Continuum Mechanics

ONLINE LIBRARY
Physics and Astronomy
http ://www.springer.de/phys/

Springer-Verlag Berlin Heidelberg GmbH

Advanced Texts in Physics
This program of advanced texts covers a broad spectrum of topics which are of
current and emerging interest in physics. Each book provides a comprehensive and
yet accessible introduction to a field at the forefront of modern research. As such,
these texts are intended for senior undergraduate and graduate students at the MS
and PhD level; however, research scientists seeking an introduction to particular
areas of physics will also benefit from the titles in this collection.
I-Shih Liu

Continuum Mechanics
With 28 Figures and Numerous Exercises

Springer
Professor I-Shih Liu
Universidade Federal do Rio de Janeiro
Inst ituto de Matematica
C.P.6853°
21945-970 Rio de Janeiro, Brasil

Library of Congress Cataloging-in-Publication Data

Liu, I-Shih, 1943-

Continuum mechanics / I-Shih Liu.
p. em. -- (Advanced texts in physics, ISSN 1439-2674)
Includes bibliographical references.

I. Continuum mechanics . I. Title . II. Series.

QA808 .2 .L58 2002

531--dc21
2002017040
ISSN 1439-2674
ISBN 978-3-642-07702-9 ISBN 978-3-662-05056-9 (eBook)
DOI 10.1007/978-3-662-05056-9

This work is sub ject to copy right. All rights are reserved, whether the whole or part of the material
is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broad-
casting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of
this publication or parts thereof is permitted only under the provisions of the German Copyright Law
of September 9, 1965, in its current version, and permission for use must always be obtained from
Springer-Verlag. Violations are liable for prosecution under the German Copyright Law.

http://www.springer.de
© Springer-Verlag Berlin Heidelberg 2002
Originally published by Springer-Verlag Berlin Heidelberg New York in 2002 .
Softcover reprint of the hardco ver I st edition 2002
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
impl y, even in the absence of a specific statement, that such names are exempt from the relevant pro-
tective laws and regulations and therefore free for general use.

Typesetting by the authors using a Springer TI;X macro package

Cover design: design & production GmbH , Heidelberg
Printed on acid-free paper SPIN: 10858104 54!3141/mf 54 3210
To my t eacher
Ingo Mtiller
Preface

In t his book the basic principles of cont inuum mechanics and thermodynam-
ics are treated in t he tradition of the rational framework established in the
1960s, typically in the fundament al memoir "T he Non-Linear Field Theories
of Mechan ics" by Truesd ell and Noll. The theoreti cal aspect of constitutive
t heories for mater ials in general has been car efully developed in mathemati-
cal clarity - from general kinem atics, balance equat ions, material obj ectivity,
and isotropic representations to the framework of rational thermodynamics
based on the ent ropy principle. However , I make no claim that the subjects
are covered complete ly, nor do es this book cover solutions and examples that
can usually be found in textbooks of fluid mechanics and linear elast icity.
However , som e of the int eresting examples of finit e deformations in elastic
materials , such as biaxial stret ching of an elast ic membran e and inflation of
a rubber balloon , ar e discussed .
In the last two chapte rs of t he book, some recent developments in ther-
modynamic theori es are considered . Sp ecifically, t hey emphasize the use of
Lagrange multipliers , which enables the exploi t ation of t he entropy principle
in a systematic manner for const it ut ive equations, and introduce some basic
notions of ext ended thermodynamics. Although extended thermodynamics is
closely relat ed to the kineti c t heory of ideal gases , very limit ed knowledge of
kinetic t heory is needed .
Earlier versions of this book have been used over the years, in the Insti-
tut e of Mathematics at the Federal University of Rio de Janeiro as well as in
the Institute of Applied Mechani cs at the National Taiwan University, in an
introductory cou rse on continuum mechanics at the graduate level, and at
the advanced undergraduat e level with a simplified version. The readers ar e
not required to have a good knowled ge of either solid mechanics or fluid me-
chanics, but, of course, some prior acquaint ance with them would be helpful.
An appendix is written at the end to provide a review of basic notions
in linear algebra and t ensor an alysis as mathematical pr eliminaries for the
subject s, and occasionally cross-references to it (e.g. (A.32)) are used in the
text. The reader who already has a reasonable mathematical knowledge may
refer to it for refer ence and notations. However, in introductory cour ses I
have oft en put the appe ndix before t he first chapte r because most of the
st udents may not be fami liar with the notations and some basic notions. No
VIII Preface

effort has been made to compile an extensive bibliography on related works

in continuum mechanics. Only those cited in the book are listed.
Examples and exercises are given to supplement the understanding of the
material and sometimes to provide further insights into the subjects. Usually
my students are asked to do most of the exercises to accompany the progress
of learning. Their feedback on the errors and the difficulties has resulted in
considerable improvement of the manuscript. Their participation is greatly
appreciated.
The endeavor of writing this book depended on many ideas and work in
the scientific literature. To many of the relevant researchers, acquaintances or
not, are due my grateful acknowledgements for their contributions. My special
acknowledgement is due to Prof. Muller for his friendship and inspirations on
many of my scientific trajectories. Finally, I would like to thank my family,
especially my wife Lu Ping, for their understanding and patience during many
long hours of preparing the manuscript over the years.

Rio de Janeiro,
March 2002 I-Shih Liu
Contents

1. Kinematics . . .. . .. . .. . . . . . . .. . . . .. .. . ... .. ... . . .. . .. . .. . . . 1

1.1 Configuration and Deform ation . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Chan ge of Reference Configuration 4
1.2 Strain and Rotation 4
1.3 Linear Strain Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Motion . . . .. . . . .. . . . . .. . . . .. . ..... . . .. .. . .. . . . . . . . .. . . . 13
1.4.1 Material and Spatial Descriptions . . . . . . . . . . . . . . . . . . 14
1.5 Relat ive Deforma tion 17
1.6 Rat e of Deformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.7 Change of Fram e and Obj ect ive Tensors . . . . . . . . . . . . . . . . . .. 22
1.7.1 Transformation Property of Motion. . . . . . . . . . . . . . . . . 25
1.7.2 Property of Some Kinemati c Quantities . . . . . . . . . . . .. 26

2. Balance Laws . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . .. . . . 31
2.1 General Bal ance Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1.1 Field Equation and Jump Condition . . . . . . . . . . . . . . . . 35
2.1.2 Balan ce Equations in Mat erial Coord inat es . . . . . . . . .. 36
2.2 Conservation of Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 38
2.3 Laws of Dyn am ics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.3.1 Forces and Moment s. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 42
2.3.2 Stress Tensor 43
2.3.3 Cons ervation of Linear and Angular Momenta 50
2.4 Cons ervation of Energy 51
2.5 Summar y of Basi c Equ ations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.5.1 Basic Equ ations in Material Coordinates. . . . . . . . . . . . 56
2.5.2 Boundar y Conditions of a Material Body . . . . . . . . . . .. 57
2.6 Field Equations in Arb itrary Fram es . . . . . . . . . . . . . . . . . . . . .. 58

3. Basic Principles of Constitutive Theories 63

3.1 Constitutive Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.2 Principle of Material Obj ectivity 65
3.2.1 In Referential Descrip tion 68
3.2.2 An Example: a P articular Class of Materials . . . . . . . . . 70
3.3 Simple Mat erial Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 72
X Contents

3.4 Redu ced Constitutive Relations 75

3.5 Material Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.5.1 Constitutive Equation for a Simple Solid Body . . . . . . . 81
3.5.2 Constitutive Equation for a Simple Fluid. . . . . . . . . . . . 82
3.5.3 Fluid Crystal with an Intrinsic Direction . . . . . . . . . . . . 84
3.6 Isotropic Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.6.1 Constitutive Equation of an Isotropic Material . . . . . .. 88
3.7 Fading Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.7.1 Linear Viscoelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.7.2 Boltzmann-Volterra Theory of Viscoelasticity . . . . . . . . 92
3.7.3 Linear Viscoelasticity of Rate Type " 93
3.7.4 Remark on Obj ectivity of Linear Elasticity . . . . . . . . . . 94

4. Representation of Constitutive Functions . . . . . . . . . . . . . . . .. 97

4.1 Materials of Grade n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 97
4.2 Isotropic Functions 98
4.2.1 Isotropic Elastic Materials and Linear Elasticity 107
4.2.2 Reiner-Rivlin Fluids and Navier-Stokes Fluids 109
4.2.3 Elastic Fluids III
4.3 Representation of Isotropic Funct ions 112
4.3.1 Isotropic Thermo elastic Solids
and Viscous Heat-Conducting Fluids 118
4.4 Hemitropic Invar iants 119
4.5 Anisotropic Invar iants 122
4.5.1 Transverse Isotropy and Orthotropy 124
4.5.2 On Irreducibility of Invariant Sets 126

5. Entropy Principle 129

5.1 Entropy Inequality 129
5.2 Entropy Principle 131
5.3 T hermodyna mics of Elastic Materials 132
5.3.1 Linear Thermo elasticity 135
5.4 Elastic Materials with Internal Constraints 139
5.5 St ab ility of Equilibrium 144
5.5.1 Thermodyn ami c St ability Criteria 148
5.6 Phase Equilibrium 149

6. Isotropic Elastic Solids 153

6.1 Constitutive Equations " 153
6.2 Boundary Value Problems in Elasticity 155
6.3 Homogeneous Stret ch 157
6.3.1 Uniaxi al Stret ch 158
6.3.2 Biaxi al Stretch 159
6.4 Symmetric Loading of a Square Sheet 160
6.4.1 Stability of a Squ are Sheet 162
 Contents XI

6.5 Simpl e Shear 166

6.6 Pure Shear of a Squ are Block 169
6.7 Finit e Deform ation of Spherical Shells 173
6.7.1 Eversion of a Spherical Shell 175
6.7.2 Inflation of a Spherical Shell 176
6.8 St ab ility of Spherical Shells 179
6.8.1 St ability under Cons tant Pressures 180
6.8.2 Stability for an Enclosed Spherical Shell 181

7. Thermodynamics with Lagrange Multipliers 183

7.1 Supply-Fr ee Bodi es 183
7.2 Viscous Heat-Conducting Fluid 184
7.2.1 General Results 186
7.2.2 Navier- Stokes- Fourier Fluids 188
7.3 Method of Lagrange Multipliers 189
7.3.1 An Algebraic Problem 190
7.3.2 Local Solvability 191
7.4 Relation Between Entropy Flux and Heat Flux 194
7.4.1 Theorem of Parallel Isotropic Vector Functions 194

8. Rational Extended Thermodynamics 199

8.1 Introduct ion 199
8.2 Formal Structure of Syst em of Balan ce Equations 200
8.2 .1 Symmetric Hyp erbolic Syst em 201
8.2.2 Galilean Invari an ce 204
8.3 Syst em of Moment Equ ations 207
8.4 Closure Problem 213
8.4.1 Ent ropy Principle 214
8.4.2 Form al Procedures 216
8.5 Thirt een-Moment Theory
of Viscous Heat-Conduct ing Fluid 217
8.5.1 Fi eld Equa t ions 223
8.5.2 Entropy and Entropy Flux 225
8.6 Monatomic Ideal Gases 226
8.6.1 Thirteen-Moment Theory 227
8.6.2 Constitutive Equations 228
8.7 St ationary Heat Conduction in Ideal Gases 228
8.7.1 Fouri er's Law and Heat Conduction 229
8.7.2 Heat Conduction in Thirteen-Moment Theory 229
8.7.3 Remark on Boundary Valu e Problems 232
XII Contents

A. Elementary Tensor Analysis 233

A.l Linear Algebra 233
A.I.l Inn er Product 234
A.I.2 Dual Bases 235
A.I.3 Tensor Product 238
A.I.4 Transformation Rules for Components 243
A.I.5 Determinant and Trace 245
A.I.6 Exterior Product and Vector Product 251
A.I.7 Second-Order Tensors . . , 254
A.I.8 Some Theorems of Linear Algebr a 256
A.2 Tensor Calculus 262
A.2.1 Euclidean Point Space 262
A.2.2 Differentiation 263
A.2.3 Coordinat e System 272
A.2.4 Covariant Derivatives 275
A.2.5 Other Differential Operators 277
A.2.6 Physical Components 281
A.2.7 Orthogonal Coordinate Systems 282

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289

Index 293
1. Kinematics

1.1 Configuration and Deformation

In continuum mechani cs, we are int erest ed in material bodies that can un-
dergo motions and deformations. The stage for such ph enom ena to occur is,
of cour se, t he four-dimensional space-time . In this book, we shall restrict our
a ttent ion to the Newtonian space - time of classical mechanics. The Newtonian
space-time W can be regarded as a product space of a three-dimensional Eu-
clidean space E and t he one-dimensional space of real numbers JR through a
one-to-one mapping
¢ : W -+ E x JR.
Such a mapping is called a fra m e of referen ce. To set the st age for any dis-
cussion hereafter a frame of reference is always chosen, eit her explicitl y or
impli citl y. We will usu ally think of a frame of reference as an "observer" .
Different observers will measure spac e-t ime event s in different ways. The
relat ion between two different fram es of reference will be discussed later .
In order to describe the presence of a body B in space, mathematically
we sha ll identify it with a region in a three-dimensional Euclidean space £
relativ e to a fram e of reference. We call a one-to-one mapping from B into £
a configuration of B. Such an identification endows the (physical) body with
the mathematical structure of a Euclidean space.
It is mor e convenient to choose a par ticular configuration of 5 , say K" as
a reference,
K, : B -+ E, K,(X) = X . (1.1)
We call K, a referen ce configuration of B . The coordinat es of X , (X C> , a =
1,2,3) are called the referential coordin ates, or more commonly the material
coordinates, since the point X in t he reference configuration is often identifi ed
with the material point X of the body. The body 5 in the configurat ion K,
will be denoted by BK •
Let K, be a reference configuration and X be an arbit rary configurati on of
B. Then the mapping

(1.2)
is called the deformation of B from K, t o X (Fig . 1.1) . In t erms of coordi-
nate syste ms (xi ,i = 1,2,3) and (X C> , a = 1,2 ,3) in t he deformed and t he

I.-S. Liu, Continuum Mechanics

© Springer-Verlag Berlin Heidelberg 2002
2 1. Kin ematics

reference configurations, respe ctively, the deformation XI< can be expressed

as
Xi = Xi(X Q ) , (1.3)
where Xi are called the deformation functions.
The deformation gradient of X relative to K , denoted by FI< is defined by

(1.4)
When the reference configurati on K is chosen and und erstood in the context,
FI< will be denoted simply by F . Since the mapping XI< in (1.2) is one-to-
one and onto, F is non-singular. Therefore, the determinant of F must be
different from zero ,
J = detF =I- O. (1.5)

.
X

BI< ;/ ~ Bx

X . XI<
.
X

Fig. 1.1. Refer ence configurati on

Relative to the coordinate systems (X and (Xi) in the reference and the
Q
)

deformed configurat ions, respectively, F can be expressed in the following

component form ,

. aXi
F 'Q = aXQ' (1.6)

The deformation gradient F in this expression is a two-point t ensor, becaus e

it is expressed in t erms of two different bases for a cont ravariant component
at x in the deformed configurat ion and a covariant component at X in the
reference configur at ion (see Example A.2.7 on p. 274). This component form
is particularly simple, because it only involves partial derivative s, even if the
deformation function (1.3) is not given in the Cartesian coordinate systems.
By definition, the deformation gradient F is a linea r transform ation on
the translation space V of E (see Sect . A.2.1) , F : V -+ V , such t hat

(1.7)
 1.1 Configuration and Deformation 3

wher e 0(2) repr esent s the higher-order terms in E = IX - X ol, Le.,

lim 0(2) = o.
<-+ 0 E

In other words, the value of F( X 0) determ ines X" (X) to within an error of
the second-order near X o relative to the valu e of X,,(X o).
Let dX = X - X o, a vector in V , be regarded as a small (infinitesi-
mal) mat erial line element in the reference configuration. Since F is a linear
tran sformation on V , the image of the line element in t he deform ed state
F(Xo)dX , denot ed by the vector dx E V , is t he first-ord er approximat ion of
X,, (X ) - X,,(X o) at X o from the relation (1.7) . We writ e

dx = FdX . (1.8)

Similar relations for material surface elements and volum e elements in the
deformed state are derived in t he following example.

Example 1.1.1 Let da.; and n" be a small material surface element
and its unit normal in the reference configurat ion and da and n be the
corre sponding ones in the deformed configur ation . Also, let do; and dv
be small material volume elements in the reference and the deform ed
configurat ions, respectively. Then we have

nda = JF- T n"da" , dv = IJI dv" . (1.9)

Proof Let the small rect angular surface element da.; be formed from
the two line elements dX 1 and dX 2 . Then we have from (1.8)
dX 1 X dX 2 FdX 1 X FdX 2
n=-------
n " = IdX 1 x dX 2 1' IFdX 1 x FdX 2 1'
do.; = IdX 1 x dX 21, da = IFdX 1 x FdX 21.
Therefore, for any vector v , we have from (A.32)

v · nda = v · FdX 1 x FdX 2 = F(F-1v) . FdX 1 X FdX 2

= J F -1v ' dX 1 X dX 2
= J F -1v . n "da" = v . J F - T n "da" .

Similarly, let t he small volume element dv" be formed from three line
elements dX 1 , dX 2 , and dX 3 . Then we have

dv = Idxl . dX2 x dX31 = IFdX l ' FdX 2 x FdX 3 1

= IdetFlldX 1 ·dX 2 x dX 3 = IdetFldv" .

1

Not e that if det F = 1, the deformation is volume preserving. 0

4 1. Kinematics

1.1.1 Change of Reference Configuration

Let k be another reference configuration of B. The map

>. = k 0 ", -1 : E --+ E;

is called a change of reference configuration from", to k , and the deformations

from the two different reference configurations r; and k to X are related by
(Fig . 1.2)

x x x

.j
x .x

Fig. 1.2. Change of reference configuration

XK = Xi< 0 >. ,
which implies that

In other words , we have

P=V'X(kO",-l) . (1.10)

These are the relations between the deformation gradients with respect to
two different reference configurations.

1.2 Strain and Rotation

The deformation gradient is a measure of local deformation of the body. In
this section we shall introduce other measures of deformation that have more
suggestive physical meanings, such as change of shape and orientation.
Since the deformation gradient F is non-singular, by the polar decompo-
sition theorem (see Sect . A.1.S) , there exist two positive definite symmetric
 1.2 Strain and Ro t ation 5

t ensors U and V , and an orthogonal tensor R , uniquely det ermined by F ,

such that
F= RU = VR (1.11)
hold s.
To int erpret this decomposition in physical terms, we observe that a pos-
it ive definite symmetric tensor represents a st at e of pure st retches along three
mutually orthogon al axes (sp ectral theorem , Sect . A.1.8) and an orthogonal
te nsor represent s a rotation . Therefore, (1.11) assures that any local defor-
mation is a combination of a pure stretch and a rot at ion : first stret ch U then
rot ation R or first rotation R then stretch V .
We call R the rotation tensor, while U and V are called the right and
t he left stretch tensors, resp ectively. The right and left , of course, refer to the
position t hey appear at the "right" and "left" in the resp ectiv e express ions of
the decomposition (1.11) . Both stretch t ensors measure the local st rain, the
cha nge of shape, while the t ensor R describes the local rot ation , the change
of orientation, expe rienced by material element s of t he body.
Clearly we have

(1.12)
det U = det V = Idet Fl .
Let t he eigenvalues and the eigenvectors of U be Vi and e. , i = 1,2 ,3,
(no sum) .

Then , since V = RU R T , we have

In other words , V and U have the same eigenvalues and their eigenvect ors
differ only by t he rotation R ..T he eigenvalues Vi are called t he prin cipal
stretches, and the corre sponding mutually orthogonal eigenvect ors are called
t he prin cipal directions.
Given a deformation gradient F , the stret ch t ensors U and V ar e de-
termined from the relation (1.12h by taking t he "square roo t " (see p. 259)
of the positive definite symmetric t ensors F T F and F F T , respect ively. In
pr acti cal calculat ions, it is mor e convenient to introduce the right and t he
left Cauchy-Green strain tensors defined by

c = U 2 = F TF, B = V 2 = FF T , (1.13)

resp ectively, as alte rnat ives to the st rain measures U and V. In component s,
they ar e given by

B ij = g o:{3 F i0:
F j
{3'
6 1. Kinemat ics

where gi j and g Qf3 are the covariant and the contravariant component s of the
corresponding metric tensor of the coordinate syst ems in the cur rent and the
reference configurations. In matrix form , they can be written as

Example 1.2.1 Consider a deformation x = X,, (X) given in Car t esian

coordinat es, in both the reference and the deformed configurations, by

X = X + KY ,
y=Y, (1.14)
z = Z.

This deformation is called a simple shear and K > 0 is called the amount
of shear (see Fig. 1.3). We have

(1.15)

and sin ce det F = 1, simple shear is volume preserving. We can eas ily
obtain

y ,Y

T1

1 t - . . -_ _ -L-----... x, X

Fig. 1.3. Simple shear

 1.2 St rain and Rotation 7

From t he square root of C , we obtain

J4 '+" ",2
2
o
J4 + ",2
[Ua (3 ] = 2 + ",2
'"
J4 + ",2 J4 + ",2 o '
0 0 1

and t he principal st retches,

V3 = 1,
wit h the correspo nding principal direct ions,

Note that in t he pr incipal direction el (with t he posit ive sign), t he prin-

cipal stretch, Vl > 1, is an extensio n, while in the direct ion e2 (wit h the
negative sign), the stretch, V2 < 1, is a contraction. Similarly, we have

r.
2 + ",2
J4 '+" ",2
0
J4 + ", 2
[B ij] = ~
'1"
0
~] , [Vij] =
J4 '+" ",2
2
J4 + ",2
0

0 0 1

T he rotation te nsor can be calc ulated from R = FU- 1 ,

2
J4 '+" ",2
0
J4 + ",2
[Rial = -'" 2
0
J4 + ",2 J4 + ",2
0 0 1

Note t hat if we denot e e = tan- 1 (",/2), t hen R becomes

cose sin e 0]
[Ria] = -sin e cose 0 ,
[ o 0 1

which is a clockwise rotation about the z-axis by t he angle e. 0

8 1. Kinematics

Exercise 1.2.1 Consider a torsion and extension of a cylinder given by

the deformation function in cylindrical coordinates (R, 8 , Z) and (r, (), z)
in the reference and the current configurations, respectively,
1
r=J(j,R, ()=8+TZ, z= -Z. (1.16)
a
Determine the matrices [Pia]' [Bij], [Ca l3] and the matrices of their cor-
responding physical components. Moreover, verify that the deformation
is volume preserving. Note that the pure torsion (for a = 1) is locally a
simple shear of amount -rr,

1.3 Linear Strain Tensors

The strain tensors in the previous section are introduced for finite deforma-
tions in general. In the classical linear theory, only small deformations are
considered. For the passage from the general theory to the linear theory,
we shall first give some geometrical meanings of the Cauchy-Green strain
tensors, C and B .
Consider two infinitesimal material line segments dX 1 and dX 2 in the
reference configuration and their corresponding ones dXl and dX2 in the
current configuration. From (1.8), we have
dXl . dX2 = PdX 1· PdX 2 = (pTp)dX 1 • dX 2 = CdX 1 · dX 2 •
Therefore we can consider the following quantity of change between the ref-
erence and current configur at ion for length and angle,
(1.17)
where
1
E=-(C-1) (1.18)
2
is called the Green-St. Venant strain tensor, or the finite strain tensor in the
reference configuration.
Similarly, since dX 1 = p -1dxl, we also have
dX i : dX 2 = dXl . (F-T P -1)dx2 = B -1dxl . dX2,
and hence

where
e=~(1-B-l) (1.19)

is called the Almansi-Hamel strain tensor, or the finite strain tensor in the
current configuration (see Sect. 31 of [72]).
 1.3 Lin ear Strain Tensors 9

Both the st rain t enso rs E and e vani sh when t here is no deformation,

i.e., F = 1 . For small deform at ions, t hese strain t ensors are therefore sm all
qu antities.
In order to consider small deform ations, we introduce the displa cem ent
vector from the reference configuration K, (see F ig. 1.4),

and its referential gradi ent,

H(X) = \l x u(X) .
On t he other hand , we ca n also write t he displacement vect or as
u(x) =x- X~l (x) ,

and its spa t ial gradient as

h(x) = \l",u( x) .
Obviously, we have
F = 1 +H, F - 1 = 1 - h,
and hence in te rms of t he displacement gradients , we obtain

E = ~ (H + H T + H T H) ,
(1.20)
e = ~(h + hT - hTh) .

x = X,,(X)

Fig. 1.4. Displ acem ent vector

For small deformations, let us write

F = 1 + Fe '
where Fe is a small qu antity, lFe I « 1. Hence
H=Fe ,
10 1. Kinematics

and since H T H is then a second-order quantity, we can write from (1.20h

E = E + 0(2) ,
where
(1.21)

is called the infinitesimal strain tensor. This linear strain tensor is introduced
by Cauchy in the classical theory of elasticity.
For small displacement gradients, the right stretch tensor U and the
rotation tensor R can be approximated by

U = -1FT F = 1 + ~(H + H T) + 0(2) = 1 + E + 0(2) ,

(1.22)
= FU- = 1 + ~(H - H ) + 0(2) = 1 + R+ 0(2),
I T
R
where
(1.23)

is called the infinitesimal rotation tensor. Note that infinitesimal strain and
rotation are the symmetric and skew-symmetric parts of the displacement
gradient.
In view of (1.17) we can give geometrical meanings to the components
of the infinitesimal strain tensor E relative to a Cartesian coordinate system.
First, let dX I = dX 2 = soel be a small material line segment in the direction
of the unit base vector el and s be the deformed length. Then we have
2 2 2 ~
S - So = 2s o (Eel · ej ) ,
which implies that

Ell = s2 - s~ = (s - so)(s + so) ~ s - So

2s~ 2s~ - So

In other words, Ell is the change of length per unit original length of a
small line segment in the ej-direction. The other diagonal components, E22
and E33 have similar interpretations as elongation per unit length in their
respective directions.
Now, let dX I = soel and dX 2 = Soe2 and denote the angle between
the two line segments after deformation bye. Then we have

So2 IF ellIFe2 I cose - So2 cos "2

7f 2-
= 2s o (Eel · e2),

from which, if we write, = 7f /2 - e, the change from its original right angle,
then
sin, EI 2
2 IFeIIIFe21·
 1.3 Linear Strain Tensors 11

Since lEd « 1 and IFeil ':0:' 1, it follows t hat sin j ':0:' I and we conclude that

Therefore, the component E 12 is equal t o one-half the change of angle between

the two line segment s 2riginall,t along the el- and e2-directions. Other off-
diagonal components, E 23 and E 13 have similar int erpret ations as the change
of angles.
Moreover , since det F = det (1 + H) ':0:' 1 + tr H for sm all deform ations,
by (1.9h we have , for a small material volume,

-
E ii = tr H dv -dv",
':0:' - - - - , - - -
dv",

Thus t he sum of the diagonal components of E i j measures the ~nfinitesimal

change of volum e per unit original volum e. We call t he trace of E the dilata-
tion.
Furtherm ore, we can also int erpret the component s of the infinitesim al
rotation t ensor Ii as "average" local rot ations. Let dX = so(cos 0 e l -l-sin 0 e 2)
be a small line segment in the Xl-X2 plan e making an angle 0 with the Xl
axis. We ar e interested in the change of angle after the deformation in the
plan e. Aft er deformation, its image dx = FdX ca n be written as

dx = dX + EdX + RdX .
The segm ent dx , in gener al, is not a vector in the Xl- X2 plan e. Let its pro-
jection vect or onto the plan e be denoted by dXh and its length by s. If we
denote the angle of rotation from dX to dXh by w, then we have the following
relation :
dX X dx · e3 = dX X dXh . e3 = sos sin w.
Therefore, we obtain

sos sinw = So2 (cos2 0 E- 2l - 2 - --

sin 0 E 12 - sin Ocos O(E l l - E 22)
+ cos 2 0 R- 2l - sin20 -
R 12)
= So (R 2l + cos 20E 12 - 21sin
2- - · -
20(Ell -
-
E 22)) .

Since s ':0:' So and t he angle w is small, we obtain

Not e that the rotation angle w in the Xl- X2 plan e depends on t he orientation
of the segme nt dX . However , if we define the average rotation in the plan e
by

<w> = 27r
1
w(O) dO ,
t"
la'
12 1. Kinem at ics

then after integration of the above expression, it follows that

Therefore, even though the rotation of an individual line segment depends

on R, as well as E , the average local ro,..!;ation at a material point in the
X I -X2 plan e ~ equal !.O t he component R 2 1 • The other two non-van ishing
component s R 13 and R 32 have similar int erpret ations.

Remark. A similar treatment for small deformations, F = 1 + FE,

can also be based on the spat ial displacement gra dient h . Indeed , since
F - 1 = 1 - FE+ 0(2), we have h = FE + 0(2) or

h = H + 0(2).
In other word s, t he two displacement gradient s

and

have approximat ely the sa me valu e for small deform ations. Therefore,
since in the classical linear theory, t he nonlinear terms are insignificant ,
it is usu ally not necessar y to introduce t he reference configurat ion in
t he linear theory. The classical infinitesim al strain and rot ation , in the
Cartesian coordinate system, are oft en defined as

(1.24)

in the current configurat ion. 0

Exercise 1.3.1 Consider the simple shear deform ation given in (1.14) .
1) Det ermine th e displacement vector u .
2) Assuming K « 1, det ermine the infinitesimal strain and the infinites-
imal rotation t ensors for simple shear of the classical theory.
3) Compare with t he pr evious results for finite shear st rains .

Exercise 1.3.2 Consider the pure torsion (a = 1) given in (1.16) .

1) Determine t he finite Gre en-St . Venant strain te nsor.
2) Det ermine the infinitesimal strain and the infinitesimal rotation t en-
sors, assuming a small displacement gradi ent .
3) Compare the results between finit e and small strain s.
 1.4 Motion 13

Exercise 1.3.3 Show that the infinitesim al strain t ensor E ij satisfies

the so-called com pati bility con diti on,
"k qJ"1 -
e pt e Eij ,kl = 0,

wher e the permutation symbols have been used (see p. 250) . In other
word s, since the linear st rain t ensor is the symmetric par t of the displ ace-
ment gradi ent , its components are not ind ependent fun ctions. Their sec-
ond partial derivatives must satisfy the compat ibility condit ion for the
existe nce of t he displ acement function . In Car t esian coordinates, they
read
- - -
E ll ,22 + E 22 ,1l - 2 E 12 ,12 = 0,
-
E 22 ,33 + E-33 ,22 - 2 E- 23 ,23 = 0,
- - -
E 33 ,1l + E ll ,33 - 2 E 3 1,3 1 = 0,
-- -- --
E 12 ,23 + E 23 ,12 - E 22 ,3 1 - E 3 1,22 = 0,
,.....,

-- --
E 23 ,3 1 + E 3 1,23 - E 33 ,12 - E 12 ,33 = 0,
-.... ""'-'

-- -- - --
E 3 1, 12 + E 12 ,3 1 - E ll ,23 - E 23 ,1l = 0.

1.4 Motion

A motion of l3 ca n be regarded as a cont inuous sequence of configur at ions in

time, i.e., we call X = {Xt , t E JR I Xt : l3 -+ £} a motion of l3. The body in
t he configuration at time t will be denoted by l3t .
The motion X of l3 can be expressed as a map,

X : l3 x JR -+ E, x = X(X , t) = Xt(X) ,
and given a reference configurat ion K, by (1.2) it can also b e expre ssed as

(1.25)

so that
XK ( " t) = X t 0 K-
1
: l3K -+ l3t
is a deform ation of the body B from K to Xt . Therefore, a motion can also
be regarded as a cont inuous sequence of deformation from a reference config-
ur ation in t ime.
For a fixed mat erial point X ,

is a curve called the path (or traj ectory) of the material point X . The vel ocit y
v and t he accelerat ion a are defined as the first and the second time derivative
14 1. Kinematics

of the position as it moves along the path of the material point X .

aXI«X , t)
v : HI< X JR --+ V v = at '
(1.26)
a2XI«X , t)
a= at 2 '

where V is the transl ation space of E. The velocity and the acce leration ar e
vector qu antities. Here , of cour se, we have assumed that XI«X , t) is twi ce
differentiable with resp ect to t. For simplicity, hereaft er we shall assume that
all functions ar e smooth enough for the conditions needed in t he conte xt ,
without their regulariti es explicit ly specified.

1.4.1 Material and Spatial Descriptions

A mat er ial body is endowed with some physical properties whose values may
chan ge along with the deformation of the body in a motion. A qu antity
defined on a motion can be described in essentially two different ways : eit her
by t he evolut ion of its value along the path of a material point or by the
change of its valu e at a fixed location in the current configurat ion of t he
body. The form er is called a material descr iption and the latter a spat ial
description. We shall make t hem mor e precise below.
For a given motion X and a fixed referenc e configurat ion «, consider a
quantity, with its valu e in some space W , defined on the motion of H by a
function
f : HI< X JR --+ W. (1.27)
Then it can also be defined on the current configuration at any time t ,

j( . , t ) : B, --+ W,
by
j(x ,t) = j(XI«X , t) ,t) = f(X ,t). (1.28)
As a custom in continuum mechanics, on e usu ally denotes the fun ctions
f and j by the sa me symbol, since they have the sa me value at the corre-
sponding point, and write, by an abuse of not ations,

f = f(X , t) = f(x, t) ,

and call the form er t he mat erial description (or referential description to
emphasize the pr esence of a reference of configurat ion ) and t he latter t he
spatial description of the fun ction f. Sometimes the mat erial descript ion is
referred to as the Lagrangian description and the spatial descrip tio n as the
Eulerian description.
Po ssibl e confusions may arise in this abuse of not ations, espec ially when
differentiations ar e involved . Of cour se, one way of avoiding this is t o write
 1.4 Motion 15

out explicit ly the variables concerne d , for example, otf (X , t) to mean the
time derivative with X held fixed in the mat erial description and otf(x, t)
with x held fixed in the spati al descrip tion. In cont inuum mechanics, however ,
it is usually pr eferabl e to avoid such confusions by usin g different not ations
for differentiation in t hese sit uat ions.
In the material description, t he time derivative is denot ed by a dot or by
dj dt while the differential operators such as gradient , divergence and curl are
denoted by Grad, Div and Curl, respectively, beginning with capi t al letters:

j= df = of (X, t ) Gr adf = \lxf(X ,t) , etc .

dt at '
In the spat ial description, t he time derivative is the usual partial derivative
aj at and the differential operators are indicated by lower-case letters , grad,
div and curl:
of of (x , t)
grad f = \l ref(x , t) , etc .
at at
The relations between these notations can easily be obtain ed . Indeed ,
let 'IjJ be a scalar field and u be a vector field. we have
. 0'IjJ
'IjJ = at + (grad 'IjJ) · v , it = ~~ + (grad u)v , (1.29)

and
Gr ad 'IjJ = p T grad 'IjJ , Gr adu = (gradu)F. (1.30)
In par ti cular , t aking the velocity v for u in the last relation we have the
following formul a for the spatial velocity gradient :

grad v = FP- l, (1. 31 )

sin ce Grad v = Gr ad x = F.
We call j the mat erial tim e derivative of I , which is the time derivative
of f following t he path of the material point. Therefore, by t he definition
(1.26), we can write the velocity v and the accelerat ion a as

v =x , a= x ,
and henc e by (1.29h ,

a = v= ~: + (gradv)v , (1.32)

t he acceleration is the material time derivat ive of the velocity.

16 1. Kinemat ics

Example 1.4.1 We have the following identities:

Gr ad J = J div(pT) ,
(1.33)
div(J- 1 p T) = o.
Proof Regarding J as a funct ion of P and employing t he definition of
gradient (A.45) , we have, for an arbit ra ry vector a ,

a - Grad J = (Grad J)[a] = J(F(X + a)) - J(F(X)) + o(lal)

= (oFJ)[F(X + a) - F(X)] = (OFJ) [Grad (X" (X + a) - X,,(X))]
= (oFJ)[Grad(GradX,,[a])] = (oFJ)[Grad(Fa)]
= J (F-T) [Grad(Fa)] = J tr(F- l Gr ad(Fa))

= J tr(grad(Fa)) = J a · div(F T) ,
which proves the first relation. In the above derivation , we have made
suc cessive use of the relations, F = Grad X" , (A.48) , (1.30) and (A.75) .
The err or terms, like t he one shown in the first line, are omit t ed for
simplicity.
To prove the second relation of (1.33) , we obtain from (A.76)

a · div (J- 1 F T ) = div(J- 1 Fa)

= Fa - gra d J- 1 + J- 1 div(Fa)
= -J- 2a · Gr ad J + J- 1a · div(F T) ,
which is equa l t o zero by th e first identity. 0

Exercise 1.4.1 Verify the following ide nti t ies similar to (1.33) :

Div(JF- T ) = 0,
(1.34)
gra d J = -J Div(F- T ).

Exercise 1.4.2 Derive the physical component s of velocity and accel-

erat ion in (use Exercise A.2.11)
1) cylindrical coordinate system: x = (r(X , t) , O(X , t) , z(X, t))
v =f e(r ) + rOe (o ) + z e(z),

a = (r - r 02) e(r ) + (rO + 2fO) e(O) + i e(z);

2) spherical coordinate system: x = (r(X , t) , O(X , t) , </J(X , t))
v = r e(r) + r Oe(O ) + r¢ sin 0 e(q,),
a = (r - r 02 - r¢2 sin 2 0) e(r ) + (rO + 2fO - r ¢2 sinO cos O) e(O)

+ (r¢sinO + 2f ¢ sinO + 2rO¢ cos 0) e(q,).

 1.5 Rela tive Deformation 17

Exercise 1.4.3 Given a motion

XK(X, t) = XO(t) + Q(t)(X - X 0), (1.35)
where Q(t) is a time-dep end ent orthogon al te nsor.
1) Show that t he velocity and the acceleration are given by
v=xo+w x(x-x o) ,
a = xo + W x (x - x o) + w x (w x (x - x o)) ,
where the angular velocity w is defined as the axi al vector of the
skew-symmetric t ensor QcF (see (A.29) , w = (QcF) ).
2) Show that , for t his motion , t he shap e (length and angle) of any
material element does not change. It is called a rigid motion.

Exercise 1.4.4 Given an accelerat ion field in the spat ial description,
(1.36)
where {ed is the natural basis of t he Cartesian coordinat e syste m.
1) Det ermine t he deform ation x = XK(X, t) of this motion, with the
following initial condit ions:
x (X, Y,O) = X, ±(X , Y, 0)= kX,
y(X, Y, 0) = Y, iJ(X, Y,0) = -kYo
2) Find t he velocity field in the mat erial and the spatial descriptions.
3) Sket ch t he path of this motion .

1.5 Relative Deformation

In practi ce, for a given motion, the reference configurat ion K is oft en chosen
as the configuration at some inst an t t = to, K = X(·, to). Bu t such a choice is
unnecessary, in general. The configurat ion K can be chosen indepe ndent ly of
any motion.
We may even choose t he (current) configurat ion Xt as the reference con-
figuration so t hat past and future deformat ions can be described relative t o it .
Such a choice is sometimes desirabl e, especially for fluids , and it is worthwhile
also to introduce it here.
We denot e the posit ion of the material point X at t ime T by ~,

(1.37)

Then , we ca n write
(1.38)
18 1. Kin em atics

where X(t) ( . , r ) = X T 0 X;- l : 13t --+ 13T is t he deformation at time r rela-

t ive t o the configuration at t ime t , or simply called the relative deformation
(Fig. 1.5) . The relative deformation gradient Ft is defined by

(1.39)

t hat is, the deform ation gradi ent at t ime r with respect to t he configurat ion
at time t. Of course, if r = t,

(1.40)

B
X
.

Bt ~ BT

X(t)(r)
X
.e

Fig. 1.5. Relative deformation

Let '" be a fixed reference configurat ion, then we can also express the
relative deformation as

(1.41)

from which we can show that

F,,(X , r) = Ft( x , r)F,,(X , t) . (1.42)

This rela tion is best exemplified in the schematic diagram Fig. 1.6.

Example 1.5.1 Consider a motion in the x- y plane relative to t he

reference configur ation r: in the Car t esian coordinate syste m,

x = X,,(X, Y, t ) = (X et , Y(t + 1)), (1.43)

or , in terms of deformation functions,

y=Y(t+1) ,
 1.5 Relative Deformat ion 19

F K( t) / ~ F« ( T)

/ F,( T) ~
x E Bt eE B T

Fig. 1.6. Relative deformation gradient

which yield
x = XC - t , y=-y-
t+i '
or
x = X;:; 1 (x , y, t) = t
( X C- , -y-) .
t+1
T herefore, from (1.41) the relative deformation is given by

e = X",(X;:;l( X, y , t) , 7) = X",(x c- t , -y_

t+1
, 7)
t 7+1
= (x c
T
- , t + 1 y).
From (1.43) and the above function, t he deformation gradi ents can be
calculate d:
F",(t) = ct e x (>9 e x + (t + 1) e y (>9 e y ,
7 + 1
F; ( 7) = C
-r -st.
e x (>9 e x +- - e y (>9 e y ,
t+1
where {e x , e y} is t he natural basis of the Cartesian coordinate system .
The relations (1.40) and (1.42) can be checked easily.
We can also obtain the path of t he mat erial point (Xo, Yo) in this
motion, by eliminating time t from t he deform at ion function,

for X o =I 0,
for Xo= o.
Not e that , for thi s motion, the reference configurat ion is the configura-
tion of the body at the inst ant t = o. (What is the imag e of a squar e in
the reference configurat ion at any inst ant in the motion?) 0

Exercise 1.5.1 Given a velocity field

= u(y)e x . v(x , t) (1.44)
1) Determine the relative deform ation e = X(t )(X , 7) of this motion.
2) Show that
Ft (7) = 1 + (7 - t)/iN, (1.45)
where", = du /dy , and N = e x (>9 e y.
20 1. Kinematics

Exercise 1.5.2 At fixed time t , the solution curve s of the velocity field
v(x , t) are called t he streamlines of the motion at time t , i.e., if x = x(s) ,
x(O) = X o is the st reamline passing through X o at time t , then

dx( s)
~ = v(x(s) , t).

Det ermine the st reamlines of the motion given in the Cartesian coordi-
nate syst em by
2
x=(Xet , Y et , Z ).

1.6 Rate of Deformation

Whereas t he deformation gradi ent measures the local deformation, the mat e-
rial time derivative of deformation gradi ent measures t he rate at which such
changes occur. Another measure for t he rate of deformation, mor e commonly
used in fluid mechani cs, is the spat ial gradient of velocity. T hey are related
by (1.31) , grad v = FF - 1 .
The material time derivative of deformation gradi ent , F, is the rate of
change of deformation relative to the reference con figurat ion. Similarly, we
can define
(1.46)

to be the rate of change of deformation relative to the cur rent configurat ion.
From (1.42) , Ft(T) = F(T)F(t)-l , by taking the derivativ e with resp ect to T,
we have

Ft(T) = F(T)F(t) -l
= (grad v(T))F(T)F(t) -l = (grad v(T))Ft( T),

and sin ce Ft(t) = 1, we conclude that

L = gradv. (1.47)

In other words, t he velocity gradient , grad v , can also be int erpret ed as the
rate of change of deformation relative to the cur rent configur at ion.
Mor eover , if we apply the pol ar decomposition to the relative deform a-
t ion gradie nt Ft (x, T),
F; = Rt Ut = 1ItRt ,
by holding x and t fixed and t aking the derivative of F t with resp ect to T,

we obtain
 1.6 Rate of Deformation 21

and hence by putting T == t , we have

(1.48)

If we denote
~(t) == llt(t) , (1.49)
we can show easily that

~T==_w. (1.50)

Therefore, the relation (1.48) is just a decomposition of the tensor L into its
symmetric and skew-symmetric parts, or from (1.47) we have

1
D == 2(grad v + grad v T ) ,
(1.51)
1 T
~ == 2(grad v - grad v ).

In view of (1.49) the symmetric part of the velocity gradient, D , is called the
rate of strain tensor or simply the stretching tensor, and the skew-symmetric
part of the velocity gradient, ~ , is called the rate of rotation tensor or simply
the spin tensor.
Since the spin tensor ~ is skew-symmetric, it can be represented as an
axi al vector w (see Sect . A.1.6) . The components of ware usually defined by
Wi == eij k ~kj, henc e it follows t hat

w == curlv. (1.52)

The vector w is usually called the vort icity vector in fluid dynamics.I
Other t ensors of interest, in t erms of the relative deformation, are
the Rivlin-Ericksen tensors by taking the time derivatives of Ct(T) ==
Ft(T) T Ft(T) ,

An(x ,t) == ct)(x ,t) == ~n

UTn
Ct(x ,T)1
r=t
' ti == 1,2,3, · · · . (1.53)

In particular, for n == 1,
. . T ' T
Al (x , t) == Ct(x, t) == Ft(x , t) + Ft(x , t) == L + L ,

which impli es that

(1.54)
Therefore, Al is just twice the strain rate tensor. More generally, the Rivlin-
Eri cksen t ensor of order n is a measure for the strain rate of higher order.
1 In t erms of the notation (A.29) , the vorticity vector w and the spin t ensor ~
are related by w == (- 2~ ) and not by w == (W), as introduced in Sect. A.1.6.
22 1. Kinem atics

Exercise 1.6.1 Show that the spin ten sor W can also b e defined as

Exercise 1.6.2 Verify that

or in t erms of the vorticit y vector w ,

. ~
av + -1 grad(v · v ) +w x v .
v = (1.55)
vt 2
Exercise 1.6.3 Consider the velocity field given by (1.44) (Ex er-
cise 1.5.1), Show that

A l = ",(N + N T ),
A 2 = 2 ",2 N T N,
(1.56)
A3 = o.
Exercise 1.6.4 Show t hat for the rigid motion given by (1.35) , the
st re tching t ensor D = 0 and the angular velocity w is t he negative axial
vector of the spin t ensor W (see (A .29) , w = - (W )).

1. 7 Change of Frame and Objective Tensors

A fram e of reference ca n be int erpreted as an observer who observes an event

in terms of position and t ime with a rul er and a clock. Different observers
may use different rul er s and clocks and come up with different results for
the sa me event. However , if t he sa me un its of measure for their rul ers and
clocks are used , t hey should obtain the same distance and the sa me time
laps e between any two event s under obs ervation, even though the values of
their observations may still be different. We shall impose these requi rements
on a change of fra me from on e t o anot her.
Let ¢ and ¢* be two frames of reference. We call

* = ¢* 0 ¢-l : £ x JR --+ E x JR
* : (x ,t) t-+ (x* ,t*)
a change of fram e from ¢ t o ¢* (Fi g. 1.7) . For convenience , we have denot ed
the map ¢* 0 ¢- l by t he symbol *.
Since we shall restrict ourselves to the use of frames of reference t hat
yield the sa me dist ance and the sa me t ime lap se between two events, a change
 1. 7 Ch ange of Frame and Objective Tenso rs 23

e x IR

(x, t)
* .
(x' ,t')

F ig . 1. 7. Change of frame

of fram e *, in general, must be of the following form :

x· = Q(t)(x - x o) + c(t) ,
(1.57)
t* = t + a ,

for some a E JR, X O E E, c(t) E E; and Q(t) E O(V) , where OW) is the
orthogonal group on V .
A change of frame * defined by (1.57) is a time-dependent rigid transfor-
mation , also referred to as a Euclidean transformation . It is the most general
form of the change of frame. The collecti on of all Euclidean tran sform ations
will be called t he Euclidean class.
A change of fram e gives rise to a linear map on V in the following way:
Let * be a change of frame from <p t o <p' defined by (1.57) . For any vector
u E V , there exist Xl , X2 E E in the frame <p, such that

Let xi and X2 be the corre sponding points in the fram e <p', and denote
u* = x ; - xi ,
then from (1.57)
u" = Q(t)(X2 - xd = Q(t)u . (1.58)
Therefore , we ca n define a linear map Q* : V -+ V taking u to u' by

Q '(u) = u * = Q(t)u.
That is,
Q' = Q(t) . (1.59)
More generally, let S n denote the n-th-order tensor space of V , i.e.,
n
S n = ®V, n = 1,2 ,3, · · ·. Then a change of fram e * gives rise to a linear map
24 1. Kinematics

on Sn, which will also be denoted by Q* ,

Q * : s; -+ s;
in t he following way: For any UI , .. . , Un E V , we define
Q *(U I 181· · · 181 u n) = QUI 181 · ·· 181 Qu n , (1.60)
and extend it linearly to the sp ace Sn' In particular , for n = 1 we have (1.59) ,
and for n = 2 we have
Q *(U I 181 U2) = QUI 181 QU2 = Q (UI 181 U2) QT.
Hence for any T E V 181 V , we have
T*= Q*(T) = QTQT . (1.61)
For M E Sn the condit ion M* = Q *(M) in components is given by
Mt1 ... i n = Q i ; l . . . Q.: M j 1 .. .i-. .
In the physical interpret ation we ca n say that U and U" , T and T* so
related, are the same vector and t he same te nsor as obser ved in two different
fram es ¢ and ¢*. Mor eover , a scalar ind ependent of observers should have
the same value in any fram e, hence we may also define Q* on Sn for n = 0,
i.e., So = IR, by
Q * : IR -+ IR Q* = 1. (1.62 )
Note that Q* : Sn -+ Sn dep ends only on the rot at ional part Q(t) of the
change of t he fram e *. Ther efore we may also refer to Q * as the induced
lin ear m ap on Sn by t he rotati on Q(t) (in the change of fram e) .
Let <P denot e t he set of all frames. Suppose f is a map
f : <P -+ s:
We call f( ¢) the value of f observed in the fram e ¢. Let E be the Euclidean
class of change of frame and tJt be a subclass of E .
Definition. An observable qu an tity f is said to be frame-indifferen t with
resp ect to tJt, if
f( ¢*) = Q* f( ¢) (1.63)
for any change of frame from ¢ t o ¢ * belon g to the class tJt .
If tJt = E , we simply say t hat f is fram e-ind ifferent or objective (with
resp ect to Euclidean transformations) . Mor e specifically, a time-d ep endent
scalar s, vect or U or t ensor T is ca lled an objective scalar, vector or t ensor
qu anti ty, resp ect ively, if, relative to any change of frame given by a Euclidean
t ransformation (1.57) ,
= s(t ),
s* (t * )
u*(t*) = Q(t)u(t) , (1.64)
T *(t*) = Q(t) T(t) Q(tf .
We say t hat objec t ive qu antiti es are invari an t under a change of obse rvers .
 1. 7 Change of Frame and Objective Tensors 25

Example 1.7.1 Let W be an objective skew-symmet ric tensor, show

that its associated axial vector (W), (see (A.29)) , is not an objective
vector.
Since W is objective, we have

W* = QWQT VQ E O(V) .
Let the associated axial vectors be

w = (W), w* = (W *) ,

and express W = ~Wijei 1\ ej . Then we have, for any U E V,

w* . Qu = (QW QT) . Qu = ~Wij (Q(ei 1\ ej)QT ). Qu

= ~ w ij (Qei 1\ Qej ) . Qu = ~ w ij Qei x Qej . Qu
= ~(detQ)Wijei x ej · u = ~(detQ)Wij (ei 1\ ej ) · u
= (det Q) (W ) . u = (det Q)w . u .

It follows t hat
w* = (detQ)Qw = ±Qw,
since Q is orthogonal , det Q = ± 1. Therefore, the associate d axial vector
w is not objective. 0

1. 7.1 Transformation Property of Motion

In mechani cs, we know that motions depend on observers and, consequently,

t he velocity and the acceleration of a motion are not obj ecti ve, in general.
We shall see how they transform under a change of observers.
Let X be a motion, and * be a change of frame from ¢ to ¢ *, then

x = X(X, t) , x* = X*(X , t*) , X E 13.

From (1.57) , we have

X*(X, t *) = Q(t)(X(X, t) - xo) + c(t) ,

(1.65)
t* = t + a.
Therefore, it follows that

oX* Q. ( )
.*
x = ot* = x - Xo + Q x. + c.
or
x. * - Qx. = n(
t X *
J - C) + c. , (1.66)
26 1. Kin em ati cs

where
D(t) = Q(t)Q(tf (1.67)
is called t he angular velocit y tenso r or spin tensor of 4>* relative to 4>. The
tensor D is skew-symmet ric, i.e.,

(1.68)

Moreover , taking again the derivative of (1.66) , we ob t ain

x * - Qx = c + 2D(x* - c) + (D - D 2)(x* - c) . (1.69)

The relations (1.66) and (1.69) show that t he velocity and t he acceler ation
ar e not object ive vectors.
Note that if the rotation of the change of fram e Q(t) is const ant in time
and c(t) is linear in time, t hen we have x* = Qx . Such a change of fram e is
called a Galilean tran sformation , which is given by

x*= Q(x - x o) + Vt + co,

(1. 70)
t* = t + a,

for any constant c.,, V, and Q. The relation (1.69) shows t hat t he acce leration
is frame indifferent with resp ect to Galil ean t ran sformations.
On t he ot her hand , the relation (1.66) shows t hat t he velocity is not
frame indifferent with resp ect t o Galile an tran sformations. However , it also
shows that the velocit y is frame indifferent with respect to t ime-independent
rigid transformation s, defined as changes of frame given by

x* = Q(x - xo) + co,

(1. 71)
t* = t + a,
where Co and Q ar e time ind ep endent.

1.7.2 Property of Some Kinematic Quantities

Let r: b e a reference configurati on, and X be a motio n. Relative to a chan ge

of frame (1.57 ), we can write

X* = ""<P' (X) , X EB. (1.72)

In order to see how a referen ce configurat ion may be affecte d by a change

of frame, we choo se the configurati on x as the configuration occupied by the
body at some inst ant to ,

x = X(X ,to) , X* = X*(X , t*)

a ,
 1. 7 Change of Frame and Objective Tensors 27

where the X and X* are related by (1.65) . It follows immediately that

X* = Q(to)(X - x o) + c(to) .

Therefore, in general, we may denote the change from 1£4> to 1£4>' by the
following transformation in a change of frame ,

(1. 73)

where K is a constant orthogonal te nsor.

On the other hand , the motion relative t o t he change of frame is given
by
x = XK(X , t) , x* = X*K (X* , t*) ,
and from (1.57) we have

X~(X* , t*) = Q(t)(XK(X , t) - x o) + c(t) .

Therefore, we obtain for the deformation gradient,

F *(X * ,t*) = Q(t)F(X ,t)KT ,

or simply
(1.74)
where K is a fixed orthogonal tensor du e to the change of frame for the
reference configurat ion.f This shows that the deformation gradient is not an
objective tensor.
Wi th polar decompositions of F and F * , (1.74) gives

R*U* = QRUKT , V *R* = QVRKT .

By t he un iqueness of such decompositions, we conclude that

(1. 75)

and hence
(1. 76)
Therefore, we have shown that V and B ar e obj ective t ensors, while R, U ,
and C are not obj ective qu antities.
If we differentiate (1.74) with respect t o time, we obtain

2 It is often assumed that the referenc e configur a tion is un affect ed by the change
of frame, in the sens e t hat there is some instant to such that Q(to) is a n identity
tensor and hence F* = QF holds . On the other hand, if we define P* = \7 x X:
(compar e with F * = \7x 'X: ) then it follows t hat P* = QF.
28 1. Kinematics

With P = LP by (1.31), we have

L*P* = QLPKT + QPKT = QLQTp* +QQTp* ,

and since P* is non-singular, it gives

(1.77)

Moreover, with L = D + W , it becomes

D* + W* = Q(D + W)QT + il,
and by separating symmetric and skew-symmetric parts, we obtain

W* = QWQT +il. (1.78)

Therefore, while the tensors Land Ware not objective, the rate of strain
tensor D is an objective quantity.
Now, let us consider an objective vector field u(x , t). For any change of
frame *,
u*(x* ,t*) = Q(t)u(x,t). (1.79)
By taking the gradient with resp ect to x, we obtain

(\7x·u*)(\7 xX*) = Q\7 xu,

and by \7xX* = Q(t) from (1.65) , this implies that
(grad u)* = Q (grad u) QT. (1.80)
Therefore, grad u is an objective t ensor field. Similarly, we can show that if f
is an obj ective tensor field of order n then grad f is an obj ective tensor field
of order n + 1.
On the other hand, if we express the relation (1.79) in the material
coordinate,
u*(X* ,to) = Q(t)u(X ,t),
then by taking the material time derivative and the gradient with respect to
X , we obtain

iJ: = Qit + o«, (Grad u)* = Q (Grad u) K T , (1.81)

where K = \7 x X* from (1.73). Therefore, even though u is an obj ective

vector, its material time derivative it is not objective, in general, and its ref-
erential gradient Grad u being a second-order tensor, is not objective either.
Similarly, if 'ljJ is an objective scalar field, we have

~* = ~, (grad 'ljJ)* = Q grad e, (Grad 'ljJ)* = K Grad e . (1.82)

In other words , ~ and grad e are objective scalar and vector fields , respec-
tively, while Crad e is not an objective vector field .
 1. 7 Change of Frame a nd Objective Tensor s 29

Example 1.7.2 For an obj ective vector field u , if we defin e

o
U=
.
u- W u , (1.83)

where W is the spin tensor of the motion , t hen ~ is objective.

Not e t hat from (1.81) and (1.78h ,

(u0)* =U.* - W *U *
= Qit + Qu - (QWQT + QQT)QU
= Q(it - Wu) =Q~ .
T his derivative is called the corotational tim e derivative, which mea-
sures the time rate of change experienced by mat eri al par ticles rotat-
ing along with t he motion . Indeed , we ca n define t he corotat ional time
derivative as

~(t) = h-t
lim -hI (u(t + h) -
O
P(t + h)u(t)) , (1.84)

where the linear transformation P : V -+ V is chosen to be t he relative

rotation of the motion , P( T) = R t (T) with resp ect to the time t. There-
fore, Rt(t + h) rotat es the vector u(t) with the motion to a vector at
time (t + h) , which can then be compared with the vector u(t + h) at
t he same inst an t . Since

and from (1.49)

we have

~(t) = lim _,I (u(t + h) - u(t) - h W(t)u (t)) ,

h-t O t

and hence (1.83) follows. D

Exercise 1.7.1 Show t hat the material time derivative is ind ep endent
of the frame, by verifying dir ectly in Ca rtes ia n component s that
8 8 8 * 8
n ox , = ~
ut + Vi-;:;- ot: + vi~'
ux i
Exercise 1.7.2 Let 7jJ and S be obj ective scalar and objective second-
ord er te nsor fields , respectively.
1) Show t hat their sp ati al gra dients, grad e and gradS, are also objec-
t ive.
2) Show t hat "j; is object ive but 5 is not .
30 1. Kinematics

Exercise 1.7.3 Show that the corotational time derivative of an ob-

jective t ensor field S defined as
1
S(t) = lim -h ts« + h) - P(t
o

h--+O
+ h)S(t)pT(t + h)) , (1.85)

with P(T) = Rt(T) , impli es that

o .
S=S-WS+SW,

and show that it is obj ective.

Exercise 1.7.4 Let u and S be obj ective vector and obj ective second-
order t ensor fields, resp ectively. Show that t he convected time derivatives
defined by
uL:o. = u-
• Lu ,
L:o. . T
S= S-LS-SL ,
are objective. Show that the convect ive time derivative can be defined
similar t o (1.84) and (1.85) with P(T) = Ft(T) , the relative deformation
gradi ent from time t to T .

Exercise 1.7.5 Show t hat

1) the relative right stretch t ensor Ut is obj ective, while the relative
rotation tensor R; is not obj ective;
2) the Ri vlin-Ericksen tensor An is objective.
2. Balance Laws

2.1 General Balance Equation

T he basic laws of mechani cs can all be expressed, in general, in the following

form,
~ r 'ljJ dv = }aPt
}Pt
r 1>1/Jnda+ i;r (J1/Jdv , (2.1)

for any bounded regular subregion of the body, called a part PcB and
the vector field n , the outward unit normal t o the boundary of the region
P t in the curre nt configurat ion. T he qu antities 'ljJ and (J1/J are t ensor fields of
certain order m , and 1>1/J is a t ensor field of order m + 1, say m = 0 or m = 1,
so t hat 'ljJ is a scalar or vector qu antity, and , resp ectively, 1>1/J is a vector or
second-order te nsor qu an tity.
The relat ion (2.1) , called the general balance of 'ljJ in inte gral form , is
interpret ed as asserting t hat t he rate of increase of the quantity 'ljJ in a part
P of a body is affected by the inflow of 'ljJ through the boundar y of P and t he
growth of 'ljJ within P . We call 1>1/J the flux of 'ljJ and (J1/J the supply of 'ljJ . In
general , t he supply (J,p may contain cont ributions of supplies from exte rnal
sources and inte rn al productions du e to t he motion of t he bod y.
It follows from t he balance equation (2.1) that ifthe body is isolat ed, i.e.,
1>1/J = 0 on aB , and is free from supplies, (J1/J = 0, then the total amount of the
qu antity 'ljJ is cons tant in time. In other word s, t he qu antity 'ljJ is conserv ed
and t he equat ion is called the conser vation law of 'ljJ .
We ar e interested in the local form s of the int egral bal an ce (2.1) at
a point in the region Bt . T he derivation of local forms rests upon certain
assumpt ions on the smoot hness of the tensor fields 'ljJ, 1>1/J ' and (J1/J . Here not
only regular point s, where all t he tensor fields are smooth, bu t also singular
points, where they may suffer jump discontinui ties, will be considered.
F irst of all, we need the following theorem , which is a three-dimensional
version of the formula,

d l!(t)
-d
t get)
'ljJ (x, t) dx =
get)
at dx + 'ljJ (J (t ), t) f(t)
l!( t) a'ljJ .
- 'ljJ (g(t ), t) g( t) ,

I.-S. Liu, Continuum Mechanics

© Springer-Verlag Berlin Heidelberg 2002
32 2. Balance Laws

in calculus for differentiation under the integral sign on a moving interval.

Note that the last two terms are the product of the values of the function '¢
and the outward velocity at the end points.

Theorem 2.1.1 (Transport Theorem). Let V(t) be a regular region in E and

un(x, t) be th e outward normal speed of a surface point x E 8V(t) . Then for
any smooth tensor field ,¢(x , t) , we have

d
dt
r'¢ dv = l,r ~'¢ dv + r '¢u n da .
Jv dt J av
(2.2)

Proof By definition,

d
dt
r'¢ dv = lim -hI { r
Jv h-tO JV(t+h)
'¢(x , t + h ) dv- r
JV(t)
'¢(x , t ) dv }

= lim -hI {
h -tO
r JV(t+h)
'¢(x , t + h) dv - r
JV(t)
'¢(x, t + h) dV}
(2.3)
+ lim -hI {r '¢(x, t + h) dv - r '¢(x, t) dV}

. 11 1
h-tO JV(t) JV(t)

= lim -h ,¢(x, t + h ) dv + 8'¢

--;:}(x,t)dv .
h -tO V(t+h)-V(t) V( t) dt

The region V(t + h) - V(t) is swept out by 8V in the time int erval (t, t + h) .
In other words, a small volume element Llv , as shown in Fig. 2.1, is equ al to

a V (t+h )

Fig. 2.1. Region with moving boundary

 2.1 General Balance Equation 33

Therefore , the first term on the right-hand side of (2.3h becomes

lim _II r
h --+O t }V(t+h) - Vet)
'I/J (x , t + h) dv = lim -hI
h--+O
r
} aV(t)
'I/J (x , t + h)un(x, t)h da

= r 'I/J (x , t + h)un(x , t) da
lim
h --+O}aV(t)

= r 'I/J (x , t )Un(x , t) da,

u;
which proves (2.2). 0
In this theorem, the surface speed un(x , t) needs only to be defined on
the boundary aV . If V(t) is a material region Pt , i.e., it always consists of
the same material points of a part P c 13, then u., = x · nand (2.2) becomes

d
dt
r 'I/J dv = I-,r ~'I/Joi dv + }apr 'I/J x .
l-, t
n da. (2.4)

Now we shall extend the above transport theorem to a material region con-
taining a surface across which 'I/J may suffer a jump discontinuity.
An oriented smooth surface S in a material region V is called a singular
surface relative to a field 'I/J defined on V, if 'I/J is smooth in V - S and suffers
a jump discontinuity across S. The jump of 'I/J is defined as

(2.5)

wher e 'I/J+ and 'I/J - are the one-side limits from the two regions of V separated
by S and designated as V+ and V - , respectively.

V-

Fig. 2.2. Singular surface

34 2. Balance Laws

Let u., be the normal speed of S with the direction pointing into V+ and
n be the outward unit normal of 8V (Fig. 2.2) . Denote

Since both (8V) + and (8V)- are material surfaces, their normal surface speed
is x.
n . Clearly,

(2.6)

Since S need not be a material surface, V + and V - need not be material

regions in general.
Suppose that 7/J (x, t) and x(x , t) are smooth in V+ and V-, then the
transport theorem (2.2) implies that

d
dt
r
l.:
ib du > r
Jv+ t
887/J dv+l
(OV)+
7/J x· n da + r 7/J+(- u n ) da,
Js
(2.7)

d
dt
r
Jv-
ib d» = r
Jv -
887/J dv
t
+ 1(av) - 7/J x · nda + r 7/J - u n da .
Js
(2.8)

Adding (2.7) and (2.8) , we obtain, by the use of (2.6) the following transport
theorem in a material region containing a singular surface:

Theorem 2.1.2. Let V(t) be a material region in E and S(t) be a singular

surface relative to th e ten sor field 7/J (x , t) that is smooth elsewhere. Then we
have
d
d t Jv
r
7/J dv =
Jv t
r
887/J dv +
Jav
r
7/J x . n da -
Js
[7/J] Un da , r
(2.9)

wh ere u n(x , t) is the normal speed of a surface point x E S(t) and [7/J] is the
jump of 7/J across S .

