Solid Mechanics: Theory, Modeling, and Problems: January 2015

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/292643555
Solid mechanics: Theory, modeling, and problems
Book · January 2015

DOI: 10.1007/978-3-319-19566-7
CITATIONS READS
8 3,404
2 authors:
Albrecht Bertram R. Glüge

Technische Universität Berlin Otto-von-Guericke-Universität Magdeburg
166 PUBLICATIONS 1,416 CITATIONS 41 PUBLICATIONS 121 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Dissertation View project
Experimental Investigations and Numerical Simulations of Lamellar Copper-silver Composites View project
All content following this page was uploaded by R. Glüge on 04 April 2016.
The user has requested enhancement of the downloaded file.

Albrecht Bertram · Rainer Glüge
Solid
Mechanics
Theory, Modeling, and Problems
Solid Mechanics
Albrecht Bertram Rainer Glüge
•
Solid Mechanics
Theory, Modeling, and Problems
123
Albrecht Bertram Rainer Glüge
Fakultät für Maschinenbau Fakultät für Maschinenbau
Otto-von-Guericke University Magdeburg Otto-von-Guericke University Magdeburg
Magdeburg, Sachsen-Anhalt Magdeburg, Sachsen-Anhalt
Germany Germany
First German edition Febr. 2013
ISBN 978-3-319-19565-0 ISBN 978-3-319-19566-7 (eBook)

DOI 10.1007/978-3-319-19566-7
Library of Congress Control Number: 2015948416
Springer Cham Heidelberg New York Dordrecht London

© Springer International Publishing Switzerland 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made.
Printed on acid-free paper
Springer International Publishing AG Switzerland is part of Springer Science+Business Media

(www.springer.com)
Preface
Do not say anything against Mechanics!

Isn´t it the most beautiful science?
Honoré de Balzac (The Magic Skin, 1831)
This book results from a two-semester course that the authors offered for many
years at the University of Magdeburg. Its understanding requires some basic
knowledge in Mechanics and Mathematics as they are taught in the first semesters
for practically all engineering students of universities.
The entire book is restricted to small deformations (geometrically linear theory).
For the general (non-linear) theory including large deformations, the reader is
referred to further literature1.
The focus of this book lies on material modelling. This is probably a topic that
cannot be substituted and will never be completed. Consequently, we start with an
overview of the most important branches of material theory, initially in a one-
dimensional form. For extending this into a fully three-dimensional one, there is a
need for appropriate mathematical and notational tools. For Continuum
Mechanics, these are the calculus of tensors, for which we give a brief
introduction. Here we limit our considerations to the most simple form of it by
using exclusively orthonormal bases ("Cartesian tensors"). Afterwards we can
outline Continuum Mechanics in a direct tensor notation.
In the three-dimensional material theory, we will again introduce elasticity,
viscoelasticity, and plasticity, all in a more general format. We also have to
include thermodynamics, as far as it is needed for material theory.
The content of the whole book is classical, little is really new. It only differs
from other literature in its selection of topics and, to some extent, their
representation. As EUGEN ROTH already remarked:
Die Wissenschaft, sie ist und bleibt,
was einer ab vom andern schreibt -
doch trotzdem ist, ganz unbestritten,
sie immer weiter fortgeschritten.
1
such as A. Krawietz: Materialtheorie. Springer-Verlag, Berlin (1986) and A.
Bertram: Elasticity and Plasticity of Large Deformations - an Introduction.
Springer-Verlag, 3. edt. (2012)
vi Preface
In each chapter of the book, we give selected references to the most important
literature in the field. For Chapter 4.1 (elasticity) the author has been mainly
inspired by GURTIN (1972), and for Chapter 4.2 (thermoelasticity) by
CARLSON (1972) (both in the Encyclopedia of Physics VIa/2). ARNOLD
KRAWIETZ contributed particularly to Chapter 4.4.6 (thermoplasticity). He also
critically reviewed the entire manuscript and gave numerous suggestions, which
shall be gratefully acknowledged here.
The understanding of many parts of the book is additionally supported by a
number of Problems, which are due to the second author R.G. They contain a
selection of applications, which have been worked out and collected by different
tutors through the years. These were in chronological order WOLFGANG LENZ,
THOMAS BÖHLKE, MICHAEL SCHURIG, GERRIT RISY and RAINER
GLÜGE. In particular, we have to mention MICHAEL SCHURIG for the
Problems 12, 16, 17, and 18. Problems 12, 16, and 20 have been adopted from the
textbooks by GÖLDNER (1991) and IEŞAN (2004). As is common for textbooks
on classical material like Continuum Mechanics, not all sources can be mentioned.
The authors want to apologize for sources that they used without mention.
In all the years in which we gave this course, this manuscript has been
continuously reworked and improved. The authors would like to thank all the
students and readers who have helped us in this process by their suggestions and
comments. In addition, such suggestions will be highly welcome in the future and
can be directly given to the authors.
The German version of this book can be freely downloaded from the authors´
internet site.
Albrecht Bertram, Rainer Glüge Magdeburg, 2015

Contents
1 ONE-DIMENSIONAL MATERIAL THEORY 1
1.1 Deformations, Stresses, and Work 1
1.2 Elasticity 6
1.3 Viscoelasticity (Rheology) 9

1.3.1 The NEWTON Element 10
1.3.2 The MAXWELL Model 13
1.3.3 The KELVIN Model 17
1.3.4 The POYNTING Model 20
1.3.5 The BURGERS Model 22
1.3.6 Viscoelastic Models of Differential-Type 24
1.3.7 Viscoelastic Models of Integral-Type 25
1.3.8 Creep Damage 27
1.3.9 Fatigue 31
1.4 Plasticity 34
1.4.1 Rigid-Plastic Models 34
1.4.2 Elastic-Plastic Models 38
1.5 Viscoplasticity 40
2 INTRODUCTION TO TENSOR CALCULUS 43
2.1 Vector and Tensor Algebra 44

2.1.1 Summation Convention 44
2.1.2 Vectors 46
2.1.3 Dyads and Tensors 49
Problem 1. Linearity 49
2.1.4 The Inverse of a Tensor 55
2.1.5 The Transpose of a Tensor 56
2.1.6 Square Forms and Tensor Surfaces 59
2.1.7 Cross-Product between Vectors and Tensors 59
2.1.8 Orthogonal Tensors 62
2.1.9 Transformations under Change of Basis 64
2.1.10 Eigenvalues and Eigenvectors 65
2.1.11 Spectral Forms of Symmetric Tensors 70
Problem 2. Products between Vectors 72
Problem 3. Direct Notation and Index Notation 74
Problem 4. Orthogonal Tensors 75
Problem 5. Eigenvalues and Invariants 77
Problem 6. Spectral Form 80
viii Contents
2.1.12 Time-Dependent Vectors and Tensors 81

2.1.13 Rigid Body Dynamics 82
2.1.14 Bending of Bars 89
2.1.15 Higher-Order Tensors 93
2.1.16 Tetrads 98
Problem 7. Multiple Contraction 102
2.2 Vector and Tensor Analysis 104

2.2.1 The Directional Differential 104
2.2.2 The Nabla Operator 109
2.2.3 Cylindrical Coordinates 113
Problem 8. Cylindrical Coordinates I 115
Problem 9. Cylindrical Coordinates II 117
2.2.4 Integral Transformations 118
3 FOUNDATIONS OF CONTINUUM MECHANICS 121
3.1 Kinematics 121

3.1.1 Compatibility Conditions 134
Problem 10. Compatibility Conditions 137
Problem 11. Torsion 138
3.2 Stress Analysis 140

3.2.1 The Principles of Mechanics 147
3.2.2 Stress Functions 154
Problem 12. Stress Functions 159
4 THREE-DIMENSIONAL MATERIAL THEORY 164
4.1 Elasticity 165

4.1.1 Material Symmetry 169
4.1.2 Isotropic Elasticity 179
4.1.3 Projection Method 187
4.1.4 Identification of the Elastic Constants 190
4.1.5 Elastic Energy 192
4.1.6 Boundary Value Problems in Elastostatics 197
Problem 13. Elastostatic Boundary Value Problem 203
Problem 14. Pull-out of a Bar 206
Problem 15. Rotating Compact Disc 208
4.1.7 Variational Principles in Elastostatics 211
4.1.8 Displacement Functions 218
Problem 16. Force on an Elastic Half-Space 223
Problem 17. HERTZean Contact 2 39
Problem 18. Comparison of Variational Principles 233
Problem 19. Principle of Minimum Strain Energy 240
4.1.9 Wave Propagation in Elastic Media 2 44
Contents ix
4.2 Thermomechanics 249

4.2.1 Thermodynamic Balances 249
4.2.2 Thermoelasticity 252
4.2.3 Linear Thermoelasticity 255
4.2.4 Isotropic Linear Thermoelasticity 259
4.2.5 Boundary and Initial Value Problems of Thermoelasticity 261
Problem 20. Thermoelastic Boundary Value Problem 262
4.3 Linear Viscoelasticity 266
4.4 Plasticity 271

4.4.1 Yield Criteria 272
4.4.2 Flow Rules 281
4.4.3 Hardening Rules 284
4.4.4 Consistency Condition 286
4.4.5 DRUCKER´s Postulate 288
Problem 21. SCHMID´s Law 2 91
Problem 22. J2Plasticity 293
4.4.6 Thermoplasticity 301
5 INDEX 312
List of the Most Important Notations
The important equations have been numbered chapterwise. The equations of the
Problems have been separately numbered with a preceding “P” followed by the
number of the Problem.
The following acronyms have been used for fundamental statements, which occur
frequently in different versions.
C1 1st equation of CAUCHY (3.2.7)

C2 2nd equation CAUCHY (3.2.8)
CDI CLAUSIUS-DUHEM inequality (4.2.9)
FOURIER FOURIER´s law of heat conduction (2.2.6), (4.2.32)
HOOKE HOOKE´s law (4.1.1)
INC compatibility conditions (3.1.34)
WB work balance (3.2.14)
ND NAVIER´s displacement equation (4.1.37)
PVF principle of virtual force (3.2.20)
PVP principle of virtual power (3.2.13)
DSR displacement strain relations (3.1.23)
RDI residual dissipation inequality (4.4.65)
1LT 1st law of thermodynamics (4.2.3)
2LT 2nd law of thermodynamics (4.2.6)
Frequently used abbreviations
COOS coordinate system

ODE ordinary differential equation
PDE partial differential equation
ONB orthonormal basis
List of frequently used symbols
A , A0 content of area (of cross-section), current and initial

A surface area
a acceleration
Aa , aa work of external loads (global and specific)
Ai , ai stress work (global and specific) (1.1.14)
b body force (mass specific) (3.2.1)
B body
C stiffness tensor, elasticity tensor (4.1.1)
d diameter of sample, damage parameter (1.3.26)
dO angular momentum w.r.t. O (2.1.110)
D rate of deformation tensor (3.1.31)
E strain tensor with components ij (3.1.23)
E' strain deviator
E YOUNG´s modulus (1.2.1)
Ei internal energy
List of the Most Important Notations xi
f force vector
F deformation gradient (3.1.16)
g temperature gradient (2.2.5)
G shear modulus (4.1.14)
H displacement gradient (3.1.14)
I identity tensor (2.1.34)
JM tensor of inertia w.r.t. M with components Jij (2.1.115)
K heat conduction tensor
k modulus of compression (4.1.13)
K kinetic energy (2.1.19), (3.2.14)
L velocity gradient (3.1.30)
l , l0 length, initial length
m vector of torque
m mass
M centre of mass (2.1.107)
n normal vector
0 zero tensor (2.1.37)
o zero vector
p pressure (3.2.11)
p linear momentum vector (2.1.109)
Pi projection tensor (4.1.21)
q heat flux vector (2.2.6)
q local heat supply per unit time and unit volume (4.4.74)
Q global heat supply per unit time (4.2.1)
Q rotation tensor (2.1.74)
r heat source per unit mass (4.2.1)
r position vector
rM position vector to centre of mass (2.1.107)
R real numbers
S compliance tensor (4.1.2)
T stress tensor (3.2.5)
T' stress deviator (3.2.12)
t time
t traction vector (3.2.2)
u displacement vector (3.1.3)
V content of volume
V vector space, volume
v velocity vector
W vorticity tensor (3.1.32)
W, w stored energy, strain energy (global and spec.) (4.1.30)
W*, w* complementary stored energy, stress energy (global and
specific)
X position vector in reference placement (3.1.2)
x position vector in current placement (3.1.1)
X1, X2, X3 material Cartesian coordinates (3.1.5)
x1, x2, x3 spatial Cartesian coordinates (3.1.4)
Z hardening variable in Chapt. 5.4
xii List of the Most Important Notations
 coefficient of thermal expansion

 motion (3.1.1)
 LAPLACE operator (2.2.54)
 variation, virtual
 specific dissipation (4.2.11)
ij KRONECKER symbol (2.1.1)
 dilatation of volume (4.1.10)
 specific internal energy (4.2.3)
ij strains (3.1.23)
ijk permutation symbol (2.1.2)
 shear
 specific entropy (4.2.6)
 flow criterion (4.4.3)
 coefficient of heat conduction (2.2.6)
 LAMÉ´s constant (4.1.17),
in Chap. 4.4 plastic parameter (4.4.29)
 LAMÉ´s constant (4.1.17)
 POISSON´s ratio (1.1.7), (4.1.16)
 temperature
Πa , πa power of external loads (global and specific) (3.2.15)
Πi , πi stress power (global and specific) (3.2.16)
 mass density
 stress (1.1.9)
 shear stress, time variable
 vector of angular velocity (2.1.114)
 specific free HELMHOLTZ energy
 nabla operator (2.2.20)
Literature on the History of Mechanics
Bell, E. T.: Men of Mathematics. Simon and Schuster, New York, 1937.
Benvenuto, E.: An Introduction to the History of Structural Mechanics. Part I, II.
Springer-Verlag, New York, 1991.
Capecchi, D.; Ruta, G.: Strength of Materials and Theory of Elasticity in 19th
Century Italy. Springer-Verlag, 2015.
Dugas, R.: A History of Mechanics. Dover Pub., New York, 1988.
Fierz, M.: Vorlesungen zur Entwicklungsgeschichte der Mechanik. Springer-
Verlag, Berlin, 1972.
Hund, F.: Geschichte der physikalischen Begriffe. BI Wissenschaftsverlag,
Mannheim, 1968.
Kurrer, K.-E.: The History of the Theory of Structures. From Arch Analysis to
Computational Mechanics. Ernst & Sohn, Berlin, 2008.
Stein, E. (edt.): The History of Theoretical, Material and Computational
Mechanics - Mathematics Meets Mechanics and Engineering. Springer, Berlin
(2014)
Szabo, I.: Geschichte der mechanischen Prinzipien. Birkhäuser, Basel, 1977.
Timoshenko, S. P.: History of Strength of Materials. McGraw-Hill, 1953.
Todhunter, I.; Pearson, K.: A History of the Theory of Elasticity and of the
Strength of Materials: From Galilei to the present time. University press, 1893.
Truesdell, C. A.: Essays on the History of Mechanics. Springer-Verlag, Berlin,
1968.
1 ONE-DIMENSIONAL MATERIAL THEORY
Literature
Besseling, J. F.; van der Giessen, E.: Mathematical Modelling of Inelastic

Deformation. Chapman & Hall, London (1994)
Besson, J.; Cailletaud, G.; Chaboche, J.-L.; Forest, S., Blétry, M.: Non-Linear
Mechanics of Materials. Springer, Dordrecht (2010)
Betten, J.: Kontinuumsmechanik. Springer, Berlin (1993), 2. edition (2001)
Fung, Y. C.; Tong, P.: Classical and Computational Solid Mechanics. World
Scientific, Singapore (2003)
Krawietz, A.: Materialtheorie. Springer-Verlag, Berlin (1986)
Lemaitre, J. (edt.): Handbook of Materials Behaviour Models. Academic Press,
San Diego, vol. 1 (2001)
Skrzypek, J. J.: Plasticity and Creep. CRC Press, Boca Raton (1993)
1.1 Deformations, Stresses, and Work

We consider at first a uniaxial tensile or compression test. For such a test, one
often uses samples with rotational symmetry, which are clamped at the (thicker)
ends, and have a middle region that is long enough to apply the measuring device
so that we can assume a uniaxial state of stress. A cylindrical form shall keep the
inhomogeneities as low as possible so that we can assume a homogeneous state of
stresses and strains.
Deformations cannot be measured directly in most cases, but only indirectly as
changes of the length of the control region
l : = l – l0
with initial length l0 and final length l . The (linear) strain is defined as the
derivative of the displacement u with respect to the axial coordinate
du( x )
(1.1.1)  := ,
dx
which coincides only for homogeneous deformations with
l l  l0 l
(1.1.2)  = = = – 1.
l0 l0 l0
Strains are dimensionless and, if their values are small, often given in percent, i.e.,
multiplied by 100. As an example, let the initial length be 30 mm being extended
to 31 mm. We obtain
 = 0,033 = 3,3 % .
Ó Springer International Publishing Switzerland 2015 1

A. Bertram and R. Glüge, Solid Mechanics,
DOI 10.1007/978-3-319-19566-7_1
2 1 One-Dimensional Material Theory
Positive values of  stand for elongations, negative ones for compressions of the
sample.
The above definition of strain is by no means the only one. Another definition is
the logarithmic or HENCKY strain or true strain (IMBERT 1880, LUDWIK
1909)
dl
(1.1.3) dln : =
l
so that
l
dl l
ln =  l
= ln( ) .
l0
l0
The transformation between these two strain measures is

(1.1.4) ln = ln (1 + )
or inversely
(1.1.5)  = exp (ln) – 1 .
0.5
0.4
0.3
0.2
ln
0.1
 ln 0
0.1
0.2
0.3
0.4
0.5
0.5 0.4 0.3 0.2 0.1 0 0.1 0.2 0.3 0.4 0.5
2
Heinrich Hencky (1885-1951)
If we expand this relation into a TAYLOR series

1 1 3
 = exp (ln) – 1 = ln + ln2 + ln + ...
2! 3!
it is obvious that the two measures coincide for small absolute values of   ln ,
i.e., if we take only the linear terms and neglect the higher-order terms.
The advantage of the logarithmic strain is its additivity. In fact, if we pull the
sample first to a length l1 and then to l2 , we obtain
l1  l0 l2  l1
1 = and 2 =
l0 l1
while
l2  l0
 =  1 + 2 .
l0
On the other hand we get
l1 l2
ln1 = ln( ) and ln2 = ln( )
l0 l1
which give the additivity
l2 l l
ln = ln( ) = ln( 2 1 ) = ln1 + ln2 .
l0 l1 l0
Thus, the total logarithmic strain is the sum of the partial logarithmic strains.
In our example we have
 = 0.0333
ln = ln (1 + 0.033) = ln (1.033)  0.0328
ln2 = 0.0011
Let d0 denote the initial diameter, A0 its initial cross-section area, and V0 its
initial volume, and d , A and V the diameter, the cross-section and the volume in
the deformed state, respectively. During the tensile test, not only axial but also
transversal deformations occur, which we measure as
d  d0
(1.1.6)  := .
d0
If the material is isotropic and the sample circular, then we can expect the same
value in all radial directions. POISSON's3 ratio is defined as the quotient

(1.1.7) v := – .

