxide, for example:
178 16 Chemico: Chemical Reaction

2H2 þ 2NO ! N2 þ 2H2 O, ð16:11Þ

the observed rate equation is

r ¼ k½H2 ½NO2 : ð16:12Þ

Just like many reactions, the above rate equation does not correspond to the
stoichiometric coefficients in the overall reaction – H2 is the first order while the
stoichiometric coefficient in the overall reaction is 2. This is because not all of
the elementary steps affect the rate of reaction. We need to investigate the elementary
steps to understand the overall reaction rate:

2NO ! N2 O2 ðfast equilibriumÞ

N2 O2 þ H2 ! N2 O þ H2 O ðslowÞ
N2 O þ H2 ! N2 þ H2 O ðfastÞ

Usually, the slowest elementary step controls the reaction rate. In this case, the
second reaction is the slowest one and thus controls the overall rate. This is a
bimolecular elementary reaction whose rate is given by the second-order equation:
r ¼ k2[H2][N2O2], where k2 is the rate constant for the second step. However, N2O2
is an unstable intermediate, and its concentration depends on the first action:
[N2O2] ¼ k1[NO]2, in which k1 is the equilibrium constant for this first step.
Substituting this equation into the overall reaction rate equation leads to

r ¼ k1 k2 ½H2 ½NO2 : ð16:13Þ

As can be seen, only one H2 appears in the rate equation. This is because the other H2
reacts in the third step, which is fast and thus does not affect the overall reaction rate.

Integration with Mass Transfer

Although a closed system, or reactor, is assumed to obtain the above reaction rate,
the real computational domain consists of a number of such mini “closed” systems
involving mass transport. Therefore, these mini systems are not really closed due to
the mass transport between them. However, we assume the chemical actions occur in
a way similar to that in a closed system, i.e., not obviously impacted by the mass
transport, within a short time span or a time step in numerical analysis. Within this
step, the system is closed. However, the system is open between steps so that the
concentrations are updated by solving the mass transfer equation including the
source term, which is calculated with the chemical reaction results. The full mass
balance must be taken into account: accumulation ¼ in – out + generation –
consumption, or mathematically:
Reaction Rate Coefficient 179

ZV
dc
¼ qin þ qout þ rdV, ð16:14Þ
dt
0

where c is the concentration, i.e., [A], qin is the inflow rate of a material in molecules
per second, qout is the outflow, and r is the instantaneous reaction rate of A. When
½A
applied to a closed system with a constant volume, this equation reduces to r ¼ ddt .

Reaction Rate Coefficient

Temperature Dependence

The temperature dependence of the reaction rate coefficient can be considered using
the Arrhenius equation:

k ¼ k0 eRT ,
Ea
ð16:15Þ

where Ea is the activation energy and R is the gas constant. Since at temperature T,
the molecules have energies given by the Boltzmann distribution, one can expect the
number of collisions with energy greater than Ea to be proportional to eRT . k0 is a
Ea

pre-exponential factor or frequency factor.

The values of k0 and Ea are dependent on the reaction. There are also more
complex equations, which describe the temperature dependence of other rate con-
stants that do not follow this pattern.
A chemical reaction takes place only when the reacting particles collide. How-
ever, not all collisions are effective in causing the reaction. Products are formed only
when the colliding particles possess a certain minimum energy called the threshold
energy. Basically, the number of activated particles nearly doubles for a temperature
of T + 10 K. For convenience, the temperature coefficient, which is the ratio of its
rate constant at a higher temperature to its rate constant at a lower temperature, is
usually adopted. For example, Q10 is commonly used as the ratio of rate constants
that are 10  C apart.

Pressure Dependence

The pressure dependence varies with different materials and with the ranges of
volume. For condensed-phase reactions, i.e., solids and liquids, the dependence is
usually sufficiently weak in the range of pressures that we normally encounter.
Therefore, the pressure dependence of the reaction constant in such materials is
180 16 Chemico: Chemical Reaction

negligible. For gas reactions, the pressure dependence of the rate constant is asso-
ciated with the activation volume. For the following reaction,

A þ B Ð jA  Bj{ ) AB,

the activation volume, ΔV{, is

ΔV { ¼ V {  V A  V B , ð16:16Þ

where V is the partial molar volumes of the reactants and products and { indicates the
activation-state complex.
The change of the reaction rate constant in the above equation with pressure at a
constant temperature can be formulated as
 
∂ ln kx
RT ¼ ΔV { : ð16:17Þ
∂P T

The conditions in real reactions can be complicated because the partial molar
volumes and the activation volume can also vary with the pressure.
The reaction rate can increase or decrease with the pressure, depending on the
value of ΔV{. As an example of the possible magnitude of the pressure effect, the
reaction rate in some organic reactions can be doubled when the pressure is increased
from one atmosphere, i.e., 0.1 MPa, to 50 MPa.
Chapter 17
Electro: Electrostatics

Introduction

Electrostatics is the study of the processes and properties of stationary or

low-velocity electric charges with negligible acceleration. Traditionally, electrostat-
ics includes both the electric field and the forces on the objects in the field. This
definition and scope are fixed in the above way attributed to the way we discovered
and understood the electric field. From the perspective of multiphysics, the above
definition of electrostatics involves both the electric field and forces. From this
viewpoint, electrostatics can be categorized as a multiphysics topic. However, we
still discuss it as monolithic physics considering the fact that the force is just one way
we understand the electric field and that our interest is more about the origin and
nature of the force rather than the mechanical behavior of the objects under the
influence of the forces.
As mentioned above, our interest in and understanding of the electric field arose
from forces that electrically charged objects exert on each other. This type of
phenomenon is especially common in tribomaterials. For example, plastic wrap
can be easily attracted to our hands. It also can be observed that the attractive
force between the hands and the plastic wrap will become stronger as they move
closer to each other. For many materials, such attractive forces only become apparent
after a material gets into contact with and separated from another material that is
usually different from the first material. This contact-and-separation process enables
electron transfer between the materials, and as a result, the materials are oppositely
charged. That is, one material loses electrons, while the other receives electrons.
Then we can understand that one of the two materials, say material 1, generates an
electric field, and the other material, say material 2, as an ensemble of charged
particles, suffers from the forces due to the existence of the electric field generated by
material 1. But to be more accurate, this electric field is just a superposition of the
numerous electric fields generated by the material points comprising material 1.
On the contrary, we can also understand the process as that the material points in

© Springer International Publishing AG, part of Springer Nature 2018 181

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_17
182 17 Electro: Electrostatics

material 2 also generate a total electric field, which exerts forces on the material
points in material 1. From this description, we can infer three things:
1. The occurrence of such forces requires the presence of both charged particles and
an external electric field.
2. The forces are mutual, and thus the description is also mutual when there are two
charged objects.
3. The forces on macroscopic objects stem from the forces on the material points.
Thus an explanation of electrostatics using such point charges is more
fundamental.
Without involving magnetics and heating, the electric field is mostly noticeable
via the forces it generates. In many cases, electrostatically induced forces could
appear to be rather weak, such as the force between the plastic wrap and hands.
However, when compared with the gravitational force, the electrostatic force
between an electron and a proton, which together make up a hydrogen atom, is
about 36 orders stronger than gravity in magnitude. Another factor which makes the
effects of charge exchanges during contacts less noticeable is that the electrons
gaining or losing in the contact-and-separation process remain at site only in
materials with a high resistance to electric currents. This is because only such
materials can prevent the electrons or cavities from moving away or being filled
within a time which is long enough to be observed. These electrons have a tendency
to move away also because of electric fields: each electron generates an electric field
and pushes away other nearby electrons. In some cases, quick neutralization could
happen if two oppositely charged objects with high resistance meet, leading to a
dramatic discharging process such as the static ‘shock’ or lightning.
In this chapter, we will first introduce the forces between electrically charged
particles. Coulomb’s law for describing these forces will be discussed. Next, we will
move to the forces that a charged particle undergoes in an electric field. Then,
Gauss’s law will be introduced to link the amount of the electric charge to the
electric field generated by the charged particles. Followed will be the derivation of
the governing equation of electrostatics in terms of the electric potential. We will
also show the energy associated with the electric field. Electric displacement field
and dielectric materials will be explained to guide the application of electrostatics in
materials. A practice problem will be given in the final section to demonstrate the
numerical simulation of electrostatics.

Coulomb’s Law

Let us first take a look at the electrostatic force between two objects. Charles-Augustin
de Coulomb reported that the magnitude of the electric force between two point
charges is directly proportional to the product of the charges and inversely propor-
tional to the square of the distance between them based on his test for studying the
repulsion and attraction forces of charged particles using a torsion balance in 1785.
Electric Field 183

Fig. 17.1 Forces on two q1 q2

r21
close point charges F1 F2
r1 r2
o

For two charges q1 and q2, Coulomb’s law can be written out using the following
equation:

1 q1 q2 qq
F¼ ¼ ke 1 2 2 , ð17:1Þ
4πε0 r 2 r
where ε0 is the vacuum permittivity,r is the distance between two charges, and ke is
Coulomb’s constant. The vacuum permittivity and Coulomb’s constant have the
following values:

109 2 -1 -2
ε0 ¼ C N m  8:854187817  1012 C2 N-1 m-2 ð17:2Þ
36π
1
ke ¼  8:987551787  109 C-2 N m2 : ð17:3Þ
4πε0
This equation also can be written in terms of vectors considering that forces are
vectors as shown in Fig. 17.1
q1 q2
F1 ¼ ke r21 , ð17:4Þ
jr21 j3

where F1 is the force applied on q1 by q2 and r21 is the vectorial distance pointing
from q2 to q1:

r21 ¼ r1  r2 , ð17:5Þ

where r1 and r2 are the vectors representing the position of the two point charges in a
given frame of axes. Therefore, if q2 to q1 are positive, the direction of the force on q1
will be the direction of the vector pointing from q2 to q1.
The rule of superposition can be applied to the above electrostatic forces. For
example, if there exist other point charges, q3,. . ., qn, the electrostatic force on the
point q1 is

X
n
q1 qi
F1 ¼ ke ri1 : ð17:6Þ
i¼2 jri1 j3

Electric Field

The electric field can be defined based on the above understanding of the electro-
static force. In the above example, if the point charge q1 changes its position, then its
position and ri1‘s will also change. As a result, the force on q1 will change. But if the
184 17 Electro: Electrostatics

point charge remains at the same location, q1 will undergo the same force no matter
how the charge moves to the location. Therefore, each point in the space has a unique
quantity which determines the magnitude and direction of the force acting on q1 and
this quantity will change with locations. This quantity thus demonstrates the char-
acteristic of a physical field. Therefore, we can define a physical field for this
quantity in the following way:

X
n
qi
Ε¼ ke ri1 : ð17:7Þ
i¼2 jri1 j3

This is, in fact, the definition of the electric field caused by all the point charges
except q1. We can see that this electric field is independent of q1. Therefore, an
electric field is independent of the charge that we place in the field if we overlook the
electric field generated by this charge. To reach a more general definition, let us take
a look at the electric field generated by n charges: point charges q1,. . .,qn. The total
electric field generated by these charges is

X
n
qi
Ε¼ ke ri , ð17:8Þ
i¼1 j ri j 3

where ri is the unit vector pointing from point qi to the point of interest. The above Ε
is the total electric field: the superposition of the electric fields generated by all the
point charges. In a similar way, we can define the electric field generated by the point
charge qi as
qi
Εi ¼ k e ri : ð17:9Þ
j ri j 3

If qi is a positive source point charge, then the direction of the electric field points
along lines directed radially outward from the charge. On the contrary, a negative qi
generates an electric field with a radially inward direction.
With the above definition of the electric field Ε, we then can easily write down the
force acts on any given point charge q as

F ¼ Eq: ð17:10Þ

The above definition of the electric field can be graphically represented using
electric field lines as shown in Fig. 17.2. Electric field lines are lines with arrows
pointing from positive and/or pointing to negative charges. Electric field lines are
parallel to the direction of the electric field, and the density of the lines at any point
indicates the magnitude of the electric field therein. We can further define the electric
flux as a measure of the flow of the electric field through a given area. Graphically,
the electric flux is proportional to the number of electric field lines going through a
normally perpendicular surface. Another field can be developed by combining this
proportionality and the electric field: electric displacement field D, which will be
introduced later. D quantifies the direction and density of the electric field lines
Gauss’s Law 185

Fig. 17.2 Conductors in the

electric field generated by a
point charge

directly without a proportionality. As a result, D is also called the electric flux

density with a unit of C/m2. Since the electric field, E, is a vector field defined
everywhere except at the location of point charges, electric lines theoretically should
not get into these points.
One thing that needs to be pointed out is that there are no electric field lines inside
conductors even if they are placed in an electric field. This reason is that perfect
conductors have no resistance to the movement of electrons. Therefore, when it is
placed in an electric field, the electrons in the conductors will feel the electric force
and move immediately. Based on the direction of the force, the positive charges will
move along the electric field lines while negative charges move against the direction
of the lines. As shown in the above figure, they will move to the peripheries of the
conductors. The movement of charges will cease when the charges on the periphery
of a conductor are able to generate an electric field which is strong enough to cancel
out the external electric field within the conductor. As a result, there is no electric
field or electric field lines in the conductor.
The above electric field is defined with discrete charges. In continuous materials,
it is more common to use the charge density ρ, which could change with position r.
In this case, we can define the electric field generated by a continuous charge “cloud”
with a charge density ρ at a point with a position vector r0 as
Z
1 1
EðrÞ ¼ ρðrÞdr: ð17:11Þ
4πε0 kr 0  r k2

Gauss’s Law

In the above section, we showed how to define the electric field based on the forces
between point charges and extended this definition to continuous materials. In fact,
the above definition can be further generalized into a relationship between charges
186 17 Electro: Electrostatics

and the electric field: Gauss’s law. Gauss’s law can serve as a key relationship in
establishing the governing equation for the electric field. The law relates the electric
charge, which serves as the source in the electric field, to the electric field, either E or
D, which serves as the flux.
Gauss's law states that the total electric flux through any closed surface of any
shape in an electric field is proportional to the total electric charge enclosed within
the surface. This law can be formulated into the following form:
Z Z
q ρ
E  ndΩ ¼ ¼ dV: ð17:12Þ
ε0 ε0
Ω Ω

Gauss’s law applies to both 2D and 3D, so Ω can be a 2D domain (area) or a 3D

one (volume). Accordingly, ρ can be both a surface charge density and a volume
charge density. The above equation in the integral form can also be reformatted into
an equivalent equation in the differential form:
ρ
∇E¼ : ð17:13Þ
ε0
This theorem was proposed by Gauss based on his study of the electric flux. The
Gauss theorem states a physical and mathematical law which has more general
implications in physical fields: the outward flux of a vector field through a closed
surface is equal to the volume integral of the divergence over the region inside the
surface.

Governing Equation and Electric Potential

The above equation for Gauss’s law cannot be used as the governing equation for the
electric field. From the physical perspective, as mentioned in the previous section,
Gauss’s law serves more as a constitutive relationship for relating the source to the
electric field flux. From the mathematical perspective, the equation relates a vector,
i.e., E or D, to a scalar, i.e., ρ. Gauss’s law provides one equation for multiple
dependent variables. Each component of the vector-dependent variable counts as one
dependent variable, and the order of quantity of either the left- or right-hand side of
the equation determines the number of governing equations (although in many cases,
we treat the number of PDEs as the number of governing equations).
The governing equation of the electric field stems from the observation that the
electric field is irrotational or equivoluminal, mathematically, that is,

∇  E ¼ 0, ð17:14Þ

where “0” is a vector. So in a 3D case, the above PDE actually represents three
dependent variables, i.e., three components of E, and three governing equations for
Governing Equation and Electric Potential 187

the three components. The above equation is a special case of Faraday’s law of
induction in electrostatics. We can simulate the electric field with the above
governing equation (here we still count it as one following the convention used in
other places of this book) and the Gauss’s law. However, the practice may be not
easy because we are dealing with two separate equations containing the same vector.
We usually need to calculate E caused by a separate charge or charge cluster at
special locations and then use these E values as boundary conditions together with
the governing equation to calculate E throughout the domain.
A more convenient and common way for the use of the governing equation in the
electric field is to combine the above two first-order PDEs into one second-order
PDE with a new scalar-dependent variable. This is, in fact, a very common practice
in both mathematics and physics. First, for any irrotational physical field, we can
relate the vector field, i.e.,E in this example, to a scalar variable, which usually has
the meaning of a potential, in the following way:

E ¼ ∇ψ, ð17:15Þ

where ψ has a physical meaning electric potential with a unit of volt (V) in this case.
The electric potential difference is the “voltage” that we use widely in daily life.
Therefore, the electric field has a unit of V/m. The magnitude of the electric field at a
particular location is usually called the electric field strength. The negative sign is
added considering that the electric field lines point from locations with a high
potential to those with a low potential.
If we write the above equation in terms of a line integral, then we can obtain the
conclusion that the voltage is the amount of work required to move a unit charge
from the start point to the end point of the line:
Z b
 E  dl ¼ ψ ðbÞ  ψ ðaÞ, ð17:16Þ
a

where a and b are the start and end points.

In fact, the above assumption that the electric field is proportional to the gradient
of another scalar field is not just a postulate. It can be deducted based on the
observation that the electric field is irrotational, mathematically, that is,

∇  E ¼ 0: ð17:17Þ

This can also be seen from a key feature of the electric field: work is path
independent, or the choice of any path between two points does not change the
value of the line integral.
The definition of the electric potential, combined with the differential form of
Gauss’s law, provides the following governing equation for the electric field:
ρ
∇2 ψ ¼  : ð17:18Þ
ε0
188 17 Electro: Electrostatics

This equation is in the form of Poisson’s equation. In the absence of electric

charges, the equation degenerates into Laplace’s equation:

∇2 ψ ¼ 0: ð17:19Þ

This PDE can be solved with appropriate boundary conditions. A Dirichlet

boundary condition for an electric field just gives out the electric potential on the
boundary. For example, when the boundary is far enough from charges, we can
assume the boundary is infinitely far and thus has an electric potential of zero. By
contrast, a Neumann boundary condition places a constraint on the gradient of the
electric potential, i.e., electric field. For example, for any conductor in an electric
field, the direction of the electric field needs to be parallel to the outward normal of
the surface. Therefore, the component of the electric field parallel to the surface
(or perpendicular to the outward normal vector) will be zero. More complicated
boundary conditions will be discussed in the chapter for electromagnetics in the next
part of the book.

Electrostatic Approximation

From another perspective, electrostatics can be viewed as a special case of electro-

magnetics which is more common in reality. The validity of the electrostatic
approximation rests on the assumption that the electric field is irrotational. From
Faraday’s law, this assumption implies the absence or near absence of time-varying
magnetic fields:

∂B
¼ 0, ð17:20Þ
∂t
where B is the magnetic field.
Therefore, electrostatics does not require the magnetic field or electric current to
be totally absent. Rather, the magnetic field or electric current, if exists, must not
change with time. When the magnetic field or electric current changes with time very
slowly, we may also be able to simplify the problem into electrostatics with an
acceptable loss in accuracy.

Electrostatic Energy

Energy is an important aspect of a physical field. The energy of the previously

introduced physical fields is mostly straightforward, either in terms of mechanical
work, potential energy, strain energy, or chemical potential. However, the energy of
the electric field is slightly different and needs discussions. This energy is usually
needed when the electric field is coupled to other fields such as the thermal field.
Electrostatic Energy 189

To obtain a better understanding, let us still start with point charges. The simplest
scenario is to consider the energy of a point charge q in an external electric field. The
total energy or, more accurately, the electric potential energy, owned by this point
charge is equal to the energy that is needed to move this charge from an infinitely far
location (with zero electric potential) to the current location:

U ¼ qψ: ð17:21Þ

If ψ is positive, i.e., higher than the potential of the start point, then extra work is
needed to make the charge move against the electric field line. Accordingly, the
potential energy owned by the point charge will be positive.
In the above example, if the external field is generated by another point charge q2
at r2, then the energy needed to move q, which is marked as q1 for convenience, from
an infinitely far location to a location r1 will be
q2
U ¼ q1 ϕ 1 ¼ q1 k e , ð17:22Þ
jr12 j

where r12 ¼ r1  r2. In fact, this is not merely the energy of charge q1 but that of the
system consisting of q1and q2. In the above example, we would do the same amount
of work for moving q2 to the same relative position of q1. In short, we need to do this
amount of work to the system consisting of the two charges by moving them to
positions with a distance of |r12|. Therefore, U in the above equation is the potential
energy of the whole system. In the above equation, we can also write the system’s
electric potential energy as
0 1
1X 2
1X 2
@ qi
X2
qj
A:
U¼ qψ ¼ ð17:23Þ
2 i¼1 i i 2 j¼1 r
j¼1ðj6¼iÞ ij

In order to extend the above equation to a more general condition, then let us
consider a collection of n charged particles, i.e., q1, q2, . . .,qi,. . .qn. qi is located at the
point with a coordinate vector of ri. Then the electric potential energy owned by the
whole system is
0 1
1X n
1X n
@ qi
Xn
qj
A:
U¼ qψ ¼ ð17:24Þ
2 i¼1 i i 2 j¼1 r
j¼1ðj6¼iÞ ij

Then we can extend the above energy calculation from discrete systems to
continuous systems. In a continuous system, the charges are distributed in a domain
consisting of elementary volumes. Then for each elementary volume dV, the energy
can be calculated as

1
dU ¼ ðρdV Þψ: ð17:25Þ
2
The total energy of the total domain can be obtained via integration
190 17 Electro: Electrostatics

Z Z  
1
U¼ dU ¼ ρψ dV: ð17:26Þ
2
Ω V

The electric potential energy of a continuum can also be formulated using the
electric field and permittivity. To derive the equation, we first substitute Poisson’s
equation into the above equation:
Z Z  
1
U¼ dU ¼ ðε∇  EÞψ dV: ð17:27Þ
2
Ω V

Then we can take advantage of the following divergence vector identity:

∇ðEψ Þ ¼ ð∇  EÞψ þ E∇ψ: ð17:28Þ

Substituting Eq. 17.28 into Eq. 17.27, we obtain

Z Z Z
ε ε
U ¼ dU ¼ ∇  ðψEÞdV  ∇ψ  EdV: ð17:29Þ
2 2
Ω V V

Then, we apply the divergence theory:

Z Z Z
ε ε
U ¼ dU ¼ ðψEÞ  ndS  ðEÞ  EdV: ð17:30Þ
2 2
Ω V V

If we set the electric potential at the boundary of the domain to zero, for example,
by assuming the boundary is at infinity, then the first term on the right-hand side of
the above equation will be zero. The energy then has the following formulation:
Z
ε
U¼ E  EdV: ð17:31Þ
2
V

Based on the above introduction, the electric potential energy density of the
electric field can be formulated using either ρψ/2 or εE2/2.

Electric Displacement Field and Dielectric Materials

As introduced above, the electric field E multiplied by another scalar, e.g., ε, will
become a new field: the electric displacement field. This field is not just randomly
created with two existing physical quantities but rather a very useful quantity with a
physical meaning in dielectric materials. Nevertheless, the physical meaning is not
as clear as that of the measurable quantity E. In vacuum, E is caused by free charges
only. However, when we consider the electric field in a normal dielectric (material),
Electric Displacement Field and Dielectric Materials 191

we will find the dielectric material will resist the external electric field by moving its
electric dipoles. As a result, the external electric field will be weakened.
To better understand this, let us introduce polarization which quantifies this
resistance. When an external electric field is applied to a dielectric material, bound
charged elements within a molecule will move in response to the external electric
field: positively charged elements move in the direction of the field while negatively
charged elements move opposite to the direction of the field. These movements are
confined in a limited distance as these elements are bound to a molecule and are thus
not free to move around in the material. Therefore, the molecule remains neutral as a
whole even with the movement of the charged elements within it. This movement
can be quantified with a vector, polarization density, which is defined as

dqb
P¼ d ¼ ρb d, ð17:32Þ
dV
where the amount of the bound charge qb within an infinitesimal volume gives out
the bound charge density ρb and the displaced distance vector d gives out the
direction of the polarization vector. qbd (unit: C  m) is also called the electric dipole
moment, and accordingly P has the physical meaning of the density of the electric
dipole moment (unit: C  m/m3). The displaced charges will also generate a second-
ary “bound electric field,” which is opposite to the original external field. We can
consider this field as one generated by the bound charges, so we have

∇  P ¼ ρb : ð17:33Þ

The negative sign is given to indicate the opposite direction (resisting to the
external electric field). This is Gauss’s law for bound charges. We can obtain the real
electric field by combining the original external field and the resistance field caused
by the bound charges. In vacuum (ρb ¼ 0), there is no bound charge, and thus
Gauss’s law gives

∇  ðε0 E0 Þ ¼ ρ0 ¼ ρ f , ð17:34Þ

where E0 and ρ0 are the electric field and total charge density in vacuum. If we
replace the vacuum with a dielectric without altering the charges, then we have

∇  ðε0 EÞ ¼ ∇  ðε0 E0 Þ  ∇  P ¼ ρ f þ ρb ¼ ρ, ð17:35Þ

where E and ρ are the electric field and total charge density, respectively, in the new
domain with the dielectric. As P is usually in the direction of E0, the electric field is
weakened compared to that in the vacuum:

∇  ð ε0 E Þ þ ∇  P ¼ ρ f : ð17:36Þ

A more common way to write the equation is to develop a new quantity, the
electric displacement field:
192 17 Electro: Electrostatics

D ¼ ε0 E þ P ¼ εE, ð17:37Þ

so that

∇  D ¼ ρ f: ð17:38Þ

Therefore, D does not change when an external field is placed on different

dielectrics. We can further define the dielectric constant εr to quantify the permit-
tivity of any material with reference to the permittivity of vacuum:

D ¼ εE ¼ ε0 εr E ð17:39Þ

Most minerals in geomaterials have εr values ranging from 1 to 5, while water has
a εr value of 80. Permittivity is the amount of charge needed to generate one unit of
electric flux in a particular medium. Thus, permittivity measures a material’s ability
to resist an electric field, not its ability to “permit” the field, as the name “permittiv-
ity” might suggest.
As εr is usually larger than 1, while D does not change in different materials, the
electric field E will be smaller (more weakened) in materials with a higher dielectric
constant.

Practice Problem

Please simulate the electric field generated by a metal cylinder whose electric
potential is 10 V. The metal cylinder has a radius of 0.2 m, and the length is much
larger than the radius.

Results

Based on the above description, the problem can be simplified into a 2D case, in
which the metal piece is a circular area. The permittivity of the metal can be
represented using a very large number such as 1  1012. Theoretically, the electric
potential is zero at an infinitely far location. However, we can use a computational
area that is much larger than the size of the metal area to approximate the theoretical
condition. The following results (Fig. 17.3) were obtained by using a square
computation area with a side length of 1 m. Considerable error can be generated
due to the size and shape of the domain. Please try to conduct a sensitivity analysis to
investigate the influence of the size and shape of the computational domain on the
results. The area outside of the metal area is assumed to be vacuum.
Practice Problem 193

Fig. 17.3 Electric potential (upper) and electric field (lower)

194 17 Electro: Electrostatics

Appendix: FlexPDE Script File

{ Electric field generated by a ring with 10 V}

TITLE '2D electric field'
VARIABLES
psi { Electric potential }
! SELECT
DEFINITIONS
E=-GRAD(psi) { Electric field }
rho=0
epsilon=8.654e-12
! INITIAL VALUES { Not time-dependent }
EQUATIONS
DIV(E)=rho/epsilon
BOUNDARIES
REGION 1 { Vacuum }
START(-1,-1) VALUE(psi)=0
LINE TO (1,-1) VALUE(psi)=0
LINE TO (1,1) VALUE(psi)=0
LINE TO (-1,1) VALUE(psi)=0
LINE TO CLOSE
REGION 2 { Charge area: a metal cylinder - a circular plate in 2D }
epsilon=1e12 { Use a large permittivity value to simulate
conductors }
START(0.2,0) VALUE(psi)=10 ARC(CENTER=0,0) ANGLE=360 TO CLOSE
! TIME 0 TO 1 { Not time-dependent }
MONITORS
PLOTS
CONTOUR(psi)
VECTOR(-GRAD(psi))
END
Chapter 18
Magneto: Magnetostatics

Introduction: Magnetostatics and Magnetic Field

The study of the magnetic field originated from the observations on the magnets and
the forces they place on objects made of iron and other metals. The study of these
phenomena, i.e., magnetism, has extended far beyond magnets and has been
entangled with the study of the electric field in electromagnetics. Similar to the
electrostatics, we here try to separate the magnetic field from electromagnetics by
confining the systems to a stationary state (no change with time) or changes with low
velocities, i.e., currents do not alternate rapidly. With the inclusion of forces,
magnetostatics can then be viewed as the magnetic analogy of electrostatics.
Similar to the electric field, there are also two physical fields defined to describe
the magnetic field, H field and B field, which correspond to the E field and D field in
the electrostatics, respectively. However, the use of H field and B field is possibly
more confusing than their compartments in the electric field. Both H field and B field
can be called the magnetic field, depending on the disciplines. It is more common to
call H the magnetic field (or magnetic field strength/intensity) because the term
“magnetic field” was historically reserved for the early proposed H. However, many
researchers, especially in physics, believe B is more fundamental than H, so B is
called the magnetic field. H has a unit of A/m, which can be compared to the unit of
E: V/m. Similar to D, which is the electric flux density with the unit of C/m2, B is the
magnetic flux density with the unit of Wb/m2 (webers per square meter, or tesla). The
magnetic field lines are a tool to visualize the magnetic field, and their density is
measured in terms of Wb.
The magnetic field is generally more difficult to understand than the electric field.
In the special relativity of Albert Einstein, the electric field and the magnetic field are
just different aspects of one thing, i.e., electromagnetics, and thus one may turn into
the other depending on the observational frame of reference. But from a simpler way,
we can understand the magnetic field as the result of two things: moving electrons
and spin magnetic moments of elementary particles (particles whose substructure is

© Springer International Publishing AG, part of Springer Nature 2018 195

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_18
196 18 Magneto: Magnetostatics

unknown, e.g., quarks and leptons). The latter is in general much less obvious than
the former except for applications such as nuclear magnetic resonance. For electrons,
a rotating electron creates a magnetic dipole with two magnetic poles of equal
magnitude but opposite polarity. Assuming an electron rotates about an axis with
an angular momentum of L, the generated magnetic dipole moment m is
e
m¼ L, ð18:1Þ
2me
where e and me are the charge (elementary charge) and rest mass of the electron. The
magnetic dipole moment can also be equivalently defined with a sufficiently small
loop of current, I, and loop area, A:

m ¼ IA: ð18:2Þ

The above two definitions are equivalent if the loop of current is caused by the
charge: I ¼ Te ¼ 2πe
ω , in which ω is the angular velocity. The direction of the vectoral
area A needs to be the same as the direction determined using the right-hand rule
with the loop current, as shown in Fig. 18.1.
Usually, electrons in a material are arranged in a way that their magnetic
moments, both orbital and intrinsic, cancel out when the ensemble of the electrons
are viewed as a whole. But in some conditions, such as in the mineralization of
magnetite or under an applied external magnetic field, the electron magnetic
moments will be lined up toward one direction to produce a strong net magnetic
field. The induced magnetic field could remain to some extent even after the removal
of the external magnetic field, leading to permanent magnets.
Magnetic dipoles are commonly introduced when discussing magnetism. This
concept is analogous to yet much different from the positive and negative charges in
the electric field. One major difference is that the existence of the magnetic mono-
pole has still not been validated. A common observation is that every magnet comes
with a dipole: a “south pole” and a “north pole.” If we cut a magnet into two pieces,
then each piece will also own its “south pole” and a “north pole.” This can be
explained with the origin of the magnetic dipoles from the rotating charges
(Fig. 18.1). Each rotating charge develops a dipole on two sides of the plane of
rotation. We thus cannot separate the two poles as a plane always has two sides. A
macroscopic magnetic material consists of numerous dipoles like this. If we cut the

Fig. 18.1 Move electric

charges (current) and
resultant magnetic dipole
Magnetic Field 197

material into two pieces, the dipole ensembles in each piece will still own a magnetic
field with the same direction.
In this chapter, we will first discuss the forces associated with a charge in a
magnetic field to introduce the concept of the magnetic field (B field) just like what
we did in the previous chapter for the electric field. Then, we will present a
mathematical description of the magnetic field using two different methods relying
on the H field and B field. Gauss’s law for magnetism will be given to reach the
governing equation for the magnetic field. A vector potential will be introduced to
turn the governing equation into a second-order partial differential equation, which is
useful and common in numerical analysis. The energy associated with the magnetic
field will be introduced afterward. A practice problem will be provided in the final
section for simulating the magnetic field generated by a permanent magnet.

Magnetic Field

The magnetic field can be defined in many equivalent ways. One common way is via
the study of the force on a moving charged particle in the vicinity of a current-
carrying wire. Based on the previous chapter, we know that a particle with charge
q in an electric field E experiences a force F ¼ qE. However, we can also easily
discover via simple experiments that the force also depends on the velocity of the
particle. If we incorporate this additional part in the force, we would obtain the
Lorentz force law:

F ¼ qE þ qv  B, ð18:3Þ

where v is the particle’s velocity. We can understand the first term as the forces
between static charges while the second term as the forces between non-stationary
charges. For the second term, the electric current in the electric wire generates a
magnetic field, which exerts a force on the moving charge. If the field is defined
using the second term in the above Lorentz law, then this field, i.e., B field, is the
way to quantify the magnetic field.
Another common way to define the magnetic field is to study the torque that a
magnetic field applies on a magnetic dipole.
As shown in the above figure, we conceptualize the magnetic dipole as two
connected magnets with a distance d and an equal pole strength of qm, within
which one represents the south pole and the other represents the north pole. As
shown in Fig. 18.2, these two monopoles can be understood as charges with different
signs in an electric field: north corresponds to positive, whereas south corresponds to
negative. Therefore, the positive (north) pole will move along the direction of the
magnetic field lines, while the south will move against the direction of the lines. As a
result, a torque develops on the dipole. This torque, in fact, explains why a compass
rotates to align itself with Earth’s magnetic field. In this example, we find that the
torque is proportional to the pole strength qm (unit: A  m), the distance between the
198 18 Magneto: Magnetostatics

Fig. 18.2 Torque on a

magnetic dipole in a
magnetic field

poles |d|, and the external magnetic field. The magnetic field defined using the
following equation is also the B field

τ ¼ qm d  B, ð18:4Þ

where d is a vectoral distance pointing from the south (negative) pole to the north
pole. Accordingly, based on the right-hand rule for the cross product, the direction of
the torque is pointing inward.
We can observe a major difference between the electrostatics and magnetostatics
in the above two ways of defining the magnetic field: Eis the field directly relating an
electric charge to a force in electrostatics, and B is the field directly relating a
magnetic dipole to a force in the magnetostatics. However, as mentioned, E is
analogous to H instead of B. As the counterpart of D, B is more fundamental and
does not change in different materials. Therefore, for a magnet in vacuum, B field
lines start from the north pole, extend into the vacuum, and return to the south pole;
the lines will continue without any change inside the magnet (a material) until they
get back to the north pole. A more accurate and common statement is that “Bfield
lines neither start nor end.” In electrostatics, charges of different signs can exist
separately, i.e., no requirement of electric dipoles, so D field lines, i.e., electric field
lines, do not have the above feature. However, just like the E field, the H field will
alter in materials due to the resistance of the material to the external magnetic field.
Inside the above magnetic, the magnetic materials will raise resistance to alter the
B field. The part of the B field excluding the resistance raised by the magnetic
materials is the H field. Outside of the magnetic field, there is no magnetic material,
neither is resistance; therefore, the H field is identical to the B field outside of the
magnet. The resistance depends on the type and geometry of the magnetic materials
and is thus very difficult to calculate.
This resistance varies from location to location, so it is also a physical field:
magnetization. Magnetization M is also called magnetic polarization. M is analo-
gous to the electric polarizationP: P is the density of the electric dipole moment
(unit: C  m/m3), while M is the density of the magnetic dipole moment (unit:A  m2/
m3). The relationship between the two types of magnetic fields and magnetization
can be mathematically formulated in the following way:
Magnetic Field 199

B ¼ μ0 ðH þ MÞ, ð18:5Þ

where μ0 is the permeability in vacuum, in which μ0 ¼ 4π  107.

In nonmagnetic materials, in which paramagnetism or diamagnetism dominates,
M is often proportional to the applied magnetic field. Accordingly, we define

M ¼ χH, ð18:6Þ

where χ is the magnetic susceptibility, which is a dimensionless quantity character-

izing a magnetic material by telling whether a material is attracted into or repelled to
the magnetic field. The magnetic susceptibility can be defined in many equivalent
forms; while the definition in the above equation is the most common form: volume
magnetic susceptibility. A positive value of χ means the magnetic field will be
strengthened in the material and the negative is verse visa. With the magnetic
susceptibility, we have the following relationship between B, H, and M:

B ¼ μ0 ðH þ χHÞ ¼ μ0 ð1 þ χ ÞH ¼ μ0 μr H, ð18:7Þ

where μr is the relative permeability.

In a simple way, we can just write

B ¼ μH, ð18:8Þ

where μ is the absolute permeability of the material. The above two equations are
introduced and referenced widely without specifying the underlying assumption: a
relationship, mostly linear, between B(H) and M. It is needed to point out this
assumption works for most cases. However, for permanent magnets, who owes
M without external magnetic field, we need to start from Eq. 18.5.
Magnetic materials are materials that are attracted to a magnet strongly enough
that the attraction can be felt. Magnetic materials can be classified as hard and soft
magnetic materials – the former can gain a permanent magnetic field after being
placed in an external magnetic field, while the latter cannot. A permanent magnet is
made by placing a hard magnetic material such as alnico and ferrite into a strong
magnetic field, whose magnetic dipoles will be aligned to a direction (on average)
and maintain part of the aligned dipoles after the external magnetic field is removed.
Soft magnetic materials like annealed iron do not tend to stay magnetized after the
external magnetic field is removed. Magnetic materials can also be categorized into
ferromagnetic, ferrimagnetic, and antiferromagnetic according to their electron
configuration and consequently the way how the magnetic moments of atoms or
molecules (usually related to the spins of electrons) response to an external magnetic
field. Magnetic materials like iron can have a volume magnetic susceptibility of
200,000, whereas common nonmagnetic materials have absolute values of χ ranging
from 1010 to 104. In nonmagnetic materials, a positive χ indicates a paramagnetic
material, while a negative χ indicates a diamagnetic material, though χ is very small
in both cases.
200 18 Magneto: Magnetostatics

Gauss’s Law for Magnetism

Gauss’s law can also be applied to the magnetic field. The Gauss’s law in the electric
field states that the electric density flux is caused by the charges which serve as
sources/sinks emitting/adsorbing the flux. When we move to the magnetic field, one
major difference is that there is no magnetic monopole like the positive or negative
charges which can exist independently. As a result, the B field lines never end. In
other words, the B field is a sourceless or a solenoidal field. For a sourceless field
such as B, the following equation is valid:

∇  B ¼ 0: ð18:9Þ

In vacuum, since the H field is related to the B field via a constant proportionality,
the above equation is also valid for the H field as follows:

∇  H ¼ 0: ð18:10Þ

However, when getting into materials, the proportionality μ may vary from point
to point depending on the materials and geometry. It is also noted that μ is not
necessarily a constant depending on the type of the magnetic material and the range
of B. Consequently, in the presence of magnetic materials, Eq. 18.10 is not valid
anymore, and we can only rely on Eq. 18.9 for B.
As explained in the previous chapter, Eq. 18.9 (or Eq. 18.10 in vacuum) is more
like a constitutive equation and cannot be used as the governing equation. To
describe the magnetic field without any considerable changes in the electric field,
i.e., magnetostatics, we need to refer to the following equation as the governing
equation:

∇  H ¼ J, ð18:11Þ

where J is the current density. This equation presents a special case of Ampère’s law,
in which any changing external electric field is excluded. In the case of permanent
magnets, if no electric currents exist, then the following governing equation can be
used:


B
∇ M ¼ 0: ð18:12Þ
μ0

This is because magnetization (magnetic polarization) exists without external

electric currents or a magnetic field. This permanent magnetization is quantified
with M, which usually has different magnitudes along different directions as magnets
are polarized. A good bar magnet may have an average magnetization magnitude of
100,000 A/m along one direction. However, magnetic materials such as ion can have
a magnetization of around a million A/m under external magnetic fields.
Governing Equation and Auxiliary Equations in Terms of Vector Potential 201

However, it is also very common to convert the above equation into a second-
order PDE by means of introducing a potential, just like the electric potential for the
electric field. This will be described in the following section.

Governing Equation and Auxiliary Equations in Terms

of Vector Potential

The magnetic field can be formulated using the magnetic vector potential. Different
from the electric field, which is irrotational, the magnetic field is equivoluminal
(sourceless or solenoidal). That is, the magnetic flux density satisfies a zero-
divergence condition; therefore, B can be represented as the curl of another vector
such that

B ¼ ∇  A, ð18:13Þ

where A is the magnetic vector potential. Substituting the above equation into
Gauss’s law for the magnetic field, we obtain

∇  ð∇  AÞ ¼ 0: ð18:14Þ

As can be seen, this equation is automatically satisfied, which confirms the

effectiveness of the use of the magnetic vector potential.
The above equation needs to be substituted into Ampère’s law to yield the
governing equation in terms of the vector potential:


∇A
∇ þM ¼ J: ð18:15Þ
μ0

If the linear relationship between Band H can be assumed, we can obtain


∇A
∇ ¼ J: ð18:16Þ
μ0 μr

As we know, except in metals, electric currents can be ignored in materials. Then

the above equation is reduced into


∇A
∇ ¼ 0: ð18:17Þ
μ0 μr

The above standard governing equation can be further simplified by introducing

extra constraints on A. First, let us use the absolute permeability:


1
∇ ∇A ¼ 0: ð18:18Þ
μ
202 18 Magneto: Magnetostatics

Then we apply Lagrange’s formula to transform the above equation into




1 1 1
∇ ∇A ¼∇ ∇  A  ∇  ∇A ¼ 0: ð18:19Þ
μ μ μ

As A is a quantity that we construct, we can impose any further constraints when

constructing A as long as the above conditions are not violated. For example, we can
set —  A ¼ 0, which is called the Coulomb gauge condition. Then, we have


1
∇ ∇A ¼ 0: ð18:20Þ
μ

For permanent magnets, permanent magnetization exists, while electric currents

are absent, so we have


∇A
∇  M ¼ 0, ð18:21Þ
μ0

or with the Coulomb gauge condition as


1
∇ ∇A  ∇  M ¼ 0: ð18:22Þ
μ

The introduction of this vector potential does not appear as useful and common as
the electric potential in the electric field. However, the vector potential will help a lot
when we move from magnetostatics to electromagnetics.

Magnetostatic Energy

The energy of the magnetic field can be obtained via the analogy between the electric
and magnetic fields. The magnetic field also has qm and A. However, these two
quantities are not as widely used as q and ϕ in the electric field, so here we do not
discuss the counterpart of ρϕ/2 in the magnetic field. Instead, we can still easily
derive the energy density function in Eq. 18.23 based on the analogy between E/D
and H/B. Corresponding to the electric field, the magnetic potential energy of a
continuum can be formulated as
Z Z Z
1 μ 1
U¼ H  BdV ¼ H  HdV ¼ B  BdV: ð18:23Þ
2 2 2μ
V V V

The electric potential energy density of the magnetic field can be formulated
using either μH2/2 or B2/(2μ), in which H and B are the magnitudes of the H and
B field, respectively.
Practice Problem 203

Practice Problem

Problem Description

Please simulate the magnetic field generated by a bar magnet that is 0.4 m long and
0.2 m wide (2D) in vacuum. The magnet has magnetic polarization along the length
direction whose magnitude is 10,000 A/m. The relative permeability of the magnet
material is 5000.

Results

The magn etic field generated by the magnet will be symmetric. So we can
analyze only one half or one fourth of the magnet to obtain the magnetic field.
Here, if we analyze one half of the magnet, we will obtain the following results for
the magnitude of the vector magnetic potential A and magnetic flux density
B (Fig. 18.3).

Appendix: FlexPDE Script File

{ Magnetic field generated by a permanent magnet with magnetization

(M) along one direction }
TITLE '2D Magnetic Field'
VARIABLES
A { Use magnetic potential }
DEFINITIONS
mu0=4*pi*1e-7
mu=1*mu0
px=0
py=0
P=VECTOR(Px,Py)
B=CURL(A)
EQUATIONS
CURL(B/mu-P)=0
BOUNDARIES
REGION 1 { Vacuum }
mu=1*mu0
START(0,0) NATURAL(A)=0
LINE TO (1,0) VALUE(A)=0
LINE TO (1,1) NATURAL(A)=0
LINE TO (0,1) VALUE(A)=0
LINE TO CLOSE
204 18 Magneto: Magnetostatics

Fig. 18.3 Magnitude of vectoral potential (upper) and B field (lower)

Practice Problem 205

REGION 2 { Magnet }
mu=5000*mu0
px=0
py=10000
START(0.4,0)
LINE TO (0.6,0) TO (0.6,0.2) TO (0.4,0.2) TO CLOSE
MONITORS
PLOTS
CONTOUR(A)
VECTOR(CURL(A)) AS 'FLUX DENSITY B'
END
 Part IV
Multiphysics
Chapter 19
Thermomechanics: Non-isothermal
Mechanics

Introduction

The coupling between the thermal and mechanical fields leads to one of the most
common types of multiphysics: thermomechanics. The coupling could be two-way.
However, analysis with one-way coupling is more common. If a thermomechanical
process is simply analyzed using a one-way coupled field, then it is most likely that it
is about the changes in the mechanical field caused by heat transfer. These changes
could be either thermal stresses or thermal strains, depending on the constraints in
the mechanical field. This is because most thermomechanical processes of concern
are caused by the changes in the thermal field, e.g., heat flux, heat sources, and
uneven temperature distributions. Such excitations originating in the thermal field
will change the way that the molecules or atoms constituting the material interact
with each other and consequently the stresses and/or strains at the macroscopic scale.
Two-way couplings between thermal and mechanical fields are also possible but
much less common. This two-way coupling requires to additionally consider the fact
that mechanical energy can convert into heat via friction or plastic deformation. As
for friction, we may all have the experience that our palms become warmer when
being rubbed against each other. This can also occur in fluids as the viscosity of the
fluids could turn the kinetic energy into thermal energy. This process is usually
termed dissipation. As for plastic deformation, one typical example is that noticeable
heat can be generated by repeatedly bending and restoring a thin aluminum plate.
In this chapter, basic concepts in thermomechanics will be introduced first
according to the direction of the coupling. The mathematical descriptions of two
different types of couplings in thermomechanics will be laid out: thermal expansion/
stress as the coupling from the thermal field to the mechanical field and heat
generated by the mechanical processes as the coupling from the mechanical field
to the thermal field. After that, a key material property in the thermomechanical field,

© Springer International Publishing AG, part of Springer Nature 2018 209

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_19
210 19 Thermomechanics: Non-isothermal Mechanics

i.e., thermal expansion coefficient, will be introduced. A practice problem modified

from the two relevant monolithic physical fields, i.e., thermal and mechanical fields,
will be provided in the end.

Thermal Expansion and Thermal Stress

Thermal expansion is the tendency of a material to change its shape, area, and
volume in response to a change in the temperature. It is known from the statistical
mechanics that temperature is a macroscopic indicator of the average kinetic energy
of the molecules/atoms of a material (Fig. 19.1, left). The kinetic energy of these
particles increases as the temperature increases. Accordingly, the particles move
more dramatically and consequently tend to maintain a greater average separation.
As a result, the material will expand if there are no mechanical constraints (Fig. 19.1,
right). The magnitude of this expansion is usually proportional to the changes in the
temperature. This proportionality is a significant material property to characterize the
material’s ability to mechanically response to temperature changes: thermal expan-
sion coefficient. Though unusual, some materials may contract as the temperature
increases. However, such phenomena are usually limited to specific temperature
ranges due to other less common mechanisms.
In reality, a non-constrained mechanical field is usually an idealization as con-
straints almost always exist. Therefore, compressive (or tensile) stresses will develop
if the thermal expansion (or contraction) caused by temperature change is
constrained. This can be explained by treating stresses as a density of the energy
of the particles (Fig. 19.1, middle). In fact, the equivalence of the virial stress from
energy density (Marc and McMillan 1985; Swenson 1983; Tsai 1979) and the
Cauchy stress has been investigated using both theoretical arguments and numerical
simulations (Subramaniyan and Sun 2008). In the above thermal expansion example,
a temperature increase leads to an increase in the average kinetic energy of the
particles. However, if the volume does not change, then the energy density will
increase, leading to an increase in the Cauchy stress. In fact, depending on the
material type, the temperature change may result in changes in other parts of the

Fig. 19.1 Microscale mechanisms for thermal stresses and strains

Introduction 211

particle’s energy such as the bond energy and angle energy. The thermal stress is an
overall indication of the above changes in the energy density.
Therefore, the thermal stresses and thermal strains are two outcomes produced via
the same microscopic mechanisms. Both thermal stresses and thermal strains can
cause detrimental or beneficial results, depending on the purpose of the application.
Thermal stresses can significantly impact structure’s strength and stability and even
lead to cracks or breaks. Two typical examples are the residual stress in welding and
the longitudinal thermal cracking in the paved roads. Thermal strains are not desired
as they can lead to excessive deformation, which will compromise the serviceability
of a structure or a component. By contrast, the beneficial use of the thermal
expansion process has also led to many significant innovations in the history of
science and engineering such as mercury thermometers. Both thermal stresses and
strains can be well considered with extensions to the governing equation of the
mechanical field, which will be introduced in the following section.

Heat Generated by Mechanical Processes

Friction and plastic deformation are two predominant mechanisms in the conversion
of mechanical energy into thermal energy. In both cases, the work done on the
material(s) will be transferred to the kinetic energy of the molecules/atoms, which
becomes part of the internal energy and exhibits as a temperature increase.
When explaining this conversion process via friction, we usually need to differ-
entiate two different conditions: the friction between two solid objects or that within
a fluid object. As for the former, the surfaces between two solid objects are rough
like canyons at very small scales. During contacts, some atoms could be brought
very close together, and as a result, the kinetic energy of the macroscopic movement
will be stored as potential energy of the particles. This potential energy will then be
converted to the kinetic energy of the particles when the particles move apart. Real
friction also involves a lot of plastic deformation. In a fluid, the resistance of the
material to gradual deformations stems from the interparticle forces. When such
deformations such as shearing flow happen, the interparticle forces will transfer
energy from the motion of the fluid, i.e., the kinetic energy of fluid parcels, into the
kinetic energy of the molecules, which is part of the internal energy of the fluid. As
the result, the fluid is heated up. This process is partially irreversible and is thus
referred to as dissipation or viscous dissipation. Turbulent flows can significantly
expedite or even dominate this energy conversion process. Dissipation is high in
regions with large gradients, e.g., boundary layers and shear layers, and in regions
with very high turbulence levels, e.g., wakes.
The mechanisms associated with plasticity are complicated and vary from mate-
rial to material. The plasticity in metals is much different from that in amorphous
materials, and the plasticity in porous materials could be even more complicated.
However, despite those complicated and distinct intermediate steps and mechanisms
involved in the process, the energy conversion still can be simplified into the process
212 19 Thermomechanics: Non-isothermal Mechanics

described above: from external work to the kinetic energy of the material/object and
then to the kinetic energy of the atoms/molecules.
As indicated above, the heat generated in the mechanical processes is less
common than the thermal stresses and thermal strains in the context of multiphysics.
When moving to the porous material, applications with heat generated by a mechan-
ical process are even rarer. Friction between solid objects is the interest of multibody
mechanics instead of continuous mechanics, based on which multiphysics is built.
The heat caused by plasticity in porous materials is usually negligible. For fluids, the
viscous dissipation is an indispensable part of fluid dynamics. However, we mostly
focus on the energy loss rather than its conversion to heat and the consequent effect
on the heat transfer in the fluids. Due to these considerations, the mathematical
formulation of the energy conversion with viscous dissipation and plasticity will be
only briefly introduced in the following section.