Example 2.1.1 Let V(t) be the volume of a part PcB in a motion,

then
V(t) = dv , r
JPt
and by taking 7/J = 1 in (2.9), it follows that

dV
dt
= r
Jap t
x ' nda = r
JPt
div z d». (2.10)

Therefore, if the motion is incompressible, i.e., the volume of any part P

in B remains constant in the motion , then the divergence of the velocity
must vanish,
div z = O.
D
 2.1 General Balance Equation 35

2.1.1 Field Equation and Jump Condition

For a mater ial region V containing a sing ular sur face S , the equat ion of
general balan ce in int egral form (2.1) becomes

r aa1/Jt dv + i;r
lv
1/J x . n da - r [1/JJ
ls
Un da = r
l ev
lfJ",n da + r
lv
(J", dv . (2.11)

A point x is called regular if there is a mat erial region containing x in

which all the t ensor fields in (2.1) ar e smooth . And a point x is called singular
if it is a point on a singular sur face relative t o 1/J and 1fJ",.
We can obt ain the local balance equat ion at a regular point as well as
at a singular point from t he above int egral equation. First, we consider a
small material region V containi ng x , such t hat V n S = 0. By the use of t he
divergence theorem , (2.11) becomes

Iv {~~ + div(1/J 0 x - 1fJ", ) - (J", } dv = O. (2.12)

Since the integrand is smoot h and (2.12) hold s for any V, such that x E V,
and vnS = 0, the integra nd must vani sh at x (see t he proposition on p. 280).
Therefore, we obtain the balan ce equ at ion at a regular point , usually called
t he field equat ion.

Field equation. At a regular p oint x , th e general balan ce eq ua tion (2 .11)

reduces to
~~ + div( 1/J 0 x -1fJ",) - (J", = O. (2.13)

The qu antity 1/J 0 x in (2.13) is called t he convec tive flu x of 1/J and t his
notat ion should be understood as 1/Jx when 1/J is a sca lar quantity.

n
s

x v

Fig. 2.3. At a singular point

Next, we consider a singular point x , i.e., XES . Let V be an arbit rary

material region around x , and s = V n S (F ig. 2.3). We shall t ake the limit
36 2. Bal ance Laws

by shrinking (OV)+ and (OV) - down to s in such a way that the volume
of V t ends to zero, while the area of s remains unchanged . If ot'¢ and a'l/J
are bounded in V then the volume integrals vanish in the limit and (2.11)
becomes
1 {[7/; (x· n - Un)] - [<p'l/J]n} da = O. (2.14)

Since the integrand is smoot h on sand (2.14) holds for any s containing x ,
the int egrand must vanish at x . We obtain

Jump condition. A t a sing ular point x , the general balance equation (2.11)
reduces to
[7/; (x · n - un)] - [<p'l/J]n = 0, (2.15)

if, in addition, ~~ and a'l/J are bounded in th e neighborhood of x.

The jump condit ion (2.15) is also known as the Rankin e-Hugoniot equa-
tion. It can also be writt en as

[7/; U] + [<p'l/J] n = 0, (2.16)

where
U± = Un -x±· n (2.17)
are called the local speeds of propagation of S relative to the motion of the
body. If S is a material surface then U± = O.

2.1.2 Balance Equations in Material Coordinates

Sometimes, for solid bodies, it is mor e convenient to use the material de-
scription. The corresponding relations for the balan ce equa t ion (2.13) and
the jump condition (2.15) in the reference coordinates can be derived in a
simil ar manner . We begin with the int egral form (2.1), now written in t he
reference configuration K,

(2.18)

In view of the relations for volume clement s and surface element s (1.9) , the
corresponding quantities are defined as

(2.19)

The tran sport theorem (2.2) rem ains valid for 7/;", (X ,t) in a movable region
V(t) ,
(2.20)
 2.1 Genera l Balance Equation 37

where U,,(X , t) is the outward normal speed of a sur face point X E 8V( t) .
Therefore, for a singular surface S,,(t) moving across a material region V" in
the reference configurat ion, by a simil ar argument , we can obtain from (2.11)

1 ~"dv" -1
v, s;
[1/J,,] u.a«; = insv, <p~n"da" + 1v, (j~dv", (2 .21 )

where U" is t he normal speed of the sur face points on S" , while the normal
speed of the surface points on 8V" is zero, since a material region is a fixed
region in the reference configurat ion. From this equat ion, we obtain the local
bal anc e equat ion and the jump cond ition in t he referen ce configuration ,

[1/Jd U" + [<p~] n " = 0, (2.22)

where U"and n " are the norm al speed and the unit normal vect or of the
singular sur face.

Example 2.1.2 Another way of obt aining the balan ce equat ion (2.22h
in t he reference configuration is t o derive from (2.13) dir ectl y. Multiply-
ing (2.13) by J , with the definition (2.19) , we obtain imm ediately t he
desired equation (2.22h by t he use of the following relations:

~" = J(~~ + div( 1/J Q9 x)), Div<p~ = J div<P,p . (2.23)

T he first relation follows from (1.29h and

j = J F - T . P = J div x
from the use of (A.48) and (1.31) . The second relation is an identity. By
t aking <P~ to be a vecto r qu antity u , we have

J div(J- 1 Fu) = ii« . div (J- 1 F T ) + t r (J - 1 F grad u))

= t r(Grad u ) = Divu,
by the successive use of (A.76h, (1.33)z and (1.30 )z. If t his result is
applied to u = ST v for any constant vector v , the identity (2.23)z is
proved for a tensor quantity <P~ = S.

Example 2.1.3 Kin em atic com patibility condition

From the deformation gradient F = Gr ad X" and the velocity x = 8 tX" ,
the existence of the deformation funct ion X" (X , t) requires the following
integrability condit ion:

P = Crad e . (2 .24)
38 2. Balance Laws

If we int egrate the above equation over an arbitrar y part P I< of t he body
in the reference configurat ion K, and use the divergence theorem (see
Exercise A.2.7), we obtain

:t 1P"
Fdvl< = 1 x0
e»;
n l< da l< ,

whi ch can be regarded as a sp ecial case of (2.18). If the region contains

a singular surface, then at a singular point, it follows from the jump
condition (2.22h that

(2.25)
This equat ion is usually called the kin ematic compatibility condit ion at
the singular sur face. It is equivalent to the following relations:

(2.26)
where a = [Fn l<~ is the normal jump of F at the singular surface. 0

Exercise 2.1.1 Let a moving singular surface S(t) in the reference

configur at ion K, be characte rized by the condition f(X , t) = O.
1) Es t ablish the following relat ions concerning t he unit normals and
the normal sp eeds in the referen ce and the spat ial configur at ions as
given in (2.22h and (2.15):

Gr adf j
n l< = IGrad f l' UI< = -IGrad fl '
(2.27)
gradf
n = "':=:---::-'-:-:-
af /at
Igradfl ' Un = -Igradfl '

2) The following relations hold :

FTn
nl< = IF T n l ' (2.28)

where U is the local sp eed in the spatial configurat ion defined by

(2.17).

2.2 Conservation of Mass

In cont inuum mechani cs, we assume that for any configurati on K, of a body
H, there exists a positive integrable fun ction

PI< : HI< -+ IR+ ,

 2.2 Conservation of Mass 39

such that for any part PcB,

(2.29)

We call M(P) the mass of t he part P , and M the mass distrib ution of B.
PI< is called t he ma ss density in the configurat ion n , We assume that the
mass distribution , and consequently the mass density, is a frame-indifferent
quantity.
By assuming t he existe nce of a mass-density function , we ar e int erested
in bodies that can be regarded as con tinuous m edia only. Discrete mass dis-
t ributions, such as concentrat ed masses, will not be considered in this book.
Since mass density depend s on the configurat ion, let "'1 and "'2 be two
configur ations, and deno te

t he change of configur at ion from "'1 t o "'2 . We then have

which implies that

Since this is valid for any part P cB, it follows that

(2.30)

Thus, the density in one configuration det ermines the density in any other
configuration .
Now, consider a motion X of a body B and let p(x, t) be the mass den-
sity in t he current configurat ion. Since the mater ial is neither destroyed nor
create d in any motion , in the absence of chemical reactions, we have

Conservation of mass. The total mass of any part P cB does not change
in any m otion,
d
dt M(P) = dt
d r
}Pt pdv = O. (2.31)

By comparison, it is a spe cial case of the general bal an ce equ ation (2.1)
with no flux and no supply,

'l/J = p, <P1/J = 0,
40 2. Balance Laws

and hence from (2.13) and (2.15) we obtain the local expressions for mass
conservat ion and the jump condition at a singular point ,

a;: + div(px ) = 0, (2.32)

The equat ion (2.31) states that the total mas s of any part is constant in
time. In particular , we have

M(P) = 1 P"
PKdvK = 1r , pdv, (2.33)

which implies that

PK(X) = p(X , t )IJ (X , t)l,
or
PK
(2.34)
P = Idet FI'
This is anot her form of the conservation of mass in t he material descrip-
tion , which also follows from the general expression (2.22) and (2.19) . This
equatio n is equivalent to
d
dt (pJ) = 0,
which , by t he use of the relation fJF J = JF- T (see (A.48)) and (1.31) , can
be written as
p + p div e = O. (2.35)
In fluid mechan ics, t his equation is also referr ed to as t he equation of conti-
nuity .

Exercise 2.2.1 Show that the mass conservat ion (2.35) implies t hat
for any t ensor qu an ti ty sb ,
. fJp'ljJ
p'ljJ = 7ft + div(p'ljJ (9 v). (2.36)

Exercise 2.2 .2 Let 'ljJ (x, t ) be an arbitrary function. Show that for any
P cB,
:t l-.r 'ljJp dv =
l-,
r ~p dv, (2.37)

provid ed t hat t he mass is cons erved . This is anot her version of the trans-
port formula (2.4) .
 2.3 Laws of Dyn amics 41

2.3 Laws of Dynamics

Let X be a motion of B. We define the lin ear momentum P of a part PcB

in the motion Xt by
P(P , t) = rp
i:
x du, (2.38)

which is a vector qu antity, and the angular m omentum H Xo of P with respect

to X o E E in t he motion Xt by

H xJP ,t) = r p(x-xo) l\xdv.

}Pt
(2.39)

The angular momentum, also called the m om ent of momentum, is a skew-

symmet ric te nsor qu antity, and henc e, more commonly, it is represented as
an ax ial vector qu an ti ty. Therefore, the op erator 1\ will be regarded eit her
as the ext erior tensor product or as t he vect or product , dep ending on t he
int erpret ation .
In pos tulat ing t he laws of dynamics, we follow the classical approac h
est abli shed by Newt on and Eul er , according to which motions are produced
by the action of forces and moments upon t he bod y. We denote

f (P , t) t otal force act ing on P cB at ti me t ,

m x o (P , t) total moment act ing on PcB with resp ect to Xo at t ime t.

T he force is a vector qu antity, while the moment is an axial vect or or a

skew-symm etric te nsor qu anti ty. The properties of forces and mom ents will
be discussed later.

Laws of dynamics.
1) Th ere exists a frame of reference 4>, called an inertial frame , such tha t
for any mo tion relative to which if f(P , t) = 0 th en F(P ,t) = 0 for any
PcB .
2) For any mo tion relative to an inertial frame,

F(P , t ) = f(P , t) , (2.40)

H XO (P , t) = m x o (P, t) . (2.41)

Equations (2.40) and (2.41) are called Euler's first and second laws, resp ec-
t ively. Note t hat from (2.37)

F(P , t) = r p iii d»,

i; (2.42)
H xo(P ,t) = r p(x- x o) l\ xdv ,
i;
provided that the mass is conserved .
42 2. Balan ce Laws

Since the velocity is not fram e indifferen t, it is easy t o see that both
P and H are not fram e indifferent and hence Euler 's laws dep end on t he
fram e of refere nce. Indeed , the first part of the laws of dynamics assures the
existence of inertial frames (an equivalent of Newton's first law) , relat ive to
which Eul er 's laws take t he above simple form s - the motion has a pr eferred
frame. In view of the results from Sect. 1.7.1 that acceleration is invari an t
with respect to Galilean t ransformation, t he class of inertial fram es can b e
characterized by t he following proposition:

Proposition 2.3.1 Let ¢ be an inertial frame and assum e that th e force f

is a frame-indifferent vector . Th en a frame ¢ * is an inertial frame if and only
if ¢ and ¢* are relate d by a Galilean transformation.
As a consequence, we have the following
Proposition 2.3.2 Th e laws of dyn amics relative to inertial fram es are
Galil ean invariant .
In other words, the dynamic laws have the sam e expressions in any fram e
relat ed to an inertial fram e by a Galilean tran sformation.

2.3.1 Forces and Moments

Forces and moments will be treated as primitive concept s of mechani cs. We

ass ume t hat the t ot al force f(P , t) acting on P c H, consists of two parts:
the body force ri» , t) , which acts within the body, and t he con tact fo rce
f C(P, t) , which acts on the boundar y of th e body,

f(P , t) = rir.t) + r(p , t ), (2.43)

and they can be represent ed in the configurat ion Ht as

ri». t) = }pr , pbdv ,

(2.44)
r(p ,t) = r tda ,
lap,
where b is called the body force dens ity (per unit mass) , and t is called the
surfa ce traction (per unit surface area). Both b and t are assumed to be
fram e-indiffer ent vector quantities.
We shall also assume t hat the total moment m ""o(P , t) act ing on P with
respec t to X o E [ can be represented by

m""JP , t) = r (x -
i; x o) /\ pbdv + r (x - x o) /\ tda.
lap,
(2.45)

In other words, only the moments due to forces ar e considered. Indeed , by

adopt ing the definitions (2.39) and (2.45) , we have essentially neglect ed the
 2.3 Laws of Dynamics 43

effect of any spin of material par ticl es, caused by polarized molecular struc-
tures, for example. Materials with microstructures are called polar materials.
We shall be concerned with non -polar materials only.
Euler's first and second laws, (2.40) and (2.41) , ca n now be written as

: { p x dv = { p b dv + { t da ,
t }Pt }Pt }aPt
(2.46)
: {p(x -x l\xdv= { (x-x o) l\pbdv+{ (x-x o) l\tda.
t t; o)
}Pt i;
2.3 .2 Stress Tensor

The body force b is regarded as an ext ernal force du e to sources outside the
body, such as the gravitational force. In saying so, we have tacitly assumed
that the interaction between different parts of the body (the internal forces) is
manifested through the contact forces alon e. Therefore, we shall assume that
t he body force density b = b(x, t) is a fun ction of position and time only. On
the other hand , at an interior point x E apt , the surface traction t is the force
exert ed on the part P by the remaining part of the body through the surface
apt at x . For two different parts with a common surface point at x , the
surface tract ions t at x on t he two different surfaces are, in general, different .
In other words , the surface traction t at x dep ends, in gener al, on the surface
apt on which x lies. The basi c assumption for sim plifying the dependence of
the traction on ap in classical cont inuum mechanics st ates that the surface
tractions on all like-orient ed sur faces wit h a common t an gent plan e at x ar e
the same (see Fig . 2.4) . This is known as Cauchy's Postulate.

P ost ula t e (Cau chy). If x E apt n ap t, and apt and apt have a com m on
orient ed norm al a t x , th en

(2.47)

F ig. 2.4 . Surfaces with common tangent plane

44 2. Balan ce Laws

In ot her words, Ca uchy' s postulate assert s that t dep ends on apt only
through the oriented unit normal n of apt at x ,

t(x , t, apd = t(x, t, n) . (2.48)

An immediate consequen ce of t his postulate is the following theorem .

Theorem 2.3.3 (C au chy 's lemma) . Suppose tha t t( " n) is a continuous

fun cti on of x, and X, b are bounded in Ht . Then Euler 's first law implies th at

t(x, - n ) = - t (x ,n) , (2.49)

for any x E Ht and any unit vector n E V .

... . ... ... .•. . ......

x . . ..
. ...

-n
Fig. 2 .5. Pillbox argume nt

Proof For any x E Ht and any unit vector n E V . We consider a small

pillbox P, C Ht of t hickness E and cent ered at x , such that one of it s flat
sur faces is normal to n (Fi g. 2.5) . Applyin g Eul er 's first law t o t he pillbox,
we obtain
r
I-,p x dv =
l-.
r
p b dv +
ev;
t( n) da. 1
If we flatten t he pillbox down to the middle sur face S then , sin ce pii: and
pb are both bounded, the two volume integrals will approac h zero, while the
sur face integral remain s finit e. Therefore we have

lim r
,~o l ap,
t(n) da = 0,

which by our const ruct ion reduces to

is (t(n) + t( -n)) da = O.
 2.3 Laws of Dynamics 45

Since the surface int egra l along the circular boundary of t he pillbox ap-
pro aches zero as E ---7 0, and the two fla t ends have exte rior normals n and
-n , resp ectively. Finally, the cont inuity of t he int egrand implies that it must
van ish identically at x ,

t(x , n) + t(x , - n ) = 0,
which proves the lemma . D
This theorem , usually known as Cauchy 's Fundamental Lemma, an
equivalent of Newton' s law of action and reaction, ensure s the existence of a
st ress tensor , as shown in the following theorem .

Theorem 2.3.4 (Cauchy's Theorem) . Under the same conditions assumed

in Cauchy 's lemma, there exists a second-order tensor field T, such that

t(x , t , n) = T(x, t) n . (2.50)

Proof From (2.48) , for fixed x E B t , t is a function defined on the set of unit
vectors in V . We can extend this function to the whole space, t( . ) : V ---7 V ,
in t he following manner ,

if v =f. 0,
(2.51)
if v = O.
To prove the existe nce of a te nsor field T sa tis fying (2.50) , it is sufficient to
show that t( . ) is a linear transformation on V . Therefore we have only to
show the following two relations:
Va E JR , v
(1) t(av) = a t (v ), E V,
(2) t(vl+vz)=t(vd+t(vz) , VV1, vz E V.
Not e that if a = 0 or v = 0, or if a > 0 and v =f. 0, then (1) follows directl y
from the definition (2.51) . Therefore, we consider the case t hat a < 0 and
v =f. O. Since a = - Ial, we have, by t he use of Cau chy's lemma (2.49)

t(av) = t(lal (-v)) = lalt(-v) = -Ialt(v) = at(v) ,

which proves (1).
To prove (2) , since if Vi and V z are linearl y dep end ent then it reduces
to t he case (1) , we assume that Vi and V z are linearly ind ep end ent and let

(2.52)

We consider a small triangular block P C Bt containing x E Bt , as shown

in Fig. 2.6, with its three faces Ai , A z , and A 3 normal to V i , v z, and V 3 ,
respectively, and the two parallel end triangles A 4 and A 5 at a distance J
46 2. Balance Laws

apart. Let the height of the triangle from the side A 3 be e and the areas of
A i be cu , i = 1,2,3. Then , from the const ruc t ion, one can easily see that

(2.53)

while the area of t he end triangle is fl£2 and the volume of the block P is
fl £28 , where A and fl are some cons tants.

Fig. 2.6. A tri angular region with chosen normals

If we apply Eul er 's first law to t he blo ck P , we can obtain

~ (rp(x -
£ Jp
b) dv - r
JA 4 UA s
t(n) da) = ~ t r
£ i=1 JA i
t(n) da.

Since the integrands are bounded, from the above estimates of areas and
volum e for P , the left-hand side is of the order of £, while the right-hand side
is independ ent of £. Therefore, we conclude that

1
lim- L 1t(x ,n)da=O,
3
<---+ 0 £ i = 1 Ai

and sin ce t is a cont inuous fun ction of x , by the mean value t heorem of
integral calculus, it follows t hat

where Xi , i = 1,2, 3 are some points on the surfaces A i , resp ectively. From
 2.3 Laws of Dynamics 47

the relation (2.53) , this gives

3

L IVil t(X i' 1::1) = O.

i =l

Now, when we shrink the block P to the point x by t aking E and 6 down to
zero, then Xi ---+ X and the above relation by t he use of the definition (2.51) ,
becomes
t(x, vi) + t (x ,V2) + t(x ,V3) = 0,
which , from (2.52) and using Cauchy's lemma, proves (2) . 0
The t ensor field T(x, t) in (2.50) is called the Cauchy stress tensor or
simply stress tensor. In the above theorem , the hypothesis of continuity of
traction field is often to o st ringent for the existe nce of a stress tensor in some
applicat ions, for example, in problems of shock waves. However, it has been
shown that the theorem remains valid under a much weaker hypothesis of
integrability, which would be satisfi ed in most applications [25, 26].
If , for some unit vector n ,

Tn = (In , (2.54)

then (J is called a principal stress and n a principal direction of T . Principal

st resses and principal directions ar e eigenvalues and eigenvectors of T . Since
T is symmet ric, there exist three mutually perpendicular principal dir ections
and three corresponding principal stresses.
The surface traction t on an oriented surface with unit normal n can be
decomposed into the sum of a normal traction,

(t · n)n = (n · Tn)n = (n ® n)Tn ,

and a sh ear traction,

t - (t · n)n = (1 - n ® n)Tn .
Let T i j be t he components of T relative to a Cartesian coordinate system ,
then
T i j = the component of the traction t in the dir ection of
i-th coordinat e line on the j-th coordinat e sur face.
Hence, we call r.:
T 22 , T 33 the normal stress components and T 12 , Tn , T 3 1
the shear stress component s on the coordinat e surfaces.

E x ample 2. 3 .1 We consider some simple st ates of stress:

1) Hydrostatic pressure
The st ress is given by T = - p l. In this state of stress , every direction
is a principal dir ection, and - p is the principal stress. p is called the
hydrostatic pressure. Therefor e, the traction on any surface with normal
n is the normal traction - p n .
48 2. Balance Laws

2) Pure tension or compression

T he stress given by T = 0" (e 181 e) is called a state of pure tension, or un i-
axial tension , in the directi on of t he unit vector e . The traction, t = 0" e ,
on the surface wit h norm al e is a normal traction. It is called a tensile
st ress if 0" is positive, and a compressive stress if 0" is negative.
3) Pure shear
The stress given by T = T(e 181 d + d 181 e) is called a state of pure shear
in the plane spanned by two vectors (e, d). Let (e*, d*) be t he pai r of
their du al vectors. Then the tractions on the surface planes parallel to e
and d , respecti vely are given by (Fig. 2.7)

They are purely shear tractions and, in addit ion, t hey are of t he same
magnitude . Indeed , one can easily ver ify t hat t he angle between t he
vectors e and e * is equal to t hat between d an d d" , and

e · e* d · d*
cos B(e, e") = lell e *I' cosB( d, d*) = Idlld*I'

which impli es t hat lel le* 1 = Idlld*l. Hence, it follows that Itll = It21.
4) Plane stress
If, relative to an orthogonal bas is {e . , e2 , e 3}, the stress components
T 13 , T23 , and T33 vanish, it is called a state of plane stress in the plane
spanned by el and e2 . Clearly pure shear and pure tension are plane
stresses. 0

0"2

tttttttt
---
-- ---
---
0"1

-- ---
---
---
0"1

- tttttttt
0"2
Fig. 2 .7. Pure shear Fig . 2.8. Biaxial tension

Example 2.3 .2 Maximum shear stress in biaxial tension

Consider a state of st ress T = 0"1 e l 181 el + 0"2e2 181 e2 and let n =
cos Bel + sin Be2 for 0 :::; B :::; 7r be the normal of a plane (Fig. 2.8) . T hen,
 2.3 Laws of Dynamics 49

the surface t ract ion on the plane is given by

which can be decomposed into normal and shear tractions. The normal
st ress,
t« = t . n = 0'1 cos 8 + 0' 2 sin 8,
2 2

by the use of trigonometrical identities, ca n be writt en as

0'1 + 0'2 0'1 - 0' 2 28

t; = 2 + 2 cos .

Therefore, since the valu e of cosine lies between 1 and -1 , we conclude

t hat the valu e of the norm al st ress lies between t he two principal st resses

°
0'1 and 0'2 , and the max imum and minimum normal stress acts in t he
principal dir ection (8 = and 8 = 7r/2).
On the other hand , the shear st ress t., satisfies t he relation

t; = O'i cos2 8 + O'~ sin 2 8 - (0'1 cos 2 8 + 0' 2 sin 2 8)2.

Note that , if 0'1 = 0' 2 , then t, = °

for any 8. That is, there is no shear
st ress at all. Therefore , we assume t hat 0' 1 i- 0'2 , in order to find the plan e
on which t he shear stress is maximum. The condit ion for an extreme
valu e for Itsl is dt ; /d8 = 0, which aft er a little rearr an gement, gives

2(0' 2 - O'd sin 8 cos 8( 0'2 + 0'1 - 2(0'1 cos 2 8 + 0' 2 sin 2 8)) = 0,
or, by the use of t rigonometrical ident it ies, it becomes

(0'2 - 0'1)2 sin 28 cos 28 = 0.

Since 0'1 i- 0' 2, t here are two solutions to t his condit ion. The first one,
°
sin 28 = 0, gives 8 = or 8 = 7r / 2, which leads to t s = 0, t he minimum
shear st ress. The second one, cos 28 = 0, gives 8 = 7r/ 4 or 8 = 37r/ 4,
which lead s to

Therefore, we conclude that the maximum shear st ress is equa l t o one

half of th e differen ce between t he two principal stresses and act s on t he
plan e t hat bisect s the angle between t he two principal dir ections. 0

Exercise 2.3.1 Consider a rect angular block with its sides parallel
to t he coordinate axes in the reference configuration . Suppose that the
st ress is given by T = -pi + p.B, Det ermine t he normal and shear
component s of the t ractions on t he surface of the blo ck in t he deformed
configuration, und er th e simple shear deformation given by (1.14) .
50 2. Balance Laws

2.3.3 Conservation of Linear and Angular Momenta

By Cauchy's theorem, Euler's first law (2.46h becomes

:t i:r p ic dn =
i;
r pbdv + }aP,
r Tnda . (2.55)

This equation is often referred to as Cauchy 's first law. Comparison with the
general balance equat ion (2.1) leads to

'ljJ = p x, <fJ", = T, fJ", = pb,

in this case, and hence from (2.13) and (2.15) we obtain the balance equat ion
of linear momentum and its jump condition,

%t (p x) + div(p x ® x - T) - P b = 0,
(2.56)
[px(x · n - un)] - [T~n = O.
The first equat ion, known as the conservation of linear momentum, also
known as the equation of motion, can be rewri tten in the following more
familiar form by the use of (2.32) ,

px - divT = pb. (2.57)

Similarly, Eul er 's second law (2.46h becom es

:t t;r p(x-x o) l\xdv = t-.

r (x-x o) l\pbdv+ }aP,
r (x-x o) I\Tnda. (2.58)

This equati on is referred to as Cauchy 's second law, also known as the conser-
vation of angular momentum. Compar ison with the general balance equation
(2.1) leads to

'ljJ = (x - x o) 1\ p x; <fJ", = (x - x o) 1\ T , fJ", = (x - x o) 1\ pb .

We have the following identity,

Indeed, in component s, the left-hand side gives

where p = x - Xo and therefore p i ,k = 8\ . Hence, from (2.13) and the above

 2.4 Conservation of Energy 51

identity, aft er simplifi cation by the use of (2.57), we obtain the following local
form of Cauchy's second law,
(2.59)
In other words , the sym me try of the st ress te nsor for non-polar materials
ensures the conservat ion of angular momentum.
On the other hand , from (2.15) , we also obtain the jump condition of
angular momentum, which reads
[(x - x o ) 1\ px(x · n - un)] - [( x - X o ) 1\ T ]n = O.
Since [x - x o ] = 0, the above condit ion is merely a consequence of the jump
condit ion (2.56)z of linear mom entum.

Exercise 2.3.2 Prove B erno ulli 's Th eorem: Consider a flow with hy-
drostatic pressure T = -pi and conservative body force b = - grad ¢>.
1) If the flow is st eady, i.e., OtV = 0, t hen
v2 1
v · grad ( - + ¢» + -v · gradp = O. (2.60)
2 P
2) If t he flow is st ead y and irrotational, i.e., curl v = 0, then
v2 1
grad ( 2" + ¢» + Pgrad p = o. (2.61)

(Use the relation (1.55) .)

2.4 Conservation of Energy

For material bodies, besides the mechani cal properties associat ed with the
motion , we are also interested in properties associated with heat and te m-
perature. In t his sect ion we sha ll consider the balan ce of energy. Other ther-
modyn ami c propertie s will be discussed later in Ch ap . 5.
We define the kinetic energy K of a part P in the motion Xt by

K(P , t) =~
i;r px · x dv , (2.62)

and t he m echanical power P by

P(P ,t) = r
}aP,
x · t da + rp
i:
x r b d», (2.63)

For material bodies, t he kineti c energy is only a par t of the to tal energy.
We call the other par t the internal energy E , and assume that it can be
represented as
E (P, t) =
t;r pe dv , (2.64)

where e (x , t) is called t he specific intern al energy density.

52 2. Bal ance Laws

We denote the energy supply (or heat supply) to the part P by Q , and
assume that it can be given by

Q(P ,t) = r
l ap,
hda+ r
i:
prdv , (2.65)

where h = h(x , t , aPt) is called the contact heat supply and r = r (x, t) is
called the energy supply density (or heat supply) du e to ext ernal sources,
such as radiation.
Both int ernal energy and energy supply are assumed to be frame-
indifferent scal ar qu antities, and cons equently, 10 , h, and r are all frame-
indifferent scalars. Obviously, kineti c energy is not fram e ind ifferent .
The conservation of energy, also known as the first law of thermodynam-
ics, can be state d as follows:

Conservation of energy. R elative to an inertial fram e,

E(P, t) + K(P , t) = P(P, t) + Q(P , t) , (2.66)

for all PcB . That is, the rate of change of th e total energy of the body is
due partly to the mechanical power and partly to the energy supply added
to the body.

The counterpart of Cauchy's postulate and Cauchy's theorem for cont act
heat supply is the Fourier-Stokes heat flu x prin ciple, which can be stated as
the following relation,

h(x , t , aPt) = h(x, t, n) = - q(x , t) . n, (2.67)

where q(x , t) is called the heat flux vector (or energy flu x), and n is t he
ext erior unit normal of aPt at x. The above st atement can be proved in a
similar manner .
With (2.67) and the definitions (2.62) through (2.65) , t he equat ion of
energy balan ce (2.66) becom es

~
dt
r (p c+~x ,x)dv= lap,
l-, 2
r (x .Tn-q .n)da+ }p,r (px .b+pr)dv . (2.68)
On comparison with the general balance equation (2.1), we have

7jJ = (p 10 + ~ X . x), P1f; =Tx-q, a1f;=p(x ·b+ r) ,

and hence we have the following local balance equation of total energy and
its jump condit ion,

ut
~ (pc + ~2 x . x)+div((pc + ~x
2
· x)x + q - TX) - p(r + x · b) = 0,
(2.69)
[(p c + ~ x · x)(x· n - un)] + [q - Tx~ · n = O.
 2.4 Conservation of Energy 53

On the other hand , if we t ake the inner product of the velocity x and
the equation of motion (2.57) , we obtain

p!i (~ x . x) = div(Tx) + p x . b - T . grad X. (2.70)

dt 2
This equat ion can be regarded as the balance of kin eti c energy. Not e that the
first two terms on the right-hand side ar e mechanical powers as defined in
(2.63) , t he last term (T · grad x), which represent s the work don e du e to the
int ernal stress, will be called the stress power.
The energy equat ion (2.69h t akes a much simpler form after b eing sub-
tract ed from (2.70),

8~s + div(psx + q) - T . grad x - p r = 0, (2.71)

or
p i. + divq =T · grad x + pro (2.72)
This equat ion is also referred to as t he balance of internal energy.
Compari son of (2.71) with (2.13) shows that one can get the balan ce
equa t ion for the internal energy in integral form (2.1) with

'ljJ =p s , P ,p = -q , f>,p = T · grad x + pro (2.73)

There are two terms in the supply of inte rnal energy, namely, the exte rnal
energy supply pr and the stress power (T . grad x), which can be regarded as
the interna l production of energy. The pr esen ce of the int ernal production du e
to the motion of the body shows that the int ern al energy is not a conserved
qu antity in general. By the sam e token, the kin et ic energy is not a conserved
quantity either .
Furth ermo re, we have to point out that the jump condition (2.15) can
not be applied to the balan ce of int ernal energy, becaus e the supply term
contains t he velocity gradient , which will become unbounded at the singular
surface where the velocity suffers a jump discontinuity. Therefor e, there is no
jump condit ion for the int ernal energy alone.
Another important quantity, associated wit h the energy and t he heat
conduction, is the temperature, alt hough it does not even appear in the energy
equa t ion expli citl y. The te mpe rature B(x , t) is a measure of how warm a body
is. Like the mass measure of t he body, the te mperat ure will be treat ed as a
primitive fram e-indep end ent scalar qu antity and is assumed to be positive
for convenience. 1

1 In classical thermost atics , the (a bsolute ) t emperature is oft en regarded as a de-

rived concept based on certain hypothesis on ph ysical ex pe riences, instead of the
feelin g of hot a nd cold .
54 2. Balance Laws

2.5 Summary of Basic Equations

We can summarize the balance equat ions of the previous sections in Table 2.1,
from whi ch we can write down immediately the corresponding field equat ions
and jump condit ions by the use of (2.13) and (2.15) .

Balance of m ass

c::: + div(p x) = 0,

[p(x . n - unH = o. (2.74)

Balance of linear momentum

ap x diIvpx
8i+ . T ) =p b,
( . 0x-

[px(x · n - unH - [T]n = O. (2.75)

Balance of angular momentum

T=TT ,
[(x - x o) 1\ (px(x· n - un) - T n )] = O. (2.76)

Balan ce of energy
o (p e + "2p x. . x. ) + diIV (( Pe + "2p x. . x. ) x. + q -
{)t T x. ) = pr + p x. · b,
1
[p(r:: + "2 x · x)(x · n - un)] + [q - Tx] . n = O. (2.77)

Balance of int ernal energy

Bpe
{)t + diIV (pex. + q ) - T . grad x. = pr. (2.78)

The balance of angular momentum reduces to the symmetry of the st ress

tensor (2.76) by the use of (2.75h , while it s jump condit ion is obviously a
cons equ ence of (2.75)z. Mor eover , sinc e the balance of kineti c energy is a
consequence of t he balance of linear momentum, its field equat ion has not
been given above. The balance of internal energy, of cour se, follows from the
balanc e of energy and linear momentum. Since it is simpler , the balan ce of
internal energy is oft en used in plac e of the balanc e of total energy. Mor eover ,
we have already pointed out t hat the jump condit ion (2.15) can not be applied
to the balan ce of int ernal energy nor to the bal an ce of kinetic energy du e to
the presence of t he int ernal production (T ·grad x) . We also rem ark that since
 2.5 Summary of Basi c Equations 55

Table 2 .1. Qu antiti es for balan ce equat ions

1/J 1>", 17",

Mass P 0 0

L. Momentum px T pb

A. Momentum (x - x o) 1\ px (x-x o) I\T (x - x o) 1\ pb

E nergy pc+ ~x·x -q+Tx pr+px ·b

Int. E nergy pc -q pr+ T · grad x

Kin. E nergy ~x ·x Tx px . b - T . grad x

both the balance equ ations of internal energy and kine tic energy contain t he
int ernal production (T . grad x) , int ernal energy and kineti c energy are not
conse rvat ive quantities. On the other hand, mass , linear momentum, angular
mom entum, and total energy are all conservative qu antiti es. Their supplies
are due to ext ern al sources only.
By t he use of material t ime derivative (1.29) , the field equa t ions can also
be written (see (2.36)) as follows:
p+ p divv = 0,
pi; - div T = p b, (2.79)
p€ + div q - T . grad v = p r,

and t he jump condit ions in te rms of local speed U of the singular sur face
become
[pU] = 0,
[pUv] + [T]n = 0, (2.80)
1
[p U(c + 2V2)] + [Tv - q]n = 0,
where v 2 stands for v . v .

Exercise 2.5.1 Suppose that the stress is in a hydrost atic state , T =

-pi , and q = 0 (perfect fluid) . Show that t he following jump conditions
hold ,
[p + p U 2 ] = 0,
[v-(v ·n)n]=O,
(2.81 )
p 1 2
[c + - + - U ] = O.
p 2
56 2. Balance Laws

Exercise 2.5.2 Suppose that the tangential component of the velocity is

cont inuous at the singular surface. Show that the energy jump condition
can be written as
11
[q] . n = m [c- - - (n · Tn) + - U 2 ] , (2.82)
P 2
where m = (pU)- is the mass flux at the surface.

2.5.1 Basic Equations in Material Coordinates

It is sometimes more convenient to rewrite the basic equat ions in material

description relative to a refer ence configuration K. They can easily be obtained
from (2.22h ,
P = IJI -1pK'
P,Jx = Div T K + PKb, (2.83)
PKE + Div qK = T K . P + PKr,
where the following definitions have been introduced according to (2.19):

(2.84)

T K is called the Piola-Kirchhoff stress tensor and qK is called the mat erial
heat flu x. Note that unlike the Cauchy stress t ensor T , the Piola-Kirchhoff
stress tensor T K is not symmetric and it must satisfy

(2.85)

The definition has been introduced according to the relation (1.9) , which
gives the relat ion,
r
is
Tnda = r
iS K
TKnKda K. (2.86)

In other word s, Tn is the surface traction per unit ar ea in t he cur rent con-
figur ation, while TKn K is the surface traction measured per uni t area in the
reference configuration. Similarly, q -ti and qK·nK are the contact heat supplies
per unit area in the cur rent and the reference configurations resp ectively.
We can also write the jump conditions in the reference configuration by
the use of (2.22)z,

[PK] UK = 0,
[PKx] UK + [TK] n.; = 0, (2.87)
[PK(c- + ~ x· x)] UK + [TZ'x - qK] . n K = 0,

where n K and UK are the unit normal and the normal sp eed at the singular
surface in the reference configuration.
 2.5 Summary of Basic Equations 57

Exercise 2.5 .3 By the use of the kinem atic compat ibility condition
(2.26) , show that the jump condit ions (2.87) can be written as

[PI<] = 0,
[Tl<nl<] = pI<U~[Fnl<] ' (2.88)
[ql<] . n l< = UI< ([Pl<c] - (Tl< nl<) . [Fnl<]) '

where we have introduced the not ation (A) = ~(A + + A-) for the mean
value of A over the singular surface.

2.5.2 Boundary Conditions of a Material Body

The boundar y of a mat erial body is a material sur face . Therefore, at the
boundar y U± = 0, the jump conditi ons (2.80h,3 of linear momentum and
energy become
[T]n = 0,
[q] . n = [v . Tn].
Suppose that the body is acted on the boundar y by a force f per unit area,
then it requires the st ress at t he boundar y to satisfy t he following condit ion,

Tn=f , (2.89)

and the heat flux to sa t isfy

[q] . n = [v] . f .

In particular , if eit her the boundar y is fixed (v = 0) or free (f = 0) t hen the

normal component of heat flux must be conti nuous at t he boundar y,

[q]. n = O. (2.90)

Therefore, if the body has a fixed adi ab ati c (meaning t hermally isolated)
boundar y then the normal component of t he heat flux must vani sh at the
boundary.
Since the deformed configurat ion is usu ally unknown for t raction bound-
ary value problems of a solid body, t he above condit ions, which involved the
unknown boundar y, are sometimes inconvenient.
Of course, we can also express the boundar y condit ions in the reference
configurat ion. From the jump condit ions (2.87h,3, we have

(2.91)
58 2. Balance Laws

and in a fixed adi ab atic boundary,

(2.92)

where t; is the force per unit ar ea acting on as".

The relations (2.91) and (2.92) ar e the usu al boundary condit ions in
elasticity and heat conduction.

2.6 Field Equations in Arbitrary Frames

In the previous sections, we have laid down the basi c laws relative to an
inertial frame . They can easily be extended to an arbitrary frame of reference
from the tran sformation properties under a change of frame considered in
Sect. 1.7.
The field equa t ions relative to the inertial frame 1> can be expressed in
the following form (see (2.79)) :

p+ P div z = 0,
px - divT = pb, (2.93)
pi + divq - T · grad x = p ro
Let 1> be an inertial frame and 1>* be an arbitrary fram e. The change of
frame from ¢ to ¢*, (x , t) M (x* , t *) is given by a Euclidean transformation,

x * = Q(t)(x - x o) + c(t ),
(2.94)
t* = t + a.
We have required that the density p, the forces t (therefore the stress tensor
T) and b, the internal energy 6 , and its supplies h (therefore the heat flux q)
and r are all frame-indifferent quantities, i.e .,

p* = p, 6* = 6, r" = r,
q*=Qq , b*=Qb,
T* = QTQT .
We have also shown that the velocity and the acceler ation are not frame
indifferent with respect to the Euclidean transformations. They satisfy the
following transformation relations, (see (1.66) and (1.69))

x* - Q x = c + D(x* - c) ,
(2.95)
x* - Q x = c + 2D(x* - c) + (D - D 2)(x * - c),
 2.6 F ield Equations in Arbitrary Frames 59

where f2 = {JQT is the spin t ensor of the arbit rary frame ¢ * relative to the
inertial frame ¢, and for the velocity gradient (see (1.77))

(grad e )" = Q(gradx)QT + f2,

which implies t hat t he divergenc e of velocity is an obj ect ive scal ar quantity
and so is the st ress power , because the stress tensor is symmetric,

(div z )" = div re, (T · grad c}" =T · grad re.

Moreover , we ca n also show that t he spatial grad ient of an objective field is

obj ective, in particul ar (see (1.80)) ,

(divq) * = divq , (divT)* = QdivT,

and the mat erial time derivative of an obj ective scal ar field is also object ive
(not true for vector and tensor fields , see (1.81)), therefore,

(p)* = p, (e)" = e.
From the above t ransformat ion properties, it follows immediately that
the balance of mass and internal energy are object ive scal ar equations so t hat
they are Euclidean-invariant, namely, they are valid in arbit rary frames. The
balance of linear mom entum, on the other hand, is on ly Galilean invariant
because it contains the accelera ti on, whi ch is frame indifferent with respect to
Galilean t ransformat ions onl y. In an arbit rary frame ¢ *, it t akes the following
form ,
p*(x* - i *) - (divT)* = p*b*, (2.96)
where
i* = c + 2f2(x* - c) + (s2 - f22)(X * - c), (2.97)
stands for t he right-hand side of (2.95) 2. Equation (2.96) is obtained from
(2.93h by multiplyin g by Q and applying the above transformation proper-
t ies.
Just as t he acceleration, t he vect or field i * is also not frame indiffer ent
with respect to Euclidean tran sformations. In fact , i * = 0, if ¢* is an iner ti al
frame (f2 = 0 and c = 0) . Nevert heless, we can show that the bal ance of
linear momentum in the form (2.96) is Euclidean invariant . Indeed , we have
t he following proposition .

Proposition 2.6.1 The quantity (x * - i*) is an obj ective vector field, i.e.,
it is invariant under any E uclidean transforma tion.
Proof Let ¢ ' be an arbit rary frame and change of frame from ¢* to ¢ ',
(x* , t*) H (x' ,t') be given by a Euclidean transformation,

x' = Q(t*)(x* - x~) + c(t* ),

(2 .98)
t' = t* + Ii.
60 2. Balance Laws

combining (2.94) and (2.98), we obtain the change of frame from the inertial
frame ¢ to the frame ¢' ,

X' = Q'(t)(X - x o) + c'(t),

t' = t + a',
where
a' = a+u,
Q'(t) = Q(t + a)Q(t),
c'(t) = Q(t + a)(c(t) - x~) + c(t + a).
Therefore, in the refer ence frame ¢ , we have

X' - Q' X = c' + 2S?I(X' - c') + (!i' - S?12)(X ' - c') ,

where S?' = QIQIT . Denoting the right-hand side by ii, we obtain

x' - i' = Q' X = Q (Q x) .

On the other hand, from (2.95h and (2.97) , we have

x* - i* = Qx.
Therefore, it follows that

X··1 -1.· 1 = Q' (X"* -1..*) ,

which proves t hat the vector field (x - i) is objective under any Euclidean
transformation (2.98) . 0
Dropping the subscript * for simplicity, we can rewrite the field equat ions
in an arbit rary fram e.

Field equations. Relative to an arbitrary frame ¢, the balance of mass ,

linear momentum, and energy take the following forms :

p+ P div z = 0,
px - divT = p(b + i),
pi + divq - T · grad x = pro
(2.99)
or

~ + div(px) = 0,

:t (px) + div(px 18) x - T) = p(b + i) ,

ata(p e+"2P x. , x.) + diIV (( pe+"2p.x . x. ) x+

. q- T x. ) =pr-px. ' (b +1..) ,
 2.6 Fi eld Equations in Arbitrary Frames 61

where

i = c + 2D(x - c) + (D - D 2 )(x - c) , (2.100)

and D is t he spin t ensor and c is the tran slation of the fram e 4> relative to
an inerti al frame. Moreover , t hese equations are Euclidean invari ant.
We ca n int erpret the vector field i as a force field du e to the motion
of t he fram e relative to an inert ial frame and call it the in ertial body fo rce,
which consists of a number of contributions, usu ally known as

2D(x - c) Corio lis force,

-D (x - c)
2
cent rifuga l force,
D(x - c) Eul er force,
inertial force of relative translation.

By comparison of (2.98) and (2.93), we have seen t hat t he field equat ions
relative to an arbit rary fram e have exactl y the same form s as that relative
t o an inertial frame, except t he body force b is replaced by (b + i) , which we
shall call the apparent body fo rce T herefore, from now on , we sha ll simply
writ e b for (b + i) in any fram e of reference, and keep in mind that when the
inertial force i do es not vanish it must be regarded as an (apparent) part of
t he body force b.
Finally, we rem ark that the corresponding int egral laws of balan ce also
have the sam e forms in an inertial fram e as in an arbitrary frame by replac-
ing t he body force with the appare nt body force. Since t he inertial force is
bounded , provided t hat t he velocity is bounded , therefore, t he jump condi-
t ions for mass, linear mom entum and energy of pr evious sections are valid in
a ny frame o f reference .
3. Basic Principles of Constitutive Theories

3.1 Constitutive Relation

T he balanc e laws int roduced in Ch ap . 2 are the fund am ent al equ ations com-
mon t o all mat erial bodies. However , these laws are insufficient to fully char-
acterize t he behavior of mat erial bodies, becau se physical experiences have
shown that two bodies of exa ct ly the same size an d sha pe , in general, will not
have the sa me behavior when they are subject ed to exactly t he same envi-
ronment (ext ernal supplies and boundar y conditions) . For example, the sam e
force will not produce the sa me elongation on wires of rubber and aluminum
of the same diamet er and length.
Physically, the behavior of a body is characterized by a description of
the fields of
p(X, t ) the density,
X(X, t) the motion , (3.1)
()(X , t) the t emperat ure ,
called t he basic fields. The field equat ions are based on the system of equations
consist ing of balan ce of mass, mom entum and energy, and t hey alone are not
sufficient t o determine t hese fields , becau se t he system (2.79) contains other
field qu antities, namely, the st ress t ensor T (X, t) , the heat flux q(X, t) , and
t he int ernal energy c(X , t) , as well as external supplies: t he body force b(X, t)
and t he energy supply r(X, t) . These quantities must be specified in order
t ha t (2.79) becomes a system of field equat ions for the basic fields (3.1).
Exte rnal supplies will be regarded as known funct ions , det ermined from
the env ironment that the body encounters. T herefore, it is usu ally assumed
that material prop erties are independent of exte rnal supplies. While t he stress
t enso r, the heat flux, and the int ernal energy,

{T(X , t ), q(X, t ), c(X, tn, (3.2)

will dep end not only on t he behavior of t he body but also on t he kind of
material that const it utes the body. Such qu antiti es are called constitutive
qu antities , for which additional hypotheses must be introduced to char act er-
ize t hermomechanical responses of a particular material body.
The mat erial response of a body generally dep end s on the past history
of its th ermomechan ical behavior. Therefore, let us introduce the notion of

I.-S. Liu, Continuum Mechanics

© Springer-Verlag Berlin Heidelberg 2002
64 3. Bas ic Principles of Constitutive Theories

t he past history of a func tion . Let 'l/J (.) be a funct ion of time. The history of
'l/J up to tim e t is defined by

(3.3)

where s E [0, 00) denotes the time- coordinate pointed into the past from the
pr esent time t. Clearly s = 0 corresponds to the pr esent time, therefore

Mathem atical descriptions of material response are called constitutive rela-

tions or const it ut ive equat ions. We pos tulate that the history of the beh avior
up to the prese nt time det ermines the pr esent response of the body.

Principle of determinism. Let C denot e a constit utive quantity, then the

constitutive relation for C is given by a functi onal of th e form,

C(X, t) = F (pt(y, s), Xt(y, s), et(y, s), X , t ). (3.4)

Y ES
O:$ s < cx:>

We call F the constitutive functi on or response fu nction of C. Not e t hat

we have indi cated the domains of the argume nt functions as underscripts in
the notati on of t he fun ctional F . Such a fun ctional allows the description of
arbit rary non-local effects of any inhomogeneous body with a p erfect memory
of the past history.
Constitutive relations can be regarded as mathem at ical mod els of ma-
t erial bodies. The validity of a model can be verified by experime nts on t he
results it predicts. On the cont rary, some expe riments may suggest certain
functional dep end ence of the const it ut ive functi on on its vari ables to within
a reason abl e satisfact ion for certain mat erials. However , expe riments alone
are rar ely, if ever, sufficient t o det ermine constitutive fun ctions of a mate-
rial body. On t he other hand , t here are some uni versal requirements that a
model should obey lest it s consequences be contradictory t o some well-known
physical exper iences. Therefore, in search of a correct formul ation of a math-
emat ical model, in general, we shall first impose these requirements on the
proposed model. T he most important universal requirem ent s of t his kind are :
- principle of m aterial objectivity,
- material symmetry,
- th ermodyn amic considerations.
These require me nts impose severe restrictions on the model and henc e lead
to a great simplification for general constitutive relations. T he reduction of
consti tuti ve relations from very general to more specific and mathem atically
simpler ones for a given class of mater ials is the main obj ective of constitutiv e
theories in continuum mechanics.
 3.2 Principle of Material Objectivity 65

In t his chapter we shall restrict our at t ent ion to the discussion of the
principle of mat erial object ivity, which deals with the transformation proper-
ti es of const it ut ive fun ctions, and the material symmetry, which characte rizes
t he specific symmetric properties of material particles. Thermodyn amic con-
siderations , which govern t hermody namic aspec ts of material bodies will be
postponed until Chap. 5.

3.2 Principle of Material Objectivity

In writing (3.4) , we have t aken for granted t hat the const it ut ive qu an tity
C is an observabl e qu an tity and the constitutive relation has been expressed
relative to a certain fram e of reference. Now we shall examine the consequence
of a change of fram e upon the constitutive functions .
Let ¢ be a frame of reference and C(X, t ; ¢) be the value of t he const it u-
tive qu antity C at the material point X and t ime t in t he fram e ¢ . We rewrite
t he const it utive relat ion (3.4) in the followin g form ,

C(X , t ; ¢) = :Fq, (/(Y, s), Xt(y, s), et(y, s), X , t) . (3.5)

YEB
O ~ s < oo

In other words, the const it utive function dep end s on t he choice of frame , in
general, so t hat we have also indic at ed the frame ¢ on :F as a subscript.
Since any int rin sic property of materials should be ind ep endent of ob-
servers, we postulate that for any obj ective constit utive quantity, its con-
st it utive fun ction must be invariant with resp ect to any change of frame.
Mathem atically, we pos tulate

Principle of material objectivity. The response fun ction of an objective

(frame indifferent with respect to Euclidean transformation s) constit utive
quantity C mus t be ind epend en t of th e frame, i.e.,

for any frames of reference ¢ and ¢* .

More specifically, suppose that C is an Sn- valu ed obj ective const itutive
qu an tity (n = 0, 1,2 for scalar- , vector-, and te nsor-valued, resp ectively) then ,
for any change of frame from ¢ to ¢*, we have the obj ectivity proper ty (see
(1.63) )
C(¢*) = Q *C(¢) , (3.6)
66 3. Basic Principles of Constitutive Theories

where Q* is the induced linear transformation on the tensor space Sn from

the change of frame ,

x* = Q(t)(x - x o ) + c(t),
(3.7)
t* = t + a.
The principle of material objectivity imposed the following conditions:

F (p*(Y, t* - s), X*(Y, t" - s), B*(Y, t* - s) , X , to)

Y ES
O ~ s < <Xl
(3.8)
= Q* F (p(Y, t - s), X(Y, t - s) , B(Y, t - s) , X , t) ,
Y ES
0 ::;8 < 00

where we have used F = F</> = F </>* . Since both density and temperature
fields are objective scalars, we have

p*(Y, t* - s) = p(Y, t - s),

B*(Y,t* - s) = B(Y,t - s) ,

and the above condition of material objectivity can be written as

F (p(Y, t - s), X*(Y, t* - s), B(Y, t - s) , X , to)

Y ES
0 ::; 8< 00
(3.9)
= Q* F (p(Y, t - s) , X(Y , t - s) , B(Y, t - s), X , t) ,
Y ES
o::; s<oo

where
X*(Y, t" - s) = Q(t - s)(X(Y, t - s) - Xo) + c(t - s) .
This equation is the restriction imposed on any constitutive function F by
the requirement of material objectivity. Note that it can be regarded as a
condition expressed in terms of any change of frame * relative to an arbitrary
frame of reference and hence it is not necessary to mention the frame </>
explicitly.
The principle of material objectivity is often referred to as the princi-
ple of material frame indifference. We emphasize that for a non-objective
quantity, for example, the total energy density (e + v 2 / 2), the condition (3.9)
need not hold . Therefore, to apply the principle of material objectivity, it
is essential to replace a non-objective constitutive quantity by an objective
one, which can usually be done by adding, subtracting or multiplying some
non-objective terms. In our later discussions, we shall always assume that C
is objective.
Since (3.9) holds for any change of frame , an immediate consequence
follows from choosing Q(t) = 1 and c(t) = X o , such that

x* =x, t" = t + a,
 3.2 P rinciple of Material Objectivity 67

for arbit rary const ant a E JR . Clearly, t he induced linear transformation on

any t ensor space in this case is

Q* = 1 .

Therefore, the condition (3.9) implies that

F(/ , Xt,Bt , X , t + a) = F(/ , Xt , e,X , t).

Since t his is t rue for arbitrar y a, t aking the derivat ive of this equat ion with
resp ect to a, we conclude that

of = 0
ot '
i.e., F ca n not dep end on t explicitly. Therefore, we can rewrite the consti-
tutive relation (3.4) as

C(X , t) = F (pt(y, s), Xt(y, s), Bt(y, s), X) , (3.10)

Y EB
O ~ s < oo

and the condit ion (3.9) becomes

F(pt , (Xt )* , «,X) = Q* F(/ , Xt , «,X) , (3.11)

where

We sha ll often omit the dom ain and the dep end ence of the argument fun ctions
in F as ind icated in (3.9) for simplicity.

Remark. Comparing t he argume nts of the two sides of the above equa-
t ion, we not e that t he condit ion (3.11) only imposes restrictions on the
dep end ence of the response function upon the histories of motion , and
nothing on that of density and t emp erature fields. In view of this, it leads
to a different int erpret ation of the condit ion of material obj ectivity. From
(1.65), for a motion X and a cha nge of fram e *, we have

X*(Y, t*) = Q(t)( X(Y, t) - x o) + e(t) ,

t* = t + a.

If we define a new motion X* associate d with the motion X by

X*(Y,t) = Q(t)(X(Y,t) - x o) + e(t), (3.12)

then t he conditi on (3.11) can be written as

F(pt , X: , e. X) = Q* F(pt , Xt , o' , X) . (3.13)

68 3. Basic Principles of Constitutive Theories

Two motions X and X* related by (3.12) are called equivalent mo-

tions. Equivalent motions are motions that differ by an arbitrary rigid
transformation. We say that a sca lar s, a vector u , or a tensor T is an
invariant quantity under equivalent motions, if it satisfies

s(X*) = s(X) ,
u(X*) = Q(t) u(X) ,
T(X*) = Q(t) T(X) Q(tf ,

where X* and X ar e related by a rigid transformation given by (3.12) .

Physically, an invariant quantity is a qu antity that accompanies the ro-
tation of the rigid transformation. In this sense, the condition (3.13) is
sometimes invoked in postulating the princip le of material obj ectivity,
especially when only mechanical theories ar e concerned .

Principle o f mat erial obj ectiv ity (another version). The response
function of an obj ective constitutive quantity must be invariant for any
equivalent motions.

From this point of view, no change of frame of reference is involved .

These two versions of t he principle of mat erial obj ectivity are equivalent ,
nevertheless, we shall ret ain to our original interpretation of material
objectivity. 0

3.2.1 In R efe r ential D escription

Let", be a reference configuration of the body B. For a motion X, we have

Similarly, we can also write

p(X, t) = p,,( X , t) , B(X, t) = B,,(X , t) .

We could have abused the notation by writing p(X, t) = p(X , t) and B(X, t) =
B(X , t) as we have don e usually, but for clarity, more precise notations will
be used in this section.
In terms of referential des cription, we can rewrite the constitutive rela-
tion (3.10) relative to r: at a point X E B" and time t as

C(X , t) = F" (p~( Y, s), X~( Y, s) , B~(Y, s) , X ), (3.14)

Y ESI'\;
o ~ s < oo
 3.2 Principle of Material Objectivity 69

where F ", is related to F ¢ of (3.5) by

F",(p~(Y, s) , X~(Y , s), e~(Y , s), X)

(3.15)
= F¢(p~(Y , s), X~(Y , s), e~(Y , s), l>;-l(X)) .

F", is called the constitutive function of C relative to the configuration 1>;.

The change of reference configuration for the cons titutive function will be
considered later in the discussion of material symmetry.
To express the condition of material objectivity in the referential de-
scription, let ¢ and ¢* be two arbitrary frames , and denote the refer ence
configurat ion in these fram es by I>; and 1>;*, resp ectively,

X = I>;(X), X* = I>;*(X).

From (1.73) , the map

(3.16)

is given by a fixed rigid transformation,

X* = K(X - x o) + Co , (3.17)

where K is a constant orthogonal transformat ion du e to the change of frame.

For an observabl e qu antity C given by (3.14) , the obj ectivity property
(3.6) for change of fram e now takes the form,

F",. (p: . (Y *, t* - s), X:. (Y * , t* - s), e~. (Y* , t " - s), X *)

Y * E l3 K *
0 :$ .'1 < 00
(3.18)
= Q* F", (p",(Y, t - s), X",(Y, t - s), 8",(Y , t - s), X) .
Y ES"
0::;.<1 < =

We can also express the left-hand side in terms of the constitutive function
F", by the principle of material objectivity. Indeed, from the relation (3.15),
without writing out all the argument s, we have

F",- (X: . (Y* , t" - s), X*)

=F¢_(X:_(Y* ,t*-S) ,I>;* -l(X*))
= F ¢(X: _('r(Y),t * - s), 1>;* - l('r(X)))
= F", (X:_ ('r(Y) , t * - s), X),
where we have applied the principle of material objectivity that F ¢ = F ¢• .
The arguments for the density and the te mperat ure are similar, and since
they are objective scalar quantities, we have

P: ' ('r(Y) , t o) = P:_(Y * , to) = p",(Y , t) ,

(3.19)
e~ _('r(Y) ,t*) = e~_(Y*,t*) = e",(Y,t).
70 3. Basic Principles of Co nst it ut ive Theories

Therefore, from (3.18), we have t he following condit ion for material obj ectiv -
ity in the reference configuration ,

F " (p(Y , t - S), X:. (-r(Y ), t* - S), B(Y, t - S), X)

Y E 13K
o ~ s < oo
(3.20)
= Q* F" (p(Y , t - S), X,,(Y , t - S), B(Y , t - S), X) ,
Y E13K
o :$. s < oo

where
X: . (-r(Y ), t*) = Q(t)(X,,(Y , t) - Xo) + C(t). (3.21)
Not e that the condit ion (3.20) is expressed in t he reference configuration r:
in t he frame ¢ only.
In the above condition , we have again adopte d the usual not at ion by
writing p(X ,t) for p,,(X ,t) , B(X ,t) for B,,(X ,t) . Similar to the condition
(3.11) , t he condit ion of material obj ectivity (3.20) imposes restrictions on
the dep endence of response fun ction upon the history of motion onl y and not
on the histories of density and temperature of the body. We can write t he
condit ion (3.20) simply as

(3.22)

where

3.2.2 An Example: a Particular Class of Materials

Besides general consequences imposed by t he principle of mat erial obj ect ivity,
to be considered further in later sections, one can also apply the principle of
material obj ectivity dir ectly t o const it ut ive equat ions of a particular class of
materials, as illustrated in the following example.
Suppose t hat a given material is characterized by the consti t uti ve equa-
tion for t he st ress t ensor in the following form :

T = T(p ,v ,L) ,
where v is the velocity and L = grad v , t hen the condit ion of material obj ec-
tivity implies that

T(p* ,v* ,L*) = QT(p,v ,L)QT ,

for any chan ge of fram e given by (1.57) . From (1.66) and (1.77) and p* = p,
since the density is an objective scalar, the above condit ion becomes

T(p , Qv + (j(x - x o) + c, QLQT + (jQT) = QT(p, v, L)QT ,

for any orthogon al tensor Q(t) and any vector c(t) . In particul ar , if we choose
 3.2 Principle of Material Objectivity 71

Q(t) = 1 , then we have

+ c, L) = T(p , v , L) .
T(p , v
Since c(t) is arbit rary, it impli es that T can not depend on t he velocity v .
Hence, from L = D + W , where D is the st retc hing tensor and W the skew-
symmet ric spin te nsor, the obj ecti vity condition becomes
T(p , QDQT + QWQT + QQT) = QT(p, L)QT ,
for any orthogonal te nsor Q(t) . For any given skew-symmet ric te nsor W , we
can define an orthogonal t ensor Q(t) such t hat
Q(to) = 1 , Q(to) = -W,
and hen ce evalu ation of the object ivity condit ion at t = to leads to
T(p , L) = T(p , D) ,
which impli es that T can not dep end on the spin tensor W . Therefore, the
constitutive equ ation for the st ress tensor must reduce to
T = T(p , D) .
In addition, the obj ectivity condit ion still requires the function T to sa tisfy
the following relation,
(3.23)
for any orthogonal transformation Q. A representation of the fun ction T
satisfying this condition will be given later.

Exercise 3 .2.1 Let t he exponential of a matrix A be defined by the

series
A2 A3
ex p(A ) = 1 + A + -2! + -3! + ... .
Show t hat
1) if AB = BA , then exp(A) ex p (B ) = ex p (A + B) ;
2) if A is skew-symmetric, then ex p(A ) is orthogonal ;
3) for any skew-symmet ric t ensor W , the tensor function Q(t)
exp ( -(t - to)W) satis fies the following differential equ ation,
Q+QW =O and Q(to)=1.

Exercise 3.2.2 Consider a material (a linear viscous fluid) with t he

stress t ensor, relative to a Cartesian coordinat e system , represented by
T ij = - pbij + Vij klDkl,
where p and the fourth-order te nsor Vijkl are funct ions of the density
p. Show that the obj ectivity condition impli es t ha t for any orthogon al
tensor Q,
Vijkl = QimQjnQkpQlqVmnpq. (3.24)
Such a tensor is usually called a fourth-order isotropic tensor.
72 3. Basic P rin cipl es of Const it ut ive Theories

3 .3 Simple Material Bodies

Let p" (Y) be the density field in the refer ence configurat ion K" then it deter-
mines the density field p(Y , t) from t he deformation field X,,(Y , t) according
to the balanc e of mass (2.34) . T herefore, we can repl ace pt(y , s) by p,,(Y) as
t he independent vari able in the const it ut ive equat ion (3.14) , which ca n then
be written as

C(X , t) = F" (p,,(Y) , X~(Y , s), Bt(y , s), X) . (3.25)

Y EBK
o ~ s < CXJ

For constit ut ive equat ions of this form , thermomechanical histories of

any part of the body can affect the response at any point of t he body. In
most applications, such a non-local property is rarely relevant and may onl y
ca use unnecessary complications. Ther efore, it is usually assumed that only
t hermomechanical histories in an arbit rary small neighborhood of X affect
the mater ial response at the point X , and hence t hey can be approximat ed
at X by Taylor ser ies up t o certain order. In particular , if only the first
approximation is concern ed, we have

p,,(Y) = p,,(X) + Gradp,,(X)(Y - X) + 0(2),

X,,(Y , t) = X,,(X , t) + F(X , t)(Y - X) + 0(2),
B(Y , t) = B(X , t) + Gr adB(X , tHY - X) + 0(2),

and the constitutive relation (3.25) can be written as

C(X , t ) = F" (X~(X , s), Ft(X , s ), Bt(X , s), g ~(X, s), X) , (3.26)
O ~ S < CXl

where g " = Gr ad B is t he te mpe ra t ure gradient with respect t o the refer ence
configurat ion. Note that , sin ce p,,(X) and Gradp,,(X) are merely functions
of X, their pr esence has been absorbed into the explicit dep endence of t he
fun ct ional F" on X.
For a change of fra me, from (3.21) we have!

\7 x( X: . (r(X) ,t*)) = Q(t)F(X ,t) .

