Larry A Taber - Nonlinear Theory of Elasticity - Applications in Biomechanics (Revised Edition) - WSPC (2023)

This page intentionally left blank
Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
NONLINEAR THEORY OF ELASTICITY

Applications in Biomechanics
Revised Edition
Copyright © 2023 by World Scientific Publishing Co. Pte. Ltd.
All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.
ISBN 978-981-127-051-2 (hardcover)

ISBN 978-981-127-052-9 (ebook for institutions)
ISBN 978-981-127-053-6 (ebook for individuals)
For any available supplementary material, please visit

https://www.worldscientific.com/worldscibooks/10.1142/13254#t=suppl
Printed in Singapore
To the memory of Charles Steele,
eminent teacher, researcher, and mentor
Preface to the Revised Edition
Since the time the first edition of this book was published, the field of
biomechanics has been going through a kind of transformation. For many
researchers, the main focus has shifted from traditional biomechanics to
the rapidly growing discipline of mechanobiology. Whereas biomechanics
typically involves studies of mechanical behavior of living cells and tissues,
mechanobiology deals with the biological response to mechanical stimuli.
Since the 1980s, considerable effort has been devoted to developing new
theories for modeling mechanobiological processes, such as growth and re-
modeling. As models have become increasingly realistic, investigators have
met new challenges. In addition to nonlinear mechanical behavior, modern
models in mechanobiology increasingly include mechanical, chemical, and
genetic feedback.
If you are looking for a book on mechanobiology, this may not be the
book for you, as it addresses only nonlinear biomechanics. (Shameless plug:
my recently published book Continuum Modeling in Mechanobiology may
be helpful.) On the other hand, a firm foundation in nonlinear elasticity
is essential for developing computational models, not only in biomechanics,
but also in mechanobiology of soft tissue.
When the first edition of this book appeared, I wondered whether anyone
would buy it, including my own students. (The same is true for this revised
edition.) Over the years, more copies sold than I expected, and some people
have told me they even read it. Unfortunately, however, it never quite made
the New York Times best-seller list.
Several readers have requested solutions to the problems given at the end
of chapters, and this became my main motivation for writing this revision.
Most of the old problems have been revamped, new problems have been
added, and answers and detailed solutions to most are included. Since the
vii
viii Preface to the Revised Edition
number of graduate courses devoted to nonlinear elasticity has dwindled,

my hope is that this volume would be useful for self-study. In addition,
numerous errors have been corrected (some embarrassing), and some parts
have been revised for clarity, e.g., the sections on polar decomposition and
physical tensor components. (Regrettably, I am sure there still are errors.)
The first edition went to press before I could acknowledge Pat Alford,
who created the beautiful artwork on its cover. (By the way, those draw-
ings represent the developmental processes of gastrulation in the sea urchin
embryo and neurulation in the chick.) I also want to thank Ashok Rama-
subramanian, who provided solutions to many of the problems in the first
edition and alerted me to several errors. Finally, I thank my wife, Char,
once again for her editorial assistance. When I asked for her help with
this book, she was absolutely thrilled to have something more interesting
to read than her mystery novels. Sadly, however, since her policy is to not
do equations, I will not be able to blame her for any math errors.
Larry A. Taber
St. Louis, Missouri
Preface to the First Edition
Problems involving the deformation of soft biological tissues are among

the most challenging in biomechanics. Soft tissues are viscoelastic com-
posite materials composed of cells and extracellular matrix, are frequently
arranged in complex geometries, and routinely experience large strains un-
der normal and pathological conditions. As the discipline of biomechanics
matures, new problems continually appear that demand more and more
sophisticated analytical and numerical solution techniques. For example,
the field of tissue engineering, which deals with in vitro construction of nat-
ural replacement tissues, could benefit greatly from analyses that integrate
nonlinear mechanical behavior at the molecular, cellular, tissue, and whole
organ levels.
Although soft tissues actually are not elastic, investigators often incor-
porate elasticity as a simplifying assumption. In fact, many tissues exhibit
approximately elastic (pseudoelastic) behavior under a fairly wide range of
conditions. Hence, knowledge of the fundamental principles of nonlinear
elasticity is crucial to understanding the biomechanics of soft tissues.
Unfortunately, however, many students trained in biomechanics are ex-
posed only to relatively elementary concepts of nonlinear elasticity during
the course of their studies. Later, these students may need to learn this
daunting topic on their own. Having been in this boat myself at one time,
I know just how difficult this task can be.
One reason this subject can be so difficult to learn is that the various
methods, notations, and definitions used in the classical literature can be
confusing. While this situation has improved in recent years, one goal of
this book is to unify the principal approaches used by researchers in the
field of nonlinear elasticity. Some authors, for example, work in convected
coordinates, while others define separate coordinate systems for the unde-
ix
x Preface to the First Edition
formed and deformed bodies. This book uses and compares both methods.
A hallmark of this book is an extensive use of tensor and dyadic analysis
in general curvilinear coordinates. Some may question the need for such
complexity in this age of high-speed computers, especially in biomechanics,
where most problems of interest require computational methods, in which
case Cartesian coordinates often are sufficient. While this is true, much of
the classical and even the current literature is littered with general curvi-
linear coordinates, and so reading and understanding these papers requires
knowing the language of general tensor analysis.
The emphasis on dyadic notation was motivated by courses I took many
years ago on linear elasticity and shell theory, respectively, from Professors
George Herrmann and Charles R. Steele at Stanford University. These
courses introduced me to the systematic elegance of dyadics. Over the
years, my students and I have found that learning the intricacies of nonlin-
ear elasticity is aided by routine use of dyadics, which form a link between
direct and indicial notation. Direct notation lends insight and clarity to
the basic physical principles, while indicial notation often makes algebraic
manipulations easier (and is needed to solve problems). Judicious use of
dyadics sometimes can clarify subtle differences, as well as similarities, be-
tween the various approaches used in solving nonlinear problems.
Throughout the book, I have tried to present as many details as possible
for each derivation. While this feature may be tedious to some readers, I
know how frustrating it can be when steps are omitted or glossed over. In
some instances, however, derivations are left to exercises at the end of the
chapter. Another possible source of irritation is the way many problems
are solved using more than one approach. The purpose of doing this is to
emphasize that there is more than one way to skin a cat, as well as to bring
out some subtleties that may be missed otherwise.
Readers of this book would benefit by prior experience with linear elas-
ticity. For those readers not already familiar with the linear theory, Ap-
pendix A provides an introduction. This book can be used as a textbook
for a course on the nonlinear theory of elasticity, as a reference for a course
on biomechanics, or as a reference for researchers trying to learn the subject
on their own.
After a brief introduction in Chapter 1, Chapter 2 establishes the math-
ematical background that is needed to navigate the remainder of this book.
This chapter covers general tensor and dyadic analysis, introducing the ba-
sic concepts of changes in coordinates and reference frames. Next, Chapters
3 and 4 develop and compare the basic measures of deformation and stress,
Preface to the First Edition xi
and present the field equations of solid mechanics. Chapter 5 then discusses
constitutive theory and develops stress-strain relations for nonlinear mate-
rials with various material symmetries. Finally, Chapter 6 solves several
specific problems in soft tissue mechanics using the theory and techniques
presented in the previous chapters.
The emphasis of this book is on learning the engineering fundamentals
needed to solve nonlinear problems in biomechanics. Some mathematical
and many biological details are omitted. For more in-depth modern treat-
ments of some of these issues, I highly recommend the books by Gerhard
A. Holzapfel and Jay D. Humphrey that are listed in the bibliography.
I am greatly indebted to Ashok Ramasubramanian, who diligently
worked through most of the in-text derivations and the problems, pointing
out numerous errors. I also would like to thank Millard F. Beatty and one
of my students, Evan Zamir, who provided helpful comments on earlier ver-
sions of this book. Finally, I thank Charlene Taber for her valuable editorial
assistance. Any remaining errors are my own, and I would appreciate any
feedback.
Larry A. Taber
St. Louis, Missouri
Contents
Preface to the Revised Edition vii
Preface to the First Edition ix
1. Introduction 1
1.1 Elasticity and Pseudoelasticity . . . . . . . . . . . . . . . 1
1.2 Biomechanics and Mechanobiology . . . . . . . . . . . . . 3
2. Vectors, Dyadics, and Tensors 5

2.1 Reference Frames and Coordinate Systems . . . . . . . . . 5
2.2 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.1 Base Vectors . . . . . . . . . . . . . . . . . . . . . 7
2.2.2 Reciprocal Base Vectors . . . . . . . . . . . . . . . 13
2.3 Dyadics and Tensors . . . . . . . . . . . . . . . . . . . . . 17
2.3.1 Dyadics . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.2 Tensors . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.3 Matrix Representation of Tensors . . . . . . . . . 21
2.3.4 Some Properties of Tensors . . . . . . . . . . . . . 22
2.4 Coordinate Transformation . . . . . . . . . . . . . . . . . 24
2.4.1 Transformation of Base Vectors . . . . . . . . . . 24
2.4.2 Transformation of Vector Components . . . . . . . 29
2.4.3 Transformation of Tensor Components . . . . . . 29
2.5 Tensor Invariants . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.1 Trace . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5.2 Determinant . . . . . . . . . . . . . . . . . . . . . 33
2.5.3 Eigenvalues and Eigenvectors . . . . . . . . . . . . 33
2.6 Special Tensors . . . . . . . . . . . . . . . . . . . . . . . . 35
xiii
xiv Contents
2.6.1 Metric Tensor . . . . . . . . . . . . . . . . . . . . 35

2.6.2 Permutation Tensor . . . . . . . . . . . . . . . . . 37
2.6.3 Orthogonal Tensors . . . . . . . . . . . . . . . . . 40
2.7 Physical Components . . . . . . . . . . . . . . . . . . . . . 41
2.7.1 Physical Components of Vectors . . . . . . . . . . 42
2.7.2 Physical Components of Tensors . . . . . . . . . . 42
2.8 Vector and Tensor Calculus . . . . . . . . . . . . . . . . . 44
2.8.1 Differentiation with Respect to Spatial Coordinates 44
2.8.2 Differentiation with Respect to Vectors and Tensors 50
2.8.3 Gradient Operator . . . . . . . . . . . . . . . . . . 52
2.8.4 Integral Relations . . . . . . . . . . . . . . . . . . 58
2.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3. Analysis of Deformation 67
3.1 Deformation in One Dimension . . . . . . . . . . . . . . . 68
3.2 Coordinate Systems and Base Vectors . . . . . . . . . . . 69
3.3 Deformation Gradient Tensor . . . . . . . . . . . . . . . . 72
3.4 Deformation and Strain Tensors . . . . . . . . . . . . . . . 74
3.4.1 Deformation Tensors . . . . . . . . . . . . . . . . 74
3.4.2 Strain Tensors . . . . . . . . . . . . . . . . . . . . 75
3.4.3 Physical Components of Strain . . . . . . . . . . . 77
3.5 Strain-Displacement Relations . . . . . . . . . . . . . . . . 88
3.6 Geometric Measures of Deformation . . . . . . . . . . . . 91
3.6.1 Stretch . . . . . . . . . . . . . . . . . . . . . . . . 91
3.6.2 Shear . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.6.3 Volume Change . . . . . . . . . . . . . . . . . . . 98
3.6.4 Area Change . . . . . . . . . . . . . . . . . . . . . 102
3.7 Principal Strains . . . . . . . . . . . . . . . . . . . . . . . 103
3.7.1 The Eigenvalue Problem . . . . . . . . . . . . . . 104
3.7.2 Strain Invariants . . . . . . . . . . . . . . . . . . . 105
3.8 Stretch and Rotation Tensors . . . . . . . . . . . . . . . . 108
3.8.1 Polar Decomposition . . . . . . . . . . . . . . . . 108
3.8.2 Principal Stretch Ratios . . . . . . . . . . . . . . . 113
3.9 Approximations . . . . . . . . . . . . . . . . . . . . . . . . 119
3.9.1 Small Displacement . . . . . . . . . . . . . . . . . 120
3.9.2 Small Deformation . . . . . . . . . . . . . . . . . . 121
3.9.3 Small Deformation and Moderate Rotation . . . . 122
3.9.4 Small Rotation . . . . . . . . . . . . . . . . . . . . 122
3.10 Deformation Rates . . . . . . . . . . . . . . . . . . . . . . 122
Contents xv
3.10.1 Time Rates of Change . . . . . . . . . . . . . . . . 122

3.10.2 Rate-of-Deformation and Spin Tensors . . . . . . . 126
3.11 Compatibility Conditions . . . . . . . . . . . . . . . . . . 132
3.12 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4. Analysis of Stress 139

4.1 Body and Contact Forces . . . . . . . . . . . . . . . . . . 139
4.2 Stress Tensors . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.2.1 Cauchy Stress Tensor . . . . . . . . . . . . . . . . 141
4.2.2 Pseudostress Tensors . . . . . . . . . . . . . . . . 145
4.3 Relations Between Stress Components . . . . . . . . . . . 146
4.4 Physical Components of Stress . . . . . . . . . . . . . . . 153
4.5 Geometric Interpretation of Stress Components . . . . . . 154
4.6 Principal Stresses . . . . . . . . . . . . . . . . . . . . . . . 155
4.7 Equations of Motion . . . . . . . . . . . . . . . . . . . . . 156
4.7.1 Principle of Linear Momentum . . . . . . . . . . . 156
4.7.2 Principle of Angular Momentum . . . . . . . . . . 157
4.7.3 Lagrangian Form of Equation of Motion . . . . . . 159
4.7.4 Component Forms of Equations of Motion . . . . 162
4.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5. Constitutive Relations 173

5.1 Thermodynamics of Deformation . . . . . . . . . . . . . . 173
5.1.1 First Law of Thermodynamics . . . . . . . . . . . 173
5.1.2 Second Law of Thermodynamics . . . . . . . . . . 179
5.2 Fundamental Constitutive Principles . . . . . . . . . . . . 181
5.2.1 Principle of Coordinate Invariance . . . . . . . . . 182
5.2.2 Principle of Determinism . . . . . . . . . . . . . . 182
5.2.3 Principle of Local Action . . . . . . . . . . . . . . 183
5.2.4 Principle of Equipresence . . . . . . . . . . . . . . 183
5.2.5 Principle of Material Frame Indifference . . . . . . 184
5.2.6 Principle of Physical Admissibility . . . . . . . . . 194
5.3 Strain-Energy Density Function . . . . . . . . . . . . . . . 200
5.3.1 Orthotropic Material . . . . . . . . . . . . . . . . 201
5.3.2 Transversely Isotropic Material . . . . . . . . . . . 203
5.3.3 Isotropic Material . . . . . . . . . . . . . . . . . . 206
5.4 Constitutive Relations for Some Common Material Sym-
metries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
xvi Contents

5.4.4 Global Curvilinear Material Symmetries . . . . . . 216
5.5 Incompressibility . . . . . . . . . . . . . . . . . . . . . . . 219
5.6 Linear Elastic Material . . . . . . . . . . . . . . . . . . . . 223
5.7 Determining W for Soft Tissues . . . . . . . . . . . . . . . 225
5.7.1 Microstructural and Phenomenological Approaches 225
5.7.2 Experimental Considerations . . . . . . . . . . . . 225
5.7.3 Some Types of Experiments . . . . . . . . . . . . 226
5.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
6. Biomechanics Applications 233

6.1 Boundary Value Problems . . . . . . . . . . . . . . . . . . 233
6.2 Extension and Compression of Soft Tissue . . . . . . . . . 236
6.2.1 Governing Equations . . . . . . . . . . . . . . . . 237
6.2.2 Biaxial Stretching of a Membrane . . . . . . . . . 238
6.2.3 Uniaxial Extension of a Bar . . . . . . . . . . . . 243
6.3 Simple Shear of Soft Tissue . . . . . . . . . . . . . . . . . 243
6.3.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . 249
6.4 Extension and Torsion of a Papillary Muscle . . . . . . . . 251
6.4.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . 258
6.5 Extension, Inflation, and Torsion of an Artery with Resid-
ual Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
6.5.1 Artery Model . . . . . . . . . . . . . . . . . . . . . 265
6.5.3 Solution . . . . . . . . . . . . . . . . . . . . . . . . 271
6.5.4 Geometric and Material Properties . . . . . . . . . 272
6.5.5 Residual Stress in Unloaded Artery . . . . . . . . 273
6.5.6 Loaded Artery . . . . . . . . . . . . . . . . . . . . 275
6.6 Passive Filling of the Left Ventricle . . . . . . . . . . . . . 277
6.6.1 Model for the Left Ventricle . . . . . . . . . . . . 277
6.6.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . 279
Contents xvii
6.6.4 Results . . . . . . . . . . . . . . . . . . . . . . . . 283

6.7 Blastula with Internal Pressure . . . . . . . . . . . . . . . 285
6.7.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . 288
6.7.4 Results . . . . . . . . . . . . . . . . . . . . . . . . 290
6.8 Bending of the Embryonic Heart . . . . . . . . . . . . . . 291
6.8.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . 295
6.8.4 Results . . . . . . . . . . . . . . . . . . . . . . . . 296
6.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
Appendix A Linear Theory of Elasticity 309

A.1 Mathematical Preliminaries . . . . . . . . . . . . . . . . . 309
A.1.1 Base Vectors . . . . . . . . . . . . . . . . . . . . . 311
A.1.2 Vectors, Dyadics, and Tensors . . . . . . . . . . . 312
A.1.3 Coordinate Transformation . . . . . . . . . . . . . 315
A.1.4 Vector and Tensor Calculus . . . . . . . . . . . . . 317
A.2 Analysis of Deformation . . . . . . . . . . . . . . . . . . . 318
A.2.1 Strain . . . . . . . . . . . . . . . . . . . . . . . . . 318
A.2.2 Rotation . . . . . . . . . . . . . . . . . . . . . . . 327
A.2.3 Principal Strains . . . . . . . . . . . . . . . . . . . 329
A.2.4 Compatibility Conditions . . . . . . . . . . . . . . 333
A.3 Analysis of Stress . . . . . . . . . . . . . . . . . . . . . . . 335
A.3.1 Body and Contact Forces . . . . . . . . . . . . . . 335
A.3.2 Stress Tensor . . . . . . . . . . . . . . . . . . . . . 335
A.3.3 Principal Stresses . . . . . . . . . . . . . . . . . . 340
A.3.4 Equations of Motion . . . . . . . . . . . . . . . . . 340
A.4 Constitutive Relations . . . . . . . . . . . . . . . . . . . . 347
A.4.1 Thermodynamics of Deformation . . . . . . . . . . 347
A.4.2 Strain-Energy Density Function . . . . . . . . . . 350
A.4.3 Linear Elastic Material . . . . . . . . . . . . . . . 351
A.5 Boundary Value Problems . . . . . . . . . . . . . . . . . . 353
A.5.1 Torsion of a Circular Cylinder . . . . . . . . . . . 354
A.5.2 Torsion of a Noncircular Cylinder . . . . . . . . . 358
Appendix B Special Coordinate Systems 363

xviii Contents
B.1 Cylindrical Polar Coordinates . . . . . . . . . . . . . . . . 363

B.2 Spherical Polar Coordinates . . . . . . . . . . . . . . . . . 365
B.3 Toroidal Coordinates . . . . . . . . . . . . . . . . . . . . . 366
Problem Solutions 369
Bibliography 437
Index 447
Chapter 1
Introduction
Problems involving deformation of soft biological tissues are among the

most challenging in applied mechanics. In addition to undergoing large
strains, soft tissues exhibit nonlinear time-dependent behavior similar to
that of viscoelastic materials (Fung, 1993). Moreover, biological tissues are
not simply passive materials like those used in traditional engineering appli-
cations. Rather, they actively contract, grow, and remodel. Further compli-
cating matters, problems in soft tissue biomechanics often involve complex
three-dimensional geometry, dynamic loading conditions, and fluid-solid in-
teraction. These problems are highly nonlinear in general.
1.1 Elasticity and Pseudoelasticity
This book deals with nonlinear analysis of elastic materials, i.e., materi-
als that deform without loss of energy. At first glance, it may seem that
limiting the discussion to elastic materials severely restricts the range of
biomechanics applications. However, the characteristic behavior of many
soft tissues suggests that nonlinear elasticity can be used as a first ap-
proximation. This behavior, called pseudoelasticity, was first recognized
by Y.C. Fung during the 1970s (Fung et al., 1979). Even today, with the
rapidly increasing power of finite element and other computational meth-
ods, researchers take advantage of this feature to justify using elasticity
theory to study the mechanics of soft tissues.
The concept of pseudoelasticity is based on the following experimental
observation. During a tensile test, a soft tissue specimen ordinarily follows
separate loading and unloading stress-strain curves, which form a hystere-
sis loop (Fig. 1.1). Such behavior is typical of viscoelastic materials. If
the loading/unloading cycle is repeated at the same rate, the curves shift,
1
2 Introduction
presumably due to passive microstructural remodeling. After several load-

ing cycles, however, the stress-strain curves become repeatable, and the
response is relatively insensitive to loading rate. To describe this behavior,
Fung at al. (1979) suggested that the tissue can be treated as two separate
elastic materials — one during loading and another during unloading, i.e.,
the tissue is pseudoelastic.
1
2
3
STRESS
STRAIN
Fig. 1.1. Loading and unloading curves for a typical soft tissue. After the three cycles
shown, the curves show relatively little additional change.
This is one possible approach, but separate loading and unloading con-
stitutive relations have seldom been used in practice. Rather, most re-
searchers either use viscoelasticity theory or, more commonly, simply as-
sume that the material is elastic with the constitutive relation based only
on the loading curve. This approach, although not strictly correct, has pro-
vided useful approximate results in a large number of problems. One feature
that usually cannot be ignored, however, is nonlinear behavior, which is the
focus of this book.
Early development of the nonlinear theory of elasticity is attributed
largely to R.S. Rivlin, whose work on this topic began during the 1940s.
Many of his early papers have been republished together as a collection
(Rivlin et al., 1997). Rivlin initially worked without tensors, which are a
staple of this book, but later embraced it as a convenient analytical tool.
In addition, the classic work of Truesdell and Noll (2004), which helped
put nonlinear elasticity on a firm theoretical foundation, exudes tensors.
Today, it is difficult to imagine learning this subject without the use of
tensor analysis.
1.2. Biomechanics and Mechanobiology 3
1.2 Biomechanics and Mechanobiology
Biomechanics can be defined as the study of mechanical function and mal-

function in biological systems (Taber, 2020). By this definition, studies of
biomechanics go back many centuries (Humphrey, 2002; Innocenti, 2017).
While biomechanics always involves both biology and mechanics, the term
can represent different things to different people. Most people are per-
haps most familiar with the mechanics of motion, e.g., how humans walk,
run, jump, and hit a baseball. These problems deal primarily with rigid-
body dynamics. However, a large number of highly important problems
in biomechanics involve the mechanics of deformable bodies. Examples in-
clude cardiovascular dynamics, bone fracture, traumatic brain injury, and
the mechanics of hearing.
More recently, interest in the related field of mechanobiology has grown
rapidly. Mechanobiology can be defined as the study of how biological sys-
tems maintain or restore normal function in the face of abnormal mechanical
loads or damage caused by disease or trauma (Taber, 2020). These prob-
lems typically involve growth (change in volume), remodeling (change in
material properties), and morphogenesis (change in shape). Bone fracture
and the cardiac function are classical problems in biomechanics, while frac-
ture healing and adaptation of the heart to high blood pressure fall under
the realm of mechanobiology. Most of the problems in this book do not
deal directly with mechanobiology, which is considered in detail elsewhere
(Fung, 1990; Taber, 1995, 2020; Humphrey, 2002; Carter and Beaupre,
2007; Cowin and Doty, 2007; Epstein, 2012; Jacobs et al., 2013; Goriely,
2017; Verbruggen, 2018).
In modeling the biomechanics and mechanobiology of soft tissue, it is
essential to have an in-depth understanding of the theory of nonlinear elas-
ticity. Hence, Chapters 2 through 5 present the fundamental equations of
the nonlinear theory. Chapter 2 presents the mathematical background
needed to understand the material, and Chapters 3, 4, and 5 deal with
the analysis of deformation, stress, and constitutive behavior, respectively.
Some examples from biomechanics are discussed in these chapters, but the
emphasis is on physical and mathematical concepts. Then, with the theo-
retical foundation in place, Chapter 6 examines applications to a relatively
wide range of specific problems in soft tissue mechanics.
Chapter 2
Vectors, Dyadics, and Tensors
Physical quantities can be represented by tensors. Zeroth-order tensors

(scalars) have only a magnitude (e.g., temperature), first-order tensors (vec-
tors) have a magnitude and direction (e.g., velocity), second-order tensors
(dyadics) have a magnitude and two associated directions (e.g., stress1 ), and
so on. The literature on nonlinear (finite) elasticity is filled with tensors,
which provide a convenient tool for deriving the governing equations. For
solving these equations, tensor quantities then can be expressed in terms of
components relative to any convenient coordinate system. Because much
of the historical literature on finite elasticity deals with general curvilinear
coordinates, such generality is a main feature throughout this book.
2.1 Reference Frames and Coordinate Systems
Often the terms “reference frame” and “coordinate system” are used syn-
onymously. There is, however, an important distinction. A reference
frame is attached to an observer, and it may be in motion relative to other
frames with other attached observers. In each frame, there can be one
or more coordinate systems that can be used to locate a point in that
frame. The coordinate systems in a particular frame are fixed relative to
each other.
Consider a vector a fixed in reference frame A (Fig. 2.1), which contains
a Cartesian (x, y, z) and a cylindrical polar (r, θ, z) coordinate system (pos-
sibly with different origins). Relative to the Cartesian system, a has the
components {ax , ay , az }, and, relative to the polar system, a has the compo-
nents {ar , aθ , az }. These different components describe the same vector a.
1 A stress component depends on its direction of action and the orientation of the area
on which it acts.
5
6 Vectors, Dyadics, and Tensors
The transformation of components from the Cartesian to the polar system

is an example of coordinate transformation.
z
a z
y
y
r
x q
A B
x
Fig. 2.1. Reference frames A and B with attached coordinate systems. Frame B is in
motion relative to frame A.
Now, consider another reference frame B that is translating and ro-

tating relative to A (Fig. 2.1). The vector a is fixed, i.e., geometrically
invariant, in A, but it appears to be changing direction to an observer in
B. Relative to a Cartesian system (x̄, ȳ, z̄) fixed in B, the components of a
change with time. A coordinate transformation again can be used to relate
the components of a from either coordinate system in A to the Cartesian
system in B, but now the transformation coefficients are time-dependent.
Thus far, the only difference between reference frames and coordinate
systems is the potential time dependency of the transformation of vector
components. The distinction between the two becomes more significant in
physical problems. For example, consider Newton’s second law of motion
f = ma, where f is the force vector acting on a particle of mass m and accel-
eration vector a. If A is an inertial (nonrotating, nonaccelerating) reference
frame, then this equation is valid in any coordinate system in A, although
its components may change. Relative to another inertial frame B, f = ma
remains valid (if relativistic effects are ignored.) If B is a noninertial (rotat-
ing or accelerating) frame, however, the form of Newton’s law changes. The
reason for this lies in the acceleration vector a. Transforming f from frame
A to frame B involves only a time-dependent coordinate transformation,
but the transformation of a is not so direct. For instance, if f = a = 0 in A,
f also is zero as seen by an observer in B, but the mass moves with a = 0
relative to this accelerating observer. (So-called fictitious or pseudo forces,
2.2. Vectors 7
such as the Coriolis force, enter the analysis.) The force vector is an exam-
ple of a frame indifferent or objective quantity, while the acceleration
vector a is coordinate invariant but not frame indifferent.
Tensor analysis deals with coordinate transformation. In this chapter,
we assume that we are dealing with various coordinate systems in a single
frame of reference at a given time. Although the development is valid for
n-dimensional Euclidean space, we restrict our attention to (at most) three
dimensions.
2.2 Vectors
A vector a has a magnitude and direction. Relative to a given coordinate

system, a possesses a unique set of components. If the coordinate system
in a given reference frame changes, the components of a change, but the
vector a does not. A vector, therefore, is geometrically invariant, and the
form of a vector equation is independent of any coordinate system in that
frame.
Due in part to their invariance, vector formulations of physical prob-
lems often provide intuitive clarity. Solving these problems, however, usu-
ally requires expressing equations in terms of scalar components in a given
coordinate system. Many problems simplify when the governing equations
are written in a coordinate system other than Cartesian, e.g., polar or ellip-
soidal coordinates. A general curvilinear coordinate system does not even
have to be orthogonal. In this section, we examine how vectors can be
described in various coordinate systems.
2.2.1 Base Vectors

We begin by introducing a notational convention. In rectangular Cartesian
coordinates, we can write
a = â1 e1 + â2 e2 + â3 e3
3
= âi ei ≡ âi ei , (2.1)
i=1
where the âi are the components of a with respect to the orthogonal triad
of unit vectors ei , which are tangent to the coordinate axes z 1 , z 2 , and z 3 .2
The second line of Eq. (2.1) defines the summation convention, with
2 Henceforth, unless stated otherwise, the e represent unit vectors and zi represent
i
Cartesian coordinates, a special case of the general curvilinear coordinates xi .
summation implied over 1,2,3 for each repeated index in a single term.3
Because any letter can replace it without changing the meaning of an ex-
pression, a repeated index also is called a “dummy” index. For example,
aii = akk = a11 + a22 + a33 ,
while
3

ai bij = ak bkj = ai bij
i=1
gives a separate equation for each j = 1, 2, 3. Note that the number of upper
and lower indices in an equation must “balance,” i.e., after all summations
are carried out, each term must be left with the same subscripts and super-
scripts. Moreover, in general curvilinear coordinates, a single term cannot
contain, for example, three i’s or even two i’s on the same level; hence, ai bi
and aii are meaningless expressions.4 Throughout the remainder of this
book, the summation convention holds unless stated otherwise or unless
the indices are placed in parentheses, e.g., the term a(i) b(i) is not summed
over i but rather represents a separate expression for each i = 1, 2, 3.
In a Cartesian coordinate system, the ei are called base vectors. As
shown by Eq. (2.1), any vector can be expressed as a linear combination
of the base vectors. If the z i -axes and base vectors ei undergo a rigid-
body rotation to give a new coordinate system z̄ i with new base vectors ēi
(Fig. 2.2), then a can be resolved along these new base vectors as
a = ¯âi ēi , (2.2)
¯i
where â = â . A primary objective of tensor analysis is to relate compo-
i
nents and base vectors in different coordinate systems.

Similarly, in general curvilinear coordinates (x1 , x2, x3 ), we intro-
duce a set of base vectors gi , which are tangent to the xi -curves at each
point in space (Fig. 2.3). But now, as we will see, these vectors are not
necessarily unit vectors or dimensionless. Moreover, in moving along the
coordinate curves, the gi can change in both magnitude and direction. In
terms of these base vectors, similar to Eq. (2.1), the vector a can be written
a = ai g i , (2.3)
3 Einstein
(1916) introduced the summation convention as shorthand notation in devel-
oping the general theory of relativity.
4 A notable exception is Cartesian coordinates, where it usually is more convenient to
use only subscripts. In this case, the summation convention apples to repeated subscripts
in a single term (see Appendix A).
2.2. Vectors 9
z2
z2 a = aiei = aiei
a2
z1
a2
e2 a1
e2 q e1
q
z1
e1 a1
Fig. 2.2. Components of vector a relative to two Cartesian coordinate systems, zi and
z̄i .
where the ai are called the contravariant components (superscript i)

of a relative to the covariant base vectors gi (subscript i) of the xi
coordinate system.5 It is important to emphasize that we now have written
the same vector a in terms of three different sets of components, giving
a = âi ei = ¯âi ēi = ai gi ,
¯i = ai .
where âi = â
We now show how the covariant base vectors can be computed for a
general xi system. The position vector from the origin to an arbitrary
point P (xi ) in space is (Fig. 2.3)
r = r(xi ) = r i gi , (2.4)
and the differential position vector is
∂r
dr =dxi = r,i dxi , (2.5)
∂xi
where r,i ≡ ∂r/∂xi . (Note that a superscript in the denominator becomes
a subscript in the numerator.) In the vicinity of P , a small movement of
the tip of r along the x1 coordinate curve corresponds to a change dx1 in
the coordinate, with similar results for the other directions. We take dxi
5 Actually, any set of three linearly independent, i.e., noncoplanar, vectors can be used
as “base vectors.” The gi can be considered the “natural base vectors” for a given
coordinate system.
g3
x3
g2
x2
P g1
r dr
x1
r+dr
0
Fig. 2.3. Position vector r and differential position vector dr in curvilinear coordinate
system with base vectors gi .
as the component of dr in the gi direction. Then, resolving the vector dr

into components relative to the basis {gi } gives
dr = gi dxi , (2.6)
and comparison with (2.5) shows that
gi = r,i . (2.7)
According to Eq. (2.6), the dxi are the contravariant components of dr

with respect to the gi . Note, however, that the xi are not the contravariant
components of r in general; only in Cartesian coordinates can we set r i =
xi . Moreover, at a given time in general curvilinear coordinates, the base
vectors and, therefore, the components of a given vector depend on position,
while in Cartesian coordinates, the base vectors and the components of the
vector are independent of its location in space.
Example 2.1 Determine the covariant base vectors for each of the follow-
ing coordinate systems:6
(1) Cartesian coordinates: (x1 , x2 , x3 ) = (x, y, z).

6 To avoid confusion, powers are accompanied by parentheses. Thus, (y)2 is “y squared,”
whereas y2 is a contravariant vector component.

2.2. Vectors 11
(2) Cylindrical polar coordinates: (x1 , x2 , x3) = (r, θ, z) with x = r cos θ

and y = r sin θ (Fig. 2.4a).
(3) Skew coordinates: (x1 , x2 , x3 ) with x = x1 + x2 sin α, y = x2 cos α, and
z = x3 (Fig. 2.4b).
(a) y (b)
gq = req y x2
x P
gr = er
a x1
r
q x2 y
x ey g2
a
x,x1
g1 = ex
Fig. 2.4. Cylindrical polar (a) and skew (b) coordinates.
Solution. Relative to these coordinate systems, the position vector to a

given point P in space can be written in the forms
r = xex + yey + zez
= r cos θ ex + r sin θ ey + zez
= (x1 + x2 sin α)ex + (x2 cos α)ey + x3 ez . (2.8)
Then, Eq. (2.7) gives the base vectors
gx = r,x = ex
gy = r,y = ey
gz = r,z = ez (2.9)
in Cartesian coordinates,
gr = r,r = cos θ ex + sin θ ey = er
gθ = r,θ = r(− sin θ ex + cos θ ey ) = reθ
gz = r,z = ez (2.10)
in cylindrical polar coordinates (er and eθ are unit vectors, see Fig. 2.4a),
and
g1 = r,1 = ex
g2 = r,2 = sin α ex + cos α ey
g3 = r,3 = ez (2.11)
in skew coordinates. These sets of base vectors are illustrated in Fig. 2.4
(in two dimensions).
Here, we note several things. First, we have expressed r in terms of
the Cartesian base vectors in each case; since these vectors are constant,
this simplifies the algebra. Second, gθ is not a unit vector and it carries
the dimension of length. Third, since gx · gy = gr · gθ = 0, Cartesian and
polar coordinates are orthogonal. (Obviously, all these coordinate systems
are orthogonal relative to the z-direction.) However, g1 · g2 = sin α = 0,
indicating the nonorthogonality of skew coordinates (for α = 0). Finally,
for future reference, Eqs. (2.10) yield
∂er ∂eθ ∂er ∂eθ

= = 0, = eθ , = −er (2.12)
∂r ∂r ∂θ ∂θ
for the derivatives of the unit base vectors in cylindrical polar coordinates.

Example 2.2 For cylindrical polar coordinates, determine gr and gθ with-

out introducing a Cartesian system.
Solution. In polar coordinates, the position vector is
r = r er (θ),
and differentiation yields
gr = r,r = er
∂er
gθ = r,θ = r .
∂θ
The first expression agrees with Eq. (2.10)1. To find ∂er /∂θ directly, we
examine the geometry (Fig. 2.5). Consider two points in the rθ-plane, P1
and P2 , that are located the same distance from the origin but are separated
by the small angle dθ. In moving from P1 to P2 , only the orientation of
er changes. To a first approximation, this change is (∂er /∂θ) dθ. The
geometry of Fig. 2.5 gives

∂er

∂θ dθ = |er | · dθ
or

∂er

∂θ = 1,
2.2. Vectors 13
which shows that the derivative has unit magnitude. Moreover, as dθ →

0, the vector ∂er /∂θ becomes perpendicular to er , i.e., it points in the
direction of eθ . Hence,
∂er
= eθ
∂θ
and so gθ = reθ , in agreement with Eq. (2.10)2.
der
er + dq
dq
P2 der
dq
dq
er
dq P1
r
Fig. 2.5. Geometry for the differential change in the unit vector er with θ.
2.2.2 Reciprocal Base Vectors

The dot product of the two vectors a = ai gi and b = bj gj is7
a · b = (ai gi ) · (bj gj ) = ai bj gi · gj
= a 1 b 1 g 1 · g 1 + a1 b 2 g 1 · g 2 + a1 b 3 g 1 · g 3
+a2 b1 g2 · g1 + a2 b2 g2 · g2 + a2 b3 g2 · g3
+a3 b1 g3 · g1 + a3 b2 g3 · g2 + a3 b3 g3 · g3 . (2.13)
If the coordinates are orthogonal, then gi ·gj = 0 for i = j, and this equation
simplifies considerably. For general coordinates, however, this expression is
quite a mess.
To render this situation less painful, we introduce a new set of base
vectors gi , which are called reciprocal or contravariant base vectors
7 Note that, in the representation for b, we have replaced the i’s by j’s without altering
the meaning. This procedure is necessary since more than two of any index is not allowed
in any single term.
in the xi coordinate system. [Simmonds (1994) calls the gi and gi “cellar”

and “roof” base vectors, respectively, according to the location of the index.]
The gi are defined by the relation
gi · gj = δ ji , (2.14)
where

1 for i = j
δ ji = (2.15)
0 for i = j
is the Kronecker delta. Thus, g1 is orthogonal to both g2 and g3 , and

so on (Fig. 2.6).
a2g2
g2
x2
g2
a2g2
a = aigi = aigi
a1g1 g1 x1
g1
a1g1
Fig. 2.6. Covariant and contravariant base vectors and components of a vector a in a
two-dimensional nonorthogonal coordinate system.
Of course, a vector can be written in terms of components relative to

any basis. Hence, if b is expressed relative to the contravariant basis, i.e.,
2.2. Vectors 15
b = bi gi , where the bi are the covariant components of b, then

a · b = (ai gi ) · (bj gj ) = ai bj (gi · gj ) = ai bj δ ji
= ai b i
= a 1 b 1 + a2 b 2 + a3 b 3 . (2.16)
This relation, which takes the form of the usual dot product in Cartesian
coordinates, is considerably more palatable than Eq. (2.13). Note that
the effect of δ ji in a term is simply to replace j by i or, equivalently, i by
j, with δ ji then removed. This mathematical procedure is called tensor
contraction.
Figure 2.6 illustrates a set of covariant and contravariant base vectors
at a point in two-dimensional space. The vectors g1 and g2 are orthogonal
to g2 and g1 , respectively. Moreover, the magnitude and directions of g1
and g2 are determined by the requirement g1 · g1 = g2 · g2 = +1.
The gi can be computed in terms of the gi as follows. First, consider
1
g . Since it is orthogonal to both g2 and g3 , we can write
√ 1
g g = g2 × g3
√
according to the properties of the cross product. The scaling factor g,
found by dotting both sides of this equation with g1 , is
√
g = g1 · (g2 × g3 ), (2.17)
which is the scalar triple product of the covariant base vectors. From
√
elementary vector analysis, Eq. (2.17) implies that g represents the vol-
ume of the parallelepiped with the vectors gi as edges. The reason for the
square root is discussed later (Sec. 2.6.2). This and similar manipulations
yield
g2 × g3 g3 × g1 g1 × g2
g1 = √ , g2 = √ , g3 = √ . (2.18)
g g g
Any vector can be expressed in terms of either its contravariant or co-

variant components (Fig. 2.6), i.e.,
a = ai g i = ai g i . (2.19)
The components of a can be extracted by dotting this relation with gj or
gj to get
a · gj = (ai gi ) · gj = ai δ ji = aj
a · gj = (ai gi ) · gj = ai δ ij = aj . (2.20)
These equations show that dotting a vector with a base vector gives the
component of the same variance.
The physical meaning of Eqs. (2.20) can be seen using elementary vector
analysis (see Fig. 2.6). First, we can write gj = |gj |ej (j not summed),
√
where |gj | = gj · gj is the magnitude of gj and ej is the unit vector in
the direction of gj . Thus, aj = a · gj = |gj |(a · ej ) (j not summed), which
shows that aj represents the product of |gj | and the orthogonal projection
of a in the direction of gj . Similarly, the relation aj = a · gj = |gj |(a · ej )
(j not summed) indicates that aj is the product of |gj | and the orthogonal
projection of a in the direction of gj .
Judicious use of covariant and contravariant components and base vec-
tors leads to numerous simplifications in vector and tensor analysis.
Example 2.3 Determine the contravariant base vectors for each of the
coordinate systems of Example 2.1 (Fig. 2.4).
Solution. Recall that the scalar triple product of the vectors a, b, and
c can be written

a 1 a2 a3

a · (b × c) = b · (c × a) = c · (a × b) = b1 b2 b3 , (2.21)
c c c
1 2 3
where ai , bi , and ci are Cartesian components. (The determinant represen-

tation for the scalar triple product is not valid in every coordinate system.)
Using Cartesian components of the natural base vectors, Eqs. (2.9)–(2.11),
and (2.17) give

1 0 0
√
g = gx · (gy × gz ) = 0 1 0 =1
(Cartesian)
0 0 1

cos θ sin θ 0

= gr · (gθ × gz ) = −r sin θ r cos θ 0 = r (cylindrical)
0 0 1

1 0 0

= g1 · (g2 × g3 ) = sin α cos α 0 = cos α (skew).
0 0 1
(2.22)
2.3. Dyadics and Tensors 17
Equations (2.18) give the reciprocal base vectors

g x = gy × gz = ey × ez = ex
g y = gz × gx = ez × ex = ey
g z = gx × gy = ex × ey = ez (2.23)
in Cartesian coordinates,
1 1
gr = (gθ × gz ) = (reθ × ez ) = er
r r
1 1 eθ
gθ = (gz × gr ) = (ez × er ) =
r r r
1 1
gz = (gr × gθ ) = (er × reθ ) = ez (2.24)
r r
in cylindrical polar coordinates, and
1 1
g1 = (g2 × g3 ) = (cos α ex − sin α ey ) = ex − tan α ey
cos α cos α
1 ey
g2 = (g3 × g1 ) =
cos α cos α
1 1
g3 = (g1 × g2 ) = (cos α ez ) = ez (2.25)
cos α cos α
in skew coordinates. It is easy to show that Eq. (2.14) is satisfied in each
case.
Note that, in orthogonal coordinates (e.g., Cartesian or cylindrical po-
lar), gi is oriented in the same direction as gi , but the magnitudes of these
base vectors may differ. Furthermore, in Cartesian coordinates, gi = gi ,
and so there is no difference between covariant and contravariant quan-
tities. Because the placement of the index is immaterial in this case, the
components of Cartesian tensors often are written with subscripts only (see
Appendix A).
2.3 Dyadics and Tensors
2.3.1 Dyadics
The tensor product of two vectors a and b, denoted by ab, is called a
dyad.8 Since each of the vectors in a dyad is geometrically invariant in a
8 Some authors use the notation a ⊗ b for the tensor product.
given frame of reference, it follows that the dyad also is invariant. The order
of the vectors composing a dyad is significant, and so the tensor product is
not commutative, i.e., ab = ba. The same is true of higher-order polyads,
such as abc or abcd, which also are useful in some applications. Here,
we are concerned primarily with dyads and linear combinations of dyads,
which are called dyadics. Equations written in terms of vector and tensor
components (e.g., fi = mai ) are written in indicial notation, whereas
those written without components (e.g., f = ma) are said to be written
in direct notation. As we will see later, dyadics link indicial and direct
notation.
Algebraic manipulation of dyads is straightforward so long as care is
taken to maintain the order of the component vectors. For example, the
operations
ab · c = a(b · c)
c · ab = (c · a)b (2.26)
show that the dot product of a dyad with a vector yields another vector.
Similarly, the equations
ab · cd = a(b · c)d = (b · c) ad
cd · ab = c(d · a)b = (d · a) cb (2.27)
show that the single-dot product of two dyads produces another dyad.
These equations also show that the single-dot product of two dyads is not
commutative in general. (Recall that the dot product of two vectors is
commutative.)
Two types of double-dot (scalar) product are useful in manipulating
dyads. The vertical double-dot product is defined by
ab : cd = (a · c)(b · d) = c · (ab) · d
= (c · a)(d · b) = cd : ab, (2.28)
which shows that this operation is commutative for dyads. For polyads,
however, this may not be the case, e.g.,
abc : defg = a(bc : de)fg
= a(b · d)(c · e)fg
= (b · d)(c · e)afg
= defg : abc.
Again, with care taken to maintain order, the algebra is straightforward.
Table 2.1. Vector and dyadic formulas
a+b = b+a
(a + b) + c = a + (b + c)
a·b = b·a
a · (b + c) = a · b + a · c
φ(a + b) = φa + φb
(φ + ψ)a = φa + ψa
a × b = −b × a
a × (b × c) = (a · c)b − (a · b)c = a · (cb − bc)
(a × b) × c = (a · c)b − (b · c)a = c · (ab − ba)
a · (b × c) = b · (c × a) = c · (a × b)
φ(ab) = (φa)b = a(φb) = φab
a(bc) = (ab)c = abc
a(b + c) = ab + ac
φ(ab + cd) = φab + φcd
The other type of double-dot product is the horizontal double-dot

product defined by
ab · · cd = (a · d)(b · c) = cd · · ab, (2.29)
which also is commutative for dyads. These operations can be extended to

.
polyads by adding more dots, e.g., abc .. def = (a · d)(b · e)(c · f ). Table
2.1 contains some useful relations for vectors and dyadics.
2.3.2 Tensors
A linear vector function T satisfies the relations
T · (a + b) = T · a + T · b and T · (φa) = φT · a (2.30)
for any vectors a and b and scalar φ. A second-order tensor T can be de-
fined as a linear vector function that, when dotted with a vector, transforms
the vector into another vector, i.e.,9
a = T · b. (2.31)
Since the vectors a and b are invariant, T also must be invariant. A third-
order tensor is a linear vector function that transforms a vector into a
second-order tensor through a single-dot product or a second-order tensor
into a vector through a double-dot product, and so on. The equation
(ab) · (φc + ψd) = φa(b · c) + ψa(b · d)
= [φ(b · c) + ψ(b · d)]a, (2.32)
with φ and ψ being scalars, shows that a dyad ab is a linear vector function
that satisfies the definition of a second-order tensor. (Of course, the dot
products b · c and b · d also are scalars.) In this way, we can show that a
vector is a first-order tensor, a dyad is a second-order tensor, a triad is a
third-order tensor, and so on.
Any second-order tensor can be represented by a dyadic. Because they
span the space, base vectors are convenient for forming the component
dyads. In three-dimensional space, for example, we can write
T = T 11 g1 g1 + T 12 g1 g2 + T 13 g1 g3
+T 21 g2 g1 + T 22 g2 g2 + T 23 g2 g3
+T 31 g3 g1 + T 32 g3 g2 + T 33 g3 g3
= T ij gi gj , (2.33)
where the T ij are the contravariant components of the second-order tensor
T with respect to the covariant basis {gi gj }. Like a vector, a tensor can be
expressed in terms of components referred to any basis. Also like a vector,
under a change of basis in a given frame of reference, a tensor does not
change, but its components do. For a vector, the base vectors form a basis;
for a tensor, the dyads (or polyads) composed of base vectors form a basis.
In terms of natural base vectors, T has four fundamental component
representations:
T = T ij gi gj = Tij gi gj = T·ji gi gj = Ti·j gi gj . (2.34)
The Tij are covariant components, and the T·ji and Ti·j are mixed com-
ponents of T. By convention, the first index on a tensor component is
9 In this chapter, lower-case bold letters denote first-order tensors (vectors), and upper-
case bold letters denote higher-order tensors.

associated with the first vector of the dyad, and the second index is as-
sociated with the second vector. Since T·ji = Tj·i in general, the dots in
the mixed components are needed for clarity to keep the place of the first
index. Again, we emphasize the importance of the order of the base vectors
in Eq. (2.34).
The single-dot product of a vector with a base vector was used previously
to extract the component of the vector with the same variance as the base
vector [see Eqs. (2.20)]. Analogously, vertical-double-dotting a second-order
tensor T with a dyadic basis gives the components of T with the same
variance as the basis. For example, extending Eqs. (2.20) gives10
T : gi gj = (Tk·l gk gl ) : gi gj
= Tk·l (gk · gi )(gl · gj )
= Tk·l δ ki δ jl
= Ti·j , (2.35)
which is of the same variance as gi gj . The other components of T can be

found similarly, giving
T : gi gj = gi · T · gj = Tij
T : gi gj = gi · T · gj = T ij
T : gi gj = gi · T · gj = Ti·j
T : gi gj = gi · T · gj = T·ji . (2.36)
Unlike vector components, the physical meaning of the projection of a ten-

sor on a dyad is not intuitively obvious.
2.3.3 Matrix Representation of Tensors
Matrix algebra is useful for computing with the components of tensors.

However, it must be kept in mind that these components are attached to
base vectors. For example, [Tij ] and [Ti·j ] are matrix representations of the
same tensor T, but their full meanings are not known without defining their
bases.
In this book, we write the matrix form of a second-order tensor T in
10 We could have used any of the four representations for T in Eq. (2.34) and ended up
with the same result. The reader may find it instructive to try this.
Table 2.2. Tensor formulas
T+U = U+T
(T + U) + V = T + (U + V)
T · (U + V) = T · U + T · V
φ(T + U) = φT + φU
(φ + ψ)T = φT + ψT
T · (U · V) = (T · U) · V
T : U = TT : UT = U : T
T : (U + V) = T : U + T : V
T : (U · V) = U : (T · VT ) = V : (UT · T)
φ(T : U) = (φT) : U = T : (φU)
the following ways:

⎡ ⎤
T·11 T·21 T·31
T(gi gj ) = T·ji (g j) = ⎣ T·12 T·22 T·32 ⎦ .
ig
T·13 T·23 T·33 (g j)
ig
j
The subscript (gi g ), which identifies the basis, can be omitted if the ba-
sis is specified. According to convention, the first index (or base vector)
corresponds to the row and the second to the column of the matrix.
2.3.4 Some Properties of Tensors

Tables 2.2 and 2.3 give some useful relations involving tensors. In this sec-
tion, we focus on the transpose and inverse. Other properties are discussed
in Sec. 2.5.
Transpose. The transpose of a tensor T, denoted by TT , is defined by

the relation
b · T · a = a · TT · b, (2.37)
where a and b are arbitrary vectors. Because the dot product of a second-
order tensor and a vector is a vector, we can write this equation in the form
Table 2.3. Formulas for transpose and inverse
(ab)T = ba
(T + U)T = TT + UT
(TT )T = T
(T · U)T = UT · TT
(TT )−1 = (T−1 )T ≡ T−T
(T · U)−1 = U−1 · T−1
(T−1 )m = (Tm )−1 ≡ T−m
b · (T · a) = (a · TT ) · b = b · (a · TT ). Hence, Eq. (2.37) implies
T · a = a · TT . (2.38)
To find the component representation for TT , we substitute, for example,

Eq. (2.34)1 to get
T · a = (T ij gi gj ) · a = T ij gi (gj · a)
= T ij (a · gj )gi = a · (T ij gj gi ) = a · TT , (2.39)
T ij
and so T = T gj gi . These and similar manipulations with the other
component forms of T show that
TT = T ij gj gi = Tij gj gi = T·ji gj gi = Ti·j gj gi . (2.40)
Comparison with Eq. (2.34) reveals that the transpose of a second-order
tensor is given by interchanging the base vectors in any of its dyadic rep-
resentations (or by interchanging the rows and columns in a matrix repre-
sentation).
A second-order tensor T is symmetric if T = TT ; if T = −TT , the
tensor is antisymmetric or skew symmetric. For a symmetric tensor,
Eqs. (2.36) and (2.37) provide the following relations between the compo-
nents:
gi · TT · gj = gj · T · gi → T ij = T ji
gi · TT · gj = gj · T · gi → Tij = Tji
g · T · gj = gj · T · g
i T i
→ T·ji = Tj·i . (2.41)
Note that symmetry of T does not imply T·ji = T·ij . Thus, in general, the
matrix of mixed components with T·ji in the ith row and jth column is not
equal to the transpose of the matrix with T·ij in the jth row and ith column.
In other words, symmetry of the tensor T does not imply that the matrix
of mixed components is symmetric.
Inverse. The inverse of a tensor is defined by
T · T−1 = T−1 · T = I, (2.42)
where T−1 is the inverse of T and I is the identity tensor, which satisfies
T · I = T. It turns out that the mixed components of I are the Kronecker
delta δ ij , and so
I = δ ij gi gj = δ ij gj gi = gi gi = gi gi , (2.43)
which shows that I = IT . These representations can be verified, for exam-

ple, by the manipulation
T · I = (T ij gi gj ) · (gk gk ) = T ij gi δ kj gk
= T ij gi gj = T. (2.44)
A similar manipulation confirms that a · I = a.
2.4 Coordinate Transformation
Deriving governing equations for physical problems is usually easiest in

Cartesian coordinates. When symmetry is present, however, it may be more
convenient to solve a specific problem in some other coordinate system, such
as cylindrical or spherical polar coordinates. Tensor analysis offers a way
to take advantage of both situations. First, the equations can be derived in
Cartesian coordinates. Next, they can be expressed in direct (coordinate-
free) notation. Finally, components relative to the specialized basis can
be extracted. To do this, we need to relate quantities in different bases
(coordinate systems).
2.4.1 Transformation of Base Vectors

Coordinate Transformation Tensors. Let gi and ḡi represent the co-
variant base vectors and gi and ḡi the contravariant base vectors in the
2.4. Coordinate Transformation 25
curvilinear coordinate systems xi and x̄i , respectively. The base vectors

can be converted from the unbarred to the barred system through the re-
lations
ḡi = A · gi , ḡi = B · gi , (2.45)
where A and B are coordinate transformation tensors. Convenient

mixed-component representations for these tensors are
A = Ai·j gi gj , B = Bi·j gi gj (2.46)
relative to the unbarred base vectors. Substituting these equations into
(2.45) yields
ḡi = (Aj·k gj gk ) · gi = Aj·k gj δ ki = Aj·i gj
ḡi = (Bj·k gj gk ) · gi = Bj·k gj δ ik = Bj·i gj , (2.47)
which show that Aj·i is the component of the vector ḡi in the direction
of gj , and Bj·i is the component of ḡi in the direction of gj . Combining
Eqs. (2.46) and (2.47) gives
A = ḡi gi , B = ḡi gi , (2.48)
which also can be seen by directly substituting these relations into (2.45).
In general, the only components of A and B we need are those indicated
in Eqs. (2.47). Therefore, without confusion we set11
Aj·i ≡ Aji , Bj·i ≡ Bji (2.49)
with the understanding that dyadic operations are to be carried out with
A and B in the forms of Eqs. (2.46) or (2.48). Then, dotting Eqs. (2.47)1,2
with gk and gk , respectively, yields
Aji = ḡi · gj , Bji = ḡi · gj , (2.50)
which can be considered defining relations for the mixed components of

the transformation tensors. Since the order of the dotted vectors can be
switched, these equations are consistent with the order of the indices in
Aji and Bji being immaterial. If the barred and unbarred systems coincide,
11 Note, however, that the order of the base vectors corresponding to Aj and B ·i are
·i j
in the opposite order, see Eqs. (2.46). Therefore, in the matrix representation for A,
the superscript and subscript correspond to the row and column, respectively, but the
opposite is true for B.
Eqs. (2.14) and (2.50) give Aij = Bji = δ ij , and Eqs. (2.43) and (2.48) show
that A = B = I, as expected.
The transformation tensors A and B are related. Since the base vectors
in any coordinate system must satisfy Eq. (2.14), then gi · gj = ḡi · ḡj = δ ji .
Thus, Eqs. (2.38) and (2.45) give
gi · gj = ḡi · ḡj = (A · gi ) · (B · gj )
= (gi · AT ) · (B · gj ) = gi · (AT · B) · gj ,
which shows that AT · B = I or
B = A−T (2.51)
with A−T ≡ (AT )−1 as defined in Table 2.3. Similarly, Eqs. (2.47) yield
ḡi · ḡj = (Aki gk ) · (Blj gl ) = Aki Blj δ lk = δ ji
or
Aki Bkj = δ ji . (2.52)
Example 2.4 Compute the tensors A and B for a transformation from

Cartesian coordinates [(x1 , x2, x3 ) = (x, y, z)] to cylindrical polar coordi-
nates [(x̄1 , x̄2, x̄3 ) = (r, θ, z)].
Solution. Let gi and ḡi represent base vectors in the Cartesian and
cylindrical coordinate systems, respectively. Then, Eq. (2.48)1 gives
A = ḡi gi = ḡ1 g1 + ḡ2 g2 + ḡ3 g3

= ḡr gx + ḡθ gy + ḡz gz . (2.53)
Substituting Eqs. (2.10) for the ḡi and (2.23) for the gi yields
A = (cos θ ex + sin θ ey )ex + r(− sin θ ex + cos θ ey )ey + ez ez

= cos θ ex ex − r sin θ ex ey + sin θ ey ex + r cos θ ey ey + ez ez ,
or, in matrix form,

⎡ ⎤
cos θ −r sin θ 0
A(ei ej ) = [Aij ] = ⎣ sin θ r cos θ 0 ⎦ . (2.54)
0 0 1
(For a Cartesian basis, the up-down positions of the indices are immaterial.)
The tensor B can be found from Eq. (2.48) or Eq. (2.51). Here, we use
the latter equation and first compute
⎡ ⎤
cos θ sin θ 0
AT(ej ei ) = ⎣ −r sin θ r cos θ 0 ⎦ . (2.55)
0 0 1
The inverse of a tensor T can be computed from
(cof T)T
T−1 = , (2.56)
det T
where the components of cof T are the cofactors of the matrix [T·ji ], and
det T is the determinant of [T·ji ] (see Sec. 2.5.2). Thus, we have
⎡ ⎤
r cos θ − sin θ 0
1
B = A−T = (AT )−1 = ⎣ r sin θ cos θ 0 ⎦ . (2.57)
r
0 0 r (e e )
i j
These expressions for A and B can be used to check our results from
Examples 2.1 and 2.3. For instance, using Eq. (2.45)1 , we can transform
g2 = gy to ḡ2 = ḡθ by
⎡ ⎤⎡ ⎤
cos θ −r sin θ 0 0
ḡθ = A · gy = A · ey = ⎣ sin θ r cos θ 0 ⎦ ⎣ 1 ⎦
0 0 1 0
⎡ ⎤
−r sin θ
= ⎣ r cos θ ⎦ = −r sin θ ex + r cos θ ey ,
0
which agrees with gθ in (2.10).
Determining Transformation Components from Coordinates. In

practice, the components of the transformation tensors usually are com-
puted from the equations that relate the unbarred and barred curvilinear
coordinates. For a unique mapping of one system to the other, we can write
x̄i = x̄i (xj ), xi = xi (x̄j ). (2.58)
For example, if (x1 , x2 , x3 ) = (x, y, z) represent a set of Cartesian coor-
dinates and (x̄1 , x̄2 , x̄3 ) = (r, θ, z) are cylindrical polar coordinates, then
Eqs. (2.58)2 become x1 = x̄1 cos x̄2 , x2 = x̄1 sin x̄2 , and x3 = x̄3 .
The differential position vector, which is the same in both coordinate

systems, can be written
∂xi j
dr = ḡi dx̄i = gi dxi = gi
dx̄ , (2.59)
∂ x̄j
where the chain rule for differentiation has been used. In this equation,
dxi and dx̄i are the contravariant components of dr in the two coordinate
systems. This equation gives the relation
∂xj
ḡi − gj dx̄i = 0.
∂ x̄i
For arbitrary dx̄i , the expression in parentheses must vanish, and compar-
ison with Eq. (2.47)1 shows that
∂xj
Aji = . (2.60)
∂ x̄i
The components of B can be found from Eq. (2.52), i.e.,

∂xk j
Aki Bkj = B = δ ji .
∂ x̄i k
Since the chain rule gives
∂xk ∂ x̄j ∂ x̄j
= = δ ji ,
∂ x̄i ∂xk ∂ x̄i
comparing these expressions and changing indices yields
∂ x̄i
Bji = . (2.61)
∂xj
Often the xi are known in terms of the x̄j (or vice versa), and the relations
between coordinates are difficult to invert. In this case, Eq. (2.60) can be
used to compute Aji , and then Eq. (2.51) or (2.52) gives the Bji .
Example 2.5 Use Eqs. (2.60) and (2.61) to compute the components of
the tensors A and B for a transformation from Cartesian coordinates to
cylindrical polar coordinates.
Solution. Let (x1 , x2 , x3) = (x, y, z) and (x̄1 , x̄2 , x̄3) = (r, θ, z); then
x1 = x̄1 cos x̄2 , x2 = x̄1 sin x̄2 , x3 = x̄3
1 x2
x̄1 = (x1 )2 + (x2 )2 2 , x̄2 = tan−1 , x̄3 = x3 .
x1
We compute, for instance,

∂ x̄1 x2 x̄1 sin x̄2
B21 = 2
= 1 = = sin θ,
∂x [(x1 )2 + (x2 )2 ] 2 x̄1
which agrees with Eq. (2.57) of Example 2.4. (Recall that the superscript in
Bji corresponds to the column and the subscript to the row; see footnote on
page 25.) The other components of A and B can be determined similarly
(see Problem 2.8).
2.4.2 Transformation of Vector Components

The coordinate transformation relations for the components of a vector
follow directly from those for the base vectors. Relative to the unbarred
and barred coordinate systems, the vector a can be written
a = ai gi = ai gi = āi ḡi = āi ḡi , (2.62)
and applying Eqs. (2.20) and (2.50) gives the components
āi = ḡi · a = ḡi · (aj gj ) = Bji aj

āi = ḡi · a = ḡi · (aj gj ) = Aji aj
ai = gi · a = gi · (āj ḡj ) = Aij āj
ai = gi · a = gi · (āj ḡj ) = Bij āj . (2.63)
Comparison with (2.47) reveals that the transformations of vector compo-

nents parallel those of the base vectors.
2.4.3 Transformation of Tensor Components

Like those for vector components, the coordinate transformation relations
for tensor components fall out naturally when dyadic notation is used. For
example, consider the transformation of the covariant components of the
tensor
T = Tij gi gj = T̄ij ḡi ḡj . (2.64)
Extracting the components relative to the barred system yields [see

Eq. (2.36)]
T̄ij = ḡi · T · ḡj = ḡi · (Tkl gk gl ) · ḡj .

With Eqs. (2.50), this and similar expressions for the other components
give
T̄ij = Aki Alj Tkl

T̄ ij = Bki Blj T kl
T̄i·j = Aki Blj Tk·l
T̄·ji = Bki Alj T·lk . (2.65)
Inspecting Eqs. (2.63) and (2.65) reveals a pattern. Transformation

from the unbarred to the barred system requires an Aij for each covariant
index and a Bji for each contravariant index. The pattern reverses when the
transformation is the other way. It is more consistent with our approach,
however, to memorize Eqs. (2.50) and simply dot the vector and dyadic
representations with the appropriate base vectors. This is how we obtained
Eqs. (2.63) and (2.65).
Expressing the above equations in matrix form can simplify computa-
tion. The coordinate transformation matrices are
⎡ 1 ⎤ ⎡ ·1 ⎤
A·1 A1·2 A1·3 B1 B1·2 B1·3
[Aj·i ] = ⎣ A2·1 A2·2 A2·3 ⎦ , [Bj·i ] = ⎣ B2·1 B2·2 B2·3 ⎦ , (2.66)
A3·1 A3·2 A3·3 B3·1 B3·2 B3·3
where including the place-holding dots in the superscripts and subscripts
helps determine which index is the row (first index) and which is the column
(second index).
For the vector a, expanding Eqs. (2.63)1,2 yields
āi = Bj·i aj = B1·i a1 + B2·i a2 + B3·i a3

āi = Aj·i aj = A1·i a1 + A2·i a2 + A3·i a3 .
With the vector components represented by column vectors, these relations

show that the appropriate matrix equations are
[āi ] = [Bj·i ]T [aj ], [āi ] = [Aj·i ]T [aj ]. (2.67)
Consider now a second vector b subject to the same coordinate trans-

formations, i.e.,
[b̄i ] = [Bj·i ]T [bj ], [b̄i ] = [Aj·i ]T [bj ], (2.68)
and, for simplicity, let the tensor T be represented by the dyad ab, i.e.,
T = ab.
2.5. Tensor Invariants 31
For covariant components of T in the unbarred and barred systems, the

corresponding matrix equations are
[Tij ] = [ai ][bj ]T , [T̄ij ] = [āi ][b̄j ]T . (2.69)
Note that the matrix for a dyad is computed by dotting a column vector
(3 × 1) with a row vector (1 × 3) to obtain a 3 × 3 matrix. Substituting
(2.67)2 and (2.68)2 into (2.69)2 yields12
[T̄ij ] = [Aj·i ]T [aj ][bj ]T [Aj·i ] = [Aj·i ]T [Tij ][Aj·i ],
where (2.69)1 has been used. This result and similar manipulations give
[T̄ij ] = [Aj·i ]T [Tij ][Aj·i]
[T̄ ij ] = [Bj·i ]T [T ij ][Bj·i ]
[T̄i·j ] = [Aj·i ]T [Ti·j ][Bj·i ]
[T̄·ji ] = [Bj·i ]T [T·ji ][Aj·i ], (2.70)
which are the matrix forms of Eqs. (2.65).
Before leaving our discussion of coordinate transformation, we point out
an important feature of tensor analysis. As Eqs. (2.34) and (2.65) show,
if the components of a tensor with respect to one basis vanish, then the
components with respect to any basis vanish. Thus, an equation written
in tensor components, like an equation written in direct notation, holds in
any coordinate system in a given reference frame.
2.5 Tensor Invariants
2.5.1 Trace
The trace of a dyad is defined by
tr ab ≡ a · b. (2.71)
This operation contracts the dyad ab, since inserting a dot between the
vectors lowers the order of ab by two, transforming it into a scalar. For a
second-order tensor T, the trace is defined in terms of mixed components
as
tr T ≡ tr (T·ji gi gj ) = T·ji gi · gj = T·ji δ ji = T·ii .
12 For matrices, ([A][B])T = [B]T [A]T and the usual tensor rules for balancing indices
are not necessarily followed here.

Table 2.4. Formulas for trace (n = dimension of the space)
tr I = n
tr T = T : I
tr (T · UT ) = tr (TT · U) = T : U
tr TT = tr T
tr (φT) = φ tr T
tr (T + U) = tr T + tr U
tr (T · U) = tr (U · T)
tr (T · U · V) = tr (U · V · T) = tr (V · T · U)
Equivalently, the trace of T is the trace of the matrix of mixed components,

i.e.,
tr T = tr [T·ji ] = T·ii . (2.72)
In the barred coordinate system, Eqs. (2.52) and (2.65)4 give
T̄·ii = Bki Ali T·lk = δ lk T·lk = T·k

k
= T·ii . (2.73)
Thus, tr T does not change with a change of basis, and the trace is called a
scalar invariant of the tensor T. Some useful properties of the trace are
given in Table 2.4.
Here, we note that any tensor T can be decomposed as
T = TS + TA , (2.74)
where
TS = 12 (T + TT ) = TTS
TA = 12 (T − TT ) = −TTA (2.75)
are symmetric and antisymmetric tensors, respectively. These tensors have

the properties
tr TS = tr T, tr TA = 0. (2.76)
2.5. Tensor Invariants 33
Table 2.5. Formulas for determinant (n = dimension of the space)
det I = 1
det TT = det T
det(φT) = φn det T
det(T · U) = det T det U
∂ det T
= (det T) T−T
∂T
2.5.2 Determinant
The determinant of a tensor T is defined as the determinant of the matrix
of mixed components, i.e.,
det T ≡ det[T·ji ]. (2.77)
Table 2.5 contains some useful properties of the determinant. Equa-

tions (2.51), (2.65), (2.77), and Table 2.5 give
det T̄ = det[T̄·ji ] = det[Bki Alj T·lk ]
= det[Bki ] det[Alj ] det[T·lk ] = det B det A det T
= det A−T det A det T = det A−1 det A det T
= det(A−1 · A) det T = det I det T
= det T, (2.78)
which shows that the scalar det T is invariant under a change of basis.
2.5.3 Eigenvalues and Eigenvectors

In the eigenvalue problem for a tensor T, we wish to find a vector a such
that the transformation of a by T produces another vector in the direction
of a. Mathematically, this can be stated as
T · a = λa
or
(T − λI) · a = 0, (2.79)
where λ is the eigenvalue (principal value) and a is the eigenvector (prin-

cipal direction) of T. For a nontrivial solution (a = 0) in n-dimensional
space, we must have
det(T − λI) = 0, (2.80)
which gives the characteristic equation of order n to be solved for the

λi (i = 1, 2, . . ., n). With the λi known, Eq. (2.79) then provides the
corresponding ai . Since Eqs. (2.79) and (2.80) are geometrically invariant,
both λi and ai are invariants of the tensor T.
In mixed-component form, Eq. (2.79) can be written
(T·ji gi gj − λδ ij gi gj ) · (ak gk ) = (T·ji − λδ ij )gi δ jk ak = 0
or, for arbitrary gi ,
(T·ji − λδ ij )aj = 0. (2.81)
In three-dimensional space, expanding det[T·ji − λδ ij ] = 0 gives the charac-

teristic equation
−λ3 + I1 λ2 − I2 λ + I3 = 0 (2.82)
for the second-order tensor T, where
I1 = tr T
I2 = 12 [(tr T)2 − tr T2 ]
I3 = det T (2.83)
in which T2 = T · T. Because λ is invariant, it follows from Eq. (2.82) that

I1 , I2 , and I3 also are invariant, and they are called the principal invari-
ants of the tensor T. Of course, any function of these invariants is also
invariant. A useful relation is given by the Cayley-Hamilton theorem,
which states that a tensor satisfies its own characteristic equation, i.e.,
−T3 + I1 T2 − I2 T + I3 I = 0. (2.84)
2.6. Special Tensors 35
2.6 Special Tensors
2.6.1 Metric Tensor

The square of the length of a differential line element is
ds2 = dr · dr = (dxi gi ) · (dxj gj ) = gij dxi dxj , (2.85)
where Eq. (2.59) has been used and
gij ≡ gi · gj . (2.86)
The scalar invariant given by Eq. (2.85) is called the metric of the space,
and the gij characterize a particular coordinate system in the space.
The quantities gij have a useful mathematical interpretation. Since
gi gj : I = gi · I · gj = gi · gj = gij , Eq. (2.36)1 shows that the gij are
the covariant components of the identity tensor. For this reason, I also
is called the metric tensor. Like any second-order tensor, I has four
representations in any coordinate system:
I = gij gi gj = gij gi gj = g·j

i
gi gj = gj·i gj gi . (2.87)
Extracting the components of I gives the relations
gi · I · gj = gi · gj = gij
gi · I · gj = gi · gj = g·j
i
gj · I · gi = gj · gi = gj·i , (2.88)
which show that gij = gji, gij = gji, and g·ji

= gj·i . Thus, I = IT by
Eqs. (2.41). Moreover, comparison with Eq. (2.43) shows that
i
g·j = gj·i = δ ij . (2.89)
For a physical interpretation of the components of the metric tensor, we

dot I with the base vectors as follows:
gi = gi · I = gi · (gkj gk gj ) = gkj δ ki gj = gij gj

gi = gi · I = gi · (gkj gk gj ) = gkj δ ik gj = gij gj . (2.90)
These expressions show that gij is the component of gi in the direction of

gj , and gij is the component of gi along gj . Another useful relation is
given by
δ ji = gi · gj = (gik gk ) · (gjl gl ) = gik gjl δ kl
or
gik gjk = δ ji . (2.91)
In orthogonal curvilinear coordinates, Eqs. (2.88) show that gij = gij = 0

for i = j, and so Eq. (2.91) gives g(ii) = 1/g(ii). More generally, Eq. (2.91)
implies the matrix equation
[gik ][gjk ] = [δ ji ],
which provides the formula
[gij ] = [gij ]−1 . (2.92)
This relation can be confirmed rather easily by multiplying 2 × 2 matrices
consisting of metric components.
The covariant and contravariant components of I can be used to raise
and lower indices. One example of this is illustrated by Eqs. (2.90), where
the indices of the base vectors are raised or lowered. For vector components,
we find
ai = gi · a = gi · (aj gj ) = gij aj
ai = gi · a = gi · (aj gj ) = gij aj . (2.93)
ij
In general, raising an index requires a g and lowering an index requires a
gij . (Similar to contraction, the raised or lowered index also is replaced by
i or j.) This applies also to tensor components. For example,
T·ji = gi · T · gj = gi · (Tk·l gk gl ) · gj
= gik glj Tk·l , (2.94)
ik
where g raises one index, while glj lowers the other. Similarly, we get
Tij = gik glj T kl
T ij = gik glj Tkl
Tj·i = gik glj T·k
l
. (2.95)
Note that the components of the metric tensor I relate quantities in the
same coordinate system, while the components of the coordinate transfor-
mation tensors A and B relate quantities in different coordinate systems
[compare Eqs. (2.47) and (2.90)].
2.6.2 Permutation Tensor

According to Eq. (2.77), det T = det[T·ji ]. In this section, we write det T =
det[Tji ] for convenience, where Tji is the component in the ith row and the
jth column of the matrix; hence,
1
T1 T21 T31
2
T = det T = T1 T22 T32 . (2.96)
T3 T3 T3
1 2 3
Expanding the determinant in the usual manner shows that it can be writ-
ten
T = T1i T2j T3k eijk = Ti1 Tj2 Tk3 eijk (2.97)
or, equivalently,
epqr T = eijk Tpi Tqj Trk , epqr T = eijk Tip Tjq Tkr , (2.98)
where
⎧
⎨ +1 if (i, j, k) is an even permutation
eijk = eijk = −1 if (i, j, k) is an odd permutation (2.99)
⎩
0 if two or more indices are equal
are permutation symbols. (An even permutation also is called cyclic

and an odd permutation is anticyclic.)
Permutation symbols are useful in manipulations involving cross prod-
ucts. For example, consider the cross product of the unit vectors given
by
gi gi
ei = = √
|gi | g(ii)
gi gi
ei = = , (2.100)
|gi | g(ii)
√ √
where |gi | = g(i) · g(i) = g(ii) is the magnitude of the base vector gi ,
with a similar meaning for |gi |. Since e1 is orthogonal to e2 and e3 , we
have e2 × e3 = e1 , e3 × e2 = −e1 , e2 × e2 = 0, etc., or
ei × ej = eijk ek (2.101)
in general, which shows that
(ei × ej ) · ek = eijk . (2.102)

Extending this concept to general base vectors, we write

gi × gj = ijk g
k
(gi × gj ) · gk = ijk (2.103)

and
gi × gj = ijk
gk
(g × g ) · g =
i j k ijk
, (2.104)
ijk
where ijk and are components of the third-order permutation tensor
= ijk gigj gk = ijk

gi gj gk . (2.105)
In polyadic notation, Eqs. (2.103) and (2.104) can be written
gi × gj = gi gj :
.
(gi × gj ) · gk = gi gj gk .. (2.106)
and
gi × gj = gigj :
.
(gi × gj ) · gk = gi gj gk .. , (2.107)
as can be verified easily using Eq. (2.28) and its extension, the triple-dot
.
product given by abc .. def = (a · d)(b · e)(c · f ).
In Cartesian coordinates, ijk = eijk , but since the gi are not always
unit vectors, ijk = eijk in general. To find the relation between these
components, we let the unit vectors ei of Eqs. (2.100)–(2.102) be Cartesian
base vectors (ei = ei ) and set ei = ḡi and eijk = ¯ijk . Then, the permuta-
tion tensor can be written in terms of the Cartesian (barred) system and a
general (unbarred) system as
= eijk ei ej ek = ¯ijk ḡi ḡj ḡk = ijk gigj gk . (2.108)
Relative to the natural basis, the covariant components are
.
ijk = gi gj gk ..
.
= gi gj gk .. (¯lmn ḡl ḡm ḡn )
= (gi · ḡl )(gj · ḡm )(gk · ḡn ) ¯lmn
= Bil Bjm Bkn elmn , (2.109)
in which Eq. (2.50)2 has been used. This transformation and Eq. (2.106)1,
which shows that converts a dyad into a vector via the double-dot product,
demonstrate the tensor character of .
The relation (2.109) between ijk and eijk can be simplified as follows.
First, for (i, j, k) = (1, 2, 3), comparison of Eqs. (2.97) and (2.109) reveals
that 123 = det[Bji ], with a similar result for other even permutations of
the indices. (The sign changes for odd permutations.) Next, we evaluate
det[Bji ] = det B. In terms of unbarred and barred base vectors, the identity
tensor can be written as
I = gij gi gj = ḡij ḡi ḡj . (2.110)
Extracting the covariant components and using (2.50) yields
gij = I : gi gj = (ḡkl ḡk ḡl ) : gi gj
= ḡkl (ḡk · gi )(ḡl · gj )
= Bik Bjl ḡkl . (2.111)
Therefore,
det[gij ] = det[Bik ] det[Bjl ] det[ḡkl] = (det B)2 det[ḡkl ]
or
g = (det B)2 ḡ, (2.112)
where
g ≡ det[gij ], ḡ ≡ det[ḡij ]. (2.113)
Finally, since ḡ = 1 in Cartesian coordinates, Eq. (2.112) gives

√
det B = g. (2.114)
Because 123 = det B, 321 = − det B, etc., this result and a similar analysis
for ijk imply that (2.109) and its contravariant counterpart become
√
ijk = g eijk
ijk √
= eijk / g. (2.115)
√
Note that Eq. (2.103)2 gives (g1 × g2 ) · g3 = 123 = g, which is the same
√
g found in Eq. (2.17), i.e., it represents the volume of the parallelepiped
with the gi as edges. Equations (2.115), therefore, explain the square root
in Eq. (2.17).
Now, we list a couple of useful formulas involving the permutation ten-

sor. First, with Eqs. (2.103) and (2.106), the cross product of two vectors
can be written
a × b = (ai gi ) × (bj gj )
= ai b j ijk g
k
= ai bj gi gj : . (2.116)
Second, the -δ identity is (see Problem 2.7)
ijk
irs
= δ rj δ sk − δ sj δ rk . (2.117)
2.6.3 Orthogonal Tensors

Consider the transformation of two vectors a and b according to the rela-
tions
ā = Q · a, b̄ = Q · b, (2.118)
where Q is an orthogonal tensor defined by
QT · Q = Q · QT = I, (2.119)
i.e., QT = Q−1 . Then,

ā · b̄ = (Q · a) · (Q · b) = (a · QT ) · (Q · b)
= a · (QT · Q) · b = a · I · b
= a · b, (2.120)
which shows that the angle between the vectors ā and b̄ is the same as
that between a and b. Moreover, because det(QT · Q) = det QT det Q =
det Q2 = det I = 1, we have det Q = ±1. If det Q = |Q| = +1, Q is
called a proper orthogonal tensor, and Eq. (2.118) gives |ā| = |Q| |a| =
|a|, and similarly |b̄| = |b|. In this case, therefore, a and b retain their
lengths and relative orientations in space; thus, these vectors undergo a
rigid-body rotation due to the transformation by Q. For this reason, a
proper orthogonal tensor also can be called a rotation tensor.
Example 2.6 Consider rigid-body rotation about the z-axis of the Carte-
sian base vectors ex and ey into the new unit vectors ēx and ēy [Fig. 2.2
2.7. Physical Components 41
(page 9) with (e1 , e2 ) = (ex , ey ) and (ē1 , ē2 ) = (ēx , ēy )]. With this rotation
given by
ēx = Q · ex , ēy = Q · ey , ēz = Q · ez , (2.121)
determine the Cartesian components of Q.
Solution. In this relatively simple problem, we can write by inspection
Q = ēx ex + ēy ey + ēz ez , (2.122)
where
ēx = ex cos θ + ey sin θ
ēy = −ex sin θ + ey cos θ
ēz = ez . (2.123)
Comparing Eqs. (2.48)1 and (2.122) shows that this rotation can be consid-
ered a coordinate transformation from the unbarred to the barred system.
In matrix form, inserting Eqs. (2.123) into (2.122) yields
⎡ ⎤
cos θ − sin θ 0
Q(ei ej ) = ⎣ sin θ cos θ 0 ⎦ . (2.124)
0 0 1
Finally, Q · QT = exex + ey ey + ez ez = I can be easily verified.
2.7 Physical Components
In general curvilinear coordinates, the components of vectors and tensors do

not always have a direct physical meaning. If the base vectors have different
dimensions, for example, then the components have different dimensions,
since their combination must be consistent with the dimensions of the vector
or tensor. For instance, the velocity vector in cylindrical polar coordinates
can be written
v = v r gr + v θ gθ + v z gz ,
where gr = er , gθ = reθ , and gz = ez (see Example 2.1, page 10). Since
gr and gz are dimensionless, vr and vz have units of velocity, but vθ has
units of (time)−1 . Moreover, if dimensionless base vectors are not unit vec-
tors, the magnitudes of the components are not consistent with the physical
quantity actually observed. The physical components of a vector or ten-
sor are components that have physically meaningful units and magnitudes.
Sometimes it is convenient to derive the governing equations for a prob-
lem in terms of tensor components but to solve the problem using physical
components.
2.7.1 Physical Components of Vectors

The physical components of a vector a are the components of a relative to
a set of unit vectors. With the unit vectors ei defined by Eq. (2.100)1 , we
have
√
a = ai gi = ai g(ii) ei ≡ âi ei , (2.125)
where
√
âi = ai g(ii) (2.126)
are the physical components of a. Since the ei are dimensionless, the âi have
the same units as a. Moreover, âi is the length of the vector a(i)g(i) (see
Fig. 2.6, page 14). In terms of the covariant components of a, Eqs. (2.93)1
and (2.126) give
√
âi = gij aj g(ii) . (2.127)
2.7.2 Physical Components of Tensors

For second-order tensors, physical components can be defined in various
ways. Here, we base the definition on the mixed tensor components T·ji of
T, which were used to define the trace and determinant [Eqs. (2.72) and
(2.77)]. The T·ji can be transformed to or from other components using the
metric tensor to raise or lower indices.
Consider the transformation
b = T · a, (2.128)
which gives the components
bi = g i · b = g i · T · a
= gi · (T·lk gk gl ) · (aj gj )
= T·lk aj δ ik δ lj
= T·ji aj . (2.129)
We define the physical components T̂·ji of T through the analogous relation
b̂i ≡ T̂·ji âj , (2.130)
where Eq. (2.126) gives
√ √
âj = aj g(jj), b̂i = bi g(ii). (2.131)
2.7. Physical Components 43
Since b̂i and âj possess physically meaningful units, T̂·ji does also. Inserting
Eqs. (2.131) into (2.129) yields

i i g(ii) j
b̂ = T·j â ,
g(jj)
and comparison with (2.130) gives

g(ii)
T̂·ji ≡= T·ji . (2.132)
g(jj)
At this point, we make the following observations:
• In the matrix [T̂·ji ], the first index (i) denotes the row and the second
index (j) denotes the column. Interchanging rows and columns (making
j first and i second) gives the transpose matrix

T g(ii)
i ·i
T̂·j = Tj .
g(jj)
If the tensor T is symmetric, then Tj·i = T·ji by (2.41)3 and so

T
T̂·ji = T̂·ji ,
where Eq. (2.132) provides the right-hand side. In other words, if the
tensor T is symmetric, so is its matrix of physical components.
• Equations (2.132) give
T̂·ii = T·ii = tr T
det[T̂·ji ] = det[T·ji ] = det T. (2.133)
(The second relation is easy to show using a 2x2 matrix.) Hence, the
invariants given by Eq. (2.83) have the same values when expressed in
terms of either tensor or physical components.
• The physical components of T can be written in terms of the contravari-
ant components T ij by substituting T·ji = T ik gkj into Eq. (2.132). In
orthogonal coordinates, gkj = 0 for k = j, giving T·ji = T ij g(jj) for
i, j = 1, 2, 3 (not summed). In this case, (2.132) becomes

g(ii) √
T̂·j ≡= T g(jj)
i ij
= T ij g(ii) g(jj).
g(jj)
Note that, since this is not a tensor equation, the upper and lower
indices on the two sides of the equation need not match. In terms of
natural base vectors, T can be written
√ √
T = T ij gi gj = T ij g(ii) ei g(jj)ej ,
where Eq. (2.100)1 has been used. In orthogonal coordinates, therefore,
it is convenient to redefine the physical components of the second-order
tensor T as
√
T̂ ij = T ij g(ii) g(jj), (2.134)
which gives the dyadic

T = T̂ ij ei ej , (2.135)
in which the usual summation convention holds. This equation implies
that components of tensors relative to unit dyads in orthogonal curvi-
linear coordinates are physical components. Note also that T̂·ji = T̂ ij
in orthogonal coordinates.
These considerations make it easier to set up certain problems in terms of
physical components. Being components of a dyadic, the T̂ ij obey the rules
for tensor transformation (2.65), with the coordinate transformation tensors
computed using unit base vectors for orthogonal coordinates. On the other
hand, the T̂·ji do not obey these rules in general curvilinear coordinates.
2.8 Vector and Tensor Calculus
In Cartesian coordinates, differentiation of vectors and tensors with respect

to scalars is straightforward. Since the base vectors are constant, only differ-
entiation of the components is required. In general curvilinear coordinates,
however, the base vectors can change from point to point in space. Thus,
differentiation of base vectors may enter the analysis, complicating the sit-
uation. In addition, we often need to differentiate quantities with respect
to vectors and tensors. This section addresses these operations.
2.8.1 Differentiation with Respect to Spatial Coordinates

First, we consider differentiation of vectors and tensors with respect to
the coordinates xi . For any function φ(xi ), this operation is denoted by a
comma, i.e., ∂φ/∂xi ≡ φ,i .
2.8. Vector and Tensor Calculus 45
Differentiation of Base Vectors. Since the derivative gi ,j is a vector,

it can be resolved into components relative to the base vectors as
gi ,j = Γijk gk = Γkij gk , (2.136)
where the components Γijk and Γkij are called Christoffel symbols. It is
important to note that these quantities are not components of a third-order
tensor. Dotting this relation with the base vectors yields
gi ,j · gk = (Γijl gl ) · gk = Γijl δ lk
gi ,j · gk = (Γlij gl ) · gk = Γlij δ kl
or
Γijk = gi ,j · gk
Γkij = gi ,j · gk . (2.137)
Christoffel symbols appear whenever we differentiate general vectors and

tensors.
The derivative of a contravariant base vector can be found by differen-
tiating the relation
gi · gj = δ ji
k
with respect to x to get
gi ,k · gj + gi · gj ,k = 0.
Substituting Eq. (2.137)2 gives
gj ,k · gi = −Γjik (2.138)
or
gi ,j = −Γikj gk , (2.139)
which can be verified by dotting both sides of this relation with gl and
changing the indices.
Two useful properties of the Christoffel symbols warrant attention.
First, since gi = r,i , then gi ,j = r,ij = r,ji = gj ,i . Thus, from Eq. (2.136),
we find
Γijk = Γjik , Γkij = Γkji. (2.140)
Second, dotting (2.136) with gl gives

Γijk gk · gl = Γkij gk · gl
or, since gk · gl = δ kl and gk · gl = gkl ,

Γijl = Γkij gkl . (2.141)
l
Similarly, dotting both sides of (2.136) with g yields
Γlij = Γijk gkl . (2.142)
Hence, although Γijk and Γkij are not tensor components, the components
of the metric tensor can be used to raise or lower the third index.
A useful formula for computing the Christoffel symbols in a given coor-
dinate system can be found by differentiating the expression
gij = gi · gj
k
with respect to x to obtain
gij ,k = gi ,k · gj + gi · gj ,k
= Γikj + Γjki,
where Eq. (2.137)1 has been used. Permuting the subscripts and using
(2.140)1 yields the sequence of equations
gij ,k = Γkij + Γjki
gjk ,i = Γijk + Γkij
gki ,j = Γjki + Γijk . (2.143)
Now, adding the last two equations and subtracting the first gives
2Γijk = gjk ,i +gki ,j −gij ,k . (2.144)
Thus, if the metric for a coordinate system is known, this equation gives the
Γijk and then Eq. (2.142) gives the Γkij . In Cartesian coordinates, because
the base vectors are constant, all the Christoffel symbols are zero.
Example 2.7 Compute the Christoffel symbols for cylindrical polar coor-
dinates.
Solution. With (x1 , x2 , x3 ) = (r, θ, z), Eqs. (2.10) and (2.24) give the
base vectors
g1 = gr = ex cos θ + ey sin θ = er
g2 = gθ = r(−ex sin θ + ey cos θ) = reθ
g3 = gz = ez
g 1 = g r = er
g2 = gθ = r −1 eθ
g 3 = g z = ez , (2.145)
and the non-zero components of the metric tensor (gij = gi ·gj , gij = gi ·gj )
are
g11 = 1, g22 = r 2 , g33 = 1
g11 = 1, g22 = r −2 , g33 = 1. (2.146)
As the only non-zero derivative of the gij is g22 ,1 = 2r, the only non-zero
Christoffel symbols given by Eq. (2.144) are
Γ221 = − 12 g22 ,1 = −r
Γ122 = Γ212 = 12 g22 ,1 = r (2.147)
and from Γlij = Γijk gkl ,
Γ122 = Γ221 g11 = −r

Γ212 = Γ221 = Γ122 g22 = Γ212 g22 = r −1 . (2.148)
For a check, we now compute

gr ,θ = g1 ,2 = Γ12k gk = Γk12 gk
as given by (2.136). Using the above equations, we find
gr ,θ = Γ122 g2 = r(r −1 eθ ) = eθ
= Γ212 g2 = r −1 (reθ ) = eθ ,
which agrees with the result given by taking the derivative of Eq. (2.145)1
with respect to θ.
√
Example 2.8 Compute the derivative of g = g1 · (g2 × g3 ) with respect
to xi . Express the result in terms of g and Christoffel symbols.
Solution. Differentiating a product provides

√
( g),i = g1 ,i · (g2 × g3 ) + g1 · (g2 ,i ×g3 ) + g1 · (g2 × g3 ,i )
= g1 ,i · (g2 × g3 ) + g2 ,i · (g3 × g1 ) + g3 ,i · (g1 × g2 ),
where the vectors have been permuted according to the scalar triple product
formula in Table 2.1 (page 19). Next, substituting Eq. (2.136) gives
√
( g),i = Γj1i gj · (g2 × g3 ) + Γj2i gj · (g3 × g1 ) + Γj3i gj · (g1 × g2 ).
Since gj · (g2 × g3 ) = 0 only for j = 1, and so on for the other terms, this
expression can be written
√
( g),i = Γ11i g1 · (g2 × g3 ) + Γ22i g2 · (g3 × g1 ) + Γ33i g3 · (g1 × g2 )
√
= g (Γ11i + Γ22i + Γ33i )
or
√ √
( g),i = g Γjji. (2.149)
Differentiation of Vectors. Differentiating the vector a and using

Eqs. (2.136) and (2.139) gives
a,j = (ai gi ),j = ai ,j gi + ai gi ,j
= ai ,j gi + ai Γkij gk
= (ai gi ),j = ai ,j gi + ai gi ,j
= ai ,j gi − ai Γijk gk .
Interchanging the dummy indices i and k in the terms involving Christoffel
symbols yields
a,j = ai |j gi = ai |j gi , (2.150)
where
ai |j ≡ ai ,j +ak Γijk
ai |j ≡ ai ,j −ak Γkij (2.151)
are called the covariant derivatives of ai and ai , respectively. Later, we

will show that ai |j and ai |j are mixed and covariant components, respec-
tively, of a second-order tensor (Sec. 2.8.3). In Cartesian coordinates, the
vanishing of the Christoffel symbols implies that covariant and ordinary
differentiation are the same. In general curvilinear coordinates, however,
ai |j = 0 implies that the vector a is constant, but the components ai are
not (ai ,j = 0). (Consider, for example, a vector that translates in a po-
lar coordinate system.) This is consistent with the idea that a vector is
geometrically invariant.
From Eqs. (2.139), (2.150), and (2.151), the second derivative of the
vector a is
a,jk = (ai |j gi ),k = (ai |j ),k gi + (ai |j )gi ,k
= (ai ,j −al Γlij ),k gi − ai |j Γikl gl
= (ai ,jk −al ,k Γlij − al Γlij ,k −al |j Γlik )gi
= [ai ,jk −al ,k Γlij − al Γlij ,k −(al ,j −am Γm l
jl )Γik ]g
i
(2.152)
with a similar relation for a,jk = (ai |j gi ),k . Thus, taking higher derivatives
of a vector is straightforward, but labor-intensive. Symbolic manipulation
computer programs, however, can make tensor calculus virtually painless.
Differentiation of Tensors. Differentiation of tensors also is straightfor-

ward when they are expressed in the form of dyadics (or polyadics). For ex-
ample, for a second-order tensor in terms of mixed components, Eqs. (2.136)
and (2.139) give
T,k = (T.ji gi gj ),k
= T.ji ,k gi gj + T.ji gi ,k gj + T.ji gi gj ,k
= T.ji ,k gi gj + T.ji Γlik gl gj − T.ji gi Γjkl gl .
On interchanging dummy indices, these and similar computations involving
the other component representations of T produce
T,k = Tij |k gi gj = T ij |k gi gj = T.ji |k gi gj = Ti.j |k gi gj , (2.153)
where
Tij |k = Tij ,k −Tlj Γlik − Til Γljk

T ij |k = T ij ,k +T lj Γikl + T il Γjkl
T.ji |k = T.ji ,k +T.jl Γikl − T.li Γljk
Ti.j |k = Ti.j ,k −Tl.j Γlik + Ti.l Γjkl (2.154)
are covariant derivatives of the indicated components.