Mathematical Formulation

Coupling from Thermal Field to Mechanical Field

Both thermal stresses and thermal strains can be considered as the sequence of the
coupling from the thermal field to the mechanical field. The coupling can be
formulated with an additional term to Navier’s equation. Let us first take a look on
how to formulate thermal expansion before we introduce how to add in this term.
The coefficient of thermal expansion quantifies a material’s ability to change its
size with a change in temperature. This ability can be defined according to the
dimensions of the deformation, i.e., 1D, 2D, and 3D. In 1D, such as a rod, the linear
coefficient of thermal expansion is defined as

1 dL
α¼ , ð19:1Þ
L dT
where L is the length and dL/dT is the rate of the change of the dimension L per a unit
change in temperature. Based on the definition, the change in the dimension can be
measured in terms of
ΔL
¼ αΔT, ð19:2Þ
L
where α is the linear thermal expansion coefficient. The word “linear” here stands
for 1D.
This equation works well as long as the linear expansion coefficient does not
change too much over the temperature change range ΔT and the fractional change in
length is small ΔL/L << 1. When moving to the small strain theory, the above
equation can be written as
Mathematical Formulation 213

ε ¼ α  dT: ð19:3Þ

In a 3D case with homogeneous and isotropic materials, we will also have strains
along the other two perpendicular directions. Therefore, the strains caused by the
temperature change can be expressed as follows using the matrix notation:
2 3
α  dT 0 0
εT ¼ 4 0 α  dT 0 5: ð19:4Þ
0 0 α  dT

There are only normal strain components because the temperature change only
causes changes along normal directions in the absence of mechanical constraints.
If the above object is completely constrained, the following stress increment will
be caused:
2 3
3Kα  dT 0 0
σT ¼ 4 0 3Kα  dT 0 5: ð19:5Þ
0 0 3Kα  dT

where K is the bulk modulus. Accordingly, we also only have normal stress
components. The negative sign implies compressive stresses.
In order to derive the governing equation, let us get back to the equation of
equilibrium:

∇  σ þ f ¼ 0: ð19:6Þ

The stress σ in the above equation is the elastic stress. However, the strain
obtained from the strain-displacement relationship is the total strain ε instead of
the elastic strain εe. Therefore, the above equation with the thermal strain term can be
rewritten considering
σ ¼ D : εe ¼ D : (ε  εT):

∇  ½D : ðε  εT Þ þ f ¼ 0: ð19:7Þ

Substituting the formulation of the stiffness matrix for linear elasticity into the
above equilibrium equation (Eq. 19.7), we can obtain the following non-isothermal
Navier’s equation by following the derivation procedure used in the chapter for the
mechanical field:

μ∇2 u þ ðμ þ λÞ∇ð∇  uÞ þ f  3αK∇½ðT  T 0 ÞI ¼ 0, ð19:8Þ

or equivalently
 
1 
∇  D : ∇u  ð∇uÞT  3αK ðT  T 0 ÞI þ f ¼ 0, ð19:9Þ
2

where D is the stiffness tensor (fourth-order tensor) and I is

214 19 Thermomechanics: Non-isothermal Mechanics

2 3
1 0 0
I ¼ 40 1 0 5: ð19:10Þ
0 0 1

Once the displacement is obtained by solving the above governing equation, the
stress will need to be calculated with the following equation:

1 
σ¼D: ∇u  ð∇uÞT  3αK ðT  T 0 ÞI: ð19:11Þ
2
Please note that the stress and displacement in the above equation are written
using the tensor notation instead of the Voigt notation.

Coupling from Mechanical Field to Thermal Field

The heat generated by the mechanical field serves as the major coupling from the
mechanical field to the thermal field. The governing equation of the thermal field,
i.e., heat equation, is first recalled to mathematically formulate this coupling:

∂ðρCT Þ
¼ ∇  ðλ∇T Þ  ∇  ðρCTvÞ þ g, ð19:12Þ
∂t
where ρ is the density, C is the gravimetric heat capacity, T is temperature, λ is the
thermal conductivity, v is the velocity of material particles, and g is the heat source.
For both friction and plasticity deformation, the heat source can be related to the
rate of the work done on the material point of interest:

dW
g¼η , ð19:13Þ
dt
where η is the ratio between the generated heat and the total mechanical energy input
or energy dissipation, W. For plasticity, the above total mechanical energy dissipa-
tion in the plastic deformation can be formulated as

W ¼ σ : εp , ð19:14Þ

where ε_ p is the plastic strain rate. The heat source in the thermal field then becomes

g ¼ ησ : ε_ p : ð19:15Þ

The value of η is usually low in friction. However, it could be much higher in the
plastic deformation. For example, experiments with an alloy made of Ti, Al, and Cu
showed that η could be as high as 70% during high strain rate plastic deformation
(Kapoor and Nemat-Nasser 1998). In a liquid, it is common to assume that most of
the energy dissipation goes to the internal energy of the liquid, especially the kinetic
Coefficient of Thermal Expansion 215

energy of the liquid molecules. Therefore, η is approximately 1. The heat source

attributed to the viscous energy dissipation in the fluid can be formulated as

g ¼ σ : ∇u_ , ð19:16Þ

where u_ is the velocity vector. For compressible Newtonian fluids, a detailed

formulation of the viscous dissipation is
2  2  2 3
∂u2 ∂v ∂w
2
6 ∂x þ 2 ∂y
þ 2 ∂z 7 2
g ¼ μ4  2  2  2 5 þ λð∇  uÞ , ð19:17Þ
∂v ∂u ∂w ∂v ∂u ∂w
þ ∂x þ ∂y þ ∂y þ ∂z þ ∂z þ ∂x

where μ and λ are the viscosities associated with the shearing deformation and
volume change, respectively, and u, v, and w are the components of the flow velocity
along three mutually perpendicular directions in the Cartesian coordinate system.

Coefficient of Thermal Expansion

Two other types of thermal expansion coefficients have been developed in addition
to the linear one: area and volumetric thermal expansion coefficients. The choice of
the coefficient type depends on the application and dimensions of interest. Mathe-
matical definitions of these coefficients are defined below for solids, liquids, and
gasses. Substances that expand at the same rate in every direction are called
isotropic. For isotropic materials, the area and volumetric thermal expansion coef-
ficients are, respectively, approximately twice and three times of the linear thermal
expansion coefficient.

Area Expansion Coefficient

The area thermal expansion coefficient is defined as the ratio between the fractional
change in the area and the change in the temperature:

1 dA
αA ¼ , ð19:18Þ
A dT
where A is the area of interest in the object and dA/dT is the rate of the area change per
unit change in the temperature. The fractional change in the area can be estimated as

ΔA
¼ αA ΔT: ð19:19Þ
A
216 19 Thermomechanics: Non-isothermal Mechanics

This equation is valid when the area expansion coefficient does not change
significantly over the range of ΔT and the fractional change in the area is small,
i.e., ΔA/A << 1.

Volume Expansion Coefficient

The volumetric thermal expansion coefficient is the most basic thermal expansion
coefficient and the one most relevant to fluids. In general, substances expand or
contract when their temperature changes, with expansion or contraction occurring in
all directions. The volume thermal expansion coefficient is defined as the ratio
between the fractional change in volume and the change in temperature:

1 dV
αV ¼ , ð19:20Þ
V dT
where V is the volume of the material and dV/dT is the rate of volume change with
the temperature. If we already know the expansion coefficient, then we can calculate
the change in volume as

ΔV
¼ αV ΔT, ð19:21Þ
V
where ΔV/V is the fractional change in volume and ΔT is the change in the
temperature.

Practice Problem

Problem Description

A 2D (plane strain) square object with a side length of 1 m has an initial temperature
of 273.15 K. The top boundary gets into contact with a water bath with a temperature
of 293.15 K, while all the other boundaries are connected to a water bath with a
temperature of 273.15 K. The material of the object has a thermal diffusivity of
1  107m2/s. The two sides of the object are fixed, while the top and bottom are
constrained using rollers. Young’s modulus is 10 GPa and Poisson’s ratio is 0.2. The
linear thermal expansion coefficient of the material is 1  1051/K. Please compute
σ y in the domain at t ¼ 10 (Figs. 19.2 and 19.3).

Results
Practice Problem 217

Fig. 19.2 Computational

domain and boundary
conditions

Fig. 19.3 Distribution of vertical stresses

218 19 Thermomechanics: Non-isothermal Mechanics

Appendix: FlexPDE Script File

{ Thermomechanical process in a plane strain case }

TITLE 'Thermomechanics'
VARIABLES
u=VECTOR(ux,uy)
Temp
DEFINITIONS
E=10e9
nu=0.2
mu=E/2/(1+nu)
lambda=E*nu/(1+nu)/(1-2*nu)
alpha=10e-6 C=1e7
delta=VECTOR(1,1)
INITIAL VALUES
u=VECTOR(0,0)
Temp=273.15
EQUATIONS
u: mu*DIV(GRAD(u))+(lambda+mu)*GRAD(DIV(u))-3*alpha*GRAD(T-273)=0
temp: C*dt(temp)=DIV(GRAD(temp))
BOUNDARIES
REGION 1
START(0,0)
NATURAL(ux)=0
VALUE(uy)=0
NATURAL(temp)=0
LINE TO (1,0)
VALUE(ux)=0
VALUE(uy)=0
NATURAL(temp)=0
LINE TO (1,1)
NATURAL(ux)=0
VALUE(uy)=1
VALUE (temp)=293.15
LINE TO (0,1)
VALUE(ux)=0
VALUE(uy)=0
NATURAL(temp)=0
LINE TO CLOSE
TIME 0 TO 10 by 0.01
MONITORS
PLOTS
FOR CYCLE=1
SURFACE(lambda*(dx(ux)+dy(uy))+2*mu*dx(uy)) ! sigma_y
SURFACE(uy)
END
Chapter 20
Hydromechanics: Poroelasticity
as a Simple Case

Introduction

Poroelasticity is traditionally used to describe the coupling between the solid and the
interpenetrating fluid in their mixture, in which the solid is believed to be saturated
by the fluid. In other words, it is an area of studying the movement/flow of material
points in both the liquid phase and the solid skeleton. This is distinct from both the
groundwater movement and mechanical/dynamic responses of the skeleton. For
example, the hydraulic field based on Darcy’s law does not take into account the
fact that the solid skeleton can be altered. A correction is needed even in the case of
steady flow of the pore water under the influence of external forces, let alone
problems such as the propagation of elastic vibrations in the porous materials, in
which the problems could become untenable (Frenkel 2005).
There are two major parts for poroelasticity: static (or quasi-static) and dynamic
theories. The static poroelasticity accounts for a process in which water movement
and solid skeleton deformation occur simultaneously and affect each other. There-
fore, it can be termed as hydromechanics according to the convention adopted in this
book. This static poroelasticity theory is a generalization of the consolidation theory
in soil mechanics. In fact, this static theory has been more extensively studied. As a
result, poroelasticity is frequently used interchangeably with the consolidation
theory or to refer to the static poroelasticity in a narrow sense (Detournay and
Cheng 1993; Verruijt 2013). The dynamic poroelasticity is developed for the wave
propagation in both the liquid and solid phases of saturated porous materials. So, it is
only about dynamics, but the monolithic field involves two different phases. The
dynamics here requires using the comprehensive momentum conservation equation
for each phase to describe the elastic waves in the representative elementary volume
as a mixture of the solid and liquid. It is actually an extension of the mechanical field
for mixtures by including the momentum of each phase.
According to Wang (2000), the term “poroelasticity” was first coined by
Geertsma as a footnote in his 1966 paper, entitled “Problems of Rock Mechanics

© Springer International Publishing AG, part of Springer Nature 2018 219

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_20
220 20 Hydromechanics: Poroelasticity as a Simple Case

in Petroleum Production Engineering.” In the footnote, Geertsma referred to Biot’s

work on “the theory of the elasticity and viscoelasticity of fluid-saturated porous
solids.” Geertsma also pointed out the analogy between poroelasticity and
thermoelasticity. The development of the static poroelasticity is primarily attributed
to the relevant advances in geomechanics, hydrogeology, and petroleum engineer-
ing, for practical problems involving the coupled fluid and mechanical behavior.
Typical problems include soil consolidation, water level change in wells due to tides
or external loading, oil fields that sink into the sea, and stress around a borehole
(Wang 2000). The dynamic poroelasticity is advanced in virtue of the needs from
ultrasonic and seismic applications in porous materials such as geological formations
(Carcione et al. 2010) and biomaterials (Cowin 1999).
The modern framework of poroelasticity was built on Biot’s work on static
poroelasticity (1941) and dynamic poroelasticity (1956a, b), though it could trace
back to earlier work such as the 1D soil consolidation theory of von Terzaghi (1925).
Later work usually focused on reformatting the equation system, validating relation-
ships or parameters with experiments, adding extra terms to account for special
effects, relating parameters with microscopic mechanics, and applications to more
materials and real problems (Gassmann 1951; Geertsma 1957; Biot and Willis 1957;
Detournay and Cheng 1993; De Boer 2000; Wang 2000; Rudnicki 2001; Coussy
2004; Verruijt 2013). The theory of linear poroelasticity (or consolidation) has now
reached a stage where there is a consensus on the basic equations. However, there is
still no general agreement on the definitions and the notations of the basic physical
parameters (Verruijt 2013). In this book, the notation used by Wang (2000) for static
poroelasticity and those adopted by van Dalen (2013) will be employed.
As mentioned, poroelasticity has gained a significant development attributed to
needs from different disciplines. Due to the same reason, it has widespread applica-
tions in various areas. Typical topics include the seepage of liquid waste disposed of
underground, oil and gas recovery, soil consolidation, and glacier dynamics (Penta
et al. 2014). In recent years, research on the transport, mass exchange, and solid
stress in biological tissues has also gained momentum. It is noted that both the static
and dynamic poroelasticity theories have been attracting increasing interest in bio-
materials (Moeendarbary et al. 2013; Penta et al. 2014). The historical development
and future directions of poroelasticity will continue to be interwoven with the
advances in both engineering approaches following the work of von Terzaghi and
the mixture theory following the work of Fillunger (de Boer 2000; Penta et al. 2014).
It is also necessary to mention that most of the work in poroelasticity, in fact, is
devoted to linear poroelasticity (Schanz and Diebels 2003). As commented by
Rudnicki (2001), although departures from linear behavior are common in real
cases, linear poroelasticity still presents significant improvements to overcome
over-idealizations in monolithic physics studies in which either the liquid or the
solid is neglected. Moreover, parameters describing the nonlinear material behavior
are usually difficult to obtain with accuracy. Therefore, linear poroelasticity often
offers a more practical approach than complex nonlinear models. In fact, the insights
gained from linear poroelasticity provide valuable assistance to interpret and build
more complex models.
Definitions of Phase Composition, Stress, and Strain 221

In this chapter, the definitions of the basic terms in poroelasticity, including phase
compositions, stresses, and strains, will be discussed first. Then static poroelasticity
and dynamic poroelasticity will be introduced consequently. A practice problem
derived from a classic poroelasticity problem for static poroelasticity will be
presented for the numerical implementation of the theory.

Definitions of Phase Composition, Stress, and Strain

The derivation of the mathematical equation system includes the definitions of phase
compositions, stresses, and strains, identification of constitutive relationships, estab-
lishment of governing equations, and acquisition of relationships between proposed
parameters and well-defined properties of porous materials. For the definitions of
phase compositions, stresses, and strains, there are usually two ways due to the
historical development of poroelasticity: the classic way in soil mechanics and the
way in the mixture theory. Most recent poroelasticity studies are formulated using
the mixture theory notation, possibly considering its rigorous rationality in physics
and potential to be extended to more complicated multiphase and multicomponent
systems. Here, we will start the derivation with the mixture theory and then explain
its relationship to the classic way in soil mechanics.

Phase Composition

The mixture theory relies heavily on the assumption of the REV and the concept of
volume fraction. Every point in a domain of interest is considered as a well-mixed
REV, within which any phase i takes a volume fraction of θi. In a saturated porous
material, there are only two phases, i.e., a solid (skeleton) and a liquid (water).
Because the sum of the volume fractions should be 1, in saturated porous materials,
only 1 parameter is needed to describe the volume fractions. We can use the porosity
which is widely used in most disciplines related to soils and rocks:

Vf
ϕ¼ , ð20:1Þ
V
where ϕ is the porosity, Vf is the fluid (water) volume, and V is the total volume.
Therefore, the volume fractions can be calculated using the porosity:

θ f ¼ ϕ, ð20:2Þ
θs ¼ 1  ϕ, ð20:3Þ

where θw and θf are the volume fractions (or volumetric water content) of the fluid
and solid, respectively.
222 20 Hydromechanics: Poroelasticity as a Simple Case

Stress

The total stress is defined for the REV, which is a pseudo bulk material obtained by
mixing the solid particle and pore liquid. As a result, the total stress is the sum of the
solid stress and fluid stress. Fluid instead of water is used here for a more general
framework. To obtain the total stress, let us first obtain the solid and fluid stresses.
The fluid stress is the stress caused by the pore fluid pressure, which only has normal
components but no shear components:
2 3
ϕp 0 0
σ f ¼ ϕpI ¼ 4 0 ϕp 0 5, ð20:4Þ
0 0 ϕp

where I means only the normal stresses are non-zero. The pore fluid pressure is a
compressive stress, whereas a stress is positive when it is a tensile stress. Therefore,
there is a negative sign. As can be seen, the pore pressure in the fluid volume fraction
of ϕ is averaged with respect to the whole domain to obtain the fluid stress. The solid
stress has two parts: the intergranular interaction and the pressure coming from the
liquid. Therefore, it is equal to

σs ¼ σ∗ þ ð1  ϕÞðpÞI, ð20:5Þ

where σ∗ is the intergranular stress and can be defined as

2 3
σ 11 σ 12 σ 13
σ∗ ¼ 4 σ 22 σ 22 σ 23 5: ð20:6Þ
σ 31 σ 32 σ 33

The total stress σt is then obtained as

2 3
σ 11  p σ 12 σ 13
σt ¼ σ f þ σs ¼ 4 σ 22 σ 22  p σ 23 5, ð20:7Þ
σ 31 σ 32 σ 33  p

that is,

σt ¼ σ∗ þ p: ð20:8Þ

The above equation is very similar to the principle of effective stress in the binary
system (Fig. 20.1) as follows:
0
σt ¼ σ þ p: ð20:9Þ

In fact, the effective stress σ' is assumed to be approximately equal to σ∗ in

traditional soil mechanics. However, it has been shown by both theoretical and
experimental studies that a more accurate definition is
Definitions of Phase Composition, Stress, and Strain 223

Fig. 20.1 Stresses in binary (two-phase) porous materials

0 K 0
σt ¼ σ þ p  p ¼ σ þ αp, ð20:10Þ
Ks
where α is the Biot-Willis coefficient. Other formulations which include volume
fractions are also available. In soft soils, the bulk modulus of the solid-fluid mixture,
K, is much smaller than that of the solid grains Ks, so α is close to 1. That is why the
stress principle is applicable to most geotechnical applications involving soft soils.
But in stiff soils or rocks, the effective stress may deviate from the value calculated
with the effective stress principle significantly. In later constitutive relationships, σ
will be used to represent an isotropic stress field applied to the mixture, which is
equal to the average of the normal terms in the total stress:

1 1
σ ¼ traceðσt Þ ¼ σ ii , ð20:11Þ
3 3
where trace(σt) is the trace of the total stress tensor. The above mean normal stress or
hydrostatic stress is used because the stress interaction between the solid and fluid
phases resides mostly between the hydrostatic stress terms (i.e., pressure) rather than
deviatoric stress terms.

Strain and Displacement

For strains, we will define the strain of the solid, ε; the strain of the fluid, e; and the
increment of the fluid content, ζ. The solid constitutes the skeleton of the porous
material; therefore the solid deformation reflects what we observe on the solid
skeleton. The volume of the solid is assumed to be evenly distributed across the
whole volume. The fluid strain is the strain of a pseudo liquid which is evenly
distributed over the whole REV. This definition of the fluid strain is similar to
Darcy’s velocity, which is defined over the whole sectional area including both the
liquid and solid, instead of the true velocity (analogous to the true fluid deformation),
which is defined over the liquid sectional area (real flow passageway). If we define
the displacements of liquid and solid on these pseudo phases, then the solid and fluid
strains can be related to the displacement as
224 20 Hydromechanics: Poroelasticity as a Simple Case

1h i
ε¼ ∇u þ ð∇uÞT , ð20:12Þ
2
and

1h i
e¼ ∇U þ ð∇UÞT , ð20:13Þ
2
where u and U are the displacements of the solid and liquid, respectively.
That is to say, we treat a saturated porous material, which is a two-phase
composite, as a continuum. Each REV is a well-mixed region which is large enough
compared to the solid grains and interconnected pores but small enough compared to
the domain of interest. The increment of fluid content, ζ, can be accurately defined as
(Wang 2000)

δm f
ζ¼ , ð20:14Þ
ρ f0

where δmf is the change in fluid mass content and ρf0 is the fluid density in the
reference state. Or it can be equivalently defined as

δV p  δV f
ζ¼ , ð20:15Þ
V
where Vp is the pore volume. This increment can be related to the deformations of the
solid and liquid via (Biot and Willis 1957)

ζ ¼ ε jj  eii ¼ ∇  u  ∇  U, ð20:16Þ

where e and ε are the scalar terms of the liquid and solid strains, respectively. In
many cases, the volume strain of the solid skeleton is represented using another
symbol:

εv ¼ εii : ð20:17Þ

The theories of static and dynamic poroelasticity usually adopt different forms of
constitutive relationships. Therefore, the constitutive relationships will be intro-
duced in the following sections for static and dynamic poroelasticity.

Static Poroelasticity

Constitutive Relationships

The establishment of the constitutive relationships for saturated porous materials

needs to consider both the solid and fluid. It is not difficult to imagine that these two
Static Poroelasticity 225

phases will interact with each other in the stress-strain relationship. Since the fluid
stress/strain only has normal components, the couplings between the solid and fluid
only occur via these normal terms. As a result, we usually identify the couplings
between normal and bulk/volume stress/strain first and then add in the shear terms in
the formulation. The following constitutive relationship can be obtained:
σ p
εv ¼ þ , ð20:18Þ
K H
σ p
ζ¼ þ , ð20:19Þ
H R
where K is the bulk modulus, H is the poroelastic expansion coefficient, and R is the
reciprocal of the unconstrained storage coefficient. Their physical meanings of these
parameters can be described using the following equations:

δσ 
K¼ p¼0 , ð20:20Þ
δεv
δp δσ 
H¼ jσ¼0 ¼ p¼0 , ð20:21Þ
δεv δζ
δζ
R¼ jσ¼0 : ð20:22Þ
δp
The above constitutive relationships are sometimes written in the following form:

σ ¼ Kεv  αp, ð20:23Þ

1
ζ ¼ αεv þ p, ð20:24Þ
M
where α ¼ HK is the Biot-Willis coefficient introduced above and M1 ¼ R1  HK2 is the
constrained specific storage coefficient. It is seen that, in either form of the consti-
tutive relationship, only three independent mechanical properties are needed. In
order to use the above relationships to derive the governing equation, the shear
stress/strain needs to be included. The complete constitutive relationship for the
static poroelasticity is

σ ij ¼ λεkk δij þ 2Gεij  αpδij ¼ λ  traceðεÞ þ 2Gε  αpI, ð20:25Þ

1 1
ζ ¼ αui, i þ p ¼ α∇  u þ p: ð20:26Þ
M M
226 20 Hydromechanics: Poroelasticity as a Simple Case

Governing Equation

Then let us recall the Navier’s equation for the momentum equilibrium of the solid
and continuity equation of the pore fluid:

ðλ þ μÞ∇ð∇  uÞ þ μ∇2 u þ g ¼ 0, ð20:27Þ

 
1 ∂p k
þ ∇   ∇ðp þ ρgzÞ ¼ 0, ð20:28Þ
M ∂t η

where k is the permeability.

Substituting the constitutive equations (Eqs. 20.25 and 20.26) into the governing
equations (Eqs. 20.27 and 20.28), we obtain the final governing equations:

ðλ þ μÞ∇ð∇  uÞ þ μ∇2 u þ f ¼ α∇p, ð20:29Þ

 
1 ∂p k ∂
þ ∇   ∇ðp þ zÞ ¼ α ð∇  uÞ: ð20:30Þ
M ∂t η ∂t

In poroelasticity, we may also see some other material properties, which can be
formulated using K, H, R, and so on. These include the undrained modulus

δσ  1
Ku ¼ ζ¼0 ¼ 1 , ð20:31Þ
δεv K  R
H2

undrained Poisson’s ratio

3ν þ αBð1  2νÞ
νu ¼ , ð20:32Þ
3  αBð1  2νÞ

and first Skempton’s coefficient

δp  R
B¼ ζ¼0 ¼  ¼ αM=K u : ð20:33Þ
δσ H
Further micromechanical relationships are sometimes needed to relate the mate-
rial properties of the porous material to its phases (Merxhani 2016):

K
α¼1 ðBiot and Willis 1957Þ ð20:34Þ
Ks
 
1 1 α K 1 1
¼  ¼ 1 ϕ þ ϕ ðVerruijt 1969Þ ð20:35Þ
M R H Ks Ks Kf
Dynamic Poroelasticity 227

This will leads to

1 1 1
¼  , ð20:36Þ
H K Ks
 
1 1 1 1 1
¼ þϕ   , ð20:37Þ
R K K f Ks Ks

where Ks and Kf are the bulk moduli of the solid and fluid phases, respectively. In
many documents, compressibilities (reciprocals of bulk moduli) instead of bulk
moduli are used in the above equations.

Dynamic Poroelasticity

Constitutive Relationships

In the previous section for static poroelasticity, the inertial and associated kinetic
energy was not considered. However, when the speed of the movement of the phases
in the porous material is appreciable, e.g., when there is vibration or stress waves, the
theory in the previous section cannot well describe the behavior of the porous
materials. To include these dynamic effects, constitutive relationships are slightly
modified from those adopted in the previous section:

σs ¼ Aεv δ þ Qe þ 2Gε ð20:38Þ

p ¼ Qεv δ þ Re ð20:39Þ

1
E p ¼ ðσs  ε þ p  eÞ, ð20:40Þ
2
where G, A, Q, and R are another four parameters quantifying the elastic properties
of the porous material. They are not conventional moduli but can be obtained using
the moduli of the phases and the bulk mixture.

Governing Equation

In order to derive the governing equation, it is better to resort to the Lagrangian

equation. The Lagrangian mechanics is just another description of the classic
mechanics which works better for complicated systems involving a lot of con-
straints. The Lagrange equation is
228 20 Hydromechanics: Poroelasticity as a Simple Case

d ∂ ∂
ðE k  E d Þ  ðEk  E d Þ ¼ F, ð20:41Þ
dt ∂x_ ∂x
where x is the generalized coordinate(s), Ek is the kinetic energy, Ed is the energy
dissipation, and F is the total force. The kinetic energy is

1 
Ek ¼ ρ11 u_  u_ þ ρ11 U_  U_ þ 2ρ12 u_  U_ , ð20:42Þ
2
where ρ11, ρ22, and ρ12 are the effective density of the solid, liquid, and coupling
between the solid and the liquid, respectively. These density terms can be related to
the densities of the solid and liquid via the following equations (Biot 1956a, b):

ρ11 ¼ ð1  ϕÞρs  ρ12 , ð20:43Þ

ρ22 ¼ ϕρ f  ρ12 , ð20:44Þ

ρ12 ¼ ðτ  1Þϕρ f : ð20:45Þ

The energy dissipation function can be expressed in terms of the velocities of the
solid and fluid:

1    
E d ¼ b0 u_  U_  u_  U_ , ð20:46Þ
2
where the coefficient b0 can be related to the permeability, dynamic viscosity, and
porosity via the following equation:

ηϕ2
b0 ¼ : ð20:47Þ
k
Substituting the above constitutive relationships and energy functions into the
Lagrange equation, we can obtain the following governing equations:
 
€ þ b0 u_  U_ ¼ ðA þ GÞ∇ð∇  uÞ þ G∇2 u
€ þ ρ12 U
ρ11 u
ð20:48Þ
þ Q∇ð∇  UÞ þ Fs  Q∇θs
 
ρ12 u €  b0 u_  U_ ¼ Q∇ð∇  uÞ þ R∇2 U þ F f  R∇θ f :
€ þ ρ22 U ð20:49Þ

Practice Problem

Problem Description

In linear poroelasticity, Mandel’s problem (Mandel 1953; Abousleiman et al. 1996)

is usually used to validate linear poroelastic models. Mandel (1953) presented an
Practice Problem 229

Fig. 20.2 Computational

domain and boundary
conditions

analytical solution to predict the coupled hydromechanical response of porous

materials under a set of prescribed boundary conditions (Ferronato et al. 2010). In
the Mandel’s problem, a 2D rectangular object (plane strain) is subjected to a vertical
distributed load of p0, while the two side boundaries are free to move. The top and
bottom boundaries are hydraulically insulated, while the side boundaries are exposed
in air (zero pore water pressure). Due to the symmetry in the geometry and loading,
we can only analyze one fourth of the object. Accordingly, a 2D domain that has
dimensions of 1 m by 1 m (ab) is adopted here (Fig. 20.2). Let us consider the
upper right fourth of the 2D object. Therefore, the left and bottom boundaries will be
symmetric. That is, there is no normal displacement and no shear stresses for the
mechanical field, while no flux is allowed through the boundaries (hydraulic
insulation).
The analytical solution for such a problem is
1 
X  2 
sin αn αn x αn ct
p ¼ 2p0 cos  cos αn exp , ð20:50Þ
n¼1
αn  sin αn cos αn a a2

where αn is the nth positive solution to the following equation:

λ þ 2μ
tan α ¼ α, ð20:51Þ
μ
and c is the consolidation coefficient and can be calculated using poroelastic and
other material properties as
κ
c¼ , ð20:52Þ
α
ρw g þ λþ2μ
1 2
M

in which κ is the hydraulic conductivity.

230 20 Hydromechanics: Poroelasticity as a Simple Case

Fig. 20.3 Analytical results

for the pore pressure
variation

In our example, the two Lamé constants are both 40 MPa, Poisson’s ratio is
0.3, the Biot-Willis coefficient is 1.0, the porosity is 0.375, the hydraulic conduc-
tivity is 1  105 m/s, and the compressibilities of the pore water and solid grain are
4.4 1010 1/Pa and 2.81010 1/Pa, respectively. Please calculate the pore water
distributions at 0.05 s, 0.5 s, 5 s, and 10 s and compare them against the analytical
solution.

Results

The analytical solution is shown in Fig. 20.3. MATLAB code for the implementation
of the solution is attached in the next subsection.
The above figure clearly shows the peculiar non-monotonic pressure
behavior, i.e., the so-called Mandel-Cryer effect, in which p rises above the initial
p0 at small time values. This non-monotonic pressure change occurs because the slab
contraction at the drained faces induces an additional buildup of pressure in the
interior.
The following figures show the numerical solution. As can be seen, the numerical
solution reproduces a similar process. However, the default iteration and error
control offered in the student version of FlexPDE failed to handle the high
nonlinearity residing in the beginning of the process. The Mandel-Cryer effect was
not perfectly simulated (Fig. 20.4).
Practice Problem 231

Fig. 20.4 Numerical results for pore pressure variation

232 20 Hydromechanics: Poroelasticity as a Simple Case

Fig. 20.4 (continued)

Practice Problem 233

Appendix: FlexPDE Script File

Analytical Solution

MATLAB script file 1: main.m

%Analytical solution to Mandel's problem
clear;

mu=40e6; %E/2/(1+nu)
lambda=40e6; %E*nu/(1+nu)/(1-2*nu)
alpha=1;
kappa=1e-5;
phi=0.375;
beta=4.8e-10;
c_s=2.8e-10; % Compressibility of solid grain
M=1/(phi*beta+(alpha-phi)*c_s);
p0=1e4;

K=lambda+2/3*mu;
R=1/(1/M+alpha^2/K);
H=K/alpha;
Ku=1/(1/K-R/H^2);
B=alpha*M/Ku;
nu=lambda/2/(lambda+mu);
nu_u=(3*nu+alpha*B*(1-2*nu))/(3-alpha*B*(1-2*nu));
c=kappa/(9800*(M^(-1)+alpha^2/(lambda+2*mu)));

a=1;

time=[0.05,0.500,50,10];
x=0:0.02:1;
n_time=0;
p=zeros(length(time),length(x));
for t=time
n_time=n_time+1;
for n=1:100
fun=@f;
alpha_n=fzero(fun,[0+(n-1)*pi+0.1 pi/2+(n-1)*pi]);
%We can also use vpasolve to calculate alpha_n, but it is much
slower!
%{ syms xxx
% alpha_n=vpasolve(-abs(tan(xxx))+(lambda+2*mu)*xxx/mu==0,xxx,
[0+(n-1)*pi+0.1 pi/2+(n-1)*pi]);
%}
%p(x,z,t), x=0~1, z=1 t=0.5
p_n=2*p0*(lambda+2*mu)*cos(alpha_n)/(mu-(lambda+2*mu)*(cos
(alpha_n))^2)*(cos(alpha_n*x/a)-cos(alpha_n))*exp
(-alpha_n^2*c*t/a^2); %Original
234 20 Hydromechanics: Poroelasticity as a Simple Case

%p_n=2*p0*sin(alpha_n)/(alpha_n-sin(alpha_n)*cos(alpha_n))*
(cos(alpha_n*x/a)-cos(alpha_n))*exp(-alpha_n^2*c*t/a^2); %
Equation in Ferronato et al. (2010)
p(n_time,:)=p(n_time,:)+p_n;
end
end

plot(x,p(:,:)/p0);

MATLAB script file 2: f.m

function y=f(x)
mu=40e6; %E/2/(1+nu)
lambda=40e6; %E*nu/(1+nu)/(1-2*nu)
y=-abs(tan(x))+(lambda+2*mu)*x/mu

Numerical Solution

{ Basic linear elastic mechanical problems for 2D (plane strain)

homogenous cases }
TITLE '2D Plane Strain'
SELECT
!PREFER_STABILITY
ERRLIM 0.01
REMATRIX=ON
NEWTON=3
VARIABLES
u=VECTOR(ux,uy)
p
DEFINITIONS
mu=40e6
lambda=40e6
alpha=1
kappa=1e-5
phi=0.375
beta=4.8e-10 ! Compressibility of fluid
c_s=2.8e-10 ! Compressibility of solid grain
M=1/(phi*beta+(alpha-phi)*c_s)
p0=1e4
INITIAL VALUES
ux=0
uy=0
p=p0
EQUATIONS
u: mu*DIV(GRAD(u))+(lambda+mu)*GRAD(DIV(u))=alpha*GRAD(p)
p: dt(1/M*p+alpha*DIV(u))+DIV(-kappa/9800*GRAD(p))=0
BOUNDARIES
REGION 1
START(0,0)
NATURAL(ux)=0
VALUE(uy)=0
NATURAL(p)=0
Practice Problem 235

LINE TO (1,0)
NATURAL(uy)=0
NATURAL(ux)=0
VALUE(p)=0
LINE TO (1,1)
NATURAL(ux)=0
NATURAL(uy)=-p0
NATURAL(p)=0
LINE TO (0,1)
VALUE(ux)=0
NATURAL(uy)=0
NATURAL(p)=0
LINE TO CLOSE
TIME 0 TO 10
MONITORS
FOR CYCLE=1
ELEVATION(p/p0) FROM (0,0) TO (1,0)
VECTOR(u)
VECTOR(-GRAD(p))
CONTOUR(p/p0)
PLOTS
FOR t=0.05
ELEVATION(p/p0) FROM (0,1) TO (1,1)
FOR t=0.5
ELEVATION(p/p0) FROM (0,1) TO (1,1)
FOR t=5
ELEVATION(p/p0) FROM (0,1) TO (1,1)
FOR t=10
ELEVATION(p/p0) FROM (0,1) TO (1,1)
HISTORY(p/p0) at (0,0.05)
END
Chapter 21
Thermohydromechanics

Introduction

In porous materials, thermohydromechanics is one of the most common types of

multiphysical phenomena in addition to the static poroelasticity discussed in the
previous chapter. Thermohydromechanics is responsible for many applications
related to unsaturated porous materials such as hydro-diffusion and subsidence,
drying/hydration and shrinkage, frost heave, spalling, geothermal applications, and
hydraulic cracking. This is because, in nature or in labs, most processes occur in
non-isothermal conditions. Therefore, the existence of heat transfer is more a rule
than an exception. Also, theoretically, the water flow and the accumulation or
dissipation of strain energy can alter or trigger a heat transfer process. For the
water movement, any type of potential difference within porous materials, e.g.,
matric, osmotic, and pneumatic, will give rise to water movement. These differences
can result from a temperature difference or form a change in the mechanical field.
For the mechanical field, either thermal transfer or water movement can explicitly
change the mechanical properties or the mechanical balance. This process could also
occur in the opposite direction, that is, changes in the structure or the existence/
changes of stress and strain can also influence the heat transfer and water movement.
In summary, the couplings are strong in the abovementioned phenomena, and these
strong couplings make the studies of thermohydromechanics (THM) a very popular
topic in the multiphysics in porous materials.
The development of the modeling techniques for the THM benefited from a lot of
studies for geomaterials. For example, there are a great number of papers dedicated
to the coupled thermohydromechanical model or thermohydromechanochemical
(THMC) model for rocks and soils (Thomas and King 1991; Thomas and He
1995, 1997; Thomas et al. 2009; Sahimi 2011; Noorishad et al. 1992; Stephanson
et al. 1997; Rutqvist et al. 2001). Noteworthily, these models were developed either
from the non-isothermal consolidation theory of deformable porous materials or the
extension to Biot’s phenomenological model. More recent models include the

© Springer International Publishing AG, part of Springer Nature 2018 237

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_21
238 21 Thermohydromechanics

phase change of water (i.e., freezing/thawing phenomena) in porous media. These

THM models are usually solved numerically, because an analytical solution is
usually unavailable. High nonlinearity is common because of the complicated
couplings between the involved monolithic physical fields.
Thermohydromechanics is very closely related to thermohydraulics and hydrome-
chanics. In many studies, thermohydromechanics is extended from thermohydraulics
or hydromechanics. For example, thermohydraulics models are widely adopted in soil
sciences to predict water movement under varying ambient temperatures. In this case,
moisture content is the major interest, while the temperature changes are the cause,
though the influence of the water movement on the temperature variation could also be
appreciable. “Thermal” is thus placed in front of “hydraulic.” While in disciplines that
more focus on stresses and strains such as geotechnical engineering or geophysics, the
THM models are also obtained by adding the thermal field into existing hydrome-
chanical models, i.e., poroelasticity theory.
In this chapter, the establishment of a typical THM model with the phase change
of the pore liquid will be presented for unsaturated porous materials (Liu and Yu
2011). We will show the establishment of the model using a bottom-up approach in
contrast to the top-down approach using in the previous chapter for poroelasticity.
Accordingly, the governing equations and auxiliary relationships for the thermal
field, hydraulic field, and mechanical field will be introduced sequentially. Boundary
conditions will be briefly mentioned. Finally, a practice problem for the use of a
THM model for predicting the behavior of a soil in a non-isothermal environment
will be presented.

Thermal Field

The thermal field is usually the main cause of the whole thermohydromechanical
process. For example, during a frost heave process, the subfreezing temperatures
cause the advancement of the frost front, which in return induces fluid migration and
soil deformation. This process generated by energy dissipation within a multiphase
material can be alternatively explained as the result of the varied surface tension, soil
moisture suction, kinetic energy changes associated with the hydrogen bond distri-
bution, and thermally induced osmotic gradients (Cary 1965, 1966). Besides the
temperature-induced moisture flux or deformation, the energy carried by the migra-
tory fluid and the heat parameters influenced by the fluid transport and metamorphic
solid skeleton are also assumed to be “reactions” to the thermal field. To precisely
formulate the energy transport in porous materials, a modified Fourier’s equation
with both conduction and convection terms can be adopted:

∂T
Ca þ Cw ∇ðJT Þ ¼ ∇  ðλ∇T Þ, ð21:1Þ
∂t
Hydraulic Field 239

where Cw is the heat capacity of unfrozen water, Ca is the apparent heat capacity, λ is
the thermal conductivity, T is the temperature, t is time, and J is the water flux from
the hydraulic field. Both Ca and λ are coupling variables. The moisture migration
changes the soil composition and consequently Ca and λ, which in turn affects the
heat transfer process.
The effects of the air phase and radiation are usually neglected as the phase
transition of water dominates the process of energy conversion. The apparent
volumetric heat capacity Ca in Eq. 21.1 takes into account the energy released/
absorbed by the phase change of the pore water. Instead of being treated as an energy
sink or source on the right-hand side of Fourier’s equation, it is a common practice in
partially frozen porous materials to incorporate the enthalpy change due to the phase
change into the heat capacity to reduce the nonlinearity (Anderson et al. 1973)

dθi
C a ¼ C s θs þ C w θw þ C i θi þ Cv ðn  θw  θi Þ þ L f , ð21:2Þ
dt
where θs, θw, and θi denote the volumetric content of soil mass, unfrozen water, and
ice, respectively (the volume change of soil skeleton is neglected here) and Lf is the
latent heat. The same convention on subscripts applies to other parameters.
The thermal conductivity λ in Eq. 21.1 can be approximated by empirical
relationships such as Eq. 21.3 (McInnes 1981; Hansson et al. 2004):
n o
λ ¼ C1 þ C2 ðθw þ θi F Þ  ðC1 -C4 Þ  exp ½C3 ðθw þ θi Þ þ Fθi C5 ð21:3Þ

F ¼ 1 þ F1 θ i F 2 , ð21:4Þ

where C1, C2, C3, C4, C5, F1, and F2 are empirical curve fitting constants. There are
also other similar relationships such as the equations proposed by Gardner (1958),
Sawada (1977), and Campbell (1985).

Hydraulic Field

For variably unsaturated porous materials, the fluid movement is generally described
by mixed-type Richards’ equation, which has been proven to perform well in
ensuring mass conservation (Celia and Binning 1992). To extend Richards’ equation
to allow for freezing/thawing, a term related to ice formation needs to be added to the
left-hand side of Richards’ equation to obtain

∂θw ρi ∂θi
þ ¼ ∇  ðκLh ∇h þ κ Lh i þ κLT ∇T Þ, ð21:5Þ
∂t ρw ∂t

where θw is the volumetric content of water, θi is the volumetric content of ice, ρw is

the density of water, ρi is the density of ice, κLh is the hydraulic conductivity, κLT is
240 21 Thermohydromechanics

the hydraulic conductivity due to a thermal gradient, i is the unit vector along the
direction of gravity, and h is the matric potential head (or pressure head). The matric
potential head is the equivalent water head (unit: m) of the matric potential (negative
water suction, unit: Pa). It is common to formulate the above equation in terms of the
matric potential head, matric potential, total soil potential head, or total soil potential.
But attention is needed to make sure the units of other terms are consistent with that
of the selected dependent variable.
Based on the analogy between the freezing process and the drying process, it has
been accepted that the concept of the soil-water characteristic curve (SWCC) can be
extended to describe the relationship between the unfrozen water content and the
matric potential (Koopmans and Miller 1966; Spans and Baker 1996). The hydraulic
conductivity can be estimated by integrating the SWCC (Childs and Colis-George
1950; Mualem 1986; Fredlund et al. 1994). Most THM studies employed either
simplified van Genuchten’s equation (van Genuchten 1980, Hansson et al. 2004)
(Eq. 21.6) or Fredlund and Xing’s equation (Fredlund and Xing 1994) to mathe-
matically describe the SWCC. Corresponding approaches are then adopted to predict
the hydraulic conductivity with the SWCC. Taking van Genuchten’s equation, for
example, we have the following equation for the SWCC:

θ  θr m
Se ¼ ¼ ð1 þ jαhjn Þ , ð21:6Þ
θs  θr
where Se is the effective saturation; θs and θr are the saturated and residual water
content, respectively; and α, m, and n are empirical parameters.
κLh and κLT are hydraulic conductivities related to the pore water head and temper-
ature, respectively. One set of widely accepted relationships for these parameters is
h   i
1 m 2
κ Lh ¼ κs Se 1  1  Se m , ð21:7Þ
 
1 dγ
κ LT ¼ κLh hGwT , ð21:8Þ
γ 0 dT

where γ denotes the surface tension of soil-water, which is temperature-dependent

and can be approximated using the following equation:

γ ¼ 75:6  0:1425T  2:38  104 T 2 , ð21:9Þ

where γ 0 is the value of γ at 25 ∘C, i.e., γ 0 ¼ 71.89  103 kg  m3.

As shown in Eq. 21.7, the hydraulic conductivity of a partially saturated or
partially frozen soil, κLh, is obtained by multiplying the saturated conductivity
with a saturation-dependent “relative conductivity” term. The thermal-induced
hydraulic conductivity κLT in Eq. 21.8 is developed from thermodynamics (Philip
and de Vries 1957). GwT is a gain factor, which has a value around 7 for coarse-
grained soils (Noborio et al. 1996). The dependence of the viscosity on temperature
Mechanical Field 241

can be neglected here to unify the equations for the intrinsic hydraulic conductivity
and the hydraulic conductivity.
The role of ice as an impedance for fluid migration was first proposed by Harlan
(1973) while studying the unsaturated hydraulic conductivity in partially frozen
soils. This viewpoint was confirmed by subsequent researchers such as Jame and
Norum (1980) and Hansson et al. (2004). An impedance factor was adopted to
describe the effects of ice on the fluid migration. However, a few more recent
publications reported that the impedance factor is unnecessary as long as the
SWCC is precisely determined (Newman and Wilson 1997; Watanabe and Flury
2008).
When phase changes are involved, the generalized Clapeyron equation needs to
be employed to describe the thermodynamic condition for the coexistence of water
and ice. The local freezing point of the pore fluid can be obtained from the
generalized Clapeyron equation:

dh L f
¼ , ð21:10Þ
dT gT

where h is the water head, Lf is the latent heat of water, and g is gravitational
acceleration.
Assuming thermodynamic equilibrium conditions are maintained at the ice-pore
water interface within infinitesimal time intervals, the Clapeyron equation will deter-
mine the liquid water content via the phase composition curves (Liu and Yu 2013),
while the frozen water content can be determined by the governing equation. There-
fore, a change to the governing equation is needed at the highest freezing temperature
of pore water. When such a numerical treatment is not feasible, the following
approximation can be employed to determine the ice content via Eq. 21.11:

dθi L f dθ
¼ : ð21:11Þ
dT gT dT

Mechanical Field

The governing equation for the mechanical field is Navier’s equation, which, as
introduced in the mechanical field, incorporates the equation of motion, the strain-
displacement correlation, and the constitutive relationship. The equation equilibrium
is introduced in a general tensor format as

∇  σ þ F ¼ 0, ð21:12Þ

where u is the displacement vector and F is the body force vector.

The strain-displacement equation is
242 21 Thermohydromechanics

1 
ε¼ ∇u þ ð∇uÞT : ð21:13Þ
2

The constitutive equation is

σ ¼ C:ε, ð21:14Þ

where σ is the Cauchy stress tensor, C is the fourth-order tensor of material stiffness,
ε is the infinitesimal strain tensor, and the symbol “:” stands for double contraction.
In order to consider the influence of the thermal field and the hydraulic field on the
mechanical field, the constitutive relationship for porous materials is formulated as

σ ¼ Dεel þ σ0 , ð21:15Þ

where D is the stiffness matrix of soil skeleton, σ0 is the initial stress vector, and εel is
the elastic strain which can be obtained from the following relationship:

ε ¼ εel þ εth þ εtr þ εhp þ ε0 , ð21:16Þ

where εth is the strain caused by thermal expansion, e.g., [α(T  Tref), α(T  Tref), 0]T
in 2D; εtr is the strain caused by the phase change of water, which was approximated
as [0.09Q, 0.09Q, 0]T when a unit localization tensor in mixture theory is followed,
where Q is the degree of water phase transition and 0.09 is the relative change of
volume when water turns into ice; ε0 is the initial strain; and εhp is the strain resulting
from the change in the matric potential, which is calculated using [h/H, h/H, 0]T. H is
the poroelastic expansion coefficient in unsaturated soils. The value of H can be
obtained through experimental measurement. The use of H casts light on the
independent role of matric potential in the constitutive relationship of unsaturated
porous media as indicated in Biot’s model for an unsaturated fluid with air bubble
and in Fredlund’s method to address volume change of unsaturated soil (Biot 1941;
Fredlund and Rahardjo 1993). The exact formulation of the stress or strain terms to
couple different phases in unsaturated porous materials or materials with even more
fluid phases is still under debate. The formulations given above are just simple
examples to illustrate the coupling of the hydraulic field with the phase change of the
pore liquid to the mechanical field.

General Boundary Condition

The treatment of the boundary conditions in the thermohydromechanical field is not

different from the ensemble of the treatments of the boundary conditions of the
individual monolithic physical fields. However, in a multiphysical field involving
several physical fields, it would be beneficial to treat all the boundary conditions in a
general way. The multiphysics simulation package COMSOL provides a very
convenient way for such a purpose.
Practice Problem 243

The general boundary condition, which includes the special cases such as the
Dirichlet (first-type), Neumann (second-type), and Robin (third-type) boundary
conditions, is formulated by Eq. 21.17:

n  ðc∇u þ ςu  γ Þ þ qu ¼ δ  hT μ, ð21:17Þ

where n is the outward normal unit vector of a boundary, u is the dependent variable
of the field under consideration (temperature, matric potential, displacements, etc.),
c is a conductivity term, ς is the conservative flux convection coefficient, γ is the
source in the subdomain, q is the boundary absorption coefficient, δ is the boundary
source, hT is a matrix designated for the flexibility of the constraint type, and μ is the
matrix of Lagrange multiplier.

Practice Problem
Problem Description

A cylindrical soil sample with a diameter of 0.06 m and a height of 0.26 m wrapped by
thermal and hydraulic insulation materials on its side. The top and bottom are connected
to solid objects at temperatures of 293.15 K and 343.15 K, respectively (Fig. 21.1).
The soil has a thermal diffusivity of 1  107m2/s; a saturated hydraulic
conductivity of 3.2  106m/s; a residual and saturated water content of 0.05 and
0.535, respectively; a gain factor for thermally induced water flux of 5; Young’s
modulus of 0.1 MPa; Poisson’s ratio of 0.3; a poroelastic expansion coefficient of
7653 Pa; and a thermal expansion coefficient of 0.8  1061/K. The SWCC and

Fig. 21.1 Experimental

setup
244 21 Thermohydromechanics

unsaturated water content are formulated using van Genuchten’s model, and the
constants in the model are n ¼ 1.48, m ¼ 0.2, and l ¼ 0.5. The initial temperature and
volumetric water content are 293.15 K and 0.5, respectively. The constants for the
computation of the thermal diffusivity are C1 ¼ 0.55, C2 ¼ 0.8, C3 ¼ 3.07,
C4 ¼ 0.13, Cs ¼ 2106, Cw ¼ 4.2106e6, and Cv ¼ 1.2103. The initial soil-
water potential head is 7 m, and the initial temperature is 293.15 K. Please simulate
the THM field to predict the temperature and water distributions at the end of 5 days,
and compare them against the following experimental results. Also, simulate the
variation of the axial deformation of the cylindrical sample within the 5 days.

Results

Figure 21.2 shows the measured distributions of volumetric water content and
temperature at the time of 5 days.
Figure 21.3 illustrates the results from the numerical simulation.