Therefore, the cond it ion of material ob jectivity (3.20) now t akes the form ,

F" (x *(X , t " - s) , Qt(s)Ft(X , s), «(x, s), g ~(X , s), X)

O ~ S < CXl
(3.27)
= Q*(t) F" (x (X , t - s ), r'tx, s), o'(x. s) , g~(X , s), X),
O ~S < CXl

1 Note that t he left-hand side is not F * (X, t* ), in gen er al , (see the remark
on p . 27). T he ass um ption t hat the referenc e configurati on is un affect ed by
t he change of fram e will give this relation as V'x(X ~(X,t *)) = F* (X ,t* ) =
Q(t) F(X , t ), and hen ce will lead t o the same cond it ion of material objectivity.
 3.3 Simple Material Bodies 73

where
x(X, t) = X,,(X, t) , x*(X ,t*) = X:.(r(X) ,t*) ,
denote the position of the material point X in the two frames , respectively.
We claim that the dependence of const it ut ive function on the position
is not allowed by the principle of materi al objectivity. Indeed, since (3.27) is
valid for any change of frame, in particular, let us consider a change of frame
given by
x* =x-xo+c(t) , t" = t.
In this case, Q(t) is an identity and the condition (3.27) becomes

which proves our claim, since c(t) - Xo E V is arbitrary. Therefore , the

constitutive relation (3.26) reduces to

C(X , t) = F" (Ft(X , s), e'ix, s) , g~(X , s), X) , (3.28)

o:Ss<oo

and the condition of material obj ectivity (3.27) becom es

F" (Qt(s)Ft(X, s) , «ix; s ), g~(X , s ), X)

o:S s< oo
(3.29)
= Q*(t) F" (Ft(X , s) , e'ix, s) , g~(X , s) , X) ,
o:S s< oo

or simply
F,,(Q tF t , et , g~, X) = Q*F,,(F t , et , o': X) .
Note that , as we have remarked before, this condit ion imposes no restrictions
e
on thermal histories, t (X , s) and g~ (X , s) .
A body with it s const it ut ive relation defined by (3.28) is called a simple
material body (due to Noll).2 For simple material bodies, the value of the
const it ut ive quantities at the point X depend on the histories of deformation
gradient , t emperature, and temperature gradi ent a t the point X only. This
class of materials may have a perfect temporal memory but only a very limited
non-local effect, nam ely, the neighboring points are taken into account, in the
first gr adi ent of defo rm ation and temperature, only as a first approximation.
Nevertheless, it t urns out that this class is still general enough t o include
most material bodies of pr actical int erest. Non-simple material bodies are by
no means unimportant , as an example, in theories of mixtures and porous
media [51] the density gradi ent (which is related to the second gradient of
deformation) must be taken into account in constitutive equat ions in order
to hav e a consist ent theory.
2 Noll, in his memoir [58], considered purely mechanical theory only.
74 3. Basic Principles of Constitutive Theories

A bo dy is called homog eneous if there exists a reference configurat ion

called a homog eneous configuration, such that t he const it ut ive fun ction
/'i"
F" do es not dep end on X expli citly. Therefore for a homogeneous simple
material body (3.28) reduces to

C(X , t) = F" (Ft(X , s) , et(X , s), g~(X , s)). (3.30)

0 ::;8< 00

In our later discussions , explicit dependence on X of response functions is

mostly irr elevant , t herefore we shall omit it from t he argument s of response
fun ctions for simplicity.

Exercise 3.3.1 Consider an elastic material with the stress tensor given
by T = ~(F) relative to a refer ence configuration «. A configurat ion r;
is called a natural configur at ion if

~(1)=0 ,

where 1 is the identity tran sformation. Show that any rotation has no
effect on a natural configuration. In other words, if k is anot her reference
configuration that differs from /'i, by a rigid rotation , then k is also a
natural configur at ion.

Exercise 3.3.2 A hyperelasti c mat erial is defined by the const it uti ve

equat ion
(3.31)
where o = u(F) is called the stored energy function . Show that if a
satisfies t he obj ectiv ity condition, then so does T.

Exercise 3.3.3 Let g = grad e be the spat ial gradient of t emperature

and the const it ut ive relation (3.28) be rewritten as

then the condit ion of material obj ectivity becomes

Note that this condition imposes restrictions on the response function j:

not onl y on t he history of the deformation gradi ent but also on that of
t he spatial te mpe rat ure gradient , since g" = F T g .
 3.4 Reduced Constitutive Relations 75

3.4 Reduced Constitutive Relations

For simple material bo dies , response functions for the constitutive quantities
(3.2), t he stress tensor, the heat flux vector and the int ernal energy, are given
by
T( X , t) = T( Ft ,fl ,g ~ ) ,
q (X , t)= Q (F t , (}t , g ~ ) , (3.32)
c(X , t) = £ (F t
, e g~ ) .
Here , we have assumed that all constitutive quantities of a material bo dy
depend on t he same set of ind ependent vari ables. This is referred to as t he rule
of equipresence (Truesdell). Even though this ru le is not to be considered as a
physical principle, it may serve as a starting assumption for the formu lati on
of a constitutive t heory before any ot her req uirements are analyzed, so as
not to impose unnecessary restrictions or excl ude some possible (but perhaps
still unknown) cross-effects prematurely.
The constitutive functions (3.32) must satisfy the condition of mate-
rial objectivity (3.29) . By t he definit ion (1.60) of Q*, we have t he following
conditions,

T( Q t Ft ,(}t ,g ~ ) = Q T(Ft , (}t, g~)QT ,

Q(QtFt ,(}t , g~ ) = QQ(Ft , (}t ,g~) , \;j Q(t) E OW ). (3.33)
£ ( Qt Ft ,(}t , g~ ) = £ ( Ft ,(} t ,g ~ ) ,
These conditions of material objectivity are typical for tensor, vector, and
scalar objective constitutive quantities, respectively.
We have seen t hat t he response functions must satisfy the conditions
(3.33) imp osed by the principle of material objectivity. Such restrictions ca n
be made explicit in a couple of different ways , as we shall see now . Since the
condition of material objectivity does not impose any restrictions on thermal
histories , for simplicity, we shall omit them and write, typically for t he stress
tensor

\;j Q(t) E O(V ), (3.34)

and for all non-singular F( t) E £ (V ). Solutions of t his condition are given in

the following proposit ion. The extension to other vector or scalar constitutive
functions is straightforward.

Proposition 3.4.1 For a simple material body, the constitutive functio n T

satisfies the condition of material objectivity (3.34) if and only if it can be
represented by
(3.35)
where F = RU is the polar decomposition of F . The restriction of T to the
positive symm etric stretch history ir is arbitrary.
76 3. Bas ic P rinciples of Constitutive T heories

Proof Suppose t ha t T(Ft) satisfi es (3.34) , then by the pol ar decomp osition
F = RU, we have

\;f Q(t ) E O(V) ,

and (3.35) follows imm ediately by choosing Q = R T . Conve rsely, if (3.35)

hold s, then for any Q E O(V) , QF is non-singular and

QF = QRU = (QR) U
is its polar decomposition. Hence, (3.35) implies that

Therefore, (3.35) is also sufficient for (3.34) . 0

From the representation (3.35) , we note t hat the mat erial response de-
pends only on the cur rent valu e of rotation par t of deformation. The past
history of t he rotation R t (s) for all s > 0 do es not influence the material
resp ons e at all.

Corollary 3.4.2 T he repres entati on (3.35) can be pu t into th e following

form :
(3.36)
where 5 is an arbit rary fun ction of th e Cauchy-Green tensor C = FT F .

Proof From (3.35) and F = RU , we have

F - 1T F - T = U - 1RT(RT(U t ) RT) RU - 1 = U -1T(U t ) U- 1 = 5(C t ) ,

where we have redefined the fun ction T by the new func tion 5 dep ending on
C = U 2 = F TF. 0

The const it ut ive relations (3.35) and (3.36) are the solutions of the con-
dition (3.34) . Therefore, they ar e not subjected to any further restrictions
from the objectivity condi t ion. Con stitutive relations of this kind are said to
be in reduced form s.
We can obtain a different vers ion of the reduced const it ut ive relation
(3.36) , for qu antities in material description introduced in (2.84) , nam ely,
t he material heat flux vector qK = 1J1F-1q and the (first) Piola-Kirchhoff
stress t ensor T K = IJIT F - T . In addit ion, we shall define the second Piola-
Kirchhoff stress tensor as

which , unlike the (first) Piola-Kirchhoff stress t ensor, is a symmetric tensor .

As a consequence of (3.36) we have the following
 3.5 Material Symmetry 77

Proposition 3.4.3 Constitutive relations for the second Piola-Kirchhoff

stress tensor 5,,, the material heat flux q" and the internal energy density c
of a simple material body satisfy the condition of material objectivity if and
only if they can be represent ed in th e following forms,

5" = S,,(ct ,()t ,g~) ,

q" = Q ,,(ct ,()t ,g~), (3.37)
E = [,,(ct ,()t ,g~) ,

where C is the right Cauchy-Green strain tensor , and g" = Gr ad() is the
temperature gradient with respect to the reference configuration.

Exercise 3.4.1 Consider the following constitutive relation for the

stress tensor,
T = tal + tIB + t2B2 ,
where B is the left Cauchy-Green tensor, and to , h , and t2 are arbit rary
functions of tr B , t r B 2, and t r B 3 . Verify that this constitutive relation
satisfies the obj ectivity condit ion and is a special case of (3.35) .

Exercise 3.4.2 Prove that for a simple material body, the stress tensor
has a reduced cons titutive relation of the form

where F = RU, C = U 2 , C, = Ft Ft and F, is the relativ e deformation

gradient (see (1.39) and (1.42)) .

Exercise 3.4.3 Verify that for a simple material body, one has the
following reduced constitutive relations:

T = RT(U t , «' gt)K, R T

,
q = RQ(Ut ,()t ,g~),

c = [(ut ,()t ,g~) .

3 .5 Material Symmetry

The const it ut ive relations of a simple material relative to a reference config-

uration K, from (3.28) can be written in t he form ,

C(x , t) = f"" (Ft(X , s), «ex , s), g~(X , s)), x = X,,(X , t) . (3.38)
0 ::0; 8 < 00
78 3. Basic Principles of Constitutive Theories

Suppose that k is another reference configuration and the constitutive rela-

tions relative to k can be written as

C(x ,t) = F" (pt( X,s) , ()t(X,s), g~ (X , s) ) , x = X,,( X, t). (3.39)

O:'::s<oo

Let P = V'x k 0 K-
1,
then from (1.10)

F=FP,

By comparing (3.38) and (3.39) , we have the following relation between the
response functions F" and F",
(3.40)

Therefore, a response function relative to a reference configuration deter-

mines that relative to any other reference configuration. Note that P is time
ind ependent since the change of refer ence configuration is.
A material body subjected to the same thermomechanical history at two
different configurations K and k , in general, have different results. However, it
may happen that t he results are exactly the same if the material possesses a
certain symmetry that makes it unable to distinguish K and k: For example,
one can not distinguish the material body with a cubic crystal structure
before and after a rotation of 90° about one of its crystal axes.
Definition. Two reference configurations K and k are said to be materially
indistinguishable if
(3.41)

If we denote the gradient of the change from K to k by G = V'x (k 0 K - 1) ,

then by (3.40) , the condition (3.41) is equivalent to
(3.42)

We call a transformation G E L:(V) that satisfies (3.42) a material symmetry

transformation with respect to K.

Ft ,et ,g~
K(8) Resu lts

la
F' 1 0' , g 'K
n
k(8)
)
Results

F ig . 3 .1. Material symmetry transformation

Physically, this means that the response of the material with respect
to the configuration K can not be distinguished from that with respect to
 3.5 Material Symmetry 79

the configuration F., obtained from K by a linear trans form ation G , by any
possible (or ideal) exp erim ent . The idea is schemat ically represent ed in the
diagram (Fi g. 3.1) .
We shall assume that a mat erial symmetry t ransfo rm ation is volume pre-
serving (consequ ently, by (2.34) also density preserv ing) , since otherwise, a
mat erial body could suffer arbit rarily lar ge dilat ation with no change in ma-
terial response - a conclusion that seems physically un acceptabl e (for another
just ificat ions, see [27, 75]). Therefore, we must require that Idet GI = 1, i.e., a
material transformation G must be a unimodular transformat ion, G E U(V).

Proposition 3.5.1 Th e set of all ma terial sym me try transformations with

respect to K , denoted by YK ' is a subgroup of the unimodular group ,

YK <;;; U(V) . (3.43)

Proof. We need only t o show that OK is a group:

1) It is obvious that 1 E YK'
2) For any G I , G z E YK' by definition , we have

:FK(p t , et , g~) = :FK(ptG I , e,Gf g~)

= :FK((ptGdG z , et , Gr(Gf g ~)) = :FK(pt(G IG z) , et , (G I Gz f g~) ,
which impli es that G I G z E YK'
3) For any G E YK' since Idet GI = 1, G - I
exists , and

which implies G- I E YK' Therefore, YK is a group. D

We call YKt he material symme try group of C with respect to the reference
configurat ion K. It is clear that YK dep ends on K as well as t he const it ut ive
qu antity C. More specifically, we can denote it by YK(C) , In other words, we
may have different symmet ry groups for different const it ut ive qu antities of
t he same material body. T he largest group contained in the symmetry groups
of all const it ut ive qu an tities of B is called the material sym me tr y group of B .
Suppose t hat F. is ano ther reference configur at ion and P = 'VX(F.o K- I) ,
t hen for any G E YK ' we have from (3.40)

:Fk(pt , et , g~) = :FK(ptP, «,p Tg ~) = :FK((pt P)G, et , GT (pTg~))

= :FK(pt(PGP- I)P, et , p T(PGP -I f g~)

= :Fk(pt(PGP- I) , et , (PGP -If g ~) ,

which implies t hat PGP- l E Yk. Therefore, we have proved the following
proposition.
80 3. Basic Principles of Constitutive Theories

Proposition 3.5.2 For any r: and ;:., such that P = \7",(;:'0 Ii-I) , the follow-
ing relation holds,
(3.44)
The relation (3.44) is referred to as Noll's rule, which establishes the
relation between symmetry groups with respect to two different reference
configurations. Note that since P is a change of reference configuration it is
non-singular, but need not be unimodular. In particular, if P = o l , a :f= 0,
then 9ft = 9"" i.e., dilatation (a > 1), contraction (0 < a < 1), and central
inversion (a = -1) leave the material symmetry group invariant.
Physical concepts of real materials such as solids and fluids, are usually
characterized by their symmetry properties. For example, one such concept
can be interpreted as saying that a solid has a preferred configuration such
that any non-rigid deformation from it alters its material response, while for a
fluid any deformation that preserves the density should not affect the material
response. Based on this concept, we shall give the following definitions of
solids and fluids due to Noll [58] .
Definition. A material is called a solid if there exists a reference configura-
tion n ; such that 9", is the full orthogonal group or a subgroup of it, i.e.,

(3.45)

Such a configuration is called an undistorted configuration for the solid .

Definition. A material is called a fluid if, for a reference configuration «,

the symmetry group is the full unimodular group, i.e. ,

9", = U(V). (3.46)

For a fluid , Noll's rule (3.44) implies that

Vii, k : (3.47)

Therefore, a fluid has the same symmetry group with respect to any config-
uration. In other words, a fluid does not have a preferred configuration.
Definition. A material that is neither a fluid nor a solid will be called a
fluid crystal.
In other words , for a fluid crystal, there does not exist a reference con-
figuration r: for which either 9", ~ O(V) or 9", = U(V).

Exercise 3.5.1 Consider a constitutive equation for the stress ten-

sor given by T = T..(F,B ,g) . Use the material objectivity condition
to show that if G E 9",(T), then -G E 9",(T). Moreover, does this
conclusion remain valid for the symmetry group 9",(q) of the heat flux
q = Q",(F,B,g)?
 3.5 Material Symmetry 81

Exercise 3.5.2 A mat erial for which 9" = {1, -1} is a triclinic solid .
Show that the relation (3.47) also holds for t his solid .

Exercise 3.5.3 Let '" and k be two undistorted configurat ions of a

solid, and let P = V,,( k 0 ", -1). Suppose that t he polar decomposition
of Pis P = RU . Show t hat
(3.48)

Moreover , if 9" = O(V) t hen 9k = 9" and U = 0:1 for 0: E JR.

3.5.1 Constitutive Equation for a Simple Solid Body

Combining the reduced constitutive relations (3.37) and the symmetry condi-
t ion (3.42) , we have the following theorem that characte rizes the most general
constitutive relations for simple solid bodies.
Proposition 3.5.3 For a simple solid body, the constit utive relations for
the second Piole-Kirchhoii stress, the m aterial heat flux and th e internal
energy, relative to an un distorted configuration ", with symmetry group 9" ,
satisfy the conditions of material objectivity and symmetry if and only if the
constit utive functions for C" = {3", q" , E} can be represented by
C" = F,,(Ct ,f)t ,g ~) .
Moreover, F" = {S" , Q" ,£,,} mu st satisfy the following conditions,
S,,(QctQT ,f)t ,Qg~) = QS,,( ct ,f)t ,g~)QT ,
Q,,(QCtQT , e' , Qg~) = Q Q,,(Ct , f)t , g~ ), v Q E 9" s: O(V) . (3.49)
£K(Q CtQT ,f)t ,Qg ~) = £K( Ct ,f)t ,g~) ,

Proof By the Proposition 3.4.3, we only have t o prove the condit ions (3.49) .
We shall prove for the stress tensor T , from t he definit ion (2.84) and t he
redu ced const it utive relation (3.37) , we have
T(F t ,f)t ,g ~) = Idet FI - 1FS,,( Ct ,f)t ,g ~) FT .

s:
For a solid , we have Q E 9" OW) , and by taking G = QT = Q-1 E 9" in
the condit ion (3.42) , we obtain, from the above relation , t ha t

I det FI- 1F S,,( C t , f)t ,g~) F T

= Idet(FQT)I - 1(FQT) S,,((FtQTf(FtQT) , f)t , Qg ~) (FQT)T .
Since Idet QI = 1, it becomes
S ,,(C t , f)t , g~) = QTS ,,(QCtQT , f) t , Qg~) Q,
which proves (3.49) 1. The proofs for Q" and £" are similar. 0
82 3. Basic Principl es of Consti tutive Theories

Relations of t he typ e (3.49) for constit utive functions are of particular

inter est in the light of representations to be considered in the next chapt er.
We note that , in general, G E QK do es not impl y GT E QK' but for
G E O(V) n QK ' since GT = G- 1 , GT must also belong to QK ' Therefore, t he
condit ion of material objectivity (3.33) and the mat erial symmetry condition
(3.42) imply t he following condit ion,

for all Q E O(V) n QK' In particular, it is valid for solids . However , unlike the
constitutive function 5 K of (3.49) , the fun ction T is still required t o satisfy
t he objectivity condit ion (3.33) .

3.5.2 Constitutive Equation for a Simple Fluid

In order to derive t he general const it utive relations of simple fluids, let us

rewrite the const it ut ive relation (3.38) in t er ms of relative description (see
Fig. 1.5) . From (1.42) , the relative deformation gradient Ft, defined by (1.39),
is relat ed to the deformation gradient F by the relation,

x = XK(X , t) . (3.50)

We can also define a relative t emperature gradient gt in a simil ar manner ,

which implies that

(3.51)

where g K = Grad e is t he referenti al temperature gradi ent in the configur at ion

K, . From (3.50) and (3.51) we have

g~( s) = F(t) -T g~(s) ,

and

where g = grad e is the spatial t emper ature gradient .

In t erms of relative gradient of deformation and t emperature, we can
rewrite (3.38) as

C(x ,t) = j: (Ff(x , s) , et(x, s), g~(x , s); F(x , t)). (3.52)
0 :'08< 00
 3.5 Material Symmetry 83

The constitutive functions F and r K of (3.38), from (3.50) and (3.51), are
related by

from which one can verify that the condition of material objectivity (3.33)
and material symmetry (3.42) become

F(QtFttQT,(}t,Qg~ ;QF) = Q* F(Fi ,(}t,g~ ;F) , VQ E OW) ,

(3.53)
F(Ftt, (}t, g~; FC) = F(Ftt , (}t , g~; F) , VC E 9K'

respectively. To see this, note that we have

with
- t P-T -t
gt = gK'
By taking first P = QF and fJK = gK for Q E O(V), we have
Pi = QtFtF-1QT = QtFttQT ,
fJ~ = Q F - T g~ = QgL

which lead to (3.53h from (3.33). Then, by taking P = FC and fJK = C T gK

for C E 9K' we have
Pi = FtCC -1F - 1 = FtF - 1 = Ftt,
- t = F-TC -TCT gK
gt t
= F -T gKt
= gtt ,

which lead to (3.53h from (3.42).

We can obtain general constitutive relations for a simple fluid , with sym-
metry group 9K = U(V), from the conditions (3.53).

Proposition 3.5.4 For a simple fluid, the constitutive function satisfies the
conditions of objectivity and symmetry (3.53) if and only if it can be repre-
sented by
c = j(Ci,(}t ,g~;p), (3.54)
where C, = Ft Ft is the relative Cauchy-Creen strain tensor and p is the mass
density in the present configuration. Moreover, the constitutive functions
j = {T, Q, £} for the stress tensor, heat flux vector, and internal energy
must satisfy the following conditions:

v Q E OW). (3 .55)
84 3. Basic P rinciples of Constitutive Theories

Proof Since for a simple fluid, the symmetry group gK = U(V) , by taking
the unimodular tensor G = [det F11 / 3 F- 1 the condition (3.53h gives

Since [det F[ = PK / P, where PK is th~ mass density at the reference configu-

ration , we can rewrite the function :F as

The condition (3.53h for the stress tensor now becomes

T(Ftt , ot , g~ ; F) = QTT(Qt F:QT , 0L Qg~ ; p) Q.

With the decomposition F: = RWL taking Qt(s) = RH s f , hence Q(t) =

R~(O) = 1 , we obtain (3.54) by replacing U: with c:
= (Uf) 2. Similarl y,
choo sing Qt(s) = (J(t) R~(s) T , hence Q(t) = (J(t) , for any (J(t) E O(V), we
obtain
~ - t -T t - t - ~ t t t -T
,(QUtQ , 0 , Qgt; p) = Q:F( Ut ,O , gt ;p) Q ,
which implies (3.55h . The proof of sufficiency is straightforward. D
From this representation , we conclude that , for a simple fluid , the consti-
t uti ve dependen ce on the deformation gradient F reduces to the dependence
on it s det erminant det F only.

3.5.3 Fluid Crystal with an Intrinsic Direction

We shall not consider fluid cryst als in more det ail in t his book, except for
the following interest ing example.
Consider a fluid crys tal with the symmetry group gK relative to an or-
t honormal basis {e l' e2, e3} in the reference configurat ion /'i, given by

c; = { [ b
a d 0]
e O, Ig(ae - bd)1 = 1} .
c f 9

On e ca n easily verify that this is a subgroup of U(V) . Moreover , for any

G E gK' we have

In other word s, t his material has an intrinsic dire ction e 3 in the reference
configurat ion n. A material with needle-like crystals that can flow in such a
way that they are always aligned in on e direction (therefore, it is called a
fluid crystal) is an exa mple of materials that possess such a symmetry.
 3.5 Material Symmetry 85

Let the constitutive equat ion for the stress tensor be given by T = T(F) ,
then the objectivity and the symmetry conditions require that
T(QF) = QT(F)QT , \/Q E O(V),
(3.56)
T(FG) = T(F), \/G E 9K .

In order to analyze these conditions, first we note that they imply

(3.57)

In particular , for F = 0: 1 , it leads to

T(o:1) = RT(o:1)RT .
That is, the orthogonal tensor R commut es with the symmetric tensor T( 0: 1).
By the commutation theorem (see Sect. A.1.8), R must preserve all the char-
act eristi c spaces of T(o:l) . Note that the general form for R E O(V) n 9K
is
COS sin o 0] o
R = ± - sin () cos () 0 ,
[
o 0 1
which preserves two characteristic spaces, a one-dimensional one, spanned
by e3, and a two-dimensional space, spanned by {e 1 , e2} . Therefore, by the
spectral theorem, T( 0: 1) must be of the form,

T(o:1) = T(o:)(el ® el + e2 ® e2) + T'(O:) e3 ® e3

(3.58)
= T(o:)l + T3(0:)e3 ® e3,
where T3 = T' - T.
For an arbitrary F, suppose we can decompose F in the form ,

F = o:K G, K E O(V) , G E 9K.

Then, since Ge 3 II e3, we have Fe3 II Ke3 and hence

Fe3
Ke3 = IF e 31·

Therefore , this decomposition is possible if we define K as a rotation that

takes the uni t vector e3 into the unit vector in the direction of Fe3. Moreover,
sin ce Idet KI = I det GI = 1, we must have 0: = [det FI 1 / 3 . Substituting this
decomposition of F into (3.57), we obtain

T(F) = RTT(o:RKGRT)R = RTT(o:RK)R

= T(o:K) = KT(o:1)K T
= T(o:)l + T3(0:)K(e3 ® e3)KT,
where we have mad e repeated use of (3.56) and, in the last ste p, of the relation
86 3. Basic Principles of Constitutive Theories

(3.58) . Since I det FI = Pt< / P, we can write the constitutive equation as

Fe3 Fe3
T(F) = -p(p) 1 + a-(p) lFe 31® IF e 31' (3.59)

Conversely, on e can easily show that if (3.59) holds then the condit ions (3.56)
are satisfied . Therefore, we have shown that the st ress of this fluid cryst al
consists of a hydrostatic pressure and a pure te nsion in the deformed intrinsic
direction. Mor eover , the pr essure p and t he tensile stress a- are functions of
density only.
Not e that in this exa mple, we are able t o reduce the const it utive function
for the stress T = T(F) , a symmetric tensor fun ct ion of a tensor variable
(equivalent t o six scalar fun ctions of nin e vari abl es) to a representation that
is explicit to within two scalar functions of one vari abl e only. Represent ations
of this kind are among the most useful results of the const it ut ive theory. Other
representations of constitutiv e functions for solid s and fluids will be discussed
in the next chapter .

Exercise 3.5.4 Relative to an orthonormal basis, let the material sym-

metry group of a fluid crys t al be given by

9t< ={ d
a b
e
C] ,
f Ig(ae - bd)1 = 1} .
[
o 0 9
Show that it is a group and an alyze the const it ut ive properties of t his
material.

3.6 Isotropic Materials

Definition. A mat erial body is called isotropic if there exists a configur at ion
K" such t hat
(3.60)
Such a configur at ion is called undisto rt ed for the isotropic mat erial body, or
simply an isotropic configuration.
Physically, we can interpret the above definition as saying that any ro-
tation (orthogonal transformation) from an isotropic configurat ion do es not
alte r the material response. The following theorem char acteri zes isotropic
mat erials [59] .
Theorem 3.6.1. The orthogonal group is m aximal in the unimodular group,
i.e., iI9 is a group such th at
OW) c 9 c U(V) ,
then either 9 = O(V) or 9 = U(V) .
 3.6 Iso tropic Materi als 87

Therefore, any isotropic mat erial is either a fluid , 9K = U(V) , or an

isotropic solid , 9K = O(V) . Any other materials are anisot ropic. Anisotropic
materials include fluid crystals and anisot ropic solids.
It is obvious t hat, for an isotropic solid , an isotropic configurat ion is
also an undistort ed configuration, defined earlier in (3.45) . Conversely, if k
is anot her undistorted configuration, i.e., 9k ~ O(V) , then from (3.48) , it
follows that

in other word s, for an isotropic solid , an undistorted configurat ion is also an

isotropic configuration. Moreover , by Noll's rule (3.44) and the relation (3.48)
two isotropic configur at ions r: and k can differ by a dilatation and a rotat ion
only (see Ex ercise 3.5.3) .
In general , only certain particular reference configur at ions are undis-
torted for a given solid bod y. For a distorted configur at ion of the solid ,
the symmetry group neither cont ains the orthogonal group nor is contained
within it .
For an anisot ropic solid , the symmetry group relative t o an undistorted
configurat ion is a proper subgroup of t he or thogonal group, 9K c O (V ) and
9K =I=- O(V) . Its const it utive relation is already given in Proposition 3.5.3,
no further general reductions can be made unle ss t he symmetry group 9K is
spe cified .

Example 3.6.1 Transversely isotropic and ortliotropic solid bodies.

Relative to an ort honormal basis {nl ' n 2, n 3} in a refer ence configur a-
tion n; let the mate rial symmetry groups of two anisotropic solids be
given by

91 = {Q E O(V ), Qnl = nl} ,

92 = {Q E O(V) , Q(nl 0 ndQT = nl 0 n l , (3.61)
Q(n2 0 n2)QT = n 2 0 n2 , Q(n3 0 n 3)QT = n 3 0 n 3}.

The group 9 1, containing any rotation t hat preserves the vector n 1,

is called a tran sverse isotrop y group, and a material body having this
symmetry is called a transversely isotropic solid , while th e group 92,
containing any rot ation that pre serv es t hree mu tually orthogonal direc-
tions, is called an orthotropy group, and a mat erial poss essing such a
symmet ry is called an orthotropic solid .
These groups are merely examples of anisotropic solids . In particular,
we rem ark that 91 is not t he only group that charac te rizes the so-called
transverse isotropy, other char act eri zations are possibl e.
Som e representations for the constitutive functions of these material
bodies will be discussed in the next chapte r. D
88 3. Basi c Principles of Constitutive Theories

3.6.1 Constitutive Equation of an Isotropic Material

For isotropic materials, besides the general const it ut ive equat ions for solid
bodies given in Proposition 3.5.3 for g" = O(V) , we also have the following
representation.
Proposition 3.6.2 For an isotropi c simple mat erial body, th e constitutive
function satisfies the condition of material objectivity and symmetry if and
only if it can be represent ed by
~ t t t
C=:F(Ct,()t ,gt ;B) , (3.62)
where C; = Fl Ft and B = F FT . Moreover, the function F = fT, Q, £}
for th e stress tensor, heat flux vector and internal energy must satisfy the
following conditions:
T(QciQT ,aLQg~ ;QBQT) = QT(Ci ,aLg~ ;B)QT ,
= Q Q(Ci , aLg~ ; B) ,
Q(Q CiQT , a~ , Qg~ ; QBQT) 'tj Q E O(V) . (3.63)
£(QciQT ,aLQg~ ;QBQT) = £(Ci ,aLg~ ;B) ,

Proof For isotropic materials, 9 = O(V) , by t aking F = R U and G = R T ,

the condit ion (3.53)z becomes

F(Ftt , at , g~ ; F) = F(Ftt , at , g~ ; R U R T) .
Since V = RURT and B = V 2 , the function F can be rewritten as
F(Fi , at,g~ ; F ) = F(Fi , at ,g L B).
The rest of the proof is similar to that of Proposition 3.5.4. 0
From t his pr oposi tion we see that the rotation does not affect the re-
sponse of isot ropi c materials at all, which is an expected result , of course.
Since fluids and isotropic solids are the only isotropic materials, t he above
representation (3.62) is essent ially a representation of isotropic solid bodies.
The representation (3.54) for fluids is a sp ecial case of (3.62), in which the
dependence on the left Cauchy -Green t ensor B reduces t o the dependence of
detB only.

Exercise 3.6.1 Consider a thermo elastic mat erial with the const it ut ive
relation for the stress t ensor given by T = T(F,a ,g) . Prove dir ectly
from the obj ectivity and the symmetry condit ions that if the material is
isotropic then the constitutive relation must reduce to the form
T = T(B ,a,g) ,
where r: is an isotropic configuration and T is an isotropic function.
 3.7 Fading Memory 89

3.7 Fading Memory

In this sect ion, we shall give a bri ef discussion of memory effect s of simple
materials and the linear approximation that gives rise to the Boltzmann-
Volterra theory of viscoelasticity.
We shall consider only a mechanical theory. From the general solut ion
(3.35) for simple materials, the st ress t ensor can be represented by

F=RU.

The depend ence on X is not indicated , for simplicity. We emphas ize that in
this representation, T(F t ) and T(U t ) are t he values of the same fun ctional
T( ·) evaluate d at two different functions F t(s) and Ut( s) for s E [0, (0) ,
resp ectively.
It is useful t o put the functional T(U t ) in a slightly different form by
writing it as a sum of a te rm at t he present value of deformation and a term
that vanishes when the bod y has always been at rest ,

(3.64)

where
and K(O ;U(t)) = O.
G t (s) is the past history of the digression from t he pr esent st ate. We shall now
make an addit ional physical assumpt ion that t he memory fad es in time. In
order to do this we shall first introduce the concept of an influence 'funct ion
that is used to cha ract erize the memory of past histories the material can
recall."

Definition. We call h( s), defined for 0 :::; s < 00, an influence jun ction if it
is monotonically decreasing and

h(s) > 0, 1 00

h(s) ds < 00 . (3.65)

The influen ce func tion is material dep endent. h(s) = (s+ 1)-P for p > 1,
and h(s) = e -{3s for f3 > 0, where both p and f3 are material par am et ers , are
some examples of such funct ions. We can then define a norm to measure the
recollection of past histories.

Definition. The norm IIGII of a history G(s) relative to an influence function

h( s ) is defined as
IIGII = CorOO IG(s)1 2h(
s) dS)
1/2
, (3.66)

where IGI is the norm of the t ensor G, i.e., IGI 2 = t r GT G.

3 See [la , 71] and for more general concept of fading memory sec [9, 73].
90 3. Basic Principles of Constitutive Theori es

The value of IIGII will be referred to as the recollection of G. The influence

function h(s) is regarded as the weight in computing the norm of G(s) . Since
h(s) decre ases monotonically, the values of G(s) for small s (recent past)
have a greater weight than the values for large s (distant past) . Not e that by
(3.65) all the constant histories, have finit e recollections. A constant history
AC(s) is a history such t hat AC(s) = A(O) for 0 :::; s < 00 .
A deform ation history that has always b een near ly at a const ant history,
or that may have suffered some large departures from a const ant history only
in t he distant past , has a small recollection of past digressions in t he norm
(3.66) . These are those circumstances that should not much affect t he st ress
in a ma terial with fading memory, as we expect from our experience.
The set of all histories with finit e recollections forms a Hilb ert sp ace IH
with the inner product defined by

(G, H)JlI = 1
00

tr(G(sf'H(s)) h(s) ds (3.67)

for any G , HE IH.

We call a material defined by the cons titutive equation

F=RU, (3.68)

where

(3.69)

a material with fading memory, if both the function T and the functional JC
are smooth in some proper sense in t he usual norm in L:(V) and t he fading
memory norm (3.66) for past histories in IH.

3.7.1 Linear Viscoelasticity

We need t he notion of differentiability in the history space IH.

D efini t io n . A functional F defined on IH is said to be Frechet differentiable
at the zero history if there is a conti nuous linear functional 6F such that

F(G) = F(O) + 6F(G) + O( IIGII),

where the remainder is of the order O(IIGII) in the sense t hat

lim O(IIGII) = O.
IIGII--+o IIGII

In other words , it is a second-order term in IIGII, and it will b e denoted simply

by 0(2).
 3.7 Fading Memory 91

If we assume that the functional K of (3.69) is Frechet differentiable, we

have
K(G ;U) = 5K(G;U) + 0(2),
because K(O , U) = O. A theorem of the theory of Hilbert spaces (Riesz repre-
sentation theorem, e.g. see [5]) states that every continuous linear functional
may be represented by an inner product. In the present case, with the inner
product (3.67) by allowing for the fact that the funct ional K is tensor-valued
rather than scalar-valued , it follows that

5K(G;U) = 1 00

K(s ;U) [G(s) ]h(s) ds.

Here , the kernel K E Sym(V) ®Sym(V) is a linear transformation of the space

of symmetric tensors into its elf, namely in component forms, (K [G])ij =
K ijklGkl , and ~nce, for any indic es i and j fixed, K i j . . must belong to lEI .
It follows that K must have the property:

We then have the following linear approximat ion,

K(G t ;U) = 1 00

K (s;U)[Gt(s) ]ds + 0(2), (3.70)

where K (s;U) = K( s;U)h( s) have the property:

1 00

IK (s; U) 12 h(s) -1 ds < 00 . (3.71)

By introducing the stress relaxation junction M ,

M (t , U) = -1 00

K (s;U) ds, M (s, U) = K (s;U), (3.72)

and substituting (3.70) into (3.69), we obtain

T (Ut ) = T(U(t)) + 1 00

M(s; U(t ))[U(t - s) - U(t) ]ds + 0(2),

which, upon int egration by parts, gives

T (U t ) = T(U(t)) + 1 00

M (s;U(t)) [U(t - s)]ds + 0(2). (3.73)

92 3. Basic Principles of Constitutive Theori es

The const it utive equ ation, T = 7(Pt), given by the above integral equa-
tion up to the first gradi ent in history t akes the followin g form,

R(t f7(p t)R(t) = T(U(t)) + 100

M(s ;U(t))[ U(t - s )]ds . (3.74)

This const it ut ive equat ion defines the so-called finit e linear viscoelasticity,
which is a theor y of finit e deformation with linear depend ence on the history
of st rai n rat e.
Note that the relax at ion function M E Sym(V) ® Sym(V) is mater ial
dependent and must satisfy the condit ion rel ative to the influence function
h(s) of the fading memory norm (3.66) ,

lim M( s; U)
8 --+00
= 0, (3.75)

3.7.2 Boltzmann-Volterra Theory of Viscoelasticity

We furt her consider the case of small deformations. The displ acement gradi-
ent H = P - 1 is assumed to be an infinitesim al qu antity with IH(t)1 of t he
order f « 1 for all time t. We use t he order symbol o(n) in the sense of the
usual norm as well as the fading memory norm,

Ilo(n)11 < k' f n ,

for some constants, k and k' , as a consequence of (3.65) and (3.66). We have
from (1.22)
U = 1 + E + 0(2), R = 1 + R + 0(2),
- -
where E and R ar e the infinitesim al strain and rot ation tensors, resp ecti vely.
The linear appr oximation for the function T( U) ca n be written as

T(U) = To + £[ E] + 0(2), (3.76)

where
To = T(1)
is called the residual stress, i.e. , the stress the mat erial sustains in the refer-
ence configuratio n, and

is a fourth-order tensor in Sym(V) ® Sym(V) . Here, of course, we have as-

sumed that t he function T(U) is differentiable in the usu al sense.
Mor eover, we have

U = E + 0(2),
M( s, U) = M( s, 1) + ouM( s, l)[E] + 0(2) = M(s) + 0(1).
 3.7 Fading Memory 93

Finally, putting (3.76) and (3.73) into (3.68) and neglecting the higher-
order t erms, we obtain

T(t) = To + R(t)To - ToR(t) + L[E(t)] + 1 00

M(s)[ E(t - s )]ds. (3.77)

We remark that the st ress relaxation funct ion M depends not only on
the material property but also on t he configuration that has been taken as
the reference. The sp ecial case M( s) = 0 gives

T(t) = To + R(t)To - ToR(t) + L[E(t)],

which corresponds to the theory of infinitesimal elastic deformation supe r-
posed on a large deformation, taking as the reference configuration with the
non-vanishing stress field To .
When the reference configur at ion is a natural st ate , i.e., To = 0, (3.77)
reduces to
T(t) =
00

L[E(t)] + 1 M(s)[E(t - s)]ds. (3.78)

This is t he classic al equation of linear viscoelasticity of the Boltzmann-

Volterra theory.
The special case M (s) = 0 corr esponds to t he classical t heory of linear
elasticity.
T=L[ E].

We call L the elasticity tensor of the material. Since both the elast icity
tensor L and the stress relaxat ion function M ar e fourth-order te nsors in
Sym(V) @ Sym(V), t heir component s satisfy the following symmetry condi-
tions:
L ij kl = L j i k l = L i j lk,
(3.79)
M i j kl = M j i kl = M ij lk .

3.7.3 Linear Viscoelasticity of Rate Type

From the functional T(U t) given in (3.73) - by simply replacing U t with F t -

the const it ut ive equ ation for the stress tensor, T(t) = T(F t ) , of finite linear
viscoelas ti city ca n also be expressed as

T(t ) = T(F(t)) + 1 00

M(s ;F(t))[F(t - s)]ds,

or
T(t) = T(F(t)) + [too M(t - s ; F(t))[F( s)]ds. (3.80)
94 3. Basic Principl es of Constitut ive Theories

This is a const it ut ive equat ion of int egral type with stress relaxation fun ction
M(s , U) and the influence function h(s) of fading memory. Both are material
functions .
In this section, we shall consider a material class char act eri zed by t he
following fun ctions,

M(s , U) = M(U)h(s) , (3.81)

where T is called the relaxation time of the material. On e can easily check
that t he condit ions (3.75) ar e sa tisfied.
Taking t he time derivative of (3.80) , we obtain t he following equ ation,

T(t) = (o/t(F) + M(F)) [p] - ~ [too M(t - s;F(t))[P(s)] ds ,

where we have neglected the second-order terms in P. By the use of (3.80)

again, it becomes

T - L'(F)[P] = -~ (T - T(F)) , (3.82)

where L'(F) = oFT(F) + M(F) is called the dynamic elast icity t ensor .
This is a different kind of constitutive equat ion that contains t he stress rate.
The history dependence of the mat erial response becomes implicit in the
solut ion of the first-order differential equat ion for the stress. The material
characte rized by this equat ion is called a linear viscoelastic material of rate
type. In the one-dimensional case, such a material model has been proposed
in [28] .
For small deformations, from t he pr evious section, (3.82) reduces to

(3.83)

where the dyn ami c elast icity tensor is given by L ' = L + M(O) . It has the
sam e symmetry properties as the elast icity t ensor L in (3.79). The dyn amic
elasticity tensor cha racte rizes the instan taneous response of the mater ial sub-
ject t o a sudden change of strain . In the one-dimensional case, M(O) is usually
assumed t o b e positive, and hence L' > L .

3 .7.4 Remark on Objectivity of Linear Elasticity

We rem ark that alt hough the constitutive equa t ion (3.68) satisfies the prin-
ciple of mater ial obj ectivity, the const it ut ive equation of linear elasti city de-
rived from it do es not satisfy this principle becaus e, in such a theor y, arbitrary
chan ge of fram e is not allowed if the deformation gradient s ar e to b e small
before and afte r the change of fram e.
 3.7 Fading Memory 95

Let H = F - 1 and H * = F* -1 be t he displacement gradi ents in the two

frames differed by an orthogonal t ransformat ion Q , and we have F * = QF ,
then
H * = Q(H + 1) - 1 .
If we take Q as the rotation of 1800 around the el axis,

(3.84)

we have
H * = QH - 2(e 2 ® e2 + e 3 ® e3),
in which the second t erm on the right-hand side is no longer a small quantity.
The displ acem ent gradient H * will rem ain of 0(1) only if Q is an infinitesim al
rotation, i.e., Q = 1 + R for R = -RT = 0(1).
For linear elasticity, the stress dep ends linearly on t he infinitesimal st rain
tensor,
T(F) = L[E],
where
~ 1 1
E = "2((F + 1) + (F T + 1)) = "2(F + F T) - 1.
The mat erial obj ect ivity requ ires t hat

T(QF) = Q T(F)QT ,
for any Q E OW) and FE Inv(V) . In particular, for F = 1, it requires that

T(Q) = 0,
becaus e T (1 ) = L[O] = O. However , if we t ake Q as the rotation given in
(3.84) , t hen
T(Q) = L [~(Q + QT) - 1] = L[Q - 1J
= -2L[e2 ® e2 + e3 ® e3] -=I O.
Hence the condit ion of mat erial obj ectivity is not satisfied .
Therefore the linear elasticity can only be regarded as an approximat ion
of a phy sically meaningful theory, and it should not be adopted as a physical
model as far as large deformations ar e concerned .
4. Representation of Constitutive Functions

4.1 Materials of Grade n

We have seen that the constitutive functions have to satisfy both the condi-
tion of material objecivity an d t he condition of material symmetry. In partic-
ular , with respect to an undist ort ed configuration, t he constitutive functions
of an isotropic material are isotropic fun ctions. The main difficulty in finding
representations for constitutive functions lies in the fact that t hey generally
depend on the past values of thermomechanical histories. Fortunately, in most
practical problems, long-range memory effect s are mostly irre levant. In ot her
words , for most materials only short memories ar e of practical interest , in
the sense that we can assume that the hist ory be approximated by a Tay lor
series expansion ,

(4.1)

and conseq uently, t he depend ence of t he const it ut ive funct ions on the his-
to ry ht(s) red uces to the dependence on the derivat ives of ht(s) up to t he
n-t h-order at the pr esent time, s = o. A material body wit h such an infinites-
imal memory will be called a material of grade n relative to t he constit utive
var iable h .
Since we have

Ctt (s ) >:::; Ct(O)

t
oq(s) I s + · ··+-
+--- 1 onq(s) I s n
as 8=0 n ! osn 8=0

where A i is t he Rivlin-Ericksen t ensor defined in (1.53) , for a fluid, wit h in-

finitesi mal memory of grade n an d grade 0 relative to mechanical and t hermal
hist ories, respect ively, from (3.54) , we can write t he constitutive relation in
the form

I.-S. Liu, Continuum Mechanics

© Springer-Verlag Berlin Heidelberg 2002
98 4. Representation of Constitutive Functions

(4.2)
In particular, for n = 1, we have
C = F(p , B,g, D) , (4.3)

where we have used the relation Al = 2D from (1.54) . We call this material
a viscous heat- conducting fluid .
For an isot ropic solid of grade 0, from (3.62), we have

(4.4)

where K is an undistort ed configur at ion. We call this material an isotropi c

thermo elastic solid.
Mor eover , the above functions are all isotropic fun ctions, nam ely

T(B ,Qg,QBQT) = QT(B,g,B)QT,

Q(B, Qg, QBQT) = Q Q(B,g , B) , 't/Q E OW) , (4.5)
E(B, Qg , QBQT) = E(B, g , B),

for isotropic t her moelast ic solids, and similar relations for viscous heat-
conduct ing fluids . Functions that satisfy su ch relations are called tensor,
vector, and scalar isotropic functions , respectively.
The condit ions (4.5) impose restrictions on the generality of the consti-
tutive functions on their argument s. Gen eral solutions in the form of rep-
resent ation for isotropic functions are well known and will be given in the
following sections.

4.2 Isotropic Functions

We shall give some represent ations of isot ropic functions in t his section. Let
¢ , hand S be scalar-, vector- and t ensor-valued functions defined on JR x
V x £(V) , resp ectively.

Definition. We say that ¢, h, and S are scalar-, vector-, and tensor- valued
isotropic fun ctions , resp ectively, if for any s E JR , v E V , A E £(V) , they
satisfy t he following condit ions:

¢ (s, Qv, QAQT) = ¢ (s, v , A) ,

h( s, Qv , QAQT) = Q h(s, v , A) , 't/Q E OW) . (4.6)
S( s,Qv ,QAQT) = QS(s,V ,A)QT ,
 4.2 Isotrop ic Functions 99

Isotropic fun ctions are also called isotropic invariants. The definition can
easily be exte nded to a ny number of sca lar, vecto r, and te nsor vari abl es.
From the definition, obviously the condit ions (4.6) impose no restrictions
on the scalar vari abl es of an isotropic fun ction, and hence sca lar vari ables are
irr elevant as far as t he represent ations of isotropic invariants ar e concerned .
Some general solut ions of t he condit ion (4.6) ar e given in the following rep-
resent ation theorems.

Theorem 4 .2.1. Let 1>, h, and S be isotropic scalar-, vector- , and symmetric
tensor- valued functions of a vector variable v , respectively, th en it is necessary
and suffi cient th at they have th e following represent ations ,

1) 1>(v ) = f(v · v) ,
2) h(v) = h(v · v)v , (4.7)
3) S(v)= so(v·v)1+s 1(v ·v)v @v,

where I, h, So, and Sl are arbitrary scalar fun ctions.

Proof. The sufficiency can easily be checked. We shall only show that the
above representations ar e necessary.
To prove (1) it is sufficient to show that if 1> is a scalar isotropic fun ction,
then 1>(v) = 1>(u) whenever

v ·v = u · u.
In fact , since two vectors of the same length can always be brought int o each
ot her by a rot ation, there exists an orthogon al t ensor Q such that u = Qv .
Therefore, by the cond it ion of isotropy,

1>(u ) = 1>(Qv ) = 1>(v ),

which proves (1).

For the vector isotropic fun ction h( v) , if v = 0, then we must have

h(O) = Qh(O) , VQ E OW) ,

which implies t hat h(O) = 0 and hence (2) is satisfied . Now, suppose that
v-I- 0, then we can write t he vector h as
h(v) = o:(v)v + f3(v )u , (4.8)

where u is som e vect or or thogonal t o v . Since h is isot ropi c we have

h(Qv) = Qh(v) , VQ E OW) . (4.9)

If we consider a rotation Q of 1800 about the vect or v , clearly, Qv = v

and Qu = -u , then it follows from (4.8) and (4.9) t hat the second t erm in
100 4. Rep resen t ation of Consti tutive Functions

(4.8) must vani sh . Moreover , from (4.9) the fun ction a:(v) must be a scalar
isotropic fun cti on , therefore, by ite m (1) it mu st reduce to a funct ion of v . v .
The representation (2) is proved .
Finally, for the tensor isotropic function S (v ), we consider a vector-
valu ed fun ction g( v) defined by

g(v) = S(v)v.
On e ca n eas ily check t ha t g( v) is an isotropic vector fun cti on , therefore, by
representation (2) we have

S(v)v = g(v) = a: (v )v .
In ot her words, v is an eigenvect or of t he tensor S. Since S is a symmetric
t ensor , by the spec tral t heorem, there exist two unit vecto rs u and w such
that S ca n be represented as

S(v) = a: (v )v ®v + j3(v )u ®u + l' (v)w ®w, (4.10)

where {v , u , w} is an orthonormal set of eigenvect ors of S . Moreover , since

S (Q v ) = QS(v)QT , VQ E O(V) , (4.11)

if we choose an orthogon al t ensor Q, such that

Qv=v , Qu=w , Qw=u ,

then (4.10) and (4.11) imply that j3(v ) = l' (v ). Therefore, we ca n rewrite
(4.10) as
S(v) = so(v )l + Sl(V)V ® v .
Mor eover , (4.11) implies t hat So an d S l mu st be scalar isotropic fun ct ions .
This completes t he proof of the theorem . 0

Theorem 4 .2.2 (Ri vlin and Ericksen) . Let ¢, h , and S be isotropic scalar-,
vector- , and sym me tric tensor-valued functions ofa symmetric tensor variable
A , respectively, then it is necessary and sufficient that they have the following
represent ations,

1) = f(a1 , a2, a3),

¢ (A )
2) h(A) = 0, (4.12)
3) S (A ) = sol + Sl A + s2A2,

where I, as well as 8 0, Sl, and 8 2, are arbit rary scalar functions of th e three
eigenvalues {aI , a2, a3} of A.
 4.2 Isotropic Functions 101

Proof Again, t he proof of sufficiency is trivial , and we shall only prove the
necessity of the above representations.
First of all, for the vector isotropic fun ction h ,

h(QAQT) = Qh(A) , VQ E OW).

By t ak ing Q = - 1, it follows t hat h(A) = -h(A) , t herefore, it must vanish .
For the scalar isotropi c fun ct ion ¢ (A ), it suffices to show that ¢ (A ) =
¢ (B ), whenever A and B have the same set of eigenvalues.

Thus, let A and B be two symmet ric tensors and ass ume that their eigen-
valu es are t he same. Then , by t he spectral theorem , there exist orthonormal
bases {e i} and {d i} such that

Let Q be t he orthogonal tensor carrying t he basis {d i } into the basis {e.} ,

Qd i = e. , Then , since

Q(d i ® di)QT = (Qd i) ® (Qd i) = ei ® e. ,

it follows that QBQT = A . But since ¢ is isotropic,

¢ (A ) = ¢ (QB QT ) = ¢(B ),
which proves (1).
To show the representation (3) for the symmetric t ensor isotropic fun c-
tion S, we need the following lemma.
Lernrna, Let S(A) be an isotropic te nsor-valued fun ction of a symmetric
tensor A , then every eigenvector of A is an eigen vector of S(A) .
To prove the lemma, let [e, e2, e3} be a principal orthonormal basis of
A . We consi der a rot ation Q of 1800 about the el-axis, therefore,

Qe l = el, Qe2 = -e2 , Qe3 = -e3·

Clearly, we have QAQT = A and since S(A) is isotropic, it follows that

QS(A) = S(QAQT)Q = S(A)Q .

Therefo re, we have

QS(A) el = S(A)Qel = S(A)el ,

i.e., S(A)el remain s un changed under Q . By our choice of Q , this can happen
only if S(A)el is in the direct ion of el . In other words, el is an eigenvect or
of S(A) . For other eigenvect ors, we ca n use similar ar guments. Hence the
lemma is proved . 0
102 4. Representation of Constitutive Functions

From t he lemma, if we express A as

(4.13)

then S(A) can be expressed as

(4.14)

where bl , bz , and b3 ar e functions of A .

Suppose that the t hree eigenvalues aI , az, and a3 are distinct . We con-
sid er t he following simultaneous equat ions for 80 , 8 1 , and 8 Z,

80 + a18 l + ai 8 z = bi ,
8 0 + aZ8l + a~ 8Z = bz ,_

80 + a38l + a58z = b3·

Since the determinant of the coefficient matrix do es not vanish,

we ca n solve for 8 0 , 8 1, and 8Z , which are fun ctions of A of cour se. Then , from
(4.13) , (4.14) can be writt en in the form

S(A) = 80(A)1 + 8 1 (A)A + 8z(A )A z . (4.15)

Furthermore, since S(A) is isotropic,

'rIQ E OW),
it follows from (4.15) that

80(QAQT)1 + 8l (QAQT )A + 8z(QAQT )A Z = 80(A )1 + 81 (A) A + 8z(A)A z .

On the other hand, one can show that {l, A , A Z } is linearl y independent ,
therefore, we conclude that s, are all scalar isotropic functions,

Henc e, the representation (3) for the case of distinct eigenvalues of A is

proved .
With similar arguments one can show that when A has exac tly two
distinct eigenvalues, S(A) admits the representation (3) with 8Z = O. For the
case when A = al , every vector is an eigenvect or, and hence by the lemma,
we must have S(A) = 8 0 1 , which is a special case of the representation (3) .
D
 4.2 Isotropic Functions 103

Recall that the Cayley-Hamilton t heore m states t hat for any te nsor A ,

(4.16)

where I A, II A, and III A are called the prin cipal invariant of A . They ar e the
coefficients of the char acteristic polynomial of A , i.e.,

(4.17)

Since eigenvalues of A ar e the roots of the characterist ic equa t ion (4.17) ,

if A is symmet ric and a1 , a2 , and a3 are t hree eigenvalues of A , distinct or
not , t hen it follows that

I A = a 1 + a2 + a3,
II A = a1a2 + a2a3 + a3a 1, (4.18)
IlIA = a1a2a3.
They can be written in terms of another three invariants of A , namely,

IA = tr A,
IIA = H(tr A)2 - tr A 2), (4.19)
IlIA = i-((tr A) 3 - 3 tr A tr A 2
+ 2 tr A3) ,
or conversely,

tr A = lA ,
2
tr A = I~ - 2IIA, (4.20)
3
t r A = I~ - 3IAIIA + 3IIIA .
From the above relations we can see that the three sets {a1 ' a2 , a3},
{IA,IIA,IIIA} , and {t rA ,trA 2 ,trA 3} are equivalent in the sense t hat they
uniquely det ermine one anot her . Moreover , they are all invariant under any
chan ge from A to QAQT for any orthogonal tensor Q . Therefore, from The-
orem 4.2.2, any scalar isotropic function ¢ of A may also be expressed as :

The proof of t he following corollary is straightforward.

Corollary 4 .2.3 Let S be an isotropi c symmetric tensor-valued function of

a symmetric tensor variable A . If, in addition, the function S is linear in th e
variable A , then it has the following represent ation:

S(A) = (}(trA)l + j3A, (4.21)

where (} and j3 are independent of A .

104 4. Representation of Const it utive Functions

As we have seen from the above represent ation t heorems, an isotropic

fun ction is quite rest ricte d in its dependence on the ind ep end ent vari abl es.
From a different point of view, we ca n also derive some such restrictions in
the form of differential equat ions. First, we need the following lemma:
Lemma 4.2.4 Let F(Q) be a function [sceler-, vector-, or tensor-valued)
defined on £(V) and suppose that F(Q) = 0 for any orthogonal Q E O(V) .
Th en the gradient of F a t the identi ty tensor is symmetric, i.e.,
oQF(1 )[W] = 0 (4.22)
for any skew-symmetric W E Skw(V).

Proof For W = -WT and 0 < E « 1, since (1 + EW)(1 + EW)T = l - E2W2 ,

t he t ensor 1 + EW is orthogonal to within second-order t erms 0(2) in E. In
ot her word s, there exists an Qw E O(V) such t hat
1 + EW = Qw + 0(2).
By t he ass umption we have F(Qw) = 0 and F( 1) = O. Consequent ly, by the
definition of the gradient (A.46) ,
F(1 + EW ) -
F(1) = EoQF (1 )[W] + 0(2),
= F( Qw + 0(2)) - F(Qw) = oQF( Qw )[0(2)] + 0(2) = 0(2),
which implies t hat t he first-o rd er t erm EoQF(1) [W] must vanish, and (4.22)
is proved. 0

Example 4.2.1 Let ¢i(A, v) be a sca lar isot ropic function. T hen the
fun ction ¢i satisfies the following relat ion ,

Eirnn ( :¢i V n + !:lA o¢i Apn) = 0,

o¢i A np + !:lA (4.23)
u V rn U rnp U pm

relative to a Cartes ian coordinate syst em.

Proof Let
F(Q) = ¢i(QAQT, Qv) - ¢i(A,v) If Q E £(V) .
By the chai n rul e, one can easily verify that
oQF (1 )[W ] = 0A<P[(1 + W)A(1 + W) T - A] + ov¢i[(1 + W)v - v]
T]
= oA¢i [W A + AW + ov¢i[Wv]

= (oA1»AT[W] + (oA ¢if A[W] + ov¢i (9 v[W] .

Since 1>(A, v) is an isotropic fun cti on , we have F(Q) = 0 for any Q E
O(V) , and hence, by t he above lemma
oQF (1 ) = (oA¢i) AT + (oA ¢ifA + ov1> (9 v
mus t be sym metric. T herefore its skew-symmetric part mu st vani sh ,
wh ich, in component form s, gives the relation (4.23) . 0
 4.2 Isotropic Functions 105

From Lemma 4.2.4 we can prove t he following representation theorem for a

sca lar isotropic fun ction of vector variables as the solut ion of a system of
first-order partial differential equa t ions.
Theorem 4.2.5. Let ¢ (v , u ) be a scalar isotropi c function of two vector
variables, then it has th e following representation ,
¢ (v , u ) =f(v ,v,u ,u,v,u) .

Proof By the use of Lemma 4.2.4, if we define

F(Q) = ¢ (Qv, Qu) - ¢ (v, u ) 't/Q E L:(V),
then , similar to t he pr eviou s example, we have
8¢ 8¢ 8¢ 8¢
- V2 - - VI + - U2 - -Ul = 0,
8 Vl 8V2 8Ul 8U2
8¢ 8¢ 8¢ 8¢
- V3 - - VI + - U3 - - Ul = 0, (4.24)
8 Vl 8 V3 8Ul 8U3
8¢ 8¢ 8¢ 8¢
- V3 - - V2 + - U3 - -U2 = O.
8 V2 8 V3 8U2 8U3

From the first-order partial differential equat ion (4.24h for ¢ with V3 and U3
as paramet ers, one ca n eas ily obtain the general solut ion (e.g., see [29])

¢= f (ci , C2 , C3 ),
where f is an arbit rary fun ction, and
Cl = vi + v~ + gl (V 3 ' U 3) ,
C2 = ui + u~ + g2( V3 , U3 ),
C3 = Vl U I + V2 U2 + 9 3(V3 , U3 ),
are the t hree int egrals of (4.24h . Substituting into the second equa t ion
(4.24h, one obtains

I 3 - V I 8g 1 _ Ul 8 9 1) 8f
( 2V V
+ (2 U I U3 _ Ul 8 g 2 _ V I 8 9 2) 8f
8 V3 8U3 8 Cl 8 U3 8 V3 8 C2

+ ( V I U 3 + V3 U l 3
- V I 8g _ Ul 8
9 3)
8f = O.
8 V3 8 U3 &3
Since f is arbit rary, so are 8 f / 8 Ci , and hence their coefficients in the above
equation must vani sh inde pende nt ly,

V I ( 2V3 - 8
9 1)
= Ul 8g 1 ,
8V3 8 U3

VI 8 g
2
= Ul ( 2 U 3 _ 8
9 2)
,
8 V3 8 U3
93
V I (U 3 - 8 ) = Ul (89 3 - V3 ) '
8 V3 8U3
106 4. Representation of Constitu tive Functions

Note that in these equat ions, the left-hand sides do not dep end on Ul , while
t he right-hand sides do not dep end on VI, t herefore, both sides mu st vani sh .
From whi ch one can easily show that

Therefore, it follows that

which also satisfies (4.24h and the t heorem is proved . 0

This representati on theorem can be extended to isot ropic functions of
an arbit rary number of vector var iables.
Before we consider general representations for sca lar, vector, and t ensor
isot rop ic fun ctions in the next sect ions , we shall give an example to illustrat e
how to obt ain a representat ion for higher-order te nsor isot ropic functions .

Example 4.2.2 Let Vi jkl be a fourth-order isotropic tensor , i.e.,

for any ort hogonal t ensor Q relat ive to a Cartesian coordinate syste m
(see Ex ercise 3.2.2), wit h the addit ional symmet ry prop erty:

Vij kl = VJikl = Vijlk .

T hen it has the following representation:

(4.25)

Proof For any symmetric A, let

Then it is easy to ver ify that S = S(A) is an isotropic symmet ric tensor-
valued fun ction and is linear in it s symmetric te nsor variable A . There-
fore, from Coroll ar y 4.2.3, we can represent S(A) as an expression linear
in A in the following form ,

S(A) = A(tr A)l + 2flA,

where A and fl are two scalar constants. In component form , it gives

Sij = A8ij A kk + 2flAij

= A 8i j8kl A kl + fl (8i k8j l + 8il8j k) A kl ,
which proves (4.25) by comparison. 0
 4.2 Isotropic Functions 107

Exercise 4.2.1 Verify the relations (4.19) and (4.20).

Exercise 4.2.2 Suppose that M(v) is a third-order t ensor isotropic

function of a vector vari abl e, i.e., for any Q E O(V) ,

relativ e to a Cartesian coordinate system. Find a representation formul a

for M . (Note that if S(v ,u) is defined by Sij = Mi jk Uk, then S is a
second-order te nsor isotropic function of two vector var iabl es).

Exercise 4.2.3 Suppose that A E Sym(V) and (1 + A) E Inv(V) .

1) Let S(A) = (1 + A) -i . Show t hat S(A) is an isotropic function.
2) By t he use of representat ion (4.12) , show that (see also Exer-
cise A.1.9)

where
S2 = (1 + I A + II A + III A) -l ,
Sl = -(1 + I A )S2'
So = (1 + IA + II A) s2.

4.2.1 Isotropic Elastic Materials and Linear Elasticity

For an isotropic material with no memory of past deformation histories, from

the reduced constitut ive rela tion (3.62) , we have the st ress t ensor T = T(B) ,
where T is an isotropic function of t he left Cauchy-Green tensor B. By
represent ation (4.12) , it follows that

T=tol +tlB+t2B2 ,
(4.26)
ti = ti(IB, II B, III B) , i = 0,1 ,2,

or equivalently, by the use of the identity (4.16)

T = so l + Si B + s _ l B - l,
(4.27)
s , = Si(IB ,11B ,1I1B ), i = 0, 1,2,

where
So = to - II Bt2,
Sl = tl + I Bt2 , (4.28)
8- 1 = III Bt 2.
This is the general constitutive equation for an isotropi c elastic material body
with finit e deformations.
108 4. Representation of Constitutive Functions

For small deformations, by assuming that the displ acement gradient H =

F - 1 is a small qu an tity of 0(1), we have

B-1 = 2E + 0(2) ,
where E = (H + H T) /2 is the infinitesim al strain te nsor. A linear approxi-
mation of 7(B) at B = 1 can be written as

T = 7(B) = 7(1) + oB7(1)[B - 1] + 0(2) .

With t he addit ional assumpt ion that the reference configurat ion is a natural
state, 7( 1) = 0, and by neglecting the second-order t erms, it gives

T = L[ E],

where L is the elasticity tensor (see (3.79)) given by

Using the representation (4.26) and ca rrying out the gradient explicit ly by
the use of (A.54) , we obtain

(4.29)

where >. and J-l are called the Lame elastic moduli and are relat ed t o the
material parameters to , t 1, and t 2 of (4.26) by the following relations:

>. = 2 (01at + 2 OIlat

B B
ot)1
+ 0111
B (3 ,3 ,1) (4.30)
J-l = t 1(3, 3,1) + 2 t 2(3, 3,1) .
Equati on (4.29) is the consti t ut ive relation of the classical theory of isotropic
lin ear elastici ty, also known as Hooke 's law. The two material constant s A
and J-l can be det ermined in simple experime nt s, such as uni axial te nsion and
simple sh ear. The Lam e const ant J-l is also known as t he shear modulus.
Two more commonly used mater ial constants in linear elasticity, Young 's
modulu s and Poisson's ratio are usually introduced in uni ax ial t ension tests:
The state of st ress in such a tes t is given by

and hence t he linear st rain t ensor has the following form :

 4.2 Isotropic Funct ions 109

Young 's modulus E and Po isson 's rat io v are defined as

_ al EZ
E- , v=-- ,
EI EI

i.e., E is the stress per unit strain in t he longitudinal dir ection of the uni axi al
t est , while v is the ratio of the transverse compression to the longitudinal
exte nsion of the st rain. It follows imm ediately from (4.29) that

E _ fl(3A + 2fl)
(4.31)
- A+/l. '

and conversely,

A- vE E
(4.32)
- (1-2v)(I+v) ' fl= 2(1 + v)'
Hooke's law can also be written in a different form,

- u 1+v
E = - - (tr T) l + - - T (4.33)
E E '
expressing the linear strain in t erms of the stress.