It is important to note that taking the ordinary derivative of a vector
or tensor is equivalent to taking the covariant derivative of the components
while holding the base vectors constant [Eqs. (2.150) and (2.153)]. We again
emphasize that a vanishing covariant derivative implies that the vector or
tensor is constant, but generally the components of the vector or tensor are
not constant [Eqs. (2.151) and (2.154)].
The tensor character of the covariant derivative can be used to derive

an interesting feature of the components of the special tensors I and . In
Cartesian coordinates, gij ,k = ijk ,l = 0, which can be written
gij |k = ijk |l =0 (2.155)
since the Christoffel symbols vanish (see also Problem 2.19). These last
equations are in tensor form and, therefore, hold in any coordinate system,
indicating that the tensors I and are constants in Euclidean space.
2.8.2 Differentiation with Respect to Vectors and Tensors

The derivative of a scalar function φ(a, T) with respect to the vector a is
defined by
∂φ
dφ = da · , (2.156)
∂a
which indicates that ∂φ/∂a is a vector. If the differential of a is written in

terms of its contravariant and covariant components as
da = gi dai = gi dai , (2.157)
then substitution into Eq. (2.156) shows that
∂ ∂ ∂
= g i i = gi . (2.158)
∂a ∂a ∂ai
The first of these representations is verified by
∂φ ∂φ
dφ = da · = (gi dai ) · gj j
∂a ∂a
∂φ ∂φ i
= dai δ ji = da = dφ,
∂aj ∂ai
where the chain rule for differentiation has been used. The second of (2.158)
can be confirmed similarly.
Analogously, the derivative of φ with respect to the second-order tensor
T is defined by
∂φ
dφ = dT : . (2.159)
∂T
Differentiation with respect to higher-order tensors follows by adding more
dots. If the differential of T is written, for example, in the form
dT = dT ij gi gj , (2.160)
comparing these equations gives

∂ ∂
= gi gj ij (2.161)
∂T ∂T
as confirmed by
∂φ ∂φ
dφ = dT : = (dT ij gi gj ) : gkgl
∂T ∂T kl
∂φ ∂φ
= dT ij (gi · gk )(gj · gl ) = dT ij kl δ ki δ lj
∂T kl ∂T
∂φ
= dT ij = dφ.
∂T ij
Equations (2.158) and (2.161) indicate that the component form of a
derivative operator containing a vector or tensor is obtained by moving the
base vectors of the differential to the numerator and raising or lowering
their indices. In summary, therefore, we can write
∂ ∂ ∂
= g i i = gi
∂a ∂a ∂ai
∂ ∂ ∂ ∂ ∂
= gi gj ij = gi gj = gi gj i = gi gj .j . (2.162)
∂T ∂T ∂Tij ∂T.j ∂Ti
Note that the base vectors and the components in each of these expressions
have the same variance.
The above observation is useful in differentiating vectors and tensors
with respect to other vectors and tensors. The basic definitions, which
follow from Eqs. (2.156) and (2.159), are
∂b ∂b
db = da · + dT :
∂a ∂T
∂U ∂U
dU = da · + dT : (2.163)
∂a ∂T
for b = b(a, T) and U = U(a, T). In these expressions, the order of the
factors is important, e.g.,
∂b ∂ ∂ ∂b
= (b) = (b) = gi gj ij . (2.164)
∂T ∂T gi gj ∂T ij ∂T
2.8.3 Gradient Operator

The gradient (“del”) operator ∇ is defined by
dφ = dr · ∇φ, (2.165)
where dφ is the differential of the scalar φ in the direction of the differential

position vector dr = gi dxi . Comparison with Eqs. (2.156)–(2.158) shows
that
∂ ∂
∇≡ = gi i , (2.166)
∂r ∂x
with the invariance of ∇ indicated by the form ∂/∂r. This invariance also
can be demonstrated by the manipulation
∂ ∂ ∂
∇ = gi = δ ij gj i = (Aik Bjk )gj i
∂xi ∂x ∂x
∂xi ∂
= (Bjk gj )
∂ x̄k ∂xi
∂ ∂
= ḡk k
= ḡi i ,
∂ x̄ ∂ x̄
which involves Eqs. (2.47), (2.52), and (2.60) and the chain rule for dif-
ferentiation. Thus, ∇ has the same indicial form in both the xi and x̄i
coordinate systems. Equation (2.165) also holds if φ is replaced by a vector
or a tensor.
Example 2.9 Using the results from Example 2.3 (page 16), write the
gradient operator in Cartesian and cylindrical polar coordinates.
Solution. With (x1 , x2 , x3 ) = (x, y, z) and Eqs. (2.23), Eq. (2.166) gives
∂ ∂ ∂
∇ = ex + ey + ez (2.167)
∂x ∂y ∂z
in Cartesian coordinates. Similarly, with (x1 , x2 , x3 ) = (r, θ, z) and
Eq. (2.24), we get
∂ eθ ∂ ∂
∇ = er + + ez (2.168)
∂r r ∂θ ∂z
in cylindrical coordinates.
Gradient, Divergence, and Curl. Using the representation for ∇

given by Eq. (2.166), we can compute the gradient, divergence, and curl of
a function for any coordinate system. The gradient of a scalar function φ
is
∂φ
∇φ = gi i = gi φ,i
∂x
or
∇φ = gi φ|i (2.169)
since ordinary and covariant differentiation of a scalar are the same. The
gradient of vector a is
∂a
∇a = gj j = gj a,j ,
∂x
and substituting Eqs. (2.150) yields
∇a = ai |j gj gi = ai |j gj gi , (2.170)
where care has been taken to maintain the order of the base vectors. This
expression shows that the covariant derivatives ai |j and ai |j are components
of the second-order tensor ∇a.
The divergence of vector a, defined by ∇ · a, can be obtained by a
contraction of the gradient, i.e., by inserting a dot between the base vectors
of (2.170). Hence,
∇ · a = ai |j gj · gi = ai |j δ ji = ai |i
= ai |j gj · gi = ai |j gij = ai |i
or
∇ · a = a i | i = ai | i , (2.171)
where the contravariant components of the metric tensor have been used
to raise one of the indices on the tensor component ai |j . The form of ai |i
appears to indicate contravariant differentiation, but this operation is of
limited use since the components dxi of dr = gi dxi needed for ∂/∂xi are
not along the coordinate curves in general. The components ai |i , therefore,
are considered shorthand for ai |j gij .
The curl of vector a is given by
∂
∇ × a = gj j × a = gj × a,j
∂x
= ai | j g j × g I ,
where Eq. (2.150) has been used. With Eqs. (2.104)1 and (2.107)1, this
expression assumes the forms
∇ × a = ai | j g j g i :
= ai | j jik
g k = aj | i ijk
gk . (2.172)
Second Derivative. If we let

T = (∇a)T = ai |j gi gj (2.173)
from Eq. (2.170), then Tij = ai |j , and (2.154)1 gives
Tij |k = (ai |j )|k ≡ ai |jk
= ai |j ,k −al |j Γlik − ai |l Γljk . (2.174)
With Eq. (2.151), this relation becomes
ai |jk = (ai ,j −al Γlij ),k −(al ,j −am Γm

lj )Γik − (ai ,l −am Γil )Γjk ,
l m l
(2.175)
which is the second covariant derivative of ai . It is important to note,

by comparing Eqs. (2.152) and (2.175), that a,jk = ai |jk gi in general. The
difference is the last term in Eq. (2.175), which comes from differentiating
the second base vector in the dyadic representation of (2.173).
The second covariant derivatives of Eq. (2.175) represent the compo-
nents of a third-order tensor. This can be shown by combining Eqs. (2.153),
(2.166), and (2.170) to produce the curvature tensor13
∂
∇(∇a)T = gk (ai |j gi gj ) = ai |jk gk gi gj
∂xk
or
∇(∇a)T = aj |kigi gj gk . (2.176)
This tensor has important implications that depend on the geometry of the
space. For example, interchanging j and k in Eq. (2.175) to produce ai |kj
and using (2.140) gives
ai |jk − ai |kj = am (Γm

ik ,j −Γij ,k +Γlj Γik − Γlk Γij )
m m l m l
≡ am R m
.ijk , (2.177)
13 The name curvature tensor stems from the physical meaning of the second derivative.
where Rm .ijk are components of the Riemann-Christoffel tensor. In

Cartesian coordinates, since the Christoffel symbols are zero, Rm .ijk = 0, and
so ai |jk = ai |kj . The tensor character of Rm .ijk indicates that this relation
holds in all coordinate systems in three-dimensional (flat) Euclidean space,
and thus the order of covariant differentiation is interchangeable. On the
two-dimensional space of a curved surface, however, a Cartesian system is
not possible, and so Rm .ijk = 0 in general. Hence, covariant differentiation is
not always interchangeable on a curved surface.14 The Riemann-Christoffel
tensor is important in shell theory (Flugge, 1972), as well as the general
theory of relativity (Einstein, 1916).
Next, we compute the Laplacian of a scalar φ by
∇2 φ ≡ ∇ · ∇φ. (2.178)
Application of Eqs. (2.169) and (2.171) yields
∇2 φ = ∇ · (gi φ|i ) = φ|i |i

or
∇2 φ = φ|ii = φ|ij gij . (2.179)
Substituting Eqs. (2.139) and (2.166) into (2.178) gives the explicit form
(see Problem 2.16)
∇2 φ = gij φ,ij −Γjki gik φ,j . (2.180)
Operations with the Gradient Operator. With polyadic analysis,

many operations involving the gradient operator are straightforward. For
example, Eqs. (2.153) and (2.166) yield the gradient of the tensor T =
Tij gi gj as
∇T = gk T,k = Tij |k gk gi gj (2.181)
with Eq. (2.154)1 providing Tij |k . This equation shows that the covariant
derivatives of the second-order tensor components Tij are components of
the third-order tensor ∇T.
Other operations can be a little tricky, however. Consider, for instance,
the gradient of the dot product a · b. Directly differentiating the product
14 A curved surface formed by curling an initially flat surface is an exception that allows
interchangeable differentiation. This type of surface is called developable.

gives ∇(a · b) = (∇a) · b + a · (∇b), which happens to be incorrect. The

correct relation can be derived formally by first taking the differential
d(a · b) = (da) · b + a · (db). (2.182)
Next, using Eq. (2.165) and treating ∇ as a vector gives
d(a · b) = (dr · ∇a) · b + a · (dr · ∇b)

= (dr · ∇a) · b + (dr · ∇b) · a
= dr · (∇a) · b + dr · (∇b) · a
= dr · [(∇a) · b + (∇b) · a]
≡ dr · ∇(a · b), (2.183)
and thus
∇(a · b) = (∇a) · b + (∇b) · a. (2.184)
Note that we have used the fact that dr · ∇b is a vector in the second line
of (2.183).
Another method, which often is easier to apply, involves using the com-
ponent form of ∇ given by Eq. (2.166), which we write in shorthand form
as (Drew, 1961)
∂
∇ = gi ≡ g ∂. (2.185)
∂xi
In manipulating equations with this expression for ∇, we must take care to

keep the derivatives on the proper terms and to preserve the order of vector
products. Thus, ∂ must remain in front of its proper operand at all times,
but we can move ∂ (not g) across dots and crosses, e.g., a · ∂b = a∂ · b.
Note also that, since ∂ is simply a scalar differentiator, products can be
differentiated without ambiguity. Now,
∇(a · b) = g ∂(a · b)
= g[(∂a) · b + a · (∂b)]
= g[(∂a) · b + (∂b) · a]
= (g ∂a) · b + (g ∂b) · a
= (∇a) · b + (∇b) · a, (2.186)
which agrees with Eq. (2.184).

Example 2.10 Compute the gradient, divergence, and curl of the dyad
ab.
Solution. With care taken to preserve order, the gradient is
∇(ab) = (g ∂)(ab)
= g[(∂a)b + a(∂b)]
= (g ∂a)b + ga ∂b.
= (∇a)b + (a∇)T b. (2.187)
Similar manipulations yield the divergence as
∇ · (ab) = (g ∂) · (ab)
= g · [(∂a)b + a(∂b)]
= (g ∂) · ab + g · a ∂b
= (g ∂ · a)b + a · g ∂b
= (∇ · a)b + a · (∇b), (2.188)
and the curl is
∇ × (ab) = (g ∂) × (ab)
= g × [(∂a)b + a(∂b)]
= (g ∂ × a)b + g × a ∂b
= (g ∂ × a)b − a × (g ∂b)
= (∇ × a)b − a × (∇b). (2.189)
Example 2.11 Compute the divergence of T · a.

Solution. Substituting Eq. (2.185) yields
∇ · (T · a) = (g ∂) · (T · a)
= g · [(∂T) · a + T · (∂a)]
= (g ∂) · T · a + g · T · (∂a)
= (∇ · T) · a + T : g ∂a
= (∇ · T) · a + T : (∇a), (2.190)
where the relation a·T·b = T : ab has been used. Other formulas involving
the gradient operator are given in Tables 2.6–2.8.
Table 2.6. Gradient formulas
∇(φψ) = (∇φ)ψ + φ(∇ψ)

∇(φa) = (∇φ)a + φ(∇a)
∇(a · b) = (∇a) · b + (∇b) · a
∇(ab) = (∇a)b + (a∇)T b
∇(T · a) = (∇T) · a + (∇a) · TT
∇r = I
∇(r · r) = 2r
Table 2.7. Divergence formulas
∇ · a = tr ∇a
∇ · (φa) = (∇φ) · a + φ(∇ · a)
∇ · (ab) = (∇ · a)b + a · (∇b)
∇ · (T · a) = (∇ · T) · a + T : ∇a
∇ · (a × b) = (∇ × a) · b − a · (∇ × b)
∇ · (∇ × a) = 0
∇·r = 3
2.8.4 Integral Relations

In continuum mechanics, we often need to convert volume integrals into
surface integrals, surface integrals into line integrals, and vice versa. This
section considers three useful theorems for these conversions. Although
these relations can be derived in general curvilinear coordinates, it is simpler
to derive them in Cartesian coordinates, as we do here.15 The results then
are written in direct notation for later specialization to any coordinate
system.
15 Since we are working in Cartesian coordinates, we do not distinguish between sub-
scripts and superscripts in this section.
Table 2.8. Curl formulas
∇ × (φa) = (∇φ) × a + φ(∇ × a)

∇ × (ab) = (∇ × a)b − a × (∇b)
∇ × (a × b) = a(∇ · b) + b · (∇a) − a · (∇b) − b(∇ · a)
∇ × (∇ × a) = ∇(∇ · a) − ∇2 a
∇ × (∇φ) = 0
∇×r = 0
Gradient Theorem. For simplicity, we consider a two-dimensional re-

gion A bounded by a contour C, such that any line through A parallel to
either one of the coordinate axes intersects C in only two points (Fig. 2.7).
The curve C is divided by its leftmost and rightmost points (x = a, b) into
a lower segment C1 , described by y = y1 (x), and an upper segment C2 ,
described by y = y2 (x). With the position vector to point P on C given by
y
C2
y2(x)
C1 P
r t
n y1(x)
0 x
a b
Fig. 2.7. Planar region A enclosed by curve C = C1 + C2 .
r = xex + yey , (2.191)

the unit tangent (“base”) vector at P is [see Eq. (2.7)]
dr dx dy
t= = ex + ey , (2.192)
ds ds ds
where ds is the differential length along C, and the unit normal vector is
dy dx
n = t × ez = ex − e y = nx e x + ny e y . (2.193)
ds ds
For a function φ(x, y) defined in A, consider the area integral
b y2 (x)
∂φ ∂φ
dA = dy dx
A ∂y a y1 (x) ∂y
b
= {φ[x, y2 (x)] − φ[x, y1(x)]} dx
a
b
= {[φ]C2 − [φ]C1 } dx. (2.194)
a
As indicated in Fig. 2.7, a positive contour integration corresponds to a

counterclockwise traversal of C. To make the first integral in Eq. (2.194)
consistent with this convention, we write
a b
∂φ
dA = − [φ]C2 dx − [φ]C1 dx
A ∂y
b a
dx
=− φ dx = − φ ds.
C C ds
Finally, using Eq. (2.193) yields

∂φ
dA = φ ny ds (2.195)
A ∂y C
and a similar analysis gives

∂φ
dA = φ nx ds. (2.196)
A ∂x C
Equations (2.195) and (2.196) are the basic relations used to develop the
integral theorems.
Multiplying Eqs. (2.195) and (2.196) by ey and ex , respectively, and
adding provides

∂φ ∂φ
(ex + ey ) dA = φ (nx ex + ny ey ) ds
A ∂x ∂y C
∂ ∂
or, since ∇ = ex + ey in two-dimensional Cartesian coordinates,
∂x ∂y

∇φ dA = nφ ds (2.197)
A C
with n given by Eq. (2.193). This relation, which is written in direct no-
tation, is called the gradient theorem or Green’s theorem. In three
dimensions, it becomes

∇φ dV = nφ dA, (2.198)
V A
where V is the volume enclosed by a surface of area A. Since it is written

in coordinate-free form, this equation is valid for any coordinate system.
Divergence Theorem. Consider a vector a = ax ex + ay ey . If we let

φ = ax in Eq. (2.196) and φ = ay in Eq. (2.195) and add the results, we
obtain

∂ax ∂ay
+ dA = (ax nx + ay ny ) ds.
A ∂x ∂y C
In direct notation, this equation is

∇ · a dA = n · a ds, (2.199)
A C
which is called the divergence theorem or Gauss’s theorem. In three
dimensions, this relation becomes

∇ · a dV = n · a dA. (2.200)
V A
Curl Theorem. If we set φ = ay in Eq. (2.196) and φ = ax in (2.195)

and subtract the results, we get

∂ay ∂ax
− dA = (ay nx − ax ny ) ds.
A ∂x ∂y C
This equation can be written

∇ × a dA = n × a ds (2.201)
A C
or, in three dimensions,

∇ × a dV = n × a dA, (2.202)
V A
which is called the curl theorem.

In summary, Eqs. (2.198), (2.200), and (2.202) can be written in the

form

∇ ∗ φ dV = n ∗ φ dA, (2.203)
V A
where ∗ is a blank, a dot, or a cross; and φ can be a scalar, a vector, or

a tensor, so long as the expression is meaningful. (For example, ∇ × φ is
meaningless if φ is a scalar.)
2.9 Problems
Unless stated otherwise, work the following problems using base vectors for
general curvilinear coordinates (gi and gi ). Lower- and upper-case bold
letters represent vectors and second-order tensors, respectively.
2.1(a) Evaluate δ ij δ ji .
(b) Show that mki jki = 2δjm . (Hint: Try using the -δ identity.)
(c) Simplify the expression ijk ai aj as much as possible.
2.2 Let
a = 3ex + ey + ez
b = ex + 2ez
c = ex − 3ey + ez ,
where the ei are Cartesian base vectors. Compute (ab) · c and (ba) · c.
2.3 Expand and simplify as much as possible the derivative
∂ i j
xx .
∂xi
2.4 For the vectors a and b and second-order tensors A and B, show the
following:
(a) (a × b) · a = 0
(b) I : A =trA
(c) A : B = 0 if A is symmetric and B is antisymmetric
(d) (AT )−1 = (A−1 )T
−1
(e) (A · B) = B−1 · A−1
2.5 If T and U are second-order tensors, write A = T · U in terms of
the following dyadic bases: (a) gi gj , (b) gi gj , and (c) gi gj . Find two
dyadic representations for each basis in terms of components of T and
U only.
2.9. Problems 63
T
2.6 Use dyadic analysis to show T : (U · V) = U : (T · V ).
2.7 Use the vector formula
a × (b × c) = (a · c)b − (a · b)c
to derive the -δ identity
ijk
irs
= δ rj δ sk − δ sj δ rk .
Hint: Working in general curvilinear coordinates, set a = gi , b = gj ,
and c = gk .
2.8 Cylindrical polar coordinates (r, θ, z) and Cartesian coordinates (x, y, z)
are related by
x = r cos θ
y = r sin θ
z = z.
(a) Determine the components of the tensors A and B for a transfor-
mation from Cartesian to cylindrical coordinates using Eqs. (2.60)
and (2.61). Write both tensors in matrix form. Hint: Be careful
when assigning rows and columns.
(b) Determine B using Eq. (2.48)2.
2.9 Spherical polar coordinates (r, θ, φ) and Cartesian coordinates (x, y, z)
are related by
x = r sin θ cos φ
y = r sin θ sin φ
z = r cos θ.
(a) For a transformation from Cartesian to spherical coordinates, use
Eq. (2.60) to find the coordinate transformation components Aji .
(b) Show that ḡi = A · gi , where ḡi = (gr , gθ , gφ), gi = (ex , ey , ez ),
and A is the coordinate transformation tensor.
(c) Relative to spherical base vectors, write the physical components
of the second-order tensor T in terms of the components T·ji and
in terms of T ij .
2.10 Parabolic cylindrical coordinates (u, v, z) and Cartesian coordinates
(x, y, z) are related by
x = 21 (u2 − v2 )
y = uv
z = z.
(a) Plot curves for constant values of u and constant values of v in the
xy-plane. Plot for u, v = 1, 2, and 3. Hint: Find equations for
x(y; u) and x(y; v).
(b) Determine the base vectors ḡi and ḡi (i = u, v, w) for parabolic
coordinates in terms of Cartesian base vectors.
(c) Compute the matrices [ḡij ] and [ḡij ]. Are the parabolic coordinates
orthogonal?
(d) For transformation from Cartesian to parabolic coordinates, deter-
mine the coordinate transformation matrices [Aij ] and [Bji ] using
Eqs. (2.50).
(e) Determine [Aij ] using Eq. (2.60) and [Bji ] using (2.51).
2.11 If T, U, and V are orthogonal tensors, show that T · U · V is an or-
thogonal tensor.
2.12 Consider the vector
a = 2ex − 5ey + 3ez
and the base vectors
g1 = ex − 2ey + ez
g2 = ex − ez
g3 = 3ex − 2ey + 2ez ,
where the ei are Cartesian base vectors.
(a) Compute the gi and the natural components ai and ai of a.
(b) Suppose a tensor T has the matrix representation
⎡ ⎤
−2 5 1
T(ei ej ) = ⎣ 3 0 −2 ⎦ .
1 −1 4
Compute the principal invariants of T and the natural components
T 11 , T 23 , T22 , T13 , T2·1 , and T·21 .
2.13 Consider the eigenvalue problem for the second-order tensor
T = 2g1 g1 + g2 g2 − 5g3 g3 + 3g2 g3 + 3g3 g2 .
(a) Determine the characteristic equation for T and compute its eigen-
values.
(b) Determine the eigenvector corresponding to each eigenvalue. Show
that the eigenvectors are mutually orthogonal.
2.14 If A, B, and C are second-order tensors, show the following:
2.9. Problems 65
∂
(a) trA = I
∂A
∂
(b) trA2 = 2AT
∂A
∂ ∂A ∂B
(c) (A : B) = :B+A:
∂C ∂C ∂C
2.15 Determine the Christoffel symbols Γkij for the following coordinate sys-
tems:
(a) Spherical polar coordinates (see Problem 2.9). Then, differentiate
the base vectors gr , gθ , and gφ with respect to θ and φ by taking
straight derivatives of the gi , as well as using the Γkij .
(b) Parabolic cylindrical coordinates (Problem 2.10). Then, differenti-
ate the base vectors gu and gv with respect to u and v by taking
straight derivatives of the gi , as well as using the Γkij .
2.16 Consider the Laplacian operator defined by ∇2 = ∇ · ∇.
(a) Use Eqs. (2.139) and (2.166) to show that the Laplacian of the
scalar function σ(xi ) can be written in the form
∇2 σ = gij σ,ij −Γjki gik σ,j .
(b) Use the equation in part (a) to derive the explicit form for ∇2 σ in
toroidal coordinates, where σ depends only on r and φ (axisymme-
try, see Appendix B).
2.17 Consider the following vectors in cylindrical polar coordinates (r, θ, z):
sin θ
a = cos θ gr − gθ + z gz
r
1 1
b= gr + 2 gθ .
r r
Using Eq. (2.150)1, compute the derivative of each vector with respect
to r and with respect to θ. Express the results in terms of the unit base
vectors er , eθ , and ez .
2.18 Derive the expression
T ij |k = T ij ,k +T lj Γikl + T il Γjkl .
2.19 Working in general curvilinear coordinates, prove the relation gij |k = 0.

2.20 For the scalar function φ(r, θ, z), write ∇2 φ in cylindrical polar coordi-
nates using (a) Eq. (2.178) and (b) Eq. (2.180).
2.21 If a and b are vector functions and T is a second-order tensor function,

use the shorthand notation ∇ = g∂ to show the following:
T
(a) ∇(T · a) = (∇T) · a + (∇a) · T
(b) ∇ · (a × b) = (∇ × a) · b − a · (∇ × b)
(c) ∇ · (∇ · aI) = ∇(∇ · a)
Chapter 3
Analysis of Deformation
The mechanics of hard tissues, such as bones, teeth, and horns, can be ana-
lyzed using the linear theory of elasticity, in which deformation is assumed
to be “small.” Under this condition, the distinction between the geometries
of the undeformed and the deformed body can be ignored. Soft tissues,
in contrast, often undergo “large” (finite) deformation. In this case, geo-
metric nonlinearity enters the analysis, even if the material properties
(stress-strain relations) are linear. This chapter deals only with geometry.
Nonlinear stress-strain relations characterize material nonlinearity, as
discussed in Chap. 5.
One manifestation of geometric nonlinearity is nonlinear strain-displace-
ment relations. Another can be seen through the example of a cantilever
beam of length L with an end load P (Fig. 3.1). For small deflection, the
peak bending moment is approximately P L at the support. Classical lin-
ear beam theory is based on this assumption. When the deflection is large,
however, the load moves toward the support by a significant distance, re-
ducing the bending moment. Throughout most of this chapter, we make no
restrictions on the magnitudes of deformation (i.e., strain), displacement, or
rotation. Approximations for various restrictions are examined in Sec. 3.9.
P
L
Fig. 3.1. Geometric nonlinearity in a cantilever beam.
67
68 Analysis of Deformation
Deformation of a body is measured relative to a reference configu-

ration, which may or may not be stress free. In fact, most unloaded soft
biological structures contain residual stress (Fung, 1990). Still, it often is
convenient to choose the unloaded state as the reference configuration, even
if it contains residual stress. In this chapter, unless stated otherwise, we
refer to the reference configuration as the undeformed configuration, keep-
ing in mind that the term “undeformed” is relative. The term “deformed”
then refers to any other configuration.
When possible, we use upper-case letters to represent Lagrangian
quantities, referred to the undeformed configuration, and lower-case letters
to represent Eulerian quantities referred to the deformed configuration.
This convention also applies to indices, as we distinguish between I and i,
so that aiI = a11 + a22 + a33 . In addition, note that some upper-case bold
letters may be vectors, rather than tensors, a change from the convention
in the previous chapter.
3.1 Deformation in One Dimension
To illustrate the basic measures used to describe the deformation of a solid

body, we first consider uniaxial stretching of a bar (Fig. 3.2). We choose the
unloaded bar of length L as the reference configuration. When subjected to
end forces, the length of the bar changes to = L + ΔL, and the classical
engineering strain or the stretch is defined as
ΔL −L
E∗ = = = − 1. (3.1)
L L L
In addition,
λ= = 1 + E∗ (3.2)
L
is called the stretch ratio, being the ratio of the deformed to the unde-
formed (reference) length.
The deformation measures defined above are valid regardless of the mag-
nitude of deformation. When deformation is large, however, it generally is
more convenient to define strain measures based on changes in squared
length, rather than changes in length. A commonly used measure for large
deformation is the Lagrangian strain, which is defined to be
2
− L2
E= (3.3)
2L2
for uniaxial stretching of a bar. In terms of E ∗ , this relation can be written
E = E ∗ (1 + 21 E ∗ ), (3.4)
3.2. Coordinate Systems and Base Vectors 69
l = L + DL
Fig. 3.2. Uniaxial stretching of a bar.
which implies
λ = 1 + E ∗ = (1 + 2E)1/2
E = 12 (λ2 − 1). (3.5)
Equation (3.4) can be verified by substituting (3.1).
The Lagrangian strain is referred to the undeformed configuration
(length L of the bar). In some formulations, it is useful to define a strain
measure relative to the deformed configuration. For the one-dimensional
case, the Eulerian strain is defined as
2
− L2
e= . (3.6)
2 2
In terms of the other strain measures, this relation can be written
E E ∗ (1 + 12 E ∗ )
e = 12 (1 − λ−2 ) = = , (3.7)
1 + 2E (1 + E ∗ )2
which can be confirmed by substituting Eqs. (3.2), (3.4), and (3.5).
Any one of the three strain measures (E ∗ , E, e) can be used to charac-
terize a given deformation, although they differ numerically. If the defor-
mation is small (E ∗ << 1), however, then Eqs. (3.4) and (3.7) show that
E ∼= e ∼= E ∗ , i.e., all the strain measures are essentially equivalent. The
development in the following sections extends these ideas to three dimen-
sions.
3.2 Coordinate Systems and Base Vectors
Consider a deformation that sends a point P in the undeformed body B at

time t = 0 into the point p in the deformed body b at time t (Fig. 3.3). The
point P is located at the material, or Lagrangian, curvilinear coordinates
X I , and p is located at the spatial, or Eulerian, curvilinear coordinates
xi . In general, the coordinate systems X I and xi may be attached to two

different reference frames, O and o, which may be in motion relative to
each other (Fig. 3.3).
b
B x3 g3
g2
G3
G2 x2
X2 u
X3 p g1
x1
P X1 r
G1
R
b o
Fig. 3.3. Coordinates and base vectors in undeformed body B and deformed body b.
Some authors [e.g., Eringen (1962; 1980)] prefer to keep the X I and
i
x coordinate systems independent of each other. In this way, the coor-
dinates can be chosen to take advantage of any symmetries in B and b.
Other authors [e.g., Green and Zerna (1968)] assume that the X I system
is imbedded in B so that it deforms with the body, and the xi system is
then taken to coincide with the X I system at all times during the defor-
mation. During the motion, each point keeps the same coordinate label,
and xi = X I are called convected coordinates. Although the flexibility
in choosing the deformed coordinate system is lost in this case, the basic
equations assume a somewhat simpler form. In this chapter, we consider
both choices.
Let R(X I , t) and r(xi , t) be the position vectors to P and p, respectively.
Then, the mapping of P into p and the inverse mapping of p into P are
described by
r = r(R, t), R = R(r, t), (3.8)
respectively, which imply
r = r(xi , t) = r(X I , t), R = R(X I , t) = R(xi , t). (3.9)

3.2. Coordinate Systems and Base Vectors 71
The covariant base vectors in B are

∂R
GI = = R,I (3.10)
∂X I
and, in b, two sets of covariant base vectors are defined by
∂r
gI = = r,I
∂X I
∂r
gi = = r,i . (3.11)
∂xi
The GI are tangent to the X I coordinate curves at P , the gi are tangent

to the xi coordinate curves at p (Fig. 3.3), and the gI are tangent to the
convected X I coordinate curves at p.
Before proceeding further, we need to introduce a new notation. Since
there are no upper- or lower-case numerical indices, we use an asterisk
to denote an “upper-case number.” This allows us to distinguish between
gI = {g1∗ , g2∗ , g3∗ } and gi = {g1 , g2, g3 }. (The gi are illustrated in
Fig. 3.3.) If there is no confusion, however, the asterisk is omitted, i.e.,
GI = {G1 , G2 , G3 }. Moreover, a similar confusion may arise when vector
and tensor components are referred to the bases GI and gI . Here again,
and following Atluri (1984), we attach asterisks to indices associated with
gI or gI . For example, we write
T = TIJ GI GJ = TI ∗ J ∗ gI gJ = TI ∗ J gI GJ .
The identity tensor, however, is written
I = GIJ GI GJ = gIJ gI gJ = gij gi gj (3.12)
since there is no confusion between GIJ and gIJ . Finally, note that the
summation convention applies to repeated indices even if only one of them
contains an asterisk, e.g., aIJI ∗ = a1J1∗ + a2J2∗ + a3J3∗ .
Relations can be found that link the two sets of base vectors in b. Equa-
tions (3.10) and (3.11) give the differential length vectors
dR = GI dX I
dr = gI dX I = gi dxi . (3.13)
If xi = X I (convected coordinates), then gi = gI . Otherwise, Eq. (3.13)2

gives
gI = FIi gi , gi = FiI gI , (3.14)

where
∂xi ∂X I
FIi ≡ , FiI ≡ . (3.15)
∂X I ∂xi
These last two sets of equations can be considered a coordinate transfor-

mation between the xi and the X I coordinates, as given by Eqs. (2.47)1
and (2.60). In this regard, using Eqs. (2.47)2 and (2.61), we also can write
gI = FiI gi , gi = FIi gI . (3.16)
Moreover, combining these relations gives

δ JI = gI · gJ = (FIi gi ) · (FjJ gj ) = FIi FjJ δ ji
δ ji = gi · gj = (FiI gI ) · (FJj gJ ) = FiI FJj δ JI
or
FIi FiJ = δ JI , FiI FIj = δ ji , (3.17)
which also can be found from Eqs. (3.15) and the chain rule for differenti-
ation, or from Eq. (2.52). Physical meanings for FIi and FiI are discussed
in the next section.
3.3 Deformation Gradient Tensor
In taking gradients, we must distinguish between gradients with respect to

the undeformed and deformed configurations. With the gradient operator
being ∇ in B and ∇ ¯ in b, Eqs. (2.158), (2.166), and (3.13) yield
∂ ∂
∇≡ = GI
∂R ∂X I
¯ ≡ ∂ = gi ∂ = gI ∂
∇ (3.18)
∂r ∂xi ∂X I
since ∂/(GI ∂X I ) = GI ∂/∂X I , etc.

The deformation gradient tensor F of the deformed body b relative
to the undeformed body B is defined to be the tensor that transforms a
differential line element dR into dr (Fig. 3.4), i.e.,
dr = F · dR = dR · FT . (3.19)
3.3. Deformation Gradient Tensor 73
Similarly, the spatial deformation gradient tensor F−1 of B relative

to b is defined by the inverse transformation
dR = F−1 · dr = dr · F−T . (3.20)
Using these equations, we can write
∂r
FT = = ∇r
∂R
∂R
F−T = ¯
= ∇R. (3.21)
∂r
N
dS ds
dR dr
F
n
B b
Fig. 3.4. Deformation of length element dR into length element dr.
These tensors also can be expressed in terms of base vectors alone.

Substituting Eqs. (3.13) into (3.21) yields
∂r gI dX I
FT = = = GJ gI δ IJ = GI gI
∂R GJ dX J
∂R GJ dX J
F−T = = = gI GJ δ JI = gI GI
∂r gI dX I
or, with Eqs. (3.14)1 , and (3.16)1,
F = gI GI = FIi gi GI
−1
F = GI gI = FiI GI gi . (3.22)
The expression for F shows that the FIi are actually the mixed components
F·Ii of the deformation gradient tensor. We drop the dot with the under-
standing that the deformed base vector (lower case g) always appears first
in F; in matrix form, therefore, i denotes the row and I the column.
We also could derive F = gI GI from Eqs. (3.11)1 , (3.18)1 , and (3.21)1,
i.e., FT = ∇r = GI r,I = GI gI . With this relation, it is straightforward to
show
gI = F · G I , (3.23)
illustrating how deformation convects GI into gI .

The mixed components FIi = xi ,I and FiI = X I ,i are called deforma-
tion gradients. Since the bases GI and gi are chosen independently of
the deformation, the FIi or FiI contain the deformation part of the tensor
F. In convected coordinates, setting xi = X J gives FIJ = X J ,I = δ JI and
gi = gJ , and so F = δ JI gJ GI = gI GI , which is consistent with Eq. (3.22)1.
In this case, therefore, deformation is shifted into the convected base vector
gI . Tensor components such as FIi and FiI , which depend on base vectors in
two different reference frames, are called two-point tensor components.
Equations (3.18) and (3.22) provide a useful identity. Since
¯ = (GI gI ) · (gJ ∂ ) = GI δ JI ∂ = GI ∂ ,
FT · ∇
∂X J ∂X J ∂X I
we have
¯
∇ = FT · ∇. (3.24)
3.4 Deformation and Strain Tensors
The deformation gradient tensor can serve as the primary deformation vari-
able in solid mechanics problems. It is relatively easy to compute, but it
contains two features that may complicate analyses. First, as shown by
Eqs. (3.22), F is not symmetric in general, and so all nine components rela-
tive to a given basis may be unknown in some problems. Second, as shown
by Eq. (3.19), F deforms and rotates dR into dr. Since rigid-body rotation
of an element should produce no stress, certain restrictions must be placed
on how F appears in constitutive relations (see Sec. 5.2.5).1
One way around these problems is to define deformation variables based
only on changes in length, and not orientation, of line elements. There are
several ways to do this; here, we focus on the most common approaches.
3.4.1 Deformation Tensors

Consider an arbitrary differential element dR in B of length dS that de-
forms into dr of length ds in b (Fig. 3.4). In terms of F, Eqs. (3.19) and
1 In certain instances, rotation information may be useful. One example is torsion of
the left ventricle during the cardiac cycle (Arts et al., 1984; Hansen et al., 1988; Ingels
et al., 1989; Taber et al., 1996).
3.4. Deformation and Strain Tensors 75
(3.20) give
ds2 = dr · dr = (dR · FT ) · (F · dR) ≡ dR · C · dR

dS 2 = dR · dR = (dr · F−T ) · (F−1 · dr) ≡ dr · B−1 · dr. (3.25)
In these equations, since (F−T · F−1 )−1 = (F−1 )−1 · (F−T )−1 = F · FT (see
Table 2.3, page 23),
C = FT · F
B = F · FT (3.26)
are the right and left Cauchy-Green deformation tensors, respec-

tively.2 (F is on the right or left.) Because CT = (FT · F)T = FT · (FT )T =
FT · F = C and similarly for B, C and B are symmetric tensors.
3.4.2 Strain Tensors

The Lagrangian strain tensor E and the Eulerian strain tensor e are
defined by the relations
ds2 − dS 2 = 2 dR · E · dR
= 2 dr · e · dr, (3.27)
which extend (3.3) and (3.6) to three dimensions. In the literature, vari-
ous names are attached to these strain tensors. For instance, Green and
Saint-Venant are often associated with E, and Almansi and Hamel with e.
Equations (3.25) yield
ds2 − dS 2 = dR · C · dR − dR · dR = dR · (C − I) · dR
= dr · dr − dr · B−1 · dr = dr · (I − B−1 ) · dr
and comparison with Eqs. (3.27) reveals
2E = C − I = FT · F − I
2e = I − B−1 = I − F−T · F−1 . (3.28)
Because E and e are tensors, the components of strain relative to any co-
ordinate system can be obtained by double-dotting with the corresponding
2 In Chap. 2, B was defined as a coordinate transformation tensor. From this point
forward, unless stated otherwise, B represents a deformation tensor.

base dyad. Substituting Eqs. (3.22) into (3.26) gives

C = FT · F = (GI gI ) · (gJ GJ ) = gIJ GI GJ
= (FIi GI gi ) · (FJj gj GJ ) = FIi FJj gij GI GJ
B−1 = F−T · F−1 = (gI GI ) · (GJ gJ ) = GIJ gI gJ

= (FiI gi GI ) · (FjJ GJ gj ) = FiI FjJ GIJ gi gj . (3.29)
Next, inserting these equations, Eqs. (3.12), and the relations
E = EIJ GI GJ , e = eij gi gj (3.30)
into Eqs. (3.28) yields the strain components
2EIJ = CIJ − GIJ

2eij = gij − (B −1 )ij , (3.31)
where
CIJ = FIi FJj gij = gIJ

(B −1 )ij = FiI FjJ GIJ . (3.32)
These equations show that the components of the strain tensors reflect
differences between the metrics of the deformed and undeformed bodies.
Here, we also note that substituting Eqs. (3.13) and (3.30) into (3.27) gives
the alternate definitions
ds2 − dS 2 = 2EIJ dX I dX J
= 2eij dxi dxj (3.33)
for the Lagrangian and Eulerian strains.
Now, we pause to mention two important points. First, since E and e
are tensors, all the tools of tensor analysis in Chap. 2 can be used, including
the computation of principal strains and strain invariants (see Sec. 3.7). To
illustrate the second point, we examine the components of e defined by
e = eI ∗ J ∗ gI gJ .
Equations (3.12), (3.28)2 , and (3.29)2 give
2e = I − B−1 = gIJ gI gJ − GIJ gI gJ
and, therefore,
2eI ∗ J ∗ = gIJ − GIJ .
Comparison with Eqs. (3.31)1 and (3.32)1 shows that EIJ = eI ∗ J ∗ . Thus,
although the strain tensors E and e are fundamentally different, the compo-
nents of E with respect to the basis {GI GJ } are equal to the components
of e with respect to the basis {gI gJ }. This equivalency of components
of different tensors in appropriate bases will play a major role in our un-
derstanding of the relations between the components of the various stress
tensors discussed in the next chapter.
Example 3.1 In Cartesian coordinates, let the position of a point be Z I

and z i before and after deformation, respectively. Derive expressions for the
Lagrangian and Eulerian strain components in terms of these coordinates.
Solution. In Cartesian coordinates, we have GIJ = δ IJ and gij = δ ij .
Thus, Eqs. (3.15), (3.31), and (3.32) give
1 ∂z k ∂z k
EIJ = − δ IJ
2 ∂Z I ∂Z J
1 ∂Z K ∂Z K
eij = δ ij − . (3.34)
2 ∂z i ∂z j
3.4.3 Physical Components of Strain

As mentioned in Sec. 2.7.2, physical components of tensors can be defined
in various ways, with one possibility given by Eq. (2.132) for second-order
tensors. For strain, we use a somewhat different approach, which is guided
by the defining relation for the strain tensor itself.
We define the physical components of the Lagrangian strain tensor E
by writing Eq. (3.33) in terms of physical quantities. Doing this yields
ds2 − dS 2 = 2EIJ dX I dX J = 2ÊIJ dX̂ I dX̂ J , (3.35)
where hat denotes a physical component. Since
dR = GI dX I = eI dX̂ I , (3.36)
where
GI
eI = (3.37)
G(II)
is the unit vector in the direction of GI , we have

dX̂ I = dX I G(II). (3.38)
Inserting this relation into Eq. (3.35) shows that the physical Lagrangian
strain components are given by
EIJ
ÊIJ = . (3.39)
G(II) G(JJ)
Since E is a symmetric tensor, the matrix [EIJ ] is symmetric, and this

equation shows that [ÊIJ ] also is symmetric.
In terms of the reciprocal unit vectors
GI
eI = √ , (3.40)
G(II)
the Lagrangian strain tensor can be written

E = EIJ GI GJ = ÊIJ G(II) G(JJ)G(II) G(JJ) eI eJ . (3.41)
In orthogonal curvilinear coordinates, because eI = eI and G(II) = 1/G(II),
this expression becomes
E = ÊIJ eI eJ (I, J summed), (3.42)
which is consistent with Eq. (2.135). This result is useful for setting up
problems in terms of physical components, if the undeformed body is re-
ferred to an orthogonal coordinate system.
In the remainder of this section, we examine some simple deformations
through examples.
Example 3.2 Soft biological tissues rarely are devoid of stress in vivo. In
fact, most tissues experience ever-changing three-dimensional forces and de-
formations. To a first approximation, however, many tissues are subjected
to simple tension or compression in one, two, or three dimensions. For ex-
ample, papillary muscles stretch uniaxially to prevent the heart valves from
inverting during systole, and articular cartilage in the knee is compressed
between the femur and tibia to provide a cushion between the surfaces of
the bones. An example of multiaxial loading is that experienced by epithe-
lia, which are thin sheets of cells covering surfaces of tissues and organs.
The endothelium is a specialized epithelium that lines the inner surfaces
of blood vessels. When subjected to blood pressure, endothelia undergo
circumferential and longitudinal stretching with radial compression.
Experiments designed to determine material properties of soft tissues
commonly use uniaxial or biaxial loading protocols. Thus, simple extension
and compression of tissues is of fundamental importance in biomechanics.
Consider uniform triaxial extension of a rectangular block. With

the coordinates X I of P and xi of p referred to the same Cartesian system,
the deformation is described by (see Fig. 3.5)
x1 = λ1 X 1 , x2 = λ2 X 2 , x3 = λ 3 X 3 , (3.43)
where the λI are stretch ratios in the coordinate directions. Compute the
Cartesian components of E and e in terms of the λI .
X2,x2
P p
X1
l1X1
X1,x1
Fig. 3.5. Uniform extension in X 1 -direction.
Solution. The position vectors to P and p, respectively, are
R = X I eI , r = x i ei , (3.44)
where the eI = ei are unit vectors along the coordinate axes. Thus, the base
vectors are GI = R,I = eI and gi = r,i = ei , and so GIJ = eI ·eJ = δ IJ and
gij = ei · ej = δ ij . In the following, we keep subscripts and superscripts
distinct, although this is not really necessary, as eI = eI in Cartesian
coordinates.
Equations (3.18) and (3.21)1 give
FT = ∇r
∂ ∂ ∂
= e1 1
+ e2 2
+ e3 (λ1 X 1 e1 + λ2 X 2 e2 + λ3 X 3 e3 )
∂X ∂X ∂X 3
= λ1 e1 e1 + λ2 e2 e2 + λ3 e3 e3 . (3.45)
Therefore, F = λ1 e1 e1 + λ2 e2 e2 + λ3 e3 e3 , which shows that the stretch

ratios are the only non-zero Cartesian components of F. (Note that F = FT
in this problem, since eI = eI .) In matrix form, we can write

F(ei eI ) = FIi = diag [λ1 , λ2 , λ3 ], (3.46)
where
⎡ ⎤
a 0 0
⎣
diag [a, b, c] ≡ 0 b 0 ⎦. (3.47)
0 0 c
The use of a Cartesian basis for both the undeformed and deformed co-
ordinates allows us to employ matrix algebra without worrying about pick-
ing up extra factors that general base vectors may include. Thus, with all
components referred to the unit base vectors eI = eI = ei = ei , Eqs. (3.26)
give
[CIJ ] = [FIi ]T [FJj ] = diag [λ21 , λ22 , λ23 ]
[Bij ] = [FIi ][FJj ]T = diag [λ21 , λ22 , λ23 ]

−1
[Bij ] = diag [λ−2 −2 −2
1 , λ2 , λ3 ]. (3.48)
It is easy to show that computing C = FT · F and B = F · FT , with F
given by Eq. (3.45), yields the same results. Finally, Eqs. (3.31) and (3.48)
provide the strain components
1
[EIJ ] = [CIJ − GIJ ] = 12 [CIJ − δ IJ ]
2

= diag 12 (λ21 − 1), 12 (λ22 − 1), 12 (λ23 − 1)

[eij ] = gij − (B −1 )ij = 12 δ ij − (B −1 )ij
1
2
1
= diag 2 (1 − λ−2 1 −2 1 −2
1 ), 2 (1 − λ2 ), 2 (1 − λ3 ) , (3.49)
which are consistent with Eqs. (3.5) and (3.7).
Example 3.3 The importance of shear deformation in the mechanics of

red blood cells is well known (Fung, 1993). The wall of the heart also un-
dergoes significant shearing as it beats (Waldman et al., 1985), and flowing
blood exerts shear stresses on vascular endothelial cells. While mechanical
testing of shear deformation in soft tissue is more difficult technically than
extension and compression, some researchers have explored this problem
using laboratory experiments (Dokos et al., 2002; Sacks and Sun, 2003).
Consider simple shear of a unit cube (Fig. 3.6). During deformation,
a point originally at (X 1 , X 2 , X 3 ) moves parallel to the X 1 -axis to the new
coordinates (x1 , x2 , x3 ). With the coordinates X I and xi measured rela-
tive to the same Cartesian system, the undeformed and deformed position
vectors are
R = X I eI , r = x i ei , (3.50)
g2 g2*
g1= g1*
Fig. 3.6. Simple shear.
in which
x1 = X 1 + kX 2 , x2 = X 2 , x3 = X 3 . (3.51)
In these equations, k is a constant and eI = ei = GI = gi are base vectors.
(a) Determine F and then the convected base vectors gI .

(b) Using an analysis based on convected coordinates (xi = X I ), determine
the gI and then F.
(c) Determine the deformation tensor components CIJ and Bij .
Solution. (a) The deformation gradients defined in Eq. (3.15)1 are given
by the matrix
⎡ ⎤
i i 1 k 0
FI = x , I = ⎣ 0 1 0 ⎦ . (3.52)
0 0 1
Then, Eq. (3.22)1 gives
F = FIi gi GI = e1 e1 + ke1 e2 + e2 e2 + e3 e3 (3.53)
since GI = eI . Next, (3.14)1 provides the convected base vectors

g1∗ = F11∗ g1 + F12∗ g2 + F13∗ g3 = e1
g2∗ = F21∗ g1 + F22∗ g2 + F23∗ g3 = ke1 + e2
g3∗ = F31∗ g1 + F32∗ g2 + F33∗ g3 = e3 , (3.54)
which are illustrated in Fig. 3.6. Note that the gI = {g1∗ , g2∗ , g3∗ } point
along the edges of the deformed cube.
(b) In the convected-coordinate approach, the xi coordinates do not

enter the analysis directly. Rather, we compute the base vectors gI directly
from Eq. (3.11)1 with r(X I ) = (X 1 + kX 2 )e1 + X 2 e2 + X 3 e3 , i.e.,
∂r
g1 ∗ = = e1
∂X 1
∂r
g2 ∗ = = ke1 + e2
∂X 2
∂r
g3 ∗ = = e3 ,
∂X 3
which agree with Eqs. (3.54). These relations and (3.22)1 yield
F = gI G I = g1 ∗ G 1 + g2 ∗ G 2 + g3 ∗ G 3
= e1 e1 + (ke1 + e2 )e2 + e3 e3
= e1 e1 + ke1 e2 + e2 e2 + e3 e3 , (3.55)

(c) As in the previous example, since the basis of FIi is the set of
Cartesian dyads {ei eI }, Eqs. (3.26) and elementary matrix operations give
⎡ ⎤⎡ ⎤ ⎡ ⎤
i T j 1 0 0 1 k 0 1 k 0
[CIJ ] = FI FJ = ⎣ k 1 0 ⎦ ⎣ 0 1 0 ⎦ = ⎣ k 1 + k 2 0 ⎦
0 0 1 0 0 1 0 0 1
⎡ ⎤⎡ ⎤ ⎡ ⎤
T 1 k 0 1 0 0 1 + k2 k 0
[Bij ] = i
FI FJj = ⎣ 0 1 0 ⎦⎣ k 1 0 ⎦ = ⎣ k 1 0 ⎦.
0 0 1 0 0 1 0 0 1
(3.56)
Note that CIJ = gIJ = gI · gJ [see Eq. (3.32)1 ] also could be computed
directly using Eqs. (3.54).
Example 3.4 Blood vessels, ureters, intestines, tracheas, and plant stems
are examples of biological tubes. Even the left ventricle has been modeled
as a thick-walled tube (Arts et al., 1979; Chadwick, 1982; Tozeren, 1983;
Guccione et al., 1991; Taber, 1991; Taber et al., 1996). Thus, deformation
of a tube, i.e., a hollow cylinder, is an important problem in biomechanics.
Consider combined extension, inflation, and torsion of a circular
tube (Fig. 3.7). For convenience, choose (X 1 , X 2 , X 3 ) = (R, Θ, Z) and
(x1 , x2 , x3 ) = (r, θ, z) as cylindrical polar coordinates for a point before and
after deformation, respectively. In addition, assume that the deformation

is described by the mapping
r = r(R), θ = Θ + ψZ, z = λZ, (3.57)
where ψ is the angle of twist per unit undeformed length, and λ is the axial
stretch ratio. Determine the tensor (natural) and physical components of
F and E.
ez
eZ
y
Y z
Z
eq er
eQ eR r
R q
Q x
X
Undeformed
Deformed
Fig. 3.7. Combined extension, inflation, and torsion of a cylinder.
Solution. For an arbitrary material element in the tube, the undeformed

and deformed position vectors are
R = ReR + ZeZ
r = rer + zez , (3.58)
where the geometry of Fig. 3.7 gives (with ez = eZ = constant)
eR = ex cos Θ + ey sin Θ
eΘ = −ex sin Θ + ey cos Θ
er = ex cos θ + ey sin θ
eθ = −ex sin θ + ey cos θ. (3.59)
In these equations, (ex , ey , ez ) are Cartesian unit vectors fixed in the frame
of reference, while (eR , eΘ , eZ ) and (er , eθ , ez ) are cylindrical unit vectors
associated with the undeformed and deformed positions of the element, re-
spectively. With (3.57)–(3.59), Eqs. (3.10) and (3.11) give the base vectors
G1 = R,R = eR
G2 = R,Θ = ReΘ
G3 = R,Z = eZ
g1 = r,r = er
g2 = r,θ = reθ
g3 = r,z = ez
g1 ∗ = r,R = r (R) er
g2 ∗ = r,Θ = reθ
g3 ∗ = r,Z = ψreθ + λez . (3.60)
In terms of the convected base vectors gI , we can compute the defor-
mation gradient tensor F = gI GI after determining the GI . Since the
cylindrical polar coordinates (R, Θ, Z) are orthogonal, this is a relatively
simple matter. In fact, we already have found the GI for cylindrical coor-
dinates in Example 2.3; Eqs. (2.24) give
G 1 = eR , G2 = R−1 eΘ , G 3 = eZ . (3.61)
For simplicity in this example, we write all the (orthogonal) unit vectors
with subscripts. Thus,
F = g1 ∗ G 1 + g2 ∗ G 2 + g3 ∗ G 3
= r (R) er eR + (reθ )(R−1 eΘ ) + (ψreθ + λez )eZ
= r (R) er eR + (r/R)eθ eΘ + λez eZ + ψreθ eZ , (3.62)
and the matrix of components relative to the basis {ei eI } is
⎡ ⎤
r 0 0
F(ei eI ) = ⎣ 0 r/R ψr ⎦ . (3.63)
0 0 λ
Note that, because the basis is composed of unit dyads, i.e., dyads consisting
of unit vectors, all the components of F in these equations are nondimen-
sional. In fact, as shown next, they also are the physical components of
F.
Relative to the natural basis, the tensor components of F are given by
[see Eqs. (3.15)1 and (3.22)1 ]
⎡ ⎤ ⎡ ⎤
i i r,R r,Θ r,Z r 0 0
F(gi GI ) = FI = x ,I = ⎣ θ,R θ,Θ θ,Z ⎦ = ⎣ 0 1 ψ ⎦ , (3.64)
z,R z,Θ z,Z 0 0 λ
where Eqs. (3.57) have been used. Adapting Eq. (2.132) gives the physical
components

i i
g(ii)
F̂I = FI (3.65)
G(II)
with Eqs. (3.60) giving

[GIJ ] [GI · GJ ] = diag 1, R2 , 1
=

[gij ] = [gi · gj ] = diag 1, r2, 1 . (3.66)
Combining Eqs. (3.64)–(3.66) and comparing with Eq. (3.63) shows that

F(ei eI ) = F̂Ii . (3.67)
The deformation tensor components CIJ and (B −1 )ij now can be com-
puted using Eqs. (3.32), (3.64), and (3.66). Alternatively, Eqs. (3.32)1 and
(3.60) give the CIJ directly as
⎡ 2 ⎤
r 0 0
[CIJ ] = [gIJ ] = [gI · gJ ] = ⎣ 0 r2 ψr 2 ⎦. (3.68)
2 2 2 2
0 ψr λ + ψ r
Equation (3.31)1 then gives the Lagrangian strain components
⎡ 2 ⎤
r −1 0 0
1
[EIJ ] = 12 [CIJ − GIJ ] = ⎣ 0 r 2 − R2 ψr 2 ⎦ , (3.69)
2 2 2 2 2
0 ψr λ +ψ r −1
and Eq. (3.39) gives the physical strain components
⎡ 2 ⎤
r −1 0 0
⎢ ⎥
⎢ ⎥
1⎢⎢ r 2
ψr 2 ⎥
0 − 1 ⎥
[ÊIJ ] = ⎢ R2 R ⎥. (3.70)
2⎢ ⎥
⎢ ⎥
⎣ ψr 2 ⎦
0 λ2 + ψ 2 r 2 − 1
R
As a check, these components should agree with those computed directly
from Eq. (3.28)1 using the physical components of F and I. Using Eq. (3.62)
and I expressed in terms of unit dyads, we get
E = 12 (FT · F − I)
= 1
2 {[r eR er + (r/R)eΘ eθ + λeZ ez + ψreZ eθ ]
· [r er eR + (r/R)eθ eΘ + λez eZ + ψreθ eZ ]
−(eR eR + eΘ eΘ + eZ eZ )}
2
= 1
2 (r − 1)eR eR + (r 2 /R2 − 1)eΘ eΘ + (ψr 2 /R)eΘ eZ

+ (ψr 2 /R)eZ eΘ + (λ2 + ψ2 r 2 − 1)eZ eZ
= ÊIJ eI eJ . (3.71)
The last line follows from Eq. (3.42), since the eI = eI are orthogonal unit
vectors. The ÊIJ in this expression do indeed agree with those of Eq. (3.70).
Finally, we note one last point. The matrix [EIJ ] is always symmetric
since CIJ = gIJ = gJI . Equation (3.70) shows that the matrix of physical
components also is symmetric, consistent with the discussion in Sec. 2.7.
However, the matrix of mixed components, given by EJI = EJK GKI , is
⎡ ⎤
r 2 − 1 0 0
⎢ ⎥
I 1 ⎢ r2 ψr 2 ⎥
EJ = ⎢ ⎢ − ⎥, (3.72)
2⎣
0
R2
1
R2 ⎥
⎦
0 ψr 2 λ2 + ψ 2 r 2 − 1
which is not symmetric. (Here, we have used the relation G(II) = 1/G(II)
for orthogonal coordinates.) This demonstrates that, even though E = ET ,
EJI = EIJ in general [see text following Eqs. (2.41)].
Example 3.5 If the tube of the previous example is elastic, it does not
matter whether the cylinder is extended, inflated, or twisted first (or all
three simultaneously). The final state of stress is the same in all cases, and
so we did not consider the order of the imposed deformations. If dissipa-
tion occurs, however, the path the cylinder takes to its final configuration
may affect the stresses, even though the final strain state is the same. Be-
cause viscoelastic effects often are important in biomechanics, it is useful
to illustrate how such an analysis may proceed. In this example, therefore,
consider the following two paths to the same final deformed configuration:
Path A:
1 Extension: r1 = R, θ 1 = Θ, z1 = λZ
2 Inflation: r2 = r(r1 ), θ 2 = θ 1 , z2 = z1
3 Torsion: r = r2 , θ = θ2 + ψZ = θ 2 + ψz2 /λ, z = z2
Path B:
1 Torsion: r1 = R, θ1 = Θ + ψZ, z1 = Z
2 Extension: r2 = r1 , θ 2 = θ 1 , z2 = λz1
3 Inflation: r = r(r2 ), θ = θ 2 , z = z2
In both cases, the final location of an arbitrary element is given by
Eqs. (3.57). Show that both paths produce the same total deformation
gradient tensor of (3.64).
Solution. For each step in Paths A and B, the following relative defor-
mation gradients are computed [see Eq. (3.64)]:
⎡ ⎤(A) ⎡ ⎤
i (A) r1 , R r1 , Θ r1 , Z 1 0 0
FI 1 = ⎣ θ 1 ,R θ 1 ,Θ θ1 ,Z ⎦ =⎣ 0 1 0 ⎦
z1 ,R z1 ,Θ z1 ,Z 0 0 λ
⎡ ⎤(A) ⎡ ⎤
i (A) r2 ,r1 r 2 , θ1 r 2 , z1 r (R) 0 0
FI 2 = ⎣ θ 2 ,r1 θ 2 , θ1 θ 2 , z1 ⎦ =⎣ 0 1 0 ⎦
z2 ,r1 z2 ,θ1 z2 ,z1 0 0 1
⎡ ⎤(A) ⎡ ⎤
i (A) r,r2 r,θ2 r,z2 1 0 0
FI 3 = ⎣ θ,r2 θ,θ2 θ,z2 ⎦ =⎣ 0 1 ψ/λ ⎦
z,r2 z,θ2 z,z2 0 0 1
⎡ ⎤(B) ⎡ ⎤
i (B) r1 , R r1 , Θ r1 , Z 1 0 0
FI 1 = ⎣ θ 1 ,R θ 1 ,Θ θ1 ,Z ⎦ =⎣ 0 1 ψ ⎦
z1 ,R z1 ,Θ z1 ,Z 0 0 1
⎡ ⎤(B) ⎡ ⎤
i (B) r2 ,r1 r 2 , θ1 r 2 , z1 1 0 0
FI 2 = ⎣ θ 2 ,r1 θ 2 , θ1 θ 2 , z1 ⎦ =⎣ 0 1 0 ⎦
z2 ,r1 z2 ,θ1 z2 ,z1 0 0 λ
⎡ ⎤(B) ⎡ ⎤
i (B) r,r2 r,θ2 r,z2 r (R) 0 0
FI 3 = ⎣ θ,r2 θ,θ2 θ,z2 ⎦ =⎣ 0 1 0 ⎦ (3.73)
z,r2 z,θ2 z,z2 0 0 1
These matrices give the deformation gradients for each step relative to the
geometry of the previous step.
With each of these matrices, we can compute a deformation gradient
tensor Fk = (FIi )k gi GI , where k indicates a step in the deformation path.
[Note that the gi and GI given by Eqs. (3.60) remain fixed.] Then, the
successive deformations of a line element are given by
dr1 = F1 · dR
dr2 = F2 · dr1
dr = F3 · dr2
and, therefore,
dr = F3 · [F2 · (F1 · dR)] .
This equation can be written
dr = F · dR,
where
F = F3 · F2 · F1 (3.74)
is the total deformation gradient tensor. Alternatively, the chain rule gives
the equivalent expression
i ∂xi ∂xi ∂xj2 ∂xk1
FI = = j k ∂X I
= [Fji ]3 [Fkj ]2 [FIk ]1 . (3.75)
∂X I ∂x
∂x2 1
The reader may verify that substituting Eqs. (3.73) for either Path A or
Path B reproduces Eq. (3.64) (see Problem 3.5). This exercise demonstrates
that the final deformation is independent of the deformation path; but
again, if the material is inelastic, the stresses generally depend on the path.
3.5 Strain-Displacement Relations
In this section, we write the strains in terms of displacements. Let u(X I , t)

be the displacement vector from the point P in body B to its image p in b.
The reference frames associated with B and b may be in relative motion,
with the vector b(t) locating the position of reference frame o with respect
to frame O (see Fig. 3.3, page 70). Then, the position vectors of P and p
are related by
r(X I , t) + b(t) = R(X I , t) + u(X I , t). (3.76)
Note that, since it is independent of the point P , b does not depend on the
coordinates X I of P .
Substituting Eq. (3.76) into (3.21) and noting Eqs. (3.18) gives
∂r ∂u
FT = = I+ = I + ∇u
∂R ∂R
∂R ∂u
F−T = = I− ¯
= I − ∇u (3.77)
∂r ∂r
or, since IT = I,
F = (FT )T = I + (∇u)T
F−1 = (F−T )T = I − (∇u)
¯ T. (3.78)
Inserting these equations into Eqs. (3.28) yields
2E = FT · F − I

= [I + ∇u] · I + (∇u)T − I
3.5. Strain-Displacement Relations 89
E = 12 [∇u + (∇u)T + (∇u) · (∇u)T ] (3.79)
2e = I − F−T · F−1

= ¯ · I − (∇u)
I − I − ∇u ¯ T
e = 12 [∇u
¯ + (∇u)
¯ T − (∇u)
¯ · (∇u)
¯ T ]. (3.80)
These strain-displacement relations illustrate the symmetry of the ten-

sors E and e. Moreover, they show that geometric nonlinearity comes from
terms quadratic in the displacement gradients. This nonlinearity leads to
substantial complexity when deformation is large; thus, when possible, it is
advantageous to seek approximations to Eqs. (3.79) and (3.80), as will be
discussed in Sec. 3.9.
We next derive component forms for the strain-displacement relations.
Because the Lagrangian and the Eulerian strain tensors contain derivatives
with respect to the undeformed and deformed bodies, respectively, the most
convenient representations for these tensors are
E = EIJ GI GJ
e = eij gi gj = eI ∗ J ∗ gI gJ , (3.81)
and the displacement vector has the representations
u = uI G I = uI G I
∗
= u i gi = u i g i = u I gI = u I ∗ g I . (3.82)
The strain tensors depend on the displacement gradients given by
∇u = GI u,I
¯
∇u = gi u,i = gI u,I , (3.83)
in which Eqs. (3.18) have provided the gradient operators. Now, substitut-
ing the covariant-component forms of Eqs. (3.82) and noting Eq. (2.150)
yields
∇u = GI (uJ GJ ),I = GI (uJ |I GJ ) = uJ |I GI GJ
¯
∇u = gi (uj gj ),i = gi (uj ||i gj ) = uj ||i gi gj
= gI (uJ ∗ gJ ),I = gI (uJ ∗ ||I gJ ) = uJ ∗ ||I gI gJ , (3.84)
where the single vertical lines and double vertical lines indicate covariant
differentiation with respect to B and b, respectively. In particular, the three
relations in (3.84) involve differentiation of the base vectors GI , gi , and gI ,
respectively. Thus, Eq. (2.151)2 gives
uI |J = uI ,J −uK ΓK
IJ
ui ||j = ui ,j −uk Γ̄kij
uI ∗ ||J = uI ∗ ,J −uK ∗ Γ̄K
IJ , (3.85)
where suitable modification of Eq. (2.137)2 yields the Christoffel symbols
IJ = GI ,J · G
ΓK K
Γ̄kij = gi ,j · gk
IJ = gI ,J · g ,
Γ̄K K
(3.86)
with bar indicating a symbol defined in the deformed body b. In summary,

()|I represents differentiation with respect to X I and GI in B, ()||i rep-
resents differentiation with respect to xi and gi in b, and ()||I represents
differentiation with respect to X I (convected) and gI in b. In convected
coordinates, xi = X I , gi = gI , and ui ||j = uI ∗ ||J ; and the two forms for
¯ in Eq. (3.84) are equivalent.
∇u
Getting back to the strain-displacement relations, we substitute
Eqs. (3.84) into (3.79) to obtain
2E = uJ |I GI GJ + uJ |I GJ GI + (uJ |I GI GJ ) · (uL |K GL GK )
= uJ |I GI GJ + uI |J GI GJ + uJ |I uL |K GJLGI GK
= uJ | I G I G J + uI | J G I G J + uJ | I uJ | K G I G K
= (uI |J + uJ |I + uK |I uK |J )GI GJ , (3.87)
and, similarly, Eq. (3.80) gives
2e = uj ||i gi gj + uj ||i gj gi − (uj ||i gi gj ) · (ul ||k gl gk )
= (ui ||j + uj ||i − uk ||i uk ||j )gi gj (3.88)
or
2e = uJ ∗ ||I gI gJ + uJ ∗ ||I gJ gI − (uJ ∗ ||I gI gJ ) · (uL∗ ||K gL gK )
∗
= (uI ∗ ||J + uJ ∗ ||I − uK ∗ ||I uK ||J )gI gJ . (3.89)
Note that GJL in the second line of Eq. (3.87) was used to raise the subscript
on uL, and the indices J and K were interchanged in the last term of the
3.6. Geometric Measures of Deformation 91
last line. In Eqs. (3.88) and (3.89), gjl and gJL raised the subscripts on
ul and uL∗ , respectively. To show that gJL is needed to raise the index of
uL∗ , we consider the representations
∗
u = u I ∗ g I = u I gI .
Dotting with gJ yields
∗
(uI ∗ gI ) · gJ = (uI gI ) · gJ
or
∗ ∗
uI ∗ gIJ = uI δ JI = uJ , (3.90)
which is the desired result.
Finally, comparing Eqs. (3.81) with (3.87)–(3.89) reveals that
EIJ = 12 (uI |J + uJ |I + uK |I uK |J )
eij = 12 (ui ||j + uj ||i − uk ||i uk ||j )
∗
= 12 (uI ∗ ||J + uJ ∗ ||I − uK ∗ ||I uK ||J ). (3.91)
These equations provide the strain-displacement relations in terms of scalar

components.
3.6 Geometric Measures of Deformation
The deformation gradient, deformation, and strain tensors characterize the

deformation of a body. In general, they have no direct physical interpreta-
tions like those of the stretch or stretch ratio. In practice, however, more
physically meaningful measures often are useful. For example, regional de-
formation of a tissue or organ, such as the beating heart, can be estimated
by tracking the motions of markers (Waldman et al., 1985; 1988; McCul-
loch et al., 1989; Villarreal et al., 1991; Hashima et al., 1993; Taber et al.,
1994; Alford and Taber, 2003). This section explores the relations between
strain and changes in length, angle, volume, and area.
3.6.1 Stretch
Let dS and ds be the lengths of a differential line element before and after
deformation, respectively (see Fig. 3.4, page 73). The stretch ratio ds/dS
depends on the specified orientation of either the undeformed or the de-
formed element. In other words, we can compute a Lagrangian stretch
ratio Λ(N) for an element that occupies a specified direction N in the un-
deformed body B or an Eulerian stretch ratio λ(n) for an element that
occupies a specified direction n in the deformed body b, with N and n
being unit vectors (Fig. 3.4). In general, N and n can be chosen arbitrar-
ily in B and b, respectively. However, if N and n are bound to the same
material line element, i.e., N is convected into n during deformation, then
Λ(N) = λ(n) . On the other hand, if we choose N = n, then the material
elements associated with each of these unit vectors differ if local rotation
occurs, and so Λ(N) = λ(n) in general.
Noting the geometry of Fig. 3.4, we write
dR = N dS, dr = n ds. (3.92)
Substitution into Eq. (3.19) yields
n ds = F · N dS or N dS = F−1 · n ds,
which give the relations
n Λ(N) = F · N = N · FT
N λ−1
(n) = F−1 · n = n · F−T . (3.93)
Because N and n are unit vectors, we can write

Λ2(N) = n Λ(N) · n Λ(N) = N · FT · (F · N)

λ−2(n) = N λ−1 −1
(n) · N λ(n) = n · F
−T
· F−1 · n ,
and using Eqs. (3.26) gives
Λ2(N) = N · C · N
λ−2
(n) = n · B
−1
· n. (3.94)
In addition, analogous to the engineering strain (stretch) of Eq. (3.1), the

Lagrangian and Eulerian extension ratios are defined by
∗ ds − dS
E(N) = = Λ(N) − 1
dS
ds − dS 1
e∗(n) = =1− . (3.95)
ds λ(n)
Example 3.6 Suppose the components of the strain tensors E and e are
known in the Cartesian coordinates X I = xi . Compute the extension ratios
in terms of strain components for an element aligned with the X 1 -axis in the
undeformed body and for an element aligned with this axis in the deformed
body.
Solution. In the first case, N = e1 , and in the second case, n = e1 .
Thus, Eqs. (3.28) and (3.94) yield
Λ2(1) = e1 · C · e1 = e1 · (I + 2E) · e1 = 1 + 2E11
λ−2
(1) = e1 · B−1 · e1 = e1 · (I − 2e) · e1 = 1 − 2e11 , (3.96)
which show that Λ(1) = λ(1) in general. In this problem, the stretch ratios
depend only on normal strains. However, if either N or n is not aligned
with a coordinate axis, then shear strains would be involved. The extension
ratios, from Eqs. (3.95), are
1
∗
E(1) = Λ(1) − 1 = (1 + 2E11 ) 2 − 1
1
e∗(1) = 1 − (λ(1) )−1 = 1 − (1 − 2e11 ) 2 , (3.97)
and these equations give
∗ ∗2
E11 = E(1) + 12 E(1)
e11 = e∗(1) − 12 e∗2
(1) . (3.98)
These last relations show that if the extension ratios are small, then E11 =∼
∗ ∼ ∗
E(1) and e11 = e(1) , i.e., the linear normal strain components have the usual
physical interpretation of change in length divided by undeformed length.
Example 3.7 For the case of simple shear (see Example 3.3 and Fig. 3.6,
page 81), compute the stretch ratios for elements that are parallel to the
X 2 -axis (a) before deformation and (b) after deformation.
Solution. (a) On setting N = e2 , Eqs. (3.56)1 and (3.94)1 give the
stretch ratio
Λ2(2) = e2 · C · e2 = C22 = 1 + k 2 .
Note that
√ the geometry of Fig. 3.6 shows that the deformed length of side
AB is 1 + k 2 , in agreement with this result.
(b) Setting n = e2 in Eq. (3.94)2 gives
λ−2
(2) = e2 · B
−1
· e2 = (B −1 )22 .
Inverting the matrix of Eq. (3.56)2 yields

⎡ ⎤
1 −k 0
−1
[Bij ] = ⎣ −k 1 + k 2 0 ⎦,
0 0 1
and so
λ2(2) = (1 + k 2 )−1 .
∼ λ(2) ∼
Again, we see that Λ(2) = λ(2). For k << 1, however, Λ(2) = = 1, i.e.,
to a first approximation, simple shear involves no extension in the linear
case (relative to the coordinate axes).
Example 3.8 Wound healing involves considerable mechanics. As a

wound heals in adults, the shape of its boundary affects the extent of scar-
ring, possibly caused by stresses generated during wound closure. Fibrob-
lasts play a major role in this process, as these cells generate contractile
forces and tractions that draw the edges of the wound together (Murray,
2003). In contrast, wounds in the embryo heal without scarring, as an ac-
tomyosin cable quickly forms around the wound edge and contracts to close
the opening (Jacinto et al., 2001; Redd et al., 2004).
Consider a simple model for skin with a puncture wound. Before healing,
the model consists of an annular membrane with an inner radius a and outer
radius b (Fig. 3.8a). With the membrane fixed at R = b, the healing process
moves the inner (cut) edge symmetrically inward toward the center. If a
point originally a distance R from the center moves to the location
log(R/a)
r(R) = b , (3.99)
log(b/a)
compute the radial and circumferential stretch ratios.3
Solution. The geometry for this problem is like that for a circular tube
(see Example 3.4, page 82). In two dimensions, the position vectors are
R = R eR , r = r er . (3.100)
Because there is no torsion in the present problem, the unit vectors eR = er

and eΘ = eθ are given by Eqs. (3.59)1,2.
3 The deformed geometry described by Eq. (3.99), chosen only for illustration, likely is
not the actual solution to this problem. The true deformation field must be found by
solving the full boundary value problem of nonlinear elasticity (see Chap. 6).
(a)
Fig. 3.8. (a) Annular membrane model for wound healing. With outer edge fixed, inner
edge moves inward to close the wound. (b) Stretch ratios in model after healing. (inner
edge: R = 1; outer edge: R = 2)
With the gradient operator for cylindrical coordinates provided by

Eq. (2.168), Eq. (3.21)1 yields
∂ eΘ ∂
FT = ∇r = eR + (r er )
∂R R ∂Θ
∂r r ∂er
= eR er + eΘ
∂R R ∂Θ
∂r r
= eR er + eΘ eθ
∂R R
or
∂r r
F= er eR + eθ eΘ . (3.101)
∂R R
To determine the stretch ratios, we compute the two-dimensional defor-
mation tensor
C = FT · F
∂r
2 r 2
= eR eR + eΘ eΘ , (3.102)
∂R R
and Eq. (3.94)1 gives the stretch ratios
∂r
ΛR = (eR · C · eR )1/2 =
∂R
r
ΛΘ = (eΘ · C · eΘ )1/2 = . (3.103)
R
These stretch ratios are also the physical components of F. Finally, insert-
ing Eq. (3.99) yields
1 b
ΛR =
log(b/a) R
1 b R
ΛΘ = log . (3.104)
log(b/a) R a
N2
n2
p dr2
dR2 F
P dr1
dR1
N1 n1
B b
Fig. 3.9. Change in angle (shear) between two differential line elements.
Stretch ratio distributions for a = 1 and b = 2 are shown in Fig. 3.8b.

Note that, for complete wound closure, the computed value ΛΘ = 0 at
the center is not physically possible, as a length cannot shrink to a point.
Hence, the assumed deformation is possible only if the wound does not close
completely.
3.6.2 Shear
Shear strains characterize the change in angle, due to deformation, between
two line elements. The elements
dR1 = N1 dS1 = F−1 · dr1 = dr1 · F−T
dR2 = N2 dS2 = F−1 · dr2 = dr2 · F−T (3.105)
at a point P in B deform into the elements
dr1 = n1 ds1 = F · dR1 = dR1 · FT
dr2 = n2 ds2 = F · dR2 = dR2 · FT (3.106)
at the point p in b (Fig. 3.9), where the Ni and ni are unit vectors and F
is the local deformation gradient tensor [see Eqs. (3.19) and (3.20)].
The cosines of the angles between the undeformed and the deformed
elements are
dR1 · dR2 (dr1 · F−T ) · (F−1 · dr2 )
cos(N1 , N2 ) = N1 · N2 = = 1 1
dS1 dS2 (dr ·B−1 ·dr ) 2 (dr ·B−1 ·dr ) 2
1 1 2 2
dr1 · dr2 (dR1 · FT ) · (F · dR2 )

cos(n1 , n2 ) = n1 · n2 = = 1 1,
ds1 ds2
(dR1 · C · dR1 ) 2 (dR2 · C · dR2 ) 2
(3.107)
in which Eqs. (3.25) have been used. With Eqs. (3.26), (3.94), (3.105), and
(3.106), these relations can be written
n1 · B−1 · n2
N1 · N2 = 1 1
= λ(n1 ) λ(n2 ) (n1 · B−1 · n2 )
(n1 · B−1 · n1 ) 2 (n2 · B−1 · n2 ) 2
N1 · C · N2 N1 · C · N2
n1 · n2 = 1 1 = . (3.108)
Λ(N1 ) Λ(N2 )
(N1 · C · N1 ) 2 (N2 ·C· N2 ) 2
If n1 and n2 are given, the first equation provides N1 · N2 ; if N1 and N2
are given, the second yields n1 · n2 . Of course, if there is no deformation,
then C = B = I and both equations give n1 · n2 = N1 · N2 . Finally, we
define the shears
Γ(N1 N2 ) = γ (n1 n2 ) = n1 · n2 − N1 · N2 , (3.109)
and Eqs. (3.108) give

C
Γ(N1 N2 ) = N1 · − I · N2
Λ(N1 ) Λ(N2 )

γ (n1 n2 ) = n1 · I − λ(n1 ) λ(n2 ) B−1 · n2 . (3.110)
The shears provide measures of the change in angle between undeformed

and deformed line elements.
Example 3.9 Suppose the strain components EIJ are known in Cartesian
coordinates. Compute the shear in the x1 x2 -plane in terms of the EIJ for
(a) an element that is rectangular in the undeformed body B, with edges
parallel to the X 1 and X 2 axes; and (b) an element that is rectangular in
the deformed body b, again with edges parallel to the X 1 and X 2 axes.
Solution. In the first case, N1 = e1 and N2 = e2 , while in the second
case, n1 = e1 and n2 = e2 (e1 · e2 = 0). Thus, Eqs. (3.28), (3.96), and
(3.110) yield

I + 2E 2E12
Γ(12) = e1 · − I · e2 = 1
Λ(1)Λ(2)
[(1 + 2E11 )(1 + 2E22)] 2
2e12
γ (12) = e1 · I − λ(1) λ(2)(I − 2e) · e2 = 1
[(1 − 2e11 )(1 − 2e22 )] 2
(3.111)
for cases (a) and (b), respectively. Note that the shears depend on the shear
strains, E12 and e12 , as well as the normal strains. If the normal strains
are small compared to unity, then
Γ(12) ∼
= 2E12 , γ (12) ∼
= 2e12 ,
which agree with the usual interpretation for linear shear strain compo-
nents.
Example 3.10 For simple shear (see Example 3.3 and Fig. 3.6, page 81),
Γ(12) provides a measure of the change in angle between line elements orig-
inally aligned with the X 1 and X 2 axes. Determine Γ(12) in terms of k.
Solution. With N1 = e1 and N2 = e2 , Eq. (3.110)1 gives

C C12 C12
Γ(12) = e1 · − I · e2 = = √ ,
Λ(1)Λ(2) Λ(1) Λ(2) C11 C22
where we have inserted Λ2(1) = e1 · C · e1 = C11 and Λ2(2) = e2 · C · e2 = C22.
Using Eq. (3.56)1 , we find
k
Γ(12) = √ .
1 + k2
A glance at the geometry of Fig. 3.6 reveals that Γ(12) = sin γ.
3.6.3 Volume Change

Consider a differential volume element with edges parallel to dR(1), dR(2),
and dR(3) that deforms into an element with edges parallel to dr(1), dr(2),
and dr(3), where
I
dR(k) = GI dX(k) , dr(k) = gi dxi(k). (3.112)
Inserting these relations and Eq. (3.22)1 into
dr(k) = F · dR(k) (3.113)
yields
gi dxi(k) = (FIi gi GI ) · (GJ dX(k)
J
)
= FIi gi δ IJ dX(k)
J
= FIi gi dX(k)
I
or
dxi(k) = FIi dX(k)
I
= xi ,I dX(k)
I
. (3.114)
This relation, which involves Eq. (3.15), is an expression of the chain rule
for differentiation.
The volumes of the parallelepiped before and after deformation, respec-
tively, are
dV = dR(1) · (dR(2) × dR(3))
dv = dr(1) · (dr(2) × dr(3)). (3.115)
Substituting Eqs. (3.112) and noting Eqs. (2.103)2 and (2.115)1 gives
dV = GI · (GJ × GK ) dX(1)
I J
dX(2) K
dX(3)
√ I J K
= G eIJK dX(1) dX(2) dX(3)
dv = gi · (gj × gk ) dxi(1)dxj(2)dxk(3)
√
= g eijk dxi(1)dxj(2)dxk(3), (3.116)
where eIJK and eijk are the permutation symbols defined by Eq. (2.99),
and
G = det [GIJ ] , g = det [gij ] . (3.117)
Now, using Eqs. (3.114), (2.98)1 , and (3.116)1 in order, we can write
Eq. (3.116)2 as
√
dv = g (eijk FIi FJj FK
k I
) dX(1) J
dX(2) K
dX(3)
√ I J K
= g (eIJK j) dX(1) dX(2) dX(3)
√ √
= g (dV / G) j,
where

j ≡ det FIi . (3.118)
Finally, the dilatation ratio is defined to be
dv g
J≡ = j, (3.119)
dV G
which is the ratio of the deformed to the undeformed volume of the ele-
ment. An alternate form for J can be found by taking the determinant of
Eq. (3.65) and comparing the result to (3.118) and (3.119) to obtain
J = det[F̂Ii], (3.120)
where the F̂Ii are physical components of F. (Note the distinction between
j and J.)
Here, we mention two important special cases:
• Convected coordinates. Setting xi = X I and gi = gI gives g ≡ g∗ =

det gIJ and FIi = xi ,I = δ iI . In this case, j = det FIi = 1 and J =

g∗ /G.
• Same coordinate system for undeformed and deformed configurations.
Taking gi = GI gives g = G, and hence J = j = det FIi .
Most authors use one or the other of these specializations of Eq. (3.119).
Before leaving our present discussion, we derive yet another representa-
tion for J. Equations (2.133)2 and (3.120) suggest that J = det F, which
turns out to be true. But since the FIi are two-point tensor components (re-
ferred to base vectors in two different reference frames), Eq. (2.132) suggests
that we must exercise caution in making this statement. In this regard, we
write
det C = det FT · F = det FT det F = (det F)2 . (3.121)
In addition, Eqs. (2.77), (3.32), (3.117), and (3.118) yield
I
det C = det C·J = det GIK CKJ

= det GIK FK i
FJj gij
i
= det GIK det FK det FJj det [gij ]
= (g/G) j 2 ,
and thus
det C = J 2 . (3.122)
In the manipulations above, we also have used the fact that det[GIK GJK ] =
det[δIJ ] = 1, and so det[GIK ] = 1/ det[GIK ] = 1/G. Now, Eqs. (3.121) and
(3.122) give
J = det F. (3.123)
At first glance, it appears that Eqs. (2.77) and (3.118) imply j = det F,
which is generally inconsistent with Eqs. (3.119) and (3.123). Similar to the
above discussion, the explanation is that j is expressed in terms of mixed
componentsof F involving both undeformed and deformed base vectors.
The factor g/G in (3.119) links these bases. These subtle distinctions in
computing the quantities j and J are easy to overlook when reading the
literature.
Soft biological tissues usually are assumed to be incompressible. In this
case, J = 1 (dv = dV ) must be enforced as a material constraint, and the
resulting deformation is called isochoric. As will be shown in Chap. 6,

this assumption sometimes simplifies analyses.
The incompressibility assumption for soft tissues is based on the fact
that most tissues are composed primarily of water by volume fraction. This
assumption is reasonable if extracellular fluid is relatively immobile. In
some tissues, including articular cartilage, however, significant fluid flow
can occur as the tissue deforms. In this case, the solid skeleton should be
considered compressible, and the problem can be analyzed using a biphasic
theory such as poroelasticity or mixture theory (Armstrong et al., 1984;
Holmes, 1986; Mow et al., 1986; Simon et al., 1988; Huyghe et al., 1992).
Example 3.11 Compute the dilatation ratios for the problems discussed
in Examples 3.2–3.4, i.e., uniform extension, simple shear, and axisymmet-
ric deformation of a tube.
Solution. For uniform extension, Eq. (3.45) expresses F in a Cartesian
dyadic basis. Since the components are mixed and because the undeformed
and deformed bases coincide, the definition for the determinant of a tensor
(2.77) gives

λ1 0
0
J = det F = 0 λ2 0 = λ1 λ2 λ3 .
0 0 λ3
(Here, we note that the distinctions between covariant, contravariant, and
mixed components vanish in Cartesian coordinates.) If a block of material
is stretched by λ1 = λ in the X 1 -direction and deforms freely in the other
two directions, then symmetry demands that λ2 = λ3 . If, in addition, the
material is incompressible, then
J = λλ22 = 1
−1/2
which gives λ2 = λ3 = λ .
For simple shear, Eq. (3.55) gives

1 k 0

J = det F = 0 1 0 = 1.
0 0 1
Hence, this deformation is isochoric for all values of the shear parameter k.
Lastly, combined extension, inflation, and torsion of a tube is treated in
three ways. First, Eqs. (3.64) and (3.66) give

j = det FIi = λr
G = det [GIJ ] = R2
g = det [gij ] = r 2 ,
and so Eq. (3.119) yields

g r
J= j = λr . (3.124)
G R
Second, Eq. (3.122) gives

J2 = det C = det CJI = det CJK GKI

= det [CJK ] det GKI
1
= det [CJK ] .
G
With Eq. (3.68), this expression becomes
2
r 0 0
1 λ2 r 2 r 2
J 2 = 2 0 r2 ψr 2 =
,
R R2
0 ψr 2 λ2 + ψ2 r 2
which agrees with Eq. (3.124). Finally, we compute J = det F = det[F̂Ii ].
With Eq. (3.63) providing the physical components of F, we have

r 0 0
λrr
J = 0 r/R ψr = ,
0 R
0 λ
which again agrees with the previous result.
3.6.4 Area Change

Consider a differential area element dA with edges parallel to dR(1) and
dR(2) that deforms into an element da with edges parallel to dr(1) and dr(2)
(Fig. 3.10). Before and after deformation, respectively, we use elementary
vector mechanics to write
dA = N dA = dR(1) × dR(2)
da = n da = dr(1) × dr(2), (3.125)
where N and n are unit vectors normal to the elements with areas dA and
da.
Substituting Eqs. (3.112) into (3.125) and noting Eq. (2.103)1 yields
dA = I
(GI dX(1) ) × (GJ dX(2)
J
) = GI × GJ dX(1)
I J
dX(2)
= IJK GK dX(1)
I J
dX(2)
da = (gi dxi(1)) × (gj dxj(2)) = gi × gj dxi(1)dxj(2)

= ijk gk dxi(1)dxj(2). (3.126)
3.7. Principal Strains 103
N n
dR(2)
dr(2)
dA F
da
dR(1)
dr(1)
Fig. 3.10. Differential area element before and after deformation.
Next, using Eqs. (2.115)1 , (3.16)2 , and (3.114) to replace ijk , gk , dxi(1),
and dxi(2) in da gives
√
da = k K
g eijk (FK g )(FIi dX(1)
I
)(FJj dX(2)
J
)
√ i j k K I J
= g (eijk FI FJ FK )g dX(1)dX(2)
√
= g (eIJK j)gK dX(1) I J
dX(2) . (3.127)
In the last line, we have used the determinant formula (2.98)1, with j
defined by Eq. (3.118). Finally, this expression can be written in the form
da = J dA · F−1 , (3.128)
as verified by the manipulation
da = J( IJK GK dX(1) I J
dX(2) ) · (GL gL)
√
= J( G eIJK )δ K L I
L g dX(1) dX(2)
J
√
= j g eIJK gK dX(1) I J
dX(2) ,
which employs Eqs. (3.126)1 , (3.22)2 , (2.115)1, and (3.119), in this order.
This expression agrees with Eq. (3.127). Equation (3.128) is known as
Nanson’s formula.
3.7 Principal Strains
At any point in a body, we always can find a unique set of orthogonal

axes so that the shear strains relative to those axes vanish. These axes
are called principal axes of strain, and the normal strain components
relative to this system are called principal strains. In addition, the planes
orthogonal to the principal axes are called principal planes. The principal
axes (principal directions) and the corresponding principal strains at a point
are determined by solving an eigenvalue problem.
3.7.1 The Eigenvalue Problem

Consider a line element that is oriented in the direction of the unit vector
N in the undeformed body and n in the deformed body. In addition, let T
and t be arbitrary unit vectors orthogonal to N and n, respectively. The
vectors N and n correspond to principal axes if
ENN = E : NN = N · E · N = 0
ENT = E : NT = N · E · T = 0 (3.129)
and
enn = e : nn = n · e · n = 0
ent = e : nt = n · e · t = 0, (3.130)
where ENN and enn are principal Lagrangian and Eulerian strain compo-
nents along N and n, respectively, and ENT and ent are shear strains.
Relative to the local orthogonal principal axes, GIJ = δ IJ and gij = δ ij ,
and so Eqs. (3.31) give
1
EIJ = 2 (CIJ − δ IJ )
1 −1
eij = 2 [δij − (B )ij ]. (3.131)
These equations show that the shear components of E and e vanish when
the corresponding components of C and B−1 vanish. In other words, the
principal directions of E and C coincide, as do those of e and B−1 . Thus,
Eqs. (3.129) and (3.130) can be replaced by the conditions N · C · N = 0,
N · C · T = 0 and n · B−1 · n = 0, n · B−1 · t = 0, which are satisfied if
N·C = Λ2(N) N
n · B−1 = λ−2
(n) n. (3.132)
The stretch ratios on the right-hand sides of these relations follow from
Eqs. (3.94). Rewriting them yields the eigenvalue problems
(C − Λ2(N) I) · N = 0
(B − λ2(n) I) · n = 0. (3.133)
Solving these equations (see Sec. 2.5.3) provides the principal stretch
ratios Λ(Ni ) and λ(ni ) (i = 1, 2, 3) and the corresponding principal axes
Ni and ni . Equations (3.5)2 and (3.7) then give the principal strains
E(Ni ) = 1
2
(Λ2(Ni ) − 1)
1 −2
e(ni ) = 2 (1 − λ(ni ) ). (3.134)
It can be shown that since C and B are symmetric tensors, their eigen-
values are real and their eigenvectors are mutually orthogonal. Also, as
n is taken in the direction of the deformed element with undeformed ori-
entation N, λ(n) = Λ(N) , and so the eigenvalues of C and B are equal.
Thus, the orthogonal principal axes Ni at the point P in the undeformed
body translate and rotate into the orthogonal principal axes ni at the point
p in the deformed body. In addition, since the eigenvalues represent the
deformation or strain components relative to the principal axes, we can
write
F = Λ1 n1 N1 + Λ2 n2 N2 + Λ3 n3 N3
C = Λ21 N1 N1 + Λ22 N2 N2 + Λ23 N3 N3
B = λ21 n1 n1 + λ22 n2 n2 + λ23 n3 n3
E = E1 N1 N1 + E2 N2 N2 + E3 N3 N3
e = e1 n1 n1 + e2 n2 n2 + e3 n3 n3 , (3.135)
where we have used the invariance properties of these tensors, along with
the definitions Ei ≡ E(Ni ) and ei ≡ e(ni ) . The relation for F follows from
Eqs. (3.93)1 and is consistent with the definitions C = FT ·F and B = F·FT .
Equations (3.135), containing tensors written in terms of their eigenvalues
and eigenvectors, are called spectral representations of these tensors.
3.7.2 Strain Invariants

Each of the deformation and strain tensors possesses principal invariants as
defined by (2.83). Because C and B have the same eigenvalues, Eq. (2.82)
indicates that they also have the same invariants, i.e.,
I1 = tr C = tr B

I2 = 12 (tr C)2 − tr C2 = 1
2 (tr B)2 − tr B2
I3 = det C = det B. (3.136)
In solid mechanics problems, the Lagrangian description often is more
convenient than the Eulerian description. Thus, we now give component
forms for the invariants of C in terms of the components of C and E, where
C = I + 2E. Note that the invariants contain the trace and determinant,
which Eqs. (2.72) and (2.77) define in terms of mixed tensor components.
(Since C and E are symmetric tensors, C·J I
= CJ·I ≡ CJI and E·J
I
= EJ·I ≡
EJI .) Equations (3.31) and (3.32) give
C = CIJ GI GJ = gIJ GI GJ
E = EIJ GI GJ = 21 (CIJ − GIJ )GI GJ ,
and the mixed components are

CJI = CKJ GKI = gKJ GKI
EJI = EKJ gKI = 12 (CKJ − GKJ )GKI = 12 (CJI − δ IJ ), (3.137)
where (2.91) has been used. With Eq. (2.72), the first strain invariant
becomes

I1 = tr CJI = CII = gKI GIK = gIJ GIJ

= tr δ IJ + 2EJI = δ II + 2EII = 3 + 2EII . (3.138)
The second invariant requires
C2 = C · C = (CIJ GI GJ ) · (CKL GK GL )
= CIJ CKL GJK GI GL
= CIJ CLJ GI GL
= CIK CJK GI GJ ,
and thus, by Eq. (2.71),
tr C2 = CIK CJK GI · GJ = CIK CJK GIJ
J
= CK CJK = CJI CIJ . (3.139)
Combining these relations yields
I2 = 12 (I12 − CJI CIJ )

= 12 I12 − (gJK GIK )(gIL GJL )

= 12 (3 + 2EII )(3 + 2EJJ ) − (δ IJ + 2EJI )(δ JI + 2EIJ )
= 3 + 4EII + 2(EII EJJ − EJI EIJ ). (3.140)
Finally, with Eqs. (2.77) and (3.121)–(3.123),
we can write
I
I3 = det CJ = det δ IJ + 2EJI
= (det F)2
= (g/G)j 2 = J 2 . (3.141)
In summary, the strain invariants have the indicial forms
I1 = CII = gIJ GIJ
= 3 + 2EII
I2 = 12 (CII CJJ − CJI CIJ )
= 12 (I12 − gIL gJK GIK GJL ) = gIJ GIJ I3
= 3 + 4EII + 2(EII EJJ − EJI EIJ )

I3 = det CJI = (det F)2 = (g/G)j 2 = J 2

= det δ IJ + 2EJI . (3.142)
By definition, these invariants have the same values when computed in any
coordinate system. In principal coordinates, the shear strains vanish and
the invariants become
I1 = 3 + 2(E1 + E2 + E3 )
I2 = 3 + 4(E1 + E2 + E3 ) + 4(E1 E2 + E2 E3 + E3 E1 )
I3 = (1 + 2E1 )(1 + 2E2 )(1 + 2E3 ), (3.143)
where the Ei are principal Lagrangian strains.
Example 3.12 Compute the strain invariants for Examples 3.2–3.4, i.e.,
uniform extension, simple shear, and axisymmetric deformation of a tube.
Solution. For uniform extension, Eqs. (3.48)1 and (3.142) yield (recall
that CJI = CIJ in Cartesian coordinates)
I1 = C11 + C22 + C33 = λ21 + λ22 + λ23
1
! 1 2 3 2
1 2 2 2 3 2
"
I2 = 2 (C1 + C2 + C3 ) − (C1 ) + (C2 ) + (C3 )
= C11 C22 + C22 C33 + C33 C11
= λ21 λ22 + λ22 λ23 + λ23 λ21

I3 = det CJI = λ21 λ22 λ23 . (3.144)
For simple shear, Eq. (3.56)1 gives (again, CJI = CIJ )
I1 = C11 + C22 + C33 = 3 + k 2
I2 = C11 C22 + C22 C33 + C33 C11 − (C21 )2 = 3 + k 2

I3 = det CJI = 1. (3.145)
Finally, for combined extension, inflation, and torsion of a cylinder,
Eqs. (3.72) give the mixed strain components EJI . Putting these into
Eqs. (3.142) yields
r2
I1 = 3 + 2(E11 + E22 + E33 ) = r 2 + + λ2 + ψ2 r 2
R2
I2 = 3 + 4(E11 + E22 + E33 + E11 E22 + E22 E33 + E33 E11 − E32 E23 )
r2 2 2 2 λ2 r 2
= r 2 + λ + ψ r +
R2 R2
λ2 r 2 r 2
I3 = det δ IJ + 2EJI = . (3.146)
R2

3.8 Stretch and Rotation Tensors
As discussed in Sec. 3.4, the deformation gradient tensor F includes in-

formation on both deformation and rigid-body rotation. In contrast, the
deformation tensors, C and B, and the strain tensors, E and e, contain
only deformation information. In this section, we decompose F into its
component parts of deformation and rotation.
3.8.1 Polar Decomposition