3
Siméon Denis Poisson (1781-1840)
It is also dimensionless. For linear-elastic materials this ratio is a material constant

with values between 0 and ½ for most materials. However, there are also cases
with negative values. For incompressible materials the deformation must be
isochoric (volume preserving)
V d 2l d  d0 l  l0
1 = = 2 = ( + 1) 2 ( + 1)
V0 d0 l0 d0 l0
= (  +1)2 ( + 1)  2  +  + 1
 2  +  = (1 – 2 v)   0
so that v = ½ . In this case, we obtain for the ratio of the cross-section area
A0 l
(1.1.8) A0 l0 = A l  = = 1+
A l0
A
  1– .
A0
In addition, measurement of the stresses cannot be achieved directly, but only

through the resulting force F acting on the sample at both ends. This force is the
integral of the axial stresses  over the cross-section
(1.1.9) F =   dA .
A
If  is homogeneous (i.e. constant in space), we obtain
F
(1.1.10)  = .
A
This stress is sometimes called the true stress. In contrast to this we obtain the
nominal stress or engineering stress if we relate it to the initial cross-section
F
(1.1.11) 0 : = .
A0
The dimension of all these stresses is [force / area]. The international unit for
stress is Pascal 4
Pa : = N m–2.
With (1.1.6) we obtain the relation between the two types of stress
F F A d2
0 = = =  2 =  (  +1)2   (2  + 1)
A0 A A0 d0
4
Blaise Pascal (1623-1662)
=  (1 – 2 v )
and in particular for the incompressible case (1.1.8)
(1.1.12) 0 =  (1 –  ) .
For a stress-strain diagram it is therefore important to know which types of
stresses and strains are used. The larger the strains are, the larger become also the
differences between them. If the strains are small, then not only do the different
strain measures coincide, but also the different stresses, i. e.,   0 . Since this
book is almost exclusively dedicated to small deformations, we can concentrate on
relations between  and  without having to explain which type of strain and
stress measure is meant.
During a deformation / stress process, stresses and strains are functions of time:

 (t) and  (t) . We use a dot to indicate the derivative with respect to time.
The stress power density is defined as
(1.1.13) πi : =   
with the dimension [stress / time]. The time integral of the power between two
instants is the density of the stress work
t t  t 
  

(1.1.14) ai : = πi () d =  ()  () d =  d
0 0 0
with dimension [stress].
By interchanging the roles of stress and strain in these expressions, we obtain
the complementary power (indicated by an asterisk)
(1.1.15) πi * : =   
and the specific complementary work
t  t 
 

ai* : =  ()  () d =  d .
0 0
ai*
ai

The name is due to the fact that

ai + ai* =  (t)  (t)
called the final work.
The global power/ work of the body is then the volume integral of the specific
power/ work over all of its points.
1.2 Elasticity
Mechanical material theory addresses the relationships between stresses and

strains for different materials. Materials also differ in their ability to remember
past events. If the past has an effect on the present material behaviour, we call this
memory. There are materials with a long-range memory, those with a fading
memory, and others with no memory at all. These latter materials define elastic
behaviour. This gives rise to the following definition.
Elasticity: The current stresses depend only on the current deformations, and vice
versa.
We consider the control range of a sample made of some elastic material under
arbitrary loadings. The stresses  (x0) at a section x0 depend on the
displacements u(x) , 0  x  l , which gives rise to a functional
 (x0) = F(u(x) xx 0l ) .
x0
We expand u(x) into a TAYLOR series at x0

1.2 Elasticity 7
u(x) = u(x0) + u(x0)' (x – x0) + ½ u(x0)'' (x – x0)2 + ...

Then we can consider the stresses as a function of all the derivatives at the same
point x0
 (x0) = f (u(x0) , u(x0)' , u(x0)'' , ...) .
When shifting the sample as if it were rigid, only u(x0) is affected. However,
since rigid body motions shall not influence the stresses (principle of invariance
under superimposed rigid body motions), we can drop the first argument u(x0)
from the list of variables. Since the remaining variables are still infinitely many,
one has to truncate at some order for practical reasons. An elastic material of the
form
 = fn(u' , ... , u(n))
(dropping the spatial coordinate x0 on both sides) is called a gradient material of
order n . In most cases, however, one considers the simplest case with n  1, a
simple elastic material
 = f1(u' ) = : f () .
E f ()

So far the function f () can still be arbitrarily non-linear. For stability reasons,
one would expect that f is monotonously increasing. For small deformations,
however, we can linearise this function by substituting it by its tangent in the
origin with incline
df   
E :=
d
 0
and obtain the constitutive law for a simple linear elastic material called
HOOKE´s5 law (1678)
(1.2.1)  = E
For stability reasons, the YOUNG's6 modulus E is supposed to be positive. It is
of the dimension [stress] .
5
Robert Hooke (1635-1703) "ut tensio sic vis"
6
Thomas Young (1773-1829)
The symbol of a linear elastic material is a spring, called a HOOKE element
HOOKE element
The specific stress power of this material possesses a potential w

(1.2.2) πi =    = E    = (½ E  2)
dw( ) 
= : w() =  = (½  )
d
whose time integral gives the stress work
t t 
ai : =  πi d =   ()  () d =   d
0 0 0

(1.2.3) =  E  d = ½ E  2 = : w () = ½  
0
which is stored as elastic or potential energy or strain energy w ().

The complementary work is
t t 
ai* : =  πi * d =   ()  () d =   d
0 0 0

(1.2.4) =  E –1  d = ½ E –1  2 = : w *() = ½  
0
with the complementary stress power

(1.2.5) πi * =    = E –1    = (½ E –1  2) = : w*()
dw* ( ) 
=  = (½  )
d
with the complementary elastic energy or stress energy w*() : = ½   .
Because of the linearity of HOOKE´s law, ai and ai* have equal values.
After (1.2.2) and (1.2.5), we obtain the potential relations
dw( ) dw* ( )
(1.2.6)  = and  = .
d d
1.3 Viscoelasticity (Rheology)

Literature
Bert, C. W.: Material damping: An introductory review of mathematical models,

measures and experimental techniques. J. Sound Vibr., 29 (2), 129-153 (1973)
Bland, D. R.: The Theory of Linear Viscoelasticity. Pergamon Press, Oxford
(1960)
Christensen, R. M.: Theory of Viscoelasticity. Acad. Press (1971)
Clough, R. W.; Penzin, J.: Dynamics of Structures. McGraw-Hill (1982)
Creus, G. J.: Viscoelasticity – Basic Theory and Applications to Concrete
Structures. Springer, Berlin (1986)
Drozdov, A. D.: Mechanics of Viscoelastic Solids. J. Wiley & S., New York,
London (1998)
Ferry, J.: Viscoelastic Properties of Polymers. J. Wiley & S., New York, London
(1961)
Findley, W. N.; Lai, J. S.; Onaran, K.: Creep and Relaxation of Nonlinear
Viscoelastic Materials. North-Holland Pub., Amsterdam (1976)
Flügge, W.: Visco-Elasticity. Springer, Berlin (1975)
Francois, D.; Pineau, A.; Zaoui, A.: Mechanical Behaviour of Materials, vol. II,
Kluwer, Dordrecht (1998)
Freed, A. D.: Soft Solids. Birkhäuser (2014)
Giesekus, H.: Phänomenologische Rheologie. Eine Einführung. Springer, Berlin
(1994).
Gurtin, M. E.; Sternberg, E.: On the linear theory of viscoelasticity. Arch. Rat.
Mech. Anal. 11 (1962)
Krawietz, A.: Materialtheorie. Springer, Berlin (1986)
Lazan, B. J.: Damping of Material and Members in Structural Mechanics.
Pergamon Press (1968)
Leitmann, M. J.; Fisher, G. M. C.: The Linear Theory of Viscoelasticity.
Encyclopedia of Physics VI a/3. Edt. S. Flügge. Springer, Berlin (1973)
Naumenko, K.; Altenbach, H.: Modeling of Creep for Structural Analysis.
Springer, Berlin (2007)
Nowacki, W.: Theorie des Kriechens. Franz Deuticke, Vienna (1965)
Odqvist, F. K. G.; Hult, J.: Kriechfestigkeit metallischer Werkstoffe. Springer,
Berlin (1962)
Phan-Thien, M.: Understanding Viscoelasticity. Basics of Rheology. Springer,
Berlin (2002)
Pipkin, A. C.: Lectures on Viscoelasticity Theory. Springer, Berlin (1972)
Rabotnov, Y. u. N.: Creep Problems in Structural Members. North-Holland
(1969)
Rabotnov, Ju. N.; Iljuschin, A. A.: Methoden der Viskoelastizitätstheorie. Carl
Hanser Verlag. München (1968)
Schlimmer, M.: Zeitabhängiges mechanisches Werkstoffverhalten. Springer,
Berlin (1984).
Tschoegl, N. W.: The Phenomenological Theory of Linear Viscoelastic

Behaviour. An Introduction. Springer, Berlin (1989)
Among the inelastic materials, we find rate-independent (scleronomous) ones,

for which the stresses do not depend on the rate of the deformation process, and
also rate-dependent (rheonomous) ones, for which this is not the case. Rate-
independent materials are considered in elasticity and plasticity. We will first
consider rate-dependent models, again preferably the linear ones (linear
viscoelasticity).
1.3.1 The NEWTON Element

The most simple ansatz for a rate-dependent model is the linear viscous law
t   
(1.3.1)  (t) = D  (t)   (t) =  d +  (0)
0 D
with a positive viscosity or damper constant D of dimension [stress  time]. This

is symbolised by a damper
NEWTON element
called NEWTON7 element. Under constant load, this material creeps at a

constant rate. If one stops the deformation, the stresses will abruptly drop to zero.
Thus, the NEWTON element does not possess an instantaneous elasticity. Neither
does it exhibit a natural length, and can be infinitely deformed under enduring
loads. In this it behaves fluid-like.
The stress power of the NEWTON element is positive semidefinite
2
(1.3.2) πi =    = = D  2 = :  ( )  0 for D > 0
D
being completely dissipated. It can be transformed into heat and cannot be directly
recovered.
Other linear-viscoelastic models can be constructed by parallel arrangements or

by a series of springs and dampers. For such arrangements the following rules
hold:
7
Isaac Newton (1643-1727)
 for parallel arrangements:
E1
E2
(1.3.3)  = 1 + 2 The stresses are additive.

(1.3.4)  = 1 = 2 The strains are equal.
 for arrangements in series:
E1 E2
(1.3.5)  = 1 = 2 The stresses are equal.

(1.3.6)  = 1 + 2 The strains are additive.
With these rules one can construct and handle arbitrary rheological models. It
can be the case that different models describe identical material behaviour (for an
appropriate choice of the material constants).
Before describing common features of all of these models, we will consider the
most simple ones in more detail.
MAXWELL model
KELVIN or
VOIGT model
POYNTING model
BURGERS model
1.3.2 The MAXWELL Model
E D
The MAXWELL8 model consists of an arrangement of a spring and a damper in

series. After (1.3.6) we obtain for the strains in the different elements
 = E + D
with E = E E and D = D D
 E
(1.3.7)    = E + D =  D
E D
and after (1.3.5)
 = E = D
and finally the ODE
 
(1.3.8)   =  .
E D
We will now consider finite stress processes and the accompanying strain
processes between two instants 0 and t , for which we need initial values for
 (0) and  (0) . If the deformation process is prescribed, one wants to determine
the stress process, or vice versa. By integration of (1.3.8) we obtain a form explicit
in the strains
t
1 1
(1.3.9)   t    0     t     0       d .
E D
0
In order to find a form which is explicit in stresses, we multiply (1.3.8) by E

E 
exp  t  and obtain as a result an expression that is a complete time-derivative
D 
after the product rule
E  E  E E    E 
E    t  exp  t      t  exp  t    ( t ) exp  t     t  exp  t    .
D  D  D D    D 
The time integral in the interval from 0 to t is
8
James Clerk Maxwell (1831-1879)
t t
E  E 
 E    exp  D   d     exp   

0    D   0
E 
   t  exp  t     0 
D 
which gives the form explicit in the stresses
t
 E   E 
(1.3.10)   t     0  exp   t   E     exp    t   d .
 D  0 D 
The integral can be reformulated by partial integration
t t
E  E  E t E 
    exp  D   d     exp  D       exp  D   d


0      0 D 0  
so that we get
 E   E   E  
  t     0  exp   t   E exp   t    t  exp  t     0  
 D   D   D  
E2 t E 

D 0
    exp  D (  t )  d
 
 E 
 E  t    0   E   0   exp   t 
 D 
(1.3.11) E2 t E 

D 0
    exp  D   t   d .
 
The first term in the second line represents the elastic part of the stresses, as if the
model only consisted of a spring. The second term represents the influence of the
initial values, which diminishes with time because of the exponential function.
The third term is a convolution integral, which weights the strains in such a way
that more remote events have less influence on the present than recent ones
(fading memory).
k
k
E
t
creeptest
creep test

The stress power of the MAXWELL model is composed of elastic and viscous
parts after (1.3.7) and (1.3.8)
(1.3.12) πi =    = E E + D D = (½ E E 2) + D D 2
= w(E )  +  (D ) .
To characterise the material behaviour of this model, we submit it to a creep test.
We assume that, until a time t  0 , the stress is  (t)  0 , and afterwards  (t) 
k = constant. After (1.3.9) for  (0) = 0 we obtain
k k
   t for t>0
E D
i.e., we have (unbounded) linear creep. This is why the model is sometimes called
a MAXWELL fluid. Sudden loadings are elastically absorbed by the model
(instantaneous elasticity).
If one submits the model to a sudden and maintained constant strain r
(relaxation test), the damper will instantaneously act rigidly and the spring
elastically, so that an initial stress 0 = E r is needed. Afterwards we have by
(1.3.10)
 E 
 (t) = E r exp   t 
 D 
i.e., the stress approaches zero since the spring unloads through the damper.
The material constant tr : = D / E (dimension [time]) is called the relaxation

time. For t  tr we obtain
 (tr) = E r e–1  0,368 0 .
and after (1.3.8) for t  0
 (t) = – tr  (t) 
and particularly for t  0 the initial slope
0
 (0)  = – .
tr


r
 = E  r
 ( tr )
tr t
relaxation test

Hence, the relaxation time is an inverse measure for the speed of the relaxation.
There are many non-linear generalisations of the (linear) MAXWELL model.

An example is the non-linear DISCHINGER9 law for concrete for which the
viscosity is time-dependent
9
Franz Dischinger (1887-1953)
D(t) = R exp(C t) R, C > 0 constants,

describing an increasing stiffening of the concrete.
1.3.3 The KELVIN Model
The KELVIN10 model is defined as a parallel arrangement of spring and damper.
KELVIN model
After the rules (1.3.3) and (1.3.4) we have

 = E = D
E = E 
 = E + D
D = D  
1 E 
(1.3.13)   = E  + D   exp  t 
D D 


1 E  E E  E    E 
  t  exp  t     t  exp  t      t  exp  t     ( t ) exp  t  
D D  D D  D    D 
t t
1 E  E  E 
   exp    d   (  ) exp    =   t  exp  t     0 
D 0 D  D  0 D 
t
 E  1 E 
  t     0  exp   t  +   τ  exp    t   d .
D
(1.3.14) 
 D  0 D 
10
Lord Kelvin, Sir William Thomson (1824-1907)
Vk
H
Hf
H
tk
tk t
Also for this model, the initial disturbance decays exponentially in time, and the
stress process enters as a convolution integral with an exponential kernel.
The stress power is with (1.3.13)
(1.3.15) πi =    = E E  + D D = (½ E E 2)  + D D 2
= w( )  +  ( ) .
The creep test with a creep load k from t  0 and an initial strain  (0)  0
gives with (1.3.14)
t
1 E   E 
 t  
D 
 k exp  τ  dτ exp   t 
0
 D   D 
1 D E  t  E     E 
  k exp  τ  exp   t   k 1  exp   D t  
D E D  0  D  E   
k  E 
  (0) = exp   t  .
D  D 
Hence, there is no instantaneous elasticity. The creep rate slows down, and the
creep strains converge to the elastic strain  : = k / E .
The creep time (or retardation time) tk : = D / E (dimension [time]) is that
particular time, where the strains have grown until
k
 (tk) = (1 – e–1)  0,632 k / E .
E
The relation with the initial slope is

 (0) =  / tk .
A relaxation test is not feasible for this model since finite loads cannot introduce
discontinuous strains. One can, however, study the behaviour under constant strain
after an arbitrary but continuous strain history. In this case, the stress is constant
 = E r in the spring, while the damper is immediately stress-free.
The stress response for an enforced harmonic vibration
 (t) = 0 sin ( t)
is with (1.3.13)

 (t) = E 0 [sin ( t) + D cos ( t)] .
E
In the  - -diagram, this gives an elliptic hysteresis.
The work done during one cycle with a period T = 2 / is completely dissipated
T T T T
 2
 D  dt   0 D   cos   d
2 2 2
ai     dt    dt 
0 0 0 0
2 / 
 1 
  02 D  2   sin( 2 )   02 D   .
 2 4 0
The maximum of the elastic energy of the spring is wmax = ½ E 02 , and the ratio
of dissipated work and maximal stored energy is
 02 D   2D  
(1.3.16)  :  
1 E
E  02
2
and thus frequency-dependent, but not amplitude-dependent. The ratio  /  is

called the loss factor and is a criterion for the damping capability. For other such
criteria, see BERT (1973).
1.3.4 The POYNTING Model
The POYNTING11 model (MALVERN model) is composed of two springs and

one damper.
K D POYNTING model
We have
 = K + C = D + C C = C C
K = K K
 = C = K + D D = D D 
   = C = K + D
 C  K D
  
C K D
  C
 
 C
 
K D
 C   C
(1.3.17)      
K K D D
D  CK 
(1.3.18)      C  D 
K K
This leads to the following explicit forms
11
John Henry Poynting (1852-1914)
(1.3.19)
 K2
t
K    K 
 t    C  K  
 t    0     exp    d   exp   t 

 D  D    D 
0
 t 
 t   
CK
 0  K KC  t  KC  
         exp    d  
  
C  K  D K  C D( K  C )2   D K  C 
  0   
 KC 
exp   t
 D K  C  
 
with the initial conditions (0)  0 and (0)  0 .
The POYNTING model shows both creep and relaxation properties and
instantaneous elasticity.

 k
C
creep curve k
C+K
t

(C+K) r
relaxation curve
C r
t
1.3.5 The BURGERS Model
The BURGERS12 model is introduced as an arrangement in series with a

MAXWELL model and a KELVIN model.
C D
R BURGERS model
Equivalently, it can be introduced as a parallel arrangement of two MAXWELL

models. Here we have
 = C = D = K + R C = C C
K = K K
R = K D = D D
 = C + D + K,R R = R R
   = K K + R R = K (  – C – D) + R (  – C – D)
           
 K       R    
 C D   C D 
 
  1 K 1 K K 
(1.3.20)              .
C  D CR R  DR R
This model shows instantaneous elasticity (because of the spring C ).
For the numerical integration, one can transform this 2nd-order ODE into a
system of two 1st-order ODEs by introducing the stress in the spring K as an
internal variable
   K
    
C D R R
K  K
 K  K      
C D
12
Johannes Martinus Burgers (1895-1981)
or as an explicit version in the stress increments

    
   C     K
 D R R 
K
 K     K  .
R
The creep curve of the BURGERS model under a creep load k results from a
superposition of the creep curves of the KELVIN and the MAXWELL models
k   K  k k
 t   1  exp   R t    t  .
K    D C
In the beginning of the creep process, the damper R contributes to the creep
deformation (primary creep). This contribution, however, decreases more and
more in time, while in the long range the damper D dominates with the stationary
creep rate k / D . This is called the secondary or steady creep phase.
For many real materials, this creep phase is continued by an overlinear tertiary
creep phase, which cannot be described by the BURGERS model alone, in
contrast to the first and second creep phases.
In a relaxation test, the stresses in the spring C relax through the damper D
and converge to zero.
H
VN VN
+
creep curve C K
VN
C
CHU
relaxation curve
t
1.3.6 Viscoelastic Models of Differential-Type
In the following table, the differential equations of the aforementioned rheological

models are listed and brought into a common form as linear and homogeneous
ODEs of the order p in the time derivatives of the stresses, and of the order q in
the strains.
(1.3.21)
HOOKE P0  = Q0 
NEWTON P0  = Q1  
MAXWELL P0  + P1   = Q1  
KELVIN P0  = Q0  + Q1  
POYNTING P0  + P1   = Q0  + Q1  
BURGERS P0  + P1   + P2   = Q1   + Q2  
(p , q) -type P0  + ... + Pp  (p) = Q0  + ... + Qq  (q)
The constants P0 , ... , Pp , Q0 , ... , Qq are determined by those of the springs

and dampers. In all cases, one of the non-zero constants can be normed to 1. These
differential equations completely describe the behaviour of these models, i.e., for
given initial conditions, a prescribed stress process  (t) can be uniquely
integrated for the strain process  (t) , or vice versa (see GURTIN/ STERNBERG
1962). For complicated models and/or complicated processes one would prefer a
numerical integration.
There exist generalisations of these (p, q)-type ODEs to fractional derivatives, i.e.
derivatives of any real order. If one considers this order as a fit parameter, one can
often obtain good approximations of the real behaviour by a small number of
elements compared to integer orders.
If one is particularly interested in the vibrational or damping behaviour of these
models, the use of complex numbers often leads to a more compact model. If such
a system is excited harmonically with a frequency  , the (p, q)-type ODE can be
solved by the functions
 (t) = 0 exp(i  t)
 (t) = 0 exp(i  t)
where 0 and 0 become complex. By inserting these ansatz functions into the
ODEs, we obtain
p q
j k
 Pj  0 i  exp  i t    Qk  0  i  exp  i t 
j 0 k 0
and a pseudo-elastic relation

p
 Pj  i 
j
j 0
(1.3.22) 0 = q
0 = G( ) 0 .
k
 Qk  i 
k 0
Both the real and the imaginary parts of  (t) and  (t) are solutions of the forced
vibration. The quotient G() is the complex compliance, and its inverse is the
complex stiffness. We obtain for the
- KELVIN model G() = (E – i D  ) / (E2 + D2  2)
- MAXWELL model G() = 1/E – i /(D  ) .
One can also introduce a complex viscosity and obtain a pseudo-viscous relation.
There also exist suggestions for generalising the (p, q)-models for cases with
infinitely many elements (see KRAWIETZ 1986 p. 13, BETTEN 2001 p. 257).
1.3.7 Viscoelastic Models of Integral-Type
Another approach to linear viscoelasticity is based on the following assumption,

which generally characterises linear viscoelastic behaviour.
BOLTZMANN13´s Principle of Superposition (1876)
If strain processes  i (t) lead to stress processes  i (t) , i = 1, 2, then the
superposition 1 (t) + 2 (t) leads to 1 (t) + 2 (t) .
If we consider a constant creep load i (t) applied at a time ti , then we can
represent this stress-process by the HEAVISIDE14 function H(t) defined as
H(t) = 0 for t < 0
H(t) = 1 for t  0
in the form
i (t) =  (ti) H(t – ti)
where  (ti) is the creep stress being constant from ti onwards. Let J(t – ti) be
the strain-process resulting from a unit load H(t – ti) starting at ti . Then
according to the superposition principle,  (ti) H(t – ti) induces a strain-process
13
Ludwig Boltzmann (1844-1904)
14
Oliver Heaviside (1850-1925)
i(t) =  (ti) J(t – ti) .