Exercise 4 .2 .4 Verify the relat ions (4.31) , (4.32) and (4.33) .

Exercise 4.2.5 From Hooke's law (4.29) and Cauchy 's first law (2.57) ,
deri ve the following equa t ion of linear elast icity,

pu = (A + fl) grad (div u ) + fl '\7zu + pb ,

where u is the displacement vecto r.

4.2.2 Reiner-Rivlin Fluids and Navier-Stokes Fluids

A simple fluid of grade 1 is defined by t he const it ut ive equation

T = T(p ,D) .
Since it is an isotropic function, by the representation theorem (4.12) it can
be exp ressed as

T = a oi + a ID + az D z,
(4.34)
a i = a i (p, I D,II D , III D), i = 0, 1,2.

A fluid characterized by t his const it ut ive equat ion is called a Rein er-Riolisi
fl uid. This is the most general form of constit uti ve equat ions for simple fluids
110 4. Representation of Constitutive Functions

of grade 1. Even though this seems to be a general model for nonlinear vis-
cosity, it has been pointed out that this model is inadequate to describe some
obs erved nonlinear effects in real fluids (See Sect. 119, [71D. Nevertheless,
by the use of (4.21), the special case , for which the viscous stress depends
linearly on the stretching tensor D , leads to the most well-known model , the
Navier-Stokes fluid. This is given by the following constitutive equation,

T = -p(p) 1 + A(p)(tr D)l + 2JL(p)D , (4.35)

where p is the pressure, while A and JL ar e called the coefficients of viscosity.

A Navier-Stokes fluid is also known as a Newtonian fluid in fluid mechanics.
Note that from Exercise 3.2.2, and the representation (4.25) in Example 4.2.2,
also proves (4.35) .
We can rewrite (4.35) in a different form ,
2
= -pl + (A + "3 JL )(tr D)l + 2/l.D,
A

T (4.36)

where
1
D = D - "3 (tr D) 1 ,
A

called the deviatoric part of D, is traceless, i.e., tr fJ = O. The parameters JL

and (A + 2JL/3) are also known as the shear and the bulk viscosities, respec-
tively. It is usually assumed that

2
A+ "3 JL 2: 0. (4.37)

These inequalities ensure that that the fluid particles t end to flow in the
direction of a shear force (for the proof see Sect . 7.2.2). When the bulk vis-
cosity vanishes identically, it is known as a Stokes fluid, a model adequate to
describe some real fluids and frequently used in numerical calculations,

(4.38)

The devi atoric part fJ in component form is given by

D ')
A

-
~(avi
£:l
+ aVj )
£:l
_ ~ aVk
£:l
5:..
U,).
2 U Xj U Xi 3 UXk

A Navier-Stokes fluid is governed by the system of equations that con-

sists of the conservation of mass (2.32) ,

(4.39)
 4.2 Isotropic Functions 111

and the equat ion of motion (2.57) ,

aVi
(-+Vk- aVi) ap a (' aVk)
p
at aXk + -aXi - -Bx, "'-
aXk (4.40)
a ( aVk) a ( aVi)
- aXk J1 aXi - aXk 11aXk = pb. .
The last equat ion (4.40) is known as the Navi er-Stokes equation. The pres-
sure p and the viscosities, A and J1 , are, in general, functions of the density
p. Equations (4.39) and (4.40) form a system for the fields (p(x , t) , v(x , t)) .
For Stokes fluids, the governing equat ions ar e obtained from above by sub-
st it ut ion of A = -2J1/3 .
Note that aside from the similarity between the constitutive equ ations
of linear elast icity (4.29) and the Navier-Stokes fluids (4.35) , there is a fun-
damental differen ce between the two theories. Unlike linear elast icity, which
is physically meaningless for finit e deformations because it does not sat isfy
the requirement of the principle of material obj ectivity (see the rem ark on
Sect . 3.7.4), the Navier-Stokes fluids do sa tisfy such a requirement. Therefore
the lineariz ation of the const it ut ive relation in (4.35) need not be regarded as
an approximat ion of the more genera l const it ut ive relation (4.34) . Any par-
t icular form of a const it ut ive relation (4.34) cha racterizes a particular class of
simple fluids. Thus, it is conceivable that there are some fluids tha t obey the
constit ut ive equat ion (4.35) for arbitrary rate of deformation. Indeed, wat er
and air are usu ally treated as Navier-Stokes fluids (with addit ional thermal
effect s) in most practical applicat ions with very sat isfact ory results, even in
rapid deform ation pro cesses.

4.2.3 Elastic Fluids

The simplest constituti ve equa t ion used in cont inuum mechani cs is that of a
fluid that is inviscid , i.e., non-viscous , and a non- conductor of heat ,
T = -p(p , B) 1 , q =0. (4.41 )
This fluid is also an elasti c material, i.e., its const it ut ive function dep ends on
the deformation gradient (through det F) only. Therefore, it is also called an
elastic flu id.
For an elastic fluid , from the bal an ce equa t ions of mass, linear momen-
tum, and energy, we have the following governing equat ions in component
form s:

(4.42)
112 4. Representation of Constitutive Funct ions

The int ernal energy E = ei p, ()) and t he pressure p = p(p , ()) have to be
sp ecified , for instance by the ideal gas laws, and t he equations becom e a
hyp erbolic system for t he fields (p(x , t) , v(x , t) , ()(x, t)) , usually known as
the Euler equat ions for compressible flows of ideal gases.

4.3 Representation of Isotropic Functions

In t his section, we sh all give a mor e general discussion on isotropic invaria nts,
and rest ate the well-known resul t s for the representation of isotropic fun ctions
of any number of vecto r and tensor vari abl es.
Let £(V) be the sp ace of second-order t ensors on V . Let

'D = V m X £(V) n = V x . .. x V x £(V) x . .. x £(V) , (4.43)

~, v '
m times n t imes

and
¢ : V -+ IR,
h :V-+V, (4.44)
S : V -+ £(V).
Definition. We say that ¢, h , and S are scalar, vector, and tensor invariants
relative to t he group Q ~ O(V) , resp ectively, if for any v E V'" , A E £(V) n,
we have
¢(Qv, QAQ T) = ¢(v , A) ,
h(Qv, QAQT) = ou» ,A), VQ E Q, (4.45)
S(QV,QAQT) = QS(V,A)QT .
In t he above definition, we have introduced the following abbreviations:

v= (VI , · · · ,V m ) ,
(4.46)
A = (A I , · · · , A n) ,
and
Qv= (QVI , ·· · , Qv m ),
QAQT = (QAIQT, .. . , QAnQT) ,
for E V and A j E £(V) .
Vi
If the group Q = O(V) , the invariant s are usually called isotropic invari-
ants or isotropic ju nctions, ot herwise, t hey are called anisotropic invariants.
 4.3 Representation of Isotropic Functions 113

Example 4.3.1 The following functions are

1) isotropic scalar invari ants: v· u , det A , tr(Am B n) , Am v · B' t u ;
2) isotropic vect or invari ants: Amv , Amtr» ,
3) isotropic t ensor invariants: Am , Amv 0 Br» ,
for any u , v E V and A , BE .c(V) . 0

Example 4.3.2 The vector product v x u is not an isotropic vector

invariant, bu t it is a vector invariant relati ve to the proper orthogon al
group O+(V) . To see this, we have

QT(QV X Qu) · w = Qv x Qu · Qw = (detQ)(v x u) · w ,

for any w E V . Hence we obtain

(Qv x Qu) = (detQ)Q(v x u ).

Since det Q = ±1 for Q E O(V) and det Q = 1 for Q E O+(V) , therefore,

it is not an isotropic vector invari ant , but it is an anisot ropic vector
invar ian t relat ive to the proper orthogon al gro up O+(V) .
The same is t rue for the vector fun ction h(W) = (W) , where (W)
denotes the ax ial vector associate d with the skew-symmetric t ensor W
(see (A.29)) , since we can also show

(QW QT) = (det Q) Q(W )

easily. 0

Suppose that Y s , Yv, and Y t are sets of scalar, vector, and te nsor invari ants,
resp ectively, then it is obvious that t he following functions ar e isotropic:

cP = cP(Y s ),

h= L cPa (Ys) u a,
(4.47)
u a EYv

S= L cPb(Y s ) n,
nEY,

where cP, cPa' and cPb are arbit rary functions of the sca lar invari ants in Y s'
The purpose of the representation problems is t o find the set s of invari-
ants Y s , Y v , and Y t , so that isotropic functions can be expressed in the above
forms (4.47) . We call T; t he set of basic invariants, and say t hat Tv and Y t
are the generating sets for isotropic vector, and t ensor fun ctions, resp ect ively.
A set of basic invari ants or a generating set is called a functi onal basis if
it is irreduci ble. By irr edu cible we mean t hat elements of the basic invar iant s
are not fun ctionally relate d , and elements of the generating set are linearly
114 4. Representation of Constit utive Function s

indepe ndent with resp ect to isotropic functions, i.e., no elements can be ex-
pressed as a linear combination of other elements with coefficients of sca lar
isot ropi c functi ons.
Fun ct ional bases for isotropic functions have been ext ensively st udied in
the literature (see [7, 66, 74]). For read y reference, we reproduce the complete
results from [74] for isotropic invari an t s in Tabl e 4.1 t hrough Table 4.4, in
which v stands for vecto r , A for symmetric tensor, W for skew-symmet ric
t enso r.
From these tables one can eas ily construc t functi onal bases Ys , Yv , and
Y t of invari ant s for any given domain V of vari abl es and thus obtain the
representat ion of isot ropi c funct ions from (4.47). The representat ion theorems
4.2.1, 4.2.2, and 4.2.5 proved before are just the simplest cases given by these
t abl es.
In t he following examples we sha ll give t he functional bases Y s , Yv, and
Yt for isotropic functi ons of some different cases of vari abl es.

Example 4 .3 .3 For V = {(V I, ' " ,V m ) E Vm }, we obtain

{VI 'VI ,V I · Vz,· · ·, Vn- I· Vn, Vn · Vn },
{v , · · · ,v n },
{I , V I 0 VI , ' " ,Vn 0 v n, (VI 0 v z + Vz 0 V I ) ,"' ,
(Vn-I 0 Vn + Vn 0 v n-d} ·
The above funct ional bases for scalar, vector, and symmetric tensor in-
vari ants are const ructe d by taking all possible combinat ions of vari ables
according t o t he t abl es. 0

The above result for scalar invari ants of any number of vector vari abl es -
Theorem 4.2.5 is a special on e with two vectors - is a well-know n result due
t o Euler in the classical invariant theor y.

Example 4.3.4 For V = {(v ,A) E V x £(V) ,A = AT} , we obt ain t he

following funct ional bases:
Y s = {trA , trA Z,trA 3 ,v ·v ,v ·Av,v ·A Zv} ,
Y v = {v ,Av,AZv} ,
Y t = {1 ,A,AZ,v 0v ,(Av 0v+v 0Av) ,Av 0 Av }.
for scalar , vector , and symmet ric t enso r invari ant s, resp ectively. 0

If we examine Tabl e 4.1 t hrough Table 4.4, we can find that every sca lar in-
varian t is t he trace of some tensor , e.g., v ·Au = tr(v 0Au) , and every vector
or te nsor invarian t is some vector or t ensor form ed from t he vari abl es intrinsi-
cally. Mor eover , scalar invari ant eleme nt s involvin g mor e than four variables,
an d vect or or t ensor generat or eleme nts involving more t ha n three vari abl es,
 4.3 Representation of Isotropic Functions 115

Table 4. 1. Isotropic scalar invariants

One variable Invaria nt element s

v V 'V
A trA , trA Z, trA 3
W trW Z
Two variables Invariant elements
VI , Vz VI'VZ
v, A v · Av , v · AZv
v, W v· Wzv
AI , A z tr AIA z, tr AIA~, tr AzAi , tr AiA~
WI , W z trWIWz
A, W tr AW z , tr AZW z , tr AWAzW z
Three var iables Invariant elements
VI , Vz, A VI . Avz , VI . AZvz

VI, Vz, W VI' Wvz , VI ' WZvz

V, AI, A z v · AIAzv
V, WI , W z v · WI Wzv , v · W I wiv, v · W zwfv
V , A, W v · WAv , v · WAzv , v · WAWzv
AI , A z , A 3 tr A IA zA 3
WI , W z , W 3 trWI W ZW3
AI, A z, W tr AIAzW, tr AIA~W, tr AzAiw, tr Al WAzW z
A , WI , W z t r AWl W z , tr AWl wi , tr AWzwf
Four variables Invariant elements
VI , Vz, AI , A z VI ' AIAzvz , VI ' AzA lvz
VI , Vz, W I , W z VI ' WIWZVz , VI ' WZWlvz
VI, Vz, A , W VI ' AWV2 , VI ' WAV 2

Table 4 .2 . Isot ropic vector invariants

One variable Generator element s

V V
Aor W o
Two variables Generator elem ents
V, A
v, W
Three variables Generator elements
V, AI, A z AIAzv , AzA I V
V, WI , W z WI Wzv , W zWI V
V , A, W AWv , WAv
116 4. Representation of Constitutive Functions

Table 4.3. Isotropic symmetric tensor invariants

No variable Generator elements

o 1
One variable Generator elements
v
A
W
Two vari ables Generator elem ents
VI, V2 VI @ V2 + V2 @ VI
V, A v @Av+Av @v, Av @Av
v, W v @Wv+Wv @v , Wv @Wv ,
Wv @W 2v+ W 2v @Wv
A IA2 + A 2A I, A IA2A I , A 2A IA 2
W IW2 + W2WI , WI wi - wi WI , wlw2 - w2wl
AW - WA , WAW, A 2W - WA 2, WAW 2 - W 2AW
Three variables Gen erator elements
VI @ AV2 + AV2 @ V I, V2 @ AVI + AVI @V2
VI @W V2 + WV2 @VI, V2 @WVI + WVI @V2

Table 4.4. Isotropic skew-symmetric tensor invariants

One variable Generator elem ents

vorA o
W W
Two variables Generator elements
VI, V2 VI @ V2 - V2 @ VI
V, A V @ Av - Av @ v , V @ A 2v - A 2v @ V,
Av @ A 2v - A 2v @ Av
v @Wv- Wv @v , v @w 2v- w 2v @v
A IA2 - A 2A I , AIA~ - A~AI , AiA 2 - A 2Ai ,
A IA2Ai - AiA2A I , A2AIA~ - A~AIA2
W IW2 - W2WI
2
AW + WA , AW - W 2A
Three variables Generator elements
VI , V2, A VI @ AV2 - AV2 @ VI , V2 @ AVI - AVI @ V2
VI , V2, W VI @ WV2 - WV 2 @ V I, V2 @ WVI - WVI @V2
V, AI , A 2 A lv @A 2v - A 2v @A lv, A IA2v @v - V @A I A 2v,
A 2A lv @v - V @A2 AIV
A IA2A3 - A 3A2A I, A 2A3A I - A IA3A2,
A 3A IA2 - A 2A IA3
 4.3 Representation of Isotropic Functions 117

ar e not present in the t ables. In ot her words , such elements are proved to be
redundant. These observations are very helpful in formin g various invariants
from a given set of vari ables without consult ing the tables. In fact, we can
form infinitely many such invari ants. However , there exist some identities,
such as t he Cayley-Ham ilton theorem (4.16) , which enable us to eliminat e
many of them and thus we are left with a finit e set of invari ants. Of cour se,
such eliminations usin g known identities would not give us an irre ducible set
of invariants, in general. Nevertheless, since general representations are oft en
too complicate d to be useful as far as pr acti cal applicati ons ar e concerned ,
we often need only certain linear or qu adratic representations. Such repre-
sentations ar e relatively easy to work out based on t he above observation,
even in t he absence of those t ables .

Example 4.3.5 Consider variables (v , A).

1) The isotropic fun ctions lineariz ed in A can be written as

rP = rP1 + rP2 tr A + rP3 (v · Av) ,

h = (h 1 + h2 t r A + h3 (v· Av))v + h4Av ,
S = (81 + 82 tr A + 83 (v · Av))l + 84A
+ (85+ 86 trA+ 87 (v ·Av))V Q9V+ 88(V Q9Av+Av Q9v),

where the coefficients rPi , hj , and 8 k are scalar functions of (u v).

2) The isotropic fun ctions lineari zed in both v and A can b e written as

rP = rP1 + rP2 tr A ,
h = lu» , (4.48)
s =(81 + 82 trA)l + 83A ,
where the coefficients rP1 through 83 are ind epend ent of v and A. D

There is an identity t hat is someti mes very useful in reducing som e

complicated invari ants. Let A and B be two symmetric tensors , then they
must satisfy (for a pro of of thi s identity see [62])

ABA + A 2B + BA 2
I
= (tr A 2B - tr A t r AB - "2 t r B (tr A 2 - (tr A) 2)) 1
(4.49)
1
+ (tr AB - tr A t r B)A + "2(tr A 2 - (tr A) 2)B
+ t r A (AB + BA) + (tr B) A 2.

Not e that when B = A , t his identity redu ces t o (4.16) of the Cayl ey-Hamilton
theorem .
118 4. Representation of Constitutive Functions

As an example, let us t ake B =v ® v , then the left-hand side of (4.49)

becomes

Ex amining the terms on the right-hand side, one ca n conclude that the gener-
ator element Av ®Av , for two vari abl es v and A in Tabl e 4.3, can be replaced
by the element (v ® A 2v + A 2v ® v).

Exercise 4.3.1 Con sid er the vari abl es (u , v , A) .

1) F ind t he general represent ations of sca lar, vector, and symmet ric
tensor isotropic functions.
2) Fi nd their represe ntat ions that are linearized in the var iables v and
A.

Exercise 4 .3.2 Show that if h(A , v) is a vector isot ropic fun ction,
then ovh is a t ensor isot ropic function, (1) dir ectly from definit ion ; (2)
by using the represent ation formul a.

Exercise 4.3.3 Derive the explicit relation between Av ® Av and (v ®

A 2v + A 2v ® v) , by the use of (4.49) . Therefore, justify the replacement
of the generator eleme nt Av ® Av for two variabl es v and A by the
eleme nt (v ® A 2v + A 2v ® v) in Tabl e 4.3.

Exercise 4 .3.4 By t he use of (4.49) , show that both (Av ®A 2v+A2v ®

Av) and A 2v ® A 2v are not needed as generator elements for isotropic
sy mmetric tensor invari ants of vari abl es v and A .

4 .3.1 Isotropic Thermoelastic Solids

and Viscous H eat-Conducting Fluids

For an isot ropic thermoelastic solid defined by (4.4), from Exa mple 4.3.4,
we have the following constitut ive equ ations,

T = tol + tlB + t2B2

+ t3g ® 9 + t4(B g ® 9 + 9 ® Bg) + tsBg ® Bg ,
(4.50)
q = kig + k 2Bg + k 3B 2g ,
E: = E: (B, tr B , tr B 2, tr B 3,g . g , g . Bg , g . B 2g) ,
where t he coefficients ti and kj , as well as E:, are sca lar funct ions of seven
vari abl es indi cat ed in t he argume nt of E: .
 4.4 Hemitropic Invariants 119

Simil arl y, one can immediat ely wr it e down t he const it ut ive equat ions for
a viscous heat-conducting fluid defined by (4.3) ,

T = (Yo 1+ (Y I D + (Y2D2
+ (Y3 g ® 9 + (Y4(Dg ® 9 + 9 ® Dg) + (Y5 D g ® Dg ,
(4.51)
q = /31 g + /32 D g + /33 D 2g,
e = ei p, e , tr D , tr D 2, t r D 3,g . g , g . Dg , g . D 2g),

where t he coefficients (Yi and /3j, as well as c, are scalar fun ctions of eight
vari abl es indicated in the arguments of c.
If we linear ize the const it ut ive equations (4.51) of a viscous heat-
conducting fluid in both D and g , from Example 4.3.5, we ob t ain t he following
represent a tions for t he st ress tensor and the heat flux ,

T= (-p+>. t rD)l +2JLD ,

(4.52)
q = - K,g ,

where the coefficient s ar e functions of (p, e). These are the classical Navier-
Stokes theory and Fourier's law, and we shall call it a Naoier-Siokes-Fourier
fluid. The coefficient s A, JL are the viscosities and K, is the thermal conduc-
t ivity.

4.4 Hemitropic Invariants

For thermoelastic solid s, the consti t ut ive relations for the second Piola-
Kir chhoff st ress tensor , the mat erial heat flux, and the internal energy, rela-
t ive to an undistort ed configurat ion K, with symmet ry group 9"" must sa t isfy
t he following sym met ry condit ions , according to (3.49) ,

S",(QcQT ,e,QgJ = QS",(C,e ,g", )QT ,

Q",(QCQT , e , Qg",) = Q Q",(C , e ,g",) , VQ E Q",. (4.53)
[",(Q CQT , e , Qg",) = [",(C, e ,g",) ,

In other words, t he fun ctions S"', Q"" and [ '" are invar iants relative t o the
group Q"" according to the definition (4.45) .
If the group Q", = O(V ), the solid is isotropic and the functions are
isotropic. For anisot ropic solids in general, Q", =1= O(V) . In this and the follow-
ing sect ion we shall consider representation t heorems for invari ants relative
to some subgr oups of t he orthogonal group O(V).
Definition. We call an invariant relative t o the prope r orthogonal group
0 +(V) a hemitropic function.
Since a prop er or t hogonal tran sformation preserves t he orientat ion, one
can always identify a vector with a skew-symmet ric te nsor relative to the
120 4. Representation of Constitutive Funct ions

act ion of any proper orthogonal transformation (see Sect . A.1.6) . Let v be a
vector and W be its associated skew-symmetric tensor, v = (W) . On e can
eas ily verify t hat (see Ex ample 4.3.2), for any Q E O+(V) ,

We have the following representation theorems for hemitropic fun ctions.

Theorem 4.4.1. For any scalar-valued (or tensor-valued) function 'l/J (iJ, A),
define
(4.54)
Then 'l/J (iJ, A) is a scalar (or tensor) hemitropic functi on if and only if ~( W, A)
is a scalar (or tensor) isotropic function.

Theorem 4.4.2. For any vector-valued function h( iJ, A), let H be the skew-
symmetric tensor-valued function , such th at h = (H) and define
H(W,A) = H(iJ, A) , iJ = (W) . (4.55)

Th en h(iJ, A) is a vector hemitropic function if and only if H(W, A) is a

skew-symmetric tensor isotropic function .

In these st at ements we have used t he notation introduced in (4.46) . The

proof of these theorems follows from the simple observation that a scalar or
t ensor hemitropic fun ction of only tensor va riables is also isotropic because
the transformation QAQT is un changed if Q is repl aced by -Q . Based on
t hese t heorems , one ca n obtain representations for any hemitropic fun ctions
by simply replacing vect ors with their associate d skew-symmetric te nsor s.
From Tabl es 4.1, 4.3, and 4.4 one ca n immedi ately obtain irreducible fun c-
tional bases for scalar, te nsor, and vector hemitropic fun ctions .1 They are
given in Tables 4.5, 4.6, and 4.7, resp ectively.

Example 4.4.1 Let V = {(v , A) E V x L(V) , A = AT} and h : V -+ V

be a vector hemitropi c fun ction. Suppose that v = (W), then from Tabl e
4.1 and Tabl e 4.4 for vari abl es Wand A, we have the following irreducible
bases for scalar and skew-symmet ric tensor isotropic invari ants:
T; = {trA , trA 2 , trA3, trW 2 , trAW 2 , trA 2W2 , trAWA 2W 2 } ,
Y t = {W, AW + WA , AW 2 - W 2A} .
Therefore, by Theorem 4.4.2, we can write the rep resentation of h( v , A)
as
(4.56)

1 Integrity bases for hemitropic invariant s can be found in [19].

 4.4 Hemitropic Invariants 121

Table 4.5. Hemi tropic scala r invari ants

On e variable Invariant elements

A t r A, trA 2, trA 3
v V,V
Two vari ables Invariant elements
trA IA2, trAIA ~ , t r A 2A i , trAiA~
VI , V2 VI 'V2
A, V V . Av, V . A 2v , v · Av X A 2v
Three variables Invari ant eleme nt s

VI , V2, V3 VI'V2 X V3
AI , A 2, V V · (A I A2) , V · (A I A~) , v · (A 2Ai), v · Alv X A 2v
A , V I , V2 V I ' AV2, V I ' V2 X AVI , V2 ' VI X AV2

Table 4 .6. Hemitropic vector invariants

On e variable Generator element s

V V
A o
Two va riables Gener ator element s
(A I A2) , ( A I A~) , (A 2Ai), (A I A 2Ai), ( A2 A I A~)
VI x V2
Av , V x Av
T hree var iables Gen erator element s

Table 4.7. Hemitropic symmetric tensor invar iants

No vari abl e Gener ator element s

o 1
One variable Gen erator elements
A
V

Two va riables Generator elements

A IA 2 + A 2A I , A IA2A I , A 2A IA2
VI 0 V2 + V2 0 VI ,
v l 0(vl X V2) + (VI xV2) 0v I ,
v 2 0(VI XV2)+(VI x V2)0 v 2
A, V = (W) AW - WA , WAW, A 2W - WA 2, WAW 2 - W 2AW
122 4. Representation of Const itutive Functions

where for i = 1,2 ,3,

The above results can also be read off from Tabl es 4.5 and 4.6.
In obtaining (4.56) , we have changed the skew-symmetric t ensor W
back to it s corresponding vector v . For exa mple, with W ij = CijkVk from
(A.30), we have

tr W 2 = W ij W j i = Cij kCji lVkVl = -26klVkVl = -2vj vj ,

(A W + W A )i = 2CijkAj i W 1k = 2Ci j kClknAjl Vn
= 2(6in6jl - 6il6j n)Aj1vn = 2Ajj Vi - 2A j i vj ,
or

tr W 2 = -2v · v , (A W + WA ) = 2(tr A)v - 2Av .

Therefore, we can replace tr W 2 by v ·v and (A W + W A ) by Av , bearing

in mind t hat t he rem ainder of the expression consist s of elements already
in t he list of invariants. 0

Exercise 4.4.1 Find the general represent ations of scalar , vect or , an d

symmetric tensor hemitropic fun ctions of two vect or vari abl es (u , v) .

4.5 Anisotropic Invariants

Many anisot ro pic solid bodies poss ess symmet ries that can b e charac te rized
by certain direct ions, lines or planes, more specifi cally, say, characterized by
som e unit vectors m = (mI , ' " , rna ), and some tensors M = (MI , ' " , Mb)'
Let g be the group t hat preserves these charac te rist ics, i.e.,

g = {Q E G , Qm = m, QMQT= M} , (4.57)

wher e G is a subgroup of O(V) , usually O(V) its elf or O+(V) . In (4.57) we

have also used the not ations introduced in (4.46) .
Obviously, not every anisotropi c solid can be specified by a symme try
group of the typ e (4.57). However , many of them do , among t hem, t rans-
versely isot ropic, orthotropic bodi es, and som e class es of crystalline solid .
For such materi al bodies, we can obt ain represent ations of invariant func-
t ions relative to symmet ry group g in terms of invari ant functions relative to
t he group G, which is usu ally O(V) or O+(V) , from the following t heorem .
 4.5 Anisotropic Invariants 123

Theorem 4.5.1. Let f(v , A) be a sceler-, vector-, or tensor-valued function

and 9 be a group of th e type (4.57) . Th en f(v, A) is invariant relative to 9
if and only if it can be represented by
f(v , A) = f(v ,A,m,M) , (4.58)
where the function f(v, A,m, M) is invariant relative to G .
Proof We sha ll prove the case for a scalar-valued fun ction only. The proofs
for the other cases are simil ar . For sufficiency, note that if
f(Qv ,QAQT,Qm,QMQT) = f(v ,A,m,M), V Q E G,
then for any Q E 9 c G, we have Qm = m, and QMQ T = M , hence
f(Qv ,QAQT,m,M) = f(v ,A,m,M) , V Q E 9,
which from (4.58) implies that f(v , A ) is invariant relativ e to 9.
Conversely, let M be the orbit of (m , M) in G, that is,
M = {(Qm ,QMQT), V Q E G} .
Then for f(v , A) invariant relative to 9, we can define a function f on V x M
by
f(v ,A,ii,N) = f(RTv ,R TAR) , (4.59)
for any (v , A) E V and (ii , N) E M, where REG is such that
Rm = ii , RMRT = N . (4.60)
In gener al, R is not uniquely det ermined by the condition (4.60). However,
if R' E G also satisfies t he condit ion, then
Rm = R'm , RMRT = R'MR,T,
which implies t ha t Q = R T R ' E O, an d
f(RTv ,R TAR) = f(RT(R'R' T)v ,RT(R'R 'T)A(R'R'T)R)
= f(Q(R'TV), Q(R'TAR')QT) = f(R'TV, R'TAR') ,
since f is invari an t relati ve to 9. Therefore , t he definition (4.59) is well de-
fined .
Now, for any Q E G, by definition
f (Qv, QAQT,Qm, QMQT) = f(RT(Qv) , RT(QAQT)R) , (4.61)
wher e Rm. = Qm and RMRT = QMQT , or equivalently
RTQm = m, (RTQ)M(RTQf = M ,
which implies RTQ E 9. Therefore, sin ce f is invari ant relative t o 9, (4.61)
reduces t o
V Q E G,
which proves that f is invar ian t relative to G. 0
124 4. Representation of Constitutive Functions

4 .5.1 Transverse Isotropy and Orthotropy

As examples, we consider the symmetry groups given by (3.61)
91 = {Q E O(V) , Qn = n} ,
92 = {Q E O(V) , Q(nl 0 nt}QT = nl 0 nl , (4.62)
Q(n2 0 n 2)QT = n 2 0 n 2, Q(n3 0 n 3)QT = n 3 0 n 3},
where n is a unit vector, and {nl ' n2 , n 3} is a set of orthonormal unit vectors.
The group 91 has a pr eferr ed dir ection charact erized by the uni t vector
n . A material body that has such a pr eferr ed dir ection in an undistort ed
configur at ion is usu ally called a transversely isotropic solid.
By the commutat ion theorem , since Q(n1 0nl)QT = n1 0nl, the trans-
form ati on Q preserves the charac te ristic space of nl 0 nl , which is the line
in the dir ection of n l ' Therefore, the group ~h pr eserves t hree mutually or-
thogonal lines in the dir ections of {nl ' n 2, n 3}' A body with this symmetry
relative to an und istorted configurat ion is usu ally called an orthotropic solid
body'

Example 4.5.1 For a tran sversely isotropic elastic solid , from (4.53) the
second Piol a-Kirchhoff st ress te nsor S ,,(C) is a t ensor invar iant relative
to the symmetry group ~h in the undistort ed configurat ion r: By The-
orem 4.5.1, we can derive the representation for S" by considering t he
isotropic invari ants of two vari abl es C and n . For this case, from Tab les
4.1 and 4.3, we obtain, for the second Piela-Kirchhoff stress t ensor
S" = sol + S IC + S2C2 + S3n 0 n
(4.63)
+ s4(n 0 Cn + Cn 0n) + ssCn 0 Cn ,
where for i = 0, . .. , 5,
Si = si (tr C, trC 2 , trC 3 , n · Cn , n · C 2n ).
The sca lar invari ant n . n has been eliminated since it is equa l to t he
constant 1.
Therefore, from (4.63) we obtain the representation for the Cauchy
stress t ensor T = J-1FS"F T ,
T = tal + t 1B + t2B2 + t3Fn 0 Fn
(4.64)
+ t4(Fn 0 BFn + BFn 0 Fn) + t sBFn 0 BFn ,
where for i = 0, . .. , 5
ti =ti(trB , t rB 2, trB 3, F n· Fn, Fn·BFn) . (4.65)
In the above transitions, we have used the relations (4.16) and (1.13)
and the fact that C and B have t he sa me principal invariants . 0
2 For more discuss ions on characte rizat ion of some a nisot ropic materi al bodies see
[38] .
 4.5 Ani sotropic Inva riants 125

Example 4.5.2 Consider an orthotropic elastic solid body with the

symmetry group 9z defined in (4 .62) in an undistorted reference con-
figur ation n : Similarly, the representation for S",.(C ) can be obtain ed
from isotropic invari ants of four symmetric tensor variables C , nl ® nl ,
n z ® n z, an d n 3 ® n 3. For t his case, we have
Y = {trC, trC Z, t r C 3, trC(nl ®n d, t r C (n z ® nz ), tr C(n3®n3),
s

trCz(nl ® nl) , trCz(nz ® n z) , tr C z(n 3 ® n 3)}'

and
Yt = {L c , C z, nl ®nl , nz ®nz , n 3®n3, C(nl ®nd+(nl ®ndC,
C(nz ® n z) + (n z ® n z)C, C(n3 ® n 3) + (n3 ® n 3)C,
C(nl ® nd C, C(nz ® nz) C, C(n3 ® n 3)C} ,
where we have alre ad y elimina t ed many redundant eleme nts by in-
spection, using the identities such as (n l ® nd z = n l ® nl and
(nl ® nd(nz ® nz) = O. Furthermore, by the use of t he identities:
nl ' Cru , +nz · Cnz + n 3' Cn3 = trC,
nl ® n l + n z ® n z + n 3 ® n 3 = 1 ,
we can easily see t hat t he underlined elements tr C , tr C Z, 1 , C , and C Z
are also redundant in the above list s. Therefore , we ca n express SK as
SK = 81 n l ® n l + 8zn z ® n z + 83n 3 ® n 3
+ 84(nl ® Cn l + Cn., ® nd + 85 (n Z ® Cnz + Cn z ® nz)
(4 .66)
+ 86(n 3 ® Cn3 + Cn 3 ® n 3)
+ 87Cn l ® Cn ; + 8sC nZ ® Cn z + 8gCn3 ® Cn3,
where s; depends on (det C, nl ' Cnl , nz · Cnz , n 3' Cn3, nl ' C Znl , nz ·
CZn z, n 3' CZn3)' We have also replaced tr C3 by det C by the use of
relation (4 .19) . One can easily turn (4 .66) into a representation for the
Cau chy st ress t ensor T . D

Exercise 4 .5.1 Find t he general repr esentations of the heat flux,

q = Q(O,g) , g = grad O,

for a rigid heat conduct or with t he following different symmetry groups .i'
1) 91 = {Q E O(V) , Qn = n} ,
2) 9z = {Q E O+(V) , Qn = n} ,
3) 93 = {Q E O(V) , Q(n ® n)QT =n ® n},
resp ectively.
3 T he groups 91, 92, a nd 93 characterize three different classes of tran svers e
isot ropy, see [38].
126 4. Represent a tion of Co nstit ut ive Functions

Exercise 4.5.2 From (4.66) , write out the repr esentation for the
Cau chy st ress te nsor T of an or thotropic elastic b ody.

4.5.2 On Irreducibility of Invariant Sets

From t he above exa mples, we not e that t he invari an t sets obtained in this
manner are in general not irr educible sets. In fact , some trivial redundant ele-
ments ca n b e eliminate d by insp ection using known identities . However , elim-
ination of such redundant eleme nts need not render irreducib ility of t hese set s
du e to the constancy of t he added vari abl es 'Iii and M. Proof of irr educibility
is not trivial in general. We give a sim ple example below.

Example 4.5.3 Consider a vect or invari ant h(v , A) of a symmetric

t ensor vari abl e A and a vector vari abl e v relative to the group 9 1 defined
in (4.62). We obtain from Tabl e 4.2,

Appar ently, there are no trivial redundant eleme nts in this list. However ,
we ca n show that A 2 v is redundant and by removing it we have ind eed
an irr educible gener ator set ,

i; = {v , n , Av , An, A 2n} .
F irst , let us prove that A 2 v is redundant . Clearly, if v = a n, t he
redundancy is obvious. So let us assume that {v , n} is linearly indepe n-
de nt and that y~ is not a generat or set . In ot her words , we ass ume t ha t
A 2 v ca n not be expressed as a linear combinat ion of the element s in Y~.
Since t he space V is three-dimensional, t he set y~ spans at most a two-
dimensional space. Since {v , n} is linearly independent , y~ must span
t he subspace [v, n], where we use this not ation for t he space generate d
by the vect ors in t he squar e bracket s. In particular, we have

Av E [v, n] and An E [v, n].

Therefore, we have
Av = av+bn ,
for som e a, b E JR . Consequently,

A 2v = aAv + bAn,
and hence A 2 v mu st belong to the subspace [v, n], which is a contradic-
t ion to our ass umpt ion, and t he redundan cy is proved .
Now let us prove that y~ is irr educible. Note that if v = a n, t hen y~
reduces t o the set {n , An, A 2 n }. Since the sp ace is t hree -dimensional,
 4.5 Anisotropic Invariants 127

this set is irr educible. Now, suppose that {v, n} is linearly independent
and n is an eigenvect or of A , then

[n, An, A2n] = [n].

Suppose that Av 1. [v, n], then the set {v ,n , Av} is linearl y independent.
In these case, the elements of {v ,n , Av} are irr educible. Therefore , y~
is an irr educible generator set. 0
5. Entropy Principle

5.1 Entropy Inequality

In t his chapte r, we shall give an int ro duction to thermodynamic const it uti ve

t heories. We have already mentioned the first law of thermodynamics, the
energy bal an ce, in Ch ap . 2. Now we ar e going to consider the second law for
which the essential qua nt ity is the ent ropy .
We denote t he ent ropy of a par t P by H and assume that it is given by

H(P , t) = r
}p t
pn do, (5.1)

where ry(x, t) is called the specific entropy density .

Associated with t he cha nge of ent ropy, there is a qu antity called entropy
supply t:P. We assume that t:P is given by

t:P(P ,t) =- r q, .nda+ i;r psdv ,

i; (5.2)

where we call q,(x , t) t he entropy fl ux and s the entropy supply density due
to exte rn al sources. It is assumed that ry and s are obj ective scalar qu an tities,
while t:P is ass umed t o be an obj ective vector quantity.
Let the difference between the rate of change of ent ropy and its supply
be called the entropy production and denoted by

E(P , t) = H(P , t) - t:P(P, t) . (5.3)

Unlike the energy, the ent ropy is not a conservative qu antity. However , it is
commonly assumed that the entropy production is a non-negative quantity .
Such a state ment is oft en referred t o as the second law of thermodynamics,
i.e.,
E(P , t) ~ 0 \:IP c B,
or
:t r pry dv + }aP,
i;
r q, . n da - }Ptr ps dv ~ O. (5.4)

This relation is called the entropy inequality.

I.-S. Liu, Continuum Mechanics

© Springer-Verlag Berlin Heidelberg 2002
130 5. Entropy P rinciple

Similar to t he balan ce laws, from (2.13) we have the local forms of t he

ent ropy inequality,
pi] + div P - ps 2 0, (5.5)
or
8;2+ div(p1]x + p) - ps 2 0, (5.6)

and from (2.15) t he ent ropy jump condit ion at a singular surface,

(5.7)

In deriving t his jump condition, we do not rul e out the possibili ty that there
might be an entropy production on the sur face and it is assu med to be a
non-negati ve qu antity, in general, as well.
In the reference configuration , from (2.22) t he ent ropy inequality and it s
j ump condit ion can be written as

(5.8)

where the material ent ropy flux P " = 1J1F- 1 p .

The ent ropy inequ ality in the form (5.4) is qui t e general. Motivat ed by
the results of class ical thermost atics, it is oft en assumed that the ent ropy flux
and t he ent ropy supply ar e proportional to the heat flux and the heat sup-
ply, resp ectively, in classical t hermodyna mic t heories. Moreover, both propor-
ti onal cons t ant s are assumed to be t he reciprocal of the absolute temp erature
(),
1 1
P = B q, s = B r. (5.9)
T he res ulti ng entropy inequ ality is called t he Clausius-Duhem inequality,

r pi] dv + ir; ~ q . n da - i;r ~ pr dv 2 0,

i; (5.10)

and the local form (5.5) becomes

P1]. + diIV (jq - P (jr >°

_ . (5.11)

The role of the Clausius-Duhem inequality in the development of modern ra-

ti onal thermod yn am ics will be illustrat ed in this chapte r. However, we should
rem ark t hat alt hough theories based on the Clausius-Duhem inequ ality seem
t o be acceptable in various situations, the assumptions (5.9) are known to b e
inconsist ent with the results from the kin et ic t heory of gases . We shall post -
pone t he discussion of rational thermodyn amics based on the general ent ropy
inequ ality (5.5) to Ch ap . 7.
 5.2 Entropy Principle 131

5.2 Entropy Principle

On e of the principal obj ectives of cont inuum mechan ics is to determine or

pr edict the behavior of a body once the ext ern al causes are specified. Math-
ematically, t his amount s to solving an initial boundar y valued problem gov-
erne d by t he balan ce laws (2.79) ,

p+ pdivx = 0,
pii: - divT = pb,
(5.12)
T=TT ,
pi + div q - T . grad x = pr ,
when the external supplies b and r are given. Aft er introducing the consti-
t ut ive relations for T , c, and q , the balan ce laws (5.12) become a system of
partial differential equations for t he determination of the fields p, X, and B.
We call a set of fields {p , X, B, b, r } a thermodynam ic process of 13, if it satisfies
the balance laws in 13.
On the other hand, the behavior of a body mus t also obey the second
law of thermodyn ami cs, i.e., t he thermodyn ami c pro cess must also sa t isfy
t he ent ropy inequ ality (5.5),

pi] + div ep - ps ~ o.
Classical const it ut ive relations were usually proposed bas ed on physi-
cal experiences or experimental observations. Thermodynamic proc esses, ob-
tained as solut ions of t he bal an ce laws and t he proposed const it ut ive rela-
tions, are further required to be consist ent with the second law of thermo-
dyn ami cs so as to be regarded as physically admissible pro cesses. This is the
usu al role of the second law of thermodynamics in the traditional formul ation
of physi cal problems .
Following the idea set forth by Coleman and Noll [11]' in modern rational
thermodynami cs, the second law of thermodyn ami cs plays a more important
role:

Entropy principle. It is required that constitutive relations be such

th at the entropy inequality is satisfied ident ically for any th ermodynamic
process.

From this point of view, the ent ropy principle imposes restrictions on const i-
t ut ive fun ctions, just as the principle of mat erial objectivity and the material
symmetry do, so that the resulting const it utive fun ctions can be established
in a physical mod el for the material body. Consequently, with such restric-
tions taken care of, all thermodyn ami c processes will always be consist ent
with the second law.
132 5. Entropy Principle

To find the restrictions imposed on const it ut ive functions by the en-

t ropy principle is one of the major obj ectives in modern cont inuum thermo-
dyn amics. In this chapter, we shall illust rate the procedure for exploiti ng this
requirem ent for t hermoelast ic materials.

5.3 Thermodynamics of Elastic Materials

We shall now exploit the ent ropy principle based on the Clau sius-Duhem in-
equality for thermoelastic mat erials. Following the method of its exploitation
sugges ted by Coleman and Noll [11], we define the free energy density 'l/J by

'l/J = c - ()1], (5.13)

and use the energy balan ce (5.12)4 t o eliminat e r , so that t he Clausius-

Duhem inequ ality becomes

(5.14)

where (1.31) has been used .

The constitutive relations for thermoelastic materials are given by

T = T(F,(),g) ,
1] = 1](F,(),g) ,
q = q(F,(},g) , (5 .15)
'l/J = 'l/J (F, (} , g ).
e = c(F, (} ,g ),

We assume that the response functions a re smooth, then

and hence (5.14) becomes

o'l/J)
p (1 ] + . o'l/J . (
- ()+p_.g- TF -
T
- po'l/J)
- · F. + -1 q· g < O (5.16)
o() og of () - '

The entropy principle requires that the above inequ ality must hold for
any t hermodynamic pro cess {p, X,(), b, r ] . Now we sha ll evalua te the restric-
t ions imposed by t his requirement .
Not e that for any given fields {X,()}, one can det ermine the fields {p, b, r}
from the balan ce laws (5.12) , so that {p, X, (), b, r } is a thermodynamic pro-
cess. In particular , for any given valu es of {F, (), g , F, 0,g} at (X , t) such
that det F i- 0 and () > 0, t here exist s a thermodynami c pro cess having t hese
valu es at (X ,t) . In other word s, for any given {F , () ,g} the inequ ality (5.16)
 5.3 Thermodynamics of Elastic Materials 133

mus t hold for arbitrary values of P, 0, and 9, in which the inequality is linear.
Cons equently, their coefficients mus t vanish identically,
81j!
71 + 8B = 0,

81j! = 0 (5.17)
8g ,

-T 81j!
TF -P8F =0 ,

or equivalent ly,
1j! = 1j!(F, B),
81j! (5.18)
TJ = - 8B '
Moreover , we have from (5.12h the relation,

(5.19)

and from (5.16) the rem aining inequality,

q(F,B,g) · g :::; O. (5.20)

The above relations represent the most general forms of the constitutive
relations for thermoelastic mat erials . Compared to the consti t ut ive relations
(5.15) they are great ly simplified . Indeed , t he stress, the int ernal energy and
t he ent ropy ar e complete ly det ermined once the free energy function 1j! (F , B)
is known.
T he inequality (5.20) is called Fouri er 's inequality, which st ates that
the heat flux always points in the dir ection from a hot sp ot to a cold one .
Furthermore, since the left-hand side of (5.20) , deno t ed by !(F,B,g) , is a
smooth function of 9 and it assumes its maximum , namely zero , a t 9 = 0,
we can conclude that

8g!l g = 0 = (q + ~q9 T g) Ig =O = q(F, B, 0) = 0,

(5.21)
8~!lg=0 = (~q + ~q T) I is negative semi-definite,
9 9 g =O

for any (F, B). If we assume t he classical Fouri er's law of heat conduct ion,

q = -K(F,B)g, (5.22)

where K denotes the thermal conductivity tenso r, then (5.21h implies that
the symmetric part of K(F, B) is positive semi-definite for any (F, B) .
134 5. Ent ropy Principle

From (5.18) we can writ e t he differenti al of 7jJ as

1 T
d7jJ = -TF - . dF - 1] dB , (5.23)
p
or equi valently,
d1] = ~ (d e - ~TF-T . dF) . (5.24)

This is known as the Gibbs relation for elast ic mat erials. In particular , for
a thermoelasti c fluid , which dep end s on the deformation gradient F only
t hrough it s dep end enc e on the density p, we have
T = -pi , dp = _pF - T . dF,
where p = p(p , B) is t he pressure and the relation (5.24) reduces to t he well-
known Gibbs relation for fluids in classic al thermost atics:

d1]
1( p) lOc
= B de - p2 dp = BaB dB + B ap - p2 dp.
1 (Oc p) (5.25)

An immediate consequence of (5.25) is the following int egrability condit ion

for 1](p, B),
Oc p B ap
(5.26)
ap p2 - p2 aB '
Simil ar integrability conditi ons can be obtained from (5.23) and (5.24) .
From the relation (5.18)4 and t he relation (3.31), it follows t hat in any
isotherm al pro cess, a thermoelastic materi al is a hyp erelastic mat erial , with'
free energy functi on 7jJ served as t he st ored energy fun ction ,

CT(F) = 7jJ(F, B)II!=constant , (5.27)

and
T aCT F T .
= PaF (5.28)

Not e that if we introduce the Piola-Kirchhoff stress tensor T"" the relation
between the stored energy function and t he stress becomes even simpl er ,
aCT
T", = P"' aF ' (5.29)

where p", is the densi ty in t he referen ce configuration r: and we have used the
relations (2.84h and (2.34) .
On e may notice that the momentum and the energy balance impose no
restrictions in t he evaluation of the inequ ality (5.16) , simpl y because one ar-
gues that the external supplies b and r can be suitabl y adjuste d so as to sat isfy
the momentum and the ener gy balan ce. This argument may sound wishful
in t he real world! but it is usu ally regarded as acceptable. Nevertheless, we
shall see in Ch ap . 7 that such an argument can be avoided .
1 This argument has been criticized by Woods [79] .
 5.3 Thermodynamics of Elastic Materials 135

Exercise 5.3.1 Show that the Gibbs relation (5.25) ca n also be written
in the following forms :
P
d7/J = 2 dp - 7] dO ,
p
(5.30)
1
d(p7]) = (j (d(p e:) - g dp) ,

where 7/J is the free energy and g is t he free enthalpy defined as

7/J = e: - 0TJ, (5.31)

Exercise 5.3.2 For an isotropic thermoelastic solid , the stress tensor

(5.18)4 and the Gibbs relation (5.24) become

(5.32)
d7] = -1 ( de: - -1 T B - 1 . dB ) .
o 2p

From the representation theorem (4.26) , we can write

T = tol + tIB + t 2B2 ,

(5.33)
7/J = 7/J (O,IB ,II B, III B )'
Show that the par am et ers t, ar e given by

0 '1/;
to = 2pIII B alIIB '

(5.34)

5.3.1 Linear Thermoelasticity

First, we not e t hat for thermoelastic materials , by the use of the Gibbs rela-
t ion (5.24) ,
· e:. - -1 T F - T . F',
O7]=
p
136 5. Entropy Principle

and hence the energy equation (5.12)4 can be written as

pB", + div q = pro (5.35)

For the classical linear theory, the displacement gradi ent H and the
e
temperature increment = B- Bo are assumed to be small quantities, where
Bo is the refer ence t emperature of the body. On the other hand, similar to
(3.35) the principle of material objectivity requires that the depend ence of
the free energy function on the deformation gradient F must reduce to the
depend ence of the right stretch tensor U . Since U = 1 + E + 0(2) from
(1.22) , in the linear theory we have 'Ij; = 'Ij; (E ,B) . Therefore, let us express
the function 'Ij; up to the second-order t erms in E and in the following form, e
- 1 cv -2 1 - 1
'Ij; = 'lj;o - TloB + M ijEij - -2 -B B - - Pij Eij B + -2 LijklEijEkl. (5.36)
o Po Po
The relations (5.18h,4 now take the form,

and if the reference state is assumed to be a natural state, then M ij in (5.36)

mus t vanish and we obtain

Tij = LijklEkl - Pije,

c - 1 (5.37)
TI = Tlo + -Bv B + - Pij Eij .
o Po
The fourth-order tensor L is the elast icity t ensor (see (3.79)) and c; is called
the specific heat (at constant volume) because, from (5.18h it follows that

8c
Cv = 8B'
For the linear theory, we shall assume in addition that Fourier 's law of
heat conduction holds, so that the linear expression for the heat flux is given
by
q = - Kg , (5.38)
where K is the thermal conduct ivity tensor.
Therefore, we can summarize the constitutive equ ations of linear ther-
moel asticity for anisotropic materials in the following :

T = L[E] - pe,
q = - Kg ,
(5.39)
P -
Cv -
TI = Tlo + - B + - . E .
Bo Po
 5.3 Thermodynamics of Elastic Materials 137

In general, the coefficients ar e functions of the temperature eo in the reference

configur at ion. Mor eover , the coefficients must satisfy the following condit ions:

L ij kl = L j ikl = L i j 1k = L k 1i j , Pi j = Pj i ,
K is posi ti ve semi-definite, (5.40)
c., > O.
The first two conditions follow fro m the symmetry of the stress and the linear
strain t ensors, and the last inequality is a consequence of thermal stability,
which will be discuss ed later .
If the material is isotropic, then we have, for any orthogon al tensor Q,

T(QFQT,e) = QT(F,e)QT ,
q(QFQT,e,Qg) = Qq(F,e,g) .
Since
- 1 T 1 T
E = 2(H + H ) = 2(F + F ) - 1 ,
from (5.39) it follows immediately t hat

L[QEQT] = QL[E]QT,
p =QPQT,
KQ = QK.

Therefore, we conclude that , in component form s,

and from t he first relation, we have , for any orthogon al te nsor Q,

This implies the following representation (see Ex ample 4.2.2) ,

where). and J-L are the Lame elastic moduli (see also the derivation given in
Sect. 4.2.1) .
In summary, the const it ut ive equati ons of linear thermoelasticity for
isotropic mat erials are given in the following :

T = ). t r E 1 + 2J-L E - Q: e1 ,
q = - K g, (5.41)
T/ = 770 + -eo e + -Po
Cv - Q:
tr E.
-
138 5. Entropy P r inciple

T he field equations for thermo elastic ity consist of the mom entum equa-
tion and the energy equa t ion (or the equivalent equ ation (5.35)) for t he dis-
placem ent u(x , t) and the te mperat ure B(x, t). In component forms, from
(1.24)
E.. _
tJ -
~( aUi
2 !'l
aUj )
+!'l '
u Xj UXi

and the constitutive equat ions (5.39) , we have the following field equations
for anisot ro pic t herm oelastic materi als:

(5.42)

The material coefficients ar e, in general, functions of the referen ce te mpe r-

ature . In particular, if t he reference t emperature is uniform t hroughout the
body so that the material coefficients are constants, then t he field equat ions
(5.42) become

(5.43)

or, if the body is rigid and fixed , then P ij = 0 an d the only field equat ion is
t he classical equat ion of heat conduction,

(5.44)

For isotropic t hermoelasti c materials, from (5.41) the field equat ions
become

(5.45)
 5.4 Elastic Materials with Internal Constraints 139

Exercise 5.3.3 Suppose that the entropy function 'fl(F, e,g) is invertible
in e for each fixed F and g . Then we can regard (F,'fl,g) as the field
variables for a thermoelastic material body with constitutive equations
of the form
c = :F(F,'fl,g) .
1) Evaluate the consequence of the entropy principle with the Clausius-
Duhem inequality using (F, 'fl, g) as field variables.
2) Show that for an isentropic process, i.e., constant entropy, a ther-
moelastic material is hyperelastic, and the stored energy function (J
is given by
(J(F) = s (F , 'fl) I'1 = const a nt . (5.46)

Exercise 5.3.4 Evaluate the consequence of the ent ropy principle with
the Clausius-Duhem inequality,
1) for viscous heat-conducting fluids defined by (4.3) ;
2) for Navier-Stokes-Fourier fluids with constitutive equat ions (4.52),
and show that the viscosities A, J.l, and the thermal conductivity K,
satisfy
(5.47)
while the free-energy density 't/J and the pressure p are related by

't/J = 't/J (p, e) , (5.48)

5.4 Elastic Materials with Internal Constraints

In the previous section, to obtain the main conclusion (5.17) from the in-
equality (5.16) by being able to choose arbitrary values of F and P, we have
tacitly assumed that the material body is capable of undergoing any defor-
mation. However, for some material bodies, such an assumption may not be
appropriate. For example, an incompressible material body can not undergo
any deformation that changes its volume.
If the class of possible deformations for the body is limited, the body is
said to be subjected to internal constraints. Such materials are called con-
strained materials. In this section, we shall consider mech anical constraints/
for thermoelastic bodies.
Let V = {F E L(V)I J.l(F) = O}, where J.l is a scalar-valued function . We
say that V is a mechanical constraint if the body admits only local deforma-
tions compatible with V. For bodies with such a constraint, the values of F
2 For some discussions on thermomechanical const raints, see [8] .
140 5. Entropy Prin ciple

and P can no longer be assigned arbitrary, since F must belong to D and P

must satisfy the following relation,

. Of-l .
f-l(F) = of . F = O. (5.49)

This relation has a simple geometrical int erpretation. If we call Z = Of-l/oF

a vector in £(V) , then the vector P must lie in the plan e orthogonal to Z .
Let us denote this plan e by Z 1., then we must have P E Z 1..
For thermoelastic materials with such a const raint, we can reconsider
the consequ ences of the inequality (5.16) ,

o'l/J) e+p-
p (17+- ' o'l/J 'g-
. (TF - T - po'l/J)
- · F' + -1 q· g < O
oe og 0F e - '

For any given e, g , and FED , this inequ ality must st ill hold for any values
of iJ and g. Ther efore, the first two relations of (5.17) remain valid, and the
inequ ality reduces to

o'l/J) . F'
- ( T F - T - p of 1
+ (jq . g :s; 0, (5.50)

which mu st hold also for any PE Z1. . We claim that this implies

(5.51)

where A is an arbit rary par am et er. In other word s, TF - T - po'l/J/oF must

be par allel to Z . For if it do es not , t hen there exists a vector P in the plane
Z 1. such that (TF - T - po'l/J/oF) · P =f. 0, and it can also be chosen in such a
way t hat t he inequ ality (5.50) is violated . Therefore, we conclude from (5.51)
that the stress tensor for a thermoelastic body must be given by

T = N + 7(F, e), (5.52)

where
(5.53)

is called the reaction stress due to the constraint.

Note that there is no work don e due to the reaction stress. Indeed , the
stress power produced by the reaction stress is

Of-l T
N ·L = A-F .L = AOf-l- . F. = 0 (5.54)
of of '

by the use of (5.49) and L = PF - 1 • Conversely, if N . L = 0 for any F

satisfying f-l(F) = 0, then one can show that N must b e of the form (5.53) .
Therefore, we can say that the constraint is maintained by the reaction stress,
 5.4 Elastic Materi als with Internal Constraints 141

which does not do any work in the actual deformation. This observation is
often t aken as a postulate':'
The stress, for a m ateri al body with mech anical constraint, is det ermined
by th e history of motion only to within an arbitrary tensor N that does
no work, i.e., N . L = 0, in any motion compatible with th e constraint.

Since int ernal constraints are intrinsic properties of material bodies, we

assume that the constraint sat isfies the condition of material obj ectivity,
namely,
/-L(QF) = /-L(F) , V Q E OW) , (5.55)
which impli es that

Therefore, (5.53) can be written as

N = 2>.F oj], F T (5.56)

oC '
from which we conclude that the reaction stress N is symmetric and obj ective
like the st ress T itself. Moreover , the condit ion (5.54) can also be written as

N ·D=O, (5.57)

for any D , the symmetric part of L , compatible with the constraint. Further-
mor e, one ca n show that the reaction stress N is ind ependent of the reference
configuration. Indeed , if Pi, is another reference configuration, then we have

°
where P = 'V( K 0 ",-1) and F" = F",P . Hence, the cons traint is given by
P(F;J = in K an d one can easily show t hat

oj], = op, p T.
OFk er;
By (5.53) , the reaction st ress N (F k ) relative to Pi, is

Since the reaction stress is symmetric, obj ective, and indepe ndent of the ref-
erence configuration, the const it ut ive function T(F, B) in (5.52) is subject ed
to the sam e conditions of mat erial obj ectivity and material symmetry for the
stress tensor itself.
3 T his postulat e is referred to as t he principle of determinism for const rained
simple material bodies [71] .
142 5. E ntropy Principle

Suppose t hat t he constraint D is specified by mul tiple condit ions,

D = {F E £(V) IJLa(F) = O,a = 1, ·· · , m}. (5.58)

Then with similar argument s leadi ng to (5.53) , we can obtain the reaction
st ress
(5.59)

where >. a, a = 1, " ' , m are arbitrar y parameters. We rem ark t hat these pa-
ram et ers can not be det ermined from the thermomechani cal st ate var iabl es
alone, therefore, t he reacti on st ress N is also "undet erminat e". Never t heless,
the react ion stress N , as a part of the to t al stress T , ca n be det ermined by
solving t he field equations with appropriate boundary condit ions.

Example 5.4.1 Incompressibility

We now consider t he most importan t internal const raint - incompress-
ibility. This requires that t he deformation is volume pr eserving, which
ca n be descri bed by

D = {F E £(V) II det F] = I} = U(V) , (5.60)

or JL(F) = Idet FI-l. Therefore, from (5.53) and the relati on OF det F =
F - T det F , we concl ude that the react ion st ress N is an arbitrary pres-
sure,
N = -pl. (5.61)
Con sequ ently, the const it utive equ at ion for an incompressible therm o-
elastic mat erial bod y mus t be of the form ,

T = - pi + T(F, ()), \IF E U(V), (5.62)

where p , the undet er min ate hydrostat ic pr essure, is the react ion force of
the body to maint ain incompressibili ty. 0

Example 5.4.2 Rigidity

A mate rial is called rigid if it admits on ly rigid motions. i.e.,

XK (X, t ) = Q(t)(X - X o) + c , Q E O(V).

Therefore, t he const raint is given by

D = O(V) . (5.63)
 5.4 Elastic Mat erials with Internal Constr aint s 143

For any F E V , we have

L = FF -I, F E O(V),

which is a skew t ensor, and henc e its symmetric part D must vanish.
Therefore, the condit ion (5.57) is identically satisfied for any tensor N .
In other words, the st ress of a rigid material body is totally un affect ed
by t he motion. The reaction st ress N is an arbit rary symmet ric tensor .
D

Example 5.4.3 Inextensibility

Suppose that the body is inextensible in the direction parallel to a unit
vector e in a reference configuration K , then the const raint can be ex-
pressed by
'0= {F E £(V) I IFel = lei}· (5.64)
In this case, we ca n t ake p,(F) = Fe -Fe-s e -e and hence 8 F P, = 2Fe 0e .
It follows immediately from (5.53) that

N = 2>"Fe 0 Fe . (5.65)

The reaction stress N is a pure t ension in the dir ection of Fe . D

Exercise 5.4.1 Det ermine t he reaction stress N for the constraint

'0= {F E U(V) Ie ·Ce - e·e = OJ,

wh ere e is a fixed unit vector in the reference configuration. Interpret
this const raint physically.

Exercise 5.4.2 Show t hat the reaction stress N is an arbit rary sym-
metric t ensor for t he rigidity cons t raint

'0= {F E OW)} ,
by taking t he te nsor equation /1(C) = C - l = 0 as a const raint (equiv-
alent t o a multiple condit ion of six equati ons).

Exercise 5.4.3 By using a similar argument employing the Clausius-

Duhem inequality, show t hat for incompressible Navier-Stokes fluids the
st ress tensor is given by

T = -pi + 2p,D, p, ~ 0,

with arbit ra ry pressure p and t r D = o.

144 5. Entropy Principle

5.5 Stability of Equilibrium

For analyzing thermodynamical stability of equilibrium, we consider a supply-
free body occupying a region V with a fixed adiabat ic boundary. We have

v = 0, q .n = 0, ip . n =0 on 8V,

and hence the ent ropy inequality (5.4) and the energy bal an ce (2.68) become

:t fv pn do :::: 0,
d
dt
r
Jv p( c + i1 V . v) dv = O. (5.66)

In other words, the total ent ropy must incre ase in time while the total energy
rem ains constant for a body with fixed adiabat ic boundar y. St atements of this
kind are usually called stability cri teri a, as we sha ll explain in the following
example .
We say that an equilibrium state is stable if any small disturban ce away
from it will event ually die out and thus the original st ate will be rest ored .
Suppose that the region V is occupied by a thermoelastic fluid in an
equilibrium state at rest with const ant mass density Po and int ernal energy
density co' Now let us consider a small disturban ce from the equilibrium state
at the initial time such that

p(x ,O) = p(x) , c(x , O) = i (x ), v(x ,O) = 0,

and Ip - Pol and Ii - col are sm all qu antities. If we assume that the original
st ate is st abl e then the perturbed st ate will eventually return to t he original
state at a later time. Therefore, since t he total ent ropy must increase we
conclude t hat
fv Po'r/o dv :::: fv PfJ dv , (5.67)

where 'r/o = 'r/(co, Po) and fJ = 'r/(£, p) are the final equilibrium entropy and
the per turbed initial entropy resp ect ively. Here we have regarded (10, p) as
ind ep endent variables instead of (p, B) for convenience. Expanding fJ in Taylor
series ar ound the equilibrium st ate, we obtain from (5.67)

(5.68)

Since total mass and total energy rem ain constant , we have
 5.5 Stability of Equilibrium 145

and hence

Therefore, up to t he second-order t erms (5.68) becomes

2"1
J{
v
)2 0 1 ' (A
2" OE2 / E - Eo + OEOP / E - Eo )( P' - Po)
102"1 I ' (A

Ej2"1 + -10"1)1
+ ( -l -- - '('
P P - Po)2}dv::::: 0.
2 op 2 pop 0

By the mean valu e theorem for integrals, it reduces to

where V is the volum e of t he region V while p* = p(x*) and E* = t (x * ) for

some point x* in V.
Since the non-posit iveness of the above qu adratic form must hold for any
small disturbance and Po and Eo are arbit rary, it follows t hat the matrix

must be negative semi-definite, or equivalentl y

(5.69)

(5.70)

In these relations, partial derivatives ar e t aken with resp ect to the vari ables
(E, p).
In order to give mor e suggestive meanings to the above conditions for
st ability, we shall reit erat e them in terms of the indepe ndent vari abl es (p,O) .
In making this change of var iables we must admit the invert ibility of eip, 0)
with resp ect to the temperature, i.e., OE/OO -I O. To avoid confu sion , vari abl es
held constant in partial differentiations will be indic at ed .
146 5. E nt ro py Principle

From the Gibbs relation (5.25) , we have

ory I 1 Oryl
u£:lE p 8' op e

and by the use of the chain rule for t he change of vari abl es (see Exercise
5.5.1) , we have

oB I
OE P
_&
- oB
1-1 p ,
OBI OEI OE I-
op e = - op IJ oB p
1

It follows imm ediat ely that

and
02ry I 2 ory I p oe I 1 op I
Op2 e + Pop e = p2B2 op e - p2B op € '

which, by the relation

and t he integ ra bility condit ion (5.26), becomes

T herefore, in terms of the vari abl es (p, B) , th e st abil ity condit ion (5.69)
reduces to
Be
oB > 0, (5.71)

since we have already admit ted that &/oB =f. 0, and the conditi on (5.70),
afte r simplific ations, impli es
(5.72)

Obviously, these condit ions are restrictions on the const it ut ive fun ctions
ei p , B) and p(p, B).
We have shown that the st ability of equilibrium requires t hat: (1) the spe-
cific heat at constant volume (equivalently at constant densit y) , c; = oE/oB,
must be positive; (2) the isothermal compressibil ity (see Exercise 5.5.3) must
be non-negative. Some other impli cat ions of st abili ty condit ions are given in
t he exercises.
 5.5 Stability of Equilibrium 147

Exercise 5.5.1 Let u, v , and w be three variables and assume there

is a relation between them so that we have the functional relations:
u = u(v,w), v = v(u,w), and w = w(u,v). Show that

1) oulovl -1
ov w ou w - ,

2) au I ov I ow I --1
ov wow uou v - .