According to the polar decomposition theorem, the deformation gra-
dient tensor F can be decomposed into (1) deformation followed by rigid-
body rotation or (2) rigid-body rotation followed by deformation. For small
displacements and rotations, these motions are additive and commutative
(see Sec. A.2.2). For large displacements and rotations, this is not the case;
rather, rotation and deformation are multiplicative.
Mathematically, the theorem states that the deformation gradient ten-
sor can be decomposed uniquely into the two forms
F = Θ · U = V · Θ, (3.147)
where U and V are positive definite symmetric tensors, and Θ is a proper

orthogonal tensor (since det F = J > 0, see Sec. 2.6.3). Since a proper
orthogonal tensor produces rigid-body rotation, Θ is called the rotation
tensor. All deformation is contained in U and V, which are called the
right and left stretch tensors, respectively.
Notably, the term “stretch tensor” strictly applies only to line elements
oriented in principal directions of strain. Otherwise, U and V generally
include both deformation and at least some rigid-body rotation (Fig. 3.11).
Relative to principal axes, the relation dr = F · dR shows that the first
decomposition represents a stretch (U) of dR followed by a rigid-body
rotation (Θ) into dr, whereas the second decomposition represents a rigid-
body rotation (Θ) of dR followed by a stretch (V) into dr (Fig. 3.12).
Although the rotation Θ is the same in both cases, U = V in general.
In the following, we prove the polar decomposition theorem in two ways:
first through a formal mathematical derivation and then via a relatively
informal physical argument.
Mathematical Derivation. We begin our formal derivation by consid-

ering the deformation tensor C in principal coordinates. Relative to the
3.8. Stretch and Rotation Tensors 109
y F
1 1
4 T
x
1 O
Fig. 3.11. Uniaxial stretch without rotation (F = U = V) of a unit cube with dimensions
fixed in the y and z directions. Line elements parallel to principal axes (e.g., edges)
stretch but do not rotate. Other line elements (e.g., the diagonal) both stretch and
rotate.
(a)
dR U dr Q dr
(b)
dR Q dR V dr
Fig. 3.12. Polar decomposition of deformation in principal directions. (a) Stretch U

followed by rotation Θ. (b) Rotation Θ followed by stretch V.
principal axes Ni , Eq. (3.135)2 gives

C = Λ21 N1 N1 + Λ22 N2 N2 + Λ23 N3 N3 , (3.148)
where Λi ≡ Λ(Ni ) . We already have seen that C is a symmetric tensor.
Moreover, Eq. (3.25)1 shows that it also is positive definite, i.e.,
C : aa = a · C · a > 0 (3.149)
for all a = 0. (Here, a = dR and dR · C · dR = ds2 > 0.) Next, we define
the tensor
U = Λ1 N1 N1 + Λ2 N2 N2 + Λ3 N3 N3 , (3.150)
so that
C = U · U = U2 . (3.151)
Since C is symmetric and positive definite, so is U. Now, if we can show
that Θ as defined by
Θ = F · U−1
is an orthogonal tensor, then the existence of the first decomposition of

(3.147) will be proven. This equation and Table 2.3 (page 23) give
ΘT · Θ = (F · U−1 )T · (F · U−1 )
= (U−T · FT ) · (F · U−1 )
= U−1 · (FT · F) · U−1
since UT = U. Substituting FT · F = C = U2 yields

ΘT · Θ = U−1 · (U · U) · U−1
= (U−1 · U) · (U · U−1 )
= I,
which shows that Θ is orthogonal under the definition of Eq. (2.119). It

also can be shown that Θ is a proper orthogonal tensor.
The existence of the second decomposition of Eq. (3.147) follows from
the definition of a second-order tensor V in the form
V = Θ · U · ΘT . (3.152)
The manipulations
VT = (Θ · U · ΘT )T = (ΘT )T · UT · ΘT
= Θ · U · ΘT
= V
show that V is a symmetric tensor, and the rigid-body rotations of U to
produce V in (3.152) preserve its positive definite character. To obtain the
second decomposition, we substitute for U from the first decomposition of
(3.147) to get
V = Θ · (Θ−1 · F) · ΘT
= (Θ · Θ−1 ) · F · ΘT
= I · F · Θ−1
= F · Θ−1 ,
as orthogonality implies ΘT = Θ−1 . This relation proves the existence of

the second decomposition of (3.147).
To prove uniqueness of the first decomposition, we suppose that another
decomposition exists such that
F = Θ · U ,
where
U = Λ1 N1 N1 + Λ2 N2 N2 + Λ3 N3 N3
and Θ is a rotation tensor. Then,
C = FT · F = (Θ · U )T · (Θ · U )
= U T · (Θ T · Θ ) · U
= U · U = U 2 .
Since the eigenvalues of C are unique, the Λi must equal the Λi in
Eq. (3.150). Thus, U = U which proves uniqueness. The uniqueness
of Θ and V follows from their definitions.
Finally, we note that substituting Eq. (3.147) into (3.26) yields
C = FT · F = (Θ · U)T · (Θ · U) = (UT · ΘT ) · (Θ · U)
= UT · (ΘT · Θ) · U = UT · I · U
B = F · FT = (V · Θ) · (V · Θ)T = (V · Θ) · (ΘT · VT )
= V · (Θ · ΘT ) · VT = V · I · VT .
Because U and V are symmetric tensors, we have
C = U2 , B = V2 . (3.153)
The first relation again confirms Eq. (3.151).
Physical Derivation. Consider translation and rotation (without defor-

mation) of the differential line element dR, which is oriented in a principal
direction, into the element dR̂ (Fig. 3.12b) according to
dR̂ = Θ · dR + D = dR · ΘT + D, (3.154)
where D is a translation vector and Θ is the rotation tensor. For this
motion, the deformation gradient tensor, given by Eqs. (2.163) and (3.21)1,
is
∂ R̂
F̂T = = ΘT → F̂ = Θ. (3.155)
∂R
Because Eq. (3.154) represents a rigid-body motion, the deformation tensors
must equal the identity tensor. Hence,
Ĉ = F̂T · F̂ = ΘT · Θ = I
B̂ = F̂ · F̂T = Θ · ΘT = I
or
Θ · ΘT = ΘT · Θ = I, (3.156)
which shows that Θ does indeed satisfy the criterion for an orthogonal
(rotation) tensor. Next, the rotated element dR̂ is stretched (or shortened)
into the element dr (Fig. 3.12b) according to
dr = V · dR̂ = V · (Θ · dR + D)
= (V · Θ) · dR + V · D
= dR · (V · Θ)T + V · D, (3.157)
and the deformation gradient tensor for the entire motion is

∂r
FT = = (V · Θ)T → F = V · Θ. (3.158)
∂R
Now, consider a deformation that first transforms the line element dR
into dr̂ (Fig. 3.12a), without rotation, through
dr̂ = U · dR.
Then dr̂ translates and rotates into dr by
dr = Θ · dr̂ + D
= (Θ · U) · dR + D = dR · (Θ · U)T + D, (3.159)
where we stipulate that the rotation tensor Θ is the same one introduced
above. Thus,
∂r
FT = = (Θ · U)T → F = Θ · U. (3.160)
∂R
Equations (3.158) and (3.160) are consistent with the polar decomposition
theorem (3.147).
In summary, the deformation gradient tensor F can be decomposed
into a deformation U followed by a rigid-body rotation Θ or a rigid-body
rotation Θ followed by a deformation V. The stretch tensors U and V
generally provide all the deformation and at least part of the rotation.
In principal directions, however, these tensors include only deformation
(stretch or shortening), with the rotation tensor Θ providing all the rotation
(see Problem 3.14). Although the rotation tensors are the same, U = V in
general. We next investigate some properties of the stretch tensors.
3.8.2 Principal Stretch Ratios

Consider again the mapping of the undeformed line element dR into the
deformed element dr according to
dr = F · dR. (3.161)
The representations
dR = N dS, dr = n ds (3.162)
separate the motion into the rotation of the unit vector N into n and the
stretching of the length dS into ds (see Fig. 3.4, page 73). Substituting
Eqs. (3.162) into (3.161) and dividing by dS yields
Λ(N) n = F · N or λ(n) n = F · N, (3.163)
where Λ(N) = λ(n) = ds/dS is the stretch ratio of the line element originally
pointing in the direction of N, as defined in Sec. 3.6.1.
If N and n correspond to principal directions of strain, then N rotates
into n by
n = Θ · N, (3.164)
where Θ is the rotation tensor. Direct substitution into this equation veri-
fies the dyadic representation
Θ = nN. (3.165)
Now, inserting Eqs. (3.147) and (3.164) into (3.163)1 gives
Λ(N) Θ · N = (Θ · U) · N
or
Θ · (U − Λ(N) I) · N = 0.
For an arbitrary rotation Θ, this relation provides the eigenvalue problem
(U − Λ(N) I) · N = 0 (3.166)
for the principal stretch ratios Λ(N) in the unrotated principal directions
N at R in the undeformed body B. Recall, from Eq. (3.147), that the
deformation tensor U operates on dR before the rotation due to Θ.
Similarly, substituting Eq. (3.147) into (3.163)2 yields
λ(n) n = (V · Θ) · N = V · (Θ · N) = V · n,
in which Eq. (3.164) has been used. This equation gives the eigenvalue
problem
(V − λ(n) I) · n = 0 (3.167)
for the principal stretch ratios λ(n) in the rotated principal directions n at
r in the deformed body b. This is consistent with the fact that V gives a
deformation after the rotation due to Θ.
Here, we make several observations. First, comparing Eqs. (3.166) and
(3.167) with Eqs. (3.133) suggests that the eigenvalues of the strain tensors
C and B are equal to the squares of the eigenvalues (principal stretches) of
the stretch tensors U and V, respectively. Equations (3.148) and (3.150)
also are consistent with this assertion. To show that this is indeed the case,
we substitute Eqs. (3.153) into (3.133) to get
(U2 − Λ2(N) I) · N = 0
2
(V − λ2(n) I) · n = 0.
These equations can be written
(U + Λ(N) I) · (U − Λ(N) I) · N = 0
(V + λ(n) I) · (V − λ(n) I) · n = 0,
which are satisfied by Eqs. (3.166) and (3.167). In addition, since the
eigenvalues of C and B are equal, so are the eigenvalues of U and V.
Moreover, the principal directions of C and B correspond to those of U
and V, respectively. Finally, as shown by Eq. (3.164), the tensor Θ rotates
the principal directions of U (and C) into those of V (and B).
In summary, the tensors C, B, U, and V are all symmetric and positive
definite. The eigenvalues of C and B are Λ2(N) = λ2(n) , and those of U and
V are Λ(N) = λ(n) , which are the principal stretch ratios. Moreover, Θ
rotates the equal and mutually orthogonal principal directions N(i) of C
and U at point P in the undeformed body into the equal and orthogonal
principal directions n(i) of B and V at the image p of P in the deformed
body. Thus, we can extend Eqs. (3.164) and (3.165) to write
n(i) = Θ · N(i)

Θ= n(i)N(i) . (3.168)
i
Then we have, for example, n(1) = Θ · N(1) = [n(1)N(1) + n(2)N(2) +

n(3)N(3) ] · N(1) = n(1), since N(1) · N(2) = N(1) · N(3) = 0.
Example 3.13 Consider constrained uniaxial deformation of a unit cube

given by F = λ ex ex + ey ey + ez ez . For the decomposition F = Θ · U,
find Θ and U and determine how much each of these tensors stretches and
rotates a cube diagonal (dashed line in Fig. 3.11, page 109).
Solution. For the prescribed deformation, the principal directions align
with the coordinate axes, giving Ni = ni = ei (i = x, y, z). Thus,
Eq. (3.168)2 gives
Θ = ni Ni = ex ex + ey ey + ez ez = I,
and we find
U = Θ−1 · F = ΘT · F = F.
For the undeformed diagonal in Fig. 3.11 (left) defined by the vector a0 =
ex + ey , the right stretch tensor transforms a0 into the vector
a = U · a0 = λex + ey ,
which corresponds to the diagonal in the deformed cube (Fig. 3.11, right).
The rotation tensor, being the identity tensor, has no effect. In this simple
problem, therefore, U provides the entire stretch and rotation of the diag-
onal, illustrating how U can both deform and rotate line elements oriented
in non-principal directions (see also Problem 3.14).
Example 3.14 For simple shear of a cube (Example 3.3, page 80), deter-
mine the following:
(a) The eigenvalues and eigenvectors for the deformation tensors C and B.
(b) The rotation tensor Θ and the stretch tensors U and V.
Solution. (a) The eigenvalue problems are given by Eqs. (3.133). For
nontrivial solutions, we must have
det(C − Λ2(N) I) = 0
det(B − λ2(n) I) = 0 (3.169)
or

det CJI − Λ2(N) δ IJ = 0

det Bji − λ2(n) δ ij = 0. (3.170)
Because the GI and gi are Cartesian (unit) base vectors in this problem
(see Example 3.3), we can write CJI = CIJ and Bji = Bij . Thus, Eqs. (3.56)
and (3.170) give

1 − Λ2 k 0
(N)
2 2
k 1 + k − Λ(N) 0 = 0

0 0 1 − Λ(N)
2

1 + k 2 − λ2
(n) k 0

k 1 − λ2(n) 0 = 0. (3.171)

0 0 1 − λ2(n)
These characteristic equations provide the same eigenvalues, and so Λ(Ni ) =
λ(ni ) ≡ λ(i) for the ith eigenvalue. This is expected, because the λ(i) rep-
resent principal stretch ratios for an element that rotates from orientation
N(i) to orientation n(i) during deformation. We also note that expanding
(3.171)1 gives
−Λ6(N) + I1 Λ4(N) − I2 Λ2(N) + I3 = 0
with the invariants given by Eq. (3.145). This agrees with the general
characteristic equation (2.82) with λ = Λ2(N) .
For convenience, we now let

a = 1 + 14 k 2 , b = a−1 .
Then, Eqs. (3.171) yield the principal stretch ratios
λ2(1,2) = 1 + 12 k 2 ± ka, λ(3) = 1, (3.172)
and Eqs. (3.56) and (3.133) provide the corresponding principal directions

e1 + 12 k ± a e2
N(1,2) = , N(3) = e3
2 + 12 k 2 ± ka

e1 + − 12 k ± a e2
n(1,2) = , n(3) = e3 . (3.173)
2 + 12 k 2 ∓ ka
(The denominators make N(1,2) and n(1,2) unit vectors.) It is easy to show
that the N(i) are mutually orthogonal, as are the n(i).
For the special case k = 0.5, these equations give λ(1) = 1.28, λ(2) =
0.781, N1 = 0.615e1 + 0.788e2, N2 = 0.788e1 − 0.615e2, n1 = 0.788e1 +
0.615e2, and n2 = 0.615e1−0.788e2 . The eigenvectors (principal directions)
are shown in Fig. 3.13.
X2,x2
N1
n1
N2
n2
X1,x1
0
Fig. 3.13. Principal directions for simple shear of a unit cube (k = 0.5).
(b) Equations (3.168)2 and (3.173) provide the rotation tensor Θ =

n(1)N(1) + n(2)N(2) + n(3) N(3) , or
⎡ 1 ⎤
b 2 kb 0
Θ(ei ej ) = ⎣ − 12 kb b 0 ⎦. (3.174)
0 0 1
This tensor rotates the orthogonal triad N(i) into n(i) (see Problem 3.6).
The above matrix indicates that the rotation is about the X 3 -axis.
With Θ known, Eqs. (3.147) yield
U = Θ−1 · F = ΘT · F
V = F · Θ−1 = F · ΘT (3.175)
as Θ−1 = ΘT due to orthogonality. Substituting Eqs. (3.55) and (3.174)

gives
⎡ ⎤⎡ ⎤
b − 12 kb 0 1 k 0
U(ei ej ) = ⎣ 1 ⎦ ⎣ 0 ⎦
2 kb b 0 0 1
0 0 1 0 0 1
⎡ 1 ⎤
b 2
kb 0
= ⎣ 12 kb (1 + 12 k 2 )b 0 ⎦ , (3.176)
0 0 1
which is symmetric. The left stretch tensor V can be found similarly (see
Problem 3.6).
Finally, we mention some other calculations that can be carried out as
checks. Equation (3.56) can be used to show that C = U2 = U · U. Also,
since the N(i) and n(i) correspond to the directions of the principal stretch
ratios, Eqs. (3.56), (3.94), and (3.173) should give
−1
λ2(i) = N(i) · C · N(i) = n(i) · B−1 · n(i) .
Moreover, the shears should vanish in the principal directions, and

Eqs. (3.110) should give Γ(12) = Γ(23) = Γ(31) = 0 and γ (12) = γ (23) =
γ (31) = 0, which are true if N(i) · C · N(j) = n(i) · B−1 · n(j) = 0 for i = j.
Verifying these relations is left to the interested, diligent, or skeptical reader.
Example 3.15 An incompressible tube with undeformed inner and outer

radii a0 and b0 , respectively, is everted, i.e., turned inside out, with the
inner surface becoming the outer surface and vice versa. This geometry
can be created by cutting the tube longitudinally, bending it to reverse
the curvature, and gluing the cut surfaces together (Fig. 3.14). With end
effects neglected, this deformation is described by
r = r(R)
θ = π−Θ
z = λZ
in cylindrical coordinates. Determine the deformation gradient tensor F,

the stretch tensor U, and the rotation tensor Θ.
ST
P
R p r
T
P cut glue P
a0 b
b0 a
Undeformed Everted
Fig. 3.14. Turning a tube inside out.
Solution. Using the deformed position vector
r = rer + zez ,
3.9. Approximations 119
we get
∂ 1 ∂ ∂
∇r = eR + eΘ + eZ [r (R)er (θ) + z(Z)ez ]
∂R R ∂Θ ∂Z
∂r r ∂er ∂θ ∂z
= eR e r + eΘ + eZ .
∂R R ∂θ ∂Θ ∂Z
The given mapping and Eq. (2.12)2 yield
F = (∇r)T = λR er eR − λΘ eθ eΘ + λZ ez eZ ,
where
∂r r
λR = , λΘ = , λZ = λ.
∂R R
Because F contains no shear terms, the base vectors ei in the deformed
configuration correspond to principal directions of strain. In most cases,
the components of F in principal coordinates are stretch ratios. Stretch
ratios, however, must always be positive, whereas the eθ eΘ component of
F is negative.
The principal stretch ratios are provided by solving the eigenvalue prob-
lem (3.169)1 for the right Cauchy-Green deformation tensor
C = FT · F = λ2R eR eR + λ2Θ eΘ eΘ + λ2Z eZ eZ .
Doing this reveals that the principal stretch ratios are the λI given above.
Consequently, in principal coordinates, the stretch tensor U = C1/2 takes
the form
U = λR eR eR + λΘ eΘ eΘ + λZ eZ eZ .
With this result, the rotation tensor is given by
Θ = F · U−1

= (λR er eR − λΘ eθ eΘ + λZ ez eZ ) · λ−1 −1 −1
R eR eR + λΘ eΘ eΘ + λZ eZ eZ
= er e R − eθ e Θ + ez e Z .
3.9 Approximations
Geometric nonlinearity in the exact strain-displacement relation

E = 21 [∇u + (∇u)T + (∇u) · (∇u)T ] (3.177)
can be a source of substantial analytical complication. Thus, it is advanta-
geous to seek simplifying approximations when possible.
This section considers the following cases:
• Small displacement. Example: Any problem in the linear theory of

elasticity, e.g., bone.
• Small deformation, arbitrary displacement and rotation. Exam-
ple: The “elastica” problem, i.e., large bending of thin beams such as
cilia or actin filaments.
• Small deformation, moderate displacement and rotation. Ex-
ample: Deflection a few times the thickness in a thin plate or shell, e.g.,
deformation of the cornea by an indenter (tonometer) to measure intraoc-
ular pressure.
• Small rotation, arbitrary deformation. Example: Local load on a
half space, e.g., a tumor pressing against an organ.
The approximations considered here are valid for general 3-D geometries.
Of course, further simplification may be possible by taking advantage of
symmetry, thinness, or other features of a particular problem. Note that
we consider “deformation” equivalent to “strain.”
The meanings of the terms “small” and “moderate” are not always ob-
vious. Thus, we give general rules of thumb, based on the magnitudes of
the strain and rotation tensors, |E| = (E : E)1/2 and |Θ| = (Θ : Θ)1/2 ,
which are analogous to the magnitude of a vector, |a| = (a · a)1/2 . In addi-
tion, since displacement can include rigid-body translation, the magnitude
of the “displacement” is actually based on the magnitude of the displace-
ment gradient, |∇u| = [(∇u) : (∇u)]1/2. With these definitions, “small”
means 0.01 and “moderate” means 0.2.
3.9.1 Small Displacement

If displacements are small, then the convected coordinate system differs
little from the undeformed system, and we can take Gi ∼ = gI ∼ = gi and
I ∼ i
X = x . For |∇u| << 1, Eq. (3.177) provides the linear strain tensor
1

E∗ = 2 (∇u)T + (∇u) . (3.178)
In addition, as shown in Appendix A, the linear rotation tensor is given

by
1

Θ∗ = 2
(∇u)T − (∇u) , (3.179)
which represents the average rotation of all line elements passing through
a point.
3.9. Approximations 121
These relations show that E∗ = E∗T , Θ∗ = −Θ∗T , and
(∇u)T = E∗ + Θ∗
∇u = E∗ − Θ∗ . (3.180)
Thus, Eq. (3.77)1 gives
F = I + (∇u)T ∼
= I + E∗ + Θ∗ . (3.181)
With the 1-D stretch ratio being λ = 1 + E ∗ by (3.2), this equation shows
that the deformational and rotational parts of F add, rather than multiply,
in the linear theory.
3.9.2 Small Deformation

So far, we have examined the two extremes of strain analysis, i.e., the
full nonlinear theory and the linearized version. Developing intermediate
theories is facilitated by expressing E in terms of E∗ and Θ∗ . Substituting
Eqs. (3.180) into (3.177) yields
E = E∗ + 12 (E∗ − Θ∗ ) · (E∗ + Θ∗ ), (3.182)
which is an exact expression. Here, we consider the case where deformation

is small, but no restrictions are made on the magnitudes of displacements
or rotations.
For small deformation, we assume |E∗ · E∗ | << |E∗ |, and (3.182) gives
the approximation
E∼
= E∗ + 12 (E∗ · Θ∗ − Θ∗ · E∗ − Θ∗ · Θ∗ ). (3.183)
In addition, (3.142)3 provides the dilatation ratio, with

J 2 = det δ IJ + 2EJI = 1 + 2(E11 + E22 + E33 ) + · · ·
giving
J∼
= 1 + EII . (3.184)
3.9.3 Small Deformation and Moderate Rotation

If rotation is “moderate” and strain is “small,” terms like |E∗ ·Θ∗ | (∼ 0.002)
could be ignored compared to |E∗ | (∼ 0.01), but |Θ∗ · Θ∗ | (∼ 0.04) cannot.
In this case, Eq. (3.182) reduces to
E∼
= E∗ − 12 Θ∗ · Θ∗ = E∗ + 12 Θ∗ · Θ∗T . (3.185)
3.9.4 Small Rotation

If rotations are small but deformation is arbitrary, then |Θ∗ · Θ∗ | << |E∗ |
and Eq. (3.182) gives the approximation
E∼
= E∗ + 12 (E∗ · E∗ + E∗ · Θ∗ − Θ∗ · E∗ ). (3.186)
3.10 Deformation Rates
The state of stress in an elastic body depends only on its deformed configu-
ration, regardless of how or how fast it got there. Soft tissues generally are
viscoelastic, however, with viscous losses accompanying the deformation. In
such materials, the final stress state depends on the entire history and rate
of the deformation. Even for elastic bodies, moreover, rate formulations
often are advantageous computationally, and we will use rate equations to
develop constitutive relations for elastic materials in Chap. 5. This section
presents measures for the rate of deformation relative to a given frame of
reference. First, however, we discuss time differentiation of Lagrangian and
Eulerian quantities.
3.10.1 Time Rates of Change

The mechanics of differentiating a physical quantity, such as density or
temperature, with respect to time depends on whether the quantity is con-
sidered a Lagrangian or an Eulerian variable. In the Lagrangian descrip-
tion, a particle originally located at position R is followed through space
and time, while the Eulerian description monitors time-dependent changes
at a particular spatial position r. In general, as a body deforms, different
particles pass through r. Since it is relatively easy in general to identify
a reference configuration in a solid body, Lagrangian variables usually are
3.10. Deformation Rates 123
used in solid mechanics problems. Researchers in fluid dynamics, however,

usually prefer to work with Eulerian variables.
Consider a scalar quantity φ(R, t) = φ(r, t). In general, the partial
derivatives ∂φ(R, t)/∂t and ∂φ(r, t)/∂t differ. The former derivative is the
time rate of change of φ following a particle (R = constant), while the latter
is the rate of change of φ at a fixed point in space (r = constant). As a
particle passes through the point located at r, it instantaneously possesses
the property φ(r, t).
Usually, we want to know how φ changes for a given particle. Since R
is a constant, the Lagrangian description gives the time derivative
dφ(R, t) ∂φ(R, t)
= . (3.187)
dt ∂t
Following the motion of a particle in the Eulerian description means that
its position r changes with time. In this case, the chain rule yields
dφ(r, t) ∂φ(r, t) ∂r ∂φ(r, t)
= + · ,
dt ∂t ∂t ∂r
where Eq. (2.156) has been used in taking the derivative with respect to
the vector r. Since r = r(R, t), the velocity of a particle is
dr ∂r
v= = , (3.188)
dt ∂t
¯ Thus,
and Eq. (3.18)2 gives ∂/∂r = ∇.
dφ(r, t) ∂φ(r, t) ¯
= + v · ∇φ(r, t), (3.189)
dt ∂t
in which the first term represents the time rate of change of φ at a fixed
position r, and the second term gives the rate of change as the particle is
convected through space.
The physical meaning of the convection term in (3.189) is illustrated by
considering the following. Suppose φ has a different value but is constant at
each point in space. Then, ∂φ/∂t = 0 at each point, but a particle traveling
through space takes on different local values of φ as it passes through each
¯ = 0.
point, i.e., dφ/dt = v · ∇φ
With the definition
d ∂ ¯
≡ + v · ∇, (3.190)
dt ∂t
dφ/dt becomes either Eq. (3.187) or Eq. (3.189), depending on whether

¯
φ = φ(R, t) or φ = φ(r, t) [since ∇φ(R, t) = 0]. In component form, with
v = v gi and ∇
i ¯ = g ∂/∂x , Eq. (3.190) becomes
i i
d ∂
= + vi ( ),i . (3.191)
dt ∂t
Consider now the time derivative of a vector. We compute, for example,
the acceleration of a particle as
dv
a= . (3.192)
dt
For the Lagrangian and Eulerian description, respectively, Eq. (3.190) gives
dv(R, t) ∂v
a= =
dt ∂t
dv(r, t) ∂v ¯
a= = + v · ∇v. (3.193)
dt ∂t
Vector and tensor components are written most conveniently in terms of

the undeformed basis {GI } (coordinates X I ) for Lagrangian quantities and
in terms of the deformed basis {gi } (coordinates xi ) for Eulerian quantities.
Both of these coordinate systems are fixed in the reference frame and so
are independent of time. In contrast, since the base vectors gI move with
the deformation, they are time-dependent. Thus, we take
v(R, t) = vI (X I , t) GI , v(r, t) = vi (xi , t) gi
(3.194)
a(R, t) = aI (X I , t) GI , a(r, t) = ai (xi , t) gi .
Substituting these expressions into Eq. (3.193)1 gives simply
∂vI
aI = , (3.195)
∂t
but finding ai requires a little more work. First, we note that Eq. (2.170)
gives
¯ = v i | j g j gi ,
∇v
and inserting this relation and Eq. (3.194) into (3.193)2 yields
∂ i
a = (v gi ) + (vk gk ) · (vi |j gj gi )
∂t
∂vi
= gi + vk vi |j δ jk gi
∂t
∂vi
= + v j v i | j gi .
∂t
Thus, we can write
dv Dvi
a= = gi , (3.196)
dt Dt
where
D ∂
≡ + vj ( )|j (3.197)
Dt ∂t
is called the material derivative or the convected derivative. As

Eqs. (3.192) and (3.196) indicate, the material derivatives of vj are the
contravariant components of dv/dt with respect to the deformed basis. Fi-
nally, Eqs. (3.194) and (3.196) give
Dvi
ai =. (3.198)
Dt
Comparing Eqs. (3.191) and (3.197) reveals that the operators d/dt and
D/Dt are not quite the same. The former is to be used when differenti-
ating vectors and tensors, whereas the latter is for differentiating vector
or tensor components. They are equivalent, however, in Cartesian coordi-
nates or when the argument is a scalar function. In these cases, there is no
distinction between covariant differentiation and partial differentiation.
Example 3.16 At t = 0, a particle is located at the Cartesian coordinates

XI , and for t ≥ 0, it occupies the coordinates
x1 = X1 (1 + at2 X2 )
x2 = X2
x3 = X3 , (3.199)
where a is a constant. (For convenience, only subscripts are used here.)
Compute the displacement, velocity, and acceleration vectors for the parti-
cle.
Solution. With the position vectors being
t=0: R = X1 e1 + X2 e2 + X3 e3
t≥0: r = x 1 e1 + x 2 e2 + x 3 e3 , (3.200)
the displacement vector is
u = r − R = at2 X1 X2 e1 . (3.201)
In terms of the material coordinates XI , the velocity vector is

dr du
v= = = 2atX1 X2 e1 . (3.202)
dt dt
The velocity can be expressed in terms of the spatial coordinates xi by
solving Eqs. (3.199) for XI (xi ) and substituting into (3.202) to obtain
2atx1 x2
v= e1 . (3.203)
1 + at2 x2
Finally, we compute the acceleration of the particle by differentiating
both the Lagrangian representation (3.202) and the Eulerian representation
(3.203) for the velocity. We should obtain the same answer in both cases.
In the first case, Eq. (3.193)1 gives
dv ∂v
a= = = 2aX1 X2 e1 . (3.204)
dt ∂t
In the second case, Eq. (3.193)2 gives
dv ∂v ¯
a= = + v · ∇v,
dt ∂t
where Eq. (3.203) provides

∂v 1 2at2 x2
= 2ax1 x2 − e1
∂t 1 + at2 x2 (1 + at2 x2 )2

¯ 2atx1 x2 ∂ 2atx1 x2 4a2 t2 x1 x22
v · ∇v = 2 2
e1 = e1
1 + at x2 ∂x1 1 + at x2 (1 + at2 x2 )2
¯ = e1 ∂/∂x1 + e2 ∂/∂x2 + e3 ∂/∂x3 . Combining these relations yields
since ∇
2ax1 x2
a= e1 , (3.205)
1 + at2 x2
and substituting Eqs. (3.199) for x1 and x2 shows that this relation is
equivalent to (3.204).
3.10.2 Rate-of-Deformation and Spin Tensors

We now examine the deformation rate in the current configuration, i.e.,
using the Eulerian description. We begin by taking the time derivative
d/dt, denoted by a superposed dot, of the deformed element dr = F · dR
to get the differential velocity vector
dv ≡ dṙ = Ḟ · dR = Ḟ · (F−1 · dr)
since dR is constant. This expression can be written

dv = L · dr = dr · LT , (3.206)
where
L ≡ Ḟ · F−1 . (3.207)
In addition, with (2.163)1 and (3.18)2, Eq. (3.206) gives
∂v ¯
LT = = ∇v. (3.208)
∂r
The tensor L is called the velocity gradient tensor.

In terms of components with respect to the deformed body, the velocity
vector is given by
∗
v = vi gi = vi gi = vI gI = vI ∗ gI , (3.209)
and Eqs. (2.150) and (3.18)2 yield, for example,
¯ = gj v,j = gj (vi ||j gi ) = vi ||j gj gi ,
LT = ∇v
where modifying Eq. (2.151)1 gives
vi ||j = vi ,j +vk Γ̄ijk .
This and similar manipulations yield
L = vi ||j gi gj = vi ||j gi gj
∗
= vI ||J gI gJ = vI ∗ ||J gI gJ . (3.210)
The rate-of-deformation tensor (or stretch rate tensor) and the spin
tensor (or vorticity tensor) are defined to be the symmetric and antisym-
metric components of L, respectively, i.e. [see Eqs. (2.75)]
D ≡ 12 (L + LT ) = DT
Ω ≡ 12 (L − LT ) = −ΩT . (3.211)
Substituting Eq. (3.208) gives

1

D ≡ (∇v) ¯
¯ T + (∇v)
2
1

Ω ≡ ¯ T − (∇v)
(∇v) ¯ , (3.212)
2
which resemble the linear strain and rotation tensors of Eqs. (3.178) and
¯ ∼
(3.179). In fact, in the linear case with ∇ = ∇ and v ∼
= u̇, we find
D∼
= Ė∗ , Ω∼
= Θ̇∗ . (3.213)
There is, however, an important distinction. Whereas Eqs. (3.178) and
(3.179) are approximate relations when displacement gradients are small,
Eqs. (3.212) are exact for arbitrarily large velocity gradients. Also, adding
Eqs. (3.211) gives
L = D + Ω, (3.214)
which agrees with a time derivative of Eq. (3.181)1, with Ḟ ∼= (∇u̇)T =

T
(∇v) = L in the linear case.
Consider now the time derivative of ds2 = dr · dr given by
d
(ds2 ) = dṙ · dr + dr · dṙ = 2 dr · dv
dt
= 2 dr · L · dr, (3.215)
in which Eq. (3.206) has been substituted. Inserting Eq. (3.214) yields
d
(ds2 ) = 2 dr · D · dr (3.216)
dt
since
2 dr · Ω · dr = dr · (L − LT ) · dr
= dr · (L · dr) − (dr · LT ) · dr
= dr · dv − dv · dr
= 0,
where Eqs. (3.206) and (3.211)2 have been used. Thus, D characterizes the
rate of change of ds2 . Next, substituting Eq. (3.19) into (3.216) gives
d
(ds2 ) = 2 (dR · FT ) · D · (F · dR)
dt
= 2 dR · (FT · D · F) · dR,
and differentiating (3.27) yields
d d
(ds2 − dS 2 ) = (ds2 ) = 2 dR · Ė · dR.
dt dt
Comparing these relations defines the Lagrangian strain-rate tensor as
Ė = FT · D · F. (3.217)
Similarly, Eqs. (3.27)2 and (3.206) give

d d d
(ds2 − dS 2 ) = (ds2 ) = (2dr · e · dr)
dt dt dt
= 2 (dṙ · e · dr + dr · ė · dr + dr · e · dṙ)
= 2 (dr · LT · e · dr + dr · ė · dr + dr · e · L · dr)
= 2 dr · (LT · e + ė + e · L) · dr,
and comparison with Eq. (3.216) gives the Eulerian strain-rate tensor
ė = D − e · L − LT · e. (3.218)
Note that ė = de/dt, with d/dt defined by Eq. (3.190).
Example 3.17 Of course, deformation rate tensors are particularly useful

in rate-dependent problems, e.g., if the material is viscoelastic. For a cir-
cular tube undergoing combined extension, inflation, and torsion (Example
3.4, page 82), derive (a) the velocity vector v in terms of natural com-
ponents, (b) the velocity gradient tensor L, and (c) Ḟ by differentiating
Eq. (3.62) directly, as well as using Ḟ = L · F from Eq. (3.207).
Solution. (a) Including time dependency, we write Eqs. (3.57) as
r = r(R, t), θ = Θ + ψ(t)Z, z = λ(t)Z, (3.219)
which can be inverted to give
R = R(r, t), Θ = θ − ψz/λ, Z = z/λ. (3.220)
Taking time derivatives of Eqs. (3.219) yields
ṙ = ṙ(R(r, t), t) = ṙ(r, t)
θ̇ = ψ̇(t) z/λ(t) = θ̇(z, t)
ż = λ̇(t) z/λ(t) = ż(z, t) (3.221)
to be used in an Eulerian formulation.
Now, with the deformed position vector given by r = rer + zez , the
velocity of a material element is
v = ṙ = ṙer + r ėr + żez ,
where the time derivatives of the unit vectors er and eθ , found from
Eqs. (3.59), are (ez =constant)
ėr = θ̇eθ , ėθ = −θ̇er . (3.222)

Combining these relations gives

v = ṙer + θ̇r eθ + ż ez = vi gi . (3.223)
With the gi given by Eqs. (3.60), the natural velocity components are
v1 = ṙ, v2 = θ̇, v3 = ż. (3.224)
(b) The velocity gradient tensor can be computed by inserting
Eq. (3.223) into (3.208) and taking derivatives of the components and base
vectors directly (see Problem 3.19). For a change of pace, however, we use
Eq. (3.210) in the form
L = vi ||j gi gj , (3.225)
where modifying Eq. (2.151)1 gives
vi ||j = vi ,j +vk Γ̄ijk (3.226)
with the Γ̄ijk defined by Eq. (3.86)2 . Since the base vectors gi correspond
to the cylindrical polar coordinate system (r, θ, z), Eqs. (2.148) provide the
only non-zero Christoffel symbols:
Γ̄122 = −r, Γ̄212 = Γ̄221 = r −1 . (3.227)
Inserting these relations and Eqs. (3.224) into (3.226) yields
⎡ ⎤
ṙ,r −r θ̇ 0
⎢⎢ ⎥
⎥
L(gigj ) = vi ||j = ⎢ θ̇ ṙ
θ̇,z ⎥. (3.228)
⎣ r r ⎦
0 0 ż,z
This equation can be used in Eqs. (3.211) to compute D and Ω.
(c) In the first approach, differentiating Eq. (3.62) gives (eR , eΘ , and
eZ = ez are constants with respect to time)
ṙ r
Ḟ = ṙ,R er eR + r,R ėr eR + eθ eΘ + ėθ eΘ
R R
+(ψ̇r + ψṙ)eθ eZ + ψr ėθ eZ + λ̇ez eZ ,
and substituting Eqs. (3.222) produces
⎡ ⎤
r θ̇
⎢ ṙ,R − R −ψr θ̇ ⎥
⎢ ⎥
Ḟ(ei eI ) = ⎢
⎢ r,R θ̇ ṙ ⎥. (3.229)
⎣ (ψ̇r + ψ ṙ) ⎥
⎦
R
0 0 λ̇
In the second approach, substituting Eqs. (3.22)1 and (3.225) yields

Ḟ = L · F = (vi ||j gi gj ) · (FK
k
gk G K )
= vi ||j FK
k
gi δ jk GK
j
= vi ||j FK gi G K .
Thus, Eqs. (3.64) and (3.228) give
j
Ḟ(gi GK ) = vi ||j FK
⎡ ⎤⎡ ⎤
ṙ,r −r θ̇ 0 r,R 0 0
⎢ ⎥⎢ ⎥
⎢ ⎥⎢
= ⎢ θ̇ ṙ
θ̇,z ⎥⎣ 0 1 ψ ⎥
⎦
⎣ r r ⎦
0 0 ż,z 0 0 λ
⎡ ⎤
ṙ,r r,R −r θ̇ −rψ θ̇
⎢ ⎥
⎢ ⎥
= ⎢ θ̇r,R ṙ ṙ
ψ + λθ̇,z ⎥ . (3.230)
⎣ r r r ⎦
0 0 λż,z
Showing that Eqs. (3.229) and (3.230) are identical requires some work.
They first need to be expressed in terms of the same basis. One way to do
this is to write the latter equation in the form
⎡ ⎤
Ḟ11 g1 G1 Ḟ21 g1 G2 Ḟ31 g1 G3
⎢ ⎥
Ḟ = ⎢ 2 1 2 2 2
⎣ Ḟ1 g2 G Ḟ2 g2 G Ḟ3 g2 G ⎦ .
3 ⎥
Ḟ13 g3 G1 Ḟ23 g3 G2 Ḟ33 g3 G3

Then, on substituting Eqs. (3.60) and (3.61), Eq. (3.230) becomes
⎡ r ⎤
r,R ṙ,r − θ̇ −rψ θ̇
⎢ R ⎥
⎢ ṙ ⎥
Ḟ(ei eI ) = ⎢ r,R θ̇ ṙψ + λr θ̇, ⎥. (3.231)
⎣ R
z ⎦
0 0 λż,z
With the exception of Ḟ11 , Ḟ32 , and Ḟ33 , this expression is the same as
Eq. (3.229). These other terms can be shown to be equivalent by noting
that
∂ ṙ ∂r ∂ ṙ
r,R ṙ,r = = = ṙ,R
∂r ∂R ∂R
λr θ̇,z = λr(ψ̇/λ) = rψ̇
λż,z = λ(λ̇/λ) = λ̇, (3.232)

where the chain rule and Eqs. (3.221) have been used. Putting these rela-
tions into (3.231) reproduces Eq. (3.229).
3.11 Compatibility Conditions
In Eqs. (3.91), six independent components of the strain tensor E (or e)

are related to three components of the displacement vector u. If u is given,
then all the strain components can be computed directly. However, if E is
prescribed, the strain-displacement equations provide a system of six dif-
ferential equations for only three unknown displacements. In this case, as
in the linear theory of elasticity, the strain components must satisfy com-
patibility conditions to ensure a single-valued continuous displacement
field.
In the linear theory, the compatibility conditions can be found by
eliminating the displacements from the strain-displacement relations (see
Sec. A.2.4). In the nonlinear theory, this procedure is virtually hopeless.
An alternate derivation uses a theorem of Riemann (Eringen, 1962), which
states that for a symmetric tensor T to be a metric tensor for a Euclidean
space, it is necessary and sufficient that T be nonsingular and positive defi-
(T) m
nite and that the Riemann-Christoffel tensor R·ijk formed from it vanish
identically. We noted previously [see remarks following Eq. (2.177)] that
this condition implies that the order of covariant differentiation is imma-
terial. Since C is a positive definite tensor, this theorem and Eq. (2.177)
yield
(C) M (C) M (C) M (C) M (C) L (C) M (C) L
R.IJK = ΓIK ,J − ΓIJ ,K + ΓLJ ΓIK − ΓLK ΓIJ = 0,
(3.233)
where modifying Eqs. (2.142) and (2.144) gives
(C)
2ΓIJK = CJK ,I +CKI ,J −CIJ ,K
(C) L
ΓIJ = C KLΓIJK . (3.234)
If the CIJ satisfy Eq. (3.233), then they also represent the components of
the deformed metric tensor, as indicated by Eq. (3.32)1. Thus, Eq. (3.233)
provides 34 = 81 compatibility conditions. Fortunately, using CIJ = CJI
and ΓK K
IJ = ΓJI reduces the number of independent and nonidentically van-
ishing equations to six, as in the linear theory [see Eqs. (A.86)]. Similar
equations can be found for EIJ (CIJ = δ IJ + 2EIJ ) and eij (in terms of
Bij ).
3.12. Problems 133
3.12 Problems
Most of the following problems involve orthogonal coordinate systems in

which there is no distinction between covariant and contravariant compo-
nents. Unless the problem statement explicitly references contravariant
components or base vectors, please feel free to use only subscripts, as well
as the summation convention for repeated indices defined in Appendix A.
Also note that, for simplicity, units are ignored, so do not be alarmed by
finding unrealistic values for some quantities or terms like 1 + x with in-
consistent dimensions (unless x is defined as a dimensionless coordinate).
Finally, symbolic manipulation software may be useful for certain problems.
3.1 Consider a body in a state of plane strain relative to the xy-plane,

i.e., Ezx = Ezy = Ezz = 0. Assume that all components of the strain
tensor E are known relative to the Cartesian axes (x, y, z). If another
set of axes (x̄, ȳ, z̄) is defined by rotating the (x, y, z) axes by an angle
θ about the z-axis, use the general formula Eij = gi · E · gj to compute
the strain components relative to the rotated axes.
3.2 During axisymmetric deformation in two dimensions, a point in a body
moves from polar coordinates (R, Θ) to (r, θ). The displacement vector
can be written
u = uR (R)eR + uΘ (R)eΘ ,
and the gradient operator is
∂ eΘ ∂
∇ = eR +
∂R R ∂Θ
relative to the undeformed configuration. Determine the dyadic repre-
sentations of the deformation gradient and Lagrangian strain tensors
in terms of the displacement components uR and uΘ .
3.3 During deformation, the points in a body move from (X1 , X2 , X3 ) to
x1 = X 1 + 2X 2 X 3 , x2 = X 2 − 2X 1 X 3 , x3 = X 3 ,
where X I and xi are Cartesian coordinates.
(a) Determine the base vectors gi , gi , GI , GI , gI , and gI . Show that
gI · gJ = δ JI .
(b) Use Eq. (3.15)1 to compute the deformation gradients FIi and show
that F = FIi gi GI = gI GI .
(c) Compute the covariant components CIJ of C using each of the
relations C = FT · F, CIJ = gIJ , and CIJ = FIi FJj gij .
(d) Compute the stretch ratio for a fiber that originally is located at the
coordinates (-0.2,0.1,-0.1) and points in the direction of the vector
e1 + 2e3 .
(e) For the same pointas in (d), compute the dilatation ratio using the
expressions J = j g/G and J = det F.
3.4 In material Cartesian coordinates, the displacement field in a body is
u = X2 X3 e1 + X32 e2 .
Compute the strain tensors E(Xi ) and e(xi ).
3.5 In Example 3.5 (page 86), two deformation paths were considered for
extension, inflation, and torsion of a circular tube. Show that both
paths lead to the same total F as given by Eq. (3.62).
3.6 Consider simple shear of a cube (see Example 3.14, page 115).
(a) Show that the rotation tensor of Eq. (3.174) transforms the eigen-
vectors N(i) of Eqs. (3.173) into the vectors n(i).
(b) With C given by (3.56)1 and U by (3.176), confirm the relation
C = U2 .
(c) Compute the left stretch tensor V. Confirm that the relation B =
V2 is consistent with Eq. (3.56)2 .
(d) Show that the shear strains E12 and e12 vanish relative to the prin-
cipal axes of strain.
3.7 A thin membrane with middle surface lying in the xy-plane under-
goes deformation with a displacement field given by u = u(x, y)ex +
v(x, y)ey . For small deformation, use Eq. (3.183) to write the approxi-
mate 2-D strain-displacement relations in Cartesian coordinates.
3.8 In Cartesian coordinates, a deformation is defined by
x1 = X1 cos θ + X2 sin θ
x2 = −X1 sin θ + X2 cos θ
x3 = X3 .
Compute C, E, U, and Θ. Explain the result.
3.9 Relative to Cartesian coordinates, the state of strain at a point in a
body is given by
⎡ ⎤
0.5 0.3 0
E = ⎣ 0.3 0.4 −0.1 ⎦ .
0 −0.1 0.2 (e e ) i j
Determine the change in angle (in degrees) between two differential line
segments, emanating from the point, that are parallel to the vectors
2e1 + 2e2 + e3 and e1 − 5e3 in the undeformed body.
3.12. Problems 135
3.10 The displacement field in a body is

u = [x − (xy)1/2 ]ex + (y − y1/2 )ey + [z − (xz)1/2 ]ez
relative to spatial Cartesian coordinates (x, y, z) > 0. Determine the
stretch ratio and Eulerian extensional strain of a fiber that lies in the
direction ex + ey + ez at the point (1, 2, 0.5) in the deformed body.
3.11 The deformation of an object is described by the relations
x = 2X 2 + XY + Z
y = XY − Y 2 + Z 2
z = Z(1 + X),
where (X, Y, Z) and (x, y, z) are Cartesian coordinates of a point before
and after deformation, respectively. Consider a differential area element
with area da that is perpendicular to the vector 2ex + 3ey − 2ez in
the deformed configuration. Prior to deformation, the element was
located at the coordinates (−1, 2, 1), and its area was dA. Compute
the ratio da/dA and determine the unit normal to the area element in
the undeformed configuration.
3.12 A circular tube undergoes the deformation given by
r = R
θ = Θ + φ(R)
z = Z + w(R),
where (R, Θ, Z) and (r, θ, z) are cylindrical polar coordinates of a point
in the tube before and after deformation, respectively, and φ and w are
scalar functions of R.
(a) Sketch and describe in words the deformation of the tube for the
cases (i) φ = 0 and (ii) w = 0.
(b) Compute F and E using dyadic analysis.
3.13 Consider a circular disk with undeformed radius a0 and thickness h0 .
Suppose the disk undergoes a deformation described by
r = R(1 + w cos mΘ)
θ = Θ
z = Z
for 0 ≤ R ≤ a0 and −h0 /2 ≤ Z ≤ h0 /2, where w is a constant and
(R, Θ, Z) and (r, θ, z) are respective material and spatial cylindrical
polar coordinates.
(a) Determine the deformation gradient tensor.

(b) Compute the ratio of the deformed to undeformed volume of the
disk.
(c) For w = 0.4 and m = 3, plot the Lagrangian strain components
ERR , EΘΘ , and ERΘ at the outer edge of the disk as functions of
Θ for 0 ≤ Θ ≤ 2π. Explain the shape of the curve for EΘΘ . Hint:
Consider the problem of simple shear (Example 3.3).
3.14 Consider deformation of a differential line element a0 in a cube under-
going simple shear (see Fig. 3.6, page 81). For the calculations below,
set k = 0.6 and take (i) a0 = −e1 + e2 , (ii) a0 = e1 , (iii) a0 = e2 , and
(iv) a0 = N1 , where the ei are Cartesian base vectors, and N1 is the
eigenvector given by Eq. (3.173)1. Use vector mechanics and the results
of Examples 3.3 and 3.14 to compute the stretch ratio and the angle of
rotation in degrees about the X 3 -axis (positive counterclockwise) for
the following transformations:
(a) Total deformation of a0 (rotation = θ): a = F · a0
(b) Deformation of a0 by right stretch tensor (rotation = (θ 0 )str ): astr =
U · a0
(c) Rotation of a0 by rotation tensor (rotation = (θ 0 )rot ): (a0 )rot =
Θ · a0
(d) Rotation of astr by rotation tensor (rotation = θ rot ): arot = Θ · astr
Create a table listing the rotations and stretch ratio for each a0 (cases
i–iv). Then, for each case, discuss whether the computed values make
sense for simple shear.
Suggestions:
• Compute the rotation angles and stretch ratios for the edges of the
cube by considering the deformation geometry for k = 0.6.
• As a check, compute the stretch ratio for part (a) as the ratio of
the length of a to the length of a0 , as well as using Eq. (3.94)1.
• Determine whether adding the rotations computed in (b) and (c)
produces the total rotation computed in (a) or (d). Here, it is
important to note that all rotations are about the same axis (X 3 ).
• Use Eq. (3.172) to compute the principal stretch ratio correspond-
ing to N1 .
3.15 In Cartesian coordinates, the deformation of a thin rectangular sheet
is given by
r = (α1 X1 + β 1 X2 )e1 + (β 2 X1 + α2 X2 )e2 + α3 X3 e3 ,
3.12. Problems 137
where αi and β i are constants.

(a) Compute the tensors F, C, and B.
(b) Assuming the sheet is incompressible, determine α3 in terms of the
other quantities. Then, for α1 = 1.25, α2 = 0.9, β 1 = 0.2, and
β 2 = 0.5, compute the principal stretch ratios, principal strains Ei
and ei , and the corresponding principal directions Ni and ni .
(c) For the same parameter values, compute the rotation tensor and
then the right and left stretch tensors, all in matrix form.
(d) Compute the strain invariants using Eqs. (3.136) and show that
they are consistent with the characteristic equation for the eigen-
values of both C and B.
3.16 Consider a deformation described by the relations
x1 = X1 (1 + X2 tet )
x2 = X2
x3 = X3 ,
where the Xi and xi are material and spatial Cartesian coordinates,
respectively.
(a) Determine the velocity field in both material and spatial coordi-
nates.
(b) Determine the acceleration field using both the Lagrangian and Eu-
lerian forms of dv/dt. Show that the two formulas are equivalent.
(c) Determine the rate-of-deformation tensor D in spatial coordinates.
3.17 In Cartesian coordinates, the motion of a particle is described by
r = (X1 + X2 t)e1 + (X2 + X1 t2 )e2 + X3 e3 .
In the spatial coordinates x1 and x2 , the temperature distribution is
θ = (x21 + x22 )t.
For a particle located at (2, −3, 1) at t = 0, compute dθ/dt at t = 1.5.
3.18 In Cartesian coordinates, the motion of a particle is described by the
equations
x1 = X1 + 0.2X2 t4
x2 = X2 − 0.5X1 t3
x3 = X3 .
At t = 2.0, the particle is located at the point (2, 3, 4). Determine the
components of its position, velocity, and acceleration at time t = 1.5.
3.19 Derive Eq. (3.228) by substituting (3.223) into (3.208).

3.20 Consider the velocity field
x 1 e1 + x 2 e2 + x 3 e3
v= ,
α+t
where the xi are Cartesian spatial coordinates and α is a constant.
Compute the acceleration field and explain the result.
3.21 The dilatation rate is given in terms of the velocity v by
J˙ = J ∇
¯ · v.
Derive this equation using the following steps:

(a) Noting that the relation J = det F implies J = J(F), use the chain
rule and a formula in Table 2.5 (page 33) to show
J˙ = JF−T : Ḟ.
(b) Using (3.207), substitute for Ḟ in the result of part (a). Then,
use formulas from the tables in Chap. 2, as well as Eq. (3.208), to
manipulate the equation into the desired form.
Chapter 4
Analysis of Stress
The classical definition of stress is force per unit area. But which area:
the undeformed area or some other reference area? For small deformation,
the distinction is immaterial, since changes in area are negligible. For large
deformation, however, we must distinguish between the undeformed and
deformed geometries. In this case, the only physically meaningful defini-
tion for stress is force per unit deformed area, which is called true stress.
But the deformed geometry of a solid body is not known a priori, often is
difficult to measure, and may be time-dependent. Experimentalists, there-
fore, find it useful to define a Lagrangian stress or engineering stress
as the force per unit undeformed (reference) area. In addition, the gov-
erning equations sometimes have simpler forms or may be easier to solve
when written in terms of yet another type of stress (to be defined later).
We call any stress that is not the true stress a pseudostress, since it has
no physical significance. Pseudostresses always must be converted into true
stress for physical interpretation. This chapter introduces various types of
stress tensor and discusses the relations between them.
4.1 Body and Contact Forces
External forces on an object consist of body forces, which act at a dis-

tance, and contact forces, which act directly on surfaces. In an Eulerian
formulation, we denote the body force per unit deformed volume by f (r,t);
in a Lagrangian formulation, the body force per unit undeformed volume
is f0 (R,t). Then, with dV and dv being the respective undeformed and
deformed volumes of a material element, we have
f0 = J f , (4.1)
where J = dv/dV .
139
140 Analysis of Stress
Contact forces act either on the external surfaces of a body, including

the surfaces of cavities, or on the fictitious internal surface enclosing a
volume element. Internal contact forces are exerted on the element by the
material adjacent to its surface. The intensity of a contact force, i.e., the
force per unit area, is represented by the traction vector. The following
paragraphs define three types of traction vector, which will be used later
to define three types of stress tensor.
Consider an area element dA in an undeformed body B that, under the
action of applied loads, deforms into the area element da in the deformed
body b (Fig. 4.1). As in Eqs. (3.125), we can write
dA = N dA
da = n da, (4.2)
where dA and da are the physical areas of the elements, and N and n are
unit normals to the elements. By convention, the normal to a surface points
outward from the material enclosed by the surface. If dP is a contact force
acting on the deformed area da, then the true traction (stress) vector
is defined as
dP
T(n) (r,t) = (4.3)
da
in the limit as da → 0. In general, T(n) varies from point to point in a
body. Moreover, as the orientation of da(n) changes at a point, so does dP
and therefore T(n) .
n
N dP
dA F da
B b
Fig. 4.1. Deformation of area element dA into area element da. The contact force dP
acts on da.
Next, two pseudotraction (pseudostress) vectors are defined. The

first, which is an engineering-type traction, is simply the force per unit
undeformed area, i.e.,
dP
T(N) (R,t) = . (4.4)
dA
4.2. Stress Tensors 141
Note that, even though T(N) is referred to the undeformed area dA, dP
=da, then T(N) ∼
still acts on the deformed area da (Fig. 4.1). If dA ∼ = T(n) ,
an approximation used in the linear theory (see Sec. A.3.1).
The second pseudotraction vector is defined in terms of a fictitious
(pseudo) force dP̃ acting on dA as defined by the relation
dP = F · dP̃ = dP̃ · FT . (4.5)
This transformation is analogous to the deformation of the undeformed
length element dR into the deformed element dr, i.e., dr = F · dR. In
terms of this pseudoforce, the second pseudotraction vector is
dP̃
T̃(N) (R,t) = , (4.6)
dA
and combining Eqs. (4.3)–(4.6) yields
dP = T(n) da = T(N) dA = T̃(N) · FT dA. (4.7)
4.2 Stress Tensors
Using the traction vectors of Eq. (4.7), we will define three stress tensors:
the Cauchy stress tensor and the first and second Piola-Kirchhoff stress
tensors. The relations between these tensors are discussed in this section,
while the relations between their various components are derived in Sec. 4.3.
4.2.1 Cauchy Stress Tensor

When subjected to applied forces and moments, an object and all its com-
ponent parts must obey Newton’s laws of motion. Consider a tetrahedron
at a point p in a deformed body that consists of three faces carved out
by the surfaces of a general curvilinear coordinate system and an arbitrary
fourth face (Fig. 4.2a). In the limit as the dimensions of the tetrahedral
element shrink to zero, its edges become approximately straight and its
faces planar (Fig. 4.2b).
The coordinate system that we use to define the element is irrelevant.
For convenience, the present derivation is based on a system corresponding
to the natural base vectors gi . The three coordinate faces of the element
then correspond to the surfaces xi = constant (Fig. 4.2).
Let dâi and ni denote the physical area and unit normal of the element
face xi = constant (the xi -face), and let da and n denote the area and normal
of the arbitrary (oblique) face (Fig. 4.2b). The outward-directed normal to
(a) (b)
g3 dP
x3 dP
f n f n
x2 dv dP-1
dv da da
g2
dP-2
dP-3
x1 g1
Fig. 4.2. Differential volume element. (a) Element carved out by general curvilinear
coordinate surfaces. (b) Element in limit as volume approaches zero.

3
the x -face, for example, is n3 = −g1 × g2 /|g1 × g2 | = −g3 / g3 · g3 =
−g3 / g33 by Eq. (2.18), and in general
gi
ni = − . (4.8)
g(ii)
Thus, the vector representations of the four
faces of the tetrahedron can be
written n da and n(i)dâ(i) = −g(i)dâ(i) / g(ii) (i = 1, 2, 3).
To relate these areas, we note that they form a closed surface, and it
can be shown that the vectorial sum of the areas must vanish, i.e.,
n da + ni dâi = 0. (4.9)
This can be seen by considering the limiting case where the element col-
lapses into a triangle in the plane x3 = constant (see Fig. 4.2). In this case,
dâ1 = dâ2 = 0 and Eq. (4.9) gives n da = −ni dâi = −n3 dâ3 . Inspection
shows that n = −n3 , and so da = dâ3 as expected. Thus, Eqs. (4.8) and
(4.9) yield

n da = −ni dâi = gi dâi / g(ii). (4.10)
Setting
da = n da = gi dai (4.11)
and comparing with Eq. (4.10) shows that the covariant components of the
area vector da are related to the physical areas of the xi -faces by
dâi
dai = . (4.12)
g(ii)
Now, let dP and dP−i be the contact forces acting on the element faces
with unit normals n and ni respectively (Fig. 4.2b). The element also is
subjected to a body force f per unit deformed volume. Then, Newton’s
second law of motion for the element gives
dP+ dP−1 + dP−2 + dP−3 + f dv = (ρ dv)a, (4.13)
where ρ is the mass density and dv the volume of the element, and a is
the acceleration of its center of mass. The minus sign in the superscript
of dP−i indicates that the force acts on a negative face of the element,
i.e., the normal to the face makes an acute angle with the −xi -direction
(Fig. 4.2). The face of the element adjacent to this surface is positive,
with a normal making an acute angle with the +xi -direction, and so the
contact force on this adjacent area is denoted dPi .1 By Newton’s third law
of action-reaction, dP−i = −dPi and Eq. (4.13) becomes
dP− dP1 − dP2 − dP3 + f dv = ρa dv. (4.14)
The true traction vector acting on the area da is defined by Eq. (4.3), and
similarly
dP(i)
T̂i = (4.15)
dâ(i)
is the true traction vector acting on the xi -face. Substituting Eqs. (4.3)
and (4.15) into (4.14) yields
T(n) da − T̂i dâi + f dv = ρa dv, (4.16)
in which the summation convention applies to the index i.
Next, we examine the limit as the tetrahedral element shrinks to a point.
Dividing Eq. (4.16) through by da and letting da and dv approach zero gives
dâi
T(n) = T̂i (4.17)
da
since the linear dimension dv/da of the element approaches zero. Note that,
in this limit, all four faces of the tetrahedron pass through the same point.
The area ratio dâi /da is obtained by taking the dot product of Eq. (4.10)
with gj to get
dâi dâi dâj
(n da) · gj = gi · gj = δ ij = ,
g (ii) g (ii) g(jj)
1 In nonorthogonal coordinates, the normals to the element faces may not point in the
coordinate directions (hence the acute angle reference).

which gives
dâi
= (n · gi ) g(ii) . (4.18)
da
Combining this relation with Eq. (4.17) yields

T(n) = (n · gi ) g(ii) T̂i . (4.19)
With the definition of yet another pseudotraction vector

Ti ≡ g(ii) T̂i , (4.20)
Eq. (4.19) becomes
T(n) = (n · gi ) Ti = ni Ti , (4.21)
where the ni are the covariant components of n. This equation shows that
if the traction vectors Ti are known across three different planes passing
through a point, then the traction across any plane through that point can
be computed.
Next, we express the pseudotraction vector of Eq. (4.20) in the form
Ti = σ ij gj , (4.22)
where the σ ij are contravariant components of the vector Ti . Substituting

this relation into Eq. (4.21) yields the Cauchy stress formula
T(n) = n · σ, (4.23)
where
σ = σ ij gi gj . (4.24)
Because T(n) is a true traction vector, σ is called the true stress tensor
or the Cauchy stress tensor. Of course, the Cauchy stress components
with respect to any basis can be obtained by double-dotting σ with the
appropriate dyad.
4.2.2 Pseudostress Tensors

Since it is referred to the deformed body, the Cauchy stress tensor is an ap-
propriate stress measure to use in an Eulerian formulation. In a Lagrangian
formulation, on the other hand, there often are advantages to working with
a pseudostress tensor referred to the undeformed, or a reference, configura-
tion. Thus, we now define two commonly used pseudostress tensors based
on the pseudotraction vectors T(N) and T̃(N) of Eqs. (4.4) and (4.6).
In analogy with Eq. (4.23), the first Piola-Kirchhoff stress tensor
t and the second Piola-Kirchhoff stress tensor s are defined through
the relations
T(N) = N · t
T̃(N) = N · s, (4.25)
where N is the unit normal to the undeformed area element dA (Fig. 4.1).
Substituting Eqs. (4.23) and (4.25) into (4.7) yields
dP = n · σ da = N · t dA = N · s · FT dA, (4.26)
and using Eqs. (4.2) gives
dP = da · σ = dA · t = dA · s · FT . (4.27)
The undeformed and deformed areas are related by Eq. (3.128), which gives
da · σ = (J dA · F −1 ) · σ,
and thus Eq. (4.27) implies
J F−1 · σ = t = s · FT
or
σ = J −1 F · t = J −1 F · s · FT . (4.28)
This relation links the three stress tensors through the deformation gradient
tensor. Solving for the pseudostress tensors yields
t = J F−1 · σ = s · FT
s = J F−1 · σ · F−T = t · F−T . (4.29)
Which stress tensor to use in formulating a given mechanics problem is

a matter of convenience and personal preference. Each has both advantages
and disadvantages, which will become clear later. Later, we will show that
σ is a symmetric tensor (Sec. 4.7.2), i.e., σ = σ T , and so Eq. (4.29)2 gives
sT = J(F−1 · σ · F−T )T = J(F−T )T · (F−1 · σ)T
= JF−1 · (σ T · F−T ) = JF−1 · σ · F−T
= s.
Thus, s also is a symmetric tensor. However, because F is generally not
symmetric, Eq. (4.29)1 shows that t is not symmetric in general. For
solid mechanics problems involving large deformation, therefore, the second
Piola-Kirchhoff stress tensor s often is the tensor of choice for two primary
reasons: (1) It is referred to the undeformed configuration (known a priori);
and (2) it is symmetric. Once s is determined, the other stress tensors can
be computed from Eqs. (4.28) and (4.29). However, only the physical stress
components provide physically meaningful results (see Sec. 4.4).
Atluri (1984) discusses several other stress tensors that are useful in
certain situations. Equation (4.28), for example, suggests defining another
stress tensor τ = Jσ, which is called simply the Kirchhoff stress tensor,
but we do not use τ in this book. For small displacement, all these ten-
sors are essentially equal to the linear stress tensor described in Appendix
A. Indeed, Eq. (4.28) shows that σ ∼ =t∼ = s when J → 1 and F → I. We
note, however, that since F contains both deformation and rigid-body ro-
tation, stipulating small strain alone is not a sufficient condition for this
approximation.
4.3 Relations Between Stress Components
The components of stress can be a source of confusion in learning nonlinear

elasticity. Different authors use different notations, different coordinate
bases, and different derivations. Often a certain set of stress components is
referred to as a “stress tensor,” suggesting that other types of components
of the actual stress tensor do not exist. Working with the stress tensors in
dyadic form alleviates much of this confusion.
Like any tensor, the stress tensors can be expressed in terms of compo-
nents relative to any dyadic basis, with the corresponding set of components
defining the state of stress at a point in a body. Our choice of basis in-
cludes dyads composed of the natural base vectors GI , gi , or gI and their
reciprocals, with some stress components offering certain advantages over
others. Here, only the most commonly used stress components are con-
sidered, which happen to be contravariant components of the three stress
4.3. Relations Between Stress Components 147
tensors. When desired, covariant and mixed components can be computed

by lowering superscripts in the usual manner.
Because the Cauchy stress tensor σ is referred to the deformed configu-
ration, it seems natural to represent this tensor in terms of the base vectors
gi or gI in the deformed body. Likewise, since t and s are referred to the
undeformed body, expressing these tensors in terms of the undeformed base
vectors GI is warranted. Moreover, Eq. (4.28) indicates that t falls between
σ and s in some respects. Thus, components of t relative to mixed dyads
of undeformed and deformed base vectors also are useful.
With these considerations in mind, we examine stress components given
by the following representations:
∗
J∗
σ = σ ij gi gj = σ I gI gJ
∗
t = t GI GJ = t GI gj = tIJ GI gJ
IJ Ij
s = sIJ GI GJ . (4.30)
Table 4.1 lists the notations used for stress components in two other popular
texts. Relations between the various stress components can be found by
substituting Eqs. (3.22) and (4.30) into (4.28). For example, we can write
σ = σ ij gi gj = J −1 F · t

= J −1 FJi gi GJ · tIj GI gj
= J −1 FJi tIj δ JI gi gj
= J −1 FIi tIj gi gj , (4.31)
which implies
σ ij = J −1 FIi tIj .
Similarly,
σ = σ ij gi gj = J −1 F · s · FT
i
= J −1 FK gi GK · sIJ GI GJ · FLj GL gj
= J −1 FK
i
FLj sIJ δ K L
I δ J gi gj
= J −1 FIi FJj sIJ gi gj (4.32)
gives
σ ij = J −1 FIi FJj sIJ .
Combining these expressions yields
σ ij = J −1 FIi tIj = J −1 FIi FJj sIJ , (4.33)

Table 4.1. Stress components used by other authors
Current Green & Zerna (1968) Eringen (1962)

σ ij tij
∗ ∗
σI J τ ij
tIJ tij
sIJ sij T IJ
∗
σI J π ij
tIj T Ij
which relates the stress components used by Eringen (1962; 1980) (see Table
4.1).
∗ ∗
Relations for the convected stress components σ I J can be found by
i I
setting x = X in Eq. (4.33), but it is not clear where to place aster-
isks. One way to clarify matters is to substitute Eq. (4.33) into the dyadic
representations (4.30)1 . This procedure gives, for example,
σ = σ ij gi gj = J −1 FIi tIj gi gj
∗
= J −1 FK
I KJ
t gI gJ ,
where the dummy index I has been changed to K, i and j have been set
to I and J (xi = X I ), and the asterisk indicates that the superscript J ∗ is
attached to the base vector gJ (not GJ ). Since FK I
= ∂X I /∂X K = δ IK by
Eq. (3.15), the above expression can be written
∗
J∗ ∗
σ = σI gI gJ = J −1 δ IK tKJ gI gJ
∗
= J −1 tIJ gI gJ ,
which implies
∗
J∗ ∗
σI = J −1 tIJ .
Perhaps a more straightforward (and safer) method is to substitute

F = gK GK from Eq. (3.22)1 , along with Eqs. (4.30), into (4.28) to get
∗
J∗
σ = σI gI gJ = J −1 F · t
∗
= J −1 gK GK · tIJ GI gJ
∗
= J −1 tIJ δ K
I gK gJ
∗
= J −1 tIJ gI gJ ,
which agrees with the expression derived above. Similarly,

∗
J∗
σ = σI gI gJ = J −1 F · s · FT

= J −1 gK GK · sIJ GI GJ · GL gL
= J −1 sIJ δ K L
I δ J gK gL
= J −1 sIJ gI gJ
gives
∗
J∗
σI = J −1 sIJ .
∗
J∗ ∗
σI = J −1 tIJ = J −1 sIJ , (4.34)
∗
which indicates that tIJ = sIJ . Again, working with dyadics is recom-
mended to avoid confusion.
Comparing Eqs. (4.33) and (4.34) reveals that the relations between
stress components simplify when convected base vectors are used. Most of
the finite elasticity (and biomechanics) literature, however, uses expressions
of the form (4.33), sacrificing the added complexity for the convenience of
describing the deformed configuration in terms of an independent coor-
dinate system (xi ). Furthermore, we see that calling stress components
∗ ∗
“tensors” can be misleading; σ ij certainly is quite different from σ I J in
general. In fact, for an incompressible material (J = 1), Eq. (4.34) gives
∗ ∗ ∗
σ I J = tIJ = sIJ , suggesting that all these stress “tensors” are the same.
Clearly, this is not true for the actual stress tensors σ, t, and s.
Finally, the relation between sIJ and tIJ is derived. Substituting
Eqs. (3.22)1 and (4.30)3 into (4.29)1 yields
t = s · FT
IJ
= s G I G J · G K gK
= sIJ δ K
J G I gK
= sIJ GI gJ,
and extracting the components gives
tIJ = GI · t · GJ

= GI · sKL GK gL · GJ
= sKL δ IK gL · GJ
= sIL gL · GJ
or

tIJ = sIK gK · GJ . (4.35)
This expression can be written in another form by noting r = R + u, and
so Eqs. (3.10) and (3.11) give
gI = r,I = R,I + u,I = GI + u,I

= G I + uK G K , I
= G I + uK | I G K ,
where the covariant derivative comes from Eq. (2.150). Thus,

gI · G J = G I + u K | I G K · G J
= δ JI + uK |I δ JK
= δ JI + uJ |I
and Eq. (4.35) becomes
tIJ = sIK (δ JK + uJ |K ). (4.36)
Example 4.1 Consider uniform extension of a rectangular block of tissue

subjected to an applied load P in the x1 -direction (Fig. 4.3). The face
on which the load acts has cross-sectional areas A1 and a1 before and
after deformation, respectively, and the stretch ratios in the (Cartesian)
coordinate directions are λ1 , λ2 , and λ3 . If the load is distributed uniformly
over the cross section, determine the following:
(a) The relation between the Cauchy (true) stress σ 11 and the first Piola-
Kirchhoff (engineering) stress t11 in terms of the stretch ratios. Do this
first using direct geometrical analysis and then using Eq. (4.33).
(b) The relation between the three stress components σ 11 , t11 , and s11 in
terms of λ1 only, assuming the block is composed of incompressible
material.
Solution. (a) By definition, the Cauchy stress is

σ 11 = P/a1 , (4.37)
and the first Piola-Kirchhoff stress is
t11 = P/A1 . (4.38)
According to the geometry,
a1 /A1 = λ2 λ3 ,
P P
a1
x2
A1
x1
Fig. 4.3. Extension of a rectangular block.
where λ2 and λ3 are stretch ratios. Combining these equations gives

A1 11 t11
σ 11 = t = . (4.39)
a1 λ2 λ3
Next, we derive this relation from Eq. (4.33). For uniform extension,
Eq. (3.46) gives
[FIi ] = diag [λ1 , λ2 , λ3 ], (4.40)
and (4.33) yields
1 t11
σ 11 = J −1 F11 t11 = · λ1 t11 = ,
λ1 λ2 λ3 λ2 λ3
which agrees with (4.39).
(b) Equation (4.33) gives the second Piola-Kirchhoff stress
λ1 λ2 λ3 11
s11 = J(F11 )−2 σ 11 = σ . (4.41)
λ21
If the block is incompressible, then det[FIi ] = λ1 λ2 λ3 = 1, and Eqs. (4.39)
and (4.41) give
σ 11 = λ1 t11 = λ21 s11 . (4.42)
Example 4.2 An aneurysm is a local bulge in the wall of a pressurized

shell-like biological structure. Regional weakening of the wall due to injury
or disease can cause bulges in the heart and arteries, with abnormally high
wall stress likely playing a major role in aneurysm formation and its sub-
sequent development (Zhao et al., 1987; Whittaker et al., 1991; Fillinger et
al., 2002, 2003).
In arteries, a local loss in stiffness is caused by degradation of elastin
and functional smooth muscle (Humphrey and Holzapfel, 2012; Wagenseil,
2018). In the left ventricle, an aneurysm may develop after a myocardial

infarction, which is typically triggered by blockage of a coronary artery
(Schlichter et al., 1954; Whittaker et al., 1991). The blockage cuts off
blood flow to the myocardium downstream, resulting in a region of dead
muscle. Because the affected muscle no longer contracts, it may bulge
outward during systole. As the infarct heals, the dead muscle is replaced
by relatively stiff scar tissue that inhibits bulging. In large transmural
infarcts, however, the bulge may transform into a permanent aneurysm
with potentially life-threatening consequences.
Consider a model for an aneurysm consisting of a thick-walled hemi-
spherical shell fixed around its base (Fig. 4.4). When subjected to blood
pressure p, the shell expands. The boundary conditions for this problem
include zero traction on the outer surface, and a specified pressure on the
inner surface. Write the boundary condition for the inner surface.
Fig. 4.4. Inflation of a hemispherical dome. (a) Undeformed configuration. (b) Deformed
configuration.
Solution. In setting up the pressure boundary condition, it is important

to note that the pressure exerts a compressive force per unit area normal
to the deformed surface (Fig. 4.4b). Thus, the force on an area element da
of the deformed inner surface is given by
dP = −(p da)n, (4.43)
where n is the outward-directed unit normal to da (toward the lumen). The
boundary condition can be written in terms of any of the stress tensors.
But since the pressure follows the surface, it is convenient to express the
∗ ∗
condition in terms of the convected Cauchy stress components σ I J , which
then can be transformed if desired. Equations (4.26) and (4.43) give the
boundary condition
On a: n · σ = −p n, (4.44)
4.4. Physical Components of Stress 153
∗ ∗
which next is written in terms of the σ I J .
Suppose a point in the undeformed shell is located at the spherical polar
coordinates X I , where X 1 , X 2 , and X 3 are the radial, meridional, and
circumferential coordinates, respectively (Fig. 4.4a). During deformation,
the undeformed base vectors GI are convected to the base vectors gI , with
g2∗ following the changing contour of the shell meridian (Fig. 4.4b).
∗
The normal to the inner surface n points along the vector −g1 , since
∗
it is normal to both g2∗ and g3∗ (Fig. 4.4b). Thus, n = −g1 / g1∗ 1∗ and
Eq. (4.44) gives
# $ # $
−g1
∗ ∗ ∗ −g 1∗
∗ ∗ · σ I J gI gJ = −p ∗ ∗
g1 1 g1 1
or
∗
J ∗ 1∗ ∗
J∗ ∗
σI δ I gJ = σ1 gJ = −p g1 .
Dotting both sides of this expression with gK yields the boundary condition
∗
K∗ ∗
On a: σ1 = −p g1 K
(K = 1, 2, 3), (4.45)
∗ ∗ ∗ ∗ ∗ ∗
which provides conditions on σ 1 1 , σ 1 2 , and σ 1 3 , i.e., on the com-
ponents of the pseudotraction vector T1 acting on the inner surface [see
Eq. (4.22)].
4.4 Physical Components of Stress
Clearly, physical components of stress should be based on true (Cauchy)

stress. Accordingly, Eqs. (2.132) and (2.134) give
g(ii)
σ̂ i·j ≡= σ i·j (4.46)
g(jj)
in general curvilinear coordinates and
√
σ̂ ij = σ ij g(ii) g(jj) (4.47)
in orthogonal coordinates. Thus, in an orthogonal coordinate system, the

Cauchy stress tensor can be written in the forms
σ = σ ij gi gj = σ̂ ij ei ej , (4.48)
where the ei are unit base vectors.
4.5 Geometric Interpretation of Stress Components
Recall that the stress components are components of the traction vector,
which depends on the orientation of the plane on which it acts. Consider,
for example, the pseudotraction vector of (4.22), i.e.,
Ti = σ ij gj ,
which acts on an area element with a normal oriented in the direction of
gi .2 This equation shows that the first index of σ ij corresponds to the
plane (orthogonal to gi ) on which the stress acts, and the second index
corresponds to the direction of action (gj ). Interpreting other types of stress
components follows similar reasoning.
Consider now the dyadic representations
σ = σ ij gi gj = σ ij gi gj = σ i·j gi gj = σ ·j i
i g gj . (4.49)
ij
The relation involving σ indicates that the direction of the surface normal
(gi ) is given by raising the index of the first base vector of the dyadic, and
the stress acts in the direction of the second base vector. In fact, for all
components of the stress dyadic, changing the position of the index of the
first base vector (up or down) gives the direction of the surface normal.
Thus, σ ij acts on a surface normal to gi and points along gj , and so on.
All four sets of stress components are illustrated for two-dimensional
elements in the x1 x2 -plane (Fig. 4.5). In each case, the shape of the element
is dictated by the orientations of the gi for σ ij and σ i·j and by the gi
for σ ij and σ ·j i
i , i.e., the faces of the element are perpendicular to g and
gi , respectively. The figure suggests that the σ ij may be the easiest to
visualize, since the faces of the element align with the coordinate curves,
and the stresses point along the coordinate directions. This is one reason
we focused on contravariant stress components in the previous section.
Similar interpretations apply to components relative to the {GI } and
{gI } bases. These alternative base vectors simply carve out differently
shaped material elements. Being referred to areas and directions that are
∗ ∗
convected with the deformation, the Cauchy stress components σ I J are
advantageous in visualizing the mechanics. Of course, only the physical
components
I∗ I∗ g(II)
σ̂ ·J ∗ = σ ·J ∗ , (4.50)
g(JJ)
given by modifying Eq. (4.46), have a true physical meaning.
2 As shown in Fig. 4.2 (page 142), dP−i acts on the element face with a normal in the
direction −gi , and, therefore, dPi acts on a face with a normal in the direction +gi .
4.6. Principal Stresses 155
s22 s2. 2
s2. 1
s21
s12
s1. 1
s11
g2 s1. 2
g2
.
s 22 s22
g1
s2. 1 s21
g1 .
s12 s12
s1. 1 s11
Fig. 4.5. Stress components on two-dimensional differential elements.
4.6 Principal Stresses
In general, the traction vector acting on an area element in a deformed body

does not align with the normal to the area. If these vectors do align, then
no shear stresses act across the area. At any point, it always is possible to
find a unique set of three mutually orthogonal surfaces on which all shear
stresses vanish. These surfaces are called principal planes, the normals to
these surfaces define the principal axes of stress, and the normal stress
components acting on these surfaces are called principal stresses.
Consider a plane with unit normal n passing through point p in a de-
formed body. If the traction vector T(n) points in the direction of n, then
we can write
T(n) = σn, (4.51)
where σ is the normal component of stress on the plane. Because σ is a

symmetric tensor, n · σ = σ T · n = σ · n, and substituting Eq. (4.23) into
(4.51) gives the eigenvalue problem
(σ−σI) · n = 0. (4.52)
Solving this equation provides the principal Cauchy stresses σ i and the
corresponding principal directions ni (i = 1, 2, 3). Since σ is symmetric,
the eigenvalues are real and the eigenvectors are mutually orthogonal.
Principal stresses, therefore, are determined by solving an eigenvalue

problem similar to that used to find the principal strains. The principal axes
of stress and strain, however, do not always coincide. In general anisotropic
materials, for example, normal stresses alone can induce shear strains.
4.7 Equations of Motion
Every particle of a solid body, as well as the entire body itself, must obey
Newton’s laws of motion. Section 4.2 explored the consequences of these
laws for a volume element that shrinks to a point. That analysis led to the
definition of the Cauchy stress tensor. Here, we use Newton’s second law
in the form of the principles of linear and angular momentum to write the
global equations of motion for a solid body. Manipulating these relations
then provides local equations of motion for an infinitesimal element.
4.7.1 Principle of Linear Momentum

Consider a deformed body of mass density ρ(r, t) that is subjected to a
traction T(n) (r, t) acting over its surface area a and a body force f (r, t)
acting over its volume v (Fig. 4.6). Applying the law of conservation of
linear momentum to the entire body gives
d
T(n) da + f dv = v ρ dv, (4.53)
a v dt v
where v(r, t) is the velocity of the mass center of the volume element dv.
This relation represents the Eulerian form of the global equation of motion
for the body, with the various quantities being functions of the deformed
position r.
Substituting Eq. (4.23) into the area integral of (4.53) and applying the
divergence theorem (2.200) yields

T(n) da = n · σ da = ¯ · σ dv,
∇ (4.54)
a a v
where ∇ ¯ is the gradient operator in the deformed body b, as defined by
Eq. (3.18)2. In addition, carrying out the differentiation on the right-hand-
side of Eq. (4.53) gives

d d
v ρ dv = (v ρ dv)
dt v v dt

dv d
= ρ dv + v (ρ dv)
dt dt
v
= a ρ dv, (4.55)
v
4.7. Equations of Motion 157
in which d(ρ dv)/dt = 0 by conservation of mass and a = dv/dt is the

acceleration of the element dv. With these expressions, Eq. (4.53) can be
written
(∇¯ · σ + f −ρ a) dv = 0. (4.56)
v
Finally, because it must hold for an arbitrary volume in the deformed body,
this relation implies the local equation of motion
¯ · σ + f = ρ a,
∇ (4.57)
where the acceleration a(r, t) is provided by Eq. (3.193)2 . If inertia ef-
fects can be neglected, then we can set a = 0, and Eq. (4.57) becomes an
equilibrium equation.
f
n
(n)
T
v
da
r
dv
Fig. 4.6. Forces acting on deformed body.
4.7.2 Principle of Angular Momentum

Equation (4.53) governs the linear motion of a body. The rotational motion
must satisfy the
law of conservation
of angularmomentum
d
r × T(n) da + r × f dv = (r × v) ρ dv, (4.58)
a v dt v
where r is the position vector to a point in the deformed body (Fig. 4.6).
To set the stage for converting the surface integral in the above equation
into a volume integral, we first note that the Cauchy stress tensor can be
written in the form
σ = σ ij gi gj = gi (σ ij gj ) = gi Ti ,
where Eq. (4.22) has been used. This relation and (4.23) give
r × T(n) = r × (n · σ) = r × (n · gi Ti ) = (n · gi ) r × Ti
= n · (gi r × Ti ).
With this equation and the divergence theorem (2.200), the first term in
Eq. (4.58) becomes

r × T(n) da = n · (gi r × Ti ) da = ∇¯ · (g r × Ti ) dv.
i (4.59)
a a v
Next, differentiating the integrand, setting r,i = gi , and using (2.137)2
yields
¯ · (g r × Ti ) ∂
∇ i = gk · (gi r × Ti )
∂xk
= gk · (gi ,k r × Ti + gi r,k × Ti + gi r × Ti ,k )
= Γkik r × Ti + δ ki gk × Ti + δ ki r × Ti ,k
= Γkik r × Ti + gi × Ti + r × Ti ,i ,
which can be simplified by noting
¯ ·σ ∂
∇ = gk · (gi Ti )
∂xk
= gk · (gi ,k Ti +gi Ti ,k )
= Γkik Ti + δ ki Ti ,k
= Γkik Ti + Ti ,i .
Combining these expressions shows that Eq. (4.59) can be written

(n) ¯ · σ)] dv.
r × T da = [gi × Ti + r × (∇ (4.60)
a v
In addition, the right-hand-side of Eq. (4.58) is

d d
(r × v) ρ dv = [(r × v) ρ dv]
dt v v dt

d d
= ρ dv (r × v) + (r × v) (ρ dv)
v dt dt

d
= ρ dv (ṙ × v + r × v̇) + (r × v) (ρ dv)
dt
v
= (r × a) ρ dv. (4.61)
v
Note that ṙ × v = 0 since ṙ = v, and d( ρ dv)/dt = 0 by conservation of
mass.
Substituting Eqs. (4.60) and (4.61) into (4.58) and rearranging yields

[gi × Ti + r × (∇¯ · σ + f −ρ a)] dv = 0. (4.62)
v
The term in parentheses vanishes by the equation of motion (4.57), and,
because the integral must be zero for an arbitrary volume, the principle of
angular momentum reduces to
gi × Ti = 0. (4.63)
The consequence of this result is revealed by substituting Eq. (4.22) and
using (2.103)1 to obtain
gi × Ti = gi × σ ij gj = ijk σ
ij k
g = 0,
which can be written
ij k ji k
ijk σ g =0 or jik σ g = 0.
Since jik =− ijk , adding these expressions yields
ijk (σ
ij
− σ ji)gk = 0, (4.64)
ij ji
which implies that σ = σ , i.e., the Cauchy stress tensor is symmetric
(σ = σ T ). In Sec. 4.2.2, we used this fact to show that the second Piola-
Kirchhoff stress tensor s also is symmetric.
4.7.3 Lagrangian Form of Equation of Motion

For small deformation, the gradient operator ∇ ¯ in Eq. (4.57) can be re-
placed by ∇, so that derivatives are taken with respect to undeformed co-
ordinates. When deformation is large, however, derivatives must be taken
with respect to the deformed coordinates, which are not known a priori.
Because this can complicate matters considerably, it is useful to express the
equations of motion for arbitrarily large deformation in a Lagrangian, or
material, form that involves ∇ rather than ∇.¯ A similar concern motivated
the introduction of the pseudostress tensors in Sec. 4.2.2.
A relatively painless way to derive the appropriate equation is to rewrite
the principle of linear momentum (4.53) in terms of Lagrangian quantities.
For a body that originally occupied a volume V enclosed by a surface area
A, we have

(N) d
T dA + f0 dV = v ρ0 dV, (4.65)
A V dt V
where the surface traction T(N) (R, t) and body force f0 (R, t) are defined
by Eqs. (4.4) and (4.1), respectively, and ρ0 (R, t) is the mass density of
the undeformed volume element dV . Note that, although the integrals are
taken over the undeformed body, the forces actually are applied to the
deformed body.
From here, the derivation follows that given in Sec. 4.7.1. First, with
Eqs. (4.25)1 , and (2.200), the surface integral of (4.65) becomes

T(N) dA = N · t dA = ∇ · t dV. (4.66)
A A V
Second, as Eq. (4.55) shows,

d
v ρ0 dV = a ρ0 dV (4.67)
dt V V
since ρ0 dV = ρ dv is the constant mass of an element. Substituting these

relations into (4.65) yields

(∇ · t + f 0 −ρ0 a) dV = 0, (4.68)
V
which implies the local equation of motion
∇ · t + f 0 = ρ0 a. (4.69)
Although Eqs. (4.57) and (4.69) appear similar, there are significant
differences. Since the derivatives are computed relative to the known coor-
dinates in the undeformed configuration, Eq. (4.69) is easier to solve in prin-
ciple. Moreover, with v = v(R, t), the acceleration is given by Eq. (3.193)1,
rather than (3.193)2. In contrast to σ, however, the first Piola-Kirchhoff
stress tensor t is generally not symmetric. Thus, working with the sym-
metric second Piola-Kirchhoff stress tensor s often is more convenient, and
putting Eq. (4.29)1 into (4.69) gives
T
∇ · (s · F ) +f 0 = ρ0 a. (4.70)
But this equation is more complicated than Eq. (4.69). As is becoming ap-
parent, all large-deformation formulations possess both inherent advantages
and disadvantages. While the stress and strain measures of choice vary, the
pseudostress tensor s and the Lagrangian strain tensor E are often used in
finite elasticity problems. Besides their symmetry and being referred to the
undeformed configuration, these tensors lead to a convenient form of the
material constitutive relations (see Chap. 5).
Example 4.3 As we have just seen, deriving the Lagrangian form of the
equation of motion from a direct application of the principle of linear mo-
mentum is straightforward. Deriving Eq. (4.69) from Eq. (4.57), or vice
versa, is a more complicated task. Show that these two forms of the equa-
tions of motion are indeed equivalent.
Solution. Beginning with Eq. (4.69), we substitute Eqs. (4.1) and (4.29)1
and use the relation
dv ρ
J= = 0 (4.71)
dV ρ
to obtain
J −1 ∇ · (J F−1 · σ) + f = ρ a. (4.72)
A glance at Eq. (4.57) reveals that our task boils down to showing that the
¯ · σ. But this is easier said than done.
first term in (4.72) is equivalent to ∇
With ∇ given by Eq. (3.18)1 , expanding this term gives
∂
J −1 ∇ · (J F−1 · σ) = J −1 GI · (J F−1 · σ)
∂X I
= J −1 GI · (J,I F−1 · σ+J F−1,I · σ + J F−1 · σ,I ).
(4.73)
In the following, we examine separately each term of this equation.
I
J,I = ∂J/∂X in the first term, we use Eq. (3.119), which
To compute
gives J = j g/G. Then, working in convected coordinates for convenience,
we can set j = 1 (see Sec. 3.6.3) to obtain
√ √ √ √
g G( g),I − g( G),I
J,I = =
G ,I G
√ √
g ( g),I ( G),I
= √ − √ . (4.74)
G g G
Specializing Eq. (2.149) gives
√ √ J
( g),I = g Γ̄
√ √ JI
( G),I = G ΓJJI , (4.75)
where the Christoffel symbols are defined by Eqs. (3.86), and Eq. (4.74)
can now be written
J −1 J,I = Γ̄JJI − ΓJJI . (4.76)

With the help of Eq. (3.22)2 , the first term on the right-hand-side of
Eq. (4.73) becomes
J −1 GI · (J,I F−1 · σ) = J −1 J,I GI · (GK gK ) · σ
= J −1 J,I δ IK gK · σ
= (Γ̄JJI − ΓJJI ) gI · σ. (4.77)
The second term on the right-hand-side of Eq. (4.73) requires
F−1,I = (GK gK ),I = GK ,I gK + GK gK ,I
= (ΓJKI GJ )gK − GK (Γ̄K J
JI g ),
in which Eqs. (2.136) and (2.139) have provided the derivatives of the base
vectors. Inserting this expression into the second term of (4.73) gives
J −1 GI · (J F−1,I · σ) = GI · (ΓJKI GJ gK − Γ̄K
JI GK g ) · σ
J
= (ΓJKI δ IJ gK − Γ̄K I J
JI δ K g ) · σ
= (ΓIKI gK − Γ̄IJI gJ ) · σ
= (ΓJIJ − Γ̄JIJ )gI · σ, (4.78)
where dummy indices have been redefined.
Next, using Eq. (3.18)2 in the last term of (4.73) yields
J −1 GI · (J F−1 · σ,I ) = GI · (GK gK ) · σ,I
= δ IK gK · σ,I = gI · σ,I
∂
= gI ·σ
∂X I
= ¯ · σ.
∇ (4.79)
Finally(!), substituting Eqs. (4.77)–(4.79) into (4.73) gives (since ΓJIJ =
ΓJJI )
J −1 ∇ · (J F−1 · σ) = ∇
¯ · σ, (4.80)
which proves that Eq. (4.72), or Eq. (4.69), is equivalent to Eq. (4.57).
4.7.4 Component Forms of Equations of Motion

In terms of the three stress tensors, the equations of motion (4.57), (4.69),
and (4.70) are
∇¯ · σ + f = ρa
∇ · t + f0 = ρ0 a
T
∇ · (s · F ) +f 0 = ρ0 a. (4.81)
When dyadic notation is used, obtaining the component forms of these

equations is straightforward. Here, we write the equations in terms of the
following sets of components:
∗
J∗
j J∗
f = f gj = f gJ
j J∗
a = a gj = a gJ (4.82)
t = tIJ GI GJ
f0 = f0J GJ
a = aJ G J (4.83)
s = sIJ GI GJ
∗
f0 = f0J GJ = f0J gJ
∗
a = aJ G J = a J g J . (4.84)
The derivation requires the gradient operators
¯ = gi ∂ = gI ∂
∇
∂xi ∂X I
∂
∇ = GI , (4.85)
∂X I
as provided by Eqs. (3.18) and the Christoffel symbols
Γ̄kij = gi , j · g k
Γ̄K
IJ = gI , J · g K
ΓK
IJ = GI ,J · GK (4.86)
which are given by Eqs. (3.86).
Consider first the divergences in Eqs. (4.81). For the first of these
expressions, Eqs. (4.82)1 and (4.85)1 give the two forms
¯ ·σ = ∂
∇ gk k · σ = gk · (σ ij gi gj ),k
∂x
∂ ∗ ∗
= gK K
· σ = gK · (σ I J gI gJ ),K
∂X
and noting Eq. (2.153) yields
∇¯ · σ = gk · (σ ij |k gi gj ) = σ ij |k δ k gj
i
= σ ij |i gj
∗
J∗ ∗
J∗
= gK · (σ I | K gI gJ ) = σ I |K δ K
I gJ
∗ ∗
= σI J
| I gJ . (4.87)
In these equations the vertical bar indicates covariant differentiation with

respect to the coordinates in the deformed body. Adapting Eq. (2.154)2
gives
σ ij |i = σ ij ,i +σ kj Γ̄iik + σ ik Γ̄jik
∗
J∗ ∗
J∗ ∗
J∗ ∗
K∗
σI |I = σ I ,I +σ K Γ̄IIK + σ I Γ̄JIK . (4.88)
Similarly, for the second of (4.81), Eqs. (2.153), (4.83)1 , and (4.85)2 give
∂
∇·t = GK · t = GK · (tIJ GI GJ ),K
∂X K
= GK · (tIJ ||K GI GJ ) = tIJ ||K δ K
I GJ
= tIJ ||I GJ , (4.89)
where
tIJ ||I = tIJ ,I +tKJ ΓIIK + tIK ΓJIK . (4.90)
The double vertical bar denotes covariant differentiation with respect to

the coordinates in the undeformed body.
The last of Eqs. (4.81) requires a little more work. With the relation
F = gI GI , Eqs. (4.84)1 and (4.85)2 give
T
∇ · (s · F ) = ∇ · (sIJ GI GJ · GK gK )
= ∇ · (sIJ GI δ K
J gK )
∂
= GK · (sIJ GI gJ )
∂X K
= GK · (sIJ ,K GI gJ + sIJ GI ,K gJ + sIJ GI gJ ,K )
= sIJ ,K δ K IJ K
I gJ + s G · G I , K gJ + s δ I gJ , K
IJ K
= sIJ ,I gJ + sIJ GK · GI ,K gJ + sIJ gJ ,I .