J(t) is called the creep function.
If an arbitrary stress-process  (t) is given, we can approximate it as a sum of
such step-functions
n
 (t) =   (ti) H(t – ti)
i 1
and obtain the strain response
n
 (t) =   (ti) J(t – ti) .
i 1
If we make this approximation finer and finer, the strain function converges in the
limit n   to the STIELTJES15 integral
 t  t
(1.3.23)  (t) =  J(t – ) d () =  J(t – )  () d .
 0  0
By interchanging the roles of stresses and strains, we obtain analogously the

resulting stress-process
 t  t
(1.3.24)  (t) =  R(t – ) d () =  R(t – )  () d
 0  0
with the relaxation function R(t) . The latter is not independent of the creep
function. In fact, between the two there is the relation
t t
t =  R(t – ) J () d =  J(t – ) R () d .
0 0
Since all rheological models of (p, q)-type fulfil the above principle of
superposition, we can represent them all by such STIELTJES integrals.
All these viscoelastic models have the following features:

 They are linear. Their mathematical properties are well-known (see GURTIN/
STERNBERG 1962).
 Their behaviour is rate-dependent or viscous (rheonomous).
 They all describe a behaviour with fading memory. This means the following:
If an event happened a sufficiently long time ago, its effect on the present
becomes negligibly small. In the limit, the initial conditions get forgotten and
15
Thomas Jan Stieltjes (1856-1894)
one obtains a history functional, which contains convolution integrals with

obliviator functions (creep or relaxation functions) of exponential type.
 Under very fast or very slow processes, these models behave elastically since
all dampers become rigid or compliant, respectively.
 Their behaviour is reasonable in the sense that no contradiction to the
dissipation postulate (  0) can occur if the ODEs and in particular their
coefficients are derived from reasonable rheological models. If one, instead,
starts with the ODE, one cannot exclude the possibility of unphysical
behaviour.
The range of all these models is limited by their linearity. In reality, the creep
and the damping behaviour of many materials like metals or polymers is strongly
non-linear. Accordingly, the literature is full of non-linear generalisations of such
rheological models like, e.g., the introduction of a frequency-dependent damping
R0
R    R0 = constant

(KIMBALL/ LOVELL 1927), or of stress-dependent damping (BERTRAM/
OLSCHEWSKI 1990)
R() = R0 exp(– A ) .
Other historical creep functions are
H (t)x = k (~V~ / V0)N NORTON (1929), BAILEY (1929)
H (t)x = k [exp(~V~ / V0) – 1] SODERBERG (1936)
H (t)x = k sinh(~V~ / V0) PRANDTL (1928), NADAI (1938)
H (t)x = k sinh(~V~ / V0)N GAROFALO (1965)
In particular the NORTON law has gained much popularity. For other creep
functions see SZKRYPEK p. 363 ff.
1.3.8 Creep Damage

Literature
Betten, J.: Kontinuumsmechanik. Springer, Berlin (1993), 2nd ext. edt. (2001)
Betten, J.: Creep Mechanics. Springer, Berlin (2002)
Chaboche, J.-L.; Lemaitre, J.: Mechanics of Solid Materials. Cambridge Univ.

Press (1990)
Danzer, R.: Lebensdauerprognose hochfester metallischer Werkstoffe im Bereich
hoher Temperaturen. Gebr. Borntraeger, Stuttgart (1988)
Kachanov, L. M.: Introduction to Continuum Damage Mechanics. Kluwer
Academic Publ., Dordrecht (1986)
Krajcinovic, D.; Lemaitre, J.: Continuum Damage Mechanics. Springer, Vienna
(1987)
Lemaitre, J.: A Course on Damage Mechanics. Springer, Berlin (1992)
Lemaitre, J. (edt.): Handbook of Materials Behaviour Models. Academic Press,
San Diego, vol. 2 (2001)
Lemaitre, J.; Desmorat, R.: Engineering Damage Mechanics - Ductile, Creep,
Fatigue and Brittle Failures. Springer, Berlin (2007)
Murakami, S.: Continuum Damage Mechanics. Springer, Berlin (2012)
Naumenko, K.; Altenbach, H.: Modeling of Creep for Structural Analysis.
Springer, Berlin (2007)
Rösler, J.; Harders, H.; Bäker, M.: Mechanisches Verhalten der Werkstoffe. B.
G. Teubner, Stuttgart (2003)
Saanouni, K.: Damage Mechanics in Metal Forming. ISTE, London/ Wiley,
Hoboken (2012)
Voyiadjis, G. Z.; Kattan, P. I.: Advances in Damage Mechanics: Metals and
Metal Matrix Composites. Elsevier, Amsterdam (1999)
 
primary secondary tertiary primary secondary tertiary
t t
typical creep phases typical creep phases
A creep curve typically starts with a phase of primary creep with high creep
rates. Later it turns into a secondary creep phase with a minimal and constant
creep rate (steady creep). Finally the creep rate increases again (non-linear creep).
This tertiary creep phase finally ends with creep rupture.
While the primary and secondary creep phases can be described by rheological
models within linear viscoelasticity, e.g., by the BURGERS model, this does not
hold for the tertiary creep phase because of its non-linear behaviour. This can be
caused by different mechanisms, such as
 reduction of the cross-sectional area due to transverse contraction, which leads
to an increase of the true stresses;
 initiation and growth of microcracks and pores;
 changes of the microstructure;
and much more. The effects of all these mechanisms are summarised under the
label creep damage in phenomenological theory.
The term “damage“ will be used to designate the general destructive

effect of the various changes occurring in the material, which
eventually limit its usefulness. Thus, the visible damage may consist of
an accumulation of permanent deformation from creep (viscous flow),
plastic flow, initiation of corrosion pits, or development of cracks.
These visible effects are the result of submicroscopic changes in the
crystalline lattice or in the intercrystalline boundary material. The
processes designated by the terms slip, diffusion, recrystallization,
rotation of grains, phase change, and relaxation involve varying
degrees of atomic rearrangement that takes place within the material;
any of these processes may constitute the initial stages of “damage“.
DOLAN (1952)
A very successful suggestion for the description of damage is due to

KACHANOV16 (1958). His leading idea is that only a part of the cross-section A
of the creep sample has to bear the force, namely the effective or reduced cross-
section Ae , reduced by cracks, pores, etc. This leads to effective stresses
F
(1.3.25) e 
Ae
larger than the true stresses
F
  .
A
A Ae
undamaged damaged
16
Lazar M. Kachanov (1914-1993)
If the effective stresses enter the creep law, this causes to a growth of the creep
rate.
For this aim one introduces a damage parameter d as the ratio
A  Ae Ae
(1.3.26) d := = 1   Ae = (1 – d) A
A A
which starts with an initial value d  0 for the undamaged material, and grows
monotonously during creep. If its value eventually approaches 1, the effective
stresses grow unboundedly. In reality, creep rupture will already occur at a critical
value of d which is smaller than 1
0 < dkr < 1
which is considered as a material constant. This damage theory is based on the
Principle of stress equivalence (LEMAITRE 1971)
The damaged material behaves like the undamaged one if the stress in the creep
law is substituted by the effective stress.
The damage parameter d is considered as an internal variable, for which one
needs an evolution equation. By analogy to the BURGERS model, one may use
the ansatz
d  =  e +  e +  d
with constants  ,  and  (BERTMAM 1991). These must be determined in a
way that d in fact monotonously increases during creep.
Another ansatz for this equation is due to RABOTNOV (1963)
d  = (e / ) N
with two positive constants  and N , which gives good results in many
applications. This equation together with (1.3.25, 26) constitute the
KACHANOV-RABOTNOV damage theory.
The rate of the effective stress is then

 F  F F
(1.3.27)  e      Ae .
 Ae  Ae Ae2
In a monotonous creep test, the applied load F is constant, so that F  = 0. The

growth of the effective cross-section area is fed by different sources after (1.3.26)
(1.3.28) Ae = – d  A + (1 – d) A
namely by the growth of the damage parameter d  and the change of the nominal
cross-section area A due to transverse contraction. Often one assumes that the
creep deformation (not the damage) occurs isochorically, which is a common
assumption for metals. If we, moreover, neglect elastic volume changes after
(1.1.8), then with the strains
l  l0 l
    
l0 l0
and the current volume V = l A , we obtain for the change of volume of a
prismatic sample
V  = 0 = l  A + l A =   l0 A + l A.
With
  l0 A 
A     A
l 1
we finally obtain for the increment of the effective stress
(1.3.29)
F F    F F  d  
 e   2  d  A  1  d  A     .
Ae Ae  1    Ae Ae  1  d 1   
After this, it grows due to

 an increment of the applied load F
 the growth of the damage variable d
 the transverse contraction.
Since the strains in the tertiary creep phase may become rather large, it may
occasionally become necessary to distinguish between different stress and strain
measures.
1.3.9 Fatigue
Apart from damage under tensile loads, there is also another type of failure
which plays an important role in technical applications. It occurs under alternating
loads (tension/ compression), and is called fatigue after PONCELET17 (1839). Its
mechanisms are rather different from creep damage. Typically such fatigue
processes initiate under low frequent alternating loads in metals at the surface,
where micro-cracks are already present, and which play the role of germs for
fatigue cracks. These cracks grow very slowly in the beginning, but later may
accelerate, mainly along grain boundaries, until different cracks unite
17
Jean Victor Poncelet (1788-1867)
(coalescence) and finally weaken the cross-section to such an extent that rupture
occurs.
For studying this behaviour, one performs tests with alternating loads (stress
controlled) or with alternating strains (strain controlled). The alternating processes
are in most cases periodic and can be harmonic, triangular, or of some other form.
One measures the number of load changes N until failure. Such tests are needed
for lifetime prediction for technical parts.
WÖHLER curve of a construction steel: stress amplitude [MPa] vs. cycles

to failure at room temperature (from DANZER, p. 13)
If one depicts the stress amplitude over the number of cycles until failure in a
double logarithmic diagram for many sinusoidal tests, one obtains the
WÖHLER18 curve (1862). For large stress amplitudes, it forms almost a straight
line, while it approaches a horizontal line for small amplitudes, which
characterises the regime of endurance strength.
to
m
tu t
18
August Wöhler (1819-1914)
Nowadays one often prefers to perform (strain-controlled) LCF tests (low cycle
fatigue). Here the sample is submitted to a cyclic strain process around a middle
strain m which may consist of intervals of linear strain changes and holding
times with constant strains, as can be seen in the above diagram.
The plot of the measured stresses gives in the initial (unfatigued) regime a
typical hysteresis curve. If fatigue occurs, this hysteresis curve tends to flatten in
its tensile part.
One measures the axial force necessary for such a strain process, by which one
can determine the nominal stress after (1.1.11). After a short shakedown process
of few cycles, the stress will cycle between two extremes in the tensile and
pressure regime, the absolute values of which are initially almost equal. If the
sample becomes finally fatigued, one will observe a decrease of the curve in the
tensile regime, which starts slowly but becomes more and more pronounced close
to final failure. In contrast to the tensile regime, the extreme stresses in the
compression regime will remain almost constant, since the micro-cracks will tend
to close under compression and thus will not be visible.
The prediction of fatigue failure is rather difficult in reality, since it not only
depends on the material, but also on
 the holding times, the middle strains, the form of the load ramps etc.
 the form of the cross-section of the sample
 the treatment of the surface of the sample
 the environment like chemical influence, humidity, temperature, etc.
and much more.

max
N n, t
min
Typical extremal stress curve vs. number of cycles n

in an LCF test
For these reasons, fatigue prediction is still rather difficult and highly empirical.
A historical example of a fatigue rule is due to MANSON (1953) and COFFIN

(1954)
 =  (2N) –  ,  > 0 constants
wherein  is the amplitude of the inelastic deformation.
The treatment of this highly important but rather complex topic is the subject of
fatigue analysis.
1.4 Plasticity
Literature
Backhaus, G.: Deformationsgesetze. Akademie-Verlag, Berlin (1983)

Besson, J.; Cailletaud; G.; Chaboche, J.-L.; Forest, S. ; Blétry, M.: Non-
Linear Mechanics of Materials. Springer, Dordrecht (2010)
Betten, J.: Kontinuumsmechanik. Springer, Berlin (1993), 2nd edt. (2001)
Burth, K.; Brocks, W.: Plastizität. Vieweg, Braunschweig/ Wiesbaden (1992)
Chen, W. F.; Han, D. J.: Plasticity for Structural Engineers. Springer, New
York, Berlin (1988)
Francois, D.; Pineau, A.; Zaoui, A.: Mechanical Behaviour of Materials, vol. I,
Kreißig, R.: Einführung in die Plastizitätstheorie. Fachbuchverlag, Leipzig, Köln
(1992)
Lubliner, J.: Plasticity Theory. Macmillan, New York (1990)
Mang, H.; Hofstetter, G.: Festigkeitslehre. Springer (2000)
Negahban, M.: The Mechanical and Thermodynamical Theory of Plasticity. CRC
Press, Boca Raton (2012)
Yu, M.-H.: Generalized Plasticity. Springer, Berlin (2006)
1.4.1 Rigid-Plastic Models

A completely different behaviour than that of the viscoelastic models, is
performed by the COULOMB19 (ST. VENANT) element (a block with dry
friction).
Here the rate of strain equals zero as long as the absolute value of the stress is
below the yield stress
19
Charles Augustin Coulomb (1736-1806)
1.4 Plasticity 35
(1.4.1)   < Y   = 0 .
When reaching the yield limit, yielding of any amount in the direction of the
applied stress can occur
 = + Y    0 arbitrary

 = – Y    0 arbitrary

Upon unloading, we have immediately again  = 0 and the block is stopped.
COULOMB element

Y
Y
During yielding the entire power will be dissipated in a rate-independent way

(1.4.2) πi =    = Y   =   0 .
Such behaviour is called rigid-perfect plastic. There is no functional relation
between the current stress and strain. The behaviour is piecewise linear and thus
non-linear. Additionally, it is rate-independent (scleronomous).
For real materials, the yield limit is not constant but evolves during plastic
deformations. If the yield stress increases under straining, we talk about
hardening, and if it eventually decreases, it is softening. For many materials, one
observes first a hardening behaviour until some saturation strain, and then
softening until rupture. Under softening, the material becomes unstable and is
prone to localisations like shear banding or necking, so that the measurement of
the stress-strain relation becomes rather complicated in the softening regime.
Since the transformation from hardening to softening behaviour often takes place
at large deformations, the choice of the strain and stress measures becomes
important.
The determination of the ultimate stiffness also becomes difficult. Therefore it is
customary to limit the considerations to the hardening regime.
Suggested hardening rules are
 the linear hardening law

Y = K   + Yo K positive constant
 or the potential hardening law
Y = K  n + Yo K, n positive constants
If one wants to include load reversals so that yielding in the other direction may
also occur, it is better to use incremental laws like
(1.4.3) Y = K   K positive constant
shear banding
Scherbandbildung necking
Einschnürung
If one does not allow for infinite hardening, one can use rules with saturation like
Y = K (Y – Y)   K , Y positive constants
for which Y is limited from above by the saturation value Y .
Another hardening rule with saturation has been suggested by VOCE (1955) in the
form
Y =  – ( – 0) exp(– /0)
with constants  , 0 , and 0 .
Upon reversal of the load from pressure to tension, or vice versa, one can often
observe that the difference between the yield stress in both regimes becomes
smaller during plastic deformations, which is called the BAUSCHINGER20 effect
(1886).
20
Johann Bauschinger (1834-1893)
1.4 Plasticity 37
Y

Y
Hardening with saturation
Y
2Y
BAUSCHINGER effect
In the literature two types of hardening can be found, the difference of which can
only be observed under load reversal:
1. isotropic hardening, for which the yield limit under tension and compression
increases in parallel; and
2. kinematic hardening, for which the yield limit under tension increases by the
same amount as it increases under compression.
In reality the softening behaviour often lies between these two extremes, which
can be described by a combination of the two hardening models. For that purpose
we need two hardening variables, namely one for the isotropic hardening Y and
a back stress B for kinematic hardening. For both variables evolution equations
are needed.
Yo Y > Yo Y Y
B Y
 
Y
isotropic hardening kinematic hardening
1.4.2 Elastic-Plastic Models
PRANDTL model
E Y

Y
E E

Y
Many real materials behave initially elastically as long as the deformations are
below the yield limit. Only upon reaching the yield limit Y (which is also called
the elastic limit), does the material deform plastically (plastic yielding). This
1.4 Plasticity 39
behaviour can be modelled by an arrangement in series of a COULOMB element

and a HOOKE element, which is called a PRANDTL21 or JENKIN model. We can
decompose the total strain into an elastic and a plastic part
(1.4.4)  = e + p with e =  / E .
In the hardening rules only the plastic deformation p appears. The stress power
of the PRANDTL model is
(1.4.5) πi =    =  (e + p) = E e e + Y p = w + 
being composed of an elastic and a dissipative part.
One uses the label materials with elastic ranges, which particularly applies to
many metals at low temperatures. In contrast, at high temperatures (near to the
melting temperature) viscous effects can be important, which is the subject of
viscoplasticity.
An elastoplastic model with kinematic hardening is the MASING model.
MASING model
K Y

Y E
E+K E+K

Y
In order to estimate the size of the elastic range, we give the following example:
21
Ludwig Prandtl (1875-1953)
material: Chrome-Nickel steel

YOUNG´s modulus: 208 GPa
yield stress: 120 MPa
 yield strain: 0.058 %
So for metals the elastic ranges in the strain space are often rather small, which
justifies the use of linear elastic laws.
If one considers elastoplastic material behaviour under monotonous loading
without reversal of the load, one cannot distinguish its behaviour from a non-
linear elastic behaviour. This fact inspired HENCKY (1924) to suggest a finite
deformation rule which allows for treating the material behaviour in the context of
(non-linear) elasticity as long as no load reversal occurs. A similar approach is the
RAMBERG-OSGOOD law (1945)
N
  
   k2   k1 , k2 , k3 , N constants
k1  k3 
Other historical suggestions in this field are
 = 0 + K  N LUDWIK (1909)
 = Y tanh (E  /Y ) PRAGER (1938)
 = KN HOLLOMAN (1944)
 = K (0 + ) N
SWIFT (1947)
(see KHAN/ HUANG p. 11).
Because of the strong restrictions to the range of application, the deformation
method of plasticity has only gained limited consideration.
Besides its non-linearity, the rate-independency is characteristic of all
(elasto)plastic models, which excludes creep and relaxation. In viscoplasticity this
constraint is relaxed, in order to combine plastic and viscous behaviour.
1.5 Viscoplasticity
Literature
Betten, J.: Kontinuumsmechanik. Springer, Berlin (1993), 2. ext. edt. (2001)
Christescu, N.; Suliciu, I.: Viscoplasticity. M. Nijhoff Publ., Den Haag (1982)
Lemaitre, J.; Chaboche, J.-L.: Mechanics of Solid Materials. Cambridge Univ.
Press (1990)
1.5 Viscoplasticity 41
Miller, A. K.: Unified Constitutive Equations for Creep and Plasticity. Elsevier,
London (1987)
A simple method to include viscous effects is by adding viscous elements like

dampers in series (V1) or in parallel (V2) (or both) to plastic elements.
V2
V1
K
Y
The effect of V1 is that there are no more elastic ranges but rather viscoelastic
ones. Under deformation cycles even below the yield limit one will observe a
hysteresis (like for the MAXWELL model).
D
BINGHAM model
K
Y
D K
SCHWEDOFF model
C
Y
Even more important is the application of a viscous element V2 in parallel with

the COULOMB model (like the BINGHAM22 model). V2 is only activated under
inelastic deformations. Therefore, elastic ranges may still exist. During yielding,
the stresses are composed by the yield stress Y and a viscous overstress o
22
Eugene Cook Bingham (1878-1945)
 = Y + o .
As a consequence, the stresses can be arbitrarily large or small, which is perhaps
more realistic than the restriction to Y for a perfect plastic material. If one
combines such models with complex hardening mechanisms, one can describe
rather complex material behaviour, which is particularly needed in the high-
temperature regime and the subject of intensive research activity. The following
models have been suggested in this field:
 the PERZYNA model (1963)
 the BODNER-PARTOM model (1972)
 the HART-MILLER model (1976)
 the CHABOCHE model (1977)
 the ROBINSON model (1978)
 the overstress-model by CHOI/ KREMPL (1985)
 the EURATOM model by BRUHNS (1987)
among others.
Common to all such suggestions is the intention to describe the material behaviour
under essentially all kinds of loadings (creep, cyclic loads with or without holding
times, relaxation, etc.) as realistically as possible.
Due to the actuality of research activities in viscoplasticity, the number of
overview textbooks is still rather limited, in contrast to those on plasticity.
2 INTRODUCTION TO TENSOR CALCULUS
In the first part we have considered one-dimensional models to describe material

behavior. The application of such models, however, is rather limited, since in the
general case, both the stress and the strain state are three-dimensional. For the
generalization of such concepts to higher dimensions than one, an appropriate
mathematical instrumentarium is needed. In mechanics, the appropriate objects for
the mathematical description are vectors and tensors. For readers not familiar with
such objects, we will next give a brief introduction into tensor algebra and
analysis.
Literature
Akivis, M. A.; Goldberg, V. V.: Tensor Calculus with Applications. World
Scientific Pub., Singapore (2003)
Betten, J.: Tensorrechnung für Ingenieure. Teubner, Stuttgart (1987)
De Boer, R.: Vektor- and Tensorrechnung für Ingenieure. Springer, Berlin (1982)
Duschek, A.; Hochrainer, A.: Tensorrechnung in analytischer Darstellung.
Springer, Vienna (1961)
Iben, H. K.: Tensorrechnung. Teubner. Stuttgart, Leipzig (1995)
Hackbusch, W.: Tensor Spaces and Numerical Tensor Calculus. Springer Series
in Computational Mathematics. Springer, Berlin (2012)
Itskov, M.: Tensor Algebra and Tensor Analysis for Engineers. Springer, Berlin
(2007)
Lagally, M.: Vorlesungen über Vektor-Rechnung. Akad. Verlagsgesellschaft,
Leipzig (1928)
Lebedev, L. P.; Cloud, M. J.: Tensor Analysis. World Scientific, New Jersey
(2003)
Marsden, J., Tromba, A. J.: Vektoranalysis. Spektrum Akad. Verlag,
Heidelberg, Berlin, Oxford (1995)
Ruiz-Tolosa, J. R.; Castillo, E.: From Vectors to Tensors. Spinger, Berlin (2005)
Schade, H.: Tensoranalysis. de Gruyter, Berlin (1997), 2. Aufl. (2006) together
with K. Neemann.
Shouten, J. A.: Tensor Analysis for Physicists. Dover Pub., New York (1990)
Trostel, R.: Mathematische Grundlagen der Technischen Mechanik I, Vektor- and
Tensoralgebra. Vieweg, Braunschweig, Wiesbaden (1993)
Trostel, R.: Mathematische Grundlagen der Technischen Mechanik II, Vektor-
and Tensoranalysis. Vieweg, Braunschweig, Wiesbaden (1997)
Many textbooks on continuum mechanics contain introductions to tensors, like:
Gonzalez, O.; Stuart, A. M.: A First Course in Continuum Mechanics.