We pu t the vari abl e as a subscript only to emphasize its constancy held

in the partial differentiation. (Hint : Write u = u(v(u ,w) ,w) = u(u,w)
and compute au/au and au/ow.)

Exercise 5.5.2 Let v = 1/ P be the sp ecific volume, t he volume per

uni t mass. Show t hat
1) The Gibb s relation can be writt en as

d1j; = -ryd8 - pdv .

2) The condition (5.72) impli es

Therefore, the st ability of equilibrium requires that the free energy

be a concave-upward function of the specific volume.

Exercise 5.5.3 With pressure and temperature as vari ables, let v =

v (p, 8) and ry = ry(p, 8). Define

cp aryl p
= 8 08 spe cific heat at constant pr essure,

a - ~ ov I
- v 08 p
coefficient of thermal expansion,

KT = - ~ ov I isothermal compressibility.
v op 0

Derive t he following expression from the Gibbs relation

op OV 2

Cp = Cv - 8 OV 0 08
I 1
p'

in which the condit ion (5.26) may be used. Furthermore, from (5.72)
148 5. E nt ropy Principle

show that
2 1
cp ~ c v , 0::::; (}v Cp liT'

Therefore, both cp and liT are positive, and 0: must vanish if liT te nds to
zero . Not e t hat the specifi c heat at constant volume can also be defined
as

5.5.1 Thermodynamic Stability Criteria

To est ablish a stability crite rion for a material syste m under a different condi-
tion , on e may try to find a decreasing function of t ime A(t) from the balan ce
laws and the ent ropy inequ ality in integral forms . Such a function is called
the availabilit y4 of the syste m, since it is t he qu antity available to the system
for its expe nse in the course toward equilibrium. In the previous example,
from (5.66h one may define the availability A of the syste m as

A(t) = -l pn d», dA < 0.

dt -

As a second exa mple, we consid er a supply-free body with a fixed isother-

mal boundary,
v = 0, () = ()o on av,
and ass ume that t he relation if! = q /() holds. Then the energy balanc e (2.68)
and the ent ropy inequ ali ty (5.4) lead to

ddt lv p(€ + ~ v ' V) dv + Jr fJV

q ·nda=O,

d
dt v
1
pn d» + (}1
0 J8 V
r
a nda ~ 0.
Elimination of the terms cont aining surface int egrals from ab ove, gives

~ :::;O, A(t) = lp(€-(}or,+~V 'V)dV. (5.73)

In this manner we have found a decreasing fun ction of time, t he availability

A(t ), which charac te rizes t he stability for this syste m . Note that

4 Such a funct ion is a lso kn own as a Liapoun ov fu nction in t he stability theory of

dyn amic sys tems, see [33].
 5.6 Phase Equilibrium 149

is the total free energy if () = ()o throughout the body. Therefore, it follows
that for a body with constant uniform temperature in a fixed region the
availability A reduces to the sum of the free energy and the kin et ic energy.
Summari zing the above two sit uat ions, we can st at e the following ther-
modynamic stability criteri a of equilibrium:

1) For a body with fixed adiabati c boundary and constant energy, th e en-
tropy tends to a m aximum in equilibrium .

2) For a body with fixed boundary and constant uniform temperature, th e

sum of th e free energy and the kin etic energy tends to a minimum in
equili brium .

We have seen in this sect ion that thermodyn ami c st ability crite ria, like
the ent ropy principle, impose further restrictions on properti es of the consti-
tutive functions, nam ely, sp ecific heat and compressibility must be posi tiv e.
On the other hand , such crite ria, besides being used in analyzing st ability of
solutions, are the basic principles for the formul ation of equilibrium solut ions
in t erms of minimization (or maximi zation) problems.

Exercise 5.5.4 Consider a material region V subjec t to a uniform

t emperature ()o and pr essure Po on the boundary avoShow that the
linear momentum of t he body is const ant and the availability of t he
system ca n be defined as

A= r
Jv
p(C:-() oTJ+
Po
P
+~v .v)dv.

The relation (2.10) can be used in the derivation .

5.6 Phase Equilibrium

We consider a body consist ing of two ph ases wit h different material prop-
ert ies. The int erface will be considered as a movin g singular sur face so that
the phase t ransit ion may progress within t he body with a conti nuous and
piecewise smooth deform ation and t emperature (see [41]). Accepting the as-
sumption (5.9) 1 , the jump condit ion (5.8h of ent ropy becomes

-p"U,,[TJI + [~" ] . n " ~ O.

Since the temperature is cont inuous across the int erface, [()I = 0, by elimi-
na ting the heat flux from the jump condit ion of energy (2.87)s and the use of
150 5. Entropy Principl e

the kin ematic compatibility condition (2.26), the jump condition of ent ropy
t akes the form (see (2.88) of Exercise 2.5.3) ,

(5.74)

where (A) = ~(A+ + A-) is the mean value of A at the inter face. We can
rewrit e the last t erm in the above expression as

(TKn K,) • [Fn"J = [TKn K . Fn K] - ~ [TKn K] . F +n K - ~ [TKn"J . F-n K

= [TKn K . Fn K] - ~ PKU~ [Fn K . Fn K] ,

where in the second passage (2.88h has been used. The cond it ion (5.74) now
becomes
(5.75)

Here, we have introduced Eshelby 's ener gy-momentum ten sor ([20]), or sim-
ply the Eshelby tensor, in the reference configuration by

1 T
PK = 'lj;1 - - F TK , (5.76)
PK
where
'lj; = c:-Bry
is the free energy densi ty .
The cond it ion (5.75) characterized the process of phase t ransition , and
we can regard t he left-hand sid e as a function of t he interface velocity UK' the
normal n K , and other properties of the materials adjacent to the interface.
Hence,

wh ere X stands for the other variables involved. Assuming that the fun ction
is smooth, then sin ce 1(0 , X) = 0, we have t he following cond it ion for 1 to
at tain its minimum at UK = 0,

from which (5.75) implies that

n K ' [PK]n K = 0. (5.77)

We say t hat an interface is in equilibrium if there is no sur face production

of ent ropy. Therefore, we conclude that the cond it ion of phase equilibrium is
characterized by the cont inuity of the t emperature B and the Eshelby t en sor
PK across t he interface.
 5.6 Phase Equilibrium 151

We can rewrite the Eshelby t ensor in terms of the Cauchy stress t ensor
T and the mas s density p in the present configuration, from (2.84h

Not e that the Eshelby t ensor is not symmet ric, in general, but, of cour se, only
the symmetric part is involved in the condit ion (5.77) of phase equilibrium.
Moreover, at the singular surface with unit normal n in the present
configur at ion, since from (2.28h ,
pT n
nt<. = IF T n j '
the condit ion of phase equi librium (5.77) becomes

01, _ ~ r« . Bn ] = 0,
[ 'f/ P n . Bn (5.78)

where B =P p T is the left Cau chy-Green te nsor.

For fluids , the stress T reduces to a hydrostatic pressure, T - pl .
Hence, the phase equilibrium condit ion (5.77) for fluids becom es

Therefore, in ph ase equilibrium, the fr ee enthalpy

g = 1/J+ P- (5.79)
P
is continuous at the fluid-fluid interface. The free enthalpy is also known as
the Gibbs fr ee ene rgy.
For solids , if the normal n is in the princip al direction e, of the stress
tensor T,
i = 1, 2, 3,
where Pi is the pr essure on the surface perp endicular to the dir ection e. .
Then the condition (5.78) becomes

This result was est ablished by Gibbs [23] as a generaliza tion of t he phase
equilibrium condition of fluids to non -hydrostatic ally st ressed solids .
6. Isotropic Elastic Solids

6.1 Constitutive Equations

In the pr evious chapte rs, we have discussed the basic equations and consti-
tutive relations for material bodies in general. Now we shall consider some
interesting st atic problems for isotropic elast ic bodies.
For elast ic materials, the constitutive equation of the stress tensor can
be written as
T = T(F) ,
where F is the deformation gradient relative to some reference configuration
K" which is chosen so that det F > O.
A deformation in an elastic body can be regarded as an isothermal defor-
mat ion for a thermoelasti c body, for which the temperature is constant and
uniform throughout the ent ire body. From this point of view, the material is
also hyp erelastic with the free energy 'ljJ as the st ored energy fun ction (see
(5.28) ),
T(F) = o'ljJ(F) F T
P of ' (6.1)

for compressible elast ic mat erials or

for incompressible elastic materials .

Moreover , if the material is isotropic, then by (4.26) we have the following
representation for the stress tensor,

T = tal + hB + t 2B2,
ti = ti(IB, II B,IIIB),
or , alte rnat ively, from (4.27) ,

T = sol + sI B + s- IB- I ,
(6.2)
s, = si(IB ,IIB ,III B ),
where I B, II B, III B are the three principal invari ants of the left Cauchy-
Gr een te nsor B.

I.-S. Liu, Continuum Mechanics

© Springer-Verlag Berlin Heidelberg 2002
154 6. Isotropic El astic Solids

From (6.1), we can also write

T(B) = 2 o¢(B) B
p oB '

wher e the stored energy fun ction ¢ (B ) = ¢ (I B, II B,JIIB) ' By performing

the tensor gradient of ¢ in the above express ion and comparing it with (6.2) ,
we can obtain
o¢ o¢)
80 = 2p ( III B 8111 B + II B oIl B '

o¢
81 = 2p oI B ' (6.3)

o¢
8- 1 = - 2pIII B""~'"
uIIB
Similar relations for t i have been given in (5.34) .
For incompressibl e isot ropic elast ic materials, we can write

T = -pi + 81B + 8_1B- 1,

(6.4)
8i = 8i(IB,JI B) ,
where B is required to be unimodular, i.e., III B = 1. Not e that in this case,
we have lIB = trB - 1 (see (A.36)) . Moreover , we have ¢ = ¢ (IB, lIB) and

o¢ o¢
81 = 2p 01 B ' 8_ 1 = - 2P o II ' (6.5)
B

The undeterminat e pre ssure p can not be det ermined from t he deformation.
The simplest model for incompressible isotropic mat erials is to t ake the
linear form for the st rain energy fun ction

¢ = a(I B - 3) + (3 (II B - 3), (6.6)

where a and (3 are constants. This defines a Moon ey-Rivlin material. The
stress tensor becomes

(6.7)

where, from (6.5) , 8 1 = 2pa and 8- 1 = -2p(3 are const ant s. This incom-
pressible material model is often adopted as the constitutive equat ion for
rubber-like materials . Experimental data indicate that bo th a and (3 are pos-
iti ve.! and the numerical valu e of (3 is much smaller t han t hat of a , typic ally,
say (3 ~ O.la . Therefore , experimental evidence seems to support

81 > 0, 8 -1:::; O. (6.8)

1 More discussions on ex pe rime ntal data can b e found in Sect . 55 of [71].
 6.2 Boundar y Value Probl em s in Elasticity 155

They are known as (empirical) E-in equalities in elast icity.

If, in addition , the material constant (3 is zero, then t he stress tensor is
given by
T = -pl +s 1 B . (6.9)
T his defines a neo-Hookean mat erial. This mat erial model is also pr edi ct ed
by t he kineti c theor y of rubber from molecular calculat ions in the first ap-
proximation (see [68]). It provides a reasonable account of the behavior of
natural rubber for modest st rains.
Before closing this sect ion, we shall add an int eresting rem ark. From the
represent at ions (6.2) and (6.4) , it is obvious t hat for compressible or incom-
pr essibl e isotropic elastic mat erials, the stress tensor and t he left Cau chy-
Gre en t ensor commut e (see also [4]),

TB=BT. (6.10)
In particular , for a deformation such that the physical components B (13) =
B (23) = 0, since , according to (6.2) or (6.4) , T( 13) an d T (23) also vanish, the
only non-trivial relatio n of (6.10) is t he expression for the (12)-COmponent,
which reads
T (11) - T (22) B (11) - B (22)
(6.11)
T (12) B (12)

T he relations (6.10) and (6.11) between st ress and deformation do not dep end
on any partic ular constit uti ve funct ion, and thus t hey are called un iversal
relation s of isotropic elastic materials, compressible or incompressible . Rela-
tions of this kind ar e very import ant for expe rimental verificati on of mat er ial
models, since they reflect a dir ect consequence from t he material symmet ry
withou t having to know the const it utive fun ction its elf.

6.2 Boundary Value Problems in Elasticity

For elast ic solid bodies, the equat ion of mot ion is given by

divT(P) + pb = px , (6.12)

or , in te rms of referential description,

(6.13)
where T" = J T p - T is t he Piol a-Kirchhoff st ress te nsor and p" is the mass
density in the reference configurati on K . The exte rnal body force b is usually
given in a specific problem .
A boundary valu e pr oblem in elastic bodies is a problem of finding so-
lutions, x = X(X , t) , of (6.12), or (6.13) , with certain boundar y conditions.
The following t hree ty pes of boundar y condi tions are ofte n consid ered .
156 6. Isotropic Elastic Solids

1) Traction boundary condition: The forces acting on the boundary are pre-
scribed,
(6.14)
where i; denotes external surface forces exerted on the boundary and
n", denotes the outward unit normal in the reference configuration (see
(2.91)) .
2) Place boundary condition: The position of the boundary is prescribed,

X(X) = xo(X), (6.15)

where xo(X) is a given function .

3) Mixed boundary condition: The traction is prescribed on a part of the
boundary, while on the other part of the boundary the position is pre-
scribed.

Boundary value problems for incompressible elastic bodies can be simi-

larly formulated . From (5.62) the constitutive equation for the stress tensor
can be written as

T = -pi + T(F) , detF=I,

and the equation of motion (6.12) becomes

- gradp + divT(F) + p b = px, (6.16)

or, in referential description ,

(6.17)

where p is the undeterminate hydrostatic pressure. The pressure p is not

directly related to F and should be determined in order to satisfy both the
equation of motion and the boundary conditions.
We shall not discuss the existence and uniqueness of boundary value
problems defined above except to make an opportune remark on unique-
ness . A thin hemisphere, like a tennis ball cut in half, for example, can be
turned inside out, so that there are at least two different solutions to the
boundary value problem of a hemisphere with zero traction on the boundary.
Therefore, it is clear that an unqualified uniqueness is not to be desired for
boundary value problems in general. In the later sections, multiple solutions
for traction boundary value problems - biaxial stretching of a square sheet
and deformation of a hollow sphere - will be discussed in detail.
We shall cons ider some solutions, called controllable deformations. Such a
solution is specified by a certain deformation function satisfying the equation
of motion such that the body can be maintained in equilibrium by applying
 6.3 Homogeneou s St retch 157

suitable surface traction on the boundar y alone. In ot her words , a controllabl e

deformation is a solut ion of t he equilibrium equati on,

divT( F) + pb = 0, (6.18)

for compressible elast ic bodies, or

- grad p + divT(F) + pb = 0, (6.19)

for incompressibl e elast ic bodies. No addit ional boundar y condit ions are pre-
scrib ed , instead , the boundary tractions are to be det ermined from (6.14) . In
the case of incompressibl e bodies, the pr essure field must be suitably chosen
so as t o satisfy the equilibrium equa t ion.
In general, a controllabl e deform at ion for a certain elast ic mat erial may
not be controllable for a different elast ic mater ial , since the equilibrium equa-
tion dep end s on the constit ut ive equat ion. If a deformation fun ction is con-
trollabl e for a certain typ e of elast ic mat erials, it will be called a universal
solution of such materials. It has been shown by Ericksen ([16] or Sect . 91 of
[71]) t ha t homo geneous deform ations are the only class of universal solutions
for compressible isotropic elast ic bodies. However , being allowed t o choose a
suitable pressure field in order to satisfy the equilibrium equation gives an
additiona l freedom for possible solu tions in the case of incompressible bod ies.
And ind eed , there are several other well-known classes of universal solut ions
for incompressible elastic bodies. The sear ch for universal solutions is known
as "Ericksen' s problem" in t he lit erature.? We shall consider som e of these
solut ions in the succeeding sect ions.

6 .3 Homogeneous Stretch

Consider a homogeneous deformat ion for which the deformation gradient F

is a constant te nsor,
x = F(X - X 0) + x o ,
where X 0 and X o denot e t he origins in the reference and the deformed con-
figurations, resp ectively. Since F is a constant te nsor the stress T(F) is also
constant . Therefore, in the absence of external body force b, the equilibrium
equation (6.18) is identically sat isfied and the boundar y condit ion (6.14) can
be fulfilled by suitabl y applied sur face forces on the bod y. For incompressibl e
materials, the equilibrium equation (6.19) can also be satisfied by a constant
pr essure field p . Therefore, a hom ogeneous deform ation is a controllable de-
format ion for any elast ic body with no body force. Moreover , it is a univers al
soluti on of elastic mater ials , since it do es not depend on any particular form
of const it ut ive equa tions.
2 Man y such solutions are disc ussed in [71]. See also [1 ,4, 76].
158 6. Isotropi c Elastic Solid s

6.3.1 Uniaxial Stretch

We cons ider an isotropic elast ic body subject t o a homogeneous stret ch of

t he form
x = )'1X, Y = A2Y , Z = A3Z , (6.20)
where AI, A2' A3 are constants. Then the component of F is given by

and if we writ e the st ress t ensor as

T = 80 i +7, T = -pi +7,

for compressible and incompressibl e bodies, resp ect ively, t hen

o
A~
o
where t he coefficient s 8i = 8 i(AI , A~ , A5) and for incompressible bodies,
A1A2A3 = 1. In t he following we shall consider the cases of uni axi al and
biaxi al st retc hes .
For uni axial st retch in the x-direct ion, the lat er al st resses must vanish.
Therefore, for compressible bodies,

T ( yy ) = 8 0 + 8 1 A~ + 8-1A22 = 0,
(6.21)
T (z z ) = 8 0 + 81 A5 + 8 _1 \3 = o.
2

Taking the difference between T(x x) and T (yy ) , we then obt ain the axial st ress,

(6.22)

More over , from T (y y ) - T (z z ) , we also have

which implies a symmetric solut ion A2 = A3' and perhap s an asy mmet ric
solution A2 :f. A3' The asy mmet ric solut ion would be impossibl e if t he E-
inequ alities, 8 1 > 0 and 8 - 1 ::::: 0, are valid.
If 8i = 8i ( AI, A~ , A5) are known, the equat ions (6.21) may be solved for A2
and A3 in t erms of AI. Then , (6.22) det ermines the axi al st ress T (x x ) uniquely
in te rms of t he axi al exte nsion AI. On the other hand , since the stress T (x x )
an d the extensions AI, A2 are measur able quantiti es, t he relation (6.22) may
be used for t he experime ntal determination of 8 1 and 8 - 1, while 8 0 ca n be
det ermined dir ectly from (6.21).
 6.3 Homogeneou s Stretch 159

For incompre ssible bodies, since Al A2A3 = 1 and if, in addit ion, we
assume A2 = A3, then we have

The axial stress T (xx) becomes

For Moon ey-Rivlin material , it asserts a linear relation between Al l and

T(xx)/(Ai - Al l) . Such a relation is verified expe riment ally for rubber within
mod erate strain s. The hydrostatic pr essure p , which helps t o main t ain t he
lateral surface free of st ress, is given by

6.3.2 Biaxial Stretch

We consider an incompressible body subject to homogeneous stret ches in the

x- and y-dir ect ions only, hence
1
T (z z ) = 0, and A3 = AI A2 '

which impli es that the undeterminat e pr essure has the valu e,

The st resses T(xx) and T(yy) are given by

T(xx) = (Ai - A2\2) (SI - S-I A~),

I 2

T(yy) = (A~ - A2\ 2)(SI- S-IAi),

I 2

where s, = si (Ai, A~) . Since cross-s ect ional areas change aft er deformation ,
it is mor e convenient to measure Piola-Kirchhoff st resses as forces per unit
reference area. From (2.84), T" = JTF - T , t he Piola-Kirchhoff stresses are
given by
160 6. Isotropic Elastic Solids

t tt tt tt tt tt
+- ,------------. ~
+- ~

+-

+-
+-
+- ' - - - - - - - - - - - ' ~

F2
Fig. 6 .1. Biaxial st retc h

If we consider the body as a square sheet of unifo rm thickness, subject ed

to biaxi al stretching , then the equilibr ium of the sheet must be maint ain ed
by the dist ributed forces along the four lateral surfaces (Fig. 6.1) . The total
forces on the lat er al sur faces in the x- and y-directions are given by F l
AT",(x x ) and F 2 = A T ", (yy ), or

Fl = A(81 - 8- 1AD(AI - A3\2) '

1 2
(6.23)
F2 = A(81 - S-I Ai) (A 2 - A2\ 3) ,
] 2

where A is the reference cross-sectional area.

In biaxi al stretching expe riments, for som e pr escribed values of ext en-
sions A] and A2' the required axial forces F] and F2 can be measured . There-
fore, from the above equations, t he mat erial par am et ers 8 ] and 8- 1 ca n be
det ermined as funct ions of (A] , A2). On e may also determine the st ored en-
ergy fun ction 'l/J (I B,IIB ) from the relation (6.5), by suitably cont rolling t he
exte nsions A] and A2 so as to main tain eit her constant I B or constant II B
while measuring the ax ial forces.

6.4 Symmetric Loading of a Square Sheet

Now, consid er the body as a squar e sheet of rubber, load ed on four lateral
surfaces with equa l forces, F l = F2 . Intuitively, one would expec t t he rubber
sheet to rem ain square at leas t if the forces are not t oo lar ge. However , not
onl y shall we see that it may also become a rect angular one, but also t he
square one is unst able as long as the loading is large enough.
 6.4 Symmetric Loading of a Square Sheet 161

Let us consider rubber as a Mooney-Rivlin material , with const ant s

Sl > 0 and 8- 1 < 0, and denote h =
- S- l/S l . Then from (6.23) t he condition
F 1 = F 2 can be rewri tten as

(6.24)

which immediate ly gives the symmetric solution, )'1 = A2. Since AI , A2'
and h ar e positive qu antities, no other solut ion exists if hAl A2 < 1, which
incid entally rul es out the possibility for an asymmet ric solut ion in t he case
of neo-Hookean mat erials (h = 0) .

o !l

7 7

s s

.~

(1)) FI > F2

Fig. 6.2 . Biaxial st retching of a squa re sh eet

The asy mmetric solution may exist and ca n be found from t he equa t ion

(6.25)

For such a solution, in genera l, Al and A2 ar e different, and the square sheet
becomes rectangular after st retching . In Fig. 6.2(a) we have plotted Al again st
A2 for a typi cal valu e h = 0.1 (say) . The straight line represents the symmet ric
soluti on Al = A2' and the curve corresponds to the asy mmet ric solut ion
Al -I=- A2· The point, the bifurcation po int B , wher e the two intercept each
other has the value AB = Al = A2 = 3.1685.
In t he linear theory of elast icity, solutions of traction boundary valu e
problems, (6.18) and (6.14) , ar e unique up to a rigid deformation , and hence
t he st ress fields are unique. However , for large enough forces, there are two
possible solutions with different st ress fields , corres ponding to the one with
Al = A2 and the other with Al -I=- A2' to the t raction boundar y valu e problem
of a square sheet with biaxial symmetric loading. Therefore, it is clear that ,
unlike t he linear theory, one should not expect unqualified uniqueness of
162 6. Isotropic Elastic Solids

solu tions to the boundary valu e problems for theories of finit e deformation ,
in general.
When a squar e sheet is slowly st ret ched with equal forces F 1 = F 2 , it
will remain square when Al < AB. As the stretch Al reaches the valu e AB'
on e may ask whether the sheet will remain square or become rect an gular, the
two possible solut ions. In other words, which one is a stable solution?
In order to answer this qu estion , we shall consider an imp erfect problem
for which the square sheet is stretched with unequal forces F1 = (1 + E)F2 ,
namely, F 1 is slightly larger than F 2 • From (6.23) we obtain the following
equat ion,

This is a relation between A2 and Al for a given valu e of E. The equ ation ca n
be solved numerically and the results are plotted in Fig. 6.2(b) for E = 0.005
and h = 0.1. It is interesting to note that there ar e also two solution curves,
but they do not intercept each other . They approach the two int ercepting
curves for F 1 = F 2 when E approaches zero. Therefore, in stret ching a square
sheet by equ al forces, it is clear that beyond the bifurcation point B, any
t iny imbalance of forces will ca use the sheet to change into a rectan gular
shape with the long side along t he dir ection of the slightly larger force, i.e.,
Al > A2 for F 1 > F 2. This is what would happen most probabl y because it is
extremely difficult to keep the forces "exactly" equal by any practical means .
Therefore, we conclude that beyond the bifurcation point, the symmet ric
solution is unstable.

6.4.1 Stability of a Square Sheet

The st ability of a squar e sheet under symmetric biaxial stret ching may also
be analyzed via a t hermodynamic crit erion following the pro cedure discussed
in Sect . 5.5.
Con sider a body in a region V at a uniform constant temperature and
free of ext ernal supplies, t hen we have from the energy equat ion (2.68) ,

~lp(E+~x ,x)dV+
dt v 2
r (q-Tx) ·nda=O,
Jav
(6.26)

and the ent ropy inequ ality (5.4),

~lp7]dv+-(}l
dt v
r
Jav
q ·nda20. (6.27)

By eliminating the heat flux , we obtain

~ 1
dt v
p('lj! + ~x . x) dv -
2
r x · Tn da <
Jav
0, (6.28)
 6.4 Symmetric Load ing of a Squa re Sheet 163

where 'ljJ = E - {}ry is the free energy. If t he region occupied by the body in
t he reference state is denot ed by VI< then the above conditio n can be written
in t he reference st ate as

(6.29)

Let t he region VI< occupi ed by t he squ ar e sheet in the reference st at e be

given by 0 < X ::::; 1,0 ::::; Y ::::; 1, and -D ::::; Z < D. We have the following
boundar y condi t ions:

and from (6.20),

xlx =o = 0, xl X= i
='\ie x ,
XI = 0, x IY =i = ).2e y ,
y=o
for which the biaxi al st retching is regarded as a continuous pro cess, while
F, = 2DTI< (xx) an d F 2 = 2DTI< (yy) are the total forces prescribed at the
four lat eral sur faces (see Fig. 6.1). Therefore, it follows that

where Ai and A 2 are t he later al sur faces of t he region at X = 1 and Y = 1,

respectively. Therefore, assuming t he pro cess is qu asi-st atic (with negligible
acceleration), t hen from (6.29) we have

which, upon int egration gives

since t he deformation is homogeneous . T herefore , we can define t he availabil-

ity function of the square sheet as

(6.30)
164 6. Isotropi c E las t ic Solids

where Fl = FI/2D e F2 = F2 /2D , so that the last relation becomes

dA < 0. (6.31)
dt -
T his allows us to establish a stability criterion for th e deform ation of the
square sheet. The availability A(t) is a decreasin g function of time. Therefore,
if we assume the deformed state characterized by t he stret ches (>'1 , 'x2) is a
stable equilibrium st ate, then any small perturbation from this state will
event ually return to this st at e as t ime t ends to infinity. Suppose t hat such a
perturbation is represented by a pro cess ()\1 (t) , A2(t)) , it follows that

and hence by the conditi on (6.31) the availability A(t) = A(Al(t), A2(t)) will
at tain its minimum at ('xl, 'x2). This crite rion is equivalent to the following
conditi ons:
(6.32)

and t he matrix

is positive semi-definite or, equivalent ly,

8 A
8 Ar
2
I 8 AI
2
OA~
E E -
8
2A
1
2
OA l OA2 E 2: 0,
(6.33)

where E denotes the evaluation at the equilibrium state ('xl, 'x2).

Now we have t he free energy 'l/J = 'l/J (IB ,II B ), where from (6.20)

By (6.5) and (6.30), the condit ions (6.32) become

(6.34)

where t he equa t ions are evalua t ed at the equilibrium state and t he overhead
bar s are suppressed for simplicity. Not e that they are identi cal to th e relati ons
 6.4 Symmetric Loading of a Squar e Sheet 165

O l--------------~Ioo<:::-------I

-2
-4 l....-_ _...L-_ _.....L..._ _---'-_ _ - - - l _" - -_ ..l--_ ------l

1 1.5 2 2.5 3 A8 3.5 4

A
Fig. 6 .3. f(A , A) - sy m metric solution

(6.23) . Meanwhile the condit ion (6.33) leads to

3
1 + >.4>.2 + h ( >.4
1
+ >'22) ;:::: 0, (6.35)
1 2 1

which is identically satisfied since h = -3 - d 31 > 0, and

3 3 3 3
(1 + >.4>.2
1 2
+ h( >.41 + >.D) (1 + >.4>.2
2 1
+ h( >.42 + >.i))
- C3 3+ 2h>'I>'2f ;: : o.
(6.36)
2>.
1 2

Let t he left-hand side of t he relat ion (6.36) be denoted by f(>'I' >'2), t hen we
have
(6.37)
which is the condit ion for an equilibrium st ate (>'1, >'2) to be st able.
For symmet ric loading, putting F1 = F2 , from the equat ions (6.34) we
recover the relation (6.24) , which gives t he symmetric solution >'1 = >'2 and
the asy mmet ric solu tion >'1 =I- >'2 given by (6.25) .
We have plotted the fun ction f(>', >') against >. for t he symmetric solut ion
in Fig. 6.3 for h = 0.1. This shows that t he function f(>', >') is posi tiv e for
>. < >'B = 3.1685, which corresponds to the bifurcation point in Fig. 6.2(a) ,
and therefore, according to the condit ion (6.37) , t he symmet ric solut ion is
stable. However , beyond the value >'B, the fun ct ion f(>', >') becomes negativ e
and hence the square sheet is no longer stable.
For an asymmet ric solution >'1 =I- >'2 , from the condition (6.25) one can
solve for >'2 in t erms of >'1 so that >'2 = g(>'d and hence f(>'I ,g(>'d) becomes
166 6. Isotropic Elastic Solids

Gn-- .--- --r....--- ...--- --,-- - ,--- ---r-- - .--___.--,

0 '--- -'-- ......-'-""""-.........- ........---'-------'---'---.....

1 2 3AB 4 5 6 7
)q

Fig. 6.4. f(Al , A2) - asymmetric solution

a function of Al only. Doing this numerically, we can easily verify the condition
(6.37) by plotting the function f(AI' A2) against AI , as shown in Fig. 6.4, from
which we conclude that the asymmetric solution is always stable. Hence , a
square sheet turns into a rectangular one when symmetric loading gradually
increases beyond the bifurcation point."

6.5 Simple Shear

We consider an isotropic elastic body subject to a deformation of simple shear

given by (see Example 1.2.1, Fig. 1.3)

x =X + KY, Y = Y, z = Z,

where the amount of shear K is a constant, we have the left Cauchy-Green

tensor

~ ~], (6.38)
o 1
and its inverse

3 See also the stability analysis in [32, 56] and more discussions in [18].
 6.5 Simple She ar 167

so that its principal invariants are given by

I B = 3 + K,2, II B = 3 + K,2 , III B = 1.

Since the simple she ar is a homogeneous deformation, it sati sfies the

equilibrium equation with no body force. The stress t ensor is a constant
tensor given by

(6.39)

where s, = si (3 + K,2 , 3 + K,2 , 1) = Si(K,2) . Note that simple sh ear is a volume-

pr eserving deformation and the above results are also valid for an incom-
pressible elastic body, provided that the material parameter So is replaced by
- p, the undeterminate pressure.
The shear stress T (xy ) on the sur face, Y = Yo, has the valu e

(6.40)

where
(l(K,2) = Sl (K,2) - 8- 1 (K,2)

is called the shear modulus of the material. For a small K" then

where /1 = p.(O) is the classical sh ear modulus. Therefore, any discrepancy

from the classical result for the shear st ress is at least of third-order in the
amount of she ar K, . Mor eover , unlike elast ic fluids , for which (l(K,2) = 0 for any
K" elast ic solids with this trivial property will be ignored in our discussion .
Therefore , by (6.40) the shear stress T (xy ) does not vanish if K, -I- o.
For simple shear, from (6.11) and (6.38) we have the following univers al
relation ,
T (x x ) - T (yy ) = K,T(xy ) , (6.41)
which can be checked immediately from (6.39). If we denote the normal stress
on the slanted sur face of the block , corresponding to the plane in the reference
state X = X o, by N (see Fig. 6.5), the universal relation (6.41) can be
rewritten as (see Exercise 6.5.1)

(6.42)

From this expression, it is clear that in order to effect a simple she ar on a

rectangular block, besides shear st resses, normal stresses must also be applied
168 6. Isotropic Elastic Solids

on the surfaces of the block, since from the above relation the two normal
stresses can not vanish simultaneously or even be equal to each other unless
there is no shear at all. Moreover, the normal stress difference is a second-
order effect in the amount of shear according to (6.40) and (6.42) . Therefore,
the normal stress difference is more significant than the discrepancy in the
shear stress as an indication for the departure from the classical theory. The
existence of a normal stress difference is usu ally known as the Poynting effect
or simply as the normal stress effect.

y
r(yy~

T
T (x y )

1 Fig. 6.5 . Simple shear

x

For compressible bodies, there is another way of looking at the impossi-

bility of having a simple shear deformation by shear stress alone. From (6.39)
the mean hydrostatic pressure is given by
1 1
- 3 trT = -(so + Sl + 3-d - 3(Sl + s-d,,; 2 .
If we assume that the reference configuration is a natural state, namely, no
stress when x = 0, then from (6.39)
so(O) + Sl (0) + 3-1 (0) = 0,
and hence the mean hydrostatic pressure is of second-order in the amount of
shear. Therefore a non-vanishing hydrostatic pressure (or tension according
to the sign) is needed to effect a simple shear in a compressible isotropic
elastic body. This is called the Kelvin effect.

Exercise 6.5.1 Let the normal stress and shear stress on the slanted
surface, corresponding to the surface in the reference state X = X o, of
the block be denoted by Nand S, respectively (see Fig . 6.5) , show that
r:
N = T (yy) - 1 + ,,;2 T (xy ) ,
1
S = --2T(xy ) '
1+,,;
 6.6 Pure Shear of a Sq uare Block 169

For infinitesimal 1'\" if the second-order terms in I'\, are neglected, it follows
t hat N c::' T (yy ) and S c::' T (xy ) .

E x ercise 6.5 .2 Consider a torsion an d extension of a cylinder given by

the deformation function (1.16),
1
r = v'aR, () = e + TZ, Z= - Z.
a
1) Show t hat it is a controllable universal solution for incompressible
isotropic materials.
2) Determine the stress fields of a hollow cylinder free of stress at t he
outer surface for Mooney-Rivlin materials.
3) Discuss the presence of the normal stress effect.

6.6 Pure Shear of a Square Block

We have noticed that by app lying shear stresses alone on the surface of a rect-
angular block, the body will tend to contract or expand if normal st resses
were not supplied properly. To examine such changes quantitatively, we con-
sider a deformation that consists of a homogeneous stretch followed by a
simple shear ,
x = A1 X + I'\,A2Y, Y = A2Y , Z = A3Z.
Since t his is a homogeneous deformation , it is a controllable universal solu-
tio n for an elastic body. The deformation gradient relative to t he Cartesian
coordinate system is given by

(6.43)

T he left Ca uchy-Green tensor is given by

Ai + 1'\,2 A~ I'\,A~ 0]
[B (ij )] = I'\, A~ A~ 0 , (6.44)
[
o 0 A~
and its inverse
1 I'\,
0
A21 - A2
1
I'\, 1 1'\,2
[(B-1) (ij )] = - A21 A22 + A21 0
1
0 0
A23
170 6. Isotropic Elastic Solids

For isotropic elastic body, the stress tensor T (ij ) can be calculated from (6.2)
for a compressible body or from (6.4) for an incompressible body. In partic-
ular, the shear stress on the surface, Y = Yo , is given by

(6.45)

where s, = s, (I B, II B, III B). This holds for either compressib le or incom-

pr essible (III B = 1) bodies. Moreover, from (6.44) , the universal relation
(6.11) takes the following form,

(6.46)

Unlike the case of simple shear discussed in the previous section, for a
fixed n ; it is now possible to det ermine the three constants AI , A2 ' and A3
in such a way that three normal stresses vanish on the surface of the block.
In the case of an incompressib le body, the three conditions for vanishing
normal stresses can be used to det ermine AI, A2 and the pr essure p, while
the condition of incompressibility, A1A2A3 = 1, determines A3'

L
y
",>'2 82

f,
>'2
e €2
B

N1
€1 A
x

>'1
€~
Fig . 6.6 . Pure shear

We consider a square block, with sides of unit length, and introduce a

new basis in the deformed configuration (see Fig. 6.6) defined by

and its dual basis

 6.6 Pure Shear of a Square Blo ck 171

Note that le 1 1 = le21= VI + ",2 and t he base vectors e1 and e2 ar e normal

to the deformed sur faces of X = X o and Y = Yo , resp ectively. Therefore,
normal st resses N 1 and N 2 on these surfaces are given by

(6.47)

where we have denot ed the contravariant components of T relative to the new

j
basis by Tij = ei . Te according to (A .5). The component s Tij can be ob-
tained from the Cartesian component s T(mn) by the following transformation
rul e (see (A.14)) ,
[T ij ] = [Mmi r T[T(mn)][Mn j ]- l , (6.48)
where [M k i] is the transformat ion matrix of chan ge of basis from {ed to
{ed, from (A.H) it is given by M k i = ek ' e i or

We obtain

T (XX) - 2",T(xy ) + ", 2T(yy ) T (xy ) - ",T(yy )

[Tij] = T (xy ) - ", T(y y) T (yy ) (6.49)

[
o 0

From (6.47) the vani shing of normal stresses N 1 = N 2 = N 3 = 0 would re-

quire T (yy ) = T (z z ) = 0 and T (x x ) - 2",T(x y } = O. The last condition combined
with the relat ion (6.46) leads to"

(6.50)

It is interesting to point out that this relat ion is also a universal relation for
elastic bodies and it admits a very simple geometric int erpret ation, nam ely,
OA = OB , as shown in F ig. 6.6. Furthermore, from (6.49) t he stress tensor
relative to t he product basis {ei ® ej} t akes t he following simple form ,

where from (6.45) and (6.50), the shear stress T is given by

Ai 1
T = T (xy ) = "'(81 -1 +-'"2- 8-1\2)
A1
' (6.51)

4 See a lso t he derivation in [61] and more d iscussion on pure shear in [48].
172 6. Isotropic Elastic Solids

Besides the vanishing of normal stresses, the shear stresses on the surfaces,
X = X o and Y = Yo , are equal, i.e., 8 1 = 8 2 = T. Such a state of stress is
called a pure shear. Thus we have seen that to effect a state of pure shear on
a square block , it is only necessary to apply equal shear stresses on the four
surfaces. The amount of shear I'\, and the stretches Al and A2 are adjusted
in such a way that the length of the four sides remains the same. A square
block becoming a rhombic block is also what one would expect intuitively in
a pure shear.
To determine the stretches for a given amount of 1'\" explicit constitutive
expressions would be needed. As an example, for a neo-Hookean material
given by (6.9) , we have the contravariant components of the stress

fij = _p gi j + 81 jjij ,
where the metric tensor gi j = ei . ej is given by

-I'\,
1
0]
0 ,
o 1

and the contravariant components of B can be calculated similar to (6.48),

Therefore, by putting r» = fn = f33 = 0 and by the incompressibility

condition AIA2A3 = 1, we obtain
p = 81(1 + 1'\,2)-1 /3 ,
Al = (1 + 1'\,2)1/3,
A2 = A3 = (1 + 1'\,2) -1 /6.

Moreover, from (6.51) the shear stress T has the value,

T = 811'\,(1 + 1'\,2) -1 /3 .
Note that the value of Al is greater than 1, while that of A2 is less than 1.
Therefore the block lengthens in the x-direction and shortens in the y- and z-
directions. This is a quantitative description of the Poynting effect mentioned
in the case of simple shear.

Exercise 6.6.1 The relation (6.10) can be written in component form

as
Tij 9jk B k i = Bij 9jk T k l.

Show that the universal relation (6.50) follows directly from this equation
relative to the basis {eJ and the conditions i» = f22 = O.
 6.7 Finite Deform ation of Spher ical Shells 173

Exercise 6.6 .2 Show that the right stretc h tensor U and the rotation
te nsor R of t he deformation gradient P in (6.43) with the condit ion
(6.50) are given by

)'1
)'1
sin 0
cos O [R(i{3 )]
cos O sinO
= -sin 0 cos 0 0 ,
0]
o [ o 0 1

where 0 is half the angle between t he base vectors e2and e2 (see Fig. 6.6)
or
t an 20 = n:
Verify that the vectors (1,1 ,0) and (-1 ,1 ,0) are eigenvect ors of U and
hen ce, this deformation is a biaxial stret ch along two diagonals of the
undeformed square (such a deformation is often called a "pure shear" in
the liter ature) followed by a rotat ion that brings the base back to the
hori zontal position. Note the difference between this and t he results for
t he simple shea r in Example 1.2.1.

6.7 Finite Deformation of Spherical Shells

We shall consider some problems of spherical shells. In t erms of spherical

coordinates (R, 8 , p) and (r, 0,¢) in the reference and the deformed config-
ur ations, given a deformation of the form ,

r = r (R ), 0 = 8, ¢ = c P, (6.52)

where c is a constant , the matrix of the deformation gradient is

[pial =
r
0
f
01 00] ,
[ o 0 c

which in t erms of phy sical components becom es

[ ~ ~ ~] .
r
R
o ]
r sinO
o c .
8
o 0 c-
R Slll R
The matrix of the left Cauchy- Green tensor is t hen
174 6. Isotropic Elastic Solids

and its inverse,

2 2
1 R R ]
[B0])]= diag [ r'2' --;:2' c2r2 '
are functions of r only.
For incompressible isotropic materials, det F = 1, and shells for which
T (oo ) = T (<!><!» , we have from the above expressions and (6.4) that

r2
c R2 r' = 1,

Hence we obtain
c = ±1. (6.53)
wh ere A is an integration constant.
It is easy to verify that the deformation given by (6.52) and (6.53) is
a controllable universal solution for incompressible isotropic materials. The
equilibrium equation (6.19) in the spherical coordinate system (r,B,¢) now
reads (see (A.83))
8T(rr) 2
->:>- = --(T(rr) - T(oo»),
or r
(6.54)
8p 8p
()(} = 0, 8¢ = O.

From this and (6.4) we obtain

(6.55)

Hence we have

and

(6.56)

All the other stress components are zero . Moreover, the material parameters
81 and 8 -1 ar e functions of r ,

where
(6.57)
 6.7 Finite Deformation of Spher ical Shells 175

For Mooney-Rivlin materials, with cons tant S1 and 8- 1 , (6.55) can be

int egrated. Let ~ = r f R, t hen by (6.53)

4 dr
d~ = ( ~ - c~ )- , (6.58)
r
and hence, f (r) becomes

f (r) = 2 J(e - C 4 )(S1 - 8-1 e )(~ - C~4 ) - 1 d~

= -2 J(l + ce )h - 8- 1e)C5d~
1
= S1("2 C 4 + 2cC 1) - 8- 1(C 2 - 2c~) + K.
where K is an int egration const ant . Therefore , t he stress fields for Mooney-
Rivlin materials are given by

(6.59)

6.7.1 Eversion of a Spherical Shell

For t he case c = - 1, we have

This deformation is an eversion that turns the shell inside ou t . Since for
Ro < R 1 we have A - R~ > A - Rr , and henc e r o = r(Ro) > r 1 = r(Rd , in
other words, the inn er sur face becomes the outer sur face aft er deformation
and vice versa . Moreover , for t his to be physically possible, it is necessar y
that r1 be positiv e, or th e constant A must satisfy

(6.60)

To effect such a deformation for a complete spherical shell, it would be nec-

essar y to make a cut some where first and then join together again after t he
eversion .
If we require t hat the shell be free of traction at the inn er and t he outer
sur faces in the everte d state , T (r r )(rd = T (r r ) (ro) = 0, t hen (6.55) impl ies
4
j
2 2
r ! (
-r- -R4 ) ( S1 - S 1 -r ) -dr -_ 0 (6.61)
ro R2 r - R2 r '
176 6. Isotropic Elast ic Solids

where the materi al paramet ers 81 and 8 -1 are fun ctions of R = (A - r 3)1/3
and r .
The constant A is to be det ermined as a solution of t he above equation
for a tracti on-free everted state. We can show that such a solut ion exists for
any incompressible isotropic materials if the E-inequaliti es (6.8) hold and the
sh ell is not too thick. Indeed , by the mean value theor em for definit e integrals,
we have

from (6.61) , for some value f su ch that r o ~ f ~ rl' Since 81 > 0 and 8 -1 :::; 0,
it implies
f2 R(f) 4
R(r) 2 - ~ =0 .

The only real positive root of this equat ion is R(f) = i', or (A - f3 )1/3 = i',
yielding

and hen ce A must satisfy

or equivalent ly,
2R~ :::; A :::; 2Rr.
Since, from (6.60) , the constant A mu st also be greater t han Rt , we conclude
that the sh ell can always be everted if Rt < 2R~ . In other words, if t he
t hickness (R 1 - R o) is less t han (ij2 - 1) times (or about 26% of) t he inner
radius R o, it is po ssible t o turn the shell inside out freely for any incompress-
ible isotropic elastic materi als . It has been proved by Ericksen [17] t hat the
eversion is possible for Mooney-Rivlin materi als.
We have seen here another example of non-uniquen ess of a traction
boundary value problem , that for a spherical sh ell with null traction, there
are at least two pos sible solutions, namely, the undeformed state with zero
stresses and the everted equilibrium st ate, with non-vanishing stresses in the
interior.

6.7.2 Inflation of a Spherical Shell

Consid er the case c = 1,

(6.62)

For a shell of radius Ro :::; R :::; R 1, t he inner and the outer surfaces mu st be
 6.7 Finite Deformation of Spherical Shells 177

maintained by suitable pressures, i.e.,

(6.63)

where ro = r(Ro) and rl = r(Rd are the inner and the outer radii of the
shell in the deformed configuration. Let [P] = Po - PI denote the pressure
difference between the inner and the outer surfaces, then

(6.64)

We define the thickness parameter a and the expansion ratio A by

rl
and A= R ' (6.65)
1

respectively, with a > 1 and A > 1. We have

rl = aARo, (6.66)

where the second relation is obtained by the elimination of the constant A

from (6 .62h . Therefore, for a given thickness parameter a, the pressure differ-
ence [P] is a function of the expansion ratio A. For Mooney-Rivlin materials,
this function can be written out explicitly from (6 .59) . Such pressure-radius
relations are shown in Fig. 6.7, for a typical value s., d 81 = - 0.1 and several
values of a .

1.5

[p~
-;;v
. JRl

0.5
a = 1.500 ......
u = 1.200 -
u = 1.001
O L.....L._ _---L_ _ ---I. ' - -_ _..L...-_ _....L..-_ _....J

2 3 4 5 G 7
A

Fig. 6 .7 . Pressure-radius relation

We notice that the pressure-radius curves for a thick shell, a = 1.5, a

thickness, D = R 1 - R o, half the value of the inner radius, and for a thin
178 6. Iso tropic Elastic Solid s

shell , a = 1.001, a thickness one thousandth of that of the inn er radius,

are cha rac te rist ically similar. After a st eep rise, the cur ve reaches a local
maximum and t hen decr eases to a local minimum before it slowly rises again.
Such a non-monotone func tion may usu ally lead to certain un st abl e behavior,
which we shall consider in the next section.
For t hin shells, t he express ion (6.64) for the pressure difference can be
approx imate d by the Taylor expansion

where d = r 1 - ro. By using (6.54h through (6.56) , and by neglecting t he

higher-order t erms, this last equa t ion takes t he form

wh ich can also be written as

D R R7 r2
[P] = 2- (- - -7) (8 1 - 8_1- ) . (6.67)
R r r R2
In t he ab ove expression , since t he shell is thin , we have dropped the subindex
o in both rand R for simplicity, and employed t he relation r 2 d ~ R 2 D for in-
compres sibility. Note that the derivation of the pr essure-radius relation (6.67)
for a t hin spherical shell is valid for any incompressibl e isotrop ic mater ial.
This relation has been used as t he const it utive equation for spherical rubber
balloons (for a different derivation see [47, 55]) and its cur ve for Moo ney-
Rivlin mat eri als is almost ide nt ical to t ha t sh own in Fi g. 6.7 for a = 1.001.
Sin ce T (Be) = T (q,q, ) the tange nt ial st ress in the sh ell is a pure tension .
Let the surface tension (p er un it length in t he inflated state ) be de not ed by
a, then from t he classical formula ,
a
[p] = 2-;:- ,
by comparison with the rela tion (6.67) , t he sur face t ension for a spherical
rubber balloon can be expressed as

a = D (1 - ~:) ( 8 1 - 8- 1 ~: ) .
Note that even though the pressure cur ve is non-monotone, one ca n easily
verify t hat the sur face tension is a monotone increasin g fun ction of radius if
the E-inequa lities (6.8) hold.
 6.8 Stability of Spherical Shells 179

Exercise 6.7.1 Consid er the deformation of a cylinde r given by

1
r = r(R), 0 = e, Z = -Z,
c
where c is a constant , for incompressible isotropic elast ic mat erials.
1) Show that incompressibility implies that r(R) = (cR 2+ A )I/ 2, where
A is a constant .
2) Show that it is a controllable universal solut ion and det ermine the
stress fields.
3) Discuss t he inflation and the po ssibility of eversion for hollow cylin-
ders .

6.8 Stability of Spherical Shells

The pr essure-radius relation for the inflation of a spherical shell, as shown
in Fi g. 6.7, is non-monotone, a feature that may lead to unstable behavior.
We shall investigate the stability of thin spherical shells under two different
boundary cond it ions .

Fig. 6.8. Spherical shell

Let the spheri cal shell lie in a region V between avo and avl , where
Vo and VI ar e spherical balls with radii ro and rl , resp ect ively (Fi g. 6.8) .
Assume that the shell is subject t o uniform temperature 0 and free from
ext ernal sup plies. Then with t he same consi deration leading to (6.28) , we
have
r
~ p('¢ + ~v . v)dv -
dt Jv 2 J&V
rv · Tnda < 0, (6.68)

where '¢ = E - 0TJ is t he free energy. Moreover , from the boundar y condit ions
180 6. Isotropic Elastic Solids

(6.63) , we have

r
lev
v . Tn da = Pa r
i;
r v . n da
v · n da - PI
i; (6.69)
= [P] ~ r dv - PI ~ r dv = [P] dVa ,
dt lvo dt lVI- vo dt

In this derivation , we have used the relation (2.10) and the incompressibility
condition that the volume of V = VI - Va does not change. We have also
denoted Va as the volume of the spherical region Va .
Combining (6.68) and (6.69) , we obtain for quasi-st atic problems (with
negligible accelerat ion)

-d
dt v
J dVa < O.
p'l/Jdv - [P]-
dt -
(6.70)

Since the pressure-radius relations are essentially the same irrespective

of thickness, for simplicity in the following analysis we shall consider only
thin shells of Moon ey-Rivlin materials. The free energy ca n be expressed as

(6.71)

where I B and II B are given by (6.57) .

6.8.1 Stability under Constant Pressures

We shall an alyze t he stability of a spherical shell under pr escribed const ant

int ernal and exte rn al pr essures. In order to be abl e to maintain constant
int ernal pressure, we have tacitly assumed that a suitabl e device is provided,
such as a tube int erconnecting the interior to a const ant pr essure chamber.
However , as long as the interior can be maintained at const ant pr essure, such
a device is irr elevant to the problem .
The relation (6.70) can now be written as

Therefore, we can define the availability A , a monotone decreasing function

of time, for a thin shell as
 6.8 Stability of Spherical Shells 181

According to the st ability criter ion, if r = r corresponds to a stable

equilibrium st at e, then A( r) must at tain its minimum at r = i . The necessary
and sufficient conditions for this to happen are

From the expressions (6.71) and (6.57) for the free energy, we obtain afte r
simple differentiations,

where
7 5
D { 8 1 (R-
F(r) = 2- -
R
-7) -
r
8- 1 ( - -
R
-5)
}
. (6.72)
R r r R r
Therefore , we have the following st abili ty condit ions:

[P] = F(f) , d~~) ~ O. (6.73)

The first condition is merely the pressure-radius relation (6.67) , while the
other condition impli es that the pr essure-radius curve F(r) must have a
positive slope at a stable equilibrium st at e.
Therefore, the branch with negative slope in the pr essure-rad ius cur ve
shown in Fig. 6.7 (with a = 1.001 for thin shells) corresponds to uns t abl e
equilibrium st ates under pr escribed constant int ernal and external pr essures.
A simil ar ana lysis for thick spherical shells also leads t o t he same conclusion
for stability [42] .

6.8.2 Stability for an Enclosed Spherical Shell

For an enclosed spherical shell, if we assume t ha t the air inside the shell is an
ideal gas, then the int ernal pressure is inversely proportional to the int erior
volume un der isothermal conditions (Boyle's law) . Therefore, for the pr esent
case we suppose t hat t he exte rnal pr essure PI is const ant while t he int ernal
pr essure Po is given by
k
Po = Vo ' (6.74)

where k is a const an t .
We can rewrite the condit ion (6.70) as

~ {4Jr 1:' p'l/J r

2dr-klog
vO+ PI VO} :':::: 0.
182 6. Isotropic Elastic Solids

Therefore, for a thin shell, we can define the availability A as

A = 41rr 2dp 7/J(r) - k log Vo + PIVo·

Since Vo = 11rr3, we obtain

dA
dr = 41rr 2(F(r) - Vk +Pl ) ,
o
where F( r) is given by (6.72) . The equilibrium condit ion becomes
k
V - PI = Po - PI = F(f) , (6.75)
o
by (6.74) . This is again the pressure-radius relation (6.67) .
Moreover , the second derivative of A is
2
d A ( ) 2 (dF (r) k dVO )
dr 2 = 81rr F( r) - Po + PI + 41rr ~ + V02 dr .

Hence, the st ability condition requires that

2
d AI = 41rr 2 (dF( r) + ~po) I

tc
d r? dr r
r=r r=r (6.76)
~ 12"G d~~) + F(r) + p, :> 0,

where (6.74) and (6.75) have been used . By the use of (6.72) , aft er a simple
calculat ion, we obtain the expression
7 5
rdF(r)
- 4D
- - +F(r) = - - {81( R R )
- +2- - 8- 1( r + -R
2- ) } ,
3 dr 3R r r7 R 5 r
whi ch is always positive if the E-inequalities (6.8) hold s. T herefore, for this
case t he st ability condition (6.76) is identically satisfied at any equ ilibrium
st ate . In other word s, every equilibrium st at e is st abl e.
It is worthwhile to emphasize that t he st abili ty condit ions depend on
the par ti cular sit uat ion a body happens to encounter. For different sit uations
t he st ability condit ions are, in general, differ ent as we have seen in the ab ove
examples.P

Exercise 6 .8 .1 Proceed with a similar ana lysis to det ermine the equi-
librium and t he stability conditions for
1) thick spheric al shells subject to pr escribed constant pressures,
2) enclosed spherical shells of finite thickness.

5 A d iffer ent stability a nalysis for these examples can be found in Sect. 8.3, [55].
See a lso [56].
7. Thermodynamics with Lagrange Multipliers

7.1 Supply-Free Bodies

T he entropy principle based on the Claus ius- Duhem inequa lity has been
widely adopted in t he development of mod ern rational t hermody na mics , after
the fund am ent al memoir of Colem an and Noll [11]. T he main assumptions,
1
s= O r, (7.1)

while at least tacit in all classical t heories of cont inuum mechani cs, are not
par ti cularl y well motivat ed for mat erials in general. In fact, the relation (7.1h
is known t o be inconsist ent with the result from the kinetic theory of ideal
gases, and is also found to be inappropriate to account for thermodynam-
ics of diffusion . Other assumptions an d formul ations of the second law of
t hermody namics have been proposed elsewhere, " In [50] Muller proposed to
abandon t he assumpt ion (7.1h by t reating t he ent ropy flux cP and the energy
flux q as ind ep end ent constit ut ive quantities and hence leaving the ent ropy
inequ ality in it s general form ,

pi] + div P - ps ::::: o. (7.2)

Bu t , in addition, he also proposed t ha t if t he body is free of extern al supplies,

the ent ropy supply must also vani sh , i.e.,

s=0 if r = 0 and b = 0,
which is certainly much weaker t ha n the ass umption (7.1h . He argued t hat
since constit ut ive prop erti es of a mat erial should not dep end on external
supplies, in exploit ing const it ut ive restrictions it suffices to consider only
supply-free bodies. For supply-free bodi es, the ent ropy inequality becomes

prj + div cP 2: o. (7.3)

It has been shown that the ent ropy pr inciple imposes severe restricti ons
on constitut ive functions and the exploit at ion for such restricti ons based on
I For more genera l discu ssion s a nd historical not es on the second law of t he rmo-
dyn am ics, see [69].

I.-S. Liu, Continuum Mechanics

© Springer-Verlag Berlin Heidelberg 2002
184 7. Thermodynamics with Lagrange Multipliers

the Clausius-Duhem inequality are relatively easy. On the other hand , the
exploitation of the entropy principle based on the general entropy inequality
(7.3) for supply-free bodies, first considered by Muller ([52, 53]) was much
more difficult and later its procedure has been improved greatly by the use
of Lagrange multipliers in a more systematic manner proposed by Liu [37] .
In this chapter we shall illustrate the procedure for exploiting the entropy
principle by the use of Lagrange multipliers for viscous heat-conducting fluids.
For convenience, we shall use component notations in a Cartesian coordinate
system for mathematical manipulations, which are often simpler.

7.2 Viscous Heat-Conducting Fluid

The determination of the basic fields {p(x , t), Vi(X, t), B(x , t)} are based on
the balance laws, which, for a supply-free body, can be written as

p + PVk ,k = 0,
PVi - Tik,k = 0, (7.4)
pi + qk,k - TikVi,k = 0.

We consider the viscous heat-conducting fluid , for which the constitutive

quantities, the symmetric stress tensor T ik, the heat flux qk, and the internal
energy € are given by constitutive equations of the form

C = C(p, Vi , B, B,i' vi,d .

The principle of material objectivity requires that the constitutive function

of an objective constitutive quantity must be independent of the observer. As
we have shown earlier (see for example, Sect . 3.2.2), this requirement implies
that the constitutive functions are independent of the velocity Vi, and depend
on the velocity gradient only through its symmetric part Dik . Therefore, they
can be written as
Tik = Tik(P, B, B,n' D mn),
qk = qk(p, B, B,n' D mn ), (7.5)
E = € (p, B, B,n' D mn ).
Moreover, the constitutive functions are isotropic functions .
The field equations are obtained by introducing constitutive equations
tn
(7.5) into the balance equations (7.4) . Any solution {p(x , t), Vi(X, t) , B(x ,
of the field equations is called a thermodynamic process in the supply-free
body.
Since our objective is to derive restrictions imposed on the constitutive
equations that are independent of external supplies, we need only consider
supply-free bodies without loss of generality. Accordingly, the entropy prin-
 7.2 Viscous Heat-Conduct ing Fluid 185

ciple requires t hat t he entropy inequa lity

(7.6)

mus t hold for every thermodyn am ic pro cess in supply-free bodies. T he spe-
cific ent ropy density 1] and t he ent ropy flux If>i are also const it ut ive qu antities,

1] = 1] (p, B, B,n, D mn) ,

(7.7)
If>k = If>k(p , B, B,n, D mn ).

T his const raint on t hermody na mic processes can be state d in a different

way, nam ely, t he fields that satisfy the entropy inequality are const rained by
the requirement t hat they must be solutions of t he field equat ions . We can
take care of this requirement by the use of Lagran ge multipliers much like
that in the classical problems of finding t he extrema with constraint s:
Method of Lagrange multipliers. T here exist Lagrange multipliers
AP, AVi and N' suca that the inequality

pi] + If>k,k - AP(p + pD kk) - AVi (PVi - 7';k ,k)

(7.8)
- AC(pi + qk,k - TikD ik) 2: 0

is valid under no constraints , in oth er words, it mu st hold for any

tn.
fields {p(x m , t) ,Vi (Xm , t) , e(x m , Moreover, the Lagrange multipliers
are functions of (p, B, B,m, D mn ).
T he validi ty of this statement will be justified in the next section.

To exploit the consequence of the inequ ality (7.8) , we first observe t hat
since constit utive functions are indep endent of the velocity Vi, the inequ ality
contains an explicit te rm linear in Vi. It s coefficient, the Lagrange multiplier
AVi, mus t vanish because the inequality must hold for any field Vi( Xm , t) ,
and , in particular , for any valu e of Vi at any par ti cular position and t ime .
Therefore, we have
(7.9)
and t he inequ ality afte r t he const it ut ive equations are introduced becom es
186 7. Th ermodynamics with Lagrange Multipliers

Note that t he above inequality is linear in p, iJ , (e,k)", Dkl ' and P,k' e,lk, D jl ,k
and by the same ar gument, it must hold for an y values of t hese qu antities.
Therefore, t he coefficients of these derivatives mus t vanish, which lead s t o
the following resul ts :

(7.10)

and
OPk _ At:oqk =0
op op ,
OPk
( Oe,l
+ OPI) _ At:(Oqk + Oql) _ 0 (7.11)
Oe,k Oe,l Oe,k - ,

OPk
( oD
+oD
OPI ) _ At: (Oqk
oD j l
+~)
oD j k
-- 0 .
jl jk
In obtaining (7.11h,3' we hav e taken into account the symmetry of t he second
derivatives, therefore only t he symme t ric parts of t he ir coefficients need t o
vanish. Mor e specifically, the t erm involvin g Djl ,k ca n be writ t en as

since D jl = D lj . Du e to t he sy mmetry of D jl ' t he relation (7.11h ca n fur ther

be reduced to
OPk At: oqk - 0 ( )
ODjl - oD - .
jl
7.12

Indeed , if we denote the left-hand side of (7.12) by Hlt then H;l = H~ and
the relation (7.11h becomes

which implies t ha t
l -
H jkl -- - HIjk -- - H kj - n!kl -- tt:lk -- - H lj -- -
k n:j l '
and hen ce H;l = o.

7.2.1 General Results

We can now obtain the consequence of t he relations (7.10) t hrough (7.12).

First of all, we claim that the relations (7.11h and (7.12), for t he ent ro py flux
 7.2 Viscous Heat-Conducting Fluid 187

and the heat flux, together with the fact that they are isotropic functions,
imply the following parallel relation,

(7.13)

and the Lagrange multiplier AE: does not depend on e,i and D ik. The proof
of this claim is not trivial, in general , and will be given in Sect. 7.4.
Since now we have
AE: = AE:(p ,e) ,
from (7.13), the relation (7.llh implies that

aN
ap qk = O.

Of course, qk -I- 0, in general, therefore AE: must be ind ep endent of the density,
hence
(7.14)
Furthermore, from (7.1Oh and (7.1Oh , the integrability condition for 1] with
respect to e and e,k gives

which shows that E is independent of e,k since we will not consider the possi-
bility of AE: being a constant , in general. Similarly, the integrability condition
with respect to e and D k l from (7.10h and (7.10)4 leads to the independence
of E from D k l . Therefore, we conclude that

E = eip, e) , 1] = 1](p, e),

and the first two relations of (7.10) can then be put together to give the total
differential of 1],
AP
d1] = NdE + -dp, (7.15)
p
which also shows that
AP = AP(p, e).

Aft er elimination of those linear t erms, the inequ ality (7.8) becomes

(7.16)

The left-hand side , denoted by 0", is the entropy production density, which is
a non-negative quantity for any values of (p, e, e,k , D ik ). We call a state with
no entropy production an equilibrium state. Since 0" t akes its minimum value
188 7. Thermodynamics with Lagrange Multipliers

at the equilibrium state, namely zero, at B,k = 0 and D i k = O. One of the

necessary conditions for the minimum is

where the index 0 indicates the evaluation at the equilibrium state and the
equilibrium pressure p is defined as

Hence , we have
(7.17)

Insertion of this into (7.15) gives

dry = AE (de - ; dp). (7.18)

Therefore, by comparison with the Gibbs relation (5.25) for elastic fluids in
thermostatics, we can identify the Lagrange multiplier AE as the reciprocal
of the absolute temperature B,

(7.19)

An immediate consequence follows from (7.13),

tf>k = e1
qk, (7.20)

which assures that the entropy flux is the heat flux divided by the absolute
temperature for the fluid under consideration. This is what is usually taken
as an assumption for Clausius-Duhem inequality.
With (7.19) and (7.20) the remaining inequality (7.16) becomes

a =- ;2 qkB,k +~ (Ti k + P8i k ) Di k ~ O. (7.21)

7.2.2 Navier-Stokes-Fourier Fluids

Further consequences from the remaining inequality (7.21) will only concern
the linear terms of the constitutive functions . Therefore, up to the first-order
terms in the gradient B,rn and D rn n , we may write

Tik = -p8i k +).. D n n8i k + 2J-l D i k + 0(2),

(7.22)
qk = -Jo\, B,k + 0(2),
where the viscosities ).. and J-l, as well as the thermal conductivity Jo\" are
functions of (p, B) . The inequality (7.21) with the linear constitutive relations
 7.3 Method of Lagr an ge Multipliers 189

now takes the form

where Dik = D ik - 1DnnOik is the traceless part (or deviatoric part , see
(4.36)) of o.;
Since the trace D kk and the traceless part Dik are mutually ind ependent ,
the left-hand side of t he inequality is quadratic in three indepe ndent variables
{O,k, Dkk, Did , and hence it follows that t he thermal conduct ivity, and t he
bulk and the shear viscositi es, are positive (see (4.37)) ,

(7.23)

These inequalit ies ensure t hat t he heat flux points in the dir ection to t he
colder region and that the fluid par ticl es t end to flow in t he dir ection of a
shear force.