Simplifying this expression using Eqs. (2.136) and (4.86)2,3 yields
T
∇ · (s · F ) = sIJ |||I gJ , (4.91)
where
sIJ |||I = sIJ ,I +sIJ ΓK

IK + s
IK J
Γ̄IK . (4.92)
Here, the triple vertical bar indicates a mixed covariant derivative with
respect to both the undeformed and deformed configurations.
Substituting all these results into Eqs. (4.81) yields the equations of
motion
σ ij |i + f j = ρaj
∗
J∗ ∗ ∗
σI |I + f J = ρaJ
tIJ ||I + f0J = ρ0 aJ
∗ ∗
sIJ |||I + f0J = ρ0 aJ . (4.93)
An alternate equation involving sIJ can be found by substituting Eq. (4.36)

into the third of these relations to get
[sIK (δ JK + uJ |K )]||I + f0J = ρ0 aJ . (4.94)
Example 4.4 Consider a cylindrical polar coordinate system with unit

base vectors {er , eθ , ez } (Fig. 4.7).
(a) Derive the equations of motion for large deformation in terms of the
tensor components σ ij .
(b) Write the equations of motion in terms of physical stress components σ̂ ij .
x3, z
e3, ez
e2
x2
r eq
e1
q
x1
er
Fig. 4.7. Cylindrical polar coordinate system.
Solution. (a) The tensor-component form of the equation of motion,

provided by Eqs. (4.88)1 and (4.93)1, is
σ ij ,i +σ kj Γ̄iik + σ ik Γ̄jik + f j = ρaj . (4.95)
With (x1 , x2 , x3 ) = (r, θ, z), the only non-zero Christoffel symbols for cylin-
drical polar coordinates are
Γ̄122 = −r, Γ̄212 = Γ̄221 = r −1 (4.96)
as given by Eqs. (2.148). Hence, expanding Eq. (4.95) yields the relations
1
σ 11 ,1 +σ 21 ,2 +σ 31 ,3 + σ 11 − rσ 22 + f 1 = ρa1
r
3
σ ,1 +σ ,2 +σ 32 ,3 + σ 12 + f 2
12 22
= ρa2
r
1 13
σ ,1 +σ ,2 +σ ,3 + σ + f 3
13 23 33
= ρa3 . (4.97)
r
(b) Since cylindrical coordinates are orthogonal, Eq. (4.47) gives the
physical stress components
√
σ̂ ij = σ ij g(ii)g(jj).
With the components of the metric tensor being [see Eqs. (2.146)]
[gij ] = diag [1, r2, 1],
this equation gives

⎡ ⎤ ⎡ rr ⎤
σ̂ 11 σ̂ 12 σ̂ 13 σ̂ σ̂ rθ σ̂ rz
σ (ei ej ) = ⎣ σ̂ 21 σ̂ 22 σ̂ 23 ⎦ = ⎣ σ̂ θr σ̂ θθ σ̂ θz ⎦
σ̂ 31 σ̂ 32 σ̂ 33 σ̂ zr σ̂ zθ σ̂ zz
⎡ 11 ⎤
σ rσ 12 σ 13
= ⎣ rσ 21 r 2 σ 22 rσ 23 ⎦ . (4.98)
σ 31 rσ 32 σ 33
In addition, the physical components of the body force and acceleration are
defined by
f = f i gi = fˆr er + fˆθ eθ + fˆz ez

a = ai gi = âr er + âθ eθ + âz ez . (4.99)
Since Eqs. (2.145) give
g1 = er , g2 = reθ , g3 = ez ,
we have
fˆr = f1, fˆθ = rf 2 , fˆz = f 3
âr = a1 , âθ = ra2 , âz = a3 . (4.100)
4.8. Problems 167
Finally, after substituting (4.98) and (4.100) into (4.97) and differentiating,
we obtain the equations
∂ σ̂ rr 1 ∂ σ̂ θr ∂ σ̂ zr σ̂ rr − σ̂ θθ
+ + + + fˆr = ρâr
∂r r ∂θ ∂z r
∂ σ̂ rθ 1 ∂ σ̂ θθ ∂ σ̂ zθ 2σ̂ rθ
+ + + + fˆθ = ρâθ
∂r r ∂θ ∂z r
∂ σ̂ rz 1 ∂ σ̂ θz ∂ σ̂ zz σ̂ rz
+ + + + fˆz = ρâz , (4.101)
∂r r ∂θ ∂z r
which agree with those given in Appendix B for cylindrical coordinates.
4.8 Problems
As in the previous chapter, unless the problem statement explicitly refer-

ences contravariant components or base vectors, please feel free to use only
subscripts and the summation convention defined in Appendix A. Here
again, units are ignored, and any relations listed in Appendix B can be
used without derivation. Finally, to focus on basic ideas, no attempt has
been made to make deformation and stress components consistent within a
given problem. Doing this would require constitutive relations, the subject
of the next chapter.
4.1 Relative to Cartesian coordinates (x, y, z), the Cauchy stress tensor at
a point is given by the matrix
⎡ ⎤
40 30 −10
[σ ij ] = ⎣ 30 60 25 ⎦ .
−10 25 80
(a) Determine the true traction vector across a plane defined by the
relation
F = 2x + 2y + z − C = 0,
where C is a constant. Hint: A vector normal to a surface with
equation F = 0 is given by a = ∇F.
(b) The traction vector can be written in the form T(n) = σ nn n+ σ ns s,
where n and s are unit vectors normal and tangent to the plane
F = 0 (n not summed). Determine the stress components σ nn and
σ ns .
(c) Determine the principal Cauchy stresses and the unit vectors defin-
ing the directions in which they act.
4.2 Direction cosines of a unit vector n relative to Cartesian axes (x, y, z)
are given by p = n · ex , q = n · ey , and r = n · ez . Suppose (p, q, r) are
known for the unit normal at a point on the boundary of a deformed
body.
√ At this point, the true surface traction vector has magnitude
3a and is oriented at the same angle relative to all three positive
coordinate axes. If σ yy = σ xz = σ yz = 0, determine the remaining
Cauchy stress components.
4.3 In Cartesian coordinates xi , the Cauchy stress tensor at a given point
is
σ = a e1 e1 − b e2 e2 + c e3 e3 + d(e1 e3 + e3 e1 ) − e(e2 e3 + e3 e2 ),
where a, b, c, d, and e are positive constants.
(a) Determine the unit normal vector for a plane parallel to the x3 -axis
on which the true traction vector is tangent to the plane.
(b) Consider a coordinate system x̄i that is obtained by rotating the xi
system through an angle θ about the x1 -axis. Compute the stress
components σ̄ 13 and σ̄ 23 relative to the rotated system.
4.4 Relative to spherical polar coordinates (r, θ, φ), the physical compo-
nents of the Cauchy stress tensor at a point in a body are
⎡ rr ⎤
σ̂ 0 0
σ (er ,eθ ,eφ ) = ⎣ 0 1 2 ⎦ .
0 2 σ̂ φφ
If the principal stresses at this point are σ̂ 1 = 1, σ̂ 2 = −1, and σ̂ 3 = 3,
find σ̂ rr and σ̂ φφ .
4.5 In Cartesian coordinates (x, y, z), the Cauchy stress distribution in a
body is given by
⎡ ⎤
xy −y 0
σ (ei ej ) = ⎣ −y x − y 0 ⎦.
3
0 0 y + 2z
Determine the physical components of σ relative to cylindrical coordi-
nates (r, θ, z).
4.6 In cylindrical coordinates, the first Piola-Kirchhoff stress tensor at the
point (r, θ, z) = (5, π/3, −2) is
t = 2er er + 3eθ eθ − eθ ez − (2/3)ez eθ ,
4.8. Problems 169
and the deformation gradient tensor is
F = er er + 2eθ eθ + 3ez ez − eθ ez .
Find the Cauchy stress components σ ij , σ i·j , σ ·j
i , and σ ij (i, j = r, θ, z)
relative to the natural base vectors for cylindrical coordinates.
4.7 A circular tube with deformed inner radius a and outer radius b is
loaded by a uniform circumferential Cauchy shear stress τ 0 on its outer
surface.
(a) Determine the shear stress τ i that must be applied to the inner
surface to maintain equilibrium.
(b) Write the dyadic form of the Cauchy stress tensor σ for an arbi-
trary point in the tube in terms of the base vectors {er , eθ , ez } for
cylindrical polar coordinates.
(c) Find the Cartesian components of σ for an arbitrary point in the
tube.
4.8 In Cartesian coordinates, the deformation of a solid body from config-
uration B(X I ) to configuration b(xi ) is defined by the relations
x1 = X 1 + 0.2X 2 X 3
x2 = X 2 + 0.2(X 3 )2
x3 = X3.
In addition, the second Piola-Kirchhoff stress tensor in b is given by

s = 2X 1 e1 e1 − X 1 X 2 e1 e2 − X 1 X 2 e2 e1 + (3X 1 + X 2 )e2 e2 ,
where the ei are Cartesian base vectors.

(a) Determine the Cauchy stress tensor σ in terms of the basis {ei ej }
for the point in b that was located at (−1, 2, 1) in B.
(b) Consider natural stress components defined by
∗
J∗ ∗ ∗
σ = σI gI gJ = σ ·J I ·J I
I ∗ g gJ = σ I G gJ .
∗ ∗ ∗ ∗
Determine σ I J , σ ·JI ∗ , and σ I
·J
at the same point using dyadic
analysis, i.e., by dotting σ with appropriate base vectors.
∗ ∗ ∗
(c) Determine σ I J and σ ·J I ∗ using matrix analysis, i.e., by using ap-
propriate forms of Eqs. (2.70). Hint: Treat the change in basis as
a transformation from Cartesian to convected coordinates.
(d) Determine the pseudotraction vector T(N) on a plane that passes
through this point and is normal to the vector 3e1 − e3 in B.
4.9 At a given point in a solid body, the deformation gradient tensor is

F = 4e1 e1 − e1 e2 + 2e2 e1 + 3e2 e2 + e3 e3 ,
and the Cauchy stress tensor is
σ = −2e1 e1 + 3e1 e2 + 3e2 e1 + 4e2 e2 + e3 e3 ,
in which the ei are Cartesian base vectors.
(a) Compute the Piola-Kirchhoff stress tensors t and s.
(b) Compute the contravariant components (σ ij ) and the mixed com-
ponents (σ i·j and σ ·i
j ) of σ relative to the basis defined by the
vectors
g1 = e1 − 3e2
g2 = 2e1 − e2
g3 = e1 + e2 + e3 .
Find these components using (i) dyadic analysis by dotting σ with
appropriate base vectors, and (ii) using matrix analysis based on
Eqs. (2.70). Show that the matrices are consistent with σ being a
symmetric tensor.
(c) Compute the physical stress components relative to the basis of
part (b).
(d) Sketch the stress components σ ij and σ i·j on planar elements de-
fined by g1 and g2 . Show the correct orientations, but not neces-
sarily the correct relative magnitudes.
4.10 Consider the Cauchy and second Piola-Kirchhoff stress tensors written
in the dyadic forms
∗
σ = σI J
gI G J
∗
IJ
s = s G I gJ .
Here, the GI are base vectors in the initial configuration, and the gI
are convected base vectors in the deformed body. Derive a relation for
∗ ∗
σ I J as a function of sIJ and components of the deformation gradient
I
tensor. Hint: Try using F = F·J GI GJ .
4.11 Consider a circular tube with a uniform internal pressure. Let (R, Θ, Z)
and (r, θ, z) be cylindrical coordinates of a point before and after de-
formation, respectively. With ei = eI , the deformation gradient tensor
is
F = λr er er + λθ eθ eθ + λz ez ez ,
4.8. Problems 171
where the λi are stretch ratios. The axial stretch ratio λz is constant,
and, with inflation moving a point in the wall radially from radius R to
r(R), Eq. (3.63) with ψ = 0 (no torsion) gives the other stretch ratios
dr r
λr = , λθ = .
dR R
Finally, assume the Cauchy and first Piola-Kirchhoff stress tensors are
given by
σ = σ r (r)er er + σ θ (r)eθ eθ + σ z ez ez
t = tr (R)er er + tθ (R)eθ eθ + tz ez ez .
(a) Show that the radial equilibrium equation takes the following
forms:
dσ r 1
+ (σ r − σ θ ) = 0
dr r
dtr 1
+ (tr − tθ ) = 0.
dR R
(b) Show that the two equilibrium equations are equivalent.
4.12 Consider a solid circular cylinder with deformed radius a and length d.
In Cartesian coordinates, the Cauchy stress tensor is
σ = x2 ex ex + y2 ey ey + xy(ex ey + ey ex).
Determine the physical components of the body forces and surface trac-
tions (on all surfaces) that must be applied to the cylinder to maintain
equilibrium with the given stress distribution. Express the results rel-
ative to spatial cylindrical coordinates (r, θ, z).
4.13 Consider torsion of a spherical shell about the z-axis (see Fig. B.2 in
Appendix B). Assume the Cauchy stress tensor can be written in the
dyadic form
σ = σ̂ θφ (eθ eφ + eφ eθ ),
where σ̂ θφ = σ̂ θφ (r, θ) is the only non-zero physical stress component
in spherical polar coordinates (r, θ, φ).
(a) Using ∇ ¯ · σ = 0, derive the non-trivial equilibrium equation(s) in
terms of σ̂ θφ .
(b) Using σ ij |i , derive the non-trivial equilibrium equation(s) in terms
of σ θφ .
(c) Show that the equations found in parts (a) and (b) are equivalent.
4.14 A solid circular cylinder has undeformed radius a0 . Let the undeformed
and deformed configurations be described by the cylindrical coordinates
(R, Θ, Z) and (r, θ, z), respectively. Suppose a surface traction T(n) =
−pn + τ s acts on the curved surface, where p(Θ) is a pressure, τ (Θ) is
a shear stress, and n(Θ) and s(Θ) are normal and circumferential unit
vectors tangent to the deformed surface. This loading causes a point
in the cylinder located originally at the position R = ReR + ZeZ to
move to r = R(1 + b cos2 Θ)eR + ZeZ , where b is a positive constant
and the eI are unit vectors.
(a) Determine the deformation gradient tensor in the dyadic form F =
FIJ eI eJ (I, J = R, Θ, Z; summed).
(b) Write n, s, the undeformed base vectors GI and GI , and the con-
vected base vectors gI and gI in terms of the non-zero FIJ .
(c) Consider the dyadic representations
∗
J∗
σ = σ IJ GI GJ = σ I gI gJ
for the Cauchy stress tensor. Write the boundary conditions on the
∗ ∗
curved surface in terms of (i) σ IJ and (ii) σ I J .
Chapter 5
Constitutive Relations
The analyses of deformation and stress in the previous chapters are valid
for any solid body that can be represented as a continuum, regardless of the
type of material that comprises the body. Deformation and stress, however,
are linked by constitutive relations that must be found experimentally for
each type of material. In biomechanics, determining constitutive relations
is complicated by material nonlinearity, complex geometry, the composite
nature of biological tissues, and the influence of a wide range of environ-
mental variables. In general, even the functional forms of these relations
are unknown, but thermodynamic considerations place important restric-
tions on their form. This chapter considers some of these restrictions and
discusses some strategies for determining constitutive relations for soft tis-
sues.
5.1 Thermodynamics of Deformation
At the outset, we consider the first and second laws of thermodynamics

for a thermomechanical continuum in which thermal and mechanical
effects dominate the behavior, with electrical, chemical, and other effects
ignored. Note that the present treatment is not exhaustive, and the reader
is expected to be familiar with basic concepts such as entropy.
5.1.1 First Law of Thermodynamics

For a thermomechanical continuum, the principle of conservation of energy
(balance of energy) can be written in the form
K̇ + U̇ = P + Q, (5.1)
where K is kinetic energy, U is internal energy, P is mechanical power
input, and Q is the rate of heat input. The left-hand-side of this equation
173
174 Constitutive Relations
represents the time-rate of change of the total energy in the system, and
the right-hand-side is the rate at which energy is added to the system. The
various quantities in this equation are written below in both Eulerian and
Lagrangian forms.
Consider a body of mass density ρ0 (R) and volume V in the reference
configuration. After deformation, these quantities become ρ(r, t) and v,
respectively. The total kinetic energy for the body is

1 1
K = 2 ρ v · v dv = 2 ρ0 v · v dV, (5.2)
v V
where v(R, t) = v(r, t) is the velocity vector for a point that moves from
the undeformed location R to the deformed location r. In addition, the
total internal energy contained in the body is

U = ρu dv = ρ0 u dV, (5.3)
v V
where u is the internal energy per unit mass. Here, the internal energy
consists of thermal energy and strain energy.
The power input is the rate of work done on the body by applied loads,
which include surface and body forces. In terms of the body force vectors
in Eq. (4.1) and the surface traction vectors defined by Eqs. (4.3) and (4.4),
the total power input is

P = T(n) · v da + f · v dv
a v

(N)
= T · v dA + f0 · v dV, (5.4)
A V
in which A and a are the surface areas of the body in the undeformed and
deformed configurations, respectively.
Finally, the body gains thermal energy by heat flowing across its surface,
as well as heat generated by internal sources. Let q and q0 be the outward-
directed heat flux vectors per unit deformed and undeformed surface area,
respectively, and let r be the rate of heat production per unit mass due to
internal sources. Then, the rate of heat added to the body is

Q = − q · n da + ρr dv
a v

= − q0 · N dA + ρ0 r dV, (5.5)
A V
where n and N are the respective (outward-directed) unit normals to the

surface before and after deformation.
5.1. Thermodynamics of Deformation 175
The various terms in Eq. (5.1) are next expressed in alternate forms so
they can be combined. First, Eq. (5.2) gives

1 d d
K̇ = 2 (ρ dv) v · v+ (v · v)ρ dv
dt dt
v
1
= 2 ( v̇ · v + v · v̇)ρ dv = v · v̇ ρ dv
v v
= v · (∇¯ · σ + f ) dv, (5.6)

v
in which mass conservation gives d(ρ dv)/dt = 0, and substituting for ρv̇ =
ρa from the equation of motion (4.57) produces the third line. Similar
manipulations with the Lagrangian form for K in Eq. (5.2) yield

K̇ = v · (∇ · t + f 0 ) dV, (5.7)
V
where Eq. (4.69) has been used.

Next, we recast the terms v · (∇ ¯ · σ) and v · (∇ · t) in the last two
equations. The complexities of the following operations make it convenient
to work in Cartesian components and then convert the results back to direct
notation. With all indices written temporarily as subscripts, we have
¯ · σ) ∂
v · (∇ = v · ek · σ ij ei ej
∂xk
= v · (σ ij ,k δ ki ej ) = v · (σ ij ,i ej )
= (vk ek ) · (σ ij ,i ej ) = vk σ ij ,i δ kj
= vj σ ij ,i
= (vj σ ij ),i −vj ,i σ ij . (5.8)
To convert this expression back to direct notation, we note the following:
¯ · (σ · v)
∇ = ∇¯ · (σ ij ei ej · vk ek ) = ∇ ¯ · (σ ij vk δ jk ei )
∂
= ek · (σ ij vj ei ) = (σ ij vj ),k δ ki = (σ ij vj ),i
∂xk
¯ ∂
σ : ∇v = (σ ij ei ej ) : ek vl el = σ ij vl ,k (ei · ek )(ej · el )
∂xk
= σ ij vl ,k δ ik δ jl = σ ij vj ,i .
Thus, Eq. (5.8) and similar manipulations for v · (∇ · t) yield
¯ · σ)
v·(∇ = ¯ · (σ · v) − σ : ∇v
∇ ¯
v·(∇ · t) = ∇ · (t · v) − t : ∇v. (5.9)
¯ = D − Ω and ∇u = FT −
In addition, Eqs. (3.212) and (3.77)1 give ∇v
I, where D is the symmetric rate-of-deformation tensor and Ω is the anti-
symmetric spin tensor. Thus, the last terms of Eqs. (5.9) can be written
¯
σ : ∇v = σ : (D − Ω) = σ : D
T
t : ∇v = t : ∇u̇ = t : Ḟ , (5.10)
where σ : Ω = 0 since the double-dot product of a symmetric tensor (σ)

and an antisymmetric tensor (Ω) is zero. Now, substituting Eqs. (5.9) and
(5.10) into (5.6) and (5.7) yields

K̇ = [∇¯ · (σ · v) − σ : D + f · v] dv
v
T
= [∇ · (t · v) − t : Ḟ +f 0 · v] dV. (5.11)
V
The Eulerian form of the second term in Eq. (5.1), given by differenti-
ating (5.3), is

d
U̇ = (ρ dv) u + u̇ρ dv
dt
v
= u̇ρ dv. (5.12)
v
Similarly, the Lagrangian form is

U̇ = u̇ρ0 dV. (5.13)
V
Next, the surface integrals in Eqs. (5.4) and (5.5) are converted into
volume integrals. Substituting Eqs. (4.23) and (4.25)1 into (5.4) and using
the divergence theorem (2.200) yields

P = n · σ · v da + f · v dv
a v
= ¯ · (σ · v) + f · v ] dv
[∇ (5.14)
v
in terms of Eulerian quantities and

P = N · t · v dA + f0 · v dV
A V
= [∇ · (t · v) + f 0 · v ] dV (5.15)
V
in terms of Lagrangian quantities. Similarly, applying the divergence the-

orem to Eq. (5.5) gives

Q = (ρr − ∇ ¯ · q) dv
v
= (ρ0 r − ∇ · q0 ) dV. (5.16)

V
Finally, inserting Eqs. (5.11)–(5.16) into (5.1) and simplifying yields

(ρu̇ − σ : D−ρr + ∇ ¯ · q) dv = 0
v
(ρ0 u̇ − t : ḞT −ρ0 r + ∇ · q0 ) dV = 0, (5.17)

V
which are the global forms of the first law of thermodynamics (balance of
energy). For an arbitrary volume element that deforms from dV into dv,
these relations imply
¯ ·q=0
ρu̇ − σ : D−ρr + ∇
ρ0 u̇ − t : ḞT −ρ0 r + ∇ · q0 = 0, (5.18)
which are the local forms of the first law. According to these equations, the
rate of increase in internal energy of a volume element (ρu̇ or ρ0 u̇) is equal
to the sum of the rate of work done by the stresses on the element, i.e., the
stress power (σ : D or t : ḞT ), the rate of internal heat production (ρr
or ρ0 r), and the rate of heat flow into the element (−∇ ¯ · q or −∇ · q0 ).
The stress power appears in Eqs. (5.18) in terms of the Cauchy and first
Piola-Kirchhoff stress tensors. We now derive the relation between these
two forms for the stress power. First, since σ : Ω = 0, we can write
−1
σ : D = σ : (D + Ω) = σ : L = σ : (Ḟ · F ), (5.19)
where Eqs. (3.214) and (3.207) have been substituted. Next, applying for-
mulas from Table 2.2 (page 22) with T = σ, U = Ḟ, and V = F−1 gives
−T
σ:D = Ḟ : (σ · F )
−T
= (σ · F ) : Ḟ = (σ · F−T )T : ḞT
= (F−1 · σ T ) : ḞT = (F−1 · σ) : ḞT ,
where the third line uses the symmetry property of the Cauchy stress tensor.
Finally, substituting Eq. (4.29)1 gives the result
T
σ : D = J −1 t : Ḟ . (5.20)
A third form for the stress power can be written in terms of the second
Piola-Kirchhoff stress tensor s. Substituting Eq. (3.217) for D into (5.20)
yields
T −1
t : Ḟ = J σ : D = Jσ : (F−T · Ė · F ).
This expression is transformed by setting T = F−1 , UT = F−T · Ė, and
V = σ in the next-to-last formula of Table 2.2 to get
T
t : ḞT = J ( F−T · Ė) : (F−1 · σ T )
T
= J (F−1 · σ T ) : (Ė · F−1 )
= J (F−1 · σ) : (Ė · F−1 )
since σ = σ T and E = ET . Now, on setting T = Ė, U = F−1 · σ, and
VT = F−1 , the next-to-last line in Table 2.2 gives
t : ḞT = J Ė :(F−1 · σ · F−T )
= Ė : s = s : Ė, (5.21)
in which Eq. (4.29)2 has been used. Thus, Eqs. (5.20) and (5.21) give the
stress power in the three forms
Ps = σ : D = J −1 t : ḞT = J −1 s : Ė. (5.22)
Therefore, an alternative Lagrangian form for the first law of thermody-

namics (5.18)2 is
ρ0 u̇ − s : Ė−ρ0 r + ∇ · q0 = 0. (5.23)
Example 5.1 Show by direct manipulation that the two equations in

(5.18) are equivalent.
Solution. Multiplying the second relation by J −1 gives
ρu̇ − σ : D−ρr + J −1 ∇ · q0 = 0,
where ρ0 = ρJ and Eq. (5.22) have been used. All that is left to do is to
show that J −1 ∇ · q0 = ∇
¯ · q. To do this, we first note that q0 is the heat
flux per unit undeformed area dA. Since the area vector dA deforms into
da, the total amount of heat per unit time passing across the area element
is
q0 · dA = q · da. (5.24)
Next, substituting Eq. (3.128) gives

q0 · dA = q · (J dA · F−1 )
= J q · (F−T · dA)
= J (q · F−T ) · dA
or
q0 = J q · F−T = J F−1 · q. (5.25)
Thus, we have
J −1 ∇ · q0 = J −1 ∇ · (J F−1 · q).
Finally, replacing σ by q0 in Eq. (4.80) shows that this expression becomes
J −1 ∇ · q0 = ∇
¯ · q, (5.26)
which is the desired result.
5.1.2 Second Law of Thermodynamics

The first law of thermodynamics states that work can be converted into
heat and vice versa, so long as the total energy contained in a closed sys-
tem remains constant. However, dissipated heat energy, such as that due
to friction or viscosity, cannot be converted into work. The second law of
thermodynamics restricts the direction of energy conversion processes. This
principle is based on the concept of entropy, which is a measure of the
disorder in a system due to an increase in heat energy. During a reversible
process, the total entropy in a system is conserved, but during an irre-
versible process, the system gains entropy due to heat input. Friction is
an example of an irreversible process.
In general terms, the specific entropy η (entropy per unit mass) is defined
through the inequality
2
dQ̄
Δη = η 2 − η 1 ≥ (5.27)
1 T
between two thermodynamic states, where Q̄ is the heat input per unit
mass and T is the absolute temperature. Equality holds for a reversible
process, whereas the inequality indicates that entropy is generated during
an irreversible process. Equation (5.27) defines change in entropy as a
measure of the change in energy dissipation with respect to temperature.
For a solid body, the second law of thermodynamics states that the time-
rate of change of the total entropy in the body is greater than or equal to
the sum of the influx of entropy through the surface of the body and the
entropy generated by internal heat sources. In mathematical terms, using
Eq. (5.5) to generalize Eq. (5.27) gives the second law in Eulerian rate form
as

d q ρr
ρη dv ≥ − · n da + dv (5.28)
dt v a T v T
and in Lagrangian form as

d q0 ρ0 r
ρ η dV ≥ − · N dA + dV. (5.29)
dt V 0 A T V T
The local forms of the second law follow by applying the divergence
theorem (2.200) to the area integrals in the above equations to obtain

ρr ¯ · q
ρη̇ − +∇ dv ≥ 0
v T T
q
ρ r 0
ρ0 η̇ − 0 + ∇ · dV ≥ 0. (5.30)
V T T
These integral relations imply the local equations
r 1 ¯ q
η̇ − + ∇ · ≥0
T ρ T
r 1 q
0
η̇ − + ∇· ≥ 0, (5.31)
T ρ0 T
which are the Eulerian and Lagrangian forms of the Dissipation

(Clausius-Duhem) inequality. These equations can be written in al-
ternate forms by expanding the terms involving the gradient operators. To
do this, we use the defining relations (3.18) to find
∂ q
¯ · q
∇ = gi i ·
T ∂x T
q,i T − qi T,i
= gi ·
T2
1 T,i
= gi · q,i − 2 gi · q
T T
1 ∂ 1 ∂T
= gi i · q − 2 q · gi i
T ∂x T ∂x
1 ¯ 1 ¯
= ∇ · q − 2 q · ∇T
T T
5.2. Fundamental Constitutive Principles 181
with a similar expression for ∇ · (q0 /T ). Thus, Eqs. (5.31) become
r 1 ¯ 1 ¯ ≥0
η̇ − + ∇ · q− 2 q · ∇T
T ρT ρT
r 1 1
η̇ − + ∇ · q0 − q0 · ∇T ≥ 0. (5.32)
T ρ0 T ρ0 T 2
Since heat does not flow naturally from a cooler toward a warmer part
of a body, it is always true that
¯ ≤0
q · ∇T and q0 · ∇T ≤ 0,
i.e., the temperature gradient and the heat flux have opposite signs unless
¯ = 0, in which case q = 0. These terms represent the entropy intro-
∇T
duced into the body by heat conduction. This observation and Eqs. (5.32)
imply that the second law of thermodynamics also can be written in the
stronger forms
r 1 ¯
η̇ − + ∇·q ≥ 0
T ρT
r 1
η̇ −
+ ∇ · q0 ≥ 0, (5.33)
T ρ0 T
where equality holds for a reversible system
5.2 Fundamental Constitutive Principles
The possible forms for constitutive equations are restricted by fundamental

postulates based on physical and intuitive arguments. The constitutive
principles for a thermomechanical material include the following (Truesdell
and Noll, 2004; Malvern, 1969):
1 Coordinate Invariance. Constitutive equations must be independent
of the coordinate system used to describe the motion of a body.
2 Determinism. The stress in a body depends on (at most) the entire
thermomechanical history of the body.
3 Local Action. The stress at a point p in a body is not affected signif-
icantly by motion outside an arbitrarily small neighborhood of p.
4 Equipresence. An independent variable that appears in one constitu-
tive equation for a material must be present in all constitutive equations
for the material, unless its presence violates some other fundamental
principle.
5 Material Frame Indifference (Material Objectivity). Constitu-

tive equations must be form invariant under rigid motions of the spatial
(observer) reference frame.
6 Physical Admissibility. Constitutive equations must be consistent
with the fundamental physical balance laws (mass, momentum, energy)
and the law of entropy (second law of thermodynamics).
7 Material Symmetry. Constitutive equations must be form invariant
under certain rigid transformations of material frames, depending on
the symmetries inherent in the material.
The remainder of this section uses the first six of these principles, in
the listed order, to develop general constitutive relations for a compressible
thermoelastic material. Consideration of the principle of material symmetry
is deferred to Sec. 5.3, which examines specific forms for an elastic mate-
rial in which thermal effects are ignored. The issue of incompressibility is
addressed in Sec. 5.5.
5.2.1 Principle of Coordinate Invariance
A deforming body is oblivious to any coordinate system used to follow its

motion. Thus, like the other basic equations of mechanics, constitutive
equations should be developed in tensor form. Doing this ensures satisfac-
tion of the principle of coordinate invariance.
5.2.2 Principle of Determinism
In a general material, the stress distribution at a given time depends not

only on the instantaneous deformation and temperature, but also on the
entire history of the deformation and temperature up until this time. De-
forming a viscoelastic material, for example, involves viscous dissipation of
energy, which cannot be recovered. Since the amount of energy lost de-
pends on the path taken to a particular configuration, the stresses depend
on the entire deformation history.
During the deformation of an ideal thermoelastic material, no energy
is dissipated. Thus, the state of stress at time t depends only on the de-
formation and temperature at t. In an ideal elastic material, temperature
effects are neglected, and so the stress depends only on the instantaneous
deformation.
5.2.3 Principle of Local Action

To determine the theoretical implications of the principle of local action,
we consider the motion r(R,t) and temperature T (R,t) at a point p in a
body. In addition, the motion and temperature of an arbitrary point within
a small neighborhood of p are given by r(R̄,t) and T (R̄,t), where R and R̄
are the position vectors to the two nearby points in the undeformed body.
For a sufficiently smooth deformation in the vicinity of p, Taylor series
expansions yield
∂r
r(R̄,t) = r(R,t) + (R̄ − R) · +···
∂R
∂T
T (R̄,t) = T (R,t) + (R̄ − R) · +··· . (5.34)
∂R
Through the principle of local action, these equations indicate that the mo-
tion and temperature at a point in a thermoelastic body can be determined
from the values of these quantities and their spatial derivatives at a nearby
point.
In this book, we consider only simple materials, for which derivatives
higher than the first are ignored in Eqs. (5.34). Then, these equations and
the principle of determinism suggest that the constitutive equations for a
thermoelastic material can be written in the form
σ = σ(R, F, T, ∇T )
q = q(R, F, T, ∇T )
u = u(R, F, T, ∇T )
η = η(R, F, T, ∇T ) (5.35)
since ∂/∂R = ∇ and ∇r = F . The dependency on R allows for an in-
T
homogeneous material. In an initially homogeneous thermoelastic body,

therefore, the dependent constitutive variables at a point depend only on
the instantaneous values of the deformation gradient tensor, the tempera-
ture, and the temperature gradient at that point.
5.2.4 Principle of Equipresence

The constitutive relations (5.35) satisfy the principle of equipresence, since
the same independent variables appear in each equation. The usefulness of
this principle can be appreciated when other effects are to be included. For
example, if chemical effects are important, then chemical potentials can be
added to the list of dependencies for each constitutive variable.
In some instances, it is useful to use entropy as an independent vari-

able, rather than temperature. Then, by the principle of equipresence,
Eqs. (5.35) can be replaced by the alternate constitutive relations
σ = σ(R, F, η, ∇η)
q = q(R, F, η, ∇η)
u = u(R, F, η, ∇η)
T = T (R, F, η, ∇η). (5.36)
5.2.5 Principle of Material Frame Indifference

The fundamental idea behind the principle of material frame indifference
is that, if relativistic effects are ignored, the events occurring at a point
in a body must be independent of the motion of the observer. Because
internal forces, lengths, and temperature are frame indifferent (objective)
quantities (see Sec. 2.1), the form of the constitutive equations must be
invariant under rigid motions of the reference frame. Here, ignoring shifts
in the time frame, we derive a mathematical formulation of this principle.
As discussed in Sec. 2.1, a change in coordinates is not the same as
a change in reference frame. There it was pointed out that the form of
Newton’s second law of motion f = ma is invariant under a change of coor-
dinates in a given frame of reference, but it is not invariant under a general
change of frame. The reason for this is that acceleration is not frame indif-
ferent (see below). Writing an equation in tensor form ensures invariance
under a time-independent change in coordinates, but ensuring invariance
under a time-dependent change between frames in relative motion requires
that the tensors satisfy certain transformation relations. (Note that a scalar
point function, such as temperature, is inherently frame indifferent.) These
transformation relations are derived in the following paragraphs.
Position Vectors. Consider two reference frames A and A∗ that are in

relative motion (Fig. 5.1). A set of Cartesian coordinate axes is fixed in
each frame, with observers stationed at the respective origins O and O ∗.
It is important to note that these observers are “glued” in position so that
they move with their respective frames. Thus, a point p fixed in frame
A appears stationary to the observer at O, but to the observer at O ∗ , p
appears to be moving.
The position vectors r and r∗ that locate p with respect to points O and
∗
O , respectively, are related (Fig. 5.1). Let the translation and rotation of
a Q
p
r* r
O
b
O* A
A*
Fig. 5.1. Two reference frames (A and A∗ ) in relative motion. The point p and the
vector a are fixed in frame A, and r and r∗ are position vectors of p relative to coordinate
systems fixed in A and A∗ , respectively.
frame A relative to frame A∗ be described by the displacement vector b(t)

and the rotation tensor Q(t), with the latter satisfying the conditions
QT · Q = Q · QT = I, det Q = 1 (5.37)
for a proper orthogonal tensor (see Sec. 2.6.3). Then, while r is constant
relative to the observer in A, the observer in A∗ sees r as the rotating vector
Q(t) · r. Thus, as seen from O ∗ , the position of p is
r∗ (t) = Q(t) · r + b(t). (5.38)
Note that this equation does not seem to agree with our global view (relative
to the “fixed” stars) of the geometry in Fig. 5.1, which suggests that r∗ =
r + b. The reason for this seeming contradiction is that Eq. (5.38) expresses
a relation between position vectors as they appear to observers in relative
motion. Although not frame indifferent, motions that satisfy Eq. (5.38)
are called equivalent motions, as they represent the motion of the same
point as seen by observers in two different reference frames.
Vectors. Equation (5.38) links the two distinct position vectors r and r∗,
which locate the same point in space. We now examine how the observers
in frames A and A∗ view a single vector a that is fixed in A (Fig. 5.1).
Obviously, a does not appear to move relative to the observer in A, just
as a tree does not appear to move to a person standing on the surface of
the earth. But relative to the observer in A∗ , the vector a translates and
rotates. Translation does not change a vector, but rotation does. Thus,
even though a is not changing in frame A, the relative motion of frame A∗
makes a appear to be changing according to the relation
a∗ = Q · a (5.39)
as viewed by the observer at O ∗ . This transformation ensures that the

vector a is frame indifferent.
In other words, if the observers at O and O ∗ simultaneously measure
the length and orientation of the vector a, the vectors constructed from
these measurements must satisfy Eq. (5.39) at each instant of time. Thus,
although a appears to be changing to the observer in A∗ , it is constant
relative to the observer (or a body) moving with frame A.
Example 5.2 Using the transformation (5.39), show that (a) the length of
a vector and the angle between two vectors fixed in frame A are objective
quantities; and (b) that velocity is not an objective quantity.
Solution. (a) Clearly, the length of a vector and the angle between two
fixed vectors are scalar invariants under (non-relativistic) changes in the
frame of reference. As measured in A, the squared length of a is ds2 = a · a.
As measured in A∗ , the squared length of a is
ds∗2 = a∗ · a∗ = (Q · a) · (Q · a)
T
= (a · QT ) · (Q · a) = a · (Q · Q) · a
= a·I·a = a·a
= ds2 , (5.40)
where Eqs. (5.37) and (5.39) have been used. Thus, as expected, the length
of a is independent of the motion of the observer.
Next, consider the angle between two vectors a and b fixed in frame A.
As seen from A∗ , Eq. (5.39) gives
a∗ = Q · a, b∗ = Q · b, (5.41)
and so
a∗ · b∗ = (Q · a) · (Q · b) = (a · QT ) · (Q · b)
T
= a · (Q · Q) · b = a · I · b
= a · b. (5.42)
∗
Thus, cos(a, b) = cos(a∗ , b ), which implies that angles are invariant under
changes of reference frame.
(b) If the velocity of a point in frame A is ṙ, then Eq. (5.38) shows that
the velocity of the point as seen from A∗ is
v∗ = ṙ∗ = Q · ṙ + Q̇ · r + ḃ
= Q · v + Q̇ · r + ḃ. (5.43)
Clearly, the velocity vector is not objective, as it does not satisfy v∗ = Q·v.
A similar calculation shows that acceleration also is not frame indifferent.
Example 5.3 Use Eq. (5.38) to derive (5.39).

Solution. Consider the position vectors to the ends of the vector a,
which is fixed in frame A (Fig. 5.2). Relative to the observers in A and A∗ ,
respectively, the geometry gives
a = r2 − r1 , a∗ = r∗2 − r∗1 , (5.44)
r∗1 = Q · r1 +b
r∗2 = Q · r2 +b. (5.45)
a∗ = r∗2 − r∗1 = Q · (r2 − r1 ) = Q · a,
which is Eq. (5.39).
Q
a
r2* r2
r1
r1*
b O
O* A
A*
Fig. 5.2. Position vectors to the ends of vector a fixed in reference frame A, which is in
motion relative to frame A∗ .
Tensors. The objectivity condition for a second-order tensor T follows

from its defining relation (2.31). Suppose the observer in frame A witnesses
the transformation
b = T·a (5.46)
between the two vectors a and b fixed in A. Frame indifference requires

that this transformation must have the same form relative to the observer
in A∗ , i.e.,
b∗ = T∗ · a∗ , (5.47)
∗ ∗ ∗
where a and b satisfy Eqs. (5.41), and T is the tensor T as seen from
A∗ .
Inserting Eq. (5.46) into (5.41)2 yields
b∗ Q · b = Q · (T · a) = Q · T · (Q · a∗ )
T
=
(Q · T · Q ) · a∗ ,
T
=
in which the substitution a = QT · a∗ follows from Eqs. (5.41)1 and (5.37).
Comparing this expression with Eq. (5.47) reveals that
T∗ = Q · T · QT (5.48)
is the appropriate transformation relation.
Example 5.4 Suppose the governing equation for a physical system has
the form
a = A·B·b+C·c (5.49)
relative to an observer in frame A, where a, b, and c are vectors and A,
B, and C are second-order tensors. According to the principle of material
frame indifference, this equation must be seen as
a∗ = A∗ · B∗ · b∗ + C∗ · c∗ (5.50)
to an observer in frame A∗ . Show that the vector and tensor transformation
relations ensure that Eqs. (5.49) and (5.50) are equivalent.
Solution. With Eqs. (5.37), (5.39), and (5.48), Eq. (5.50) takes the form
T T
Q·a = (Q · A · Q ) · (Q · B · Q ) · (Q · b)
T
+(Q · C · Q ) · (Q · c)
T T
= Q · A · (Q · Q) · B · (Q · Q) · b
T
+Q · C · (Q · Q) · c
= Q·A·I·B·I·b+Q·C·I·c
= Q · (A · B · b + C · c).
For arbitrary Q, this relation implies Eq. (5.49), confirming its objective
character.
dr
A
dR O
O, O*
A, A*
t=0
O* A*
t>0
Fig. 5.3. Deformation of a body in reference frame A as seen by observers stationed in

frames A and A∗ , which are in relative motion. At t = 0, these frames coincide.
Deformation Gradient Tensor. The deformation gradient tensor F

contains information about the deformation and rigid-body rotation of a
material element in a body. Deformation is an objective quantity, since
it describes changes in lengths and angles, and, relative to a given refer-
ence configuration, rotation also is frame indifferent. These considerations
suggest that F is an objective measure. While it may seem that, being a
second-order tensor, F should transform according to Eq. (5.48) under a
change of frame, the need for a reference configuration alters the transfor-
mation relation.
To determine the proper transformation, we consider two reference
frames A and A∗ that coincide at t = 0, when the body is undeformed
(Fig. 5.3). At this instant, the observers stationed in both frames view the
undeformed line element dR in the same way. For t > 0, the two frames
move apart, with the deformed body following the rigid-body motion of
frame A (Fig. 5.3). The vector dr, the deformed image of dR, then appears
differently to the observers in A and A∗ , with Eq. (5.39) giving
dr∗ = Q · dr. (5.51)
Material frame indifference demands that the forms of the defining rela-
tions for F be the same in all reference frames. Since dR appears the same
to both observers in the coinciding frames at t = 0, Eq. (3.19) gives

dr = F · dR
∗
dr = F∗ · dR (5.52)
as seen later by the observers in the separated frames A and A∗ , respec-
tively. Substituting Eq. (5.52)1 into (5.51) yields
dr∗ = Q · (F · dR) = (Q · F) · dR,
and comparison with (5.52)2 shows that
F∗ = Q · F. (5.53)
This relation, which has the form of Eq. (5.39), indicates that the defor-
mation gradient tensor transforms like a vector under a change of reference
frame.
Rather than the mapping of dR into dr, we could consider the mapping
of the base vectors GI , assumed to be embedded in the undeformed body,
into the convected base vectors gI in the deformed body. In this case, both
observers view the GI the same way at t = 0, but the gI are seen differently
in the two frames after deformation. Thus, Eq. (3.22)1 gives
F = gI G I
F∗ = gI∗GI (5.54)
∗
as seen in frames A and A , respectively. Substituting the transformation
relation
gI∗ = Q · gI
for the convected base vectors into (5.54)2 yields
F∗ = (Q · gI )GI = Q · (gI GI ) = Q · F,
Summary. Under a change of reference frame, the following transforma-

tions apply:
Motion: r∗ = Q · r + b
Scalar: φ∗ = φ
Vector: a∗ = Q · a
Second-Order Tensor: T∗ = Q · T · QT
Deformation Gradient Tensor: F∗ = Q · F (5.55)
Scalars are always frame indifferent,1 but objective vectors and tensors must
either remain unaffected or transform according to the above relations.
Within a single frame of reference, equations are coordinate invariant if
they are written in tensor form. They also are frame indifferent if all their
variables transform according to Eqs. (5.55)2-5 under a change of frame.
Example 5.5 Determine whether the velocity gradient tensor L, the rate-
of-deformation tensor D, and the spin tensor Ω are objective quantities.
Solution. As seen in frame A, Eqs. (3.207) and (3.211) give
L = Ḟ · F−1
1
D = 2
(L + LT )
1
Ω = − LT ).
2 (L
Likewise, the observer in frame A∗ sees
L∗ = Ḟ∗ · (F∗ )−1
∗T
D∗ = 1 ∗
2 (L +L )
Ω∗ = 1
2
(L∗ − L∗T ).
The transformation relation F∗ = Q · F yields
Ḟ∗ = Q̇ · F + Q · Ḟ
(F∗ )−1 = F−1 · Q−1 = F−1 · QT ,
which are substituted into the above expression for L∗ to obtain
L∗ = (Q̇ · F + Q · Ḟ) · (F−1 · QT )
= Q̇ · QT + Q · L · QT .
Because of the term Q̇ · QT , we find that L is not frame indifferent.
Next, inserting this result into the relation for D∗ gives
T T
D∗ = 12 (Q̇ · QT + Q · Q̇ ) + 12 Q · (L+LT ) · Q .
Since Q is an orthogonal tensor, the first term in parentheses becomes
T d d
Q̇ · QT + Q · Q̇ = (Q · QT ) = (I) = 0.
dt dt
Thus,
D ∗ = Q · D · QT ,
which shows by (5.55)4 that D is objective. A similar analysis yields
T
Ω∗ = 12 (Q̇ · QT − Q · Q̇ ) + Q · Ω · QT ,
in which the first term generally does not disappear. Hence, Ω is not frame
indifferent.
1 Accordingly, scalars such as energy, ∇ · a, and A : B are objective quantities.
Implications for Constitutive Equations. Contact forces within a

body are assumed to be frame indifferent, as are the differential areas on
which they act. Thus, the Cauchy stress tensor is an objective quantity.
Since deformation and temperature also are objective, it follows that the
constitutive relations (5.35) for a thermoelastic material are frame indif-
ferent. This idea, that the intrinsic physical properties of a material are
invariant relative to a change in observer, is a fundamental principle of
mechanics.
Consider a simple material with a constitutive relation of the form2
σ = g(σ) (F) (5.56)
where the tensor function g(σ) is called the response function. Here,
we assume that the response function is determined through experiments
conducted in frame A. Then, if the functional form of the constitutive
equation is independent of the motion of the observer, Eq. (5.56) must
appear as
σ ∗ = g(σ) (F∗ ) (5.57)
to an observer in frame A∗ , where σ ∗ and F∗ are related to σ and F
through the appropriate objectivity relations. Since the Cauchy stress ten-
sor depends only on the forces and geometry of the deformed body, σ
transforms according to Eq. (5.55)4 under a change of frame.3 Hence, us-
ing Eqs. (5.55)4-5 in (5.57) yields
Q · σ · QT = g(σ) (Q · F). (5.58)
This equation must hold for an arbitrary rotation Q. Convenient forms
for the constitutive relation are obtained by choosing Q = ΘT , where Θ
is the rotation tensor defined by the polar decomposition (3.147). In this
case, (5.58) becomes
ΘT · σ · Θ = g(σ) (ΘT · F). (5.59)
Moreover, since ΘT = Θ−1 , Eq. (3.147) gives U = ΘT ·F, and (5.59) yields
T
σ = Θ · g(σ) (U) · Θ ,
which separates the contributions of deformation U and rotation Θ. Other
forms follow by noting Eqs. (3.28) and (3.153), which give U2 = C = I+2E.
2 Forconvenience, the other independent variables of Eq. (5.35)1 are not listed explicitly.
3 Equation (5.55)4 may not apply to the Piola-Kirchhoff stress tensors t and s, since
these tensors depend on reference configurations in addition to that of the deformed
body (see Problem 5.3).
Hence, Eq. (5.56) and the principle of material frame indifference provide
constitutive equations in the forms
T
σ = Θ · g(σ) (U) · Θ
T
or σ = Θ · f (σ) (C) · Θ
or σ = Θ · h(σ) (E) · ΘT . (5.60)
To write the constitutive relations for the Piola-Kirchhoff stress tensors,

we use Eq. (4.28), i.e.,
σ = J −1 F · t = J −1 F · s · FT . (5.61)
Substituting Eq. (5.60)1 gives
J −1 F · t = Θ · g(σ) (U) · Θ
T
or
J −1 (ΘT · F) · t = g(σ) (U) · Θ .
T
Since ΘT · F = U and J = det F = det(Θ · U) = det Θ det U = det U, we

have
t = g(t)(U) · ΘT ,
where
g(t) (U) = (det U) U−1 · g(σ) (U).
Thus, in terms of the first Piola-Kirchhoff stress tensor, the constitutive
relations take the forms
t = g(t) (U) · ΘT
T
or t = f(t) (C) · Θ
T
or t = h(t) (E) · Θ . (5.62)
For the second Piola-Kirchhoff stress tensor, Eqs. (5.61) and (5.62)1 give
s = t · F−T = g(t)(U) · Θ · F−T ,
T
and the symmetry property of s allows us to write

s = sT = (ΘT · F−T )T · [g(t)(U)]T
= (F−1 · Θ) · [g(t)(U)]T
= U−1 · [g(t)(U)]T ≡ g(s) (U).
The last expression follows from the equation U−1 = F−1 · Θ, as verified
by the manipulations
U · U−1 = (Θ−1 · F) · (F−1 · Θ)
= Θ−1 · I · Θ = I,
where Eq. (3.147) has been used. Thus, the constitutive equations also can
be written in the forms
s = g(s)(U)
or s = f(s)(C)
or s = h(s)(E). (5.63)
These relations for the second Piola-Kirchhoff stress tensor are particularly
convenient since rotation does not appear explicitly. For small rotation
(Θ ∼
= I), Eqs. (5.60), (5.62), and (5.63) are all approximately the same.
5.2.6 Principle of Physical Admissibility

The principle of physical admissibility stipulates that constitutive relations
must not conflict with any of the fundamental laws of continuum mechanics.
For example, we already have seen that the principle of angular momentum
leads to the conclusion that the Cauchy stress tensor is symmetric. Thus,
the constitutive equations must not violate this symmetry. Here, we exam-
ine the consequences of the first and second laws of thermodynamics on the
form of the constitutive equations.
The present development focuses on a Lagrangian formulation for ther-
moelastic materials. The second Piola-Kirchhoff stress tensor s and the
Lagrangian strain tensor E are taken as basic variables. Thus, the consti-
tutive equations (5.35) and (5.36) are replaced by the alternate relations
s = s(R, E, T, ∇T )
q0 = q0 (R, E, T, ∇T )
u = u(R, E, T, ∇T )
η = η(R, E, T, ∇T ) (5.64)
and
s = s(R, E, η, ∇η)
q0 = q0 (R, E, η, ∇η)
u = u(R, E, η, ∇η)
T = T (R, E, η, ∇η). (5.65)
The state of stress in a thermoelastic body depends only on the instan-

taneous deformation and temperature (or entropy) fields. Since no energy is
dissipated during loading, the deformation is a reversible process governed
by the thermodynamic relations
ρ0 u̇ − s : Ė−ρ0 r + ∇ · q0 = 0
r 1
η̇ − + ∇ · q0 = 0 (5.66)
T ρ0 T
as given by Eqs. (5.23) and (5.33)2 . Eliminating r between these two equa-
tions yields
ρ0 (T η̇ − u̇) + s : Ė = 0. (5.67)
In the following, the general constitutive relations (5.64) and (5.65) are sub-
stituted into this equation to examine the cases for which (1) temperature
is an independent variable and (2) entropy is an independent variable.
Case 1: When temperature is taken as an independent variable, it is
convenient to introduce the Helmholtz free-energy function
ψ = u − Tη (5.68)
as a replacement for the internal energy u. Then, Eq. (5.64)3 is replaced

by
ψ = ψ(R, E, T, ∇T ), (5.69)
and using Eq. (5.68) to eliminate u in (5.67) gives
−ρ0 (ψ̇ + η Ṫ ) + s : Ė = 0. (5.70)
Next, substituting Eq. (5.69) into (5.70) and using the chain rule yields
∂ψ ∂ψ ∂ψ
s − ρ0 : Ė − ρ0 η + Ṫ − ρ0 · ∇Ṫ = 0.
∂E ∂T ∂∇T
For independent strain and temperature distributions, this equation implies
the relations
∂ψ
s = ρ0
∂E
∂ψ
η = −
∂T
∂ψ
= 0. (5.71)
∂∇T
Thus, stress and entropy can be computed directly from ψ, and the last
equation shows that ψ is independent of the temperature gradient. The
first two of (5.71) represent the constitutive relations for a thermoelastic
material with temperature taken as an independent variable.
If the deformation is isothermal (Ṫ = 0), then ψ = ψ(R, E) and the
material behaves elastically. In this case, we define the strain-energy
density function
W (R, E) = ρ0 ψ (5.72)
per unit undeformed volume. Since the density ρ0 of the undeformed body
is independent of the deformation, Eq. (5.71)1 gives the constitutive equa-
tion
∂W
s= . (5.73)
∂E
Case 2: If entropy is taken as an independent variable, then the internal
energy is retained in the formulation. Substituting Eq. (5.65)3 into (5.67)
and using the chain rule gives
∂u ∂u ∂u
s − ρ0 : Ė + ρ0 T − η̇ − ρ0 · ∇η̇ = 0.
∂E ∂η ∂∇η
For independent strain and entropy fields, this equation yields
∂u
s = ρ0
∂E
∂u
T =
∂η
∂u
= 0. (5.74)
∂∇η
Thus, stress and temperature can be computed directly from u, and the
last equation shows that u is independent of the entropy gradient. The
first two of (5.74) represent the constitutive relations for a thermoelastic
material with entropy taken as an independent variable.
If the deformation is isentropic (η̇ = 0), then u = u(R, E) and the
material behaves elastically. In this case, we take
W (R, E) = ρ0 u, (5.75)
and Eq. (5.74)1 takes the form of Eq. (5.73).
A material possessing a constitutive relation of the form (5.73) is called
a hyperelastic material. In other words, the mechanical properties of a
hyperelastic material are characterized completely by a scalar strain-energy
density function W . During an isothermal deformation, W is associated

with the free energy per unit undeformed volume. During an isentropic
deformation, W corresponds to the internal energy per unit undeformed
volume.
The constitutive relation (5.73) can be written readily in terms of the
Cauchy and first Piola-Kirchhoff stress tensors. Substitution into (4.28)
and (4.29)1 yields
∂W
σ = J −1 F · · FT
∂E
∂W
t = · FT . (5.76)
∂E
An alternate form for t can be found by noting that, via Eq. (5.22), the
T
stress power s : Ė in Eq. (5.67) can be replaced by t : Ḟ . Moreover, we
replace s by t and E by FT in Eqs. (5.64) and (5.65). Then, manipulations
similar to those discussed above lead to
∂W
t= . (5.77)
∂FT
In summary, the constitutive equations for a compressible hyperelastic
material can be written in the following forms:
∂W
σ = J −1 F · · FT
∂E
∂W ∂W
t= · FT =
∂E ∂FT
∂W
s= . (5.78)
∂E
Note that these relations are consistent with σ and s being symmetric
tensors (since E = ET ), while t is not symmetric in general.
Scalar forms of these equations can be found by direct substitution of
Eqs. (3.22), (3.30), and (4.30). Thus, Eq. (5.78)1 gives
∂W
σ kl gk gl = J −1 (FK
k
gk G K ) · GI GJ · (FLl GL gl )
∂EIJ
∂W K L
= J −1 FK
k l
FL δ δ gk gl
∂EIJ I J
∂W
= J −1 FIk FJl gk gl (5.79)
∂EIJ
or
∗
J∗ ∂W
σI gI gJ = J −1 (gK GK ) · GI GJ · (GL gL )
∂EIJ
∂W K L
= J −1 δ δ gK gL
∂EIJ I J
∂W
= J −1 gI gJ ; (5.80)
∂EIJ
Eq. (5.78)2 gives
∂W
tIk GI gk = GI GJ · (FK k
G K gk )
∂EIJ
∂W
= F k δ K G I gk
∂EIJ K J
∂W
= F k G I gk (5.81)
∂EIJ J
or
∂W ∂W
tI·j GI gj = = GI g j ; (5.82)
∂(FIj GI g j) ∂FIj
and Eq. (5.78)3 gives

∂W
sIJ GI GJ = GI GJ . (5.83)
∂EIJ
From these equations, component forms of the constitutive relations for
a compressible hyperelastic material can be written as
∂W
σ ij = J −1 FIi FJj
∂EIJ
∗
J∗ ∂W
σI = J −1
∂EIJ
∂W
tIj = FJj
∂EIJ
∂W
tI·j =
∂FIj
∂W
sIJ = . (5.84)
∂EIJ
Other forms can be found similarly.

Example 5.6 Derive Eq. (5.84)3 directly from (5.84)4 .

Solution. To do this, we need the relation
EIJ = 12 (FIi FJj gij − GIJ ) (5.85)
as given by Eqs. (3.31)1 and (3.32)1 . Then, with the chain rule and W =
W (EIJ ), Eq. (5.84)4 can be written
∂W ∂W ∂EKL
tI·j = =
∂FIj ∂EKL ∂FIj
∂W ∂ 1 m n
= F F gmn
∂EKL ∂FIj 2 K L
# $
m n
1 ∂W ∂FK n m ∂FL
= gmn FL + FK
2 ∂EKL ∂FIj ∂FIj
1 ∂W
= gmn δ m I n m n I
j δ K FL + FK δ j δ L .
2 ∂EKL
Contractions over the Kronecker deltas convert this expression into
1 ∂W ∂W
tI·j = gjnFLn + gmj FK m
.
2 ∂EIL ∂EKI
Renaming indices and noting the symmetry of EIJ and gij yields
∂W
tI·j = gjnFJn .
∂EIJ
Finally, using contravariant components of the metric tensor to raise a
subscript gives
∂W
tIk = gjk tI·j = gjk gjn FJn
∂EIJ
∂W n
= δ kn F
∂EIJ J
∂W k
= F ,
∂EIJ J
where Eq. (2.91) has been used. This relation agrees with Eq. (5.84)3.
It is important to note that the strain components in the strain-energy

function must be treated independently in evaluating Eqs. (5.84). In other
words, we cannot set EJI = EIJ in the expression for W before carrying
out the differentiation ∂W/∂EIJ .
Consider, for example, a material characterized by the strain-energy
density function
2 2
W = C(E12 + E21 ) (5.86)
with C being a material constant. Equation (5.84)5 gives

∂W
s12 = = 2CE12
∂E12
∂W
s21 = = 2CE21,
∂E21
and setting E21 = E12 gives s21 = s12 as it should. However, if we set
E21 = E12 a priori to obtain
2
W = 2CE12 , (5.87)
then we get
∂W
s12 = = 4CE12
∂E12
∂W
s21 = = 0,
∂E21
which is not correct. Some authors, therefore, replace Eq. (5.84)5 by the
equivalent expression
1 ∂W ∂W
sIJ = + , (5.88)
2 ∂EIJ ∂EJI
which gives the correct result using either (5.86) or (5.87).
5.3 Strain-Energy Density Function
Thus far, we have explored the implications of six of the seven princi-
ples listed in Sec. 5.2. The seventh, the principle of material symmetry,
is used herein to deduce the form of the strain-energy density function
for orthotropic, transversely isotropic, and isotropic hyperelastic materi-
als. These types of material symmetries often are used to characterize soft
tissues. The approach used here follows Green and Adkins (1970). An
alternative method, involving theorems for tensor invariants, can be found
elsewhere (e.g., Spencer, 1984; Zheng, 1994).
This section deals with how W must depend on the strain components
for each type of material, and some specific functional forms are listed.
Section 5.7 discusses strategies for determining material parameters. It is
convenient in this section to work in local Cartesian coordinates using only
subscript notation.
5.3. Strain-Energy Density Function 201
5.3.1 Orthotropic Material

A material is orthotropic if each point of the unstressed material possesses
three mutually orthogonal planes of symmetry. The symmetry is a function
of the microstructure, which may consist of aligned, orthogonally oriented
fibers embedded in an isotropic matrix. Skin is but one example. Local
Cartesian axes (Z1 , Z2 , Z3 ) are chosen to lie in the undeformed body along
the intersections of the local planes of symmetry, making these planes the
local coordinate planes. We call the ZI axes principal material axes.
In general, the orientation of these axes can vary from point to point in a
body, but the principle of local action ensures that we need only consider
a representative point in developing constitutive relations.
Due to the material symmetry, the constitutive relations must be in-
dependent of a reflection of any of the principal material axes. In other
words, the constitutive equations must be form-invariant if the ZI axes are
replaced by the Z̄I = −ZI axes. It then follows that the strain-energy
function must be independent of this coordinate transformation.
Let IJ and ¯IJ be the components of the Lagrangian strain tensor with
respect to the ZI and Z̄I coordinates, respectively. Then, we assume that
the strain-energy density function has the functional dependencies4
W = W( 11 , 22 , 33 , 12 , 23 , 31 )
= W (¯11 , ¯22 , ¯33 , ¯12 , ¯23 , ¯31 ), (5.89)

1 ∂zk ∂zk
IJ = − δ IJ
2 ∂ZI ∂ZJ
1 ∂zk ∂zk
¯IJ = − δ IJ . (5.90)
2 ∂ Z̄I ∂ Z̄J
Here, the zi are Cartesian coordinates of the deformed image of the point
located at ZI (or Z̄I ) in the undeformed body. Note that the zi coordi-
nate system is independent of both ZI and Z̄I . With these definitions, we
consider sequentially the following symmetry transformations:
T1 : Z̄1 = −Z1 , Z̄2 = Z2 , Z̄3 = Z3
¯11 = 11 , ¯22 = 22 , ¯33 = 33
¯12 = − 12 , ¯23 = 23 , ¯31 = − 31
4 For convenience, we set
JI = IJ a priori and use Eq. (5.88) to compute stress
components.
T2 : Z̄1 = Z1 , Z̄2 = −Z2 , Z̄3 = Z3

¯11 = 11 , ¯22 = 22 , ¯33 = 33
¯12 = − 12 , ¯23 = − 23 , ¯31 = 31
(5.91)
T3 : Z̄1 = Z1 , Z̄2 = Z2 , Z̄3 = −Z3
¯11 = 11 , ¯22 = 22 , ¯33 = 33
¯12 = 12 , ¯23 = − 23 , ¯31 = − 31 .
Under the transformation T1 , Eq. (5.89)2 becomes
W = W( 11 , 22 , 33 , − 12 , 23 , − 31 ),
and consistency with (5.89)1 requires that W take the form

2 2
W = W( 11 , 22 , 33 , 23 , 12 , 31 , 12 31 ). (5.92)
Next, applying T2 to this relation gives

2 2
W = W( 11 , 22 , 33 , − 23 , 12 , 31 , − 12 31 ),
and consistency with (5.92) requires

2 2 2
W = W( 11 , 22 , 33 , 12 , 23 , 31 , 12 23 31 ). (5.93)
Furthermore, Eq. (3.142)3 gives the third strain invariant
I3 = det(δ IJ + 2 IJ )
in Cartesian coordinates. Expanding the determinant reveals that 12 23 31

can be expressed in terms of I3 and the other arguments listed in Eq. (5.93),
which therefore can be written in the form
2 2 2
W = W( 11 , 22 , 33 , 12 , 23 , 31 , I3 ). (5.94)
Applying T3 to this relation results in no further modification. Due to their

high water content, soft tissues often are assumed to be incompressible. In
this case, the strain invariant I3 = J 2 = 1 is deleted from the list of
arguments.
As a specific example, consider blood vessels. To a first approximation,
arteries can be treated as orthotropic circular cylinders, with the principal
material directions defined by the cylindrical polar coordinates (R, Θ, Z).
Experiments indicate that arterial tissue is nearly incompressible (I3 = 1)
and stiffens with increasing strain. Thus, one possible form for the strain-
energy function is (Humphrey, 2002)

W = C(eQ − 1)
2 2 2
Q = a1 RR + a2 ΘΘ + a3 ZZ + 2a4 RR ΘΘ + 2a5 ΘΘ ZZ
2 2
+2a6 ZZ RR + a7 ( RΘ + ΘR ) + a8 ( 2ΘZ + 2ZΘ ) + a9 ( 2ZR + 2
RZ ),
(5.95)
where C and the ai are material constants, and the IJ are physical compo-
nents of strain relative to the polar system.5 This form for W is consistent
with Eq. (5.94). Once W is known, stress-strain relations can be obtained
using Eq. (5.88).
5.3.2 Transversely Isotropic Material

Each point in an undeformed transversely isotropic material possesses a
single axis of symmetry, taken here as the Z3 -axis. In the Z1 Z2 -plane, the
material is isotropic, i.e., the material properties in this plane are the same
in all directions, while the properties in the Z3 direction are different. This
type of material may be composed, for example, of fibers (oriented in the
Z3 direction) embedded in an isotropic matrix. Skeletal muscle often is
treated as transversely isotropic material.
Since the Z3 -axis is an axis of symmetry, W must be invariant under
the transformation T3 of Eq. (5.91). Moreover, since the properties are
isotropic in the Z1 Z2 -plane, W also must be invariant for any rotation θ of
the Z1 and Z2 axes about the Z3 -axis. The appropriate transformation for
this case is
T4 : Z̄1 = Z1 cos θ + Z2 sin θ
Z̄2 = −Z1 sin θ + Z2 cos θ
Z̄3 = Z3
2 2
¯11 = 11 cos θ + 22 sin θ +2 12 sin θ cos θ
2 2
¯22 = 11 sin θ + 22 cos θ −2 12 sin θ cos θ
¯33 = 33
2
¯12 = ( 22 − 11 ) sin θ cos θ + 12 (cos θ − sin2 θ)
¯23 = 23 cos θ − 31 sin θ
¯31 = 23 sin θ + 31 cos θ. (5.96)

5 In the neighborhood of a point, polar coordinate axes approximate Cartesian axes.
The basis for writing W in terms of physical strain components is discussed in Sec. 5.4.4.
These relations represent the standard equations for strain transformation

under a rotation about the Z3 -axis [see Eqs. (A.53) in Appendix A].
For the transformation T4 , our main task is to find combinations of the
strain components that are independent of the rotation θ. It is relatively
easy to deduce that Eqs. (5.96) satisfy
¯11 + ¯22 = 11 + 22
¯231 + ¯223 = 2
31 + 2
23
¯33 = 33 , (5.97)
but other combinations are more difficult to find. Fortunately, Green and
Adkins (1970) have done the job for us. Direct substitution of Eqs. (5.96)
verifies the relations
Ē12 + Ē22 = E12 + E22

F̄12 + F̄22 = F12 + F22
Ē1 F̄1 + Ē2 F̄2 = E1 F1 + E2 F2
Ē1 F̄2 − Ē2 F̄1 = E1 F2 − E2 F1 , (5.98)
where
E1 = 11 − 22
E2 = 2 12
2 2
F1 = 31 − 23
F2 = 2 31 23 , (5.99)
and Ē1 , Ē2 , F̄1 , and F̄2 are given by placing bars over all quantities in
Eqs. (5.99).
For W to be independent of rotation about the Z 3 -axis, Eqs. (5.97) and
(5.98) indicate that the strain-energy function has the form
2 2 2
W = W( 11 + 22 , 31 + 23 , 33 , E1 + E22 , F12 + F22 ,
E1 F1 + E2 F2 , E1F2 − E2 F1 ), (5.100)
which also satisfies transformation T3 of (5.91). Further manipulations
show that this expression can be written more compactly. First, Eqs. (5.99)
give
E12 + E22 = ( 11 + 22 )
2
− 4( 11 22 − 2
12 )
F12 + F22 = ( 2
31 + 2 2
23 )
2 2
E1 F1 + E2 F2 = 2 det[ IJ ] +( 11 +
22 )( 31 + 23 )
2
−2 33 ( 11 22 − 12 )

E12 + E22 E1 F1 + E2 F2
(E1 F2 − E2 F1 ) 2
=
E1 F1 + E2 F2 F12 + F22 ,
and comparison with the terms in (5.100) reveals that W also can be written
as
2 2 2
W = W( 11 + 22 , 31 + 23 , 33 , 11 22 − 12 , det[ IJ ]). (5.101)
Now, comparing these terms with those in the strain invariants I1 , I2 , and
I3 of Eq. (3.142) shows that the strain-energy function for a transversely
isotropic material can be simplified to6
W = W (I1 , I2 , I3 , I4 , I5 ), (5.102)
where
I4 = 33
2 2
I5 = 31 + 23 . (5.103)
In direct form, it is easy to verify that these invariants can be written
I4 = e3 · E · e3
I5 = e3 · E2 · e3 − I42 , (5.104)
where e3 is the unit vector in the fiber direction.

Heart muscle consists of layers that contain aligned fibers embedded
in an isotropic matrix. Many researchers, therefore, treat each layer as
transversely isotropic. For passive heart muscle, possible forms for the
strain-energy density function include
μ b1 b2 (I4 −1)2
W = (I1 − 3) + e −1 , (5.105)
2 2b2
as proposed by Holzapfel et al. (2000), and

n
n
W = cij (I1 − 3)i (λf − 1)j , (5.106)
i=0 j=0
where μ, bi and cij are material constants and λf is the fiber stretch ratio
(λ2f = 1 + 2 33 = 1 + 2I4 ). Equation (5.105) separates the effects of the
isotropic matrix (first term) from those of the fibers (second term). In
contrast, Eq. (5.106) includes coupling between the matrix and fibers.
6 It is important to note here that any combination of invariants also is invariant.
5.3.3 Isotropic Material

In the undeformed configuration, the mechanical properties of an isotropic
material are independent of direction.7 Thus, interchanging any two coor-
dinate axes should not affect the form of W . Setting Z̄1 = Z2 , Z̄2 = Z3 ,
and Z̄3 = Z1 does not affect the three invariants defined by Eq. (3.142), but
I4 and I5 of Eq. (5.103) change, i.e., they are no longer invariants. Thus,
Eq. (5.102) reduces to
W = W (I1 , I2 , I3 ) (5.107)
for an isotropic material.

Although most soft tissues are not isotropic, some undifferentiated tis-
sues in the early embryo, can be treated as approximately isotropic and
incompressible (I3 = 1), with mechanical properties similar to those of rub-
ber (Taber, 2020). In this case, the strain-energy density function reduces
to the form
W = W (I1 , I2 ). (5.108)
A commonly used strain-energy function for rubber-like materials is the
Mooney-Rivlin form (Mooney, 1940; Rivlin, 1947)
W = b1 (I1 − 3) + b2 (I2 − 3), (5.109)
where b1 and b2 are material constants.8 If b2 = 0, this expression re-

duces to the so-called neo-Hookean form. More general forms that contain
Eq. (5.109) as special cases include (Rivlin and Saunders, 1951)
W = b1 (I1 − 3) + f(I2 − 3), (5.110)
with f being a general function, and (Rivlin, 1956)

n n
W = cij (I1 − 3)i (I2 − 3)j . (5.111)
i=0 j=0
Note the similarity of Eqs. (5.106) and (5.111). Another form is the Ogden
material defined by (Ogden, 1997)
n
an b n
W = (λ1 + λb2n + λb3n − 3), (5.112)
i=1
b n
7 When an initially isotropic, nonlinear material is deformed, its tangent moduli

change. Thus, if the deformation is not isotropic, the deformed material itself becomes
anisotropic. For this reason, material symmetries are defined in the unstressed material.
8 The 3s in (5.109) are needed to make W = 0 when all strains are zero. These constants
do not affect the stresses, however, which are given by differentiating W .

where the λi are principal stretch ratios. This last expression takes advan-
tage of the fact that the strain invariants always can be written in terms
of principal strains. A disadvantage is that, except for some geometrically
symmetric problems, the eigenvalue problem for the principal strains must
be solved at each point as part of the solution procedure.
If the water in a tissue is mobile, such as in articular cartilage, then the
effects of bulk material compressibility should not be ignored. For com-
pressible materials, W must include the effects of local changes in volume,
as characterized by the strain invariant I3 . One possible form for W , given
by generalizing Eq. (5.111), is

n
n
n
W = cijk (I1 − 3)i (I2 − 3)j (I3 − 1)k , (5.113)
i=0 j=0 k=0
where the cijk are constants. Another form was proposed by Blatz and Ko
(1962), who studied the strain-energy density function for foam rubbers.
Using theoretical and experimental arguments, they suggested the function

μα 1 − 2ν −ν/(1−2ν)
W = I1 − 3 + I3 −1
2 ν

μ(1 − α) 1 − 2ν ν/(1−2ν)
+ I2 /I3 − 3 + I3 − 1 , (5.114)
2 ν
where μ, ν, and α (0 ≤ α ≤ 1) are constants. For small strain, μ becomes
the shear modulus and ν becomes Poisson’s ratio. (These parameters have
no physical meaning for large strain.) In addition, for I3 = 1, W takes
the Mooney-Rivlin form (5.109) for an incompressible rubber-like material.
Finally, setting α = 0 and ν = 0.25 yields results in good agreement with
some experimental data for foam rubber (Blatz and Ko, 1962). In this case,
Eq. (5.114) becomes
1/2
W = (μ/2)(I2 /I3 + 2I3 − 5). (5.115)
Most soft tissues stiffen with increasing strain levels. This behavior is
captured, for example, by a function of the form
W = C(eQ − 1), (5.116)
where Q is given by the right-hand side of Eq. (5.109) or (5.114) for incom-
pressible or compressible tissue, respectively.
5.4 Constitutive Relations for Some Common Material

Symmetries
In the previous section, the strain-energy density function is expressed in

terms of Lagrangian strain components referred to a local set of Cartesian
axes oriented along the principal material directions. Equation (5.84)5 or
(5.88) can be used to obtain the second Piola-Kirchhoff stress components
relative to these material axes. The resulting stress-strain relations then can
be expressed in terms of stress and strain components relative to any desired
curvilinear coordinate system through appropriate tensor transformations.
This procedure is recommended, since it promotes understanding of the
fundamentals and allows for relatively straightforward adaptation to gen-
eral types of materials. Some authors [e.g., Green and Adkins (1970)],
however, prefer working with equations that are specialized to particu-
lar material symmetries. Indeed, this approach has advantages in some
problems. Thus, we now derive constitutive relations for hyperelastic com-
pressible materials possessing the symmetries considered in the previous
section.
2
X2 Z
X2'
X1
P
1' Z1
X
Fig. 5.4. Three coordinate systems defined at a point in an undeformed orthotropic ma-
terial (two-dimensional case). The dashed lines represent principal material directions.
First, we introduce some notation. At a given point P in the unde-

formed body, three coordinate systems are defined (Fig. 5.4): (1) Cartesian
coordinates Z I with base vectors eI along the principal material axes (as-
sumed to be mutually orthogonal); (2) orthogonal curvilinear coordinates

X I with base vectors GI tangent to the eI at P ; and (3) general curvi-
linear coordinates X I with base vectors GI . This last coordinate system
5.4. Constitutive Relations for Some Common Material Symmetries 209
is chosen for convenience in formulating a particular problem. Relative to

these systems, the Lagrangian strain tensor and the second Piola-Kirchhoff
stress tensor have the dyadic representations

I J
E = IJ e e = EI J GI GJ = EIJ GI GJ

IJ J
s = S eI eJ = sI GI GJ = sIJ GI GJ , (5.117)
in which contravariant base vectors are defined in the usual manner.

Double-dotting with appropriate dyads yields the transformation relations

IJ = AK L K L
I AJ EK L = AI AJ EKL

IJ L
s = AIK AJL sK = AIK AJL S KL , (5.118)
where

∂X J
AJI = eI · G J =
∂Z I
J
∂X
AJI = eI · G J =
∂Z I
∂X J
AJI = GI · GJ = . (5.119)
∂X I
With Cartesian coordinates chosen as the barred system, these relations

are consistent with the transformation components in Eq. (2.50)1 .
With this groundwork, we now derive explicit constitutive relations for
isotropic, transversely isotropic, and orthotropic materials. The experi-
mentally determined strain-energy density function is expressed most con-
veniently in terms of Cartesian strain components relative to the local ma-
terial axes. Our goal, however, is to obtain constitutive relations for stress
in general curvilinear coordinates.
In deriving constitutive relations for an isotropic material, we first work

in Cartesian coordinates. Then, the equations are converted into direct
notation, valid for any coordinate system. And finally, the components
with respect to general curvilinear coordinates are extracted.
For an isotropic material, Eq. (5.107) gives
W = W (I1 , I2 , I3 ) (5.120)
with Eqs. (3.142) providing the strain invariants9

I1 = cII
1 2
I2 = 2 (I1 − cIJ cIJ )
I3 = det[cIJ ]. (5.121)
Here, the cIJ are Cartesian components of the right Cauchy-Green defor-
mation tensor C; thus, Eq. (3.31)1 yields
IJ = 12 (cIJ − δ IJ ). (5.122)
Constitutive relations in Cartesian coordinates, given by Eq. (5.84)5,
are
∂W ∂W
SIJ = =2
∂ IJ ∂cIJ
∂W ∂I1 ∂W ∂I2 ∂W ∂I3
= 2 + + , (5.123)
∂I1 ∂cIJ ∂I2 ∂cIJ ∂I3 ∂cIJ
for which the required derivatives of the strain invariants are computed as
follows. The first two are relatively straightforward, with Eqs. (5.121)1,2
giving
∂I1 ∂cKK
= = δ IK δ JK = δ IJ (5.124)
∂cIJ ∂cIJ
∂I2 1 ∂
= (I 2 − cKL cKL )
∂cIJ 2 ∂cIJ 1
∂I1 1 ∂cKL ∂cKL
= I1 − cKL + cKL
∂cIJ 2 ∂cIJ ∂cIJ
= I1 δ IJ − cKL δ KI δ LJ
= I1 δ IJ − cIJ. (5.125)
Computing ∂I3 /∂cIJ begins with the Cayley-Hamilton theorem (2.84) in
the form
C3 − I1 C2 + I2 C−I3 I = 0. (5.126)
In terms of components, substituting C = cIJ eI eJ and I = δ IJ eI eJ yields
cIL cLK cKJ − I1 cIK cKJ + I2 cIJ − I3 δ IJ = 0, (5.127)
and setting J = I produces the contracted form
cIL cLK cKI − I1 cIK cKI + I2 cII − 3I3 = 0.
9 Since we are working in Cartesian coordinates, the use of superscript notation is sus-
pended until Eq. (5.138).

Finally, solving this equation for I3 , computing ∂I3 /∂cIJ , and noting the
above relations for the invariants and their derivatives yields
∂I3
= I2 δ IJ − I1 cIJ + cIK cKJ . (5.128)
∂cIJ
In direct notation, Eqs. (5.124), (5.125), and (5.128) become
∂I1
= I
∂C
∂I2
= I1 I − C
∂C
∂I3
= I2 I−I1 C + C2 . (5.129)
∂C
The last derivative can be put in an alternate form by first writing it as
∂I3
= C−1 · (I2 C−I1 C2 +C3 )
∂C
and then using Eq. (5.126) to get
∂I3
= I3 C−1 . (5.130)
∂C
Now, substituting Eqs. (5.129) and (5.130) into (5.123) yields
∂W
s= = 2[W1 I + W2 (I1 I − C)+W3 I3 C−1 ], (5.131)
∂E
where
∂W
Wi ≡ . (5.132)
∂Ii
With Eqs. (3.136) providing the strain invariants in terms of C = I + 2E,

double-dotting Eq. (5.131) with appropriate dyads yields the constitutive
relations in any coordinate system of interest.
For the Cauchy stress tensor, Eqs. (4.28) and (5.131) give
σ = 2J −1 F · [W1 I + W2 (I1 I − C) + W3 I3 C−1 ] · FT . (5.133)
This expression can be simplified using the relations
FT · F = C = CT
F · FT = B = BT . (5.134)
In addition, since C−1 = F−1 · F−T , direct substitution shows that
F · C · FT = B2
F · C−1 · FT = I. (5.135)
Inserting these equations into (5.133) yields the tensor equation
σ = α0 I+α1 B+α2 B2 , (5.136)

1/2
where the response functions are (since J = I3 )
1/2
α0 = 2I3 W3
−1/2
α1 = 2I3 (W1 + I1 W2 )
−1/2
α2 = −2I3 W2 . (5.137)
Lastly, we derive a component form of Eq. (5.136) for general curvilinear

coordinates. Green and Zerna (1968) and Green and Adkins (1970) work
in terms of the convected Cauchy stress components defined by
∗
J∗
σ = σI gI gJ , (5.138)
in which the gI = F · GI are base vectors convected from a set of specified
∗ ∗
base vectors GI . To obtain explicit constitutive relations for the σ I J , we
first use F in the form given by Eq. (3.22)1 to get
J
B = F · FT = (gI GI ) · (G gJ )
= GIJ gI gJ
2
B = B · B = (GIJ gI gJ ) · (GKLgK gL )
= GIJ GKLgJK gI gL
= GIL GKJ gLK gI gJ . (5.139)
With these expressions, (5.136) gives
∗
J∗
σI = σ : gI gJ = gI · σ · gJ
= gI · (α0 I+α1 B+α2 B2 ) · gJ
= α0 gIJ +α1 GIJ +α2 GIL GKJ gLK .
Finally, substituting Eqs. (5.137) and rearranging yields
∗
J∗
σI = ΦGIJ + ΨH IJ − pgIJ , (5.140)
where
−1/2
Φ = 2I3 W1
−1/2
Ψ= 2I3 W2
1/2
p = −2I3 W3
H IJ
= I1 G IJ
− GIK GJL gKL . (5.141)

For a material that is transversely isotropic relative to the local Z 3 -
direction, Eq. (5.102) gives
W = W (I1 , I2 , I3 , I4 , I5 ). (5.142)
Equations (5.121) provide the strain invariants I1 , I2 , and I3 , while (5.103)
gives
I4 = 33
2 2
I5 = 31 + 23 , (5.143)
with the IJ being Cartesian strain components in local material coordi-
nates. Equation (5.88) yields
1 ∂W ∂W
S IJ = + (5.144)
2 ∂ IJ ∂ JI
for the corresponding Cartesian components of the second Piola-Kirchhoff
stress tensor.
Differentiating Eq. (5.142) yields
∂W ∂W ∂I1 ∂W ∂I2 ∂W ∂I3
= + +
∂ IJ ∂I1 ∂ IJ ∂I2 ∂ IJ ∂I3 ∂ IJ
∂W ∂I4 ∂W ∂I5
+ + . (5.145)
∂I4 ∂ IJ ∂I5 ∂ IJ
The first three terms of this expression are equivalent to the right-hand-side
of (5.123) and, therefore, lead to the stresses of Eq. (5.140). Thus, we only
need to deal here with the last two terms, with Eq. (5.143) giving
∂I4 ∂ 33
= = δ I3 δ J3
∂ IJ ∂ IJ
∂I5 ∂ 31 ∂ 23
= 2 31 + 23
∂ IJ ∂ IJ ∂ IJ
I J I J
= 2( 31 δ 3 δ 1 + 23 δ 2 δ 3 ). (5.146)
Now, we are in position to compute the stress components relative to a

general curvilinear coordinate system. First, substituting Eq. (5.144) into
(5.118)2 and noting (4.34) yields
∗
J∗ 1 ∂W ∂W
σI = J −1 sIJ = + AIK AJL . (5.147)
2J ∂ KL ∂ LK
Next, using Eqs. (5.140), (5.145), and (5.146) transforms this relation into
∗
J∗
σI = ΦGIJ + ΨH IJ − pgIJ + (2J)−1 [W4 (δ K L L K
3 δ3 + δ3 δ3 )
K L K L L K L K I J
+2W5 ( 31 δ 3 δ 1 + 23 δ 2 δ 3 + 31 δ 3 δ 1 + 23 δ 2 δ 3 )]AK AL
= ΦGIJ + ΨH IJ − pgIJ + (2J)−1 [2W4 AI3 AJ3

I J I J I J I J
+2W5 ( 31 A3 A1 + 23 A2 A3 + 31 A1 A3 + 23 A3 A2 )]
with W4 and W5 defined by Eq. (5.132). Thus, we can write

∗
J∗
σI = ΦGIJ + ΨH IJ − pgIJ + ΘM IJ + ΛN IJ , (5.148)
where Φ, Ψ, p, and H IJ are defined by Eqs. (5.141) and

−1/2
Θ = I3 W4
−1/2
Λ= I3 W5
IJ I J
M = A3 A3
2

IJ
N = (AI3 AJα + AIα AJ3 ) α3 . (5.149)
α=1
In direct form, Eq. (5.136) gives the first three terms of (5.148). To deal
with the other terms, we note that Eq. (5.119)2 gives
AI3 = e3 · GI . (5.150)
With this equation, it is straightforward to show that (5.148) can be written
σ = α0 I + α1 B + α2 B2 + ΘM + ΛN, (5.151)
where
M = F · (e3 e3 ) · FT
2

N= F · (e3 eα + eα e3 ) · FT α3 . (5.152)
α=1
∗ ∗ ∗ ∗ ∗ ∗
If we take σ = σ I J gI gJ , M = M I J gI gJ , and N = N I J gI gJ , then
Eqs. (5.149) give the components of M and N (see Problem 5.7).
We again emphasize that these constitutive relations are based on the
requirement that the fibers (i.e., axis of symmetry) are aligned in the Z 3 -
direction. Otherwise, the equations would need to be modified accordingly.
The analysis for an orthotropic material is similar to that used for a trans-
versely isotropic material. The strain-energy density function
2 2 2
W = W( 11 , 22 , 33 , 12 , 23 , 31 , I3 ) (5.153)
of Eq. (5.94) is substituted into (5.147). With the chain rule, the necessary
derivatives are
∂W ∂W ∂ 11 ∂W ∂ 22 ∂W ∂ 33
= + +
∂ KL ∂ 11 ∂ KL ∂ 22 ∂ KL ∂ 33 ∂ KL
∂W ∂ 212 ∂W ∂ 223 ∂W ∂ 231 ∂W ∂I3
+ + + + , (5.154)
∂ 212 ∂ KL ∂ 223 ∂ KL ∂ 231 ∂ KL ∂I3 ∂ KL
where
∂ 11
= δK L
1 δ1
∂ KL
∂ 212 ∂ 12 K L
= 2 12 =2 12 δ 1 δ 2 (5.155)
∂ KL ∂ KL
and so on, with the I3 term leading to the last term in Eq. (5.140). Substi-
tuting these and related expressions into (5.154) and the result into (5.147)
yields

I∗J∗ 1 ∂W K L ∂W K L L K

σ = δ δ + · · · + 12 2 δ 1 δ 2 + δ 1 δ 2 + · · · AIK AJL −pgIJ ,
J ∂ 11 1 1 ∂ 12
which simplifies to

∗
J∗ −1/2 ∂W I J ∂W I J ∂W I J
σI = I3 A A + A A + A A
∂ 11 1 1 ∂ 22 2 2 ∂ 33 3 3
∂W I J
+ 12 2 A1 A2 + AI2 AJ1
∂ 12
∂W I J
+ 23 2 A2 A3 + AI3 AJ2
∂ 23

∂W I J
+ 31 2 A3 A1 + A1 A3 − pgIJ .
I J
(5.156)
∂ 31
5.4.4 Global Curvilinear Material Symmetries

Thus far, the strain-energy density function has been expressed in terms of
strain components referred to Cartesian axes (Z I ) that align locally with
the principal material directions, which may vary from point to point in an
undeformed body. Sometimes, however, certain symmetries make it possi-

ble to introduce a single system of orthogonal curvilinear coordinates (X I )
that are tangent to the principal material directions throughout the body
(Fig. 5.4). One example is an orthotropic artery subjected to internal pres-
sure, for which cylindrical polar coordinates are useful. In such problems,
it is convenient at the outset to express W as a function of strains relative
to the global curvilinear coordinate system. This matter is examined here.
Within an infinitesimal neighborhood of a point in the undeformed con-

figuration, the Z I and X I axes are approximately the same (Fig. 5.4).
According to the principle of local action, it would seem that we can ob-
tain the appropriate form for W simply by replacing the Cartesian strains
by their corresponding curvilinear strain components. However, since the
Cartesian strains IJ are dimensionless, they must be replaced by the di-

mensionless physical strain components ÊI J relative to the X I system.
Thus, Eq. (5.89)1 becomes
W = W (Ê11 , Ê22 , Ê33 , Ê12 , Ê23 , Ê31 ), (5.157)
where (3.39) gives
EI J
ÊI J = (5.158)
G(I I ) G(J J )
with G(I J ) = GI ·GJ and the EI J defined by Eq. (5.117)1.
With this background, it is straightforward to obtain a general consti-
tutive relation for hyperelastic materials in terms of physical components
of stress and strain. In Sec. 4.4, physical stress components are defined
based on Cauchy (true) stress. However, although they have no direct
physical meaning, physical components also can be defined for the pseu-
dostress tensors using variations of Eqs. (2.132) and (2.134). For orthogonal
curvilinear coordinates, we define the physical components of the second
Piola-Kirchhoff stress tensor as

ŝI J = sI J G(I I ) G(J J ) . (5.159)
Then, in terms of the mutually orthogonal unit vectors
GI
eI = eI = ,
G(I I )
which correspond to principal material directions, Eqs. (3.42) and (2.135)

give

E = ÊI J eI eJ

J
s = ŝI eI eJ . (5.160)
Inserting these expressions into (5.78)3 yields

J ∂W
ŝI = (5.161)
∂ ÊI J
or

J 1 ∂W ∂W
ŝI = + . (5.162)
2 ∂ ÊI J ∂ ÊJ I
With s given by (5.160)2, constitutive relations for other stress components

can be found in the usual way.
Next, we consider the constitutive relations for specific material symme-
tries discussed earlier in this section. For isotropic material, no modifica-
tion is necessary. As discussed in Sec. 2.7.2, the invariants I1 , I2 , and I3 do
not change when expressed in terms of physical components, i.e., I1 = I1 ,
I2 = I2 , and I3 = I3 , with prime denoting invariants expressed in terms

of strains relative to the curvilinear material axes X I . Thus, Eqs. (5.136)
and (5.137) remain valid.
The relations for orthotropy and transverse isotropy require some re-
vision, however. While the derivations follow the same basic path, care
must be taken to properly include the components of the metric tensor
that enter through the definitions of physical components. First, combin-
ing Eqs. (5.84)5 and (5.118)2 gives

sIJ = sK L AIK AJL
∂W
= AI AJ ,
∂EK L K L
and substituting (5.158) yields
∂W AIK AJL
sIJ = .
∂ ÊK L G(K K ) G(L L )
This relation also can be written in the form
∗ ∗ 1 ∂W ∂W
σ I J = J −1 sIJ = + ÂIK ÂJL , (5.163)
2J ∂ ÊK L ∂ ÊL K
where Eq. (5.119)3 gives

AI 1 ∂X I
ÂIJ ≡ J = . (5.164)
G(J J ) G(J J ) ∂X J
Comparing Eqs. (5.147) and (5.163) reveals that the constitutive relations
for global curvilinear orthotropy can be obtained from those written in
Cartesian material coordinates by making the following replacements:
IJ → ÊI J and AIJ → ÂIJ . (5.165)

For illustration, we now examine how these observations affect
Eq. (5.148) for transversely isotropic materials. As mentioned above, the
first three stain invariants are not affected if Cartesian strains are simply re-
placed by the corresponding tensor components for curvilinear coordinates.
The same is not necessarily true for I4 and I5 defined in (5.143), e.g., the
units of E31 and E23 could differ. Thus, we need to replace Cartesian
strains by physical strains in I4 and I5 , and Eq. (5.142) takes the form
W = W (I1 , I2 , I3 , I4 , I5 ), (5.166)
where
I4 = Ê3 3
I5 = Ê32 1 + Ê22 3 . (5.167)
With (5.165), therefore, Eqs. (5.148) and (5.149) become
∗
J∗
σI = ΦGIJ + ΨH IJ − pgIJ + Θ M IJ + Λ N IJ , (5.168)
where
−1/2 ∂W
Θ = I3
∂I4
−1/2 ∂W
Λ = I3
∂I5
M IJ = ÂI3 ÂJ3

2
N IJ = (ÂI3 ÂJα + ÂIα ÂJ3 )Êα 3 . (5.169)
α=1
Note again that these equations would need to be modified for fibers ori-

ented in any direction other than X 3 . For orthotropic materials, the revi-
sions required for Eq. (5.156) are similar.
5.5. Incompressibility 219
5.5 Incompressibility
As discussed earlier, soft tissues usually are assumed to be incompressible,

with deformation constrained by the condition J = 1. Thus, the strain
components are not all independent, and the constitutive relations must be
modified. To determine the required revision, we first derive an alternate
form for the incompressibility condition.
As a body deforms, a surface element da sweeps out a volume v · n da
per unit time, where v is the velocity of the element and n is the unit
normal to da. Thus, the time rate of change of the volume v of the body is

dv
= v · n da,
dt a
which becomes

dv ¯ · v dv
= ∇ (5.170)
dt v
with the divergence theorem (2.200). For an infinitesimal volume element

δv, the dilatation rate is defined as

1 d δv 1 ¯ · v dv
Δ = lim = lim ∇
δv→0 δv dt δv→0 δv δv
or
¯ · v.
Δ=∇ (5.171)
In an incompressible body, Δ must vanish at each point for all time. More-
over, applying the definition of the trace (2.71) to Eq. (3.212)1 yields
1 ¯ ¯ · v]
tr D = 2
[(∇ · v)T + ∇
= ¯ · v.
∇ (5.172)
Thus, the incompressibility condition can be written in the form
¯ · v = tr D = 0.
Δ=∇ (5.173)
Consider now the stress power. If we define a modified Cauchy stress

tensor by
σ ∗ = σ − pI, (5.174)
where p is an arbitrary function of position analogous to a hydrostatic
pressure, then substitution into Eq. (5.22) yields the modified stress power
σ ∗ : D = (σ−pI) : D. (5.175)
Because I : D = tr D = 0 (see Table 2.4, page 32), this equation becomes

σ ∗ : D = σ : D. (5.176)
In other words, hydrostatic pressure does no net work on an incompressible
material. The physical reason for this result is that the work done by a pres-
sure load is due to the pressure acting through a change in volume. Since
an incompressible material undergoes no volume change, the hydrostatic
pressure imparts no net energy.
Consequently, for a given strain field in an incompressible material, the
constitutive relations determine the state of stress only up to an arbitrary
function, −p I. Therefore, Eq. (5.76)1 must be modified to read
∂W
σ = J −1 F · · FT − p I, (5.177)
∂E
which, with J = 1, provides the general constitutive relation for an in-

compressible hyperelastic material. Inserting this equation into Eqs. (4.29)
gives the Piola-Kirchhoff stress tensors
∂W
t = JF−1 · σ = · FT − Jp F−1
∂E
∂W
s = JF−1 · σ · F−T = − Jp F−1 · F−T . (5.178)
∂E
For incompressible material, we set J = 1 in these equations. On the

other hand, setting p = 0 (and J = 1) yields the constitutive relations for
compressible materials.
It is important to note that, although it may be useful to the think of
the function p as being in some ways analogous to a hydrostatic pressure,
this terminology often has been a source of confusion in the biomechanics
literature. Several authors have associated p with a “tissue pressure” that
can be measured with an appropriate transducer. In fact, however, p gener-
ally does not have a direct physical interpretation. Rather, it is a Lagrange
multiplier that is needed to enforce the incompressibility constraint.10
Component forms of the constitutive relations for a compressible mate-
rial were derived in Sec. 5.2.6. To obtain the corresponding equations for
an incompressible material, we need only examine the terms involving the
10 To better understand this point, try computing p for an unloaded rectangular block
composed of neo-Hookean material with W = C(λ21 + λ22 + λ23 − 3). Is the “hydrostatic
pressure” zero, as may be expected for this case?
5.5. Incompressibility 221
Lagrange multiplier. The stress components of Eqs. (5.84) are defined by

the dyadics
∗
J∗
Ij
t = t G I gj = t·j GI gj
I
s = sIJ GI GJ. (5.179)

Thus, the terms involving p in Eqs. (5.177) and (5.178) require the relations
gI · I · gJ = gI · gJ = gIJ
GI · F−1 · gj
K j
= GI · (Fk GK gk ) · g
I kj
= FK
k δK g = F Ij
GI · F−1 · gj
K
= GI · (Fk GK gk ) · gj
I k
= FK I
k δ K δ j = Fj
−1
· F−T ) · GJ
L J
GI · (F = GI · (GK gK ) · (g GL) · G
= δ IK gKL δ JL = gIJ ,
in which Eq. (3.22)2 has been used. Using these results in (5.177) and
(5.178) and noting Eqs. (5.84) yields
∂W
σ ij = J −1 FIi FJj − pgij
∂EIJ
∗ ∗ ∂W
σ I J = J −1 sIJ = J −1 − pgIJ
∂EIJ
∂W
tIj = FJj − JpF Ij
∂EIJ
∂W
tI·j = − JpFjI . (5.180)
∂FIj
Again, we recover the constitutive relations for both incompressible and

compressible materials by setting J = 1 or p = 0, respectively.
We now make some observations. First, consistent with Eq. (4.34),
I∗J∗
σ = sIJ for an incompressible material. Second, comparing the above
∗ ∗
expression for σ I J with Eqs. (5.140) and (5.141) reveals that the p term
due to incompressibility replaces the ∂W/∂I3 term in the constitutive rela-
tions for a compressible material. Thus, with p defined as the Lagrange
multiplier, Eqs. (5.140), (5.148), and (5.156) apply also to incompress-
ible materials with isotropic, transversely isotropic, and orthotropic sym-
metry, respectively. Third, since I3 = 1, incompressibility renders the
p = −2 ∂W/∂I3 term indeterminate from constitutive behavior alone. In

general, p is a function of position to be determined from the equilibrium
equations and boundary conditions (see Chap. 6).
Finally, we consider the modified strain-energy density function
1/2
W ∗ = W − p(J − 1) = W − p(I3 − 1). (5.181)
Since J = 1, the added multiplier term does not alter the strain energy
stored in an incompressible material. With this expression, we can use
Eq. (5.78)2 to compute a first Piola-Kirchhoff stress tensor
∂W ∗ ∂W ∂J
t∗ = T
= T
−p . (5.182)
∂F ∂F ∂FT
The last term of this equation can be transformed by noting that, because
J = det F, the last formula in Table 2.5 (page 33) gives
∂J ∂ det F
= = (det F)F−T = JF−T ,
∂F ∂F
which, with J = 1, yields
∂J
= F−1 . (5.183)
∂FT
Accordingly, Eq. (5.182) becomes
∂W
t∗ = − pF−1 ,
∂FT
or, by (5.78)2 ,
∂W ∂W
t∗ = T
− pF−1 = · FT − pF−1 .
∂F ∂E
This result agrees with that for the first Piola-Kirchhoff stress tensor of
Eq. (5.178)1 with J = 1.
Thus, the constitutive relations for an incompressible material also can
be obtained by setting J = 1 and replacing W by W ∗ in Eqs. (5.78) to get
∂W ∗
σ = F· · FT
∂E
∂W ∗ ∂W ∗
t= · FT =
∂E ∂FT
∂W ∗
s= (5.184)
∂E
with W ∗ defined by (5.181).