Cambridge University Press (2008)
Irgens, F.: Continuum Mechanics. Springer (2008)
Kienzler, R., Schröder, R.: Einführung in die Festigkeitslehre. Springer,
Dordrecht (2009)

DOI 10.1007/978-3-319-19566-7_2
44 2 Introduction to Tensor Calculus
Lai, W. M., Rubin, D., Krempl, E.: Introduction to Continuum Mechanics.

Pergamon Press, Oxford (1978, 1993)
Parisch, H.: Festkörper-Kontinuumsmechanik: Von den Grundgleichungen zur
Lösung mit Finiten Elementen. Teubner (2003)
Slaughter, W. S.: The Linearized Theory of Elasticity. Birkhäuser, Boston (2002)
2.1 Vector and Tensor Algebra

2.1.1 Summation Convention
Let us consider the following sum
N
s = a1 b1 + a2 b2 +  + aN bN =  ai bi .
i 1
We could also have written

N
s =  ak bk ,
k 1
i.e., the choice of the summation index (i or k) does not influence the result and is
therefore called a dummy index. Since we often need sums, we introduce the
following abbreviation:
Summation convention
If an index appears twice in a product term, one has to sum over this index from 1
to N .
Or: one has to sum over dummy indices.
The number N results from the context. In what follows, it is usually 3 , equal to
the dimension of the geometrical space.
Consequently, we can write for the previous example
s = ai bi = ak bk etc.
We will need the component representation of a vector v with respect to some
vector basis {gi}
v = v 1 g1 + v 2 g2 + v 3 g3 = v i gi .
If aij are the elements of a 3  3–matrix, the sum of the diagonal elements
aii = a11 + a22 + a33
is called the trace of the matrix. The product of such matrices with elements aij
and bij is
aij bjk = ail blk = ai1 b1k + ai2 b2k + ai3 b3k .
Note that only j and l are dummy indices, while i and k are not. i and k are
called free indices, taking arbitrary values between 1 and N. Free indices stand on
both sides of the equation only once in each term.
The trace of the resulting matrix aik bki is the double sum over both i and k , and
the order of the summation does not matter
N N N N
  aik bki =   aik bki = aik bki .
i 1 k 1 k 1 i 1
It is always important to distinguish between dummy and free indices. We will
show this for another example. One can present the three equations
y1 = a11 x1 + a12 x2 + a13 x3
y2 = a21 x1 + a22 x2 + a23 x3
y3 = a31 x1 + a32 x2 + a33 x3
more briefly as
yi = aik xk
or equivalently as
yp = apm xm .
Here k and m are dummy, while i and p are free indices.
Very often a situation occurs where some term equals 1 if two indices coincide,
or 0 otherwise, like in the following example
a1 b1 = 1 a1 b2 = 0 a1 b3 = 0
a2 b1 = 0 a2 b2 = 1 a2 b3 = 0
a3 b1 = 0 a3 b 2 = 0 a3 b3 = 1 .
In order to abbreviate this notation, one introduces the KRONECKER23 symbol
 ij as the components of the unity matrix
 11  12  13  1 0 0 
  0 1 0  ,
  
 21 22 23    
 31  32  33  0 0 1
i.e.
1 for i  j
(2.1.1)  ij =   =  ji =  i =  i .
j j
0 for i  j
23
Leopold Kronecker (1828-1891)
It shall not matter here whether the indices are notated in the upper or in the lower
position. For the above 9 equations we can briefly write
ai bj =  ij .
In reverse, one can use the extensions
aik xk = ail  lk xk
and
ai bi = ai bk  ik
as well as
aik xk –  xi = aik xk –   ik xk = (aik –   ik) xk .
In addition, we will introduce in three dimensions the permutation symbol or
LEVI-CIVITA24 symbol ijk with three indices as
 1 if ijk is an even permutation of 1, 2, 3

(2.1.2) ijk =  1 if ijk is an odd permutation of 1, 2, 3
 0 if ijk is no permutation von 1, 2, 3

such as
231 = + 1 132 = – 1 122 = 0
Consequently
ijk = kij = – ikj = – kji
etc.
2.1.2 Vectors
Definition. A vector space or linear space is a set V (of “vectors“), in which

addition and multiplication with scalars or real numbers are defined in accordance
with the following rules
a+b = b+a (commutative)
(a + b) + c = a + (b + c) (associative)
a+o = a (zero vector)
a + (–a) = o (negative element)
(  ) a =  ( a) (associative)
24
Tullio Levi-Civita (1873-1941)
1a = a (unit element)
 (a + b) =  a +  b (distributive)
( +  ) a =  a +  a (distributive)
 a , b , c  V ,  ,   R .
A maximal system of linear independent vectors {gi} : = {g1 , g2 , ..., gN} forms a
vector basis in V . With respect to such a basis one can represent each vector as a
linear combination
a = ai g i .
The scalars ai are the components of the vector with respect to this basis. The
following rules hold for the addition of two vectors
(2.1.3) a + b = ai g i + b j g j = (ai + bi) g i
and the scalar multiplication
(2.1.4)  a =  (ai g i) = ( ai) g i .
Some vector spaces have in addition an inner product or a scalar product
 : V  V  R  (a , b)  a  b
with the following rules
ab = ba (commutative)
( a)  b =  (a  b) (associative)
(a + b)  c = (a  c) + (b  c) (distributive)
a  a > 0 for a  o (positive–definite)
 a , b , c  V ,   R .
With such a scalar product we obtain
(2.1.5) a  b = (ai gi)  (bk gk) = ai bk gi  gk .
A scalar product induces a norm a: = a  a (length) of a vector. One can
also define an angle  between two vectors as a solution of the equation
a  b = ab cos  .
If the vectors a and b are mutually orthogonal we have
ab = 0.
Definition. Two vector bases {gi} and {gk} are called dual if
(2.1.6) g i  g k = ik
Consequently, all base vectors of {g i} are orthogonal to those of {g k} for k  i.
For a given basis {g i} , (2.1.6) is an inhomogeneous system of linear equations
for the components of the dual basis {gk} , whose coefficient determinant cannot
be singular, so that it always possesses a unique solution.
Theorem. For every vector basis {g i} a unique dual basis {g j} exists.
The use of dual bases is always possible, and in many cases rather convenient. For
example, for the inner product we have
a  b = (ai g i)  (bk g k) = ai bk g i  g k
(2.1.7) = ai bk ik = ai bi = ak bk .
Here we posed the dummy indices in counterpositions in order to indicate to
which basis they are referred. We also have
a  b = (ai g i )  (bk g k) = ai bi ,
but
a  b = (ai g i )  (bk g k ) = ai bk (g i  g k ) .
If a basis {ei} coincides with its dual {e j}
ei  ei for i = 1, 2, 3,
there is no need to distinguish between upper and lower indices anymore.
Definition. A vector basis {ei} is called an orthonormal basis (ONB) if
(2.1.8) ei  ek =  ik .
In each vector space with inner product there are infinitely many vector bases or
ONBs. If one uses simultaneously more than one basis, it becomes necessary to
indicate to which basis the components are referred, like
a = ai ei = ai ei
with
ai = a  ei and ai  a  ei .
With respect to an ONB we have simply
(2.1.9) ek  ei =  ki
(2.1.10) a  b = ai bi
(2.1.11) a = ai ai  a12  a22  a32 .
In three dimensions one can further introduce the vector product or cross-
product with respect to a positively-oriented ONB with the aid of the permutation
symbol by
(2.1.12) a  b = ai ei  bj ej = ai bj ijk ek
and the triple product between three vectors as
(2.1.13) [a , b , c] = [ai ei , bk ek , cl el] = ai bk cl ikl .
2.1.3 Dyads and Tensors
Definition. A mapping between vectors

F:V  V  x  F(x)
is called linear if
(2.1.14) F(x1 +  x2) = F(x1) +  F(x2)
holds for all vectors x1 , x2  V and all scalars   R .
Problem 1. Linearity
Check the real function
y = f (x) : = m x + n
for linearity. m and n are real constants.

Solution
The above-introduced definition of linearity can be written for a real
function f as
(P1.1) f (α a + b) = α f (a) + f (b) α,a,bR
The condition P1.1 gives
αn = 0.
In general, this holds only for n  0. The concept of linearity used in

algebra is stronger than the one used in real analysis or calculus. In the
latter context such functions are sometimes called affine or quasilinear,
whereas the present linearity is called proportionality in the context of
calculus.
Let a and b be arbitrarily chosen, but fixed vectors. With their aid, one can
construct a special linear mapping that assigns to each vector x another vector y
after
y = a (b  x) = (x  b) a = (b  x) a .
Thus, the resulting vector y is always parallel to a .
Definition. The dyadic product (tensor product) between two vectors a and b
or the simple or collinear dyad a  b is the mapping
a  b : V  V  x  a (b  x)
so that
(2.1.15) a  b (x) : = a (b  x)
This mapping is linear since
a  b (x1 +  x2) = a [b  (x1 +  x2)]
= a [b  x1 +  b  x2]
= a  b (x1) +  [a  b (x2)]
holds for all vectors x1 , x2  V and all scalars   R .
The dyadic product is in general not commutative since a  b  b  a as long as
both vectors do not happen to be collinear (parallel).
One can define the sum of two dyads as that particular linear mapping that gives
for all vectors x
(2.1.16) (a  b + c  d) (x) : = a  b (x) + c  d (x)
= a (b  x) + c (d  x)
and a multiplication of a dyad a  b with a scalar   R as
(2.1.17) ( a  b) (x) : =  [(a  b) (x)] =  a (b  x) .
These operations fulfil the axioms of the addition and multiplication with a scalar
of vectorspaces. Moreover, the following rules hold for all vectors a , b , c  V
and all scalars   R
(2.1.18) (a + b)  c = a  c + b  c
(2.1.19) a  (b + c) = a  b + a  c
(2.1.20)  (a  b) = ( a)  b = a  ( b)
Accordingly, we can drop the brackets in the last line. The dyadic product is thus
linear in both involved vectors. It follows that
(2.1.21) (a +  b)  (c +  d) = a  c +  a  d +  b  c +   b  d .
If we represent a and b with respect to a basis {gi} , we obtain
a  b = (ai g i)  (bk g k)
(2.1.22) = ai bk g i  g k
and
(2.1.23) (a +  b)  c = (ai +  bi ) c k g i  g k
and
(2.1.24) a  (b +  c) = ai (bk +  c k ) g i  g k .
For an ONB, these expressions do not become shorter. Only if we are evaluating
scalar products, is the use of an ONB or the use of dual bases recommended.
Because of the linearity we have with respect to an ONB {ei}
a  b (x) = [(ai ei)  (bk ek)] (xl el)
= xl ai bk (ei  ek) (el)
(2.1.25) = ai bk xl ei (ek  el)
= ai bk xl ei  kl
= ai bk xk ei .
Since the simple dyad a  b maps all vectors x in the direction of a , it is a
special linear mapping, called a collinear dyad.
The sum of two dyads (a  b) + (c  d) maps all vectors x into a linear com-
bination of a and c , i.e., into a plane spanned by a and c . Therefore, one calls
such a mapping a planar dyad.
Definition. The general linear mapping of a vector into a vector is called a tensor
or a (complete) dyad.
If T is such a tensor, we have
(2.1.26) y = T(x) = T(xi ei) = xi T(ei) .
A tensor is therefore completely determined if its action on every base vector is
given. Since the brackets are not needed, we will no longer use them in what
follows and instead use a dot which stands for the scalar product in the definition
of the dyad
(2.1.27) y = Tx.
Since T  ei is a vector, we can represent it with respect to an ONB {ek} as
T  ei = Tki ek
with the components
Tki : = ek  (T  ei) .
In this expression the brackets are not needed since there is no danger of
confusion. Thus
T  x = T  (xi ei) = xi T  ei = xi Tki ek
= xi Tkm ek  em  ei
= (Tkm ek  em)  x .
By comparison one obtains
Theorem. Each tensor T can be uniquely represented with respect to an ONB

{ei} as
(2.1.28) T = Tki ek  ei
with the nine (base) dyads
e1  e1 e1  e2 e1  e3
e2  e1 e2  e2 e2  e3
e3  e1 e3  e2 e3  e3
and nine tensor components
(2.1.29) Tki : = ek  T  ei for i, k = 1, 2, 3 .
{ek  ei} forms a tensor basis. The nine tensor components with respect to this
basis can be assembled in the matrix of components
T11 T12 T13 
[Tij] : = T T T 
 21 22 23  .
T31 T32 T33 
As a special case, a simple dyad gives

a  b = ai bk ei  ek
and therefore for the matrix of components
 a1 b1 a1 b2 a1 b3 
a b a2 b2 a2 b3  .
 2 1
 a3 b1 a3 b2 a3 b3 
The dyadic product between two vectors can be performed by a matrix product in
the FALK´s scheme
b1 b2 b3
a1 a1 b1 a1 b2 a1 b3
a2 a2 b1 a2 b2 a2 b3
a3 a3 b1 a3 b2 a3 b3
This is the matrix product between the column vectors of a and the row vectors
of b .
Note that the components of a tensor depend on the choice of the basis, the same
as vector components. In fact, if {ei} and {ei} are two bases, then
T = Tik ei  ek = Tik ei  ek
with
Tik = ei  T  ek
Tik = ei  T  ek ,
so that Tik and Tik are in general not equal. However, the tensor T itself is
independent of the basis with respect to which it is represented.
If one wants to determine the value y of a vector x under the mapping of a
tensor, one represents the tensor as before and also x = xn en and obtains
y = Tx
= yk ek = (Tki ek  ei)  (xn en)
= Tki xn ek (ei  en)
(2.1.30) = Tki xn ek in
= Tki xi ek
which gives the equation for the components
yk = Tki xi for k = 1, 2, 3.
In matrix form this is
 y1  T11 T12 T13   x1 
     
y
 2  T21 T22 T23   x2  .
 y3  T31 T32 T33   x3 
Accordingly, one can reduce the tensor operation to matrix operations after
choosing a basis. This holds also for the sum of two tensors
S = Ski ek  ei
T = Tki ek  ei
as
S + T = Ski ek  ei + Tki ek  ei
(2.1.31) = (Ski + Tki) ek  ei
and the multiplication of a tensor T with a scalar 
 T =  (Tki ek  ei) = ( Tki) ek  ei
which are defined in analogy to the same products between dyads after (2.1.16)
and (2.1.17).
If S and T are two tensors and x a vector, then S  x is a vector, upon which
we can apply T
T  (S  x) .
Since the composition of linear mappings is again linear, T  S stands for
another tensor after
T  (S  x) : = (T  S)  x .
Its components can be obtained by the following calculation
T  (S  x) = (Tpi ep  ei)  [(Skl ek  el)  (xm em)]
= (Tpi ep  ei)  [(Skl xm ek) (el  em)]

= (Tpi ep  ei)  [Skl xm ek lm]
= (Tpi ep  ei)  [Skm xm ek]
= Tpi Skm xm ep (ei  ek)
= Tpi Skm xm ep ik
= Tpi Sim xm ep
= Tpi Siq qm xm ep
= (Tpi Siq ep  eq)  (xm em) .
Here all brackets are unnecessary and are only used to emphasise the connections.
By comparison we obtain the representation of the composed tensor
(2.1.32) T  S = (Tpi ep  ei)  (Skl ek  el) = Tpi Skl ik ep  el
= Tpi Sil ep  el
i.e., one contracts the neighbouring base vectors by a scalar product, and the
remaining ones by a dyadic product. Its component matrix is the result of a matrix
product between the two component matrices. This operation is called a simple
contraction.
It is important to note that this product between two tensors is in general not
commutative, i.e., T  S does not equal S  T. It is associative
(2.1.33) (T  S)  R = T  (S  R)
so that we can drop the brackets. And it is linear in both factors
(T +  S)  R = T  R +  S  R
R  (T +  S) = R  T +  R  S .
The particular tensor that maps every vector into itself is the unit or identity
tensor. With respect to any ONB {ei} this tensor has the unity matrix
representing the coefficients after (2.1.1)
(2.1.34) I = ik ei  ek = ei  ei
since we have for every vector x = xm em
Ix = (ik ei  ek)  (xm em)
= ik xm ei (ek  em)
= xm ek km = xm em = x .
If T = Tki ek  ei is an arbitrary tensor and I the identity tensor, then we have
T  I = (Tki ek  ei)  (lm el  em)
= (Tki ek  ei)  (el  el)
= Tki ek  el (ei  el)
= Tki ek  el il
= Tkl ek  el
= T
the same as the other way round
(2.1.35) I T = T
for all tensors T.
Scalar multiples  I of I are called spherical tensors (or isotropic tensors).
They multiply each vector
(2.1.36) Iv = v
keeping its direction constant. In particular, 0 : = 0 I is the zero tensor, which
maps every vector into the zero vector. Its coefficients are all zero with respect to
whatever basis. For all tensors T we obtain
(2.1.37) 0 T = T 0 = 0.
2.1.4 The Inverse of a Tensor
One may ask the question of whether a tensor possesses an inversion, i.e., if
there exists for a tensor T an inverse tensor T –1 such that
T –1  (T  x) = x
holds for all vectors x . If such a mapping exists, then it must also be linear (and
therefore also a tensor, notated as T –1 ). This is equivalent to
(2.1.38) T –1  T = I .
In components with respect to an ONB this gives
(T –1
kl ek  el)  (Tmn em  en) = kn ek  en
–1
= T kl el  em Tmn ek  en
–1
= T km Tmn ek  en
(2.1.39)  T –1
km Tmn = kn .
Accordingly, the matrix of components [T –1ik] of the inverse tensor T –1 is the
inverse matrix of [Tmn] in the sense of matrix algebra if for both tensors the same
ONB is used.
We know from matrices that only the non-singular matrices are invertible.
These are characterized by the property that their determinant is non-zero
det[Tin]  0 .
A tensor is in fact invertible if and only if the determinant of its matrix of
components is non-zero with respect to an arbitrary basis (and hence for all bases).
Otherwise it is called singular. We will later define the determinant of a tensor,

which will lead to an invertibility rule which is independent of the used basis.
A collinear or complanar dyad is always singular.
A spherical tensor  I is invertible if and only if   0 . Its inverse is then –1 I .
If S and T are both invertible tensors, then the composition S  T is also
invertible, and vice versa, and we obtain
(2.1.40) (S  T) –1 = T –1  S–1
since
(S  T) –1  (S  T) = T –1  S–1  S  T = T –1  T = I .
2.1.5 The Transpose of a Tensor
For every tensor T the transposed tensor TT is defined through the bilinear
form
(2.1.41) a  (TT  b) = b  (T  a)
for arbitrary vectors a and b . It is sufficient to postulate this for two arbitrary
base vectors
ek  (T  ei) = ei  (TT  ek)
or
(2.1.42) Tki = (T T )ik i, k = 1, 2, 3
which means that with respect to an ONB the matrix of the components of the
transposed tensor equals the transposed matrix of the original tensor
T = Tik ei  ek  TT = Tik ek  ei
= Tki ek  ei = Tki ei  ek .
The following rules hold for all tensors T and S , all vectors a and b , and all
scalars  :
(2.1.43) (a  b)T = (b  a)
(2.1.44) (T  S)T = ST  TT
(2.1.45) (T + S)T = TT + ST
(2.1.46) ( T)T =  (TT)
(2.1.47) (TT )T = T
(2.1.48) (TT ) –1 = (T –1)T = : T –T for invertible tensors T
IT = I
0T = 0
One can also define a left-product between a vector and tensor

v  T : = TT  v
(2.1.49) = (Tkm em  ek)  (vi ei) = Tkm vi em (ek  ei) = Tkm vi em ki
= vi Tim em = (vi ei)  (Tmk em  ek) = vi Tmk (ei  em) ek
= vi Tmk im ek = vi Tik ek .
For all vectors v and w we have then
(2.1.50) v  (T  w) = (v  T)  w
so that the brackets are not needed. Since we can express the left-product through
the usual product from the right, it does not show any new properties.
If a tensor equals its transposed tensor we call it symmetric. As an example,
every spherical tensor is symmetric. The definition of the symmetry coincides
with the symmetry of the matrix of components with respect to an ONB.
If a tensor equals its negative transposed tensor
(2.1.51) T = – TT
it is called anti(sym)metric or skew. Consequently, its components with respect
to an ONB obey
(2.1.52) Tik = – Tki
and in particular (no summation)
Tii = – Tii = 0 .
While a symmetric tensor possesses six independent components
T11 T12 T13 
 