Remark. Even though we have shown t hat for t he viscous heat-conducting

fluids the general ent ropy inequality do es reduce t o the Clausius- Duhem in-
equality, we have seen in this example that the method of Lagrange multipli-
ers enables us t o avoid the unnecessar y ass umpt ions <P = q/O - qu estionabl e
in general - and t he wishful argument that t he external supplies ca n be arbi-
t rarily adj uste d in the exploitat ion pr ocedure based on the Clausius- Duhem
inequality (see the rem ark on p. 134) .

7.3 Method of Lagrange Multipliers

We consider a system of linear algebraic equa tions with an inequ ality con-
st raint . Suppose t hat we have a linear syste m of m equations and n unknowns

(7.24)

and a linear inequ ality

(7.25)
where a = 1" ", m , b = 1" " , n, and t he summat ion convent ion is used . As
indi cated , A is an m x n matrix, and B E JRm, a E IRn , are vectors. If A
and B are given, of course , in general, the solut ion X E JRn of (7.24) may
not sat isfy t he inequality (7.25) .
We have seen from t he exa mple in the pr evious section that afte r in-
troducing the constitutive relations into the syste m of balan ce laws and the
ent ropy inequality, we can put them in t he form of t he above relations in
spite of t he fact that they are par tial differential relations.
190 7. Thermodynamics with Lagrange Multipliers

7.3 .1 An Algebraic Problem

The ent ro py principle requires that the const it ut ive fun ctions b e imposed in
such a way that t he ent ropy inequ ality is satisfied for any solutions of t he
balan ce equations. For the algebraic relations (7.24) and (7.25) , we ca n also
pose t he following problem :
Wh at conditions mu st be imposed on th e m atrix A and th e vector B
so th at every solution of th e linear sys tem (7.24) will always satisfy the
inequality cons traint (7.25) ?

This is simil ar to the problem of exploit ing t he ent ropy princip le, bu t is much
simpler of course , becau se it is purely an algebraic one . The answer is state d
in t he following lemma [37] .

Lemma 7.3 .1 Let S = {X E JRn I AabXb + B a = A}, th e solution set

of (7.24) , be neith er empty nor th e whole space JRn and 0: is no t a zero
vector (otherwise the cases are t rivial). Th en, th e following two statements
are equivalent:

1) D:bXb + ,6 ~ 0 'V XES ; (7.26)

2) Th ere ex ists a vector A E JRm such th at
(7.27)

Proof It is obvi ous that t he condit ion (2) im plies the condition (1) . To prove
the converse, let us observe that the condit ion (7.27) can be written as

(7.28)

We claim that the coefficient of X must vanish, and hence

D:b - AaA ab = 0,
(7.29)
,6 - AaBa ~ O.
Indeed , if this do es not hold , then for some index b, sup pose that D:b- AaA ab =j:.
O. Since X E JRn is arbitrar y, we can choose X i = 0 for any i =j:. b and Xb in
such a way that the inequality (7.28) is violated . Therefore (7.28) and (7.29)
are equivalent . To complete the proof of the lemma, we need only show that
(1) implies the existence of A so t hat (7.29) hold . The proof can better be
given in geomet ric te rms.
Let H = {X E JRn I D:bXb + ,6 ~ O} denote the half-space defined by t he
inequ ality (7.25) . Then the state ment (1) means t hat t he hyp erplan e S must
lie inside t he half-space H, or simply S c H . Let H; = {X E IRn I D:bX b = O}
be t he (n - 1)-d imension al hyp erplan e par allel to H and passing t hrough the
origin , and So = {X E JRn I AabXb = O} be the hyp erplan e par allel to S
and passing through the origin . Not e that bo th S o and Hi; are subspaces, and
 7.3 Met hod of Lagrange Multipliers 191

dim So is smaller t han dim Hi; We claim that if the hyp erplane S is contained
in the half-space H then t he subspace So is also contained in the subspace
Hi; Since if this is not true, t hen there is some vector Z in So but not in H o.
T hen for any constant c, if Y E S , Y + cZ is also in S . But since Z tf- H o ,
which means t hat (XbZb i= 0, then it is possible to choose a value of c such
that (Xb(Yb + CZb) + j3 becomes negative. T his means that Y + cZ is not in
H , which contradicts the assumption S c H .
If we denote t he orthogonal complements of So and H o by S;- and H;- ,
respect ively, then it follows from So C Ho that S;- =:> H;-. Now, since a is in
H;-, therefore a is also a vector in S;-.
Since we know from linear algebra that
dim S;- = rank A , and t he row vectors of the matrix A a b with a = 1, · . . , m
are in S;-, this set of vectors span the space S;-. T herefore, there exists a A
such that a can be expressed as
(Xb - AaA a b = 0,

and the first part of (7.29) is proved . T he second part follows easily, because
for any X E S , we now have

which implies j3 - A a B a ;:::

t he proof of the lemma. 0
°
by the ass umption that S is in H . This complet es

If the m row-vectors of the matrix A a b are linearly independent, then

t hey form a basis of S;- . In this case, the components A a are uniquely deter-
mined by a . On the ot her han d , if the i-th row-vector A i b is zero, then Ai
is ar bitrary, which implies from (7.29h that B , must vanish. In t his case , we
can simply eliminate the i-th equation from the system (7.24), because it is
merel y B, = 0, which contains no X at all.
We call A a Lagrange multipliers, because t he replacement of the state-
ment (1) by t he statement (2) of the lemma is sim ilar to t he use of Lagrange
multipliers in classical problems of maximizing or minimizing a function with
constraints.

7.3 .2 Local Solvability

Now let us t ake viscous heat-conducting fluids considered in t he previous

section as an example. Incorporated with the constitut ive equations (7.5) ,
the system of balance laws (7.4) takes the following quasi- linear for m,

(7.30)

and so does t he ent ropy ineq uality (7.6),

(7.31)
192 7. Thermodynamics with Lagrange Multipliers

where
(Xb) = (p, Vi, iJ, (B ,i)", b.; P,i,{J,ij, Dij ,k),
(7.32)
(Yc) = (p,vi ,B,B,i,Dij) ,
for a = 1, .. . ,5, b = 1, . .. ,41, and c = 1, ... , 14. In counting the numbers of
corresponding components, symmetry of indices is taken into account. The
matrix A ab as well as B a , (Xb, 13 can all be written out easily. Their explicit
expressions are left as an exercise.
The entropy principle requires that the inequality (7.31) be satisfied
for any solution {p(x , t) , Vi(X, t), £(x, tn of the system of partial differential
equat ions (7.30) . This requirement will reduce to a purely algebraic problem
stated above if the following assumption can be justified.
Local solvability. The system (7.30) is locally solvable at (XO, to) iffor
any values of (Yc, Xb) satisfying (7.30) algebraically, there is a solution
{p(x, t), Vi(X, t) , £(x, t)} of the system of differential equations (7.30) in
a neighborhood of (XO, to) at which the solution is consistent with the
given values.

The local solvability assumption is the key point in employing the method
of Lagrange multipliers for the exploitation of the entropy principle. In most
"normal" situations, like the one we consider here, this assumption can be
justified. The justification is usually based on the existence of solutions for
initial-value problems.
In the present example, the balance laws (7.4) or (7.30) can be written
as a system of partial differential equations in the form :

(7.33)

where the explicit expressions of the functions Ip , I« and Ie on the right-

hand side are left as an exercise. By the Cauchy-Kowalewsky theorem (see,
for example, [12]) for initial value problems, there exists a unique analytic
tn
solution {p(x , t) , Vi(X, t), £(x , of the system (7.33) in a neighborhood of
a point, say (0,0), for any analytic initial values {p(x, 0) , Vi(X, 0) , £(x , On.
Since the initial values are arbitrary, by Taylor series expansion such as
1
p(x ,O) = p(O, 0) + p,k(O, O)Xk + "2 p,kl(O, O)XkXI + " ' ,
(p, B,Vk) and their gradients of any order, (p,k' B,k' B,kl' Vk,l, Vk,lm, Vk,lmn , " ')
can be assigned any values at (0,0).
 7.3 Method of Lagrange Multipliers 193

In order to show that the syste m (7.33) is locally solvab le at (0,0) , we not e
that for the existence of a solution, we can give any values to all the elem ents
of (Yc' X b) except those five wit h time derivative, namely (p, Vi , il, ((},k)", Dij ),
aP av ' a(} ne . B»,
or equivalentl y, (at ' at' , at ' at ' a~'J) · T he first three of them can be cal-
culated so that (Yc, Xb) sat isfies (7.33). Therefore, t o complete our argument
av . a(} .
we on ly have to show t hat t he values of a~'J and at can also be given ar -
bitrarily at (0,0) . To see this , we take t he derivatives of the equat ion (7.33h,3
to obtain
OVi,j a IVi a IVi a IVi
7ft = 8 Pk P,kj + W(},kl
,klj + -a -- Vk,lmj + ... ,
Vk,lm

-ee, = ( -OleP,kj + -
ole ole )(
- (},klj + - - - Vk,lmj + ... P-
&)-1 + ....
at 0Pk O(},kl OVk ,lrn oe
OV . o(} .
This equat ion can be satisfied for any given value of o~J and at by an ap-
propriate choice of values of the higher gradients of {p ,kj , (}klj , Vk,lrnj} , which
does not belong to the chosen values of (Yc, Xb) but for the existence of a
solution it is complet ely arbitrary. T his proves our claim that indeed for any
given values of (Yc, X b ) satisfying the system (7.33) algebraically at a point ,
there is a solution of the system in the neighborhood of that point. There-
fore, our problem of explo iting the entropy principle in this case red uces to
t he algebraic problem posed at t he beginning of this section, and the use of
Lagrange multipliers in the pr evious section is justified.
To justify the local solvab ility assumption for t he use of Lagrange mul-
tipliers in the exp loitation of the entropy principle usually requ ires a careful
analysis of t he system of field equat ions, and sometimes addit ional condi-
tions between the derivatives of the vari ables must be consi dered in order to
prove the local solvabi lity.e Of course, the exploit at ion of the entropy prin-
cip le can also be don e by dir ect eliminat ion of quantities imposed by the
equations of balan ce, albeit very tediously (see, for example, [53]). T he use
of Lagrange mu ltipliers greatly simplifies this task and makes the procedure
for the exploitation of the ent ropy princip le systematic and straightforward .
T hermodynamics with Lagrange mu ltipliers has sin ce become the routine
procedure in formulat ing constitutive theories.

E x e rcise 7.3.1 Write out explicit ly the matrix A a b as well as B a , (tb,

and f3 in the relations (7.30) .

Exe rcise 7 .3 .2 Give the explicit expressions of the right-hand side of

the field equations (7.33).
2 For more details , the reader is referred to the original paper [37].
194 7. Thermodynamics with Lagrange Multipliers

Exercise 7.3.3 Derive thermodynamic restrictions on consti t ut ive

equat ions (5.15), C = F(F, B, g) , for thermoelastic materials based on
the general ent ropy inequ ality (7.6) by the use of Lagrange mul tipli-
ers. (Formulate the probl em in material description and t hink abo ut the
problem caused by the use of deformation gradient as a var iabl e - the
compat ibility condition P = Gr ad v.)

7.4 Relation Between Entropy Flux and Heat Flux

First of all , it is easy to prove t he parallel rela tion (7.13) between the en-
tropy flux and the heat flux from the condit ion (7.llh and (7.12) if only
linear const it uti ve equations ar e concern ed. Indeed, since qk and tJ>k a re vec-
tor isotropic fun ctions of (p, B, B,m, D mn) , their linear representations in these
variabl es are given by

qk = k B,k'
where k and ¢ are inde pe ndent of B,m and D mn . Then , t he condition (7.llh
implies t ha t
¢ = AEk,
which immediate ly lead s to the relation,

tJ>k = A qk,
E

and AE is independent of B,k and Dkj by int roducing t he above relation into
(7.11h and (7.12).
This is ind eed almost trivial , nevertheless, we rem ark that alt houg h gen-
eral represent ations for isotropic functions are well known, a dir ect proof of
t he parallel relation based on such representations (see (4.51)) would be too
complica te d if not ent irely impracti cal (such a pr oof has been given in [36]).

7.4.1 Theorem of Parallel Isotropic Vector Functions

To prove the par allel relation between t he entropy flux and the heat flux,
from condit ions simil ar to the equat ions (7.11), is a ty pical probl em in ex-
ploiting thermodyn amic restrictions by the use of Lagrange multipliers with
the general ent ropy inequality, see for example: for viscous fluids [52], for
thermoelast ic solids [53] , for fluids and thermoelastic solids in elect romag-
net ic field [30,45], for rigid heat conduct ors [3] and for mixtures of fluids [54] .
This problem is usually trivial if linear const it ut ive relations are ass umed, as
we have seen ab ove. Otherwise, it can be quite difficult, even though such a
relation may st ill be expec te d . In the following , we shall treat this problem
without recourse to general representations for isotropic fun ctions considered
in Ch ap. 4. The proof will be based on the following lemma:
 7.4 Relation Between Entropy Flux and Heat Flux 195

Lemma 7.4.1 Let h(A , v) be an isotropic vector function . Th en th e function

h satisfies th e following relation,

Proof If we define

F(Q) = h(QAQT , Qv) - Q h(A, v) "I Q E L:(V) ,

then by t aking t he gr adient with respect to Q we obtain

Since h(A, v) is an isotropic fun ction, we have F( Q) = 0 for any Q E O(V) ,

and hen ce from Lemma 4.2.4 , proved in Chap. 4, it follows that

oQF(l)[W] = 0,
for any W E Skw(V). Therefore we have, in the terms of component s,

which proves t he lemma. D

On e can eas ily ext end the above lemma to the case for an isotropic vect or
fun ction of an arbit rary number of vect or and tensor variables.
Now we shall state the problem in the followin g theorem , but first let us
introduce the followin g abbreviations :

(7.34)

We shall cons ide r const it ut ive class es dep ending on an arbit rary number of
vector and tensor variables. Dependen ce of scalar variables is irrelevant in
isotropic fun ctions and hence will not be mentioned explicitl y.

Theorem 7.4.2. Let CP and q be isotropic vector fun ctions, and A be an

isotropic scalar f unc tion, of an arbit rary number of vector and tensor vari -
ables. A ssume that
i) for N vector vari ables v a, a = 1, .. . , N ,

OPi
( o va
+ OPj ) _ A( Oqi + Oqj ) = 0 (7.35)
ova o vC! o va '
J' J '
196 7. Thermodynamics with Lagrange Multipliers

ii) for every other vector variable u ,

OPi _ A Oqi - 0 (7.36)

OUj OUj - ,

iii) for every tensor variable A ,

(7.37)

Then for N = 1, the function A is independent of all vector and tensor

variables, and

p=Aq.

The above conclusion also holds for N = 2 if q and Vi x v 2 are

functionally independent.3

Proof With the abbreviations (7.34), the assumption (i) implies that H Vi ;
is skew-symmetric,
tr:
ij + n-;
ji = 0, (7.38)
and the assumptions (ii) and (iii) give

Applying Lemma 7.4.1 to the vectors P and q, we have

where the summations are taken over all vector variables (including the vec-
tors va and all other vectors u) and all tensor variables, respectively.
Multiplying the second equation with A and subtracting it from the first
one, we obtain, by the use of the abbreviations (7.34) and the assumptions
(ii) and (iii) ,
N
(Oijkk - Oikkj) = L(H ai j vk - H ai k vj) , (7.39)
a=i

3 For the case N > 2 and some other more general conditions see [43].
 7.4 Rel ation Between E nt ro py Flux a nd Heat Flux 197

in which for simplicity, we have writ ten H ai j for H Vi ; . This impli es that
N N N
a a H a31 vf,
k1 = L H 12V2 ' k2 = L H 23 v!3, k3 = L (7.40)
a=1 a=1 a=1
and the following system of six equa t ions:
N N

L(H aI2 v2 + H a31 v!3) = 0, L(H aI2 v!3 + H a31V2 ) = 0,

a=1 a=1
N N
L(Ha23 v!3 + H a12 vf) = 0, L(H a23 vf + H a12 v!3 ) = 0, (7.41 )
a=1 a=1
N N
L(H a31 vf + H a23 V2) = 0, L(Ha31 v2 + H a23 vf) = 0,
a= 1 a=1
by the use of the relation (7.38) . The three equ ations on the right of (7.41)
reduce to
N N N
LHa12 v!3 = 0, LHa23Vf = 0, L H a31V2 = 0. (7.42)
a=1 a= 1 a=1
We shall now pro ceed t o prove the t heorem for N = 1 and for N = 2
separat ely.
For N = 1, since VI =I- 0 in general , from (7.42) we obtain

Therefore, from (7.40) it follows that k must vani sh , in other word s, the
relation P = A q hold s.
By the substi tution of t he relation P = A q into t he condit ions (7.35) ,
(7.36) , and (7.37) , it follows imm ediately that the partial derivatives of A
with resp ect to all t he vecto r and the t ensor variabl es must vani sh , since, in
general, the vector function q need not vanish , and t he t heorem is proved for
N= 1.
For N = 2, from (7.41) and (7.42) we have t he following linear syst em
of six equat ions for six vari abl es (Hi 3' H jl ' H f2' H'i3' H§I' H r2)'

H I12 v21 + H I3 1 v31 + H 212 v22 + H 23 1 v32 = 0,

H 1n vj + H\ 2 v} + H 223 v~ + H 212 v~ = 0,
H 131 v} + H 1n v~ + H 231 v~ + H 223 v~ = 0,
(7.43)
H \ 2 vj + H 212 v~ = 0,
H 123 v } + H 223 v ~ = 0,
H 131 v~ + H 231 v~ = O.
198 7. Thermodynamics with Lagrange Multi pli ers

The coefficient mat rix of t his syste m is of rank equal to 5, and the syst em
admit s a one-par am et er solu tion given by

which imply from (7.40) t hat

k=iP-Aq= /,(v I xv 2 ) , (7.44)

where v is a scalar function of the vector and t he t ensor vari abl es.
By the relation (7.44) , t he ass umpt ions (ii) and (iii) lead t o

where X stands for the components of any vector variable u and any te nsor
vari abl e A . Since q and VI x v 2 are functionally independent by assumpt ion,
the above relations are possibl e only if both A and ry are ind ependent of u
and A. Therefore , A and v are funct ions of VI and v 2 only,

Consequently, we also have k = k(v I , V 2 ) .

On the other hand , since k = iP - A q is an isotropic vector fun ct ion , it
can be represent ed by,

or we have
/' v I X v 2 = k I VI + k 2 v 2 ,
where k I and k 2 are isot ropic scal ar funct ions of (VI , v 2 ) . Taking t he inner
product of t his relation with VI x v 2 we obtain

which implies that v must vanish. Therefore, k = 0 and the relation iP = A q

hold s.
Finally, by t he substitution of iP = A q into (7.35) , it follows that t he
partial derivatives of A with resp ect to VI and v 2 must vanish. Therefore, A
is ind ep endent of any vector and tensor var iables. T his com pletes the proof
for N = 2.0

T he problem of proving the relation (7.13), iP = A E: q, for t he viscous

heat-conducting fluids considered in this chapte r is a sp ecial case of the above
theorem (N= 1) with one vect or var iable grad e and one symmetric t ensor
vari abl e D .
8. Rational Extended Thermodynamics

8.1 Introduction
In the const it ut ive theories of materials we have considered so far , there is one
essential feature, namely, the basic equat ions ar e based upon the principle of
balance of mass , momentum, and energy, while the diversity of materials is
charact erized solely by the fun ctional complexity of constitutive equations.
A different approach, known as Ext ended Thermodynamics, has been
proposed by Liu and Muller in [46], in which , in addition to the densities
of mass, mom entum, and energy, the mom entum flux and t he energy flux
are also t aken as basic field qu antities. .For these exte nded field qu antities,
balance equations of momentum flux and energy flux are postulated and the
theory is formul at ed in the fram ework of rational thermodynamics laid down
in the pr evious chapters.
Unlike the cons titutive equations of ordinary thermodynamics, in which
cons titutive functions may depend on the history and space gradients of the
basic field vari ables, t he constitutive functions of extended thermodynamics
are assumed to be instantaneous and local , i.e., they ar e functions of basic
field variables only. As a consequence, the resulting field equ ations form a
system of first-order quasi-linear partial differential equations - in contrast
to the ordinary theory, for example, the system is of second-order in space
and first-order in time in the case of Navier- Stokes equ ations (see (4.40)).
Owing to the simple const it ut ive equations in the extended theories, the
complexity of materials is no longer characterized by constitutive equations
alone, rather it is also characterized by t he choice of differ ent basic fields . In
other words , the complexity of materials rests upon the formulation of the
system of balanc e equations.
Extended t hermodyna mics has been an ar ea of active research in the
last two decades. It has been found that it not only has mor e systematic
mathem atical structures - symmetric hyp erbolic systems - it also has yielded
very explicit constitutive relations, in complete agreement with the results of
the kin etic t heory of gases, in spite of being a phenomenological theory based
on the fram ework of rational thermodynamics considered in this book. In this
chapter , we shall give an outlook of rational extended thermodynami cs. Most
of the main results and refer ences of recent developments can be found in the
book of Muller and Ru ggeri [57J.

I.-S. Liu, Continuum Mechanics

© Springer-Verlag Berlin Heidelberg 2002
200 8. Ration al Extended Thermodynamics

8 .2 Formal Structure of System of Balance Equations

We consider a syste m of balan ce equat ions relative to an inertial frame
au
at + div. H(u) = g(u) (8.1)

for the basic field u(x , t) in JR N. The flux H(x , t) and the production density
g( x, t) ar e assumed t o be fun ctions of t he basic field u(x , t) . Following the
idea of rational thermodyn ami cs - the ent ropy principle - t he constit utive
fun ction of H(u) and t he production g(u) have to be determined in such a
way that the syst em (8.1) is consist ent with t he ent ro py condit ion: (also see
[22])
ah(u)
----at + div II(u) = 17(u) ~ 0, (8.2)

where h ,17 E JR and II E JR3. Note that all the const it ut ive qu an tities are
assumed to b e functi ons of the basic vari abl es u only.
This requ irement lead s to t he existe nce of Lagran ge multiplier A in JR N
(see Sect . 7.3 for justification) , such that

{ ----at
ah(U) + div
. II(u) - at
17(u) } - A · {au . H(u)
+ div - g(u) } = 0

hold s for any u(x , t) E JRN. In particular , it must hold for arbit rary values
of ~u and ~u . Since t he a bove relat ion is linear in bo th of t hem, it follows
ut UXi
imm ed iat ely t hat

ah = A 17 =A ·g. (8.3)
au '
We ca n rewri te the relations (8.3) 1,2 in the differential for m:

dh = A ·du, (8.4)
which we shall refer to as the entropy-entropy flux integrability relations,
since t he Lagran ge multiplier A appears as t he common integration fact or of
the right-hand sides for h and Il. , i = 1,2, 3.
Furthermore, if we assume the invertibility of A(u) , we ca n write t he
syste m (8.1) in terms of t he variable A as

(8.5)

Now, if we introduce h(A) and ll(A) defined as

h= A ·u-h, (8.6)
 8.2 Formal Structure of System of Balance Equations 201

which by the relation (S.4) lead to

dh =u ·dA,
or equivalent ly,
8k (S.7)
U= 8A '
Therefore , the syst em (S.5) becom es

8 2 ft 8A 3 8 2 it, 8A
8A8A 7ft +L 8A8A 8Xi = g .
(S.S)
i= l

In other words, the differential operator of the syst em is determined solely by

a pair of a scalar and a vector function. For thi s reason, the functions (ft, Ii)
have been called the generator junctions of the system [65]. Note that the
coefficient matrices of this system are all symmet ric N x N matrices.

Remark. From (S.3), it follows that

8IIi 8h 8Hi (S.9)

8u = 8u · Bu:'
In t he lit erature of mathematical theory of hyp erbolic conservation laws
(see [13]) , a pair of scalar and vect or functions (h( u) , II( u)) is called an
entropy-entropy flux pair for the system of conservat ion laws ,

~~ + div H(u) = 0, (S.lO)

if it satisfi es (S.9).
For a given syst em of the form (S.lO) , if an entropy-entropy flux pair
exists , it is usu ally not unique. Even though such a pair may not bear
any physical significance , in general, it certainly reflects the no tion of an
ent ropy ba lance equat ion in continuum thermodyn amics. Indeed , in ex-
tended thermodynamics, besides a system of balance equat ions, an addi-
tional ent ropy inequ ality is also postulated . The system is t hen required
to be consist ent with the ent ropy inequ ality, and henc e the ent ropy and
the entropy flux form a pair in the above sense. Cons equ ently, extended
thermodynamics cons titutes a class of physical models for mathematic al
ana lysis of hyperbolic conservat ion laws. D

8 .2.1 Symmetric Hyperbolic System

Con sider a qu asi-linear syste m of variable u E JRN ,

8u 8u
+ L Bi(U)~ = g(u) ,
3
A(u)-;:l (S.l1)
at i= l u X,

where A and B, are N x N matrices.

202 8. Rational Extended Thermodynami cs

Definition. The system (8.11) is called symmetri c if A and B , are all sym-
metric matrices.

Definition. The system (8.11) is called hyperbolic if det A -I 0 and for any
v E JR 3 , Ilvll = 1, the eigenvalue problem
3
(2: s», - AA)v = 0
i= l

has real eigenvalues A and admits N linearly independent eigenvect ors v .

In the theory of wave propagation, v and A have the physical meaning of
unit normal and normal speed of the wavefront, respectively. Moreover, from
the spectral theorem (see p. 258), it follows that if the system is symmetric
and det A -I 0, then it is also hyp erbolic.
Now, returning to our syst em (8.8), we have the following theorem (see
[22]):
Theorem 8.2.1. Suppose that h(u) is a concave function of u, then the
system (8.1) , consist ent with th e entropy condition (8.2), can be reduced to
a sy mmetric hyperbolic system.
Proof Not e that the concavity of h( u) is equivalent to the condition that
t he Hessian matrix
[P h
is negative definit e.
8u8u
Therefore , from (8.3) t he Jacobian matrix ~~ is non-singular, which ensures
t he invertibility of A(u) . By t he previous argument s leading to the syste m
(8.8) , which is symmetric, it is sufficient to show that it is also hyp erbolic.
Since h(u) is a concave fun ction, for any variation OU -I 0, we have

8 2h
OU' 8u8u OU < O. (8.12)

By the use of (8.3) and (8.7), we can write

which implies that

8 2h
A(A) = 8A8A is also negative definite.

Therefore, det A(A) -10 and the syst em (8.8) is hyp erbolic. D
Some important properties of symmetric hyp erbolic systems include
well-pos edness of Cauchy initial value problems and finite speeds of wave
 8.2 Formal Structure of System of Balance Equations 203

propagation [21, 63]. Such properties ar e almost essent ial for material behav-
iors that we might regard the symmet ric hyperbolicity as a desirable property
for material models.
In the following, we shall see that the Euler equa tions of an elast ic fluid is
an example of a symmetric hyp erbolic syste m with concave ent ropy function.

Example 8.2.1 Elastic fluids :

From (4.42) we have the following syst em of bal an ce equat ions:

(8.13)

For an elast ic fluid , the internal energy density eip, B) and the pr essure
pip, B) also sa t isfy the Gibbs relation (5.25) ,

dry = ~ (de: - ; dp) , (8.14)

where B is the absolute temperature and r] is the entropy density.

In order t o show that the system (8.13) can be reduced to a sym-
metric hyperbolic syste m, it suffices to show the exist ence of an ent ropy
condition (8.2) and the concavity of h according to the above theorem .
In this case, we have

u = (p , PVi, p(e: + V;)) E IR5,

(8.15)
n, = (PVj, PViVj + p8ij , p(e: + ~2 + ~)Vj) ,
and the Lagrange mul tipliers are defined as

(8.16)

where g is the free ent ha lpy defined in (5.79) ,

p
g = e: - Br] + -.
p

Taking the inner product of the system (8.13) with A , we obtain

A . ( -aU
at
aH-j )
+-
aXj
_
-
0, (8.17)
204 8. Rational Extended Thermodynamics

which , after simplifi cation using the Gibbs relation (8.14) , becomes

(8.18)

This is the entropy condition (8.2) for an elastic fluid by setting

h = pry, E=O.

Finally, in order to show the concavity of h(u), let us consider the

quadratic form ,

(8.19)

In the above calculation the relations (8.14) , (8.15) and (8.16) are used.
Since the density and the temperature are positive quantities, the
function h(u) is concave if and only if

8E op
o(} > 0, op > 0,
i.e., the specific heat and the compressibility are positive, which are the
conditions (5.71) and (5.72) of thermodynamic stability. In other words,
the concavity condition is equivalent to the thermodynamic stability
conditions in this case. D

Theories of mixtures and granular materials can also be formulated with hy-
perbolic field equations [54, 77, 78]. But field equat ions of ordinary thermody-
namic theories ar e generally not even hyperbolic. Indeed, the Navier-Stokes
equat ion (4.40) is parabolic. However , we shall see later that it is closely re-
lated to the exte nded theory with hyperbolic field equat ions, consisting of
additional balance equations for the stress and the heat flux.

Exercise 8.2.1 Derive the entropy condition (8.18) from (8.17) .

Exercise 8.2.2 Verify the calculations in (8.19) .

8.2.2 Galilean Invariance

It is known that relative to inertial frames , dynamic laws, such as balance

equat ions of mass, momentum, and energy, are Galilean invariant (see p . 42,
Prop. 2.3.2) . Likewise , in formulating a general system of balance equations
 8.2 Formal Structure of System of Bal ance Equations 205

relative to inertial frames, Galil ean invari ance must be regarded as a funda-
mental requirement.
Recall that a change of frame from (x , t) to (x* , t *) is a Galilean trans-
formation if
x* = Q(x - x o ) + Vt + Co ,
(8.20)
t* = t + a,
where V is a constant vector and Q is a constant orthogonal tensor.
The requirement of Galilean invari ance imposes rather a specific dep en-
dence of the flux and the production on the velocity field as we shall see. To
begin with, let us express the basic field u in t erms of the velocity v and the
variables w,
u = F(v,w) E IRN , wE IRN - 3 . (8.21)
The variables ware assumed to be objective quantities (with respective to
an arbitrar y change of fram e, i.e., Euclidean transformations, see Sect. 1.7)
of var ious tensorial order . We also split the flux H k into the convective and
non-convective parts,

(8.22)

Let the velocity-independent part be denoted by

F(w) = F(O, w), g(w) =g(O,w) , (8.23)

called the internal parts of the respective qu antities. The internal qu antities
are assumed to be objective.
The bal an ce equat ion (8.1) can now be written as

:tF(v ,w) + 8~k (F(v ,W)Vk + Gk(v ,w)) = g(v ,w). (8.24)

Proposition 8.2.2 If the syst em of balance equations (8.24) is Galilean in-

variant, then th ere ex ists an N x N m atrix X (v) such th at ([64])

F(v ,w) = X(v)F(w) ,

Gk(V , w) = X(V)Gk(W) , (8.25)
g (v , w) = X(v) g (w) .

Moreover, X (v) has th e following prop erties: For any VI , v 2 E IR3

X(O) = 1. (8.26)

Proof Consider a Galilean transformation (8.20) with Q being the identity

t ensor , then we have

v * =v+V , w* =w , (8.27)
206 8. Rat ion al Extende d Thermodynamics

and
a
ax";
a
Bx , '
~ = ~ _ v;~
at * at ' aXi .
(
8.28
)

Since the system (8.24) is Galil ean invari an t , it takes t he same form in the
frame (x * , t*) ,

a~*F(V* ,w*)+ a~* (F(v* ,W*)Vk+Gk(V * ,w*)) =g(v* ,w*) ,

k

which from (8.27) and (8.28) reduces to

:tF(V+V,w)+a~k (F(V +V ,W) Vk+G k(V+V,W)) = g (v + V , w ). (8.29)

This system mus t be equivalent to the syst em (8.24) and hence by compar-
ison , it requires that one syste m must be a linear combination of the other.
Con sequ ently, t here exists a non-singular N x N matrix X(V) such that

F(v + V , w) = X(V)F(v , w) ,
Gk(v + V , w) = X(V)Gk(v , w) ,
g(v + V , w) = X(V)g(v , w) ,
hold s for any v and V .
In particul ar , for v = °we have for any vector V
F(V , w) = X(V)F(O , w) ,
Gk(V , w) = X(V)Gk(O , w) ,
g (V , w) = X(V)g(O , w) .
T herefore, the relations (8.25) hold and from which the properties (8.26)
follow immediately. D

The properties (8.26) define X(v) as an exp onent ial op erato r. Indeed, by
taking the derivativ e of (8.26h with respe ct t o VI and evaluat ing at V I = 0,
we ob t ain
aX
-;::;- = AkX, X(O) = 1, (8.30)
UVk
where A k for k = 1,2,3 are constant N x N matrices defined by

A k = aX I . (8.31)
aVk v=o
The solution of (8.30) is given by an exponent ial op erator (see p. 71)
1
X(v) = exp(vkAk) = 1 + VkAk + "2 Vk VlAkAI + .... (8.32)

We shall give explicit form s of t hese matrices for an elast ic fluid in t he fol-
lowing example.
 8.3 System of Moment Eq uations 207

Example 8 .2.2 For an elastic fluid, let w = (p, c) E IR2 , then from
(8.15h we can write
2)
F (v , w ) = (p, PVi, p(c + V
2
) ,

F (w ) = (p,O,O,O,pc) ,
and from (8.15h we have t he non-convect ive flux

G k ( V , w) = (0, pOjk, pVk) ,

Ch (w ) = (0, pOjk, 0).

l
Hence, by com parison with the relation (8.25) we obtain

X(v) ~[
2
:;
Iv 2
s.,°
Vj

and from (8.31), we have A k for k

~] ~ ~;1
1 V3
Iv
2

= 1,2,3,
2
VI
1

°° °1
V2
1
V3 J (8.33)

°1 ° °1
°° °
°°
AI =
°°° A2 =
°°
°°°°
A3 =
°°°
°°°1 °
°°°°°
1
1
°°° °° 1 °°
Note that X(v) and A k are all lower triangular matrices. 0
°°°
Exerci se 8 .2 .3 From (8.26), show that

X(v)X(u) = X(u)X(v), X(V)-I = X( - v ),

for any v, u E IR3 .

8 .3 System of Moment Equations

We have alr eady mentioned some material models in mathematical physics

that exhibit the elegant structure of symmetric hyp erbolic systems (for some
others, see [13]). Nevertheless , t he most systematic models ar e the theory of
moments ak in to Grad's theory of moments in the kinetic theory of gases [24] .
Our disc uss ion of general balance equations will be centered on the structure
of systems of fields consisting of mom ent densities of various orders.
208 8. Rational Extended Thermodynamics

We shall use t he following not ation: Round bracket s indicat e sym-

met rization of all indices within the brackets, i.e., t erms are summed over all
permutations of n indi ces within the br ackets and divided by n!, irr espective
of whether the indi ces are distinct or not , e.g .,

Definition of Moment Densities

In the kinetic theor y of gases, physical quantities, such as density, stress,

energy, and energy flux are associate d with moment s of a phase-density func-
ti on f(x , c , t) . The phase density gives t he number density of molecules with
velocity c at place x and t ime t.
Macroscopic thermodynamic qu antities ca n be defined , from the kinet ic
theor y, as expectation valu es of a fun ction 'I/J(x, c, t) ,

~(x, t) = J 'I/J(x, e,t)f(x , e, t) de. (8.34)

In particul ar , the densities of mass, momentum and momentum flux are de-
fined , resp ectively, with 'I/J = m, 'I/J = mCi and 'I/J = mCiCj ,

Fo=m ,
where m is the molecular mass. They are also called the zeroth-, t he first -
and the second-order moment densi ti es. More generally, we define the mom ent
density of order n as
(8.35)
The mean velocity v of the gas can be defined as
Fi
Vi = F '
o
The velocity of a molecule relative to the mean velocit y of the gas is called
the peculi ar velocity of t he molecule,

(8.36)

In the sa me manner, we may define moment s with peculiar velocity,

(8.37)

called t he int ernal moment densities of order n . We remark t hat the int ernal
moment Pi, ...in is an object ive tensor qu antity owing to the fact that t he
relative velocity is an obj ect ive vector (see Sect . 1.7.1 for verification) .
 8.3 System of Moment Equations 209

With simple algebraic manipulations, it is easy to verify the following

relations:
Fo =P,
F i = Pi + PV i , (8.38)
F i j = P i j + 2P(i Vj) + PVi V j ,

F i j k = Pijk + 3P( i j Vk) + 3P(i V j Vk ) + PViVj Vk ·

It is not a surprise to see that t he righ t-hand sides look like a binomial
expansion because of (8.36). Indeed , one can easily see that the following
general relation holds :
n
F i l" ' i n = L (~) Veil . . · V i kPi k+l ···i n) ' (8.39)
k =O

where
(n)
k
n!
k!(n-k)! '
and we have adopte d the following convent ions:

for k =n P ik +l ,··i n = P,
for k =0 F i l · .. i k = F o, ViI' . . V i k = 1.
Moment Equations

We shall consider a system of balance equati ons (8.1) for the densities of
mom ent F = (Fo, F i l , F i 1i 2 , F i d 2i 3" " ) E N , N ;::: 5, m
of en;
7ft + O Xk = g . (8.40)

We denote t he non-convective par t of the flux by Gk ,

(8.41)
We shall denote the internal parts of moments, their fluxes, and produc-
t ion densities by

F = (p, P i l , Pil i 2 ' P il i2i 3 " ") ,

c, = (Pk ' tu,» , P il i2k , P i d 2i 3k , " .) , (8.42)

ii = (7l'0, 7l'i l ' 7l'i l i 2 ' 7l'i l i2 i3 ' " . ).

Taking the relation (8.39) between F i l " ' i n and P i l ... i n for granted, one
can det ermine t he N x N matrix X(v) by comparison of (8.39) and (8.25h .
The matrix, too cumbe rsome to writ e out here , is a lower trian gular ma trix
with identity diagonal elements (a special case for N = 5 is given in (8.33)) .
For general discussions on the properties of t hese matrices for syste ms of
moment equations see [57, 64].
210 8. R a t ion al Extende d T herm ody namics

Since t he moment s, t he non-convecti ve fluxes, and the product ion den-

sities have t he same st ructure given in (8.25), withou t explicit ly using t he
mat rix X (v ), we ca n eas ily obtain t he following expressions:
n

Fi ," .i n = L.( ~) V(il " · Vi k P i k+ l ,, ·i n ) ,

k=O
n

G i l " ·i n j = L (~) v«. ...Vi kPik+l· ··in)j, (8.43)

k =O
n

g il " ·i n = L ( ~) v«. ...Vi k 7l'i k+ l " ·i n) ·

k=O

Note t hat the above st ru ct ure impli es t hat Fil " .i n and gil" .i n are com-
pletel y symmet ric t ensors, while G i l " ' i n j is symmetric in the first n indices
onl y. Since P i l ." i n is completely symmetric by kinetic definition, we shall also
ass ume tha t 7l'i l ". i n is complet ely symmet ric, and P i l ."inj is symmetric in t he
first n indices.
We can rewrit e the balance equ ations (8.40) in terms of t he internal
part s of mom ent s P i l . " t n • It reads

(8.44)

for ti = 0,1 ,2" " , where t he dot deno tes the mat erial ti me derivati ve. The
proof of (8.44) using the relat ions (8.43) is st raight forwa rd algebraic calcula-
t ions (see [39]) and will be left for the read er to verify.

Hierarchy of Systems of Moments

The first five equat ions of the moment equat ions can be ident ified with con-
servation laws of mass, momentum, and energy, by int roducing the following
conventional qu antities:

P mass densi ty,

1
E =- Pi i spec ific int ernal energy,
2p
(8.45)
st ress te nsor,
1
qi = '2 P ij j hea t flux vector,
 8.3 System of Moment Equations 211

and require

Pi = 0, Pi = 0, 1fo = 0, 1fi = 0, 1fii = 0. (8.46)

They read
ap a
at + oxJ (pVj) = 0,
a a
at (PVi) + ox"J (PVi Vj - 1';j) = 0, (8.47)

a 1 2 a( 1 2
-(pc+-pV ) + - (pc+-pV )v - T· v + q " =0 .
at
r )
2 oxJ" 2 J 'J' J

From (8.40) we have a system of equations for N moments of increasing

order. With the conditions (8.46) , explicit expressions of (8.43) for moments
up to the fourth-order are given in the following:

Moment densities:
Fo =P,
F; = PVi,
F i j = Pij + PViVj,
Fi j k = Pijk + 3P(ij Vk) + PViVjVk ·
Non-convective fluxes of moments:
G, =0,
G ij = Pij ,
Gijk = P ijk+ 2v(iPj)k,
Gijkl = Pijkl + 3V(iPjk)1 + 3V( i VjPk)I '
Production densities of moments:
go = 0,
gi = 0,
gij = 1f (i j ) ,

gijk = 1fijk + 31f(ij Vk)'

We have introduced the notation ,
1
A (""
'J
) = A ' J" - -AkkO
3 'J'"
to represent the traceless part of a symmetric tensor A i j .

We shall consider systems consist of N moment equations with increasing

tensorial order. The simplest one, N = 5, is a system consisting of conserva-
tion laws of mass, momentum, and energy given by (8.47) .
212 8. Rational Extended Thermodynamics

In the framework of exte nded thermodynamics, a theory bas ed on this

system of five fields (F, Fi , Fii) , or (p, Vi, E) , must be completed with consti-
tutive equations:
T ij = Tij(p, Vi, E),
qj = qj(p, Vi, E) .
Since the int ernal parts of moment and flux of moment are assumed to be
obj ective, by the principle of material objectivity the above constitutive equa-
tions can not dep end on Vi , and from the representation theorem (Sect . 4.3)
they must reduce to

Tij = -p(p, E )c5i j ,

Therefore, the syst em (8.47) is identical to the Eul er equat ions (8.13) . In
other words, an extended theory of five fields is necessarily a theory of an
elast ic fluid.
To account for a mor e accurate description of material behaviors, more
basic field vari abl es are needed and systems of balance equations involving
higher moments will be requ ired . Let

FA = (Fo,Fi ,Fij ,Fijk , " ') E JRN.

The constitutive equations for the non-convective fluxes and the productions
are functions of FA,

GAk = GAk(FB),
A,B = 1,2, ... , N . (8.48)
gA = 9A(FB) ,
Equivalently, in terms of internal parts of moments,

Pa = (p,Pij ,Pijk , " ') E JRN- 3,

we can repl ace (8.48) by the constitutive equat ions for internal parts of fluxes
and productions,

PAk = PAk(Pa) , A = 1,2 , · · · ,N,

(8.49)
1l'A = 1TA(Pa) , Q = 1,2 , . . . , N - 3.

The possible dep endence on the velocity Vi is not allowed by the principle
of material objectivity. Mor eover , sinc e Pa , PAk, and 1l'A ar e all obj ective
quantities, the constitutive functions PAk and 1TA must be isot ropic functions.
A severe restriction is placed on exte nded theories of moment equat ions
by material symmetry. Recall that material symmetry is an invarianc e re-
quirem ent of constitutive functions with resp ect to certain volume-preserving
transformation of reference configurations (see Sect. 3.5) . Since the quantities
Pa, PAk, and 1l'A are invariant with resp ect to any such transformations, the
const it ut ive functions PAk and 1TA ar e invari ant under any volume-preserving
 8.4 Closure Problem 213

change of reference configurations. Therefore by the definition (3.46) of a

fluid, we conclude that exte nded theories , based on the moment equations
considered here, are applicable to flu ids (or gases) only. 1
We have seen that for N = 5 it is a theory of elastic fluids . In general,
we have a hierarchy of fluid models dep ending on the number of fields N . For
examples:

N=5 FA = (Fo, r; Fii),

N = 10 FA = (Fo,Fi ,Fij) ,
N= 13 FA = (Fo, Fi ,Fij ,Fiij) ,
N = 14 FA = (Fo,Fi , Fij ,Fiij, Fiijj) ,
N=· · · FA = . ...
For a theory of fluids with viscosity and heat conduct ion, the minimum
number of fields in the hierar chy is N = 13 , which consists, in addition to
the densities of mass, mom entum, and energy, the fluxes of momentum and
energy, so that the viscous stress and the heat flux are among the basic field
variables.
More higher-order moments can be added to the basic fields , so as to
obtain a better description of fluid behaviors. The higher-order moments do
not usually have immediate physi cal interpretations, however , the presence of
such fields may cont ribute some additional improvements to the fluid models.

Exercise 8.3.1 From the relations (8.25h and (8.39), determine the
mat rices X(v) for the system of moments FA = (F o, Fi , Fij , Fijk) ; and
by (8.31) show that Ak is given by

J k = 1,2,3. (8.50)

8.4 Closure Problem

We consider a material st ate charact erized by the fields u E JRN with their
balance equat ions of the form (8.1),

~~ + div H(u) = g(u) (8.51)

without ext ernal supplies. In ord er to be able to det ermine the stat e functions
u for some initi al boundary value problems, we need the const it uti ve equa-
tions for the flux H(u) and the production density g(u), so that (8.51) will
1 Attempts have been mad e to generali ze mom ent theori es of extende d thermody-
nami cs to solid s in [39, 40].
214 8. Rational Extended Thermodynamics

becom e a closed system of partial differential equations for the fields u(x , t) .
Completion of balance equations with proper cons titutive equ ations to yield
a field theory is one of the main obj ectives of rational thermodyn amics. This
problem is also referr ed to as the "closure problem" , a te rm frequently used
in the kin etic theory of gases . The basic st rate gy of this problem has already
been addressed ea rlier in Sect . 8.2. In t he following sections , the closure pro-
cedur e will be treated in mor e detail by the use of Lagrange multipliers.

8.4.1 Entropy Principle

Besides the Galilean invari anc e discussed in the preceding sections, the en-
tropy principle is also imposed on the system of bal ance equat ions.
Entropy principle. The system of balance equations (8.51) must be
consistent with th e entropy inequality,

8h(u)
----at + div II(u) = E(u) 2: 0, (8.52)

and the entropy density h( u) is assumed to be a concave function .

From the analysis in Sect . 8.2, the ent ropy principle implies the existe nce
of the Lagrange multipliers A( u), which is invertible so t hat we can int er-
change u and A as vari ab les. The existence of the Lagrange mul tipliers for
the ent ropy-ent ropy flux int egrability relations (8.4) imposes great restric-
tions on the const it ut ive fun ctions. Moreover , in order to t ake into account
the formal st ruct ure of the system and the principle of material obj ecti vity,
the entropy-entropy flux int egrability relations (8.4) will be reformulated.
From the form al st ru ct ure of balan ce equat ions, it is evident that the
system dep ends on the velocity in a very explicit manner. The actual calcu-
lations in the closure procedure obviously can take this specific feature into
account by sepa rat ing velocity-dependent expressions in order to obt ain the
obj ective consti tutive equations in terms of velocity-independent variables.
This can be don e by rewriting the entropy- entropy flux integrability relations
(8.4) in terms of velocity-indep endent variables.
With the separ ation of t he fields u into the velocity and the obj ective
fields win (8.21) ,

u = F(v ,w) E JRN,

and the fluxes into convective and non- convective pa rt s (8.22) ,

H(u) = F(v , w) ® v + G(v , w) , (8.53)

 8.4 Closure Problem 215

we sh all also express the entropy flux II in a similar manner ,

II(u) = h(w)v + p(w) . (8.54)
In this express ion, both the ent ropy dens ity h and the (internal) entropy flux
p are not allowed to depend on t he veloci ty, as a consequence of the principle
of materi al objectivity because they are regarded as objective quantities.
By the use of (8.25h , t he relation (8.4h gives
- - aXAB -
dh = AAdFA = AAd(XABFB) = AAXABdF B + AA -:::>-FBdvk
- .
U Vk

Sinc e the entropy h is ind ependent of the velocity Vi , it follows that

aXAB -
dh = ABdFB , - B = 0,
AA -:::>-F (8.55)
UVk

where
A = X(vf A (8.56)
is called the int ernal Lagrange multiplier. Sinc e, from (8.26b X(O) = 1, it
follows that
A(w) = A(O, w) ,
is t he velocity-indep endent par t of A.
Simil arly, t he relation (8.4h with (8.54) and (8.53) gives
d.Il, = d(h vi + cPi) = Vi dh + h do, + dcP i
= AA dHA; = AA d(FAVi + C Ai)
= AAXAB d(FB Vi + C Bi)
+ AA(FBVi + C Bi) dX AB
= A B dC Bi + viAB dFB + ABFB dn, + AA(FBVi + C Bi) dXA B,
and it follows that
- -
dcP i = A B dC Bi ,
- - - - aX-AB = 0.
(A BFB - h)8ik + AA(FBVi + C Bi) -:::>- (8.57)
UVk

Therefore, the abo ve resul ts can be st at ed in t he following

Proposition 8.4.1 If th e syst em of balance equations (8.51) is Galilean in-

variant, then the m at erial objectivity and th e entropy prin ciple imply that
there exists an internal Lagrange multiplier A satisfying
(8.58)
such th at
(8.59)
hold and th e entropy production is given by
17 = A .9 2 0. (8.60)
Moreover, h(F), p(F) and 17(F) as well as G(F) and g(F) are isotropic
functions.
216 8. Rational Extended Thermodynamics

Proof The relations (8.58) follow from (8.55h and (8.57h evaluate d at v = 0
and the definition of the matrix A k defined in (8.31). The entropy production
(8.60) is a trivial consequence of (8.3h, (8.25h , and (8.56) . 0

We sh all call the relations (8.58) the Lagrange multiplier 's identities and
(8.59) the entropy-entropy flux integrability relations. The integrability rela-
tion can be written in an equ ivalent form through a Legendre transformation
similar to (8.6) . We define h and ip as
(8.61 )
then the integrability relations (8.59) become
- -
dip = G ·dA.
~

(8.62)
Not e that both h(A) and ~(A) are isotropic functions of A.
8.4.2 Formal Procedures
The ext ended theory, in the framework of rational thermodynamics, was first
formulated by Liu and Muller in [46], in which the closur e procedure was
based on the ent ropy- ent ropy flux int egrability relations (8.59) ,
(8.63)
by usin g the constitutive representations of h, ip and G in t erms of F in a
rather complicate d calculat ion (also see [55]).
A different pro cedure, based on the relation (8.62)
~ - -
dh = F ·dA , (8.64)
was proposed later (see [57]). It is clear that the fluxes G(A) , as functions of
the internal Lagran ge multiplier A, ar e complete ly det ermined by the knowl-
edge of the vector function ~(A). On the other hand, the knowledge of the
scalar fun ction h(A) gives the functions F(A), from which the Lagrang e mul-
tiplier A in terms of F can be obtain ed by a functional inversion . Then , by
a simple substitution , t he const it ut ive equat ions of G(F) = G(A(F)) can
be obtain ed . T his procedure based on the pai r of functions (h, ~) , which are
equivalent to the generator functions (h,'ii; is very elegant . The inversion
is relatively easy for a linear theory, however, for higher-order const it ut ive
equ ations, the inversion of a system of nonlinear equat ions may become mor e
difficult if it can be don e at all.
Her e, we sh all employ an alte rn at ive procedure for which there is no
need for a functional inversion of the Lagrange multipliers but still maintains
the elegant feature of the generator functions. It relies on the hybrid pair
of generat or functions (h(F) , ~(A)), with their inte grabili ty relations (8.63h
and (8.64)z,
(8.65)
 8.5 Thirteen-Moment Theory of Viscous Heat-Condu cting Fluid 217

From which we can obtain

A = A(F) = : ; ,

and by a simple substitution of the first relation into t he second, we imm e-

diately obtain the constitutive equat ions of the flux G in terms of t he field
vari abl es F,
G = G(A(F)).
This pro cedure will be illustrated by the theory of 13 fields for viscous heat-
conducting fluids in the following section. However, for simplicity, only linear
const it ut ive equat ions will be derived . The procedure can be applied to ob-
t ain higher-order const it uti ve equat ions, for which calculat ions are generally
cumbersome but mostly st raight forward (see [44]).

8.5 Thirteen-Moment Theory

of Viscous Heat-Conducting Fluid

We consider the st ate of a viscous heat-conducting fluid to be charac-

t erized by the fields of 13 mom ent s, (Fo, Fi ,Fij , Fiij) , and from (8.42) , we
have
F = (p,O,Pij ,Piij) ,
G = (O ,Pik ,Pijk ,Piijk) ,
g = (0,0 , 7f(i j ) , 7fiij) .
The ent ropy principle impli es t he existe nce of the internal Lagrange mul-
tiplier denoted as
A = (A ,Ai ,Aij ,Aj) ,
for the integrability relations (8.59) , which now t ake the form s:

dh = A dp + Aij dPij + Aj dPiij , (8.66)

dih = Ai dPik + Aij dPijk + Aj dPiijk . (8.67)

The Lagrange multipliers must also satisfy the identities (8.58) , which by the
use of (8.50) can be written as

Aj P + 3A(i Pij) = 0, (8.68)

(A P + Aij Pij + Aj Piij - h )151k + 2A1j Pjk + 3A(j Pjl)k = 0. (8.69)

From (8.60), we also have the inequality for t he ent ropy production density,

(8.70)
218 8. Rational Extended T her mody namics

Exercise 8 .5.1 Verify the identities (8.68) and (8.69) by direct calcu-
lations for the veloci ty ind ep endenc e of h and P i from (8.4) by the use
of explicit expressions for FA and G A i .

Absolute Temperature and Non-Equilibrium Variables

We define equilibrium as a process with no ent ropy production. From (8.70) ,
the entropy production has a minimum in equilibrium for any process with
no productions , i.e., E = 0 when Jr (ij ) = 0 and Jr i i j = O.
Sin ce we can regard E , Jr (i j ) and Jri i j as fun ctions of A, therefore, if we
assume the invert ibility so t hat we can repl ace (A (ij ), Aj) by (Jr(i j ) , Jr i i j) as
variables in E , then from (8.70) the necessary cond it ions for t he minimum of
the ent ro py product ion imply t hat

~EI
uJr,] E
=A(ij )IE=O , oE I
OJriij E
= Aj IE = O, AjI E =O.

The last condition follows from (8.68) . In other words, t he Lagr an ge multi-
pli ers A(ij ), Aj and A j must also vanish in equilibri um.
With the convent ional physical qu antities (8.45) , (!ii = 2pc, and t he
spec ific entropy density 7J = hlp , in equilibrium t he relation (8.66) reduces
to
2
dhl E = d(p7JI E) = Al E dp + "3AiilE d(pc) . (8.71)
By reference to t he Gibb's relation of t hermostat ics (5.30h , it follows t hat
1 2 1
o = "3AiiI E' Al E = -o glE' (8.72)
where {} is the abso lute t emper ature, g is the free ent halpy func t ion ,
P
g =c- (}7J + -, (8.73)
P
and P = Pii!3 is the mean pressure.
We shall be conte nt with formula ting t he theor y for thermod yn amic pro-
cesses "near" equilibrium . Mathematically, this means that non- equilibrium
variables will be regarded as sm all qu anti ti es of the first-order and constitu-
t ive fun ct ions will be represented in terms of them up t o cert ain order.
Since t he Lagrange multipliers A(ij ), Aj , and Aj vani sh in equilibrium,
they are non-equilibrium qu ant ities. We can also decompose t he Lagrange
multipliers A ii and A into the equilibrium and non- equilibrium parts,
3 1
Aii = 2{} +A 6 , A = - OglE + Ap . (8.74)

Therefore, (Ap , A6 , Ai , A(ij ), Ai) form s the set of non- equilibrium variables of
the Lagr ange mul t ipliers .Ii.
On t he other hand, what are the non-equilibrium variables of t he mo-
men ts F? To answer t his , we st art by regarding F as funct ions of .11, and
 8.5 Thir teen-Moment Theory of Viscous Heat-Conducting Fluid 219

hence, from the decomposition (8.74), we can write

Pij = Pij(P, 0, Ap, Ao, Am , A (mn), Am) ,

Piij = Piij (p, 0, Ap, Ao, Am , A(mn), An) .

Since all qu an t iti es in these express ions are objective, t he principle of materi al
object ivity requires t hat for any orthogonal t ensor Q,

Pij(QmpAp, QmpQn qA (pq ), QmpAp) = QikQ jl Pkl(A m , A (mn), An) ,

Piij(QmpAp, QmpQnqA (pq), QmpAp) = Q jl Piil(A m , A (rnn), An) ,
in which t he irr elevant scalar var iables have not been included . After evalu-
ating the above relations in equilibrium, we obtain

Pij(O, 0, 0) = QikQjl Pkl(O , 0, 0),

Piij(O, 0, 0) = Qjl Piil(O, 0, 0),

i.e. , for any orthogon al te nsor Q,

Therefore, it follows t ha t

P(ij ) IE = 0, Piijl E = 0. (8.75)

The second one follows from the fact t hat not every orthogonal t ensor has t he
same eigenvalue, while the first one follows from the commut ation theorem
(see p. 258), which asserts that Pij IE must be a multiple of t he identity tensor.
Therefore , we have t he following two set s of non- equilibrium variables
for t he moments and the Lagran ge multipliers, respect ively,

(8.76)

Representations of Hybrid Generator Functions

From (8.66) we have t he int egrability rela tion ,

2
dh = A dp + 3" A ii d(pc) + A (ij ) dp (ij ) + Aj dPiij ,
which , by the use of (8.71) , (8.72) , and (8.74), can be wri tten as

(8.77)
220 8. Rational Extended Thermodynamics

According to (8.61), we now define

(8.78)

and from (8.67), we have the relations,

~ 1 2
dtPk = - B2 qk dB + "3qk dA£ + Pik dAi + P(ij )k dA (ij ) + Piijk dAj . (8.79)

The two relations (8.77) and (8.79) are in the form of the hybrid pair (8.65) ,
from which we can regard hand ¥k as functions of

h = h (p, B, P(ij ), Piij) ,

(8.80)
¥k = ¥k(p, B, A£ , Ai, A (ij ), Aj) .

Of course, from (8.79) we have

a¥k = 0 (8.81)
ap .
We remark that although, from (8.68) ,
10 2
Ai = -3c Ai - pP (ij)Aj , (8.82)

and hence Ai can be eliminated from the list of independent variables for the
function ¥k' we have decided not to do so in order to maintain the simplicity
of the integrability relation (8.79) .
From the requirement of material objectivity, the const it ut ive functions
for hand ¥k are isotropic functions . Since (P(ij) , Piij) and (A£ , Ai, A (ij ), Aj)
ar e small qu antities of the first-order, we can easily write down their repre-
sentations up to the second-order terms:

h = hiE + hI PiijPnnj + h2 P(ij)P(ij ) + 0(3) , (8.83)

¥k = a Ak+(3 Ak+aI A£Ak+bI A£A k+a2 A (kj )Aj+b2 A (kj)Aj+o(3) , (8.84)

where all the coefficients are functions of (p, B) .

Our objective is to determine the constitutive equations of the fluxes
(qk' Pik , P(ij )k' Piijk) as functions of the fields (P,B,P(ij ),Piij) . For the lin-
ear constitutive equations, we shall need only the representations (8.83) and
(8.84) up to the second-order terms. The entire calculations are based on the
four relations, namely, the entropy-entropy flux integrability relations (8.77),
(8.79), and the Lagrange multiplier's identities (8.68) (or (8.82)) and (8.69).
We shall introduce the following abbreviations to be used later,

(8.85)
 8.5 Thirteen-Moment Theory of Viscou s Heat-Conducting Fluid 221

Entropy-Entropy Flux Integrability Relations

First we obtain t he Lagrange mul t ipliers from the int egrability relation (8.77)
and the expression (8.83),
A€ = 0(2),
Ap = 0(2),
Aj = 2h1 Piij + 0(2), (8.86)
A (i j ) = 2hz P (ij ) + 0(2).
Similarl y, from (8.79) and (8.84) we have t he the following expressions for
the fluxes,
2
3" qk = a l Ak+ b1Ak + 0(2),
Pik = f3 bik + b2A (i k ) + 0(2), (8.87)
P (ij )k = a2 bk (iA j ) + bz bk (iA j ) + 0(2),
Pi i j k = a bj k + a2A (jk ) + 0(2),
and
(8.88)

(8.89)

Lagrange Multiplier's Identities

From (8.82) , (8.86) leads to

20
A j = - 3ch1 Piij + 0(2). (8.90)

The identity (8.69) , afte r subst it ut ion of the Lagran ge multipliers, gives an
identity in the field var iabl es. In particular , the first-order t erms of the t race
par t are identically satisfi ed , while the traceless part lead s to

e1 P (kl ) + 2p A (kl ) = 0,
from which the relation (8.87) impli es that

f3 = P , bz = -2p8. (8.91)
From the above results, we can obtain, from (8.88) and (8.89),
000 10 op 3
08 - 3 c 08 = - 28 2 AI, (8.92)
000 _ 10 e op =0 (8.93)
op 3 op .
222 8. Ration al Exte nded Thermodynamics

By t he use of (8.82) , (8.93) permits the relat ion (8.81) to b e fur ther evaluate d
up to the second-order terms. It read s
1 op
or -- (8.94)
ph 2 op '

Summary of Results
Summar izing t he above results, we can now obt ain the linear const itutive
equations for the fluxes:

Pik = POik - 4pf)h 2 P(ik ) + 0(2),

qk = 3A Ih IP i i k+ O (2), (8.95)
P(ij )k = 2 A 2h I Ok(i Pj )nn + 0(2),
Pi i jk = a Oj k + 2 a2 h 2 P(jk ) + 0(2),
where
A = _~f)2(Oa _ 10 OP)
I 3 of) 3 Eo f) ,
20
A2 = a2 + :3 Bpe, (8.96)

h
2
= ~ OP(OA 2) -I.
pop Bp
On inspection of these const it utive equat ions, we conclude that besid es
the equilibrium stat e functions E(p, f)) and p(p, f)) , only three unknown mate-
rial functions, hI (p , f)) , a2(p, f)) , and otp, ()), are needed . Note that the fun c-
t ion a can be det ermined from the equilibrium st ate functions to wit hin a
fun ct ion of f) by int egrating (8.93) .

Concavity of Entropy Function

°
In the ent ropy principle , t he ent ropy density is assumed to b e a concave
fun ct ions of u. By evaluat ing at v = and ov = 0, the condit ion (8.12) can
now be expressed as

(8.97)

for any variations of (p , f), P(i j ) , Pii j) .

In par ticular, If we consider a vari ation with op i=- 0, of) i=- and OP(i j ) =
0, 0Pi i j = 0, then by the use of t he relations (8.71) t hrough (8.74) , we ob t ain
°
from (8.97)
lop 2 (}& 2
pf)oP(Op) + f)2 > 0, (8.98)
of)(O{))
which implies that
op &
op > 0, of) > 0. (8.99)
 8.5 Thirteen-Mom ent Theory of Viscou s Heat-Conducting Fluid 223

The first inequality states t hat the isothermal compressibility is pos itive,
while the second one ensures the positiveness of the sp ecific heat at constant
volume. These conditions ar e equivalent to the condit ions of thermodynamic
stability cons idered in Sect . 5.5.
Similarly, if we consider a variation with op = 0, oB = 0 and OP(ij ) =F 0,
0P i ij =F 0, then from (8.86) it follows t ha t

2h 2 o p (i j )OP (i j ) + 2h 1o Pi i j OPn nj < 0,

which leads to
hI < 0, h 2 < O. (8.100)
Of course, t here ar e other restrictions imposed by the inequality (8.97) ,
but they are of no imm edi ate int erest here.

Exerci se 8.5. 2 Verify the inequality (8.98).

P r o ductio n D ensities
The production densities can also be given by linear representations:
1
1rij = - - P (i j) + 0(2),
Ts
(8.101)
1
1riij = - - Piij + 0(2),
Tq

since they are isot ropic functions and vanish in equilibrium. Both T s (p, B) and
Tq(p, B) are called relax ation t imes . Accordingly, from the relation (8.70) the
ent ropy production density is given by
h2 hI
E = - 2 -T P (ij )P (i j ) - 2-
Tq
Pi i jPi ij + 0(3), (8.102)
s

and since it is non -negative, from (8.100) we have

Ts 2: 0, Tq 2: o. (8.103)

8.5. 1 F ield Equations

The system of balanc e equa tions for 13 moments can be written in t erms of
internal mom ents from (8.44) ,
· a Vi
P +P - = 0,
a Xi
. api j
PVi + - a =0 ,
Xj
(8.104)
· aVk apijk a V(i
Pi j + Pi j -
a Xk
+ - a- -
Xk
+ 2-a Xk Pj)k = 1ri j ,
aV
·
Pi i j
api i j k
+ P i i j -aa XVkk + -a--
Xk
+ 3 -a Xk(i Pij)k + 3P(ii Vj)
'
= 1ri ij ·
224 8. Rational Extended Thermodynamics

With the constitutive equations, one can get a system of field equ ations for the
13 fields (p , Vi, Pij , Pi i j). However , it is more convenient to repl ace (Pi j , Pii j) by
the conventi onal variables ((), Si j , qj) , namely, the t emperature, the viscous
stress t ensor , and the heat flux, resp ectively. The viscous stress t ensor is
defined as the traceless part of the stress t ensor,
Ti j = - Pij = -P Di j + Si j , S(ij ) = O. (8.105)
From (8.95) and (8.101) we obtain
Pii = 2 pe,
1
P(i j ) = 4p()h Sij
2
+ 0(2) ,
1
Pi i j =3AIhIqj+0(2) , (8.106)
2A 2
P(i j )k = 3" Al Dk (i Qj) + 0(2) ,
a2
Piijk =a Djk + 2p() Sjk + 0(2) ,
and
1
1rij 4 ()h S,)
P 2Ts
+ 0(2) ,
1
3A h
I I Tq
q) + 0(2) .
ON subst it ut ion of the above constitutive equat ions into (8.104) , we obtain a
system of qu asi-linear first-order partial differential equations for the 13 field
variables (p , (), Vi , Si j , qj) ' T his is a hyp erbolic system of field equa t ions for
viscous heat-conducting fluids.