5.6. Linear Elastic Material 223
5.6 Linear Elastic Material
For a linear material, the strain-energy density function is a quadratic func-

tion of the strain components. As shown in Appendix A (Sec. A.4.3), char-
acterizing the mechanical behavior of a general linear anisotropic material
requires 21 independent elastic constants. In the following, the results of
Sec. 5.3 are used to derive the appropriate stress-strain relations for iso-
tropic, transversely isotropic, and orthotropic materials. For simplicity,
only Cartesian coordinates are considered.

For an isotropic material, W depends only on the strain invariants I1 , I2 ,
and I3 of Eqs. (3.142). However, because I3 = det[δ IJ + 2EJI ] is a cubic
function of the strains, it can be deleted from the list of arguments for a
linear material. For the subsequent analysis, it is convenient to replace I1
and I2 by the alternate strain invariants
J1 = 21 (I1 − 3)
= 11 + 22 + 33
1
J2 = 4 (I2 − 2I1 + 3)
= 11 22 + 22 33 + 33 11 − 212 − 223 − 231 , (5.185)
which are linear and quadratic functions of the strain components, respec-
tively.
In terms of these alternate strain invariants, the quadratic strain-energy
density function has the form
W = C1 J12 + C2 J2 , (5.186)
where C1 and C2 are material constants. Since all stress and strain ten-
sors are essentially the same in the linear theory, Eq. (5.88) provides the
constitutive relation
1 ∂W ∂W
σ ij = + , (5.187)
2 ∂ ij ∂ ji
which gives
∂W
σ 11 = = 2C1 ( 11 + 22 + 33 ) + C2 ( 22 + 33 )
∂ 11
..
.
1 ∂W ∂W
σ 12 = + = −C2 12
2 ∂ 12 ∂ 21
..
.
On setting
1
C1 = 2
(λ + 2μ)
C2 = −2μ,
where λ and μ are the Lamé constants, the above stress-strain relations
become
σ ij = λΔδij + 2μ ij (5.188)
or
σ = λΔI + 2μE, (5.189)
where
Δ = ii = tr E (5.190)
i j
is the dilatation, and E = ij e e is the strain tensor. Equations (5.188)
and (5.189) are Hooke’s law for a linear isotropic material [see Eq. (A.146)].
The two material constants λ and μ completely characterize the mechanical
behavior of a linear isotropic elastic material.
Note that Eq. (5.189) is valid for large strains if the material behavior
remains linear. In general, however, materials behave nonlinearly when
deformation becomes large. Thus, Eq. (5.189) usually is restricted to small
strains.

According to Eq. (5.102), the strain-energy density function for transversely
isotropic material depends on the quantities I4 and I5 defined by (5.103), in
addition to the strain invariants I1 , I2 , and I3 (or J1 , J2 , and I3 ). Modifying
Eq. (5.186) to include all quadratic combinations of the strain components
gives
W = C1 J12 + C2 J2 + C3 J1 I4 + C4 I42 + C5 I5 . (5.191)
Thus, a linear transversely isotropic material is characterized by the five
elastic constants Ci , with Eq. (5.187) providing the stress-strain relations.

Equation (5.94) gives the strain dependencies of W for an orthotropic ma-
terial. Writing all quadratic combinations of these terms yields
W = C1 211 + C2 222 + C3 233
+C4 11 22 + C5 22 33 + C6 33 11
+C7 212 + C8 223 + C9 231 , (5.192)

and Eq. (5.187) again provides the stress-strain relations. The above equa-
tion indicates that defining the mechanical behavior of a linear orthotropic
material requires nine independent elastic constants.
5.7. Determining W for Soft Tissues 225
5.7 Determining W for Soft Tissues
Determining constitutive relations for soft biological tissues has been and,
in some cases, remains a subject of considerable ongoing research. This
section discusses some of the issues involved in this process.
5.7.1 Microstructural and Phenomenological Approaches

In general, two approaches are used to model the mechanical behavior of
soft tissues: microstructural and phenomenological (Humphrey, 2002). In
the microstructural approach, the geometric and mechanical properties
of individual tissue components, e.g., collagen and elastin, are measured di-
rectly. Then, with assumptions concerning the interactions between these
components, a macroscopic constitutive relation for the composite tissue is
derived from first principles. This approach has several advantages, includ-
ing providing insight into the mechanisms of tissue behavior and the ability
to describe remodeling due to microstructural changes under changing load-
ing conditions (Humphrey, 1999; Humphrey and Rajagopal, 2002, 2003).
On the other hand, attempts to adequately describe such complex mate-
rials have met with relatively limited success, as judged by the predicted
response in comparison to experimental measurements.
In the phenomenological approach, macroscopic constitutive rela-
tions are determined directly by fitting computed stress-strain curves to
measured tissue-level data. As discussed below, this process usually involves
postulating a functional form for W and then determining the material con-
stants by minimizing the difference between computed and measured data.
This approach is relatively simple mathematically, but the global nature
of the method may mask underlying microstructural mechanisms. Often,
however, microstructure is considered in postulating the functional form of
the constitutive relation. For example, tissue may be assumed to be trans-
versely isotropic or orthotropic, depending on the observed arrangement of
constituent fibers.
Due to its relative simplicity and consistency within a continuum frame-
work, the remainder of this section focuses on the phenomenological ap-
proach.
5.7.2 Experimental Considerations

For a linear elastic material, depending on the specific material symme-
try, the general form of W is known a priori, i.e., W must be a quadratic
function of the strains. For a linear isotropic material, the two material con-
stants can be determined from uniaxial testing alone. On the other hand,
determining constants for linear anisotropic materials, e.g., bone, generally
requires various sets of experiments, such as tensile tests on samples cut
from different directions relative to the material axes. These experiments,
however, are still relatively straightforward.
In contrast, the functional form of W is generally not known a pri-
ori for nonlinear elastic materials, and determining mechanical properties
requires data from various loading protocols. Unless the structure is one-
dimensional, e.g., an actin microfilament, uniaxial testing alone is not ade-
quate. At the very least, biaxial testing is required (Humphrey, 2002; Sacks
and Sun, 2003).
In most studies of soft tissue, the functional form of W is postulated,
guided by microstructural considerations and the principles discussed ear-
lier in this chapter. (Some popular choices for W are given in Sec. 5.3.)
Then, for the postulated W , the boundary value problem is solved for an ex-
perimental loading protocol, and fitting theoretical results to experimental
data provides values for the unknown material coefficients.
A disadvantage of this approach is that it is difficult to know whether
the chosen form for W is correct even if it appears to fit the data well. One
reason for this is that the range of strain combinations that can be covered
experimentally is limited, and some physiologically significant deformations
may occur outside this realm for which the postulated W is not appropriate.
In biomechanics applications, however, the choice for W may not need
to be precise if it predicts approximately the correct behavior within the
deformation range of interest. Another approach is to use experimental data
to determine the functional form of W , as well as the material parameters
(Humphrey, 2002; Sacks and Sun, 2003).
5.7.3 Some Types of Experiments
Due to the complexity of the problem, most material testing of soft tis-
sues has involved relatively simple geometry and loading protocols. To
enable precise control of loading conditions, these tests often are conducted
on specimens that are cut to convenient geometries. Experiments on my-
ocardium, for example, have used thin rectangular samples cut from the
heart (Demer and Yin, 1983; Humphrey and Yin, 1987; Yin et al., 1987;
Humphrey et al., 1990a; 1990b; Humphrey, 2002; Sacks and Sun, 2003). To
facilitate biaxial stretching in the fiber and cross-fiber directions, the edges
of the specimen are cut approximately parallel to and normal to the local
5.8. Problems 227
muscle fiber direction.

In addition to experimental convenience, there is another reason that
simple geometries often are preferred. Determining material coefficients
generally involves an optimization procedure to find the best fit of theo-
retical results to experimental data. Hence, numerous iterations may be
required to find convergence, and the availability of an analytic solution
can greatly speed the process.
One disadvantage of using dissected specimens for experimentation is
that the act of cutting introduces unphysiological boundary conditions that
may alter tissue microstructure. Preconditioning by repeated loading and
unloading, however, likely restores the structure in part, and parameters
such as temperature and pH can be controlled. But some factors that may
influence tissue behavior in vivo are difficult to simulate.
For this reason, considerable effort has been devoted to developing meth-
ods for determining material properties in vivo. For myocardium, material
coefficients for a postulated form for W have been determined by fitting re-
gional wall strains measured by tracking physical markers to those predicted
by a computational model for the heart (Guccione et al., 1991; Omens et
al., 1993). Such studies provide important understanding of cardiac me-
chanics. Clinical use, however, requires a noninvasive approach to quantify
deformation, such as by tracking artificial markers produced by magnetic
resonance imaging (MRI) with tissue tagging (Axel et al., 2005). Stud-
ies that combine this method with 3-D finite-element models have yielded
promising results in both animals and humans (Li et al., 2009; Wang et al.,
2009).
Nevertheless, simple problems remain useful. They provide insight into
the fundamental behavior of tissues undergoing large deformation, and an-
alytical solutions for benchmark problems always will be needed to test the
accuracy of computer codes. Hence, the next chapter presents solutions for
several problems that are commonly used in experiments.
5.8 Problems
5.1 Consider a differential element for a bar undergoing uniaxial deforma-

tion. In the deformed configuration, the element has length dx and
cross-sectional area da, and is subjected to the following thermome-
chanical loads: Cauchy stress σ(x, t); body force f(x, t) per unit vol-
ume; heat flux q(x, t) per unit area; and heat production rate r(x, t) per
unit mass. Let ρ(x) be the mass density and v(x, t) the axial velocity.
In general, the stress, heat flux, and velocity differ slightly between the
ends of the element (Fig. 5.5).
da wq
q+ dx
q wx
V + wV dx
f
V r
wx
dx
v v + wv dx
wx
Fig. 5.5. Differential element for a bar with thermomechanical loads (Problem 5.1).
(a) By summing forces on the element, derive the 1-D differential equa-
tion of motion.
(b) Using conservation of energy, derive the 1-D form of Eq. (5.18)1 .
5.2 Consider a stress tensor τ defined by
τ = 12 (s · U + U · s),
where s is the second Piola-Kirchhoff stress tensor, and U is the right
stretch tensor. Show that the stress power can be written in the form
Ps = J −1 τ : U̇.
5.3 Determine the transformation relations for the following quantities in
frame A as viewed by an observer in frame A∗ :
(a) The strain tensors E and e as defined by Eqs. (3.28).
(b) The stress tensors σ, t, and s, as defined by Eqs. (4.23) and (4.29).
(c) The gradient operator ∇ ¯ in the deformed configuration.
Hint: Consider the relations
∂r ¯ = ∂
dr = dr∗ · ∗ , ∇
∂r ∂r
as given by Eqs. (2.163)1 and (3.18)2 .
5.4 A bar is in a state of uniaxial stress σ = σ 0 exex relative to Cartesian
coordinates (x, y, z) fixed in reference frame A. Another Cartesian
system (x̄, ȳ, z̄), fixed in reference frame A∗ , initially coincides with the
system in A. Suppose the bar rotates through an angle θ in the xy-
plane while maintaining the same uniaxial state of stress (Fig. 5.6). If
frame A∗ rotates with the bar, determine the stress dyadic σ ∗ observed
in A∗ in terms of the Cartesian base vectors ēi fixed in frame A∗ .
5.8. Problems 229
y
y
T x
x
Fig. 5.6. Rotation of a bar under constant uniaxial stress (Problem 5.4).
5.5 For α = 1, Eq. (5.114) becomes

μ 1 −β
W = I1 − 3 + I −1 , (5.193)
2 β 3
where I1 and I3 are given by (3.142) and
ν
β= .
1 − 2ν
(a) For small deformation (|Eij | << 1) and plane strain (E13 = E23 =
E33 = 0), write W as a quadratic function of the Cartesian strain
components Eij .
(b) Use the result from part (a) to determine the stresses σ 11 , σ 22 , σ 12 ,
σ 21 in Cartesian coordinates. Show that these stress components
agree with those given by Hooke’s law (5.189). Note: The Lamé
constant λ is related to the shear modulus μ and Poisson’s ratio ν
by
2μν
λ= .
1 − 2ν
5.6 Show that the expressions for I5 given by Eqs. (5.103) and (5.104) are
equivalent.
5.7 Consider the tensors M and N defined by Eqs. (5.152). Show that
∗ ∗ ∗ ∗
the components M I J and N I J of these tensors are given by the
right-hand sides of Eqs. (5.149)3,4.
5.8 Suppose an incompressible material is characterized by the Mooney-
Rivlin strain-energy density function
W = b1 (I1 − 3) + b2 (I2 − 3),
where I1 and I2 are stain invariants, and b1 and b2 are constants.
(a) Show that Eq. (5.178)2 can be written as

∂W
s=2 − p C−1 .
∂C
(b) Using the given form for W and the equation from part (a), derive
the constitutive relation for a Mooney-Rivlin material in terms of
the Cartesian components of s and C.
5.9 Consider a general strain-energy density function W (Λ1 , Λ2 , Λ3 ) for a
compressible isotropic material, where the Λi are principal stretch ra-
tios. Show that the constitutive relations for the stress tensors can be
written in the spectral forms
3 3 3
1 ∂W ∂W ∂W
s= Ni Ni , t = Ni ni , σ = J −1 Λi ni ni ,
Λi ∂Λi ∂Λi ∂Λi
i=1 i=1 i=1
where the unit vectors Ni and ni are principal directions relative to
the undeformed and deformed configuration, respectively. Hint: See
Eqs. (3.135).
5.10 A compressible material is characterized by the specialized Blatz-Ko
strain-energy density function
1/2
W = (μ/2)(I2 /I3 + 2I3 − 5).
Write W in terms of principal stretch ratios Λi (i = 1, 2, 3). Then, find
the constitutive relations for the principal stresses σ 2 , t2 , and s2 . Note:
For an isotropic material, the principal directions of stress and strain
coincide.
5.11 For the Blatz-Ko material of Problem 5.10, show that the constitutive
relation for the Cauchy stress tensor can be written in the form

−1/2 −1
σ = μ I − I3 B ,
where B is the left Cauchy-Green deformation tensor. Hint: Work with
invariants of C; the Cayley-Hamilton theorem (2.84) could be useful.
5.12 Consider a passive muscle composed of compressible, transversely iso-
tropic material consisting of aligned fibers embedded in an isotropic
matrix. For fibers that are aligned in the X 1 -direction prior to defor-
mation, the strain-energy density function has the form
1/2
W = a(I1 − 2I3 − 1) + b(I2 − 3)2 + cI42 + dI52 ,
where a = 0.5, b = 0.1, c = 5, and d = 4. Suppose the muscle undergoes
deformation defined by
x1 = 0.2X 1 (1 + 2X 1 + X 2 )
x2 = 0.1X 2 (1 + X 2 )
x3 = 0.8X 3 ,
5.8. Problems 231
where X I and xi are material and spatial Cartesian coordinates, respec-

tively. For the point originally located at (1,2,1), compute the Cauchy
∗ ∗
stress components σ I J using (a) Eq. (5.148) and (b) Eq. (5.151). For
each case, write the equations needed for the computation, including
any modifications needed for the fiber direction being X 1 , rather than
X 3 as assumed in the text.
5.13 A simple model for the left ventricle consists of a thick-walled circular
tube composed of layers of incompressible, orthotropic myocardium.
At an arbitrary point in the wall, the principal material directions of
the undeformed myocardium are defined by the Cartesian unit vectors
e1 = ēR
e2 = ēΘ cos β + ēZ sin β
e3 = −ēΘ sin β + ēZ cos β,
where the ēI are unit base vectors in global cylindrical coordinates
(R, Θ, Z), and β is the muscle-fiber angle relative to the circumferential
direction (see Fig. 6.16b, page 278 in next chapter).
In terms of Cartesian strain components relative to the local material
directions, the strain-energy density function is given by
W = c(eQ − 1)
2 2 2
Q = a1 11 + a2 22 + a3 33 + 2(a4 11 22 + a5 22 33 + a6 33 11
+a7 212 + a8 223 + a9 231 ),

where the ai are constants. Ignoring the Lagrange multiplier and taking
the second Piola-Kirchhoff stress tensor in the forms s = sIJ eI eJ =
s̄IJ ēI ēJ , determine the stress components sIJ in terms of the strains
IJ
IJ and s̄ in terms of sIJ . This is a convenient way to determine
constitutive relations for anisotropic materials in global coordinates.
5.14 Consider in-plane deformation of a flat plate composed of incompress-
ible, transversely isotropic material with aligned fibers. Let (x, y, z) be
Cartesian “plate” coordinates with axes parallel to the edges of the un-
loaded plate. In the undeformed configuration, the fibers are oriented
parallel to the middle surface (i.e., the xy-plane, halfway between the
upper and lower plate surfaces) at the angle β relative to the x-axis
(Fig. 5.7). The strain-energy density function is
W = c[I1 − 3 + α(Λ2f − 1)4 ],
where I1 is the first strain invariant defined by Eq. (3.142)1 , Λf is
the Lagrangian stretch ratio in the fiber direction, and c and α are
constants.
(a) For deformation without transverse shear (Exz = Eyz = 0), write
W as a function of the Cartesian strain components Eij relative to
the plate coordinates.
(b) Suppose the plate is in a state of plane stress (σ zz = 0) and con-
strained to undergo biaxial stretch (specified Exx and Eyy ) with-
out shear (Exy = 0). Determine the constitutive relations for the
Cauchy stress components σ xx, σ yy , and σ xy in plate coordinates.
Hint: Use incompressibility to determine Ezz and the condition
σ zz = 0 to find the Lagrange multiplier p.
(c) For c = 1 kPa, α = 0.1, β = 60◦ , and Eyy = 0, plot the stress
components from part (b) as functions of the stretch ratio Λx for
0.5 ≤ Λx ≤ 3. Do the principal axes of stress and strain coincide?
Fibers
y x-
y- E
x
Fig. 5.7. Transversely isotropic plate (Problem 5.14).

Chapter 6
Biomechanics Applications
Finite element and other computational methods now make it possible to

solve virtually any problem in biomechanics, no matter the complexity of
the geometry, deformation, and material properties. Nevertheless, “exact
solutions” remain useful for several reasons. First, they provide insight into
the fundamental nonlinear behavior of soft tissues. Second, they provide
benchmarks for checking the accuracy of numerical solutions. Third, as dis-
cussed in the previous chapter, exact solutions can be used in combination
with experiments to determine material properties. Such experiments often
involve relatively simple geometries and deformations, making the use of
exact solutions convenient.
This chapter considers a number of fundamental problems in soft tissue
biomechanics. Besides integrating the material presented in previous chap-
ters, we use these problems to illustrate the effects of compressibility and
anisotropy. Although soft tissues usually are assumed to be incompress-
ible, extracellular fluid flow can produce significant compressibility effects,
and microstructural considerations suggest that many soft tissues can be
treated as orthotropic or transversely isotropic materials. Realistic models
must take these features into account. For simplicity, gravity and other
body forces, as well as inertia effects, are ignored in all problems of this
chapter (with the exception of Problem 6.9).
6.1 Boundary Value Problems
The governing equations, along with appropriate boundary and initial con-
ditions, define a boundary value problem in the theory of elasticity. As
derived in the previous chapters, the tensor equations of nonlinear elastic-
ity consist of the following:
233
234 Biomechanics Applications
Kinematic Relations:
T
F = (∇r)
E = 12 (FT · F − I)

= 12 ∇u + (∇u) + (∇u) · (∇u)T
T
(6.1)
Stresses:
σ = J −1 F · t = J −1 F · s · FT (6.2)
Equations of Motion:
¯ · σ + f = ρa
∇
∇ · t + f 0 = ρ0 a
∇ · (s · FT )+f 0 = ρ0 a (6.3)
Constitutive Relations:
∂W
σ = J −1 F · · FT − p I
∂E
∂W
t= · FT − JpF−1
∂E
∂W
s= − JpF−1 · F−T (6.4)
∂E
Compressible material: p = 0; incompressible material: J = 1.
Incompressibility:
J = det F = 1 (6.5)
The boundary conditions include specified tractions and displacements

over the surface of the body. In dynamic problems, initial conditions also
must be stipulated. These consist of the specified displacement and velocity
of all points in the body at t = 0.
In principal coordinates, the governing equations simplify considerably.
These versions are commonly used in relatively simple problems that do
not involve shear stress or shear strain in the chosen global coordinates. In
these cases, tensors can be written in terms of physical components with a
single subscript. With the summation convention suspended, the equations
take the following form:
6.1. Boundary Value Problems 235
Kinematic Relations:
Fi = λ i
Ei = 12 (λ2i − 1) (6.6)
Stresses:
σ i = J −1 λi ti = J −1 λ2i si (6.7)
Depend on the specific coordinate system.
λ2i ∂W λi ∂W
σi = −p = −p
J ∂Ei J ∂λi
∂W J ∂W J
ti = λi − p= − p
∂Ei λi ∂λi λi
∂W J 1 ∂W J
si = − 2p= − 2p (6.8)
∂Ei λi λi ∂λi λi
Compressible material: p = 0; incompressible material: J = 1.
Incompressibility:
J = λ1 λ2 λ3 = 1 (6.9)
In the constitutive relations, the equation
∂W ∂W ∂Ei ∂W
= = λi (i not summed)
∂λi ∂Ei ∂λi ∂Ei
has been used.
The specific equations and variables used to formulate and solve a
boundary value problem depend on the geometry, the types of loads, the
solution method (e.g., numerical or analytical), and personal preference.
Moreover, elasticity problems can be formulated either by specifying the
applied loads and computing the resulting deformation or by specifying the
deformation and computing the loads needed to produce it. The latter
method, called the inverse method, often is more convenient in solving non-
linear problems. Sometimes, however, it is not reasonable to specify the
entire deformation a priori. In such cases, a semi-inverse method may be
useful in which only part of the deformation is specified, with the rest to
be determined as part of the solution procedure.
In linear elasticity, any method gives the same solution, as guaranteed by

the uniqueness theorem. In nonlinear elasticity, however, multiple solutions
are possible, and different approaches may yield different solutions, each
of which may be valid. For example, investigating bifurcations in stability
problems may benefit from combining various solution methods. In general,
problems in this chapter are solved using either inverse or semi-inverse
methods.
6.2 Extension and Compression of Soft Tissue
For simplicity, most experimental studies of the constitutive behavior of

soft tissues involve uniaxial or biaxial tests. Example 3.2 (page 78) ex-
amined the kinematics of a rectangular block of tissue undergoing uniform
extension. Here, we compute the tractions that must be applied to the
block to maintain this deformation. For homogeneous stress and strain,
these tractions need to be uniform over the faces of the block (Fig. 6.1),
and the equilibrium equations (6.3) (with a = 0) are satisfied identically if
there are no body forces.
t22
X2, x2
Fibers
t11 X1, x1
Fig. 6.1. Uniform extension of a transversely isotropic block.
We assume that the tissue is transversely isotropic, with fibers aligned

in the X1 -direction, which also is taken as a loading direction (Fig. 6.1).
Skeletal muscle, heart muscle, ligaments, and tendons often are considered
as transversely isotropic. Due to symmetry, the coordinate axes correspond
to the principal directions of stress and strain. For convenience and since we
are working entirely in principal Cartesian coordinates, tensor components
are written here with a single index, and we do not distinguish between
subscripts and superscripts, nor between lower and upper case indices.
6.2. Extension and Compression of Soft Tissue 237
6.2.1 Governing Equations

Kinematics. Relative to the Cartesian coordinates Xi and xi
(Fig. 6.1), Eqs. (3.46) and (3.49)1 give the deformation gradients and La-
grangian strain components
[Fi ] = diag [λ1 , λ2 , λ3 ]

[Ei ] = diag 12 (λ21 − 1), 12 (λ22 − 1), 12 (λ23 − 1) , (6.10)
where the λi are stretch ratios. Here, we assume the λi are specified vari-
ables.
Constitutive Relations. Since the xi are principal axes, there is no

particular disadvantage in working with the first Piola-Kirchhoff stress ten-
sor, which generally is not symmetric if shear stresses enter the analysis. In
fact, t often is the stress tensor of choice in problems not directly involving
shear. (Of course, shear stresses may occur relative to axes not aligned
with the principal directions.) In this case, the constitutive relation (6.8)2
gives
∂W J
ti = − p (i not summed). (6.11)
∂λi λi
For a compressible material, we set p = 0. In this case, if W (λi ) is known
and all the λi are specified, then the force applied to each face of the block
is given by multiplying ti by the corresponding undeformed area of the face,
thereby completing the solution. For an incompressible block, however, all
the λi cannot be specified independently, since they are constrained by the
incompressibility condition
J = det F = λ1 λ2 λ3 = 1. (6.12)
For example, if λ1 and λ2 are specified, then this relation gives λ3 indepen-
dently of the forces applied in the x3 -direction. Thus, a well-posed problem
for this case requires specifying one of the surface tractions, say t3 , rather
than λ3 . (For the membrane to remain planar, t3 must be applied to both
faces normal to x3 , in opposite directions.) Then, Eq. (6.11) gives (with
J = 1)
∂W
p = λ3 − t3 , (6.13)
∂λ3
which is uniformly valid for homogeneous deformation. The other stress
components now can be determined using (6.11).
6.2.2 Biaxial Stretching of a Membrane

Next, consider the special case of biaxial stretching of a thin homogeneous
membrane in the x1 x2 -plane (Fig. 6.2). Biaxial loading has been used to
determine the material properties of skin, pericardium, and myocardium,
for example. These tissues experience multiaxial loading in vivo. We as-
sume that the in-plane stretch ratios λ1 and λ2 are known, and the surfaces
at X3 = ±H/2 are traction free (H = undeformed membrane thickness).
In the following, we examine separately the problems for membranes com-
posed of incompressible and compressible materials.
X3, x3
X2, x2
t22
Fibers
t11
X1, x1
Fig. 6.2. Biaxial stretching of a transversely isotropic membrane.
Incompressible Membrane. If the membrane is composed of an

incompressible material, the Lagrange multiplier p can be determined from
the boundary conditions. Since the membrane is not loaded in the x3 -
direction, we assume a condition of plane stress (t3 = 0 everywhere), and
Eq. (6.13) gives
∂W
p = λ3 . (6.14)
∂λ3
Substitution into Eq. (6.11) then yields
∂W λ3 ∂W
t1 = −
∂λ1 λ1 ∂λ3
∂W λ3 ∂W
t2 = − . (6.15)
∂λ2 λ2 ∂λ3
If the membrane is transversely isotropic relative to the X1 -axis

(Fig. 6.2), then W is a function of the five strain invariants
I1 = 3 + 2(E1 + E2 + E3 ) = λ21 + λ22 + λ23
I2 = 3 + 4(E1 + E2 + E3 + E1 E2 + E2 E3 + E3 E1 )
= λ21 λ22 + λ22 λ23 + λ23 λ21
I3 = (1 + 2E1 )(1 + 2E2 )(1 + 2E3 ) = λ21 λ22 λ23 = 1
I4 = E1 = 21 (λ21 − 1)
I5 = 0, (6.16)
as provided by Eqs. (3.143), (5.103), and (6.10)2 . Since I3 = 1 (incom-

pressibility) and I5 = 0, Eq. (5.102) reduces to
W = W (I1 , I2 , I4 ). (6.17)
These equations give the derivatives (with Wi ≡ ∂W/∂Ii )

∂W ∂I1 ∂I2 ∂I4
= W1 + W2 + W4
∂λ1 ∂λ1 ∂λ1 ∂λ1
= W1 (2λ1 ) + W2 (2λ1 λ22 + 2λ1 λ23 ) + W4 (λ1 )
= λ1 [2W1 + 2W2 (λ22 + λ23 ) + W4 ]
∂W
= λ2 [2W1 + 2W2 (λ21 + λ23 )]
∂λ2
∂W
= λ3 [2W1 + 2W2 (λ21 + λ22 )].
∂λ3
Inserting these expressions and λ3 = 1/λ1 λ2 into Eqs. (6.15) yields
1
t1 = 2λ1 1 − (W1 + λ22 W2 ) + λ1 W4
λ41 λ22
1
t2 = 2λ2 1− 2 4 (W1 + λ21 W2 ). (6.18)
λ1 λ2
The Cauchy stress components, given by (6.7), are σ i = λi ti (i not
summed).
For illustration, suppose the membrane is composed of a material with
W = C1 (I1 − 3 + βI42 ), (6.19)

E E
O1 O
Fig. 6.3. Stress-stretch curves for equibiaxial stretching of a transversely isotropic, in-
compressible membrane for various values of the fiber modulus (β). The fibers are
oriented in the x1 -direction.
where β correlates with the fiber stiffness. Substitution into Eqs. (6.18)
gives the stresses as functions of λ1 and λ2 . For equibiaxial stretch
(λ1 = λ2 ), t1 is plotted in Fig. 6.3 as a function of the stretch ratio for
various values of β. Since the fibers do not contribute to the stress in the
x2 -direction, the β = 0 curve for t1 also represents the curve for t2 , but for
all values of β. For relatively small strain (λ1 = λ2 < 1.05), all the solutions
nearly coincide. For large strain, however, the curves diverge significantly,
with the curve shape changing from primarily concave downward to con-
cave upward as β increases. The curves for β = 0 and 0.5 are typical for
incompressible rubber, while the curves for larger values of β are similar to
those generated by tests on most soft biological tissues (Fung, 1993).
Results also are shown for an isotropic membrane (β = 0, Fig. 6.4).
Here, the membrane is stretched in the x1 -direction, while the x2 -direction
is held fixed at λ2 = 1, 1.5, and 2. Note that t2 remains relatively constant
as λ1 increases while λ2 is held at the value 2.
Compressible Membrane. If the membrane is composed of com-

pressible material, then the deformation is not constrained by J = 1, and
we set the Lagrange multiplier p = 0. In this case, the plane stress condition
t3 = 0 provides an equation to be solved for λ3 (rather than p). However,
since t3 depends on W , the value of λ3 depends on the specific properties
O
O
O
Fig. 6.4. Stress-stretch curves for biaxial stretching of an isotropic incompressible mem-
brane (β = 0). The stretch ratio λ2 is held fixed at the three values indicated.
of the membrane material.

Consider, for example, a membrane composed of isotropic tissue char-
acterized by the Blatz-Ko strain-energy density function

1 − 2ν ν/(1−2ν)
W = C2 I2 /I3 − 3 + I3 −1
ν
%
−2 −2 −2 1 − 2ν 2ν/(1−2ν)
= C2 λ1 + λ2 + λ3 − 3 + (λ1 λ2 λ3 ) −1 ,
ν
(6.20)
as provided by Eqs. (6.16) and (5.114) with α = 0 and μ = 2C2 . With
p = 0, Eq. (6.11) yields

t1 = 2C2 λ2 λ3 (λ1 λ2 λ3 )−(1−4ν)/(1−2ν) − λ−3
1

t2 = 2C2 λ3 λ1 (λ1 λ2 λ3 )−(1−4ν)/(1−2ν) − λ−3
2

t3 = 2C2 λ1 λ2 (λ1 λ2 λ3 )−(1−4ν)/(1−2ν) − λ−3
3 , (6.21)
and Eq. (6.7) gives the Cauchy stress components σ 1 = t1 /λ2 λ3 , σ 2 =

t2 /λ3 λ1 , and σ 3 = t3 /λ1 λ2 . For t3 = 0, Eq. (6.21)3 gives
λ3 = (λ1 λ2 )−ν/(1−ν) (6.22)

and then the other relations in (6.21) give

1/(1−ν)
t1 = 2C2 λ−1+3ν
1 λ2ν
2 − λ−3
1

−1+3ν 1/(1−ν)
t2 = 2C2 λ2ν 1 λ2 − λ−3
2 . (6.23)
For small strain, the material parameter ν can be identified with Pois-
son’s ratio. For the special case ν = 1/4, the above relations become
⎡ 1
⎤
2 3
λ 1
t1 = 2C2 ⎣ 2 − 3⎦
λ1 λ1
⎡ 1
⎤
λ2 3
1⎦
t2 = 2C2 ⎣ 1 − 3 . (6.24)
λ2 λ2
On the other hand, setting ν = 0.5 and I3 = 1 yields the incompressible

case. In this instance, Eq. (6.22) gives λ3 = (λ1 λ2 )−1 and Eq. (6.20) reduces
to W = C2 (I2 − 3). Substitution into (6.18) then reproduces Eqs. (6.23) if
we set ν = 0.5.
O O
Fig. 6.5. Stress-stretch curves for equibiaxial stretching of an isotropic compressible

membrane for various values of the material constant ν.
It is important to note that for some forms of W , the equation t3 = 0

may have multiple roots. Thus, it is theoretically possible that, for the
same λ1 and λ2 , the membrane can have different thicknesses, although
it is unlikely that all solutions would be stable. The possibility of mul-
6.3. Simple Shear of Soft Tissue 243
tiple solutions always must be kept in mind when dealing with nonlinear
problems.
Equibiaxial loading curves computed from Eqs. (6.23) for various values
of ν show that the effects of compressibility can be quite large (Fig. 6.5).
The membrane stiffness increases markedly with ν, especially as ν ap-
proaches 0.5, corresponding to an incompressible membrane.
6.2.3 Uniaxial Extension of a Bar

Papillary muscles, actin microfilaments, ligaments, and articular cartilage
are examples of biological structures that are subjected primarily to uniaxial
loading conditions. If a tissue or cytoskeletal component is stretched by
tractions applied only in the x1 direction, then t2 = t3 = 0. In addition,
symmetry demands that λ2 = λ3 . For an incompressible material, setting
−1/2
J = λ1 λ2 λ3 = 1 gives λ2 = λ3 = λ1 , and substitution into (6.18)2
reveals that the condition t2 = 0 is satisfied identically. (The condition
t3 = 0 already has been used to determine p.) Hence, Eq. (6.18)1 gives
1 1
t1 = 2λ1 1 − W1 + W2 + λ1 W4 . (6.25)
λ31 λ1
For a compressible (Blatz-Ko) material, setting t2 = 0 in Eq. (6.23)2
gives λ2 = λ−ν
1 . Then, (6.23)1 yields

t1 = 2C2 λ−1+2ν
1 − λ−31 . (6.26)
For ν = 1/4, the Cauchy stress is

t1
1/2 −5/2
σ1 = = λ1 t1 = 2C2 1 − λ1 . (6.27)
λ2 λ3
The loading behavior for various values of ν are illustrated in Fig. 6.6. In
contrast to the biaxial stress-stretch curves (Fig. 6.5), the curves for ν = 0.5
and ν = 0.49 are practically indistinguishable. (The curve for ν = 0.49 is
not shown.) Note also the strong asymmetry in the curves for tensile versus
compressive loading.
6.3 Simple Shear of Soft Tissue
For many years, shear deformation received relatively little attention in the
biomechanics literature. Nevertheless, shear is an important issue in nu-
merous problems. For example, torsional and transverse shear of the heart
wall play major roles in proper functioning of the left ventricle (Waldman
Fig. 6.6. Stress-stretch curves for uniaxial loading of an isotropic compressible bar for
various values of the material constant ν.
et al., 1985; Hansen et al., 1988; Moon et al., 1994; LeGrice et al., 1995,
1997).
In this section, we examine the forces required to sustain simple shear
of a homogeneous block of tissue (see Example 3.3 on page 80). As in the
stretching problem, surface tractions are assumed to be distributed uni-
formly, giving a homogeneous deformation. Hence, the equilibrium equa-
tions (6.3) are satisfied identically (with f = a = 0). The problem of simple
shear illustrates some important points in using nonorthogonal coordinate
systems to solve problems in nonlinear elasticity.
Fig. 6.7. Simple shear of a unit cube composed of transversely isotropic material.
Kinematics. For convenience, we list some of the results from Exam-

ple 3.3, which considered shear of a unit cube. In terms of the undeformed
coordinates X I , the Cartesian coordinates of a point in the deformed block
are
x1 = X 1 + kX 2 , x2 = X 2 , x3 = X 3 , (6.28)
where the parameter k defines the magnitude of the shear (Fig. 6.7). This
mapping produces the convected base vectors
g1 ∗ = e1 , g2∗ = ke1 + e2 , g3 ∗ = e3 , (6.29)
as given by (3.54), and Eqs. (2.18) provide the contravariant base vectors
∗ ∗ ∗
g1 = e1 − ke2 , g 2 = e2 , g 3 = e3 . (6.30)
The directions of these base vectors are shown in Fig. 6.7. In addition, with
(3.56)1 , the Cartesian components of the Lagrangian strain tensor are
⎡ ⎤
0 k/2 0
[EIJ ] = 12 [CIJ − δ IJ ] = ⎣ k/2 k 2 /2 0 ⎦ . (6.31)
0 0 0
Note that Eq. (3.52) yields J = det F = 1 for any value of k. Thus, the
deformation described by (6.28) is isochoric, regardless of the composition
of the block.
Later, we will need components of the metric tensors. Since GI = GI =
eI ,
⎡ ⎤
1 0 0
[GIJ ] = [GIJ ] = [eI · eJ ] = ⎣ 0 1 0 ⎦, (6.32)
0 0 1
and Eqs. (6.29) and (6.30) give

⎡ ⎤
1 k 0
[gIJ ] = [gI · gJ ] = ⎣ k 1 + k 2 0 ⎦
0 0 1
⎡ ⎤
1 + k 2 −k 0
[gIJ ] = [gI · gJ ] = ⎣ −k 1 0 ⎦. (6.33)
0 0 1
Constitutive Relations. We want to compute the stresses in the

block and the surface tractions required to sustain the specified deforma-
∗ ∗
tion. The geometry suggests that the stress components σ I J are appro-
priate for this problem, as they act in the directions defined by the faces of
the deformed block (Fig. 6.7). This can be seen from the dyadic representa-
∗ ∗
tion σ = σI J gI gJ , with gJ giving the direction of action (see Sec. 4.5). If
we stipulate that no loads are applied on the ±X 3 faces of the block, then
∗ ∗ ∗ ∗ ∗ ∗
we have the plane stress conditions σ 3 1 = σ 3 2 = σ 3 3 = 0, which must
hold throughout the block because the stress field is homogeneous. It is im-
portant to note that not all materials can satisfy plane stress and the plane
strain condition E33 = 0 simultaneously. For example, if the anisotropy is
such that the given shear in the X 1 X 2 -plane generates stresses on planes
normal to X 3 , then traction must be exerted on the ±X 3 faces.
Here, we consider a block composed of transversely isotropic material
with fibers oriented originally in the X 2 -direction (Fig. 6.7). For such a
material, Eq. (5.102) gives
W = W (I1 , I2 , I3 , I4 , I5 ). (6.34)
IK
Because EJI = GIK EKJ = δ EKJ = EIJ , Eqs. (3.142), (5.103), and
(6.31) yield the strain invariants
I1 = 3 + 2(E11 + E22 + E33 )

= 3 + k2
I2 = 3 + 4(E11 + E22 + E33 + E11 E22 + E22 E33 + E33E11
2 2 2
− E12 − E23 − E31 )
= 3 + k2
I3 = det(δ IJ + 2EIJ ) = 1
I4 = E22 = k 2 /2
2 2
I5 = E12 + E23 = k 2 /4. (6.35)
Equations (4.34) and (5.180)2 provide the constitutive equations

∗ ∗ ∂W
Jσ I J = sIJ = − pgIJ . (6.36)
∂EIJ
If the material is incompressible, then the Lagrange multiplier p can be
∗ ∗
determined using the condition σ 3 3 = 0, i.e.,
1 ∂W ∂W
p= = . (6.37)
g 3∗ 3∗ ∂E33 ∂E33
Given the strains of (6.31), all stress components can now be computed from
(6.36). On the other hand, if the block consists of compressible material,
then p = 0 and Eq. (6.36) stipulates that the functional form of W must
∗ ∗
be compatible with the plane-stress condition σ 3 3 = ∂W/∂E33 = 0 when
E33 = 0.
With W expressed in terms of the strain invariants, Eq. (5.148) gives
the constitutive relation
∗
J∗
σI = ΦGIJ + ΨH IJ − pgIJ + ΘM IJ + ΛN IJ (6.38)
for a transversely isotropic material. The H IJ in this equation are com-

puted from (5.141)4 . Moreover, because the principal material directions
coincide with the X I -axes, we have IJ = EIJ and AJI = ∂X J /∂Z I =
∂X J /∂X I = δ JI , and Eqs. (5.149)3,4 give (for fibers in the X 2 -direction)
M IJ = δ I2 δ J2
N IJ = (δ I2 δ J1 + δ I1 δ J2 )E12 .
In matrix form, the results are

⎡ ⎤
2 + k2 −k 0
[H ] = ⎣ −k
IJ
2 0 ⎦
2
0 0 2+k
⎡ ⎤ ⎡ ⎤
0 0 0 0 1 0
[M IJ
]=⎣ 0 1 0 ⎦, [N IJ
]=⎣ 1 0 0 ⎦ E12. (6.39)
0 0 0 0 0 0
With these relations, Eq. (6.38) yields the Cauchy stress components
∗ ∗
σ1 1
= Φ + (2 + k 2 )Ψ − (1 + k 2 )p
∗ ∗
σ2 2
= Φ + 2Ψ − p + Θ
∗ ∗
3 3
σ = Φ + (2 + k 2 )Ψ − p
∗ ∗
σ1 2
= k(−Ψ + p + Λ/2)
∗ ∗ ∗ ∗
σ1 3
= σ2 3
= 0, (6.40)
where Eqs. (5.141) and (5.149) give

∂W ∂W ∂W
Φ=2 , Ψ=2 , p = −2
∂I1 ∂I2 ∂I3
∂W ∂W
Θ= , Λ= (6.41)
∂I4 ∂I5
since I3 = 1 for simple shear.
∗ ∗
Surface Tractions. With the components σ I J of the Cauchy stress
tensor given above, we can compute the true traction vector on the surfaces
of the block using
T(ni ) = ni · σ, (6.42)
where ni is the unit vector normal to the deformed xi -surface. The geom-
etry of Fig. 6.7, along with Eqs. (6.30) and (6.33), gives
∗ ∗
g1 e1 − ke2 g2
n1 = ∗ ∗ = √ , n 2 = ∗ ∗ = e2 . (6.43)
g1 1 1 + k2 g2 2
Hence, we have
g1
∗ ∗ ∗ ∗ ∗
σI J
∗ ∗
σ1 J
(n1 ) 1
T = ∗ ∗ · σ I J
gI gJ = ∗ ∗ δ I gJ = ∗ ∗ gJ
g1 1 g1 1 g1 1
1 ∗ ∗ ∗ ∗

= ∗ ∗ σ 1 1 g1 ∗ + σ 1 2 g2 ∗
g 1 1
g2
∗ ∗ ∗ ∗ ∗
σI J
∗ ∗
σ2 J
T(n2 ) = ∗ ∗ · σ I J gI gJ = ∗ ∗ δ 2I gJ = ∗ ∗ gJ
g2 2 g2 2 g2 2
1 ∗ ∗ ∗ ∗

= ∗ ∗ σ 2 1 g1 ∗ + σ 2 2 g2 ∗ . (6.44)
g 2 2
Because the T(ni ) are vectors, the stress components normal and tan-
gential to the surfaces of the block can be determined by dotting with the
corresponding unit vectors (see the geometry in Fig. 6.7). On the x1 -face,
these components are
∗ ∗ ∗
(n1 ) g1 σ1 1
Tnorm = ∗ ∗ · T(n1 ) = 1∗ 1∗
g1 1 g
∗ ∗
σ1 1
=
1 + k2
g2 ∗ 1 ∗ ∗ ∗ ∗

(n1 )
Ttang =√ · T(n1 ) = ∗ ∗ σ 1 1 g2 ∗ 1 ∗ + σ 1 2 g2 ∗ 2 ∗
g2 ∗ 2 ∗ g 1 1 g2 ∗ 2 ∗
k ∗ ∗ ∗ ∗
= 2
σ 1 1 + σ1 2 , (6.45)
1+k
while on the x2 -face,
∗ ∗ ∗
(n2 ) g2 σ2 2
Tnorm = ∗ ∗ · T(n2 ) = ∗ ∗
g2 2 g2 2
∗ ∗
= σ2 2
g1 ∗ 1 ∗ ∗
(n ) 2∗ 2∗
2
Ttang = √ · T(n2 ) = σ 2 1
g 1 ∗ 1∗ + σ g 1 ∗ 2∗
g1 ∗ 1 ∗ g1 ∗ 1 ∗ g 2 ∗ 2 ∗
∗ ∗ ∗ ∗
= σ2 1
+ kσ 2 2
. (6.46)
The applied forces needed to produce the specified deformation can be

computed by multiplying these tractions by the appropriate deformed sur-
face areas. (Note that, for our solution to be “exact,” these forces must be
distributed uniformly over the surfaces of the block.) Suppose the areas of
the undeformed and deformed faces of the block are dAi and dai , respec-
tively. Then, a glance at the geometry (Fig. 6.7) reveals that √dA2 does not
change during deformation, but dA1 increases by a factor of 1 + k 2 , i.e.,
by the stretch ratio λ2 of the originally vertical side. Alternatively, we can
use Eq. (3.128), which gives (for J = 1)
dai = dAi · F−1 , (6.47)
where
dA1 = e1 dA1 , dA2 = e2 dA2

da1 = n1 da1 , da2 = n2 da2 . (6.48)
With Eqs. (3.22)2 and (6.30) and GI = eI , the inverse of the deformation
gradient tensor for this problem can be written in the form
F−1 = GI gI = e1 (e1 − ke2 ) + e2 e2 + e3 e3 . (6.49)
With the above results, (6.47) gives

da1 = 1 + k 2 dA1 , da2 = dA2 , (6.50)
in agreement with our observation.

Next, we examine results for incompressible and compressible blocks
composed of tissues with specific properties.
6.3.2 Solution
Incompressible Tissue. For an incompressible material, the condi-
tion I3 = 1 renders ∂W/∂I3 indeterminate, and p becomes a Lagrange
∗ ∗
multiplier in Eqs. (6.40). Setting σ 3 3 = 0 in (6.40) yields
p = Φ + (2 + k 2 )Ψ, (6.51)
and so the non-zero components of σ become

∗ ∗
σ1 1
= −k 2 [Φ + (2 + k 2 )Ψ]
∗ ∗
σ2 2
= −k 2 Ψ + Θ
∗ ∗
σ1 2
= k[Φ + (1 + k 2 )Ψ + Λ/2]. (6.52)
Substituting these relations into (6.45) and (6.46) yields the surface trac-
tions
(n1 ) k2
Tnorm =− [Φ + (2 + k 2 )Ψ]
1 + k2
(n )
1
k kΛ
Ttang = 2
(Φ + Ψ) +
1+k 2
(n2 )
Tnorm = −k 2 Ψ + Θ
(n )
2
Ttang = k(Φ + Ψ + Λ/2 + Θ). (6.53)
To illustrate the behavior of this solution, consider a block composed of

soft tissue with the strain-energy density function of Eq. (6.19), where the
strain invariants are given by (6.35). In this case, Φ = 2C1 , Θ = C1 βk 2 ,
and Ψ = Λ = 0. For this material, Eqs. (6.52) give the stresses
∗ ∗ ∗ ∗ ∗ ∗
σ1 1
= −2C1 k 2 , σ2 2
= C1 βk 2 , σ1 2
= 2C1 k, (6.54)
and Eqs. (6.53) yield the surface tractions
(n1 ) 2C1 k 2 (n ) 2C1 k

Tnorm =− , 1
Ttang =
1 + k2 1 + k2
(n2 ) (n )
Tnorm = C1 βk 2 , 2
Ttang = C1 k(2 + βk 2 ). (6.55)
Note that, unlike in the linear theory, normal forces must be exerted on
the block to maintain simple shear. For the material considered here, the
normal stress is compressive on the x1 -face and tensile on the x2 -face. How-
ever, if the material is isotropic (β = 0), the normal force on the x2 -face
(n2 )
is zero. In fact, Eq. (6.53)3 shows that Tnorm can be compressive for cer-
tain transversely isotropic materials. For k << 1, these results reduce to
those of the linear theory, with the tangential tractions becoming equal to
∗ ∗
the shear stress σ 1 2 , and the normal stresses [O(k 2 )] can be neglected
compared to the shear stress [O(k)].
Compressible Tissue. If the block is composed of compressible ma-

terial, Eqs. (6.40)3 and (6.41) give the plane-stress condition

3∗ 3∗ ∂W 2 ∂W ∂W
σ =2 + (2 + k ) + = 0. (6.56)
∂I1 ∂I2 ∂I3
This relation constrains the composition of the block to materials that can
∗ ∗
sustain both plane strain (E33 = 0) and plane stress (σ 3 3 = 0) conditions
6.4. Extension and Torsion of a Papillary Muscle 251
simultaneously for shear in the X 1 X 2 -plane. One material that does satisfy
these conditions consists of fibers initially aligned in the X 2 -direction that
are embedded in a matrix of the Blatz-Ko type. Here, we consider a material
characterized by

1 − 2ν −ν/(1−2ν)
2
W = C1 I1 − 3 + I3 − 1 + βI4 , (6.57)
ν
which is an extended form of Eq. (5.114) with α = 1 and reduces to
Eq. (6.19) when I3 = 1. Direct substitution shows that Eq. (6.56) is satis-
fied identically, and Eqs. (6.40) yield stress components identical to those of
Eq. (6.54). This is the expected result, because the specified deformation is
isochoric, and hence material compressibility should not affect the solution.
6.4 Extension and Torsion of a Papillary Muscle
Papillary muscles play a vital role in the pumping efficiency of the heart.
These cylindrically shaped muscles are attached at one end to the wall of
the left or right ventricle and at the other end to the mitral or tricuspid
valve, respectively. During ventricular systole, increasing blood pressure
in the ventricles tends to push the valves upward into the atria, which
would allow backflow of blood during ejection. The papillary muscles pre-
vent this inversion by contracting and holding the valves closed. Due to
their relatively simple geometry and muscle fibers aligned along their axes,
papillary muscles have long been popular in studies of the passive and ac-
tive mechanical properties of heart muscle (Pinto and Fung, 1973; Fung,
1993; Criscione et al., 1999; Humphrey, 2002). For determining extensional
and shear properties, combined extension and torsion is a useful loading
protocol.
As a model for a papillary muscle, consider a solid circular cylinder
of undeformed radius b0 and length 0 (Fig. 6.8). The unloaded cylinder
is composed of material that is transversely isotropic relative to the axial
direction Z. The ends are subjected to tractions that exert a net twisting
moment M and normal force N , while the curved surface is traction free.
Given the angle of twist per unit undeformed length ψ and the (uniform)
axial stretch ratio λ, we want to compute M , N , and the stress distribution
in the cylinder. Note that, unlike the previous problems in this chapter,
torsional deformation of a cylinder is not homogeneous.1
1 This book considers only passive behavior of heart muscle. The mechanics of active
contraction are treated elsewhere (see, e.g., Fung, 1993; Humphrey, 2002; Taber, 2020).
Z
N
Q
b0 b
R
l0 l g3
G3 g3*
Fibers
Undeformed Deformed
Fig. 6.8. Extension and torsion of a cylindrical model for a papillary muscle loaded by
an axial force N and twisting moment M.

Kinematics. This problem is a special case of that considered in Ex-
ample 3.4 (page 82), where we analyzed the deformation of a tube under-
going simultaneous extension, inflation, and torsion. The deformation is
described by
r = r(R), θ = Θ + ψZ, z = λZ, (6.58)
1 2 3 1 2 3
where (R, Θ, Z) = (X , X , X ) and (r, θ, z) = (x , x , x ) are cylindrical
polar coordinates. The covariant base vectors, provided by Eqs. (3.60), are
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
eR er ṙer
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
[GI ] = ⎣ ReΘ ⎦ , [gi ] = ⎣ reθ ⎦ , [gI ] = ⎣ reθ ⎦ , (6.59)
eZ ez ψr eθ + λez
and Eqs. (2.18) give the contravariant base vectors
⎡ ⎤
⎡ ⎤ ⎡ ⎤ 1
eR er er
⎢ ṙ ⎥
⎢ 1 ⎥ ⎢ 1 ⎥ ⎢ ⎥
[GI ] = ⎢ ⎥ i ⎢ ⎥ I ⎢ 1 ψ ⎥ . (6.60)
⎣ R eΘ ⎦ , [g ] = ⎣ r eθ ⎦ , [g ] = ⎢ r eθ − λ ez ⎥
⎣ 1 ⎦
eZ ez ez
λ
To prevent confusion with primed quantities used later in this section, we
use dot to denote differentiation with respect to R. With these equations,
the various components of the metric tensor are

⎡ ⎤ ⎡ ⎤
1 0 0 1 0 0
[GIJ ] = [GI · GJ ] = ⎣ 0 R2 0 ⎦ , [gij ] = [gi · gj ] = ⎣ 0 r2 0 ⎦,
0 0 1 0 0 1
⎡ ⎤
ṙ 2 0 0
[gIJ ] = [gI · gJ ] = ⎣ 0 r2 ψr 2 ⎦ (6.61)
2 2 2
0 ψr 2 λ +ψ r
⎡ ⎤ ⎡ ⎤
1 0 0 1 0 0
IJ ⎢ 1 ⎥ ij ⎢ 1 ⎥
G = [GI · GJ ] = ⎣ 0 0 ⎦, g = [gi · gj ] = ⎣ 0 0 ⎦,
R2 r2
0 0 1 0 0 1
⎡ ⎤
1
⎢ ṙ 2 0 0 ⎥
IJ ⎢ ⎥
⎢ 1 ψ2 ψ ⎥
g = [gI · gJ ] = ⎢ 0 + − 2 ⎥. (6.62)
⎢ r2 λ2 λ ⎥
⎣ ψ 1 ⎦
0 − 2
λ λ2
The deformation gradient tensor can be written in several forms. With
the above relations, Eq. (3.22)1 gives
F = gI G I
= g1 ∗ G 1 + g2 ∗ G 2 + g3 ∗ G 3
r
= ṙer eR + eθ eΘ + λez eZ + ψreθ eZ
R
= ṙg1 G1 + g2 G2 + λg3 G3 + ψg2 G3 . (6.63)
Note that, with the FIi given by Eq. (3.64), the last expression is consistent
with the form F = FIi gi GI . (Recall again the distinction between the base
vectors g1∗ and g1 , etc.) For relatively simple problems such as this, many
authors find it convenient to work entirely with physical components, as
given by the third line of (6.63). For illustration, however, we stay here
with tensor components. Of course, physical components always can be
computed as desired.
Finally, Eq. (3.69) gives the Lagrangian strain components
⎡ 2 ⎤
ṙ − 1 0 0
1⎣ ⎦,
[EIJ ] = 0 r 2 − R2 ψr 2 (6.64)
2 2 2 2 2
0 ψr λ +ψ r −1
and (3.146) and (5.167) provide the strain invariants

r2
I1 = ṙ 2 + + λ2 + ψ2 r 2
R2
r2 2 2 2 λ2 r 2
I2 = ṙ 2 + λ + ψ r +
R2 R2
λ2 ṙ 2 r 2
I3 =
R2
I4 = Ê33 = 12 (λ2 + ψ 2 r 2 − 1)

ψ 2r4
I5 = Ê13
2 2
+ Ê23 = . (6.65)
4R2
Recall that the modified invariants I4 and I5 are needed for a curvilinear
transversely isotropic material, with the physical strain components Ê13
and Ê23 given by Eq. (3.70).
Constitutive Relations. For the present problem, Cauchy stress

components are defined by the dyadic representations
∗
J∗
σ = σ ij gi gj = σ I gI gJ . (6.66)
The stress analysis can be simplified by taking advantage of the inherent
symmetry in this problem. For example, from the symmetry of the ge-
ometry and loading, we can show that the conditions σ 12 = σ 13 = 0 or
∗ ∗ ∗ ∗
σ 1 2 = σ 1 3 = 0 must be satisfied. Further simplification can be achieved
∗ ∗
by working with the stress components σ ij , rather than σ I J (or sIJ ). The
reason for this will become clear when we write down the equilibrium equa-
tions.2 In addition, the twist causes t to be asymmetric, making tIJ an
inconvenient choice.
∗ ∗
To determine σ ij , however, we effectively compute σ I J first, and then
Eqs. (4.33) and (4.34) give σ ij . Equation (5.168) provides the constitutive
relation
∗
J∗
σI = ΦGIJ + ΨH IJ − pgIJ + Θ M IJ + Λ N IJ (6.67)
where
−1/2 ∂W −1/2 ∂W 1/2 ∂W
Φ = 2 I3 , Ψ = 2 I3 , p = −2 I3
∂I1 ∂I2 ∂I3
∗ ∗
2 Green and Zerna (1968) circumvented complications in using σ I J by taking the un-
deformed base vector G3 = GZ to be at an angle relative to the vertical axis so that

g3 = gz is vertical in the deformed cylinder.
−1/2 ∂W −1/2 ∂W
Θ = I3 , Λ = I3 . (6.68)
∂I4 ∂I5
The M IJ and N IJ are functions of ÂIJ defined by Eq. (5.164). Because

the principal material axes X I coincide with the X I -axes, we find
⎡ ⎤
I 1 0 0
1 ∂X 1
ÂIJ = = δ I = ⎣ 0 R−1 0 ⎦ . (6.69)
G(J J ) ∂X J G(J J ) J
0 0 1
Hence, Eqs. (5.141)4 and (5.169)3,4 yield
⎡ 2 ⎤
r 2 2 2
⎢ R2 + λ + ψ r 0 0 ⎥
IJ ⎢ ⎢ 1 2 r 2 ⎥
⎥
=⎢ 2 2 2 ⎥
H
⎢ 0 ṙ + λ + ψ r − ψ
⎣ R 2 R 2 ⎥
2
r 2
r ⎦
0 − 2ψ ṙ 2 + 2
R R
⎡ ⎤
⎡ ⎤ 0 0 0
IJ 0 0 0 IJ ⎢ ⎢ ψr 2 ⎥ ⎥
M = ⎣ 0 0 0 ⎦, N =⎢ 0 0
2R2 ⎥ , (6.70)
⎣ 2 ⎦
0 0 1 ψr
0 0
2R2
which are valid for a cylinder with fibers oriented originally in the X 3 -
direction. The solution for an isotropic cylinder can be obtained by setting
Λ = Θ = 0. Note also that Eq. (3.70) provides the ÊI J = ÊIJ required
by N IJ . Inserting these relations into (6.67) gives
∗ ∗ r2 1
σ1 1
= Φ+ + λ2 + ψ 2 r 2 Ψ − 2 p
R2 ṙ
∗ ∗ 1 1 1 ψ2
σ2 2
= 2 Φ + 2 ṙ 2 + λ2 + ψ 2 r 2 Ψ − + p
R R r2 λ2
∗ ∗ r2 1
σ3 3
= Φ + ṙ 2 + 2 Ψ − 2 p + Θ
R λ
∗ ∗ r2 1 r2
σ2 3
= ψ − 2Ψ+ 2 p+ Λ
R λ 2R2
∗ ∗ ∗ ∗
σ1 2
= σ1 3
= 0, (6.71)
1∗ 2∗ 1∗ 3∗
which are consistent with the symmetry requirement σ =σ = 0.
Cauchy stress components relative to the {gi } basis now can be com-
puted using Eqs. (4.33) and (4.34), which give
∗
J∗
σ ij = FIi FJj σ I .
With the FIi given by the last of (6.63), this equation yields
⎡ ⎤
2 1∗ 1∗
⎢ ṙ σ 0 0 ⎥
⎢ ⎥
[σ ] = ⎢
ij
⎢ 0 σ
∗
2 2 ∗
+ 2ψσ
∗
2 3 ∗
+ψ σ
∗
2 3 3 ∗
λ(σ
∗
2 3 ∗
+ ψσ 3∗ 3∗ ⎥ .
) ⎥
⎣ ⎦
∗ ∗ ∗ ∗ ∗ ∗
0 λ(σ 2 3 + ψσ 3 3 ) λ2 σ 3 3
(6.72)
Substituting the convected stress components from (6.71) gives

11 2 r2 2 2 2
σ = ṙ Φ + +λ +ψ r Ψ −p
R2
1 2 ṙ 2 λ2 2 2 p ψ2 r2
σ 22 = + ψ Φ + + + ψ ṙ Ψ − + Λ + ψ 2 Θ
R2 R2 R2 r2 R2

r2
σ 33 = λ2 Φ + ṙ 2 + 2 Ψ + Θ − p
R
r2
σ 23 = λψ Φ + ṙ 2 Ψ + Λ + Θ
2R2
σ 12 = σ 13 = 0. (6.73)
Equilibrium. Since the stress field in this problem is not uniform,

the equations of equilibrium are not all satisfied trivially as they were in
the previous problems. Several choices are available for the form of these
equations [see Eqs. (6.3)]. All these forms are equivalent, but some may be
more convenient than others. For example, since t is not symmetric for this
problem (due to the shear), we would like to take advantage of the symmetry
in the stress tensor σ or s. In many problems, the form ∇ · (s · FT ) = 0
has advantages, because derivatives are taken with respect to undeformed
coordinates, which are known a priori. In the current problem, however, the
relatively simple geometry makes the choice ∇ ¯ · σ = 0 more convenient. In
this case, the scalar equations (4.97) for cylindrical coordinates apply with
f i = ai = 0.
Symmetry demands that the Cauchy stress components be functions
of r alone. Hence, for σ 12 = σ 13 = 0, the last two of (4.97) are satisfied
identically, and the first equation becomes
dσ 11
r + σ 11 − r 2 σ 22 = 0, (6.74)
dr
which corresponds to radial equilibrium.
Boundary Conditions. We seek a solution for which tractions are

applied only at the ends of the cylinder. On the deformed curved surface,
the unit normal is
g1
n1 = = g1,
g 11
and the true traction vector on this surface is

T(n1 ) = n1 · σ
= g1 · (σ ij gi gj ) = σ ij δ 1i gj
= σ 1j gj .
Thus, for a stress-free surface at the deformed radius b, the appropriate
boundary conditions are
r = b : σ 11 = σ 12 = σ 13 = 0. (6.75)
12 13
As discussed earlier, σ and σ already are assumed to be zero everywhere,
and the condition on σ 11 provides the boundary condition required by the
differential equation (6.74).
Similarly, on the ends of the cylinder, the traction vector is
T(n3 ) = n3 · σ = g3 · σ = σ 3j gj ,
and the resultant force and moment applied to the ends are3
2π b
N= T(n3 ) r dr dθ
0 0
2π b
M= r × T(n3 ) r dr dθ, (6.76)
0 0
where r = rer is the radius vector from the x3 -axis (i.e., the z-axis). With
the gi given by (6.59)2 , these expressions become
2π b
31
N= σ er + σ 32 reθ + σ 33 ez r dr dθ
0 0
2π b
M= (rer )× σ 31 er + σ 32 reθ + σ 33 ez r dr dθ.
0 0
These equations simplify if we note that the stresses are functions of r and
&that er = ex &cos θ + ey sin θ and eθ = −ex sin θ + ey cos θ. Then, since
2π 2π
0 sin θ dθ = 0 cos θ dθ = 0, we have
b
N = 2πez σ 33 r dr ≡ N ez
0
b
M = 2πez σ 32 r 3 dr ≡ M ez , (6.77)
0
3 Here, N is a force rather than a unit normal vector.

which provide the resultant axial force N and twisting moment M applied
at the ends of the cylinder.
6.4.2 Solution
In the following, we develop explicit solutions for cylindrical specimens
composed of incompressible and compressible transversely isotropic muscle.
Incompressible Muscle. If the papillary muscle is assumed to be

incompressible, the deformation must satisfy the condition
1/2 r dr
J = I3 = λ =1 (6.78)
R dR
as provided by Eq. (6.65)3 . Integrating this equation yields
r2 R2
λ = + C,
2 2
and the condition r(0) = 0 gives C = 0. Thus,
R
r(R) = √ , (6.79)
λ
√
which provides the deformed radius b = r(b0 ) = b0 / λ. With λ and ψ
specified, this relation completes the solution for the deformation field, and
Eq. (6.64) gives the Lagrangian strain components
⎡ ⎤
1
⎢ −1 0 0 ⎥
⎢ λ ⎥
1⎢⎢ 2 ⎥
1 ψR ⎥
[EIJ ] = ⎢ 0 −1 R 2 ⎥. (6.80)
2⎢ λ λ ⎥
⎢ ⎥
⎣ ψR2 ψ 2 R2 ⎦
0 λ2 + −1
λ λ
√
In addition, substituting ṙ = r/R = 1/ λ into (6.73) gives the non-zero
stress components

11 1 1 2 2 2
σ = Φ+ +λ +ψ r Ψ −p
λ λ
1 1 1
r 2 σ 22 = + ψ2 r2 Φ + + λ2 + ψ2 r 2 Ψ
λ λ λ
1
+ψ 2 r 2 Λ + Θ − p
λ
2
σ 33 = λ2 Φ + Ψ + Θ − p
λ
1 1
σ 23 = λψ Φ + Ψ + Λ + Θ . (6.81)
λ 2λ
As discussed in Sec. 5.5, the constraint I3 = 1 for incompressible mate-

rial means that p cannot be determined using Eq. (6.68)3. In this case, p
is a Lagrange multiplier, which can be determined by first integrating the
equilibrium equation (6.74) to get

dr
σ + (σ 11 − r 2 σ 22 )
11
= 0. (6.82)
r
Substituting (6.81)1,2 reveals that p drops out of the integrand, and the
resulting equation can be solved for p to obtain
b
1 1 1
p(r) = Φ+ + λ2 + ψ2 r 2 Ψ + ψ2 Φ + Λ + Θ r dr.
λ λ r λ
(6.83)
The limits of integration are chosen to satisfy the boundary condition
σ 11 (b) = 0, as can be shown by direct substitution into Eq. (6.81)1 . Note
also that p would not disappear from the integral if pseudostress compo-
nents are used instead of Cauchy stresses.
Given W , Eq. (6.83) can be integrated to obtain the Lagrange multiplier
at any point in the cylinder. Then, with p known, stress components are
given by (6.71) or (6.81). In the second case, the non-zero σ ij become
b
1
σ̂ 11 = σ 11 = −ψ 2 Φ + Λ + Θ r dr
r λ
1
σ̂ 22 = r 2 σ 22 = σ 11 + ψ2 r 2 Φ + Λ + Θ
λ
1 1 ψ2 r2
σ̂ 33 = σ 33 = σ 11 + λ2 − Φ+ λ− 2 − Ψ + λ2 Θ
λ λ λ
1
σ̂ 23 = rσ 23 = ψr λΦ + Ψ + Λ + λΘ , (6.84)
2
where hat indicates physical components defined by Eq. (4.47). Finally, the
resultant force and moment at the ends of the cylinder can be computed
using (6.77).
For the special case of muscle modeled as a modified Mooney-Rivlin
material with
W = C1 (I1 − 3) + C2 (I2 − 3) + C4 I42 , (6.85)

√
Eqs. (6.68) and (6.84)1 yield (with R = r λ)
σ 11 = −ψ 2 (b2 − r 2 )[C1 + C4 (λ2 − 1)/2 + C4 ψ2 (b2 + r 2 )/4], (6.86)

and the other stress components can be computed easily from the rest of
Eqs. (6.84). Substitution into (6.77) then yields the axial force and twisting
moment

1 ψ 2 b20 1 ψ2 b20
N = 2πb20 C1 λ − 2 − + C 2 1 − −
λ 4λ2 λ3 2λ3
%
C4 3 1 ψ 2 b20 ψ 4 b40
+ λ −λ+ 1+ 2 −
2 λ 4 6λ3

πb40 ψ C2 C4 2ψ2 b20
M = C1 + + λ2 − 1 + . (6.87)
λ λ 2 3λ
Note that, for an isotropic cylinder (C4 = 0), M is proportional to ψ, as
in the linear solution. Such is not the case, however, if the material is
transversely isotropic relative to the axial direction (C4 = 0).
Stress distributions for the special case of a neo-Hookean material
(C2 = C4 = 0) reveal clear differences between the linear and nonlinear
solutions (see Fig. 6.9a,b). According to the linear theory, for example, the
only non-zero stress component is the shear stress σ̂ 23 . In contrast, for large
angles of twist, the nonlinear theory predicts comparable magnitudes for
all three normal stress components (Fig. 6.9a). The results in Fig. 6.9a are
for a cylinder held at its undeformed length. As shown in Fig. 6.9b, axial
stretch increases the magnitude of the torsional shear stress σ̂ 23 . Moreover,
the distribution of axial stress σ̂ 33 , which is independent of r for stretch
without twist, becomes more and more nonuniform as the torsion increases
(Fig. 6.9a). (Note also that σ̂ 33 = σ̂ 11 when λ = 1.) Finally, the pres-
ence of axial fibers (C4 = 0) increases the magnitudes of the peak stresses
(Fig. 6.9c).
If the cylinder is held at its undeformed length (λ = 1), the axial stress
33
σ̂ is compressive at all r (Fig. 6.9a). This stress distribution produces
a net compressive force N that increases in magnitude with the twist
(Fig. 6.10).4 Hence, if N is not supplied, the cylinder elongates when
twisted. This phenomenon is known as the Poynting effect. For λ = 2, the
effect of twist on N is not nearly as large (Fig. 6.10).
Compressible Muscle. If the muscle is assumed to be compressible,

then the problem contains one fewer unknown (p). But since the deforma-
tion is no longer constrained by the incompressibility condition, the function
r(R) is not found as easily as in the incompressible case. For a given W ,
4 For comparison with the following solution for a compressible cylinder, the results in
Fig. 6.10 are for an isotropic material with C1 = C4 = 0. The case C2 = C4 = 0 is
similar.
Fig. 6.9. Distributions of Cauchy stress (physical components) in a solid circular cylinder
(papillary muscle) undergoing combined extension and torsion. (a,b) isotropic cylinder
with C2 = C4 = 0; (c) transversely isotropic cylinder with C2 = 0 and C4 = C1 /2.
Eqs. (6.73) still apply, but p is given by (6.68)3 rather than (6.83). If λ and
ψ are specified, then substituting the constitutive relations into the equilib-
rium equation (6.74) provides a differential equation to be solved for r(R).
In general, the resulting equation must be solved numerically, e.g., by finite
differences, which may be significantly more involved than the solution by
quadratures in the incompressible case.
A relatively simple solution, however, can be found for passive muscle
modeled as a specialized Blatz-Ko material without fibers (Carroll and
Fig. 6.10. Axial force generated in a solid isotropic circular cylinder undergoing extension
and torsion (C1 = C4 = 0). The incompressible and compressible solutions are identical
for λ = 1.
Horgan, 1990). In particular, we consider an isotropic material with

I2 1/2
W = C2 + 2I3 − 5 (6.88)
I3
as given by Eq. (5.115) with μ = 2C2 . Note that this relation is the
compressible counterpart of Eq. (6.85) if C1 = C4 = 0. With Eq. (6.68)3
providing p, Eqs. (6.81) yield the non-zero physical stress components
R
σ̂ 11 = σ 11 = 2C2 1 −
λr ṙ 3
R3
σ̂ 22 = r 2 σ 22 = 2C2 1 −
λr 3 ṙ
R(1 + ψ 2 R2 )
σ̂ 33 = σ 33 = 2C2 1 −
λ3 rṙ
ψR3
σ̂ 23 = rσ 23 = 2C2 . (6.89)
λ2 r 2 ṙ
Substituting these relations into Eq. (6.74) gives the nonlinear differen-
tial equation
3 3
dR d2 R dR R
3R + − =0 (6.90)
dr dr 2 dr r
to be solved for the undeformed coordinate R(r). This equation, written
in Eulerian form for convenience, admits a solution of the form
R = cr, (6.91)
where c is a constant. Inserting this relation into (6.89)1 and applying the
boundary condition σ 11 (b) = 0 yields
c = λ1/4 , (6.92)
and the stress components become

σ 33 = 2C2 1 − λ−5/2 (1 + λ1/2 ψ2 r 2 )
ψr
rσ 23 = 2C2
λ
σ 11 = σ 22 = σ 12 = σ 13 = 0. (6.93)
Finally, using these expressions in Eqs. (6.77), we compute the resultant
force and twisting moment

2π b0 33 1 1 ψ2 b20
N= √ σ R dR = 2πC2 b20 √ − 3 −
λ 0 λ λ 2λ3

2π b0 23 3 πC2 b40
M= σ R dR = ψ, (6.94)
λ 0 λ2
in which Eq. (6.91) has provided the transformation from r to R.
Fig. 6.11. Effects of material compressibility on Cauchy stresses (physical components)

in a solid isotropic circular cylinder (papillary muscle) undergoing extension and torsion
(C1 = C4 = 0).
Results are compared to those for the corresponding incompressible

cylinder with C1 = C4 = 0, i.e., W = C2 (I2 − 3).5 For λ = 1, the defor-
mation is isochoric (r = R and I3 = 1) even for the compressible cylinder,
5 Note the differences between the curves for an incompressible, isotropic cylinder with
C1 = 0 (Fig. 6.11) and C2 = 0 (Fig. 6.9b).

and so compressibility has no effect. Otherwise, compressibility decreases

the magnitudes of the stresses and the applied axial force (Figs. 6.10 and
6.11).
6.5 Extension, Inflation, and Torsion of an Artery with

Residual Stress
Arteries are the blood vessels that carry blood away from the heart. The
wall of an artery consists of three layers. The inner layer (intima) consists
of a single layer of epithelial cells, the middle layer (media) contains elastin
and smooth muscle, and the outer layer (adventitia) is composed mainly of
connective tissue.
Reflecting differences in function, the size and composition of arteries
change with distance from the heart. Arteries near the heart, including the
aorta, are relatively large and serve primarily as conduits of blood. The
walls of these “elastic arteries” are dominated by elastin. Downstream,
arteries gradually become smaller in diameter, and smooth muscle becomes
more prominent. The small-caliber arterioles (“muscular arteries”) regulate
the flow of blood to the tissues via smooth-muscle contraction or relaxation.
This response changes the diameter of the vessel, thereby altering resistance
to flow.
Hence, arteries are not merely passive tubes. Besides actively chang-
ing their diameter, arteries grow (change size) and remodel (change ma-
terial properties) to adapt to changes in flow and pressure. Developing
theoretical models for these processes, while accounting for heterogeneity
and viscoelasticity, is a challenging and ongoing endeavor (Rachev, 1997;
Humphrey, 2002; Taber, 2020). In this section, however, we ignore these
complications and consider arteries as thick-walled passive, homogeneous,
pseudoelastic tubes. We do include anisotropy, however.
Another feature that we include is residual stress, i.e., the stress that
remains in a body when all external loads are removed. Residual stress in
artery walls can be generated by nonuniform growth, swelling, turnover of
wall constituents, or other means. A popular method used to characterize
the magnitude of residual stress in soft tissue is to cut the tissue and mea-
sure the resulting deformation caused by the release of stress (Fung, 1990;
Humphrey, 2002; Taber, 2020). For example, when an unloaded section
of an artery is cut transmurally, the section typically springs open as cir-
cumferential residual stress is relieved (Fig. 6.12). Measuring the opening
angle φ provides an approximate way to characterize the amount of residual
6.5. Extension, Inflation, and Torsion of an Artery with Residual Stress 265
strain that was in the section before the cut. Residual stress then can be
computed if material properties are known.
Most investigators have assumed that a single radial cut is enough to
render an artery nearly free of stress. Some studies have suggested, however,
that a combination of radial and circumferential cuts may be needed (Vos-
soughi et al., 1993; Greenwald et al., 1997; Taber and Humphrey, 2001).
For simplicity, we assume here that the zero-stress state is produced by one
radial cut.
6.5.1 Artery Model
Consider the following experiment. A straight section of an artery is dis-

sected and mounted in a testing machine that subjects the vessel to simul-
taneous extension, inflation, and torsion. The inner and outer radii, axial
stretch, and angle of twist are recorded, as well as the applied internal
pressure P , axial force N , and twisting moment M . After testing under
various loading protocols, a thin slice of the artery is dissected and cut once
transmurally, and the opening angle is measured, along with the inner and
outer radii of the opened section. One way to determine material properties
would be to vary the material coefficients in a model until the theoretical
results match as closely as possible the values of all the measured variables
under various loading conditions.
F = F2 . F1
f0
F1 F2 a b
a0 f
b0
B0 BU BL
(R,Q,Z) (r,J,z) (r,q,z)
Fig. 6.12. Artery (or left ventricle) cross section in zero-stress (B0), unloaded (BU ), and
loaded (BL) configurations.
The artery is modeled as a straight thick-walled circular tube com-

posed of homogeneous, incompressible, cylindrically orthotropic material
(Fig. 6.12). For simplicity, our analysis ignores all end effects near the sup-
ports in BL and near the cut edges in B0 . In the unloaded configuration

BU , the principal material directions coincide with the radial, circumferen-
tial, and longitudinal directions of the vessel. The zero-stress configuration
B0 , obtained by cutting BU transmurally, is assumed to approximate the
shape of a circular sector with an inner radius a0 , outer radius b0 , and
opening angle φ. The inner and outer radii of the loaded configuration BL
are a and b, respectively.
This problem bears many similarities to the papillary muscle problem.
Hence, many aspects of the analysis are similar. However, to illustrate again
that multiple approaches usually are possible in nonlinear elasticity, we vary
some key aspects. For example, the formulation is based on a general strain-
energy density function, rather than assuming a specific material symmetry
at the outset. In addition, the entire analysis uses physical components of
stress and strain, rather than tensor components.
Kinematics. The deformation for this problem, excluding residual

stress, was discussed in Example 3.4 (page 82). For clarity, however, we
present herein a complete analysis from scratch. It is convenient to divide
the total deformation into two parts. First, the deformation F1 carries a
point from the cylindrical coordinates (R, Θ, Z) in B0 to (ρ, ϑ, ζ) in BU , and
then F2 takes it to (r, θ, z) in BL (Fig. 6.12). For axisymmetric deformation,
the position vectors to the point in B0 , BU , and BL , respectively, are
R = R eR + Z eZ
ρ = ρ eρ + ζ e ζ
r = r er + z ez . (6.95)
In terms of a set of Cartesian base vectors, the unit vectors for each coor-
dinate system are
eR = ex cos Θ + ey sin Θ
eΘ = −ex sin Θ + ey cos Θ
eZ = ez
eρ = ex cos ϑ + ey sin ϑ
eϑ = −ex sin ϑ + ey cos ϑ
eζ = ez
eθ = −ex sin θ + ey cos θ
ez = ez . (6.96)
The deformation F1 describes bending of a circular bar, and the map-
ping from B0 to BU is assumed to have the form
ρ = ρ(R), ϑ = πΘ/φ0 , ζ = ΛZ, (6.97)
where φ0 is the half-sector angle defined in Fig. 6.12, and Λ is the axial
stretch ratio of BU relative to B0 . Note that ϑ = 0 when Θ = 0, ϑ = π
when Θ = φ0 , and the opening angle is given by φ = 2(π − φ0 ). As in
Eqs. (3.57), combined extension, inflation, and torsion from BU to BL is
defined by the relations
r = r(ρ), θ = ϑ + ψζ, z = λζ, (6.98)
where ψ is the angle of twist per unit axial length of BU , and λ is the axial
stretch ratio in the loaded tube BL relative to the unloaded tube BU .
Equations (6.95) and (6.96) give the natural base vectors for the coor-
dinate systems in B0 and BU as
B0 : GR = R,R = eR BU : Gρ = ρ,ρ = eρ
GΘ = R,Θ = R eΘ Gϑ = ρ,ϑ = ρ eϑ
GZ = R,Z = eZ Gζ = ρ,ζ = eζ
(6.99)
G R = eR G ρ = eρ
GΘ = R−1 eΘ Gϑ = ρ−1 eϑ
G Z = eZ G ζ = eζ ,
where Eq. (2.18) is used to compute the contravariant base vectors. In
addition, with Eqs. (6.95)–(6.98), the convected base vectors are
B0 → BU : gR = ρ,R = ρ eρ
π
gΘ = ρ,Θ = ρ eϑ
φ0
gZ = ρ,Z = Λeζ
(6.100)
BU → BL : gρ = r,ρ = ṙer
gϑ = r,ϑ = reθ
gζ = r,ζ = ψreθ + λez ,
where prime and dot denote differentiation with respect to R and ρ, respec-
tively.
The deformation gradient tensors now can be computed by way of

Eq. (3.22)1, which yields
F1 = gR G R + gΘ G Θ + gZ G Z
πρ
= ρ eρ eR + eϑ eΘ + Λeζ eZ (6.101)
φ0 R
F2 gρ G ρ + gϑ G ϑ + gζ G ζ
=
r
= ṙer eρ + eθ eϑ + λez eζ + ψr eθ eζ . (6.102)
ρ
The total deformation from B0 to BL is described by (see Fig. 6.12 and
Example 3.5, page 86)
F = F2 · F1
= λR er eR + λΘ eθ eΘ + λZ ez eZ + γeθ eZ , (6.103)
where
λR = r
πr
λΘ =
φ0 R
λZ = λΛ
γ = ψrΛ. (6.104)
Here, we have used the relation ṙρ = (∂r/∂ρ)(∂ρ/∂R) = ∂r/∂R = r .

Note that the λs in these terms represent stretch ratios only if γ = 0. Also,
if there is no residual stress in BU , then B0 and BU are identical. In this
case, φ0 = π and Λ = 1, and Eq. (6.103) reduces to Eq. (3.62) of Example
3.4, as it should.
Finally, with (6.103) and I = GR GR + GΘ GΘ + GZ GZ = eR eR +
eΘ eΘ + eZ eZ , the Lagrangian strain tensor of BL relative to B0 is
1
E = 2 (F
T
· F − I)
= ÊRR eR eR + ÊΘΘ eΘ eΘ + ÊZZ eZ eZ
+ÊΘZ eΘ eZ + ÊZΘeZ eΘ , (6.105)
where

ÊRR = 1
2
λ2R − 1

ÊΘΘ = 1
2 λ2Θ − 1

ÊZZ = 1
2
λ2Z + γ 2 − 1
ÊΘZ = ÊZΘ = 12 γλΘ . (6.106)
As indicated by Eq. (3.42), the ÊIJ are physical components of E.
Constitutive Relations. With the artery assumed to be incompress-

ible and pseudoelastic, Eq. (6.4)1 with J = 1 gives the constitutive relation
∂W
σ = F· · FT − p I. (6.107)
∂E
As in the papillary muscle problem, we take advantage of simplifications
by writing the equilibrium equations in terms of stress components relative
to the basis associated with the coordinates (r, θ, z) = (x1 , x2 , x3 ) in BL .
Since these coordinates are orthogonal, Eq. (4.48) gives
σ = σ̂ ij ei ej , (6.108)
in which the σ̂ ij are physical stress components. Substituting Eqs. (6.103),

(6.105), and (6.108), along with I = er er + eθ eθ + ez ez , into (6.107) yields
σ̂ rr = σ̄ rr − p
σ̂ θθ = σ̄ θθ − p
σ̂ zz = σ̄ zz − p
∂W ∂W
σ̂ θz = λZ λΘ +γ
∂ ÊΘZ ∂ ÊZZ
∂W ∂W
σ̂ rθ = λR λΘ +γ
∂ ÊRΘ ∂ ÊZR
∂W
σ̂ zr = λR λZ , (6.109)
∂ ÊZR
where
∂W
σ̄ rr = λ2R
∂ ÊRR
∂W ∂W ∂W ∂W
σ̄ θθ = λ2Θ + γλΘ + + γ2
∂ ÊΘΘ ∂ ÊΘZ ∂ ÊZΘ ∂ ÊZZ
∂W
σ̄ zz = λ2Z , (6.110)
∂ ÊZZ
Note the coupling between normal stress and shear strain, as well as between
normal strain and shear stress, for a general anisotropic material.
Equilibrium. With end effects ignored, the symmetry in this problem

demands that stress must be independent of θ and z. In the absence of
inertial and body forces, Eqs. (4.101) reduce to
∂ σ̂ rr σ̂ rr − σ̂ θθ
+ = 0
∂r r
∂ 2 rθ
(r σ̂ ) = 0
∂r
∂
(rσ̂ rz ) = 0. (6.111)
∂r
Boundary Conditions. The inner wall of the artery is subjected

to an internal pressure P , the outer wall is free of loads, and the testing
machine exerts a resultant force N and twisting moment M at the ends.
With the exception of the pressure, the boundary conditions match those
of the papillary muscle problem.
At the deformed inner surface, the unit normal is n = −er = −e1 =
−e1 , and the true traction vector is
T(n) = n·σ
= −e1 · (σ̂ ij ei ej )
= −σ̂ ij δ 1i ej
= −σ̂ 1j ej .
This traction must match the applied traction P er = P e1 at r = a, i.e.,

−σ̂ 1j ej = −(σ̂ 11 e1 + σ̂ 12 e2 + σ̂ 13 e3 ) = P e1 . Adding this condition to those
given by slightly modified versions of Eqs. (6.75) and (6.77) yields
r = a: σ̂ rr = −P, σ̂ rθ = σ̂ rz = 0
r = b: σ̂ rr = σ̂ rθ = σ̂ rz = 0 (6.112)
b
2
πa P + N = 2π σ̂ zz r dr
a
b
M = 2π σ̂ zθ r 2 dr. (6.113)
a
The pressure term in Eq. (6.113)1 comes from the force exerted by P on
the supports at the closed ends of the artery (Humphrey, 2002). Note also
that the equations for N and M are derived from (6.77) using the relations
between tensor and physical stress components of Eq. (4.98).
6.5.3 Solution
Because the stresses depend only on r, Eqs. (6.111)2,3 can be integrated to
obtain
C1 C2
σ̂ rθ = 2 , σ̂ rz = , (6.114)
r r
where C1 and C2 are constants. Applying the boundary conditions (6.112)
gives C1 = C2 = 0, and thus σ̂ rθ = σ̂ rz = 0 everywhere. In the papillary
muscle problem, we used symmetry arguments to establish this result a
priori, and so the circumferential and axial equilibrium equations (6.111)2,3
were satisfied identically. Moreover, Eqs. (6.109) show that, in order for
these shear stresses to be zero, the artery must be composed of tissue with
a strain-energy density function that satisfies the conditions
∂W ∂W
= = 0. (6.115)
∂ ÊRΘ ∂ ÊZR
Otherwise, the stipulated deformation cannot be realized without applying
shear stresses to the curved surfaces of the tube.
Because the artery wall is assumed to be incompressible, the matrix
form of Eq. (6.103) gives the constraint
J = det F = λR λΘ λZ = 1.
Substituting (6.104) yields
π r dr
λΛ = 1,
φ0 R dR
which can be integrated to obtain
φ
r 2 = 0 (R2 − a20 ) + a2 , (6.116)
πλΛ
where the integration constant was found from the condition r(a0 ) = a.
If the opening angle φ, the axial stretch ratio λ, and the deformed inner
radius a are known, then the deformed outer radius can be computed using
the relation
1/2
φ0 2 2 2
b = r(b0 ) = (b − a0 ) + a . (6.117)
πλΛ 0
With r(R) now known, Eqs. (6.104) and (6.106) provide the deformation
gradient and strain components. Then, with W (ÊIJ ) specified, Eqs. (6.109)
and (6.110) give the stress components in terms of the yet unknown La-
grange multiplier p. To determine p, we substitute (6.109)1,2 into the radial
equilibrium equation (6.111)1 and integrate to get
b
θθ dr
rr
p(r) = σ̄ + σ̄ − σ̄ rr . (6.118)
r r
The upper limit on this integral is chosen to satisfy the boundary condition
σ̂ rr = σ̄ rr − p = 0 at r = b. With p(r) known, the internal pressure
required to maintain the deformation can be found using (6.118) and the
final boundary condition on the curved surfaces, σ̂ rr (a) = −P , to get
b
θθ dr
P = σ̄ − σ̄ rr . (6.119)
a r
Finally, the applied end loads M and N can be computed using Eqs. (6.113).
For a general form of W , the above integrals must be evaluated numer-
ically. Computing N , therefore, involves evaluating the above integral for
p within the integral in Eq. (6.113)1 . Although this procedure is relatively
straightforward, it may be more convenient to write the equation for N in a
form that does not involve p. This can be done by first writing Eq. (6.113)1
as
b
b
2 zz rr rr
πa P + N = 2π (σ̂ − σ̂ ) r dr + σ̂ r dr .
a a
Integrating the last integral by parts and using (6.109) and (6.111)1 yields
b 2 b b rr 2
r rr dσ̂ r
σ̂ rr r dr = σ̂ − dr
a 2 a a dr 2
b
b2 rr a2 rr σ̄ rr − σ̄ θθ r 2
= σ̂ (b) − σ̂ (a) + dr.
2 2 a r 2
Now, applying the boundary conditions σ̂ rr (a) = −P and σ̂ rr (b) = 0 and
substituting the result into the above relation for N yields
b
πa2 P + N = πa2 P + π 2σ̂ zz − σ̂ rr − σ̂ θθ r dr
a
b
= πa2 P + π 2σ̄ zz − σ̄ rr − σ̄ θθ r dr.
a
The integral in this expression does not contain p. Thus, Eq. (6.113)1 can
be replaced by
b
zz
N =π 2σ̄ − σ̄ rr − σ̄ θθ r dr. (6.120)
a
6.5.4 Geometric and Material Properties

The results presented here are based on the parameters used by Chuong
and Fung (1986) for the rabbit aorta. They assumed that the artery wall
is cylindrically orthotropic, with the strain-energy density function having

the form of Eq. (5.95), i.e.,
W = C(eQ − 1)
2 2 2
Q = a1 ÊRR + a2 ÊΘΘ + a3 ÊZZ

+2 a4 ÊRR ÊΘΘ + a5 ÊΘΘ ÊZZ + a6 ÊZZ ÊRR

2 2
+a8 ÊΘZ + ÊZΘ . (6.121)
Note that the ÊRΘ and ÊZR terms have been dropped in order to satisfy
the requirement (6.115). Using experimental data, the authors determined
the following values for the material constants:
C = 11.2 kPa
a1 = 0.0499 a4 = 0.0042 a8 = 0.100
(6.122)
a2 = 1.0672 a5 = 0.0903
a3 = 0.4775 a6 = 0.0585.
Because Chuong and Fung (1986) did not examine torsion, they actually set
a8 = 0. The value a8 = 0.1 is used here only for illustration. In addition,
their measurements of vessel geometry before and after a transmural cut
yielded the following approximate parameter values:
a0 = 3.9 mm
b0 = 4.5 mm
a = 1.4 mm (unloaded)
b = 2.0 mm (unloaded)
φ = 220◦ (φ0 = 70◦ )
Λ = 1. (6.123)
For the unloaded intact artery with λ = 1, these values satisfy Eq. (6.117)
approximately.
6.5.5 Residual Stress in Unloaded Artery

Deformation from the zero-stress configuration B0 to the unloaded intact
configuration BU (Fig. 6.12) produces residual stress in BU . To compute
this stress, we specialize our solution by setting N = M = P = 0. In
addition, we take F2 = I, which implies that BL is the same as BU with ρ =
r, ϑ = θ (ψ = 0), and ζ = z (λ = 1). Hence, if experimental measurements
provide a0 and φ0 , Eqs. (6.97) and (6.116) give the deformation field in
terms of the pair of unknowns a and Λ. These variables can be determined

by solving the integral equations (6.119) and (6.120) simultaneously with
N = P = 0. Numerically, this essentially amounts to solving two nonlinear
algebraic equations.
Fig. 6.13. Residual stress distributions in unloaded artery (physical components).
Notably, it is simpler to compute an approximate solution by assuming

that Λ = 1 and solving Eq. (6.119) for a. This procedure yields a = 1.33
mm, which is relatively close to the measured value of 1.4 mm (Chuong
and Fung, 1986), considering all the assumptions made.
Transmural distributions of residual stress in BU (Fig. 6.13) illustrate
several points. First, the circumferential stress σ̂ θθ is the largest stress com-
ponent, being compressive in the inner region and tensile in the outer region
of the wall. This distribution agrees with the expected stress distribution
for bending of a curved beam. Second, σ̂ rr vanishes at the inner and outer
radii, consistent with the boundary conditions for zero pressure. Third, σ̂ zz
is quite small compared to σ̂ θθ , justifying the plane strain approximation
Λ = 1 in place of the average plane stress condition N = 0.
Finally, it is important to note that, because the artery is unloaded,
residual stresses must be self-equilibrating. In other words, the (integrated)
resultant forces due to σ̂ θθ and σ̂ zz must be zero. The results in Fig. 6.13
satisfy this requirement.
6.5.6 Loaded Artery
The deformation of the loaded artery is determined completely by the de-

formed inner radius a, the twist ψ, and the axial stretch λ. Figure 6.14
exemplifies the nonlinear coupling between these deformation modes.
Fig. 6.14. Pressure, axial force, and torque versus deformed inner radius in artery. Note
that, due to nonlinear coupling effects, the inflation stiffness in (a) first decreases and
then increases as the artery is stretched axially.
Fig. 6.15. Stress and strain distributions (physical components) in loaded artery with
(φ = 220◦ ) and without (φ = 0) residual stress.
Pressure-radius curves for ψ = 0 show that the apparent artery stiffness

increases with λ (Fig. 6.14a). This behavior is due to the exponential form
for W , which characterizes a material that stiffens with increasing strain.
Adding twist (ψ = 0.3 rad/mm) essentially shifts the pressure-radius curve
upward, indicating that the pressure required to maintain a given radius
increases, but the stiffness (slope) changes relatively little. This curve also
shows that the vessel radius at zero pressure decreases with twist.
For ψ = 0 and λ held at a fixed value, the axial force N remains rela-
tively constant when circumferential stretch (r/R) is small, but, depending
on the value of λ, N either increases or decreases as the radius becomes
much larger than the unloaded radius of 1.4 mm (Fig. 6.14b). This complex
behavior is caused by coupling between axial and circumferential strain.
For example, pressure induces tension in both the axial and circumferential
directions, but circumferential stretch alone is accompanied by axial short-
ening. In addition, for ψ > 0 and λ fixed, the torque M increases with the
inner radius (Fig. 6.14c). The torque-radius curve for a8 = 0 shows that the
artery can sustain a twisting moment even if all shear strains are dropped
from W [see Eq. (6.121)]. The torque depends on the shear stress σ̂ θz ,
which is given by the coupling term involving ∂W/∂ ÊZZ in Eq. (6.109)4.
The global response curves of Fig. 6.14 are relatively unaffected by the
presence of residual stress. In contrast, even though residual stresses are
relatively small (see Fig. 6.13), local stress distributions undergo dramatic
changes in the loaded vessel (Fig. 6.15). In the absence of residual stress
(φ = 0), strong concentrations of circumferential and axial stress occur in
the inner layers of the wall (Fig. 6.15a), although transmural differences in
6.6. Passive Filling of the Left Ventricle 277
strain are relatively modest (Fig. 6.15b). This behavior is caused by the
strong material nonlinearity between stress and strain, i.e., relatively small
changes in strain can produce large changes in stress. The bending that
accompanies the deformation from B0 to BU (Fig. 6.12) lowers the strain at
the inner wall and raises it at the outer wall (Fig. 6.15b, φ = 220◦), causing
corresponding but relatively large changes in stress in the loaded artery
(Fig. 6.15a). According to these results, the presence of residual stress
and strain in arteries tends to homogenize transmural stress distributions
in the loaded vessel, thereby increasing the efficiency of the artery as a
load-bearing structure (Fung, 1991).
6.6 Passive Filling of the Left Ventricle
In the heart, the left ventricle (LV) pumps blood to the tissues and organs
of the body. Due to the relatively high hydraulic resistance of the sys-
temic vascular system, the LV experiences higher pressures than the other
three chambers and, therefore, does the most work. Being composed of car-
diac muscle (myocardium), the LV responds like other muscles to increased
workload, i.e., it grows. The wall of the LV, therefore, is thicker than the
walls of the atria and right ventricle, and it becomes even thicker in response
to hypertension (high blood pressure). Because more blood is needed to
supply this greater muscle volume, the LV is more prone to myocardial in-
farction than the other heart chambers. (Infarctions usually are caused by
blockage of a coronary artery feeding the myocardium.) It is believed that
wall stress plays a role in adaptation of the heart to altered loading, as well
as in pathological conditions including infarction. For more than a century,
therefore, researchers have proposed models to compute wall stress in the
heart (Woods, 1892; Mirsky et al., 1974; Yin, 1981; McCulloch et al., 1992;
Hunter et al., 2003).
6.6.1 Model for the Left Ventricle

This section considers a model for inflation of the passive LV. Passive filling
of the LV has received a great deal of attention, as diastolic dysfunction can
lead to severe pathological consequences. However, it must be remembered
that the highest wall stresses occur during systolic contraction, a topic that
is beyond the scope of this book.
To a first approximation, the human LV can be treated as an ellipsoidal
shell of revolution (Fig. 6.16a). Near the base of the LV, a circular tube pro-
vides an even simpler representation. Here, we consider a thick-walled tube

(Fig. 6.16b) that is fixed at its upper end (base) and free at its lower end
(apex). Both ends are assumed to be closed to maintain an internal blood
pressure P , but end effects and all other external loads are ignored. In
addition, like arteries, the LV contains residual stress, as unloaded ventric-
ular cross sections open when cut transmurally (Omens and Fung, 1990).
Hence, this LV model is similar to that for the artery, and we build on the
analysis in the previous section. Like the artery, the LV is assumed to have
the zero-stress, unloaded, and loaded configurations depicted in Fig. 6.12
(page 265).
Z
(a) (b) Q cut
base R
eC
fibers eF
apex b
Fig. 6.16. Models for left ventricle. (a) Ellipsoid of revolution. (b) Thick-walled cir-
cular cylinder in approximately zero-stress configuration showing fiber orientation at
epicardium, midwall, and endocardium.
Although the geometry has been simplified considerably, it is impor-

tant to include the characteristic muscle fiber architecture. The wall of the
LV is composed of highly ordered layers of muscle connected by a network
of collagen fibers (LeGrice et al., 1995, 1997). The muscle fibers in these
layers are organized into helices that wind around the lumen (Fig. 6.16).
Relative to the circumferential direction, the fiber helix angle β varies con-
tinuously across the wall from about 60◦ at the inner radius (a0 , endo-
cardium) to nearly circumferential at midwall to -60◦ at the outer radius
(b0 , epicardium) (see Fig. 6.16b). A rough approximation for the transmu-
ral distribution in the zero-stress configuration is given by the relation
n
2R − a0 − b0
β(R) = β 0 , (6.124)
a0 − b 0
where β 0 is the endocardial value of β, and n is an odd integer. Here, we

consider the cases n = 1 and n = 3 (Fig. 6.17), with the latter value being
more consistent with the classical measurements of Streeter et al. (1979)
for the dog LV.
Fig. 6.17. Transmural fiber angle distributions for model of left ventricle.
In summary, our model for the LV is a thick-walled circular tube with

closed ends and an internal pressure P (Fig. 6.16b). The wall is assumed
to be composed of pseudoelastic muscle tissue that is transversely isotropic
relative to the local fiber direction, which varies across the wall according
to Eq. (6.124).
6.6.2 Analysis
When subjected to internal pressure, the passive LV stretches in the longi-
tudinal and circumferential directions and, due to the helical fiber geometry,
twists about its axis of symmetry. Because this type of deformation was
studied for an artery in the previous section, we assume that the deforma-
tion of the LV model also is described by Eqs. (6.97) and (6.98). Therefore,
with the following modifications, the analysis of the artery also applies to
the LV.
First, because the apex is not constrained, we set the twisting moment
and axial force to zero, i.e., M = N = 0. With these specified loads,
Eqs. (6.113)2 and (6.120) provide two equations to be solved simultaneously
for ψ and λ. Moreover, if we specify a, as in the artery problem, then
Eq. (6.119) provides P . Otherwise, we could stipulate the value of P and
add (6.119) as a third simultaneous equation for the unknown a. In practice,
solving three nonlinear equations is not much more difficult than solving
two. Hence, we choose to specify the value of P , along with M = N = 0,
and solve for a, λ, and ψ.
Second, we need to modify the constitutive relations (6.109) to account
for the fiber geometry. For this purpose, it is convenient to express W in
terms of strain components relative to local Cartesian axes (R, F, C) in the
zero-stress (cut) configuration, where F and C represent the fiber and cross-
fiber directions, respectively. The unit vectors eF and eC are obtained by
rotating eΘ and eZ about eR through the helix angle β (Fig. 6.16b), giving
eF = eΘ cos β + eZ sin β
eC = −eΘ sin β + eZ cos β. (6.125)
Relative to this local material basis, the Lagrangian strain tensor is
E = ÊRR eR eR + ÊF F eF eF + ÊCC eC eC + ÊF C eF eC + ÊCF eC eF , (6.126)
and the strain-energy density function has the form
W = W (ÊRR , ÊF F , ÊCC , ÊF C , ÊCF ). (6.127)
Note that, as in the artery problem, the stipulated geometry demands that
ÊRΘ = ÊRZ = 0, which implies ÊRF = ÊRC = 0. Similarly, the artery
analysis shows that the Cauchy stress tensor for this problem can be written
σ = σ̂ rr er er + σ̂ θθ eθ eθ + σ̂ zz ez ez + σ̂ θz eθ ez + σ̂ zθ ez eθ (6.128)
relative to the deformed coordinates (r, θ, z). Here again, hat denotes phys-
ical components.
Substituting Eqs. (6.103) and (6.126)–(6.128), along with the relation
I = er er + eθ eθ + ez ez , into (6.107) provides the appropriate constitutive
relations. The manipulations are straightforward, but lengthy because of
the coordinate transformations involved. To make the computations more
manageable, we first note that Eq. (6.126) gives
∂W
= WRR eR eR + WF F eF eF + WCC eC eC + WF C eF eC + WCF eC eF ,
∂E
(6.129)
where
∂W
WIJ ≡ . (6.130)
∂ ÊIJ
Next, Eqs. (6.125) are used to write the above dyads in terms of eΘ and
eZ , i.e.,
eF eF = eΘ eΘ cos2 β + (eΘ eZ + eZ eΘ ) sin β cos β + eZ eZ sin2 β
eC eC = eΘ eΘ sin2 β − (eΘ eZ + eZ eΘ ) sin β cos β + eZ eZ cos2 β
eF eC = sin β cos β(eZ eZ − eΘ eΘ ) + eΘ eZ cos2 β − eZ eΘ sin2 β
eC eF = sin β cos β(eZ eZ − eΘ eΘ ) − eΘ eZ sin2 β + eZ eΘ cos2 β.
(6.131)
Inserting Eqs. (6.103), (6.129), and (6.131) into (6.107) reveals that all dot
products now involve the orthogonal system (eR , eΘ , eZ ), hence eliminating
these base vectors and leaving only (er , eθ , ez ), which is the basis used for
σ in (6.128). This observation allows us to express Eq. (6.107) in matrix
form without confusion.
In cylindrical coordinates, the matrix forms of Eqs. (6.103) and (6.128)
are
⎡ ⎤ ⎡ rr ⎤
λR 0 0 σ̂ 0 0
[F] = ⎣ 0 λΘ γ ⎦ , [σ] = ⎣ 0 σ̂ θθ σ̂ θz ⎦ ,
0 0 λZ 0 σ̂ zθ σ̂ zz
and putting (6.131) into (6.129) yields
⎡ ⎤
WRR 0 0
2 % %
⎢ c WF F + s2 WCC sc(WF F − WCC ) ⎥
∂W ⎢ 0 ⎥
=⎢⎢ −sc(WF C + WCF ) +c WF C − s WCF ) ⎥
2 2
∂E % % ⎥,
⎣ sc(WF F − WCC ) s2 WF F + c2 WCC ⎦
0 2 2
+c WF C − s WCF ) +sc(WF C + WCF )
(6.132)
where WF C = WCF and
c ≡ cos β, s ≡ sin β. (6.133)
Finally, in terms of these matrices, Eq. (6.107) becomes
[σ] = [σ̄] − p[I]

∂W
[σ̄] = [F] [F]T . (6.134)
∂E
For β = 0, we have eF = eΘ and eC = eZ , and these constitutive relations
reduce to Eqs. (6.109). (Recall that σ̂ rθ = σ̂ rz = 0.)
To compute the solution, we specify φ0 and take Λ = 1 as in the artery
problem. Then, with a, λ, and ψ to be determined, Eq. (6.116) gives r(R),
and (6.104) and (6.106) give the strains relative to the (R, Θ, Z) system.
Next, after the strains in the rotated (R, F, C) system are computed by
standard coordinate transformation, Eqs. (6.118) and (6.134)2 provide p(r),
then (6.134)1 yields the stress components relative to (r, θ, z). Given the
pressure P , the values of a, λ, and ψ are found by solving the system of
equations
b
zz
2σ̄ − σ̄ rr − σ̄ θθ r dr = 0
a
b
σ̂ zθ r 2 dr = 0
a
b dr
σ̄ θθ − σ̄ rr
= P, (6.135)
a r
as provided by (6.113)2 , (6.119), and (6.120) with M = N = 0.
It often is of interest to determine the load carried by the muscle fibers.
The fiber stress σ̂ ff can be computed using the relation
σ̂ ff = ef · σ · ef , (6.136)
where ef is the unit vector in the deformed fiber direction. This vector is
given by
F · eF
ef = , (6.137)
|F · eF |
in which Eqs. (6.103) and (6.125)1 provide F and eF , respectively.