0.35
Volumetric Water Content

0.33

0.31

0.29

0.27

0.25
0 5 10 15 20 25
Distance from the hot end (cm)
75

65
Temperature (◦C)

55

45

35

25
0 5 10 15 20 25
Distance from the hot end (cm)

Fig. 21.2 Measured water content (upper) and temperature (lower) along the specimen at the end
of the lab test (5 days)
Practice Problem 245

Fig. 21.3 Simulation results for water content (upper), temperature (middle), and axial deformation
(lower)
246 21 Thermohydromechanics

Fig. 21.3 (continued)

Appendix: FlexPDE Script File

{Thermohydromechanical process in a plane strain case. This program is

developed to simulate an THM test conducted by LiuRG (www.liurg.org) }
TITLE 'Thermohydromechanics'
COORDINATES YCYLINDER
VARIABLES
u=VECTOR(ur,uz)
Temp
H
DEFINITIONS
C1=0.55
C2=0.8
C3=3.07
C4=0.13
Cs=2e6
Cw=4.2e6
Cv=1.2e3
thetar=0.05
thetas=0.535
Ks=3.2e-6
alpha=1.11
n=1.48
m=0.2
l=0.5
Practice Problem 247

gamma0=71.89e-3
rhow=1000 !rhoi=931 ! rhos=2700
E=1e5
nu=0.3
Hk=7653
al=0.8e-6
Se=(1+abs(alpha*H)^n)^(-m)
thetaw=Se*(0.535-0.05)+0.05
KH=Ks*Se*(1-(1-Se^(1/m))^m)^2
KT=KH*(H*4*1/gamma0*(-0.1425-2.38e-4*2*Temp)*1e-3) mu=E/2/(1+nu)
lambda=E*nu/(1+nu)/(1-2*nu)
K=E/3/(1-2*nu)
D_temp=(C1+C2*(thetaw)-(C1-C4)*exp(-C3*thetaw)) / (Cs*thetas
+Cw*thetaw) ! Thermal diffusivity
delta=VECTOR(1,1)
C_stiffness=TENSOR((2*mu+lambda,lambda,0),(lambda,2*mu
+lambda,0),(0,0,mu)) ! Stiffness matrix
epsilon_total=VECTOR(dr(ur),dz(uz),0.5*(dz(ur)+dr(uz)))
epsilon_thermal=VECTOR(al*(T-293.15),al*(T-293.15),0)
epsilon_H=VECTOR((H-(-55))/HK,(H-(-55))/HK,0)
epsilon_elastic=epsilon_total -epsilon_thermal-epsilon_H
sigma=dot( C_stiffness,epsilon_elastic)
sigma_tensor=TENSOR((XCOMP(sigma),ZCOMP(sigma)),(ZCOMP
(sigma),YCOMP(sigma)))
INITIAL VALUES
temp=293.15
u=VECTOR(0,0)
H=-7
EQUATIONS
u: DIV(sigma_tensor)=VECTOR(0,0)
temp: Dt(temp)+DIV(KH*GRAD(H+z)*temp)=DIV(D_temp*GRAD(temp))
h: Dt(thetaw)=DIV(KH*GRAD(H+z))+DIV(KT*GRAD(Temp))
BOUNDARIES
REGION 1
START(0,0)
NATURAL(ur)=0
VALUE(uz)=0
VALUE(temp)=293.15+70
NATURAL(H)=0
LINE TO (0.03,0)
NATURAL(ur)=0
NATURAL(uz)=0
NATURAL(temp)=1*D_temp*(temp-293.15)
NATURAL(H)=0
LINE TO (0.03,0.28)
NATURAL(ur)=0
NATURAL(uz)=0
VALUE(temp)=293.15
NATURAL(H)=0
LINE TO (0,0.28)
VALUE(ur)=0
NATURAL(uz)=0
NATURAL(temp)=0
NATURAL(H)=0
LINE TO CLOSE
248 21 Thermohydromechanics

TIME 0 TO 3600*24*5 by 3600

MONITORS
PLOTS
FOR CYCLE=1
CONTOUR(thetaw)
ELEVATION(thetaw) FROM (0,0) TO (0,0.3)
ELEVATION(temp-273.15) FROM (0,0) TO (0,0.3)
HISTORY(uz) at (0,0.26)
END
Chapter 22
Electrokinetics

Introduction

Electrokinetics includes a family of phenomena resulting from the distributions of

the charges and electric potential at interfaces between two different phases. In
porous materials, these two phases primarily refer to the solid phase such as clay
particles and the liquid phase such as water. Such two dissimilar phases usually have
different chemical potentials. Upon contact, it is very likely that one of the phases,
like water, is a polar liquid and will tend to orient its (dipolar) molecules in a
particular direction at the interface and consequently generate a potential difference.
In many cases, there are ions or excess electrons in one or both phases or ionogenic
groups present; as a result, there will be a tendency for the electric charges to
distribute in a nonuniform way in a small region close to the interface. All these
distributions of the charges and electric potential will make the electric field in that
small region close to the interface different from that inside the bulk materials. This
small region can be divided into two layers according to the mobility of the liquid. A
theory explaining such distributions of the charges and potential is called the double
layer theory. The special redistribution of charges and potential is the origin of
various electrokinetic phenomena.
The existence of this special electric field gives rise to various electrokinetic
phenomena. This is because an external electric field can cause relative movements
of phases via the electrostatic forces, whereas relative movement between the liquid
and solid phases can lead to changes in the electric field in the double layer or even in
the bulk of the liquid phase. The most common electrokinetic phenomenon is
electroosmosis, in which the solid phase remains stationary while the liquid moves
in response to an external electric field. This phenomenon usually occurs in fully or
partially saturated porous materials, when the solid is still in the form of capillaries or
porous plugs (Hunter 2013). As shown in Fig. 22.1 for soils, water moves as
polarized water molecules are attached to cations, which are mobilized by an
external electric field. Meanwhile, as the external electric field is also applied to

© Springer International Publishing AG, part of Springer Nature 2018 249

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_22
250 22 Electrokinetics

Fig. 22.1 Major electrokinetic mechanisms in porous materials

the particles with excess charges, the suspended particles can be moved, leading to
another significant electrokinetic phenomenon: electrophoresis. In addition, charged
ions and ion complexes also migrate due to the electric field, resulting in the
phenomenon of electromigration.
There are several other types of electrokinetic processes. Similar to an external
electric field, an external mechanical disturbance such as a pressure can also trigger
coupled changes due to the existence of the electric field. If a pore liquid moves as a
result of an external pressure gradient, the migratory water could carry the charges in
it and thus result in an electric current. This phenomenon, termed streaming poten-
tial, usually happens in saturated or partially saturated porous materials, because the
solid is relatively immobile compared to the liquid. By contrast, in a suspension, a
solid phase can settle down due to gravity, and this relative movement of the solid to
the liquid can cause an electric potential in a similar way, which is called the
sedimentation potential. Introduced above are the five major types of electrokinetic
phenomena.
In a broader sense, electrokinetic effects also include several other types of
electrokinetic phenomena. Diffusiophoresis refers to the motion of particles under
the influence of chemical potential gradients. Capillary osmosis describes the motion
of a liquid in a porous body under the influence of chemical potential gradients. A
colloid vibration current is an electric current generated by particles moving in a fluid
under the influence of ultrasound. The electric sonic amplitude is used to describe the
ultrasound generated by colloidal particles in an oscillating electric field.
Electrokinetics is a typical interdisciplinary subject which spans a wide breadth of
disciplines. Related areas include colloid science, microfluidics, biochemistry, envi-
ronment remediation techniques, and geo-engineering applications. The electroki-
netic phenomena were first extensively studied in colloid chemistry in the early
Double Layer Theory 251

Fig. 22.2 Application of electrokinetics for soil remediation

twentieth century, since its discovery in the early part of the nineteenth century
(Casagrande 1949). These days, these phenomena have attracted great attention in
micro- and nano-fluidic lab-on-a-chip systems for manipulating both liquids and
particles, in terms of electrokinetics over insulating surfaces and that over
conducting surfaces (Zhao and Yang 2012). In biochemistry, interest has been
placed on the theories and practice for separating a complex mixture of components
with different motilities (Righetti et al. 1979). In environmental applications,
low-level direct currents have been explored as a “cleaning agent” to remediate
polluted soils and other contaminated porous materials using a setup similar to the
one illustrated in Fig. 22.2. This technique demonstrated a great potential in saline
soil restoration, nutrient recovery from wastes, and repair/maintenance of building
structures (Huang et al. 2012). Attempts in geotechnical engineering were first
pioneered by Casagrande (1949) in the 1930s, which has been developed into a
common technique for strengthening soft clays and silts and stabilizing earth slopes
and dams, e.g., Bjerrum et al. (1967), Casagrande (1983), and Lo et al. (1991).
In this chapter, we will first discuss the double layer theory, which sets the basis
for understanding the electrokinetic phenomena. Then three sections will be devoted
to electroosmosis, electromigration, and electrophoresis for the transport of water,
free ions, and charged particles in soils, respectively, under the influence of an
external electric field. Later, we will combine these theories into a complete frame-
work for considering these three major types of electrokinetic phenomena. A
practice problem developed based on a lab test will be introduced in the last section.

Double Layer Theory

The occurrence of electrokinetic phenomena is attributed to the potential and charge

redistributions on the interface between the charged solid and its neighboring liquid
containing opposite charges, which can be described using the double layer theory.
252 22 Electrokinetics

Therefore, it is necessary to understand the double layer theory to better comprehend

electrokinetics. However, this could lead to a confusion between the double layer
theory and electrokinetics. In fact, the physical mechanisms responsible for the
double layer theory also represent a multiphysical process at the scale of the double
layer: an equilibrium diffusion field is coupled with an electrostatic field, while
adsorption and ionization reactions may also play a role. Multiphysics can also be
used to analyze the distribution of particles and charges in the double layer area.
Apparently, this type of multiphysics process occurs at the nano- or microscale. This
multiphysical field is less common though it becomes an interest in microfluidics.
Most effort that is devoted to the understanding of double layers was of an analytical
nature, forming the major body of the double layer theory. By contrast, electrokinetic
phenomena occur at a scale which is larger than the double layer: a cluster of
particles or bulk porous materials instead of a single interface. This is the
multiphysics to be discussed in this chapter. Despite the difference, electrokinetics
borrows conclusions and understandings from the double layer theory. Considering
the significance of the double layer theory, the classic theory of double layers will be
briefly introduced in this subsection for the sake of completeness.
On solid-liquid interfaces, which are common in porous materials, the redistri-
butions of electric potentials and charges on these interfaces are triggered by surface
charges via a few common mechanisms: (1) preferential adsorption of ions in the
solution, (2) adsorption-desorption of lattice ions, (3) direct dissociation or ioniza-
tion of surface groups, and (4) charge-defective lattice: isomorphic substitution
(Zhao and Yang 2012). These charges on the solid surface give rise to the redistri-
bution of charges in the liquid phase, which are needed to achieve a charge balance
in the vicinity of the solid-liquid interface.
As shown in Fig. 22.3, the liquid close to the interface can be divided into two
layers according to the mobility of the charges within each. According to Stern
(1924), the first layer, which is sometimes termed the Stern layer, includes one
layer of ions adsorbed to the solid through a specific chemical adsorption potential.
This layer is immobile and behaves more like a solid. The outer diffuse part of the
double layer, or diffuse layer, starts at a distance of one layer of ions from the wall
until a distance where the Coulombic force is negligible or where charge neutrality is
reached in the solution. Within the diffuse layer, there is an imaginary surface which
separates the diffuse layer into an inner stationary part and an outer part which could
move in response to shear forces. The electric potential at this slipping plane relative
to that at a point in the bulk fluid far from the interface is termed the electrokinetic
potential or zeta potential, which is a characteristic potential of the whole double
layer. The surface potential and the Stern potential are two other characteristic
potentials on the surface and the outer surface of the Stern layer, respectively.
The electrostatic potential distribution is assumed to obey Poisson’s equation in
the system, which represents a typical electrostatic problem. Besides, the equilibrium
diffusion field for the ion distribution is assumed to be determined by the Boltzmann
equation. Combining these two mechanisms, we can obtain the so-called Poisson-
Boltzmann equation. Detailed deductions of a typical double layer theory will be
presented in the following paragraphs.
Double Layer Theory 253

Fig. 22.3 Double layer and

electric potentials

The separation of charges that occurs at the interface between two phases is called
an electrical double layer because it consists of, ideally, two regions of opposite
charges. The simplest model of such a system is usually attributed to Helmholtz in
which both layers of charge are fixed in parallel planes to form a molecular
condenser. According to Overbeek and Kruyt (1952), this model was first explicitly
proposed by Perrin (1904). Although this model can be used to derive some
important electrokinetic relationships, it has long been recognized as an inadequate
representation of the situation. To obtain the Poisson-Boltzmann equation based on
the model, Poisson’s equation is first recalled:
ρ
∇2 ψ ¼  , ð22:1Þ
ε0 εr
where ψ is the electric potential, ρ is the volumetric charge density, ε0 is the
dielectric constant of vacuum, and εr is the relative permittivity.
At equilibrium, the diffusional force and the electric force on an ion must balance
out, which requires

∇μi ¼ zi e∇ψ, ð22:2Þ

where μi and zi are the chemical potential and valence of ion type i, respectively, e is
the elementary charge, and ψ is the electric potential.
Then we use the definition of the chemical potential per ion:

μi ¼ μi, 0 þ kB T ln ðC i Þ, ð22:3Þ
254 22 Electrokinetics

where kB is the Boltzmann constant, T is the absolute temperature, and Ci is the

number concentration (see the chapter for the concentration field). Combining the
above two equations and then integrating from a point in the bulk liquid lead to the
Boltzmann equation:


zi eψ
C i ¼ C i, 0 exp  : ð22:4Þ
kB T

The volumetric charge density in the neighborhood of the surface is given by

X
ρ¼ Ci zi e: ð22:5Þ
i

Substituting Eqs. 22.4 and 22.5 into Poisson’s equation (Eq. 22.1), we obtain the
complete Poisson-Boltzmann equation:


1 X zi eψ
∇ ψ ¼
2
Ci, 0 zi eexp  : ð22:6Þ
ε0 ε r i kB T

The major part of the double layer theory is to solve, improve, and validate this
equation for predicting the distributions of the electric potential and charges within
the double layer based on the above equations. These equations form the major part
of the DLVO theory (Gouy 1910; Derjaguin and Landau 1941; Verwey 1947),
which has been found to have many limitations, especially the limited applicability
in relatively low concentration ranges. However, the theory describes the major
mechanisms for the formation of the electric field close to the surface, which is
responsible for the electrokinetic phenomena, and identifies critical concepts such as
the slipping or shear plane.

Electroosmosis

The water flow in partially or fully saturated porous materials induced by an external
electric field is referred as electroosmotic flow. Electroosmosis is the main mecha-
nism underlying the water flow through fine-grained porous materials under the
influence of an electric field.
The direction of the water movement depends on the nature of the double layer
system around the charged particles. This water movement, or electroosmotic flow,
is caused by the Coulombic force resulting from the net charges in the water and the
external electric field. In soils, the clay particles are usually negatively charged.
Therefore, there are more positive charges than negative charges in the diffuse layer
to ensure an electrostatic balance in the double layer. The positive charges will move
to the cathode due to the Coulombic forces between the charges and the external
electric field. This movement will lead to the movement of the water, enabling water
Electroosmosis 255

Table 22.1 Values of electroosmotic hydraulic conductivity of soils

Number Soil type KEO (m2/ V. s)
1 London clay 5.8  109
2 Boston Blue clay 5.1  109
3 Kaolin 5.7  109
4 Clayey silt 5.0  109
5 Rock flour 4.5  109
6 Na-montmorillonite 2.0  109
7 Mica powder 6.9  109
8 Fine sand 4.1  109
9 Quartz powder 4.3  109
10 As quick clay (20.0–2.5)  109
11 Bootlegger Cove clay (2.4–5.0)  109
12 Silty clay, West Branch Dam (3.0–6.0)  109
13 Clayey silt, Little Pic River, Ontario 1.5  109

to move to the cathode in the above case. The water moves in the opposite direction
if the solid particles are positively charged.
The electroosmotic flow can be modeled using the Helmholtz-Smoluchowski
theory (Chung and Kang 1999). The flux of the electroosmotic flow is described by
an equation analogous to Darcy’s law (Casgrande 1949; Bjerrum et al. 1967;
Alshawabkeh et al. 1999):

JEO ¼ K EO ∇ψ, ð22:7Þ

where JEO is the electroosmotic flow flux (m/s), KEO is the electroosmotic (hydrau-
lic) conductivity m2/(V  s), and ψ is the electric potential. The coefficient of
electroosmotic conductivity is generally in the range of 109 to 1010 m2/(V  s)
as shown in Table 22.1 (Mitchell and Soga 2005).
The electromagnetic flow occurs because water molecules move with cations, to
which the water molecules are “attached” due to their polarity. It is usually accepted
to assume that the water velocity, v, is the same as the drift velocity of the ion, vd,
under the influence of an external electric field, E ¼  — ψ. In this case, the
electroosmotic flux JEO, or equivalently, a Darcy velocity due to electroosmosis, is
equal to the drift velocity:

JEO ¼ v ¼ vd ¼ μq E: ð22:8Þ

This average drift velocity of the cations in the diffuse layer can be described
using the equation proposed by Smoluchowski (1903):

εζ
μq ¼  , ð22:9Þ
η
256 22 Electrokinetics

Fig. 22.4 Zeta potential -2

values of typical soils
(Qi and Zhang 2015) -4
-6

Zeta potential (mV)

-8
Sand
-10 Silt
Clay
-12
-14
-16
-18
-20
0 2 4 6 8 10 12
pH

where ζ is the zeta potential, ε is the (absolute) permittivity of the dispersion medium
(i.e., pore water), and η is the dynamic viscosity. Please be aware of that in other
chapters of this book, the dynamic viscosity is represented using the symbol μ,
which is reserved for electrical ion mobility in this chapter. A negative sign is added
to allow for the fact that ζ values for normal porous materials such as soils are
usually negative. Typical values of the zeta potential values for different soils are
shown in Fig. 22.4.
Combining the above three equations, we can obtain the formulation of KEO:

εζ
K EO ¼  : ð22:10Þ
η
In porous materials, the formulation of the electroosmotic hydraulic conductivity
needs to take into consideration of the porosity ϕ and tortuosity τ. But it is noted that
tortuosity defined in the literature for electrokinetics is usually different from that is
used in other areas, i.e., the definition in the previous chapters. Usually, τ is defined
as the ratio between the length of the real flow streamline L and the corresponding
straight line distance, d, as τ ¼ L/d; however, the tortuosity used in electrokinetics is
defined as

τEK ¼ ðd=LÞ2 : ð22:11Þ

Therefore, the two definitions of the tortuosity can be related as

τEK ¼ ð1=τÞ2 : ð22:12Þ

The effective electrical ionic mobility in a porous material saturated with a pore
liquid can be formulated as
Electroosmosis 257

εζ
μqeff ¼  ϕτ : ð22:13Þ
η EK
Accordingly, the electroosmotic hydraulic conductivity including the porosity
and tortuosity is

εζ
K EO ¼  ϕτ : ð22:14Þ
η EK
In unsaturated soils, it is more appropriate to replace the porosity with the
volumetric water content as

εζ
K EO ¼  θτEK : ð22:15Þ
η
Despite the above analytical expression, for the numerical simulation purpose, it
seems to be more practical to measure or even assume the value of the saturated
electroosmotic hydraulic conductivity and extend this conductivity to unsaturated
soils based on the degree of saturation as

K EO ¼ K EO, 0  a  Sb , ð22:16Þ

where KEO, 0 is the electroosmotic hydraulic conductivity of the saturated soil, S is

the degree of saturation, and a and b are fitting constants. As can be seen, the above
two equations predict a linear and a nonlinear relationship between the electroos-
motic hydraulic conductivity and the water content, respectively. Despite these
equations, the real nature of this relationship is not that clear and deserves further
investigations.
Two characteristics differentiate the electroosmotic flow from the flow caused by
the hydraulic gradients in porous materials. First, the electroosmotic flow is a plug
flow, in which the velocity of the fluid is invariant across any cross section perpen-
dicular to the flow direction. This leads to much less significant dispersive effects
(see dispersion in Chap. 13). Second, the velocity of the electroosmotic flow is
independent of the size of the water passway, as long as the thickness of the electrical
double layer is much smaller than the characteristic length of the passway. This
means that the electroosmotic flow could be significant or even dominant in small
pores or fine-grained porous materials, in which the flow caused by hydraulic
gradients diminished due to the small size of the water passway. Unlike a parabolic
velocity profile generated by a pressure gradient in the flow, a plug flow’s velocity
profile is approximately planar, with slight variation near the electrical double layer.
In a simple calculation to compare the two types of flows (Huweg 2013), we can
assume they are equal:

K EO ∇U ¼ K H ∇ψ: ð22:17Þ
258 22 Electrokinetics

If a KH of 1.3  1010 m/s and a KEO of 5  109 m2/(Vs) are used, plus an
electric potential gradient of 50 V/m, which is common, the equivalence of the two
flows will require a hydraulic gradient of 1923 m/m. This value is higher than the
numbers we would encounter in the field by several orders of magnitude.
In recent years, in order to apply electroosmotic flow to electrokinetic remedia-
tion, the complex physical phenomena underlying the water flow in soils under an
external electric field have been extensively investigated (Acar and Alshawabkeh
1993; Virkutyte et al. 2002). Besides altering the pore water pressure regime by
inducing a time-dependent coupled flow and deformation (consolidation) process in
the soil mass, the application of an electric field to a fine-grained soil promotes
complex electrochemical phenomena, such as hydrolysis reactions at the electrodes,
ion migration, changes in pore water pH, aqueous phase reactions, precipitation and
dissolution, and adsorption and desorption. These phenomena can, in turn, induce
permanent modifications to the mechanical properties of the soil (Esrig and
Gemeinhardt 1900; Gray 1970). Therefore, it is worthwhile to mention that inves-
tigations into the electroosmotic phenomenon may not just be about electroosmosis
but, instead, be associated with a complex multiphysical process.

Electromigration

An electric field applied to porous materials can cause the pore liquid containing ions
to move, i.e., electroosmotic flow. Due to the same reason, the external electric field
can also lead to the movement of the ions themselves. This movement of dissolved
salt ions due to the external electric field is called electromigration. Ions can still
move due to diffusion, dispersion, and advection as we introduced in the concen-
tration field. However, at field moisture contents we mostly encounter, the dominant
ionic transport mechanism during the electrokinetic process is usually
electromigration (Alshawabkeh et al. 1999). Experiments have led to a conclusion
that the rate of electromigration is proportional to the ion concentration (Kim et al.
2005) and salt type (Cho et al. 2009). Together with the more easily observable
relationship between the magnitude of the electric field and the rate the
electromigration, we can reach the mathematical formulation of the electromigration
in a bulk solution:

JEM ¼ vC ¼ μq ∇ψc, ð22:18Þ

where JEM is the electromigration (ion) flux (mol/(m2  s)), μq is the electric mobility
of the ion (m2/(V  s)), and c is the (molar) concentration (mol/m3). As introduced in
the previous section, the ionic mobility can be defined as the average velocity of an
ion species moving under a unit electric field (1 V/m).
The above equation is valid for any ion species. Accordingly, μq is the electric
mobility of the ion species under consideration, which is different from the average
mobility discussed in the electroosmosis section. In order to use the above equation,
Electromigration 259

the electric mobility of the dispersed ion needs to be related to the material properties
that we are more familiar with, such as diffusivity. This goal can be achieved via the
Einstein relationship or Einstein-Smoluchowski relation:

μq k B T
Ddiff ¼ , ð22:19Þ
q

where Ddiff is the coefficient of diffusivity, kB is the Boltzmann constant, T is the

absolute temperature (in Kelvin), and q is the charge of each ion. The above equation
can be reformulated into

Ddiff q Ddiff  z  F
μq ¼ ¼ ,
kB T RT
where F is Faraday’s constant and F ¼ e  Na (96485.3329 C/mol), in which e is the
elementary charge and Na is Avogadro’s constant; R is the universal gas constant and
R ¼ kBNa (J /K mol); and z is the valence of the ion species under consideration and
q ¼ z  e.
Similar to the extension of the bulk solution theory to allow for porous materials
in the previous section, the effective electrical ionic mobility considering the poros-
ity and tortuosity of (unsaturated) porous materials is:

Ddiff  z  F
μqeff ¼ θτEK : ð22:20Þ
RT
Then the flux of the ion due to electromigration can be calculated as

Ddiff  z  F
JEM ¼ vc ¼  θτEK ∇ψc: ð22:21Þ
RT
In the above two equations, if the diffusivity coefficient has already considered
eff
the porosity and tortuosity, i.e., in terms of effective diffusivity Ddiff , then θτEK can
be left out:
eff
Ddiff zF
μqeff ¼ : ð22:22Þ
RT
The ionic mobility of salt ions is at least one order of magnitude higher than the
diffusion coefficient of the same ion. In addition, the value of the ionic mobility is
higher than that of the electroosmotic hydraulic conductivity, typically by two orders
of magnitude (Sullivan 2008; Huweg 2013). In other words, the velocity of
electromigration of ions is typically higher than the electroosmotic flow rate. More
significantly, the rate of electromigration is not directly affected by the characteris-
tics of the double layer, such as the zeta potential.
The direction of the electromigration is determined by the charge of the ion: an ion
moves to the electrode with the opposite charge. This direction is reflected in the
mobility term. Positively charged ions, i.e., cations, have a positive charge number,
260 22 Electrokinetics

enabling them to move in the direction of the electric field. On the contrary, nega-
tively charged ions, i.e., anions, move against the electric field. As the electroosmotic
flow can also advect particles, it is possible that the direction of this advection is
opposite to the direction of the electromigration for some ions. But as mentioned, the
velocity of electromigration usually far surpasses that of the electroosmotic flow;
therefore, the electromigration should dominate. The coexistence of multiple ions
will further complicate the formulation of the mobility due to the interaction between
different ions. Some further information will be offered in later sections.

Electrophoresis

Electrophoresis is the movement of dispersed particles relative to the fluid in contact

with them, due to the electric field generated by the double layer between the solid
and the liquid. This phenomenon usually happens in suspensions because suspended
particles are more mobile compared to the liquid. This is because the particles are the
dispersed phase, while the liquid as the dispersion medium is predominant in volume
and forms the skeleton of the composite. By contrast, in soils with a fixed solid
skeleton, especially unsaturated soils, the influence of this mechanism on the ion
transport is not appreciable.
The motion of the dispersed particles is caused by an applied electric field. The
electrostatic forces caused by this external electric field act on both the solid and bulk
liquid. From this viewpoint, electrophoresis is the “opposite” process of electroosmo-
sis: the solid particles move when they are dispersed in a relatively stagnant water, e.g.,
colloid, whereas in electroosmosis the liquid moves if the solid particles are fixed.
The suspended particles suffer from this force due to the existence of electric
charges on their surfaces. Also, considering that the Stern layer and a very small part
of the diffuse layer will also move along with the particles, the electrostatic forces
will also be affected by surface-adsorbed species. In the meanwhile, the electric field
also exerts a force on the ions in the diffuse layer which has a direction opposite to
that acting on the surface charge. Though this force is not actually applied to the
particle, it is transferred indirectly to the particle surface through viscous stress. This
force is termed the electrophoretic retardation force, which also contributes to the
movement of suspended particles. Both the direct and indirect parts contributing to
the motion of the particles are of the Coulombic force nature.
The motion caused by the electrostatic Coulombic forces will contribute to the
development of an average velocity field. Typically, the average velocity is directly
proportional to the applied force. In a colloid, the water velocity can be described
using the Smoluchowski equation or drift-diffusion equation introduced in the
section for electroosmosis:
ες
v ¼ μq E ¼ ∇ψ , ð22:23Þ
η
Mathematical Formulation 261

where ε is absolute permittivity of the dispersion medium (fluid), ς is the zeta

potential, and η is the dynamic viscosity. The negative sign in front of E indicates
that, if ς is negative, the dispersed particles move against the electric field: from the
cathode to the anode, and vice versa.
It is noted that the above formulation of the velocity overlooks many effects such
as the Debye length and surface conductivity. When the Debye length is much larger
than the particle radius, i.e., a thick double layer, the following equation can be used
instead:

2ες
v ¼ ∇ψ : ð22:24Þ
3η
This velocity field will be an independent mechanism contributing to the overall
movement of the solid particles. Usually, the movement is of interest for particle
transport in terms of a concentration field. Accordingly, the above velocity attributed
to electromigration will function in a way similar to the velocity in the ordinary
advection mechanism. Due to the same reason, the electrophoresis flux as a flux for
ion transport is
ες
JEP ¼ vc ¼  c∇ψ, ð22:25Þ
η
or

2ες
JEP ¼  c∇ψ ð22:26Þ
3η
for thick double layers. It is noted that c is the concentration of the dispersed particles
that are mobile in response to the external electric field.
As mentioned, the mechanism is not significant in porous materials as movable
solid particles with double layers may only account for a very small portion of the
solid phase with double layers. Therefore, the mathematical formulation including
the porosity (or water content) and tortuosity is usually not of interest. Here, we still
give out a general equation for the electrophoretic flux in porous materials to show
the idea:
ες
JEP ¼  c∇ψθτEK : ð22:27Þ
η

Mathematical Formulation

This section presents a general mathematical formulation of the multiphysics asso-

ciated with typical electrokinetic processes including the conservation of pore fluid
mass (h: water head), electrostatics (ψ: electric potential), and ion transport (c: molar
262 22 Electrokinetics

Fig. 22.5 Setup of an electrokinetics experiment with soils

concentration). Figure 22.5 shows a typical experimental setup to generate such

processes. The external electric field applied via the two copper electrodes on two
sides of a soil can trigger a multiphysical process including but not limited to the
above physical fields. For electrokinetics, mechanisms to be considered include
electroosmosis, electromigration, and electrophoresis in addition to other major
particle transport mechanisms, i.e., diffusion, dispersion, and advection.

Governing Equation

Unlike electrophoresis, electroosmosis contributes to the mass transfer of the continu-

ous phase instead of the dispersed phase. Accordingly, the water movement equation
should be used. For a more general variably saturated porous material, an extension to
Richards equation (Richards 1931) for unsaturated water flow could be expressed as

∂Sw ∂h
ϕ þ ∇  Jm ¼ 0, ð22:28Þ
∂h ∂t
where Sw is the degree of saturation, t is the time, ϕ is the porosity, h is the hydraulic
pressure head, and Jm is flux or the extended Darcy velocity. In this case, let us
consider the water flow caused by both the water pressure/potential gradient and the
electric potential gradient:

Jm ¼ JH þ JEO ¼ K H, 0 K Hr ∇ðh þ zÞ  K EM, 0 K EM, r ∇ψ ð22:29Þ

Mathematical Formulation 263

where KH, 0 is the intrinsic (saturated) hydraulic conductivity (due to water head),
KHr is the coefficient of relative hydraulic conductivity, z is the coordinate along the
gravity direction, KEM, 0 is the intrinsic (saturated) electroosmotic conductivity, KEM,
r is the coefficient of relative electroosmotic conductivity, and ψ is the electric
potential. Substituting Eq. 22.29 into Eq. 22.28, the governing equation for the
coupled unsaturated water flow including electroosmosis turns into

∂Sw ∂h
ϕ ¼ ∇  ½K H, 0 K Hr ∇ðh þ zÞ þ K EM, 0 K EM, r ∇ψ : ð22:30Þ
∂h ∂t
According to Ohm’s law, the electric current density JE can be expressed as (Ohm
1827)

JE ¼ σ∇ψ, ð22:31Þ

where σ is the conductivity of the soil. Due to the conservation of electric charges,
the governing equation for the electric potential could be obtained by placing a
divergence operator on the two sides of the Ampère equation, leading to

∂ψ
∇  JE ¼ ε , ð22:32Þ
∂t
where ε is the absolute permittivity of the soil, which is very small. Substituting
Eq. 22.31 into Eq. 22.32, the governing equation for the electrostatics in soils could
be obtained as follows:

∂ψ
∇  ðσ∇ψ Þ ¼ ε : ð22:33Þ
∂t
In addition to the water movement and electrostatics, the conservation of ions is
also usually of major interest in electrokinetics applications. The governing equation
for ion transport can be obtained by extending the equation introduced in the chapter
for the concentration field:

∂c
¼ ∇  ðDdiff ∇cÞ  ∇  ðJEM þ JEP Þ  ∇  ðvcÞ : ð22:34Þ
∂t |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflffl{zfflfflfflffl}
diffusionþdispersion dispersion advection

All the flux and velocity terms in the above equation have been introduced in the
previous sections.

Typical Auxiliary Relationships

One big challenge in electrokinetics simulation is the nonlinearity in the material

properties, that is, the complicated mathematical descriptions of the undetermined
264 22 Electrokinetics

parameters in the governing equations: the relationships between these parameters

and their dependence on the dependent and independent variables. Most of these
physical relationships are associated with the consideration of the unsaturated
regime in the computational domain. The (relative) hydraulic conductivity, (relative)
electroosmosis conductivity, and (relative) electric conductivity could be expressed
as a function of the degree of saturation. There is no consensus on the adoption of the
auxiliary relationships. However, to offer an example, in this subsection, some
relatively new equations for the auxiliary relationships in electrokinetics, which
were adopted in Mattson et al. (2002), Tamagnini et al. (2010), and Yuan and
Hicks (2013), together with some other physically based equations, which have
been tested by the author with numerical simulations, are presented. Typical values
for the fitting constants in the relationships are also given, which will be used in the
later section for a practice problem.
The degree of saturation could be expressed as a function of the hydraulic
pressure head:
m
S ¼ ½1 þ ðαhÞn  , ð22:35Þ

where typical values for the fitting constants are α ¼ 8.51 m1, n ¼ 1.31, and
m ¼ 1  1n.
It is also common to express the relative hydraulic conductivity as a function of
the degree of saturation. Here we use a simple power rule:

K Hr ¼ aH ðSÞbH , ð22:36Þ

where typical values for the fitting constants are aH ¼ 1.0 and bH ¼ 5.
Similarly, the electroosmotic conductivity could be expressed as a function of the
degree of saturation:

K EO, r ¼ aEO ðSÞbEO , ð22:37Þ

where typical values for the fitting constants are aEO ¼ 1.0 and bEO ¼ 3.2.
The electric conductivity deserves special attention in soil-related applications
especially electrokinetics. In some literature, the electric conductivity of a soil is
assumed to be the product of the electric conductivity of the saturated soil σ 0 and the
relative electric conductivity σ r, which is similar to the treatment of the hydraulic
conductivity. Accordingly, a relationship between the electric conductivity and the
degree of saturation has also been proposed:

σ r ¼ aσ ð Sw Þ b σ , ð22:38Þ

where aσ and bσ are fitting constants. Typical values are aE ¼ 0.109 and bE ¼ 2.
However, this relationship may not work well in electrokinetic applications, because
the variation of ion concentrations under an external electric field can also signifi-
cantly alter the electric conductivity. Therefore, a more accurate formulation of the
electric conductivity should also include the ion concentration(s). In a simple way,
Mathematical Formulation 265

we can obtain such a formulation by considering the origin of the electric conduc-
tivity. The electric conductivity of a soil has contributions from both the liquid and
the solid phases

σ ¼ σl þ σs, ð22:39Þ

where σ l and σ s are the electric conductivity components contributed by the liquid
and the solid, respectively. σ l is termed the bulk electric conductivity of the pore
liquid in some literature. This quantity changes with both the degree of saturation,
tortuosity, and the types and concentrations of the ions in the liquid. Theoretically, σ l
can be formulated based on the theory of electromigration:

Ddiff  z  F
σ l ¼ σ w θτEK ¼ μqeff cF ¼ θτEK cF, ð22:40Þ
RT
where σ w is the true electric conductivity of the pore water and μqeff c gives out the
number of ions (in mol) passing through a unit area per unit time under a unit electric
field. The use of Faraday’s constant turns the mass into a charge amount (C), leading
to the electric conductivity, i.e., amount of charge passing through a unit area per
unit time under a unit electric field. In strong electrolytes with high concentrations,
Kohlrausch’s law instead of a linear function is more appropriate for describing the
relationship between σ l and the concentration:

σ l ¼ aσ c  bσ c1:5 , ð22:41Þ

where aσ and bσ are fitting constants.

Besides, the electric conductivity in the above context needs to be differentiated
from some other electric conductivity concepts in soil-related applications. In fact,
the σ discussed above is the apparent electric conductivity of the whole soil, which is
represented using the symbol ECa in many studies (Corwin and Plant 2005). It is the
value we would obtain using an electric conductivity sensor. Another electric
conductivity that is frequently used in soil chemistry and agricultural applications
is the extract conductivity σ e. σ e is the saturation extract electric conductivity. This
electric conductivity is obtained by first saturating a soil with deionized water and
then measuring the electric conductivity of the extracted water from this saturated
soil. This is the traditional way to measure the electric conductivity of a pore liquid.
Therefore, a majority of the electric conductivity values reported in the literature are,
in fact, σ e. σ e is different from the true electric conductivity of the pore water σ w.
If the salts stay dissolved in water as the water content changes, then the following
relationship is valid:

σ e ¼ σ w  S: ð22:42Þ

One more auxiliary relationship is still needed to enable numerical simulation of

electrokinetics in porous materials for tortuosity. Mattson et al. (2002) employed a
simple relationship between the tortuosity and the water content as follows:
266 22 Electrokinetics

F λ ðθ  θr Þnþ1
τEK ¼ : ð22:43Þ
θsat
where θ, θr, and θsat are the water content, immobile water content, and saturated
water content, respectively, and Fλ is a constant, i.e., 1.58.
A more complicated way is available to consider the mobility of various ion
species in porous materials. In this way, the effective ionic mobility is related to that
in a bulk solution to consider the influence of other ions on the mobility of the ion
under consideration, i.e., species i:
 pffi
0:5058z2  I ffi
p 0:3I
μqeff
, i ¼ μq, i  θ  τEK  10
1þ I
, ð22:44Þ

where I is the ionic strength, which can be related to the electric conductivity of the
free solution σ w via the following function, and z is the valence of the ion. This
function is obtained by fitting the test results of 124 river water samples
(Ponnamperuna et al. 1966):

I ¼ 0:127  σ w  0:003: ð22:45Þ

Practice Problem

Problem Description

An experiment was conducted using a mixture of a play sand (poorly graded fine
sand) and a clay (high plasticity) with a dry mass ratio of 1:1. The mixture was
compacted gently using a miniature compactor, and the final specific gravity is 1.9.
The compacted soil sample sat for 24 h to allow for the water content to become
uniform, and the sample was also covered with a plastic sheet to reduce the effects of
evaporation (see Fig. 22.5).
Electrodes made of copper robs were placed on two sides of the plastic tank to
trigger the electrokinetic process. A voltage of 20 V was placed on the two
electrodes: 10 V and  10 V on the two electrodes. The tank had (inner) dimensions
of 37 cm (length) by 21 cm (width) by 22.5 cm (height). The electrodes extend to the
bottom of the soil layer (also tank), which filled slightly over one half of the tank
height to generate an approximately 1D problem.
Three water content sensors (Decagon GS3), connected to a data logger (Decagon
EM50), were placed along the x-axis of the tank, whose positions are displayed in
Fig. 22.6. The sensors, from the left to the right, are named Sensor 1, Sensor 2, and
Sensor 3. A program called ECH20 Utility was used to record the volumetric
moisture content and the electric conductivity automatically. The GS3 water content
sensors measure the variation of the electric conductivity between two of its probes.
Practice Problem 267

Fig. 22.6 2D simplification

of an electrokinetics
experiment using a 1:1 sand
and clay mixture

The program took measurements every minute. Data were recorded for about 417 h.
Temperatures of the soil sample were recorded using Type-K thermocouples and the
built-in temperature sensors in the water content sensor.
The saturated hydraulic conductivity and saturated electroosmotic hydraulic
conductivity are 5  108 m/s and 1.5  109 m/s, respectively. The soil water
characteristic curve, relative hydraulic conductivity, and relative electroosmotic
conductivity are formulated using Eqs. 22.35, 22.36, and 22.37 with the given
default values. The tortuosity is calculated using Eq. 22.43. The initial soil water
pressure head is 0.08 m. The saturated and residual water contents are assumed to
be 0.40 and 0, respectively. The concentrations of cations and anions in the pore
water are both 8 mol/m3. The average diffusivities of cations and anions are 0.5  10
9
m2/s and 1.5  109 m2/s. The average valence is assumed to be 1 for both the
cations and anions. The electric conductivity can be calculated using Eqs. 22.39 and
22.40 with σ s ¼ 0.0032 S/m.

Results

The results of the sand-clay mixture electrokinetics experiment were shown in terms
of variations of the volumetric water content and electric conductivity at Sensors
1, 2, and 3 in Figs. 22.7 and 22.8. Shown in the figures are the raw data. As can be
seen, the initial water contents measured by the sensors are much different from each
other. The differences were most likely caused by the lack of the calibration.
However, the trends in the curves can still reflect some major mechanisms. The
electric conductivity measurements are more consistent.
If you try a few simulations for electrokinetics, it will not be difficult for you to
find out that electrokinetics can easily generate the most highly nonlinear phenom-
ena. The involved physical fields are strongly dependent on each other in a variety of
268 22 Electrokinetics

Fig. 22.7 Measured water

0.21

Volumetric Water Content (m3/m3)

content measured at the
three sensors
0.20 Sensor 1
Sensor 2
Sensor 3
0.19

0.18

0.17

0.16

0 5000 10000 15000 20000 25000

Time (minutes)

Fig. 22.8 Measured electric 0.14

conductivity at the three
sensors Sensor 1
Electric Conductivity (mS/cm)

0.12
Sensor 2
Sensor 3
0.10

0.08

0.06

0.04

0.02
0 5000 10000 15000 20000 25000
Time (minutes)

ways. Numerical results for the water content and electric conductivity are shown in
Figs. 22.9 and 22.10. Points a, b, and c are the locations of Sensor 1, Sensor 2, and
Sensor 3, respectively. We did not conduct a quantitative analysis here due to the
missing information about the soil properties. Despite the lack of information, we
can still see the simulation with the theory presented in this chapter successfully
captured the major trends in the experiments.
Besides, a detailed explanation of the above experiment can be obtained by
analyzing the numerical results as shown in Figs. 22.11 and 22.12. The water content
variations were primarily attributed to the pore water movement toward and water
Practice Problem 269

Fig. 22.9 Simulated water content at the three sensors

Fig. 22.10 Simulated electric conductivity at the three sensors

Fig. 22.11 Final distribution of the degree of saturation

Fig. 22.12 Typical distribution of the total concentration in the process

Practice Problem 271

accumulation at the cathode. The water movement leads to a decrease in the water
content at the three sensors. As more and more water accumulates at the cathode, as
shown in Fig. 22.11, the saturated region in the vicinity of the cathode gets bigger
and bigger. Then the left boundary of this saturated region reaches Sensor 3 first and
causes a significant increase in the water content there. The redistribution of the
electric conductivity is mainly caused by the migrations of cations and anions toward
the cathode and anode, respectively. As illustrated in Fig. 22.12, the ion migrations
will produce two areas with very high ion concentrations in the neighborhoods of the
two electrodes. In addition, there will be two valleys between these two high-
concentration zones and the plateau occupying the middle area between the two
electrodes. The plateau area shrinks, and its height decreases as more and more ions
are attracted to the electrodes.

Appendix: FlexPDE Script File

{ This program is developed to simulate an electrokinetics test

conducted by LiuRG (www.liurg.org) }
TITLE 'EK'
VARIABLES
H ! Pressure head
V ! Potential, “phi” in the book
C1 ! Anion concentration (negatively charged)
C2 ! Cation concentration (positively charged)
SELECT
!ERRLIM 0.001
!NGRID 30
DEFINITIONS { parameter definitions }
zeta=-10e-3 ! Zeta potential
eta=1e-3 ! Dynamic viscosity
R=8.314 ! Universal gas constant
F=96487 ! Faraday's constant
Zn=1 ! Charge number
epsilon=8.854e-12*81 ! Permittivity of dispersion fluid (water)
Gs=1.9 ! Specific gravity of soil

Sw=(1+(-8.51*H)^(1.31))^(-0.23) { Original definition of degree of

saturation }
S= IF H>=0 THEN 1.0 ELSE (1+(-8.51*H)^(1.31))^(-0.23) { Extended Sw
to allow for full saturation }
Sr= IF H>=0 THEN 1.0 ELSE -0.23*1.31*(-8.51)*Sw*0.4 { 0.4 is
porosity }
Kw=50e-9*S^5
Ke=1.5e-9*S^3.2 { Ke0*Kre }
z=0 { Gravity is not considered }
theta=S*0.40 ! Volumetric water content
Flambda=1.58
tau=Flambda*(theta-0.0)^(1+1.31)/0.4 ! Tortuosity
Di1=1.5e-9 ! Molecular diffusion coefficient
272 22 Electrokinetics

Di_eff1=theta*tau*Di1
Di2=0.5e-9 ! Molecular diffusion coefficient
Di_eff2=theta*tau*Di2
u=VECTOR(-Kw*dx(H+z)-Ke*dx(V),-Kw*dy(H+z)-Ke*dy(V)) { Water flux }
uh=VECTOR(-Kw*dx(H+z),-Kw*dy(H+z))
Cb1=IF C1<=0 THEN 0 ELSE C1 { Use Cb to set a lower bound for C: C cannot
be negative }
Cb2=IF C2<=0 THEN 0 ELSE C2
sigma= Di_eff1*Zn*F/(R*293.15)*Cb1*F + Di_eff2*Zn*F/(R*293.15)
*Cb2*F +0.0032 { 0.0032 is the electric conductivity of solid }
INITIAL VALUES
H=-0.08
V=10-20*x/0.37
C1=8
C2=8
EQUATIONS
H: DIV(Kw*GRAD(H+z)+Ke*GRAD(V))=Sr*dt(H+z)
V: DIV(sigma*GRAD(V))=epsilon*dt(V)
C1: -DIV( -Di_eff1*GRAD(Cb1)) - DIV( Di_eff1*Zn*F/(R*293)*Cb1
*GRAD(V) )=dt(C1)+DIV(u*Cb1) { JEp=- epsilon*zeta/eta*GRAD(V)*Cb
is excluded }
C2: -DIV( -Di_eff2*GRAD(Cb1)) - DIV( Di_eff2*Zn*F/(R*293)*Cb2
*GRAD(V) )=dt(C2)+DIV(u*Cb2) { JEp=- epsilon*zeta/eta*GRAD(V)*Cb
is excluded }
BOUNDARIES
REGION 1
START (0.0,0)
NATURAL(V)=0
NATURAL(H)=0
NATURAL(C1)=0.0
NATURAL(C2)=0.0
LINE TO (0.37,0)
VALUE(V)=-10
NATURAL(H)=0.0
NATURAL(C1)=0.0
NATURAL(C2)=0.0
LINE TO (0.37,0.21)
NATURAL(V)=0.0
NATURAL(H)=0.0
NATURAL(C1)=0.0
NATURAL(C2)=0.0
LINE TO (0.0,0.21)
NATURAL(H)=0
VALUE(V)=10
NATURAL(C1)=0.0
NATURAL(C2)=0.0
LINE TO CLOSE
TIME 0 TO 3600*24*17 BY 10
MONITORS
FOR Cycle=1
CONTOUR(S)
CONTOUR(V)
CONTOUR(C1+C2)
Practice Problem 273

VECTOR(u)
VECTOR(uh)
CONTOUR(-Di_eff1*Zn*F/(R*293.15)*dx(V)/(epsilon*zeta/eta*dx(V)))
CONTOUR(-Di_eff1*Zn*F/(R*293.15)*dx(V)/ (-Kw*dx(H+z)-Ke*dx(V)))
Elevation(S) from (0.0,0.1) to (0.37,0.1)
Elevation(C1+C2) from(0.0,0.1)to (0.37,0.1)
HISTORY(theta/Gs) AT (0.06,0.105) AT (0.17,0.105) AT
(0.28,0.105) { theta/Gs is the gravimetric water content }
HISTORY(sigma) AT (0.06,0.105) AT (0.17,0.105) AT (0.28,0.105)
END
Chapter 23
Electromagnetics

Introduction

Electromagnetics, or electromagnetism, deals with physical processes in which both

the electric field and magnetic field are involved. Unlike an individual electrical field
and an individual magnetic field, which exists independently as electrostatics and
magnetostatics, respectively, the electrical field and magnetic field in electromag-
netics are varying (dynamic), and each can cause the other, leading to a fully coupled
multiphysical process. From a viewpoint of physics, electromagnetics is the study of
the electromagnetic force between electrically charged particles as one of the four
fundamental physical interactions. The electromagnetic force usually appears in the
form of electromagnetic fields, such as electric fields, magnetic fields, and light.
Aside from the electromagnetic force, the other three fundamental interactions are
the strong interaction, weak interaction, and gravitation.
Electromagnetic phenomena originate from the electromagnetic force, which is
also called the Lorentz force. The electromagnetic force includes both electricity and
magnetism as two elements of one mechanism. The electromagnetic force primarily
determines the internal properties of most objects encountered in daily life. Ordinary
materials take their form due to the intermolecular forces between molecules in the
materials which originate from electromagnetics. Even inside an atom, electrons are
bounded by electromagnetic wave mechanics within orbitals around the atomic
nuclei. However, the electromagnetics discussed in this book will be focused on
the macroscopic field theories in the context of continuum mechanics instead of pure
forces between microscopic particles.
Maxwell’s equations, together with the Lorentz force law, form the foundation of
classical electrodynamics, classical optics, and electric circuits. For the electromag-
netic field of interest here, Maxwell’s equations are the primary mathematical
descriptions. Maxwell’s equations in the tensor form include four partial differential
equations and a few auxiliary constitutive relationships between several terms in
these partial differential equations. Maxwell’s equations describe how electric and

© Springer International Publishing AG, part of Springer Nature 2018 275

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_23
276 23 Electromagnetics

Fig. 23.1 Common application areas of computational electromagnetics

magnetic fields are generated and altered by each other and by charges and currents.
These equations are named after the physicist and mathematician James Clerk
Maxwell, who published an early form of the equations between 1861 and 1862.
The numerical solution of Maxwell’s equations forms an important study area called
computational electromagnetics, which also involves the development and applica-
tion several numerical methods that are not that well known in the general scope of
numerical simulation. As shown in Fig. 23.1, computational electromagnetics has
applications spanning over most traditional areas involving the electromagnetic field.
Electromagnetics is not a major topic in the traditional scope of porous materials
and related research. However, it is still of interest as electromagnetic wave-based
sensors are widely used in geophysics, petroleum engineering, and geotechnical
engineering. For example, electromagnetic induction techniques have been used
since the late 1970s to characterize the spatial variability of soil properties including
the soil salinity, water content, flow patterns, soil texture, compaction, organic
matter content, pH, and depth to subsurface horizons and stratigraphic layers or
bedrock (Doolittle and Brevik 2014). Also, ground penetration radar, time-domain
reflectometry sensor, and frequency domain sensors have been widely used in
transportation engineering, geotechnical engineering, and soil sciences. It is inter-
esting to notice that the applications of these electromagnetic techniques have been
gaining rapidly increasing acceptance. Moreover, the electromagnetics-related
explorations in microfluidics and bio-environments will also prompt the application
of electromagnetics in porous materials. Electromagnetics is turning into an essential
part of the multiphysics in porous materials.
Conventional Maxwell’s Equations 277

In the following sections, Maxwell’s equations as the governing equation,

together with auxiliary relationships needed to close the mathematical description,
will be introduced first. Later, the governing equations will be further explained with
physics regarding polarization and magnification, several variants of Maxwell’s
equations, and an analogy between the electromagnetic fields and the mechanical
field. More discussions will be made to facilitate numerical simulation with the
equation system for both vacuum and special conditions like conductive materials
and to explain the treatments of boundary conditions. Electromagnetic waves will be
introduced by deriving the wave equation from Maxwell’s equations. To facilitate
numerical implementations, formulations of Maxwell’s equations using potentials
and phasors (for waveguides) will be detailed in two separate sections. The imple-
mentation of Maxwell’s equations using both the field vector and potentials will be
achieved via a simple practice problem in the last section.