Linearized Field Equations

In ord er to associate the t hree unknown functions tis, a2, and a and two
other coefficient s, T s and T q , with more suggestive physical par am et ers, we
shall lineari ze t he field equ ations by leaving out all nonlinear t erms, such as
the products of Sij and qj with the derivatives of P, () and Vi' They read
. oVi
P + P-;:;- = 0,
UXi
. op oSij 0
PVi + - - - - = ,
o Xi o Xj

Po()
Oc e+ ()opo() oo XiVi + ooqiXi -- 0
,
(8.107)

. o V(i oq (i
TsSi j - 2J-l-;:;-- - 2J-lK -;:;-- = -Si j ,
UXj ) UXj )

T q + K,~ - K,K ()o Si j = -q .

q ' oXi o Xj ,.
 8.5 Thirteen-Moment Theory of Viscous Heat-Conducting Fluid 225

In these equations, we have identified five material parameters that are mea-
surable, at least in principle, namely, T s and T q for the relaxation times of
viscous stress and heat flux, and

K=~A2 (8.108)
3pA 1 '

for the shear viscosity, the thermal conductivity and the thermal-viscous
coupling coefficient, respectively. From (8.100) and (8.103) , we conclude that

(8.109)

Therefore, the shear viscosity J-l and the thermal conductivity '" as well as
relaxation times T s and T« are all non-negative quantities.
Note that if the relaxation times and the thermo-viscous coupling are
neglected, the equations (8.107) 4,5 reduce to the usual Navier-Stokes and
Fourier laws with vanishing bulk viscosity (see Sect. 7.2.2).

8.5.2 Entropy and Entropy Flux

The constitutive equations for the ent ropy density TJ and the entropy flux P
can be obtained up to the second-order terms directly from the representa-
tions (8.83) , (8.84) and the definition (8.78) . With the material parameters
defined in the pr evious section, they can be written as

(8.110)

The relation (8.110h asserts that the difference between the entropy
flux P and q /() is of second-order, due to the thermo-viscous interaction.
Therefore, the assumption (5.9h that the heat flux and the entropy flux are
proportional in the Clausius-Duhem inequality, is generally not valid .
The linearized equations (8.107) and the constitutive relations (8.110)
were derived by Miiller in 1966 [49] in the first ph enomenological extended
theory of mom ents within the framework of the then pr evailing theory of
thermodynamics of irr eversible processes (see, for example, [14]). Unlike the
rational thermodynamics we have considered in this book, thermodynamics
of irr eversible pro cesses rests upon some heuristic assumptions, notably, a
modified Gibbs relation to include non-equilibrium vari ables. Although this
approach has been widely used in many branches of physics (known as ex-
te nded irreversible thermodynamics, see [31]), it lacks the rational structure
of modern thermodynamics and the results concern, at most, linear const i-
tutive equ at ions .
226 8. Rational Extended Thermodyn amics

8.6 Monatomic Ideal Gases

In the kin eti c t heory of ideal gases, the equations of moment s are obtained
from Maxwell's equati on of t ransfer for ideal gases (please refer to any st an-
dard textbook [24, 34, 55, 70] for derivation) , which, in the ab sence of an
exte rnal force, can be expressed as

87jj 81jJcj = p(n l') (8.111)

8t + D»,
J
'f' ,

where the overhead bar denotes the exp ect ation valu e defined by (8.34) . The
last t erm denotes the production density due to molecul ar collisions. The
explicit form of P(1jJ ) is irr elevant here. It suffices to know that P(1jJ ) is linear
in 1jJ and
P(m) = 0, P(mCi) = 0, (8.112)
which follow from the conservation of mass, linear mom entum and energy in
the collision pro cess.
Wi th the moment Fi, ...i n defined by (8.35) , t aking 1jJ = m Ci, .. . Ci n in
(8.111) , we obtain
8Fi, ..·in 8Fi, ···inj ( )
8t + 8x . = % ..·i n , 8.113
J
where gi, ...in = P(m Ci, ...iJ denotes the production density of the moments .
It is int eresting to note that in this equat ion, the flux of the n- th mom ent
Fi, ...in is t he same as the (n + 1)-th moment Fi , ...in+, . This forward-linking
property between the moment and its flux is a characteristic feature of ideal
gases.
Comparing this relation with (8.40) , we conclude that ideal gases is a
special case of the general structure considered in Sect 8.3 with

or, from (8.41) and (8.42) ,

(8.114)
In par ticular, Pii = Pii = 3p, and from the definitions (8.45) , it follows that
3p
c = --. (8.115)
2 P
It is known that this relation is valid for monatomic ideal gases only. In
ot her word s, the simple structure of the forward-linking property of moment
equat ions (8.113) is only applicabl e to noble gases, such as helium and ar gon .
For classical ideal gases, the equa t ion of st at e is given by the usual ideal
gas law,
P = RpB, (8.116)
where R kim is the gas cons tant , m is t he at omic mass and k is the
Boltzmann constant .
 8.6 Mon atomic Ideal Gases 227

8.6.1 Thirteen-Moment Theory

W ith the relations (8.114) through (8.116), we can further evaluat e the con-
stitutive relations in the pr evious sect ion. Surprisingly, all the unknown fun c-
t ions can be determined explicitly.
First , we int egrate (8.93) to yield

a = 5 R 2p02 .
In the int egration with respect to p, we have set the arbitrary function of 0
to zero for simplicity and we shall do so hereafter without further remarks.f
From (8.96h, we obtain

A 1 -- - 3lOR203
p ,

and since P(ik ) = P (i k ) and Pi ik = 2qk , from (8.95h,2 we have -4pOh 2 = 1

and 6A Ih I = 1, which lead to
1 1
hI = - 20 R2p03'
We can now integrate the relation (8.96h for A 2 ,

which , from (8.96h gives

a2 = -14 R 2 p03 .
The coefficients al and bI are not individually determined, instead, they
ar e related through t he value of AI . Henc e, only one of them, say bI , is
undet ermined . Nevertheless, the explicit value of bI is not needed in the
linear const it ut ive equations. Therefore, all material fun ctions for the fluxes
have been explicitly determined .
Mor eover , from (8.86) , the first-order non-equilibrium Lagrange multi-
pliers are given by

Ac = 0,
Ap = 0,
1 1
x, = -5 R2p0 3 qi ,

2 By doing so, the results are consistent with those obtain ed from the kinetic
theory.
228 8. Rational Extended T hermody nam ics

8.6.2 Constitutive Equations

From the above results, the first-order const it ut ive equat ion for t he fluxes
and P i i j k can now be calculate d,
Pi j k

Pi j = RpB8i j - S i j ,
2
P i jk = 5" (qi 8j k + qj 8k i + qk 8i j ) + 0(2) , (8.117)

Pii jk = 5 R 2 pB 2 Ojk - 7 RB Sjk + 0(2) .

Mor eover , we ca n also det ermine the cons titutive equa t ions of the ent ropy
density TJ and the ent ropy flux tPk from (8.78) ,

They are given by

(8.118)

Not e that in the linear theory, they can be det ermined up to the second-order
t erms.
We remark that all the coefficients of the representations in the above
constitutive equati ons are explicit ly det ermined . These amaz ing results of t he
rat ional exte nded t herm ody namics were first derived by Liu and Miiller [46] .
They are in complete agreement with t he resu lts from the kin eti c theory of
gases [15, 34].

8.7 Stationary Heat Conduction in Ideal Gases

Extended t hermody namics provides a hier ar chy of hyp erbolic systems of first-
order partial differential equations. Applications for ideal gases , such as heat
wave and shea r wave propagations, shock st ruc t ure , and light scattering, are
some of the phenomena success fully treated within exte nded thermodynam-
ics. The exte nded t heories have shown great improvement of results beyond
Navier-Stokes-Fourier theory in ordinar y thermodynamics. Syst ems of higher
hierarchy are usually required t o achieve bet t er results in comparison with
expe rime nt s. Discussion of various aspec ts of the theory and applications ca n
be found in the mon ograph by Miiller and Ruggeri [57] .
As a simple applicat ion of exte nded thermodyn amics, we shall consider
boundar y value problems of st ationary heat conduct ion in ideal gase s. In
thirteen-moment theory, we shall see that Fourier 's law of heat conduction is
no longer valid, in general.
 8.7 St ationar y Heat Conduction in Ideal Gases 229

8.7.1 Fourier's Law and Heat Conduction

For st ationary heat conduction in an ideal gas at rest, the equat ions of equi-
librium and energy are given by
- s'!» + S i j
,) <
,) < = 0,

where gij is the contravariant component of the metric tensor and a subscript
comma (,) denotes the covariant derivative (see Sect . A.2.4) . In ordinary
theories, Fourier's law of heat conduct ion is given by
(8.119)

where the conduct ivity r: is a constant for ideal gases. Therefore , the govern-
ing equat ion for heat conduct ion becomes
<k
gJ (},j k = O.
In other words, the te mpe ra t ure field () satisfi es the Laplace equation. On
t he other hand , the pressure p is a constant, since the viscous stress S ij = 0
in a Navier-Stokes fluid when the gas is at rest . If, at the boundar y, either
the t emperature or t he heat flux is pr escribed , it is known that the boundary
valu e problem is well-posed . For purposes of comparison, the following simple
exa mple will be considered .

Heat Conduction Between Coaxial Cylinders

The inner and outer rad ii of two cylinders are denoted by r i and 1'0 ' resp ec-
tivel y, and we consider an axial sym met ric solution ()(1' ) where r is the radial
coordina te . In this case , the Lapl acian of (}(r) becomes
2
d (} +~ d(} = 0
d r2 r dr '
for r j < r < 1'0 ' We also pr escrib ed the following boundary condit ions:
(8.120)
where q is the r- component of the heat flux. The solution is given by
j
(}(r) = () - qjr log !..- . (8.121)
o Ii 1'0

8.7.2 Heat Conduction in Thirteen-Moment Theory

Extend ed thermodyn ami cs of the lowest hierarchy for the study of heat con-
duction problem is the theory of t hirt een moments, (p , v i , pi j , pi i j ) . We con-
sider a st ationar y pro cess at rest, vi = 0, for mon atomic ideal gases. In this
case, t he equation of mass balanc e is identically sat isfied .
In the cylindrical coordina te system - (x 1 , x 2 , x 3 ) = (r,'l9 , z), for the
coordinat e, 'l9 is used to avoid confusion with t he temperature field () - with
230 8. Rat ion al Exte nded T hermody namics

(8.117) we have the following system of moment equa t ions:

OXk k,
Ok
j k Op _ oSJ _ r iSi k _ r k Sji =
9 OXk k,
°,

Oqk k i
OXk + rk iq = 0,
(8.122)
2(
-5 l ik: Oqj
-- + gJk -
Oqi
- +i kT J
° 1 'k
q + gJ r
I) 1
' q = - S' J
° ° °

ox k ox k k1 k iT '
ik:
5Rgjk O(pB) _ 7R O(BSJ ) _ 7RB(r iSik +r k Sji ) = -~ qj .
OXk OXk k, k, T

The metric te nsor and the Christoffel symbols for cylindrical coordinate sys-
t em can be found in Sect . A.2 .7 (b) .
In these equat ions, the production densit ies are given by (8.101) with a
constant relax ation time T = Ts = Tq , corresponding to the BGK model for
molecular int eraction - the simplest linear model in t he kin eti c theor y [6] .

Heat Conduction Between Coaxial Cylinders

We now consider again the boundary value problem of heat conduction be-
tween two coaxial cylinders. T he solution is assumed t o be axially symmetric
and hence all fields are fun ctions of r only. The heat flux is also assumed t o
be in t he radi al dir ection, qr = q(r) an d q{} = qZ = 0.
From (8.122h it follows that
dq 1
- + - q = 0,
dr r
and t he solu tion is given by
c
q(r) = -, (8.123)
r
where C = qirj from the boundary condit ion (8.120h .
Equation (8.122h t hen impli es t hat

s rr = ~ T dq = _~ TC ,
5 dr 5 r2
(8.124)
SM = ~ T~ q = ~ TC4
5 r3 5 r '

and the other component s of the stress are zero . Moreover , by the use of
(8.124), we obtain Sj k,k = 0, i.e.,
Ok
oSJ ik: k
+ r /is' + r k i SJ'
° 00

oxk = 0.

Therefore, (8.122h implies that the pr essure p is constant and the last equa-
tion of (8.122) leads to the following relation between the heat flux an d t he
 8.7 Sta tionary Heat Conduction in Ideal Gases 231

temperature gradient,

5
K, = 2' Rrp, (8.125)

where K, is the thermal conduct ivity. Note that the pr esence of the normal
st ress makes Fourier 's law of heat cond uct ion no longer valid in the 13-
moment theory. In the Navier-Stokes-Fourier theory, of course, S ij vanishes
identically in a gas at rest.

550

500

450
8
400

350

300
0.01 0.015 0.02 0.025 0.03
r

Fig . 8.1. Temperature 8(r) of 13-moment th eory (low er curve) and Fourier theory
(upper curve )

Finally, we can determine the temperature field. Substitution of (8.123)

and (8.124) into (8.125) gives

de e r b = 28.:::: .
dr K, r 2
+ be ' 25p

With the pr escribed boundary condit ions (8.120) of Example 8.7.1, the solu-
tion of 13-moment theory is

(8.126)

This result has been plo tted together with the result (8.121) of Fourier theory
for comparison in Fig. 8.1 for argon and the following data:
r j = O.Olm , r: = 0.03m, qj = 104W /m 2 , eo = 300K,
r = 10- 5 s, P = 100 Pa, K, = 0.5164 W /rn/K.
Note that the temperature for the 13-moment theory is slightly lower than
for the Fouri er theory, and the lar gest difference is less than 1.80% at the
inner surface.
232 8. Rational Extended Thermodynamics

8.7.3 Remark on Boundary Value Problems

Generally speaking, a system of field equat ions with mor e variables requires
more boundary condit ions in order to have unique solutions. However , in
the heat conduction problem considered in this secti on , the usual boundary
condit ions (8.120) for t he Fourier t heory of heat conducti on turn out to be
sufficient for t he determination of unique solutions within t he 13-moment
t heory. This is generally not so in exte nded thermodynami cs.
The pr esence of high er mom ents in exte nded thermodyn ami cs pr esents
an addit ional problem rega rd ing the sp ecification of their initial and bound-
ary values, becaus e usually higher moment s do not have meaningful physical
int erpret ations and can not be measured in convent iona l physical qu antities,
and hence we do not know how to prescribe their valu es. The problem with
initial values seems to be of less concern , becau se in most of the applica-
tions, initial valu es of non- equilibrium higher mom ents can be set to zero if
necessary. This has been don e frequ ently in wave and sho ck problems [57] .
However , with respect to boundary values of higher mom ents, t he problem
is more difficult . Mathem atically they have to be assigned in order to have
unique solutions. In ot her word s, the dep end en ce of such boundary values will
result in t he non-uniqueness of the solution. However , not only are t hey un-
convent ional physical qu an tities, ar bitrary ass ignment of addit ional boundar y
valu es is inconsist ent with the nature of well-posedness of the corres ponding
physical problem in ordinar y theories .
Therefore, it is necessary to est ablish some kind of criterion to determine
t he valu es of such qu antities, so that the solut ion is un ique and physi cally
relevant in comparison with the solut ion of the corresponding problem in
ordinary theories. In [67] Struchtrup and Weiss made the first atte mpt to
consider such a crite rion and they proposed to find t he uni que solu tion by
requiring the ent ropy product ion to be as small as possibl e by the so-called
Minimax Principle, similar to Prigogine's Principle of minimum ent ropy pro-
du ction [2, 60]. However , both principles do not always lead to reasonably
consiste nt results. The boundary valu e problem in exte nded thermodynami cs
has come into discus sion only recently and it is still lar gely an un explo red
problem .
A. Elementary Tensor Analysis

This appendix is int ended to provide the mathematical preliminaries needed

for a clear and rigorous pres entation of the basic principles in cont inuum
mech an ics. The reader is expect ed to be familiar with some notions of vec-
tor spaces or matrix algebra. In the first part, we shall review some basic
notions of vector sp aces and linear t ran sformations. At t he sam e time, ele-
mentar y properties of tensors as well as tensor not ations will be introduced .
All mathemati cal symbols ar e set in italic. Generally, sca lars are represented
by letters (a, b, 0:, ... ) of normal typeface, vectors by lowercase boldface letters
(u ,v, · ··) , and tensors by upper case letters (T,F, · · ·) . The identity t ensor is
represent ed by 1 (italic " one" ).

A .I Linear A lgebra
We shall consider finit e-dimensional real vector spaces only. The field of real
numbers is denoted by JR .
D efinit io n. A vector space V is a set equippe d with two op erations:
1) v + u E V , called addit ion of v and u in V ,
2) o:v E V, called scalar multiplication of v E V by 0: E IR,
which satisfy th e following rules: for any v , u , w E V , and any 0: , (3 E JR ,
1) v +u=u+v.
2) v +(u +w) =(v+u)+w .
3) Th ere exists a null vector a E V , such that v + 0= v .
4) For any v E V , th ere exist - v E V, sucli th at v + (-v) = O.
5) o:((3v) = (o:(3)v.
6) (0: + (3 )v = o:v + (3v .
7) o:(v + u) = o:v + o:u.
8) 1v =v .

D e fini t io n . A set of vectors {VI ,· · · ' v n } is said to be a basis of V , if

1) it is a linearly independent set , i.e., for any aj, · ·· , a n E IR,
if a jVj + ... + anV n = a th en a l = . . . = an = O.
2) it spans th e space V , i.e., for any u E V , the vector u can be ex pressed
as a linear com bina tion of { v I , . . . , v n } .

I.-S. Liu, Continuum Mechanics

© Springer-Verlag Berlin Heidelberg 2002
234 A. Elementary Tensor Analysis

Let {e 1 , . .. , en } be a basis of V , then any vector u E V can be expressed

as
n

u= L uiei'
i= l

where u i , ca lled the components of u , are uniquely det ermined relative to the
basis {e.}.
A vector space can have many different bases, but all of them will have
the same number of elem ents. The number of element s in a basis is called the
dimension of the space, in this case, we hav e dim V = n .

A.I.! Inner Product

We may think of a vector as a geometric object that has a lengt h and points
in a certain direction. To incorporate this notion we introduce an addit ional
structure, inner product, into t he vector space.
D efinition. An inn er product is a m ap
g :VxV-+JR

with the following properties: For any u , v, w E V , and a E JR ,

1) g(u + av, w) = g(u , w) + ag(v, w ),
2) g(u, v ) = g(v , u) ,
3) g(u , u) > 0, if u=l-O.
An inn er product is a positive-definite symmetric bilinear function on
V . We call g(u , v) the inner product of u and v. The vector space equipped
with an inner product is called an inner product space. Hereafter, all vector
sp aces considered are always inner product spaces.
Notation. g(u ,v) = u · v , if 9 is given and fixed .

D efinition. The norm of a vector v E V is defined as

Ivl= ~·
A vector space equipped with such a norm is called a Euclidean vector space.

The notion of angle between two vectors can be defined based on the
following Schwarz inequality:

lu·vl :::; lullvl · (A.I)

D efinition. For any non-zero u, v E V, the angle between u and v , B(u , v) E

[0, n], is defined by
u·v
cosB(u , v) = lullvl.
 A.I Linear Algeb ra 235

The vectors u and v are said to be orthogonal if B(u , v) = 7f 12. Obv i-

ously, u and v are orthogonal if and only if u . v = O.
A vect or v is called a un it vector if Iv I = 1. T he proj ection of a vector u on
t he vector v can be defined as lui cos B(u, v) , or as (u · e) , where e = v i Ivl
is t he unit vect or in t he dir ect ion of v . The vect or (u . e) e is called the
projection vector of u in t he direction of v .
Let {e. , i = 1," ' , n} be a basis of V . Denote t he inner product of e,
and ej by gi j,
gi j = e i . ej '

Clearl y, gij is symmet ric, gij = gji . Let u = u i ei ' v = v j ej be arbit rary
vectors in V expressed in terms of t he basis {e.} . Then

u ·v = (u i ei) . (v je j)
= u i v j(ei ' ej) = u i v jgi j ,

or
j
u ·v = gijUi V . (A.2)
Here we have used t he following summat ion convention.
Notation. (Summation convent ion) In the expression of a term , if an index
is repeated once (and only once), a summa tion over th e range of this index
is ass umed.
For example,
n
Ui " ' i e. ,
e, = 'L..tu
i= 1
n n
j j.
gijUi V =L Lgi jUi V
i= 1 j = 1

Not e that in these express ions, we purposely write the ind ices in two different
levels so that t he rep eat ed summation indi ces are always one superindex and
one subind ex. The reason for doing so will become clear in the next section.

A.1.2 Dual Bases

Let { e1, " ' , en} be a basis of V . There exists a non-zero vector orthogonal
to the plan e spanned by the n - 1 vectors {e2 , ' . . , en }, and if, in addit ion,
the proj ection of this vect or on e 1 is prescribed , then this vector is un iquely
det ermined . In this manner, for any given bas is {e 1' . . . , en }, we can const ruct
a set of vecto rs {e" , . .. , en} such t hat
i si
e · ej = u j'
236 A. Elem entary Tensor Analysis

where tS i j is called the Kronecker delta defined by

tS i . =
]
{O,1, if
if
i =I' j ,
i = j.

From t his const ru ct ion, if v = vi e i is a vector in V , then by t aking the inner

product with e i we have

Hence, the i-t h component of v relative to the basis {e l ' . . . , en } is it s inn er

pr oduct with the vector e': Therefore , this set of vectors [e'} associate d with
the basis {e.] can be regarded as linear fun ctions that map a vector to its
components. 1
We can easily show that this new set of vectors is a linearly ind ep endent
set. Indeed , if for any linear combinat ion aje j = 0, t hen it follows t hat
( aje
j)
. e , = a jtS ]i = a i =°for all i . Fur t hermore, it also sp ans the sp ace V ,
for if u = u' e, is a vect or in V , t hen for any vector v = vi ei , from (A .2) and
vj = ej . v ,
j j)
u .v = gijU iV = (gijU ie . v,
wh ich implies that u can be expressed as u = u.e' with

Therefor e we have proved t hat t his set of vectors {e '} is also a basis of V.
Definition. Let (3 = {ed and (3* = {e i } be two bases of V relat ed by the
property
ei . e j = s:i
U i:
Th ey are said to be a pair of dual bases for V , or (3* is th e du al basis of (3.
The dual bases are uniquely det ermined from each other. For this reason,
we have used the sam e notation for their element s except the different level
of indices to distinguish them . Clearly, if u is a vector in V , then we ca n
express u in t erms of component s in two different ways relative to the du al
bases,

where we have also employed different levels of component indices in order

to b e consiste nt with our summat ion convent ion , whi ch sums over repeated
1 In gene ral, the space of all linear fun ctions on V is called the du al sp ace of V a nd
is denoted by V* . In this note, for simplicity, we shall not dis tinguish vectors in
V' a nd V wh en the sp ace V is equippe d with a n inner product.
 A.I Linear Algebra 237

indic es in different levels. We call

ui the i-th contravariant component of u,
Uj th e j -t h covari ant component of u .

From the definition, it follows that

(A.3)

and they are related by

Ui - g ij uJ·'
-

where we have denoted

The two operations

g ij : uj I-t U i ,

enable us to lower and raise the component ind ex. On e can also show that

Therefore, lowering or raising the ind ex for du al bases can be made in the
same manner. It is easy to verify that
ij si
9 gj k = uk ·

A basis {e i} is called an orthogonal basis if all the elements of the basis

are mutually orthogonal, i.e.,

e, . e j = 0 if i =I- j .

If, in addit ion, leil = 1, for all i , it is called an orthonormal basis. Although
in genera l, we carefully do our bookkeepin g of super- and subindices, this
becomes unn ecessary if fJ = {e i} is an ort honormal basis . Since then g i j = Oi j ,
and

That is, t he basis fJ is identical to its du al basis fJ*. Hence, we do not have
to distinguish contravariant and covariant components . In this case, we can
write all th e indi ces at the sa me level. for exa mple,

v = Vi ei.
Of cour se, according t o our summation convention, we st ill sum over the
repeat ed indices (now in the same level) in this sit uat ion.
238 A. Elem entar y Ten sor An alysis

Exercise A.L1 Let (3' = {e- = (1,0) ,e2 = (2, I)) be a basis of JR. 2,
and v = (1, -1) be a vector in JR.2.
1) F ind the dual basis {e 1 ,e 2 } of (3' .
2) Determine t he matrix representations [gi j] and [gi j] relative to (3' .
3) Determine the cont ravariant and covariant compo nent s of v relative
to the bases and make a graphic representation of the resu lt s.

A.1.3 Tensor Product

The notion of matrix is related t o linear fun ctions on vector spaces. Let U
and V be two vector sp aces with inn er product. A fun ction T : U 4 V , is
called a linear transformation from U to V , if for any u , v E U and a E JR.,

T(u + av) = T (u ) + aT(v) .

Notation. £( U, V) = {T : U 4 V I T is linear} .
If T and S are two linear tran sformations in £(U, V) , we can define
t he addit ion T + S and the scalar multiplication o/I' , as t ransformat ions in
£( U, V) , in t he following manner , for all v E U ,

(T + S)(v) = T(v) + S(v) ,

(aT)(v) = a T(v) .

W ith t hese operations the set £( U, V) becomes a vector space.

Definition. For any vectors v E V and u E U, the tensor product of v and
u , denoted by v ® u , is defined as a linear transformation from U to V such
that
(v ® u)(w) = (u · w)v , (A A )
for any w E U.
The te nsor product of two vectors is a linear t ransformation. We ca ll
such a linear transformation a simple tensor. Of course, not every linear
tran sformation ca n b e obtain ed as a tensor produ ct of two vectors . However ,
we ca n show that , ind eed , it can always be expressed as a linear combination
of simple t ensors .
Proposition. Let {e.}, i = 1" " , n and {d a }, a = 1, " " m be bas es of V
and U , respectively. Th en th e set {e, ® d a } , i = 1, . . . , n , a = 1, . . . , m, forms
a basis of £( U, V) .
Proof Let {e i } be the dual basis of {ed and {d a } the du al basis of {d a } .
If a i aei ® d., = 0, then
 A.I Linear Algebra 239

which impli es that a if3 = 0, since {ed is a basis. Therefore, {e , (51 d a} is a

linearly independ ent set. Moreover , for any T E £(U, V) , let

Then, for any v E V and any u E U ,

a)
v · T(u) = Vi ei . T(uad
a
= ViUa ei . T (d ) = T i aViUa .
On the other hand,

v · (e, (51 d a)(u) = vj e j . (e, (51 d a)(uf3df3)

= vjuf3(e j . ed(d a . d
(3
) = ViUa'

Therefore, we have

for any v and any u , which leads to

That is, {e, (51 d a } spans the space £( U, V) . D

We may call £(U, V) the tensor product space of V and U and denote it
by V (51 U . Obviously, from t his result , we have

dim V (51 U = (dim V)(dim U ).

The basis {e , (51 d e,} is called a product basis of V (51 U. Similarl y, the sets
{e , (51 d a} , {e' (51 d a}, and {e ' (51 d a} are also product bases of V (51 U .

Notation. V (51 V = £ (V) = £(V, V) .

We shall call linear t ransformat ions in £(V) t he second-order tensors.

Let {e i} and {e j} be du al bases of V , a second-order te nsor T then has
different component forms relativ e to the different product bases.

T = T i jei (51 ej = T ijei (51 e j

-- T i j e i 'U
r»;
ej -- T ij e i 'U
r»;
ej ,

where the various components are given by

i
T ij = ei . Te j , T j = e i . Tej ,
j (A.5)
T i j = e, . Tej , Tij = ei' Te .
240 A . Elementary Tensor Analysis

These components are called the associated components of the second-order

tensor T. In classical tensor analysis, they are also called
T ij contravariant tensor of order 2,
T ij covariant tensor of order 2,
i j
T j' t; mixed tensor of order 2.

They are related by

T ij = gkjT ik = g ik T kj , etc., (A .6)

with the operations of raising or lowering the indices discussed in the previous
section.
The matrices [Tij], [T ij], [T/]' [Tij] are called the matrix representations
of T relative to the corresponding product bases. Note that the first index
refers to the row and the second index refers to the column of the matrix. It
is important to distinguish the level as well as the position order of the com-
ponent indices. In general, Tij =I- T j i, therefore it may cause some confusion
to write Tj with i and j at the same position one on top of the other. The
relation (A.6) can be written in terms of matrix multiplication, in which the
column of the first matrix is summed against the row of the second matrix,

Note that if S, T E L(V) , then the composition SoT, defined as So T(v) =

S(T(v)) for all v E V, is also in L(V) . The composition SoT will be more
conveniently denoted by ST. In terms of components and matrix operation,
we have

Example A.I.! The identity transformation, 1 v = v for any v in V ,

has the components,

1 = <5 i jei 0 ej = <5/ei 0ej i j

=gijei 0ej =gije 0e , (A .7)

since from (A.5) , we have

1ij = e i . 1 ej = ei . ej = Usi j'

l ij = ei . 1 ej = e, . ej = gij '
Therefore, the Kronecker deltas are the mixed components of the iden-
tity tensor, while gij and gij are just its covariant and contravariant
components.
Note that the identity transformation is denoted by 1 (italic "one") .
That is, 1 E L(V) , while the number, 1 E JR , is a scalar quantity. 0
 A.l Lin ear Algeb ra 241

Example A.1.2 For v = vi ei and u = uiei in V, their t ensor product

has the component form:

V Q9u= viujei Qgej'

Let v = (VI , V2 ) and u = (UI' U2) be two vectors in JR 2, then , relativ e t o

the standard basis of JR 2 , the matrix of v Q9 u is given by

This product is sometimes referr ed to as the dyadi c product of vectors v

and u. 0

In general , the te nsor products V Q9U and U Q9V belong to two different spaces,
namely V Q9 U and U Q9 V , resp ectively. Even in the case V = U , by definition,
v Q9 U and U Q9 v are different, Le., the tensor product is not symmetric.
Definition. For A E V Q9 U, th e transpose of A , deno ted by A T , is defined
as a tensor in U Q9 V su ch that

v · Au =u · A Tv , (A.8)
for any v E V and any u E U .

Example A .1.3 For simple t ensors, it follows that

(v Q9uf = u Q9v,

because for any WI , W2 E V , we have

WI ' (v Q9 uf W2 = W2 ' (v Q9 U)WI

= (W2 . v)(u · wd = WI' (u Q9 V)W2'
o

Example A.1.4 We have

A(u Q9v) = AU Q9v, (U Q9v)A =u Q9ATv .

Indeed , for any vector w E V , we obtain

A(u Q9 v)w = Au(v · w) = (Au Q9 v)w ,

and

(U Q9 v)Aw = u(v · Aw) = u(AT tz - w) = (u Q9 A Tv)w .

o
242 A. Elementary Tensor Analysis

If A is a second-order tensor in £(V) , then the components of the transpose

AT satisfy the following relations:

(AT)i j = Aji , (AT)ij = A j i,

(A .9)
( AT )i J = Ai
J' (A T)/ = Aji'

We see from these relations that for contravariant or covariant tensors the
matrix of AT is simply the transpose of the matrix of A. However, from the
second group of the relations in (A.9) for mix ed tensors, this is not valid in
general, since the matrix transpose of [Ai j ], by changing rows and columns,
is [Aji], instead of [A j 'l-
A tensor S E £(V) is called symmetric if ST = S, and is called skew-
symmetric if ST = <S , In other words , S is symmetric if v · Su = u· Sv and
S is skew-symmetric if v . Su = -u · Sv , for all u , v E V .

Notation. Sym(V) = {S E £(V) I ST = S} and

Skw(V) = {S E £(V) 1ST = -S} .

Not e that both Sym(V) and Skw(V) are subspaces of £(V) . If S E

Sym(V) , then its components satisfy

s ij = s», Sij = Sji ,

s-, = S/ = gjkgirnskm'

In terms of matrix representation we have

Note that although S is symmetric, the matrix [Sij] is not symmetric, III
general ,

A second-order tensor can also be regarded as a bilin ear function in the

following manner : For any A E £(V), define the function on V x V, also
denoted by A ,
A(u,v) = u · Av,
for any vectors u and v in V . In particular, for simple tensors, we have

(u ® v)(u' ,v') = u' . (u ® v)v' = (u· u')(v· v') .

By employing the notion of multilinear functions , we can generalize tensor

products to higher orders. For example, we can define a tensor product of
three vectors u, v , and w as a trilinear function on V by

(u ® v ® w)(u' , v' ,w') = (u · u')(v · v')(w · w')

 A.I Linear Algebra 243

for any vectors u ' , Vi and ui' in V . One ca n show as before, that if {e i} is a
basis of V , then {ei ® ej ® e k} is a product basis for t he space of all t rilin ear
fun cti ons on V. We shall denote t his space as V ® V ® V and call it the space
of third-ord er tensors. If S is a third-order t ensor, then

There ar e several different component forms relative t o the different product

bases. In a similar manner, the te nsor product of high er orders can be defined .
We writ e k ti m es
k ~
® V = V®· · ·® V
k
for t ensors of order k . Clearly, dim ® V = (dim V)k.

Exercise A.1.2 Let (3' = {e l = (1,O), e2 = (2, I)) be a basis of JR2 ,

and T E £(JR 2 ) be defined by

(A .IO)

1) Show t hat T is a symmet ric transformation.

2) Det ermine t he matrices of the associated components of T relative
to (3/:

Note t hat t he last two matrices are not symmetric.

A.1.4 Transformation Rules for Components

The components of a te nsor relat ive to a basis are un iquel y determined and
their valu es dep end on the basis. Therefore, if we make a change of basis, they
must change accordingly. In this sect ion, we shall establish t he transfo rmation
rul es for components of te nsors under a change of basis.
Con sider a change of basis from (3 = {e.} to anot her basis [3 = {ei}
given by
(A .ll)
We call Mkj t he tran sformation ma tri x for the change of basis from (3 to [3.
By the use of (A.3), we have

from which we can also obt ain t he relation between t he dual bases (3* = {e i }
and [3* = { ei } ,
(A .12)
244 A. Elementary Tensor Analysis

The above two transformation relations (A.H) and (A.12) can be schemat i-
cally represented by
M -
(3 ----+ (3 ,

(3* ~ /3* .
In ot her words, if M changes a basis (3 to another basis /3, t heir corresponding
du al bases (3* and /3* ar e changed in the opposite dir ection through M T .
The components of a vector transform in a simil ar manner. Indeed , let
v be a vector in V , and

= vJ e J = vJ iY
One ca n easily verify that the transformation rul es for the components are
V- k = M kj V j, (A .13)
which look exact ly like those for the change of basis (A.H) and (A.12) . In
matrix notations, we have

or schemat ically

. M T .
[v"] +--- [v"].
That is, the covariant component s tran sform in the sam e dir ecti on as the
change of basis by M , while the cont ravariant components transform in the
opposite dir ect ion by M T . This is why such component s are called covariant
and contravariant in classical t ensor analysis, in which tensors are defined
through their t ran sformation properties.
For a second-order t ensor A in £(V) ,
A = A ij e i I8i e j = Aij ei I8i ej
= A/ e i I8i e j = A/ ei I8i e j '

We have the following tran sformation rul es,

Aij = A mn Mi m Mj n,
. -1 (A.14)
AJ
l
= A m n M,":
tn
M j ,
- 1 .
where the matrix [Mi j] is the invers e matrix of [M/]. In matrix not ations,
the transformation rul es can be writ t en as
[Aij ] = [Mi m] [AmnJ [MjnlT ,
[A/] = [Mi '"l [Am"l [Mn jJ-1 .
 A.I Linear Algeb ra 245

Tr ansformation rul es for ot her components and for t ensors of high er orders
are similar . The general rul e can easily be obt ained by composing t he transfor-
mation rul es for covariant and cont ravariant components , as shown in (A.I 3)
or (A.I4) .

Exercise A.1.3 Let (3 = {e l = (I ,0) ,e2 = (0, I)) and 13 = {el

(l ,0) ,e2 = (2, I)) be two bases of JR2. Det ermine the tran sformation
matrix of the cha nge of basis from (3 to 13 and also the tran sformation
matrix from 13* to (3*. Let T E .c(R2 ) be defined in (A.lO) . Determine
t he various components of T relative to the two different bases and verify
t he transformation rul es.

Exercise A.1.4 For any two bases (3 = {ed and 13 = {ed of V , there
exists a linear transformation A E .c(V) such that ek = Aek' Show that
the tran sformation matrix M for t he change of basis from (3 to 13 is given
j
by M k = e j . Aek , that is, [Mk jj = [A\] .

A.1.5 Determinant and Trace

In matrix algebra, the definit ion of t he det erminant of a squ ar e matrix is

based on the noti on of permutation. Let (1"" , n) be an ordered set of nat-
ur al numbers . A reordering of the element s in (1" " , n) is called a permuta-
t ion . More pr ecisely, a permutation is a one-to-one mapping a : {I , ... , n} -+
{I , . . . , n} resul ting in the ordered set (a( 1), ... , a(n)) . There are exac tly n!
permutations of (1" " , n) . A permutation by exchan ging t he order of two
adjacent elements is called a t ransposition . It is known that any p ermut ation
can be obt ained by merely subsequent transposi tions, and alt hough the num-
ber of such transposi tions are not un ique for a given permutation, t he parity
of this number is. Hence, a permutation is called even or odd according to
the parity of the number of tran sposi tions in ord er t o restore the permuta-
tio n back t o t he natural ord er an d one can define the sign of a permutation,
deno t ed sign a , as +1 if a is even and - 1 if a is odd.
Let [Mij] be a square matrix. The first ind ex denotes the row and the
second t he column (it does not matter whether they are superindices or
subindices). The det erminant of t he matrix can be calculate d by

det [Mij] = L (sign a) M "(l) 1 .. . M,, (n) n, (A .I5)

where the summat ion is t aken over all permutations of (1" " , n ).
On the other hand , since the matrix representation of a linear t ransfor-
mation depends on the choice of basis, t he question arises of whether it is
meaningful to define the det erminant of a linear t ransformat ion as the de-
terminant of its matrix represent ation. In the following, we shall see t hat
246 A. Elementary Tensor Analysis

the notion of the determinant of a linear transformation can be defined in a

natural way, independent of the choice of basis and we see how it is related
to its matrix representations.
Definition. Let V be a vector space of dimension n. A function w :
n
~
V X . • . x V -t JR is said to be an alternating n-linear form if it is n-linear
and for all VI , ·· ·, V n E V,
W(Va(I)' · · · ' Va(n)) = (sign a)W(VI ,· ··, V n) . (A.16)
W is called non-trivial if there exist UI, ... , Un E V , such that w( UI, . .. , un) =1=
o.
It is obvious that if W is alternating then
W(· · · ,U,· · ·,V , · · ·)=O, if u=v. (A.17)
More generally, if {VI , ... , v n } is linearly dependent then w(VI , . .. , V n ) = O.
In other words, if w(VI, . . . , V n ) =1= 0 then {VI' · ··' V n } is a linearly in-
dependent set, and since the number of vectors in this set equals dim V ,
{VI ,· . . , v n } is also a basis of V .
Theorem. (uniqueness) Let wand w' be two alternating n-linear forms and
W be non-trivial. Then there exists uniquely a A E JR, such that w' = AW,
i.e., VVI , · · ·,V n E V,

Proof Since W is non-trivial, there exists a set of vectors, say {el ' ··· ' en} ,
such that w(ej , .. . , en) =1= 0, and hence it is a basis of V. Let A be the number
defined by
A = w'(el,· ··, en) .
w(el ,· · ·, en)
Suppose that VI, ·· ·, V n E V, and
a = 1, · ·· ,n.
Then, using (A.16) and (A .17) one can easily obtain
W(VI,·· ·, v n) = aw(el , · · · , en) ,
W'(VI, ·· · , v n ) = aw'(el, · ··, en) ,
where
a = l:)sign a )V~(l) . .. v~(n).
a
Therefore, we have
W'(VI , · · ·, V n) = AW(VI, · · · , v n).
Moreover , this relation also shows that A do es not depend on the choice of
basis. D
 A.l Linea r Algebra 247

Let T E £(V) be a linear transformation on V , and w be a non-trivial

alte rn at ing n-linear form on V. Define a map Tw : V x .. . x V -+ JR by

(A .18)

Clearl y it is alternat ing and n- linear , hence by t he uniqueness theor em , there

exist s a unique A E JR, such that

Tw = AW.

We can eas ily see t hat t he scalar A so defin ed does not depend on the
choice of w . For , if w' is anot her non- trivial alternat ing n-linea r form , then
by t he un iquen ess theorem ,

w' = pw ,
Therefore, we have
T w' = A'W' = A'pW.
On the other hand , we have

Tw' (Vi , " ', v n) = W'(TV i , " " Tv n) = PW(TV i " ' " Tv n)
= pTw(Vi , " " v n ) = pAW(Vi , · ·· , v n) ,

which implies that

T w' = pAW.
Con sequ ently, A = A'. Ther efore, A is uniquely det ermined by T alone an d
we can lay down the following definition.
Definition. T E £(V) , th e det erminant of T, detT E JR , is defined by th e
following relation ,

(A .19)

for any non- trivial alterna ting n-lineet form w and for any Vi , " ' , V n E V.

The fun ct ion det : £(V) -+ JR ha s t he following properties:

1) detu 181 V = O.
2) det(a1) = an . (A .20)
3) det(ST) = (detS)(detT) .
4) det ST = det S.

The first two properties are almost trivial. Here, let us verify the pr op erty
(3) . By definition,

det(ST) w(Vi , . . . , V n)
= W(S T Vi," " S T v n ) = w(S (T v d , " " S(Tv n))
= (detS) w(Tvi ,· .. , T vn ) = (detS)(det T)w(Vi , ,, .,V n ) ,
248 A. Eleme ntary Tensor Analysis

Since it hold s for any w(V I , . .. , v n ) , relation (3) follows.

We ca n calculate the det erm inant of a linear transfo rm ation in t erm of
its compo nent matrix. Let {e;} be a basis of V , and T = T ijei (9 e j. Then
by definition ,

(det T) W ( e I, ... , e n)
= W(T e l, . . . , Ten) = w(T ii ei 1 , • • • , T i;:e iJ
= 2:)sign (1)T(7~l) .. . T(7 ~n) w( el , . . . , e n) .
(7

Hence, we obtain

det T = I:(sign (1)T(7~l) ' " T(7 ~n) ,

(7

which assures that

det T = det [T i j ],

i.e., det T is equal to the determinant of the comp onent matrix [T i j ] accord-
ing to the definition (A.I5) . Similarly, one ca n show that it is also equal to
det erminant of [T/] . Therefore, we have

det T = det[Tij] = det[ T/]

= det[gikTk j] = det[ gik T kj] .

Note t hat detT is not equal to det[Tij] nor to det[Tij], unl ess det[ gij] = 1.
Simil ar to the det erminant , anot her scalar can b e associate d with a linear
transformation . Let ! E LeV) , and w be a non-trivial alte rnat ing n-linear
form. Define a map T w : V x .. . x V --+ JR by
n
Tw(VI "" 'Vn) = I:w(vr, . .. , T Vi' ·· ·' V n ),
i= 1

On e can eas ily check that Tw is alternating and n-linear, hence by t he unique-
ness theor em , there exists a J1 E JR, such that

Tw = J1W .
Moreover , J1 does not dep end on the choice of w . Therefore, we can make the
following definition.
Definition. T E £(V) , th e trace of T , tr T E JR, is defined by th e followin g
relation n
(t r T ) W(V I , . , . , V n ) = ""'
L-J W(VI ' '" , Tv z,- .. . i v n )t »
v (A.21)
i= 1

for any non-trivial alte rna ting n-linear form w and for any VI , " . , Vn E V .
 A.I Linear Algebra 249

The function t r : £(V) --+ JR has t he following properties:

1) tr( as + T) = a tr S + tr T .
2) tr 1 = n .
3) tr(v ®u)=v ·u. (A .22)
4) trS T = trS.
5) tr(ST) = t r(T S ).

The prop erty (1) st ates th at trace is a linear function on £(V). Here, let
us prove the property (3) . Suppose that v = viei , t hen
n
tr(v ® u) w(el " ' " en) = Lw(el "' " (v ® u)ei, " " en)
i= l
n n
= L(u. ei)w(el , " ' , v , " ' , en) = L(u . ei) viw(e l " '" e n) ,
i= l i= l

which impli es that

n
tr(v ® u) = L u· (viei) = u · V .
i =l

Hence (3) is proved .

In te rms of components, let T = T ijei ® e j = Tije i ® e j , t hen

That is, t r T is equa l to t he sum of dia gonal elements of t he matrix [T i j ] or

[Tj i], but, in genera l, is not equa l to that of the matrix [Tij 1 or [T ij].

Example A.1.5 Show that det(l +u ® v) = 1+u ·v.

By definition, we have

det(1 +u ® v) w(e l, " ' , en)

= w((1 +u ®v)el , · · · ,(1 +u ®v)e n)
n
= w( el , . . . , e n) +L w( e 1, . . . , ( u ® v )ei , . . . , en) + .. .
i= l

= w(e l"' " e n) + tr(u ® v ) w(el , "', en) ,

where the dots represent terms involved with mor e than one factor of
(u ® v)ei in w. Since (u ® v)ei = (v . ei)u, which is a vector in the
direction of u for any ind ex i , t hose t erms must all equa l zero because
w is an alt ernating form . 0
250 A. Element ar y Tensor An alysis

Two non-trivial alternating n-linear forms WI and W 2 are said to be equiva-

lent if WI = ..\W2 for some ..\ > O. Clearly, this is an equivalence relation that
decomposes the set of non-trivial alte rnat ing n-linear forms into two equiva-
lent classes. Each of these classes is called an orientation of V. We call one
of them , say .d, the positive orientation. A basis {ed of V is called positively
oriented if for any W E .d ,

and A E £(V) is said to be orientation preserving if A w E .d , for any W E

.d. Here, A w is defined by (A.18) . Since A w = (detA)w , A pr eserves t he
orientation if and only if det A > O.
Let {e.} and {ed be two bases such that A (ei) = e i . If det A > 0
(or < 0) , then [e.] and {ed are said to have the sam e (or the opposite)
orientation.
Suppose that V is a three-dimensional vector space and let {iI , iz , i 3}
be a positively oriented orthonormal basis of V , then there exists a unique
e E .d , called the volume elem ent , such that

Since e E £(V x V x V , JR) , it is a t hird-order t ensor and can be repre-

sented as
e = Cij k i , 181 i j 181 i k ,

where Ci j k = e ( ii, i j, i k ) are th e components of e relative to th e basis {id .

Obviously we have

if (i , j ,k) is an even permutation of (1, 2, 3) ,

if (i, j ,k) is an odd permutation of (1,2,3),
otherwise,
and we call it the permutation symbol. One can easily check t he following
identities:
Ci j kC imn = 8j m8k n - 8j n8k m ,
Ci j kCi j n = 2 8k n , (A .23)
Ci j kCij k = 6,
where 8m n is the Kronecker delta.
Let {ek} be a basis and A E £(V) be a chang e of basis from {id to
{ed, i.e., A i k = e k . Then the covariant components of the volum e element
relative to {ed are
. . k
e= eij k e l® e J® e .
 A.l Linear Algebra 251

From (A.19 ), it follows that

and also,
9iJ = ec - ej = Ai i · Ai j = (AT A) i j ,

which yields 9 = (det A)", where 9 = det[gij] . Therefore, we have

(A.24)

if A pr eserves the ori ent ation. Similarly, the cont ravariant components of the
volume eleme nt ar e

and
C
ij k
= (y'g)-l Eij k,
where E
ij k
= Eij k ' Moreover , the identities (A.23) ca n be written as

CijkCi m n = 6j m6\ - 6j n6 km ,
Ci j k C'tJn
·· -
-
2 Uskn' (A.25)
cij kCijk = 6.
If T E £(V) and T = T ije i ® el , then (A.19) leads to the following
formul a for the determinant of T ,

(A.26)

Multiplyin g by c1m n and using the last identi ty of (A.25) , we obtain another
formula for the det erminant ,

i Tj T k
det T -- (3l ICm n C'J. .k T I m n'

Exercise A.1.5 Consider the tensor defined by (A.lO) in the pr evious

exercise. Calculate det T and tr T by means of definition and also by the
use of component matrices relative to (3' .

A .1.6 Exterior Product and Vector Product

The usual vector product on a t hree-dimensional vector space can not be

generalized dir ectly to vector spaces in general. However, it ca n be associat ed
with the skew-symmet ric t ensor product in a trivial manner.
252 A. Elementary Tensor Analysis

Definition. For any, v , U E V , the exterior product of v and u , denot ed

v 1\ u , is defined by
v 1\ u =v ® u - u ® v.
It is obvious that the op eration 1\ : V x V ---+ V ® V is bilinear and
skew-symmet ric, i.e.,
v 1\ u = -u 1\ v .
The exte rior product of two vectors v 1\ u is a skew-sy mmet ric t ensor.
Suppose that {e i ® e j} , i , j = 1, .. . ,n is a product basis of V ® V , then
it is easy to verify that {e, 1\ e j} , 1 :S i < j :S n is a basis for Skw(V) .
Therefore, we have the following proposition .
Proposition. If dim V = n, then dim Skw(V) = n (n - 1)/ 2. In particular,
if n = 3, then dim Skw(V) = 3.
Now, suppose t hat V is an ori ented Euclidean three-dimensional vector
space. Sin ce t he space of skew-symmetric t enso rs is also three-dimension al ,
we can define a linear map

r : Skw(V) ---+ V

by the condition: for all u , v , wE V ,

r(u 1\ v) . w = e(u, v , w) . (A .27)

Here, e is t he volume element of V . This linear map, called the duality map,
is one-to-one and onto and hen ce establishes a one-to-one corresponde nce
between a skew-symmetric t ensor and a vector. It is easy to ver ify that

(A .28)

For an orthonormal basis {i k } the duality map r est abli sh es the following
correspondence,
i 1 l\i 2 t------+ i 3 ,
i 2 1\ i 3 t------+ i I ,
i 3 1\i 1 t------+ i 2.
For a skew-symmetric tensor W , let w = r(W) be the associated vector,
which shall be denoted more convenientl y by

w = (W). (A .29)
 A.I Linea r Algebra 253

In component form, if

or
1 ..
W = 2 W 'Je i 1\ e j '

Then , it follows from (A.28) that

If t he basis is orthonormal , it becomes

(A .30)

or , in matrix form ,

Remark. It is worthwhile to point out that the vector associate d with

a skew-symmet ric t ensor behaves differently from usu al vect ors under
linear tran sformations. To see this, let u , v E V , then for any w E V
and Q E 'c(V) , it follows from the definition that

(Qu Qv ) . Qw = e(Qu, Qv , Qw)

1\
= (detQ) e(u ,v,w) = (detQ) (u l\v ) · w,
which implies that

(Qu 1\ Qv ) = (det Q)Q (u 1\ v ).

In other word s, as the vect ors u , v ar e transformed into Qu , Qv , resp ec-

t ively, the vector (u 1\ v ) is tran sformed into Q (u 1\ v ) only to within a
sca lar constant, or, into a vecto r that may point in one or t he opposite
sense of t he sa me axial direction of (Qu 1\ Qv ). For this reason , a vect or
asso ciated with a skew-symmetric t ensor is usually called an axial vector.
o

The usu al vector product , in the t hree-dimensional vector space, can now be
defined from the exte rior product in a similar manner .
254 A. Element ary Tensor Analysis

Definition. For any u , v E V , the vector product of u and v , denoted u x v ,

is defined by
u x v = (u /\ v ). (A .31)
Clearly the operation x : V x V -----+ V is bilin ear and skew-symmetric.
In components (A.31) gives

If the basis is orthonormal, say {id, t hen it becomes

which is the usu al definition of the vector product.

The relat ions (A.27) and (A.31) imply that

e(u, v, w ) = (u x v) · w.
This is usu ally called the triple product of u, v , and w. For convenience, we
shall also use the not ation,

[u,v,w] = (u x v) · w.
With this notation, we can rewrite t he definitions (A.19) and (A.21) of the
det erminant and the trace in the following form

[Ael ' Ae2' Ae3]

d et A - -'---=---------::--=-
- [el , e2, e3] ,
(A .32)
[Ael ,e2,e3] + [el ,Ae2 ,e3] + [el ,e2, Ae 3]
tr A = -'------=..:..--=-:.-...::-'---------'-;-..::...:.....----':..:-.7'-----'.---=...:.----=."'----~
[el,e2 ,e3]

One may use the du ality map to identify a skew-symmetric t ensor with
an axial vector, as well as the exterior product with the vector product. In
other words, one may int erpret the du ality in eit her way, as far as the context
requires.

Exercise A.lo6 Verify the following relations , using ind ex notations:

1) Wv = -w x v .
2) (u xv) xw=(u ·w)v-(v ·w)u.
3) lu X vl 2 = lul 21vl 2 -Iu , v1 2 .
4) lu x vi = lullvl sinO(u ,v).

A.lo7 Second-Order Tensors

We shall review some of the imp ortant prop erties of linear transformations,
i.e., the second-order t ensors, mostly without pr oofs in this section. The
proofs can be found in most standard books in linear algebra.
 A.I Linea r Algeb ra 255

First, let us introduce an inn er product of two second-order t ensors . Let

A , B E £(V) , we can define the inner product of A and B by

which is obviously a bilin ear , symmetric and positive-definite operation. We

have
1 . A = tr A ,
where 1 is the identity t ensor , and for any A , B , C E £(V) ,

The norm of a tensor A E £(V) , can then be defined as

IAI = Jj[:""A = Jt r AAT .

Not e t hat if A i j are the components of A relati ve to an orthonormal basis ,

then the norm of A is simply

Now, suppose that A E £(V) is one-to-one (there fore, onto) , then there
is a unique A-I E £(V) , called the inverse of A , such that

If A - I exist s, A is said to be inverti ble or non- singular, otherwis e, it is said

to be singular. It can be proved that A is invertible if and only if det A -=I 0,
and for any non-singular A and B,
(AB) - l= B- 1A- 1,
(A-1)T = (AT) -l = A - T .

Notation. In v(V) = {F E £(V) I F is invertible}.

Recall t hat a set G is called a group if it has the following prop erties:
1) If A , BEG then AB E G.
2) If A , B, C E G then A(BC) = (AB)C .
3) There exist s an ident ity element 1 E G such that 1 A = Al = A , for any
AEG.
4) For any A E G , t here exists A -I E G , such that AA- 1 = A- 1A = 1.
It is easy to verify that Inv(V) forms a group under t he op er ation of com-
position. It is usu ally known as the general linear group of V , denoted by
GL(V) .
256 A. Elementary Tensor An alysis

Definition. Q E £(V) is called an orthogonal transformation if it preserves

th e inn er product of V . i.e., for all u, v E V,

Qu · Qv = u · v .

Notation. O(V) = {Q E £(V) I Q is orthogonal} .

The set O(V) form s a group and is called t he orthogonal group of V.
Orthogonal t ransformat ions have the following properti es:
1) QT = Q -l .
2) Idet QI = l.
3) IQvl = Ivl·
4) B(Qv, Qu) = B(v, u) .
The last two relations asse rt that ort hogonal transformations also preserve
norms and angles. An orthogon al transformation Q is said to be proper if
det Q = 1, and improper if det Q = - l.
Notation. O+(V) = {Q E O(V) I detQ = I} .
The set 0 + (V) also form s a group, called the proper orthogonal group of
V . It is also called the rotat ion group since its element s are rotations. Note
that the subset of O(V) with determinant equal t o -1 does not form a gro up
since it do es not have an identity eleme nt.

Notation. U(V) = {T E £(V) I [det TI = I} and

SL(V) = {T E £(V) I detT = I}.

E lements ofU(V) ar e ca lled unimodular transformations andU(V) form s

a group, called the unimodular group of V . SL(V) also forms a group , ca lled
the special lin ear group of V . Clearly, we have the following relations:

o+ (V) c SL(V)
O(V) c U(V) c GL(V) .

A.lo8 Some Theorems of Linear Algebra

We shall mention some important theor ems of linear algebra relevant to the
study of mechanics. They ar e all related to the concept of eigenva lues and
eigenvect ors.
Definition. Let A E £(V) . A scalar A E JR is called an eigenvalue of A, if
th ere exists a non-zero vector v E V , such that

Av = Av . (A.33)

v is called th e eigenvector of A associa ted with th e eigenvalue A.

 A.I Line ar Algeb ra 257

It follows from the definition that>. is an eigenv alue if and only if

det(A - >'1) = O. (A.34)

The left-hand sid e of (A.34) is a pol ynomial of degr ee n in >. , where n is the
dimension of V . We may write it in the form

It is called the characteristic equation of A . Its real roots are the eigenvalues
of A . The coefficients II , . .. , In are scalar fun ctions of A and are called the
prin cipal invariants of A .
It can be shown that the charact er istic equat ion is also satisfied by the
t ensor A it self. We have the following
Cayley-Hamilton Theorem. A second-order tensor A E .c(V) satisfies
its own characteristic equation,

Example A.1.6 For dim V = 3 and A E V , we have

(A.35)

The three princip al invariant s of A , more sp ecifically denoted by I A, II A,

and III A can be obtain ed from the followin g relations:

I A = tr A , II A = tr A-I det A, IlIA = detA. (A .36)

Of course, the second relation is valid only when A is non- singular.

Proo f From (A .32) we can write

det(A - >.1)[eI , e2 , e3] = [(A - >'1)el , (A - >'1 )e2, (A - >'1 )e3]

- >.3 [el , e2, e3]
+ >.2([Ael ,e 2,e 3] + [el ,Ae2,e3] + [el ,e 2, Ae3])
- >. ([el ,Ae2,Ae3] + [Ael ,e2,Ae3] + [Ael , Ae2, e3])
+ [Ael , Ae2, Ae3] '
Comparing t his with t he right-hand side of (A.35), we obtain (A.36h ,3
by the use of (A.32), as well as t he followin g relat ion for the second
invariant II A,

II A -
[el , Ae2, Ae3] + [Ael , e2, Ae3] + [Ael , Ae2, e3]
-'--"-'----'----'----'-'---'--;---'----7"'-'--0----"-'---"-'--=
- [el,e2,e 3] .

If A E Inv(V) , then it implies t he second relation of (A.36). In particular,

if detA = 1, we have II A = I A- l. 0
258 A. Eleme ntary Tensor Analysis

In general, the characteristic equat ion may not have real roots. However , it
is known that if A is symmetric all the roots are real and there exist s a basis
of V consist ing ent irely of eigenvectors.

Spectral Theorem. Let 5 E 5ym(V) , then there exists an orthonormal

basis {e .] of V , such that 5 can be written in the form
n
5 = L Sie i 0 e.. (A.37)
i= l

Such a basis is called a prin cipal basis for 5. Relative to this basis , the
component matrix of 5 is a diagonal matrix and the diagonal element s Si are
the eigenvalues of 5 associat ed with the eigenvect ors e. , resp ectively. The
eigenvalues s. , i = 1," ' , n mayor may not be distinct.
Definition. Let A be an eigenvalue of 5 E £(V) . We call V.\ = {v E V I 5v =
AV} th e characteristic space of 5 associated with A.
If 5 is a symmetric t ensor and suppose that v E V.\ , U E Vi" where A and
J.l are two distinct eigenvalues of 5 , then one can easily show that v . U = 0,
i.e., they are mutually orthogonal. Moreover , by the spectral theorem any
vector v can be writ t en in the form

v=Lv.\ , (A .38)
x
where the summat ion is ext ended over all charact erist ic spaces of 5 .
Commutation Theorem. Let T E £(V) and 5 E 5ym(V) . Th en
5T=T5
if and only if T preserves all characteristic spaces of 5, i.e., T maps each
characteristi c space of 5 into its elf.
Proof Suppose that 5 and T commute, and S» = Av . Then
5(Tv) = T (5 v ) = A(Tv) ,
so that both v and Tv belong to the characterist ic sp ace V.\ .
To prove the converse, since 5 is symmet ric, for any v E V, let v =
L.\ v.\ be the decomposition relative to the characte rist ic spaces of 5 as given
in (A.38) . If T leaves each characte rist ic space V.\ invari an t , then Tv.\ E V.\
and
5(Tv.\) = A(Tv.\) = T(AV.\) = T(5v.\) .
Therefore, from (A.38), we have

5Tv = L5Tv.\ = LT5v.\ = T5v,

.\ .\

which shows that 5T = T5 . 0

 A .I Lin ear Algebra 259

There is only one subspace of V that is preserved by any rotation , namely

V itself. Therefore, we have the following
Corollary. A symmetric S E £(V) com m utes with every orthogonal trans-
form ation if and only if S = ,\1 , for some ,\ E JR .

Definition. S E £(V) is said to be positive definite (positive semi-definite)

if for any v E V and v :f. 0, v . Sv > 0 ( ~ 0) . Similarly, S is said to be
negative definit e (negat ive semi-definite ) ifv · Sv < 0 (::; 0).
On e ca n easily see that if S is symmetric, then it is positive definite if
and only if all of its eigenvalues ar e pos itive. Consequ ently, for any symmet-
ric positive definit e transfo rmation S , there is a unique symmetric positive
definit e tran sformation T such that T 2 = S and the eigenvalues of T are the
positive squ ar e roots of those of S asso ciat ed with the sam e eigenvect ors.
We denote T = VB and call T the square root of S . In other words , if Sis
expressed by (A.37) in t erms of the principal basis, t hen
n
T = yIS = L VSie i Q9 e. .
i= l

Example A.lo7 Let S E £(JR 2 ) be given by S( x , y) = (3x+J2y, J2x+

2y) . Relative t o the st andard basis of JR 2 , t he matrix of S is

which has the eigenvalues S l = 4 and S2 = 1 and the corr esponding pr in-
cipal basis e1 = (fi!3, Ji73)and e 2 = (-Ji73,
fi!3) . Therefore, we
have
T = yIS = 2e 1 Q9 e1 + e 2 Q9 e2,
whose matrix, relative to the standard basis, becomes

On e can easily verify that [Tij j2 = [Sij] .

Example A.loB Let S be a positive definite symmet ric tensor in a

two-dimensional space, t hen
1
yIS = b(S + a1) ,

where a = Jdet Sand b = J2a + tr S .

260 A. Elementary Tensor Analysis

Proof Let A = VB. By the Cayley-Hamilton theorem in the two-

dimensional space, we have the identity
A2 - (tr A)A + (detA)l = 0.
Since A 2 = 5 , if we let the eigenvalues of A be al and a2, then det 5 =
aia~ and tr 5 = ai + a~. Therefore

a = Jaia~ = ala2 = detA,

which, together with the above identity, prove the result . 0

Polar Decomposition Theorem. For any F E Inv(V), th ere exist sym-

metric positive definite transformations V and U and an orthogonal trans-
formation R such that
F=RU= VR .
Moveover , the transformations U , V and R are uniquely determined in the
above decompositions.
Proof We can easily verify that F F T and FTF are symmetric positive defi-
nite. Indeed, for any v =1= 0, we have

(v · FTFv) = (Fv · Fv) > 0,

since F is non-singular.
To prove the theorem, let us define

U = VFTF, R = FU- 1, V = RURT . (A .39)

By definition, U is symmetric positive definite and R is orthogonal since

RRT = FU-1(FU -1)T = FU -1U -TF T

= FU - 2FT = F(FTF)-lF T = 1 .

Moreover, from the definition (A.39) we also have

Therefore, V is the square root of F F T and hence is itself a symmetric

positive definite transformation. Furthermore, the un iqueness follows from
the definition of a square root. 0
The polar decomposition theorem, which decomposes a non-singular
transformation into a rotation and a positive definite tensor, is crucial in
the development of continuum mechanics. The following decomposition of
 A.l Linear Algeb ra 261

a tensor into its symmet ric and skew-symmet ric parts is also import ant in
mechanics.
For any T E £(V) , let

then
T=A+B , A E Sym(V) , B E Skw(V).
This is somet imes called the Cartesian decomposition of a te nsor. Such a
decomposition is also unique.

Exercise A.l.7 Let A E £(V) be such that (1 + A) is non- singular .

Verify that

Exercise A .l.8 Let u , v E V . Show that if 1 + u · vi=- ° then

u 0v
(1 + U 0 V )- 1 = 1- .
1 +u ·v

Exercise A.l.9 For dim V = 3, let A E £(V) and B = 1 + A . Show

that
I B=3+I A,
II B = 3 + 21A + II A,
III B = 1 + I A + II A + III A,
and if a = det B i=- 0, verify that

Exercise A .l.IO Prove the Cay ley- Ha milt on theorem for the special
case that A E £(V) is symmet ric, by employing the spec t ral t heorem .

Exercise A.l.ll Let {3 = {(1 ,0,0) ,(0,1 ,0) ,(0,0,1)} be the st andard
basis of IRa and the matrix represent ation of F E £(IR 3 ) relativ e t o {3
be given by

F= [ ° 10]
J3
2 ° .
° ° 1
Suppose that F = R U = VR is the polar decomposition of F. Find t he
matrix representation of U, V, and R relative t o t he standard basis {3.
262 A. Elem ent ary Tensor An aly sis

A.2 Tensor Calculus

In the second part of this appendix, we shall discuss some basic notions of
calculus on Euclidean spaces: gradi ents and other differential op erators of
t ensor fun ctions.