Representative results are presented for the dog LV. For locally transversely
isotropic myocardium, the strain-energy density function is assumed to have
the form
W = C(eQ − 1)
Q = 2a1 (ÊRR + ÊF F + ÊCC ) + a2 ÊF2 F
+a3 (ÊF2 C + ÊCF
2
), (6.138)
which is consistent with Eq. (5.166). Considering the data of Guccione et
al. (1991), we use the following representative material constants:
C = 0.5 kPa
a1 = 2, a2 = 15, a3 = 10.
In addition, the results are based on following geometric parameters:
a0 = 1.7 cm, b0 = 2.9 cm
a = 1.4 cm, b = 2.6 cm (unloaded)
◦ ◦
φ = 50 (φ0 = 155 ), Λ = 1.
Fig. 6.18. Transmural distributions of physical stresses in model for left ventricle with
cubic fiber angle distribution and opening angle φ = 50◦ . (a) Stresses for P = 2.67 kPa.
(b) Residual stresses (P = 0).
6.6.4 Results
Stress distributions are shown for a cubic distribution of fiber angle (n = 3
and β 0 = 60◦ , see Fig. 6.17) at an inflation pressure P = 2.67 kPa = 20
mmHg (Fig. 6.18a). Unlike the results for the artery (Fig. 6.15a), the stress
components in the LV generally do not vary monotonically across the wall.
Rather, the peak fiber stress occurs in the subendocardial layers of the wall.
This behavior reflects the effects of the fiber architecture. For example,
near midwall, where the fibers are oriented essentially in the circumferen-
tial direction (Fig. 6.17), the fiber and circumferential stresses are nearly
identical. Near the endocardium and epicardium, however, longitudinal
stress contributes significantly to the fiber stress.
Exhibiting similar complexity, residual stresses are an order of mag-
nitude smaller than the stresses in the loaded ventricle (Fig. 6.18b). As
discussed in the artery problem, these stresses must be self-equilibrating.
This requirement provides a valuable check on the accuracy of our solution.
The effects of fiber angle distribution and residual strain on fiber stress
are illustrated in Fig. 6.19, where various distributions of the fiber angle β
are considered. Note the following:
1 For circumferential fibers (β = 0◦ ) and no residual stress (φ = 0◦ ),

a strong stress concentration in σ̂ ff = σ̂ θθ occurs near the endo-
cardium, similar to the behavior in an artery without residual stress
(see Fig. 6.15a).
2 Varying the fiber orientation across the wall decreases peak fiber stress
Fig. 6.19. Effects of fiber angle β and opening angle φ on transmural distributions of
physical fiber stress in model for left ventricle (P = 2.67 kPa).
by about half, regardless of the precise transmural distribution of β.

3 Because stresses are significantly reduced by fiber architecture alone
and because the opening angle (φ = 50◦ ) is smaller, residual strain has
considerably less effect on wall stress in the LV than it does in arteries
(see Fig. 6.15a).
Fig. 6.20. Effects of fiber geometry and residual stress on global behavior of model for left
ventricle. (a) Pressure-volume curves. (b) Axial stretch λ and twist ψ versus pressure.
The influence of residual stress and fiber angle distribution on global

variables is shown in Fig. 6.20. Pressure-volume curves are affected only
at relatively high pressures, and the axial stretch ratio λ changes only
6.7. Blastula with Internal Pressure 285
slightly. The twist ψ is affected more, especially by residual stress. With

increasing pressure, the LV twists in one direction (ψ > 0) and then reverses
course. Interestingly, when residual stress is included (φ = 50◦ ), the LV is
twisted even when P = 0. This behavior is caused by interaction between
fiber architecture and the redistribution of wall stresses required for self-
equilibrium (see Fig. 6.18b).
6.7 Blastula with Internal Pressure
Embryos undergo dramatic changes in form during development. In gen-

eral, the deformations that produce three-dimensional shapes are extremely
complex and require computational methods for analysis. There are, how-
ever, some relatively simple problems in morphogenesis that are amenable
to analytical solution. Here, we consider a problem that occurs in the early
embryo.
A fertilized egg initially undergoes a series of divisions that produces
a group of cells. Next, a fluid pocket forms to create a hollow ball of
cells called a blastula. The morphogenetic process of gastrulation then
rearranges the cells of the blastula into the primary germ layers, which later
form specialized tissues and organs (Gilbert, 2010). During gastrulation, a
local region of the blastula wall bends inward, moving cells from the outer
layers to the interior of the blastula, while creating the primitive gut (see
Problem 6.14). These events are crucial for proper development of the
embryo.
Considering a wide range of experimental observations, Beloussov (1998;
2015) has suggested that the fluid in the blastula exerts an outward pressure
on the cells, which then respond to the increased stress by moving and
deforming in ways that tend to restore stresses back toward their initial
levels. He speculates that this response partially drives gastrulation. This
section examines wall stress in the sea urchin blastula, which is essentially
a thick-walled spherical shell filled with fluid (Fig. 6.21).
For simplicity, we ignore residual stress and assume that the cells in the
blastula wall are isotropic and incompressible. Symmetry demands that the
shell maintains a spherical shape as it deforms under an internal pressure
P . Let a0 and b0 be the undeformed inner and outer radii, respectively,
and a and b the corresponding deformed radii.

Kinematics. Let (R, Θ, Φ) and (r, θ, φ) represent spherical polar co-
ordinates in the reference (unloaded, zero-stress) configuration and the de-
z
f
a
P
Fig. 6.21. Model for sea urchin blastula.
formed configuration, respectively. In terms of Cartesian base vectors, the

unit base vectors for these coordinate systems are (see Appendix B)
eR = ex sin Θ cos Φ + ey sin Θ sin Φ + ez cos Θ
eΘ = ex cos Θ cos Φ + ey cos Θ sin Φ − ez sin Θ
eΦ = −ex sin Φ + ey cos Φ
er = ex sin θ cos φ + ey sin θ sin φ + ez cos θ

eθ = ex cos θ cos φ + ey cos θ sin φ − ez sin θ
eφ = −ex sin φ + ey cos φ. (6.139)
For spherically symmetric deformation, the position vectors before and
after deformation are, respectively,
R = R eR (Θ, Φ), r = r er (θ, φ), (6.140)
with
r = r(R), θ = Θ, φ = Φ. (6.141)
Hence, eR = er , eΘ = eθ , and eΦ = eφ for this problem. The corresponding
natural base vectors in the reference configuration are
GR = R,R = eR G R = eR
GΘ = R,Θ = R eR ,Θ = R eΘ GΘ = R−1 eΘ (6.142)
−1
GΦ = R,Φ = ReR ,Φ = R sin Θ eΦ GΦ = (R sin Θ) eΦ ,
and the convected base vectors are

gR = r,R = r er gR = (r )−1 er
gΘ = r,Θ = r eθ gΘ = r −1 eθ (6.143)
−1
gΦ = r,Φ = r sin θ eφ gΦ = (r sin θ) eφ,
where prime denotes differentiation with respect to R. (Note that these
base vectors are mutually orthogonal.) With these relations, Eq. (3.22)1
gives the deformation gradient tensor
F = gI G I = gR G R + gΘ G Θ + gΦ G Φ
= λr er eR + λθ eθ eΘ + λφ eφ eΦ , (6.144)
where
r
λr = r , .
λθ = λφ = (6.145)
R
Because the expression for F does not contain shear terms, these compo-
nents represent stretch ratios.
Constitutive Relations. For spherically symmetric inflation, the

Cauchy stress tensor takes the form
σ = σ̂ r er er + σ̂ θ eθ eθ + σ̂ φ eφ eφ (6.146)
in terms of physical components. With W = W (λr , λθ , λφ ), Eq. (6.8)1
provides the constitutive relations
∂W
σ̂ r = λr −p
∂λr
∂W
σ̂ θ = λθ −p
∂λθ
∂W
σ̂ φ = λφ −p (6.147)
∂λφ
for incompressible material (J = 1).
Equilibrium. Spherical symmetry tells us that σ̂ θ = σ̂ φ and λθ =

λφ , but we keep these quantities distinct for now. The equation of radial
equilibrium is given by
∂ σ̂ r 1
+ (2σ̂ r − σ̂ θ − σ̂ φ ) = 0 (6.148)
∂r r
in spherical coordinates (see Appendix B). Equilibrium in the other coor-
dinate directions is satisfied identically.
Boundary Conditions. For an internal pressure P , the boundary

conditions are
r = a: σ̂r = −P
r = b: σ̂r = 0. (6.149)
6.7.2 Solution
With Eqs. (6.145), enforcing incompressibility yields
r 2 dr
J = det F = λr λθ λφ = = 1. (6.150)
R2 dR
Integrating this relation and using the condition r(a0 ) = a gives
r 3 = R3 − a30 + a3 . (6.151)
Hence, if the deformed inner radius a is known, the outer radius is
1/3
b = r(b0 ) = b30 − a30 + a3 . (6.152)
The Lagrange multiplier is determined by inserting (6.147) into (6.148)
and integrating. This procedure yields
b
∂W ∂W ∂W ∂W dr
p(r) = λr + λθ + λφ −2λr , (6.153)
∂λr r ∂λθ ∂λ φ ∂λr r
which satisfies the boundary condition σ̂ r (b) = 0. Substituting this ex-
pression into Eq. (6.147)1 and using the boundary condition σ̂ r (a) = −P
gives
b
∂W ∂W ∂W dr
P = λθ + λφ −2λr . (6.154)
a ∂λθ ∂λ φ ∂λr r
If the deformed inner radius a is specified, then Eqs. (6.145) and (6.151)
provide the stretch ratios as a function of r(R), and the above equations give
p(r) and P . Then, the stress components can be computed from (6.147).
If the shell wall is “thin,” the blastula can be treated as a pressur-
ized membrane. In this case, a0 /h0 >> 1 with h0 = b0 − a0 being the
undeformed wall thickness, and the above integrals can be simplified by
assuming that stress and strain vary relatively little across the wall. As the
deformed wall thickness h = b −a approaches zero, Eqs. (6.153) and (6.154)
become
∂W
p ∼ = λr
∂λr
h ∂W ∂W ∂W
P ∼ = λθ + λφ −2λr . (6.155)
a ∂λθ ∂λφ ∂λr
Hence, with Eqs. (6.147), the pressure can be written

h
P = (σ̂ θ + σ̂ φ ) ,
a
and using symmetry gives
Pa
σ̂ θ = σ̂ φ = . (6.156)
2h
This equation, which also can be derived from equilibrium considerations
alone, is known as the Law of Laplace for a pressurized spherical mem-
brane.
Note further that Eqs. (6.147)1 and (6.155)1 give σ̂ r ∼
= 0 for a membrane.
Actually, the radial stress is not zero; rather it is small compared to the
other stress components. We can see this by considering the boundary
conditions (6.149), which indicate that the value of σ̂ r varies from −P at
r = a to 0 at r = b, i.e., σ̂ r = O(P ). Hence, for a thin membrane with
a/h >> 1, Eq. (6.156) shows that σ̂ θ = σ̂ φ >> σ̂ r . Typically, a shell
is considered thin when the undeformed radius-to-thickness ratio a0 /h0 is
greater than about 20. For large deformation, however, an incompressible
shell becomes thinner as it inflates, i.e., a/h grows larger. Thus, even if
the undeformed shell is relatively thick, the membrane approximation may
become valid as its radius increases.

We assume that the strain-energy density function for the blastula wall has
the form
C α(I1 −3)
W = e −1 , (6.157)
α
where C and α are material constants and
I1 = λ2r + λ2θ + λ2φ
is a strain invariant. In the limit of small strain (I1 → 3), it can be shown
that C = EY /6, where EY is the small-strain elastic (Young’s) modulus.
From small-strain measurements for the sea urchin blastula (Davidson et
al., 1999), we take the following parameter values:
C = 0.2 kPa
a0 = 50 μm
b0 = 60 μm.
Because the nonlinear properties apparently are unknown, we examine var-
ious values for α. In addition, it is instructive to study the effects of the
radius-to-thickness ratio a0 /h0 .
Fig. 6.22. Pressure-radius curves for blastula model. (a) Effects of material constant α.
(b) Effects of ratio of undeformed inner radius to wall thickness. Results for a membrane
approximation also are shown.
Fig. 6.23. Computed distribution of circumferential stress across blastula wall for a =
2a0 . A stress concentration develops as the material constant α increases.
6.7.4 Results
Pressure-radius curves for three values of α are shown in Fig. 6.22a. As
α increases, the blastula stiffens considerably at large radii. For α → 0,
Eq. (6.157) reduces to the form for neo-Hookean material, i.e., W → C(I1 −
3). For this case, the pressure reaches a peak near a/a0 = 1.3, indicating a
limit-point instability. If pressure is prescribed for α = 0, then the radius
would become infinite, i.e., the blastula would burst, when the pressure is
greater than about 0.1 kPa. In contrast, the curves for larger values of α
6.8. Bending of the Embryonic Heart 291
remain stable, which should benefit the embryo. This type of analysis also
has applications in studies of aneurysm rupture (Humphrey, 2002).
For α = 0.2, Fig. 6.22b illustrates the effects of the relative wall thick-
ness. Results are shown for the “exact” thick-walled theory (6.154) and for
thin-walled membrane theory (6.155). As expected, the shell stiffens as the
shell becomes thicker (a0 /h0 decreases). Also as expected, the accuracy of
membrane theory increases dramatically for thinner shells.
As in the artery without residual stress (Fig. 6.15a, page 276), circum-
ferential stress decreases from the inner to the outer radius (Fig. 6.23). The
results shown correspond to a deformation a = 2a0 .6
As the material becomes more nonlinear (increasing α), a strong stress
concentration develops in the inner layers of the wall. In early work in car-
diac mechanics, (Mirsky, 1973) modeled the left ventricle as a thick-walled
isotropic spherical shell and found an extremely large stress concentration
near the endocardium. As Fung (1991, 1997) later pointed out, however,
these results do not make much sense physiologically, because the inner
layers of the ventricular wall would need to perform much more work than
the outer layers. Such a design could significantly decrease the efficiency
of the heart as a pump. This observation led in part to Fung’s emphasis
on the importance of residual stress (Fung, 1991), which can reduce stress
concentrations dramatically in the cardiovascular system (Fig. 6.15a).
6.8 Bending of the Embryonic Heart
The heart is the first functioning organ in the embryo. During the early
stages of development, a pair of membranes, on the left and right sides
of the embryo, fold and fuse at the midline to create a single, relatively
straight tube. The primitive heart tube is composed of an outer layer of
myocardium, a middle layer of extracellular matrix (cardiac jelly), and an
inner layer of endocardium. Soon thereafter, the first contractions occur
and the morphogenetic process of cardiac looping begins, as the heart
tube bends and twists into a curved tube to lay out the basic plan of the
future four-chambered pump (Manner, 2000; Taber, 2001, 2006).
Studies suggest that the forces that drive looping arise from sources both
intrinsic and extrinsic to the heart. During the later stages of looping, for
example, the body of the embryo compresses the ends of the heart tube,
6 This exceptionally large deformation is used here only for illustration. It is not likely
that blastulae actually deform this much due to internal pressure. During gastrulation,
however, regional strains may exceed 100%.
causing it to bend (Ramasubramanian et al., 2013, 2019; Bayraktar and

Manner, 2014). This section examines a relatively simple model for bending
of the embryonic heart by end loads.
(a) (b)
Y y
b r T
a1 a2 X r1 r2 x
b
undeformed deformed
Fig. 6.24. Beam model for bending of the embryonic heart.
Bending a thick-walled tube is similar to bending a beam. As a first ap-

proximation, therefore, the heart is modeled as a rectangular “beam” that
bends into a circular sector. Furthermore, ignoring the layered structure of
the heart, we treat the beam as being composed entirely of homogeneous
myocardium.
It is convenient here to introduce two coordinate systems (Fig. 6.24).
The undeformed geometry is described in terms of Cartesian coordinates
(X, Y, Z), while the deformed geometry is referred to cylindrical polar co-
ordinates (r, θ, z). Initially, the beam is bounded by the planes X = a1 , a2 ;
Y = ±b; and Z = ±c (Fig. 6.24a). After bending, the beam is assumed to
be symmetric about the z-axis, with the planes X = a1 , a2 becoming the cir-
cular cylindrical surfaces r = r1 , r2 (Fig. 6.24b). In this simple “flattened”
representation of the heart tube, the X- and Y -directions correspond to the
circumferential and longitudinal directions, respectively, in the undeformed
heart.

Kinematics. The undeformed and deformed position vectors to a
point in the beam are
R = XeX + Y eY + ZeZ
r = rer + zez (6.158)
where the unit base vectors are related by

er = eX cos θ + eY sin θ
eθ = −eX sin θ + eY cos θ
ez = eZ . (6.159)
For simplicity, this analysis does not allow for anticlastic curvature,
i.e., the heterogeneous deformation that occurs in the z-direction during
bending. Rather, tractions are applied on the sides of the beam to maintain
a uniform stretch in the z-direction. Hence, deformation into a circular
beam is defined by the transformation
r = r(X), θ = kY, z = λZ, (6.160)
in which k and λ are constants. These relations give
∂ ∂ ∂
∇r = eX + eY + eZ [r(X)er (θ(Y )) + z(Z)ez ]
∂X ∂Y ∂Z
∂r ∂er ∂z
= eX er + eY r + eZ ez .
∂X ∂Y ∂Z
Substituting
∂er ∂er ∂θ
= = keθ
∂Y ∂θ ∂Y
yields the deformation gradient tensor
F = (∇r)T = λr er eX + λθ eθ eY + λz ez eZ , (6.161)
where
∂r
λr = , λθ = kr, λz = λ (6.162)
∂X
are stretch ratios.
Constitutive Relations. We assume that the heart is composed of

incompressible material with principal material directions that align with
the (X, Y, Z) axes in the undeformed configuration. For pure bending,
there should be no shear stress relative to the (r, θ, z) coordinates, and so
the Cauchy stress tensor has the form
σ = σ̂r er er + σ̂ θ eθ eθ + σ̂ z ez ez . (6.163)
Equations (6.8)1 with J = 1 gives the constitutive relations
∂W
σ̂ r = λr −p
∂λr
∂W
σ̂ θ = λθ −p
∂λθ
∂W
σ̂ z = λz − p. (6.164)
∂λz
Equilibrium. Relative to the cylindrical coordinates in the de-

formed configuration, the only nontrivial equilibrium equation, given by
Eq. (6.111)1, is
∂ σ̂ r σ̂ r − σ̂ θ
+ = 0. (6.165)
∂r r
Boundary Conditions. In the present model, the embryo applies

loads only at the ends of the heart tube (in addition to the tractions needed
at z = ±λc). Hence, the curved surfaces are assumed to be traction free
with boundary conditions
σ̂r (r1 ) = 0, σ̂ r (r2 ) = 0. (6.166)
Because of symmetry, we need only consider boundary conditions at one
end of the beam. At θ = kb (Y = b), Eq. (6.163) gives the true traction
vector
T(eθ ) = eθ · σ = σ̂ θ eθ . (6.167)
Hence, the resultant force N and moment M applied to the ends are given
by the integrals
λc r2
N = T(eθ ) dr dz
−λc r1
λc r2
M = r × T(eθ ) dr dz, (6.168)
−λc r1
where r is the position vector from the origin.7 We also could compute
a force resultant due to σ̂ z on the z-faces of the beam, but this is not of
interest here. Substituting Eqs. (6.158)2 and (6.167) and integrating over
z yields
r2
N = 2λc eθ σ̂ θ dr ≡ N eθ
r
1r2
M = 2λc ez r σ̂ θ dr ≡ M ez , (6.169)
r1
with N and M being the normal force and bending moment, respectively.
Note, however, that global equilibrium for the entire deformed beam de-
mands that N = 0. This observation provides a check on the accuracy of
any numerical solution.
7 Since bending moments are couples, the location of the reference point does not matter.
The origin is chosen here for convenience.

6.8.2 Solution
With (6.162), the incompressibility condition is
∂r
det F = λr λθ λz = kλr = 1,
∂X
which, upon integration, yields
r(X) = [(2/kλ)(X − A)]1/2 , (6.170)
where A is a constant. Applying the conditions r(a1 ) = r1 and r(a2 ) = r2

(see Fig. 6.24) gives
2 (a2 − a1 )
kλ =
r22 − r12
a1 r22 − a2 r12
A = . (6.171)
r22 − r12
With these relations, Eqs. (6.162) provide the stretch ratios in terms of the
variables r1 , r2 , and λ.
To solve the governing equations, we first substitute Eqs. (6.164) into
(6.165) and integrate to get
r2
∂W ∂W ∂W dr
p(r) = λr + λθ − λr , (6.172)
∂λr r ∂λ θ ∂λr r
where the limits on the integral have been chosen to satisfy the boundary
condition σ̂ r (r2 ) = 0. With this expression for p, Eq. (6.164)1 and the
boundary condition σ̂ r (r1 ) = 0 yield
r2
∂W ∂W dr
λθ − λr = 0, (6.173)
r1 ∂λ θ ∂λr r
which is to be solved for r2 if r1 and λ are specified. With the stretch ratios
now known in terms of these variables, Eqs. (6.164) and (6.172) give the
stress components, and Eqs. (6.169) provide N (which should be zero) and
M.

Experimental measurements have shown that the material properties of the
embryonic chick heart are relatively linear and isotropic during the early
looping stages (Zamir and Taber, 2004). However, the heart tube contains
circumferentially oriented actin filaments (Itasaki et al., 1989; Shiraishi
et al., 1992), suggesting that the myocardium later becomes transversely

isotropic. Therefore, we take the strain-energy density function in the form
W = C1 (I1 − 3 + βI42 ), (6.174)
where C1 and β are constants and
I1 = λ2r + λ2θ + λ2z
I4 = Êf = 12 (λ2f − 1), (6.175)
in which Êf and λf are the physical Lagrangian strain and stretch ra-
tio, respectively, in the fiber direction. Note that I4 is consistent with
Eqs. (5.167)1 and (6.6)2 .
The material constants, based primarily on estimates from pressure-
volume curves for somewhat older embryonic chick hearts (Taber et al.,
1992; Lin and Taber, 1994), are taken as C1 = 20 Pa and β = 2. In
addition, we take λ = 1 and a2 − a1 = 2c = 0.2 mm. (The circular
cross section is represented by a square.) For illustration, we consider
three cases: (1) isotropic (β = 0); (2) transversely isotropic relative to the
circumferential direction (λf = λr ); and (3) transversely isotropic relative
to the longitudinal direction (λf = λθ ).
VT
Fig. 6.25. Beam model for looping of the embryonic heart. (a) Bending moment versus
curvature κ. The linear solution for an isotropic beam is shown for comparison. (b)
Bending stress distributions.
6.8.4 Results
In Fig. 6.25a the bending moment M is plotted as a function of the average
beam curvature
2
κ= .
r1 + r2
Also shown is the bending moment given by the linear theory for cylindrical
bending of an isotropic plate.8 For a plate, the bending moment is (Szilard,
1974)
EY Iκ
(M )lin = .
1 − ν2
In the present problem with β = 0, Young’s modulus is given by EY = 6C1 ,

Poisson’s ratio is ν = 0.5 (incompressible material), and the cross-sectional
area moment of inertia is I = (a2 − a1 )3 (2c)/12. The linear theory also
gives the bending stress (Szilard, 1974)
EY κ(r − r0 )
(σ̂ θ )lin = ,
1 − ν2
where r0 = (a1 + a2 )/2 is the location of the middle surface of the plate.
For the isotropic beam, the linear and nonlinear solutions for M agree
until κ 1.5 mm−1 (Fig. 6.25a). For larger bending, however, the magni-
tude of M drops below the linear value.
As expected, including fibers increases M , with longitudinal fibers hav-
ing the greater effect. For this reason, Nakamura et al. (1980) speculated
that the circumferential fiber alignment observed in the heart tube facil-
itates looping by offering less resistance to bending than fibers of other
orientations.
Distributions of bending stress (σ̂ θ ) are shown for κ ≈ 5 (Fig. 6.25b).
The linear and nonlinear solutions for the isotropic beam are similar, but
note the shift in the location of the neutral axis, as defined by the point
where σ̂ θ = 0. In contrast, the transversely isotropic beams contain rela-
tively strong stress concentrations — at the inner curvature for circumfer-
ential fibers and at the outer curvature for longitudinal fibers.
The possible implications of these results for cardiac looping are not
clear. Studies have shown that mechanical stress plays a role in morpho-
genesis (Beloussov, 1998; 2015; Taber, 2020), but the relative simplicity of
this model must be kept in mind when interpreting these results. Never-
theless, this problem illustrates how models can suggest avenues for further
experimental investigation.
8 In cylindrical bending of a plate, the length of the plate in the z-direction constrains
the deformation so that λz ∼ = 1, consistent with our beam analysis. Thus, a plate solution
is more appropriate than a beam solution for comparison purposes.
6.9 Problems
Here again, computation and symbolic manipulation software may be useful

for some problems. Unless stated otherwise, units are ignored.
6.1 A rectangular bar is stretched by forces applied only at its ends. The
bar is composed of an isotropic compressible material with the strain-
energy density function
−1/3
W = (μ/2)(I1 + 3I3 − 6),
where the Ii are strain invariants and μ is the shear modulus in the
small-strain limit. Determine the axial Cauchy stress as a function of
the stretch ratio in the axial direction.
6.2 Consider a rectangular bar composed of anisotropic incompressible ma-
terial containing fibers embedded in an isotropic matrix. With the
fibers aligned in both transverse directions (x and y), the strain-energy
density function is given by
W = C(I1 − 3) + 21 D1 (λ2x − 1)2 + 12 D2 (λ2y − 1)2 ,
where I1 is the first strain invariant, and C, D1 , and D2 are constants.
Suppose the bar is stretched in the axial (z) direction by the specified
stretch ratio λz . Taking C = D2 = 1, plot on one graph λx and λy
versus λz (0.5 ≤ λz ≤ 4) for D1 = 0.5, 1, and 2. Hint: The symmetry
condition λx = λy holds only for the case D1 = D2 .
6.3 Consider planar deformation of an incompressible rectangular mem-
brane with undeformed edges parallel to the Cartesian axes X1 and
X2 . The membrane is composed of isotropic Ogden material with
N
an b n
W = (Λ1 + Λb2n + Λb3n − 3),
n=1
b n
where an and bn are constants, and the Λi are stretch ratios in the
principal directions of strain (and stress), with Λ3 corresponding to the
transverse (thickness) direction.
(a) Compute the principal Cauchy stresses σ i in terms of Λ1 and Λ2 .
(b) Suppose the deformation is described by the mapping
x1 = λ1 X1 + kλ2 X2
x2 = λ2 X2
x3 = λ3 X3 ,
6.9. Problems 299
where λ1 , λ2 , and k are constants. Take N = 1 in W , along with

a1 = 1, b1 = 3, λ1 = 2, λ2 = 1, and k = 0.5. Determine the Cauchy
stress components σ ij relative to the global Cartesian system.
Hint: It may be useful to write the stress tensor in the spec-
'3
tral form σ = i=1 σ i ni ni , where the unit vectors ni define the
principal stress directions in the deformed configuration.
6.4 Consider again a rectangular membrane undergoing the same deforma-
tion defined in Problem 6.3. Here, however, the membrane is compress-
ible with

μ 1 −β
W = I1 − 3 + I −1 ,
2 β 3
where the Ii are strain invariants and
ν
β= .
1 − 2ν
In the following, ei and gI represent Cartesian and convected base
vectors, respectively.
(a) For F = Fij ei ej and t = tij ei ej , write the first Piola-Kirchhoff
stresses tij in terms of the deformation gradients Fij . Then, as-
suming only the edges of the membrane are loaded, eliminate F33
from the resulting expression.
∗ ∗
(b) Determine the convected components σ I J of the Cauchy stress
∗ ∗
tensor defined by σ = σ I J gI gJ . Check your results by computing
these stresses for the case of simple shear (λ1 = λ2 = λ3 = 1) and
comparing with those found using Eqs. (6.40). Hint: First, plug in
the value of I3 for simple shear.
∗ ∗ ∗ ∗ ∗ ∗
(c) For μ = 2, λ2 = 2, and k = 0.5, plot σ 1 1 , σ 2 2 , and σ 1 2 vs. λ1
for 0.5 ≤ λ1 ≤ 2. Compare results for ν = 0, 0.25, and 0.4999 by
plotting all curves on the same graph. Note: The last value for ν
represents an approximation for the incompressible case (ν → 0.5).
6.5 Prior to deformation, a unit cube is composed of compressible material
consisting of an isotropic matrix with embedded fibers oriented parallel
to the X1 -axis. Assume the strain-energy density function has the form
μ 1/2
W = (I2 /I3 + 2I3 − 5) + cI44 + dI5 ,
2
in which the strain invariants are given by Eqs. (3.136) and (5.103)
with suitable modifications. Applied loads deform the cube with the
mapping
x1 = λ1 X1
1 3
x2 = 3 kX1 + λ2 X2
x3 = X3 ,
where Xi and xi are Cartesian coordinates, and λ1 , λ2 , and k are
positive constants.
(a) Sketch the deformed shape of the block in the x1 x2 -plane.
(b) Determine the stretch ratio for a line segment located at the center
of the cube before deformation that is parallel to the x1 -axis after
deformation.
(c) Take μ = 1, λ1 = 2, λ2 = 1.5, and k = 1. For the cases of transverse
isotropy (c = 2, d = 4) and isotropy (c = d = 0), compute the non-
zero normal and tangential Cauchy stresses acting on all surfaces,
as well as any body forces, that are needed to hold the block in
equilibrium with the specified deformation. State the values of the
surface stresses on the left and right sides of the block, as well as
on the faces normal to the X3 -axis. Plot the stresses applied to the
upper surface and any body forces as functions of X1 .
6.6 Using relations presented in Sec. 6.4, derive Eqs. (6.86) and (6.87).
6.7 Ignoring residual stress, write a computer program to solve the problem
involving a cylindrical model for the passive left ventricle discussed
in Sec. 6.6. For a baseline model, use the material properties and
parameter values given in that section; for a0 and b0 , use the unloaded
values given for a and b, respectively.
(a) Write the steps of the solution procedure, including all necessary
equations. The equations can be written in tensor or matrix form,
with reference to equations in the text for specific details. Compute
the solution for internal pressures 0 ≤ P ≤ PED , where PED = 2.67
kPa (20 mm Hg) is end-diastolic pressure.
(b) Check your calculations against the results in Figs. 6.19 and 6.20a
for φ = 0◦ (no residual stress).
(c) Examine the effects of varying the distribution of fiber angle β,
defined by (6.124), as well as the fiber stiffness by changing the
material constant a2 in (6.138). Let r be the deformed radial co-
ordinate, σ̂ ff the fiber stress (the physical Cauchy stress in the
deformed fiber direction), ψ the twist, and V0 and V the cavity
volume before and after deformation, respectively. Plot σ̂ ff vs. r
at P = PED , ψ vs. P , and P vs. V /V0 for the following cases:
6.9. Problems 301
• β 0 = 30◦, 60◦, and 90◦ with n = 3

• β 0 = 0◦, 30◦, 60◦, and 90◦ with n = 0 (constant β)
• a2 = 7.5, 15, and 30 with n = 3 and β 0 = 60◦
Briefly discuss and explain the results for each case.
6.8 A circular elastic tube is fixed to a rigid circular cylinder at its inner
surface and to a rigid wall at its outer surface. The undeformed tube
has inner radius a, outer radius b, and length L. It is composed of
incompressible material with W = C1 (I1 − 3) + C2 (I2 − 3). An axial
force P pulls the inner cylinder downward, shearing the tube (Fig. 6.26).
Assume the deformation is defined by the relations
r = r(R), θ = Θ, z = Z + w(R)
in cylindrical coordinates, where w(R) is the vertical displacement in
the tube.
Rigid cylinder Z z
4 T
Elastic tube R r
a
b P
Undeformed Deformed
Fig. 6.26. Axial shear of a tube (Problem 6.8). Modified from Taber (2020).
(a) Find r(R).

(b) Assume the Cauchy stress tensor stress has the form
σ = σ rr er er + σ θθ eθ eθ + σ zz ez ez + σ rz er ez + σ zr ez er ,
where the stress components depend only on r. Determine the
nontrivial equations of equilibrium.
(c) Given P , find σ rz (r) and w(r).
(d) Determine σ rr (r). Hint: In this problem, boundary conditions at
r = a, b have already been used and are not available for finding
the Lagrange multiplier p. One possibility is to use radial equi-

librium to find p(r) up to an unknown constant, then determine
the constant by assuming the resultant axial force Nz vanishes in
the tube. Note that the assumed deformation requires that a self-
equilibrating axial stress σ z (r) be applied at the ends of the tube.
(e) Take C1 = 1, C2 = 0.2, a = 1, b = 2, L = 5, and P = 80. Plot w
versus r, as well as σ rz , σ rr , and σ zz versus r. Explain the sign of
σ rr .
6.9 A solid incompressible circular cylinder (no hole) rotates with constant
angular velocity ω about its axis of symmetry (the Z-axis). The un-
deformed cylinder is composed of two axisymmetric regions: region 1
(0 ≤ R < a0 ) and region 2 (a0 < R ≤ b0 ). The regions have the
same mass density ρ but potentially different material properties as
characterized by the neo-Hookean strain-energy density function
W (k) = C (k)(λ2r + λ2θ + λ2z − 3),
where the λi are stretch ratios and C (k) is the modulus of region k. In
cylindrical coordinates, assume the deformation is described by
r = r(R), θ = Θ, z = λZ,
where λ is a constant.
(a) Determine r(R) and write the stretch ratios in terms of λ.
(b) A differential element located a distance r from the axis of rotation
experiences a centripetal (inward) acceleration rω2 . If the curved
surface is free of external loads, determine the Cauchy stresses σ i
(i = r, θ, z) as functions of r and λ.
(c) An exact solution requires applied loads on the upper and lower
(flat) surfaces consistent with σ z as found in part (b). Assuming the
resultant axial force is zero, determine λ as a function of ω. Discuss
what happens as ω becomes large, as well as the implications of this
finding. Hint: Consider inflation of a spherical shell composed of
neo-Hookean material (Sec. 6.7).
6.10 Consider a spherical shell composed of two regions: an inner layer
(a0 ≤ R ≤ c0 ) and an outer layer (c0 ≤ R ≤ b0 ). In the undeformed
configuration, R is the radial coordinate, a0 and b0 represent the inner
and outer radii of the shell, respectively, and c0 is the radius of the
boundary between the layers. Both layers are composed of isotropic
incompressible material with
C α(I1−3)
W = e −1 ,
α
6.9. Problems 303
where I1 is the first strain invariant. An internal pressure P inflates

the shell to a specified inner radius a.
(a) Assume the values of C and α are constant in each layer but can
differ between layers. Modify the analysis in Sec. 6.7 as needed and
write a computer program to solve for P and the stress distribution
across the wall. Hint: Define a grid for R and divide the grid and
integrals into two parts.
(b) Using the parameter values given in Sec. 6.7, check that your pro-
gram can reproduce the results for a one-layer model as shown in
Fig. 6.22a. For this case, the solution should not depend on the
value of c0 .
(c) Consider the following parameter values: (a0 , b0 , c0 ) = (1.0, 1.2, 1.1)
mm; α = 0.2 (both layers); Cinner = 0.2 kPa; Couter = 0.2, 0.4, and
0.6 kPa. For each value of Couter , plot on the same graph the
physical Cauchy stress σ̂ θθ vs. R/a0 for the case a = 4a0 .
(d) Your program also should be able to handle the case where the
material coefficients vary continuously across the wall, i.e., C =
C(R) and α = α(R), rather than being constant in each layer.
For a = 4a0 , determine approximate distributions of C and α that
produce relatively uniform stress σ̂ θθ across the wall. Try varying
C for constant α and varying α for constant C. Use trial and error
or any other method you prefer. Note: There may be multiple
solutions.
6.11 Consider a compressible spherical shell composed of Blatz-Ko material
with W given by Eq. (5.115). For inflation caused by an internal pres-
sure load, determine a single differential equation to be solved for the
deformed radial coordinate r(R).
Answer:
4
d2 r dr dr
3R2 r 3 2
− 2Rr 3 + 2R4 = 0.
dR dR dR
6.12 Consider a spherical membrane with undeformed radius a0 and wall
thickness h0 subjected to uniform internal pressure P . Section 6.7
shows how the solution for inflation of a thick-walled spherical shell
reduces to Laplace’s law (6.156) for a membrane as the ratio h0 /a0
becomes small compared to unity. Here, the problem is to be solved
by assuming, at the outset, that the shell is a membrane composed of
isotropic incompressible material with a general strain-energy density
function W (I1 , I2 ).
(a) Using Laplace’s law and the approximation σ r ∼

= 0, show that the
pressure is given by
4h0 1 ∂W ∂W
P = 1− 6 + λ2 ,
λa0 λ ∂I 1 ∂I2
where the circumferential stretch ratio is λ = λθ = λφ by symme-
try.
(b) Plot P̄ = P a0 /Ch0 vs. λ (1 ≤ λ ≤ 5) for the following cases:
(i) Mooney-Rivlin material:
W = C[I1 − 3 + α(I2 − 3)]
On the same graph, plot curves for α = 0, 0.05, 0.1, 0.2. This form
for W could represent soft tissues in the early embryo, many of
which contain relatively little collagen.
(ii) Exponential material:
C α(I1 −3)
W = e
α
On the same graph, plot curves for α = 0, 0.02, 0.04, 0.06. This
form for W could represent a first approximation for collagenous
soft tissues in the adult. In general, these tissues stiffen exponen-
tially with strain, but they usually are not isotropic.
6.13 The previous problem examined an incompressible spherical membrane
with undeformed radius a0 and wall thickness h0 that is loaded by an
internal pressure P . Consider again the same problem, but now the
membrane is composed of Blatz-Ko material with

1 −β
W = C I1 − 3 + I3 − 1 ,
β
where
ν
β= .
1 − 2ν
(a) Using the approximations σ r = 0 and Laplace’s law (6.156), write
P as a function of the stretch ratio λ = λθ = λφ .
Answer:
4C h0 1
P= 1−
λ a0 λm
In this relation, m is a constant involving ν.
(b) On a single graph, plot P̄ = P a0 /Ch0 vs. λ (1 ≤ λ ≤ 5) for ν = 0,
0.2, 0.4, and 0.5.
6.9. Problems 305
6.14 As discussed in Sec. 6.7, the cells in the early sea urchin embryo create
a spherical fluid-filled shell called the blastula. Invagination (inward
bending) soon generates a circular dimple in the shell, beginning the
process of gastrulation, which creates the primitive gut (Gilbert, 2010).
Here, we examine some of the mechanics involved in this problem.
Neglecting internal fluid pressure, consider a model for half a blastula
consisting of an unloaded hemispherical shell with inner radius a0 and
outer radius b0 that is composed of isotropic, incompressible material.
To simulate invagination, the shell is everted, i.e., it is turned inside
out. (A fully everted hemispherical shell generally stays that way when
unloaded.) In spherical coordinates, assume the everted configuration
is described by the mapping (Fig. 6.27).
f
F
q
Q r
R
Fig. 6.27. Eversion of a hemispherical shell (Problem 6.14).
r = r(R)
θ = π−Θ
φ = Φ.
(a) Working with physical components, determine the deformation gra-
dient tensor F.
(b) Derive the relation
r = (A − R3 )1/3 ,
where A is a constant. Determine the radii of the inner and outer
surfaces of the shell in the everted configuration as functions of A,
and discuss any constraints on the value of A.
(c) Suppose the strain-energy density function is
c α(I1 −3)
W = e −1 ,
α
where c and α are constants. Derive an equation to solve for A.
(d) Write a computer program to solve for the value of A. Then,

compute and plot σ r and σ θ versus the deformed radial coordinate
r. Neglecting units, plot curves for c = 1, α = 2, a0 = 1, and
b0 = 1.1, 1.2, and 1.3.
(e) Discuss the mechanics behind the computed stress distributions for
σ θ . Is the assumed deformation consistent with the everted shell
being free of all external loads?
6.15 A sector of a circular cylindrical tube with inner radius R1 , outer radius
R2 , and length L (in the Z-direction) is straightened into a rectangular
block (Fig. 6.28). The section is composed of compressible material
characterized by the strain-energy density function
μ 1/2
W = (I2 /I3 + 2I3 − 5).
2
With the undeformed and deformed configurations described in cylin-
drical coordinates (R, Θ, Z) and Cartesian coordinates (x, y, z), respec-
tively, the deformation is defined by
x = x(R), y = kΘ, z = λZ,
where k and λ are positive constants.
R
4
x1 x2 x
R1 R2
undeformed deformed
Fig. 6.28. Straightening of a cylindrical sector (Problem 6.15).
In the deformed configuration, the x-faces are located at x = x1 , x2,

the y-faces at y = y1 , y2 , and the z-faces at z = z1 , z2 . Assume the
x-faces are stress-free, while the resultant normal forces Ny and Nz are
zero on the y- and z-faces, respectively, i.e., the boundary conditions
are
σ x (x1 ) = σ x (x2 ) = 0
Ny (y1 ) = Ny (y2 ) = 0
Nz (z1 ) = Nz (z2 ) = 0.
6.9. Problems 307
(a) Determine the deformation gradient tensor.

(b) Write the non-zero Cartesian components of the Cauchy stress ten-
sor in terms of stretch ratios.
(c) Considering equilibrium, determine x(R).
(d) Derive two algebraic equations to solve for k and λ (do not solve).
Partial answer:
3 1
4/3 4/3
1/3
R2 − R1 − 3 (R42 − R41 ) = 0
4(kλ) 4k λ
Solving the two equations gives
1 2
k2 = 2
2 λ (R1 + R22 )
16(R22 − R21 )3
λ10 = 4/3 4/3
.
27(R21 + R22 )(R2 − R1 )3
(e) For R1 = 1, R2 = 2, L = 2, and μ = 1, compute the bending
moment applied on the y-faces of the block. Then, compute and
plot σ y and σ z versus x. In a separate graph, plot λx , λy , and J
versus x.
Appendix A
Linear Theory of Elasticity
The linear theory of elasticity deals with problems in which deformations,

displacements, and rotations are “small.” In such problems, the undeformed
and deformed configurations are virtually identical, the governing equations
assume relatively simple forms, and the principle of superposition applies.
In biomechanics, the linear theory usually is accurate only in the analysis of
bone and other hard tissue. However, this theory sometimes may be useful
in studying qualitative trends in problems involving large deformation of
soft tissue, if strains are no more than about 20%.
To fully appreciate the complexities of nonlinear elasticity theory, it is
useful to be familiar with the linear theory. This appendix is intended as a
review and an aid to those readers who have only a modest familiarity with
linear elasticity theory. (Some familiarity is assumed, however.) Because
some readers may find it helpful to read this material before delving into
the more involved presentation in the main text, the following development
is relatively self-contained. The order of presentation also is similar, and
it may be instructive to compare the development of the nonlinear theory
with that in this appendix.
An important restriction in this appendix is the almost exclusive use
of rectangular Cartesian coordinates. This allows us to focus on funda-
mental concepts without getting bogged down in the added complexity of
curvilinear coordinates.
A.1 Mathematical Preliminaries
The presentation begins with a brief discussion of the mathematical back-

ground required in this appendix. First, we introduce a shorthand notation
that is used throughout the book. According to the summation conven-
309
310 Linear Theory of Elasticity
tion, when an index is repeated in a term of an expression, summation over

1, 2, 3 for that index is automatically implied (unless stated otherwise or
the subscripts are placed in parentheses).9 Thus, we write
3

ai b i = ai b i ,
i=1
where the repeated index i is called a “dummy” index, since it can be

replaced by any letter without altering the meaning, i.e., ai bi = aj bj . Sum-
mation is not implied, however, for subscripts surrounded by parentheses,
e.g., a(i) b(i) . Furthermore, an index that is not repeated implies a set of
equations. For example,
3

ai bij = ai bij
i=1
gives a separate equation for each j = 1, 2, 3. Note also that the indices
in an equation must “balance” in that all terms must contain the same
number of non-dummy indices. Thus, ai + bij cj = 0 is a valid equation, but
aik + bij cj = 0 is not since the first term contains an extra k. Moreover,
three or more of the same index in a single term is not allowed (unless
some of the indices are placed in parentheses). The following development
assumes an elementary knowledge of vector analysis.
x3
a x3
e3 x2
e3 e2
e2 Rotation
x2
e1 e1
x1
x1
Fig. A.1. Cartesian coordinate systems and base vectors.
9 The summation convention introduced here is modified slightly from that used in
Chapter 2. The difference is that here, because the focus is on Cartesian coordinates,
repetition of subscripts in a term is allowed.
A.1. Mathematical Preliminaries 311
A.1.1 Base Vectors

Consider a set of Cartesian coordinate axes (x1 , x2 , x3 ) fixed in Euclidean
space, and let (e1 , e2 , e3 ) represent unit vectors directed along these axes
(Fig. A.1). Since any vector can be expressed in terms of these unit vectors,
the ei are called base vectors. Being mutually orthogonal, they satisfy
the conditions
ei · ej = δ ij
ei × ej = ijk ek , (A.1)
where

1 for i = j
δ ij = (A.2)
0 for i = j
is the Kronecker delta and
⎧
⎨ +1 if (i, j, k) is an even permutation
ijk = −1 if (i, j, k) is an odd permutation
⎩ (A.3)
0 if two or more indices are equal
is the permutation symbol. For example, Eqs. (A.1)2 and (A.3) give
e2 × e1 = 21k ek = 211 e1 + 212e2 + 213e3 = −e3 .
The permutation symbol can be expressed directly in terms of the ei by
dotting both sides of Eq. (A.1)2 by el to get
el · (ei × ej ) = el · ( ijk ek ) = ijk el · ek = ijk δ lk ,
where Eqs. (A.1) have been used. For a given l, the only term on the
right-hand-side that survives the summation over k is the term for k = l,
i.e.,
ijk δ lk = ij1 δ l1 + ij2δ l2 + ij3 δ l3 = ijl .
In other words, we can replace i by j or j by i in terms containing δ ij and

remove the Kronecker delta. This operation, which can be used generally to
simplify expressions, is called tensor contraction. Thus, after renaming
the subscripts in the above expressions, we obtain
ijk = ei · (ej × ek ). (A.4)

The Kronecker delta and the permutation symbol are related through the
-δ identity
ijk irs = δ jr δ ks − δ js δ kr . (A.5)
A.1.2 Vectors, Dyadics, and Tensors

A vector a has a magnitude and direction. Relative to the coordinate
system xi , we can write
a = ai e i , (A.6)
where the ai are the components of a in the ei directions. It is important to
note that if the coordinate system changes (see the x̄i system in Fig. A.1),
the components of a change, but the vector a does not. A vector, therefore,
is geometrically invariant, and the form of a vector equation is independent
of the coordinate system.
There are three types of vector product. In terms of components, the
dot product of two vectors a = ai ei and b = bi ei is
a · b = (ai ei ) · (bj ej ) = ai bj (ei · ej ) = ai bj δ ij
or
a · b = ai b i . (A.7)
Note that the dummy indices in b have been changed from i to j, since more
than two of any subscript in a term is meaningless. The cross product is
a × b = (ai ei ) × (bj ej ) = ai bj (ei × ej )
or
a × b = ai b j ijk ek (A.8)
by Eq. (A.1)2 . Finally, the tensor product is defined by

ab = (ai ei )(bj ej )
or
ab = ai bj ei ej . (A.9)
The tensor product of two vectors is called a dyad, and a linear combination
of dyads, e.g., the right-hand-side of Eq. (A.9), is called a dyadic. A useful
operation involving dyads is the double-dot (scalar) product, which is

defined by
ab : cd = (a · c)(b · d). (A.10)
The left-hand-sides of Eqs. (A.7)–(A.9) are written in direct nota-

tion, valid for any coordinate system. The right-hand-sides are written in
indicial notation, or component form, here being specific to Cartesian
coordinates. Direct notation promotes physical understanding, but indicial
notation often makes manipulations easier. Throughout this book, there-
fore, we emphasize the importance of being able to move back and forth
between the two approaches.
At this point, we note several things. First, the vector dot, cross, and
tensor products yield a scalar, a vector, and a dyad, respectively. Second,
the dot product is commutative, but the cross and tensor products are not.
Thus, a·b = b·a, while a×b = −b×a, and all we can say about the tensor
product is that ab = ba in general. Third, just as the ai are the components
of the vector a with respect to the basis {ei }, ai bj are the components of
the dyad ab with respect to the basis {ei ej } [see Eq. (A.9)]. Since a dyad
is composed of vectors, the components of ab change if the basis changes,
but the dyad itself does not. Finally, as long as care is taken to preserve
the order of the vectors, operations with dyadics are straightforward. For
example, (ab) · c = a(b · c) = (b · c)a because b · c is a scalar.
Next, we introduce the concept of a tensor. A second-order tensor
T is defined as a linear vector function that, when dotted with a vector,
transforms the vector into another vector, i.e., T · a = b transforms a into
b.10 A dyad satisfies this requirement; since (ab) · c = (b · c)a, the dyad ab
transforms the vector c into the vector (b · c)a. Thus, a dyad is a tensor.
In fact, any second-order tensor can be expressed as a dyadic, i.e.,
T = Tij ei ej , (A.11)
where the Tij are the components of T with respect to the basis {ei ej }.
Higher order tensors can be written analogously. Like a vector or dyad, a
tensor is geometrically invariant; when the basis changes, the components
of T change, but T itself does not. (A vector is a tensor of the first order.)
The double-dot product of two tensors yields a scalar. Consider, for ex-
ample, the scalar product of A = Aij ei ej and B = Bij ei ej . By Eq. (A.10),
10 In this section, lower-case bold letters denote first-order tensors (vectors), and upper-
case bold letters denote second-order tensors (dyadics).

we have
A:B = (Aij ei ej ) : (Bkl ek el )
= Aij Bkl (ei · ek )(ej · el )
= Aij Bkl δ ik δ jl
and contraction yields
A : B = Aij Bij . (A.12)
For computational purposes, it often is convenient to write a second-

order tensor in the form of a matrix, i.e.,
⎡ ⎤
T11 T12 T13
T = [Tij ] = ⎣ T21 T22 T23 ⎦ , (A.13)
T31 T32 T33
where the position (i, j) in the matrix corresponds to the base dyad ei ej .
From this representation, it follows that the transpose of a tensor is obtained
by switching the order of the base vectors to get
TT = Tij ej ei = Tji ei ej . (A.14)
Moreover, we can show that
T · a = a · TT (A.15)
by the manipulations
(Tij ei ej ) · (ak ek ) = (ak ek ) · (Tij ej ei )
Tij ak ei δ jk = ak Tij δ kj ei
Tij aj ei = Tij aj ei .
A tensor T is symmetric if T = TT , i.e., Tij = Tji .
Finally, we note that the identity tensor I satisfies the relations a· I =
I · a = a and T · I = I · T = T. In dyadic form, the representation
I = δ ij ei ej = ei ei (A.16)
can be verified, for example, by the manipulations

a·I = (ai ei ) · (ej ej ) = ai (ei · ej )ej
= ai δ ij ej = ai ei = a.
Eq. (A.16) shows that the Kronecker delta δ ij represents the component of
I with respect to the Cartesian dyad ei ej .
A.1.3 Coordinate Transformation

As mentioned earlier, the components of vectors and tensors change when
the base vectors, i.e., the coordinates, change. For Cartesian coordinates,
rotating the coordinate axes produces a new Cartesian coordinate system.
Consider rotation of the coordinate system xi with base vectors ei into
the system x̄i with base vectors ēi (Figs. A.1 and A.2). Any vector can be
x2
x2
a2
a
a2 x1
e2 a1
q e1
e2
q
e1 a1 x1
Fig. A.2. Components of a vector a relative to two Cartesian coordinate systems, xi and
x̄i , with base vectors ei and ēi.
written in terms of components relative to either basis. Writing the vector

ēi in terms of the ei , for example, yields
ēi = Aij ej , (A.17)
where the transformation components Aij are the components of ēi
relative to the basis {ei }. Dotting this equation with ek gives
ēi · ek = Aij ej · ek = Aij δ jk = Aik
or
Aij = ēi · ej , (A.18)
which indicates that the Aij represent the direction cosines between the
two sets of coordinate axes. For the geometry of Fig. A.2,
ē1 = e1 cos θ + e2 sin θ
ē2 = −e1 sin θ + e2 cos θ
ē3 = e3 , (A.19)
and so
⎡ ⎤
cos θ sin θ 0
[Aij ] = [ēi · ej ] = ⎣ − sin θ cos θ 0 ⎦ . (A.20)
0 0 1
With these relations, transforming vector and tensor components is
straightforward. First, however, we note a convenient method for extracting
the components of vectors and tensors. Dotting Eq. (A.6) with ej gives
a · ej = (ai ei ) · ej = ai δ ij = aj ,
while double-dotting Eq. (A.11) with ek el yields
T : (ek el ) = (Tij ei ej ) : (ek el )
= Tij (ei · ek )(ej · el )
= Tij δ ik δ jl = Tkl .
where Eq. (A.10) has been used. Thus, we have
ai = a · e i
Tij = T : ei ej = ei · T · ej . (A.21)
In other words, dotting the vector a with ei gives the component of a

along ei , and double-dotting the dyadic T with ei ej gives the component
of T along ei ej .11 The components with respect to any other basis can be
extracted similarly.
To relate the components between two bases, we use the invariance
properties and write (Figs. A.1 and A.2)
a = ai ei = āi ēi
T = Tij ei ej = T̄ij ēi ēj . (A.22)
Then, the components of a and T relative to the x̄i system are
āi = a · ēi = (aj ej ) · ēi = aj (ej · ēi )
T̄ij = T : ēi ēj = (Tkl ek el ) : (ēi ēj ) = Tkl (ek · ēi )(el · ēj ), (A.23)
and using Eq. (A.18) gives
āi = Aij aj
T̄ij = Aik Ajl Tkl . (A.24)
According to these relations, an Aij is needed to transform each subscript

to the new basis.
11 In Cartesian coordinates, the components of a vector are the orthogonal projections
of the vector on the coordinate axes (Fig. A.2). A physical interpretation for the com-
ponents of a tensor cannot be visualized as easily.
A.1.4 Vector and Tensor Calculus

In Cartesian coordinates, the gradient (“del”) operator ∇ is defined as
∂
∇ ≡ ei . (A.25)
∂xi
Then, the gradient of a scalar function φ(xi ) is

∂
∇φ = ei φ
∂xi
or
∇φ = ei φ,i (A.26)
where φ,i ≡ ∂φ/∂xi . Moreover, the gradient of a vector function a(xi ) is

∂
∇a = ei (aj ej ) = ei (aj ej ),i
∂xi
or
∇a = aj ,i ei ej (A.27)
since the ei are constant. Similarly, the divergence of a(xi ) is

∇ · a = ei · (aj ej ),i = aj ,i ei · ej = aj ,i δ ij
or
∇ · a = ai , i . (A.28)
In addition, the curl of a is
∇ × a = ei × (aj ej ),i = aj ,i ei × ej ,
and using Eq. (A.1)2 gives
∇ × a = aj , i ijk ek . (A.29)
Finally, we list some relations that are useful in transforming volume

integrals into surface integrals and vice versa. In these equations, which
are derived in Section 2.8.4, n is a unit normal to the surface area A, which
encloses the volume V .
Gradient Theorem:

∇φ dV = nφ dA or φ,i dV = ni φ dA (A.30)
V A V A
Divergence Theorem:

∇ · a dV = n · a dA or ai ,i dV = ni ai dA (A.31)
V A V A
Curl Theorem:

∇ × a dV = n × a dA or aj , i ijk dV = n i aj ijk dA
V A V A
(A.32)
Similar relations between surface and line integrals can be obtained by
replacing V by A and A by the contour C (length element ds) in these
equations.
A.2 Analysis of Deformation
When subjected to forces, a body transforms from one configuration to

another. It is convenient to characterize this transformation relative to a
reference configuration, which is the collective position of all particles
of the body at any specified instant of time. A general mapping from one
configuration to another consists of rigid-body motion and deformation.
Rigid-body motion involves translation and rotation, while deformation
involves extensional and shear strains.
A.2.1 Strain
Consider a body that deforms from the undeformed (reference) configura-
tion B into the deformed configuration b (Fig. A.3a). An arbitrary point
P in the body B is located by the position vector
R = x i ei , (A.33)
where the xi are the Cartesian coordinates of P . In the body b, the position
vector to p, the deformed image of P , is
r = R + u, (A.34)
A.2. Analysis of Deformation 319
where u is the displacement vector from P to p. For a one-to-one mapping

of B into b, we can write
r = r(R, t), u = u(R, t), (A.35)
with t being time.
Fig. A.3. (a) Geometry of deformation. (b) Differential position and displacement
vectors.
Displacement Gradient Tensor. The state of strain at a point in b is

defined in terms of the deformation of the differential length element dR
at P into the element dr at p (Fig. A.3a). Equation (A.34) relates these
elements by
dr = dR + du. (A.36)
To express du in a different form, we write u = ui ei , and then Eqs. (A.27)
and (A.33) and the chain rule yield
dR · (∇u) = (dxi ei ) · (uk ,j ej ek )
= dxi uk ,j δ ij ek = dxi uk ,i ek
∂uk
= dxi ek = duk ek .
∂xi
Thus,
du = dR · (∇u) = (∇u)T · dR, (A.37)
where Eq. (A.15) has been used. The dyad ∇u is called the displacement
gradient tensor. Now, inserting this expression into Eq. (A.36) gives,
since IT = I,
dr = dR · (I + ∇u)
= [I + (∇u)T ] · dR. (A.38)
This relation shows that ∇u characterizes the deformation of dR into dr.
Extensional Strain. To characterize the stretching of dR as it deforms

into dr, we must compare absolute lengths. Thus, we write
dR = N dS, dr = n ds, (A.39)
where dS and ds are the lengths of dR and dr, which point in the directions
of the unit vectors N and n, respectively (Fig. A.3a). In large-deformation
problems, it is convenient to define strain in terms of the difference in
squared lengths. Thus, the Lagrangian extensional strain of dR is defined
by
1 ds2 − dS 2 1 dr · dr − dR · dR
E(NN) = = , (A.40)
2 dS 2 2 dS 2
in which Eqs. (A.39) have been used, and substituting Eq. (A.38) yields
1 dR dR
E(NN) = · (I + ∇u) · (I + (∇u)T ) − I · .
2 dS dS
Expanding the term in brackets and using Eq. (A.39)1 gives
E(NN) = N · E · N, (A.41)
where
E = 12 [∇u + (∇u)T + (∇u) · (∇u)T ] (A.42)
is the Lagrangian strain tensor. Equations (A.21)2 and (A.41) show

that the extensional strain E(NN) is the component of the tensor E along
the dyad NN. In the following, we show that shear deformation also can
be computed from E.
Equation (A.42) is the direct form of the strain-displacement re-
lation, valid for arbitrarily large deformation in any coordinate system.
Writing these relations for a specific coordinate system, e.g., Cartesian or
polar, requires simply expressing ∇ in the appropriate form. As shown
later, the strain-displacement relation for the linear theory can be obtained
by linearizing (A.42).
Shear Strain. Thus far, we have shown that the extensional strain for
a line element oriented originally in the direction of N can be computed
from E. Since N is arbitrary, the extensional strain in any direction at a
point in the body can be computed from this strain tensor. A complete
description of deformation, however, also requires the shear strains, which
correspond to angle (shape) changes. Next, we show that E also contains
this information.
Consider two orthogonal line elements at P described by (Fig. A.4)
dR1 = N1 dS1 , dR2 = N2 dS2 , (A.43)
where dS1 and dS2 are the undeformed lengths of the elements oriented in
the directions of the orthogonal unit vectors N1 and N2 in the undeformed
body. According to Eq. (A.38), these elements deform into
dr1 = n1 ds1 = dR1 · (I + ∇u) = [I + (∇u)T ] · dR1
dr2 = n2 ds2 = dR2 · (I + ∇u) = [I + (∇u)T ] · dR2 (A.44)
in the directions of the unit vectors n1 and n2 , which are not orthogonal
in general (Fig. A.4).
N2
B
dR2 b n2
P p
dR1 N1 a dr2
x3 dr1
R
n1
e3 r
e2
x2
e1
x1
Fig. A.4. Geometry for two originally orthogonal line elements deforming into elements
with an enclosed angle α.
Now, if α is the angle between dr1 and dr2 , then

dr1 · dr2
cos α = n1 · n2 =
ds1 ds2
dR1 · [(I + ∇u) · (I + (∇u)T )] · dR2
=
ds1 ds2
and, since dR1 · dR2 = 0, expanding the term in brackets leads to

2N1 · E · N2
cos α = ,
ds1 ds2
dS1 dS2
where Eqs. (A.42) and (A.43) have been substituted. Finally, Eq. (A.40)
gives ds1 /dS1 = (1 + 2E(N1N1 ) )1/2 and ds2 /dS2 = (1 + 2E(N2N2 ) )1/2 , lead-
ing to the expression
cos α = 2N1 · E · N2
1/2 1/2 , (A.45)
1 + 2E(N1 N1 ) 1 + 2E(N2N2 )
where E(N1 N1 ) = N1 · E · N1 and E(N2N2 ) = N2 · E · N2 by (A.41).

Thus, given E, the angle α between the deformed images of N1 and
N2 can be computed from Eq. (A.45), and π/2 − α gives a measure of the
shear. The numerator of (A.45) is the component of E along N1 N2 , but
the denominator shows that α also depends on the extensional strains along
N1 and N2 .
In summary, we have shown that the Lagrangian strain tensor E con-
tains all the information necessary to compute the extensional and shear
strains for a differential element of any orientation. In general, the strains
vary with location in the body, with the components of E (Eij ) at a given
point defining the state of strain at that point. Note that, although the
Eij themselves do not have direct physical interpretations, they are called
Lagrangian normal strains for i = j and shear strains for i = j.
Example A.1 The nonlinear strain-displacement relations for any coordi-

nate system can be computed from Eq. (A.42). Derive the expressions for
the Cartesian components of the Lagrangian strain tensor.
Solution. For u = ui ei , Eq. (A.27) gives
∇u = uj ,i ei ej , (∇u)T = ui ,j ei ej
and so
(∇u) · (∇u)T = (uj ,i ei ej ) · (uk ,l ek el )
= uj ,i uk ,l (ej · ek )ei el
= uj ,i uk ,l δ jk ei el
= u k , i u k , l ei el
= u k , i u k , j ei ej .
Substituting these expressions into (A.42) yields
E = 12 (ui ,j +uj ,i +uk ,i uk ,j )ei ej . (A.46)
Thus, with E = Eij ei ej ,, the Cartesian strain components are given by
Eij = 12 (ui ,j +uj ,i +uk ,i uk ,j ). (A.47)
Note that Eij = Eji, and so the strain tensor is symmetric, i.e., only six of
the nine strain components are independent.
Given the Eij at a point, the deformation can be computed for elements
of any orientation. For example, the extensional strain for a fiber passing
through a point that initially is parallel to the x1 -axis is obtained by setting
N = e1 in Eq. (A.41) to get E(e1 e1 ) = e1 · E · e1 = E11 , and Eq. (A.47)
yields

2 2 2
∂u1 1 ∂u1 ∂u2 ∂u3
E11 = + + + .
∂x1 2 ∂x1 ∂x1 ∂x1
In addition, the deformed angle between fibers that are parallel to the x1
and x2 axes in the undeformed body is obtained by setting N1 = e1 and
N2 = e2 in Eq. (A.45). This procedure gives
2E12
cos α = 1/2
,
(1 + 2E11 ) (1 + 2E22 )1/2
where Eq. (A.47) provides
1 ∂u1 ∂u2 ∂u1 ∂u1 ∂u2 ∂u2 ∂u3 ∂u3
E12 = + + + + .
2 ∂x2 ∂x1 ∂x1 ∂x2 ∂x1 ∂x2 ∂x1 ∂x2
Similar relations can be obtained for the other strain components.
Example A.2 Consider the set of coordinate axes x̄i , which are obtained
by rotating the Cartesian axes xi by an angle θ about the x3 -axis (Fig. A.2,
page 315). Write the strain components Ēij relative to the x̄i system in
terms of the components Eij relative to the xi system.
Solution. One way to do this is to use Eq. (A.24)2 to write the tensor
transformation
Ēij = Aik Ajl Ekl (A.48)
with the Aij provided by (A.20). For illustration, rather than simply sum
over the repeated indices, we write Eq. (A.48) in matrix form. This requires
some care, however, since matrix multiplication is not a commutative op-

eration. To obtain the correct equation, we note that all components in a
matrix equation must be referred to the same basis. Thus, we write
E = Eij ei ej = Ēij ēi ēj
Ē = Ēij ei ej
A = Aij ei ej , (A.49)
where E is the strain tensor expressed in terms of both sets of compo-
nents, while Ē is merely a tensor defined for computational convenience.
Substituting these relations into the equation
Ē = A · E · AT (A.50)
yields
Ēij ei ej = (Aij ei ej ) · (Ekl ek el ) · (Amn en em )
= Aij Amn Ekl (ej · ek )(el · en )ei em
= Aij Amn Ekl δ jk δ ln ei em
= Aik Aml Ekl ei em
= Aik Ajl Ekl ei ej .
Because this expression agrees with (A.48), Eq. (A.50) shows that the ap-
propriate matrix transformation relation is
[Ēij ] = [Aik ][Ekl][Ajl ]T . (A.51)
Substituting Eq. (A.20) now gives
⎡ ⎤ ⎡ ⎤⎡ ⎤
Ē11 Ē12 Ē13 cos θ sin θ 0 E11 E12 E13
⎣ Ē21 Ē22 Ē23 ⎦ = ⎣ − sin θ cos θ 0 ⎦ ⎣ E21 E22 E23 ⎦
Ē31 Ē32 Ē33 0 0 1 E31 E32 E33
⎡ ⎤
cos θ − sin θ 0
× ⎣ sin θ cos θ 0 ⎦, (A.52)
0 0 1
and carrying out the matrix multiplication yields
Ē11 = E11 cos2 θ + E22 sin2 θ + 2E12 sin θ cos θ
Ē22 = E11 sin 2 θ + E22 cos2 θ − 2E12 sin θ cos θ
Ē33 = E33
Ē12 = (E22 − E11 ) sin θ cos θ + E12(cos2 θ − sin 2 θ)
Ē23 = E23 cos θ − E31 sin θ
Ē31 = E23 sin θ + E31 cos θ. (A.53)

Linear Strain Tensor. Equations (A.41), (A.42), and (A.45) are exact
expressions, valid for arbitrarily large deformation. If the strains (displace-
ment gradients) are small, then the nonlinear term of Eq. (A.42), i.e., the
last term in brackets, can be neglected to obtain the linear strain tensor
E∗ = 12 [∇u + (∇u)T ]. (A.54)
∗
Moreover, if we define the linear shear strain as E(N1 N2 )
= 21 (π/2−α) << 1,

then cos α = cos π/2 − 2E ∗ = sin 2E ∗ ∼ 2E ∗
= . Thus, for
(N1N2 ) (N1 N2 ) (N1 N2 )
∗ ∗
E(N1 N1 )
<< 1 and E(N2 N2 )
<< 1, Eqs. (A.41) and (A.45) yield
∗
E(NN) = N · E∗ · N
∗
E(N1 N2 )
= N1 · E∗ · N2 . (A.55)
Comparison with Eq. (A.21)2 reveals that the linear extensional (normal)
and shear strains are components of the linear strain tensor. (The shear no
longer depends on the normal strains.) Thus, in general we can write
∗
Ēij = E∗ : ēi ēj = ēi · E∗ · ēj , (A.56)
where the ēi are an orthogonal set of unit vectors of arbitrary orientation.
∗ ∗
If i = j, the Ēij are normal strain components; if i = j, the Ēij are shear
strain components.
The scalar form of the linear strain-displacement relation in Carte-
sian coordinates is obtained by neglecting the nonlinear term uk ,i uk ,j in
Eq. (A.47) to get
∗
Eij = 12 (ui ,j +uj ,i ). (A.57)
These components transform according to Eq. (A.24)2 .

As we have shown, the nonlinear strain components Eij characterize
the deformation at a point in a body; they have no direct physical mean-
∗
ing. The linear components Eij , on the other hand, have the usual physical
interpretations, i.e., the normal strains (i = j) represent relative exten-
sions and the shear strains (i = j) represent angle changes. It must be
remembered, however, that these interpretations are valid only when the
deformation is “small.”
Example A.3 Derive the linear strain-displacement relations in cylindri-

cal polar coordinates.
Solution. The primary complication introduced by curvilinear coor-

dinates is due to the fact that the base vectors may vary with position.
In cylindrical polar coordinates, for example, a set of unit base vectors
(er , eθ , ez ) can be defined in the (r, θ, z) directions, respectively. The vec-
tor ez is constant, but er and eθ change direction (not magnitude) as θ
changes (Fig. A.5). In terms of the Cartesian base vectors ei , we can write
Fig. A.5. Cylindrical polar coordinate system.
er (θ) = e1 cos θ + e2 sin θ

eθ (θ) = −e1 sin θ + e2 cos θ
ez = e3 . (A.58)
Thus,
∂er
= er ,θ = −e1 sin θ + e2 cos θ = eθ
∂θ
∂eθ
= eθ ,θ = −e1 cos θ − e2 sin θ = −er , (A.59)
∂θ
and all derivatives of the unit vectors with respect to r and z vanish.
In cylindrical polar coordinates, the gradient operator has the form (see
Section 2.8.3)
∂ 1 ∂ ∂
∇ = er + eθ + ez , (A.60)
∂r r ∂θ ∂z
and the displacement vector is

u = u r er + u θ eθ + u z ez . (A.61)
Combining these relations yields the displacement gradient tensor
∇u = er (ur ,r er + uθ ,r eθ + uz ,r ez )
+eθ r1 (ur ,θ er + ur er ,θ +uθ ,θ eθ + uθ eθ ,θ +uz ,θ ez )
+ez (ur ,z er + uθ ,z eθ + uz ,z ez ),
and substituting Eqs. (A.59) for the derivatives of the unit vectors gives
the matrix representation
⎡ ⎤
ur , r uθ , r uz , r
∇u = ⎣ r1 (ur ,θ −uθ ) 1r (uθ ,θ +ur ) 1r uz ,θ ⎦ (A.62)
ur , z uθ , z uz , z
with respect to the basis {er , eθ , ez }. With this expression, the linear strain
tensor is
⎡ ∗ ∗ ∗
⎤
Err Erθ Erz
E∗ = 12 [∇u + (∇u)T ] = ⎣ Eθr ∗ ∗
Eθθ ∗ ⎦
Eθz ,
∗ ∗ ∗
Ezr Ezθ Ezz
where
∗
Err = ur , r
∗ 1
Eθθ = r (ur + uθ , θ )
∗
Ezz = uz , z
∗ ∗

1 1

Erθ = Eθr = 2 r
(ur ,θ −uθ ) + uθ , r
∗ ∗

1 1

Eθz = Ezθ = 2 r uz ,θ +uθ ,z
∗ ∗ 1
Erz = Ezr = 2
(ur ,z +uz ,r ). (A.63)
A.2.2 Rotation
As Eq. (A.38) shows, the displacement gradient tensor describes the map-
ping of dR into dr. Although translation does not affect this transforma-
tion, rotation changes the vector dR and so is contained in ∇u. To separate
the effects of deformation and rigid-body motion, we write
(∇u)T = E∗ + Θ∗ , (A.64)
where
E∗ = 12 [(∇u)T + ∇u].
is the linear strain tensor of Eq. (A.54) and
Θ∗ = 12 [(∇u)T − ∇u] (A.65)
is the linear rotation tensor (see below). Substituting Eq. (A.64) into
(A.37) yields
du = (E∗ + Θ∗ ) · dR. (A.66)
∗
Consider now rigid-body motion. In this case, E = 0 and Eq. (A.66)
gives
du = Θ∗ · dR = ω × dR, (A.67)
where ω is the rotation vector. The magnitude of ω gives the rotation
angle, while its direction gives the rotation axis. The right-hand-side of
this expression follows from the geometry of Fig. A.3b (page 319), with dR
rotating through an angle |ω| as its tip moves a distance |du| = |ω| |dR|.
(Since ω is orthogonal to dR, |ω × dR| = |ω| |dR| sin π2 .) In terms of
components, Eq. (A.67) can be written
(Θ∗ij ei ej ) · (dxk ek ) = (ω k ek ) × (dxj ej )
or, with Eq. (A.1)2 ,
Θ∗ij dxk δ jk ei = Θ∗ij dxj ei = ωk dxj kji ei .
Thus,
Θ∗ij = kji ω k , (A.68)
which shows that Θ∗21 = k12 ωk = 312 ω3 = ω 3 = −Θ∗12 represents the

rotation about the x3 -axis, etc.
Another way to see the meaning of Θ∗ is to examine the component
form of Eq. (A.65). Manipulations similar to those that led to Eq. (A.57)
yield
Θ∗ij = 12 (ui ,j −uj ,i ), (A.69)
which are the familiar rotation components from linear elasticity theory,
with their usual geometric interpretations. In a deformable body, Θ∗ rep-
resents the average rotation of all fibers passing through a given point.
Note that, whereas E∗ is symmetric (Eij ∗ ∗

= Eji ), Θ∗ is antisymmetric
∗ ∗
(skew symmetric) (Θij = −Θji ).
In terms of the displacement vector, Eq. (A.65) gives Θ∗ , and we can
show that
ω = 12 ∇ × u. (A.70)
To see this, we use Eq. (A.29) to write this expression as

ωk ek = 12 um ,l lmk ek ,
and then multiplying both sides by ijk yields
1
ωk ijk = u ,
2 m l lmk ijk
= 12 um ,l klm kij .
Next, using (A.5) gives

ωk ijk = 12 um ,l (δ li δ mj − δ lj δ mi ) = 12 (uj ,i −ui ,j ).
Finally, reordering the subscripts in the permutation symbol and noting
Eq. (A.69) gives
ωk kji = 12 (ui ,j −uj ,i ) = Θ∗ij ,
which agrees with (A.68). Thus, local rigid-body rotation can be described
in terms of either the rotation tensor Θ∗ , given by Eq. (A.65), or the
rotation vector ω, given by (A.70).
A.2.3 Principal Strains

Consider the deformation part of Eq. (A.66), i.e.,
du = E∗ · dR. (A.71)
We want to find the orientation of a differential element of the body for
which the shear strains are zero. Since only extension occurs in this case,
the vectors du and dR align and we set du = E ∗ dR, where E ∗ is the linear
extensional strain. Combining this relation with (A.71) yields
E ∗ dR = E∗ · dR
or, with dR = N dS,
(E∗ − E ∗ I) · N = 0, (A.72)
which is in the form of an eigenvalue problem. The three eigenvalues Ei∗

are the principal strains, and the corresponding eigenvectors Ni define
the principal directions of strain. Since E∗ is a symmetric tensor, the
eigenvalues are real and the eigenvectors are mutually orthogonal.
Example A.4 Consider simple shear of a unit cube (Fig. A.6). During
deformation, each point of the cube moves parallel to the x1 -axis with the
displacement vector given by
u = kx2 e1 , (A.73)
where k is a constant. Compute the following:
(a) The Cartesian components of the linear strain tensor E∗ and rotation
tensor Θ∗ .
(b) The components of the nonlinear (Lagrangian) strain tensor E.
(c) The components of E relative to axes x̄i that are rotated through an
angle θ = 45◦ about the x3 -axis.
(d) The principal strains and principal directions for E∗ .
Fig. A.6. Simple shear of a cube.
Solution. (a) For this problem, the displacement gradient tensor is
∂ ∂ ∂
∇u = e1 + e2 + e3 (kx2 e1 )
∂x1 ∂x2 ∂x3
⎡ ⎤
0 0 0
= k e2 e 1 = ⎣ k 0 0 ⎦
0 0 0
and so
⎡ ⎤
0 k 0
(∇u)T = k e1 e2 = ⎣ 0 0 0 ⎦.
0 0 0
In terms of these quantities, Eqs. (A.54) and (A.65) provide the linear strain
and rotation tensors

⎡ ⎤
0 12 k 0
E∗ = 12 [(∇u)T + ∇u] = 1 ⎣ 1k 0 0 ⎦
2 k (e1 e2 + e2 e1 ) = 2
0 0 0
⎡ 1 ⎤
0 2k 0
Θ∗ = 12 [(∇u)T − ∇u] = 1
2
k (e1 e2 − e2 e1 ) = ⎣ − 12 k 0 0 ⎦ ,
0 0 0
(A.74)
which are seen to be symmetric and antisymmetric tensors, respectively.
Thus, for small deformation, the only non-zero strain components are
∗
E12 ∗
= E21 = 12 k, which is half the angle change, as expected (Fig. A.6).
Because the strains and rotations are constant, the deformation of the cube
is homogeneous.
As shown by Eq. (A.66), the deformed configuration of the cube rep-
resents the superposition of a pure deformation and a rigid-body rotation.
According to Eqs. (A.74), the deformation is a shear k/2, and the rota-
tion has a magnitude of k/2 about the x3 -axis (Θ∗12 = −Θ∗21 = k/2). The
rotation also can be computed from (A.70), i.e.,
1 1 ∂ ∂ ∂
ω = 2∇ ×u= e1 + e2 + e3 × (kx2 e1 )
2 ∂x1 ∂x2 ∂x3
1 1
= 2 k e2 × e1 = − 2 k e3 = ω 3 e3 .
The result ω 3 = −k/2 is consistent with a clockwise (negative) rotation

about the x3 -axis. Moreover, Eq. (A.68) gives Θ∗12 = −Θ∗21 = 321 ω3 =
−ω 3 , in agreement with this result. The deformational and rotational com-
ponents for this problem are illustrated in Fig. A.6.
(b) The Lagrangian strain tensor requires the additional term
(∇u) · (∇u)T = (ke2 e1 ) · (ke1 e2 ) = k 2 e2 e2 ,
and Eq. (A.42) yields

1
E = 2 [∇u + (∇u)T + (∇u) · (∇u)T ]
= 1
2
[k (e1 e2 + e2 e1 ) + k 2 e2 e2 ]
⎡ 1 ⎤
0 2
k 0
= ⎣ 1k 1 2
0 ⎦.
2 2k (A.75)
0 0 0
The component E22 characterizes the stretch of a line element originally

parallel to the x2 -axis. If the side AB of the√
undeformed cube has a length
dS = 1, then its deformed length is ds = 1 + k 2 (Fig. A.6). Equation
(A.40) then gives E22 = 12 (ds2 − dS 2 )/dS 2 = 12 k 2 , in agreement with the
above result.
x2
x2
e2 x1
e2
e1
q
e1 x1
Fig. A.7. Rotation of coordinate axes in two dimensions.
(c) For θ = 45◦ , the unit vectors along the x̄i axes in Fig. A.7 are
ē1 = √12 (e1 + e2 ) = √12 [1, 1, 0]T
ē2 = √1 (−e1 + e2 ) = √1 [−1, 1, 0]T
2 2
ē3 = e3 = [0, 0, 1]T (A.76)
in terms of the nonrotated unit vectors ei . Thus, Eqs. (A.21) and (A.75)
give
Ē11 = ē1 · E · ē1
⎡ ⎤ ⎡ ⎤
0 k 0 1
= √1 [1, 1, 0] · 1 ⎣ k k2 0 ⎦· √1 ⎣ 1 ⎦
2 2 2
0 0 0 0
= 1
2k + 14 k 2
Ē22 = ē2 · E · ē2 = − 12 k + 14 k 2
Ē12 = ē1 · E · ē2 = 14 k 2 (A.77)
and so on.
(d) To compute the principal strains and principal directions for the
linear case, we use (A.74)1 and write Eq. (A.72) in matrix form
⎡ 1 ⎤⎡ ⎤ ⎡ ⎤
−E ∗ 2
k 0 N1 0
⎣ 1 k −E ∗ 0 ⎦ ⎣ N2 ⎦ = ⎣ 0 ⎦ . (A.78)
2
0 0 −E ∗ N3 0
Taking the determinant of the 3×3 matrix gives the characteristic equation
E ∗ (E ∗2 − 14 k 2 ) = 0,
which provides the principal strains
E1∗ = 12 k, E2∗ = − 12 k, E3∗ = 0. (A.79)
The corresponding principal directions, obtained by substituting these
equations into (A.78), are
N1 = √1 [1, 1, 0]T , N2 = √1 [−1, 1, 0]T , N3 = [0, 0, 1]T . (A.80)
2 2
Note that these directions are the same as those used in part (c) to trans-
form the nonlinear strain components [see Eq. (A.76)], i.e., parallel to the
diagonals of the undeformed cube. When linearized for k << 1, Eqs. (A.77)
agree with the principal strains of (A.79), as they should, including the van-
ishing shear strain (relative to the magnitudes of E11 and E22 ). But since
the nonlinear shear strain is not zero, the principal directions for the linear
and nonlinear problems are not the same.
A.2.4 Compatibility Conditions

If the strain field is specified, then the strain-displacement relations (A.57)
provide six equations to solve for only three displacement components.
Thus, the system is over-determined. In other words, the strains cannot be
specified arbitrarily; they must satisfy some additional constraints called
compatibility conditions.
The physical basis for this problem is the following. Suppose the unde-
formed body is cut into differential elements, and then the strain for each
element is specified arbitrarily. It is quite possible that the deformed pieces
would not fit together without further deformation when the body is re-
assembled, i.e., the specified deformation field would be incompatible. The
compatibility conditions guarantee that, after deformation, the elements fit
together without holes and with no further deformation.
Our derivation of the requisite equations makes use of the following
identities, which we list without proof:
I × (∇ × a) = (∇a)T − ∇a
∇ × (I × a) = (∇a)T − I ∇ · a
∇ · (∇ × a) = 0
∇ × ∇a = 0. (A.81)
We begin with Eq. (A.64) in the form

∇u = E∗T + Θ∗T = E∗ − Θ∗
= E∗ − 12 [(∇u)T − ∇u],
where Eq. (A.65) and the symmetry properties of the strain and rotation
tensors have been used. Then, applying Eq. (A.81)1 yields
∇u = E∗ − 1
2 I × (∇ × u)
∗
= E − I × ω, (A.82)
where ω is given by Eq. (A.70). Taking the curl of this expression and
using Eq. (A.81)2 gives
∇ × ∇u = ∇ × E∗ − [(∇ω)T − I ∇ · ω].
Next, substituting (A.70) and applying Eq. (A.81)4 to the left-hand-side
and (A.81)3 to the right-hand-side shows that the terms involving ∇u and
∇ · ω vanish, giving
∇ × E∗ = (∇ω)T . (A.83)
Finally, taking the transpose and curl of this equation and using (A.81)4
yields
∇ × (∇ × E∗ )T = 0, (A.84)
which is the compatibility condition in direct notation.