T12 T22 T23 
T13 T23 T33 
a skew tensor has only three

 0 T12 T13 
 
 T12 0 T23  .
 T13  T23 0 
So a skew tensor has the same DOFs as the underlying vector space. This gives
rise to the supposition that its effect on some arbitrary vector x can also be
obtained by an appropriate operation of some vector with x . And in fact we find
for every skew tensor T a unique axial vector tA such that for all vectors x
(2.1.53) T  x = tA  x
holds. For determining tA we choose an ONB {ei} and obtain
(Tik ei  ek)  (xl el) = t Am em  xl el
= Til xl ei = t Am xl mli ei
so that
(2.1.54) Til = t Am mli
or
T12 = t A1 121 + t A2 221 + t A3 321 = – t A3
T23 = t A1 132 + t A2 232 + t A 3 332 = – t A1
T31 = t A1 113 + t A2 213 + t A3 313 = – t A2 .
Therefore
tA = – T23 e1 – T31 e2 – T12 e3
= + T32 e1 + T13 e2 + T21 e3
and
(2.1.55) t Am = ½ Til mli .
We obtain the following rules which can be easily proven.

 Any linear combination of (anti)symmetric tensors is again (anti)symmetric.
The composition of (anti)symmetric tensors, however, may lose this property.
 The inverse of symmetric invertible tensors is also symmetric.
 Skew tensors are singular.
One can uniquely decompose any tensor T into its symmetric part
(2.1.56) sym(T) : = ½ (T + TT)
and its skew part
(2.1.57) skw (T) : = ½ (T – TT )
so that
T = sym(T) + skw (T) .
In particular, we obtain for collinear dyads the symmetric part
sym(a  b) = ½ (a  b + b  a)
and the skew part
skw(a  b) = ½ (a  b – b  a)
with the corresponding axial vector
tA = ½ b  a .
2.1.6 Square Forms and Tensor Surfaces
If x is a vector and T a tensor, then x  T  x is a scalar. Such an expression is

called a square form. Obviously, in such a form only the symmetric part of T is
relevant.
A tensor T is called
 positive definite if xTx > 0
 positive semidefinite if xTx  0
 negative definite if xTx < 0
 negative semidefinite if x  T  x  0
holds for all vectors x  o , and indefinite otherwise.
If some tensor T is positive (semi)definite, then –T is negative (semi)definite.
An example for a positive definite tensor is the identity tensor or a spherical tensor
 I with some positive scalar  .
A geometrical characterization of tensors can be obtained by their tensor
surfaces which are defined in the following way: We consider all vectors x that
fulfil the quadratic equation
xTx = 1.
If we interpret the solution vectors as position vectors, then they describe a two-
dimensional subset of the three-dimensional space, which is a surface. For a
spherical tensor T , this surface forms a sphere. If T is positive definite, then this
surface is an ellipsoid. Other tensor surfaces will be considered after the treatment
of eigenvalue problems.
2.1.7 Cross-Product between Vectors and Tensors
For some applications one needs the cross-product between a vector (from the
left) and a tensor (from the right)
vT
which is defined by its action on an arbitrary vector w as
(2.1.58) (v  T)  w : = v  (T  w)
so that the brackets are not needed. With respect to an ONB this gives
v  T  w = vi ei  (Tkl ek  el)  (wm em)
= vi ei  Tkm wm ek
= vi Tkm wm ei  ek
= vi Tkm wm ikp ep
= vi Tkl (ei  ek)  el  (wm em) .
Accordingly
v  T = (vi ei)  (Tkl ek  el)
= (vi ei  Tkl ek)  el
= vi Tkl (ei  ek)  el
or for linear dyads
v  (a  b) = (v  a)  b .
The resulting tensor is linear in the three involved vectors. The brackets are not
needed.
Its transpose is
(v  a  b)T = b  (v  a)
= – b  (a  v)
= : – (b  a )  v
= – (a  b) T  v
or, in general, for all tensors T
(2.1.59) v  T : = – (TT  v)T .
In this way we have introduced the cross-product between a dyad (from the left)
and a vector (from the right) as
(a  b)  v : = a  (b  v) = – a  (v  b)
= – (v  b  a)T = – [v  (a  b)T]T
and, more generally, between a tensor (from the left) and a vector (from the right)
(2.1.60) T  v : = – (v  TT )T
which is again linear in all factors. For the components with respect to an ONB we
obtain
Tv = (Tik ei  ek)  (vm em)
= Tik vm ei  (ek  em)
= Tik vm kmp ei  ep
where all brackets are again unnecessary.
The following rules hold for all scalars  , vectors v , w and tensors S , T .
(2.1.61) (v  T)  w = v  (T  w) = : v  T  w
(2.1.62) v  (T  w) = (v  T)  w = : v  T  w
(2.1.63) (T + S)  v = T  v + S  v
(2.1.64) v  (T + S) = v  T + v  S
(2.1.65)  (T  v) = ( T)  v = T  ( v) = :  T  v
(2.1.66)  (v  T) = ( v)  T = v  ( T) = :  v  T
(2.1.67) T  (v + w) = T  v + T  w
(2.1.68) (v + w)  T = v  T + w  T
As a consequence of the rules for the triple product, we have additionally
(2.1.69) T  (a  b) = (T  a)  b
or for the product from the left
(2.1.70) (a  b)  T = a  (b  T) .
Here again all brackets are unnecessary since the operations only make sense in
the given order.
If we choose in particular for T the identity I , then we obtain for arbitrary
vectors w
(v  I)  w = v  (I  w) = v  w
= I  (v  w) = (I  v)  w = (I  v)  w
and therefore
(2.1.71) vI = Iv.
On the other hand, we obtain with (2.1.60)
(2.1.72) I  v = – (v  I)T = – (I  v)T
so that I  v = v  I must be antisymmetric. By comparison we conclude that v
is the axial vector of the skew tensor v  I .
We obtain with respect to an ONB
I  v = ei  ei  vj ej
= vj ijk ei  ek
(2.1.73) = v  I = vj ej  ei  ei
= vj jik ek  ei .
If this tensor is applied to a vector w we obtain
v  I  w = vj jik ek (ei  wm em)
= vj wi jik ek .
2.1.8 Orthogonal Tensors
are those tensors which are compatible with the inner product such that
a  b = (Q  a)  (Q  b) = a  QT  Q  b
holds for arbitrary vectors a and b . This leads to
QT  Q = I  Qmi Qmj = ij
with respect to some ONB, or
(2.1.74) Q–1 = QT.
So for orthogonal tensors, the inverse equals the transpose. All orthogonal tensors
are therefore invertible. We also have
Q  QT = I  Qim Qjm = ij
with respect to some ONB.
Accordingly, the transpose/ inverse of some orthogonal tensors is again
orthogonal. If we represent an orthogonal tensor with respect to some ONB by its
matrix of components, then this is an orthogonal matrix
Q11 Q12 Q13 
 
Q21 Q22 Q23  .
Q31 Q32 Q33 
Such orthogonal matrices have the property that both the row vectors and the
column vectors are normalized and mutually orthogonal.
Orthogonal tensors describe rotations and reflections of vectors. If {ej} is an
ONB, then {Q  ei} is also an ONB for every orthogonal tensor Q . Occasionally
Q may change the orientation of the basis.
Examples for orthogonal tensors are
  e1  e1  e2  e2  e3  e3
  e1  e2  e2  e3  e3  e1
  e1  e1  e2  e3  e3  e2
for any ONB {ej}.
If v is a vector and Q an orthogonal tensor, then Q  v is the rotated and
occasionally reflected vector. With respect to some ONB {ej} we obtain
Q  v = Q  vi ei = vi Q  ei
so that the mapped vector has the same components with respect to the ONB
{Q  ei} as the original one with respect to {ej} .
The same can be done with a tensor T . If we rotate the tensorbasis {ei  ej}
into {(Q  ei)  (Q  ej)} = {(Q  ei)  (ej  QT)} , then we obtain the rotated tensor
(2.1.75) Q  T  QT = Q  Tij ei  ej  QT = Tij (Q  ei)  (Q  ej)
again with the same components with respect to the ONB {(Q  ei)  (Q  ej)} as
the original tensor with respect to an ONB {ei  ej} .
For representing an orthogonal tensor Q we choose a particular ONB, the
e1 direction of which coincides with the rotational axis of Q , so that the matrix
of components of Q is
 1 0 0 
0 cos   sin  
 
 0 + sin  cos  
with some angle  . For +1 the tensor describes a pure rotation. In this case the
tensor is called a versor25 or proper-orthogonal, while for –1 an additional
reflection at the e2e3plane takes place. One sees easily that also in this case the
rows and columns of the matrix are mutually orthogonal.
For the rotation we obtain the following representation
Q = e1  e1 + (e2  e2 + e3  e3) cos  + (e3  e2 – e2  e3) sin 
(2.1.76) = cos  I + (1 – cos ) e1  e1 + I  e1 sin  .
For small rotations one linearises this expression in 
sin   
cos   1
and obtains the more simple representation
Qv
  v
v
(2.1.77) Q  I+I = I+I

with the skew tensor I   , the axial vector of which is  =  e1 . Then
Qv  v+v.
Note that this linearised tensor is not orthogonal anymore.
25
from lat. vertere = to turn
2.1.9 Transformations under Change of Basis
We introduced vectors and tensors without a basis. All operations and properties
of tensors can also be written without referring to a basis (direct or symbolic
notation).
On the other hand, we could see that after choosing a basis, all operations and
properties of tensors could be related to analogous ones on the matrices of the
components. This representation of tensor operations is general because this is
always possible, but also special because one could have chosen any other basis as
well.
Therefore the question arises of how the components of an arbitrary vector v or
tensor T transform under changes of the basis. We will exclusively consider
ONB, as we also did before. So letting {ei} and {ei} be such ONBs, we obtain
the representations
v = vi ei = vi ei
T = Tik ei  ek = Tik ei  ek .
It is always possible to transform one ONB into another ONB by rotations and
occasionally reflections. This can be described by an orthogonal tensor Q . If we
take
Q = ei  ei
then this tensor maps
(2.1.78) Q  ek = ek
and vice versa
QT  ek = ek .
This representation of Q is, however, trivial and not helpful for our purpose to
derive the transformations of the components. For this purpose, we choose another
representation
Q = Qrs er  es
with
Qrs = er  Q  es = er  (ei  ei)  es
= (er  ei) (ei  es) = er  es = cos (er , es) .
The component Qrs is the directional cosine, i.e., the cosine of the angle between
the base vectors er and es . With this representation of Q we obtain
v = vr er = vi ei = vi Q  ei
= vi (Qrs er  es)  ei = vi Qrs er si
= vi Qri er
and by comparing the components

(2.1.79) vr = Qri vi .
In the reverse direction one would obtain
vi = vr Qri .
Analogously one obtains for the components of a tensor
T = Tik ei  ek = Tlm el  em
= Tlm (Q  el)  (Q  em)
= Qil Qkm Tlm ei  ek
the tranformations
(2.1.80) Tik = Qil Tlm Qkm and Tlm = Qil Tik Qkm
or, if written in matrix form,
[Tik] = [Qil] [Tlm] [Qkm]T and [Tlm] = [Qil]T [Tik] [Qkm] .
2.1.10 Eigenvalues and Eigenvectors
In mechanics we are often confronted with the following problem: Find for a
given tensor T vectors that T maps into their own direction, i.e.,
(2.1.81) Ta = a
for some real  . Such a vector is called the eigenvector of T and  the
corresponding eigenvalue of T .
First of all one states that the eigenvalue equation is trivially fulfilled for the
zero vector for arbitrary  . We will therefore only look for non-zero eigenvectors
ao.
If we take the –multiple of some eigenvectors a , then
T  ( a) =  (T  a) =   a =  ( a) .
So every scalar multiple of some eigenvector is also an eigenvector with the same
corresponding eigenvalue. This gives rise to a normalisation
(2.1.82) a = 1 = a  a = ai ai
where the sense of the direction of a still remains arbitrary. It would be more
reasonable to talk about eigendirections instead of eigenvectors.
We can reformulate the above eigenvalue equation as
(2.1.83) ( T –  I)  a = o
or in component form with respect to some ONB {ei} as
(2.1.84) (Tik –  ik) ak = 0 for i = 1, 2, 3.

This gives the equations
(T11 – ) a1 + T12 a2 + T13 a3 = 0
T21 a1 + (T22 – ) a2 + T23 a3 = 0
T31 a1 + T32 a2 + (T33 – ) a3 = 0 .
These are the well-known eigenvalue equations from matrix algebra. It is a
system of three linear and homogeneous equations for the components ai of the
eigenvectors a , which allows for a non-zero solution if and only if the
determinant of the matrix of coefficients is zero
T11   T12 T13 
 
(2.1.85) det [Tik –  ik] = det T21 T22   T23  = 0.
T31 T32 T33   
This leads to a cubic polynomial in 

(T11 – ) [(T22 – )(T33 – ) – T32 T23]
– T12 [T21(T33 – ) – T31 T23]
+ T13 [T21 T32 – T31(T22 – )] = 0
called the characteristic equation of the matrix [Tin] . It can always be brought
into the form
(2.1.86) 3 – IT 2 + IIT  – IIIT = 0
with the three scalar coefficients IT , IIT , IIIT , called the principal invariants of
T. Since the eigenvalue equations (2.1.83) do not depend on a basis, this should
also hold for the characteristic polynomial and, therefore, for the principal
invariants.
The first of these is the trace of the tensor
(2.1.87) IT = tr T = Tii = T11 + T22 + T33 .
The second one is the sum of the minors of the main diagonal
T11 T12  T22 T23  T11 T13 
IIT = det    det    det  
T21 T22  T32 T33  T31 T33 
(2.1.88) = T11 T22 – T12 T21 + T22 T33 – T23 T32 + T11 T33 – T13 T31
= ½ (Tii Tkk – Tik Tki)
= 1/2 {tr2(T) – tr(T2 )} .
The third is the determinant of the tensor
T11 T12 T13 

 
IIIT = det(T) = det T21 T22 T23 
T31 T32 T33 
(2.1.89) = ikm T1i T2k T3m

= T11 T22 T33 – T11 T23 T32 – T12 T21 T33
+ T12 T23 T31 + T13 T21 T32 – T13 T22 T31
= 1/6 tr3(T) – 1/2 tr(T) tr(T2 ) + 1/3 tr(T3 ) .
All of these representations with components hold only if referred to some ONB.
If we rotate the tensor by some versor, then the characteristic polynomial
remains the same
det (Q  A  QT –  I) = det [Q  (A –  I)  QT] = det (A –  I)
and therefore the invariants and the eigenvalues remain the same. Only the
eigenvectors rotate.
For the principal invariants, the following rules hold for all scalars  , vectors a
and b , tensors A and B , and orthogonal tensors Q .
(2.1.90) tr(A +  B) = tr(A) +  tr(B) linearity
T
(2.1.91) tr A = tr(A )
(2.1.92) tr(a  b) = a  b
(2.1.93) tr( I) = 3 in three dimensions
T
(2.1.94) tr(Q A Q ) = tr A
(2.1.95) det(A) = det(AT )

(2.1.96) det( A) =  3 det(A) in three dimensions
(2.1.97) det(A  B) = det(A) det(B) determinant rule
–1 –1
(2.1.98) det (A) = det(A ) for invertible tensors
(2.1.99) det I = 1
(2.1.100) det Q = det(Q –1) =  1 for orthogonal tensors
(2.1.101) det(Q  A  QT) = det A
for orthogonal tensors Q and arbitrary tensors A
A traceless tensor is called a deviator. One can uniquely decompose every
tensor into its deviatoric and its spherical part. The deviatoric part is
(2.1.102) T' : = T – 1/3 tr(T) I .
After the theorem of VIETA26 one can represent every polynomial by its roots
(nulls)
( – 1) ( – 2) ( – 3)
= 3 – (1 + 2 + 3) 2 + (1 2 + 2 3 + 3 1)  – 1 2 3 .
By comparison with the characteristic polynomial, we obtain a representation for
the invariants by the eigenvalues i
IT = 1 + 2 + 3
(2.1.103) IIT = 1 2 + 2 3 + 3 1
IIIT = 1 2 3 .
Theorem. A tensor has
 either three real eigenvalues
 or one real and two conjugate complex ones.
Proof. From the behaviour of such cubic polynomials for very small and very
large values of  and its continuity we conclude that at least one real root 1
must exist. Let e1 be the corresponding eigenvector. Then there exists an ONB
{ei} with respect to which the tensor has the following components
1 T12 T13 
0 T 
 22 T23  .
0 T32 T33 
The trace of a tensor is real

IT = 1 + 2 + 3 = 1 + T22 + T33
and therefore also the sum of the other eigenvalues
2 + 3 = T22 + T33 .
We make for them a complex ansatz with the imaginary unity i
2 = 2 + i 2
3 = 3 + i 3 .
Its sum is only real if 2 = – 3 . The determinant of the tensor is also real
IIIT = 1 2 3 = 1 (T22 T33 – T32 T23)
 2 3 = T22 T33 – T32 T23
so that the product of the two other eigenvalues is also real
2 3 = (2 + i 2) (3 – i 2) = 2 3 + 22 + i (3 – 2) 2
and thus (3 – 2) 2 = 0 . For 2 = 0 all eigenvalues are real. If 3 – 2 = 0 ,
then the two remaining eigenvalues are conjugate complex; q. e. d.
26
Francois Viète (1540-1603)
Let T be a tensor with real eigenvalues 1 , 2 , 3 and corresponding

eigenvectors a1 , a2 , a3 . First, we consider the case in which two different
eigenvectors a1 and a2 have the same eigenvalue 
T  a1 =  a1
T  a2 =  a2 a1  a2
 T  a1 + T  a2 =  a1 +  a2
= T  (a1 + a2) =  (a1 + a2) .
So the sum of two eigenvectors corresponding to the same eigenvalue is also an
eigenvector for the same eigenvalue. The same holds for all linear combinations of
the two eigenvectors. The plane spanned by such eigenvectors is an eigenspace of
the tensor. If the same holds for three eigenvectors which are linear independent,
then the entire vector space is an eigenspace.
Examples
 For the zero tensor 0 every vector is an eigenvector with the (triple)
eigenvalue 0 since
0a = o = 0a.
 For the identity tensor I every vector is an eigenvector with the (triple)
eigenvalue 1 since
Ia = a = 1a.
 For a spherical tensor  I every vector is an eigenvector with the (triple)
eigenvalue  .
Since for spherical tensors all directions are eigendirections, we conclude
Theorem. If T is a tensor and  I a spherical tensor, then T and T +  I
have the same eigendirections.
Accordingly, the eigenvectors of a tensor depend only on its deviatoric part.
 For an orthogonal tensor (2.1.76) the axial vector e1 is an eigenvector with
eigenvalue 1. The positive sign holds for pure rotations (versors), the
negative one for additional reflections. The other two eigenvalues are
conjugate complex.
 For dyads a  b , a is an eigenvector corresponding to the eigenvalue a  b ,
and every vector perpendicular to b is an eigenvector for the double
eigenvalue 0 . If a is perpendicular to b , there is a triple eigenvalue 0 .
2.1.11 Spectral Forms of Symmetric Tensors
In mechanics, the eigenvalue problem is mainly posed for symmetric tensors, for
which the following important theorem holds.
Theorem. Symmetric tensors have three (not necessary different) real eigen-
values.
Proof. By the relation of the previous proof we obtain the equation
(2 – 3)2 = 22 + 32 – 2 2 3
= (2 + 3)2 – 4 2 3
= (T22 + T33)2 – 4 (T22 T33 – T23 T32)
= (T22 – T33)2 + 4 T232 .
Both terms on the right -hand side are non-negative, and so
0  (2 – 3)2 = (i 2 + i 2)2 = 4 i2 2 2 = – 4 22.
This is only possible if 2 = 0 , so that all eigenvalues must be real; q. e. d.
If we consider the case of two eigenvectors a1 and a2 with different eigen-
values 1  2 , then
T  a 1 = 1 a 1 and T  a 2 = 2 a 2
 a2  T  a1 = 1 a2  a1
a1  T  a2 = 2 a1  a2
and because of the assumed symmetry of T we obtain for the difference
a1  T  a2 – a2  T  a1 = 0 = (2 – 1) (a1  a2)
i.e., a1 is perpendicular to a2 . Thus, we have shown the following
Theorem. Eigenvectors of symmetric tensors with different eigenvalues are
mutually orthogonal.
We consider now the three possible cases.
1st case: three different eigenvalues.
In this case, the eigenvectors form an ONB called the eigenbasis of the tensor,
which is unique (up to changes of sign). The eigenspaces are three one-
dimensional mutually orthogonal vector spaces. With respect to the eigenbasis, the
matrix of the components has diagonal form or spectral form
 λ1 0 0 
0 λ 0 
 2 
0 0 λ3 
with the three eigenvalues 1 , 2 , 3 .