Conventional Maxwell’s Equations

Maxwell’s equations have been well developed and validated for the description of
the electromagnetic field. Considering the fact that they are so well known, a
conventional form of the equations will be given first. Then the four PDEs will be
explained one by one regarding their physical meanings and mathematical construc-
tions. Strictly speaking, Maxwell’s equations are not exact laws of the universe such
as in phenomena related to photons or virtual photons. Instead, they are a classical
approximation to the more accurate and fundamental theory of quantum electrody-
namics. However, deviations from Maxwell’s equations are immeasurably small.
These deviations are generally not considered in the context of multiphysics.
The equations in SI units can be formulated as follows:
Gauss’s law
∇
D ¼ ρef ð23:1Þ

Gauss’s law of magnetism

∇
B¼0 ð23:2Þ

Faraday’s law of induction

∂
∇E¼ B ð23:3Þ
∂t
Ampère’s circuital law

∂
∇  H ¼ Jef þ D ð23:4Þ
∂t
278 23 Electromagnetics

where D is the electric displacement (field), ρef is the free charge density, E is the
electric field, B is the magnetic field, H is the magnetizing field or the auxiliary
magnetic field, and Jef is the free current density.
The first two equations are the governing equations for electrostatics and mag-
netostatics, both of which can be called Gauss’s law. The first equation indicates that
the electrical field is a field with a source, whereas the second equation implies that
the magnetic field is a field without a source. That is, there are positive or negative
charges but no magnetic charge (or monopole). The third and fourth equations
describe how a varying magnetic field generates an electric field and how a varying
electric field generates a magnetic field, respectively. The above four equations
contain seven unknowns. Therefore, three more constitutive relationships are needed
to close the mathematical description:

D ¼ εE ð23:5Þ

B ¼ μH ð23:6Þ

Jef ¼ σ e E, ð23:7Þ

where ε is the absolute permittivity, μ is the permeability, and σ e is the electrical

conductivity. The last constitutive equation (Eq. 23.7) is Ohm’s law.
The above Maxwell’s equations were obtained based on the assumption of the
absence of magnetic charges or monopoles. However, even though the existence of
magnetic charges has not been proven, it is often convenient to introduce magnetic
charges and currents into the equations. This treatment will lead to symmetric
equations for the electric and magnetic fields. If magnetic charges do exist, Max-
well’s equations would become

∇
D ¼ ρef ð23:8Þ

∇
B ¼ ρm ð23:9Þ
∂
∇  E ¼ Jm  B ð23:10Þ
∂t
∂
∇  H ¼ Je þ D,
f
ð23:11Þ
∂t
where ρm is the magnetic charge density and Jm is the magnetic current density,
which can be related to other quantities in a way similar to their electric counterparts:

∂ρm
∇
Jm þ ¼ 0, ð23:12Þ
∂t

Jm ¼ σ m H: ð23:13Þ
Physical Understanding of Maxwell’s Equations 279

Physical Understanding of Maxwell’s Equations

The Maxwell’s equations involve complicated relationships between D, E, B, H, ρef ,

ρm, Jef , and Jm. These relationships involved both the constitutive relationships
within each field and the interaction between the two fields. To further understand
the relationships, we first need to recall more detailed constitutive relationships in the
electric and magnetic fields:

D ¼ ε0 E þ P ð23:14Þ

B ¼ μ0 H þ μ0 M, ð23:15Þ

where P and M are the electric polarization and magnification (or magnetic polar-
ization), respectively.
The above two equations can be understood in a similar way. That is, the electric/
magnetic displacement currents or electric/magnetic charges consist of two parts: the
part independent of material type, i.e., the first term on the right-hand side of each
equation, and the part depending on material properties, i.e., polarization and
magnification for electric and magnetic fields, respectively. For the former part,
only one parameter related to the vacuum, ε0 or μ0, is involved for each field. We can
substitute the second constitutive equation (Eq. 23.15) into the fourth Maxwell’s
equations to obtain


B ∂ ∂
∇ ¼ Jef þ ∇  M þ P þ ð ε0 E Þ :
μ0 |{z} |fflfflfflffl{zfflfflfflffl} ∂t ∂t
|{z} |fflfflfflffl{zfflfflfflffl}
free current magnefication
polarization vacuum ð23:16Þ
∂
¼ Jt þ ð ε 0 E Þ
∂t
The total electric current pertaining to the material is thus composed of the
following components:

∂P
Jt ¼ Jef þ ∇  M þ ¼ Jef þ Jem þ Jep , ð23:17Þ
∂t
where Jt is the total electrical current, Jem is the magnification current, and Jep is the
polarization current. The sum of Jem and Jep is the bound current. To obtain the
relationship between the current and the other terms, we can try to carry out the curl
operation on both sides of the fourth Maxwell’s equation, leading to

∂ρef
∇
Jef þ ¼ 0, ð23:18Þ
∂t
which is the variation of the first Maxwell’s equation (Gauss’s equation)
280 23 Electromagnetics

∂D
∇
Jef þ ∇
¼ 0: ð23:19Þ
∂t
However, the above equation does not imply that

∂D
Jef þ ¼ 0: ð23:20Þ
∂t
This is because we cannot remove the divergence operator (—
) from the two
sides of the equation without impairing the equality. Alternatively, we have the
following relationship:

∂D ∂ðε0 EÞ ∂P
JeD ¼ ¼ þ , ð23:21Þ
∂t ∂t ∂t
where JeD is defined as the displacement current.
Similar relationships and concepts can be derived for the magnetic field as well.
However, it is not common to do so because the magnetic charge is an “imaginary”
object and thus not widely used. Notwithstanding, the following equation is still
helpful:

∂ρm
∇
Jm þ ¼ 0: ð23:22Þ
∂t
On the other hand, an analogy can be drawn between both the electric and
magnetic fields and the stress-strain field as listed in the following table. As can be
seen in Table 23.1, D and B are quantities that can be applied without specifying the
material type, and their units imply the meaning of “quantity per unit area”; they are
thus analogous to the “stress,” i.e., force per area. E and H can be caused by D and B,
respectively, and their definitions are related to the material type. Therefore, they are
analogous to the concept of strain. The relationships between the “strains,” e.g.,
E and H, to the “stresses,” e.g., D and B, are defined as material properties, e.g.,
permittivity and permeability, respectively. These two properties are analogous to
the compressibility of the material in the mechanical field. ρef and ρef serve the roles
of sources or body forces. Jef and Jm are the “stresses” arising due to the resistance to
the external loading; therefore, they are analogous to the effective stress in porous
materials. U and Q directly correspond to E and D, respectively, and have the
meaning of “total” instead of “per unit area”; therefore, they are analogous to strain
and stress, respectively.

Table 23.1 Analogy between ρm

U Q D P B E H M ρef Jef Jm
the electromagnetic field and
V C C C N V A A C N C N
mechanical field m2 m2 Am m m m m3 Am2 m2 s Ams
L F σ σ σ ε ε ε f f σ '
σ'

N N N m m m N N N N
m N m2 m2 m2 m m m m3 m3 m3 s m3 s
Equation System for Numerical Simulations 281

These analogies could help readers who are more familiar with the concepts in the
stress and strain field to understand the concepts in the electromagnetic field.
However, it is worth pointing out that the concepts in the electromagnetic field can
be understood in many other different ways. In the table, macroscopic quantities
such as the potential and charges are mentioned for the electric field while no
corresponding quantities are listed for the magnetic field.

Equation System for Numerical Simulations

Independent Governing Equations

In the four Maxwell’s equations without magnetic charges, there are only two
independent equations after including two constitutive relationships, i.e.,
Eqs. 23.14 and 23.15. Generally, Faraday’s law and Ampère’s law are chosen as
the independent governing equations:

∂
∇  E ¼  B, ð23:23Þ
∂t
∂
∇  H ¼ Jef þ D: ð23:24Þ
∂t
The above two equations are in some cases called Maxwell’s equations. Gauss’s
equations for both the electric and magnetic fields can be deduced from the above
two equations by taking the divergence of Eqs. 23.23 and 23.24. An auxiliary
equation is required to close the equation system for solution. This equation should
be able to relate the electric charge flux and the charge density. In fact, this is the
equation describing the conservation of charges at any point, which is frequently
referred to as the equation of continuity or the equation of the electric charge density
conservation:

∂ρef
∇
Jef þ ¼ 0: ð23:25Þ
∂t
The two governing equations can also be combined to obtain a second-order
governing equation in terms of E as
2
∂ E ∂Je
εμ þ ∇  ∇  E ¼ μ , ð23:26Þ
∂t 2 ∂t
or in terms of H as
2
∂ H
εμ  ∇  ∇  E ¼ ∇  Je :
∂t 2
282 23 Electromagnetics

Conductive Materials

Special attention is needed to apply Maxwell’s equations to conductive materials

like metals. It is well known that the charges will vanish from an interior point of a
metal body and appear at the surface almost immediately. To get a better under-
standing, let us first combine the equation for the conservation of electric charge,
Ohm’s law, and Gauss’s equation into the following equation:

∂ρef ρf
þ σ e e ¼ 0: ð23:27Þ
∂t ε
The electric charge density at any time is thus equal to
σe
ρef ¼ ρef0 e ε t , ð23:28Þ

where σ e/ε characterizes the relaxation time of the material and ρef0 is the free electric
charge density at the beginning. According to the above solution, charges within a
conductive material decay exponentially with time. For metal, with a typical conduc-
tivity of the order 106 Siemens/m and a permittivity ε of 1011 As/Vm, the decaying
time will be of the order 1017 s. Even for a relatively poor conductor such as distilled
water, the relaxation time is less than 106 s. Therefore, we can believe that no charge
will be accumulated inside the conductor even at high frequencies. Consequently, the
equation of electric charge density conservation can be approximated by

∇
Jef ¼ 0: ð23:29Þ

Similarly, for Ampère’s equation, we can assume a harmonic field with an angular
frequency of ω. Then the free electric current flux and displacement current flux are
Jef and ωε/σ e, respectively. The comparison of these two fluxes (the latter one is very
small even at high frequencies) indicates that it is reasonable to neglect the displace-
ment current in a conductive material even at high frequencies in the way as

∇  H ¼ Jef : ð23:30Þ

The approximate equation, i.e., Eq. 23.29, for the electric charge density can be
obtained by taking the divergence of Eq. 23.30 because the divergence of a curl is zero.

Boundary Conditions

The boundary conditions can be derived from the governing equations by ensuring
the validity of the equations on the boundary. The derivation is more complicated
than other physical fields due to (1) the number of equations, (2) the vector form of
the dependent variables, and (3) the involvement of the cross product operators.
Equation System for Numerical Simulations 283

Fig. 23.2 Boundary

conditions of
electromagnetic field

Figure 23.2 illustrates a simple derivation for Faraday’s equation. Let us consider a
small slender rectangular area bounded by the blue lines along the boundary between
the two regions. Application of the Stokes theorem to Faraday’s equation will lead to
E1t
dl1  E1t
dl2 ¼ 0 if the area is infinitely slender, i.e., dl12  0. Because
k k
dl1 ¼ dl2, we obtain E1t ¼ E1t (or E2  E1 ¼ 0 ), which is equivalent to
n12  (E2  E1) ¼ 0. The other boundary conditions can be derived in similar ways.
Depending on the electric and magnetic properties, the electromagnetic fields can
be discontinuous or continuous on each side of the common boundary (interface)
between two different materials. In summary, on an interface between two materials,
the field must satisfy the following conditions:
k k
n12  ðE2  E1 Þ ¼ 0 , E2  E1 ¼ 0 ð23:31Þ

n12
ðD2  D1 Þ ¼ ρe , D⊥ ⊥
2  D1 ¼ 0 ð23:32Þ

k k
n12  ðH2  H1 Þ ¼ J , H2  H1 ¼ J ð23:33Þ

n12
ðB2  B1 Þ ¼ 0 , B⊥ ⊥
2  B1 ¼ 0, ð23:34Þ

where n12 is the unit normal vector pointing from Region 1 to Region 2 and uk and u
⊥
are the tangential and normal components, respectively.
The above four equations indicate that (1) the electric field’s tangential compo-
nents are continuous across the interface, (2) the electric flux’s normal component is
discontinuous across the interface between materials and the difference is the charge
density, (3) the tangential components of the magnetic field strength are discontin-
uous across the interface and the “jump” equals an electric current flux, and (4) the
magnetic flux density’s normal component is continuous across the interface (Fisk
2008). ρe and Jef vanish under the conditions of no electric charge and no electric
current, respectively.
For external boundaries, we only consider the region inside the boundary. Then
Eqs. 23.31, 23.32, 23.33 and 23.34 reduce into
284 23 Electromagnetics

nE¼0 ð23:35Þ

n
D ¼ ρe ð23:36Þ

n  H ¼ Jef ð23:37Þ

n
H ¼ 0: ð23:38Þ

Electromagnetic Waves

The four Maxwell’s equations can be consolidated into two wave equations in
special cases. Such a consolidation can be used to unify the theories of electromag-
netism and optics. The 3D diagram in Fig. 23.3 shows a plane linearly polarized
wave propagating from left to right.
The special conditions that are required for the reformulation of the original
Maxwell’s equations into wave equations are (1) no charges (ρ ¼ 0) and (2) no
currents ( Jef ¼ 0), such as the condition in vacuum. In such cases, Maxwell’s
equations reduce to

∇
D¼0 ð23:39Þ

∇
B¼0 ð23:40Þ

∂
∇E¼ B ð23:41Þ
∂t

Fig. 23.3 Visualization of magnetic wave

Formulation Using Potentials 285

∂
∇H¼ D: ð23:42Þ
∂t
Taking the curl (— ) of the curl operation on the last two equations and using the
curl of the curl identity, i.e., —  (—  u) ¼ — (—
u)  — 2u, we obtain the wave
equations:
2
∂ D 1
¼ ∇2 D ð23:43Þ
∂t 2 ε0 μ 0

2
∂ B 1
¼ ∇2 B, ð23:44Þ
∂t 2 ε0 μ 0

where
sffiffiffiffiffiffiffiffiffi
1
c¼ ¼ 2:9979  108 m=s ð23:45Þ
ε0 μ 0

is the speed of light in free space. In materials with relative permittivity εr and
relative permeability μr, the phase velocity of light becomes
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
c¼ , ð23:46Þ
ε0 εr μ 0 μ r

which is usually less than c in vacuum.

In addition, D and B are mutually perpendicular and both are normal to the
direction of wave propagation. D and B are also in phase with each other. A
sinusoidal plane wave is one special solution to these equations. Maxwell’s equa-
tions explain how these waves can physically propagate through space. A changing
magnetic field creates a changing electric field through Faraday’s law. In turn, that
electric field creates a changing magnetic field by following Ampère’s law with
Maxwell’s addition. This perpetual cycle allows these waves, which is now known
as electromagnetic radiation, to move through space at the velocity of c.

Formulation Using Potentials

Governing Equation

The formulation of the equation system using the field vectors was introduced in a
previous section. In fact, the electromagnetic field can be formulated using the
electric and magnetic vector potentials, which are more commonly used due to the
286 23 Electromagnetics

advantages of this type of formulation in boundary condition treatments and in the

convergence of the numerical solution. As introduced in the chapters for electrostat-
ics and magnetostatics, the electric and magnetic field satisfy a zero curl and zero
divergence condition, respectively; we thus can introduce a vector potential first for
the magnetic field:

B ¼ ∇  A, ð23:47Þ

where A is the magnetic vector potential. From Maxwell’s equations, it is known that

∂B ∂ ∂A
∇E¼ ¼  ð∇  AÞ ¼ ∇  ð23:48Þ
∂t ∂t ∂t
or


∂A
∇ Eþ ¼ 0: ð23:49Þ
∂t

Because —  — ψ ¼ 0, one solution to Eq. 23.48 is

∂A
E¼  ∇ψ: ð23:50Þ
∂t
The negative sign in front of — ψ implies that the electric potential decreases
along electric lines. Multiplying the electric field with σ e, we will get

∂A ∂A
Jef ¼ σ e E ¼ σ e  σ e ∇ψ ¼ σ e þ Jes , ð23:51Þ
∂t ∂t
where Jes is the source current density in the induction coil. It follows from the
Ampère’s equation with a free charge current that

1 ∂D
∇  B ¼ Jef þ : ð23:52Þ
μ ∂t
Substituting Eqs. 23.47 and 23.51 into Eq. 23.52, we obtain
2
∂A ∂ A ∂
∇  ∇  A ¼ μσ e þ μJes  εμ 2  εμ ∇ψ ð23:53Þ
∂t ∂t ∂t
for a homogeneous and isotropic medium independent of field intensity. Then let us
expand and impose the Coulomb gauge condition, i.e., —
A ¼ 0, the above
equation is simplified into
2
∂ A ∂A ∂
εμ  ∇2 A ¼ μJes  σ e μ  εμ ∇ψ, ð23:54Þ
∂t 2 ∂t ∂t
Formulation Using Potentials 287

which is called the diffusion equation. The equation can also be expressed in the
same form but for different vectors: instead of A, we write the equations in terms of
J, E, H, and B.
The relation of the vector potential to the current is
ð
μ Jes
A¼ 0 dV, ð23:55Þ
4π r
where J is the current density.
The following Poisson’s equation in terms of the electric scalar potential will
complement the above the equation:

∇2 ψ ¼ ρef : ð23:56Þ

Boundary Conditions

The treatment of the boundary conditions in the potential formulation is generally

simpler. The following equations show the boundary conditions in the potential
formulation and their counterparts in the field vector formulation:

n12  ðE2  E1 Þ ¼ 0 , ψ 1 ¼ ψ 2 ð23:57Þ

n12
ðD2  D1 Þ ¼ ρe , n12
ð∇ψ 2  ∇ψ 1 Þ ¼ ρe ð23:58Þ


1 1
n12  ðH2  H1 Þ ¼ Jef , n12
∇A2  ∇A1 ¼ Jef ð23:59Þ
μ1 μ1

n12
ðB2  B1 Þ ¼ 0 , A1 ¼ A2 : ð23:60Þ

The equivalence can be proven easily using the definitions of the potentials.
For external boundaries, Eqs. 23.57, 23.58, 23.59, and 23.60 reduce to

n  E ¼ 0 , ψ ¼ 0 ð23:61Þ

n
D ¼ ρef , n
∇ψ ¼ ρef ð23:62Þ

1
n  H ¼ Jef , n
∇A ¼ Jef ð23:63Þ
μ

n
B ¼ 0 , A ¼ 0: ð23:64Þ
288 23 Electromagnetics

Phasor Formulation (Frequency Domain) and Waveguide

Phasor Formulation

The phasor formulation of Maxwell’s equations is also called the time-harmonic

form. This formulation is not only convenient but also necessary in many cases. For
example, when considering heating due to electromagnetic waves, it is very likely to
be impractical to analyze the problem in the time domain. This is because the time-
domain electromagnetic wave simulation requires a very small time step, which is
shorter than the period of the wave, e.g., a few nanoseconds to capture the temporal
variations, while heating processes usually involve a time duration from seconds to
hours. Therefore, the consideration of the coupled process requires a very small time
step and a large number of time steps, which poses a great challenge on numerical
simulation. Despite the necessity, the phasor formulation can also bring great
advantages. For example, the formulation enables the analysis of the specific
frequencies and the search for the primary frequencies (i.e., eigenvalue problems).
Essentially, this formulation converts a time-dependent problem into an equilibrium
or an eigenvalue problem, which can remarkably reduce the level of difficulty from a
numerical simulation perspective. In addition, the phasor formulation can also
significantly simplify the consideration of the energy.
The idea of this formulation is related to the Fourier transform, in which a signal
in time is rewritten as the sum of sinusoids. When the fields and sources in
Maxwell’s equations are all monochromatic functions of time, the equations can
be reformulated using the phasor, which has a general form as

~ eiωt ,
u¼u ð23:65Þ

where u represents the field vectors, e.g., E, D, H, and B. u~ thus expresses the spatial
distribution, while eiωt stores the temporal variation information. One major advan-
tage of using this phasor form is that the temporal derivative has the following form:

∂
¼ iω: ð23:66Þ
∂t
Substituting the phasors into Maxwell’s equations, we obtain

~ ¼e
∇
D ρ ef ð23:67Þ

~ ¼0
∇
B ð23:68Þ

~ ¼ iωB
∇E ~ ð23:69Þ

~ ¼ J~ f þ iωD:
∇H ~ ð23:70Þ
e

The electromagnetic wave equations can be developed using phasors and the
general field vector as
Phasor Formulation (Frequency Domain) and Waveguide 289

~ þ εμω2 u
∇2 u ~¼0 ð23:71Þ

or

~ þ k2 u
∇2 u ~ ¼ 0, ð23:72Þ
pffiffiffiffiffi
where k ¼ ω εμ ¼ ω=c is the wave number, in which c is the wave velocity in the
medium.
Complex numbers are needed to numerically solve some of the above equations
or in the later calculations of other variables using the solutions to the equations. The
treatment of complex numbers in numerical simulation is similar to that of vectors, in
which multiple components are needed for one quantity (one complex number or
one vector). Many numerical simulation programs now come with the ability to deal
with complex-dependent variables. However, it is worthwhile to point out that the
imaginary part i is not always necessarily included for numerical simulation. This is
because the imaginary part is used to express the phase relationship between
different fields (the relative position in time). However, in some cases, we are
interested in the spatial distribution, the amplitude, or the energy dissipation, in
which the phase information or temporal variation is not needed. In these cases, we
can conduct simulations using the above equation(s) without the imaginary part.
This concept can be seen in the practice example in Chap. 26 for microwave heating.

Waveguide

Waveguides are devices for transporting waves, which can be used for both mechan-
ical and electromagnetic waves. The special dimensions and boundary conditions
associated with such devices ensure that most energy will propagate along the
longitudinal direction of the waveguides and thus achieve the goal of transmitting
either power or communication signals with minimum loss. Waveguides function as
a high-pass filter, in which most of the energy above a cutoff frequency will pass
while that below the frequency will be attenuated.
A unique feature of waveguides is that they only transmit transverse waves, such
as TEM (transverse electric and magnetic), TE (transverse electric), and TM (trans-
verse magnetic) wave. TEM, TE, and TM mean the waves have no E&M, E, and M
components along the longitudinal direction, respectively. A common waveguide is
a metal tube with a rectangular cross section. The metal sidewall of the tube
represents a perfect electric conductor boundary condition, n  E ¼ 0, which will
lead to a TE or a TM wave. To facilitate the formulation, we split the Laplacian
operator into two components for the transverse (t: x and y) and the longitudinal
directions (z):

∇2 ¼ ∇2t þ ∇2z : ð23:73Þ

290 23 Electromagnetics

Taking TE, for example, we still have Hz. The solution to the EM wave equation
in TE should have the form of

H ¼ Hðx; yÞeikz : ð23:74Þ

This is because, in the frequency domain, the H distribution on the transverse

plane does not change with time, while it changes with z as a propagating wave eikz.
The negative sign means the wave is moving forward. Substituting the solution and
the split Laplacian operator into the governing equation, we can obtain the governing
equation for the TE wave:

∇2 H z þ k2 H z ¼ 0, ð23:75Þ

or
 
∇2 H z þ k2  k 2z H z ¼ 0: ð23:76Þ

Likewise, the TM wave has the following governing equation:

∇2 E z þ k2 Ez ¼ 0, ð23:77Þ

or
 
∇2 E z þ k2  k 2z Ez ¼ 0: ð23:78Þ

Once Hz or Ez is obtained, the transverse components can be obtained as follows:

ik z iωμ
Et ¼ ∇t E z þ 2 e z  ∇t H z ð23:79Þ
k2  k2z k  k 2z

iωε ik z
Ht ¼ e  ∇t E z þ 2
2 z
∇t H z , ð23:80Þ
k  kz
2
k  k2z

where ez is the unit vector along the z (longitudinal) direction and kz is the wave
number along the z direction.
The phasor formulation using an auxiliary potential vector is also available
(Balanis 1999). The practice of the above theory will be shown in Chap. 26 with
microwave heating.

Practice Problem

This section is prepared to offer practice for computational electromagnetics using a

very simple problem: the magnetic field generated by an electric current in a metal
wire. Due to the simple PDEs involved, we here require to solve the problem using
Practice Problem 291

both the field vector formulation (original Maxwell’s equations) and the potential
formulation. A more complicated simulation of the electromagnetic field using
phasors can be found in Chap. 26.

Problem Description

As shown in Fig. 23.4, in this benchmark problem, we consider a wire, which can be
viewed as a perfect conductor, with an infinite length and a constant current of
0.01 A/m2 in it. The material surrounding this wire is air. The wire is aligned in the
z-direction and has a diameter of 1 m. Please simulate the H field using 2D numerical
simulation. Please simplify the governing equation(s) from 3D to 2D first. This is a
good practice to get familiar with Maxwell’s equations. Please compare the distri-
butions of the potential along the radial direction obtained by the potential method
and field vector method.

Governing Equations Using Vector and Potential

Formulations

The governing equations written using FlexPDE script are as follows:

Field formulation

Hx : dxðHyÞ  dyðHxÞ ¼ Jz
Hy : dxðHxÞ þ dyðHyÞ ¼ 0

Potential formulation

DEL2ðAzÞ ¼ 0:

Fig. 23.4 Computation

domain and boundary
conditions
292 23 Electromagnetics

Results

The major results are shown in the following figures. The utilization of the two
formulations of Maxwell’s equations can lead to the same simulation results.
However, from a numerical perspective, it takes much more effort for the field
vector formulation method to reach convergence, at least with the finite element
method. The treatments of the boundary conditions for both methods require special
attention. The field vector method is more convenient when the magnitudes of the
field vector components are known and easy to formulate, whereas the potential
method provides the easier boundary condition treatment when the potential is
known or can be easily calculated.
Figures 23.5 and 23.6 show the H field and potential obtained using the potential
method, which are very similar to the results obtained with the field vector method.
To quantitatively compare the two methods, the distributions of the potential along
the radial direction obtained by the potential method and field vector method are
depicted in Figs. 23.7 and 23.8, respectively. As can be seen, the results obtained by
the two methods agree very well.

Fig. 23.5 H field

Practice Problem 293

Fig. 23.6 Distribution of potential

Fig. 23.7 Radial distribution of potential obtained by the potential method

294 23 Electromagnetics

Fig. 23.8 Radial distribution of potential obtained by the field vector method

Appendix: FlexPDE Script File

Field Vector Formulation

{ Electromagnetics using E H vector formulation }

TITLE 'Electromagnetics 2D-EM' { the problem identification }
SELECT { The following error control is needed }
ERRLIM 0.00005
NGRID=35
!REGRID OFF
VARIABLES
Hx (1e-8)
Hy (1e-8)
DEFINITIONS
r1=1
r2=50
mu=4*pi*1e-7
epsilon=8.654e-12
c=1/sqrt(mu*epsilon)
B_x=Hx*mu
B_y=Hx*mu
omega=10*pi
Jz=0
Practice Problem 295

EQUATIONS
Hx: dx(Hy)-dy(Hx)=Jz
Hy: dx(Hx)+dy(Hy)=0
BOUNDARIES
REGION 1
Jz=0
START(r2,0)
Value(Hx)=-1e-4*y/(x^2+y^2)^0.5
Value(Hy)=1e-4*x/(x^2+y^2)^0.5
ARC(CENTER=0,0) ANGLE=360
REGION 2
!Inactive(Hx,Hy)
Jz=1e-2
epsilon=1e12
START(r1,0)
NATURAL(Hx)=0
NATURAL(Hy)=0
ARC(CENTER=0,0) ANGLE=360
MONITORS
FOR CYCLE=1
VECTOR(Hx,Hy)
CONTOUR((Hx^2+Hy^2)^0.5)
ELEVATION((Hx^2+Hy^2)^0.5) FROM (0,0) TO (r2,0)
PLOTS
FOR CYCLE=1
VECTOR(Hx,Hy)
CONTOUR((Hx^2+Hy^2)^0.5)
END

Potential Formulation

{ Electromagnetics 2D using potential formulation }

TITLE 'Electromagnetics 2D-Potential Formulation'
SELECT
ERRLIM 0.00005
VARIABLES
Az
DEFINITIONS
r1=1
r2=50
mu=4*pi*1e-7
epsilon=8.654e-12
c=1/sqrt(mu*epsilon)
Hx=dy(Az)/mu
Hy=-dx(Az)/mu
B_x=Hx*mu
B_y=Hx*mu
EQUATIONS
Az: DEL2(Az)=0
BOUNDARIES
REGION 1
START(r2,0)
VALUE(Az)=mu/(4*PI)*1e-2*PI*r1^2/r2
ARC(CENTER=0,0) ANGLE=360
296 23 Electromagnetics

REGION 2
START(r1,0)
VALUE(Az)=mu/(4*PI)*1e-2*PI*r1^2/r1
ARC(CENTER=0,0) ANGLE=360
MONITORS
FOR CYCLE=1
VECTOR(Hx,Hy)
CONTOUR((Hx^2+Hy^2)^0.5)
ELEVATION((Hx^2+Hy^2)^0.5) FROM (0,0) TO (r2,0)
PLOTS
FOR t=2*pi
VECTOR(Hx,Hy)
CONTOUR((Hx^2+Hy^2)^0.5)
ELEVATION((Hx^2+Hy^2)^0.5) FROM (0,0) TO (r2,0)
END
Chapter 24
Fluid Dynamics

Introduction

As a branch of fluid mechanics, fluid dynamics deals with fluid flow, that is, liquids
and gasses in motion. Fluid dynamics encompasses two major sub-areas: aerody-
namics for air and other gasses and hydrodynamics for liquids. Due to historical
reason, hydrodynamics was used interchangeably with fluid dynamics in most
studies before the twentieth century. This use can still be seen from the names of
fluid dynamics topics such as magnetohydrodynamics and hydrodynamic stability,
which in fact can also be applied to gasses. In this chapter, aerodynamics and
hydrodynamics are differentiated. However, effort will also be directed to the
introduction to aerodynamics when necessary, though this chapter is referred to as
“hydrodynamics” in many places. In fact, there is a legitimate reason for putting
hydrodynamics and aerodynamics together, that is, all the fluid dynamics problems
are associated with the same governing equations, i.e., Navier-Stokes (N-S) equa-
tions. The differences originate from the different natures of the liquid flow and gas
flow, such as flow velocity, compressibility, and viscosity. These differences lead to
different forms of the Navier-Stokes equations and, consequently, different model-
ing and solution techniques. Hydrodynamics is more associated with low-velocity
and incompressible flows, either viscous or inviscid, while aerodynamics is usually
deemed to have a relatively high velocity, high compressibility, and thus negligible
viscosity. However, these factors are not independent of each other, for example, the
consideration of compressibility and viscosity needs to be done with reference to the
velocity.
Fluid dynamics, as a study area, has a much broader scope than the fluid
dynamics discussed in the context of multiphysics. The solution to a fluid dynamics
problem typically involves the calculation of the spatial (and temporal) variations of
various properties of a fluid, such as flow velocity, pressure, density, and tempera-
ture. These calculations are built on the conservation of mass, momentum, and
energy, especially the first two. These axioms of conservation lead to the

© Springer International Publishing AG, part of Springer Nature 2018 297

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_24
298 24 Fluid Dynamics

mathematical formulation of the fluid flow: Navier-Stokes equations. As the foun-

dation of fluid dynamics, the Navier-Stokes equations can be used to model prob-
lems involving fluid flows, such as weather, ocean currents, aircraft design, scour,
design of power station, the study of blood pressure, nebulae in interstellar space,
and modeling fission weapon detonation. Some of the principles underlying the
Navier-Stokes equations could even be used in modeling traffic and crowd dynam-
ics. The equations can also be used to model and study magnetohydrodynamics
when used with Maxwell’s equations. Despite the theoretical and numerical efforts,
traditional fluid dynamics also embraces empirical and semiempirical laws derived
from flow measurements and thus enables theoretical and numerical tools for solving
practical problems. However, the fluid dynamics in the following introduction will
be confined within multiphysics.
The possible dramatic motion of liquid particles makes fluid dynamic problems
deviated from the direct use of the classic conservation equations in solid mechanics.
As we know, fluids are composed of molecules that collide with each other and with
other objects in the fluids. However, the classic conservation equations treat fluids as
a continuum rather than a discrete substance. As a result, quantities for the fluids’
properties and states such as density, pressure, velocity, and temperature are defined
at infinitesimally small points and thus are continuous from point to point. Therefore,
there is a discrepancy between the nature of the material and the way we analyze
it. In solids, this issue also exists but not as obvious as in fluids. This is because the
fast mass and momentum transfer in liquids significantly magnifies the couplings
between scales. The relative movement of fluids occurs at a wide range of scales,
from that at the molecular scale, such as the viscous dissipation discussed in
thermomechanics, to the Kolmogorov microscale which defines the smallest scales
in turbulent flow, to the scale of interest. i.e., scale comparable to that of the
computational domain (or real objects). A solution to the governing equations
merely at macroscopic scale of the object/domain (defined by the grid/mesh) will
overlook the details at smaller scales. The exclusion of the lower-scale processes will
be a big issue when the flow is turbulent. This explains why, in computational fluid
dynamics, a significant amount of effort has been made to the establishment of
various turbulent simulation techniques, such as various turbulent models, which can
be majorly classified into Reynolds average methods and large eddy methods.
In the following sections, different types of flows will be introduced first without
heavily referring to the mathematical equations. The purpose is to offer a clear
picture regarding the basic concepts in fluid dynamics and computational fluid
dynamics (CFD) simulations. Presented next will be the introduction to the
Navier-Stokes (N-S) equations. To make the introduction clear and complete, the
general equations will be explained first, followed by the derivations of various
common forms of the N-S equations, which correspond to different types of flows.
Both the mass and momentum equations will be derived from the basic conservation
law. Different constitutive relationships will be added to show different forms of the
N-S equations for different fluids. Then, a section will be used to describe many
concepts more in-depth with more mathematical details. The turbulent models will
Flow Types 299

also be briefly introduced to complete the theoretical framework. Finally, a bench-

mark problem will be provided for practicing the numerical simulation of fluid
dynamics.

Flow Types

Compressible and Incompressible Flow

All fluids are compressible to some extent, that is, the fluids’ density will change in
response to the changes in the pressure or temperature. However, in many situations,
the changes in the pressure and temperature are so small that the subsequent changes
in density are negligible. In these conditions, the flow can be simplified into the
incompressible flow. Otherwise, the compressibility needs to be considered in the
N-S equations and with an additional thermodynamic equation of state.
A thermodynamic equation of state is usually needed for the compressible flow to
relate the pressure to the other thermodynamic variables, i.e., temperature and
density, in addition to N-S equations for the mass, momentum, and energy conser-
vation. A simple example is the perfect gas equation of state:

ρRT
p¼ , ð24:1Þ
M
where p is the pressure, ρ is the density, R is the gas constant, M is the molar mass,
and T is the temperature.
By contrast, in the incompressible flow, the density of the fluid does not change
with time:

Dρ
¼ 0, ð24:2Þ
Dt
where D/Dt is the substantial (or material) derivative, which includes both the local
and convective derivatives. Further simplifications can be obtained by assuming the
fluid has a uniform and constant density.
Different criteria are used for liquids and gasses for the choice of compressible
and incompressible fluid models. For liquids, this choice depends on the fluid
properties, e.g., the critical pressure and temperature of the fluid and the flow
conditions. For example, a criterion could be constructed based on the comparison
between the critical pressure and the actual flow pressure. For gasses, the Mach
number of the flow is usually adopted. The Mach number is the ratio between the
flow velocity relative and the local speed of sound:
v
Ma ¼ , ð24:3Þ
c
300 24 Fluid Dynamics

where Ma is the Mach number, v is the local flow velocity with respect to the
boundaries, and c is the speed of sound in the material. As a rough guide, a gas
can be treated as incompressible when the Mach numbers are below 0.3. Depending
on the Mach number, there are subsonic, transonic, supersonic, and hypersonic
flows. This classification will not be discussed in this book due to its relatively
weak connections to porous materials.
It is worthwhile to point out that compressibility needs to be considered in
acoustic problems. This is because sound waves are compressive waves involving
changes in the pressure and density of a material through which the waves propagate.
Therefore, the consideration of compression and compressibility is a prerequisite.

Inviscid, Newtonian, and Non-Newtonian Fluids

In a broad scope, fluids can be categorized as in Fig. 24.1. The time-independent

fluid is of more interest in traditional computational fluid dynamics. Therefore, we
will focus on this category in the following introduction.
In these fluids, fluid flows also can be classified according to the effect of the
viscosity. Viscous fluids are those in which viscosity, which quantifies the fluid’s
friction, has a considerable effect on the fluid motion. Viscous fluids can be further
categorized into Newtonian and non-Newtonian fluids according to the relationship
between the viscosity and velocity gradient. On the contrary, the effect of the
viscosity is negligible in inviscid fluids. The above classification of the fluids is
relative and dependent on the flow conditions. Therefore, it is also common to

Fig. 24.1 Classification of fluid types

Flow Types 301

Fig. 24.2 Typical fluid

types and their constitutive
relationships

classify fluid flow into Stokes, viscous, and inviscid flows by considering both
material properties and flow conditions. These three flow types use different
governing equations which are simplified versions of the general N-S equations.
The Reynolds number is usually adopted to determine whether a flow is viscous
or inviscid. The Reynolds number, Re, is the ratio between the inertial force and
viscous force (the equation will be given in a later subsection). Stokes flow is the
flow with very low Reynolds numbers, i.e., Re 1. In Stokes flow, inertial forces
can be neglected compared to viscous forces due to the low flow velocity. As the
velocity increases, the Reynolds number increases, which leads to non-negligible
inertial forces compared to viscous forces. Stokes flow is predominant in porous
materials due to the low velocity of the flow through the pore spaces, whereas
viscous flow is the most common one that we would encounter in surface water.
In viscous flow, the stress is related to the rate of strain. As illustrated in Fig. 24.2,
in many fluids, such as water and air, the relationship is approximately linear. The
proportionality between the stress and strain rate is the viscosity. Thus, those liquids
maintaining a linear relationship or a constant viscosity are classified as Newtonian
liquids, which are named after Isaac Newton who reported the relationship. How-
ever, there are also many other liquids who maintain a much more complicated
nonlinear relationship between the stress and strain rate, such as honey, blood, and
some polymers. More complicated non-Newtonian stress-strain behaviors are stud-
ied in the area of rheology.
If the velocity keeps increasing, the inertial forces can increase to values way
above that of the viscous forces. Under such a condition, the viscous forces may be
negligible, leading to inviscid flow. However, even for inviscid flow, viscosity still
needs to be included for certain problems such as those involving solid boundaries:
viscosity often cannot be neglected near solid boundaries because the no-slip
condition can generate a thin region with large strain rates, i.e., boundary layer.
The standard equations of inviscid flow are the Euler equations. It is also common to
use the Euler equations in regions far from the boundary together with special
boundary layer equations to allow for viscosity in regions close to the solid bound-
ary. The Euler equations are a simplification of the N-S equations. The Euler
302 24 Fluid Dynamics

equations can be integrated along a streamline to get Bernoulli’s equation. When the
flow is irrotational and inviscid, i.e., potential flows, Bernoulli’s equation can be
used throughout the flow field.

Laminar and Turbulent Flow

As mentioned above, the discrete nature of the fluid flow can significantly deviate the
flow condition from what would be predicted using the N-S equations discretized at
large scales. This is because the randomness and chaos caused by the discrete nature
can change the flow patterns via couplings between different scales. The resultant
recirculation, eddies, and apparent randomness are characterized as turbulence.
Turbulence will become more obvious as the Reynolds number increases, which
accounts for both geometry and flow velocity. Flow without noticeable turbulence is
called laminar. When turbulence is noticeable to even dominant, the flow becomes
turbulent. In Fig. 24.3, a laminar flow comes from the left. When the flow encounters
an obstacle, i.e., porous fence, turbulent flow develops.
The above statement does not imply that the N-S equations are no longer valid in
turbulent flow but, instead, suggests that N-S equations need to be discretized to very
small spatial and temporal scales to capture the turbulence. This is essentially the
idea of the direct numerical simulation (DNS) (Moin and Mahesh 1998; Moser et al.
1999). Therefore, no turbulent models are needed in the DNS. For the purpose, all
the spatial scales of the turbulence must be resolved based on the computational
mesh, from the Kolmogorov microscales (smallest dissipative scales) (Kolmogorov
1962) to the integral scale L. For the time scale, the time step needs to be small
enough to ensure that a fluid particle moves only a fraction of the mesh spacing
within each step. Existing practice has enabled DNS to simulate turbulent flows at
moderate Reynolds numbers. The application is restricted by the power of the
computer and the efficiency of the solution algorithm. The DNS has been validated
against experimental data for some flows.
Other than the direct use of the N-S equations via the DNS, modeling flow with
significant turbulence with normal meshes/grids is also possible. This goal can be
achieved by introducing spatial or time averaging to decompose the flow into an

Fig. 24.3 Visualization of turbulence (behind a porous fence with 50% opening) [youtube.com/
watch?v¼bMO2IXmJa_Y]
Navier-Stokes Equations 303

average component and a perturbation component. The temporal decomposition,

i.e., Reynolds decomposition, is more common. Reynolds-averaged Navier-Stokes
(RANS) equations together with turbulent models provide an effective way to model
the turbulent flow. The additional momentum transfer and the consequent extra mass
and heat transfer are considered via the Reynolds stresses from turbulence. For space
averaging, the large eddy simulation (LES) (Sagaut 2006) ignores the smallest
length scales, which are the most computationally expensive to resolve, via
low-pass filtering of the N-S equations. This treatment can be viewed as a time
and spatial averaging. RANS and DES can be combined via the detached eddy
simulation (DES) (Spalart 2009).
Most flows of interest including those industrial applications have Reynolds
numbers far exceed which is affordable with the DNS. Therefore, the DNS is still
a high-end auxiliary tool, though it will possibly become more and more popular and
predominant as the available computational power multiplies in the incoming
decades. For example, any vehicle large enough to carry a human and moving faster
than 20 m/s will lead to a Reynolds number larger than four million, which is far
beyond the limit of DNS simulation. In the following section, the N-S equations as
the foundation of all the above simulation techniques will be introduced. Then focus
will be placed on the RANS and the turbulent models.

Navier-Stokes Equations

The Navier-Stokes equations, named after Claude-Louis Navier and George Gabriel
Stokes, describe the mass and momentum balance in moving viscous fluids. The
mass balance equation, also called continuity equation, shared the same mass
balance concept as those in groundwater and concentrations, but its formulation is
still much different. The momentum balance equation, similar to Navier’s equation
in the mechanical field, traces back to Newton’s second law. The major differences
are as follows:
1. The momentum N-S equation additionally considers advection.
2. The momentum N-S equation usually contains an accumulation (inertial) term
(time-dependent), while the Navier’s equation only carries this term in dynamic
problems.
3. The stress formulation is different: the stress in Navier’s equation (solid) is related
to the strain, while that in N-S equations is related to the strain rate. This leads to a
distinction in terminology – we talk about elastic/plastic in solids but viscous/
Newtonian/non-Newtonian in fluids.
The N-S equations can be simplified into the Stokes equations by removing the
inertial term in the momentum equation for Stokes flow with extremely low flow
velocity. On the other extreme, the N-S equations can be simplified into the Euler
equations by eliminating the viscous term for inviscid flow with a very high flow
velocity.
304 24 Fluid Dynamics

A basic solution to the Navier-Stokes equations includes a field for the flow
velocity and a field for the pressure. These two fields are highly coupled in the
governing equations and thus are considered as one field in many places. A heat or
energy balance equation and a concentration equation are also commonly added to
obtain a comprehensive understanding of the fluids and its constituents. Trajectories
of the fluid particles or solutes can be obtained once the flow velocity and its
temperature variation are obtained.
The N-S equations are established based on the following assumptions:
1. Mass and momentum are conserved. In fact, all balance laws can be expressed in
the following general form:

∂u
¼ ∇  ½uðv∂Ω  vÞ  ∇  J þ g, ð24:4Þ
∂t
where u is the conserved quantity which can be tensors of any order, v∂Ω is the
velocity of the moving boundary, J is the flux, and g is the source.
2. The fluid is continuous. Therefore, the dependent variables, i.e., flow velocity and
pressure, are continuous and differentiable, at least weakly.
3. The control volume remains fixed in space (Euler description) or can move with
the fluid (Lagrangian description). But the N-S equations that we commonly see
adopt a control volume fixed in space (Euler description).

Conservation of Mass

The mass conservation equation can be easily obtained from the above general
conservation equation by removing the flux and source terms in a computational
domain and the non-moving boundaries. Since mass is the conserved quantity, the
dependent variable will be the mass per unit volume, i.e., density. The general
conservation equation turns into

∂ρ
þ ∇  ðρvÞ ¼ 0, ð24:5Þ
∂t
where ρ is the mass density (mass per unit volume) and v is the flow velocity. In
incompressible flow, ρ is a constant and the above equation reduces to

∇  v ¼ 0: ð24:6Þ

According to the introduction in previous chapters, the above equation indicates

that the v field is an equivoluminal field.
Navier-Stokes Equations 305

Conservation of Momentum

The quantity to be conserved for the momentum balance is the momentum per unit
volume ρv. The J now is the negative of the Cauchy stress and the source is the body
force. Substituting ρv into the general balance equation, we obtain

∂ðρvÞ
þ ∇  ðρvvÞ ¼ ∇  σ þ f, ð24:7Þ
∂t

dy dz
where v ¼ ðu; v; wÞ ¼ dx dt ; dt ; dt , in which (x, y, z) are the coordinates of a fluid point
and f is a vector source: body force. Thus, when applying Newton’s second law to
this point or a REV, we will have the material derivative. However, the computation
domain is fixed as we use the Euler description. As a result, the material derivative
splits into the two terms on the left-hand side of the momentum balance equation.
For further derivations, let us first reformulate the left-hand side of the equation as
   
∂ρ ∂v
v þ ∇  ðρvÞ þ ρ þ v  ∇v ¼ ∇  σ þ f: ð24:8Þ
∂t ∂t

Recalling the mass balance equation, we have

 
∂v
ρ þ v  ∇v ¼ ∇  σ þ f: ð24:9Þ
∂t

We can also write the above equation using the material derivative as

Dv
ρ ¼ ∇  σ þ f: ð24:10Þ
Dt
As can be seen, the equation appears similar to the expression of Newton’s
second law (ma ¼ F). The sum of the two terms on the right-hand side of the
equation is the resultant force on the REV. In fact, this equation is the formulation of
Newton’s second law for continua, which thus can be applied to describe the
nonrelativistic momentum conservation of any continuum that conserves mass.
This so-called Cauchy momentum equation can be further derived into the general
N-S equation by decomposing the stress term. In fluids, it is common to differentiate
the pressure from the deviatoric stress:

σ ¼ σmI þ τ
0 1 0 1
σm 0 0 σ xx þ σ m τxy τxz
B C B C ð24:11Þ
¼ @ 0 σm 0 Aþ@ τyx σ yy þ σ m τyz A,
0 0 σm τzx τzy σ zz þ σ m
306 24 Fluid Dynamics

where I is the 3  3 identity matrix, τ is the deviatoric stress tensor, and σ m is the
mean normal stress, which is equal to the negative of the mechanical pressure
p (or hydrostatic pressure in the context of hydrodynamics):

1

σ m ¼ p ¼  σ xx þ σ yy þ σ zz : ð24:12Þ
3
The motivation for separating the two stress terms is that the pressure is typically
a variable of interest, and also, this treatment simplifies the later application to
specific fluid families because the rightmost tensor τ in the above equation must
be zero for a fluid at rest. The Navier-Stokes equation may now be written in the
most general form as
 
∂v
ρ þ v  ∇v ¼ ∇p þ ∇  τ þ f: ð24:13Þ
∂t

In the above equation, v, p, and τ are still unknown. Therefore, this equation is
still not the N-S equation that we can use to analyze practical problems. In fact, v and
p are dependent variables of the N-S equation. To derive solvable N-S equations, we
just need to find constitutive relationships to relate the stress τ to the velocity
(or pressure). This will be discussed in the following section for the N-S equations
for different fluids.

Navier-Stokes Equations for Different Fluids

The constitutive relationship to complete the general N-S equations is normally

established based on the viscous behavior of the fluid. Specific forms of the Navier-
Stokes equations will be obtained for different types of fluid flow. In the following,
the Newtonian fluids and their flow will be discussed as this is the most common
type. Non-Newtonian fluids will be briefly introduced as a more complicated and
rarer case in viscous fluids. Then Stokes equations for creeping flow and Euler
equations for the inviscid flow will be conveniently obtained by simplifying the
above N-S equations for viscous flow.

Viscous Flow of Newtonian Fluid

The constitutive relationship for Newtonian fluids is established with the following
three assumptions:
1. The fluid is isotropic.
2. The stress tensor is a linear function of the strain rate.
Navier-Stokes Equations for Different Fluids 307

3. For a fluid at rest, ∇  τ must be zero to ensure the existence of the hydrostatic
pressure.
These assumptions will lead to
 
∂ui ∂u j 2 ∂uk ∂uk
τij ¼ μ þ  δij þ δij λ , ð24:14Þ
∂x j ∂xi 3 ∂xk ∂xk

where μ is called the first coefficient of viscosity (usually just called “viscosity”) and
λ is the second coefficient of viscosity (related to the bulk viscosity). The value of λ
quantifies the viscous effect associated with the volume change and is very difficult
to determine, including its sign. This volume change effect could be negligible in
compressible flows but be significant in nearly incompressible flows, leading to a
matter of controversy. The most commonly used values of λ are 0 and 2μ/3.

Compressible Newtonian Fluid

Directly substituting the above constitutive relationship into the general N-S equa-
tions yields the N-S equations for the flow of compressible Newtonian fluids:
 
∂v   
ρ þ v  ∇v ¼ ∇p þ ∇  μ ∇v þ ð∇vÞT
∂t   ð24:15Þ
2μ
þ ∇  ∇  v þ ρg,
3

where p is used to represent the pressure and the gravity is considered as the only
body force, i.e., f ¼ ρg. The associated mass continuity equation is

∂ρ
þ ∇  ðρvÞ ¼ 0: ð24:16Þ
∂t
The above form of the N-S equations can cover the dynamics of most gasses and
liquids with a good equation of state and good functions for the dependence of
parameters such as viscosity on the variables.

Incompressible Newtonian Fluid

The N-S equations for incompressible Newtonian fluids can be derived from
Eqs. 24.15 and 24.16 for compressible fluids with two further simplifications:
1. The viscosity μ is a constant.
2. The second viscosity coefficient diminishes: λ ¼ 0.
308 24 Fluid Dynamics

As the fluid is incompressible, the mass balance equation is simply transformed into

∇  v ¼ 0: ð24:17Þ

Attention is needed for the deduction of the viscous term on the right-hand side of
the momentum balance equation. Taking the u component of the velocity v(u, v, w),
for example, we have
     
∂ ∂u 2μ ∂ ∂u ∂v ∂ ∂u ∂w
2μ  ∇v þ μ þ þ μ þ
∂x ∂x 3 ∂y ∂y ∂x ∂z ∂z ∂x
2 2 2 2 2
∂ u ∂ u ∂ v ∂ u ∂ w
¼ 2μ þμ 2 þμ þμ 2 þμ
∂x 2 ∂y ∂y∂x ∂z ∂z∂x
∂ u
2
∂ u
2
∂ u
2
∂ u ∂ v
2
∂ w
2 2 ð24:18Þ
¼μ þμ 2 þμ 2 þμ 2 þμ þμ
∂x2 ∂y ∂z ∂x ∂y∂x ∂z∂x
∂
¼ μ∇2 u þ μ ð∇  vÞ
∂x
¼ μ∇2 u:

Similarly, we can obtain μ∇2v and μ∇2w for the y and z directions. Then the N-S
equations for incompressible Newtonian fluids are

∇u¼0 ð24:19Þ
 
∂u
ρ þ u  ∇u ¼ ∇p þ μ∇2 u þ ρg: ð24:20Þ
∂t

It is also common to write the momentum equation in the following form

∂u
þ u  ∇u ¼ ∇e þ ν∇2 u þ g, ð24:21Þ
∂t
where e is the specific (per unit mass) thermodynamic work and ν ¼ μ/ρ is the
kinematic viscosity. e can also be used in the incompressible flow, in which the
following identity holds if a uniform density can be assured:
 
p 1
∇e ¼ ∇ ¼ ∇p: ð24:22Þ
ρ0 ρ0

Non-Newtonian Fluids

In non-Newtonian fluids, e.g., ketchup, custard, toothpaste, starch suspensions,

paint, blood, and shampoo, the viscosity is not necessarily dependent on the shear
rate or shear rate history. It is also possible that the viscosity is related to the
high-order temporal derivative of displacement. For those non-Newtonian fluids
Navier-Stokes Equations for Different Fluids 309

with a shear rate-dependent viscosity, the relation between the shear stress and the
shear rate is not linear anymore. The constitutive relationships for a few common and
simple non-Newtonian fluids are given here to provide basic information.
In Bingham fluids, the following relationship holds:

∂u 0 ; τ < τ0
¼ ð24:23Þ
∂y ðτ  τ0 Þ=μ; τ  τ0

Therefore, Bingham fluids, e.g., toothpaste and some saturated clays, are capable
of bearing some shear before they start flowing.
In addition, power laws are useful for approximating the behavior of many
common non-Newtonian fluids including shear thinning fluids, e.g., latex paint,
and shear thickening fluids, e.g., cornstarch-water mixture. This constitutive rela-
tionship has the following form:
 n
∂u
τ¼K : ð24:24Þ
∂y

Stokes Flow

Stokes flow, also termed creeping flow, is a type of fluid flow where advective
inertial forces are negligible compared with viscous forces. It is the flow condition in
pore water. Removing the inertial term from the momentum balance equation, the
N-S equations then reduce to the Stokes equations:

∂ρ
þ ∇  ðρvÞ ¼ 0, ð24:25Þ
∂t

0 ¼ ∇σ m þ ∇  τ þ f ¼ ∇p þ μ∇2 u þ ρg: ð24:26Þ

Figure 24.4 shows the Stokes flow passes a solid particle. This is the normal flow
condition in granular materials without other interphase phenomena such as double
layers.

Inviscid Flow

In fluid dynamics, the Euler equations are a set of quasilinear hyperbolic equations
governing adiabatic and inviscid flow. The viscous effect is negligible in the inviscid
flow. Accordingly, the incompressible Euler equations are as follows:
310 24 Fluid Dynamics

Fig. 24.4 Flow field of the Stokes flow against of solid obstacle

Fig. 24.5 Flow field of the inviscid flow against of solid obstacle

∂ρ
þ ∇  ðρvÞ ¼ 0, ð24:27Þ
∂t
 
∂v
ρ þ v  ∇v ¼ ∇p þ ρg: ð24:28Þ
∂t

Figure 24.5 demonstrates the inviscid flow passing a solid particle. The exclusion
of the viscous term, which functions as a diffusion mechanism in momentum transfer,
will raise the level of difficulty in numerically solving the equations. The solution in
Fig. 24.5 exhibits the instability and error in the area behind the solid particle.