In component form,
∂ ∗
∇ × E∗ = ek ∗
× (Eij ei ej ) = Eij ,k kil el ej
∂xk
by Eq. (A.1)2 , and so
∂ ∗
∇ × (∇ × E∗ )T = em × (Eij ,k kil ej el )
∂xm
= Eij∗ ,km kil mjn en el = 0,
which gives the nine scalar relations
∗
kim ljn Eij ,kl = 0. (A.85)
Due to the symmetry of the strain tensor, only six of these equations are in-
dependent. Moreover, using Eq. (A.3), we can show that the compatibility
conditions can be written in the alternate form
∗ ∗
Eij ,kl + Ekl ,ij − Elj∗ ,ki − Eki
∗
,lj = 0. (A.86)
A.3. Analysis of Stress 335
Equation (A.84) is a necessary condition for strain compatibility. It

also is a sufficient condition for simply connected bodies, i.e., bodies that
have no holes. For a multiply connected body with n holes, the additional
condition

du = 0 (i = 1, 2, . . ., n) (A.87)
Ci
must be satisfied to ensure single-valued displacements around the contours

Ci of the holes.
A.3 Analysis of Stress
A.3.1 Body and Contact Forces

The forces that act on a body consist of two types: body forces and con-
tact forces. Body forces act at a distance, e.g., gravitational and magnetic
forces. We denote the body force per unit deformed volume by the vector
f (r, t). Contact forces act directly on surfaces, e.g., pressure and fric-
tional forces. Internal forces in a body are contact forces generated by
material adjacent to the imaginary surface of a volume element. If we let
dP represent the force acting on a deformed surface-area element da with
an outward-directed unit normal n, then the traction (stress) vector
T(n) is defined as the force per unit deformed area, i.e.,
dP
T(n) (r, t) = . (A.88)
da
In the linear theory of elasticity, no distinction is made between the
geometries of the undeformed and deformed bodies. Thus, although the
forces actually act on the deformed body b, it is convenient to refer them to
the undeformed body B. We, therefore, can write the body force as f (R, t)
per unit undeformed volume and the traction vector as
dP
T(N) (R, t) =, (A.89)
dA
where dA is the undeformed surface area of the element with unit normal
N ∼
= n (see Fig. A.8b). The nonlinear theory distinguishes between the
undeformed and deformed geometries.
A.3.2 Stress Tensor

Consider a body B subjected to body and contact forces. Imagine that a
cutting plane divides B into two sections, B1 and B2 , exposing the forces
that B1 exerts on B2 and vice versa (Fig. A.8). If we examine the force
system on B1 alone, then the forces exerted by B2 are represented by trac-
tions T(N) acting on the cutting plane (Fig. A.8b). An infinite number
of cutting planes can pass through any given point, each with a different
traction vector. The set of tractions across all planes passing through a
point in a body defines the state of stress at that point.
Fig. A.8. Forces acting on a body before and after cutting.
To relate traction vectors at a point that act on planes of different orien-

tations, we examine a differential element of B in the shape of a tetrahedron
of volume dV (Fig. A.9). Three faces of the element are orthogonal to the
coordinate axes (unit normals −ei and areas dAi ), and the other face has
an arbitrary orientation (unit normal N and area dA). A body force f acts
on the volume of the element, and tractions T−i and T(N) act on the −ei
and N faces, respectively.
According to Newton’s second law of motion, the sum of the forces
acting on a body is equal to the product of the mass of the body and the
acceleration of its center of mass. For the differential element, Newton’s
e3
T(N)
f N
T-1
dV dh
T-2 e2
T-3
e1
Fig. A.9. Forces acting on a differential tetrahedron.
law takes the form

T(N) dA + T−1 dA1 + T−2 dA2 + T−3 dA3 + f dV = (ρ dV ) a, (A.90)
where a is the acceleration and ρ is the mass density. The surface of the
element adjacent to the −ei surface has a unit normal ei and traction Ti ,
and by Newton’s third law of action-reaction, T−i = −Ti . Thus, Eq. (A.90)
becomes
T(N) dA − Ti dAi + f dV = ρa dV, (A.91)
in which the summation convention applies. The surface areas and volume
of the element are related by the geometric relations
dAi = dA cos(N, ei ) = dA (N · ei )
1
dV = 3 dh dA, (A.92)
where dh is the distance from the origin to the N face (Fig. A.9). Substi-
tuting Eqs. (A.92) into (A.91) gives
T(N) dA − Ti dA (N · ei ) + f ( 13 dh dA) = ρ a( 13 dh dA).
Dividing through by dA and letting dh → 0 yields
T(N) = (N · ei ) Ti , (A.93)
i.e., as the volume of the element shrinks to zero, the body and inertia
forces disappear. Equation (A.93) reveals that if the traction is known on
three mutually orthogonal planes passing through a point, then it can be
computed on any arbitrarily oriented plane. Putting N = Nj ej into this
equation gives
T(N) = Ni Ti . (A.94)
Next, we write the traction vectors in the form

Ti = σ ij ej , (A.95)
where the σ ij are the components of Ti , i.e., the stress components,
relative to the ei system. Consistent with the standard convention, this
expression shows that the first subscript on σ ij indicates the orientation
of the plane on which the stress component acts, and the second subscript
indicates the direction of action. If i = j, σ ij is a normal stress; if i = j,
σ ij is a shear stress.
Inserting Eq. (A.95) into (A.93) gives the Cauchy stress formula
T(N) = N · σ, (A.96)
where
σ = σ ij ei ej (A.97)
is the (linear) stress dyadic or stress tensor. The tensor character of σ

is demonstrated by Eq. (A.96), which shows that σ transforms a vector into
another vector. Thus, the components transform according to Eq. (A.24)2 ,
which is equivalent to the expression [see Eq. (A.23)2 ]
σ̄ ij = σ : ēi ēj = ēi · σ · ēj . (A.98)
Here, the σ̄ ij are the components of σ with respect to any orthogonal basis
{ēi ēj }. The nine components of the stress tensor relative to any basis
completely define the state of stress at a point in a body.
Example A.5 Suppose the stress tensor at a point in a body is given by

σ = 3 e 1 e1 + 2 e1 e2 + 2 e2 e1 − 6 e2 e2 + 5 e3 e3
⎡ ⎤
3 2 0
= ⎣ 2 −6 0 ⎦ ,
0 0 5
where the components are expressed in appropriate units relative to the

Cartesian coordinates (x1 , x2 , x3 ). Compute the traction vector and the
normal and shear stresses across the plane passing through this point with
unit normal
N = 13 (e1 + 2e2 + 2e3 ).
Solution. Computing the traction vector is straightforward. Equation

(A.96) gives
⎡ ⎤
3 2 0
T(N) = N · σ = 13 [1, 2, 2] ⎣ 2 −6 0 ⎦
0 0 5
⎡ ⎤
7
= 13 ⎣ −10 ⎦ = 13 (7e1 − 10e2 + 10e3 ).
10
To determine the normal and shear stresses, we define S to be the unit
vector orthogonal to N, i.e., in the plane, that points along the projection
of T(N) on the plane (Fig. A.10). Then, the traction vector can be written
in the form
T(N) = σ (NN) N + σ (NS) S,
where σ (NN) is the normal stress and σ (NS) the shear stress. Thus, the
normal stress is
σ (NN) = N · T(N)
⎡ ⎤
7
1 1 ⎣ −10 ⎦ = 7 .
= 3 [1, 2, 2] · 3 9
10
Similarly, the shear stress is σ (NS) = S · T(N) , but S must be defined
carefully because an infinite number of vectors are orthogonal to N. A
straightforward way to compute σ (NS) is to note that
|T(N) |2 = T(N) · T(N) = σ 2(NN) + σ 2(NS) ,
which gives
2192
σ 2(NS) = T(N) · T(N) − σ 2(NN) = .
81

T(N)
S V(NN)
V(NS)
Fig. A.10. Normal and shear components of the traction vector.
A.3.3 Principal Stresses

Because σ is a tensor, when the coordinate system x̄i changes, the compo-
nents of σ change, but σ itself does not. As in the analysis for principal
strains, we seek an orientation of the x̄i for which the components of shear
stress vanish.
If the traction vector T(N) has no components of shear, then it aligns
with the normal N to the surface, i.e.,
T(N) = σN, (A.99)
where σ is the normal stress component on the N surface. In the next
subsection, we show that the stress tensor is symmetric, i.e., σ = σ T . Thus,
substituting Eq. (A.96) into (A.99) yields σN = N · σ = σ T · N = σ · N or
(σ − σI) · N = 0, (A.100)
which is an eigenvalue problem similar to Eq. (A.72). Here, the three eigen-
values σ i are the principal stresses and the corresponding eigenvectors Ni
correspond to the principal directions of stress. Since σ is symmetric,
the eigenvalues are real and the eigenvectors are mutually orthogonal.
It is important to note that the principal axes of stress and strain coin-
cide in isotropic solids. This is not always the case for anisotropic materials,
however, as normal stresses may produce shear strains [see Eq. (A.144) be-
low]. In this case, an element oriented in the principal directions of stress
would undergo shear deformation.
A.3.4 Equations of Motion

Thus far, we have used Newton’s second law of motion to define the stress
tensor at a point. Now, we apply Newton’s law to a body on a global scale
and then deduce the local equations of motion for an infinitesimal element.
Each element of a body must satisfy these equations.
Consider first a particle of mass m in motion relative to a Newtonian
frame of reference. The displacement of the mass at time t is u(R, t) relative
to its initial position R in the reference frame. Then, the velocity and
acceleration of m are, respectively,
du
v(R, t) = = u̇
dt
dv
a(R, t) = = v̇, (A.101)
dt
where dot denotes differentiation with respect to time. The law of conser-
vation of linear momentum for the particle is
d
F= (mv), (A.102)
dt
where F is the force acting on the particle and mv is the linear momentum.
Because m is constant (ignoring relativistic effects), this relation becomes
F = ma, i.e., Newton’s second law.
Now, consider a body of mass density ρ(R, t) subjected to a traction
T(N) (R, t) acting over the surface area A and a body force f (R, t) act-
ing over the volume V (Fig. A.11). Here again, we neglect the difference
f
N
T(N)
V
dA
R
dV
Fig. A.11. Body and surface forces acting on a body.
between the geometries of the undeformed and deformed configurations.

(The “configuration” is not affected by rigid-body motion.) Extending

Eq. (A.102) to the entire body yields

d
T(N) dA + f dV = v ρ dV, (A.103)
A V dt V
where v(R, t) is the velocity vector for the mass center of an element located
originally at position R. The left-hand-side of this equation represents the
total force acting on the body, and the right-hand-side is the time rate of
change of the linear momentum for the body.
The surface integral in the first term of (A.103) can be transformed
into a volume integral by first substituting Eq. (A.96) and then using the
divergence theorem (A.31) to obtain

T(N) dA = N · σ dA = ∇ · σ dV.
A A V
In addition, differentiating the right-hand-side of (A.103) yields

d d
v ρ dV = (v ρ dV )
dt V V dt

dv d
= ρ dV + v (ρ dV )
V dt dt

d2 u
= 2
ρ dV,
V dt
d
where Eq. (A.101)1 has been used, and dt (ρ dV ) = 0 since the mass ρ dV of
the element is constant. Thus, the global equation of motion (A.103) can
be written
(∇ · σ + f − ρü) dV = 0. (A.104)
V
Because this relation must hold for an arbitrary volume in the body, it
implies the local equation of motion
∇ · σ + f = ρü. (A.105)
If a given problem is static or if inertia effects are small, i.e., the prob-
lem is quasi-static, then we can set ü = 0 and this relation becomes an
equilibrium equation.
The component form of the equation of motion can be found by substi-
tuting σ = σ ij ei ej , f = fi ei , and u = ui ei in Eq. (A.105). Then, the first
term can be written
∂
∇·σ = ek · (σ ij ei ej )
∂xk
= σ ij ,k δ ki ej = σ ij ,i ej = σ ji,j ei ,
and (A.105) gives
σ ji,j +fi = ρüi . (A.106)
Equation (A.103) governs the linear (translational) motion of a body.

The rotational motion must satisfy the law of conservation of angular mo-
mentum. For a particle, this principle has the form
d
R × F = [m(R × v)], (A.107)
dt
where R × F is the applied moment and m(R × v) the angular momentum
about an arbitrarily chosen point. Extending this relation to a body gives

d
R × T(N) dA + R × f dV = (R × v) ρ dV, (A.108)
A V dt V
where concentrated moments are neglected (nonpolar case), and R is the
position vector to an arbitrary point in the body (Fig. A.11). The first
term can be converted into a volume integral as follows:

R × T(N) dA = R × (N · σ) dA
A
A
= (xi ei ) × (Nj ej · σ kl ek el ) dA
A
= xi Nj σ kl ei × (δ jk el ) dA
A
= xi Nj σ jl ilm em dA
A
= ilm (xi σ jl ),j em dV.
V
In these manipulations, Eq. (A.96) was substituted in the first line,
Eqs. (A.1) were used in the third and fourth lines, and Eq. (A.31) was used
in the last line (with aj → xi σ jl ilm em ). The right-hand-side of (A.108)
can be written

d
(R × v) ρ dV = (Ṙ × v) ρ dV + (R × v̇) ρ dV
dt V V
V
d
+ (R × v) (ρ dV )
V dt
2
d u
= R × 2 ρ dV,
V dt
in which the first integral on the right-hand-side of the first line vanishes
since Ṙ = v and the last integral vanishes due to conservation of mass.
With these expressions, R = xi ei , and Eq. (A.1)2 , the component form of

Eq. (A.108) becomes

ijk [(xi σ lj ),l +xi fj − ρxi üj ] ek dV = 0
V
or
ijk [xi ,l σ lj + xi (σ lj ,l +fj − ρüj )] ek dV = 0. (A.109)
V
The term in parentheses vanishes through the translational equation of
motion (A.106), while xi ,l = ∂xi /∂xl = δ il . Thus, for an arbitrary volume
element, Eq. (A.109) implies
ijk σ ij ek = 0 or jik σ ji ek = 0,
where the second equation was obtained by exchanging i with j in the first
equation. Now, since jik = − ijk , adding these equations gives
ijk (σ ij − σ ji) ek = 0, (A.110)
which implies σ ij = σ ji . Thus, the principle of angular momentum leads
to the conclusion that the stress tensor is symmetric (σ = σ T ).
Example A.6 Suppose the stresses in the wedge of Fig. A.12 are given by
⎡ ⎤
ax21 2cx1 x2 0
σ = ⎣ 2cx1 x2 bx22 0 ⎦ (A.111)
0 0 0
relative to the Cartesian coordinates (x1 , x2 , x3 ). Compute the body forces
and surface tractions required to hold the wedge in static equilibrium with
the stipulated stress distribution.
Solution. First, consider local equilibrium. With üi = 0, Eqs. (A.106)
give the equilibrium equations
∂σ 11 ∂σ 21 ∂σ 31
+ + + f1 = 0
∂x1 ∂x2 ∂x3
∂σ 12 ∂σ 22 ∂σ 32
+ + + f2 = 0
∂x1 ∂x2 ∂x3
∂σ 13 ∂σ 23 ∂σ 33
+ + + f3 = 0, (A.112)
∂x1 ∂x2 ∂x3
and substituting (A.111) yields
f1 = −2x1 (a + c)
f2 = −2x2 (b + c)
f3 = 0. (A.113)
x2
x1 + x2 = 1
x1
x3
Fig. A.12. A solid wedge.
These body force distributions are required to maintain local equilibrium

for each element of the body.
Global equilibrium of the body requires, in addition to the body
forces, the application of surface tractions consistent with the stresses of
Eq. (A.111). As σ 13 = σ 23 = σ 33 = 0, no forces are required in the x3 -
direction. On the vertical face, the unit (outward) normal is N = −e1 , and
Eq. (A.96) gives
⎡ ⎤ ⎡ ⎤
0 0 0 0
T(N) = N · σ = [−1, 0, 0] ⎣ 0 bx22 0 ⎦ = ⎣ 0 ⎦ ,
0 0 0 0
where we have set x1 = 0. Similarly, setting N = −e2 and x2 = 0 gives
⎡ ⎤ ⎡ ⎤
ax21 0 0 0
T(N) = [0, −1, 0] ⎣ 0 0 0 ⎦=⎣ 0 ⎦
0 0 0 0
on the horizontal face. These faces,√therefore, are traction-free. On the
inclined face, setting N = (e1 + e2 )/ 2 and x2 = 1 − x1 yields
⎡ ⎤
ax21 2cx1 (1 − x1 ) 0
1
T(N) = √ [1, 1, 0] ⎣ 2cx1 (1 − x1 ) b(1 − x1 )2 0 ⎦
2
0 0 0
⎡ 2
⎤
ax1 + 2cx1 (1 − x1 )
1
= √ ⎣ b(1 − x1 )2 + 2cx1 (1 − x1 ) ⎦ . (A.114)
2
0
The computed distributions of body forces and surface tractions must
be applied to the body to maintain equilibrium if the internal stresses are
to be those specified by Eq. (A.111). Overall equilibrium can be checked

by summing the applied loads. For a wedge of unit thickness, doing this in
the x1 -direction gives
1 1−x2 1
√
F1 = f1 dx1 dx2 + σ N1 ( 2 dx1 ), (A.115)
0 0 0
in which the body force f1 is integrated over the volume of the wedge,
and σ N1 , the x1 -component of T(N) , is integrated over√the area of the in-
clined surface. (Geometry gives the “arclength” ds = 2 dx1 .) Substitut-
ing Eqs. (A.113) and (A.114) into (A.115) and carrying out the integrations
' '
confirms that F1 = 0. A similar calculation shows that F2 = 0.
Example A.7 Using dyadic analysis and Eq. (A.105), derive the scalar
equations of motion in cylindrical polar coordinates.
Solution. In terms of the unit vectors (er , eθ , ez ) in the (r, θ, z) direc-
tions (see Fig. A.5, page 326), the stress tensor, body force vector, and
displacement vector can be written
σ = σ rr er er + σ rθ er eθ + σ rz er ez
+σ θr eθ er + σ θθ eθ eθ + σ θz eθ ez
+σ zr ez er + σ zθ ez eθ + σ zz ez ez
f = fr er + fθ eθ + fz ez
u = u r er + u θ eθ + u z ez . (A.116)
In addition, Eqs. (A.58) give the derivatives
er ,θ = eθ , eθ ,θ = −er
(A.117)
er ,r = er ,z = eθ ,r = eθ ,z = ez ,r = ez ,θ = ez ,z = 0.
With ∇ given by Eq. (A.60), we have
∂ 1 ∂ ∂
∇·σ = er + eθ + ez · (σ rr er er + σ rθ er eθ + σ rz er ez
∂r r ∂θ ∂z
+σ θr eθ er + σ θθ eθ eθ + σ θz eθ ez + σ zr ez er + σ zθ ez eθ + σ zz ez ez )
∂σ rr ∂σ rθ ∂σ rz
= er · er er + er eθ + er ez + · · ·
∂r ∂r ∂r
eθ
+ · (σ rr er ,θ er + σ rθ er ,θ eθ + σ rz er ,θ ez
r
∂σ θr ∂σ θθ ∂σ θz
+ eθ er + eθ eθ + eθ ez
∂θ ∂θ ∂θ
+σ θr eθ er ,θ +σ θθ eθ eθ ,θ + · · · )
∂σ zr ∂σ zθ ∂σ zz
+ez · ez er + ez eθ + ez ez + · · · , (A.118)
∂z ∂z ∂z
A.4. Constitutive Relations 347
where the dots represent terms that vanish after taking derivatives and the
dot product. To complete the analysis, we simplify this expression using
Eqs. (A.117) and then insert the result and (A.116) into (A.105) to get a
relation of the form
Fr er + Fθ eθ + Fz ez = 0.
Setting Fr = Fθ = Fz = 0 gives the equations of motion in the r, θ, and z

directions, respectively:
∂σ rr 1 ∂σ θr ∂σ zr σ rr − σ θθ
+ + + + fr = ρür
∂r r ∂θ ∂z r
∂σ rθ 1 ∂σ θθ ∂σ zθ 2σ rθ
+ + + + fθ = ρüθ
∂r r ∂θ ∂z r
∂σ rz 1 ∂σ θz ∂σ zz σ rz
+ + + + fz = ρüz . (A.119)
∂r r ∂θ ∂z r
A.4 Constitutive Relations
All the equations derived thus far are independent of material properties.
Elastic, viscoelastic, and plastic bodies obey the same geometric relations
and equations of motion. Stress and strain, however, are related through
the constitutive relations, which are material-dependent. Although these
equations must be determined experimentally, thermodynamic considera-
tions restrict their form.
A.4.1 Thermodynamics of Deformation

The first law of thermodynamics is a statement of conservation of energy
for a system. It can be written in the form
K̇ + U̇ = P + Q, (A.120)
where K is the kinetic energy ( 12 mv2 ), U is the internal energy, P is the

power input, and Q is the rate of heat input.
The kinetic energy for a body of mass density ρ and volume V is

1
K= 2 ρ u̇ · u̇ dV, (A.121)
V
where u(R, t) is the displacement vector, and the internal energy is

U= ρu dV, (A.122)
V
in which u is the internal energy per unit mass. The power input is the rate
of work done on the body. For applied surface tractions T(N) and body
forces f ,

P = T(N) · u̇ dA + f · u̇ dV, (A.123)
A V
where A is the surface area. Finally, the rate of heat added to the body is

Q=− q · N dA + ρr dV, (A.124)
A V
where q is the outward-directed heat flux vector (per unit area), r is the
rate of heat production (per unit mass) due to internal sources, and N is
the (outward-directed) unit normal to the surface.
Next, we go through a series of manipulations to express the terms of
Eq. (A.120) in alternate forms. First, Eq. (A.121) gives

d d
K̇ = 12 (ρ dV )u̇ · u̇ + (u̇ · u̇) ρ dV
dt dt
V
= 12 (u̇ · ü + ü · u̇) ρ dV = u̇ · ü ρ dV
V V
= u̇ · (∇ · σ + f ) dV, (A.125)
V
in which the second line follows from the constancy of the mass ρ dV of
an element, and substituting the equation of motion (A.105) produces the
third line. It often is convenient to develop tensor equations in component
form and then convert the result back to direct notation, especially when
the gradient operator is involved. With this procedure, the first term in
the integrand of Eq. (A.125) becomes
∂
u̇ · (∇ · σ) = u̇ · ek · σ ij ei ej
∂xk
= u̇ · (σ ij ,k δ ki ej ) = u̇ · (σ ij ,i ej )
= (u̇k ek ) · (σ ij ,i ej ) = u̇k σ ij ,i δ kj
= u̇j σ ij ,i
= (u̇j σ ij ),i −u̇j ,i σ ij . (A.126)
To convert this expression back to direct notation, we note the following:

∇ · (σ · u̇) = ∇ · (σ ij ei ej · u̇k ek ) = ∇ · (σ ij u̇k δ jk ei )
∂
= ek · (σ ij u̇j ei ) = (σ ij u̇j ),k δ ki = (σ ij u̇j ),i
∂xk
∂
σ : ∇u̇ = (σ ij ei ej ) : ek u̇l el = σ ij u̇l ,k (ei · ek )(ej · el )
∂xk
= σ ij u̇l ,k δ ik δ jl = σ ij u̇j ,i .
Thus, Eq. (A.126) can be written
u̇ · (∇ · σ) = ∇ · (σ · u̇) − σ : ∇u̇. (A.127)
Furthermore, Eq. (A.64) gives ∇u = E∗T +Θ∗T = E∗ −Θ∗ since E∗T = E∗
and Θ∗T = −Θ∗ . Therefore, the last term of (A.127) takes the form
σ : ∇u̇ = σ : (Ė∗ − Θ̇∗ ) = σ : Ė∗ , (A.128)
which is a consequence of the fact that the double-dot product of a sym-
metric tensor (σ) and an antisymmetric tensor (Θ̇∗ ) is zero. Hence, sub-
stituting Eqs. (A.127) and (A.128) into (A.125) yields

K̇ = [∇ · (σ · u̇) − σ : Ė∗ + f · u̇] dV. (A.129)
V
The second term of Eq. (A.120), given by (A.122), is

d
U̇ = (ρ dV )u + u̇ ρ dV
dt
V
= u̇ ρ dV, (A.130)
V
where conservation of mass again has been used. Adding Eqs. (A.129) and
(A.130) provides a volume integral for the left-hand-side of (A.120).
The surface integrals of Eqs. (A.123) and (A.124) now are converted
into volume integrals. Substituting Eq. (A.96) into (A.123) and applying
the divergence theorem (A.31) yields

P = N · σ · u̇ dA + f · u̇ dV
A V
= [∇ · (σ · u̇) + f · u̇] dV. (A.131)

V
Similarly, Eq. (A.124) becomes

Q= (ρr − ∇ · q) dV. (A.132)
V
Finally, inserting Eqs. (A.129)–(A.132) into (A.120) and simplifying

gives

(ρu̇ − σ : Ė∗ − ρr + ∇ · q) dV = 0.
V
For an arbitrary volume element, this relation implies
ρu̇ = σ : Ė∗ + ρr − ∇ · q, (A.133)

which is the local form of the law of conservation of energy. According to
this equation, the rate of increase in internal energy of a volume element
(ρu̇) is equal to the sum of the rate of work done by the stresses on the
element (σ : Ė∗ ), the rate of internal heat production (ρr), and the rate of
heat flow into the element (−∇ · q).
A.4.2 Strain-Energy Density Function

Equation (A.133) applies to general types of material. In the following,
we focus on the deformation of elastic bodies. In addition to the first law,
elastic deformation also must satisfy the second law of thermodynamics
(see Section 5.1.2)
ρT η̇ = ρr − ∇ · q (A.134)
for reversible processes, where η is the specific entropy (per unit mass) and
T is the absolute temperature. Combining this relation with Eq. (A.133)
to eliminate r gives
σ : Ė∗ = ρ(u̇ − T η̇). (A.135)
We consider two special cases: (1) isentropic deformation (η̇ = 0) and

(2) isothermal deformation (Ṫ = 0). For isentropic deformation, we define
W = ρu (A.136)
to be the strain-energy density function per unit volume. Then, for
η̇ = 0 and constant density, Eq. (A.135) becomes
σ : Ė∗ = Ẇ . (A.137)
Moreover, if temperature effects are ignored, W can be taken as a func-
tion of deformation alone, i.e., W = W (E∗ ) = W (Eij
∗ 12
). In this case,
Eq. (A.137) yields
∗ ∂W ∗
σ ij Ėij = ∗
Ėij
∂Eij
12 In an elastic material, stress depends only on the current deformation.
∗
by virtue of the chain rule and Eq. (A.12). Thus, for arbitrary Ėij , we have
∂W
σ ij = ∗ , (A.138)
∂Eij
∗
which is the constitutive relation for an elastic material. Once W (Eij )
is determined through appropriate experiments, this equation provides the
stress-strain relations.
In the case of isothermal deformation, we introduce the Helmholtz free-
energy function
Ψ = u − T η. (A.139)
For Ṫ = 0, this relation gives Ψ̇ = u̇ − T η̇, and Eq. (A.135) becomes

σ : Ė∗ = ρΨ̇. (A.140)
Now, if an isothermal strain-energy density function is defined by
W = ρΨ, (A.141)
Eq. (A.140) takes the form of Eq. (A.137). Thus, for an elastic material
∗
with W = W (Eij ) under isothermal conditions, the constitutive equation
again takes the form of Eq. (A.138).
In summary, when the deformation of an elastic body is isentropic or
isothermal, the stress components can be computed as the derivatives of a
strain-energy density function W with respect to the corresponding strain
components. If the deformation is isentropic, W can be identified with
the internal energy per unit volume. If the deformation is isothermal, W
corresponds to the free energy per unit volume. The specific form of W
must be determined experimentally for each particular material.
A.4.3 Linear Elastic Material

For stress to depend linearly on strain, Eq. (A.138) shows that W must be
a quadratic function of the strain components, i.e.,
1 ∗ ∗
W = 2 Cijkl Eij Ekl, (A.142)
where the Cijkl are the components of a fourth-order tensor. The 34 = 81

Cijkl are elastic constants to be determined experimentally. Fortunately,
their number can be reduced using the symmetry property of the strain ten-
sor. For example, since Eq. (A.142) can be written W = 21 CjiklEji ∗ ∗
Ekl =
1 ∗ ∗
2 C jiklE E
ij kl , then C ijkl = C jikl , and a similar argument shows that
Cijkl = Cijlk , reducing the number of independent constants to 36. Fur-
thermore, we can write W = 12 Cklij Ekl ∗ ∗
Eij , and so Cijkl = Cklij , i.e., the
matrix of 36 constants is symmetric. This observation reduces the number
of constants to 21. Thus, the mechanical properties of a general anisotropic,
linear elastic material are characterized by 21 independent elastic constants
with
Cijkl = Cjikl = Cijlk = Cklij . (A.143)
When material symmetry is present, the number of independent elastic
constants can be reduced still further (see Section 5.6). An orthotropic
material, characterized by three mutually orthogonal planes of symmetry,
has nine constants. A transversely isotropic material, which has an axis
of symmetry, has five independent constants. And an isotropic material,
with the same properties in all directions, has two constants. Note that,
for certain anisotropic materials, extensional and shear effects are coupled.
To obtain the stress-strain relations, we substitute Eq. (A.142) into
(A.138) and use (A.143) to obtain
∂ 1 ∗ ∗

σ ij = ∗ 2 Cklmn Ekl Emn
∂Eij
# $
∗ ∗
1 ∂Ekl ∗ ∗ ∂Emn
= 2 Cklmn ∗ Emn + Ekl ∂E ∗
∂Eij ij
1 ∗ ∗
= 2 Cklmn (δ ki δ lj Emn + Ekl δ mi δ nj )
1 ∗ ∗
= 2 (Cijmn Emn + Cklij Ekl )
1 ∗ ∗
= 2 (CijklEkl + CijklEkl )
or
∗
σ ij = CijklEkl , (A.144)
which is called a generalized Hooke’s law, valid for general anisotropy.

In direct notation, this equation can be written in the form
σ = C : E∗ , (A.145)
where C is a fourth-order tensor. This can be seen by the manipulations
∗
σ ij ei ej = (Cijkl ei ej ek el ) : (Emn em en )
∗
= CijklEmn ei ej (ek · em )(el · en )
∗
= CijklEmn ei ej δ km δ ln
∗
= CijklEkl ei ej
A.5. Boundary Value Problems 353
which implies (A.144)

For an isotropic material, Eq. (A.145) can be written
σ = λΔI + 2μE∗ , (A.146)
where λ and μ are Lamé constants. In addition,

Δ = tr E∗ = Eii
∗
(A.147)
∗
is the trace of the tensor E , which represents the dilatation, i.e., the ratio
of the change in volume to the undeformed volume of a material element.
Of course, Eq. (A.146) also can be written in standard engineering form in
terms of a Young’s modulus, Poisson’s ratio, and shear modulus. Again,
only two of these material constants are independent.
A.5 Boundary Value Problems
The governing equations, along with appropriate boundary and initial con-
ditions, define a boundary value problem in the theory of elasticity. As
derived in this chapter, the basic equations of linear isotropic elasticity
consist of the following:
Strain-Displacement Relations:
E∗ = 12 [∇u + (∇u)T ] or Eij

∗
= 12 (ui ,j +uj ,i ) (A.148)
∇ · σ + f = ρü or σ ji,j +fi = ρüi (A.149)
σ = λ(tr E∗ ) I + 2μE∗ or ∗
σ ij = λEkk ∗
δ ij + 2μEij (A.150)
Given the applied loads, the direct-notation forms of these relations

provide a system of three equations to be solved for u, E∗ , and σ. Due to
the symmetry of the stress and strain tensors, the scalar forms represent a
∗
system of 15 equations for 15 unknowns: ui (3), Eij (6), and σ ij (6). For
static problems, ü = 0 and Eqs. (A.149) become equilibrium equations.
Boundary conditions consist of specified tractions T̂(N) and displace-
ments û over the surface of the body, where hat indicates a prescribed
quantity. In dynamic problems, the initial conditions consist of the speci-
fied displacement û(0) and velocity v̂(0) of all points in the body at t = 0.
With Eq. (A.96) and the unit surface normal N = Ni ei , these conditions
can be written in the following forms:
Boundary Conditions (t ≥ 0):
N · σ = T̂(N) or Nj σ ij = T̂i on Aσ
u = û or ui = ûi on Au (A.151)
Initial Conditions (t = 0):

(0)
u = û(0) or ui = ûi in V
(0) (0)
u̇ = v̂ or u̇i = v̂i in V (A.152)
Here, Aσ and Au represent the portions of the surface where tractions and
displacements, respectively, are specified. In linear elasticity, the unique-
ness theorem ensures that the solution to any particular boundary value
problem is unique (Timoshenko and Goodier, 1969).
A.5.1 Torsion of a Circular Cylinder

Consider torsion of a solid circular cylinder of constant radius a and length
L (Fig. A.13). The cylinder is subjected only to static end couples M
about the longitudinal (z) axis. Our goal is to determine the stress and
strain distributions.
One technique used to solve elasticity problems is the semi-inverse
method, which involves the following steps. First, some kinematic aspects
of the solution are assumed. Next, from these assumptions, a solution is
found that satisfies all the governing equations and boundary conditions.
Finally, for the linear theory, the uniqueness theorem guarantees that the
obtained solution is the one and only solution to the given problem.
Following this procedure, we initially make some kinematic assumptions
based on symmetry. First, since the cylinder and loading are symmetric
about the z-axis, we expect the solution to be independent of the circumfer-
ential coordinate θ. Second, we note that the cylinder and loading look the
same when rotated 180◦ about the x or y-axis. Thus, plane cross sections
must remain plane, i.e., there can be no out-of-plane warping, which would
violate this symmetry. Third, we assume that each cross section rotates as
a rigid body, with the rotation angle varying linearly along the length of
the cylinder.
z
(a) (b)
M ey
eq
er
y uq v
a
q ur
N q u
L r
q
x ex
x
M
Fig. A.13. Torsion of a solid circular cylinder. (a) Three-dimensional view. (b) Cross
section.
These assumptions imply that the displacement vector can be written
u = u r er + u θ eθ + u z ez
= uex + vey + wez (A.153)
where the displacement components are
uθ = βzr, ur = uz = 0 (A.154)
relative to the cylindrical polar coordinates (r, θ, z). The variable β can
be identified as the angle of twist per unit length, which is constant for a
uniform cylinder. The geometry linking the polar and Cartesian coordinates
gives the relations (Fig. A.13b)
x = r cos θ, y = r sin θ
and
eθ = −ex sin θ + ey cos θ. (A.155)
Combining these expressions with Eqs. (A.153) and (A.154) yields the
Cartesian displacement components
u = −βyz, v = βxz, w = 0. (A.156)

With (x1 , x2 , x3 ) ≡ (x, y, z) and (u1 , u2 , u3 ) ≡ (u, v, w), substitution of

Eqs. (A.156) into (A.148) provides the strain components
∗ ∗ 1 ∂u ∂w
Exz = Ezx = + = − 12 βy
2 ∂z ∂x
∗ ∗ 1 ∂v ∂w
Eyz = Ezy = + = 12 βx
2 ∂z ∂y
∗ ∗ ∗ ∗ ∗
Exx = Eyy = Ezz = Exy = Eyx = 0. (A.157)
Inserting these relations into Eq. (A.150) then gives the Cartesian stress
components
σ xz = σ zx = −μβy
σ yz = σ zy = μβx
σ xx = σ yy = σ zz = σ xy = σ yx = 0, (A.158)
and so the stress dyadic is
σ = μβ(−y ex ez − y ez ex + x ey ez + xez ey ). (A.159)
Direct substitution of Eqs. (A.158) reveals that these stresses identically
satisfy the equilibrium equations (A.149) with fi = üi = 0. Thus, if we can
show that the boundary conditions are satisfied, then Eqs. (A.156)–(A.158)
provide the exact solution to this problem.
The only loads acting on the cylinder are twisting moments applied at
the ends. Thus, the lateral surface is stress free, and the net force on the
ends must vanish. The corresponding boundary conditions are
r=a: T(N)= 0
z = 0, L : F= T(N) dA = 0
A (A.160)
M=MN= r × T(N) dA,
A
where F is the resultant end force, M is the resultant end moment, r is the
position vector in the cross section, N is the surface normal, and A is the
cross-sectional area. We now examine each of these conditions.
On the lateral surface, N = er and Eqs. (A.96), (A.155)1 , and (A.159)
yield
T(N) = N·σ
= (ex cos θ + ey sin θ) · [μβ(−y ex ez − y ez ex + x ey ez + x ez ey )]
= μβ(−y cos θ + x sin θ)ez
= μβ[−(r sin θ) cos θ + (r cos θ) sin θ]ez
= 0.
Thus, the first boundary condition of (A.160) is satisfied.

On the end z = L (z = 0 is similar), N = ez and we get
T(N) = N · σ = ez · [μβ(−y ex ez − y ez ex + x ey ez + x ez ey )]
= μβ(−y ex + x ey ). (A.161)
Since the cross section is symmetric about the x and y axes, this expression
shows that the second boundary condition of (A.160), i.e., the vanishing end
force, is satisfied. Substitution into the third condition of (A.160) gives

M = M ez = μβ (x ex + y ey ) × (−y ex + x ey ) dA
A

= μβez (x2 + y2 ) dA.
A
This expression provides the torque-twist relation
M = μJβ, (A.162)
where
J= (x2 + y2 ) dA = r 2 dA (A.163)
A A
is the polar moment of inertia for the cross section. The quantity μJ is
called the torsional rigidity, with J = πa4 /2 for a circular cross section.
Given M , Eq. (A.162) provides β, and then Eqs. (A.156)–(A.158) give the
displacement, strain, and stress distributions.
Thus, the postulated displacement field of Eq. (A.156) gives the exact
linear solution to the torsion problem. Note, however, that this solution
requires that the tractions on the ends of the cylinder be applied exactly
according to the distribution of Eq. (A.161), which can be written (see
geometry of Fig. A.13b)
T(N) = μβ(−y ex + x ey )
= μβ[−(r sin θ)(er cos θ − eθ sin θ) + (r cos θ)(er sin θ + eθ cos θ)]
= μβr eθ . (A.164)
The end traction, therefore, is a circumferential shear stress that increases
linearly with r. Note also that we obtained this expression from the relation
T(N) = ez · σ [see Eq. (A.161)], and so the shear stress component is
T(N) · eθ = ez · σ · eθ = σ zθ = μβr.
If the applied surface traction does not satisfy Eq. (A.164) exactly but
still furnish only a twisting moment M , then the present solution is not
valid near the ends. By Saint-Venant’s principle, however, the solution
is a good approximation sufficiently far from the ends (Timoshenko and
Goodier, 1969).
A.5.2 Torsion of a Noncircular Cylinder
If the cross section of a twisted cylinder is not circular, then our prior
symmetry arguments are not valid. In this case, the problem is not ax-
isymmetric, and warping of cross sections is possible. Here, we modify the
solution of the previous example to allow a more general deformation.
Consider a solid cylinder of length L with a constant cross section that
has a general contour C (Fig. A.14). Modifying Eqs. (A.156), we assume
that the displacement components have the form
z
(a) (b)
M
C
y
N
C
L
x
x
M
Fig. A.14. Torsion of a noncircular cylinder. (a) Three-dimensional view. (b) Cross
section.
u = −βyz, v = βxz, w = β ψ(x, y), (A.165)
where β again is a constant twist per unit length and ψ is a warping

function. The strain components of Eqs. (A.157) now are replaced by
∗ ∗ 1 ∂u ∂w ∂ψ
Exz = Ezx = + = 12 β −y
2 ∂z ∂x ∂x
∗ ∗ 1 ∂v ∂w ∂ψ
Eyz = Ezy = + = 12 β +x
2 ∂z ∂y ∂y
∗ ∗ ∗ ∗ ∗
Exx = Eyy = Ezz = Exy = Eyx = 0, (A.166)
and the stress components of Eq. (A.158) become

∂ψ
σ xz = σ zx = μβ −y
∂x
∂ψ
σ yz = σ zy = μβ +x
∂y
σ xx = σ yy = σ zz = σ xy = σ yx = 0. (A.167)
Note that, for ψ = 0, these relations reduce to those of Eqs. (A.157) and
(A.158). The stresses of (A.167) satisfy the equilibrium equations σji,j = 0
identically for i = 1, 2. For i = 3, we have
∂σ 13 ∂σ 23 ∂σ 33 ∂σ xz ∂σ yz ∂σ zz
+ + = + + = 0,
∂x1 ∂x2 ∂x3 ∂x ∂y ∂z
and substituting Eqs. (A.167) yields
∂2ψ ∂2ψ
+ = ∇2 ψ = 0, (A.168)
∂x2 ∂y2
where ∇2 = ∇ · ∇ is the Laplacian operator. Thus, ψ(x, y) is a harmonic

function.
Any solution of Eq. (A.168) satisfies the field equations of linear elastic-
ity for the torsion problem, with geometric compatibility being ensured by
Eqs. (A.165). For a particular geometry, however, ψ also must satisfy the
boundary conditions, again given by Eqs. (A.160) but with the boundary
r = a replaced by the contour C.
Consider first the condition on the curved surface. With Eqs. (A.167),
the stress tensor can be written
σ = σ xz ex ez + σ zx ez ex + σ yz ey ez + σ zy ez ey , (A.169)
and for an arbitrary cross section, the unit normal to C is (Fig. A.14b)
N = Nx ex + Ny ey , (A.170)
where the components Nx and Ny are known for a given geometry. Substi-
tuting these relations into Eq. (A.96) yields
T(N) = N · σ = (Nx σ xz + Ny σ yz )ez .
Thus, with Eqs. (A.167), the first boundary condition of Eq. (A.160) be-
comes
∂ψ ∂ψ
On C : Nx + Ny = N · ∇ψ = yNx − xNy . (A.171)
∂x ∂y
For a given geometry, the right-hand-side of this equation is known, provid-

ing a Neumann type of boundary condition. Equations (A.168) and (A.171)
provide a two-dimensional boundary value problem to solve for ψ(x, y).
At the end z = L, Eqs. (A.96) and (A.169) yield
T(N) = N · σ = ez · (σ xz ex ez + σ zx ez ex + σ yz ey ez + σ zy ez ey )
= σ zx ex + σ zy ey . (A.172)
With this expression and (A.167), the second boundary condition of (A.160)
becomes

∂ψ ∂ψ
T(N) dA = μβ − y ex + + x ey dA = 0. (A.173)
A A ∂x ∂y
Consider the first term in this equation. Because ψ satisfies Eq. (A.168)
and since x and y are independent variables, this term can be written

∂ψ ∂ ∂ψ
μβex − y dA = μβex x −y
A ∂x A ∂x ∂x
%
∂ ∂ψ
+ x +x dA
∂y ∂y

∂ψ
= μβex ∇· x − y ex
A ∂x

∂ψ
+x + x ey dA,
∂y
where ∇ = ex ∂/∂x + ey ∂/∂y. Applying the divergence theorem (A.31)
now gives

∂ψ ∂ψ ∂ψ
− y dA = xN · − y ex + + x ey ds
A ∂x C ∂x ∂y

∂ψ ∂ψ
= x − y Nx + + x Ny ds
C ∂x ∂y
since Nx = N · ex and Ny = N · ey . Examining Eq. (A.171) reveals that
this line integral vanishes. A similar manipulation shows that the second
term of Eq. (A.173) also is zero, and so the second condition of (A.160) is
satisfied.
The last boundary condition of (A.160) yields

M = M ez = r × T(N) dA
A
= (xex + yey ) × (σ zx ex + σ zy ey ) dA
A

= ez (xσ zy − yσ zx ) dA,
A
where Eq. (A.172) has been used. Substituting (A.167) gives
M = μJ ∗ β, (A.174)
where

∗ ∂ψ ∂ψ
J =J+ x −y dA (A.175)
A ∂y ∂x
with J given by Eq. (A.163). The quantity μJ ∗ is the torsional rigidity of

the shaft. For ψ = 0, Eq. (A.174) reduces to Eq. (A.162), as it should.
Once ψ(x, y) is determined using Eqs. (A.168) and (A.171), Eqs. (A.174)
and (A.175) provide β for a given M . Then, Eqs. (A.165)–(A.167) give the
displacement, strain, and stress distributions. Further details on this and
other problems in linear elasticity can be found elsewhere (e.g., Sokolnikoff,
1956; Timoshenko and Goodier, 1969; Boresi et al., 2010).
Appendix B
Special Coordinate Systems
This appendix lists basic relations for some special orthogonal curvilinear
coordinate systems. Coordinates and base vectors are expressed in terms of
the Cartesian system (x, y, z). In addition, physical stress and deformation
components are used.
B.1 Cylindrical Polar Coordinates
Geometry: The geometry is shown in Fig. B.1.
(x1 , x2 , x3 ) = (r, θ, z)
x = r cos θ, r = (x2 + y2 )1/2

y = r sin θ, θ = tan−1 (y/x)
g1 = ex cos θ + ey sin θ = er
g2 = r(−ex sin θ + ey cos θ) = reθ
g3 = ez
g11 = 1, g22 = r 2 , g33 = 1

11 22 −2
g = 1, g =r , g33 = 1
Γ122 = −r
Γ212 = Γ221 = r −1
Other Γkij = 0
363
364 Special Coordinate Systems
z
y
r
q
x
Fig. B.1. Cylindrical polar coordinate system.
Differential operators:
∂ 1 ∂ ∂
∇ = er + eθ + ez
∂r r ∂θ ∂z
∂2 1 ∂ 1 ∂2 ∂2
∇2 = 2
+ + 2 2+ 2
∂r r ∂r r ∂θ ∂z
Equations of motion:
∂ σ̂ rr 1 ∂ σ̂ θr ∂ σ̂ zr 1
+ + + (σ̂ rr − σ̂ θθ ) + fˆr = ρâr
∂r r ∂θ ∂z r
∂ σ̂ rθ 1 ∂ σ̂ θθ ∂ σ̂ zθ 2σ̂ rθ
+ + + + fˆθ = ρâθ
∂r r ∂θ ∂z r
∂ σ̂ rz 1 ∂ σ̂ θz ∂ σ̂ zz σ̂ rz
+ + + + fˆz = ρâz
∂r r ∂θ ∂z r
Deformation gradient:
⎡ ⎤
∂r 1 ∂r ∂r
⎢ ∂R R ∂Θ ∂Z ⎥
⎢ ⎥
⎢ ⎥
⎢ ∂θ r ∂θ ∂θ ⎥
F(ei ej ) = ⎢ r r ⎥ (i, j = r, θ, z)
⎢ ∂R R ∂Θ ∂Z ⎥
⎢ ⎥
⎣ ∂z 1 ∂z ∂z ⎦
∂R R ∂Θ ∂Z
B.2. Spherical Polar Coordinates 365
r
y
f
x
Fig. B.2. Spherical polar coordinate system.
B.2 Spherical Polar Coordinates
Geometry: The geometry is shown in Fig. B.2.
(x1 , x2, x3 ) = (r, θ, φ)
x = r sin θ cos φ
y = r sin θ sin φ
z = r cos θ
g1 = ex sin θ cos φ + ey sin θ sin φ + ez cos θ = er

g2 = r(ex cos θ cos φ + ey cos θ sin φ − ez sin θ) = r eθ
g3 = r sin θ(−ex sin φ + ey cos φ) = r sin θ eφ
g11 = 1, g22 = r 2 , g33 = (r sin θ)2

g11 = 1, g22 = r −2 , g33 = (r sin θ)−2
Γ122 = −r, Γ133 = −r sin2 θ

Γ212 = Γ221 = r −1 Γ233 = − sin θ cos θ
Γ313 = Γ331 = r −1 Γ323 = Γ332 = cot θ
Other Γkij = 0
366 Special Coordinate Systems
Differential operators:
∂ 1 ∂ 1 ∂
∇ = er + eθ + eφ
∂r r ∂θ r sin θ ∂φ
1 ∂ ∂ 1 ∂ ∂ 1 ∂2
∇2 = r2 + 2 sin θ + 2 2
r 2 ∂r ∂r r sin θ ∂θ ∂θ r sin θ ∂φ2
Equations of motion:
∂ σ̂ rr 1 ∂ σ̂ θr 1 ∂ σ̂ φr
+ +
1 rr
+ 2σ̂ − σ̂ θθ − σ̂ φφ + σ̂ θr cot φ + fˆr = ρâr
r
∂ σ̂ rθ 1 ∂ σ̂ θθ 1 ∂ σ̂ φθ
+ +
1 rθ
+ 2σ̂ + σ̂ θr + (σ̂ θθ − σ̂ φφ ) cot θ + fˆθ = ρâθ
r
rφ θφ
∂ σ̂ 1 ∂ σ̂ 1 ∂ σ̂ φφ
+ +
1
+ (2σ̂ rφ + σ̂ φr + 2σ̂ θφ cot θ + fˆφ = ρâφ
r
Deformation gradient:
⎡ ⎤
∂r 1 ∂r 1 ∂r
⎢ ⎥
⎢ ∂R R ∂Θ R sin Θ ∂Φ ⎥
⎢ ⎥
⎢ ∂θ r ∂θ r ∂θ ⎥
F(ei ej ) = ⎢ r ⎥ (i, j = r, θ, φ)
⎢ ∂R R ∂Θ R sin Θ ∂Φ ⎥
⎢ ⎥
⎣ ∂φ r sin θ ∂φ r sin θ ∂φ ⎦
r sin θ
∂R R ∂Θ R sin Θ ∂Θ
B.3 Toroidal Coordinates
Because of the complexity of some general expressions, only basic geometric

relations are listed below (see Fig. B.3).
(x1 , x2, x3 ) = (r, θ, φ)
x = (b + r cos φ) cos θ
y = (b + r cos φ) sin θ
z = r sin φ
B.3. Toroidal Coordinates 367
y
b
q r
f
Fig. B.3. Toroidal coordinate system.
g1 = cos φ(ex cos θ + ey sin θ) + ez sin φ

g2 = (b + r cos φ)(−ex sin θ + ey cos θ)
g3 = −r sin φ(ex cos θ + ey sin θ) + ez r cos φ
g11 = 1, g22 = (b + r cos φ)2 , g33 = r 2

g11 = 1, g22 = (b + r cos φ)−2 , g33 = r −2
Γ122 = −(b + r cos φ) cos φ, Γ133 = −r

Γ212 = Γ221 = (b + r cos φ)−1 cos φ Γ223 = Γ232 = −(b + r cos φ)−1 r sin φ
Γ313 = Γ331 = r −1 Γ322 = (b + r cos φ)r −1 sin φ
Other Γkij = 0
Problem Solutions
Answers and detailed solutions are provided below for most of the end-of-
chapter problems. In some cases, only parts of the solution are presented.
Chapter 2
2.1 (a) Replacing j by i and removing δ ji (contraction) yields

δ ij δ ji = δ ii = δ 11 + δ 22 + δ 33 = 3.
(b) To solve this problem, it is convenient to use the -δ identity
(2.117), i.e.,
ijk
irs
= jki
rsi
= δ rj δ sk − δ sj δ rk ,
where cyclic permutation of indices provides the middle expression.
Replacing j by m, r by j, and s by k in this equation yields
mki
jki
= δ jm δ kk − δ km δ jk
= 3δ jm − δ jm
= 2δ jm .
(c) Summing over the repeated indices i and j gives
i j 1 1 1 2 1 3
ijk a a = 11k a a + 12k a a + 13k a a
2 1 2 2 2 3
+ 21k a a + 22k a a + 23k a a
3 1 3 2 3 3
+ 31k a a + 32k a a + 33k a a .
Because the permutation symbol is zero if any of its indices are
repeated, this equation simplifies to
i j 1 2 2 3 1 3
ijk a a =( 12k + 21k )a a +( 23k + 32k )a a +( 13k + 31k )a a .
For any k = 1, 2, or 3, 12k = − 21k and so on for the other terms
i j
in parentheses. Hence, ijk a a = 0 for any value of k.
369
370 Problem Solutions
2.2 For the first problem, expanding the dyadic ab yields

ab = (3ex + ey + ez )(ex + 2ez )
= 3ex ex + 6exez + ey ex + 2ey ez + ez ex + 2ez ez .
Dotting with c = ex − 3ey + ez gives
(ab) · c = 9ex + 3ey + 3ez .
A similar calculation yields
(ba) · c = ex + 2ez .
j j
2.3 Since ∂x /∂x = 1 for i = j and zero otherwise, we get
∂ ∂xi j ∂xj
i
(xi xj ) = i
x + xi i
∂x ∂x ∂x
i j i j
= δ i x + x δ i = 3xj + xj = 4xj .
2.4 (a) This result should be obvious, since a × b gives a vector normal
to both a and b; so the dot product with a (or b) is zero. Mathe-
matically, we note gk · a = ak to obtain
(a × b) · a = (ai gi × bj gj ) · a
= (ai bj ijk g
k
)·a
i j k
= ab a ijk .
Since ijk is non-zero only if all subscripts are different, summing

over i, j, and k yields
(a × b) · a = (a1 b2 a3 − a3 b2 a1 ) + (a2 b3 a1 − a1 b3 a2 )
+(a3 b1 a2 − a2 b1 a3 )
= 0.
(c) In terms of covariant tensor components, the given equation be-
comes
A : B = Aij gi gj : Bkl gk gl = Aij Bkl gik gjl = 0
since gi · gk = gik , etc. Using the metric tensor components to
raise the subscripts of Bkl yields
Aij B ij = 0.
For simplicity, consider now the 2-D case (i, j = 1, 2). Expanding
this equation gives
A11 B 11 + A12 B 12 + A21 B 21 + A22 B 22 = 0.
For A symmetric and B antisymmetric, Aji = Aji and B ij =
−B ij . Thus, B 11 = B 22 = 0 and B 21 = −B 12 , and it is easy to
show that the above equation, as well as its 3-D version, is satisfied.
(e) Equation (2.56) gives

(cof A)T (cof B)T cof (A · B)T
A−1 = , B−1 = , (A · B)−1 = .
det A det B det(A · B)
Dotting the first two relations yields
(cof B)T · (cof A)T
B−1 · A−1 = ,
det B det A
which should equal the third relation. Because det(A · B) =
det A det B, comparing these equations reveals that we need to
show
cof (A · B)T = (cof B)T · (cof A)T .
The adjoint of a matrix T, defined as adj T = (cof T)T , has the
known property
adj (A · B) = adj B · adj A,
which is equivalent to the above equation. The validity of this
relation can be demonstrated rather easily in 2-D by manipulating
2 × 2 matrices for A and B.
2.5(a) We begin by plugging in dyadic forms for T and U. Any choice of
basis will work; here we use contravariant components to get
A = T · U = (T ij gi gj ) · (U kl gk gl )
= T ij U kl (gj · gk ) gi gl
= T ij U kl gjk gi gl .
To write this result in terms of components of T and U only, gjk
must be eliminated. This term can be used to lower either the j in
T ij or the k in U kl and replace it with the other subscript in gjk .
This procedure provides the two representations
i
A = T·k U kl gi gl = T ij Uj·l gi gl .
Finally, renaming indices gives
A = T · U = T·k
i
U kj gi gj = T ik Uk·j gi gj .
(b) For A = Aij gi gj , the above result yields
Aij = T·k
i
U kj = T ik Uk·j ,
which can be used to compute the mixed components in A =
A·j i
i g gj . One way to do this is to use metric tensor components
to lower the first index, i.e.,
A·j
i = A
mj
gmi .
Substituting the first representation for Aij yields

A·j m kj kj
i = T·k U gmi = T ik U ,
and so
A = T · U = T ik U kj gi gj .
Computing A·j i using the second representation for A
ij
is left for
the reader’s enjoyment.
(c) Partial answer: A = Ai·j gi gj = T ik Ukj gi gj
2.7 Setting
a = gi , b = gj , c = gk
gives
a × (b × c) = gi × (gj × gk ) = gi × ( jkl g
l
)
ilm lmi
= jkl gm = ljk gm ,
in which Eqs. (2.103) and (2.104) have been used, and the indices
on the permutation symbols have undergone cyclic permutations. In
addition, with (2.14) we obtain
(a · c)b = (gi · gk )gj = δ ik gj
(a · b)c = (gi · gj )gk = δ ij gk .
After substitution, the given vector identity
a × (b × c) = (a · c)b − (a · b)c
becomes
ljk
lmi
gm = δ ik gj − δ ij gk .
Dotting both sides of this equation with gn yields
ljk
lmi n
δm = δ ik δ nj − δ ij δ nk .
Eliminating δ nm on the left-hand side by contraction, then replacing i
by r, n by s, and l by i gives
ijk
isr
= δ rk δ sj − δ rj δ sk .
isr
Anticyclic permutation of the indices in changes the sign to obtain
the final result
ijk
irs
= δ rj δ sk − δ rk δ sj .
2.8 The answers are provided by Eqs. (2.54) and (2.57).
2.9 (a) Partial answer: A13 = Arφ = −r sin θ sin φ, A31 = Aφr = cos θ.
(c) For spherical polar coordinates, Appendix B gives
grr = 1, gθθ = r 2 , gφφ = r 2 sin2 θ
with gij = 0 for i = j. Equation (2.132) yields the physical com-
ponents (1, 2, 3 = r, θ, φ)
⎡ ⎤
T·11 T·21 r −2 T·31 (r sin θ)−2
g (ii)
T̂·ji = T·ji =⎣ T·12 r 2 T·22 T·32 (sin θ)−2 ⎦ .
g(jj) 3 2 2 3 2
T·1 r sin θ T·2 sin θ T·33
To write these components in terms of T ij , we use the relation
T·ji = T ik gkj = T i1 g1j + T i2 g2j + T i3 g3j
to obtain ⎡ ⎤
T 11 T 12 T 13
T̂·ji = ⎣ T 21 r 2 T 22 r 2 T 23 r 2 ⎦.
T 31 r 2 sin θ T 32 r 2 sin θ T 33 r 2 sin2 θ
2 2
2.10 (b) For the given equations

x = 12 (u2 − v2 ), y = uv, z = z,
let x = (x, y, z) and x̄i = (u, v, z). The Cartesian base vectors are
i
g x = g x = ex , gy = g y = ey , gz = g z = ez .
With the position vector being
r = xex + yey + zez ,
the covariant base vectors in the parabolic coordinate system are
ḡi = ∂r/∂ x̄i , i.e.,
∂r
ḡu = = uex + vey
∂u
∂r
ḡv = = −vex + uey
∂v
∂r
ḡz = = ez .
∂z
These equations yield
√
ḡ = ḡu · (ḡv × ḡz ) = u2 + v2
and the contravariant base vectors
ḡv × ḡz uex + vey
ḡu = √ =
ḡ u2 + v 2
ḡz × ḡu −vex + uey
ḡv = √ =
ḡ u2 + v 2
ḡu × ḡv
ḡz = √ = ez .
ḡ
(c) For the parabolic coordinates, the components of the metric tensor
are
⎡ 2 ⎤
u + v2 0 0
[ḡij ] = [ḡi · ḡj ] = ⎣ 0 u2 + v 2 0 ⎦
0 0 1
⎡ ⎤
(u2 + v2 )−1 0 0
[ḡij ] = [ḡi · ḡj ] = ⎣ 0 (u2 + v2 )−1 0 ⎦ .
0 0 1
(d) Equations (2.50) give
Aij = ḡi · gj , Bji = ḡi · gj .
As discussed in the footnote on page 25, the upper and lower in-
dices of Aij represent the row and column, respectively, while the
opposite is true for Bji , giving
⎡ ⎤
i ḡu · gx ḡv · gx ḡz · gx
Aj = ⎣ ḡu · gy ḡv · gy ḡz · gy ⎦
ḡu · gz ḡv · gz ḡz · gz
⎡ ⎤
u −v 0
= ⎣ v u 0 ⎦
0 0 1
⎡ u ⎤
i ḡ · gx ḡv · gx ḡz ·gx
Bj = ⎣ ḡu · gy ḡv · gy ḡz · gy ⎦
ḡu · gz ḡv · gz ḡz · gz
⎡ ⎤
u −v 0
1 ⎣ v u ⎦.
= 0
u2 + v 2 2 2
0 0 u +v
(e) Equation (2.60) gives

⎡ ⎤ ⎡ ⎤
i x,u x,v x,z u −v 0
i ∂x
Aj = = ⎣ y,u y,v y,z ⎦ = ⎣ v u 0 ⎦ .
∂ x̄j
z,u z,v z,z 0 0 1
Using Eq. (2.51) to find [Bji ] is left to the reader.

2.12(a) For the given gi , Eqs. (2.17) and (2.18) give
√
g = g1 · (g2 × g3 ) = 6
and
g2 × g3 1
g1 = √ = − (2ex + 5ey + 2ez )
g 6
g3 × g1 1
g2 = √ = (2ex − ey − 4ez )
g 6
g1 × g2 1
g3 = √ = (ex + ey + ez ).
g 3
The components of a are
ai = a · gi = (15, −1, 22)

ai = a · gi = (2.5, −0.5, 0).
(b) Partial answer: T 11 = 1.83; T13 = −11; T2·1 = −3; I2 = −52

2.13 (a) The given tensor has the dyadic representation T = T·ji gi gj . The
matrix form of Eq. (2.81) is given by
⎡ 1 ⎤⎡ 1 ⎤ ⎡ ⎤
T·1 − λ T·21 T·31 a 0
⎣ T·12 T·22 − λ T·32 ⎦ ⎣ a2 ⎦ = ⎣ 0 ⎦ .
T·13 T·23 T·33 − λ a3 0
The eigenvalues are λ1 = −6.24, λ2 = 2, and λ3 = 2.24.
(b) The corresponding normalized eigenvectors are
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
0 1 0
a1 = ⎣ −0.414 ⎦ , a2 = ⎣ 0 ⎦ , a3 = ⎣ 2.41 ⎦ .
1 0 1
2.14 (a) With Eqs. (2.72) and (2.162)2, the left-hand side can be written
as
∂ ∂Ak·k
tr A = gi gj = gi gj δ ki δ kj
∂A ∂Ai·j
= gk gk = I.
(c) It is easy to show that A : B = Aij B ij (see solution to Prob-

lem 2.4c). With this result and (2.162)2 , the dyadic form of the
left-hand side of the given identity becomes
∂ ∂
(A : B) = gm gn (Aij Bij )
∂C ∂Cmn
∂Aij ∂Bij
= gm gn Bij + Aij .
∂Cmn ∂Cmn
In dyadic form, the first term on the right-hand side of the identity
is
∂A ∂
:B = gm gn Aij gi gj : Bkl gk gl
∂C ∂Cmn
∂Aij ∂Aij
= gm gn Bkl (gi · gk )(gj · gl ) = gm gn Bkl δ ki δ lj
∂Cmn ∂Cmn
∂Aij
= gm gn Bij .
∂Cmn
Exchanging A and and B in the above relation yields
∂B ∂B ∂Bij ij
A: = : A = gm gn A .
∂C ∂C ∂Cmn
Plugging these results into the identity demonstrates its validity.
2.15(a) For spherical coordinates, Appendix B gives the orthogonal base
vectors
gr = ex sin θ cos φ + ey sin θ sin φ + ez cos θ = er
gθ = r(ex cos θ cos φ + ey cos θ sin φ − ez sin θ) = reθ
gφ = r sin θ(−ex sin φ + ey cos φ) = r sin θ eφ ,
where the ei are unit vectors. The non-zero values of gij = gi · gj
are
grr = 1, gθθ = r 2 , gφφ = (r sin θ)2 ,
which give
grr = 1, gθθ = r −2 , gφφ = (r sin θ)−2 .
The Christoffel symbols Γijk are computed using the relation
2Γijk = gjk ,i +gki ,j −gij ,k .
For the present problem, the only non-zero derivatives of the gij are
gθθ ,r = 2r, gφφ ,r = 2r sin2 θ, gφφ ,θ = 2r 2 sin θ cos θ,
giving the non-zero Christoffel symbols
Γrθθ = r
Γθθr = −r
Γrφφ = Γφrφ = r sin2 θ
Γφφr = −r sin2 θ
Γφφθ = −r 2 sin θ cos θ
Γθφφ = Γφθφ = r 2 sin θ cos θ.
From these, Eq. (2.142) provides

Γlij = Γijk gkl ,
which yields
Γθrθ = Γθθr = Γrθθ gθθ = r −1
Γrθθ = Γθθr grr = −r
Γφrφ = Γφφr = Γrφφ gφφ = r −1
Γrφφ = Γφφr grr = −r sin2 θ
Γθφφ = Γφφθ gθθ = − sin θ cos θ
Γφθφ = Γφφθ = Γθφφ gφφ = cot θ.
These results agree with those listed in Appendix B.
Here, we compute only gφ ,φ for illustration. Directly differenti-
ating the equation for gφ gives
gφ ,φ = −r sin θ(ex cos φ + ey sin φ).
Alternatively, Eq. (2.136), i.e.,
gi ,j = Γkij gk ,
yields
gφ ,φ = Γrφφ gr + Γθφφ gθ + Γφφφ gφ
= −r sin2 θ gr − sin θ cos θ gθ
= −r sin2 θ (ex sin θ cos φ + ey sin θ sin φ + ez cos θ)
−r sin θ cos θ(ex cos θ cos φ + ey cos θ sin φ − ez sin θ)
= −ex r sin θ cos φ(sin2 θ + cos2 θ)
−ey r sin θ sin φ(sin2 θ + cos2 θ)
= −r sin θ(ex cos φ + ey sin φ),
which agrees with the above result. Other derivatives can be found
in a similar manner.
2.16 (a) Substituting Eqs. (2.139) and (2.166) gives
∂ ∂σ
∇2 σ = ∇ · ∇σ = gi i · gj j
∂x ∂x

= gi · gj ,i σ,j +gj σ,ij

= gi · −Γjki gk σ,j +gj σ,ij
= gij σ,ij −Γjki gik σ,j .

(b) In the following, let (r, θ, φ) = (x1 , x2 , x3 ) be orthogonal toroidal

coordinates (see Appendix B). Consider the first term on the right-
hand side of the equation in part (a). Summing over i and j,
dropping derivatives of σ with respect to θ = x2 (axisymmetry),
and setting gij = 0 for i = j (orthogonality) yields
gij σ,ij = g11 σ,11 +g33 σ,33 .
Next, consider the second term on the right. Since σ,2 = 0, we first
sum over i = 1, 2, 3 and j = 1, 3 to get
Γjki gik σ,j = Γ1k1 g1k σ,1 +Γ3k1 g1k σ,3 +Γ1k2 g2k σ,1 +Γ3k2 g2k σ,3
+Γ1k3 g3k σ,1 +Γ3k3 g3k σ,3 .
Now, summing over k = 1, 2, 3 and keeping only terms involving
g11 , g22 , and g33 gives
Γjki gik σ,j = Γ111 g11 σ,1 +Γ311 g11 σ,3 +Γ122 g22 σ,1 +Γ322 g22 σ,3
+Γ133 g33 σ,1 +Γ333 g33 σ,3 .
Appendix B provides the following relations for toroidal coordi-
nates:
g11 = 1, g22 = (b + r cos φ)−2 , g33 = r −2
Γ122 = −(b + r cos φ) cos φ, Γ133 = −r

Γ322 = r −1 (b + r cos φ) sin φ, Γ111 = Γ311 = Γ333 = 0
Combining the above relations and replacing (x1 , x2 , x3) by (r, θ, φ)

yields
∂ 2 σ 1 ∂σ 1 ∂ 2 σ cos φ ∂σ sin φ ∂σ
∇2 σ = + + + − .
∂r 2 r ∂r r 2 ∂φ2 (b + r cos φ) ∂r r(b + r cos φ) ∂φ
2.17 Differentiating a is left to the reader. Here, we consider the vector
b = b r gr + b θ gθ ,
where
1 1
br = , bθ =
r r2
gr = ex cos θ + ey sin θ = er
gθ = r(−ex sin θ + ey cos θ) = reθ .
Equations (2.150) and (2.151) give

b,j = bi |j gi
bi |j = bi ,j +bk Γijk .
Computing the covariant derivative bi |j requires the partial derivatives
1 2
br ,r = − 2 , bθ ,r = − 3 , br ,θ = bθ ,θ = 0,
r r
as well as the Christoffel symbols for cylindrical coordinates
1
Γrθθ = −r, Γθrθ = Γθθr = , other Γijk = 0,
r
which are provided in Appendix B. Hence, we get
1
br |r = br ,r +br Γrrr + bθ Γrrθ = − 2
r
1
b |r = b ,r +b Γrr + b Γrθ = − 3
θ θ r θ θ θ
r
1
br |θ = br ,θ +br Γrθr + bθ Γrθθ = −
r
1
bθ |θ = bθ ,θ +br Γθθr + bθ Γθθθ = 2 ,
r
and so the derivatives of b are
1 1
b,r = br |r gr + bθ |r gθ = − 2 gr − 3 gθ
r r
1
= − 2 (er + eθ )
r
1 1
b,θ = b |θ gr + bθ |θ gθ = − gr + 2 gθ
r
r r
1
= (eθ − er ).
r
2.19 Equation (2.154)1 gives
gij |k = gij ,k −glj Γlik − gil Γljk .
With the relations
Γlij = Γijk gkl , gik gjk = δ ji ,
as provided by (2.142) and (2.91), the above equation can be written
in the form
gij |k = gij ,k −glj Γikm gml − gil Γjkm gml
= gij ,k −Γikm δ m m
j − Γjkmδ i
= gij ,k −Γikj − Γjki .

Next, we use Eq. (2.144), i.e.,

2Γijk = gjk ,i +gki ,j −gij ,k
to obtain
1
Γikj + Γjki = 2 (gkj ,i +gji,k −gik ,j +gij ,k +gki ,j −gjk ,i )
= gij ,k ,
where gij = gji has been used. Plugging this result into the above
expression for gij |k yields
gij |k = 0.
2.20 (a) The analysis requires taking derivatives of the unit base vectors
eθ = −ex sin θ + ey cos θ
for cylindrical coordinates. Differentiating these vectors yields
∂er ∂eθ
= eθ , = −er
∂θ ∂θ
∂er ∂eθ ∂er ∂eθ
= = = = 0,
∂r ∂r ∂z ∂z
and clearly ez is constant. Substituting
∂ 1 ∂ ∂
∇ = er + eθ + ez
∂r r ∂θ ∂z
from Appendix B and using the product rule gives
∇2 φ = ∇ · ∇φ
∂ 1 ∂ ∂ ∂φ 1 ∂φ ∂φ
= er + eθ + ez · er + eθ + ez
∂r r ∂θ ∂z ∂r r ∂θ ∂z
= er · (er φ,rr −eθ r −2 φ,θ +eθ r −1 φ,rθ +ez φ,rz )
+r −1 eθ · (er ,θ φ,r +er φ,θr +eθ ,θ r −1 φ,θ
+eθ r −1 φ,θθ +ez φ,θz )
+ez · (er φ,rz +eθ r −1 φ,zθ +ez φ,zz ),
where we have dropped terms that involve vanishing derivatives of
the unit vectors. Finally, substituting for the non-zero unit-vector
derivatives and applying orthogonality to the dot products yields
∂ 2 φ 1 ∂φ 1 ∂2φ ∂2φ
∇2 φ = + + + 2.
∂r 2 r ∂r r 2 ∂θ 2 ∂z
2.21 (a) Substituting ∇ = g∂ and using the product rule gives

∇(T · a) = g∂(T · a)
= g[(∂T) · a + T · (∂a)]
T
= g[(∂T) · a + (∂a) · T ]
T
= (g∂T) · a + (g∂a) · T .
Replacing g∂ by ∇ yields
∇(T · a) = ∇T · a + ∇a · TT .
Chapter 3
3.1 For i, j = 1, 2, 3 = x, y, z, the stain tensor has the dyadic representa-

tions
E = Eij ei ej = Ēij ēi ēj .
The components relative to the rotated axes are given by
Ēij = ēi · E · ēj = ēi · (Eij ei ej ) · ēj ,
where
ēy = −ex sin θ + ey cos θ.
Partial answer:
Ēxx = Exx cos2 θ + 2Exy sin θ cos θ + Eyy sin2 θ
3.2 With Eqs. (2.12), the displacement gradient is given by
∂ eΘ ∂
∇u = eR + [uR (R)eR (Θ) + uΘ (R)eΘ (Θ)]
∂R R ∂Θ
1
= eR (uR ,R eR +uΘ ,R eΘ ) +eΘ (uR eR ,Θ +uΘ eΘ ,Θ )
R
1
= uR ,R eR eR +uΘ ,R eR eΘ + (uR eΘ eΘ −uΘ eΘ eR ).
R
With this result and I = eR eR + eΘ eΘ , Eq. (3.78)1 yields the defor-
mation gradient tensor
F = I + (∇u)T
uΘ uR
= (1 + uR ,R )eR eR − eR eΘ +uΘ ,R eΘ eR + 1 + eΘ eΘ ,
R R
and the Lagrangian strain tensor is

1
E = (FT · F − I)
2
1
= uR ,R + (u2R ,R +u2Θ ,R ) eR eR
2

uR 1 2 2
+ + (u + uΘ ) eΘ eΘ
R 2R2 R

1 uΘ 1
+ uΘ , R − + (uΘ ,R uR − uΘ uR ,R ) (eR eΘ + eΘ eR ).
2 R R
3.3 (a) In this problem, both the undeformed and deformed configura-
tions are described in the Cartesian coordinates X I and xi , re-
spectively, with base vectors
G I = G I = gi = g i = eI = ei .
Convected base vectors are given by

∂r
gI = = r,I ,
∂X I
where the deformed position vector is
r = xi gi = xi ei = (X 1 + 2X 2 X 3 )e1 + (X 2 − 2X 1 X 3 )e2 + X 3 e3 .
These relations yield the gI as
g1∗ = r,1∗ = e1 − 2X 3 e2
g2∗ = r,2∗ = 2X 3 e1 + e2
g3∗ = r,3∗ = 2X 2 e1 − 2X 1 e2 + e3 ,
where r,1∗ = ∂r/∂X 1 , etc. Equations (2.17) and (2.18) give the
contravariant base vectors
∗ g2 ∗ × g3 ∗
g1 = √ ∗ = (g∗ )−1/2 [e1 − 2X 3 e2 − (2X 2 + 4X 1 X 3 )e3 ]
g
∗ g3 ∗ × g1 ∗
g2 = √ ∗ = (g∗ )−1/2 [2X 3 e1 + e2 + (2X 1 − 4X 2 X 3 )e3 ]
g
∗ g1 ∗ × g2 ∗
g3 = √ ∗ = e3 ,
g
where
√
g∗ = g1∗ · (g2∗ × g3∗ ) = 1 + 4(X 3 )2 .
(b) In matrix form, the deformation gradients are

⎡ ⎤
i
1 2X 3 2X 2
∂x
[FIi] = = ⎣ −2X 3 1 −2X 1 ⎦ .
∂X I
0 0 1
According to the dyadic representation F = FIi gi GI = FIi ei eI ,
the scripts corresponding to ei and eI represent the row and
column, respectively. The reader can confirm that this result is
consistent with the tensor F = gI GI .
(c) Answer:
⎡ ⎤
1 + 4(X 3 )2 0 2X 2 + 4X 1 X 3
⎢ ⎥
[CIJ ] = ⎢
⎣ 0 1 + 4(X 3 2
) 4X 2 3
X − 2X 1 ⎥
⎦
2X 2 + 4X 1 X 3 4X 2 X 3 − 2X 1 1 + 4(X 1 )2 + 4(X 2 )2
(d) Answer: Λ = 1.39
(e) With g = det gij = det(gi ·gj ) = 1, G = det GIJ = det(GI ·GJ ) =
1, and j = det[FIi ], we get
g
J =j = 1 + 4(X 3 )2 = 1.04.
G
1/2
Since volume is invariant, J = I3 = det F gives the same result
independently of the coordinate system.
3.4 Answer for E = Eij ei ej :
⎡ ⎤
0 X3 X2
1⎣
[Eij ] = X3 X32 2X3 + X2 X3 ⎦
2
X2 2X3 + X2 X3 X22 + 4X32
To compute e using Eq. (3.80), we need to write u as a function of
the xi . For the given displacement
u(Xi ) = ui (Xi )ei = X2 X3 e1 + X32 e2 ,
the relation xi = Xi + ui yields
x1 = X1 + u1 = X1 + X2 X3
x2 = X2 + u2 = X2 + X32
x3 = X3 + u3 = X3 .
Inverting these equations gives
X1 = x1 − x2 x3 + x23
X2 = x2 − x23
X3 = x3 .
Substituting back into the displacement vector u(Xi ) gives

u(xi ) = (x2 x3 − x33 )e1 + x23 e2 .
Finally, this relation and
¯ = e1 ∂ + e2 ∂ + e3 ∂
∇
∂x1 ∂x2 ∂x3
are substituted into
e = 12 [∇u ¯ T − (∇u)
¯ + (∇u) ¯ T ].
¯ · (∇u)
Partial answer:
1 2
e13 = 2 (x2 − 3x3 )
e33 = − 12 [4x23 + (x2 − 3x23 )2 ]
3.6 (c) Substituting Eqs. (3.52) and (3.174) into (3.175)2 yields
⎡ ⎤ ⎡ ⎤
1 k 0 b − 12 kb 0
V = F · ΘT = ⎣ 0 1 0 ⎦ ⎣ 1 kb
2
b 0 ⎦
0 0 1 (e e ) 0 0 1 (e e )
i j i j
⎡ ⎤
b(1 + 12 k 2 ) 12 kb 0
= ⎣ 1
2 kb b 0 ⎦,
0 0 1 (e e )
i j
1 2 −1/2
where b = (1 + 4k ) .
(d) This calculation requires the strain tensors
k2 k
E = ey ey + (ex ey + ey ex )
2 2
k2 k
e = − ey ey + (ex ey + ey ex ).
2 2
3.7 Substituting u = uex + vey into Eqs. (3.178) and (3.179) yields
1

∗ u,x 2
(v,x +u,y )
E(ei ej ) = 1
2 (v,x +u,y ) v,y

0 − 12 (v,x −u,y )
Θ∗(ei ej ) = 1 .
2 (v,x −u,y ) 0
With these relations, Eq. (3.183) gives the approximate form for E.
Partial answer:
∗ ∗
Exx = Exx − Exy Θ∗xy + 21 Θ∗2
xy
3.9 In the undeformed body, unit vectors parallel to the given vectors are
2e1 + 2e2 + e3 e1 − 5e3
N1 = , N2 = √ .
3 26
Let Θ and θ represent the angles between the line segments before
and after deformation, respectively. With Eq. (3.108)2 , the cosines of
these angles are given by
cos Θ = N1 · N2
N1 · C · N2
cos θ = n1 · n2 = ,
Λ1 Λ2
where
C = I + 2E
Λi = (Ni · C · Ni )1/2 (i = 1, 2).
Computations using the given E yield
Δθ = θ − Θ = −11.7◦.
3.10 In terms of Eulerian quantities, Eq. (3.94)2 gives the stretch ratio
λ−2
(n) = n · B
−1
· n,
where
√
n = (ex + ey + ez )/ 3
B = F · FT .
To determine F, we use (3.78)2 , i.e.,
T
F−1 = I − (∇u)
¯ ,
where
∇¯ = ex ∂ + ey ∂ + e z ∂ .
∂x ∂y ∂z
For the given u, these relations yield
⎡ ⎤
(y/x)1/2 (x/y)1/2 0
1⎢ ⎥
F = ⎢
−1
0 y −1/2
0 ⎥.
2⎣ ⎦
(z/x)1/2 (x/z)1/2
(ei ej )
At the point (1,2,0.5), inverting this matrix gives

⎡ √ √ ⎤
2 −√ 2 0
F=⎣ 2 2 √0 ⎦ ,
√0 √
− 2/2 2/2 2 (e
i ej )
from which we compute B and the fiber stretch ratio

λ(n) = 1.12.
Finally, with the Eulerian strain tensor
e = 12 (I − B−1 ),
which is given by (3.28)2 , the Eulerian extensional strain for the fiber
is
enn = n · e · n = 12 (1 − n · B−1 · n)
= 1
2 (1 − λ−2
(n) ) = 0.104.
3.11 Equation (3.128) gives
da = J dA · F−1 ,
where dA and da are differential area vectors before and after defor-
mation, respectively. Dotting both sides with F gives
dA = J −1 da · F,
and substituting the component forms
dA = N dA, da = n da
yields
dA
b≡ N = J −1 n · F,
da
where the vector b is used below.
Computing b requires the unit normal to da given by
2ex + 3ey − 2ez
n= √ ,
17
as well as F and J = det F. In Cartesian coordinates, differentiating
the given expressions for xi (Xi ) yields
⎡ ⎤
4X + Y X 1
∂xi
F(ei ej ) = [Fij ] = =⎣ Y X − 2Y 2Z ⎦ ,
∂Xj
Z 0 1+X
which is evaluated at (X, Y, Z) = (−1, 2, 1). Then, we compute
J = det F = 3.
Finally, plugging all these results into the above equation for b gives
da
= |b|−1 = 0.658
dA
N = b|b|−1 = −0.905ey + 0.426ez .
3.12 In cylindrical polar coordinates, the specified deformation is defined

by
r = R, θ = Θ + φ(R), z = Z + w(R).
(a) For φ = 0, the deformation is axial shear without inflation, torsion,
or axial stretch.
(b) With the deformed position vector being r = rer + zez in cylindri-
cal coordinates, the transpose of the deformation gradient tensor
is
∂ eΘ ∂ ∂
FT = ∇r = eR + + eZ (Rer + [Z + w(R)]ez )
∂R R ∂Θ ∂Z
= eR (er + Rer ,R +w ez ) + eΘ er ,Θ +eZ ez ,
where prime denotes differentiation with respect to R. The chain
rule and Eqs. (2.12) give the derivatives
∂er ∂er ∂θ
= = φ eθ
∂R ∂θ ∂R
∂er ∂er ∂θ
= = eθ .
∂Θ ∂θ ∂Θ
Note that, since θ varies with the radius for points in the deformed
tube, the direction of er changes across the wall, i.e., er ,R = 0 if
φ (R) = 0. Combining these relations and then taking the trans-
pose yields
F = er eR + Rφ eθ eR + w ez eR + eθ eΘ + ez eZ .
With this result, the Lagrangian strain tensor is
1
2 (F · F − I)
T
E =
= 1
2
[(R2 φ2 + w 2 )eR eR + Rφ (eR eΘ + eΘ eR )
+w (eR eZ + eZ eR )].
3.13 (a) For r = rer + zez and the given mapping
r = R(1 + w cos mΘ), θ = Θ, z = Z,
we get
FT = ∇r
∂ eΘ ∂ ∂
= eR+ + eZ [r(R, Θ)er (Θ) + Zez ]
∂R R ∂Θ ∂Z
∂r eΘ ∂r ∂er
= eR er + er + r + eZ e z .
∂R R ∂Θ ∂Θ
Substituting the derivatives

∂er ∂er
= = eθ
∂Θ ∂θ
and taking the transpose yields
F = (1 + w cos mΘ)(er eR + eθ eΘ ) − wm sin mΘ er eΘ + ez eZ .
(b) The dilatation ratio is given by

dv
J= = det F = (1 + w cos mΘ)2 ,
dV
which is used to obtain the total volume
h0 /2 2π a0
v= J dV = J R dR dΘ dZ
V −h0 /2 0 0
of the deformed disk. With the undeformed disk volume V =

πa20 h0 , we find
v w2
=1+ .
V 2
(c) Partial answer:
EΘΘ = w cos mΘ + 12 w 2 cos2 mΘ + 12 w 2 m2 sin2 mΘ
The plots should show peaks in EΘΘ coinciding with peaks in the
magnitude of shear strain.
3.14 The following table lists results for the rotations and stretch ratios:
a0 −e1 + e2 e1 e2 N1
◦ ◦ ◦
θ −23.2 0 −31.0 −16.7◦
(θ 0 )str −6.5◦ 16.7◦ −14.3◦ 0◦
(θ 0 )rot −16.7◦ −16.7◦ −16.7◦ −16.7◦
θrot −23.2◦ 0◦ −31.0◦ −16.7◦
λ 0.762 1 1.17 1.34
Adding successive rotations in three dimensions generally requires

matrix multiplication, and rotation order matters. However, if the
rotations occur about the same axis, rotation angles can be added
in any order. The results in the table are consistent with this obser-
vation, as adding the rotation angles from U and Θ yields the total
rotation angle in each case, i.e., (θ 0 )str + (θ 0 )rot = θrot .
3.15 (a) For the given r, the matrix form of the deformation gradient
tensor is
⎡ ⎤
α1 β1 0
T
F = (∇r) = β 2⎣ α2 0 ⎦
0 0 α3 (e
i ej )
in Cartesian coordinates. Using this result yields

⎡ ⎤
α21 + β 22 α1 β 1 + α2 β 2 0
C = FT · F = ⎣ α1 β 1 + α2 β 2 α22 + β 21 0 ⎦
0 0 α23 (e e )
i j
⎡ ⎤
α21 + β 21 α1 β 2 + α2 β 1 0
B = F · FT = ⎣ α1 β 2 + α2 β 1 α22 + β 22 0 ⎦.
0 0 α23 (e e )
i j
(b) For an incompressible material, we set J = det F = 1 and solve

for α3 to obtain
α3 = (α1 α2 − β 1 β 2 )−1 .
Solving the eigenvalue problems of (3.133) then yields the princi-
pal stretch ratios
Λ1 = λ1 = 0.694
Λ2 = λ2 = 1.48
Λ3 = λ3 = 0.976,
where Λi ≡ Λ(Ni ) and λi ≡ λ(ni ) . These relations give the prin-
cipal strains Ei = (Λ2i − 1)/2 and ei = (1 − λ−2
i )/2, which have
the values
E1 = −0.259, E2 = −0.024, E3 = 0.590
e1 = −0.538, e2 = −0.025, e3 = 0.271.
The corresponding principal directions are
N1 = 0.465e1 − 0.885e2
N2 = −0.885e1 − 0.465e2
N3 = e3
n1 = 0.583e1 − 0.812e2
n2 = −0.812e1 − 0.583e2
n 3 = e3 .
(c) Equation (3.168)2 gives

⎡ ⎤
0.990 −0.138 0
Θ = ni Ni = ⎣ 0.138 0.990 0 ⎦ ,
0 0 1 (e
i ej )
and then Eqs. (3.175) yield
U = ΘT · F, V = F · ΘT .
(d) The characteristic equation for a 3-D problem, provided by

Eq. (2.82), has the form
−λ3 + I1 λ2 − I2 λ + I3 = 0.
Considering here only the eigenvalue problem for C, we set λ = Λ2

and (3.136) gives the strain invariants
I1 = tr C
2
I2 = 1
2 [(tr C) − tr C2 ]
I3 = det C.
A little matrix algebra shows that the above relations are con-
sistent with the usual characteristic equation of (3.133)1 given
by
det(C − Λ2 I) = −Λ6 + 3.61Λ4 − 3.58Λ2 + 1 = 0.
3.16 (a) After inserting the given expressions for xi (Xi , t) into the position
vector r = xi ei , differentiating with respect to time yields
v = ṙ = vL (Xi , t)e1 = vE (xi , t)e1 ,
where the Lagrangian and Eulerian forms of the velocity, respec-

tively, are
vL = ẋ1 = X1 X2 (1 + t)et
x1 x2 (1 + t)et
vE = .
1 + x2 tet
The equation for vE is obtained by inverting the equations for
xi (Xi , t) to get Xi (xi , t) and substituting the results into the
equation for vL .
(b) The acceleration vector is given by

dv ∂v ¯ = aL (Xi , t)e1 = aE (xi , t)e1 ,
a= = + v · ∇v
dt ∂t
where aL and aE represent respective Lagrangian and Eulerian
components given by
∂vL
aL = = X1 X2 (2 + t)et
∂t
∂vL ∂vE x1 x2 (2 + t)et
aE = + vE = .
∂t ∂x1 1 + x2 tet
(c) The rate-of-deformation tensor is defined by
D = 12 [∇v
¯ + (∇v)
¯ T ],
where the velocity gradient is
¯ ∂ ∂ ∂
∇v = e1 + e2 + e3 vE (x1 , x2 , t)e1
∂x1 ∂x2 ∂x3
∂vE ∂vE
= e1 e1 + e2 e1 .
∂x1 ∂x2
From here, it is a relatively simple matter to compute D.
3.17 For r = xi ei , the given xi are
x1 = X1 + X2 t
x2 = X2 + X1 t2
x3 = X3 .
After these equations are inverted to obtain Xi (xi , t), the velocity
components vi = ẋi can be written in the Eulerian form vi (xi , t).
With this result, we get
2 2 2 2 3
dθ
=
∂θ ¯ = x1 + x2 − (3x1 + 5x2 )t + 2x1 x2 t(1 + 2t) .
+ v · ∇θ
dt ∂t 1 − t3
For the particle initially located at Xi = (2, −3, 1), these equations
give θ̇ = 58.0 at time t = 1.5. The mathematical details are left to
the reader.
3.18 Partial answer: xi = (0.256, 1.73, 4.0), v1 = 2.15, a2 = 2.48
3.19 For combined extension, inflation, and torsion of a circular tube, sub-
stituting Eq. (3.223) into (3.208) gives
LT ¯
= ∇v
∂ eθ ∂ ∂
= er + + ez (ṙer + r θ̇eθ + żez ).
∂r r ∂θ ∂z
With the dependencies of (3.221), i.e.,
ṙ = ṙ(r, t), θ̇ = θ̇(z, t) ż = ż(z, t),
as well as the non-zero derivatives
er , θ = eθ , eθ ,θ = −er ,
differentiation yields
LT = ṙ,r er er + ṙr −1 eθ eθ + ż,z ez ez + θ̇(er eθ − eθ er ) + r θ̇,z ez eθ .
Since the matrix representation of L in Eq. (3.228) contains com-

ponents relative to the mixed dyadic basis {gi gj }, we need to replace
the unit base vectors in the equation above by the natural base vec-
tors for cylindrical coordinates. The needed relations, provided by
(3.60), have the form
gr = er , gθ = reθ , gz = ez
g r = er , gθ = r −1 eθ , g z = ez ,
where the gi satisfy the requirement gi · gj = δ ji for orthogonal coor-

dinates. With these relations, we get
L = ṙ,r er er + ṙr −1 eθ eθ + ż,z ez ez + θ̇ eθ er − θ̇ er eθ + r θ̇,z eθ ez

= ṙ,r gr gr + ṙr −1 gθ gθ + ż,z gz gz + r −1 θ̇ gθ gr
−r θ̇ gr gθ + θ̇,z gθ gz ,
which agrees with (3.228).

3.20 Answer: a = 0
3.21 (a) For J = J(F), the chain rule gives
∂J ∂J ∂F
= : .
∂t ∂F ∂t
Next, with J = det F, Table 2.5 (page 33) provides the identity
∂J ∂ −T
= det F = ( det F)F = JF−T ,
∂F ∂F
which is substituted into the first equation to obtain
J˙ = JF−T : Ḟ.
(b) Dotting both sides of

−1
L = Ḟ · F
with F gives Ḟ = L · F. After inserting this relation into the last
equation in part (a), we use a series of manipulations to obtain
J˙ = JF−T : (L · F) = JL : (F−T · FT )
= J L : I = J tr L
¯ T = J tr (∇v)
= J tr (∇v) ¯
¯ · v.
= J∇
Here, we have used the next-to-last formula in Table 2.2 (page
22) in the first line, the second formula in Table 2.4 (page 32) in
the second line, Eq. (3.208) in the third line, and the first identity
in Table 2.7 (page 58) in the last line.
Chapter 4
4.1 (a) With the unit normal to the plane given by

∇F 1
n= = (2ex + 2ey + ez ),
|∇F| 3
the true traction vector is
1
T(n) = n · σ = (130ex + 205ey + 110ez ).
3
(b) For T(n) = σ nn n + σ ns s, n and s are orthogonal, so the normal
stress is
σ nn = n · T(n) = 86.7.
The shear stress can now be computed using the relation
T(n) · T(n) = σ 2nn + σ 2ns ,
which, with T(n) known from part (a), yields
σ ns = 19.5.
(c) Partial answer:
σ 1 = 10.6, σ 2 = 70.7
n1 = −0.728ex + 0.605ey − 0.323ez

n2 = 0.657ex + 0.479ey − 0.583ez
4.2 Answer:
a q a a
σ xx = 1− , σ xy = , σ zz =
p p p r
4.3 (a) For a plane parallel to the x3 -axis, the unit normal must have the
form
n =n1 e1 + n2 e2 ,
giving n · e3 = 0. For the given σ, the true traction vector is
T(n) = n · σ = an1 e1 − bn2 e2 + (dn1 − en2 )e3 .
If this vector is tangent to the plane, we must have
n · T(n) = an21 − bn22 = 0,
and, since n is a unit vector,
n · n = n21 + n22 = 1.
Solving these equations for n1 and n2 yields
b a
n1 = , n2 = .
a+b a+b
(b) For rotation θ about the x1 -axis, the rotated system x̄i has base
vectors
ē1 = e1
ē2 = e2 cos θ + e3 sin θ
ē3 = −e2 sin θ + e3 cos θ.
The relation σ̄ ij = ēi · σ · ēj gives
σ̄ 13 = d cos θ
σ̄ 23 = (b + c) sin θ cos θ + e(sin2 θ − cos2 θ).
4.4 The characteristic equation for the given stress matrix is
rr
σ̂ − λ 0 0

det(σ − λI) = 0 1−λ 2 = 0.

0 2 σ̂ − λ
φφ
In principal coordinates, since the principal stresses σ̂ i are the only

non-zero terms in the matrix, the characteristic equation becomes

σ̂ 1 − λ 0 0

det(σ − λI) = 0 σ̂ 2 − λ 0 = 0.

0 0 σ̂ 3 − λ
As the characteristic equation is geometrically invariant, these rela-
tions can be equated to obtain a cubic equation to solve for λ. Equat-
ing the coefficients of the first and second powers of λ yields a pair of
equations to solve for σ̂ rr and σ̂ φφ , giving σ̂ rr = σ̂ φφ = 1.
4.5 In an orthogonal coordinate system, the physical components of stress

σ̂ ij correspond to the components of the Cauchy stress tensor when
written in terms of unit base vectors, i.e., σ = σ̂ ij ei ej . In dyadic
form, the components of the given stress tensor
σ = xy ex ex − y(ex ey + ey ex ) + (x − y)ey ey + (y3 + 2z)ez ez
are the physical components in Cartesian coordinates. To determine
the physical components in cylindrical polar coordinates, which also
are orthogonal, we use the relations
x = r cos θ, y = r sin θ
and the unit base vectors
eθ = −ex sin θ + ey cos θ.
Now, we can either (1) substitute these equations into the above
expression for σ and collect terms in the dyadic basis for cylindrical
coordinates or (2) use the equations for x(r, θ) and y(r, θ) along with
σ̂ ij = ei · σ · ej = ei · σ · ej ,
where ei = ei for orthogonal coordinates and i, j = r, θ, z. Both ways
yield the same results.
Partial answer:
σ̂ rr = r 2 sin θ cos3 θ − r sin2 θ(sin θ + cos θ)
σ̂ rθ = −r 2 sin2 θ cos2 θ + r sin2 θ(sin θ − cos θ)
4.6 Since the dyadic forms of t and F are expressed in terms of unit vec-
tors, their components are physical components. Using either dyadic
or matrix analysis, Eq. (4.28) gives the physical components of the
Cauchy stress tensor as
⎡ ⎤
0.33 0 0
σ = J −1 F · t = ⎣ 0 1.11 −0.33 ⎦
0 −0.33 0 (ei ej )
relative to cylindrical coordinates. For the dyadics

σ = σ ij gi gj = σ i·j gi gj = σ ·j i i j
i g gj = σ ij g g ,
Appendix B gives the base vectors

gr = er , gθ = reθ = 5eθ , gz = ez
r θ −1
g = er , g =r eθ = eθ /5, g z = ez
at the specified point (r = 5). These relations give the following

components for the first two cases:
⎡ ⎤
0.33 0 0
σ ij = gi · σ · gj = ⎣ 0 1.11/25 −0.33/5 ⎦
0 −0.33/5 0 (g g ) i j
⎡ ⎤
0.33 0 0
σ i·j = gi · σ · gj = ⎣ 0 1.11 −0.33/5 ⎦ .
0 −(0.33)(5) 0 (g j)
ig
As expected, the tensor σ is symmetric. However, while the matrices

for the physical components and the σ ij are symmetric, the matrix
for σ i·j is asymmetric [see comments after Eq. (2.41)].
4.7 (a) The shear stress τ o applied at r = b must be balanced by an equal
and opposite moment applied by the shear stress τ i applied at
r = a. The forces on differential surface elements at r = a and
r = b, respectively, are
dfi = τ i a dθdz, dfo = τ o b dθdz.
For a cylindrical segment of length , the total moment about the
center of the tube is

'
M0 = b dfo − a dfi
2π
= (b2 τ o dθdz − a2 τ i dθdz) = 0,
0 0
which gives
b2
τi =
τ o.
a2
(b) A similar analysis for an annulus with inner radius r and outer
radius b cut from the tube yields
b2
τ o = σ rθ .
τ (r) =
r2
Since this is the only non-zero stress component, the Cauchy
stress tensor is
σ = τ(r)(er eθ + eθ er )
for any point in the tube.
(c) Partial answer: σ xx = −2τ (r) sin θ cos θ
4.8 (a) In matrix form, the second Piola-Kirchhoff stress tensor can be
written
⎡ ⎤
2X 1 −X 1 X 2 0
s = ⎣ −X 1 X 2 3X 1 + X 2 0 ⎦ .
0 0 0 (e e )
i j
With r = xi ei and xi (X I ) given, the deformation gradient tensor

is
⎡ ⎤
1 0.2X 3 0.2X 2
F = (∇r) = ⎣ 0 0.4X 3 ⎦ .
T
1
0 0 1 (e e )i j
At the point X I = (−1, 2, 1), Eq. (4.28) yields

⎡ ⎤
−1.24 1.8 0
σ = J −1 F · s · FT = ⎣ 1.8 −1 0 ⎦ ,
0 0 0 (e
i ej )
where J = det F = 1.
(b) To extract the required stress components, we need the base vec-
tors GI in the undeformed configuration and the convected base
vectors gI and gI in the deformed configuration. The first two
are provided by the relations
∂R ∂r
GI = , gI = ,
∂X I ∂X I
and Eqs. (2.18) give the gI . The results are
G I = eI (I = 1, 2, 3)
g1 ∗ = e1
g2∗ = 0.2X 3 e1 + e2
g3∗ = 0.2X 2 e1 + 0.4X 3 e2 + e3
∗
g1 = e1 − 0.2X 3 e2 + [0.08(X 3)2 − 0.2X 2 ]e3
∗
g2 = e2 − 0.4X 3 e3
∗
g3 = e3 .
For X I = (−1, 2, 1), these relations yield the following compo-
nents for σ:
⎡ ⎤
−2 2 0
∗ ∗
σI J = gI · σ · gJ = ⎣ 2 −1 0 ⎦
0 0 0 (g g )
I J
⎡ ⎤
−1.6 1.8 0
∗
σ ·J
I∗ = gI · σ · gJ = ⎣ 1.7 −0.64 0 ⎦
0.16 0.32 0 (gI g
J)
⎡ ⎤
−1.6 1.8 0
∗
σ ·J
I = GI · σ · g J = ⎣ 2 −1 0 ⎦
0 0 0 (GI g
J)
(c) Additional hint: Equations

√ (2.45)1 and (3.23) may be useful.
(d) With N = (3e1 − e3 )/ 10 and (4.29)1 giving
t = JF−1 · σ,
the pseudotraction vector is
T(N) = N · t = −1.5e1 + 1.9e2.
4.9 (a) Answer:
t = JF−1 · σ = −3e1 e1 + 13e1 e2 + 16e2 e1 + 10e2 e2 + 14e3 e3
s = t · F−T = 0.29e1e1 + 4.1(e1 e2 + e2 e1 ) + 0.57e2 e2 + 14e3 e3
(b) Using dyadics to compute stress components requires the covari-
ant base vectors gi , which are given, as well as the contravariant
base vectors gi . Equations (2.18) yield
g1 = −0.2e1 − 0.4e2 + 0.6e3
g2 = 0.6e1 + 0.2e2 − 0.8e3
g 3 = e3 .
The first two sets of stress components are
⎡ ⎤
1.4 −1.4 0.6
σ ij = gi · σ · gj = ⎣ −1.4 0.8 −0.8 ⎦
0.6 −0.8 1.0 (gigj )
⎡ ⎤
5.8 0.6 −2.4
σ i·j = gi · σ · gj = ⎣ −8.4 −3.8 1.2 ⎦ .
0 0 1 (g i g j)
Computing σ ·ij is left to the reader.