A.2.1 Euclidean Point Space

Let E be a set of points and V be a Euclidean vector space of dimensi on n .

Definition. E is called a Euclidean point space of dimension n , and V is
called the translation space of E if, for any pair of points x , y E E, there is a
vector v E V, called th e difference vector of x and y , written as

v = y - x, (A.40)

with the following prop erties:

1) \f x E E, x - x = 0 E V .
2) \f x E E, \f v E V, th ere exists a uniqu e point y E E, such that (A.40) is
satisfied. We write y = x + v .
3) \fx , y,z E E, (x-y)+(y - z)=(x- z) .

Obviously, with (A.40) we can define the distance between x and y in E,

denoted d(x , y) , by
d(x , y) = lvi ,
or equivalent ly
d(x ,y) = J( x - y) . (x - y) ,
where the dot denotes the inner product on V .
Notation. Ex = {v x = (x ,v) I v =y- x, \f y E E} .

Ex denotes the set of all difference vectors at x . It can be mad e into

a Euclidean vector space in an obvious way, with the addit ion and sca lar
multiplication defined as

v x+ux=(v+u) x ,
av x = (av) x.

We call Ex the tangent space of E at x.

 A.2 Tensor Calculus 263

Clearly Ex is a copy of V, i.e., it is isomorphic to V . In other word s, for

any x E E, the map i x : V -+ Ex, called the Eucl idean parallelism , taking
v to V x trivially est ablishes a one-to-one correspondence b etween Ex and V .
The composit e map
. ·- 1 c c
Tx y = Zy 0 Zx : Gx ---+ G y

taking
Vx = (x,v) f-----+ V y = (y,v)
defines the parallel transl ation of vect ors at x to vectors at y (Fi g. A.I) .
T herefore, although Ex and Ey for x =1= y , are two different t angent spaces,
they ca n be ident ified throu gh V in an obvious manner ,

v x,y E E.
In other words , V x = (x,v) E Ex and u y = (y,u) E Ey ar e regarded as the
same vector if and only if v = u. In this manner , vectors at different tangent
spac es can be added or subtract ed as if they were in t he sa me vector space.

Fig. A .I. P ar allel translation

A.2.2 Differentiation

Before we define the derivative of tensor functions on Euclidean space in

general , let us recall the definition of derivative of a real-valu ed function of
a real vari abl e. Let f : (a, b) -+ JR be a fun ction on the interval (a, b) c JR.
The derivative of f at t E (a, b) is defined as

d~~t) = l~ ~ (J(t + h) - f(t)) ,

if the limit exists.

264 A. Elementary Tensor An alysis

This definition can easily be exte nded to t ensor-valu ed functions of a

real vari abl e. Let W be a space equipped with a norm or a dist anc e function.
As examples, we have

JR d(x ,y) = Ix - YI ,
E d(x , y) = J( x - y) . (x - y) ,
(A .41)
V lui = ..;u:u,
.c(V) , Sym(V) , Skw(V) : IAI = v'tr AAT .
It makes sense to talk about limit and convergence in the space W when a
norm or a distan ce function is defined .
Let f : (a, b) ---+ W be a function defined on an int erval (a, b) c JR. The
derivative of f at t E (a, b) is defined as

d~;t) = E~~ ~ (f(t + h) - f(t)) . (A.42)

The derivative of f at t will also be denoted by i(t) . Obviously for any

t E (a, b) we have j(t) E W .
Not e that if f is defined on a mor e general space, the expression on the
right-hand side of t he definition (A.42) may not make sense at all . However,
we can rewrite the relat ion (A.42) in a different form.
For fixed t , let D f(t) : JR ---+ W be the linear transformation defined by

Df(t)[hl = j(t) h.

Then (A.42) is equivalent t o

. 1
l~ ThT If(t + h) - f(t) - Df(t)[hll = O.

In this form , the definition of the derivative can easily be generalized to other
functions.

Tensor Fields
Now we shall consider functions on a Euclidean point space E. Let D be an
open set in E, and f be a t ensor-valued function, f : D ---+ W . Such functions
are usually called t ensor fields, mor e sp ecifically,
1) W = JR, f is called a scalar field on D ,

f : x E D t------+ f( x) E JR .

2) W = V, f is called a vector field on D ,

f : x E D t------+ f( x) E Ex ~ V.
 A.2 Tensor Calculus 265

3) W = L(V) , f is call ed a second-order tensor field on V ,

f :x E V f----+ f( x) E Ex ® Ex ~ L(V).

4) W = E, f is call ed a point field on V or a deformation of V ,

f :x E V f----+ f( x) E E.

Definition. A function f : V ~ W is said to be differentiable at x EVe E,

if there exists a linear transformation Df(x) E L(V, W) at x, such tha t for
any v E V ,
. 1
lim
111 1-+0
- IIlf( x
v
+ v) - f( x) - Df(x)[v]1 = o. (A.43)

The linear transformation D f (x) is uniquely determined by the above

relation, and it is called the gradient (or derivative) of f at x, denoted by
grad I , or \1 »L, or sim ply \1 f . By definition, \1 f( x) is a t ensor in W ® V,
or is a vector in V if W = JR .
The condition (A .43) is equivalent to

f( x + v) - f( x) = \1 f(x)[v] + o(v) ,
where o(v) is a qu antity containing t erms such t hat

lim o(v) = O.
1111-+0 Ivl
Mor eover , if we substitute tv for v for some fixed v in V , (A.4 3) is also
equivalent t o
\1f(x)[v] = lim
t -+O
~t (f( x + tv) - f( x))
(A.44)
= dd f (x + tv) I .
t t=o
The right-hand side of the above relation is usually known as the directional
derivative of f relative to the vector v . Note that for fixed x and v , f( x +tv)
is a ten sor-valued function of a real variable and its derivative can easily b e
det ermined from (A.42) .

Functions on Tensor Spaces

Let WI and W 2 be two spaces on which a norm or a distance fun ction is
defined , such as the sp aces mentioned in (A.41) , and let V C WI be an op en
subset . The grad ient of te nsor functions on V can be defined in a similar
manner .
266 A. Elem enta ry Tensor Analysis

Definition. A function F : V --+ W 2 is said to be differentiable at X EVe

WI, if there exists a linear transformation DF(X) E £(WI , W2) at X , such
tha t 'tj Y E V ,

lim ly111F(X + Y) - F(X) - DF(X)[Y]I = O.

IYI---+O

The linear transformation DF(X) is uniquely det ermined by the above

relation, and it is called the gradient of F with resp ect to X , denoted by Ox F .
We have Ox F E W2 0 WI ' The definition is equivalent to t he condit ion: for
any Y E V , we have

F(X + Y) - F(X) = OxF(X)[Y] + o(Y ), (A.45)

or
oxF(X)[Y] = ~F(X +ty)lt=o ' (A.46)
For ¢ E W20 WI , and Y E WI , the notat ion ¢ [Y ] used in the above relations
is self-evident : for ¢ = K 0 X ,

(K 0 X)[Y] = (X· Y)K ,

and for all v , u E V and A , S E £(V) , we have

v[u] =v ·u,
A[u] = Au,
A[S] = A .S = tr AS T ,
(v 0u)[Sj = v · Su .

Gradients can easily be compute d dir ectly from the definition (A.45) or
(A.46) . We demonstrate this procedure with some examples.

Example A.2.1 Let ¢ ; £(V) x V --+ JR be defined by

¢(A, v ) = v · A v .

Then , from (A.45) ,

¢(A, v + u) = (v + u) . A(v + u)
= v . Av + v . Au + u . Av + u . Au
= ¢ (A , v) + ov¢ [u j + o(u) ,

so that
Ov¢[uj = v . Au + u . Av
= ATv · U + Av· u
= (A T + A)v[uj .
 A.2 Tensor Calculus 267

Therefor e, we obtain

Mor eover , we have

4>(A + S, v) = v · (A + S)v = v · Av + v· Sv ,
which implies
OA 4>[Sj = v · Sv = (v Q9 v)[S],
so that

Example A.2.2 Let u , v E V be constant vectors, and let 4> : .c(V) -+

JR be defined by
4>(A ) = U . Av .
From (A.46) we have

OA 4>[S] = :t (u . (A + tS)v) It=o = u · Sv = (u Q9 v)[S],

for all S E .c(V) , and we obtain

OA 4> =U Q9V.

Now, suppose that A is a symmetric t ensor, hen ce the fun ction 4> is
defined on the subspace Sym(V) only,

4> : Sym(V) -+ JR,

and by definition aA 4> E Sym(V) also . In this case , we have the same
relation ,
OA 4>[S] = (u Q9 v)[S],
but it holds on ly for all S E Sym(V) . Therefore, we conclude that

after symmetrization .
Sim ilarly, if A is a skew-symme t ric t ensor, then OA4> E Skw(V) and
the result mu st be skew-symmetrized ,

o
268 A. El em entar y Tensor An alysis

Example A.2.3 We consider trace and det erminant functi ons. Since

tr(A + S) = tr A + tr S = tr A + 1 . S,
so that , trivially, the gradient of the trace is the identity tran sformation,

(A.47)

For the gradie nt of the det erminant, we have

(oAdet A)[S] = det(A + S) - det(A) + o(S) .

Let w be a non -trivial alte rnat ing n-linear form , then

W(VI, " " v n)(oAdetA)[S]

= w((A + S)vI, "" (A+ S)v n) -W(AVI , ' " , A v n ) +o(S) .

By the linearity of w, after removing all the high er-order t erms into o(S) ,
the right-hand side becomes
n
= LW(Av I " " ,SVi , · · · ,Avn)+o(S)
i= 1
n
= LW(AvI , . . . , A A- I SVi ,' " , Av n ) + o(S)
i= 1
n
= (det A) LW(VI ,·· · , A- I SVi ,"' ,Vn) + o(S)
i=1

from (A.21) . Therefore, we have

(oAdetA)[S] = (detA)(trSA- I) = (detA)A- T[S],

which impli es the following formul a ,

(A.48)

In differential calculus, we frequently differentiate a composit e fun ction by

the chain rul e. This rul e can be st ated for composite te nsor fun ct ions in
gener al. Let WI , W 2 , W 3 be normed spaces of the type (A.41) and VI C WI ,
V 2 C W 2 b e op en subs et s, and let

with ¢ (V d C V 2 . Then we have the following

 A.2 Tensor Calculus 269

Chain Rule. Let ¢ be differentiable a t X E 1)1 , and 'lj; be differentiable

a t Y = ¢ (X ) E 1)2 . Th en th e com position f = 'lj; 0 ¢ is differentiable a t X
and
Df(X)[Z] = D 'lj;(¢(X)) [D¢(X)[Z]] , (A .49)
for any Z E W I or simply
D f(X) = D 'lj;(Y) 0 D ¢(X) .

Example A .2.4 If ¢ is a scalar-valu ed function of a vector vari abl e,

g( x ) is a vector field on E; and h(v) is a vector-valu ed fun ction of a
vector variabl e, then

'Yh(g(x)) = ovhl ('Yg( x)) ,

v=g(x)

'Y¢(g(x)) = ('Yg( x)) Tov¢! .

v=g(x)
Let us verify the last one in the above formul ae . For any u E V , from
(A.49) ,

'Y¢(g(x))[u] = Ov¢lv =g(x)['Yg( x)[u]] = Ov¢ lv=g(x) . ('Yg( x ))u

= ('Yg( x)) Tov¢1 , u = ('Yg (x ))Tov¢ 1 [u],

v=g(x) v=g(x)
where in t he third step we have used the definition of t rans pose (A.8) .
Note that 'Yh , 'Yg, and ovh are all second- order t ensors, while ov¢ is a
vect or quan ti ty. 0

Another important result in differentiation is the product rul e. For tensor

functions, in general, t here are many different products available, for exa mple,
the product of a scalar and a vect or , t he inner pro du ct , the tensor product ,
t he action of a t ensor on a vector. These product s have one property in
common, nam ely, bilinear ity. Therefore, in order t o establish a product rul e
valid for all cases of interest , we consider the bilinear op eration
1r : W I x W 2 ~ W3 ,
which assigns t o each ¢ E W I , 'lj; E W 2 , t he product 1r(¢ ,'lj;) E W 3 . If ¢ , 'lj;
are two funct ions ,

where 1) is an open subset of some normed space W , then the product f =

1r( ¢, 'lj; ) is t he function defined by
f : 1) ~ W3
f(X) = 1r(¢(X) ,'lj; (X )), "IX E 1).

We t hen have the following

270 A. Elem en tary Tensor Analysis

Product Rule. Suppose that ep and 'ljJ are differentiable at XED c W,

th en th eir product f = n( ep, 'ljJ) is differentiable at X and
D f(X)[V] = n(Dep(X) [V], 'ljJ (X )) + n( ep(X) , D 'ljJ(X) [V]) , (A.50)
for all V E W .
In other words, the derivative of the product n( ep, 'ljJ) is the derivative of
tt holding 'ljJ fixed plus the derivative of n holding ep fixed .

Example A.2 .5 Let f be a scalar-valued , and h , q be vector-valued

fun ctions on DeW. For W = JR , we have

(fh)" = jh + fh,
(A .51)
(q . h)' = q . h + q . h.

For W = [ , we have
\1(fh) = h ® \1f + f\1h ,
(A .52)
\1(q . h) = (\1q)Th + (\1h) Tq.
For W = V , we have
ov(fh) = h ® ovf + f ovh ,
(A.53)
ov(q · h) = (ovq)Th + (ovh)T q.
Unlike the simple formulae in (A .51) , the relations in (A.52) and (A.53)
do not look like the familiar product rul es, becau se they have to be
consistent with our notation conventions.
Let us demonstrate the first relation of (A.52) . By the product rul e
(A .50) , for any w E V , we have
\1(fh)[w] = (\1f[w])h+f(\1h[w]) = (\1f · w )h + f (\1h )w
= (h ® \1f)w + f(\1h)w = (h ® \1f + f(\1h))[w],
wher e, in the third st ep , we have used the definit ion (A.4) . 0

If f : D c U ---7 W is differentiable and its derivative D f is cont inuous in D ,

we say that f is of class C 1 . The derivative is again a function , D f : D ---7
W ® U , for which we can talk about the differentiabili ty and cont inuity. We
say that f is of class C 2 , if D f is of class C 1 , and so forth. Frequ ently, we
say a fun ction is sm ooth to mean that it is of class c- for some k 2': 1. We
ment ion t he following

Inverse Function Theorem. Let DeW be an open set and f : D ---7

W be a one-to-on e function of class C k(k 2': 1). Assume th at th e linear
transformation Df(X) : W ---7 W is invertible at each XED , then r'
exists and is of class c».
 A.2 Tensor Calculus 271

Example A.2.6 Let VeE and ¢> : V -+ JR be of class 0 2 . Then the

second gradient of ¢> is a symmetric t ensor, that is, \l(\l¢» E Sym(V) .
Indeed , from the definition , we have

\l¢>(x + u) - \l¢>(x) = \l(\l¢>)[u] + o(u) .

Taking the inn er product with v , we obtain

\l¢>(x + u)[v]- \l¢>(x) [v] = v · \l(\l¢»u + o(u) ,

which impli es that

v· \l(\l¢»u = ( ¢>(x + u + v) - ¢>(x + u))

- ( ¢>( x + v) - ¢>(x )) + o(u) + o(v ).
Since the right -hand sid e of the last relation is symmetric in u and v , it
follows that
v · \l(\l¢»u = u · \l(\l¢»v ,
which proves that the second gradient of ¢> is symme tric . D

Exercise A .2.1 Show that if Q : JR -+ O(V) is differentiable, then

(jQ T is skew-symmetric.

Exercise A .2.2 Let h(v ,A) = (v · Av)A 2v be a vector function of a

vector v and a second-order t ensor A. Compute ovh and (oAh)[S] for
any S E £(V) .

Exercise A .2.3 If A E L(V) is invertible, show that

1) (oAA -1)[S] = -A- 1SA- 1, for any S E LeV),
2) OA t r (A- 1 ) = _(A - 2)T.

Exercise A.2 .4 Let A be a second-order tensor. Show that

1) For any posit ive int eger k ,

2) For principal invari ants I A,II A,IIIA,

oAIA=l,
oAII A = (fA 1 - A)T , (A .54)
oAlIl A = (fIAl - I AA +A 2)
T.

Hint : Calculate OA det(A + >"1) = OA(>..3 + I A>.. 2 + II A>" + III A)'

272 A. Elementary Tensor Analysis

A.2 .3 Coordinate System

Tensor fun ct ions can be expressed in terms of component s relative to smoot h

fields of bases in the Euclidean point space t: associated with a coordina t e
system .
Definition. Let D c t: be an open set. A coordinate syst em on D is a smooth
one-ta-one mapping
'ljJ : D --+ U,
where U is an open set in IRn , such that 'ljJ - l is also smooth.
Let xE D ,
'ljJ : x t----+ (xl , . .. , Xn ) = 'ljJ (X).
( Xl , .. . , x n ) is called the (curvilinear) coordinate of x , and the functions

Xi : D --+ IR
(A .55)
Xi(x)= Xi , i=l , · ··,n,
are called the i-t h coordinate junction of 'ljJ . For convenience, we call (xi) a
coordinate system on D .

ei( X)

X~
_......... ......... t

...J--------.- Xi

Fig. A.2 . Coordinate cur ve

Let X = 'ljJ- l , then

x = x (x l,· ··,x n ) . (A .56)
For X l, . . . , x n fixed , the mapping (Fi g. A.2)

Ai : IR --+ D
(A .57)
Ai(t) = X(x l, "" Xi + t , ·· · , Xn ) ,
 A.2 Tensor Calculus 273

is a cur ve in V passing through x at t = 0, called the i-t h coordinate curve

at x. We denote the tangent of this cur ve at x by ei(x) .

e i(x) = Ai(t)
. 1 = f)x . 1
-f) . (A .58)
t =o x' (xl , ··,x n )

Proposition. Th e set [e, (x) , i = 1, "' , n} forms a basis for th e tan gent
space Ex.
Proof For any vector v E Ex, we can define a cur ve through x by

A(t) = x + tv .
Let
A(t) = X(A1(t),· · · , An(t)) ,
where Ai(t) are t he coordinates of A(t) given by

(A .59)

Then , from (A.58) , t he tangent vector becomes

V
. ox a»
= A(t)1 t =O = -f)x '.[x -I
dt t=O
= -I
dAi

dt t=O
ei(x) ,

In other words , {e, (x)} spans the space Ex. 0

The set [e, (x)} is a basis of Ex for each x . This field of bas es is called
the natural basis of the coordinat e system (Xi) for V , the tran slation sp ace of
E. The corresponding dual basis of this natural basis is denoted by [e' (x)} .
Combining (A.55) and (A.56), we have

which impli es

f)x i . . f)X .
-f).
xJ
= oJ' = (V'X') . -f)J.
x
= (V'X') . ej (x) ,

by the use of (A.58) . Therefore, the two natural bases of the coordina te
syste m (Xi) are given by the following relations:

ei(x) = -ff)x
) '1 , (A .60)
x' x
The inner products,

are called the metric tensors of the coordinate system .

274 A. Elem entar y Tens or An alysis

Now let us consider a change of coordinate syst ems. Let (xi) and (xi)
be two coordinate systems on V , and {e , (x)} , {ei (x)} be the correspondi ng
natural bas es. Suppose that the coordinate tran sformations are given by
Xi = X i (x 1 , . .. , xn ) ,
xk = xk (x 1 , .. . , z" ).
Then , by t ak ing the gradi ents, one immediate ly obtains the change of the
corresponding natural bases given by
. ox i ox k
el( x) = ox k ek(x) , ei(x) = ox i ek(x) . (A.61)

Comparing t he change of bases considered in Sect . A.1.4, [ox i / ox k] plays t he

role of the t ransformation matrix [Mk i ] in (A.12), an d hence, the t ransfor-
mation rul es (A.14) for the component s of an arbitrary t ensor in t he change
of coordinate syst em becom es
-i k ox i ox l
A ) = A 1ox k ox) . (A .62)
For other components of tensors, in general, the transformation rul es are
similar .

Example A.2.7 Let us consider a deform ation r: : V ---+ E,

/'i,(x) = ii,
Let ( X i) be a coordinate system on V , a nd (i be a coordinate system
Q
)

on /'i, (V ),
x- = x- (-X 1,· ··,x-n ).
The deformation /'i, is usu ally expressed explicit ly in t he form ,
a = l ,· · · ,n. (A .63)
Using the chain rul e, we ob t ain, with Xi = Xi( x) ,

V/'i,(x) = u'ox
l-
I~O/'i,QI Vx i (x) ,
x X ux x
Q l

which, from (A.60) becomes

O/'i,QI .
V /'i,(x) =~ eQ(/'i,(x)) 0 e'{z ).
ux l x
This is the component form of t he deformation gradient V /'i,(x) in t erms
of two different coordinate syst ems (xi) and (i With resp ect to these
Q
) .

two nat ural bases at two different points, nam ely, x and /'i,(x ), t he com-
ponents of t he deformation gradi ent ar e just the par ti al derivatives of the
deformation fun ction (A.63) , which can most easily be calculated. Other
comp onent forms of V /'i, can be obtained through the met ric t enso rs and
by the change of bases relative to the coordinate syst ems . 0
 A.2 Tensor Calculus 275

A .2.4 Covariant Derivatives

We shall now consider the component form of the gradient of a t ensor field ,
in general , relative to the natural basis of a coordinate system . Let (x i ) be a
coordina te system on V eE , and {e , (x)} , {e ' (x)} be its natural bas es.
To begin with, let us consider a scalar field , f : V -+ JR, t he gradient of
f is then a vector field . From (A.44) , (A.57) , and (A.58) we have

(V'f( x)) . ei(x) = lim

t ---+ O
~t (J( x + te i) - f( x))

= l~ ~ (J(x( x l, . . . , Xi + t , ' " , x n ) ) - f(x( x l, · ·· ,xn )))

= o(~;?) l(x1,... n)' . ,X

which are the covariant components of V'f .

Usually, we sha ll write f(x( x l, . . . ,xn ) ) as f( x l, . . . ,xn ) for simplicity.
Therefore, the component form of the gradient of f( x) becomes

V'f( x) of I e i (x) .
=~ (A .54)
ux' x

In other words, for the gradient of a scalar field f , its covariant component
relative to the natural basis, (V'f ) i , is just the partial derivative relat ive to
t he coordina te xi .
Now, let us consider the gradient s of natural bases themselves. For each
i fixed , {e .] and {e '} can be regarded as vector fields on V ,

Let us deno t e the grad ients of natural bases by

r i(x) = V'ei(X) E Ex ® Ex,

(A .55)
r i(x) = V'ei(x) E Ex ® Ex.

We write
(A .55)
The component s ri jk and r i j k are called the Christoffel symbols. Not e that
ri jk and r; are not the associated components of a third-order t ensor.
By taking t he gradient of (ei(x) . ej( x)) , one can obtain the relation,

(A.57)
276 A. Elementary Tensor An alysis

r
Moreover , since i = V(VXi( x)) by (A.60h and the second gradient is a
symmetric t ensor, we have the following symmetry conditions,

(A.68)

Since both Christoffel symbols are related in such a sim ple manner , usually
r
only one is in use, namely, j ' ; and it is called the Christoffel symbol of the
second kind in classical t ensor analysis.
Now let us calcula te the gradi ent of a vector field in t erms of the coor-
dinat e system . Suppose that v(x) is a vector field and

v(x) = v i(x )ei (x ) = vi (x )ei (x ).

Then from (A .52h , (A.64), (A.65), and (A.66) , we have

Vv = V(vied
= ei0 Vv i + vi V ei

Hence, the gradient of v(x) has the component form ,

where
. &v j . .
J -
v ,k - &x k
+ V <r.i :k ' (A .69)

Similarly, we also have

Vv = v·J , k ej 0 e
k
,

where
Vj ,k =
& Vj
&x k - Vi rj i k : (A .70)

Here, the relation (A.67) has been used .

Vj, k and Vj, k are the mix ed and the covariant components of Vv . The
comma st ands for t he op eration called the covariant derivative, since it in-
crease s t he order of the covariant component s by one.
More generally, suppose that A is a second-order te nsor field , t hen V A
is a third-order t ensor field that has the following component form ,

VA = A i j, k e i 0 e j 0 e k ,
 A.2 Tensor Calculus 277

where
i OAij I i i I
A j,k= ox k +Ajrlk-Alrjk' (A.71)
Covari ant derivatives of other components ca n easily be written using the
same recipes for covari ant and cont ravariant component s, resp ectively.
We have seen in (A.7) that the components of the metric te nsor, gij (x)
and gij (x) , are also t he component s of t he identity t ensor, therefore their
covari ant derivativ es must vanish,
gij ,k = 0, gij ,k = O. (A .72)
Cons equ ently, from (A.24), the covariant derivatives of t he volume t ensor
also vani sh ,
. ·k
eijk ,1 = 0, e"J ,I = O.
In ot her words , the components of the metric t ensor and the volume t ensor
behave like constant t ensors in covariant derivation, although they are, in
general, functions of x.
From (A.72h , we can derive a formula for the det ermination of the
Christoffel symbols in terms of the metric tenso r. From (A.71) we have
Ogij T. I T. I
ox k = glj i k + gil j k :

Rotating the indices (i ,j, k) of this relation, then adding two of the three
resulting equat ions and subtracting the remaining one, we get

2 g0.r.'
"k
= (09jk
g .
+ Ogij
g kg
_ 09ik)
· .
ux" ox ux J
Hence, we have the following formul a:
1 jl (09li oglk -09ik)
r i jk -_ -g
2
-
ox k+ox- i o-
xl' (A.73)
The Christoffel symbols are not component s of a third-order tensor. For two
coordinate syst ems (x i) and (Xi) , t hey have the following t ransformat ion
rul es:

A.2 .5 Other Differential Operators

Divergence and curl of a vector field can be defined in the usu al way and
their definitions can be adopt ed also for tensor fields .
Definition. The divergence of a vector field u is a scalar field defined by
divu = tr(V'u) . (A .74)
In component form ,
diIVU=U i .i :
278 A. Elementary Tensor An alysis

Definition. Th e curl (or rotation) of u is a vector field defined by

cur l u = (\1u T - \1u).

In component form ,

Here, the duality map defined in (A .29) is employed and, according to

(A .30) , cur l u is the axi al vector of the skew-symmet ric par t of t he gradi ent
of (- 2u) . On e can easily verify the following conditi on:

v · curl u = div(u x v) ,

for any constant vector field v . This condit ion can be used as the definition
for t he curl op erator. In a similar manner , we can define the divergence of a
second-ord er t ensor in t erms of the divergence of a vector .
Definition. Th e divergence of a second-order tensor field S is a vector field
defined by th e condition : for any constant vector field v ,

v · div S = div(ST v) . (A .75)

In component form , we have

div S = S ij,jei'

Definition. Th e Laplacian of a scalar (or vector) field cP, denoted by \12cP ,

is a scalar (or vector) field defined by

In component form , if cP is a sca lar field ,

If cP =h is a vect or field,
\12h = gj khi,jke i.
In the above expressions , the comma denotes the covariant derivative.

Example A.2.8 Let I, S, and u , v be scalar, tensor , an d vector fields,

resp ectively. T hen we can show the following relations:

div(fu) = u · \11 + 1 divu,

div (S u ) = u· div ST + tr(S\1u) ,
(A .76)
div(u x v) = v · cur l u - u · curl v ,
\12(U ' v) = \12U ' V + 2\1u · \1v + u · \1 2v .
 A.2 Tensor Calculus 279

Let us verify the first relation.

div(ju) = tr(V(ju))

=tr( u 0 V f + f (V u ))
= tr(u 0 V f) + f tr(Vu) ,
which gives (A .76h. In this calculation, we have used the definition
(A.74), the relation (A .52h, and the linearity of the trace operator.
Verification of the other relations in (A.76) may not be so straight-
forward in direct notation. And more annoyingly, these relations, as well
as the relations (A.52) and (A.53), are not easy to memorize. Neverthe-
less, if we express all of these relations in index notation, they all become
trivially simple. Indeed, (A.76) may be written out directly as:

(f u ) , 2. = f , 2·u + fui "p.

i i

( S i j u J·) ,2. = Sij , 2·u J· + sv«, .) , 1,1

(gileljkujvk) ,i = gileljkuj,ivk + gileljkujvk,i '

g j k (u i v •·) ,j·k = gjkui ,jk. v •· + 2g j kui ,kv·',j· + gjkui V',jk,
· .

which are merely the usual product rules of differentiating scalar func-
tions and the symmetry of second gradient. The only difference here is
that the comma denotes the covariant derivative instead of the usual
partial derivative. D

Remark. From the observation made in the above example, the use of in-
dex notation is often encouraged, especially when complicated calculations
are involved . In arbitrary curvilinear coordinate systems, contravariant and
covariant indices must be carefully distinguished and the pair of repeated
indices, for which the summation convention is applied, must always appear
in different levels. An index can be raised or lowered to its proper level with
the metric tensor gij or gi j. Moreover, since the gradients of the metric tensor
and the volume tensor vanish, in covariant differentiation, the metric tensor
gij as well as the components of the volume element eijk can be treated as
constants. Furthermore, if the Cartesian coordinate system is used, there is
no difference between contravariant and covariant components and hence all
the indices can be written at the same level, and more conveniently, the co-
variant derivative becomes the partial derivative, and gij = oij , eijk = Cijk
are constants.
It is important to note that given an expression in index notation, one
can always turn it into an expression in direct notation or vice versa. There-
fore, in handling calculations, the choice of using direct notation or index
notation, or even using Cartesian index notation is totally up to one's taste
and convenience.
280 A. Elem entar y Tensor Analysis

We sh all also mention some important t heo rems of integr al calculus oft en
used in mechanics.

Divergence Theorem. Let n

be a bounded regular region2 in E, and let
¢ : n -+ JR, h : n -+ V , S : n -+ .c(V) be smooth fields. Th en

r
Jan
dm da = r\l¢dv ,
i;

r
Jan
v .nda=
i.
r divvdv , (A.77)

r Snda i;r divSdv ,

Jan
=

where n is th e outward unit normal field on an.

Proof The relations (A .77h ,2 are well-known classi cal resul ts. To show
(A .77h , let v b e an arbit rary const ant vector. Then

v· r
Jan
Sn da = r
Jan
v · Sn da = r
Jan
S Tv . n da

= L div(STv)dv= L v ·divSdv

= ·in
v div S dv ,

wh ere we have used (A .77h and the definition (A.75) . 0

Proposition. Let ¢ : D -+ W be a continuous function on an open set D
in E. If
L ¢dV=O ,
for any NeD, then ¢ is identi cally zero in D, i.e.,
¢ (x ) = 0, \/x E D .

Proof Suppose that ¢ (x o ) i- 0 for some X o ED, then since ¢ is cont inuous,
there exist s a small neighborhood N eD containing xc, such that ¢ (x ) i- 0,
\/x E N. Therefore, by the mean value t heorem of int egral calculus,

L ¢ dv = K ¢( x) i- 0,

2 A reg ula r region, rou ghl y sp eaking, is a closed region with a piecewise smoot h
boundary.
 A.2 Tensor Calculus 281

for some x E .N, where K denotes t he volume of .N. This cont radicts t he
hypothesis. 0
T his pro position and the divergence t heorem enable us to deduce local
field equat ions from the integral balan ce laws.

Exercise A.2 .5 Let f , u , v , and S be smoot h scalar , vector, and second-

order t ensor fields. Verify the following identi ti es:
1) div(Su) = u · div ST + tr(S\7u) ,
2) divUS) = S \7 f + f div S ,
3) div (u 0 v) = (\7u)v + u divv ,
4) div (\7u )T = \7(divu) .

Exercise A.2.6 Let f and v be smooth scalar and vector fields, re-
spectively. Show that
1) cur l \7 f = 0,

°
2) div curl v = 0,
3) If div v = and curl v = 0, then \72v = 0.
Exercise A.2.7 Let v and S be smoot h vector and te nsor field , on a
bound regular region R, resp ectively. Show that
1) r v 0 n da = i.r \l v dv ,
Jan
2) r v 0 Snda= r (v 0 div S+(\7v)ST)dv.
Jan i.

A.2.6 Physical Components

Let (xi) be a coordinate system on E and {e i (x)} and {e ' (x)} be its na tural
bases. The syst em (xi) is called an orthogon al coordin ate system if t he metric
tenso r
gij(X) = 0, for i i- i , "Ix E E:
For an orthogon al coordinate system, we can define a field of orthonormal
basis, denoted by {e (i)(x)} , by normalizing the natural basis,

(no sum).

In this expression, t he summation not ation is not invoked as ind icated ex-
plicitly. Since
(no sum),
282 A. Elementary Tensor Analysis

therefore,
e, ei
e (i ) = ..;g;i = #
= # e, = ..;g;i e i (no sum) .
Here, we have noted that normalization of the two dual natural bases of an
orthogonal coordinate system gives rise to the same orthonormal basis.
The components of a tensor field relative to the orthonormal basis
are called the physical components in the coordinate system (Xi).
{e (i )(x)}
For a vector field v ,
i i
V = V e; = Vie = V (i )e (i ) '

The physical components V(i ) are given by

i Vi
V (i ) = ..;g;i V = rx:: (no sum) (A.78)
V9ii

For a second-order tensor field T,

T = T ijei ® ej = Tije i ® e
j
= Tij e i ® e j

= T (i j )e (i ) ® e (j).
The physical components T (ij ) are given by

(no sum) . (A .79)

In particular, we have 9 (ij ) = lii j .

The advantage of using physical components is obvious in practical ap-
plications. Since the norms of the basis vectors of the natural basis, in general,
vary from point to point in E, hence it is inconvenient for the measurement
of physical quantities relative to this bas is.

A.2.7 Orthogonal Coordinate Systems

We now consider three orthogonal coordinate systems most commonly used :

the Cartesian, the cylindrical, and the spherical coordinate systems and de-
rive their basic charact eristics.

a) Cartesian Coordinate System

Fix a point 0 in E. Let {iI, i 2, i 3} be an orthonormal basis of V . For any
x E E, then x - 0 E V . We write

x- 0 = Xi i. ,
 A.2 Tensor Calculus 283

Clearl y, this defines a coordinate system

with {iI , iz , i 3} as its natural basi s, which is, of cour se, ind ep endent of x E E.
We call such a system a Cartesian coordinate system.
For a Cartesian coordinat e syste m, we have

\/X E E,

and hence, from (A.73)

r /k( X) = O.
It is also a custom to write the basis {i l ,i z ,i 3} as {e x,ey ,e z } and the
coordinat e (Xl , Xz, X3 ) as (X, y , z) for a Cartesian coordinate syst em .

b) Cylindrical Coordinate System

The cylindrical coordina t e system (r, 0, z ) is defined as

X = x( r , 0, z),
by the following coordinate tran sformation (see Fig. A.3 (a)) ,

Xl = r cosO, r >O
Xz = r sin O, 0 < 0 < 27f (A.80)
X3 = z,

where X = (XI, XZ , X3) is the Cartesian coordinate syst em .

The natural bases ar e denoted by {e r,eo ,e z } and {e r,eO,e Z } . From
(A.80) and (A.60 )z, we can determine the basis in te rms of the Car t esian
component s.

eo = ox
00 = .
- r sin O~l
' +r cos O~z
' ,

ox .
ez = oz = ~3 ·

Therefore, we obtain the mat rix represent ations of the metric t ensor in t he
cylindr ical coordinate syst em ,

[gij] =
[1 1] 'r
Z
284 A. Elem entar y Tensor An a lysis

and t he Chr istoffel sym bols from (A.73),

() _ () _ 1
T; () - r () r -- ,
r

ot hers = O.
Moreover , we have

and
e (r ) = cos 0 i l + sin 0 i z ,
e «() = - sin Oi l + cosOi 2 ,
e (z ) = i 3 ·

X3 X3

(r ,B, z) (r,8 , ¢)
r
z
B
r

X2 X2

Xl Xl

(a) Cy lind rical (b) Sp her ical

Fig. A .3 . Coordinat e system s

c) Spherical Coordinate System

T he spher ical coordinate system (r, 0, ¢) is defined as
x = x(r,O, 4» ,
by the following coordinate t ransformat ion (see Fig. A.3 (b)) ,
Xl = r sin 0 cos ¢ , r >0
X2 = r sin 0 sin 4>, 0 < 0 < 1r
X3 = r cos O, 0 < 4> < 21r
where x = (x I , X 2 , X3 ) is the Cartesian coordinate system .
 A.2 Tensor Calculus 285

The natural bases are denot ed by {e r, eo, e1>} and {e r ,eO,e1>} . We have
e; = sin O cos¢i 1 + sin O sin¢ i 2 + cos Oi 3 ,
eo = r cos O cos¢ i 1 + r cos O sin¢ i 2 - r sin Oi 3,
e 1> = - r sin O sin¢ i 1 + r sin O cos¢ i 2 ,

and
e; = e" , eo = r 2 eO, e1> = r 2 sin 2 0 e1> .
The matrix representations of t he metric tensor have the forms

[gij] = [1 r2
r 2 sirr' 0
] , [gij] = [1 r- 2
(r sin 0)- 2
] ,

and the Christ offel symbols are

rr
0_
°- r o 0 _
r -
1> _
r r 1> - r 1>
1> _ 1
r - - ,
r

r1> 1> = -r sm 20,

r .

r / 1> = r / o = cot 0,

r/1> = - sin 0 cos 0,

others = O.
Moreover , t he orthonormal bas is for t he physical components are
e (r ) = sin O cos¢ i 1 + sinO sin ¢i 2 + cos Oi 3 ,
e (O ) = cos O cos¢ i 1 + cos O sin¢ i 2 - sin 8 i 3,
e (1)) = - sin ¢ i 1 + cos ¢ i 2 .

Remark. More frequ ently, we would like to express qu an tities in these co-
ordinate syste ms in terms of t heir phy sical components. A simple way t o do
this is to derive the expressions first in te rms of contravariant or covariant
components and t hen convert them into physical component s using relations
like (A.78) and (A.79) .

Example A.2.9 Let us calculate the Lapl acian of a sca lar field I[> in
the spherical coordina t e syste m. We have
& 1[>
I[> Jo = ~ ,
, ux J
& 21[> &1[> i
I[> j k =
, &&
xJ X k - ~rJo
ux ' k»
286 A. Elementary Tensor Analysis

from which we obtain the following covariant components:

{P p
P,rr = 01'2 '
a 2p aP r a 2p aP
P ,oo = a()2 - or To 0 = a()2 + r or '
a 2p {)P r {)P 0 a 2p . 2 aP . aP
P ,</> </> = a1J2 - or T</> </> - a() T</> </> = a1J2 + r sm () or + sin ()cos () a() .
We have P ,rr = p ,(rr ), P ,90 = r 2p ,(88)' P ,</> </> = 1'2 sin 2 () p ,(</> </» in t erms
of physical component s. That is,
a2 p
p ,(rr ) = 01'2 '
1 a 2p 1 aP
P ,(09 ) = 1'2 a()2 + ~ or '
2p
P = 1 a + ~ aP + cot () ap .
, (</></» 1'2 sin 2 () a1J2 r or 1'2 a()

Therefore, the Laplacian V 2P , which is the sum p ,(rr ) + p ,(OO ) + p ,(</> </»
in physical components, becomes

V 2P = a
2p
+ ~ a P + ~ a 2p + 1 a
2p
+ cot () ap .
01'2 r or 1'2 a()2 1'2 sin 2 () a1J2 1'2 a()
o

Example A .2.10 We give the physical components of the divergence

of a symmetric tensor field T in the following coordinat e syste ms:
a) Cartesian coordi nate system (x , y , z):
8T(xx) + aT(xy) + 8T(xz)
(divT) (x) = ax ay az '
aT(xy) 8T(yy) aT(yz)
(d ivT) (y) = ---a;;- + EiY + ~' (A.81)

(di T) = aT(xz) + aT(yz) + 8T(zz)

IV (z) ax ay oz '

b) Cylindrical coordinate system (1',0, z ):

aT(rr ) 1 aT(r9) aT(rz) T (rr ) - T (oo )
(divT) (r ) = ----a;:- + ~----ae + [i;- + l'

aT(rO) 1 aT(88 ) aT(Oz) 2

(divT) (9) = ----a;:- + ~----ae + ~ + ~T( ro) , (A.82)
aT(r z) 1 aT(Oz ) aT(zz ) 1
(d ivT) (z) = -a:;:- + ~----ae + ~ + ~T(rz) .
 A.2 Te nsor Calculus 287

c) Spherical coordinate syst em (r B, ¢ ):

(divT) r = oT(rr) + ~ o T (rO) + _1_ o T( r1»

() A, r 00 r sin 0 o¢
+ ~ (2 T(rr) - T(oo) - T(1)1>) +cotBT(ro») ,

(divT) 0 = oT(rO) + ~ 8T(oo) + _ l _ oT(01>) (A .83)

() A, r 00 r sin 0 o¢
+ ~ (3 T (r O) + cot 0 (T(M ) - T (1)1>))) '

(divT) = oT(r 1» + ~ o T(o</» + _ l _ oT(</> </»

(1)) A, r oe r sin 0 o¢
+ ~ (3T(r </» + 2 co t OT( o</» ).

Exercise A.2.8 Let u be a vector field . Show that

1) in the cylind rical coordinate syst em ,

. oU (r ) 1 oU (O) oU (z) 1
div U = a:;:- + ;:----arJ + -a;- + ;:U (r ) ;
2) in the spherical coordinate syst em ,

. o U(r ) 1 oU (O) 1 oU (</» 2 cot 0

div U = - ",,-
or
+ -r ----;:;-0
u
+ -r sin
'-0 ~ + -U (r ) + - - u (O) ·
v ,/" r

Exercise A.2.9 Let u be a vect or field a nd E = !('vu +V'u T ) . Express

E in cylindrical and spherical coordinat e systems ,
1) relative to the natural basis ,
2) in terms of physical component s.

Exercise A.2.10 Let T be a symmetric t ensor field . Compute divT,

in cylindrical and spherical coordinate syst ems ,
1) relative to t he natural basis,
2) in t erms of ph ysical components. (Verify (A.82) and (A.83)) .

Exercise A.2 .11 Let <P : JR ---+ [. be a cur ve. Suppose t hat {e, (x)} is
the natural bas is and ¢i (t ) is t he coordinate of <P(t) in the coordinate
syst em (Xi) . Show that
1) d>(t) = Jyi (t ) ei (¢(t )),
2) <P(t) = ( ¢i(t) + ¢J(t )Jyk(t )rj \ (¢(t )) ei( ¢(t)) .
References

1. Antman, S. S.: Nonlinear Problem s of Elasti city, Springer-Verlag, New York

(1995) .
2. Barbera , E .: On the principle of minimal ent ropy production for Navier-Stokes-
Four ier fluid s, Continuum Mech. Th errnodyn. 11, 327-330 (1999)
3. Batra, R. C .: A thermodyn ami c theory of rigid heat cond uct ors, Arch. Rational
Mech. A nal . 53, 359-3 65 (1974).
4. Beatty, M. F .: A class of uni versal relations for cons trained isotropic elast ic
mater ials , A cta Mech. 80, 299-312 (1989) .
5. Berberi an , S. K. : Introductio n to Hilbert Space, Oxford Un iversity Press, New
York (1961) .
6. Bh atnagar , P . L.; Gross, E . P .; Krook, M.: A model for collision processes
in gases . I. Small amplitude processes in charged and neutral on e-component
systems. Phys. R ev. 94. 511-525 (1954) .
7. Boelher, J. P.: On irreducib le representations for isotropic scalar fun ct ions ,
ZAMM 57, 323-327 (1977).
8. Capriz, G .; Podio-Guidugli , P.: Internal con straints, Appendix 3A , Rational
Th ermodynamics, 2nd edit ion, ed ited by C . True sde ll, Springer , New York-
Berlin (1984) .
9. Coleman, B. D. ; Mizel, V. J .: On the general theory of fading memory, Arch.
Rational Mech. Anal. 29, 18-31 (1968) .
10. Co lem an , B. D.; Noll , W .: Foundations of linear visco elasticity, R ev. Modem
Phy s. 33, 239-249 (1961) .
11. Coleman, B. D.; Noll, W .: The thermodyn amics of elast ic materi als with heat
cond uct ion a nd viscosit y, Arch. Rational M ech. Anal. 13, 167-178 (1963) .
12. Courant , R.; Hilb ert, D.: Method of Math ematical Physics, Vol. II. Interscience
Publish ers , New York (1962) .
13. Dafermos, C .: Hyperbolic Con servation Laws in Cont inuum Physics, Springer,
Berlin-Heidelb erg (2000).
14. de Groot, S. R. ; Ma zur , P .: Non- Equilibrium Th ermodynamics , North-Holland,
Amsterdam (1963) .
15. Dreyer , W .: Maximiza tion of the ent ropy in non- equilibrium, J. Phys. A : Math.
Gen. 20, 6505-6517 (1987) .
16. Ericksen , J . L.: Deformations possible in every compressible, isotropic, perfectly
elastic material. J. Math. Phys. 34, 126-128 (1955) .
17. Ericksen , J . L.: Inversion of a perfectly elast ic spherical shell, Z. A ngew. Math.
Mech, 35, 382-385 (1955).
18. Ericksen , J . L.: Introduction to Th ermodynamics of Solids , Chapman & Hall,
London (1991).
19. Eringen , A. C .: Continuum Physics, Vol. 1, Acad emic Press, New York (1976) .
20. Eshelby, J . D.: The elastic energy-moment um t ensor. J. Elasticity, 5, 321-335
(1975) .
290 References

21. Fisher, A. E. ; Marsden, J . E.: The Einstein evolution equation as a first order
quasilinear symmetric hyperbolic system , Comm. Math . Phys. 28, 1-38 (1972).
22. Friedrichs , K. D.; Lax , P. D.: System of conservation equations with a convex
ext ens ion , Proc. Nat. Acad. Sci. 68, 1686-1688 (1971).
23. Gibbs, J. W. : On the equilibrium of heterogeneous substances. Scientific Papers
of J. Willard Gibbs. Vol. 1, Dover , New York (1961).
24. Grad , H.: Principles of the kinetic theory of gases, Handbu ch der Physik, Vol.
XII , edited by S. Fliigge, Springer, Berlin-Heidelberg-New York (1958) .
25. Gurtin, M. E.; Martins, L. C.: Cauchy's theorem in classical physics, Arch.
Rational Mech. Anal. 60, 305-324 (1976) .
26. Gurtin , M. E. ; Mizel, V. J. ; Williams, W. 0 .: A note on Cauchy's stress theorem,
J. Math . Anal. Appl. 22, 398-401 (1968).
27. Gurtin , M. E.; Williams, W . 0 .: On the inclusion of the complete symmetry
group in the unimodular group, Arch . Rational Mech. Anal. 23, 163-172 (1966).
28. Gurtin, M. E.; Williams, W . 0 .; Suliciu, I. : On rate-type constitutive equa-
tions and the energy of viscoelastic and viscoplastic materials, Int. J. Solids
Structures 16, 607-617 (1980) .
29. Hildebrand, F. B. : Advanced Calculus for Applications, 2nd edition, Prentice-
Hall , Englewood Cliffs, New Jersey (1976) .
30. Hutter, K .: On thermodynamics and thermostatics of viscous thermoelastic
solids in electromagnetic fields, Arch. Rational Mech. Anal. 58, 339-386 (1975).
31. Jou , D.; Casas-Vazquez, J .; Lebon, G .: Ext ended Irreversible Thermodynamics ,
Springer, Berlin (1993) .
32. Kearsley, E . A.: Asymmetric stretching of a symmetrically loaded elast ic she et,
Int. J. Solids Structures, 22, 111-119 (1986) .
33. Knops, R . J. ; Wilkes, E. W.: Elastic Stability, Handbuch der Physik Vol. Vla/3,
edited by S. Fliigge, Springer, Berlin (1973).
34. Kogan, M. N. : Rarefied Gas Dynamics, Plenum Press, New York (1969)
35. Kremer, G . M.: Extended thermodynamics of ideal gases with 14 fields , Ann.
In st . Henri Poincare, 45 , 419-440 (1986)
36. Liu, I-Shih: On irrev ersible thermodynamics, Ph.D Dissertation, Department of
Mechanics, The Johns Hopkins University, Baltimore, Maryland, USA (1972) .
37. Liu, I-Shih: Method of Lagrange multipliers for exploitation of the entropy
principle, Arch. Rational Mech. Anal., 46,131-148 (1972).
38. Liu , I-Shih : On representations of anisotropic invariants, Int. J. Eng . Sci . 20,
1099-1109 (1982).
39. Liu, I-Shih: On the structure of balance equations and extended field theories
of mechanics, Il Nouvo Cimento 92B, 121-141 (1986) .
40. Liu , I-Shih: Extended thermodynamics of viscoelastic materials, Continuum
Mech. Thermodyn. 1, 143-164 (1989).
41. Liu , I-Shih: On interface equilibrium and inclusion problems, Continuum Mech.
Th ermodyn. 4,177-186 (1992).
42. Liu , I-Shih: Stability of thick spherical shells, Continuum Mech. Thermody. 7,
249-258 (1995) .
43. Liu , I-Shih: On entropy flux-heat flux relation in thermodynamics with La-
grange multipliers, Continuum Mech. Thermody. 8, 247-256 (1996) .
44. Liu , I-Shih: Constitutive equations of extended thermodynamics from a hybrid
pair of gen erator functions , Continuum Mech. Thermodyn. 13, 25-39 (2001)
45. Liu , I-Shih; Miiller, I.: On the thermodynamics and thermostatics of fluids in
elect ro magnet ic fields, Arch. Rational Mech. Anal. 46, 149-176 (1972) .
46. Liu , I-Shih; Mull er , I.: Extended thermodynamics of classical and degenerate
gases, Arch . Rational Mech. Anal. 83 , 285-332 (1983) .
 Referen ces 291

47. Merritt, D. R. ; Weinhaus, F .: T he pressure curve of a rubber balloon , Am. J.

Phys . 46, 976-977 (1978) .
48. Moon , H.; Truesdell , C.: Int erpretation of adsciti t ious inequali t ies t hroug h the
effect s pure shea r produces up on an isotropic elas tic solid , Arch. Rational Mech.
Anal. 55, 1-17 (1974) .
49. Muller , I. : Zum Parad oxen der W arrneleitungstheori e. Zeitschrift fur Phy sik,
198, 329- 344 (1967).
50. Muller , I. : On t he ent ropy inequality, Arch. Ration al Mech. Anal. 26, 118-141
(1967).
51. Mull er , I. : A t he rmody na mic theor y of mixt ures of fluids , Arch. Rational Mech.
A nal. 28, 1- 39 (1968).
52. Miiller, I.: Die Kaltefunktion eine universelle Funktion in der T hermody na mik
viskoser war meleitende r Fliissigkeit en , Arch. Rat ional Mech. Anal., 40, 1-36
(1971) .
53. Mull er , I.: The cold ness, a universal function in t her moelas t ic bodies, Arch.
Rational Mech. Anal., 41, 319-332 (1971) .
54. Muller , I.: A new approac h t o thermodyn amics of simple mixtures, Zeits chrift.
fu r Natu rforschung , 28, 1801- 1813 (1973) .
55. Muller , I. : Th ermodynamics, P itman Publishing, London (1985) .
56. Mull er , I.: Two instruct ive inst abilit ies in nonlinear elasti city: biaxially load ed
membrane a nd rubber balloon s, Meccanica, 31, 387-395 (1996).
57. Mull er , I. , Ruggeri , T. : Rational Ext ended Thermodyna mics. 2nd edit ion , Tract s
in Natural Philosophy 37, Springer , New York (1998).
58. Noll, W .: A mathematical theor y of t he mechanic al behavior of cont inuous
medi a , Arch. Rational Mech. Anal. 2, 197-226 (1958) .
59. Noll, W. : Proof of t he maxim ality of the orthogon al gr oup in the un imodular
gro up , Arch. Rational Mech. Anal. 18, 100-102 (1965) .
60. P rigogine, I. : Introduction to Th ermodynam ics of Irreversible Processes, Int er-
scie nce Publish ers , New York (1967)
61. Raj agopal , K. R. ; Wineman , A. S.: New un ivers al relat ions for nonlinear
isotropic elastic mat erials, J. Elastici ty 17, 75-83 (1987) .
62. Rivlin, R. S.: Fur ther remarks on t he stress-d eformation relat ions for isot ropic
mater ials, J. Rat ional Mech. A nal., 4, 681-702 (1955) .
63. Ru gger i, T .: Symmetric-hyperbolic system of conservat ive equa tio ns for a vis-
cous heat conducting fluid , Acta Mechanica 47, 167-1 83 (1983) .
64. Ruggeri , T .: Galilean invarianc e and entropy principle for systems of balan ce
laws , Cont inuum Mech. Th emwdyn . 1, 3- 20 (1989) .
65. Ru ggeri , T .; Strumia, A.: Main field and convex covar iant density for quas ilinear
hyp erbolic syste ms. Relativist ic fluid dyn amics. Ann. In st . Henri Poin care, 34
A. 65- 84 (1981) .
66. Smith, G. F .: On isotrop ic fun ct ion s of sy mmet ric tensors, skew-sy mmetric
tensors a nd vectors, Int . J. Eng. Sci. 9, 899-916 (1971) .
67. Struchtrup, H.; Weiss , W .: Maximum of t he local ent ro py product ion becom es
minimal in stat ionary pro cesses, Phys . Rev. Lett . 80, 5048-5051 (1998)
68. Tr eloar, L. R. G .: Th e Phy sics of Rubber Elastici ty, Clarendo n Press, Oxford
(1975) .
69. Truesdell , C .: Rational Th ermodynamics, 2nd editi on, Springer-Verlag, New
York-Berlin (1984) .
70. Truesd ell , C .; Muncaster , R. G .: Fundamental s of Maxwell 's K in etic Th eory of
a Simple Monatomic Gas, Acad em ic Press, New York (1980) .
71. Truesd ell, C. ; Noll, W .: Th e Non-Linear Field Theories of Mechan ics, Hand-
bu ch der Physik , edite d by S. F liigger, Vol 111/ 3, Springer, Berlin-Heidelb erg-
New York (1965) .
292 References

72. Truesdell, C .; Toupin, R. A.: The Classical Field Theories, Handbuch der
Physik, edited by S. Fliigger, Vol III/I , Springer, Berlin-Heidelberg-New York
(1960) .
73. Wang, C.-C.: Stress relaxation and the principle of fading memory, Arch . Ra-
tional Mech. Anal. 18, 117-126 (1965) .
74. Wang, C .-C .: A new representation theorem for isotropic functions , Part I and
II , Arch. Rational Mech. Anal. 36,166-197,198-223 (1970) ; Corrigendum , ibid.
43, 392-395 (1971) .
75. Wang, C .-C .; Liu, I-Shih: A note on material symmetry, Arch. Rational Mech.
Anal. 74, 277-296 (1980) .
76. Wang, C .-C. ; Truesdell, C.: Introduction to Rational Elasticity, Noordhoff In-
ternational Publishing, Leyden (1973) .
77. Wilmariski , K. : Thermomechanics of Continua, Springer, Berlin-Heidelberg
(1998)
78. Wilmariski , K. : Toward a n extended thermodynamics of porous and granular
materials, Trends in Applications of Mathematics to Mechanics, edited by G .
Iooss, Longman, New York (1999).
79. Woods, L. C. : The bogus axioms of continuum mechanics, Bull. of Mathematics
and its Applications, 17,98-102 (1981) .
Index

Absolute temperature 53, 130, 188 - kinematic 37

Acceleration 13, 15 Components
Angular momentum 41 - contravariant 237
Angular velocity t ensor 26 - covariant 237
Anisotropic invari ant 122 - transformation ru les 243, 274
Anisotropic material 87 Configuration 1
Availability 148 - refer ence 1
- for biaxial stretching 163 Conservation of
- for spherical shell 180,182 - a ngular momentum 50
Axial vector 25,253 - energy 52
- linear momentum 50
Balance equation - mass 39
- formal structure 200 Constitutive relation 63
- general 31 - reduced 75
- of mom ents 207 Constrained materials 139
Balance of Controllable deformation 156
- internal energy 53 Convected t ime derivative 30
- kin etic energy 53 Coordinate cur ve 273
- linear mom entum 50 Coordinate system 272
- mass 39 - change of 274
Biaxial stretching 159 - metric tensor 273
Body force 42 - natural basis 273
- apparent 61 - orthogonal 281
- inertial 61 Coriolis force 61
Boundary conditions 57,155 Corotational time derivative 29
Covariant derivatives 275 ,277
Cauchy's fundamental theorem 45 Curl 278
Cauchy 's law 50
Cauchy's postulate 43 Deformation 1
Cauchy-Green tensor 5 - relative 18
Cayley-Hamilton theorem 103,257 Deformation gradient 2
Centrifugal force 61 - relative 18
Chain rul e 269 Description
Christoffel symbols 275,277 - material 14
- transformation rule 277 - spatial 14
Clausius-Duhem inequality 130, 132 Determinant 245
Closure problem - of lin ear transformation 247, 268
- formal procedure 216 Deviatoric part 110, 189
- in exte nded thermodynamics 217 Dilatation 11
Coefficient of thermal expansion 147 Dir ectional derivative 265
Commutation theorem 258 Displacement 9
Compatibility condition 13 - gradient 9,108
294 Index

Divergence 277 Fluid 80

- of t ensor field 278 ,286 - elas t ic 111
Divergen ce theorem 280 - Navi er-Stokes 110, 119, 139, 143
Du al basis 236 - Newtonian 110
- Reine r-Rivlin 109
Elastic fluid 111,203 - Stokes 110
Elastic materi al - viscous heat-conducting 98, 118
- thermodynamics of 132 Fluid crystal 80
El astic solid Flux 31
- isotropic 107 For ce 41,42
- orthotropic 125 - body 42
- transvers ely isotropic 124 - contact 42
Elasticity - moment of 42
- linear 93 Fourier 's law 119,133
Elasticity ten sor 93, 108 - heat cond uct ion 229
- dyn amic 94 Frame of referen ce 1
Energy - change of 22,65
- free 132 Frame-indifferen ce 24, 58
- internal 51 Free energy 132, 150
- kin eti c 51 - Gibbs 151
Energy flux 52 Free ent halpy 135, 151,203
Entropy inequality 129
Galilean invariance
Entropy principle 131,185,192
- of balance equation 204
- in exte nded thermodyn amics 214
- of dyn amic laws 42
Ent ro py production 129
Galilean transformation 26
- den sity 223
Gener al balanc e equa t ion
Entropy-entropy flux
- at a regular point 35
- integrability relation 200, 216
- at a singular point 36
- pair 201
- in integral form 31
Equation
- in material coordi na tes 36
- Navier-Stokes 111
Gen er al linear group 255
- of continuity 40
Generator fun ction
- of motion 50 - hybrid 216
- Rankine-Hugoniot 36 - of balanc e equa t ion 201
Equivalent motion 68 Gibbs relation 134, 188
Eshe lby te nsor 150 - for elast ic solids 135
Euclidean point space 262 Gradient
Eu clid ean transformation 23 - of t ensor field 265
Euclidean vector space 234 - of ten sor function 266
Euler equat ions 112, 203
Eul er 's laws 41 Heat conduction
Extended thermod yn amics 199 - 13-moment theory 230
- closure problem 213 - equat ion of 138
- of 13 moments 217 - Fourier's law 229
- of ideal gas es 228 Heat flux 52
Exterior prod uct 252 - materi al 56, 76
Hemitropic invariant 119
Fading memory 89 Homogeneous deformation 157
Fi eld equa t ion 54 Hooke's law 108
- gen eral 35 Hyp erbolic system
- in arbit rary fra me 60 - of balan ce equa t ions 200
- in materi al coord ina tes 37 - symmetric 201
Fi rst law of thermodynamics 52 Hyp erelastic materi al 74, 134
 Index 295

Ideal gas Lin ear t ransfor mat ion 238

- class ical 226 - grad ient of 265
- mon atomic 226 Linea r viscoelasticity 90
In compressibility 142 - of rate type 93
Incompressible materi als 142 Local speed of propaga t ion 36
Index notation 279
Inertial force 61 Mass den sity 39
Inertial fra me 41,42 , 58 Material
In extensibility 143 - of grade n 97
Inner product 234 - with fad ing memory 90
- of tensors 255 Mater ial heat flux 56
Internal const raint 139 Material objectivity
- incompressibility 142 - cond it ion 70, 73, 75
- inext ens ibili ty 143 - principle 65,68
Int ernal energ y 51 Material sy mme t ry 77
Inv ari an t - group 79
- a nisotropic 112, 122 - t ra nsfor mat ion 78
- fun ctional basis 113 Materi al time derivative 15
- hemi tropic 119 Mechanical power 51
- isotropic 99, 112 Metric t enso r 273
Inv erse 255 Mom en t density 208
Isotherm al compress ibility 146 , 147 - internal 208
Isot rop ic clas tic materi al 107 Mom ent equat ion 207
Isotropic function 98 - sys t em of 210
- representat ion theorem 99, 100, 105 Moment of momentum 41
Isotropic invariant 99 Mon atomic ideal gas 226
Iso tropic ma terial 86 - 13-moment theory 226
- reduced constitutive relati on 88 - classical 226
Isotropic solid 87 Mooney-Rivlin materi al 154
- t hermoelastic 98 ,11 8 Motion 13
- incompressible 34
Jump cond iti on 54 - rigid 17
- en t ropy 130
- general 36
Nat ural basis 273
- in mater ial coordi nat es 37
Navier-Stokes equation 111
Jump discontinuity 33
Navier -Stokes fluid 110
Kelvin effect 168 Navier--Stokes- Fou rier fluid 119,188
Kin em at ic com patibility condi t ion 37 Neo-H ook ean material 155
Kin et ic energy 51 Noll's rule 80
Kron ecker delt a 236 Norm 234
- of t ensors 255
Lagran ge multiplier 185, 200 Normal st ress effect 168
- lemma , ex istence of 190
- method of 189 Objective tensor 22,24
Lame elastic moduli 108 Orientation 250
Laws of dynamics 41 Orthogon al group 256
Line ar elas t icity 93 - proper 256
- equation 109 Or thogon al tran sformation 256
- isotropic 108 - proper 256
Linear momen tum 41 Or thotropy 87, 124
Lin ear t hermoe lasticity 135
- a nisot ropic 136 P ath 13
- isotropic 137 Permuta tion symbol 250
296 Index

Phase equilibrium 150 Simple solid 80

Physical components 281 - reduced constitutive relation 81
Piola-Kirchhoff stress tensor 56 Singular surface 33, 149
- second 76, 124 Solid 80
Poisson's ratio 108 - isotropic 87
Polar decomposition 5 Special linear group 256
- theorem 260 Specific heat
Positive definite tensor 259 - at constant pressure 147
- square root 259 - at constant volume 136,146,148
Poynting effect Spectral theorem 258
- pure shear 172 Spherical shell 173
- simple shear 168 - eversion 175
Principal invariant 103,257 - inflation 176
Principal stretch 5 Spin tensor 21
Principle of - of frame 26
- determinism 64,141 Square root
- material objectivity 65,68, 184 - positive definite tensor 259
Product Stability
exterior 252 - criterion 148
inner 234 of equilibrium 144
tensor 238 - of spherical shell 179
vector 254 - of square sheet 162
Product basis 239 - thermodynamic criteria 149
Product rule 270 Stokes fluid 110
Pure shear 172, 173 Stored energy function 74,134,139
Strain tensor
Rankine-Hugoniot equation 36 - Almansi-Hamel 8
Rate of strain tensor 21 - Green-St . Venant 8
Reaction stress 140 - infinitesimal 10, 108
Reduced constitutive relation 75 - linear 8
- anisotropic solid 81 Stress
- fluid 83 - normal 47
- shear 47
- isotropic solid 88
- simple solid 81 Stress tensor 43
- Cauchy 47
Reference configuration 1
- Piola-Kirchhoff 56
- change of 4
Stretch
Response function 64
- biaxial 159
Rigid motion 17
- uniaxial 158
Rivlin-Ericksen tensor 21,97
Stretch tensors 5
Rotation tensor 5
Summation convention 235
- infinitesimal 10
Surface traction 42
Rule of equipresence 75
Symmetric hyperbolic system 201
- of balance equations 202
Second law of thermodynamics 129 System of moment equations 207
Second-order tensor 239, 254 - hierarchy 213
Shear
- pure 48, 172 Tangent space 262
- simple 6, 166 Temperature 53
Shear modulus 108, 167 - absolute 188
Simple fluid 80 Temperature gradient 72
- reduced constitutive relation 83 - referential 72, 82
Simple material body 73 - relative 82
Simple shear 166 - spatial 74,82
 Ind ex 297

Tensor Unimod ula r group 256

- cont ravariant 240 Unimod ular transformation 79,256
- covariant 240 Universa l relat ion 155
- skew-symmetric 242 - pu re shear 171
- symmetric 242 - sim ple shear 167
Tenso r product 238 Univer sal solution 157
- space 239
Thermal conductivity 225, 231
- t ensor 133 Vector product 254
Thermodynamic process 131 Velocity 13
Thermoelasticity - grad ient 15
- linear 135 Viscoelasticity
Time derivative - Boltzmann-Volterra theory 92
- convecte d 30 - finite linear 92
- corotati onal 29 - linear 90
- materi al 15 Viscosity
Tr ace 248 ,268 - bulk 110
Tr aceless part 110, 189,211 - coefficients 110
Trajectory 13 - shear 110,225
Transformation Viscosity coefficients 189
- Euclidean 23 Viscous heat-conducting fluid 98, 118,
- Galilean 26 184
- rigid 26 - 13-moment theory 217
Tr anslation space 262 Volume element 250
Tr ansport theorem 32 Vorticity vector 21
- with a singular surface 34
Tr anspose 241
Tr an sverse isotropy 87,124 Young's modulus 108