Turbulent Models
Reynolds Number: From Laminar to Turbulent

The Reynolds number can be defined in different ways by incorporating fluid

properties including the density, viscosity, flow velocity, and a characteristic length
Turbulent Models 311

(or characteristic dimension). However, the choice of this length is a matter of

convention. For example, the radius or diameter is usually chosen to describe spheres
or circles, while the length or width is commonly used for aircrafts or ships. An
equivalent diameter is usually defined for non-spherical objects. More complicated
rules need to be applied in fluids with a variable density such as non-Newtonian
fluids. Mathematically, the Reynolds number is usually defined as below:

inertial forces ρvL vL

Re ¼ ¼ ¼ , ð24:29Þ
viscous forces μ ν

where L is a characteristic linear dimension.

For “fully developed” flow in a pipe with a diameter, D, experimental observa-
tions show that laminar flow occurs when Re ¼ ρvD μ < 2100 and turbulent flow occurs
when Re > 4000. “Transition” flows exist in the Re interval between 2300 and 4000,
in which both laminar and turbulent flows are possible, depending on other factors
such as pipe roughness and flow uniformity. Figure 24.6 shows the transition from a
laminar flow to a turbulent flow as Reynolds number increases. However, it is
worthwhile to point out that the Reynolds number in the figure is different from
what is defined in a pipe due to the existence of the obstacles and the channel.
Therefore, the threshold values are also different.
The primary use of the Reynolds number is to evaluate the relative significance of
inertial and viscous forces in fluid flow and to differentiate laminar flow from
turbulent flow based on that. Additionally, Reynolds numbers are very useful for
scaling of fluid dynamics problems to assess the dynamic similitude between two
different fluid flows.

Simulation of Turbulent Flows

Turbulent flow occurs in the high Reynolds number range. Consequently, this type
of flow is dominated by inertial forces and tends to produce chaotic eddies, vortices,
and other flow instabilities. In fact, most fluid flows occurring in nature are turbulent
to some extent. Turbulence can be described as a state of continuous instability in the
flow and is characterized by the irregularity in the flow, increased diffusivity, and
energy dissipation. Turbulent flow always has 3D and time-dependent characteris-
tics, even if the boundary conditions of the flow do not change with time.
As introduced in the previous section, the key idea to separate turbulence from the
mean flow is averaging, for which the following three different ways exist:
1. Time averaging in a fixed point of space, for stationary turbulence
2. Space averaging for a fixed moment in time in the case of homogeneous
turbulence
3. Ensemble averaging for a series of identical experiments (the most general form
of averaging)
312 24 Fluid Dynamics

Fig. 24.6 Transition from laminar to turbulent flows: the characteristic length is the width of the
flow channel (obtained using Energy2D web tool [http://energy.concord.org/energy2d/reynolds.
html])

All these methods can appear in two forms, unweighted (Reynolds) and
weighted. The Reynolds averaging technique operates as follows:
0
ϕðx; t Þ ¼ ϕðx; t Þ þ ϕ ðx; t Þ, ð24:30Þ

where ϕ(x, t) is decomposed into an average term ϕðx; t Þ and an oscillating term ϕ'(x, t).
ϕ'(x, t) denotes the fluctuation around the mean value, which is defined as

1 XN
ϕðx; t Þ ¼ lim ϕi ðx; t Þ, ð24:31Þ
N!1 N
i¼1

where N is the number of identically performed experiments.

Turbulent Models 313

Applying the above averaging procedure to the incompressible N-S equations, the
following form of the averaged equations can be obtained:

∇  v
¼ 0, ð24:32Þ
∂

v


þ ∇  v
v
¼ g  ∇
v þ v 0 v0 ,
p þ ∇  ν∇
ð24:33Þ
∂t
where the term v0 v0 is called the Reynolds stress tensor. In order to close the system,
further modeling is necessary to obtain the Reynolds stress.
The goal of Reynolds-averaged turbulence modeling is to express the Reynolds
stress tensor in terms of the known quantities. There are two widely accepted
approaches (Wilcox 1998). The first approach formulates and solves the transport
equation for the Reynolds stress tensor (and higher moments, depending on the order
of the closure). It is, however, still necessary to model some of the terms, since the
number of unknowns increases faster than the number of equations. The second and
a more popular approach prescribes a relationship between the Reynolds stress and
the mean velocity gradient. Although models prescribing nonlinear relations have
been proposed recently, the most popular approach is to use the Boussinesq approx-
imation (Boussinesq eddy viscosity hypothesis) (Schmitt 2007), which adopts a
linear relation of the form
h
T i 2
v0 v0 ¼ vt ∇

v þ ∇

v þ kI, ð24:34Þ
3
1
where k ¼ v0  v0 .
2
The kinematic eddy viscosity vt can be evaluated in many different ways, ranging
from algebraic relations and local equilibrium assumptions to the solution of trans-
port equations. The most popular approach is to express vt as a function of the
turbulent kinetic energy k and its dissipation rate ε, leading to a “two-equation”
turbulence model:

k2
vt ¼ C μ , ð24:35Þ
ε
ε ¼ vv0 v0 : ∇v0 : ð24:36Þ

In this type of turbulent model, k and ε are obtained by the solution of their
respective transport equations. The first model of this kind was proposed by Harlow
and Nakayama (1968). Derivation of the transport equations for k and ε can be found
in Launder and Spalding (1974). In fact, a wide variety of k-ε models exist, while the
most noteworthy one is the standard k-ε model proposed by Launder and Spalding
(1974) and the RNG k-ε variant proposed by Yakhot and Orszag (1986) and Yakhot
et al. (1992).
314 24 Fluid Dynamics

Standard k-ε Turbulence Model

The exact k-ε equations contain many unknown and unmeasurable terms. For a
much more practical approach, the standard k-ε turbulence model (Launder and
Spalding 1974) is used. This model was established based on our best understanding
of the relevant processes and thus minimizes the number of unknowns and leads to a
set of equations which can be applied to a large number of turbulent applications.
For turbulent kinetic energy k, we have
 
∂ðρkÞ ∂ðρkvi Þ ∂ μt ∂k
þ ¼ þ 2μt E ij E ij  ρε: ð24:37Þ
∂t ∂xi ∂x j σ k ∂x j

For dissipation ε, a similar transport equation can be used:

 
∂ðρεÞ ∂ðρεvi Þ ∂ μt ∂ε ε ε2
þ ¼ þ C1ε 2μt E ij E ij  C 2ε ρ , ð24:38Þ
∂t ∂xi ∂x j σ ε ∂x j k k

where vi is the velocity component in the corresponding direction; Eij is the compo-
nent of the rate of deformation; and μt is the eddy viscosity, which can be calculated
2
as μt ¼ ρC μ kε .
The above equation states the following balance:
Rate of change of k or ε+ transport of k or ε by convection ¼ transport of k (or ε)
by diffusion + rate of production of k (or ε) – rate of the destruction of k (or ε).
Equations 24.37 and 24.38 also consist of some adjustable constants: σ k, σ ε, C1ε,
and C2ε. The values of these constants have been obtained by numerous iterations of
data fitting for a wide range of turbulent flows. A common set of values are as
follows: Cμ ¼ 0.09, σ k ¼ 1.00, σ ε ¼ 1.30, C1ε ¼ 1.44, and C2ε ¼ 1.92.

Practice Problem

The section introduces a benchmark problem for 2D transient computational fluid

dynamics, which was described in Sharma (2012).

Problem Description

In a 1 m by 1 m square liquid container as shown in Fig. 24.7, the liquid inside the
container moves as the lid on the top boundary continuously moves from the left to
the right at a velocity of 1 m/s. Assuming the water on the top boundary moves
at the same speed as the lid, please predict the distribution of vx along x ¼ 0.5 m and
Practice Problem 315

Fig. 24.7 Computational

domain and boundary
conditions

Fig. 24.8 x-velocity along

the y-axis of the domain 1.0
(x ¼ 0.5 m)

0.8

0.6
y

0.4

Sharma (2012)
0.2 New Simulation

0.0

-0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2

vx

vy along y ¼ 0.5 m, and compare them against the numerical results of Ghia et al.
(1982). The kinematic viscosity is 10.

Results

The results obtained by the numerical simulation implemented with FlexPDE are
compared against the results given in Sharma (2012). As shown in Figs. 24.8 and
24.9, a simple CFD implementation with a very coarse mesh and the finite element
method can also generate acceptable results. Better comparisons can be obtained by
further tuning the numerical model and solution techniques.
316 24 Fluid Dynamics

Fig. 24.9 y-velocity along 0.20

the x-axis of the domain
(y ¼ 0.5 m) 0.15
Sharma (2012)
0.10 New

0.05

vy
0.00

-0.05

-0.10

-0.15

-0.20
0.0 0.2 0.4 0.6 0.8 1.0
x

Appendix: FlexPDE Script File

{ 2D CFD benchmark, Sharma, 2012, EPJ Web of Conferences 24}
TITLE 'Transient CFD-Incompressible with penalty'
COORDINATES CARTESIAN2
VARIABLES
v(1e-8)=VECTOR(vx,vy)
p(1e-8)
SELECT
ERRLIM 0.01
DEFINITIONS
rho=1
nu=10
INITIAL VALUES
p=0 v=VECTOR(0,0)
EQUATIONS
p: DIV(v)=-10e-8*DT(p)
v: -nu*DIV(GRAD(v))+DOT(v,GRAD(v))+GRAD(p)+DT(v)=0
BOUNDARIES
REGION 1
START(0,0)
VALUE(vx)=0
VALUE(vy)=0
LINE TO (1,0)
VALUE(vx)=0
VALUE(vy)=0
LINE TO (1,1)
VALUE(vx)=1
VALUE(vy)=0
LINE TO (0,1)
VALUE(vx)=0
VALUE(vy)=0
LINE TO CLOSE
Practice Problem 317

TIME 0 TO 1
MONITORS
FOR CYCLE=1
VECTOR(v)
CONTOUR(p)
PLOTS
FOR CYCLE=1
VECTOR(v)
CONTOUR(p)
ELEVATION(vx) FROM (0.5,0) TO (0.5,1)
ELEVATION(vy) FROM (0,0.5) TO (1,0.5)
END
Chapter 25
Hydrodynomechanics: Fluid-Structure
Interaction

Introduction

The term hydrodynomechanics refers to any phenomenon involving the mechanical

response of a solid and the mass and momentum transfer in a liquid. In a more
general sense, the scope can be extended from liquids to fluids. In this case,
hydrodynomechanics can refer to, or more accurately, be related to three application
areas: dynamic poroelasticity, acoustic-structure interaction, and complete fluid-
structure interaction. The relationship between hydrodynomechanics and these
areas are very complicated because the relevant applications were conducted and
named more due to practical and historical interests rather than from the perspective
of multiphysics. The following figure gives out a very typical example of fluid-
structure interaction in petroleum engineering. Similar examples are the wind tur-
bines and their foundations in lakes (Fig. 25.1).
Dynamic poroelasticity is the hydrodynamics occurring on the level of REV. It is
the dynamic extension to poromechanics. By contrast, the broad definition of the
fluid-structure interaction, including the acoustic-structure interaction and the com-
plete fluid-structure interaction, involves solid and liquid interaction on the level of
regions within the computational domain. From the perspective of the physical field,
it is very similar to the dynamic poromechanics as it also involves the
elastodynamics in the solid and acoustics in the fluid. However, they have different
natures and thus appear as different types of multiphysics. As illustrated in Fig. 25.2,
the dynamic poromechanics treats the porous material as an ensemble of REVs, each
of which is composed of a liquid fraction and a solid fraction. Therefore, the solid
and liquid are coupled at the scale of REV, which should be much smaller than the
scale of the computational domain. Accordingly, the two fields are coupled on the
level of mathematical equations. By contrast, fluid-structure interaction allows for
coupling between solid and fluid regions which are at the scale of the computational
domain. Therefore, the solid and liquid are coupled on the boundary between the
solid and liquid regions within the computational domain. Different types of

© Springer International Publishing AG, part of Springer Nature 2018 319

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_25
320 25 Hydrodynomechanics: Fluid-Structure Interaction

Fig. 25.1 Offshore oil platforms as a typical FSI example

Fig. 25.2 Two types of hydrodynomechanics

equation systems will be used for different regions, depending on whether the region
consists of a solid or a fluid. The coupling between different regions is achieved by
ensuring the mechanical balance or continuity.
In this chapter, hydrodynomechanics will be introduced as three distinct areas:
dynamic poroelasticity, acoustic-structure interaction, and complete fluid-structure
interaction. The theory of dynamic poroelasticity has been given in poroelasticity, so
this chapter only includes a short section to offer more details regarding its imple-
mentation and application. For the complete fluid-structure interaction, we will show
the idea of region coupling and give detailed procedures for implementing such
coupling. Next, the acoustic-structure interaction will be introduced in a similar way.
Dynamic Poroelasticity 321

The differences and connections between the complete fluid-structure interaction

and acoustic-structure interaction problems will be pointed out both in theory and in
their applications. A simple example to show the implementation of the complete
fluid-structure interaction will be presented in the final.

Dynamic Poroelasticity

The first application area of hydrodynomechanics is related to the dynamic

poromechanics introduced in Chapter 20 for poroelasticity. However, from a differ-
ent angle, it would be more accurate to state that dynamic poromechanics has a major
overlap with hydrodynomechanics rather than to treat it as a special case of
hydrodynomechanics. In fact, the complete equations in the dynamic poromechanics
consist of momentum balance equations for both solid and liquid, which are mixed
in any REV. The momentum balance equation for the solid is not merely the
equation for the mechanical field (static) but also includes the inertial term. There-
fore, it is the elastodynamics, or plastodynamics if plastic constitutive relationships
are involved (which is rare). However, the momentum equation for the fluid is a
wave propagation equation. We know that the wave equation in liquids can be
obtained by degenerating the comprehensive Navier-Stokes equations. Therefore,
the fluid part can still be described as hydrodynamics if acoustics in fluids can be
viewed as a part of fluid mechanics in a broad sense. The governing equations for
both the solid and liquid take the form of wave equations. These wave equations
describe the mechanical waves in the solid and liquid. From this viewpoint, dynamic
poroelasticity is the coupling between acoustic fields in two well-mixed phases.
Despite the similar nature, the wave propagation equations in solids and liquids have
similar nature but still differentiate in a few places. The major differences reside in
the constitutive relationships: solids resist shear, while fluids do not. It is also
worthwhile to mention that, in a broad sense, acoustics can also include wave
propagation in solids.
The governing equations and constitutive relationships for dynamic poroelasticity
have been introduced in Chap. 20. However, the practical treatment could be more
complicated than the original equation system. The pressure fluctuations in the
saturating fluid interact with different elastic waves (shear and longitudinal) in the
porous material. As the pores are often small, the energy losses introduce damping to
the system. Moreover, the porous domain can be coupled to the pure solid structures
or pure fluid domains, introducing a multitude of multiphysics couplings.
There are several ways of describing the propagation of waves in porous mate-
rials. This first way is to directly solve the original equation system. This way
accounts for the coupled propagation of elastic waves in the elastic porous matrix
and pressure waves in the saturating pore fluid. This equation system can incorporate
the damping effect of the pore fluid. A simpler approach is to treat the porous matrix
and saturating pore fluid as a homogenized equivalent fluid. In this case, only the
pressure waves are modeled with the consideration of losses. The loss models can
322 25 Hydrodynomechanics: Fluid-Structure Interaction

range from pure empirical models based on measurements to analytical or

semi-analytical phenomenological models; examples include the Delany-Bazley-
Miki (Jeong 2011) and the Johnson-Champoux-Allard models (Doutres et al.
2010). However, if treated in this way, the simplified physical process is more like
a monolithic physical process.

Complete Fluid-Structure Interaction

Mathematical Formulation

The above introduction indicates that the key component in the mathematical
description of the fluid-structure interaction (FSI) is essentially not the governing
equations. The governing equations for both elastodynamics and fluid dynamics/
acoustics have been introduced in Part III of this book. For the solid region, or
structure, the governing mechanisms can be formulated using the Navier’s equation
with the inertial term:

€ s ¼ ∇  σs þ f s ,
ρs u ð25:1Þ

where σs is the stress tensor, fs is the body force, ρs is the density of the solid, us the
displacement vector, and u €s is the second temporal derivative of the displacement.
The governing equation for the fluid is the Navier-Stokes equations:

∂ρ f
þ ∇  ðρ f v f Þ ¼ 0 ð25:2Þ
∂t
 
∂v f
ρ þ v f  ∇v f ¼ ∇  σ f þ f f : ð25:3Þ
∂t

The major physical mechanism to ensure the coupling of the fluid and solid
regions is to satisfy the following conditions on the boundaries between the two
regions (also frequently called interface or internal boundary), Γ:

u_ s ¼ v f , ð25:4Þ

and

σs  ns ¼ σ f  n f : ð25:5Þ

The first equation ensures the mass conservation or no separation, while the second
ensures the momentum balance or mechanical equilibrium. The first condition
(Eq. 25.4) can be replaced by the following equation when the profile of the
boundary between the fluid and solid regions is smooth in both time and space:

us ¼ u f : ð25:6Þ
Complete Fluid-Structure Interaction 323

According to the way to achieve the above couplings, the complete FSI tech-
niques can be categorized into two classes: conforming mesh methods and
nonconforming mesh methods. The conforming mesh methods track the location
of the interface and enforce Eqs. 25.5 and 25.6 explicitly. Gradual updates on the
mesh during the solution process are usually needed as the boundary between the
two regions is displaced, and therefore, the mesh within each region needs to
conform with these boundary displacements. This type of method can be
implemented conveniently with existing FDM, FVM, or FEM programs. A simple
approach is to first enforce the nonslip boundary condition for the fluid on the fluid-
solid boundary:

u f ¼ 0: ð25:7Þ

On the solid side, a Neumann boundary condition can be prescribed so that the
normal traction on the solid boundary is equal to the fluid pressure:

σs  ns ¼ σ f  n f ¼ pns , ð25:8Þ

where p is always positive and the negative sign in front of p is used to ensure that
the pressure acts as a compressive stress on the solid along its boundary with the
fluid.
The location of the boundary between the two regions is tracked using a moving
mesh algorithm, which is available in many numerical simulation programs these
days. The moving boundary is induced by the fluid pressure acting on the deform-
able solid structure. Therefore, the mesh displacement on the boundary between the
two regions is equal to the displacement of the solid,

umesh ¼ u on Γ ð25:9Þ

The movement of the mesh within the fluid and solid regions can be obtained
based on the boundary movement. A common way is to use the diffusion equation,
which “diffuses” the boundary displacement into the interior of each region:

∇2 umesh ¼ 0: ð25:10Þ

On these boundaries where no boundary displacement occurs, a Neumann boundary

condition can be used:

n  ∇umesh ¼ 0: ð25:11Þ

In addition, it is also common to use the velocity of the mesh movement as either the
dependent variable or boundary conditions for this mesh diffusion field. There are
still other ways to consider the moving mesh, which will not be detailed here. The
equation systems coupled in the above way can be solved either implicitly or
explicitly (partitioned method).
The second category is the nonconforming mesh methods, most notably the
immersed boundary method (Peskin 1977, 2002). This type of method requires
324 25 Hydrodynomechanics: Fluid-Structure Interaction

many treatments in addition to the existing discretization. These methods usually

enforce the velocity consistency on the boundary between the two regions instead of
the displacement consistency. More details can be found in Hou et al. (2012).

Application

The complete fluid-structure interaction (FSI) involves the fluid dynamics and the
mechanical field. This type of coupled phenomenon encompasses any process
involving a deformable or moving structure and a surrounding or internal fluid
flow. Slightly different from the acoustic-structure interaction, the complete FSI
theoretically involves any interaction between the fluid and solid structure, although
the focus is more placed on low frequency and large-strain applications.
One typical example is the flow around a solid obstacle as shown in Fig. 25.3. As
the fluid flow encounters a structure, the structure is deformed due to the pressure
exerted by the moving fluid. If the magnitude and variation of the deformations of
the structure are small, the fluid flow will not be greatly affected by the resistance
force from the solid. Due to this reason, we can simplify this problem as a one-way
problem in which the stresses result from the force exerted by the fluid on the solid is
a major concern.
However, if this is a relatively soft plate instead of a rigid body, then a large
deformation of the solid structure will be caused. This large deformation will in
return significantly impact the flow condition. This is what is shown in the above
figure, in which two-way fluid-structure interaction was simulated. In a turbulent
flow, the force on the solid will vary, leading to changes in the deformation of the
plates. Consequently, a very strongly-coupled two-way interaction can be produced.
It is also possible to have a condition in which the magnitude of the deformation
of the solid is small, while the time variation is large, for example, when the major

Fig. 25.3 A flexible

obstacle in a fluid channel is
bending under the viscous
and pressure forces imposed
by the fluid (flow
streamlines are colored by
flow velocity; deformed
beam is colored by the
deformation along the flow
direction; contours are for
flow velocity distribution)
[COMSOL example library]
Acoustic-Structure Interaction 325

frequency is larger than several Hz. Under such a condition, these small structural
deformations may still lead to pressure waves in the fluid. These pressure waves
result in the radiation of sound from vibrating structures. Such problems can be
treated as an acoustic-structure interaction using a wave equation rather than a
complete fluid-structure interaction with the N-S equations.

Acoustic-Structure Interaction

In the acoustic-structure interaction (ASI), we have an acoustic field in a fluid, either

a liquid or a gas, and a mechanical field in a solid which is in contact with the fluid.
There are two major differences between the ASI and the complete FSI. First, the
N-S equations in the fluid region degenerate into a wave equation. Second, the
displacement of the solid structure is generally small due to the magnitude of the
disturbance, while the frequency is higher than that in complete FSI. These two
differences lead to the different treatments in the ASI.
The governing equation for the fluid is the mechanical wave equation. We here
use the wave equation with the pressure as the dependent variable to facilitate the
following formulation of coupling,

1
∇2 p f  €p ¼ 0, ð25:12Þ
c2 f
where c is the wave velocity.
The governing equation of the solid is still the Navier’s equation with the inertial
term:

€ s ¼ ∇  σs þ f s :
ρs u ð25:13Þ

As introduced in the previous section, the couplings for FSI require the consis-
tency of the displacement (or velocity) and stress on the boundary between the fluid
and solid. In acoustic-structure interaction problems, the first requirement is auto-
matically met due to the negligible solid deformation. Therefore, we just need to
ensure that the stresses on two sides of the fluid-solid boundary are balanced.
Mathematically, this corresponds to the following two equations:

n f  ∇p f ¼ ρ€
u ð25:14Þ

ns  ∇σs ¼ pn f : ð25:15Þ

The first equation represents a Neumann boundary condition for the fluid on the
boundary, whereas the latter prescribes a Neumann boundary condition for the solid
on the boundary. The two equations will generate a very strong coupling between the
two fields by placing the dependent variable of one field into the matrix of the other
in the process of discretization. Other ways to implement the above two equations,
326 25 Hydrodynomechanics: Fluid-Structure Interaction

for example, treating them as Dirichlet boundary conditions by switching the terms
on two sides of the equations, are also possible. These ways may require more
bottom-level modifications to the discretization and solution procedures.

Application

The acoustic-structure interaction creates sound. In general, acoustics aim to under-

stand the small and rapid changes in the acoustic pressure above and below the
atmospheric pressure, which are usually described as pressure waves propagating in
space and time. The ASI problems involve two physical fields: the acoustic field and
mechanical field, if they are considered different fields. The coupling between the
two fields can be both one-way and two-way, depending on the properties of the
fluids and solids and the purpose of the analysis. Two-way couplings occur when
both the acoustic pressure waves in the fluid and the vibrations of the solid are strong
enough to affect each other significantly.
It is very common that a sound is generated as a moving structural object disturbs
a gas or a liquid carrying the acoustic pressure (or sound) wave. The pressure wave
in the fluid also produces vibrations in the solid. The interaction is by nature
two-way. However, in many cases, the interaction in one direction is more dominant.
The above one-way interaction can be observed in daily life. One example is a bass
speaker, which moves its cone back and forth to compress and decompress the air to
generate low-frequency sound. In the reserve direction, typical examples are the
acoustic pressure waves that have been employed to produce vibrations in a solid for
applications such as ultrasound imaging or nondestructive impedance testing.
ASI as a multiphysical process can further involve the thermal field. This is
similar to the energy dissipation in which mechanical energy is converted into
thermal energy. The difference is that here we focus on the energy associated with
the waves. The energy can dissipate in two ways: via the viscosity in the fluids and
via the attenuation associated with the damping in the solid. The viscous forces and
damping can cause significant energy dissipation (or loss) near the walls of the
structure in the acoustic boundary layer. In classic acoustics, the fluid is usually
assumed to be inviscid and isentropic. Accordingly, slip boundary conditions are
used on the boundaries between the solid and fluid, leading to solely coupling
normal to the surface. By contrast, if the viscous and thermal losses are considered,
both a mechanical and a thermal condition must be specified at the solid surface
together with a nonslip condition to ensure continuity in the fluid velocity. That is,
the movement of the solid is equal to the movement of the fluid. This leads to tightly
coupled acoustic and mechanical fields.
Practice Problem 327

Practice Problem

This section is about a very typical fluid-structure interaction problem in which a

deformable solid obstacle is placed in a flow channel.

Problem Description

Figure 25.4 shows the cross-section of a flow channel in which water flows from
the left to the right. The water flow velocity on the left boundary has a profile of 0.5y
(H  y)/H2. The right boundary has a pressure of zero and no shear stress. The top
and bottom sides are the walls of the flow channel, which can be approximated
using the nonslip boundary condition. The flow channel has a length of 3 m (L ) and
a width (or height) of 1 m (H ). An obstacle with a length of H/4 and a width of L/40
is fixed to the bottom wall at a location where the back face of the obstacle is L/10
from the left boundary. The deformation obstacle has the following mechanical
properties: λ ¼ 107 Pa and G ¼ 107 Pa (Lamé constants). Please simulate the fluid
flow and solid deformation in the process between the time when water starts
moving into the channel and the time when the front of the flow passes the back
face of the obstacle.

Governing Equation and Moving Mesh

In this example, the impact of the fluid movement on the solid deformation is
achieved via the balance of the normal stresses (pressures) on two sides of inter-
boundary: the solid and fluid regions. The influence of the solid deformation on the
water flow is achieved by means of moving mesh. The solid and fluid regions are
only analyzed using their own governing equations. Accordingly, in FlexPDE, the
N-S equation is inactivated in the solid region using the command “INACTIVE,”

Fig. 25.4 Computational

domain and boundary
conditions
328 25 Hydrodynomechanics: Fluid-Structure Interaction

and likewise; the Navier’s equation is inactivated in the fluid region. Four governing
equations are needed as follows:

p: rho=K ∗ DTðpÞ þ DIVðrho∗ vÞ ¼ 0

v: DTðrho∗ vÞ þ DOTðrho∗ v; GRADðvÞÞ
¼ GRADðpÞ þ mu∗ DIVðGRADðvÞÞ þ 1=3∗ mu∗ GRADðDIVðvÞÞ
xm: DIVðGRADðxmÞÞ ¼ 0
u: DIVðsigma tensorÞ ¼ VECTORð0; 0Þ

The third equation is the governing equation for the moving mesh. This diffusion
equation is used to predict the movement of the mesh within the domain based on the
moving boundary.
Unfortunately, the above governing equation system contains six PDEs according
to FlexPDE’ rules (a vector in 2D is counted as two), which exceeds the PDE number
limit in the student version. We can remove the displacement along the vertical
direction, i.e., uy, to enable the implementation of the model in the complimentary
student version. Please simulate the above problem without uy if you only have
access to the student version. Here, the script file for the complete implementation of
the above governing equation system is provided at the end of this chapter. The
removal of uy can be conveniently carried out, which is also a good practice for
deepening the understanding of the mechanical field.

Results

Figures 25.5 and 25.6 illustrate the displacement of the obstacle and the velocity
field at t ¼ 104 s. At this time, the front of the flow has not reached the front face of
the obstacle yet. However, the increase in the pressure has already led to small
displacements of the obstacle, which are not visible without magnification. Real
deformation on the obstacle can be clearly observed at t ¼ 2  104 s, when the front
of the water flow is about to reach the back face of the obstacle, as shown in
Fig. 25.7. The water pressure distribution explains the reason for this deformation:
the pressure increases on the front face is much higher than that on the back face. It is
noted that the accuracy of the simulation presented here may not be high due to the
nonlinearity residing in the problem. Better discretization and solution techniques
are desired for more satisfactory simulation results. We here just intend to use a very
simple simulation to demonstrate the idea of fluid-structure interaction.
Practice Problem 329

Fig. 25.5 Displacement field in the solid obstacle (arrows) at t ¼ 0.1 ms

Fig. 25.6 Water velocity field (arrows) at t ¼ 0.1 ms

330 25 Hydrodynomechanics: Fluid-Structure Interaction

Fig. 25.7 Pressure distribution and deformed obstacle at t ¼ 0.2 ms

Appendix: FlexPDE Script File

{ Full Fluid-Structure Interaction: Flow through a channel with a

flexible obstacle }
TITLE 'Full FSI' { the problem identification }
COORDINATES CARTESIAN2 { coordinate system, 1D,2D,3D, etc. }
SELECT
ERRLIM 0.005
VARIABLES
v(1e-8)=VECTOR(vx,vy)
p(1e-8)
xm=move(x)
u(1e-8)=VECTOR(ux,uy)
!SELECT
!NGRID =18
DEFINITIONS
G=1e7
lambda=1e7
C=TENSOR((2*G+lambda,lambda,0),(lambda,2*G+lambda,0),(0,0,G))
! Stiffness matrix
epsilon=VECTOR( dx(ux),dy(uy),0.5*(dx(uy)+dy(ux)) )
Practice Problem 331

sigma=dot( C,epsilon)
{ The above C and epsilon are defined using the Voigt notation;
sigma_tensor uses the tensor notation }
sigma_tensor=TENSOR((XCOMP(sigma),ZCOMP(sigma)),(ZCOMP
(sigma),YCOMP(sigma)))
H=1
L=3
L_front=L/10 ! x position of the front wall of the obstacle
rho=1000
mu=0.001
K=1/4.4e-10
INITIAL VALUES
p=0 ! only consider hydrodynamic pressure: gravity is excluded,
rho*9.81*(10-y)
v=VECTOR(0,0)
EQUATIONS
p: rho/K*DT(p)+DIV(rho*v)=0 !rho/K* !1e-3*0.001/0.01^2*div(grad
(p)) !VECTOR(0,0)
v: DT(rho*v)+DOT(rho*v,GRAD(v))=-GRAD(p)+mu*DIV(GRAD(v))+1/
3*mu*GRAD(DIV(v))
xm: DIV(GRAD(xm))=0
u: DIV(sigma_tensor)=VECTOR(0,0) ! CONSTRAINTS { Integral
constraints }
BOUNDARIES { The domain definition }
REGION 1 { For each material region }
INACTIVE(u)
START(0,0) { Walk the domain boundary }
!NATURAL(p)=0
VALUE(vx)=0.5*y*(H-y)/H^2
VALUE(vy)=0
VELOCITY(xm)=0
LINE TO (0,H)
!NATURAL(vy)=0
VALUE(vx)=0
VALUE(vy)=0
LINE TO (L,H)
VALUE(p)=0 { This is a key. Without it, the p distribution will be
wrong.
The reason is unknown, possibly due to FlexPDE BC treatment. }
NATURAL(vx)=0
LINE TO (L,0)
VALUE(vx)=0
VALUE(vy)=0
LINE TO CLOSE
REGION 2
INACTIVE(v,p)
START (L_front,0)
NATURAL(ux)=-p
VALUE(uy)=0
VALUE(vx)=0
VALUE(vy)=0
VALUE(xm)=L_front+ux
LINE TO (L_front,H/4)
332 25 Hydrodynomechanics: Fluid-Structure Interaction

NATURAL(ux)=0
NATURAL(uy)=-p
VALUE(vx)=0
VALUE(vy)=0
NATURAL(xm)=0
LINE TO (L_front-L/40,H/4)
NATURAL(ux)=p
VALUE(uy)=0
VALUE(vx)=0
VALUE(vy)=0
VALUE(xm)=L_front-L/40+ux
LINE TO (L_front-L/40,0)
VALUE(ux)=0
VALUE(uy)=0
NATURAL(xm)=0
LINE TO CLOSE
TIME 0 TO 1
MONITORS
FOR CYCLE=1
VECTOR(v)
CONTOUR(p) ZOOM(0,0,L_front*2,0.35*H)
HISTORY(p) AT (0,H/2)
ELEVATION(p) FROM (0,0) TO (0,H)
ELEVATION(p) FROM (H/2,0) TO (H/2,H)
VECTOR(u) ZOOM(0,0,L_front*2,0.35*H)
PLOTS
FOR CYCLE=1
VECTOR(v)
CONTOUR(p)
END
Chapter 26
Thermoelectromagnetics

Introduction

Thermoelectromagnetics covers the multiphysics involving both the thermal field

and the electromagnetic fields. This chapter introduces three major mechanisms in
thermoelectromagnetics: Joule heating, dielectric heating, and induction heating. For
each mechanism, the introduction will be carried out in the sequence of basic
mechanisms, mathematical formulations, and applications. Additional knowledge
that is needed for the implementation of thermoelectromagnetics involving these
mechanisms, such as dielectric heating types, skin depth, and Poynting vector, will
also be introduced. A practice problem will be presented in the last section for
simulating microwave heating in a waveguide.

Joule Heating

Joule heating (also referred to as resistive or Ohmic heating) describes the process
where the energy of an electric current is converted into heat as it flows through a
resistive conductor. Joule heating is caused by the interactions between the moving
particles constituting the current, which are, usually but not always, electrons, and
the atomic ions constituting the body of the conductor. The moving particles
accelerated by an electric field transfer some of their kinetic energy to the conductor
ions via collisions. This energy transfer will lead to an increase in the kinetic energy
of the conductor ions and consequently a temperature increase in the conductor. Via
such a process, energy is transferred from the external electric field to the thermal
field (Barrett and Cardello 2012). This is mostly true as collisions between the
moving particles in the current and the conductor ions are inevitable, which exhibit
as the fact that electric resistivity almost always exists at the macroscopic scale.
Figure 26.1 clearly exhibits the heat generated by a circuit using a thermal image.

© Springer International Publishing AG, part of Springer Nature 2018 333

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_26
334 26 Thermoelectromagnetics

Fig. 26.1 Thermal image of

fuses in a circuit (Fluke)

Mathematical Description

Most equations for a complete description of Joule heating have been introduced:

∂
∇
E¼ B ð26:1Þ
∂t

∂
∇
H ¼ Jef þ D: ð26:2Þ
∂t

D ¼ εE: ð26:3Þ

B ¼ μH ð26:4Þ

Jef ¼ σ e E: ð26:5Þ

An experimental observation underlying the formula for Joule heating is that

energy dissipated per unit time is equal to the product of the energy dissipated per a
unit charge passing through a resistor, i.e., a potential, and the amount of charge
passing through resistor per unit time, i.e., a current: Pef ¼ IV. When Ohm’s law is
also applicable, the formula can be written as

Pef ¼ I 2 R ¼ V 2 =R: ð26:6Þ

The differential form of the Joule heating equation calculates the power per unit
volume:
Joule Heating 335

dPef =dV ¼ Jef  E, ð26:7Þ

where Jef is the current density and E is the electric field. For a neutral plasma in a
magnetic field and with a conductivity of σ, we have Jef ¼ σE; hence


2
dPef =dV ¼ Jef  E ¼ Jef  Jef =σ ¼ ρ
Jef
¼ σ jEj2 , ð26:8Þ

where ρ ¼ 1/σ is the resistivity. This equation resembles the “I2R” term in the
macroscopic form of the Joule heating equation.

Applications and Adverse Effects

In some cases, Joule heating is utilized to design electric devices such as water
heaters, while in others, it is an unwanted effect which could be detrimental to the
safety and serviceability of electrical or electronic systems.
On one hand, the employment of Joule heating for various heating systems has a
very long history. Applications utilizing Joule heating include direct use such as hot
plates or pipes as shown in Fig. 26.2 and indirect use (via thermal expansion) such as
microvalves for fluid control. Other heating mechanisms, such as induction heating
to be introduced later, also boils down to the combination of Joule heating and
dielectric heating to be introduced later.

Fig. 26.2 Joule heating-water heating with Flow-3D [XC Engineering, 2009, online video]
336 26 Thermoelectromagnetics

On the other hand, efforts can be made to reduce the detrimental effects caused by
Joule heating. Especially, Joule heating could be a concern in electrical system
components such as conductors in electronics, electric heaters, power lines, and
fuses. The generated heat degenerates or even melts the systems or components.
Cooling designs are needed in many cases to prevent electric devices and their
components from overheating. A voltage applied to a circuit can cause a current and
consequently the Joule heating in the electrically conductive layer over the glass
plate. Bending is then caused by the thermal expansion resulting from the differential
displacements attributed to the material properties and temperature distributions.

Dielectric Heating

Dielectric heating is also known as electronic heating, RF heating, and high-fre-

quency heating. It is the conversion of electromagnetic energy into thermal energy in
dielectric materials caused by a high-frequency alternating electric field, radio wave,
or microwave electromagnetic radiation. There are two slightly different heating
types, i.e., RF heating and microwave heating. Both types of dielectric heating are
induced by dipole rotation. In the RF heating, the applied electromagnetic radiation
usually has a frequency between 3 and 300 MHz. Therefore, the heated objects are
generally smaller than the wavelength. Microwave heating, distinct from RF heating,
is associated with frequencies above 100 MHz. Due to the reason, an electromag-
netic wave can be launched from a small dimension emitter and guided through
space to the target to be heated.

Mechanism

Dielectric heating is a multiphysical phenomenon that involves electromagnetic

waves and heat transfer. Theoretically, any common material that is exposed to
electromagnetic radiation will be heated up. The rapidly varying electric and mag-
netic fields lead to four sources of heating. First, the electric field applied to a
conductive material will cause electric currents. Second, a time-varying electric
field will cause dipolar molecules, such as water, to oscillate and thus generates
heat. Additionally, a time-varying magnetic field applied to a conductive material
will also generate a current flow. Finally, there can also be hysteresis losses in certain
types of magnetic materials.
Energy losses will be generated due to the current flow and the oscillation of the
molecules in the material. These losses will dissipate in terms of heat and a
subsequent rise in the temperature. As the temperature of the material changes,
those properties characterizing the material’s ability in generating heat, such as the
electric conductivity and dielectric constants, will also change. In addition, the
temperature change will also affect the material’s ability in transferring heat such
Dielectric Heating 337

as the specific heat capacity and thermal conductivity. A comprehensive consider-

ation of dielectric heating should allow for all of these couplings between the
electromagnetic field and the thermal field. Despite the common omission of specific
secondary couplings such as the effect on material properties in simplified analyses,
the major heating mechanisms are always needed even in multiphysics simulations
involving Joule heating.
Molecular rotation occurs in materials containing polar molecules, which have an
electric dipole moment. These polar molecules will align themselves with an exter-
nal electromagnetic field. In the case of an oscillating electromagnetic field, the polar
molecules rotate continuously to align themselves with the fast-changing external
field. This process is called dipole rotation or dipolar polarization. Heat is generated
as the molecules gain kinetic energy. Moreover, collisions and other interactions will
inevitably occur as the polar molecules rotate. This will further distribute the energy
to the other molecules and atoms in the material. Dipole rotation is the major
mechanism underlying dielectric heating. The mechanism can be clearly seen in
microwave ovens, which operate most efficiently on liquid water. This is because
water has a lot of polar molecules (water) rather than fats and sugars, which have less
polar molecules.

Types of Dielectric Heating

By definition, dielectric heating refers to the heating of electrically insulating

materials via dielectric loss resulting from an external alternating electromagnetic
field. This changing field may be caused either by an electromagnetic wave propa-
gating in free space, e.g., a guided wave in a microwave oven, or by a rapidly
alternating electric field inside a capacitor. In the first condition, the freely propa-
gating wave transfers the energy. In the latter case, although the heating is accom-
plished by changing the electric field inside the capacitive cavity at radio-frequency
(RF) frequencies, no actual radio waves are generated or absorbed. Accordingly, the
changing electric field may be viewed as an analogy to the electric component of an
antenna’s near field. In other words, this effect is the direct electrical analogy of
magnetic induction heating, which is also near-field effect without classical radio
waves.
Frequencies of the alternating field can determine the dielectric heating mecha-
nism depending on the material type. This is attributed to the relationship between
the frequencies and the size of the “particles” that they can best vibrate and the
availability of different types of “particles” in the material. From a theoretical
perspective, this physical correlation turns into the relationship between the fre-
quency and the imaginary part of the permittivity as shown in Fig. 26.3, which
controls the energy dissipation of the electromagnetic wave. Although high frequen-
cies are generally better, lower-frequency dielectric heating, i.e., less than 10 MHz,
can also generate significant heating effects in some materials especially liquids via
some less common mechanisms. For example, a low-frequency alternating field can
338 26 Thermoelectromagnetics

Frequency, GHz
1000 100 10 2.45 1.0 °C
90 0
80 20
Dielectric

40
70
60
60 80
50 100
0°C
40 0°C
Dielectric loss

30
20
10
0 0°C
0.01 0.1 1 10 100
100 cm-1 10 cm-1 0.1 cm-1 0.01 cm-1

Wavelength (cm)

Fig. 26.3 Spectrum of the imaginary component of the dielectric constant. (Franks 1982; Buchner
et al. 1999; Water Structure and Science site)

drag charged ions and make them move back and forth slowly in a liquid. This
kinetic energy of these ions will be transferred to the liquid molecules via collisions,
leading to an increase in the temperature of the solution.
RF heating covers the frequencies from 10 MHz to 100–100 GHz. Excluding
special mechanisms such as the aforementioned ion-drag, frequencies larger than
10 MHz are usually required to cause efficient dielectric heating. In this frequency
range or higher, dielectric heating is caused by molecular dipole rotation. RF
dielectric heating with frequencies between 10 MHz and 100 MHz performs even
better compared to microwave heating for quickly and uniformly heating certain
food items and for killing parasites and pests in certain harvested crops. However,
RF heating involves a contact process or near-contact process, though a physical
contact is not necessary, which requires the distance between the electromagnetic
radiator and the absorber is less than about 1/6th of a wavelength. Therefore, it is
common to sandwich the material to be heated with metal plates, which generate an
external field in which the sandwiched dielectric material works as a capacitor. The
RF electric fields penetrate nonconductive materials far more deeply than higher-
frequency microwaves; RF heating can be used to rapidly heat and prepare many
non-electrically conductive food and agricultural items.
Higher frequencies, i.e., greater than 300 MHz, will enter the microwave range. In
this case, the wavelength of the external electromagnetic field becomes shorter than
the dimension of the heating space or the heating items. This is exactly what happens
inside a microwave oven. In a microwave oven, the capacitor utilized in the above
RF heating will no longer work as a heating space, but rather an antenna which
generates and transmits far-field electromagnetic waves. The waves will be guided
Dielectric Heating 339

into an insulated space, i.e., microwave heating space, and eventually absorbed by
the heating items to cause heating. The dipole-rotation mechanism in the microwave
heating remains the same as that in the RF heating. Microwave penetrates to a
distance defined by the skin effect. Microwave ovens for heating food are usually not
set to the frequency for optimal absorption by water, i.e., 10 GHz, but, instead, to
2.5 GHz to allow a relatively deep penetration into the heated item.

Mathematical Formulation

In dielectric heating, the density of generated heat per a unit volume is

0
Ped ¼ ωεr ε0 jEj2 , ð26:9Þ
0
where ω is the angular frequency of the external field, εr is the imaginary part of the
complex relative permittivity of the heated material, ε0 is the permittivity of vacuum,
and |E| is the electric field strength. The imaginary part of the relative permittivity,
which is also frequency-dependent, is a measure of the ability of a dielectric material
to convert electromagnetic energy into heat. As we know, σ determines Joule
0
heating. But in the case of a low σ and a high ω value, i.e., σ  ωεr ε0 , dielectric
heating is the dominant mechanism for energy conversion from the electromagnetic
field to the thermal field.

Application

The RF heating was proposed in the 1930s for heating dielectric materials. For
example, US Patent US2147689A filed by the Bell Telephone Laboratories in 1939
proposed RF heating at 10–20 MHz. Such wavelengths were much longer than the
dimensions of the heating space. Therefore, the near-field effects instead of electro-
magnetic waves are the primary heating approaches. In agriculture, RF dielectric
heating has been accepted as an effective way to kill pests in certain food crops after
harvest such as walnuts still in the shell. The RF heating owes the advantage that it
can heat foods more quickly and uniformly. In medicine, the RF heating of body
tissues has been used as a routine practice in diathermia and diathermy.
The most common type of microwave heating is the microwave oven. As shown
in Fig. 26.4, a microwave generated by the waveguide feeder is transmitted into a
heating space and exerted on a heated item. The distribution of the energy is
determined by the geometry and the material of the oven, which in return determine
the efficiency of the heating process. In addition, the efficiency of microwave
heating also depends upon the material properties. Usually, it depends on the content
of water, which has a lot of polar molecules accessible to dielectric heating. Besides,
the position and shape of the food also affect the distribution of the electromagnetic
340 26 Thermoelectromagnetics

Fig. 26.4 Total energy flux

(arrows) and energy
intensity (contours) for a
potato in a microwave (the
rectangular block on the
right represents a waveguide
feed) [COMSOL example
library]

field within the oven and consequently the heating result. That is why most micro-
wave ovens rotate the heated items to improve the heating result. Microwave heating
has also been used in the cancer treatment, in particular, hyperthermic oncology.
This treatment involves localized heating of tumor tissues without damaging the
healthy tissues around them. A thin microwave antenna is inserted into the tumor to
heat the tumor. A coagulated region generated by the microwave heating could kill
the cancerous cells.

Induction Heating

Mechanism

Strictly speaking, induction heating originates from both Joule heating and dielectric
heating, which correspond to the thermal energy generated by the free electric
current and bound electric current, respectively. The induction heating is triggered
by a time-varying (alternating) current, usually at a high frequency. The alternating
current will induce a time-varying magnetic field. The material to be heated is placed
inside the magnetic field so that heat can be generated without touching the coil. In
the heated item, or called workpiece here, both induced eddy electric currents and the
relaxation in the electric dipoles will lead to energy dissipation, representing a
conversion from the electromagnetic energy to thermal energy. Therefore, induction
heating involves a multiphysics process including both electromagnetic fields and
thermal field, but the key mechanisms for heat generation still boil down to Joule
heating and dielectric heating.
The heating attributed to eddy currents is the major contributor. According to
Rudnev et al. (2017), the dielectric heating effect does not exceed 7% of the heat
Induction Heating 341

generated by the eddy current in a great majority of induction heating applications.

Possibly due to this reason, induction heating is explained as the heating process
caused by the eddy currents in many places including this book. Following this
understanding, induction heating is defined as the process of heating an electrically
conducting object (usually a metal) by electromagnetic induction, through heat
generated in the object by eddy currents.
In a narrow sense, induction heating is similar to the Joule heating effect, but with
one important modification. The currents that heat the material are induced via
electromagnetic induction. Therefore, it could be a non-contact heating process.
Accordingly, only those with high electric conductivity, such as copper, gold, and
aluminum, can be directly heated via induction. Food items are then heated by the
metals.
Induction heating works better in specific metals due to additional heating
mechanisms. For instance, induction can heat ferrous metals better than other
materials. That is because the iron crystals in ferrous metals can be repeatedly
magnetized and demagnetized by the alternating magnetic field. This will lead to
hysteresis losses in addition to the normal induction heating.

Governing Equation: The Diffusion Equation

Based on the introduction in the Dielectric Heating section, we know that the
electromagnetic field can also be described using potentials as follows:

B ¼ ∇
A, ð26:10Þ

where A is the magnetic vector potential. From Maxwell’s equations, it is known that

∂A
E¼  ∇φ: ð26:11Þ
∂t
Multiplying the electric field with σ, we have

∂A ∂A
J ¼ σE ¼ σ  σ∇φ ¼ σ þ Jes , ð26:12Þ
∂t ∂t
where Jes is the source current density in the induction coil. It follows from the
Ampère’s equation with a free charge current that

1
∇
B ¼ J: ð26:13Þ
μ
Substituting the definition of the magnetic flux density with the potential into the
above equation, we obtain
342 26 Thermoelectromagnetics

∂A
∇
∇
A ¼ μσ þ μJes ð26:14Þ
∂t
for a homogeneous and isotropic medium independent of the field intensity.
Expanding and imposing the Coulomb gauge condition —  A ¼ 0, we obtain

∂A 1 2
σ  ∇ A ¼ Jes , ð26:15Þ
∂t μ
which is called the diffusion equation. The equation can also be expressed in the
same form but for different vectors instead of A, e.g., J, E, H, and B.
In the case of the magnetic vector potential, the above equation can be related for
any physically observable electromagnetic induction phenomena such as eddy
current, induced voltage, coil impedance, flaw impedance, and coil inductance. If
the excitation current is assumed to be sinusoidal, then the eddy current will be
sinusoidal as well. Consequently, a time-harmonic electromagnetic field can be
introduced:

1
iωσA  ∇2 A ¼ Jes : ð26:16Þ
μ
It should be noted that, in a ferromagnetic material, the harmonic approximation
is not valid anymore. This is due to its time-dependent relative permeability, μr,
which renders the eddy current nonsinusoidal.

Effective Volume: Skin Depth

To gain insights into the skin depth, Maxwell’s equations are expressed as two
second-order differential equations of E and H. For the purpose, let us assume that
the material is linear, isotropic, homogeneous, and charge-free (ρ ¼ 0). Then take
curl on last two equations in Maxwell’s equations
 
∂ ð∇
H Þ ∂ ∂
∇
∇
E ¼ μ ¼ μ σþε E, ð26:17Þ
∂t ∂t ∂t
   
∂ ∂ ∂
∇
∇
H¼ σþε ∇
E ¼ μ σþε H: ð26:18Þ
∂t ∂t ∂t

Applying the Jacobi identify to the two second-order homogeneous wave equa-
tions, we obtain
Induction Heating 343

∇2 E  γ 2 E ¼ 0, ð26:19Þ

∇2 H  γ 2 H ¼ 0, ð26:20Þ

where γ 2 ¼ iωμ(σ + iωε) and γ is called the propagation constant of the medium.
Frequently, the propagation constant is defined as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
iσ
γ ¼ α þ iβ ¼ iω με 1  ð26:21Þ
ωε

The solution to Eq. 26.19 for a uniform electric wave propagating in the positive
z-direction is

E ðz; t Þ ¼ E0 eαz cos ðωt  βzÞ: ð26:22Þ

Consequently, if an electric or a magnetic wave travels in a conductive medium,

its amplitude starts to attenuate exponentially by a factor eαz. The factor α is
therefore called the attenuation constant, with the unit of Np/m (Neper/meter).
A good conductor satisfies σ  ωε. Under this condition, the propagation
constant can be simplified to
pffiffiffiffiffiffiffiffiffiffi
γ ¼ ð1 þ iÞ πf μσ : ð26:23Þ

The skin depth is defined as the distance for which the amplitude of a plane wave
decreases by a factor of e1 ¼ 0.368. Thus, the skin depth becomes

1 1
δ¼ ¼ pffiffiffiffiffiffiffiffiffiffi : ð26:24Þ
α πf μσ

When the electromagnetic waves reach the surface in a conductive medium, a

current will start to flow within the surface and decrease exponentially toward the
interior of the material.

Energy Conversion: The Poynting Vector

Poynting’s theorem is a relation between the changing rate of the energy stored in the
field and the energy flow. Multiplying Faraday’s law and Ampère’s law and ignoring
the displacement current with the electric and magnetic field lead to
þ ð ð
∂B
ðE
HÞ  ndA ¼  H  dV  σE  EdV, ð26:25Þ
s v ∂t v
344 26 Thermoelectromagnetics

where the quantity E
H is known as the Poynting vector PEM in W/m2. Poynting’s
theorem states that the sum of the Ohmic loss and the power absorbed by the
magnetic field in the volume is equal to the total power input.
It is common to represent a harmonic variation in time as a complex function of
coordinates, multiplied by the factor, eiωt. The complex Poynting vector can, there-
fore, be written as

1
PEM ¼ ðE
HÞ, ð26:26Þ
2
and its integral is
þ ð ð
1 1
PEM dS ¼  jω μjHj2 dV  σ jEj2 dV: ð26:27Þ
s 2 v 2 v

The real part of the above equation determines the energy dissipated into heat
(ohmic losses) in the volume V, and the imaginary part is equal to 2ω times of the
mean value of the magnetic energy. The current from the real part is known as the
eddy current and is the main heat source in induction heating.

Applications

Induction heating requires no physical contact between the heated item and the
induction coil. This is especially suitable for heating applications in which a high
degree of cleanliness is desired, such as in semiconductor manufacturing. The
semiconductor industry has adopted induction heating for heating silicon. Another
advantage is that the heat is generated inside the heated items as opposed to those
from the surface in dielectric heating. As a result, a high heating efficiency can be
achieved by avoiding heat losses from the surfaces.
Induction stoves are a great application of induction heating, which benefit from
the above features. Figure 26.5 schematically illustrates the application of induction
heating in a typical induction stove.