In the matrix method, we use Eqs. (2.70) to transform the
given stress components relative to the Cartesian base vectors ei
to the basis defined by the gi . Hence, Eq. (2.48)1 provides the

coordinate transformation tensor
⎡ ⎤
1 2 1
A = g1 e1 + g2 e2 + g3 e3 = ⎣ −3 −1 1 ⎦ ,
0 0 1 (e
i ej )
and (2.51) gives B = A−T . Since the dyadic form of σ is given

in terms of orthogonal (Cartesian) unit vectors, the components
are physical stress components, which we denote σ̂ ij . Because it
does not matter whether the scripts are upstairs or downstairs
in Cartesian coordinates, we use σ̂ ij for all types of Cartesian
components. Thus, with Eqs. (2.70), components relative to the
{gi } basis are computed using
[σ ij ] = [Bj·i ]T [σ̂ij ][Bj·i ]
[σ ·j j T ·i
i ] = [A·i ] [σ̂ ij ][Bj ]
[σ i·j ] = [Bj·i ]T [σ̂ij ][Aj·i],

which give the same results as the dyadic analysis.
While [σ ij ] is symmetric, the other two matrices are not. How-
ever, [σ ·i
j ] = [σ·j ] , satisfying Eq. (2.41)3 for σ = σ . Thus, all
i T T
these matrices are consistent with σ being a symmetric tensor.

(c) With gij = gi · gj , Eq. (4.46) gives the physical components
⎡ ⎤
5.8 0.85 −4.4
g (ii)
σ̂ i·j = σ i·j = ⎣ −5.9 −3.8 1.5 ⎦
g(jj)
0 0 1
relative to the {gi } basis.
(d) For σ = σ ij gi gj = σ i·j gi gj , both σ ij and σ i·j act on the faces
with normal gi . On the positive faces, the positive directions of
σ ij and σ i·j are gj and gj , respectively [see discussion following
Eq. (4.49)]. For this problem, components on two faces in the
plane normal to g1 × g2 are shown in Fig. P4.9.
I
4.10 With F = F·J GI GJ and J = det F = det F·J I
by (2.77), the answer
is
∗ ∗
I −1 K J IL
σI J
= (det F·J ) F·L F·K s .
4.12 Answer:
fˆr = −3r, fˆθ = fˆz = 0
r=a: σ̂ rr = a2 , σ̂ rθ = σ̂ rz = 0
z = 0, d : σ̂ zz = σ̂ zr = σ̂ zθ = 0
g2
g2 V2.1
V21
g1
g1
V2.2
V22 V1.2
V 12
V11 V1.1
Fig. P4.9. Stress components on differential element (Problem 4.9).
4.13 (a) To compute ∇ ¯ · σ for this problem, we need derivatives of the

unit vectors (see Appendix B)
eφ = −ex sin φ + ey cos φ.
The required derivatives are given by
eθ ,r = 0, eθ ,θ = −er , eθ ,φ = eφ cos θ
eφ ,r = 0, eφ ,θ = 0, eφ ,φ = −er (sin θ)−1 + ez cot θ.
With the given stress dyadic, the equilibrium equation becomes

¯ · σ = er ∂ + eθ ∂ + eφ ∂ · σ̂ θφ (eθ eφ + eφ eθ ) = 0.
∇
Carrying out the differentiations and taking dot products gives
∂ σ̂ θφ
+ 2σ̂ θφ cot θ = 0,
∂θ
which, being the coefficient of eφ , is the only non-trivial equilib-
rium equation.
(b) In terms of tensor components, Eq. (4.88)1 gives
σ ij |i = σ ij ,i +σ kj Γ̄iik + σ ik Γ̄jik = 0.
Using the Christoffel symbols listed in Appendix B for spherical
coordinates and noting
σ 23 = σ 32 = σ θφ ,
we get
∂σ θφ
+ 3σ θφ cot θ = 0
∂θ
for j = 3.
(c) At first glance, the above results appear to differ by the constant
(2 vs. 3) in the second term, but these equations are for two differ-
ent stress components (σ̂ θφ vs. σθφ ). For orthogonal coordinates,
Eq. (4.47) gives
√
σ̂ ij = σ ij g(ii)g(jj).
With the g(ii) provided in Appendix B, this relation becomes
σ̂ θφ σ̂ θφ
σ θφ = √ = 2 .
gθθ gφφ r sin θ
After substitution into the equilibrium equation from part (b),
differentiation yields
∂ σ̂ θφ
+ 2σ̂ θφ cot θ = 0,
∂θ
which agrees with the result from part (a).
4.14 (a) With the given position vectors
R = ReR + ZeZ
r = R(1 + b cos2 Θ)eR + ZeZ
and the gradient operator
∂ 1 ∂ ∂
∇ = eR + eΘ + eZ ,
∂R R ∂Θ ∂Z
T
the relation F = (∇r) gives deformation gradient tensor
F = (1 + b cos2 Θ)(eR eR + eΘ eΘ ) − 2b sin Θ cos ΘeR eΘ + eZ eZ .
In matrix form, this tensor can be written
⎡ ⎤
FRR FRΘ 0
F = ⎣ 0 FΘΘ 0 ⎦
0 0 FZZ (e e )
I J
⎡ 2
⎤
1 + b cos Θ −2b sin Θ cos Θ 0
= ⎣ 0 1 + b cos2 Θ 0 ⎦ .
0 0 1 (e
I eJ )
(b) Deformation transforms the circumferential unit vector eΘ at a

point on the undeformed surface into a circumferential vector
tangent to the deformed surface. Therefore, the unit vector s is
given by
⎡ ⎤
FRΘ
F · eΘ 1 ⎣ FΘΘ ⎦
s= = sR eR + sΘ eΘ + sZ eZ = 2
|F · eΘ | 2
FΘΘ + FRΘ 0
in the {eI } basis. Because the deformed surface normal n is

normal to both s and eZ , we get
⎡ ⎤
FΘΘ
1 ⎣ −FRΘ ⎦ .
n = s ×eZ = nR eR +nΘ eΘ +nZ eZ = 2 2
FΘΘ + FRΘ
0
This result also can be obtained using the formula n = N ·
F−1 /|N · F−1 | with N = eR , as given by Eq. (3.128).
Equations (3.10) and (3.11) give
GI = R,I gI = r,I ,
which yield the undeformed base vectors
G R = eR , GΘ = ReΘ , G Z = eZ
and the convected base vectors
gR = FRR eR
gΘ = R(FRΘ eR + FΘΘ eΘ )
gZ = eZ ,
where the FIJ are found in part (a). With these results,
Eqs. (2.18) give the contravariant base vectors
G R = eR , GΘ = R−1 eΘ , G Z = eZ
gR = (FRR FΘΘ )−1 (FΘΘ eR − FRΘ eΘ )

gΘ = (RFΘΘ )−1 eΘ
g Z = eZ .
(c) The stress at the boundary must match the applied surface trac-
tion, i.e.,
T(n) = n · σ = −p(Θ)n + τ(Θ)s
at the material coordinate R = a0 . We need to write this equation
in terms of Cauchy stress components given by the matrices
⎡ RR ⎤
σ σ RΘ σ RZ
σ = ⎣ σ ΘR σ ΘΘ σ ΘZ ⎦
σ ZR σ ZΘ σ ZZ (G G )
I J
⎡ ∗ ∗ ∗ ∗ ∗ ∗ ⎤
σR R σR Θ σR Z
= ⎣ σΘ R σΘ Θ σΘ Z ⎦ .
∗ ∗ ∗ ∗ ∗ ∗
∗ ∗ ∗ ∗ ∗ ∗
σ Z R σZ Θ σZ Z (g g I J)
Matrix multiplication requires that all components have consis-

tent base vectors. Here, we work in terms of the unit vectors
{eR , eΘ , eZ }. As shown by the above relations, GΘ and gΘ carry
an extra R. In terms of σ IJ , the true traction vector becomes
⎡ RR ⎤
σ Rσ RΘ σ RZ
T(n) = n · σ = [nR , nΘ , 0] ⎣ Rσ ΘR R2 σ ΘΘ Rσ ΘZ ⎦ ,
σ ZR Rσ ZΘ σ ZZ (e e ) I J
yielding the boundary condition

⎡ ⎤ ⎡ ⎤ ⎡ ⎤
nR σ RR + nΘ a0 σ ΘR nR sR
⎣ nR a0 σ RΘ + nΘ a20 σ ΘΘ ⎦ = −p(Θ) ⎣ nΘ ⎦ +τ (Θ) ⎣ sΘ ⎦
nR σ RZ + nΘ a0 σ ΘZ (e )
0 (e ) 0 (e
I I I)
at R = a0 . Note that this matrix equation provides three scalar

equations to be satisfied on the boundary.
∗ ∗
For the convected stress components σ I J , it is convenient to
first use dyadic analysis, rather than matrix algebra. Since the
relations in part (b) show that n = gR /|gR|, we get
gR ∗ ∗ ∗ ∗ ∗ ∗
n·σ = · (σ R R gR gR + σ R Θ gR gΘ + σ R Z gR gZ + . . .)
|g |
R
∗
R∗ ∗
Θ∗ ∗
Z∗
= |gR|−1 (σ R gR + σ R gΘ + σ R gZ ).
The dots represent terms that vanish after taking the dot product
(gI · gJ = δ IJ ). To convert to the unit basis, we substitute the
equations from part (b) for gI to obtain the boundary condition
⎡ ∗ ∗ ∗ ∗
⎤
FRR σ R R + a0 FRΘ σ R Θ
FRR FΘΘ ⎢ ⎢ R∗ Θ∗
⎥
⎥
⎣ a 0 F ΘΘ σ ⎦
2 2
FΘΘ + FRΘ ∗ ∗
σR Z
(eI )
⎡ ⎤ ⎡ ⎤
nR sR
⎢ ⎥ ⎢ ⎥
= −p(Θ) ⎢ ⎥ ⎢
⎣ nΘ ⎦ + τ (Θ) ⎣ sΘ ⎦
⎥
0 0
(eI ) (eI )
at R = a0 . As a check, the reader may find it useful and illumi-

nating, as well as great fun, to derive this relation by transforming
∗ ∗
σ IJ into σ I J (or vice versa) and showing that the two expres-
sions for the boundary conditions are equivalent.
Chapter 5
5.1 (a) Summing forces on the element shown in Fig. 5.5 yields Newton’s
law in the form
−σ da + (σ + σ dx)da + f da dx = ρv̇ da dx,
where prime indicates differentiation with respect to x. Simplify-
ing leads to the 1-D equation of motion
σ + f = ρv̇.
(b) The kinetic and internal energies for the element are
1 2
K = 2 ρv da dx
U = ρu da dx,
where v is the average velocity and u the internal energy per unit
mass. The mechanical power input is
P = −(σ da)v + (σ + σ dx) da (v + v dx) + (f da dx)v,
where the minus sign in the first term comes from σ and v being
defined as positive in opposite directions at the left end, and the
body force acts through the average element velocity. Expanding
and neglecting terms of order greater than dx (as dx → 0) yields
P = [(σv) + fv] da dx.
Heat is added at the rate
Q = q da − (q + q dx) da + ρr da dx
= (−q + ρr) da dx.
Substituting these relations into K̇ + U̇ = P + Q yields
ρvv̇ + ρu̇ = (σv) + fv − q + ρr
or
ρu̇ = σv − q + ρr + (σ + f − ρv̇) v.
The term in parentheses vanishes by the equation of motion de-
rived in part (a), giving
ρu̇ = σv − q + ρr,
which is easily shown to be the 1-D form of (5.18)1 .
5.2 In terms of the given stress tensor τ , we need to show that the stress
power is given by
Ps = J −1 τ : U̇ = 12 J −1 (s · U + U · s) : U̇.
Table 2.2 provides the identities
A : (B · C) = B : (A · CT )
A : (B · C) = C : (BT · A),
which are used to modify the first and second terms, respectively, on
the right-hand side of the above expression for Ps . Setting B = s and
C = U in the first term, B = U and C = s in the second term, and
A = U̇ in both terms yields
1 −1 T
Ps = 2J [s : (U̇ · UT ) + s : (U · U̇)]
1 −1
= 2J [s : (U̇ · U + U · U̇)],
where the second line follows from U being a symmetric tensor. Next,
we use Eq. (3.153) to write the Lagrangian strain tensor as
E = 12 (C − I) = 12 (U · U − I),
and differentiating with respect to time gives
Ė = 12 (U̇ · U + U · U̇).
Comparing this equation with the last expression above for Ps shows
that
Ps = J −1 s : Ė,
which agrees with the last of (5.22), thus proving the given expression
for Ps .
5.3 (a) As viewed by observers in frames A and A∗ , respectively, the
Eulerian strain tensor is given by [see Eq. (3.28)2 ]
e = 1
2
(I − B−1 )
e∗ = 1 ∗
2
[I − (B∗ )−1 ],
where
B−1 = F−T · F−1
(B∗ )−1 = (F∗ )−T · (F∗ )−1 .
The transformation relation for F, provided by (5.55), is
F∗ = Q · F,
which gives
(F∗ )−1 = F−1 · Q−1 = F−1 · QT
(F∗ )−T = (F−1 · QT )T = Q · F−T .
Plugging these relations into the above equation for (B∗ )−1 yields
(B∗ )−1 = (F∗ )−T · (F∗ )−1 = (Q · F−T ) · (F−1 · QT )
= Q · B−1 · QT .
Under a change in frame, the identity tensor transforms like a
standard second-order tensor, i.e.,
I∗ = Q · I · QT .
Finally, substituting for I∗ and (B∗ )−1 into the above equation
for e∗ yields
e∗ = 12 (Q · I · QT −Q · B−1 · QT ) = Q · e · QT .
Answer for Lagrangian strain tensor: E∗ = E
(b) To determine the transformation relation for σ, we use the sym-
metry property of this tensor and the Cauchy stress formula
(4.23) to obtain
T = n · σ = σ T · n =σ · n
T∗ = σ ∗ · n∗,
where we set T(n) = T. The vectors T and n transform like any
vector, and the equation for T∗ gives
Q · T = σ ∗ · (Q · n)
or, after dotting both sides with Q−1 = QT ,
T = (QT · σ ∗ · Q) · n.
According to the Cauchy stress formula, the term in parentheses
corresponds to σ, i.e.,
σ = QT · σ ∗ · Q,
which gives the transformation relation
σ ∗ = Q · σ · QT .
Answers for other stress tensors: t∗ = t · QT , s∗ = s

(c) The length element dr in frame A is seen as the vector

dr∗ = Q · dr
in A∗ . Since QT = Q−1 , this relation also can be written in the
form
dr = QT · dr∗ = dr∗ · Q.
Equations (2.163)1 and (3.18)2 show that these elements are re-
lated by
∂r ¯ ∗ r,
dr = dr∗· ∗ = dr∗ · ∇
∂r
where ∇¯ ∗ ≡ ∂/∂r∗ . Comparing the last two equations shows
that
¯ ∗ r.
Q=∇
Now, the transformation
¯∗ = Q·∇
∇ ¯
can be confirmed by the manipulations

¯ ∗ r = Q · ∇r
∇ ¯
∂r
= Q· = Q · I = Q.
∂r
Thus, the gradient operator ∇¯ transforms as a vector under a
change of frame.
5.4 Relative to the coordinate system fixed in frame A, the base vectors
in frame A∗ , which rotate with the bar, are (see Fig. 5.6)
ēy = −ex sin θ + ey cos θ
ēz = ez .
Since the loaded bar rotates as a rigid body and frame A∗ rotates
with it, the rotation tensor Θ for the bar is equal to the rotation
tensor Q for the frame. Therefore, Eq. (3.168)2 gives
⎡ ⎤
cos θ − sin θ 0
Q = Θ = ēi ei = ⎣ sin θ cos θ 0 ⎦ .
0 0 1 (e e )
i j
With the stress tensor initially given by

σ = σ 0 ex ex
in frame A, the stress tensor in the rotated bar as seen in frame A∗

is
σ ∗ = Q · σ · QT
= σ 0 [cos2 θex ex + sin θ cos θ(ex ey + ey ex ) + sin2 θey ey ]
relative to the fixed system in A (see solution to Problem 5.3b). With
the above relations for the ēi , it is easy to show that this expression
becomes
σ ∗ = σ 0 ēx ēx
relative to the rotated system in A∗ . In other words, the stress dyadic
remains constant with respect to the bar as it rotates.
5.5 (a) For E13 = E23 = E33 = 0, Eqs. (3.142) give
I1 = 3 + 2Eii = 3 + 2(E11 + E22 )

1 + 2E11 2E12 0

I3 = det[δij + 2Eij ] = 2E21 1 + 2E22 0 = 1 + Δ,
0 0 1
where
Δ = 2(E11 + E22) + 4(E11 E22 − E12E21 ).
For |Δ| << 1, expanding I3−β in a Taylor series and omitting
2
terms of order higher than Eij yields
I3−β = (1 + Δ)−β = 1 − βΔ + 12 β(β + 1)Δ2 − · · ·
2 2
= 1 − 2β(E11 + E22 ) + 2β(β + 1)(E11 + E22 )
+4β 2 E11E22 + 4βE12 E21 .
Note that, if three terms are not retained in the series expansion
involving Δ, some quadratic terms would be lost. Substituting
these expressions into the given strain-energy density function
and simplifying gives

μ 1 −β
W = I1 − 3 + I −1
2 β 3
2 2
= μ[(1 + β)(E11 + E22 ) + 2βE11 E22 + 2E12E21 ].
(b) For small deformation, all stress tensors are essentially equal, and
Eq. (5.84)5 gives
∂W
sij ≈ σ ij = ,
∂Eij
which provides the stress components

∂W ν
σ 11 = = 2μ E11 + (E11 + E22 )
∂E11 1 − 2ν

∂W ν
σ 22 = = 2μ E22 + (E11 + E22 )
∂E22 1 − 2ν
∂W
σ 12 = = 2μE21
∂E12
∂W
σ 21 = = 2μE12
∂E21
after setting β = ν/(1 − 2ν). Investigating whether these equa-
tions are consistent with Hooke’s law is left for the reader’s en-
joyment.
5.7 Here, we consider only N. Equation (5.152)2 gives
2

N= Nα α3 ,
α=1
where
Nα = F · (e3 eα + eα e3 ) · FT .
Substituting
F = gI G I
from (3.22)1 and expanding yields
Nα = (gI GI ) · (e3 eα + eα e3 ) · (GJ gJ )
= gI [(GI · e3 )(eα · GJ ) + (GI · eα )(e3 · GJ )]gJ .
Using the relations
eα · GI = AIα , e3 · GI = AI3
as given by Eq. (5.119)2 , we find
Nα = [AI3 AJα + AIα AJ3 ]gI gJ .
∗
J∗
With this result, the components of N = N I gI gJ are
2

∗
J∗
NI = (gI · Nα · gJ ) α3
α=1
2

= (AI3 AJα + AIα AJ3 ) α3 ,
α=1
which agrees with N IJ in (5.149)4 .

5.8 (b) In terms of Cartesian components, the equation from part (a)
gives
∂W
sIJ = 2 − p(CIJ )−1
∂CIJ
∂W ∂I1 ∂W ∂I2
= 2 + − p(CIJ )−1
∂I1 ∂CIJ ∂I2 ∂CIJ
∂I1 ∂I2
= 2 b1 + b2 − p(CIJ )−1
∂CIJ ∂CIJ
for the given W . Here, (CIJ )−1 represents the IJ component of
the inverse of C, and (3.142) provides the strain invariants
I1 = CKK
1
I2 = 2 (CKK CLL − CKL CLK ).
The required derivatives are given by
∂I1 ∂CKK
= = δ IK δ JK = δ IJ
∂CIJ ∂CIJ
∂I2 1 ∂CKK ∂CLL ∂CKL
= CLL + CKK − CLK
∂CIJ 2 ∂CIJ ∂CIJ ∂CIJ
∂CLK
− CKL
∂CIJ
1
= 2 (δ IK δ JK CLL + CKK δ IL δ JL − δ IK δ JL CLK
− CKLδ IL δ JK )
1
= 2 (δ IJ CLL + CKK δ IJ − CJI − CJI )
= CKK δ IJ − CIJ ,
in which CJI = CIJ because C is symmetric. Plugging these into
the above equation for sIJ yields
sIJ = 2b1 δ IJ + 2b2 (CKK δ IJ − CIJ ) − p(CIJ )−1 .
5.9 Relative to principal axes, Eq. (3.135)4 provides the Lagrangian strain
tensor in the spectral form
3
E= Ei Ni Ni ,
i=1
where Ei and Ni are principal strains and directions, respectively.
The constitutive relation (5.78)3 for the second Piola-Kirchhoff stress
tensor yields
3 3
∂W ∂W ∂W ∂Λi
s = = Ni Ni = Ni Ni .
∂E i=1
∂E i i=1
∂Λi ∂Ei
Principal stretch ratios and strains are related by Eq. (3.134)1 , i.e.,
Λ2i = 1 + 2Ei ,
giving
∂Λi 1
= .
∂Ei Λi
Substitution into the above equation for s gives

3
1 ∂W
s= Ni Ni .
Λi ∂Λi
i=1
The relations for t and σ can be found using (4.28), (4.29)1, and the
spectral representation for F. Details are left to the reader.
5.10 Answer:
1
σ2 = μ 1 −
Λ1 Λ32 Λ3
Λ1 Λ3
t2 = Λ 1 Λ 3 σ 2 , s2 = σ2
Λ2
5.11 It is convenient to express the invariants in W in terms of the defor-

mation tensor C. For
μ 1/2
W = (I2 I3−1 + 2I3 − 5),
2
we need the derivatives
∂I2 ∂I3
= I1 I − C, = I3 C−1 ,
∂C ∂C
as provided by Eqs. (5.129)2 and (5.130). Using these relations and
E = 12 (C − I), we find

∂W ∂W ∂I2 −1 ∂I3 −1/2 ∂I3
s = =2 =μ I3 + I2 −I3−2 + I3
∂E ∂C ∂C ∂C ∂C

I1 1 1/2 I2
= μ I − C + I3 − C−1 .
I3 I3 I3
1/2
Next, computing the Cauchy stress tensor using (4.28) with J = I3 ,
C = FT · F, and B = F · FT , yields
−1/2
σ = I3 F · s · FT

−1/2 I1 1
= μI3 F · I · FT − F · C · FT
I3 I3

1/2 I2 −1

+ I3 − F·C ·F T
I3

−1/2 I1 1
= μI3 F · FT − F · FT · F · FT
I3 I3

1/2 I2
+ I3 − F · F−1 · F−T · FT
I3

−1/2 I1 1
= μI3 F · FT − F · FT · F · FT
I3 I3

1/2 I2
+ I3 − F · F−1 · F−T · FT
I3
# $
1 2
I2
= μ 3/2 I1 B − B + 1 − 3/2 I .
I3 I3
To obtain the final form, we use the Cayley-Hamilton theorem (2.84)
for the tensor B, i.e.,
B3 − I1 B2 + I2 B − I3 I = 0,
which gives
I1 B − B2 = I2 I − I3 B−1 .
Substituting this expression into the above equation for σ and sim-
plifying gives

−1/2 −1
σ = μ I−I3 B .
5.12 Rather than showing all details for each step, we list the equations
needed (roughly in order) for solving each part of the problem using
computer software, e.g., Matlab. All matrix components are Carte-
sian, so there is no need to specify the basis vectors for the matrices.
However, because the convected base vectors are not orthogonal, we
keep sub/superscript notation.
(a) For the specified xi (X I ), the deformation gradient tensor is
⎡ ⎤
0.8X 1 + 0.2X 2 + 0.2 0.2X 1 0
∂xi
F= =⎣ 0 0.2X 2 + 0.1 0 ⎦ ,
∂XI
0 0 0.8
which can be evaluated at the point (X 1 , X 2 , X 3 ) = (1, 2, 1); then

C = FT · F, E = 12 (C − I).
Deformation transforms the (Cartesian) base vectors GI in the un-
deformed configuration into the convected base vectors gI , giving
G I = G I = eI
gI = F · G I
[gIJ ] = [gI · gJ ]
[gIJ ] = [gIJ ]−1 ,
where Eq. (2.92) provides the last relation. Next, for the given W ,
we get
−1/2
W1 = a, W2 = 2b(I2 − 3), W3 = −aI3
W4 = 2cI4 , W5 = 2dI5 ,
where Wi = ∂W /∂Ii and
I1 = trC
1 2
I2 = 2 [(trC) − trC2 ]
I3 = det C
I4 = E11
2 2
I5 = E12 + E13 .
Note that I4 and I5 are modified from (5.143) for fibers oriented
in the X 1 -direction. The tensors M and N, as well as their com-
ponents, also need to be modified (see below).
Now, we are in a position to compute the stress components
using Eq. (5.148), i.e.,
∗
J∗
σI = ΦGIJ + ΨH IJ − pgIJ + ΘM IJ + ΛN IJ ,
where
−1/2
Φ = 2I3 W1
−1/2
Ψ = 2I3 W2
1/2
p = −2I3 W3
−1/2
Θ = I3 W4
−1/2
Λ = I3 W5
and, with (5.119)2 providing the AJI ,

GIJ = δ IJ
H IJ = I1 GIJ − GIK GJL gKL
M IJ = AI1 AJ1
3

N IJ = (AI1 AJα + AIα AJ1 )Eα1
α=2
AJI = eI · GJ = δ JI .
The final result is ⎡ ⎤
9.05 0.825 0
∗ ∗
σ I J = ⎣ 0.825 −4.20 0 ⎦.
0 0 −1.50
(b) In this part, we first compute the Cauchy stress tensor in Cartesian
coordinates using
σ = α0 I + α1 B + α2 B2 + ΘM + ΛN,
where
B = F · FT .
In addition to Θ and Λ from part (a), we need the coefficients
1/2
α0 = 2I3 W3
−1/2
α1 = 2I3 (W1 + I1 W2 )
−1/2
α2 = −2I3 W2
and the tensors
M = F · (e1 e1 ) · FT
3

N = F · (e1 eα + eα e1 ) · FT Eα1 .
α=2
∗ ∗
The stress components σ I J can be determined using either
∗ ∗
σI J = gI · σ · gJ
or Eq. (2.70)2 in the form
∗ ∗
[σI J ] = [B̃]T [σ][B̃],
where (2.48) gives
B̃ = gI GI
for the transformation from Cartesian base vectors GI to con-
vected base vectors gI . (Both are expressed relative to Cartesian
coordinates.) Either way, the final result is the same as that in
part (a).
5.13 Partial answer:
s22 = 2ceQ (a4 11 + a2 22 + a5 33 )

23 32 Q
s = s = 2c a8 e 23
s̄22 = s22 cos2 β + s33 sin2 β − 2s23 sin β cos β

s̄23 = s̄32 = s23 (cos2 β − sin2 β) + (s22 − s33 ) sin β cos β
5.14 (a) For transverse isotropy, the principal material directions are par-
allel and normal to the fibers. To compute deformation relative
to these directions, we introduce fiber coordinates defined by the
unit vectors
ēx = ex cos β + ey sin β
ēy = −ex sin β + ey cos β
ēz = ez ,
where the ei are base vectors in plate coordinates and ēx de-
fines the fiber direction (see Fig. 5.7). With these relations,
Eqs. (3.28)1 and (3.94)1 yield
Λ2f = ēx · (I + 2E) · ēx

= 1 + 2ēx · (Eij ei ej ) · ēx

= 1 + 2 Exx cos2 β + (Exy + Eyx) cos β sin β + Eyy sin2 β .
Using this result, along with I1 from (3.142)1, we find
W = c[I1 − 3 + α(Λ2f − 1)4 ]

= 2c[Exx + Eyy + Ezz
+8α(Exx cos2 β + (Exy + Eyx) cos β sin β + Eyy sin2 β)4 ].
(b) For in-plane stretching without shear, the deformation gradient
tensor takes the simple form
F =Λxex ex + Λy ey ey + Λz ez ez
√
in plate coordinates, where the Λi = 1 + 2Eii (i not summed)
are stretch ratios. For an incompressible material, Eq. (5.177)
provides the constitutive relation
∂W
σ = F· · FT − pI
∂E
⎡ 2 ⎤
Λx Wxx − p Λx Λy Wxy 0
= ⎣ Λx Λy Wyx Λ2y Wyy − p 0 ⎦,
0 0 Λ2z Wzz − p (e e )
i j
where
∂W
Wij ≡ .
∂Eij
Material incompressibility requires det F = Λx Λy Λz = 1, which
gives Λz = 1/ΛxΛy . Then, the plane stress condition (σ zz = 0)
yields
2c
p = 2cΛ2z = .
Λ2x Λ2y
Putting everything together, we find

2 1 2 3 2
σ xx = 2cΛx 1 − 4 2 + 4α(Λf − 1) cos β
Λx Λy

1
σ xx = 2cΛ2y 1 − 2 4 + 4α(Λ2f − 1)3 sin2 β
Λx Λy
σ xy = σ yx = 8cαΛx Λy (Λ2f − 1)3 cos β sin β.
Note that, if we set Exy = Eyx = 0 before taking derivatives of W ,
we would get σ xy = 0. This example again illustrates the hazard
of applying symmetry conditions or certain assumptions prior
to differentiating without careful consideration of the potential
consequences.
(c) Partial answer: No
Chapter 6
In the following solutions, tensor components relative to principal coordi-

nates are written as physical components with single subscripts and hats
omitted. This includes problems for which symmetry dictates the absence
of shear in global coordinates, such as uniform expansion of a spherical shell
relative to spherical polar coordinates.
6.1 With I1 and I3 given by (6.16), the given strain-energy density func-
tion becomes
μ
W = [λ2x + λ2y + λ2z + 3(λxλy λz )−2/3 − 6].
2
Since p = 0 for a compressible material, Eq. (6.8)1 gives
∂W
σ i = J −1 λi (i = x, y, z; not summed).
∂λi
Next, we set the transverse stress σ y = 0 (or σ z = 0) and use the

symmetry condition λy = λz to get
λy = λz = λ−1/5
x .
Thus,
J = λx λy λz = λ3/5
x
and
1
σ x = μ λ7/5
x − .
λx
6.2 The solution must satisfy the incompressibility condition
λx λy λz = 1.
Moreover, because forces are applied only in the axial (z) direction,
we must have
∂W
σ x = λx − p = 2λ2x C + D1 (λ2x − 1) − p = 0
∂λx
∂W
σ y = λy − p = 2λ2y C + D2 (λ2y − 1) − p = 0,
∂λy
which are found by substituting the given W into Eq. (6.8)1 . With
λz specified, these three equations can be solved for λx , λy , and p.
6.3 (a) As noted earlier, the principal directions of stress and strain co-
incide for an isotropic material. Therefore, for an incompressible
Ogden material (J = 1), Eq. (6.8)1 gives
∂W
σ i = Λi −p
∂Λi

N
= an Λbi n − p
n=1
with respect to the principal axes. For in-plane deformation, the
membrane has no transverse loading, and setting σ 3 = 0 yields

N
p= an Λb3n .
n=1
With this result, we find

N
σi = an Λbi n − (Λ1 Λ2 )−bn ,
n=1
in which incompressibility (J = 1) has provided Λ3 = 1/Λ1 Λ2 .

(b) In the spectral representation

3

σ= σ i (Λj )ni ni ,
i=1
we can determine the principal stretches Λi and principal direc-
tions in the deformed configuration (ni ) by solving the eigenvalue
problem [see Eq. (3.133)2]
(B − Λ2 I) · n = 0.
The deformation tensor B is found using the given mapping
xi (Xj ), which gives
⎡ ⎤
λ1 kλ2 0
∂xi
F= = ⎣ 0 λ2 0 ⎦
∂Xj
0 0 λ3 (e e )
i j
⎡ ⎤
λ21 + k 2 λ22 kλ22 0
B= F·F = ⎣T
kλ22 λ22 0 ⎦,
0 0 λ23 (e
i ej )
where the ei are Cartesian base vectors. Next, with the stresses
given by the expression derived in part (a), we compute the stress
components relative to the Xi coordinates using
σ ij = ei · σ · ej
3

= σ k (ei · nk )(ej · nk ).
k=1
For the specified parameter values, we find
⎡ ⎤
8.69 1.19 0
σ = ⎣ 1.19 0.944 0 ⎦.
0 0 0 (e
i ej )
6.4 (a) To express the given W in terms of the Fij , we write

⎡ ⎤
F11 F12 0
F = ⎣ F21 F22 0 ⎦ ,
0 0 F33 (e e )
i j
where F21 can be set to zero after differentiating W (see F in

solution to Problem 6.3). This representation for F is used to
compute
C = FT · F
and
2 2 2 2 2
I1 = tr C = F11 + F22 + F33 + F12 + F21
I3 = det C =(F11 F22 − F12 F21 )2 F33
2
,
which are needed for the given W . The constitutive relation
(5.77)
∂W ∂W
t= → tij =
∂FT ∂Fji
gives (after setting F21 = 0)

2 2 −(1+β)
t11 = 2F11 1 − F22 F33 I3

2 2 −(1+β)
t22 = 2F22 1 − F11 F33 I3
2 −(1+β)
t12 = 2F11F22 F33 F12 I3
t21 = 2F12,
which are the only non-zero components of the first Piola-
Kirchhoff stress tensor t. To eliminate F33, we use the plane
stress condition t33 = 0 to find
F33 = (F11 F22 )−β/(1+β) = (F11 F22 )−ν/(1−ν) .
√
(b) With J = I3 , Eq. (6.2) provides the Cauchy stress tensor,
σ = J −1 F · t.
To extract the convected stress components, we need the con-
vected base vectors given by
gI = F · eI ,
which, with Eqs. (2.18), yield
g1∗ = F11 e1 , g2∗ = F12 e1 + F22 e2 , g3∗ = F33 e3
1∗ −1 −1 2∗ −1 ∗
g = F11 (e1 − F12F22 e2 ), g = F22 e2 , g3 = F33
−1
e3 .
We now can compute
∗
J∗
σI = gI · σ · gJ .
The details are left for the reader’s entertainment.
(c) Partial answer for ν = 0.25 and λ1 = 1:
∗ ∗ ∗ ∗ ∗ ∗
σ1 1
= 0.268, σ2 2
= 1.06, σ1 2
= 0.198
This can be used as a check.
6.5 (b) For the given mapping, we get

⎡ ⎤
λ1 0 0
∂xi
F= = ⎣ kX12 λ2 0 ⎦
∂Xj
0 0 1
and ⎡ ⎤
λ21 kλ1 X12 0
B = F · FT = ⎣ kλ1 X12 λ22 + k 2 X14 0 ⎦ .
0 0 1
With n = e1 for the element parallel to the x1 -axis in the de-
formed cube, Eq. (3.94)2 gives the Eulerian stretch ratio
1 λ1 λ2
λ(n) = √ = .
n · B−1 · n 2
λ2 + k 2 X14
For the element at the center of the undeformed cube, we set
X1 = 0.5.
(c) To find the applied loads, we first need to compute the stresses
in the cube. In part (b), F is determined for the specified defor-
mation, which gives F12 = 0 and F33 = 1. If we compute stresses
by differentiating W written in terms of those specialized com-
ponents, some important terms could be missed. That is why F
is expressed more generally in the solution to Problem 6.4. Here,
we avoid this issue by using Eq. (5.151), i.e.,
σ = α0 I + α1 B + α2 B2 + ΘM + ΛN,
where differentiations of W are already included. The various
quantities in this equation can be computed using relations found
in sections 5.4.1 and 5.4.2, along with the specialized strain in-
variants
I1 = tr B = λ21 + λ22 + k 2 X14 + 1

I2 = 12 (tr B)2 − tr B2 = λ21 + λ22 + λ21 λ22 + k 2 X14
I3 = det B =λ21 λ22

I4 = E11 = e1 · E · e1 = 12 (λ21 + k 2 X14 − 1)
I5 = E12 2
= (e1 · E · e2 )2 = 14 λ22 k 2 X14 .
Here, we have used the relation E = 12 (FT · F − I), which also
tells us that E31 = 0. In addition, we need the tensors
M = F · (e1 e1 ) · FT
3

N = F · (e1 eα + eα e1 ) · FT Eα1 ,
α=2
which are modified versions of (5.152) for fibers originally parallel

to the X1 -axis.
After computing σ, we can determine the applied loads as fol-
lows. The true traction vector on each surface of the deformed
cube is given by
T(n) = n · σ,
where the unit normal to the surface is
n = N · F−1 /|N · F−1 |,
as provided by (3.128) with da = n da and dA = N dA. Along
the upper surface, N = e2 , while N = e1 on the right side of the
cube. As a check, n = −0.447e1 + 0.894e2 on the upper surface
at X1 = 1. The tangent vector m is orthogonal to n and can be
computed as m = e3 × n on the surfaces at the top, bottom, and
sides of the deformed cube. (Shear stresses in the e3 -direction
are zero.) Then, the Cauchy stresses normal and tangent to each
surface, respectively, are given by
σ nn = T(n) · n, σ nm = T(n) · m.
Note that these loads vary with X1 along the upper and lower
surfaces but are constant on the other boundaries.
Because the specified deformation is not uniform, we must con-
sider the possibility that body forces also are needed to satisfy
equilibrium. These are computed using the equilibrium equation
¯ · σ + f = 0.
∇
With Eq. (3.24), we get
1 ∂σ
f = − F−T · ∇ · σ = − e1 · .
λ1 ∂X1
Partial answer:
Upper/lower surface at X1 = 1 : σ nn = 0.798, σ nm = 3.03
Left surface (X1 = 0) : σ nn = 36.9
Right surface (X1 = 1) : σ nn = 86.2
(Surfaces normal to X3 -axis) : σ 33 = 0.667 (uniform)
body forces at X1 = 1 : f1 = −128, f2 = −110
6.7(a) Since the analysis in the text uses physical tensor components,
all hats are omitted here for convenience. For clarity, however,
because torsion causes eR and eΘ to rotate during deformation,
dyadic bases are indicated on matrices, with I, J = R, Θ, Z and
i, j = r, θ, z. Note also that the formulation in the text takes
advantage of several equations in Sec. 6.5.
First, we consider a, λ, and ψ as unknowns to be determined.
In the absence of residual stress, we set φ0 = π and Λ = 1, giving
γ = ψr and
1
2 1 2 2
2
r(R) = a + R − a0
λ
by (6.116). With this result, Eqs. (6.103)–(6.106) yield
⎡ ∂r ⎤
⎡ ⎤
FrR 0 0 0 0
⎢ ∂R ⎥
F = ⎣ 0 FθΘ FθZ ⎦ =⎢ ⎣ 0
r ⎥
ψr ⎦
0 0 FzZ (e e ) R
i J 0 0 λ (e e )
i J
and
1

E = 2 FT · F − I
⎡ 2 ⎤
FrR − 1 0 0
1⎣ 2 ⎦
= 2 0 FθΘ −1 FθΘ FθZ
2 2
0 FθΘ FθZ FzZ + FθZ − 1 (e
I eJ )
at each grid point R or r(R). Next, the Cauchy stresses are given
by
σ = σ̄ − p I,
where
∂W
σ̄ = F ·· FT = [σ̄ ij ](ei ej ) .
∂E
The matrix for ∂W/∂E relative to the basis {eI eJ } is provided by
Eq. (6.132), which requires writing W as a function of the strain
components in the fiber coordinate system. These are found using
the relations
EF F = eF ·E·eF , ECC = eC ·E·eC , EF C = ECF = eF ·E·eC ,
where
eF = eΘ cos β + eZ sin β
eC = −eΘ sin β + eZ cos β.
So now, we have σ̄ at each grid point as a function of a, λ, and

ψ. With the pressure P specified and M = N = 0, these three
unknowns are found by simultaneously solving Eqs. (6.135), i.e.,
b
zz
2σ̄ − σ̄ rr − σ̄ θθ r dr = 0
a
b
σ zθ r 2 dr = 0
a
b dr
σ̄ θθ − σ̄ rr = P,
a r
where a = r(a0 ) and b = r(b0 ). Doing this requires a nonlinear
equation solver. Finally, we compute
b
θθ dr
p(r) = σ̄ rr + σ̄ − σ̄ rr ,
r r
then the stress components σ ij and
σ ff = ef · σ · ef ,
where ef = F · eF /|F · eF |.
6.8 (a) For the given mapping, we get
∂ 1 ∂ ∂
∇r = eR + eΘ + eZ [r(R)er (θ) + z(R, Z)ez ]
∂R R ∂Θ ∂Z
in cylindrical coordinates. With ∂er /∂Θ = ∂er /∂θ = eθ , this
equation yields the deformation gradient tensor
∂r r ∂w
F = (∇r)T = er eR + e θ eΘ + ez eZ + ez eR .
∂R R ∂R
The supports at the curved surfaces of the tube prevent radial
motion, giving the boundary conditions
r(a) = a
r(b) = b.
Enforcing incompressibility leads to the simple result
r ∂r
J = det F = = 1 −→ r = R.
R ∂R
(b) Either plugging the stress tensor
σ = σ rr er er + σ θθ eθ eθ + σ zz ez ez + σ rz er ez + σ zr ez er
into ∇¯ · σ = 0 or specializing the equilibrium equations in Ap-

pendix B provides
dσ rr 1
+ (σ rr − σ θθ ) = 0
dr r
dσ rz σ rz
+ = 0.
dr r
These equations represent radial and vertical equilibrium, respec-
tively.
(c) Rather than working with the strain tensor E, we choose here to
write the constitutive relations in terms of the deformation tensor
C. For
W = C1 (I1 − 3) + C1 (I2 − 3),
the strain invariants are given by (3.136) in the form
I1 = tr C = CRR + CΘΘ + CZZ
2
I2 = 1
2 [(tr C) − tr C2 ]
= CRR CΘΘ + CΘΘ CZZ + CZZ CRR − CZRCRZ ,
where
⎡ ⎤
r 2 + w 2 0 w
C = FT · F = ⎣ 0 r 2 /R2 0 ⎦

w 0 1 (e
I eJ )
with prime indicating differentiation with respect to R. For an

incompressible material, the Cauchy stress tensor can be written
σ = σ̄ − pI,
where
∂W
σ̄ = 2F · · FT
∂C
⎡ ⎤
C1 + 2C2 0 (C1 + C2 )w
= 2⎣ 0 C1 + 2C2 + C2 w 2 0 ⎦
2
(C1 + C2 )w 0 C1 + 2C2 + (C1 + C2 )w
relative to the {ei ej } basis.
The tube displacement must be zero at the fixed outer surface.
This requirement and a force balance on the rigid cylinder inside
the tube provide the boundary conditions
P
σ rz (a) = , w(b) = 0.
2πaL
Integrating the differential equation for vertical equilibrium yields

d P 1
(rσ rz ) = 0 −→ σ rz (r) = ,
dr 2πL r
which satisfies the first boundary condition. To obtain w(r), we
first note that
dw dw
=
dR dr
since r = R. Next, because shear stresses do not involve the
Lagrange multiplier p, plugging the above expression for σrz =
σ̄ rz into the constitutive relation for this stress gives
dw P 1
σ rz = 2(C1 + C2 ) = .
dr 2πL r
Integrating this equation and using the boundary condition
w(b) = 0, we find
r
w(r) = A log ,
b
where
P
A= .
4πL(C1 + C2 )
(d) The above constitutive relation for σ̄ provides σ̄ rr = 2(C1 +
2C2 ), but determining the radial stress also requires knowing p(r).
Integrating the radial equilibrium
equation gives
dr
p = σ̄ rr + (σ̄ rr − σ̄ θθ ) .
r
After substituting the relations for σ̄ rr and σ̄ θθ , along with w =
A/r, we can integrate this equation to obtain
p(r) = 2(C1 + 2C2 ) + A2 C2 r −2 + B,
where B is a constant. To determine B, although σzz is not zero,
we assume the resultant axial force vanishes, i.e.,
2π b
Nz = σ zz r dr dθ = 0,
0 a
which gives
b
(σ̄ zz − p) r dr = 0.
a
After substituting the above equation for p and the constitutive
relation for σ̄ zz obtained in part (c), we find
log(b/a)
B = 2(2C1 + C2 )A2 2 .
b − a2
With this result, the radial stress is
A2 C2
σ rr = σ̄ rr − p = − 2 − B.
r
(e) The assumption Nz = 0 leads to p being positive, because σ̄ zz

is tensile across the wall. Thus, σ rr = σ̄ rr − p turns out to be
compressive.
6.9 (a) For axisymmetric deformation of a cylinder, Eq. (3.63) gives the
stretch ratios
dr r
λr = , λθ = , λz = λ
dR R
for no torsion (ψ = 0). Solving the incompressibility relation
dr r
J = λr λθ λz = λ=1
dR R
with the boundary condition r(0) = 0 yields
R
r= √ .
λ
With this result, the stretch ratios become
1
λr = λθ = √ , λz = λ.
λ
(b) The stress field must satisfy the equation of motion
¯ · σ + f = ρa.
∇
The effects of the centripetal acceleration can be included by set-
ting a = −rω2 er in the spinning cylinder. Otherwise, we can
consider the problem in a frame rotating with the cylinder at
speed ω. Relative to this frame, the cylinder is stationary with
body force f = ρrω 2 er . Either approach leads to the radial equa-
tion of motion
dσ r 1
+ (σ r − σ θ ) = −ρω 2 r.
dr r
The boundary condition on the curved surface is
σ r (b) = 0,
√
where b = b0 / λ. Because neo-Hookean material is isotropic and
λr = λθ , we must have σ r = σ θ , and the equation of motion can
be integrated directly to obtain
σ r = 12 ρω2 (b2 − r 2 ) = σ θ ,
which satisfies the boundary condition, as well as the condition
that the radial stress must be continuous at the boundary between
regions. Note: If the equation of motion is solved separately in
each region, enforcing continuous σ r at the boundary between

regions would lead to the same result.
Next, we use the constitutive relation to determine σ z , which
does not need to be continuous at r = a. For
W (k) = C (k)(λ2r + λ2θ + λ2z − 3),
we find
(k) (k)
σi = σ̄ i − p(k)
in region k, where
(k) ∂W (k)
σ̄ i = λi = 2C (k)λ2i (i not summed).
∂λi
At this point, it is important to realize that even if σ i is continuous
at the boundary, this does not imply continuity in σ̄ i and p. With
σ r known, the Lagrange multiplier in region k is given by
(k) 2
p(k) = σ̄ (k)
r − σ r = 2C λr − 12 ρω2 (b2 − r 2 )
= 2C (k)/λ − 12 ρω2 (b2 − r 2 ).
With this result, we find the axial stress
σ (k)
z = σ̄ (k)
z −p
(k)
= 2C (k)λ2 − p(k)
1
= 2C (k) λ2 − + 12 ρω2 (b2 − r 2 )
λ
in region k.
(c) The resultant force in the axial direction is given by
2π b
Nz = σ z r dr dθ = 0,
0 0
which leads to
a b
σ (1)
z r dr + σ (2)
z r dr = 0.
0 a
(k)
Substituting from part (c) and integrating yields
σz
3 1
λ − 1 a20 (C1 − C2 ) + b20 C2 + ρω2 b40 = 0,
8
which gives
1
ω2 3
λ= 1− 2 ,
ω0
where
8 2
ω 20 = 4 a0 (C1 − C2 ) + b20 C2 .
ρb0
According to this solution, the length of the cylinder deceases
as ω increases. This is reasonable, but the cylinder flattens com-
pletely (λ = 0) when ω = ω0 , which is not realistic. Similar
behavior occurs for inflation of a spherical neo-Hookean shell.
As shown by the curve for α = 0 in Fig. 6.22a, the pressure a
neo-Hookean shell can sustain is limited. Here, the cylinder can
sustain inertial loads only for spin rates ω < ω 0 . This unfortu-
nate behavior of neo-Hookean material becomes a serious issue in
certain cases, such as these. Other forms for W do not necessarily
have this problem (Chadwick et al., 1977).
6.10 Here, we outline one possible computational procedure to solve this
problem. After setting up a grid for R, the steps are the following:
• Given a, use Eq. (6.151) to compute the deformed coordinate r
at each grid point.
• Compute the stretch ratios at each point using (6.145). (λR also
could be computed using incompressibility.)
• Integrate (6.153) to obtain the Lagrange multiplier p at each
point. Break the integral into two parts, i.e.,
b c b
(· · · ) = (· · · ) + (· · · ), r ≤ c
r r c
b
= (· · · ), r ≥ c,
r
where b = r(b0 ) and c = r(c0 ) are found using (6.151). For

convenience, we could integrate across the undeformed radius by
using Eqs. (6.145) to write dr/r = λR dR/(λΘ R).
• Compute the Cauchy stresses using (6.147).
• Compute P using (6.154), again breaking the integral into two
parts.
6.11 The following analysis follows Chung et al. (1986). With the strain
invariants given by
I2 = λ2r λ2θ + λ2θ λ2φ + λ2φ λ2r

I3 = J 2 = λ2r λ2θ λ2φ ,
Eq. (5.115) becomes

μ 1/2

W = I2 /I3 + 2I3 − 5)
2
μ −2
= λr + λ−2 θ + λ−2
φ + 2λ λ λ
r θ φ − 5
2
in spherical coordinates. For a compressible material, (6.8)1 provides
λi ∂W
σi = = μ 1 − J −1 λ−2
i (i not summed).
J ∂λi
For spherically symmetric deformation, r = r(R) and the stretch
ratios are
∂r r
λr = = r , λθ = λφ = .
∂R R
These equations yield
rr2
J = λr λθ λφ = ,
R2
and substitution into the constitutive relation above provides the
stresses
R2
σ r (r) = μ 1 −
r 2 r 3
R4
σ θ (r) = σ φ (r) = μ 1 − .
r4r
Next, the radial equilibrium equation (see Appendix B)
dσ r 2
+ (σ r − σ θ ) = 0
dr r
for a sphere can be written as
dσr 2r
+ (σ r − σ θ ) = 0,
dR r
where we have used the relation dr = λr dR. Finally, inserting the
above relations for the stress components yields the differential equa-
tion
3R2 r 3 r − 2Rr 3 r + 2R4 r 4 = 0
to be solved for r(R).

6.13 (a) For spherically symmetric deformation, the strain invariants
I1 = λ2r + λ2θ + λ2φ

I3 = J 2 = λ2r λ2θ λ2φ
give
W = C[I1 − 3 + β −1 (I3−β − 1)
! "
= C λ2r + λ2θ + λ2φ + β −1 (λr λθ λφ )−2β − 1 ,
where β = ν/(1 − 2ν). For compressible material, (6.8)1 provides

λi ∂W
σi = .
J ∂λi
The circumferential and meridional stretch ratios, given by
(6.145), are
r
λθ = λφ ≡ λ = .
R
The radial stretch ratio is obtained by enforcing the assumption
that the transverse normal stress can be neglected, giving
σ r = 0 −→ λr = (λθ λφ )−β/(1+β) = λ−2ν/(1−ν) .
With a = λa0 and h = λr h0 , Laplace’s law (6.156) yields

2h h0 λr h0 λr λθ ∂W
P = σθ = 2 σθ = 2
a a0 λ a0 λ λr λθ λφ ∂λθ
h0 1 ∂W
= 2
a0 λ2 ∂λθ
h0 1 1
= 4C 1− m ,
a0 λ λ
where
1+ν
m=2 .
1−ν
Note that m = 6 for ν = 0.5, and the solution agrees with that
of Problem 6.12 for an incompressible membrane with α = 0 for
either W , i.e., a neo-Hookean membrane.
6.14 The ensuing solution follows that of Eringen (1962), who included
some additional modifications.
(a) In spherical coordinates, the position vector to a point in the

deformed configuration is
r = r er .
Using equations found in Appendix B, we get
∂ 1 ∂ 1 ∂
∇r = eR + eΘ + eΦ [r er (θ, φ)]
∂R R ∂Θ R sin Θ ∂Φ
∂r r ∂er ∂θ r ∂er ∂φ
= eR er + eΘ + eΦ .
∂R R ∂θ ∂Θ R sin Θ ∂φ ∂Φ
The relations
er = ex sin θ cos φ + ey sin θ sin φ + ez cos θ
eφ = −ex sin φ + ey cos φ
yield
er , θ = e θ , er ,φ = eφ sin θ.
For the assumed mapping, substitution into the above expression
for ∇r gives
F = (∇r)T = λr er eR − λθ eθ eΘ + λφ eφ eΦ ,
where
∂r r
λr = , λθ = λφ = .
∂R R
(b) The incompressibility condition gives
dr r 2
J = det F = −λr λθ λφ = − = 1.
dR R2
Multiplying by R2 dR and integrating yields
r = (A − R3 )1/3 ,
where A is a constant. In the everted shell, the original inner and
outer radii, respectively, become
a = r(a0 ) = (A − a30 )1/3
b = r(b0 ) = (A − b30 )1/3 .
Since b0 > a0 , these relations give b < a, i.e., during eversion the
outer radius becomes the inner radius and vice versa, as expected.
For the radii to be real, we must have A > b30 .
(c) For incompressible material (J = 1), the Cauchy stress tensor

can be written in the form
σ = σ̄ − pI,
where Eq. (6.4)1 gives
∂W
σ̄ = 2F ·
· FT ,
∂C
in which C = I + 2E. Since C = FT · F, we find
C(eI eJ ) = diag [Cr , Cθ , Cφ] = diag [λ2r , λ2θ , λ2φ ].
For
c α(I1 −3)
W = e −1
α
I1 = tr C = Cr + Cθ + Cφ ,
the above relations give
σ i = σ̄ i − p,
where
∂W ∂W ∂I1
σ̄ i = 2λ2i = 2λ2i = 2cλ2i eα(I1 −3) (i not summed).
∂Ci ∂I1 ∂Ci
To determine p, we use the radial equilibrium equation (see Ap-
pendix B)
dσ r 2
+ (σ r − σ θ ) = 0
dr r
for a sphere, where we have set the shear stresses to zero and
σ φ = σ θ by symmetry. Substituting σ i (r) = σ̄ i (r) − p(r) yields
d 2
(σ̄ r − p) + (σ̄ r − σ̄ θ ) = 0.
dr r
The boundary conditions are
σ r (a) = σ r (b) = 0
for no external loading. Solving for p gives
r
dr
p = σ̄ r + 2 (σ̄ r − σ̄ θ ) ,
b r
which satisfies the boundary condition at r = b. The other bound-
ary condition yields
a
dr
σ r (a) = σ̄ r (a) − p(a) = −2 (σ̄ r − σ̄ θ ) = 0,
b r
which, with the expressions for r, a, and b found in part (b), is
to be solved for A.
(d) The stress σ θ increases from negative at the inner surface to pos-
itive at the outer surface in the everted shell.
(e) The form of the stress distribution is consistent with those pro-
duced by bending the shell into its everted configuration. This
requires bending moments in both the θ- and φ-directions to re-
verse the curvature at each point. The peak stress increases with
the shell thickness, in agreement with intuition that thinner shells
would be easier to turn inside out.
However, since σ θ = 0 along the edge of the shell, the as-
sumed deformation is not consistent with the everted shell being
completely free of external loads. To eliminate these stresses,
equal and opposite bending stresses must be applied along the
edge, compressing the outer surface while causing the edge to
bend slightly outward. Because these cancelling stresses are self-
equilibrating, Saint-Venant’s principle suggests that their effects
would be significant only within a relatively narrow region near
the edge.
6.15 The solution to this problem follows Carroll and Horgan (1990).
(a) With the position vector in the deformed configuration given by
r = xex + yey + zez ,
we have
∂ 1 ∂ ∂
∇r = eR + eΘ + eZ [x(R)ex + kΘey + λZez ] ,
∂R R ∂Θ ∂Z
in which ∇ is written for the undeformed body in cylindrical co-
ordinates. This relation produces the deformation gradient tensor
F = (∇r)T = λx ex eR + λy ey eΘ + λz ez eZ ,
where
dx k
λx = , λy = , λz = λ
dR R
are stretch ratios.
(b) In terms of the stretch ratios, the given strain-energy density
function takes the form
μ −2
W = λx + λ−2 −2
y + λz + 2λx λy λz − 5 .
2
For the prescribed deformation, only normal stresses are present
relative to Cartesian coordinates, and Eq. (6.8)1 gives
λi ∂W
σi = (i not summed)
J ∂λi
for a compressible material, where
J = λx λy λz .
These relations yield

1
σx = μ 1 −
λ3x λy λz
# $
1
σy = μ 1−
λx λ3y λz
1
σz = μ 1 − .
λx λy λ3z
¯ · σ = 0 provides the scalar relations
(c) The equilibrium equation ∇
dσ x dσ y dσ z
= = = 0.
dx dy dz
Since σ x is constant in the x-direction and is prescribed as zero
on the x-faces of the straightened sector, this stress component
vanishes everywhere. Thus, the relation for σ x found in part (b)
gives
λ3x λy λz = 1.
Substituting the λi from part (a) yields

1
3
dx R
λ3x λy λz = 1 −→ = ,
dR kλ
which is integrated to obtain
1
3 R4 3
x= .
4 kλ
The constant of integration, which represents a rigid-body dis-
placement, is set to zero.
(d) The constants k and λ are determined using the boundary con-
ditions at y = y1 , y2 and z = z1 , z2 , i.e.,
z2 x2
Ny = σ y dx dz = 0
z1 x1
y2 x2
Nz = σ z dx dy = 0,
y1 x1
where dx = λx dR. Substituting the stretch ratios, stresses, and

x(R) from parts (a)–(c) yields
R2 1
R 3 R3
− 3 dR = 0
R1 kλ k λ
R2 1
R 3 R
− 3 dR = 0.
R1 kλ kλ
Integrating these equations gives
3 1
4/3 4/3
R 2 − R 1 − 3 (R42 − R41 ) = 0
4(kλ)1/3 4k λ
3 1
4/3 4/3
R2 − R1 − (R22 − R21 ) = 0,
4(kλ)1/3
2kλ3
which are solved to get (Carroll and Horgan, 1990)
1 2
k2 = 2
λ (R21 + R22 )
16(R22 − R21 )3
λ10 = 3 .
4/3 4/3
27 (R21 + R22 ) R2 − R1
(e) With the above results, the bending moment is given by

z2 x2
M = xσ y dx dz
z1 x1
x2
= λL xσ y dx = −0.276.
x1
Bibliography
Alford PW and Taber LA (2003). Regional epicardial strain in the embry-

onic chick heart during the early looping stages. J Biomech 36:1135-1141.
Armstrong CG, Lai WM, and Mow VC (1984). An analysis of the uncon-
fined compression of articular cartilage. J Biomech Eng 106:165-173.
Arts T, Meerbaum S, Reneman RS, and Corday E (1984). Torsion of the

left ventricle during the ejection phase in the intact dog. Cardiovasc Res
18:183-193.
Arts T, Reneman RS, and Veenstra PC (1979). A model of the mechanics

of the left ventricle. Ann Biomed Eng 7:299-318.
Atluri SN (1984). Alternate stress and conjugate strain measures, and

mixed variational formulations involving rigid rotations, for computational
analyses of finitely deformed solids, with application to plates and shells —
I. Theory. Comput Struct 18:93-116.
Axel L, Montillo A, and Kim D (2005). Tagged magnetic resonance imaging

of the heart: A survey. Med Image Anal 9:376-393.
Bayraktar M and Manner J (2014). Cardiac looping may be driven by com-

pressive loads resulting from unequal growth of the heart and pericardial
cavity. Observations on a physical simulation model. Front Physiol 5:112.
Beloussov LV (1998). The dynamic architecture of a developing organism:

An interdisciplinary approach to the development of organisms, Kluwer Dor-
drecht, the Netherlands.
Beloussov LV (2015). Morphomechanics of development, Springer, New

York.
437
438 Bibliography
Blatz PD and Ko WL (1962). Application of finite elasticity to the defor-

mation of rubbery materials. Trans Soc Rheology 6:223-251.
Boresi AP, Chong K, and Lee JD (2010). Elasticity in engineering mechan-
ics, Wiley, New Jersey.
Carroll MM and Horgan CO (1990). Finite strain solutions for a compress-
ible elastic solid. Q Appl Math 48 :767-780.
Carter DR and Beaupre GS (2007). Skeletal function and form: Mechanobi-
ology of skeletal development, aging, and regeneration, Cambridge Univer-
sity Press.
Chadwick P, Creasy C, and Hart V (1977). The deformation of rubber
cylinders and tubes by rotation. The ANZIAM Journal 20:62-96.
Chadwick RS (1982). Mechanics of the left ventricle. Biophys J 39:279-288.
Chung DT, Horgan CO, and Abeyaratne R (1986). The finite deforma-
tion of internally pressurized hollow cylinders and spheres for a class of
compressible elastic materials. Int J Solids Struct 22 :1557-1570.
Chuong CJ and Fung YC (1986). On residual stresses in arteries. J
Biomech Eng 108:189-192.
Cowin SC and Doty SB (2007). Tissue mechanics, Springer, New York.
Criscione JC, Lorenzen-Schmidt I, Humphrey JD, and Hunter WC (1999).
Mechanical contribution of endocardium during finite extension and torsion
experiments on papillary muscles. Ann Biomed Eng 27:123-130.
Davidson LA, Oster GF, Keller RE, and Koehl MA (1999). Measurements
of mechanical properties of the blastula wall reveal which hypothesized
mechanisms of primary invagination are physically plausible in the sea
urchin strongylocentrotus purpuratus. Dev Biol 209:221-238.
Demer LL and Yin FCP (1983). Passive biaxial mechanical properties of
isolated canine myocardium. J Physiol 339:615-630.
Dokos S, Smaill BH, Young AA, and LeGrice IJ (2002). Shear properties of
passive ventricular myocardium. Am J Physiol-Heart C 283:H2650-H2659.
Drew TB (1961). Handbook of vector and polyadic analysis, Reinhold Pub.
Corp, New York.
Einstein A (1916). The foundation of the general theory of relativity. An-
nalen Phys 14:769-822.
Bibliography 439
Epstein M (2012). The elements of continuum biomechanics, Wiley, Chich-

ester.
Eringen AC (1962). Nonlinear theory of continuous media, McGraw-Hill,

New York.
Eringen AC (1980). Mechanics of continua, R. E. Krieger Pub. Co, Hunt-

ington, N.Y.
Fillinger MF, Marra SP, Raghavan ML, and Kennedy FE (2003). Predic-
tion of rupture risk in abdominal aortic aneurysm during observation: Wall
stress versus diameter. J Vasc Surg 37:724-732.
Fillinger MF, Raghavan ML, Marra SP, Cronenwett JL, and Kennedy FE
(2002). In vivo analysis of mechanical wall stress and abdominal aortic
aneurysm rupture risk. J Vasc Surg 36:589-597.
Flugge W (1972). Tensor analysis and continuum mechanics, Springer,

New York.
Fung YC (1990). Biomechanics: Motion, flow, stress, and growth, Springer,

New York.
Fung YC (1991). What are the residual stresses doing in our blood vessels?
Ann Biomed Eng 19:237-249.
Fung YC (1993). Biomechanics: Mechanical properties of living tissues,

Springer, New York.
Fung YC (1997). Biomechanics: Circulation, 2nd ed, Springer, New York.
Fung YC, Fronek K, and Patitucci P (1979). Pseudoelasticity of arteries

and the choice of its mathematical expression. Am J Physiol 237:H620-631.
Gilbert SF (2010). Developmental biology, 9th ed, Sinauer Associates, Sun-

derland, MA.
Goriely A (2017). The mathematics and mechanics of biological growth,

Springer, New York.
Green AE and Adkins JE (1970). Large elastic deformations, Oxford Uni-

versity Press, London.
Green AE and Zerna W (1968). Theoretical elasticity, Oxford University

Press, London.
440 Bibliography
Greenwald SE, Moore JE, Jr., Rachev A, Kane TPC, and Meister JJ (1997).
Experimental investigation of the distribution of residual strains in the
artery wall. J Biomech Eng 119:438-444.
Guccione JM, McCulloch AD, and Waldman LK (1991). Passive material
properties of intact ventricular myocardium determined from a cylindrical
model. J Biomech Eng 113:42-55.
Hansen DE, Daughters GT, Alderman EL, Ingels NB, and Miller DC (1988).
Torsional deformation of the left ventricular midwall in human hearts with
intramyocardial markers: Regional heterogeneity and sensitivity to the in-
otropic effects of abrupt rate changes. Circ Res 62:941-952.
Hashima AR, Young AA, McCulloch AD, and Waldman LK (1993). Non-
homogeneous analysis of epicardial strain distributions during acute my-
ocardial ischemia in the dog. J Biomech 26:19-35.
Holmes MH (1986). Finite deformation of soft tissue: Analysis of a mixture
model in uniaxial compression. J Biomech 108:372-381.
Holzapfel GA, Gasser TC, and Ogden RW (2000). A new constitutive
framework for arterial wall mechanics and a comparative study of material
models. J Elasticity 61:1-48.
Humphrey JD (1999). Remodeling of a collagenous tissue at fixed lengths.
J Biomech Eng 121:591-597.
Humphrey JD (2002). Cardiovascular solid mechanics: Cells, tissues, and
organs, Springer, New York
Humphrey JD and Holzapfel GA (2012). Mechanics, mechanobiology, and
modeling of human abdominal aorta and aneurysms. J Biomech 45:805-
814.
Humphrey JD and Rajagopal KR (2002). A constrained mixture model
for growth and remodeling of soft tissues. Math Models Methods Appl Sci
12:407-430.
Humphrey JD and Rajagopal KR (2003). A constrained mixture model
for arterial adaptations to a sustained step change in blood flow. Biomech
Model Mechanobiol 2:109-126.
Humphrey JD, Strumpf RK, and Yin FCP (1990a). Determination of a
constitutive relation for passive myocardium: I. A new functional form. J
Bibliography 441
Humphrey JD, Strumpf RK, and Yin FCP (1990b). Determination of a

constitutive relation for passive myocardium: II. Parameter estimation. J
Humphrey JD and Yin FCP (1987). On constitutive relations and finite

deformations of passive cardiac tissue: I. A pseudostrain-energy function.
J Biomech Eng 109:298-304.
Hunter PJ, Pullan AJ, and Smaill BH (2003). Modeling total heart func-
tion. Annu Rev Biomed Eng 5:147-177.
Huyghe JM, Arts T, van Campen DH, and Reneman RS (1992). Porous
medium finite element model of the beating left ventricle. Am J Physiol
262:H1256-H1267.
Ingels NB, Hansen DE, Daughters GT, Stinson EB, Alderman EL, and
Miller DC (1989). Relation between longitudinal, circumferential, and
oblique shortening and torsional deformation in the left ventricle of the
transplanted human heart. Circ Res 64:915-927.
Innocenti B (2017). Biomechanics: A fundamental tool with a long history

(and even longer future!). Muscles Ligaments Tendons J 7:491-492.
Itasaki N, Nakamura H, and Yasuda M (1989). Changes in the arrangement

of actin bundles during heart looping in the chick embryo. Anat Embryol
180:413-420.
Jacinto A, Martinez-Arias A, and Martin P (2001). Mechanisms of epithe-
lial fusion and repair. Nat Cell Biol 3:E117-123.
Jacobs CR, Huang H, and Kwon RY (2013). Introduction to cell mechanics

and mechanobiology, Garland Science, New York.
LeGrice IJ, Hunter PJ, and Smaill BH (1997). Laminar structure of the
heart: A mathematical model. Am J Physiol 272:H2466-H2476.
LeGrice IJ, Smaill BH, Chai LZ, Edgar SG, Gavin JB, and Hunter PJ
(1995). Laminar structure of the heart: Ventricular myocyte arrangement
and connective tissue architecture in the dog. Am J Physiol 269:H571-
H582.
Li B, Li F, Puskar KM, and Wang JH (2009). Spatial patterning of cell

proliferation and differentiation depends on mechanical stress magnitude.
J Biomech 42:1622-1627.
442 Bibliography
Lin IE and Taber LA (1994). Mechanical effects of looping in the embryonic

chick heart. J Biomech 27:311-321.
Malvern LE (1969). Introduction to the mechanics of a continuous medium,
Prentice-Hall, Englewood Cliffs, NJ.
Manner J (2000). Cardiac looping in the chick embryo: A morphological
review with special reference to terminological and biomechanical aspects
of the looping process. Anat Rec 259:248-262.
McCulloch A, Waldman L, and Rogers J (1992). Large-scale finite element
analysis of the beating heart. Crit Rev Biomed Eng 20:427-449.
McCulloch AD, Smaill BH, and Hunter PJ (1989). Regional left ventricular
epicardial deformation in the passive dog heart. Circ Res 64:721-733.
Mirsky I (1973). Ventricular and arterial wall stresses based on large de-
formation analyses. Biophys J 13:1141-1159.
Mirsky I, Ghista DN, and Sandler H (1974). Review of various theories
for the evaluation of left ventricular wall stresses. In: Cardiac mechanics:
Physiological, clinical, and mathematical considerations. Wiley, New York,
pp. 381-409.
Moon MR, Ingels NB, Daughters GT, Stinson EB, Hansen DE, and Miller
DC (1994). Alterations in left ventricular twist mechanics with inotropic
stimulation and volume loading in human subjects. Circulation 89:142-150.
Mooney M (1940). A theory of large elastic deformation. J Appl Phys
11:582-592.
Mow VC, Kwan MK, Lai WM, Holmes MH, Schmid-Schonbein GW, Woo
SLY, and Zweifach BW (1986). A finite deformation theory for nonlinearly
permeable soft hydrated biological tissues. In: Frontiers in biomechanics.
Springer-Verlag, New York, pp. 153-179.
Murray JD (2003). Mathematical biology II: Spatial models and biomedical
applications, Springer, New York.
Nakamura A, Kulikowski RR, Lacktis JW, Manasek FJ, van Praagh R,
and Takao A (1980). Heart looping: A regulated response to deforming
forces. In: R. van Praagh, A. Takao (Eds.) Etiology and morphogenesis of
congenital heart disease. Futura Publishing, Mount Kisco, NY, pp. 81-98.
Ogden RW (1997). Non-linear elastic deformations, Dover, New York.
Bibliography 443
Omens JH and Fung YC (1990). Residual strain in rat left ventricle. Circ
Res 66:37-45.
Omens JH, MacKenna DA, and McCulloch AD (1993). Measurement of

strain and analysis of stress in resting rat left ventricular myocardium. J
Biomech 26:665-676.
Pinto JG and Fung YC (1973). Mechanical properties of the heart muscle

in the passive state. J Biomech 6:597-616.
Rachev A (1997). Theoretical study of the effect of stress-dependent re-

modeling on arterial geometry under hypertensive conditions. J Biomech
30:819-827.
Ramasubramanian A, Capaldi X, Bradner SA, and Gangi L (2019). On

the biomechanics of cardiac s-looping in the chick: Insights from modeling
and perturbation studies. J Biomech Eng 141:0510111.
Ramasubramanian A, Chu-Lagraff QB, Buma T, Chico KT, Carnes ME,

Burnett KR, Bradner SA, and Gordon SS (2013). On the role of intrinsic
and extrinsic forces in early cardiac s-looping. Dev Dyn 242:801-816.
Redd MJ, Cooper L, Wood W, Stramer B, and Martin P (2004). Wound

healing and inflammation: Embryos reveal the way to perfect repair. Philos
Trans R Soc Lond B Biol Sci 359:777-784.
Rivlin RS (1947). Torsion of a rubber cylinder. J Appl Phys 18 :444-449.
Rivlin RS (1956). Large elastic deformations. In: F. R. Eirich (Ed.) Rhe-

ology: Theory and applications. Academic Press, New York pp. 351-385.
Rivlin RS, Barenblatt GI, and Joseph DD (1997). Collected papers of R.S.
Rivlin, Springer, New York.
Rivlin RS and Saunders DW (1951). Large elastic deformations of isotropic

materials VII. Experiments on the deformation of rubber. Phil Trans Roy
Soc London A243:251-288.
Sacks MS and Sun W (2003). Multiaxial mechanical behavior of biological

materials. Ann Rev Biomed Eng 5:251-284.
Schlichter J, Hellerstein HK, and Katz LN (1954). Aneurysm of the heart:

A correlative study of one hundred and two proved cases. Medicine 33:43-
86.
444 Bibliography
Shiraishi I, Takamatsu T, Minamikawa T, and Fujita S (1992). 3-D obser-

vation of actin filaments during cardiac myofibrinogenesis in chick embryo
using a confocal laser scanning microscope. Anat Embryol 185:401-408.
Simmonds JG (1994). A brief on tensor analysis, Springer-Verlag, New
York.
Simon BR, Gaballa MA, and Spilker RL (1988). Finite strain poroelastic
finite element models for large arterial cross sections. In: Computational
methods in bioengineering. ASME, New York, pp. 325-334.
Sokolnikoff IS (1956). Mathematical theory of elasticity, McGraw-Hill, New
York.
Spencer AJM (1984). Constitutive theory for strongly anisotropic solids.
In: Continuum theory of the mechanics of fibre-reinforced composites.
Springer-Verlag, New York, pp. 1-32.
Streeter DD, Berne RM, Sperelakis N, and Geiger SR (1979). Gross mor-
phology and fiber geometry of the heart. In: Handbook of physiology, sec-
tion 2: The cardiovascular system, volume I: The heart. American Physi-
ological Society, Bethesda, MD, pp. 61-112.
Szilard R (1974). Theory and analysis of plates, Prentice-Hall, Englewood
Cliffs, NJ.
Taber LA (1991). On a nonlinear theory for muscle shells: II. Application
to the active left ventricle. J Biomech Eng 113:63-71.
Taber LA (1995). Biomechanics of growth, remodeling, and morphogenesis.
Appl Mech Rev 48:487-545.
Taber LA (2001). Biomechanics of cardiovascular development. Ann Rev
Biomed Eng 3:1-25.
Taber LA (2006). Biophysical mechanisms of cardiac looping. Int J Dev
Biol 50:323-332.
Taber LA (2020). Continuum modeling in mechanobiology, Springer, New
York.
Taber LA and Humphrey JD (2001). Stress-modulated growth, residual
stress, and vascular heterogeneity. J Biomech Eng 123:528-535.
Taber LA, Keller BB, and Clark EB (1992). Cardiac mechanics in the
stage-16 chick embryo. J Biomech Eng 114:427-434.
Bibliography 445
Taber LA, Sun H, Clark EB, and Keller BB (1994). Epicardial strains in
embryonic chick ventricle at stages 16 through 24. Circ Res 75:896-903.
Taber LA, Yang M, and Podszus WW (1996). Mechanics of ventricular
torsion. J Biomech 29:745-752.
Timoshenko S and Goodier JN (1969). Theory of elasticity, McGraw-Hill,
New York.
Tozeren A (1983). Static analysis of the left ventricle. J Biomech Eng
105:39-46.
Truesdell C and Noll W (2004). The non-linear field theories of mechan-
ics. In: S. A. Antman (Ed.) The non-linear field theories of mechanics.
Springer, New York.
Verbruggen SWE (2018). Mechanobiology in health and disease, Academic
Press, London.
Villarreal FJ, Lew WYW, Waldman LK, and Covell JW (1991). Transmu-
ral myocardial deformation in the ischemic canine left ventricle. Circ Res
68:368-381.
Vossoughi J, Hedjazi Z, Borris FS, Langrana NA, Friedman MH, and Grood
ES (1993). Intimal residual stress and strain in large arteries. In: Proc sum-
mer bioengineering conference. New York, ASME, pp. 434-437.
Wagenseil JE (2018). Bio-chemo-mechanics of thoracic aortic aneurysms.
Curr Opin Biomed Eng 5:50-57.
Waldman LK, Fung YC, and Covell JW (1985). Transmural myocardial
deformation in the canine left ventricle. Normal in vivo three-dimensional
finite strains. Circ Res 57:152-163.
Waldman LK, Nosan D, Villarreal F, and Covell JW (1988). Relation
between transmural deformation and local myofiber direction in canine left
ventricle. Circ Res 63:550-562.
Wang VY, Lam H, Ennis DB, Cowan BR, Young AA, and Nash MP (2009).
Modelling passive diastolic mechanics with quantitative MRI of cardiac
structure and function. Med Image Anal 13:773-784.
Whittaker P, Boughner DR, and Kloner RA (1991). Role of collagen in
acute myocardial infarct expansion. Circulation 84:2123-2134.
Woods RH (1892). A few applications of a physical theorem to membranes
in the human body in a state of tension. J Anat Physiol 26:362-370.
446 Bibliography
Yin FC (1981). Ventricular wall stress. Circ Res 49:829-842.

Yin FCP, Strumpf RK, Chew PH, and Zeger SL (1987). Quantification
of the mechanical properties of nonconducting canine myocardium under
simultaneous biaxial loading. J Biomech 20:577-589.
Zamir EA and Taber LA (2004). Material properties and residual stress in
the stage 12 chick heart during cardiac looping. J Biomech Eng 126:823-
830.
Zhao M, Zhang H, Robinson TF, Factor SM, Sonnenblick EH, and Eng C
(1987). Profound structural alterations of the extracellular collagen matrix
in postischemic dysfunctional (“stunned”) but viable myocardium. J Am
Coll Cardiol 10:1322-1334.
Zheng QS (1994). Theory of representations for tensor functions — a uni-
fied invariant approach to constitutive equations. Appl Mech Rev 47 :545-
587.
Index
Acceleration, 6, 124–126, 143, 157, cylindrical coordinates, 11, 12, 17

160, 166, 184, 336, 341 differentiation, 45–48
Aneurysm, 151, 152, 291 dyads, 20
Angular momentum, 157–159, 343, general, 8
344 natural, 20
Anisotropic material, 156, 223, 226, skew, 11, 17
269, 340, 352 Bending, 120, 274, 291–297
Antisymmetric, skew symmetric, 23, Biaxial stretching
329 experiments, 226
Area membrane, 238–243
change in, 102–103 Blastula, 285–291
Nanson’s formula, 103 Blatz-Ko material, 207, 241, 243, 251,
vector, 140, 142 262
Artery, 202 Body force, 139, 143, 156, 174, 335,
aneurysm, 151 336, 341, 344, 345, 348
extension, inflation, torsion, Boundary conditions, 152, 227, 234,
264–277 353
model, 265 Boundary value problem, 233–236,
residual stress, 264, 273–277 353–354
structure, 264
Cardiac looping, 291
Balance law Cartilage, 78, 101, 207, 243
angular momentum, 157–159 Cauchy stress formula, 144, 338
energy, 173–179 Cayley-Hamilton theorem, 34, 210
linear momentum, 156–157 Characteristic equation, 34, 116, 333
Base vectors, 7–13, 311–312 Christoffel symbols, 45–47, 55, 90,
Cartesian, 8, 10, 17 163
contravariant, 13–17 cylindrical coordinates, 46, 166
convected, 71, 153 Compatibility conditions, 132,
coordinate transformation, 24–29 333–335
covariant, 9–13, 15, 71 Compressible material, 101, 182, 207,
447
448 Index
233, 240, 243, 250, 260, 264 convected, 70

Conservation of energy, 173, 177, 347, cylindrical, 11, 12, 16, 46, 52, 165,
350 326, 363
Constitutive principles Eulerian, 70
coordinate invariance, 7, 181, 182, Lagrangian, 70
191 material, 70
determinism, 181, 182 skew, 11, 16
equipresence, 181, 183 spatial, 70
local action, 181, 183 spherical, 287, 365
material frame indifference, 182, Covariant base vectors, 9–13
184 Covariant components, 15, 20
material symmetry, 182 Covariant derivative, 48–50, 53, 55,
physical admissibility, 182, 194 90, 164
Constitutive relations, 2, 173, 201, second, 54
224, 234, 235, 347–353 Cross product, 37, 40, 312
component form, 198 Curl, 53, 57, 317
compressible material, 197 Curl theorem, 61, 318
curvilinear anisotropy, 216–218 Curvature tensor, 54
experiments, 225–227 Cylinder (solid)
fundamental principles, 181–200 extension and torsion, 251–264
general, 197 torsion, 354–361
hyperelastic material, 197, 220
incompressible material, 219–222 Deformation
isotropic material, 209–212 area change, 102–103
linear elastic material, 223–224, deformation rates, 122–132
351–353 principal strains, 103–107
material symmetry, 208–218 shear, 96–98
orthotropic material, 215 strain tensors, 75–77
second Piola-Kirchhoff stress, 196 stretch, 91–96
transversely isotropic material, volume change, 98–102
213–214, 218, 224 Deformation gradient tensor, 72–74,
Contact force, 139, 143, 192, 335 108
Contravariant base vectors, 13–17 objectivity, 189–190
Contravariant components, 9, 20 Deformation gradients, 74
Convected derivative, 125 Deformation rate, 122–132
Coordinate system, 5 Deformation tensor, 74–75
Coordinate transformation, 6, 24–31, Determinant, 33
315–316 Differentiation
base vectors, 24–29 base vectors, 45–48
tensor components, 27–31, 316 tensors, 49–50
vector components, 29, 316 vectors, 48–49
Coordinate transformation tensor, with respect to vectors and tensors,
24–27 50–51
components, 36, 315 Dilatation, 353
Coordinates rate, 219
Cartesian, 311 ratio, 99, 101
Index 449
Direct notation, 18, 313 curl, 59

Displacement gradient, 89 determinant, 33
Displacement gradient tensor, divergence, 58
319–320, 327 gradient, 58
cylindrical coordinates, 327 tensor, 22
Displacement vector, 88, 319 trace, 32
Dissipation inequality, 180 transpose and inverse, 23
Divergence, 53, 57, 317 vector and dyadic, 19
formulas, 58 Frame indifference, 7, 182, 184–194
Divergence theorem, 61, 318
Dot product, 13, 312 Gastrulation, 285
double, 18, 313 Geometric nonlinearity, 67, 89, 119
single, 18 Gradient, 53, 317
Dyadics, 17–20, 312, 313 Gradient operator, 52–57, 72, 317
cylindrical coordinates, 326
Eigenvalue, 33–34, 115, 329 Gradient theorem, 59–61, 318
problem, 33, 104, 113, 114, 155,
329, 340 Heart, 80, 91, 226, 227, 244, 251, 291
Eigenvector, 33–34, 115, 329 aneurysm, 151, 152
Elastic constants, 223, 224, 351 embryo, 291–297
Elastic material, 182, 223–226, 351 left ventricle, 244, 277–285, 291
linear, 351–353 muscle, 152, 205, 236, 251
Embryonic heart, 291–297 papillary muscle, 78
Entropy, 179, 181, 196, 350 Heat, 173, 174, 177–179, 181, 347, 348
-δ identity, 40, 312 Helmholtz free-energy function, 195,
Equation of motion, 156–167, 234, 351
340–347, 353 Hooke’s law, 224, 352
component form, 162–167 Hyperelastic material, 196–198, 200,
cylindrical coordinates, 346, 347 208, 220
Eulerian form, 156, 157
Lagrangian form, 159–162 Identity tensor, 24, 36, 71, 314
local, 342 Incompressibility, 100, 101, 219–221,
Equilibrium equation, 157, 234, 342 234, 235
Equivalent motions, 185 Index, raising and lowering, 36
Eulerian Indicial notation, 18, 313
coordinates, 70 Inflation
description, 68, 122–124, 126, 145 left ventricle, 277–285
strain, 69, 75 spherical shell, 285–291
strain tensor, 75 tube, 82, 86, 101, 107, 129, 264–277
Experiments, 1, 225–227 Initial conditions, 234, 353
Extension, 79, 101, 107, 236–243 Internal energy, 173, 196, 347, 348
ratio, 92 Invariants
strain, 105–107, 205, 223
First law of thermodynamics, tensor, 31–34, 43
173–179, 194, 347 Inverse method, 235
Formulas Inverse of tensor, 24
450 Index
Irreversible process, 179 position vectors, 184–185

Isotropic material, 209–212, 223–224, rate-of-deformation tensor, 191
352, 353 second Piola-Kirchhoff stress
strain-energy function, 206–207 tensor, 194
spin tensor, 191
Kinematic relations, 234, 235 tensors, 187–188
Kinetic energy, 173, 174, 347 vectors, 185–187
Kronecker delta, 14, 311, 312, 314 velocity gradient tensor, 191
Ogden material, 206
Lagrange multiplier, 220, 221 Opening angle, 264, 267
Lagrangian Orthogonal tensor, 40–41
description, 68, 122–124, 145 Orthotropic material, 215, 224, 352
strain, 68, 75, 320, 322 strain-energy function, 201–203
strain tensor, 75
stress, 139 Papillary muscle, 78, 243, 251–264
Lamé constants, 224, 353 Permutation symbol, 37, 311, 312
Laplacian, 55, 359 Permutation tensor, 37–40
Law of Laplace, 289 Physical components, 41–44
Left ventricle, 244, 277–285, 291 strain, 77–78, 216, 217
Ligaments, 236, 243 stress, 153, 217
Linear elastic material, 223, 351–353 tensors, 42–44
Linear momentum, 156–157, 159, 341 vectors, 42
Polar decomposition, 108–112
Material derivative, 125 Position vector, 9, 70
Material nonlinearity, 67 Power, 173, 174, 347, 348
Material symmetry, 200, 208, 216 Poynting effect, 260
Matrix, 21–22, 314 Principal directions, 34, 103, 104,
Metric tensor, 35–36 111–116, 333
Mixed components, 20, 24, 25 strain, 329
Mooney-Rivlin material, 206, 207, 259 stress, 155, 340
Muscle Principal material axes, 201, 208, 216
heart, 152, 205, 236, 251, 277, 278 Principal strain, 103–107, 329–333
papillary, 243, 251–264 Principal stress, 155–156, 340
skeletal, 203, 236 Principal stretch ratios, 104, 113–118
smooth, 151, 264 Pseudoelasticity, 1
Myocardium, 152, 205, 226, 227, 231, Pseudostress, 139, 140, 145–146
277, 278, 282, 291, 296
Rate-of-deformation tensor, 127, 176,
Neo-Hookean material, 206, 290 191
Reference configuration, 68, 189, 318
Objectivity, 7, 182, 184–194 Reference frame, 5, 6, 31, 70, 182,
Cauchy stress tensor, 193 184, 190
deformation gradient tensor, Residual stress, 68, 264, 273–278,
189–190 283, 284, 291
first Piola-Kirchhoff stress tensor, Response function, 192, 212
193 Reversible process, 179, 181, 195, 350
Index 451
Riemann-Christoffel tensor, 55, 132 modified, 222

Rotation, 108, 111, 112, 120, 122, phenomenological approach, 225
157, 189, 309, 318, 327–329 Strain-rate tensor, 128, 129
Rotation tensor, 40, 108, 111, 113, Stress components, 147, 154, 338
185, 192 Cauchy, 154
linear, 120, 328 physical, 153, 154
Rotation vector, 328 relations between, 146–153
Stress power, 177, 178, 219
Second law of thermodynamics, Stress tensor, 141–146, 234, 235,
179–182, 194, 350 335–339
Semi-inverse method, 235, 354 Cauchy, 141–144, 159
Shear, 96–98 first Piola-Kirchhoff, 145, 160
Simple material, 183, 192 Kirchhoff, 146
Simple shear, 80, 93, 98, 101, 107, pseudo, 145–146
115, 243–251, 330 second Piola-Kirchhoff, 145, 146,
Spherical shell 159, 160
inflation, 285–291 true, 144
Spin tensor, 127, 176, 191 Stress-strain relations (linear),
Strain, 318–327 351–353
components, 76 Stretch, 68, 91–96
engineering, 68 Stretch ratio, 68, 92
Eulerian, 69, 75 Stretch tensors, 108, 112
extensional, 320, 329 Summation convention, 7, 71, 310
Lagrangian, 68, 75 Surface traction, 140, 174
physical components, 77–78
shear, 96, 321–324 Tendons, 236
Strain approximation, 119–122 Tensor, 19–21, 313
small deformation, 121 calculus, 44–62
small deformation and moderate components, 20
rotation, 122 contraction, 15, 31, 311
small displacement, 120–121 coordinate transformation, 24–27
small rotation, 122 covariant components, 20
Strain invariants, 105–107, 205, 223 invariants, 31–34
Strain tensor, 75–77 inverse, 24
Eulerian, 75 mixed components, 20
Lagrangian, 75 objectivity, 187–188
linear, 120, 325–327 orthogonal, 40–41
Strain-displacement relations, 88–91, physical components, 42–44
320 symmetry, 23, 32, 43, 314
cylindrical coordinates, 325 Tensor analysis, 7
linear, 325, 353 Tensor product, 17, 312
Strain-energy function, 196, 200–207, Thermoelastic material, 182, 183, 195
350–351 Thermomechanical continuum, 173
Blatz-Ko, 207, 241, 262 Time derivative, 122–126
experiments, 225–227 Eulerian, 123
microstructural approach, 225 Lagrangian, 123
452 Index
unit vectors, 129 Two-point tensor components, 74,

Torsion 100
circular cylinder, 82, 86, 101, 129,
251–264, 354–357 Uniaxial extension, 68, 115, 243
noncircular cylinder, 358–361 Uniqueness theorem, 354
tube, 107, 129, 264–277
Trace, 31–32 Vector, 7, 312
Traction vector, 140, 154, 155, 335 base, 7–13
pseudo, 140, 141, 144, 145, 154 calculus, 44–62
surface, 174 components, 15
true, 140, 143, 144
contravariant components, 9
Transpose, 22–24, 314
covariant components, 15
Transversely isotropic material, 236,
differentiation, 48–49
246, 352
constitutive relations, 213–214, objectivity, 185–187
218, 224 physical components, 42
strain-energy function, 203–205 Velocity, 123, 125, 156, 174, 186, 342
Tube Velocity gradient, 127, 191
everted, 118 Viscoelasticity, 1, 86, 122, 129, 182
extension, inflation, torsion, 82, 86, Volume change, 98–102
101, 107, 129, 264–277
inflation, 277–285 Wound healing, 94

Larry A Taber - Nonlinear Theory of Elasticity - Applications in Biomechanics (Revised Edition) - WSPC (2023)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Larry A Taber - Nonlinear Theory of Elasticity - Applications in Biomechanics (Revised Edition) - WSPC (2023)

Uploaded by

Copyright:

Available Formats

This page intentionally left blank

British Library Cataloguing-in-Publication Data

NONLINEAR THEORY OF ELASTICITY

ISBN 978-981-127-051-2 (hardcover)

For any available supplementary material, please visit

number of graduate courses devoted to nonlinear elasticity has dwindled,

Problems involving the deformation of soft biological tissues are among

Preface to the Revised Edition vii

Preface to the First Edition ix

2. Vectors, Dyadics, and Tensors 5

2.6.1 Metric Tensor . . . . . . . . . . . . . . . . . . . . 35

3.10.1 Time Rates of Change . . . . . . . . . . . . . . . . 122

4. Analysis of Stress 139

5. Constitutive Relations 173

5.4.1 Isotropic Material . . . . . . . . . . . . . . . . . . 209

6. Biomechanics Applications 233

6.6.4 Results . . . . . . . . . . . . . . . . . . . . . . . . 283

Appendix A Linear Theory of Elasticity 309

Appendix B Special Coordinate Systems 363

B.1 Cylindrical Polar Coordinates . . . . . . . . . . . . . . . . 363

Problem Solutions 369

Problems involving deformation of soft biological tissues are among the

1.1 Elasticity and Pseudoelasticity

presumably due to passive microstructural remodeling. After several load-

1.2 Biomechanics and Mechanobiology

Biomechanics can be deﬁned as the study of mechanical function and mal-

Vectors, Dyadics, and Tensors

Physical quantities can be represented by tensors. Zeroth-order tensors

2.1 Reference Frames and Coordinate Systems

The transformation of components from the Cartesian to the polar system

Now, consider another reference frame B that is translating and ro-

A vector a has a magnitude and direction. Relative to a given coordinate

2.2.1 Base Vectors

nents and base vectors in diﬀerent coordinate systems.

where the ai are called the contravariant components (superscript i)

as the component of dr in the gi direction. Then, resolving the vector dr

and comparison with (2.5) shows that

According to Eq. (2.6), the dxi are the contravariant components of dr

(1) Cartesian coordinates: (x1 , x2 , x3 ) = (x, y, z).

whereas y2 is a contravariant vector component.

(2) Cylindrical polar coordinates: (x1 , x2 , x3) = (r, θ, z) with x = r cos θ

Fig. 2.4. Cylindrical polar (a) and skew (b) coordinates.

Solution. Relative to these coordinate systems, the position vector to a

∂er ∂eθ ∂er ∂eθ

Example 2.2 For cylindrical polar coordinates, determine gr and gθ with-

Solution. In polar coordinates, the position vector is

and diﬀerentiation yields

which shows that the derivative has unit magnitude. Moreover, as dθ →

2.2.2 Reciprocal Base Vectors

in the xi coordinate system. [Simmonds (1994) calls the gi and gi “cellar”

is the Kronecker delta. Thus, g1 is orthogonal to both g2 and g3 , and

Of course, a vector can be written in terms of components relative to

b = bi gi , where the bi are the covariant components of b, then

Any vector can be expressed in terms of either its contravariant or co-

where ai , bi , and ci are Cartesian components. (The determinant represen-

Equations (2.18) give the reciprocal base vectors

2.3 Dyadics and Tensors

Table 2.1. Vector and dyadic formulas

The other type of double-dot product is the horizontal double-dot

ab · · cd = (a · d)(b · c) = cd · · ab, (2.29)

which also is commutative for dyads. These operations can be extended to

the vector into another vector, i.e.,9

T = T ij gi gj = Tij gi gj = T·ji gi gj = Ti·j gi gj . (2.34)