2nd case: two of three eigenvalues are equal.

The eigenspace corresponding the equal eigenvalues is two-dimensional. It is a
plane perpendicular to the eigenvector of the third eigenvalue. All vectors in this
plane are eigenvectors. A diagonal form is also possible, which is (after an
appropriate ordering)
 λ1 0 0 
0 λ 0 
 1 
0 0 λ2 
3rd case: all (three) eigenvalues are equal.

In this case, all directions are eigendirections, and the eigenspace coincides with
the underlying vector space. The matrix of the components has diagonal form with
respect to each ONB
 λ 0 0
0 λ 0  .
 
0 0 λ 
The tensor is the spherical tensor  I .

Theorem. A tensor is symmetric if and only if there exists an ONB {epi} with
respect to which the tensor has a spectral form
3
(2.1.104) T   i e pi  e pi
i 1
with i : real eigenvalues,

e pi : normed eigenvectors.
Theorem. A symmetric tensor is positive definite if and only if all eigenvalues are
positive. It is positive semidefinite, if and only if all eigenvalues are non-negative.
We obtain the following classification of the symmetric tensors if we order the
eigenvalues after their value, i.e., 1  2  3 . The columns 3 - 5 in the table
contain the intersections with different planes.
sign of square e p1 - e p2 e p2 - e p3 e p1 - e p3 tensor surface

eigenvals. form plane plane plane
+++ positive def. ellipse ellipse ellipse ellipsoid

++0 pos.-semidef. ellipse straight lines straight lines ellipt. cylinder
+00 pos.-semidef. straight lines - straight lines parall. planes
++– indef. ellipse hyperbola hyperbola 1fld hyperboloid
+– – indef. hyperbola - hyperbola 2fld hyperboloid
Problem 2. Products between Vectors

Two vectors v1 = 0.6 e1 + 0.8 e3 and v2 = – 0.8 e1 + 0.6 e3 are given with
respect to an ONB {ei} . Find a third vector v3 such that {v1 , v2 , v3}
forms an ONB. Represent the tensor A : = v1  v3 + v3  v1 with respect
to {ei  ej} and calculate the product A  v2 .
Solution
An ONB is defined by the equations vi  vj = ij . We first make sure that
v1 and v2 fulfil this. In fact,
v1  v2 = – 0.48 e1  e1 + 0.36 e1  e3 – 0.64 e3  e1 + 0.48 e3  e3 = 0
v1  v1 = 0.36 e1  e1 + 0.48 e1  e3 + 0.48 e3  e1 + 0.64 e3  e3 = 1
v2  v2 = 0.64 e1  e1 – 0.48 e1  e – 0.48 e3  e1 + 0.36 e3  e3 = 1
The mixed products e1  e3 and e3  e1 are all zero for an ONB {ei}, while
e1  e1 = 1 and e3  e3 = 1 . For the determination of v3 we use the
equations v1  v3 = 0 , v2  v3 = 0 , v3  v3 = 1 . We make the ansatz
v3 = v31 e1 + v32 e2 + v33 e3
and obtain a system of equations for the components v31 , v32 , and v33
v1  v3 = 0.6 v31 + 0.8 v33 = 0
v2  v3 = – 0.8 v31 + 0.6 v33 = 0
v3  v3 = (v31)2 + (v32)2 + (v33)2 = 1
The first two equations are only fulfilled if v31 = v33 = 0 , so that v32 can
only be
(P2.1) v32 = 1 .
We want to determine v3 such that {v1 , v2 , v3} is a positively oriented
system. Thus, we postulate v3 = v1  v2 . One could also have used this
equation for the calculation of v3
v3 = v1  v2 = (0.6 e1 + 0.8 e3)  (– 0.8 e1 + 0.6 e3)
= – 0.48 e1  e1 + 0.36 e1  e3 – 0.64 e3  e1 + 0.48 e3  e3
= – 0.48 11k ek + 0.36 13k ek – 0.64 31k ek + 0.48 33k ek
The last equation has for k  2 only non-zero solutions. With 132 = –1
and 312 = 1 (11k = 0 and 33k = 0 for all k) we obtain v3 = – e2 . So we
have to choose after P2.1 –1 for a positively oriented system or +1 for a

negatively oriented one. We continue with v3  – e2 . A is then
A = v1  v3 + v3  v1
= – (0.6 e1 + 0.8 e3)  e2 – e2  (0.6 e1 + 0.8 e3)
= – 0.6 e1  e2 – 0.8 e3  e2 – 0.6 e2  e1 – 0.8 e2  e3
= A12 e1  e2 + A32 e3  e2 + A21 e2  e1 + A32 e2  e3 .
We can put the components of A with respect to this basis in a matrix
 0 0.6 0 
[A ij] =  0.6
ee
0 0.8  .

 0 0.8 0 
We now compute
b : = A  v2
(P2.2) = (– 0.6 e1  e2 – 0.8 e3  e2 – 0.6 e2  e1 – 0.8 e2  e3)
 (– 0.8 e1 + 0.6 e3) .
Since we multiply v2 from the right side to A , the right vector of the base
dyad must be contracted with v2 . The left vector of the base dyads
remains. Since {ei} is an ONB, only two terms remain
b = (– 0.6) ( – 0.8) e2 + (– 0.8) (0.6) e2 = o .
With respect to the basis {vi} we obtain the component matrix for A
0 0 1 
[Avvij] = 0 0 0  .
 1 0 0 
The calculation of b with respect to the basis {vi} is rather simple

(compare with P2.2)
b = A  v2 = ( v1  v3+ v3  v1)  v2 = o
because of the orthogonality of the basis {vi} . Since both {ei} and {vi}
are ONBs, the matrices of the components with respect to both bases must
have the same eigenvalues and principal invariants: IA = 0 , II A = –1 ,
III A = 0 after (2.1.87-89).
Problem 3. Direct Notation and Index Notation

Determine v  T  v between a vector v and a tensor T in components
with respect to an ONB {ei} . Formulate the result as compact as
possible and bring it - as far as possible - in a direct notation.
Solution
For the components with respect to an ONB we obtain
v  T  v = vi ei  Tjk ej  ek  vl el
= vi Tjk vl (ei  ej)  (ek  el) .
Since the cross-product between vectors gives a vector, we obtain as a

result dyads or second-order tensors. We use the permutation symbol of
(2.1.2) to evaluate the cross-products
(P3.1) v  T  v = ijm kln vi Tjk vl em  en .
If we want to determine the component belonging to the base dyad e1  e2

we set m  1 and n  2 and obtain the sum ij1 kl2 vi Tjk vl . We have to
sum over 4 dummy indices, so that we obtain 34 = 81 terms. Most of the
terms, however, are zero since only six index combinations in ijm out of
27 are non-zero. For the compactification, the rule
 ik  jk  mk 
 
ijm kln = det   il  jl  ml 
 in  jn  mn 

helps. One assembles a matrix with elements ij such that the indices
coincide with the three indices of the permutation symbol row-wise and
column-wise. The determinant of this matrix equals the product of the two
permutations symbols. Calculation of the determinant after the SARRUS27
rule gives
ijm kln = ik jl mn + jk ml in + mk il jn – mk jl in
– ik ml jn – jk il mn .
We insert this into P3.1

v  T  v = (ik jl mn + jk ml in + mk il jn
– mk jl in – ik ml jn – jk il mn) vi Tjk vl em  en .
27
Pierre Frédéric Sarrus (1798-1861)
We can now make reductions after rule (2.1.1) for the KRONECKER
symbols. One can rename the dummy index in one KRONECKER symbol
and thus eliminate the other KRONECKER symbol. We find
v  T  v = vj Tji vi em  em + vi Tjj vl el  ei
+ vi Tjk vi ek  ej – vi Tjk vj ek  ei
– vi Tji vl el  ej – vi Tjj vi em  em .
In each term only two dummy indices appear in the base dyads. We can
transform the result back into a direct notation
v  T  v = (v  T  v) I + tr(T) v  v
+ (v  v) TT – (v  T)  v – v  (T  v) – tr(T) (v  v) I
with
vj Tji vi = v  T  v I = em  em Tjj = tr T
vi vl el  ei = v  v vi vi = v  v Tjk ek  ej = TT.
Problem 4. Orthogonal Tensors

Given the two ONBs {ei} and {vi} from Problem 2, find the tensor Q
that describes the transformation of {ei} into {vi} . Determine the
components of Q with respect to the bases {ei  ej} , {vi  vj} , {vi  ej} ,
and {ei  vj} . Note the difference between change of basis and
transformation of a vector.
Solution
It is obvious that
(P4.1) Q = vi  ei = v1  e1 + v2  e2 + v3  e3
performs the desired transformation. The matrix of the components with
respect to {vi  ej} is then
1 0 0 
[Q ij] = 0 1 0  .
ve
0 0 1
For the representation with respect to the basis {ei  ej} we insert the
vectors vi from Problem 2
Q = (0.6 e1 + 0.8 e3)  e1 + (– 0.8 e1 +0.6 e3)  e2 – e 2  e3
and obtain the components
0.6 0.8 0 
[Qeeij] =  0 0 1 .

0.8 0.6 0 
The components with respect to {vi  vj} can be achieved by scalar pro-
ducts of the tensor by vectors vi from the left and vj from the right
Qvvij = vi  Q  vj
= (vk  vi ) (ek  vj ) = ik ek  vj = ei  vj .
With v1 = 0.6 e1 + 0.8 e3 , v2 = – 0.8 e1 + 0.6 e3 and v3 = – e2 we obtain
0.6 0.8 0 
[Qvvij] =  0 0 1 = [Qeeij] .

0.8 0.6 0 
Thus, Qeeij = Qvvij holds. In the same way we find

Qevij = Q  ei  vj = Qeekl ek  el  ei  vj
= Qeekl ik el  vj = Qeeil el  vj .
With the previous result ei  vj = Qeeij we obtain
 0.36 0.48 0.8 

[Qevij] = [Q il] [Q lj] =  0.8 0.6
ee ee
0 .

 0.48 0.64 0.6 
Since Q maps a right-hand-system into another one, it is orientation-

preserving and therefore a proper-orthogonal tensor or a versor. So its
determinant must be +1, which can easily be proven. Such tensors are pure
rotations
vi = Q  ei  ei = QT  vi .
One should note the difference between this vector transformation and a
change of basis. In the latter case, one changes both the basis and the
components simultaneously, such that the vector itself remains the same,
while only its representation changes
b = bei ei = bei QT  vi = bei [Qvvjk]T vj  vk  vi
= bei Qvvkj ik vj = bei Qvvij vj = bvi vj .
In contrast to this, a rotation of a vector would lead to a change of the

components with respect to the same basis
b* = b e*i ei = Q  b = Qeeij bek ei  ej  ek
= Qeeij bek jk ei = Qeeij bej ei .
Because of Qeeij = Qvvij we have for the change of the basis
b = bei ei = bei Qeeij (Q  ej) ,
while the rotation of the vector leads to
b* = Q  b = bie* ei = Qijee bje ei .
Problem 5. Eigenvalues and Invariants

Determine the eigenvalues and eigenvectors and the principal invariants of
a tensor, which is given with respect to an ONB {ei} by
A = 10 (e1  e1 + e2  e2) + 5 (e1  e2+ e2  e1) + 20 e3  e3 .
Represent A with respect to its eigenbasis {v i} . Find the tensor Q for
the change of the basis from {ei} to {v i} . Calculate the inverse of A with
respect to {ei} and to {vi} .
Solution
The matrix of the components of A with respect to {ei  ej} is
10 5 0 
[Aij] =  5 10 0  .
 0 0 20 
With (2.1.87) to (2.1.89) we can calculate the principal invariants as
IA = A11 + A22 + A33 = 40

IIA = A11 A22 – A12 A21 + A22 A33 – A23 A32 + A11 A33 – A13 A31
= 100 – 25 + 200 + 200 = 475
IIIA = A11 A22 A33 – A11 A23 A32 – A12 A21 A33 +
A12 A23 A31 + A13 A21 A32 – A13 A22 A31
= 2000 – 500 = 1500
The characteristic polynomial (2.1.86) is then
(P5.1) 3 – 40 2 + 475  – 1500 = 0
which can be used to determine the eigenvalues. Since A13 = A23 = A31 = A32
= 0 we see that A  e3 = 20 e3 . So  = 20 is the eigenvalue of the
eigenvector e3 . Polynomial division of P5.1 by  –20 gives
(3– 402+475 –1500) / ( – 20) = 2 – 20 + 75
3– 202
– 202+475
– 202+400
75 – 1500
75 – 1500
0
The solutions of the resulting quadratic equation are
1,2 = 10  5 .
So we have the following eigenvalues: 1 = 5 , 2 = 15 , 3 = 20 , where the
order is arbitrary. The third eigenvector is already known: v3 = e3 . We
compute v1 and v2 using (2.1.83), which results in the component form
(Aij –  ij) vj . For 1 we obtain
 v11  v 1 
10  5 5 0     5 5 0   1  0 
 5 10  5 0  v21  = 5 5 0  v21  = 0 
   1     
 0 0 20  5   v3  0 0 15   v31  0 
   
One can easily see that v31 = 0 , and that the first two equations for the
components lead to v11 = – v21. Norming v1 determines v11 and v21
only up to their sign, so that v11 = v21 = 1/ 2 . For 2 we obtain by
the same procedure v12 = v22, v12 = v22 = 1/ 2 , v13= 0. The spectral
form of A is after (2.1.104)
A = 1 v1  v1 + 2 v2  v2 + 3 v3  v3.
The sense of direction of the vi does not matter, since these vectors appear
in all base dyads twice. A has the following components with respect to
{v i  v j}
5 0 0 
[Avvij] = 0 15 0  .
0 0 20 
A change of the basis from {ei} to {vi} is performed by Q = v i  ei . For

this purpose, we have to fix the senses of v1 and v2 : v1 = 1/ 2 e1 – 1/ 2
e2 , v2 = 1/ 2 e1 + 1/ 2 e2 , v3 = e3 . Inserting them gives us the
components of Q with respect to {ei  ej}
 1/ 2 1/ 2 0 
 
[Qeeij] =  1/ 2 1/ 2 0  .
 0 0 1
 
The inverse of A can be most easily determined in the spectral form
A–1 = 1–1 v1  v1 + 2–1 v2  v2 + 3–1 v3  v3
or as a matrix
1/5 0 0 
[Avvij] –1 =  0 1/15 0 .

 0 0 1/20 
The representation with respect to {ei  ej} can be most easily calculated
by a change of the basis
A1 = [Avvij]1 v i  v j = [Avvij]1 ( Q  ei)  (Q  ej)
= [Avvij]1 Qeelk Qeemn (el  ek  ei)  (em  en  ej)
= [Avvij]1Qeeli Qeemj el  em .
A comparison of the components with the representation A1 = [Aeelm]1 el
 em shows that one can determine the components [Aeelm]1 by the matrix
multiplication
[Aeelm]1 = Qeeli [Avvij]1 QeejmT
as
 4/30 2/30 0 
 0  .
[Aeelm]1 =  2/30 4/30
 0 0 1/20 
Problem 6. Spectral Form

The well-known FIBONACCI28 series is defined by
an = an1 + an2 a1 = 1 , a2 = 1 .
This can be written in matrix form as
 an 1  0 1  an 2 
 a =   .
 n   1 1  an 1 
Find an explicit formula for an using the spectral form.
Solution
With the given initial values we can write
n2
 an 1  0 1  1 n2 1
(P6.1)  a  =    1 = [ M ]  1 .
 n   1 1    
We obtain
an = [M n2] 21 + [M n2] 22 .
By the spectral form one can determine the n–2th power of the matrix of
coefficients M. The eigenvalues and normed eigenvectors are then
 1 5 5+ 5 
1 = ½ (1+ 5 ) v =  , 
 10  20 10 

 1+ 5 2 
2 = ½ (1– 5 ) w =  , .
 10 + 20 5 + 5 
The matrices of the coefficients [Mij]n2 can be obtained from the spectral
form by applying the exponents to the eigenvalues using the scheme for
matrix multiplications
v1 v2   w1 w2 
Mijn–2 = 1 n–2
 v1    + 2
n–2
 w1    .
v    w   
 2    2  
The calculation is somewhat laborious. The result is
an = (1n – 2n) / 5 .
28
Fibonacci (Leonardo of Pisa) around 1170-1250
2.1.12 Time-Dependent Vectors and Tensors
and their derivatives will be needed in dynamical problems. So let t be the time
or a time-like parameter. Let  (t) be a scalar function of t , v(t) a time-
dependent vector, and T(t) a time-dependent tensor. The time-derivatives are
defined by the limits
d 1
 (t) = = lim [ (t + t) –  (t)]
dt t 0  t
dv 1
v(t) = = lim [v(t + t) – v(t)]
dt t 0 t
dT 1
T(t) = = lim [T(t + t) – T(t)]
dt t 0 t
assuming that they exist.