Fig. 26.5 Schematic of an induction stove (cooktop)

Practice Problem 345

In such a stove, a coil is placed beneath the stove top. The coil will apply an
electromagnetic field to a metal pot placed on the stove. This pot, which is usually
made of highly electrically conductive materials, will be heated up quickly due to the
induced electric currents in the pot. The pot then heats the food or other objects in
it. Induction heating has also been applied to sealing, heat treatment, and welding.
Despite the benefits, induction heating can also be a detrimental effect. This is
especially the case in transformers, in which induction heating leads to a waste of
power. In the design and functioning of transformers, the eddy currents flowing
inside the cores need to be eliminated to the minimum. These currents will cause
heating in the transformer’s magnetic core and other secondary problems such as
damages to the structural integrity of relevant components.

Practice Problem

This section is developed to show the numerical implementation of dielectric

heating.

Problem Description

As shown in Fig. 26.6, a long box with a cross section of 0.04 m by 0.02 m contains
water. All the side surfaces and the two end surfaces (two ends of the longitudinal
direction) are metals, which can be formulated as perfect electric conductors
(—
E ¼ 0). One side surface is perfect magnetic conductors (—
H ¼ 0). The
box contains water which has a relative permittivity of 78 + 12i. An electromagnetic
wave with a frequency of 2.45 GHz was generated, whose wave number along the
longitudinal direction is 0.4 of the total wave number. Please predict the temperature
distribution on the path of x ¼ 0.02 m at the end of 4 minutes.

Fig. 26.6 Computation

domain and boundary
conditions
346 26 Thermoelectromagnetics

Governing Equation Natural Boundary Condition

The theoretical formulation of the problem involves both the thermal field and the
waveguide theory introduced in Chapter 23 for electromagnetics. Governing equa-
tions for 2D need to be derived first following a procedure similar to the practice
problem in Chapter 23. The following listed the governing equations written in the
FlexPDE script. Please try to derive these equations first for practice.
Hz: DEL2(Hz) + k^2*Hz ¼ 0.
Temp:DT(Temp) ¼ 0.6/4.2e7*DEL2(Temp) + 0.5*omega*epsilon*epsilon_ri*
(Ex^2 + Ey^2).
An additional constraint is needed:
INTEGRAL(Hz) ¼ 0.

Results

The H field and E field are shown in Fig. 26.7 and Fig. 26.8, respectively. The
temperature distributions are depicted in Fig. 26.9 and Fig. 26.10.

Fig. 26.7 Electric field

Practice Problem 347

Fig. 26.8 Magnetic field

Fig. 26.9 Temperature distribution in the computational domain (x ¼ 0.02 m) at t ¼ 240 s

348 26 Thermoelectromagnetics

Fig. 26.10 Temperature distribution along y-axis of the computational domain (x ¼ 0.02 m) at
t ¼ 240 s

Appendix: FlexPDE Script File

{ Microwave 3D with TE wave }
TITLE "Microwave Heating"
COORDINATES CARTESIAN2
SELECT
ERRLIM 0.002
NGRID 15 ! The student version with a course mesh may lead to
incorrect solutions
REGRID off
VARIABLES
Hz ! z-component of the H field
Temp ! Temperature
INITIAL VALUES
Temp=283.15
DEFINITIONS
a = 0.04 ! Length of the wave guide
b=0.02 ! Width of the wave guide
mu=4*pi*1e-7
epsilon=8.654e-12
c=1/sqrt(mu*epsilon)
Practice Problem 349

f=2.45e10 ! Frequency of the TE microwave

omega=f*2*PI ! Radial frequency
k=omega/c ! Wave number
kc=0.6*k ! Wave number of the transverse wave components (x, y)
kz=k-kc ! Wave number of the longitudinal wave component (z)
Hy=-kz/(kc^2)*dy(Hz)
Hx=-kz/(kc^2)*dx(Hz)
Ex=-omega*mu/(kc^2)*dy(Hz)
Ey=omega*mu/(kc^2)*dx(Hz)
epsilon_rr=78 ! Real part of the relative permittivity for water at
20 degC epsilon=epsilon_0*epsilon_rr
epsilon_ri=12 ! Imaginary part of the relative permittivity
(heating source, 12 for water at 20 degC)
z=1
EQUATIONS
Hz: DEL2(Hz)+k^2*Hz = 0
Temp: DT(Temp)=0.6/4.2e7*DEL2(Temp)
+0.5*omega*epsilon*epsilon_ri*(Ex^2+Ey^2)
CONSTRAINTS
INTEGRAL(Hz) = 0 { since Hz has only natural boundary conditions, we
need an additional constraint to make the solution unique }
TIME 0 TO 240
BOUNDARIES { The domain definition }
REGION 1
START(0,0) VALUE(Hz) = 2e-5 NATURAL(Temp)=0 ! Perfect magnetic
conductor for one wall
LINE TO (a,0) NATURAL(Hz) = 0 NATURAL(Temp)=0! Perfect electric
conductor for the other walls
LINE TO (a,b) NATURAL(Hz) = 0
LINE TO (0,b) NATURAL(Hz) = 0
LINE TO CLOSE
MONITORS
FOR Cycle=1
CONTOUR(Hz)
PLOTS
FOR Cycle=1
CONTOUR(Hz)
VECTOR(Ex,Ey)
VECTOR(Hx,Hy)
ELEVATION(Temp) FROM (a/2,0) TO (a/2,b)
CONTOUR(Temp)
CONTOUR(Temp)
END
Chapter 27
Electromagnetomechanics

Introduction

This chapter for electromagnetomechanics splits into magnetomechanics and

electromechanics. Magnetomechanics deals with the steady-state force created by
a magnetic field, which usually results from a permanent magnet or a constant
current flowing through a conductor, and the mechanical responses caused by this
magnetostatic (or magnetic) force. Accordingly, we will show the formulation of the
force between a magnetic and a current using the Lorentz force law and that between
magnets using both the Ampère model and the Gilbert model. For electromechanics,
two major electromechanical mechanisms, i.e., piezoelectric effect and
piezoresistive effect, will be discussed for their mathematical formulations and
applications. In the final practice problem section, an example will be given for
the electric potential generated by a lead zirconate titanate (piezoelectric) beam.

Magnetomechanics: Magnetostatic Force

Magnetomechanics deals with the steady-state force created by a magnetic field,

which usually results from a permanent magnet or a constant current flowing through
a conductor, and the mechanical responses caused by this magnetostatic
(or magnetic) force. For example, a magnetic material such as iron suffers from a
magnetostatic force when it is placed in a magnetic field. Figure 27.1 illustrates the
deformation of a hollow nanocapsule made of sexithiophene (6 T) molecules in a
strong magnetic field (20 Tesla) at different temperatures (Manyuhina et al. 2007).
Some basics about magnetomechanics have been introduced in the electrostatics and
magnetostatics. In this section, we will focus on the two types of magnetostatic
forces.

© Springer International Publishing AG, part of Springer Nature 2018 351

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_27
352 27 Electromagnetomechanics

Fig. 27.1 Deformation of a hollow nanocapsule made of sexithiophene (6 T) molecules in a strong

magnetic field (20 T) at different temperatures. (Manyuhina et al. 2007)

Force Between a Magnet and a Current

The force between a current and a magnet is the relatively easy one of the two types
of magnetostatic forces from the perspective of numerical simulation. The force can
be obtained by integrating (adding up) the Lorentz force created by all the moving
charges within the current.
According to the Lorentz force law, the force that a magnetic field B exerts on a
charge q moving at a velocity v is

Fm ¼ qv  B: ð27:1Þ

Like the electric field E, this formula was obtained with experiments, and it
provides us with the definition of B.
In fact, the total force on a charged particle is

F ¼ QðE þ v  BÞ: ð27:2Þ

It is noted that particles moving parallel to the magnetic field do not experience a
magnetic force. Also, the magnetic force cannot do work, because it is always
perpendicular to the motion of the particle. This force cannot change the speed of
a particle but only alters its direction.
A line charge λ (unit: C/m) traveling down a thin wire produces a current I:

I ¼ λv: ð27:3Þ

For a thin wire, this direction is usually obvious; but we sometimes need to define
a current as a vector for surface or volume currents.
The force on a wire due to an external magnetic field B is
Z Z Z
Fm ¼ v  Bdq ¼ v  Bλdl ¼ I  Bdl: ð27:4Þ
Magnetomechanics: Magnetostatic Force 353

Since the current and the line element of the wire are in the same direction, we can
also write the above equation as
Z
Fm ¼ jIjdl  B: ð27:5Þ

Finally, if the current is constant along the wire, we have

Z
Fm ¼ jIj dl  B: ð27:6Þ

In 2D or 3D cases, we can define surface and volume current densities, respec-

tively. Similar to Eq. 27.4, we can obtain the equation for the magnetic force as
Z
Fm ¼ ðJ  BÞdΩ: ð27:7Þ

Force Between Magnets

Magnets exert forces and torques on each other due to the complex rules of
electromagnetism, which make this type of magnetostatic force more complicated.
The forces originate from the microscopic currents of electrically charged electrons
orbiting nuclei and the intrinsic magnetism of fundamental particles that make up the
material, such as electrons. Both of these can be well modeled as tiny loops of
current, i.e., magnetic dipoles. These magnetic dipoles produce their own magnetic
field and, in the meanwhile, are affected by external magnetic fields. Therefore, the
most elementary force between magnets is the magnetic dipole-dipole interaction.
Two models are prevalent for calculating the forces between magnets: the
Ampère and Gilbert models. The Ampère model is physically correct, while the
Gilbert model is easier to use. The Ampère model is implemented directly using the
mechanisms introduced above: all of the magnetic dipoles that make up the two
magnets are determined, and then the net force on both magnets can be obtained by
summing up all the interactions between the dipoles of the first magnet and that of
the second. By contrast, the Gilbert model assumes that the forces between magnetic
poles are caused by magnetic charges “smeared” on the poles. This model works
well for predicting the forces between simple magnets where the distribution of the
“magnetic charge” can be well modeled. However, it is worthwhile to mention that
this model excludes some characteristics of magnetism, e.g., the relationship
between the angular momentum and magnetic dipoles, and that the model is
established based on the assumed magnetic charges.
354 27 Electromagnetomechanics

Models for Magnetic Forces

Magnetic Dipole Moment

The magnetic field created by a magnet can be described using the field of the total
magnetic dipole moment, m. The direction of the magnetic moment points from the
south pole to the north pole of a magnet. Both the torque and force exerted on a
magnet by an external magnetic field are proportional to that magnet’s magnetic
moment. In the chapter for electromagnetics, we have defined the magnetic dipole
moment as the product of an infinitesimally small loop of current I with the area of
the loop, A:

m ¼ IA, ð27:8Þ

where the direction of m is normal to the area and the direction is determined by the
current and the right-hand rule. This is, in fact, the definition of the magnetic dipole
moment in the Ampère model.
In the Gilbert model, the magnetic dipole moment is assumed to be generated by
two equal but opposite magnetic charges that are separated by a distance of d. The
two separated magnetic charges form a tiny magnet. Then m can be defined in a way
similar to that for the electric dipole moment P due to electrical charges:

m ¼ qm d, ð27:9Þ

where qm is the “magnetic charge.” The direction of the magnetic dipole moment
points from the negative south pole to the positive north pole of this tiny magnet.

Ampère Model

In the Ampère model, all magnetization results from the bound currents (or called
Ampèrian currents) at the microscale. The total effect of these microscopic bound
currents is to make the magnet behave as if there is a macroscopic electric current
flowing in loops in the magnet, leading to a magnetic field normal to the loops. The
Ampère model gives out the exact magnetic field both inside and outside the magnet.
However, it is usually difficult to calculate the Ampèrian currents on the surface of a
magnet. Every magnet produces a magnetic field which is strong near its poles.
In the Ampère model, the force on a magnetic dipole caused by a nonuniform
magnetic field is essentially attributed to the Lorentz force on the “current loops”
corresponding to the magnetic dipole. This force can be calculated as

Fm ¼ ∇ðm  BÞ: ð27:10Þ

A permanent magnet is treated in the same way as an electromagnet: a magnetic

contains small current loops throughout its volume, and its magnetic strength is
Magnetomechanics: Magnetostatic Force 355

proportional to the current and the surface of each loop and proportional to the
density of current loops in the material.
It is often assumed that each small piece of a magnetic material has an equally
strong magnetization, which is, however, not always true. This is because a magnet
can change the magnetization of nearby magnets. While this effect is usually small
between permanent magnets, it could be significant for electromagnetics. For exam-
ple, if an electromagnet that consists of a wire wound around an iron core is brought
close to a permanent magnet, the magnetization of that core can change drastically.
Therefore, the Ampère model is suitable to compute the magnetic force field of
permanent magnets, while for electromagnets, a magnetic circuit approach as intro-
duced in the previous subsection may be a better option.

Gilbert Model

The Gilbert model assumes that the magnetic forces between magnets are due to the
magnetic charges near the poles. This method is proposed via the analogy between
the electric and magnetic fields. Since there is no magnetic current, the method is not
physically based. However, this model provides better predictions in magnetic fields
that are more dependent on the detailed shape and magnetization of the magnet
rather than merely the magnetic dipole contribution.
It is necessary to mention that the Gilbert model is implemented based on the
distributions of the “magnetic charges” over the magnetic poles. This fact renders
this method only useful for simple configurations. Fortunately, many useful cases
have simple configurations. For example, two poles that are small enough can be
represented as single points, and then the force between these two infinitely small
magnetic poles is given by
μqm1 qm2
Fm ¼ , ð27:11Þ
4πr 2
where qm1 and qm2 are the magnitudes of the magnetic charges on the poles, μ is the
permeability of the intervening medium, and r is the distance between the poles.

Secondary Effect of Magnetostatic Force

The magnetostatic force can result in secondary effects on both the mechanical field
and the magnetic field. As for the influence on the mechanical field, we have
magnetostriction, which is the deformation of ferromagnetic materials as the result
of an applied magnetic field. Ferromagnetic materials change their shape or dimen-
sions during the process of magnetization due to their intrinsic properties. The
magnetostrictive strain in such materials will change in response to the external
magnetic field before the strain reaches the saturation value.
356 27 Electromagnetomechanics

The intrinsic property responsible for magnetostriction is the unique structure of

ferromagnetic materials: such a material consists of domains with uniform magnetic
polarization in each domain. Upon applying a magnetic field, the boundaries
between the domains shift and the domains rotate, producing a change in the
dimensions of the material. It is common that crystals are usually nonisotropic due
to the distinct structures along different directions. Therefore, it usually takes more
energy to magnetize a crystalline material in one direction than another
(magnetocrystalline). The material will tend to rearrange its structure to minimize
the free energy of the system. This will induce a strain because different crystal
directions have different lengths.
In the opposite way, the magnetostatic force can also change the property of the
magnetic field. Especially, such a force can lead to magnetostriction and eventually
changes the susceptibility of a material to an external magnetic field. This is the
so-called Villari effect. Two other effects also exist in relation to magnetostriction.
The Matteucci effect is the generation of a helical anisotropy of the susceptibility of a
magnetostrictive material under the influence of a torque, whereas the Wiedemann
effect is the twisting of the material when a helical magnetic field is applied.

Electromechanics

The piezoelectric effect and piezoresistive effect will be discussed in this section as
the two major electromechanical mechanisms.

Piezoelectric

The piezoelectric effect states the relationship between the electric polarization and
mechanical stresses/strains. In a broad sense, this implies a two-way process, via
which the electric field and mechanical field can interact with each other. In a narrow
sense, it only refers to the influence of the mechanical stresses/strains on the electric
polarization. The piezoelectric effect relies on the structure and constituents of the
material and can be characterized as a material property. This phenomenon is
observable in many naturally occurring crystalline materials such as quartz, Rochelle
salt, and human bone. A much more pronounced piezoelectric effect can be obtained
via engineered materials, such as lithium niobate and lead (Pb) zirconate titanate
(PZT). PZT sensors have been widely used in testing. Figure 27.2 shows two
common bender element sensors made of PZT for soil testing.
Electromechanics 357

Fig. 27.2 PZT bender elements used in soil testing

Mechanism

The piezoelectric effect stems from the occurrence of electric dipole moments in
solids. Dipoles tend to be aligned in regions called Weiss domains, while the
alignments of different Weiss domains are usually random and can be much differ-
ent. However, these random alignments can be unified by applying a strong external
electric field at elevated temperatures. This process, called poling, is analogous to the
magnetic poling. Not all the piezoelectric materials can be poled. As introduced
before, the electric field in the material can be obtained via the density of electric
dipole moments. In poled piezoelectric materials, the uniformly oriented dipoles will
generate an electric field.
Under the influence of mechanical stresses/strains, the polarization P will change.
This change could be attributable to either a reconfiguration of the dipole-inducing
surrounding or a reorientation of molecular dipole moments. This change in polar-
ization, either strength, direction, or both, may then manifest as piezoelectricity. The
outcome depends on the orientation of polarization, the crystal’s symmetry, and the
applied mechanical stress. The polarization change appears as a variation of the
surface charge density on the crystal surfaces. For example, a 1 cm3 cube of quartz
with an applied force of 2 kN can produce a voltage of 12,500 V. When an external
electric field is applied, the process can also happen in the opposite way to generate
stresses/strains, which is called the converse piezoelectric effect.

Mathematical Formulation

Piezoelectricity as a reversible process requires a fully coupled constitutive relation-

ship that combines the constitutive relationships in both the electric field and
mechanical field. For the electric field, we have the following relationship between
the electric displacement field D and the electric field E:
358 27 Electromagnetomechanics

D ¼ εE: ð27:12Þ

For the mechanical field, we can recall Hooke’s law for linearly elastic materials:

S ¼ sT, ð27:13Þ

where S is the strain, s is the compliance, and T is the stress. Here symbols for the
mechanical field are different from the common ones, i.e., those in the other places of
this book, due to the conflicts in the choice of symbols in the studies of the
mechanical field and electric field.
To couple the two fields, one more material property, which is formulated as a
matrix, needs to be added to describe the couplings between the two fields. This type
of strong coupling, which is formulated using the following equations, is similar to
that in poroelasticity:

S ¼ sT þ dT E ð27:14Þ
D ¼ dT þ εE: ð27:15Þ

By employing the Voigt notation, we can obtain the following formulations for
typical piezoelectric materials such as tetragonal PZT and BaTiO3:
2 3 2 32 3
S1 s11 s12 s13 0 0 0 T1
6 S2 7 6 s21 s22 s23 0 0 0 7 6 T2 7
6 7 6 76 7
6 S3 7 6 s31 s32 s33 0 0 0 7 6 7
6 7¼6 76 T 3 7
6 S4 7 6 0 0 0 s44 0 0 7 6 7
6 7 6 76 T 4 7
4 S5 5 4 0 0 0 0 s55 0 54 T 5 5
S6 0 0 0 0 0 s66 T6
2 3 ð27:16Þ
0 0 d 31
6 0 0 d 32 7 2 3
6 7 E1
6 0 0 d 33 7
þ6
6 0
74 E 2 5
6 d 24 0 7 7 E
4 d 15 0 0 5 3
0 0 0
2 3
T1
2 3 2 36 T 2 7
D1 0 0 0 0 d15 0 66 T3 7
7
4 D2 5 ¼ 4 0 0 0 d24 0 0 56 7
6 T4 7
D3 d 31 d 32 d33 0 0 0 64 T5 5
7
ð27:17Þ
T6
2 32 3
ε11 0 0 E1
þ4 0 ε22 0 54 E 2 5
0 0 ε33 E3
Electromechanics 359

where the first equation represents the relationship for the converse piezoelectric
effect (coupling from the electric field to the mechanical field) and the latter for the
direct piezoelectric effect (coupling from the mechanical field to the electric field).
The piezoelectric coefficients, i.e., dij, can be determined as follows:
   
∂Di  ∂S j 
dij ¼ ¼ : ð27:18Þ
∂T j E¼0 ∂E i T¼0

The mathematical formulations for other piezoelectric materials will be different,

depending on the specific coupling mechanisms determined by the crystal structures.

Areas of Application

Piezoelectric materials are major players in the industries for sensors and actuators.
The materials can be used for developing actuators, such as linear motors, rotary
motors, and pumps, and sensors, such as load cells, pressure sensors, accelerometers,
and gyroscopes. In soils, the commonly used geophones and bender elements
(Fig. 27.2) are developed with piezoelectric materials. The materials have also
been widely used in acoustics for the transducers for both generating and picking
up wave signals.
However, piezoelectric materials have much wider use than sensors and actua-
tors. For example, piezoelectric materials are also commonly utilized in
microelectromechanical systems (MEMS). Piezoelectricity-based MEMS can be
made for microscale chemical and biological sensors and high-precision mass
measurement devices based on resonance. Piezoelectric materials are also used for
energy harvesting by taking advantage of the direct piezoelectric effect. Kinetic
energy can be converted to electricity via the process.

Piezoresistive

The piezoresistive effect is the change in the electrical resistivity of a semiconductor

or a metal due to mechanical strains. In contrast to the piezoelectric effect, the
piezoresistive effect causes a change in the electrical resistance (a material property)
rather than the electric potential (a state variable).

Mathematical Formulation

In electrically conductive and semiconductive materials, strains can change the

interatomic spacing and consequently the easiness for electrons to be raised into
the conduction band. This leads to a change in the resistivity of the material. A linear
relationship usually can be assumed between the resistivity and the strain within a
360 27 Electromagnetomechanics

certain range of strains. This proportionality is the piezoresistive coefficient and can
be mathematically defined as
 
∂ρ
ρ
ρσ ¼ , ð27:19Þ
ε
where ρσ is the piezoresistive coefficient, ∂ρ is the change in the resistivity, ρis the
original resistivity, and ε is the strain.

Piezoresistive Coefficient

The change of the resistance in metals is usually due to the change of the geometry
resulting from the applied mechanical stress. Accordingly, the resistivity can be
easily formulated using Ohm’s law. The piezoresistive effect in these cases is
generally small but not negligible. Some metals owe piezoresistivity values that
are much larger than what are predicted based only on the geometry. For example,
the piezoresistivity of nickel is 13 times of the prediction using the geometry.
In semiconductors, the piezoresistive coefficient can be several orders of magni-
tudes higher than what is predicted using the geometrical effect. Due to the reason, it
is common to use semiconductors to build strain gauges for the purpose of a very
high coefficient of sensitivity. However, it is more difficult to handle semiconductors
than metals for developing strain gauges. The reason is that semiconductors are
generally more sensitive to environmental conditions. Common semiconductors
with obvious piezoresistive effects include germanium, polycrystalline silicon,
amorphous silicon, silicon carbide, and single-crystal silicon.

Application

The piezoresistive effect in some semiconductive materials, such as germanium,

polycrystalline silicon, amorphous silicon, and single-crystal silicon, has been uti-
lized for fabricating a wide range of pressure and acceleration sensors. Piezoresistors
are the simplest form of piezoresistive devices. Because resistors made from
piezoresistive materials, piezoresistors are usually used for mechanical stress mea-
surements. Piezoresistors can be fabricated using a wide variety of piezoresistive
materials. In addition, the piezoresistive effect has also been used in the fabrication
of transducers, piezo-FETS, solid-state accelerometers, and bipolar transistors.

Practice Problem

The practice problem in this section requires simulating the electric potential gener-
ated by a simply supported plate made of a PZT material, which was described in
Fernandes and Pouget (2003).
Practice Problem 361

Problem Description

The plate is made of transversely isotropic PZT material. The transverse plane of the
material (direction 1 and 2 or x and y) coincides with the plane of the plate. The
constitutive relationships of the material are as follows:
2 3 2 3 2 3
S1 139 77:8 74:3 0 0 0 T1
6 S2 7 6 77:8 139 74:3 0 0 0 7 7 6 T2 7
6 7 6 6 7
6 S3 7 6 74:3 74:3 115 0 0 0 7 6 T3 7
6 7¼6 139  77:8 7
7GPa  6 7
6 S4 7 6 6 T4 7
6 7 6 0 0 0 0 0 7 6 7
4 S5 5 6
4 0
2 7 4 T5 5
0 0 0 25:6 0 5
S6 0 0 0 0 0 25:6 T6

2 3
0 0 5:2
6 0 0 5:2 7 2 3
6 7 E1
6 0 0 15:1 7 C
þ6
6 0
7  4 E2 5
6 12:7 0 7 7m
2
4 12:7 E3
0 0 5
0 0 0
ð27:20Þ
2 3
T1
2 3 2 3 6 T2 7
D1 0 0 0 0 12:7 0 6 7
C 6 T3 7
4 D2 5 ¼ 4 0 0 0 12:7 0 0 5 6 7
m2 6 T4 7
D3 5:2 5:2 15:1 0 0 0 6 7
4 T5 5
ð27:21Þ
T6
2 3 2 3
13:05 0 0 E1
nF
þ4 0 13:05 0 5  4 E2 5
m
0 0 11:5 E3

As shown in Fig. 27.3, we here simplify the problem into a 2D problem, in which
the plate is simplified into a simply supported beam with a length of 0.1 m and a
thickness of 0.005 m. Please predict the distribution of the electric potential with

Fig. 27.3 Geometry and loading conditions

362 27 Electromagnetomechanics

height (thickness) along the vertical axis of the beam (center) generated by a
downward vertical distributed load of 1 Pa on the top of the beam.

Results

A 2D geometry was selected because a 3D description would require six dependent

variables, which are not available in the student version of FlexPDE. In addition, the
mechanical boundary condition of the simply supported beam cannot be easily
implemented with given tools in FlexPDE. Due to this reason, we fixed the two
ends of the beam for simplicity.
The vertical displacement of the beam is illustrated in Fig. 27.4, which can be
validated using simple equations from the mechanics of materials. This is not done
because it is not the focus. The distributions of the electric potential are shown in the
following figures. The shapes of the curves match the normalized results obtained in
Fernandes and Pouget (2003) (Figs. 27.5 and 27.6).

Fig. 27.4 Vertical displacement of the PZT plate

Practice Problem 363

Fig. 27.5 Electric potential along the horizontal axis

364 27 Electromagnetomechanics

Fig. 27.6 Electric potential along the vertical axis

Appendix: FlexPDE Script File

{ 2D PZT Beam, Fernandes and Pouget, 2003, International Journal of
Solids and Structures, 40 }
TITLE 'Electromechanics-2D PZT'
VARIABLES
u=VECTOR(ux,uy)
psi
DEFINITIONS
! Dimensions of the beam and computational domains
a=0.1 ! Length of the beam
b=a/20 ! Thickness of the beam
! PZT constitutive relationship: sigma=C*epsilon+ee*e;
D=ee*epsilon+ep*E
c11=139e9
c12=77.8e9
c33=115e9
c13=74.3e9
c44=25.6e9
e31=-5.2
e33=15.1
Practice Problem 365

e15=12.7
ep11=13.05e-9
ep33=11.5e-9
E=-GRAD(psi) { Electric field }
C=TENSOR((C11,C13,0),(C13,C33,0),(0,0,C44)) ! Stiffness matrix
ee=TENSOR((0,e31),(0,e33),(e15,0))
ep=TENSOR((ep11,0),(0,ep33))
epsilon=VECTOR( dx(ux),dy(uy),0.5*(dx(uy)+dy(ux)) )
sigma=dot( C,epsilon) + dot(ee,E)
D=dot(epsilon,ee) + dot(E,ep)
{ The above C and epsilon are defined using the Voigt notation;
sigma_tensor uses the tensor notation }

sigma_tensor=TENSOR((XCOMP(sigma),ZCOMP(sigma)),(ZCOMP
(sigma),YCOMP(sigma)))
EQUATIONS
u: DIV(sigma_tensor)=VECTOR(0,0)
psi: DIV(D)=0
BOUNDARIES
REGION 1 { Air around the beam }
INACTIVE(u)
START(-a,-a)
VALUE(psi)=0
LINE TO (-a,a)
VALUE(psi)=0
LINE TO (2*a,a)
VALUE(psi)=0
LINE TO (2*a,-a)
VALUE(psi)=0
LINE TO CLOSE
REGION 2 { PZT beam }
START(0,0)
VALUE(ux)=0
VALUE(uy)=0
LINE TO (0,b)
NATURAL(uy)=-1
NATURAL(ux)=0
LINE TO (a,b)
VALUE(ux)=0
VALUE(uy)=0
LINE TO (a,0)
NATURAL(ux)=0
NATURAL(uy)=0
LINE TO CLOSE
MONITORS
PLOTS
VECTOR(u)
CONTOUR(XCOMP(sigma))
ELEVATION(uy) FROM (0,0) TO (0,b)
ELEVATION(uy) FROM (0,b) TO (a,b)
ELEVATION(psi) FROM (a/2,0) TO (a/2,b)
END
 Part V
Implementation Methods
Chapter 28
Finite Difference Method

Introduction: Numerical Analysis

Numerical analysis is about the use of numerical approximations as opposed to

symbolic manipulations for the solution of mathematical problems. This approach
originated from the hand calculations with large printed tables. After the birth of
computers, automated calculations with computer programs have been promoting
this type of analysis into a major approach in scientific studies and engineering
applications. The overall goal of numerical analysis is to develop and apply algo-
rithms to provide approximate but acceptable solutions to problems which are
difficult to solve with analytical techniques.
Numerical analyses can be categorized into direct and iterative methods based on
the control of calculation steps. Direct methods compute the solution to a problem
within a finite number of steps, while iterative methods do not necessarily terminate
in a finite number of steps. However, for practical purpose, analyses with iterative
methods are usually terminated when the residual is smaller than a predefined
allowable error.
The field of numerical analysis can be divided into multiple areas depending on
the purpose of the analysis. The evaluation of the function value at a given point is
one of the simplest applications of numerical analysis. Interpolation and extrapola-
tion aim to find the value of an unknown function at a point with given values of that
function in the ranges including and excluding that point, respectively. Regression is
similar to interpolation; however, it requires finding out the mathematical formula-
tion of a function instead of function values at given points. Another significant
application is to compute the solution to a given equation. Optimization is
implemented to search for the point at which a given function is maximized or
minimized. Evaluating integrals numerically, which is called numerical integration
or numerical quadrature, is the numerical analysis for calculating the value of a
definite integral.

© Springer International Publishing AG, part of Springer Nature 2018 369

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_28
370 28 Finite Difference Method

Solving differential equation, especially partial differential equations, is another

significant area of numerical analysis, as we know that many physical processes,
including most of the monolithic and multiphysical processes introduced in the
previous chapters of this book, can be described using partial differential equations.
Therefore, solving these equations with initial and boundary conditions and material
properties from real examples is like simulating or numerically reproducing the real
physical processes. As a result, the area of numerical analysis is also termed
numerical simulation. This is especially true when the solution process is enhanced
with the computer-aided model establishment and graphical user interface-based
postprocessing. These tools make numerical simulation appear as a virtual experi-
ment carried out using computers. Notwithstanding the fancy and vague appearance,
we need to bear in mind that those numerical experiments are implemented based on
physical and mathematical laws underlying the partial differential equations. In fact,
numerical simulation also contains pieces from other numerical analysis areas. For
example, numerical integrals are usually needed for the finite element method.
This chapter will introduce one of the most straightforward numerical simulation
methods: the finite difference method. We will show how to approximate derivatives
using finite differences and discretize the equation and computational domain based
on that. The discretization will be discussed for spatial and temporal derivatives
sequentially. Then the treatment of the boundary and initial conditions will be
explained with an example. A brief introduction to the error analysis in the finite
difference method will also be provided. Most of the above introductions will be
made in 2D. Therefore, the extension from 2D to 3D will be discussed to bridge the
knowledge gap. In the final, MATLAB code for implementing the example in the
boundary and initial condition section will be attached.
Following this chapter will be the introduction to the finite volume method, which
is similar to the finite difference method in many aspects. The finite element method
as a slightly more complicated and more popular type of numerical simulation
method will be introduced finally.

Spatial Finite Difference and Derivative

One essential idea behind numerical simulation is discretization. For the purpose, we
need to transform a continuous mathematical equation(s) into an algebraic equation.
Accordingly, the computational domain will be discretized into a mesh or a grid
which consists of multiple subdomains called cells or elements. The size of the
matrices in the algebraic equation is determined by the mesh size, the dimensions of
the computational domain, and the number of unknowns. Therefore, the
discretization of the equations and that of the domains is correlated. Different
ways or schemes can be used to approximate the same derivative, leading to different
levels of accuracy. In addition, discretization of spatial derivatives and temporal
derivatives is also different due to the difference between the spatial and temporal
dimensions. For example, for a spatial dimension, we always know the conditions on
Spatial Finite Difference and Derivative 371

the “start” end and need to march toward the “final” end, while for spatial dimen-
sions, boundary conditions could be given on any ends (boundary segments) or even
both. This leads to the different terms and ways used for these two types of
discretization. The following two sections will be devoted to the finite differences
for spatial and temporal dimensions sequentially.

Spatial Discretization

A finite difference is a sequence of the differences between function values as the

independent variable changes with finite increments (Fig. 28.1). It usually has the
mathematical expression of the form f(x + h)
f(x). A really useful operation is to
divide the finite difference by an increment in the independent variable, i.e., h, so that
we can obtain the following quotient

f ð x þ hÞ
f ð x Þ
D ð uÞ ¼ , ð28:1Þ
h
where D(u) is the finite difference approximation to the derivative.
The idea of the finite difference method is to use the above quotient, D(u), to
approach the gradient (derivative) at point x, point x + h, or any point between them.
The geometric analogy of this approximation is to use a secant line to approximate a
tangent line. When the interval gets infinitely small, the error of the above approx-
imation will become infinitely small as well. Therefore, the derivative of a function
f at a point x can be defined using the following limit:

f ðx þ hÞ
f ðxÞ
f 0 ðxÞ ¼ lim : ð28:2Þ
h!0 h
If h has a non-zero value instead of approaching zero, then the above approxi-
mation can be written in a general way as

f 0 ðxÞ ¼ f 0 ðxÞ þ E ðhn Þ, ð28:3Þ

Fig. 28.1 Finite differences

as approximations to
derivatives
372 28 Finite Difference Method

where E is the truncation error, which is a function of the nth power of the increment h.
Usually, there are many ways to construct D(u), and these different ways are called
schemes in the context of the finite difference method. Equation 28.3 indicates that the
scheme has an nth order accuracy. Schemes of different orders of accuracy can also be
constructed for higher-order derivatives, which can be formulated in a way similar to
Eq. 28.3.
For the first-order derivative, the most common schemes are forward, backward,
and central difference schemes. A forward scheme has an expression of the form:

f ðx þ hÞ
f ðxÞ
f 0 ð xÞ ¼ : ð28:4Þ
h
It is called “forward” because we use the function value at a point in front of
(along the direction of the corresponding axis) the point at which the derivative is
calculated. Likewise, a backward scheme uses the function values at x and x
h
instead:

f ðxÞ
f ðx
hÞ
f 0 ð xÞ ¼ : ð28:5Þ
h
We can also approximate the derivative using the function values at points on the
two sides of the point of derivative as:

f ðx þ h=2Þ
f ðx
h=2Þ
f 0 ð xÞ ¼ : ð28:6Þ
h
A second-order derivative can be approximated using first-order derivatives as we
can view the second-order derivative as a derivative of first-order derivatives. For
example, in order to obtain the second-order derivative at the point x, we can use the
derivatives at the points x
h/2 and x + h/2:

f 0 ðx þ h=2Þ
f 0 ðx
h=2Þ
f 00 ðxÞ ¼ : ð28:7Þ
h
The above approximation uses the central difference. We can also use central
differences for the first derivative in the above equation, leading to:

f ð x þ hÞ
f ð x Þ f ð x Þ
f ð x
hÞ


00
f ðxÞ ¼ h h :
h ð28:8Þ
f ðx
hÞ
2f ðxÞ þ f ðx þ hÞ
¼
h
For a derivative of any given order, we can construct different difference schemes
by employing the function values at points around the point of interest using
different weights. The general form of finite difference schemes can be formulated as
Temporal Finite Difference and Schemes 373

X
n
DðuÞ ¼ ai ðx þ ihÞ, ð28:9Þ
i¼
m

where mand n are positive integers. These two integers are the numbers of points
before and after the point where the derivative is calculated. The coefficients
(weights) ai and the corresponding values at the chosen points differentiate different
schemes. The sum of the coefficients needs to be zero. The finite difference schemes
introduced above are the most common ones and can be used to handle many
common applications. However, in a few cases, we still want to employ more
complicated schemes. This is especially necessary for problems with special features
such as high nonlinearity or when there is a specific requirement for the solution such
as accuracy. To construct more complicated schemes, we can resort to methods such
as the method of undetermined coefficients and polynomial approximation to calcu-
late the coefficients for any designed schemes. However, details for these methods
will be excluded here considering that their use is less common in entry-level finite
difference applications.

Temporal Finite Difference and Schemes

The finite difference in the time coordinate usually requires a different treatment
from that in the spatial coordinates due to the different features of time and space. In
the spatial finite difference context, forward and backward methods are usually
adopted; by contrast, in the temporal context, we talk more about explicit and
implicit methods. To differentiate the finite differences in space and time, subscripts
will be used for spatial finite differences, while superscripts will be reserved for the
temporal ones. An example is given in this subsection to introduce temporal finite
differences where the spatial finite differences are also included for a complete
introduction. Let us first reformulate the general transient partial differential equation
in the following way

ut ¼ f ðuÞ, ð28:10Þ

where f(u) in the equation contains all the terms except the temporal derivative term.
First, for the temporal difference, let us use the following most common form as our
purpose is to calculate the function value at the next time step n + 1 based on the
known function value at the current time step n

unþ1
un
¼ f ðuÞ, ð28:11Þ
k
where k is the time step. An Euler forward scheme is used for the temporal
derivative. The word “Euler” differentiates the temporal schemes from the spatial
schemes.
374 28 Finite Difference Method

The choice of u, i.e., u from which time step, determines the type of the temporal
scheme. If all the unknown values are from the current step, then the scheme is
explicit:

unþ1
un
¼ f ðun Þ: ð28:12Þ
k
The following solution can be readily obtained for this scheme:

unþ1 ¼ un þ k  f ðun Þ: ð28:13Þ

We can see that an explicit form of the solution can be obtained conveniently.
This provides convenience for the development of computer programs. Despite the
advantage, the explicit scheme needs to satisfy conditions for the adopted time steps,
i.e., k is small enough, to ensure stability. Otherwise, the numerical oscillation in the
values of the dependent variables between steps will increase and eventually fail the
solution process.
If f(u) depends only on the dependent variables at the next step, then we will have
the fully implicit scheme:

unþ1
un  
¼ f unþ1 : ð28:14Þ
k
An explicit mathematical function for the solution to the above equation is not
always available. Even if one is available, more effort is usually required to derive
the equation and to develop programs based on it. Despite the disadvantage, this
fully implicit scheme is unconditionally stable – there is no requirement on the time
step k for stability.
The above two types of schemes can also be mixed to construct other schemes

unþ1
un  
¼ ð1
θÞf ðun Þ þ θf unþ1 , ð28:15Þ
k
where θ is the ratio. θ values of 0 and 1 lead to explicit and fully implicit schemes,
respectively, while those values between 0 and 1 produce partially implicit ones.
Another common implicit scheme, the Newton-Nicolson scheme will be obtained
when θ equals 0.5. Partially implicit schemes are also unconditionally stable.
To gain a better idea of these schemes, let us revisit a simple parabolic PDE which
is common for time-dependent processes such as a transient heat transfer process:
2
∂u ∂ u
¼ : ð28:16Þ
∂t ∂x2
There is only one spatial dimension and one temporal dimension. We discretize
the space and time using a mesh, x1,. . .,xi. . .,xI, and a “time mesh,” t0,. . .,tn ,. . .,tN.
A uniform mesh size h and a constant time step k are adopted. Then the function
value at any point is:
Temporal Finite Difference and Schemes 375

Fig. 28.2 Explicit method

stencil

uðxi ; t n Þ ¼ uin : ð28:17Þ

We can discretize the governing equation with the Euler forward scheme for time
and the central difference scheme for space. The operation will lead to the following
discretized form of the governing equation:
n
unþ1
i
uin ui
1
2uin þ uiþ1
n
¼ : ð28:18Þ
k h2
The solution to the above equation is
 n 
unþ1
i ¼ uin þ r ui
1 þ uin þ uiþ1
n
, ð28:19Þ

where r ¼ k/h2.
The stencils to illustrate this method are plotted in Fig. 28.2. As can be seen, the
solution at one point at the new (next) step needs the information at this point and its
two neighboring points from the current time step.
For the fully implicit method, the discretization will be as follows:

unþ1
uin unþ1
2unþ1

i þ unþ1
i
¼ i
1 iþ1
: ð28:20Þ
k h 2

The solution is:

runþ1 nþ1
i
1 þ ð1 þ 2r Þui
runþ1 n
iþ1 ¼ ui : ð28:21Þ

The stencils for the full implicit method are shown in Fig. 28.3. Therefore, the
function values at one point and its two neighboring points at the new step rely on the
function value at the point from the current step.
Then a discretization of the above equation in the general form is

unþ1
uin u n
2uin þ uiþ1

n
unþ1
2unþ1
i þ unþ1
i
¼ ð1
θÞ i
1 þ θ i
1 iþ1
: ð28:22Þ
k h 2
h 2
376 28 Finite Difference Method

Fig. 28.3 Implicit method

stencil

Fig. 28.4 Crank-Nicolson

stencil

The solution to the above equation is

θrunþ1 nþ1
i
1 þ ð1 þ 2θr Þui
θrunþ1 n n
iþ1 ¼ ð1
θ Þrui
1 þ ½1
2ð1
θ Þr ui :
n
þ ð1
θÞruiþ1
ð28:23Þ

Substituting θ ¼ 1/2 into the above equation will lead to the solution for the
Newton-Nicolson method:

1 1 1 n 1 n

runþ1 nþ1
i
1 þ ð1 þ r Þui
runþ1 n
iþ1 ¼ rui
1 þ ð1
r Þui þ ruiþ1 : ð28:24Þ
2 2 2 2
The relationship between the function values at different steps is shown in
Fig. 28.4. The function values at one point and its two neighboring points at the
next step rely on the function values at these three points at the current step.
The above introduction is made based on the application of the Euler forward
scheme to the temporal derivative on the left-hand side of the equation. In fact, it is
also common to use in the temporal derivatives one-step multistage schemes and
multistep schemes, which will not be discussed in this book.

Boundary and Initial Conditions

The discretization will convert the mathematical description that is defined in a

continuous geometric domain (and possibly time) into an algebraic equation defined
at discrete mesh points (and possibly time points) in the domain (and time period).
However, the spatial and temporal discretizations alone are not enough to generate
such an algebraic equation. Shown in the following are the general forms of the
Boundary and Initial Conditions 377

algebraic equations for equilibrium problems, i.e., boundary value problem (BVP),
and transient problems, i.e., initial-boundary value problem (IBVP), respectively

BVP : AU ¼ F, ð28:25Þ
IBVP : Anþ1 Unþ1 ¼ An Un þ F, ð28:26Þ

where A is the stiffness matrix; Uis the array of unknowns, which are the values
of the function at the grid points; F is the force matrix; the superscripts are the time
steps: n is the current time step; and n + 1 is the next time step.
Boundary and initial conditions are needed to achieve the above algebraic
equation. For boundary conditions, the Dirichlet boundary condition can be applied
directly. For example, if the discretized domain starts with Point 1, then function
values from neighboring points are needed to calculate the derivatives at this starting
point. The Dirichlet boundary condition can then be assigned to “Point 0.” “Point 0”
is not necessarily explicitly considered in the algebraic equation, and thus its
derivative does not need to be represented using finite differences. The application
of the Dirichlet boundary condition u ¼ a at the left boundary just requires replacing
all the function value at “Point 0,” i.e., u0, with a. The application of the Neumann
boundary condition u' ¼ α on the left boundary (1D) involves the substitution of the
following equation into the discretization equation of relevant points, e.g., Point 1, to
cancel out u0:
u1
u0
¼ α: ð28:27Þ
h
The boundary conditions can be applied to other boundaries such as the right
boundary in a 1D domain in a similar way.
The treatment of the initial condition is straightforward. We just need to find out
the initial values at the grid points according to the initial condition,U0. Then the
solution of the transient problem will involve an iterative process starting from U0.
For example, the solution in the first step is obtained as

A1 U1 ¼ A0 U0 þ F, ð28:28Þ

where A0 is calculated with U0.

To better demonstrate the discretization including boundary and initial condi-
tions, let us recall the general example used in the previous section. A complete
description of the problem is as follows:
2
∂u ∂ u
¼ , ð28:29Þ
∂t ∂x2
ujx¼0 ¼ a, ð28:30Þ
378 28 Finite Difference Method


∂u
¼ β, ð28:31Þ
∂x x¼L
ujt¼0 ¼ u0 : ð28:32Þ

Let us discretize the domain with a length of L into I points, for which we have
I ¼ Lh þ 1. The physical process to be simulated lasts a time period of Nk. Then the
discretization starts with the temporal and spatial discretization at each point:

unþ1
u1n u0n
2u1n þ u2n

Point1 : 1
¼
k h2
nþ1
u
u2 u1
2u2n þ u3n
n n
Point2 : 2 ¼
k h2
...
unþ1
u n u n
2uin þ uiþ1
n
Point i : i i
¼ i
1
k h2
...
n
unþ1
uIn uI
1
2uIn þ uIþ1
n
Point I : I ¼
k h 2

If we incorporate the above boundary conditions, we will get:

unþ1
u1n a
2u1n þ u2n

Point1 : 1
¼
k h2
nþ1
u
u2 u1
2u2n þ u3n
n n
Point2 : 2 ¼
k h2
...
unþ1
u n u n
2uin þ uiþ1
n
Point i : i i
¼ i
1
k h2
...
unþ1
uIn uI
1 n

uIn þ βh
Point I : I ¼
k h2
The above series of equations can be represented using the following algebraic
equation:
2 3 02 3 2 312 n 3
unþ1
1 1
2 1 u1
6 unþ1 7 B6 1 7 6 1
2 1 7C6 u2n 7
6 2 7 B6 7 6 7C6 7
6 ⋮ 7 B6 ⋱ 7 k6 ⋱ 7C6 ⋮ 7
6 nþ1 7 ¼ B6 7þ 6 7C6 n 7
6 u 7 B6 7 6 1
2 1 7C6 u 7
7 h 6
1 2
6 i 7 B6 7C6 i 7
4 ⋮ 5 @4 ⋱ 5 4 ⋱ 5A4 ⋮ 5
nþ1
uI 1 1
1 uIn
Error Analysis 379

2 3
a
6 0 7
6 7
k 6 0 7
þ 26 7: ð28:33Þ
h 6⋮7
6
7
4 0 5
βh

The initial condition is applied in the first step of the iteration when n ¼ 1. The
discretized function value for the first step is:
2 3
u0
6 u0 7
6 7
6⋮7
U0 ¼ 6 7
6 u0 7 : ð28:34Þ
6 7
4⋮5
u0

Error Analysis

Error analysis is an essential part of numerical simulation methods. But it is also

worth pointing out that the error analysis is much less emphasized in engineering
applications than in the studies of numerical simulation methods in mathematics. In
many engineering applications, the error is examined by comparing the calculated
results against analytical or experimental data instead of directly studying the error
caused by the numerical schemes. This difference is especially obvious in the finite
difference method, where the error associated with numerical schemes has been
more extensively studied.
Because the algebraic equation is just an approximation, the numerical solution
U at the spatial grid points and time steps is different from the true solution U. b
b
Therefore, in general, U will not satisfy the above algebraic equation exactly. The
discrepancy is the local (truncation) error:

b ¼ F þ τ,
AU ð28:35Þ

where τ is the local error. τ is an array which has the same size as U (or U b ).
Therefore, τ is the ensemble of the errors at each spatial and temporal discretization
grid points. The true error is the global error, which is the difference between the
numerical solution and the true solution

b
E ¼ U
U: ð28:36Þ
380 28 Finite Difference Method

The local error and the global error are related via the following equation,
provided that A is invertible:

E ¼
A
1 τ: ð28:37Þ

The above equation implies that

   
kEk ¼ 
A
1 τ  A
1 kτk: ð28:38Þ

A finite difference method is convergent if kEk ! 0 as the mesh size or/and the
step size approaches zero. The above three equations correspond to three types of
criteria as the mesh size or/and the step size approaches zero:
Convergence: kEk ! 0
Consistency: kτk ! 0
Stability: kA
1k < a finite constant
Therefore, the convergence can be ensured by two conditions: stability and
consistency. The stability requires that kA
1k is smaller than a constant as the
mesh size or/and the step size approaches zero. The consistency requirement states
that kτk ! 0 as the mesh size or/and the step size approaches zero. The sign kk
means the norm of a matrix. For a given finite difference method, A is known. The
stability can be assessed with the second-order norm of a matrix. The method will
not be detailed here. Instead, we will investigate the consistency requirement, which
is related to the local truncation error. Another reason to look into the local error is
that the order of accuracy of finite difference schemes is determined by the local
truncation error.
The local truncation error of finite difference schemes can be obtained via the
following Taylor series

f 0 ð xÞ f ð 2 Þ ð xÞ 2
f ð x þ hÞ ¼ f ð x Þ þ hþ h þ ...,
1! 2! ð28:39Þ
f ð n Þ ð xÞ n  
þ h þ O hnþ1
n!
where n! is the factorial of n and O(hn + 1) is the difference between the Taylor
polynomial of degree n and the original function, which is a function of hn + 1. An
approximation to the first derivative of the function “f” can be obtained by truncating
the Taylor polynomial:

f 0 ð xÞ
f ð x þ hÞ ¼ f ð x Þ þ h þ :... ð28:40Þ
1!
Similarly, we can obtain an approximation to the function at any given point
around the point of interest:

f 0 ð xÞ  
f ð x þ hÞ ¼ f ð x Þ þ h þ O h2 : ð28:41Þ
1!
Extension to 2D and 3D Dimensions 381

Based on the above Taylor expansion, we can readily obtain the error for any
finite difference schemes such as the forward, backward, and central difference
schemes.
Forward:
 
f ð x þ hÞ
f ð x Þ O h2
¼ f 0 ðxÞ þ ¼ f 0 ðxÞ þ OðhÞ ð28:42Þ
h h
Backward:

f ð x Þ
f ð x
hÞ
¼ f 0 ðxÞ þ OðhÞ ð28:43Þ
h
Central:

f ðx þ hÞ
f ðx
hÞ  
¼ f 0 ð x Þ þ O h2 ð28:44Þ
h
Therefore, the forward, backward, and central difference schemes have the first-
and second-order accuracy, respectively. The order of accuracy of temporal differ-
ence schemes can be obtained in a similar way.

Extension to 2D and 3D Dimensions

The extension of the above discretization to 2D and 3D cases can be implemented in

different ways. The major challenge is that it is hard to deal with arrays more than
two dimensions. In a 2D domain, the spatial discretization will lead to function
values in the form of 2D arrays corresponding to the rows and columns of the points
in the discretized 2D domain. We thus need to convert this 2D array into a 1D array,
i.e., a column array, so that the stiffness matrices can be simplified from a 4D array to
a 2D one.
The above extension can be implemented via both element-wise operations and
matrix operations. For element-wise operations, we just need to find out the rela-
tionship between the dependent variable and the elements in the stiffness matrix.
When conducting the conversion, the stiffness matrix is converted accordingly. This
will be shown in the next chapter for the finite volume method. In a typical matrix
operation, the conversion can be finished for a 2D domain in the following way
2
∂ u
: ðI  LÞU ð28:45Þ
∂x2
2
∂ u
: ðL  IÞU, ð28:46Þ
∂y2
where I is the identity matrix and L is the matrix generated by the stiffness matrix
excluding any boundary conditions:
382 28 Finite Difference Method

2 3

2 1
6 1
2 1 7
6 7
6 ⋮ 7
L¼6
6
7:
7 ð28:47Þ
6 1
2 1 7
4 ⋮ 5
1
1

The boundary conditions can be applied either by modifying the above L matrix
or the stiffness matrix generated by the above operations specified in Eqs. 28.45
and 28.46.

Practice Problem

Problem Description

Develop FDM code to solve the problem described in Chapter 11. The material
constituting the square area (1 m by 1 m) is water. The initial temperature is
273.15 K. The top boundary is set to 293.15 K, while the other boundaries are
thermally insulated. Please calculate the temperature distribution at the end of 1 day.

Results

Figure 28.5 illustrates the temperature distribution.

Temperature field solved using FDM

286

284

282

280

278

276

274

272
15
12
10 10
8
5 6
4
2
0 0

Fig. 28.5 Temperature distribution from the numerical simulation

Practice Problem 383

Appendix: MATLAB Script File

clc;clear;
h=0.1;
k=60;
lambda=1;
rho=1000;
C=4200;
a=lambda/rho/C;
b=a*k/h^2;
m=(1/h+1);
U=273.15*ones(m^2,1);

figure(1)

Ky=[-2 1 0 0 0 0 0 0 0 0 0;...
1 -2 1 0 0 0 0 0 0 0 0;...
0 1 -2 1 0 0 0 0 0 0 0;...
0 0 1 -2 1 0 0 0 0 0 0;...
0 0 0 1 -2 1 0 0 0 0 0;...
0 0 0 0 1 -2 1 0 0 0 0;...
0 0 0 0 0 1 -2 1 0 0 0;...
0 0 0 0 0 0 1 -2 1 0 0;...
0 0 0 0 0 0 0 1 -2 1 0;...
0 0 0 0 0 0 0 0 1 -2 1;...
0 0 0 0 0 0 0 0 0 1 -1];

Kx=[-1 1 0 0 0 0 0 0 0 0 0;...
1 -2 1 0 0 0 0 0 0 0 0;...
0 1 -2 1 0 0 0 0 0 0 0;...
0 0 1 -2 1 0 0 0 0 0 0;...
0 0 0 1 -2 1 0 0 0 0 0;...
0 0 0 0 1 -2 1 0 0 0 0;...
0 0 0 0 0 1 -2 1 0 0 0;...
0 0 0 0 0 0 1 -2 1 0 0;...
0 0 0 0 0 0 0 1 -2 1 0;...
0 0 0 0 0 0 0 0 1 -2 1;...
0 0 0 0 0 0 0 0 0 1 -1];

F=[293.15;0;0;0;0;0;0;0;0;0;0]*ones(1,m);
%F=[293*ones(m,1);zeros(m^2-2*m,1);273*ones(m,1)]
F=F(:);

K_2D=kron(eye(m),Ky)+kron(Kx,eye(m));

for i=1:60*24
t=i*k
U=U+b*K_2D*U+b*F;
U=reshape(U,m,m);
384 28 Finite Difference Method

surf(U)
U=reshape(U,m^2,1);
drawnow;
title('Temperature field solved using FDM')
end
Chapter 29
Finite Volume Method

Introduction

The finite volume method (FVM) is another method for discretizing the continuous
mechanics description of a physical process in terms of partial differential equations
and other auxiliary equations into an algebraic equation(s) (LeVeque 2002; Toro
2013). Similar to the FEM, values are calculated at discrete grid points in a meshed
geometry (and time steps). The “finite volume”, or control volume, refers to the
small volume surrounding each node point on a mesh.
One major difference between the FVM and the FDM is that the FVM is based on
the integral form of the governing equations instead of the differential form. In the
FVM, this discretization is conducted over each control volume, which can ensure
that the flux entering the volume is identical to that leaving the volume. This endows
FVM an advantage over other numerical simulation methods: the conservation of the
conserved quantities can be ensured during the numerical process. This is very
attractive for conservation problems such as fluid dynamics. This is possibly one
of the major reasons why the FVM is predominant in computational fluid dynamics.
The other major advantage of the FVM is that the method can be easily formulated to
handle unstructured meshes, which provide higher efficiency and more flexibility for
meshing.
In this chapter, we will first introduce the integral form of the governing equation
and show how to formulate this equation into the form for the discretization in the
FVM. Then we will discuss the discretization process using a 2D example
by including the boundary and initial conditions. Extensions from 2D to
multidimensional problems will be made to allow for structured 3D and even more
complicated meshes. In the practice problem section, the problem solved in the FDM
chapter will be solved again with the FVM.

© Springer International Publishing AG, part of Springer Nature 2018 385

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_29
386 29 Finite Volume Method

Integral Form of the Governing Equation

The following general differential form of conservation problems has been intro-
duced at the beginning of the third part of this book:

∂u
þ∇  ðuvÞ ¼ ∇  ðα∇uÞ þ g : ð29:1Þ
∂t |fflfflfflfflfflffl{zfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |{z}
|{z} Advection Diffusion ðConductionÞ Source
Accumulation

The diffusion term in the equation also contains other mechanisms with similar
mathematical formulations such as dispersion. The FVM requires that the above
equation is satisfied over the control volume VP around the centroid of the volume,
leading to the following integral form of the governing equation:
 ð ð 
Ð tþΔt ∂
udV þ ∇  ð uv ÞdV dt,
t
∂t V P VP ð29:2Þ
Ð tþΔt hÐ Ð i
¼ t V P ∇  ðα∇uÞdV þ V P gdV dt

or we have
ð ð ð ð
∂
udV þ ∇  ðuvÞdV ¼ ∇  ðα∇uÞdV þ gdV: ð29:3Þ
∂t VP VP VP VP

Applying the general form of the Green’s theorem, we reformat the above
equation into
ð ð ð ð
∂
udV þ uv  dS ¼ α∇u  dS þ gdV, ð29:4Þ
∂t VP ∂V ∂V VP

where S is the vectoral surface area of the control volume’s boundary ∂V.
Let us first assume the variation of the dependent variable is linear over a control
volume, which can ensure a second-order accuracy:

uðrÞ ¼ uC þ ðr  rC Þ  ð∇uÞC , ð29:5Þ

where the conserved quantity u is a function of the position vector r and the subscript
C represents the centroid of the control volume under consideration.
In the above equation, the volume integral in the first term on the left-hand side of
the equation can be reformulated as
ð ð
 
udV ¼ uC þ ðr  rC Þ  ð∇uÞC dV ¼ uC V C : ð29:6Þ
VP VP

The other terms can be transformed in a similar way, leading to the following
equation:
Boundary and Initial Conditions 387

ð tþΔt hX X i
uC V C ¼ ð∇uÞc  αc S þ gC V C  vuc  S dt, ð29:7Þ
t

where the lowercase letter “c” in the subscripts indicates the centroid of the surface
around the control volume. The summation is conducted over all the boundary
segments, such as all the faces of a hexagon cell in an unstructured grid.
To numerically deal with the integration, the above equation can be further
changed into

δt hX X i
uC t 2 ¼ uC t 1 þ αc S  ð∇uÞc þ gC V C  αc S  vuc t1 , ð29:8Þ
VC
where t2 and t1 are two consecutive time points for temporal integration, for which
t2  t1 ¼ δt, and all the terms on the right-hand side of the equation are based on uC t1
or the value(s) of uC from even earlier time steps.

Boundary and Initial Conditions

This section introduces the discretization of the governing equation with the inclu-
sion of the boundary and initial conditions. The treatments of the boundary and
initial conditions are similar to those in the FDM. The boundary conditions can be
included via the discretized equation in the control volumes on the boundary. The
values of the dependent variable assigned via the Dirichlet boundary condition are
included in the uC t1 and (— u)c. The other terms can also include these boundary
function values if they are functions of the dependent variable. The values of the
derivatives on the boundary, which are given out by the Neumann boundary
condition, can be used to replace the (— u)c term directly.
A simple 1D example will be used to illustrate the discretization with the
boundary and initial conditions in detail. However, it is worthwhile to mention
that, in 1D problems or even many 2D problems with structured grids, the FVM
can lead to the same algebraic equation as the FDM. The above integral equations
use symbols that are common in the FVM. These symbols will be gradually changed
to the symbols used in the FDM to compare and relate these two methods for a better
understanding of discretization.
Let us first revisit the general parabolic equation with the typical boundary
conditions as used in the previous chapter:
2
∂u ∂ u
¼ , ð29:9Þ
∂t ∂x2

ujx¼0 ¼ a, ð29:10Þ

∂u
¼ β, ð29:11Þ
∂x x¼L
388 29 Finite Volume Method

ujt¼0 ¼ u0 : ð29:12Þ

In order to obtain an algebraic equation which is comparable to that in the FDM,

we need to discretize the domain in the following way. The 1D domain with a length
of L is divided into I  1 portions with a uniform mesh size of h: there are two
portions with a length of h/2 on the two ends (boundary) and I  3 portions in the
interior of the domain. Therefore, we have I ¼ Lh þ 1. The physical process to be
simulated lasts a time of Nk. The general form of the integral equation including an
accumulation and a diffusion term is as follows:

Δt hX i
uC t 2 ¼ uC t 1 þ αc S  ð∇uÞc : ð29:13Þ
VC
This is a 1D problem, so there is no surface area between different control
volumes (or cells). Let us use a unit area instead, thus |S| ¼ 1. S has a positive
direction if it is on the right side of the cell and negative on the left. The centroid of
the cell is the middle point of the cell. Accordingly, the above integral equation can
be transformed into
"  #
k ∂u ∂u
u nþ1
¼u þ
n
 : ð29:14Þ
h ∂x right ∂x left

The subscript “C” is removed as we use the function value in the middle of a cell
uC to represent the u of that cell. un + 1 and un are the values at the next and current
time steps, respectively. ∂u
∂x right
and ∂u∂x left
are the derivative on the right and left
ends of the cell, respectively, whose signs in the equation are determined by the
direction of S and (— u)c. The centroid of the boundary surface does not require
special treatments as the boundary segments are two infinitesimally small points in
this 1D domain.
Without considering the boundary conditions, the governing equation can be
directly discretized as

k u2n  u1n u1n  u0n
Point 1: unþ1
1 ¼ u1nþ  :
h h h

k u3n  u2n u2n  u1n
Point 2: unþ1
2 ¼ u2 þ
n
 :
h h h
...
 n
k uiþ1  uin uin  ui1
n
Point i: unþ1
i ¼ uin þ  :
h h h
...
 n
k uIþ1  uIn uIn  uI1
n
Point I: unþ1
I ¼ uin þ  :
h h h
Boundary and Initial Conditions 389

The boundary conditions are implemented by changing the discretized equations

in the boundary cells, i.e., Point 1 and Point I, with the following equations:
u0n ¼ a, at Point 1
∂u 
∂x right
¼ β, at Point I

The above set of equations with the boundary condition equations can be
transformed into the following algebraic equation:
2 3 02 3 2 3 12 3
unþ1
1 1 2 1 u1n
6 unþ1 7 B6 1 7 6 1 2 1 7 C6 u n 7
6 2 7 B6 7 6 7 C6 2 7
6 7 B6 7 6 7 C6 7
6 ⋮ 7 B6 ⋱ 7 k6 ⋮ 7 C6 ⋮ 7
6 7 B6 7þ 6 7 C6 7
6 unþ1 7 ¼B6 7 h2 6 2 7 C6 u n 7
6 i 7 B6 1 7 6 1 1 7 C6 i 7
6 7 B6 7 6 7 C6 7
4 ⋮ 5 @4 ⋱ 5 4 ⋮ 5 A4 ⋮ 5
unþ1
I 1 1 1 uIn
2 3
a
607
6 7
6 7
k6 07
þ 26 7:
h 6⋮7
6
7
6 7
405
βh
ð29:15Þ

The above equation is identical to the algebraic equation obtained using the FDM
in the previous chapter. The initial condition is applied in the first step of the iteration
with
2 3
u0
6 u0 7
6 7
6⋮7
U0 ¼ 6 7
6 u0 7 : ð29:16Þ
6 7
4⋮5
u0

Another difference can be observed between the FVM and FDM: the FVM is
discretized over the control volumes, while the FDM is discretized at the grid points.
That is the reason why the meshing is done in the above specific way to obtain the
same algebraic equation.
390 29 Finite Volume Method

Multidimensional Problems

2D Structured Grids

The above example demonstrates the concept of the FVM in a 1D case. In 2D or 3D

cases, the treatment is much more complicated, especially when complicated mesh
structures and unstructured meshes get involved.
If we still use a structured grid such as the square mesh as below, we can still
obtain the same discretization results as the FDM. However, for such a typical 2D
grid, we now have more volumes (or cells) around the volume of interest. It is
common to mark these neighboring volumes as “N” (North), “E” (East), “S” (South),
and “W” (West) in FVM textbooks. The volume VC and the vectoral area of its
boundaries S will be the same. The (— u)C at the centroid of a face or a control
volume can be determined using the function values at the centroids of its two
neighboring volumes. This is because the centroid of a boundary face is always the
middle point of the line connecting the centroids of the two volumes sandwiching
this face (Fig. 29.1).
In the above case, the integral form of the governing equation can be formulated
as follows:

αc Δt hX i
uC nþ1 ¼ uC n þ S  ð∇uÞc : ð29:17Þ
VC
The above integral governing equation can be discretized with the computational
domain illustrated in the following figure (Fig. 29.2).
If we use an element-wise treatment, the above equation will lead to the following
global algebraic equation.

Unþ1 ¼ AUn , ð29:18Þ

Fig. 29.1 Control volume

and its neighboring volumes
in a 2D structured grid
Multidimensional Problems 391

Fig. 29.2 Numbering of

grid points

in which
2 3
u1 , 1
6 ⋮ 7
6 7
6 7
6 u1 , m 7
6
U¼6 7: ð29:19Þ
7
6 u2 , 1 7
6 7
4 ⋮ 5
um , n

The value of the dependent variable at node (i,j) is located in the (i + j)th row of U.
Then the coefficients in the global stiffness matrix are a, b, d, and e corresponding to
the north, south, east, and west neighboring volumes as shown below, respectively.
The values of the coefficients are obtained from the integral form of the governing
αc Δt
equation: dxdy , in which dx and dy are the mesh size along the xand y directions,
respectively. The coefficients corresponding to the north and south control volumes
are the (i + jm)th and (i + j + m)th elements of the (i + j)th row.

Node i þ j-m iþj1 iþj iþjþ1 iþjþm

2 3
⋱
6 7
6 ⋮ 7
6 7 ð29:20Þ
6 7
A¼6 a . . . e 1 d ... b 7:
6 7
6 ⋮ 7
4 5
⋱

More Complicated Grids

However, special treatments are needed for more complicated 2D/3D structured
grids and unstructured grids. The first challenge is to determine the function values
on the boundary faces of the volumes and those parameters depending on these
392 29 Finite Volume Method

values. Several terms defined at the centroid of the boundaries are generated when
applying Green’s theorem to generate the final integral form of the governing
equation. Various convection differencing schemes have been proposed to deter-
mine the surface values. However, in arbitrarily unstructured meshes, it seems to be
impractical to use any values other than the function values at the centroid of the
control volumes to calculate these values. Therefore, the common convection
differencing schemes are limited to the centroid function values of the control
volume and its nearest (neighboring) volumes. One typical way is the central
differencing, in which the surface value is determined based on the relative position
between the two volumes on two sides of the surface, i.e., the control volume of
interest and the other neighboring volume. Based on the linear variation assumed for
the dependent variable in local volumes, the following general equation can be used
to calculate the surface values.

jcXj jcCj
uc ¼ uC þ uX , ð29:21Þ
jCXj jCXj

where X represents the centroid of the neighboring volume and |CX| is the length of the
line connecting C and X. Additionally, upwind differencing, in which the surface value
is determined according to the direction of the flow, is also common. This method
exhibits advantages in ensuring boundedness with a sacrifice of accuracy. Blended
differencing as a mixture of the above two methods has also been proposed (Fig. 29.3).
The diffusion term also requires special attention. This is because in an unstruc-
tured grid, the direction of the outward normal, i.e., S, is not necessarily the same as
the line connecting the centroids of the two volumes on two sides of the boundary
face. The S  (— u)c splits into an orthogonal part and a non-orthogonal part as
follows:
uC  uX
S  ð∇uÞc ¼ kSk þ k  ð∇uÞc , ð29:22Þ
kdk

where kSk and kdk are the magnitudes of S and the vectoral length between the
centroids of the control volume of interest and its neighboring volume (s) on the

Fig. 29.3 Unstructured grid

and treatment
Practice Problem 393

other side of the boundary face d, respectively; uX is the function value at the
centroid of the neighboring volume; and k is a vector used to represent the
non-orthogonal correction. Several decompositions are available to account for the
non-orthogonal corrections such as the minimum correction approach, the orthogo-
nal correction approach, and the over-relaxed approach (Jasak 1996).

Practice Problem

Problem Description

Develop FVM code to solve the problem described in Chap. 28. The material consti-
tuting the square area (1 m by 1 m) is water. The initial temperature is 273.15 K. The top
boundary is set to 293.15 K, while the other boundaries are thermally insulated. Please
calculate the temperature distribution at the end of one day.

Results

The following figure illustrates the temperature distribution (Fig. 29.4).

Temperature field solved using FVM

295

290

285

280

275

270
10
10
8
5 6
4
2
0 0

Fig. 29.4 Temperature distribution from the numerical simulation

394 29 Finite Volume Method

Appendix: MATLAB Script File

% The initial temperature is 273.15.

% The top BC temperature is fixed at 293.15; while all the other BCs are
thermally insulated.
% A*U^n+1=B(U^n)
% Initiation
clear;
clc;

nx=10;
ny=10; %node numbers along x and y;
theta=0.5;
alpha=1/4.2e6; %thermal diffusivity;
Lx=1.;
Ly=1.; %dimensions of domain;
dx=Lx/(nx-1);
dy=Ly/(ny-1);
tTotal=3600*24;
dt=60; %total calculation of the physical time is 10;
Tinit=273.15; %initial temperature is 700;
Tboundary=293.15; %constant boundary temperature
Tmat(:,:)=Tinit*ones(nx,ny);

k=0;
F=alpha*dt/(dx*dy); %Coefficient associated with flux
a=F;b=F;d=F;e=F; %Coefficient in the A matrix associated with
neighboring cells
c=1+a+b+d+e; %Coefficient in the A matrix associated with considered
cell
A=zeros(nx*ny,nx*ny);
B=ones(nx*ny,1)*Tinit;

for t=0:tTotal/dt
for i=1:nx
for j=1:ny
temp=(i-1)*ny+j;
%First row of cells (top, i=1) correspond to the first row of A and first
ny points of B
%Interior of the Domain
if(i~=1&i~=nx&j~=1&j~=ny)
A(temp,temp)=c; %P
A(temp,temp-ny)=-a; %N
A(temp,temp+ny)=-b; %S
A(temp,temp+1)=-d; %E
A(temp,temp-1)=-e; %W

%Different Boundaries (BCs)

elseif(i==1) %North/Top
A(temp,:)=0;
%A(:,temp)=0;
A(temp,temp)=1;
B(temp)=Tboundary;
Practice Problem 395

elseif (i==nx&j~=1&j~=ny) %South/Bottom

A(temp,temp)=c-b; %P
A(temp,temp-ny)=-a; %N
%A(temp,temp+ny)=-b; %S
A(temp,temp+1)=-d; %E
A(temp,temp-1)=-e; %W
elseif (j==1&i~=nx) %West/Left
A(temp,temp)=c-e; %P
A(temp,temp-ny)=-a; %N
A(temp,temp+ny)=-b; %S
A(temp,temp+1)=-d; %E
%A(temp,temp-1)=-e; %W
elseif (j==ny&i~=nx) %East/Right
A(temp,temp)=c-d; %P
A(temp,temp-ny)=-a; %W
A(temp,temp+ny)=-b; %E
%A(temp,temp+1)=-d; %S
A(temp,temp-1)=-e; %N
elseif (i==nx&j==1) %Left Lower Corner
A(temp,temp)=c-b-e; %P
A(temp,temp-ny)=-a; %N
%A(temp,temp+ny)=-b; %S
A(temp,temp+1)=-d; %E
%A(temp,temp-1)=-e; %W
elseif (i==nx&j==ny) %Left Right Corner
A(temp,temp)=c-b-d; %P
A(temp,temp-ny)=-a; %N
%A(temp,temp+ny)=-b; %S
%A(temp,temp+1)=-d; %E
A(temp,temp-1)=-e; %W
end
end
end
T=A\B;
B=T;
end

for i=1:nx,
for j=1:ny,
temp=(i-1)*nx+j;
Tmat(i,j)=T(temp);
end
end
%% plot Surface temperature on Z-axis
%X and Y axis as cordinates of the plate
[i,j]=meshgrid(1:1:nx,1:1:ny);
contourf(i,j,Tmat);
surf(Tmat);
title('Temperature field solved using FVM')
Chapter 30
Finite Element Method

Introduction

The finite element method (FEM) provides a much different way from the FDM and
FVM for the discretization of the continuous mechanics description of a physical
process. The discretization is made over a domain consisting of subdivisions called
elements. These elements are essentially the same as those control volumes in the FVM
from a geometric perspective. However, the FEM more emphasizes the grid points as
the apexes of the element rather than the centroids, volume, and boundary faces that are
emphasized in the FVM. The governing equation is discretized at all the grid points of
the element to form a local algebraic equation. Since different elements share grid
points, the local algebraic equation will be ensembled into the global governing
equation to add up the coefficients from different elements for the same point.
Another major difference is that the FEM adopts the weak form of the governing
equation instead of the differential form in the FDM or the integral form in the FVM.
Usually, either the Galerkin method or variational methods from the calculus of
variations are used to derive the weak form. The idea underlying the Galerkin and the
variational methods is analogous to the principle of minimum energy in physics. In
the FEM, we use polynomials of different orders, which also determine the order of
the elements, to approach the true solution. The Galerkin method or variational
methods will minimize the error of the approximation by fitting trial functions into
the governing equation.
Due to the above characteristics, the FEM owes several attractive advantages. The
FEM is a good choice for analyzing problems with complicated geometries. Second,
since the FEM uses the weak form of the governing equation, the requirement on
continuity is loosened. Therefore, it provides a better performance when handling
discontinuities compared to the FDM and FVM. The FEM also shows good flexi-
bility and good performance when dealing with moving mesh problems. In addition,
the FEM is also suitable for problems with varying precision requirements across the
computational domains. In virtue of these advantages, the FEM has been developed

© Springer International Publishing AG, part of Springer Nature 2018 397

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2_30
398 30 Finite Element Method

into a predominant numerical simulation technique in engineering. It is widely

adopted in most of the multiphysics areas. One noticeable exception is the compu-
tational fluid mechanics, in which the FVM is predominant possibly due to its ability
in ensuring conservation of water mass. The numerical simulation analysis with the
FEM is usually termed the finite element analysis (FEA).
It is interesting to note that the FEM is studied in both mathematics and engi-
neering. However, the FEM is usually introduced in significantly different ways in
these two areas. In mathematics, the FEM is introduced as a special case of the
Galerkin method. But in engineering, an introduction to the FEM is usually started
from the variational methods or even the Ritz method. In addition, emphasis is
placed on the performance of different types of elements on the overall performance
of the FEM applications.
A typical procedure for implementing the FEM is slightly more complicated than
that in the FDM and FVM. The first few steps are the establishment of the mathe-
matical model including the geometrical domain and the discretization of this
domain (meshing). This is not different from that in the FDM and FVM. The next
step is to transform the governing equation into the weak form. Some or all the
boundary conditions can be incorporated into the weak form of the governing
equation(s). Then we will choose the order of the element. The shape and order of
the element will determine the shape functions. The shape functions will be adopted
to discretize the governing equation to obtain the local algebraic equation for a single
element. The algebraic equations for all the elements within the computational
domain are ensembled to obtain the global algebraic equation. The last step is to
solve the global algebraic equation and to analyze and visualize the results.
In the following sections, at first, the focus will be placed on the Galerkin Method
for obtaining the weak form of the governing equation. Then, using thermomechanics
as an example, the discretization of the weak form of the governing equations with
Neumann boundary conditions into a local algebraic equation will be explained. The
discretization of the thermal field and mechanical field, including their couplings, will
be introduced separately. Two common types of 2D elements, more specifically their
shape functions, are given to help readers understand the discretization process. Other
significant contents of the FEM, such as different element types, stiffness matrix
ensembling for obtaining the global algebraic equation, and isoparametric elements,
will not be detailed here, as they can be easily found in many reference books for the
FEM. Finally, MATLAB FEM code is given for solving the problem that has been
solved in the FDM and FVM chapters.

Galerkin Method and Weak Form

Suppose we have a linear differential operator D acting on a function u to produce a

function, p(x),

DðuðxÞÞ ¼ pðxÞ: ð30:1Þ

Galerkin Method and Weak Form 399

In the minimal weight residual methods, which the FEM belongs to, we approx-
imate u by a function, u~, which is a linear combination of basic functions chosen
from a linearly independent set:

X
n
u ffi u~ ¼ u i φi : ð30:2Þ
i¼1

Substituting the above approximation into the differential operator, D, in general,

the result of the operation is not p(x) anymore. Accordingly, an error or residual
exists,

E ðxÞ ¼ RðxÞ ¼ Dðu~ðxÞÞ  pðxÞ 6¼ 0: ð30:3Þ

The notion in the minimal weight methods is to force the residual to approximate
zero in some average sense over the domain, that is,
ð
RðxÞW i dx ¼ 0, i ¼ 1, 2, . . . , n ð30:4Þ
X

where the number of weight functions Wi is exactly equal to the number of unknown
constants ui in u~. The result is a set of algebraic equations (n) for the unknown
constants, ui. There are five common minimal weight residual methods, which are
differentiated by the choices of the Wi’s. These methods include the collocation
method, sub-domain method, least squares method, Galerkin method, and method of
moments.
In the Galerkin Method, the derivative of the approximating function is used as
the weight function:

∂~
u
Wi ¼ ¼ φi ðxÞ: ð30:5Þ
∂ui
Then we have
ð
½Dðu~ðxÞÞ  pðxÞφi ðxÞdx ¼ 0: ð30:6Þ
X

Equation 30.6 is the weak formulation, or weak form, of the original differential
equation, Eq. 30.1. φi(x) is the shape function in the FEM. If the element type is
fixed, then all the shape functions for each point will be known prior to the solution
process. The shape functions for two simple types of elements will be given in a later
section. The outcome of using the weak form is to convert the governing PDE, which
requires the equality of the equation at every infinitesimal point within the domain
and on the boundary, to the weak form of the equation, which requires that the above
integral form of a function with the residual of the numerical approximation and the
shape functions equals zero. The integral form will bring additional benefits: a
relatively lower requirement on the continuity and differentiability of the original
400 30 Finite Element Method

differential equation. This weak form is necessary in the FEM for obtaining the
algebraic equation system. However, it is worthwhile to mention that the above way
to obtain the weak form can be reinterpreted in a more straightforward way using test
functions, which are random functions as follows:

X
n
v ffi ~v ¼ v i φi : ð30:7Þ
i¼1

The procedure is to multiply the original PDE with the above formation of the test
function and then to integrate their product over the domain. As the test function is a
random function and exists in every term of the integral form of the equation, vi can
be eliminated, and we will obtain the same weak formulation as Eq. 30.6. Detailed
operations will be demonstrated in an example to be introduced in the following
section.

Example of Discretization: Thermomechanics

Thermal Field

The governing equation of the thermal field, i.e., heat equation, is first recalled.
When material properties are constants and advection is overlooked in solids, the
heat equation can be reduced to

∂ðT Þ
ρC ¼ ∇  ðλ∇T Þ þ g, ð30:8Þ
∂t
where ρ is the density, C is the gravimetric heat capacity, T is temperature, λ is the
thermal conductivity, v is the velocity of material particles, and g is the heat source.
The above equation can be transformed into the weak form as follows:
ð ð ð ð
∂T
ρC v dΩ ¼ λ ð∇vÞ  ð∇T ÞdΩ þ v  ðλ∇T Þ  ndS þ vgdΩ, ð30:9Þ
∂t
Ω Ω ∂Ω Ω

where Ω is the computational domain, ∂Ω is the boundaries of the computational

domain, v is the test function, n is the outward unit normal vector of the boundary,
and (λ∇T)  n is the source on the boundaries.
In the FEM, the above equation in continuum mechanics will be discretized into
an algebraic equation system. Taking the bilinear quadrilateral element (Quadrilat-
eral 4 (Q4), see Fig. 30.1) element in 2D, for example, the values of the dependent
variable T and test function v can be related to the shape functions NTi (NT) and test
functions vi (v) of the element in the following way:
Example of Discretization: Thermomechanics 401

Fig. 30.1 Q4 and Q9

element

X
4
T¼ N Ti T i ¼ NT  T,
i¼1
X4
v¼ N Ti vi ¼ NT  v,
i¼1

where,

NT ¼ ½N 1 N 2 N 3 N 4 ,

T ¼ ½ T1 T2 T3 T 4 T :

The discretized governing equation is formulated as

ð ð
∂TT
ρC ðNT vÞ dΩ ¼ λ ðAT NT vÞT  ðAT NT TÞdΩ
∂t
Ω ðΩ ð ð30:10Þ
þ ðNT vÞT ðJ  nÞdS þ ðNT vÞT gdΩ,
∂Ω Ω

where N is the shape function matrix and A is the matrix of the differential operator.
The integration will be conducted using the Gauss-Legendre numerical integration.
The time differential on the left side of the equation will be handled with an iterative
process:
0 1 0 1
ð ð
ρC    
@ NT NdΩA Tnþ1  Tn ¼ @λ BT BdΩA θTnþ1 þ ð1  θÞTn
Δt
Ω ðΩ ð
þ NT ðJn  nÞdS þ NT gn dΩ,
∂Ω Ω

ð30:11Þ
402 30 Finite Element Method

where Tn+1 and Tn are the vectors of unknowns at the next and the current time steps,
respectively. It is noted that the test function is eliminated due to its presence in all
terms. Detailed
2 3expressions of the matrices in the above equation are as follows:
∂
6 ∂x 7
AT ¼ 4 ∂ 7
6
5 (for equation with scalar dependent variables in 2D, independent of
∂y
element type),
2 3
∂N 1 ∂N 2 ∂N 3 ∂N 4
6 ∂x ∂x ∂x ∂x 7
BT ¼ 6
4 ∂N 1 ∂N 2 ∂N 3
7
∂N 4 5,
∂y ∂y ∂y ∂y

where Ni’s are the shape functions corresponding to the type of the adopted element.
For example, the Q4 element has four functions N1, N2, N3, and N4.
There are two different ways to deal with the boundary source term specified by
the Neumann boundary condition in the FEM. The first way is to calculate the term
with shape functions associated with the nodes on the boundaries when discretizing
the weak form of the governing equation, e.g., Eq. 30.9, while the other one is to
establish the force (or source) vector matrix separately by multiplying the boundary
source with the boundary length (area). For both of them, the boundary force needs
to be formulated based on physical mechanisms such as convective heat transfer,
which can be predicted with Newton’s law of cooling:

J  n ¼ hðT env  TÞ, ð30:12Þ

where Tenv is the environmental temperature. The second method is employed in this
code for the practice problem. When no heat source exists, the discretized governing
equation can be expressed as
   
KTa Tnþ1  Tn ¼ KTc θTnþ1 þ ð1  θÞTn þ FTs n þ FTb n , ð30:13Þ

where KTa and KTc are the stiffness matrices relevant to the accumulation (a) and
conduction (c), respectively, and FTs and FTb are the force vectors that correspond to
source on boundary surface (s) and body source (b, corresponding to g). The term
“force” is adopted here to represent the generalized force, source. In the current
study, no source (body force) is present, so FTb is excluded. The stiffness matrices
are calculated using the following equations:
0 1
ð
ρC
KTa ¼@ NT T NT dΩA, ð30:14Þ
Δt
Ω
Example of Discretization: Thermomechanics 403

0 1
ð
KTc ¼ @λ BT T BT dSA: ð30:15Þ
Ω

When θ ¼ 0, the explicit scheme is used. The explicit scheme is believed to be

conditionally stable depending on the values of time step and mesh size. When
θ ¼ 1/2, the Crank-Nicolson scheme is produced. The full explicit scheme can be
obtained by setting θ ¼ 1. Although θ may vary between 0 and 1, the most
commonly used values are those for the fully implicit scheme and the Crank-
Nicolson scheme.
If the explicit method is adopted, the solution at the current step can be obtained
as

KTa þ ð1  θÞKTc n FTs

Tnþ1 ¼ CT Tn þ MT ¼ T þ : ð30:16Þ
KTa  θKTc KTa  θKTc

Mechanical Field

In addition to the governing equation of the thermal field, we here also show the
discretization of the mechanical field including the coupling from the thermal field.
Accordingly, the influence of the deformation on the material properties, in both the
thermal and mechanical fields, is neglected. The governing equation is the equation
of equilibrium:

∇  σe þ g ¼ 0, ð30:17Þ

where σe is the elastic stress (second-order tensor with Tensor notation) and g is the
body force. The elastic stress is the difference between the total stress and the
thermal stress as follows:

σe ¼ σe  σt ¼ ∇u  αðT  T ref ÞI, ð30:18Þ

where I is the identity matrix to allow for the fact that the thermal strain contributes
to normal strains instead of deviatoric strains. It is seen from Eq. 30.18 that the
element used for the mechanical field should be one order higher than that used for
the thermal field to obtain smooth results for the strain. For example, if a linear
element, Q4, is adopted for the thermal field, a quadratic Lagrangian element, such
as Q9 (see Fig. 30.1), is needed for the mechanical field. I takes the following form:

I ¼ 1 in 1D,
404 30 Finite Element Method

 
1 0
I¼ in 2D,
0 1

2 3
1 0 0
and I ¼ 4 0 1 0 5 in 3D:
0 0 1

Then Navier’s equation is obtained as the governing equation:

∇  fD½∇u  αðT  T ref ÞIg þ g ¼ 0, ð30:19Þ

where u is the displacement, D is the stiffness tensor, α is the coefficient of linear

expansion, Tref is the reference temperature for calculating thermal strain, and g is the
body force. The above equation is transformed into the weak form as
ð ð
∇v  ½D∇u  αDðT  T ref ÞIdΩ ¼ v  ½D∇u  αDðT  T ref ÞI  ndS
Ω ∂Ωð

þ vg:
Ω
ð30:20Þ

The discretization of the above governing equation leads to

ð ð
 
ðAM NM ÞT  ðDAM NM ÞdΩ θUnþ1 þ ð1  θÞUn ¼ NM T  ðσ  nÞdS
Ω ð ð ∂Ω
ð30:21Þ
T
þ NM gdΩ  α ðAM NM Þ DNT ðTn  T ref ÞIdΩ:
T

Ω Ω

where σ  n is the boundary force vector. The above equation is simplified into
ð ð ð
 nþ1 
BM DBM dΩ θU
T
þ ð1  θÞU ¼ NM ðσ  nÞdS þ NM T gdΩ
n T

Ω ð
∂Ω Ω
ð30:22Þ
 α BM DNT ðT  T ref ÞIdΩ:
T n

That is,
 
KM θUnþ1 þ ð1  θÞUn ¼ FM ¼ Fs þ Fb þ FT2M : ð30:23Þ

Then the solution at the current step can be obtained as follows in the explicit
method,
Appendix: Typical Finite Elements (Q4 and Q9) 405

ð1  θÞKM n Fs þ Fb þ FT2M

Unþ1 ¼ MM Un þ PM ¼ U þ : ð30:24Þ
θKM θKM
The detailed matrices in the above equation for the calculation of the matrices of
stiffness and
2 force vectors
3 are listed below:
∂
6 ∂x 0 7
6 7
6 ∂7
AM ¼ 6 6 0 7 (for vector dependent variables in 2D, independent of the
6 ∂y 7
7
4∂ ∂5
∂y ∂x
element type),
 
N1 0 N2 0 N 0
NM ¼ ... 9 ,
0 N1 0 N2 0 N9 218

2 3
∂N 1 ∂N 2 ∂N 9
∂x ∂y ∂x
6 7
BM ¼ 6 7
∂N 1 ∂N 2 ∂N 9
4 ∂y ∂y ... ∂y 5 :
∂N 1 ∂N 1 ∂N 2 ∂N 2 ∂N 9 ∂N 9
∂y ∂x ∂y ∂y ∂y ∂x 318

By solving the above coupled equation system, the numerical approximations to

the unknowns, i.e., displacement and temperature, can be obtained for a typical
thermomechanical problem. The stress (effective stress) will need to be calculated
based on the unknowns with the following equation:

σ ¼ D∇u  αDðT  T ref ÞI

ð30:25Þ
¼ D33 BM318 U181  αD33 NT14 ðT  T ref Þ41 I33 ,

where σ is the Cauchy stress tensor. But σ here it is represented by a vector due to the
use of the Voigt notation.

Appendix: Typical Finite Elements (Q4 and Q9)

Q4
1 1
N 1 ¼ ð1  ξÞð1  ηÞ; N 2 ¼ ð1 þ ξÞð1  ηÞ;
4 4
1 1
N 3 ¼ ð1 þ ξÞð1 þ ηÞ; N 4 ¼ ð1  ξÞð1 þ ηÞ:
4 4
1 1
N 1, x ¼ ½ð1  ηÞ; N 2, x ¼ ð1  ηÞ;
2a 2a
1 1
N 3, x ¼ ð1 þ ηÞ; N 4, x ¼ ½ð1 þ ηÞ;
2a 2a
406 30 Finite Element Method

1 1
N 1, y ¼ ½ð1  ξÞ; N 2, y ¼ ½ð1 þ ξÞ;
2a 2a
1 1
N 3, y ¼ ð1 þ ξÞ; N 4, y ¼ ð1  ξÞ:
2a 2a

Q9

1 1
N 1 ¼ ξð1  ξÞð1  ηÞη; N 2 ¼ ξð1 þ ξÞð1  ηÞη;
4 4
1 1
N 3 ¼ ξð1 þ ξÞð1 þ ηÞη; N 4 ¼ ξð1  ξÞð1 þ ηÞη;
4 4
1 1
N 5 ¼  ð1  ξÞð1 þ ξÞð1  ηÞη; N 6 ¼ ξð1 þ ξÞð1  ηÞð1 þ ηÞ
2 2
1 1
N 7 ¼ ð1  ξÞð1 þ ξÞð1 þ ηÞη; N 8 ¼  ξð1  ξÞð1  ηÞð1 þ ηÞ
2 2
N 9 ¼ ð1  ξÞð1 þ ξÞð1  ηÞð1 þ ηÞ:
1 1
N 1, x ¼ ð1  2ξÞð1  ηÞη; N 2, x ¼ ð1 þ 2ξÞð1  ηÞη
2a 2a
1 1
N 3, x ¼ ð1 þ 2ξÞð1 þ ηÞη; N 4, x ¼ ð1  2ξÞð1 þ ηÞη
2a 2a
2 1
N 5, x ¼ ξð1  ηÞη; N 6, x ¼ ð1 þ 2ξÞð1  ηÞð1 þ ηÞ
a a
2 1
N 7, x ¼  ξð1 þ ηÞη; N 8, x ¼  ð1  2ξÞð1  ηÞð1 þ ηÞ;
a a
4
N 9 ¼  ξð1  ηÞð1 þ ηÞ:
a
1 1
N 1, y ¼ ξð1  ξÞð1  2ηÞ; N 2, y ¼ ξð1 þ ξÞð1  2ηÞ;
2b 2b
1 1
N 3, y ¼ ξð1 þ ξÞð1 þ 2ηÞ; N 4, y ¼ ξð1  ξÞð1 þ 2ηÞ;
2b 2b
1 2
N 5, y ¼  ð1  ξÞð1 þ ξÞð1  2ηÞ; N 6, y ¼  ξð1 þ ξÞη;
b b
1 2
N 7, y ¼ ð1  ξÞð1 þ ξÞð1 þ 2ηÞ; N 8, y ¼ ξð1  ξÞη;
b b
4
N 9 ¼  ð1  ξÞð1 þ ξÞη:
a
where ξ and η are the local coordinates, while a and b are the sizes of the elements.
Practice Problem 407

Practice Problem

Problem Description

Develop FEM code to solve the problem described in Chap. 28. The material
constituting the square computational domain (1 m by 1 m) is water. The initial
temperature is 273.15 K. The top boundary is set to 293.15 K, while the other
boundaries are thermally insulated. Please calculate the temperature distribution at
the end of 1 day.

Results

Figure 30.2 illustrates the temperature distribution.

Appendix: MATLAB Script File

The following MATLAB code for the FEM implementation of the thermal field
problem was developed based on the MXFEM developed by Matthew Jon Pais
(Pais 2011). A Q4 element was adopted. The quadratic Gauss integration was
employed.

Temperature field solved using FEM

286

284

282

280

278

276

274

272
10
8 10
6 8
4 6
4
2 2
0 0

Fig. 30.2 Temperature distribution from the numerical simulation

408 30 Finite Element Method

main.m

% Main Program, Input, Calculation, and Plot

clear all; close all; clc; format compact; global DOMAIN PARA MAT SOL
IBC

%% Input
% PARA(1,2,3,4) = density(1), heat capacity(2), thermal conductivity
(3),
% coefficient of convective heat transfer(4);
% SOL(1,2) = time step(1), time range of calculation(2);
% IBC = intial temperature(1),ambient temperature in K(2).

PARA = [2700, 920, 0.75, 23];

SOL = [60,3600*24];
IBC = [273.15,293.15];

Height = 1; % Domain Height

Width = 1; % Domain Width
LElem = 1/9; % Element Length

DOMAIN = [Width/LElem,Height/LElem,LElem,LElem];
MAT = [15E9,0.3,20E6,0.3,2,1,Inf,30e-6];

connectivity;

%% Initialization
nXElem = DOMAIN(1); % Number of elements in the x-direction
nYElem = DOMAIN(2); % Number of elements in the y-direction
globalDOF_T=(nXElem+1)*(nYElem+1);
T0 = IBC(1); % Initial temperature
Tenv = IBC(2); % Environment temperature
dt = SOL(1);
tTotal= SOL(2);

T = sparse(1:globalDOF_T,ones(globalDOF_T,1),T0*ones
(globalDOF_T,1),globalDOF_T,tTotal/dt+1); % Create the
solution matrix, T0 is assigned

tstep = 1;
[globalKa_T,globalK_T]=stiffnessMatrix_T;

%% Iteration
for t = dt:dt:tTotal
tstep=tstep+1;
globalF_T=forceVector_T(Tenv,T(:,tstep-1));
T(:,tstep)=globalK_T*T(:,tstep-1)+globalKa_T\globalF_T;
end
Practice Problem 409

%% Plot
figure(1)
T_final=reshape(T(:,end),(nXElem+1),(nYElem+1));
surf(T_final)
title('Temperature field solved using FEM')

connectivity.m

% This function defines connectivity information about the problem.

This includes defining
% node numbers for the problem, defining the XYZ coordinates of the node
numbers, creating
% a connectivity matrix, and % creating matrices storing the X and Y
coordinates for each element.

function connectivity
global CONNEC DOMAIN NODES XYZ

nXElem = DOMAIN(1); % Number of elements in the x-direction

nYElem = DOMAIN(2); % Number of elements in the y-direction
lXElem = DOMAIN(3); % Elemental length in the x-direction
lYElem = DOMAIN(4); % Elemental length in the y-direction
nNode = (nXElem+1)*(nYElem+1); % The number of nodes in the domain
NODES = zeros(nNode,1); % Initialize the matrix storing nodal
information
XYZ = zeros(nNode,3); % Initialize the matrix storing nodal
position

% Create global node numbering (NN), global xyz coordinates (XYZ), and
index of enriched nodes (NODES)
% XYZ = [NodeNumber,X-Coordinate,Y-Coordinate,Z-Coordinate;...]
% NODES = [NodeNumber]
nNode = 1;
NN = zeros(nXElem+1,nYElem+1);
for iYNode = 1:(nYElem+1)
for iXNode = 1:(nXElem+1)
NN(iXNode,iYNode) = nNode;
XYZ(nNode,:) = [nNode (iXNode-1)*lXElem (iYNode-1)*lYElem];
NODES(nNode,1) = nNode;
nNode = nNode+1;
end
end

% Create a global connectivity matrix (CONNEC) and elemental

coordinate matricies (Ex, Ey)
% CONNEC = [ElementNumber,LocalNode1,LocalNode2,LocalNode3,
LocalNode4,InclusionElementNum]
nElem = 1;
CONNEC = zeros(nXElem*nYElem,5);
410 30 Finite Element Method

for iYElem = 1:nYElem

for iXElem = 1:nXElem
N1 = NN(iXElem,iYElem);
N2 = NN(iXElem+1,iYElem);
N3 = NN(iXElem+1,iYElem+1);
N4 = NN(iXElem,iYElem+1);
CONNEC(nElem,1:5) = [nElem N1 N2 N3 N4];
nElem = nElem+1;
end
end

forcevector.m

% This function creates the global force vector from the user-supplied
input file.
% Tenv is the environment temperature, T is the solution at the
previous step

function [globalF]=forceVector_T(Tenv,T)
global DOMAIN MAT FORCE CONNEC PARA

nXElem = DOMAIN(1); % Number of elements in the x-direction

nYElem = DOMAIN(2); % Number of elements in the y-direction
le = DOMAIN(3); % Length of a side of an element
globalDOF_T=(nXElem+1)*(nYElem+1);
globalF = sparse(globalDOF_T,1); % Create the global force vector

if isempty(FORCE) == 1, m = 1; else m = size(FORCE,1); end

if MAT(5) == 1, h = MAT(6); else h = 1; end

% Create the global force matrix from distributed loading

node = CONNEC((nYElem-1)*(nXElem)+1,5); % Initial node along top

edge
for i = 1:nXElem+1
Force=PARA(4)*(Tenv-T(node));
if (i == 1) || (i == nXElem+1) % Convert distributed force to nodal
force
globalF(node) = le*Force/2; % Elemental force in x-direction
for edge nodes
else
globalF(node) = le*Force; % Elemental force in x-direction
end
node = node+1; % Update node along top edge
end
end
Practice Problem 411

stiffnessMatrix_T.m

% This function calculates the global stiffness matrix for the desired
% discontinuities defined by the user-supplied input.

function [globalKa,globalK] = stiffnessMatrix_T

global CONNEC DOMAIN NODES XYZ PARA SOL %MAT

nXElem = DOMAIN(1); % Number of elements in the x-direction

nYElem = DOMAIN(2); % Number of elements in the y-direction
lXElem = DOMAIN(3); % Length of elements in the x-direction
rho = PARA(1);
Cp = PARA(2);
Lambda= PARA(3);
h = PARA(4);
dt = SOL(1);
tTotal= SOL(2);
globalDOF_T=(nXElem+1)*(nYElem+1);
globalK = sparse(globalDOF_T,globalDOF_T);

% Initialize the FE stiffness matrix

globalKa = sparse(globalDOF_T,globalDOF_T); % Define the global K of
the accumulation term
globalKc = sparse(globalDOF_T,globalDOF_T); % Define the global K of
the conduction term
globalKb = sparse(globalDOF_T,globalDOF_T); % Define the global K of
the boundary term

% Initialize the variables which will create the traditional sparse

matrix
nIndexT = 0; % Initialize index
uenrElem = nXElem*nYElem; % Number of elements
allRowT = ones(uenrElem*16,1); % Row indices
allColT = ones(uenrElem*16,1); % Column indices
allValTa = zeros(uenrElem*16,1); % Stiffness matrix values
allValTc = zeros(uenrElem*16,1);
%allValTb = zeros(uenrElem*16,1);

for iElem = 1:(nYElem*nXElem)

N1 = CONNEC(iElem,2); % Node 1 for current element
N2 = CONNEC(iElem,3); % Node 2 for current element
N3 = CONNEC(iElem,4); % Node 3 for current element
N4 = CONNEC(iElem,5); % Node 4 for current element
NN = NODES([N1 N2 N3 N4]',:); % Nodal data for current element

X1 = XYZ(N1,2); X2 = XYZ(N2,2); X3 = XYZ(N3,2); X4 = XYZ(N4,2); %

Nodal x-coordinates
412 30 Finite Element Method

Y1 = XYZ(N1,3); Y2 = XYZ(N2,3); Y3 = XYZ(N3,3); Y4 = XYZ(N4,3); %

Nodal y-coordinates
xyz = [X1 Y1;X2 Y2;X3 Y3;X4 Y4]; % Nodal coordinate
matrix

localKa = 0; % Initialize stiffness for current element

localKc = 0; % Initialize stiffness for current element
%localKb = 0; % Initialize stiffness for current element
local = [N1 N2 N3 N4]; % Traditional index locations

[gp,gw,J] = subDomain_T(3,xyz);
xi = gp(:,1);
eta = gp(:,2);

for i = 1:length(gp)
xi = gp(i,1); eta = gp(i,2); % Gauss points
W = gw(i); % Gauss weights

Ji = [J(i,1) J(i,2);J(i,3) J(i,4)]; % Jacobian of subdomain

detJ = det(Ji); % Determinant of the Jacobian

N = 1/4*[(1-xi).*(1-eta) (1+xi).*(1-eta) (1+xi).*(1+eta)

(1-xi).*(1+eta)];
% Derivative of shape functions with respect to x
Nx = 2/lXElem*1/4*[-(1-eta);1-eta;1+eta;-(1+eta)];
% Derivative of shape functions with respect to y
Ny = 2/lXElem*1/4*[-(1-xi);-(1+xi);1+xi;1-xi];

Bu = [Nx(1) Nx(2) Nx(3) Nx(4);...

Ny(1) Ny(2) Ny(3) Ny(4)];

localKa = localKa + W*N'*(rho*Cp/dt)*N*detJ;

localKc = localKc + W*Bu'*(-Lambda)*Bu*detJ;
%localKb = localKb + zeros(8,8); % W*N'*(-h)*N*detJ; localKb
is not used here, but considered in forceVector.m instead

end

for row = 1:4

for col = 1:4
nIndexT = nIndexT+1;
allRowT(nIndexT) = local(row);
allColT(nIndexT) = local(col);
allValTa(nIndexT) = localKa(row,col);
allValTc(nIndexT) = localKc(row,col);
%allValTb(nIndexT) = localKb(row,col);
end
end
end
Practice Problem 413

globalKa = sparse(allRowT,allColT,allValTa,globalDOF_T,
globalDOF_T);
globalKc = sparse(allRowT,allColT,allValTc,globalDOF_T,
globalDOF_T);
%globalKb = sparse(allRowT,allColT,allValTb,globalDOF_T,
globalDOF_T);
%globalK = eye(globalDOF_T,globalDOF_T)+globalKa\(globalKc
+globalKb);
globalK = globalK+sparse(eye(globalDOF_T,globalDOF_T))+globalKa
\(globalKc);

subdomain_T.m

% This function subdivides enriched elements and determines the Gauss

points
% and weights to be used in the integration during the assembly of the
stiffness matrix.

function [gp,gw,J] = subDomain(npt,xyz)

corner = [1 2 3 4 1];
node = [-1 -1;1 -1;1 1;-1 1];

% Find the triangles defining the subdomains

warning off MATLAB:delaunayn:DuplicateDataPoints
tri = delaunay(node(:,1),node(:,2)); % ,
{'Qt','Qbb','Qc','Qz','Q3'}

% Loop over subtriangles to get quadrature points and weights

[q,w] = gauss(npt,'QUAD');

pt = 1;
for e = 1:size(tri,1)
coord = node(tri(e,:),:);
xyzl = xyz(tri(e,:),:);
for i = 1:length(w)
xi = q(i,1); eta = q(i,2);
N = [1-xi-eta;xi;eta];
gp(pt,:) = N'*coord;
gw(pt,1) = w(i)/2;
J(pt,:) = [-xyzl(1,1)+xyzl(2,1) -xyzl(1,2)+xyzl(2,2)...
-xyzl(1,1)+xyzl(3,1) -xyzl(1,2)+xyzl(3,2)];
pt = pt+1;
end
end
414 30 Finite Element Method

gauss.m

% This function contains various gauss integration points and weights

for
% quadrilateral and triangular elements.

function [Q,W] = gauss(npt,type)

if strcmp(type,'QUAD') == 1
switch npt
case 2
np(1) = -0.577350269189626;
np(2) = 0.577350269189626;

nw(1) = 1.000000000000000;
nw(2) = 1.000000000000000;
case 3
np(1) = -0.774596669241483;
np(2) = 0.000000000000000;
np(3) = 0.774596669241483;

nw(1) = 0.555555555555556;
nw(2) = 0.888888888888889;
nw(3) = 0.555555555555556;
case 4
np(1) = -0.861136311594053;
np(2) = -0.339981043584856;
np(3) = 0.339981043584856;
np(4) = 0.861136311594053;

nw(1) = 0.347854845137454;
nw(2) = 0.652145154862546;
nw(3) = 0.652145154862546;
nw(4) = 0.347854845137454;
case 5
np(1) = -0.906179845938664;
np(2) = -0.538469310105683;
np(3) = 0.000000000000000;
np(4) = 0.538469310105683;
np(5) = 0.906179845938664;

nw(1) = 0.236926885056189;
nw(2) = 0.478628670499366;
nw(3) = 0.568888888888889;
nw(4) = 0.478628670499366;
nw(5) = 0.236926885056189;
case 6
np(1) = -0.932469514203152;
np(2) = -0.661209386466265;
np(3) = -0.238619186083197;
np(4) = 0.238619186083197;
np(5) = 0.661209386466265;
np(6) = 0.932469514203152;
Practice Problem 415

nw(1) = 0.171324492379170;
nw(2) = 0.360761573048139;
nw(3) = 0.467913934572691;
nw(4) = 0.467913934572691;
nw(5) = 0.360761573048139;
nw(6) = 0.171324492379170;
case 7
np(1) = -0.949107912342759;
np(2) = -0.741531185599394;
np(3) = -0.405845151377397;
np(4) = 0.000000000000000;
np(5) = 0.405845151377397;
np(6) = 0.741531185599394;
np(7) = 0.949109712342759;

nw(1) = 0.129484966168870;
nw(2) = 0.279705391489277;
nw(3) = 0.381830050505119;
nw(4) = 0.417959183673469;
nw(5) = 0.381830050505119;
nw(6) = 0.279705391489277;
nw(7) = 0.129484966168870;
case 8
np(1) = -0.960289856497536;
np(2) = -0.796666477413627;
np(3) = -0.525532409916329;
np(4) = -0.183434642495650;
np(5) = 0.183434642495650;
np(6) = 0.525532409916329;
np(7) = 0.796666477413627;
np(8) = 0.960289856497536;

nw(1) = 0.101228536290376;
nw(2) = 0.222381034453374;
nw(3) = 0.313706645877887;
nw(4) = 0.362683783378362;
nw(5) = 0.362683783378362;
nw(6) = 0.313706645877887;
nw(7) = 0.222381034453374;
nw(8) = 0.101228536290376;
case 9
np(1) = -0.968160239507626;
np(2) = -0.836031107326636;
np(3) = -0.613371432700590;
np(4) = -0.324253423403809;
np(5) = 0.000000000000000;
np(6) = 0.324253423403809;
np(7) = 0.613371432700590;
np(8) = 0.836031107326636;
np(9) = 0.968160239507626;
416 30 Finite Element Method

nw(1) = 0.081274388361574;
nw(2) = 0.180648160694857;
nw(3) = 0.260610696402935;
nw(4) = 0.312347077040003;
nw(5) = 0.330239355001260;
nw(6) = 0.312347077040003;
nw(7) = 0.261610696402935;
nw(8) = 0.180648160694857;
nw(9) = 0.081274388361574;
case 10
np(1) = -0.973906528517172;
np(2) = -0.865063366688985;
np(3) = -0.679409568299024;
np(4) = -0.433395394129247;
np(5) = -0.148874338981631;
np(6) = 0.148874338981631;
np(7) = 0.433395394129247;
np(8) = 0.679409568299024;
np(9) = 0.865063366688985;
np(10) = 0.973906528517172;

nw(1) = 0.066671344308688;
nw(2) = 0.149451349150581;
nw(3) = 0.219086362515982;
nw(4) = 0.269266719309996;
nw(5) = 0.295524224714753;
nw(6) = 0.295524224714753;
nw(7) = 0.269266719309996;
nw(8) = 0.219086362515982;
nw(9) = 0.149451349150581;
nw(10) = 0.066671344308688;
case 12
np(1) = -0.981560634246719;
np(2) = -0.904117256370475;
np(3) = -0.769902674194305;
np(4) = -0.587317954286617;
np(5) = -0.367831498998180;
np(6) = -0.125233408511469;
np(7) = 0.125233408511469;
np(8) = 0.367831498998180;
np(9) = 0.587317954286617;
np(10) = 0.769902674194305;
np(11) = 0.904117256370475;
np(12) = 0.981560634246719;

nw(1) = 0.047175336386512;
nw(2) = 0.106939325995318;
nw(3) = 0.160078328543346;
nw(4) = 0.203167426723066;
nw(5) = 0.233492536538355;
nw(6) = 0.249147045813403;
nw(7) = 0.249147045813403;
nw(8) = 0.233492536538355;
Practice Problem 417

nw(9) = 0.203167426723066;
nw(10) = 0.160078328543346;
nw(11) = 0.106939325995318;
nw(12) = 0.047175336386512;
end

% Adjust quadrature points based on dimension

Q = NaN(npt*npt,2);
W = NaN(npt*npt,1);
n = 1;
for i = 1:npt
for j = 1:npt
Q(n,:) = [np(i) np(j)];
W(n) = nw(i)*nw(j);
n = n+1;
end
end
end
References

Abousleiman Y, Cheng A, Cui L, Detournay E, Roegiers J (1996) Mandel’s problem revisited.