The following rules hold for all scalars  , vectors a and b , and tensors T
and S , all being differentiable functions of time. Then
(a + b) = a + b
(a  b) = a  b + a  b
(a  b) = a  b + a  b
(T + S) = T + S
(T  S) = T  S + T  S
( T) =  T +  T
(T  a) = T  a + T  a
(a  T) = a  T + a  T
(T  a) = T  a + T  a
(a  T) = a  T + a  T
(TT) = (T)T
If we represent a time-dependent vector with respect of a time-independent ONB
v = vi(t) ei ,
then by ej = o we get
v = vi ei
so that the time-derivative of a vector is reduced to the time-derivatives of its
scalar components. Analogously we obtain for a tensor
T = T ij(t) ei  ej
the time-derivative
T = T ij  ei  ej .
If T is invertible at all times, we get
(2.1.105) (T –1) = –T –1  T  T –1
since
(T –1  T) = 0 = (T –1)  T + T –1  T.
We now consider as a special case a time-dependent orthogonal tensor Q(t) . By
Q  QT = I we see that
(2.1.106) (Q  QT) = 0 = Q  QT + Q  QT 
= Q  QT + (Q  QT)T.
Thus, Q  QT is skew. If Q(t) rotates a fixed ONB {ei} into some time-
dependent ONB {ei(t) } , this allows the representation
Q = ei(t)  ei  QT = ei  ei(t)
and
Q = ei(t)  ei
and
Q  Q–1 = Q  QT
= ei(t)  ei  ej  ej(t) = ei(t)  ei .
2.1.13 Rigid Body Dynamics
As a demonstration for the application of tensors, we will next consider the laws
of motion, and specify them for rigid bodies. It will be shown that a direct tensor
notation enables us to give these laws a very clear and compact form.
Let us consider a (deformable) body B which moves in the space being
subjected to forces and torques. Let O be a fixed reference point in space and rO
the position vector of some other point with respect to O . The centre of mass of
the body B is defined by its position vector
1 1
(2.1.107) rM : =  rO  dV =  rO dm
m m
V V
with
m mass of B
 mass density in B
V the current region of space occupied by B
If we decompose the position vector of an arbitrary point of the body into

rO = rM + x ,
we obtain by the definition of the centre of mass the useful formula
(2.1.108)  x dm = o
V
as well as
 
 x dm = o ,  x dm = o
V V
etc.
The linear momentum of the body is defined as the time-dependent vector
   
(2.1.109) p :=  rO dm =  (rM + x ) dm = rM m ,
V V
and the angular momentum or moment of momentum with respect to O is the
vector

(2.1.110) dO : =  rO  rO dm .
V
The balance of linear momentum (NEWTON 1687) is in general
d
(2.1.111) p =  
 rO dm =  rO dm = rM m

dt
V V
= f = resulting force acting on B
x
M
rO
rM
V
and the balance of angular momentum (EULER29 1775) with respect to O
29
Leonhard Euler (1707-1783)
d
(2.1.112) d O =  
 rO  rO dm =  rO  rO dm
dt
V V
= mO = resulting torque with respect to O acting on the body.
We now use the above decomposition of the position vector rO = rM + x
d O =  
 rO  rO dm =  (rM + x)  (rM + x) dm
V V
=  rM  rM dm +  rM  x dm
V V
 
+  x  rM dm +  x  x dm
V V
 
= rM  rM m + rM   x dm
V

+  x dm  rM +  x  x dm
V V
and by (2.1.108)
dO = rM  rM m + 
 x  x dm
V
= mO .
By using VARIGNON´s principle, we can also refer the torques to M
mO = rM  f + mM .
If we multiply the balance of linear momentum (2.1.111) by rM
rM  rM m = rM  f
then the difference of the result and the previous equations gives the balance of
angular momentum with respect to the centre of mass
(2.1.113) dM = mM with dM : = 
 x  x dm
V
i.e., in the same form as for a fixed point (2.1.112). In contrast, if we transform the
balance of angular momentum to some other moving point, it has to be enlarged
by additional terms.
If we want to further reduce the balance of angular momentum, we restrict our
consideration to rigid bodies. The displacement of any point of a rigid body with
position vector rO = rM + x can be decomposed into the displacement of the
centre of mass
uM (t) = rM (t) – rM (0)
and a rotation of x around M , which can be performed by a versor Q(t)
x(t) = Q(t)  x(0) .
Its time-derivative is
x(t) = Q(t)  x(0) = Q(t)  QT(t)  x(t)
=  (t)  x(t)
with  (t) being the axial vector of the skew tensor Q  QT after (2.1.106)
called the angular velocity. The total velocity of an arbitrary point of the body is
thus
(2.1.114) v(t) = rO = rM (t) + x (t) = vM (t) +  (t)  x(t)

This is EULER´s velocity formula of rigid body kinematics (wherein one could
also refer to any other reference point of the body instead of the centre of mass).
Consequently, each vector x being fixed to the body rotates with the same
angular velocity x =   x .
The angular momentum with respect to M is

dM (t) =  x  x dm
V
=  x  (  x) dm
V
and after the formula for double cross-products
=  [ (x  x) – x (x  )] dm
V
= [  (x  x I – x  x) dm]  
V
(2.1.115) dM = J M  
with the tensor of inertia with respect to M
 (x I – x  x) dm .
2
JM : =
V
Its components with respect to some ONB are the moments of inertia
 [(x1 + x2 + x3 ) ik – xi xk] dm

2 2 2
Jik = ei  JM  ek =
V
 (x2 + x3 ) dm  x1 x2 dm
2 2
i.e. J11 = J12 = J21 = –
V V
 (x3 + x1 ) dm  x2 x3 dm
2 2
J22 = J23 = J32 = –
V V
 (x1 + x2 ) dm  x3 x1 dm
2 2
J33 = J31 = J13 = –
V V
The tensor of inertia is symmetric for all bodies. Accordingly, there exists an
eigenbasis {e pi} which gives the spectral form
3
JM =  J pi e pi  e pi .
i 1
These principal axes of inertia indicated by the vectors e pi are for all bodies and
all motions fixed to the body, but not fixed in space. The principal moments of
inertia J pi are time-independent and positive, so that JM is positive definite.
For evaluating the balance of angular momentum, we will need the time
derivative of the tensor of inertia, which is by use of EULER´s velocity formula
(2.1.114)
3
JM =  J pi [e pi  e pi + e pi  e pi ]
i 1
3
(2.1.116) =  J pi [(  e pi )  e pi + e pi  (  e pi )]
i 1
3
=  J pi [  (e pi  e pi) – (e pi  e pi)  ]
i 1
=   JM – JM   .
Accordingly, the rate of the angular momentum equals
dM = JM   + JM  
= (  JM – JM  )   + JM  .
The term in the middle is zero after the rules of the triple product. Therefore, the
balance of angular momentum becomes with respect to the centre of mass of a
rigid body
(2.1.117) mM = JM   +   JM  
If we also represent the angular velocity with respect to the eigenbasis {e pi}
 (t) =  pi(t) e pi(t)
then its time-derivative is
 (t) =  pi e pi +  pi e pi
=  pi e pi +  pi (  e pi)
=  pi e pi +   
=  pi  e pi .
This holds analogously for each ONB {ei} that is fixed to the body. The
component form of the balance of angular momentum with respect to such a basis
is
mMi = Jil  l  +  l Jkp  p lki
and in particular with respect to the eigenbasis
mM p1 = J p1  p1  +  p2  p3 (J p3 – J p2)
(2.1.118) mM p2 = J p2  p2  +  p1  p3 (J p1 – J p3)
mM p3 = J p3  p3  +  p2  p1 (J p2 – J p1)
These are EULER´s equations for gyroscopes (1758) with respect to the centre
of mass and to the principal axes of inertia.
The kinetic energy of the rigid body is
2
K = ½  rO dm
V
  2
= ½  (rM + x ) dm
V

 (rM +   x) dm
2
= ½
V
= ½ rM2 m + ½ 
 (  x) dm + rM     x dm
2
V V
(2.1.119) = ½ rM2 m + ½  (  x) dm
2
V
= Ktrans + Krot
i.e., a sum of translatoric and rotatoric energy. The latter can be reformulated as
Krot = ½  (  x)  (  x) dm
V
= ½    [x  (  x)] dm
V
= ½  [(x  x)  – (x  ) x] dm
V
 [x I – (x  x)] dm  
2
(2.1.120) = ½
V
= ½   JM  
= ½   dM
= ½ (J11  12 + J22  22 + J33  32)

+ ( 1 J12  2 +  2 J23  3 +  3 J31  1) .
This is a positive definite square form in the angular velocity, i.e., the rotatoric
energy is positive for all   o . With respect to the principal axes of inertia the
last bracket vanishes.
If we now consider the particular case of a load-free gyroscope (mM  o , f  o).
In this case we have
 conservation of linear momentum (2.1.121)
p = vM m = constant  vM = constant
 conservation of angular momentum (2.1.122)
dM = JM   = constant
 conservation of energy (2.1.123)
K = ½ vM2 m + ½   JM   = constant
We will later show that the conservation of energy is a result of conservation of
linear momentum and angular momentum and thus does not give another
independent balance law.
Using (2.1.123) we conclude with (2.1.121)
  JM   = constant (energy ellipsoid)
and with (2.1.122)
dM2 =   JM2   = constant (angular momentum ellipsoid)
2
The tensor surfaces of JM and JM are two ellipsoids, both fixed to the body and
with the same axes, namely the principal axes of inertia of the gyroscope. The
geometric interpretation is that  is located on both ellipsoids and, therefore, on
the intersection of them.  describes the polecone. The normal n of the energy
ellipsoid lies in the direction of
JM   = dM = constant
and is fixed in space after (2.1.122). The energy ellipsoid is rolling on a tangential
plane fixed in space (POINSOT30´s rolling motion).
On the other hand we can conclude from (2.1.122)
dM2 = dM  I  dM = constant (sphere)
and from (2.1.123)
  JM   = dM  JM–1  dM = constant
(McCULLAGH ellipsoid)
30
Louis Poinsot (1777-1859)
angular momentum ellipsoid dM

tangential plane (fixed)

energy ellipsoid
The first describes a sphere fixed in space and with respect to the body. The latter
is the McCULLAGH31 ellipsoid fixed with respect to the body. The vector of
angular momentum lies in the intersection of the body-fixed ellipsoid and a sphere
and describes the polehode. The normal to JM–1 is in the direction of
JM–1  dM = 
and swings in space.
McCullagh ellipsoid

sphere dM (fixed)
2.1.14 Bending of Bars
We consider the three-dimensional bending problem of originally straight bars.

We make the following assumptions.
 BERNOULLI´s hypothesis: plain cross-sections remain plain and
perpendicular to the central axis of the bar
 small displacements, small rotations, and small deformations
31
James McCullagh (1809-1847)
 linear-elastic material:  = E  (HOOKE´s law)

 no axial loads.
Let
s be the coordinate of the arc length of the undeformed bar
 the derivative with respect to s
 (s) the (infinitesimal) rotation vector of the cross-section in s
A the cross-section
A(s) the area of the cross-section
x the position vector of the neutral fibre to an arbitrary point of the
cross-section
t the normed tangent vector to the axis of the bar (t = 1)
z,e3

 s ds
e1, s S
x y, e2
ds(1+ )
x P
dA
ds
rs
 ds
We consider some cross-sections at a constant s . As we have already seen before,

one can represent an infinitesimal rotation by an axial vector  (s) . For the
rotation of the cross-section plane, this vector is contained in this plane. The
displacement resulting from bending is
u =  (s)  x + uo
with some constant part uo . The derivative of the tangential part of the
displacement u  t is the stretch in axial direction
d
t = (u  t) = u  t + u  t 
ds
=  (s)   x  t +  (s)  x  t +  (s)  x  t 
d
(2.1.124) = xt with  : =  (s)
ds
since x does not depend on s , t  lies in the cross-section plane, while  (s)  x
is perpendicular to it. The traction vector in some point of the cross-section
indicated by x is after HOOKE´s law
 =  t t = E t t = E {(  x)  t} t .
Since we assumed that no resulting normal force acts on the cross-section, we
obtain
  t dA = E  (  x)  t dA = E (t  )   x dA = 0
A A A
so that the neutral fibre must go through the centroid of the cross-section. The
resulting torque with respect to the centroid is
m =  x   dA = E  {  (x  t)} x  t dA
A A
= E(  (x  t)  (x  t) dA)  
A
(2.1.125) m = E JA  
with the tensor of inertia of area
JA : =  (x  t)  (x  t) dA
A
obviously being symmetric. We choose an ONB {ei} with e1  t so that
x(y, z) = y e2 + z e3
and
Jik = ei  JA  ek
=  ei  {(y e2 + z e3)  e1} {(y e2 + z e3)  e1}  ek dA
A
=  (y i21 + z i31) (y 21k + z 31k) dA

A
and also
 z dA  y dA  y z dA
2 2
Jyy = Jzz = Jyz = Jzy = –
A A A
while all other components with respect to this basis are zero
0 0 0 
 
[Jij] = 0 J yy J yz  .
 
0 J yz J zz 
e1 is eigenvector of JF with eigenvalue 0 . The two non-trivial eigenvalues J p1

and J p2 are the roots of the characteristic equation of JF
IIIJ – J pi IIJ + (J pi)2 IJ – (J pi)3 = 0
= 0 – J pi (Jyy Jzz – Jyz2) + (J pi)2 (Jyy + Jzz) – (J pi)3
with the solutions
1
J p2,3 = ½ (Jyy + Jzz)  ( J yy  J zz )2  J yy J zz  J yz
2
4
1
= ½ (Jyy + Jzz)  ( J yy  J zz )2  J 2yz
4
called the principal moments of inertia.

For the transformation of an arbitrary basis {ei} into the eigenbasis {e pi} we
use the transformations of the components
J pik = Qil Jlm Qkm
with
1 0 0 

[Qij] = 0 cos   sin  

0 sin  cos  
and the angle of rotation  . The matrix has spectral form with respect to the
principal axes of inertia
0 0 0 
p  p 
[J ij] = 0 J 2 0 .
 
0 0 J p3 
Thus
J p23 = 0 = sin  cos  (J22 – J33) + (cos2  – sin2 ) J23 .
With the trigonometric relations
2 sin  cos  = sin (2)
2 cos2  = 1 + cos (2)
2 sin2  = 1 – cos (2)
we conclude
½ sin (2) (J22 – J33) + cos (2) J23 = 0

or
2J 23
tan (2) =
J 33  J 22
as an equation to determine  and the eigenbasis

e p1 = e1
e p2 = cos  e2 + sin  e3
e p3 = – sin  e2 + cos  e3
2.1.15 Higher-Order Tensors
The scalar product of two vectors (which we will consider from now on as
1st-order tensors) gives the real number
vx.
Up to now, we have only introduced the twofold or dyadic tensor product between
two vectors v1 and v2 by its action on some vector x
v1  v2  x : = v1 (v2  x) = (v2  x) v1
(simple contraction). Analogously one defines the three-fold tensor product
between three vectors v1 , v2 , and v3 by its action on some vector x , the result of
which is the 2nd-order tensor
v1  v2  v3  x : = v1  v2 (v3  x)
(simple contraction). The three-fold tensor product is called a triad or 3rd-order
tensor.
One can continue this way up to the introduction of a K-fold tensor product of
K vectors v1 , v2 , ... , vK by its action on some vector x , the result of which is
the K–1-fold tensor product
v1  ...  vK  x : = v1  ...  vK–1 (vK  x)
(simple contraction).
We had already introduced the simple contraction of two dyads as a composition
of two linear mappings
v1  v2  x1  x2 = (v2  x1) v1  x2 .
A multiple contraction can be achieved if one contracts more couples of
adjacent vectors by scalar products. The order in which these contractions are
performed matters, and it can be defined in different ways. An example is the
double contraction of two dyads, the result of which is a scalar
v1  v2  x1  x2 : = (v1  x1) (v2  x2) .

This rule can be generalised in the following way.
Definition. The P-fold contraction of a K-fold tensor product with an M-fold
tensor product for K  P  M is the (K+M–2P)-fold tensor product
(v1  ...  vK) ... (x1  ...  xM)
(2.1.126) = v1  ...  vK–P  xP+1  ...  xM
(vK–P+1  x1) (vK–P+2  x2) ... (vK  xP) .
wherein " ..." stands for P scalar products.
Examples
K1, M1, P1 vx a scalar

K2, M1, P1 v1  v2  x = v1 (v2  x) a vector
K3, M1, P1 v1  v2  v3  x = v1  v2 (v3  x) a dyad
K1, M2, P1 v  x1  x2 = (v  x1) x2 a vector
K2, M2, P1 v1  v2  x1  x2 = (v2  x1) v1  x2 a dyad
K2, M2, P2 v1  v2  x1  x2 = (v1  x1) (v2  x2) a scalar
K3, M1, P1 v1  v2  v3  x1 = v1  v2 (v3  x1) a dyad
K3, M2, P1 v1  v2  v3  x1  x2 = v1  v2  x2 (v3  x1) a triad
K3, M2, P2 v1  v2  v3  x1  x2 = v1 (v2  x1) (v3  x2) a vector
K2, M3, P1 v1  v2  x1  x2  x3 = v1  x2  x3 (v2  x1) a triad
K1, M3, P1 v1  x1  x2  x3 = (v1  x1) x2  x3 a dyad
K3, M3, P3 v1  v2  v3  x1  x2  x3 = (v1  x1) (v2  x2) (v3  x3)
a scalar
etc.
Linear combinations of tensor products of equal order are achieved in analogy to
those of dyads.
Tensors of K-th-order with K  0 can be generated by K-fold tensor products
between base vectors of some ONB
ei1  ei2  ...  eiK
as linear combinations
(2.1.127) C = Ci1 i2 ...iK ei1  ei2  ...  eiK
with 3K components in three-dimensions. For K  1 this is a vector, for K  0 it

is defined as a scalar, i.e., a real.
Notations
tensors of 0th-order: scalars with 30 = 1 component
tensors of 1st-order: vectors with 31 = 3 components
tensor of 2nd-order: dyads with 32 = 9 components
tensors of 3rd-order: triads with 33 = 27 components
tensors of 4th-order: tetrads with 34 = 81 components.
If {ei} is another ONB, then we can represent a K-th-order tensor C as
C = Cij...kl e i  e j  ...  e k  e l .
The transformations of the components under change of ONBs results from the
orthogonal mappings
Q : = e i  ei = Qrs er  es
 e i = Q  ei = (Qrs er  es)  ei = Qri er
generalising those of vectors (2.1.79) and of dyads (2.1.80) as
(2.1.128) Cij ... kl = Qim Qjn ... Qkr Qls Cmn ... rs .
A tensor of K-th-order can be applied to some vector by a simple contraction as
C  x = (Cij...kl ei  ej  ...  ek  el)  (xm em)
= Cij...kl ei  ej  ...  ek (el  em) xm
= Cij...kl ei  ej  ...  ek lm xm
= Cij...kl xl ei  ej  ...  ek
giving a (K–1)-th-order tensor. For K  2 this operation coincides with the linear
mapping between vectors.
Similarly, one can define a simple contraction between a K-th-order tensor and a
dyad as a generalisation of the composition of two 2nd-order tensors
C  T = (Ci...jkl ei  ...  ej  ek  el)  (Tmp em  ep)
= Ci...jkl ei  ...  ej  ek  ep (el  em) Tmp
= Ci...jkl ei  ...  ej  ek  ep lm Tmp
= Ci...jkl Tlp ei  ...  ej  ek  ep
and obtains a K-th-order tensor. Thus, the two adjacent base vectors are contracted
in a scalar product.
One can also introduce a double contraction between a K-th-order tensors and a
dyad as
C  T = (Ci...jkl ei  ...  ej  ek  el)  (Tmp em  ep)
= Ci...jkl ei  ...  ej (ek  em) (el  ep) Tmp
= Ci...jkl ei  ...  ej km lp Tmp

= Ci...jkl Tkl ei  ...  ej
giving a (K–2)th-order tensor.
In analogy to (2.1.126) one can continue like this until the P-fold contraction of
a Kth-order tensor with an Mth-order tensor (K, M ≥ P), resulting in a (K+M–
2P)th-order tensor.
Theorem. A linear mapping of an M-th-order tensor into an L-th-order tensor can
be uniquely represented by an (L+M)-th-order tensor through an M-fold
contraction.
This motivates the use of higher-order tensors.
An interesting particular case is with L  0 and M arbitrary, an M-fold
contraction of two M-th-order tensors. The result is a 0-th-order tensor, i.e., a
scalar
C ... D = (Cij...k ei  ej  ...  ek)
... (Dmp...q em  ep  ...  eq)
(2.1.129) = Cij...k Dmp...q (ei  em) (ej  ep) ... (ek  eq)
= Cij...k Dmp...q im jp ... kq
= Cij...k Dij...k .
One can interpret this operation as a scalar product in the space of L-th-order
tensors, since the according axioms of a scalar product are fulfilled. This scalar
product induces a norm
(2.1.130) C : = (C ... C)
as well as lengths of and angles between tensors of arbitrary order in analogy to
those of vectors.
For L  2 this gives
S  T = (Sil ei  el)  (Tmp em  ep)
= Sil (ei  em) (el  ep) Tmp
= Sil im lp Tmp
= Sil Til
which can be expressed by the trace as
(2.1.131) S  T = tr(S  TT ) = tr(ST  T) = tr(TT  S) = tr(T  ST ) .
In the literature, this scalar product between tensors is sometimes called a double
scalar product (because of the double contraction). It commutes and is linear in
both factors. In contrast to this, one can show that tr(S  T) does not define a
scalar product, since it is not positive definite for S  T .
In particular, we obtain for T  I a representation for the trace of a tensor S
(2.1.132) S  I = tr(S  I) = tr(S) .