Geotechnique 46(2):187–195
Acar YB, Alshawabkeh AN (1993) Principles of electrokinetic remediation. Environ Sci Technol
27(13):2638–2647
Allen MB III (1988) Basic mechanics of oil reservoir flows. In: Multiphase flow in porous media.
Springer, New York, pp 1–86
Alshawabkeh AN, Yeung AT, Bricka MR (1999) Practical aspects of in-situ electrokinetic extrac-
tion. J Environ Eng 125(1):27–35
Anderson DM, Tice AR, McKim HL (1973) The unfrozen water and the apparent specific heat
capacity of frozen soils. In: Second international conference on Permafrost, Yakutsk, USSR.
North American Contribution, pp 289–295
Balanis CA (1999) Advanced engineering electromagnetics. Wiley, New York
Barrett AH, Cardello AV (2012) Military food engineering and ration technology. DEStech
Publications, Inc., Lancaster
Biot MA (1941) General theory of three-dimensional consolidation. J Appl Phys 12(2):155–164
Biot MA (1956a) Theory of propagation of elastic waves in a fluid-saturated porous solid. I.
Low-frequency range. J Acoust Soc Am 28(2):168–178
Biot MA (1956b) Theory of propagation of elastic waves in a fluid-saturated porous solid. II. Higher
frequency range. J Acoust Soc Am 28(2):179–191
Biot M, Willis D (1957) The elastic coefficients of the theory of consolidation. J Appl Mech
24:594–601
Bjerrum L, Moum J, Eide O (1967) Application of electro-osmosis to a foundation problem in a
Norwegian quick clay. Geotechnique 17(3):214–235
Buchner R, Barthel J, Stauber J (1999) The dielectric relaxation of water between 0 C and
35 C. Chem Phys Lett 306(1):57–63
Campbell GS (1974) A simple method for determining unsaturated conductivity from moisture
retention data. Soil Sci 117(6):311–314
Campbell GS (1985) Soil physics with BASIC: transport models for soil-plant systems, vol 14.
Elsevier, Amsterdam
Carcione JM, Morency C, Santos JE (2010) Computational poroelasticity-a review. Geophysics
75:75A229–75A243
Cary J (1965) Water flux in moist soil: thermal versus suction gradients. Soil Sci 100:168–175
Cary J (1966) Soil moisture transport due to thermal gradients: practical aspects. Soil Sci Soc Am J
30(4):428–433
Casagrande IL (1949) Electro-osmosis in soils. Geotechnique 1(3):159–177

© Springer International Publishing AG, part of Springer Nature 2018 419

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2
420 References

Casagrande L (1983) Stabilization of soils by means of electro-osmosis: state of the art. J Boston
Soc Civil Eng 69(2):255–302
Celia MA, Binning P (1992) A mass conservative numerical solution for two-phase flow in porous
media with application to unsaturated flow. Water Resour Res 28(10):2819–2828
Childs EC, Collis-George N (1950) The permeability of porous materials. In: Proceedings of the
Royal Society of London A: Mathematical, Physical and Engineering Sciences, vol 1066. The
Royal Society, pp 392–405
Cho J, Kim K, Chung K, Hyun S, Baek K (2009) Restoration of saline soil in cultivated land using
electrokinetic process. Sep Sci Technol 44(10):2371–2384
Chung H, Kang B (1999) Lead removal from contaminated marine clay by electrokinetic soil
decontamination. Eng Geol 53(2):139–150
COMSOL COMSOL example library. https://www.comsol.com/models/page/2
Corwin D, Plant R (2005) Applications of apparent soil electrical conductivity in precision
agriculture. Comput Electron Agric 46(1):1–10
Courtney M (2015) Multiphysics brings the real world into simulations. E&T Magazine, vol 10
Coussy (2004) The atrium, Southern gate, Chichester, West Sussex PO19 8SQ. John Wiley & Sons
Ltd, England
Cowin SC (1999) Bone poroelasticity. J Biomech 32(3):217–238
Cross M, Slone A (2005) FENet-multi-physics analysis (MPA) theme: a review of commercial
MPA capability in 2005. In: Proceedings of the FENet Project Review Meeting, St. Julians,
Malta, pp 119–130
Cyclopedia M. https://www.comsol.de/multiphysics
de Boer R (2000) Contemporary progress in porous media theory. Appl Mech Rev 53(12):323–370
De Vries DA (1963) Thermal properties of soils. In: Physics of plant environment. North-Holland
Publishing Co., Amsterdam
Dede EM, Lee J, Nomura T (2014) Multiphysics simulation: electromechanical system
applications and optimization. Springer-Verlag, London https://www.springer.com/us/book/
9781447156390
Deen WM (1998) Analysis of transport phenomena. Oxford University Press, New York
Dennis MR, Glendinning P, Martin PA, Santosa F, Tanner J (2015) The Princeton companion to
applied mathematics. Princeton University Press, Princeton
Derjaguin B, Landau L (1941) Theory of the stability of strongly charged lyophobic sols and of the
adhesion of strongly charged particles in solutions of electrolytes. Acta Physicochim URSS 14
(6):633–662
Detournay E, Cheng A (1993) Fundamental of Poroelasticity, Chapter 5. In: Fairhurst C
(ed) Comprehensive rock engineering, vol 2. Pergamon Press, Oxford
Doolittle JA, Brevik EC (2014) The use of electromagnetic induction techniques in soils studies.
Geoderma 223:33–45
Doutres O, Salissou Y, Atalla N, Panneton R (2010) Evaluation of the acoustic and non-acoustic
properties of sound absorbing materials using a three-microphone impedance tube. Appl Acoust
71(6):506–509
Energy2D web tool. http://energy.concord.org/energy2d/reynolds.html
Engineering X (2009) Simulation of a water heater with FLOW-3D
Esrig M, Gemeinhardt JP (1900) Electrokinetic stabilization of an illitic clay. J Soil Mech Found
Div 92 (SM5, Proc Paper 490)
Fernandes A, Pouget J (2003) Analytical and numerical approaches to piezoelectric bimorph. Int J
Solids Struct 40(17):4331–4352
Ferronato M, Castelletto N, Gambolati G (2010) A fully coupled 3-D mixed finite element model of
Biot consolidation. J Comput Phys 229(12):4813–4830
Fischer HB (1966) Longitudinal dispersion in laboratory and natural streams.
Fisk M (2008) Simulation of induction heating in manufacturing. Thesis, Luleå Tekniska
Universitet
Fluke Fluke Thermal Image. https://www.tequipment.net/fluke/thermal-imagers/
References 421

Franks F (1982) Water and Aqueous Solutions at Subzero Temperatures, Volume 7 in “Water, A
Comprehensive Treatise”, Springer. https://www.springer.com/series/10962
Fredlund DG, Rahardjo H (1993) Soil mechanics for unsaturated soils. Wiley, New York
Fredlund DG, Xing A (1994) Equations for the soil-water characteristic curve. Can Geotech J 31
(4):521–532
Fredlund D, Xing A, Huang S (1994) Predicting the permeability function for unsaturated soils
using the soil-water characteristic curve. Can Geotech J 31(4):533–546
Frenkel J (2005) On the theory of seismic and seismoelectric phenomena in a moist soil. J Eng
Mech 131(9):879–887
Gardner W (1958) Some steady-state solutions of the unsaturated moisture flow equation with
application to evaporation from a water table. Soil Sci 85(4):228–232
Gassmann F (1951) Elastic waves through a packing of spheres. Geophysics 16(4):673–685
Geertsma J (1957) A remark on the analogy between thermoelasticity and the elasticity of saturated
porous media. J Mech Phys Solids 6(1):13–16
Ghia U, Ghia KN, Shin C (1982) High-Re solutions for incompressible flow using the Navier-
Stokes equations and a multigrid method. J Comput Phys 48(3):387–411
Gibbard J (2013) Open Hand Project Website. http://www.openhandproject.org/
Gouy M (1910) Sur la constitution de la charge électrique à la surface d'un électrolyte. Journal de
Physique Théorique et Appliquée 9(1):457–468
Gray DH (1970) Electrochemical hardening of clay soils. Geotechnique 20(1):81–93
Groen D, Zasada SJ, Coveney PV (2014) Survey of multiscale and multiphysics applications and
communities. Comput Sci Eng 16(2):34–43
Hackett RM (2016) Stress measures. In: Hyperelasticity primer. Springer, New York, pp 27–36
Hansson K, Šimůnek J, Mizoguchi M, Lundin L-C, Van Genuchten MT (2004) Water flow and heat
transport in frozen soil. Vadose Zone J 3(2):693–704
Harlan R (1973) Analysis of coupled heat-fluid transport in partially frozen soil. Water Resour Res
9(5):1314–1323
Harlow FH, Nakayama PI (1968) Transport of turbulence energy decay rate. Report LA-3854, Los
Alamos Scientific Laboratory, New Mexico
Heil M, Hazel AL, Boyle J (2008) Solvers for large-displacement fluid-structure interaction
problems: segregated versus monolithic approaches. Comput Mech 43(1):91–101
Hou G, Wang J, Layton A (2012) Numerical methods for fluid-structure interaction-a review.
Commun Comput Phys 12(2):337–377
Huang D, Xu Q, Cheng J, Lu X, Zhang H (2012) Electrokinetic remediation and its
combined technologies for removal of organic pollutants from contaminated soils. Int J
Electrochem Sci 7(5):4528–4544
Hunter RJ (2013) Zeta potential in colloid science: principles and applications, vol 2. Academic
Press, London
Huweg AFS (2013) Modelling of electrokinetic phenomena in soils. University of Southern
Queensland
Irwin N, Botz M, Greenkorn R (1996) Experimental investigation of characteristic length scale in
periodic heterogeneous porous media. Transp Porous Media 25(2):235–246
Jame YW, Norum DI (1980) Heat and mass transfer in a freezing unsaturated porous medium.
Water Resour Res 16(4):811–819
Jasak H (2006) Multi-physics simulations in continuum mechanics. In: Proceedings of 5th inter-
national conference of croatian society of mechanics, Trogir
Jasak H (1996) Error analysis and estimation for the finite volume method with applications to fluid
flows. Ph.D. Thesis, Imperial College
Jeong C-H (2011) Guideline for adopting the local reaction assumption for porous absorbers in
terms of random incidence absorption coefficients. Acta Acustica united with Acustica 97
(5):779–790
422 References

Kanit T, Forest S, Galliet I, Mounoury V, Jeulin D (2003) Determination of the size of the
representative volume element for random composites: statistical and numerical approach. Int
J Solids Struct 40(13):3647–3679
Kapoor R, Nemat-Nasser S (1998) Determination of temperature rise during high strain rate
deformation. Mech Mater 27(1):1–12
Keyes DE, McInnes LC, Woodward C, Gropp W, Myra E, Pernice M, Bell J, Brown J, Clo A,
Connors J (2013) Multiphysics simulations: challenges and opportunities. Int J High Perform
Comput Appl 27(1):4–83
Kim S-O, Kim W-S, Kim K-W (2005) Evaluation of electrokinetic remediation of arsenic-
contaminated soils. Environ Geochem Health 27(5):443–453
Kolmogorov AN (1962) A refinement of previous hypotheses concerning the local structure of
turbulence in a viscous incompressible fluid at high Reynolds number. J Fluid Mech 13
(1):82–85
Koopmans RWR, Miller R (1966) Soil freezing and soil water characteristic curves. Soil Sci Soc
Am J 30(6):680–685
Krishna R, Wesselingh J (1997) The Maxwell-Stefan approach to mass transfer. Chem Eng Sci 52
(6):861–911
Krzhizhanovskaya VV (2007) Sun S simulation of Multiphysics multiscale systems: introduction to
the ICCS’2007 workshop. In: International conference on computational science. Springer,
Berlin/Heidelberg, pp 755–761
Laboratories BT (1939) Method and apparatus for heating dielectric materials. Google Patents
Launder BE, Spalding DB (1974) The numerical computation of turbulent flows. Comput Methods
Appl Mech Eng 3(2):269–289
Lautrup B (2005) Physics of continuous matter. Exotic and everyday phenomena in the macro-
scopic world. IOP, Bristol
LeVeque RJ (2002) Finite volume methods for hyperbolic problems, vol 31. Cambridge University
Press, Cambridge, UK
Liu Z, Yu X (2011) Coupled thermo-hydro-mechanical model for porous materials under frost
action: theory and implementation. Acta Geotech 6(2):51–65
Liu Z, Yu X (2013) Physically based equation for phase composition curve of frozen soils. http://
www.liurg.org/documents/Physically%20Based%20Equation%20for%20Phase%20Composi
tion%20Curve%20of%20Frozen%20Soils.pdf
Lo K, Ho K, Inculet I (1991) Field test of electroosmotic strengthening of soft sensitive clay. Can
Geotech J 28(1):74–83
Logan BE (2012) Environmental transport processes. Wiley, New York
Luthin JN, Orhun A, Taylor GS (1975) Coupled saturated-unsaturated transient flow in porous
media: experimental and numeric model. Water Resour Res 11(6):973–978
Mandel J (1953) Consolidation des sols (étude mathématique). Geotechnique 3(7):287–299
Manyuhina O, Shklyarevskiy I, Jonkheijm P, Christianen P, Fasolino A, Katsnelson M,
Schenning A, Meijer E, Henze O, Kilbinger A (2007) Anharmonic magnetic deformation of
self-assembled molecular nanocapsules. Phys Rev Lett 98(14):146101
Marc G, McMillan WG (1985) The virial theorem. In: Prigogine I, Rice SA (eds) Advances in
chemical physics, vol 58. Wiley, Hoboken
Massabó M, Cianci R, Paladino O (2011) An analytical solution of the advection dispersion
equation in a bounded domain and its application to laboratory experiments. J Appl Math
2011:493014
Mattson ED, Bowman RS, Lindgren ER (2002) Electrokinetic ion transport through unsaturated
soil: 1. Theory, model development, and testing. J Contam Hydrol 54(1):99–120
McInnes K (1981) Thermal Conductivities of Soils from Dryland Wheat Regions of Eastern
Washington. MS Thesis, Washington State University
Merxhani A (2016) An introduction to linear poroelasticity. arXiv preprint arXiv:160704274
Mitchell JK, Soga K (2005) Fundamentals of soil behavior. Wiley, New York
References 423

Moeendarbary E, Valon L, Fritzsche M, Harris AR, Moulding DA, Thrasher AJ, Stride E,
Mahadevan L, Charras GT (2013) The cytoplasm of living cells behaves as a poroelastic
material. Nat Mater 12(3):253–261
Moin P, Mahesh K (1998) Direct numerical simulation: a tool in turbulence research. Annu Rev
Fluid Mech 30(1):539–578
Moser RD, Kim J, Mansour NN (1999) Direct numerical simulation of turbulent channel flow up to
Re τ ¼ 590. Phys Fluids 11(4):943–945
Mualem Y (1976) A new model for predicting the hydraulic conductivity of unsaturated porous
media. Water Resour Res 12(3):513–522
Mualem Y (1986) Hydraulic conductivity of unsaturated soils: prediction and formulas. In:
Methods of soil analysis: Part 1-Physical and mineralogical methods. American Society of
Agronomy, Soil Science Society of America, Madison, pp 799–823
Nepf H (1999) Drag, turbulence, and diffusion in flow through emergent vegetation. Water Resour
Res 35(2):479–489
Newman GP, Wilson GW (1997) Heat and mass transfer in unsaturated soils during freezing. Can
Geotech J 34(1):63–70
Noborio K, McInnes K, Heilman J (1996) Two-dimensional model for water, heat, and solute
transport in furrow-irrigated soil: II. Field evaluation. Soil Sci Soc Am J 60(4):1010–1021
Noorishad J, Tsang C, Witherspoon P (1992) Theoretical and field studies of coupled hydrome-
chanical behaviour of fractured rocks—1. Development and verification of a numerical simu-
lator. Int J Rock Mech Mining Sci Geomech Abstracts 4. Elsevier:401–409
Ohm GS (1827) Die galvanische Kette, mathematisch bearbeitet. TH Riemann, Berlin
Overbeek JTG, Kruyt H (1952) Colloid science. In: Kruyt HR (ed) Kinetics of flocculation,
vol 1. Elsevier, Amsterdam, pp 278–300
Pais M (2011), http://www.matthewpais.com/
Penta R, Ambrosi D, Shipley R (2014) Effective governing equations for poroelastic growing
media. Q J Mech Appl Math 67(1):69–91
Perrin J (1904) Mécanisme de l'électrisation de contact et solutions colloidales. J Chim Phys
2:601–651
Perry RH, Chilton CH (1973) Chemical engineers’ handbook. McGraw‐Hill, New York
Peskin CS (1977) Numerical analysis of blood flow in the heart. J Comput Phys 25(3):220–252
Peskin CS (2002) The immersed boundary method. Acta Numerica 11:479–517
Philip J, De Vries D (1957) Moisture movement in porous materials under temperature gradients.
EOS Trans Am Geophys Union 38(2):222–232
Polyanin (1997), http://www.navier-stokes-equations.com/lecturenotes;focus=CMTOI_de_dtag_
hosting_hpcreator_widget_Download_14636605&path=download.action&frame=CMTOI_
de_dtag_hosting_hpcreator_widget_Download_14636605?id=176396
Ponnamperuna FN, Tianco EM, Loy TA (1966) Ionic strengths of soil solutions composition
variation with water content for desaturated soils. Soil Sci Soc Am J 35:436–442
Qi Y, Zhang TC (2015) Sorption and desorption of testosterone at environmentally relevant levels:
effects of aquatic conditions and soil particle size fractions. J Environ Eng 142(1):04015045
Rempel A, Buffett B (1997) Formation and accumulation of gas hydrate in porous media.
J Geophys Res Solid Earth 102(B5):10151–10164
Richards LA (1931) Capillary conduction of liquids through porous mediums. Physics 1
(5):318–333
Righetti PG, Van Oss CJ, Vanderhoff JW (1979) Electrokinetic separation methods. Elsevier/
North-Holland Biomedical Press, Amsterdam
Roberts PJ, Webster DR (2002) Turbulent diffusion. ASCE Press, Reston
Rudnev V, Loveless D, Cook RL (2017) Handbook of induction heating. CRC Press, Boca Raton
Rudnick JW (2001) Linear Poroelasticity
Ruge JW, Stüben K (1987) Algebraic multigrid. Multigrid Methods 3(13):73–130
424 References

Rutqvist J, Börgesson L, Chijimatsu M, Kobayashi A, Jing L, Nguyen T, Noorishad J, Tsang C-F

(2001) Thermohydromechanics of partially saturated geological media: governing equations
and formulation of four finite element models. Int J Rock Mech Min Sci 38(1):105–127
Sagaut P (2006) Large eddy simulation for incompressible flows: an introduction. Springer Science
& Business Media
Sahimi M (2011) Flow and transport in porous media and fractured rock: from classical methods to
modern approaches. Wiley, Hoboken
Sawada S (1977) Temperature dependence of thermal conductivity of frozen soil. Memoirs of the
Kitami Institute of Technology 9(1):111–122
Schanz M, Diebels S (2003) A comparative study of Biot's theory and the linear theory of porous
media for wave propagation problems. Acta Mech 161(3):213–235
Schmitt FG (2007) About Boussinesq's turbulent viscosity hypothesis: historical remarks and a
direct evaluation of its validity. Comptes Rendus Mécanique 335(9–10):617–627
Scott DM, Das DK, Subbaihaannadurai V (2006) A finite element computational method for gas
hydrate. Part I: Theory. Petroleum Sci Technol 24(8):895–909
Sharma R (2012) Viscous incompressible flow simulation using penalty finite element method. In:
EPJ web of conferences. EDP Sciences, p 01085
Spaans EJ, Baker JM (1996) The soil freezing characteristic: its measurement and similarity to the
soil moisture characteristic. Soil Sci Soc Am J 60(1):13–19
Spalart PR (2009) Detached-eddy simulation. Annu Rev Fluid Mech 41:181–202
Stephanson O, Jing L, Tsang C-F (1997) Coupled thermo-hydro-mechanical processes of fractured
media: mathematical and experimental studies, vol 79. Elsevier, Amsterdam https://www.elsevier.
com/books/coupled-thermo-hydromechanical-processes-of-fractured-media/stephanson/978-0-444-
82545-2
Stern O (1924) The theory of the electrolytic double-layer. Z Elektrochem 30(508):1014–1020
Subramaniyan AK, Sun C (2008) Continuum interpretation of virial stress in molecular simulations.
Int J Solids Struct 45(14):4340–4346
Sullivan ME (2008) Remediation of arsenic and persistent organic contaminants using enhanced
in-situ methods. Colorado State University, Fort Collins
Swenson RJ (1983) Comments on virial theorems for bounded systems. Am J Phys 51(10):940–942
Taber LA (2004) Nonlinear theory of elasticity: applications in biomechanics. World Scientific,
Singapore
Tamagnini C, Jommi C, Cattaneo F (2010) A model for coupled electro-hydro-mechanical pro-
cesses in fine grained soils accounting for gas generation and transport. An Acad Bras Cienc 82
(1):169–193
Terzaghi K (1925) Principles of soil mechanics, IV-settlement and consolidation of clay. Eng
News-Record 95(3):874–878
Thilmany J (2010) Multiphysics: all at once. Mech Eng 132(2):39
Thomas H, He Y (1995) Analysis of coupled heat, moisture and air transfer in a deformable
unsaturated soil. Geotechnique 45(4):677–689
Thomas H, He Y (1997) A coupled heat-moisture transfer theory for deformable unsaturated soil
and its algorithmic implementation. Int J Numer Methods Eng 40(18):3421–3441
Thomas HR, King SD (1991) Coupled temperature/capillary potential variations in unsaturated soil.
J Eng Mech 117(11):2475–2491
Thomas HR, Cleall PJ, Li Y, Harris C, Kern-Luetschg M (2009) Modelling of cryogenic processes
in permafrost and seasonally frozen soils. Geotechnique 59(3):173–184
Toro EF (2013) Riemann solvers and numerical methods for fluid dynamics: a practical introduc-
tion. Springer Science & Business Media, Berlin/Heidelberg
Tsai D (1979) The virial theorem and stress calculation in molecular dynamics. J Chem Phys 70
(3):1375–1382
Van Brakel J, Heertjes P (1974) Analysis of diffusion in macroporous media in terms of a porosity,
a tortuosity and a constrictivity factor. Int J Heat Mass Transf 17(9):1093–1103
References 425

van Dalen KN (2013) Multi-component acoustic characterization of porous media. Springer

Science & Business Media
Van Genuchten MT (1980) A closed-form equation for predicting the hydraulic conductivity of
unsaturated soils. Soil Sci Soc Am J 44(5):892–898
Verruijt A (1969) In: De Wiest RJM (ed) Elastic storage of aquifers in flow through porous media.
Academic Press, New York
Verruijt A (2013) Theory and problems of poroelasticity. Delft University of Technology, Delft
Verwey E (1947) Theory of the stability of lyophobic colloids. J Phys Chem 51(3):631–636
Virkutyte J, Sillanpää M, Latostenmaa P (2002) Electrokinetic soil remediation-critical overview.
Sci Total Environ 289(1):97–121
von Smoluchowski M (1903) Contribution to the theory of electro-osmosis and related phenomena.
Bulletin International de l'Académie des Sciences de Cracovie 3:184–199
Wang HF (2000) Theory of linear poroelasticity. Princeton University Press, Princeton
Watanabe K, Flury M (2008) Capillary bundle model of hydraulic conductivity for frozen soil.
Water Resour Res 44(12):1–9
Weinberg S (1977) The search for unity: notes for a history of quantum field theory. Daedalus
2:17–35
Wesseling P, Oosterlee CW (2001) Geometric multigrid with applications to computational fluid
dynamics. J Comput Appl Math 128(1):311–334
Wikipedia Multiphysics. https://en.wikipedia.org/wiki/Multiphysics
Wilcox DC (1998) Turbulence modeling for CFD, vol 2. DCW Industries, La Canada
Yakhot V, Orszag SA (1986) Renormalization group analysis of turbulence. I. Basic theory. J Sci
Comput 1(1):3–51
Yakhot V, Orszag S, Thangam S, Gatski T, Speziale C (1992) Development of turbulence models
for shear flows by a double expansion technique. Phys Fluids A Fluid Dynamics 4
(7):1510–1520
Yuan J, Hicks M (2013) Large deformation elastic electro-osmosis consolidation of clays. Comput
Geotech 54:60–68
Zhao C, Yang C (2012) Advances in electrokinetics and their applications in micro/nano fluidics.
Microfluid Nanofluid 13(2):179–203
Index

A Book organization
Acoustic field backgrounds, 35–37
speed of wave, 169–170 comprehensive introduction, 36
wave equation, 168–169 knowledge accumulation in multiphysics, 36
Acoustic-structure interaction (ASI), numerical methods, 37
325, 326 overview, 35
application suggestions to readers, 37–38
nondestructive impedance testing, 326 Bottom-up approach, 13, 238
physical fields, 326 Bottom-up coupling, 14
ultrasound imaging, 326 Boundary conditions, 21, 24, 25, 27, 28
Navier’s equation, 325 Boundary element method, 94
Neumann boundary condition, 325 Boussinesq approximation, 313
wave equation, 325
Acoustic waves, 159
Advection C
definition, 126 Capillary osmosis, 250
mathematical description, 127–128 Cartesian coordinate system, 215
physical implications, 128–129 Cauchy boundary conditions, 75
porous materials, 128 Cauchy momentum equation, 305
Algebraic equation system, 21, 25, 26 Cauchy stress tensor, 140, 142, 145, 210
Algebraic manipulations, 64–65 Chemical concentration, 174
Algorithmic and architectural perspectives, 16 Chemical field, 173, 174
Ampère model, 354 Chemical kinetics
Ampère’s equation, 282 chemical reaction, 176
Array notation, 47, 61 integration, mass transfer, 178–179
Arrhenius equation, 179 mathematical models, 176
Auxiliary equations, 21, 23, 27 reaction rate equation, 177–178
Auxiliary relationships, 23, 238 Chemical reactions, 175, 176
characteristic reaction rate, 174
parameters, 174
B reactant concentrations, 175
Bernoulli’s equation, 302 reactants/reagents, 174
Biot’s model, 242 types
Biot’s phenomenological model, 237 decomposition, 175
Biot-Willis coefficient, 223, 225 double replacement, 176

© Springer International Publishing AG, part of Springer Nature 2018 427

Z. Liu, Multiphysics in Porous Materials,
https://doi.org/10.1007/978-3-319-93028-2
428 Index

Chemical reactions (cont.) mathematical formulation, 339

single replacement, 175 mechanism, 336–337
synthesis, 175 RF heating, 336
Clapeyron equation, 241 types, 337–339
Claude-Louis Navier and George Gabriel Dielectric materials, 190–192
Stokes, 303 Diffusion
Collocation method, 94 coefficient in porous materials, 132–133
Colloid vibration current, 250 concentrations, 129
Complete fluid-structure interaction (FSI) diffusion coefficient, 131–132
application, 324–325 dynamic oscillation of molecules, 129
mathematical formulation, 322, 323 Fick’s laws, mathematical description,
Comprehensive multiphysics framework, 19 130–131
Computational complexity, 16 turbulent diffusion, 129
Computational electromagnetics, 276, 290 Diffusion equation, 287
Computational fluid dynamics (CFD), 298, 315 Diffusion-advection equation, 126
Computational fluid mechanics, 15 Diffusiophoresis, 250
Computer engineering, 3, 5, 6 Dipole-rotation mechanism, 339
Computer-aided design tools, 5 Direct numerical simulation (DNS), 302
COMSOL, 73, 79 Dirichlet boundary conditions, 74, 326
Concentration field, 123, 126, 135 Discontinuous Galerkin method, 94
Concentrato, 123 Discrete element method, 94
Conservation laws, 95, 96 Discretization
Consolidation theory, 219, 220 mathematical model
Constitutive relationship, 145, 148–151 algebraic equation system, 25
linear elastic Hooke’s law finite difference method and finite
elastic modulus, 148–149 volume method, 26
plane strain, 149–150 governing equations, 22
plane stress, 150–151 meshing in numerical analysis, 26
Voigt notation, 145 numerical computation, 25
Continuity equation, 303 numerical methods, 26
Continuum mechanics spatial finite difference and derivative,
conservation laws, 95, 96 370–373
PDEs, 93 temporal finite difference and schemes,
REV, 94 373–376
Coulomb’s Law, 182–183 Dispersed mass transport, 126–133
Count couplings, 15 advection (see Advection)
Coupled saturated-unsaturated flow definition of concentration, 124
experiment description, 115 diffusion (see Diffusion)
model development and implementation, dispersion, 133–135
116–117 equation, 126
numerical solutions, 119 fluid, 124
Couplings, 12–16 processes, 124
Crank-Nicolson stencil, 376 Dispersion, 133–135
Dissipation, 209, 211
DLVO theory, 254
D Double layer theory, 249, 251–254
Darcy’s law, 107–109, 255 Double replacement reaction, 176
Decomposition reaction, 175 Drift-diffusion equation, 260
Detached eddy simulation (DES), 303 2D Structured grids, 390–391
Dielectric heating Dynamic poroelasticity, 319
application, 339–340 constitutive relationships, 227
energy flux and intensity, 340 elastic waves, 321
imaginary component, 337, 338 elastodynamics, 319
Index 429

empirical models, 322 mathematical formulation

governing equation, 227–228 governing equation, 262–263
governing equations and constitutive typical auxiliary relationships, 263–266
relationships, 321 measured electric conductivity, 267, 268
loss models, 321 measured water content, 268
momentum balance equations, 321 micro- and nano-fluidic lab-on-a-chip
Navier-Stokes equations, 321 systems, 251
poromechanics, 319 miniature compactor, 266
saturating fluid, 321 phases/ionogenic groups, 249
wave propagation equation, 321 porous materials, 249, 250
Dynamics sedimentation potential, 250
description, 157 simulated electric conductivity, 268, 269
elastodynamics, 158 simulated water content, 267, 269
fluids, 158 soil remediation, 251
motion of entities, 157 soil water characteristic curve, 267
wave propagation theory, 158 streaming potential, 250
tortuosity, 267
total concentration, 270, 271
E water content sensors, 266
ECH20 Utility, 266 Electromagnetic radiation, 285
Einstein relationship, 259 Electromagnetics, 32, 275
Einstein-Smoluchowski relation, 259 atomic nuclei, 275
Elastic moduli, 148, 151 boundary conditions, 287
Elastic waves, 160 computation domain and boundary
Elastodynamics, 33 conditions, 291
acoustics in solids, 160–161 description, 275
governing equation, 160 distribution of potential, 293
Navier’s equation, 168 2D numerical simulation, 291
wave equations, 159, 160 electrostatics and magnetostatics, 275
Electric displacement field, 190–192 field vector method, 292, 294
Electric field, 183–185 finite element method, 292
anions, 260 governing equation, 285–287
electrokinetic phenomena, 249 H field, 292
electromigration, 258 induction techniques, 276
electroosmotic flow, 254 Lorentz force, 275
electrostatic forces, 260 Maxwell’s equations (see Maxwell’s
external mechanical disturbance, 250 equations)
oscillating, 250 microfluidics and bio-environments, 276
porous materials, 258 numerical simulation, 276
Electric potential, 186–188 phasor formulation, 288–289
Electric sonic amplitude, 250 porous materials, 276
Electrokinetic phenomena, 32 potential method, 292, 293
Electrokinetics, 262–266 vector and potential formulations, 291
chemical potentials, 249 visualization, 284
cleaning agent, 251 wave equations, 285
degree of saturation, 270, 271 waveguides, 289–290
double layer theory, 249, 251–258 Electromagnetics and elastodynamics, 4
2D simplification, 267 Electromagnetic waves, 284–285
electric field, 249 Electromagnetism, 275
electrodes, 266 Electromagnetomechanics, 34
electromigration, 258–260 See also Electromechanics;
electrophoresis, 250, 260–261 Magnetomechanics
experiment with soils, 262 Electromechanics, 34, 357–360
ionic transport mechanism, 258, 260, piezoelectric effect
261, 263 application areas, 359
430 Index

Electromechanics (cont.) 2D and 3D, 381

mathematical formulation, 357–359 error analysis, 379–381
mechanism, 357 numerical simulation methods, 370
piezoresistive effect Finite element analysis (FEA), 398
application, 360 Finite element method (FEM), 400–405,
mathematical formulation, 359–360 408–417
piezoresistive coefficient, 360 2D elements, 398
Electromigration discretization, 397, 398, 400–405
bulk solution theory, 259 FDM and FVM, 397, 398
diffusivity, 259 Galerkin method and weak form, 398–400
dissolved salt ions, 258 local algebraic equation, 397
electrokinetic process, 258 MATLAB code, 398
electroosmotic flow, 258, 260 connectivity.m, 409, 410
ionic mobility, 258, 259 forcevector.m, 410
mathematical formulation, 258 gauss.m, 414–417
negatively charged ions, 260 main.m, 408, 409
positively charged ions, 259 stiffnessMatrix_T.m, 411–413
Electronic heating, 336 subdomain_T.m, 413
Electroosmosis, 249 Neumann boundary conditions, 398
analytical expression, 257 polynomials, 397
Coulombic forces, 254 Q4, 405–406
dispersive effects, 257 Q9, 406
drift velocity, 255 temperature distribution, 407
electrochemical phenomena, 258 thermomechanics
electrokinetic remediation, 258 mechanical field, 403–405
electromagnetic flow, 255 thermal field, 400–403
hydraulic conductivity, soils, 255 Finite volume method (FVM)
linear and nonlinear relationship, 257 boundary and initial conditions, 387–389
parabolic velocity profile, 257 computational fluid dynamics, 385
tortuosity, 256, 257 discretization, 385, 387
water movement, 254 2D structured grids, 390–391
zeta potential values, 256 integral form of governing equation, 386
Electroosmotic flow, 254 unstructured grids, 391–393
Electrophoresis, 260–261 FlexPDE script file, 172, 218, 246–248,
Electrophoretic retardation force, 260 271–273, 316–317, 348–349
Electrostatic energy, 188–190 FlexPDE tool, 84–88
Electrostatics Fluid dynamics, 33, 163–168, 299–306, 385
Coulomb’s Law, 182–183 acoustics, 159
description, 181 aerodynamics for air, 297
electric displacement field and dielectric classification, 300
materials, 190–192 comprehensive derivation
electric field, 181, 183–185 conservation of mass, 166–168
electric potential, 182, 186–188 conservation of momentum, 163–166
Gauss’s Law, 185–186 computational domain and boundary
governing equation, 186–188 conditions, 315
Explicit method stencil, 375 FlexPDE, 315
flow types
compressible and incompressible
F flow, 299–300
Fick’s laws, 129, 130 inviscid, Newtonian and
Field vector formulation, 294, 295 non-Newtonian, 300–302
Finite difference method (FDM) laminar and turbulent flow, 302–303
boundary and initial conditions, 376–379 fluid types and constitutive relationships, 301
Index 431

hydrodynamics for liquids, 297 High-frequency heating, 336

kinematic viscosity, 315 Holistic multiphysics framework, 19
liquid container, 314 Hooke’s law, 145–151
liquid particles, 298 constitutive relationship
magnetohydrodynamics and elastic modulus, 148–149
hydrodynamic stability, 297 plane strain, 149–150
multiphysics, 297 plane stress, 150–151
Navier-Stokes equation, 160 Voigt notation, 145–148
N-S equations, 297 (see also Navier- Hydraulic field, 239–241
Stokes (N-S) equations) Hydrodynamics, 297
thermomechanics, 298 Hydrodynomechanics, 33, 319
turbulent simulation techniques, 298 application areas
visualization, turbulence, 302 ASI, 319
wave equations, 159 FSI, 319
x-velocity, 315 dynamic poroelasticity, 319
y-velocity, 316 FSI, 319, 320
Fluid-structure interaction (FSI), 33, 319, 320 region coupling, 320
Fluid-structure system, 15 types of hydrodynomechanics, 320
Fredlund and Xing’s equation, 240 Hydrology, 112
Fredlund’s method, 242 Hydromechanics, 32, 238
Frost heave process, 238 See also Poroelasticity
Full notation, 61 Hyperthermic oncology, 340
FVM code, 393

I
G Immersed boundary method, 323
Galerkin/variational methods, 397–400 Implementation of multiphysics, 21
Gauss-Legendre numerical integration, 401 Implicit method stencil, 376
Gauss’s divergence theorem, 152 Index notation, 148
Gauss’s law, 185–186, 277, 278 algebraic manipulations, 64–65
General boundary condition, 242–243 calculus, 63–64
Geotechnical engineering, 30, 32 rules, 61–62
Gilbert model, 355 tensor algebra, 62–63
Governing equations, 21–23, 27, 238, 241 Induction heating
Grids, 41, 43 applications, 344–345
Groundwater flow equation, 106 cooktop, 344
Groundwater hydrology, 105 eddy currents, 340
Groundwater movement energy dissipation, 340
Darcy’s Law, 108–109 governing equation, 341–342
derivation of governing equation, 109–111 non-contact heating process, 341
mass balance, 107 poynting vector, 343–344
skin depth, 342–343
time-varying (alternating) current, 340
H workpiece, 340
Heat capacity, 98–99 Industrial collaboration, 3
Heat transfer, 98–100 Integral form of governing equation, 386–387
boundary conditions, 100, 101 Interdisciplinary research, 3
continua, 96–97 Iterative methods, 27
continuum mechanics, 93–96
porous materials
heat capacity, 98–99 J
thermal conductivity, 99 Joule heating
thermal diffusivity, 99–100 applications and adverse effects, 335–336
Helmholtz-Smoluchowski theory, 255 flow-3D, 335
Higher-order tensors, 45, 50, 52, 53, 55 macroscopic scale, 333
432 Index

Joule heating (cont.) conventional, 277–278

mathematical description, 334–335 displacement current, 280
resistive conductor, 333 electric and magnetic fields, 275, 279
thermal image, 333, 334 electric polarization and magnification, 279
electromagnetic and mechanical fields, 277
governing equations, 277
K imaginary object, 280
Kinematics, 157 Lorentz force law, 275
Kohlrausch’s law, 265 numerical simulations
Kolmogorov microscale, 298 boundary conditions, 282–284
conductive materials, 282
independent governing equations, 281
L permittivity and permeability, 280
Lack of multiphysics book, 35, 36 strains, 280
Large eddy simulation (LES), 303 stresses, 280
Lorentz force, 275 waveguides, 277
Mechanical and hydraulic fields, 30
Mechanical and hydrologic process, 4
M Mechanical field, 241–242
Magnetic dipole moment, 354 coupling, 209
Magnetic field, 197–199 governing equation, 152–154
electric potential energy density, 202 Megnetostatic force, 354, 355
electromagnetics, 195 current and magnet, 352
energy, 202 force between magnets, 353
Gauss’s law, 200 magnetic dipole moment, 354
Lorentz force law, 197 models
magnetic vector potential, 201 Ampère model, 354
moving electrons and spin magnetic Gilbert model, 355
moments of elementary particles, 195 secondary effect, 355–356
nonmagnetic materials, 199 Microwave heating, 336
torque on magnetic dipole, 198 Mixed saturated-unsaturated flow, 114
types, 195, 198 Mixtures, 125
Magnetic vector potential, 201 Model-to-model coupling, 15
Magnetohydrodynamics, 12 Molar concentration, 125
Magnetomechanics, 34, 351–356 Molecular dynamics, 12, 94
megnetostatic force (see Megnetostatic Monolithic physics, 9, 13–16
force) Moving mesh algorithm, 323
steady-state force, 351 Multimodel definition, 11
Magnetostatic energy, 202 Multiphase and multicomponent flow, 30
Magnetostatics, 197 Multiphysics, 212
Gauss’s Law, 200–201 challenges, 16
governing equation and auxiliary equations, computer engineering and science, 6
201–202 couplings, 13–16
magnetic dipole moment, 196 definition, 10–11
magnetic field (see Magnetic field) example, 4, 7
Mass balance, 107 history, 4
Mathematical analysis, 16 implementation, 21
Mathematical description, 21, 24 industrial applications, 6
Mathematical model, 22–25 industrial design and product testing, 5
MATLAB PDE Toolbox, 79–86 literature, 35
Matrix notation, 54 simulation tools, 5
Maxwell’s equations, 281–284 storage of matrices, 44
analogy, 280, 281 Multiphysics applications in porous materials
components, 279 elastodynamics, 33
computational electromagnetics, 276 electrokinetic phenomena, 32
Index 433

electromagnetics, 32 boundary conditions

fluid dynamics, 33 Cauchy, 75
hydrodynomechanics, 33 confusion, 75–76
hydromechanics, 32 3D domains, 76–77
thermoelectricity, 33 Dirichlet, 74
thermoelectromagnetics, 33 mixed, 74
thermohydromechanics, 32 Neumann, 74
thermomechanics, 31 Robin, 74
thermomechanometry, 31 treatments in computer programs, 77
Multiscale, 12 classification, 69–71
engineering applications
linear equations, 71–72
N nonlinear equations, 72
Navier’s equation, 152–154, 241 system, 72–73
Navier-Cauchy equation, 153 initial condition, 78
Navier-Stokes (N-S) equations, 306–308 numerical simulation, 67–69
assumptions, 304 Particle transport, 133
conservation of momentum, 305–306 Petroleum engineering, 30, 32
continuity equation, 303 Phasor formulation, 288–289
fluid dynamics, 298 Physical fields, 16–20
governing equations, 297, 304 associated presentation of state variables, 43
inviscid flow, 309–310 computational domain, 43
mass conservation, 304 data structure and computer storage, 43
Newton’s second law, 303 definition, 43, 44
Newtonian fluid dimension of arrays, 44
compressible, 307 grid points, 43
incompressible, 307–308 materialization and storage, 41
viscous flow, 306–307 matrices, 43
non-Newtonian fluid, 308–309 spatial coordinates, 43
Navier-Stokes equations, 159, 160, 322 storage of matrices, 44
Neumann boundary condition, 74, 76, Physical phenomena, 10
323, 325, 387 Physical properties, 10
Newtonian gravitational interactions, 12 Piezoelectric effect, 356–359
Non-isothermal consolidation theory, 237 Piezoresistive effect, 359
Numerical analysis, 21, 25, 26, 369 application, 360
Numerical computation, 25 mathematical formulation, 359–360
Numerical methods, 26 piezoresistive coefficient, 360
Numerical models, 26, 27 Poisson-Boltzmann equation, 252, 254
Numerical simulations, 24, 26, 67–69 Polarization and magnification, 277, 279
Numerical solution, 27 Pore water movement, 108
error estimation, 28 Poroelasticity, 224–227
postprocessing, 27 dynamic (see Dynamic poroelasticity)
true solution, 27 linear, 220
mixture theory notation, 221
phase composition, 221
O static (see Static poroelasticity)
Ohm’s law, 263, 334 strain and displacement, 223–224
Ordinary differential equations (ODE), 69 stress, 222–223
Poroelasticity theory, 238
Porosity, 126, 128, 132, 133, 135
P Porous materials, 97–100
Partial differential equations (PDEs), 12, challenging problems, 30
36, 41, 42, 71–77 couplings, 31
434 Index

Porous materials (cont.) FlexPDE tool, 84–89

heat transfer (see Heat transfer) MATLAB PDE Toolbox
multiphysical phenomena, 29 boundary conditions, setup, 83
multiphysical processes, 31 boundary mode, 82
multiphysics, 30 development of geometry, 82
physical phenomena to coupled governing equation, setup, 83
multiphysical processes, 30 meshing, 84
physical processes, 30 postprocessing, 85
soils, 29 setup of canvas, 81
thermo-hydromechanical process, 30 visualization of results, 86
Postprocessing, 21, 27 window, 80
Potential formulation, 295, 296 Spatial discretization, 371–373
Poynting vector, 344 Specific moisture capacity, 113
Poynting’s theorem, 343 Speed of wave, 169–170
Static poroelasticity
constitutive relationships, 224, 225
R governing equation, 226–227
Radiation hydrodynamics, 12 Stoichiometric coefficients, 176–178
Reactant concentrations, 175 Strain, 142–145
Reaction rate coefficient Stress, 140–142
pressure dependence, 179 Stress-strain relation, 65
temperature dependence, 179 Surface flow, 114
Region coupling, 320 Symbolic computation, 25
Representative elementary volume (REV), Synthesis reaction, 175
93–95
Resistive/Ohmic heating, 333
Reynolds stress tensor, 313 T
Reynolds-averaged Navier-Stokes (RANS), 303 Tensor analysis, 46–50
RF heating, 336 contraction and inner product, 56–57
Richards equation, 106, 111, 112, 114, 116 definition, 53–54
Ritz method, 398 notation, 45–46
Robin boundary condition, 74 tensor product, 54–56
Rules of index notation, 61–62 vector algebra (see Vector algebra)
Tensor calculus, see Tensor analysis
Tensor notation, 61
S Tensors
Saturated flow, 105, 106, 111 algebra and calculus, 42
Scalars, 41–42 mathematical languages, 41
Script file, 80, 84, 86, 87 operations and changes, 42
Seismic waves, 160 scalars and vectors, 41
Shape functions, 398–400, 402 second-order, 42
Single replacement reaction, 175 Thermal conductivity, 96, 97, 99
Small strain theory, 140, 143, 144 Thermal diffusivity, 216
Smoluchowski equation, 260 Thermal expansion, 210
Smoothed particle hydrodynamics (SPH), Thermal expansion coefficient
94, 157 area, 215–216
Software engineering, 3, 6 linear, 212
Soil consolidation, 220 temperature changes, 210
Soil erosion, 30, 33 thermomechanical field, 209
Soils, 29, 31, 32 volume, 216
Soil-water characteristic curve (SWCC), 240 Thermal fields, 96, 238–239
Solenoidal field, 161 Thermal strains, 211
Solving PDEs, 80–86 Thermal stresses, 211
Index 435

Thermoelectricity, 33 Transverse magnetic (TM) wave, 289

Thermoelectromagnetics, 33, 333–345 Trends in multiphysics, 3, 5, 6
computation domain and boundary Turbulent diffusion, 129, 131
conditions, 345 Turbulent models
dielectric (see Dielectric heating) Reynolds number, 310–311
electric field, 346 simulation, 311–313
governing equation natural boundary standard k-ε, 314
condition, 346 Type-K thermocouples, 267
induction (see Induction heating) Types of multiphysics, 16–20
joule (see Joule heating) Typical fluid-structure interaction problem,
longitudinal direction, 345 329, 330
magnetic field, 347 computational domain and boundary
mechanisms, 333 conditions, 327
temperature distribution, 347, 348 governing equation, 327–328
Thermohydraulics models, 238 moving mesh, 327–328
Thermohydromechanics (THM), 32 non-slip boundary condition, 327
boundary conditions, 242–243 result
experimental setup, 243 pressure distribution, 330
geomaterials, 237 solid obstacle displacement field, 329
heat transfer process, 237 water velocity field (arrows), 329
hydraulic field, 239–241
measured water content and temperature, 244
mechanical field, 237, 241–242 U
multiphysical phenomena, 237 Ubiquitousness of multiphysics, 3
non-isothermal consolidation theory, 237 Unsaturated flow, 105, 106, 112, 114, 115
numerical simulation, 244, 245 Unstructured grids, 391–393
poroelasticity, 238 US Patent US2147689A, 339
thermal field, 238–239
thermohydraulics and hydromechanics, 238
unsaturated porous materials, 237 V
Thermohydromechanochemical (THMC) van Genuchten’s equation, 240
model, 237 van Genuchten’s model, 244
Thermomechanics, 31, 212–215 Vector algebra
computational domain and boundary additions and scalar multiplication,
conditions, 217 47–48
coupling operations, 46
mechanical field to thermal field, products of vectors, 48–50
214–215 Vector calculus, 57–59
thermal field to mechanical field, differential calculus and integral calculus, 50
212–214 identities
2D (plane strain) square object, 216 distributivity over vector addition, 57
mechanical processes, 211–212 multiple operators, 57
multiphysics, 209 operator and order of tensor, 57
thermal expansion/stress, 209–211 product of scalar and vector, 58
two-way coupling, 209 scalar triple product, 58
vertical stresses, 217 vector triple product, 58–59
Thermomechanometry, 31 operators, 51
Time-dependent spatial distribution, 11 theorems, 52–53
Time-harmonic form, 288 Vectors, 41–42
Top-down approach, 238 Viscosity, 307
Tortuosity, 126, 133, 135 Viscous dissipation, 211, 212, 215, 298
Transverse electric and magnetic (TEM) Voigt notation, 145–148
wave, 289 Volume concentration, 125
436 Index

W Wave theory, 159

Water flow in unsaturated zones, 112–114 Wave-based sensors, 276
derivation, 111–112 Waveguides, 289–290
formulations
head based, 112–113
saturation based, 113–114 Y
Wave equation, 168–169 Young’s modulus, 243