For simple dyads we get
(a  b)  (c  d) = (a  c) (b  d) .
With respect to this scalar product, the symmetric and the skew tensors are
mutually orthogonal. In fact, if T is an arbitrary dyad, S symmetric and A
skew, then
(2.1.133) S  A = 0
and therefore
S  T = S  ½ (T + TT ) = S  sym(T)
A  T = A  ½ (T – TT ) = A  skw (T) .
Using this scalar product, we are able to introduce a tensorial ONB.
Definition. A tensorbasis {fi  g j} is called an orthonormal basis (ONB) if
fi  g j  fk  g l = ik jl i, j, k, l = 1, 2, 3.
The following examples are given.
Let {ei} be a vectorial ONB. Then
 {ei  ej} is an ONB in the 9-dimensional space of 2nd-order tensors,
 { ei  ei and 2 sym(ei  ej) , i < j, i, j = 1, 2, 3} is an ONB in the 6-
dimensional space of symmetric tensors,
 { 2 skw (ei  ej) , i < j} is an ONB in the 3-dimensional space of skew
tensors.
Only tensors of even-order (2K) can be interpreted as linear mappings between

tensors of the same order (K). Their invertibility can be defined in complete
analogy to 2nd-order tensors. Among the 2K-th-order tensors is a distinguished
element, namely the 2K-th-order identity
I = i1 iK 1 i2 iK+2 ... diK i2K e i1  e i2  ...  e i2K
= e i1  e i2  ...  e iK  e i1  ...  e iK .
Generalising the symmetry definition of dyads to 2K-th-order (even) tensors by

K-fold contractions from both sides with arbitrary K-th-order tensors S and T
gives
S  ...  C  ...  T = T  ...  C  ...  S .
This is equivalent to the conditions for the components with respect to some ONB
C i1 i2 ... iK iK+1 ... i2K = C iK+1 iK+2 i2K i1 ... iK
for all indices. Symmetry does therefore not include the invariance under arbitrary
interchanges of indices, but only under interchanges of the first index group and
the second.
For even order (2K) tensors one can define eigenvalue problems in analogy to
Chapter 2.1.10. The eigendirections are directions in the space of K-th-order
tensors. Instead of eigenvectors, we are now looking for K-th-order eigentensors.
The resulting characteristic equation again contains principal invariants as its
coefficients. Many properties of the eigenvalue problem of dyads can also be
applied analogously to such even-order tensors. In particular, if such tensor is
symmetric, a spectral form can always be achieved with respect to its eigenbasis.
For odd-order tensors, however, an eigenvalue problem in this form can not be
defined.
2.1.16 Tetrads
Since we deal in mechanics mainly with dyads or 2nd-order tensors, we will

occasionally need 4th-order tensors or tetrads being used as linear mappings
between dyads
C  T = (Cijkl ei  ej  ek  el)  (Tmp em  ep)
= Cijkl ei  ej (ek  em) (el  ep) Tmp
(2.1.134) = Cijkl ei  ej km lp Tmp
= Cijkl Tkl ei  ej .
In the linear theory of elasticity, tetrads are frequently used for linear
dependences of the stress tensor and the deformation tensor, as we will see later.
The composition of such tensors C and D maps a 2nd-order tensor T into
(C  D)  T : = C  (D  T) = C  D  T .
So the brackets are not needed.
The 4th-order zero tensor O maps all dyads T into the second-order zero
O  T = 0 .
All of its components are zero with respect to all bases.
The 4th-order identity tensor I maps any dyad into itself
I  T = T .
One obtains its component representation with respect to some ONB as
(2.1.135) I = ik jl ei  ej  ek  el = ei  ej  ei  ej .
The eigenvalue problem of a tetrad consists of finding some 2nd-order eigen-

tensors A for eigenvalues  that fulfil the eigenvalue equation
(2.1.136) C  A =  A .
Here we expect 9 (not necessarily different) eigenvalues and 9 corresponding
eigentensors, being either real or complex.
The inverse C –1 of an invertible tetrad C gives
C –1  C  T = T
for all dyads T . This is equivalent to
(2.1.137) C –1  C = I = C  C –1.
In general, a tetrad has 34 = 81 independent components Cijkl , i, j, k, l = 1, 2, 3.
To represent them in a matrix, it is necessary to enlarge the concept of a matrix to
a hypermatrix, i.e., a matrix whose components are also matrices. This can be
achieved in the following form
 C1111 C1112 C1113  C1211 C1212 C1213  C1311 C1312 C1313  
 
 C1121 C1122 C1123  C
 1221 C1222 C1223  C
 1321 C1322 C1323  
 C1131 C1132 C1133  C1231 C1232 C1233  C1331 C1332 C1333  
 
 C2111 C2112 C2113  C2211 C2212 C2213  C2311 C2312 C2313  
 C C2122 C2123  C C2222 C2223  C C2322 C2323  
  2121  2221  2321
 C2131 C2132 C2133  C2231 C2232 C2233  C2331 C2332 C2333  
 
 C3111 C3112 C3113  C3211 C3212 C3213  C3311 C3312 C3313  
 C C3122 C3123  C C3222 C3223  C C3322 C3323  
  3121  3221  3321 
 C3131 C3132 C3133  C3231 C3232 C3233  C3331 C3332 C3333  
The number of 81 independent components can eventually be drastically

reduced by symmetries. While the components of a dyad have only two indices,
with respect to which a symmetry may exist, we have four indices in the case of
tetrads and, therefore, have to distinguish different symmetries. This will be done
next.
The transposition of a tetrad is introduced in analogy to that of a dyad (2.1.41)
by the relation
(2.1.138) S  CT  T = T  C  S
which shall be valid for all dyads S and T . A tetrad possesses the (main)
symmetry if it coincides with its transpose. For the components with respect to
some ONB this is equivalent to
Cijkl = Cklij .
These are 36 independent conditions. Consequently, a symmetric tetrad has
n2 / 2 + n / 2 = 45 independent components with n  9 in our case.
Independently from the (main) symmetry, we define two more types of

symmetries for tetrads, namely the left subsymmetry if
(2.1.139) T  C = TT  C
holds for all tensors T , and the right subsymmetry if
(2.1.140) C  T = C  TT
holds. For the components with respect to some ONB this is equivalent to the
conditions
Cijkl = Cjikl (left subsymmetry)
Cijkl = Cijlk (right subsymmetry).
These two subsymmetries are particularly helpful if a tetrad C is used to map
symmetric dyads into symmetric ones. For such applications the following
theorems hold.
 A tetrad C has the left subsymmetry if and only if C  T is symmetric for
arbitrary dyads T .
 A tetrad C has the right subsymmetry if and only if C  TA = 0 holds for all
skew dyads TA. Therefore, all skew tensors are eigentensors of C with a
triple eigenvalue 0.
 If a tetrad possesses a subsymmetry, it is singular (non-invertible).
In the linear theory of elasticity, one considers linear functions between the
(symmetric) deformation tensor and the (symmetric) stress tensor. For such linear
functions between symmetric dyads we can assume both subsymmetries without
influencing the relevant part of the function. In doing so, we can reduce the
number of independent components of the tetrad to 6  6 = 36. In this case it is
convenient to represent symmetric dyads as members of a 6-dimensional linear
space
T = T11 e1  e1 + T22 e2  e2 + T33 e3  e3
+ T23 (e2  e3 + e3  e2)
(2.1.141) + T31 (e3  e1 + e1  e3)
+ T12 (e1  e2 + e2  e1)
= T11 EV1 + T22 EV2 + T33 EV3
+ 2 T23 EV4 + 2 T31 EV5 + 2 T12 EV6
= TV EV (sum over  from 1 to 6)
with the matrix of components
{TV1 , TV2 , TV3 , TV4 , TV5 , TV6} : = {T11 , T22 , T33 ,2 T23 , 2 T31 , 2 T12}
with respect to the symmetric tensor basis
EV1 : = e1  e1
EV2 : = e2  e2
EV3 : = e3  e3
EV4 : = 1/2 (e2  e3 + e3  e2 )
EV5 : = 1/2 (e1  e3 + e3  e1 )
EV6 : = 1/2 (e2  e1 + e1  e2 )
where the square roots 2 are normalisation factors of the basis {EV}
EV  EV = 
such as
EV6  EV6 = 1/2 (e2  e1 + e1  e2)  1/2 (e2  e1 + e1  e2) = 1 .
This turns {EV} into an ONB in the 6-dimensional space of symmetric dyads.
It leads to a VOIGT32 representation (1882) of a tetrad with the two
subsymmetries
(2.1.142) C = CV EV  EV
with summation over Greek indices from 1 to 6 . The components of the tetrad can
now be given as a 6  6 matrix
 C1111 C1122 C1133 2C1123 2C1113 2C1112 
 
 C2211 C2222 C2233 2C2223 2C2213 2C2212 
 
 C3311 C3322 C3333 2C3323 2C3313 2C3312 
 
 2C2311 2C2322 2C2333 2C2323 2C2313 2C2312 
 2C 2C1322 2C1333 2C1323 2C1313 2C1312 
 1311 
 2C1211 2C1222 2C1233 2C1223 2C1213 2C1212 
 
We shall mention that in the literature the normalisation is sometimes not applied
which leads to a slightly different representation (without the 2 and the 2).
Obviously, the tetrad with the two subsymmetries possesses also the (main)
symmetry if and only if the VOIGT 66 matrix is symmetric. In this case only 21
independent components remain.
The 4th-order identity tensor possesses the (main) symmetry, but no
subsymmetry since this would lead to a loss of invertibility. The identity tensor is
invertible of course.
The identity tetrad must be distinguished from that particular tetrad I S which
possesses both subsymmetries and maps every dyad into its symmetric part. We
call it a symmetriser
I S = ¼ ik jl (ei  ej + ej  ei)  (ek  el + el  ek)
32
Woldemar Voigt (1850-1919)
= ¼ (ei  ej + ej  ei)  (ei  ej + ej  ei)

(2.1.143) = ½ (ei  ej + ej  ei)  (ei  ej)
= ½ (ei  ej)  (ei  ej + ej  ei)
with the VOIGT representation
1 0 0 0
0 0
0 1 0 0
0 0
 
0 0 1 0
0 0
  .
0 0 0 1
0 0
0 0 0 0
0 1
 
0 0 0 0 0 1
This tensor is used as the identity on the symmetric dyads.
In analogy to 2nd-order tensors we define the following properties of tetrads.

Definition. A (not necessarily symmetric) tetrad C with the property
A  C  A  0 A0 is called positive definite
A  C  A  0 A is called positive semidefinite
A  C  A  0 A0 is called negative definite
A  C  A  0 A is called negative semidefinite.
If nothing of the above holds, the tetrad is called indefinite. Evidently, only the
symmetric part of the tetrad enters into these definitions. For a symmetric C the
positive definiteness is equivalent to the positivity of all eigenvalues.
Such classifications can be made for all tensors of even-order.
Problem 7. Multiple Contraction

Between tensors of higher-order, a multiple contraction was introduced.
Simplify the following expression for two vectors a and b
(P7.1) v =   a  b
as far as possible, with  = ijk ei  ej  ek .
Solution
 is a triad, while a  b is a simple dyad. The double contraction reduces
both tensors by the order two, so that the result is a 1st-order tensor or a
vector v . By using indices, we obtain
v = ijk ei  ej  ek  am bn em  en
= ijk am bn jmkn ei = ijk aj bk ei = jki aj bk ei

= ab.
v is perpendicular to the plane spanned by a and b . The tensor a  b is
a simple dyad with IIa  b = IIIa  b = 0 .
2.2 Vector and Tensor Analysis
2.2.1 The Directional Differential
Let f be a real-valued differentiable function of a real variable. Its linear

approximation at x is the differential
df
df (x , dx) = dx = f (x)  dx
dx
with the derivative
df 1
f (x)  = : = lim [f (x + x) – f (x)] .
dx x 0 x
This concept shall now be generalised to tensor functions.
Let r be some position vector in the EUCLIDean space and  (r) a real field
(scalar field), i.e., a function that assigns to each position vector a scalar.
Examples: the temperature field, or the field of mass density or energy density.
The differential of  at r in the direction of dr is defined as the limit
1
(2.2.1) d (r , dr) : = lim [ (r + h dr) –  (r)]
h 0 h
d
=  (r + h dr) h = 0
dh
If the function  (r) is sufficiently smooth, then the differential is linear in dr ,
and therefore there exists a vector field grad  (r) , called the gradient of  (r) ,
such that
(2.2.2) d (r , dr) = grad  (r)  dr .
Other notations for the gradient are
d
grad  (r) = =  (r) .
dr
With respect to a fixed ONB {ei} we have the component representations
r = xj ej and dr = dxi ei
and for the differential of  at r in the direction dr  ei
1
(2.2.3) d (r , ei) = lim [ (r + h ei) –  (r)] = grad  (r)  ei
h 0 h
which corresponds to the i-th component of the gradient. If, e.g., i  1 , then
r + h e1 has the components {x1 + h , x2 , x3} , and the limit is the partial
2.2 Vecter and Tensor Analysis 105
derivative of  (xi ei) =  (x1 , x2 , x3) with respect to x1 , which is often denoted
by ", 1"
  x1 ,x2 ,x3 
: = d (r , e1) = grad  (r)  e1 =  , 1 .
x1
With this we obtain the component representation of the gradient with respect to
some ONB
  x1 ,x2 ,x3 
(2.2.4) grad  (r) = ei =  , i ei
xi
We consider an iso-surface of  , i.e., a surface in the EUCLIDean space on

which  is constant. If dr is tangential to this surface, then
d (r , dr) = grad  (r)  dr = 0 .
Consequently, grad  is orthogonal or normal to the tangential plane of the iso-
surface in r and points in the direction in which  has the largest increase.
 = const grad 
r dr
As an example we consider the temperature field
 (r)   (r) .
The temperature gradient
  x1 ,x2 ,x3 
(2.2.5) g(r) : = grad  (r) = ei =  , i ei
xi
points in the direction of the largest temperature increase. If there is a linear
relation between g and the heat flux vector q (FOURIER´s33 law of heat
conduction), then it can be represented by the heat conduction tensor K as
(2.2.6) q = –Kg.
After the CASIMIR-ONSAGER34 reciprocal relations, K is symmetric and
therefore allows for a spectral representation
33
Jean Baptiste Joseph de Fourier (1768-1830)
34
Hendrik Brugt Gerhard Casimir (1909-2000), Lars Onsager (1903-1976)
3
K =  i e pi  e pi
i 1
with real eigenvalues i . According to the experimental result that the heat flux is
always directed from the hot to the cold, the heat conduction tensor is positive
semidefinite (we will later see that this is a consequence of the second law of
thermodynamics), and thus
i  0 for i = 1, 2, 3 .
If the heat conduction ability is equal in all directions (isotropic heat conduction),
then
1 = 2 = 3 = : 
and
3
K =   ei  ei =  I
i 1
is a spherical tensor. In this case, the isotropic FOURIER´s law of heat conduction
is reduced to
(2.2.7) q = –g
with the (non-negative) coefficient of heat conduction  . Its components with
respect to some ONB are
  x1 ,x2 ,x3 
(2.2.8) qi = –  = –   (x1 , x2 , x3) , i .
xi
Let us next consider a vector field v(r) . Examples are the displacement field,
the velocity field, the force field, and the heat flux field. We determine the
differential of v at r in the direction of dr analogously to (2.2.1) as
1
(2.2.9) dv(r , dr) : = lim [v(r + h dr) – v(r)] .
h 0 h
This expression is again linear in dr if the field v(r) is sufficiently smooth.

Thus, there exists a tensor field grad v(r) , the gradient of v , which gives
dv  r 
(2.2.10) dv(r, dr) =  dr = grad v(r)  dr .
dr
After choosing some ONB {ei} , the differential in the direction of ei is
1
dv(r, ei) : = lim [v(r + h ei) – v(r)]
h 0 h
v  x1 ,x2 ,x3 
= = v(x1 , x2 , x3) , i
xi
= grad v(r)  ei
containing the partial derivatives v(x1 , x2 , x3) , i . If we represent the vector field
by its components with respect to a fixed ONB {ei} , which are scalar fields,
v(r) = vi(x1 , x2 , x3) ei
then we can express the partial derivatives of the vector field by those of its scalar
components as
v(x1 , x2 , x3) , i = [vj(x1 , x2 , x3) ej] , i = vj(x1 , x2 , x3) , i ej
so that
grad v  ei = v(x1 , x2 , x3) , i = vj ( x1 , x2 , x3) , i ej
= [vj (x1 , x2 , x3) , k ej  ek]  ei .
Accordingly, the gradient of the vector field v is the tensor field
(2.2.11) grad v = vj (x1 , x2 , x3) , k ej  ek
The matrix of its components is

  v1  v1  v1 
 
  x1  x2  x3 
  v2  v2  v2 
 .
  x1  x2  x3 
 v3 v3 v3 
 
 x1 x2  x3 
The transposed field of the gradient is notated as grad T v . The symmetric part
of the gradient will often be needed in kinematics. It is sometimes called the
deformator of the vector field
(2.2.12) def v : = ½ (grad v + grad T v) = ½ (vj , k + vk , j) ej  ek .
Its component matrix is with respect to some ONB
 v1 1  v1 v2  1  v1 v3  
      
 x1 2  x2 x1  2  x3 x1  
  v3  
 1  v2  v1  v2 1  v2
    .
 2  x1 x2  x2 2  x3 x2  
 
 1  v3 v1  1  v3 v2  v3 
 2  x  
x3 

2  x2
 
x3  x3 
  1 
The trace of the gradient of a vector field is called the divergence (-field)
v
(2.2.13) div v : = tr (grad v) = vj , j = v1 , 1 + v2 , 2 + v3 , 3 =  ei .
xi
Thus, the divergence of a vector field is a scalar field.
The curl of a vector field is defined by the axial vector tA of the skew part of
the gradient
W : = ½ (grad v – grad T v)
as
(2.2.14) curl v : = 2 tA .
The component matrix of W is then
 1  v1 v2  1  v1 v3  
 0      
 2  x2 x1  2  x3 x1  
 
 1  v2  v1  1  v2 v3  
 0    .
 2  x1 x2  2  x3 x2  
 
 1  v3 v1  1  v3 v2  
 2  x  
x3 

2  x2
 
x3 
0 
  1 
The component representation of the curl is
(2.2.15) curl v = (v3 , 2 – v2 , 3) e1 + (v1 , 3 – v3 , 1) e2 + (v2 , 1 – v1 , 2) e3
= vi , k kil el .
Similarly one can define the divergence of a tensor field T as that particular
vector field which acts on an arbitrary constant vector a as
a  div T = (div T)  a : = div (a  T) = div (TT  a) .
If we set a  ei , we obtain the i-th component of div T with (2.2.13)
ei  div T = div(ei  T) = div(Tim em) = Tim , k em  ek = Tim , m
and therefore
(2.2.16) div T = Tim , m ei ,
a vector field.
Similarly, one defines the curl of a tensor field T by its effect on a constant
vector a
(2.2.17) (curl T)  a : = curl(a  T) = curl(TT  a) .
If we set a  ei , then with (2.2.15)
(curl T)  ei = curl(ei  T ) = curl(Tij ej) = Tij , k kjl el
= Tmj , k lkj el  em  ei
and therefore
(2.2.18) curl T = Tmj , k lkj el  em

which is also a tensor field.
We will later need the tensor of incompatibility of a symmetric tensor field T
as
(2.2.19) inc T : = – curl curl T = ikj lmn Tjm , kn ei  el .
This gives again a 2nd-order tensor field.
One should note that the definitions of grad, div, and curl are not unique in the
literature.
We obtain the general rules for these operations:
 the gradient operation increases the order of a field by 1
 the divergence operation decreases it by 1
 the curl leaves it equal.
The chain rule also holds for gradients, divergence, and curl operations.
2.2.2 The Nabla Operator
In the literature, the calculus in linear spaces is often formalised by the nabla35
operator, which has a double function as a differential operator and as a (co-)
vector. This needs some explanation. Nabla is introduced as

(2.2.20)  := ei
xi
with respect to some fixed ONB {ei} and with tensor fields of arbitrary order
algebraically connected, i.e., by a scalar product or a simple contraction, a tensor
product, a cross-product, or another product.
The rules for the application are then:

1.) Apply the differential operator to all of these fields (if necessary by use
xi
of the product rule).
2.) Connect ei with the result according to the given algebraic product.
It is important to note that the order of these two steps is in general not
interchangeable.
As an example we choose the gradient of a scalar field after (2.2.4):
35
from greek  = harp
  x1 ,x2 ,x3 
(2.2.21) grad  (r) = ei =  , i ei =  =  .
xi
In this case, the algebraic product between  and  is a scalar multiplication of

a scalar with a (co)vector, where the scalar can be written before or after the
vector.
For a gradient of a vector field the order is no longer arbitrary, since we have to
use the tensor product in this case after (2.2.11), which does not commute,
v j  x1 ,x2 ,x3  
(2.2.22) grad v = ej  ek = vj(x1 , x2 , x3) ej  ek
x k xk
= v  .
For the divergence (2.2.13) and the curl (2.2.15) of a vector field we can write
(2.2.23) div v = tr(v  ) = v   =   v
(2.2.24) curl v =   v = – v  
and for the divergence of a tensor field (2.2.16)
(2.2.25) div T = T   =   TT
and for the curl of a tensor field (2.2.17)
(2.2.26) curl T =   TT = – (T  )T.
The tensor of incompatibility (2.2.19) is in nabla notation
(2.2.27) inc T : = – curl curl T =   T  
from which we see the identity
(2.2.28) div inc T =   T     = o .
The deformator (2.2.12) is
(2.2.29) def v : = ½ (grad v + grad T v) = ½ (v   +   v) .
If the differential operation of nabla does not apply to the whole term, but only to
a part of it, then we can indicate this by brackets or by a superimposed arrow. As
an example, we consider the gradient of two differentiable scalar fields  and 
 
( ) = ()  + ()  =    +  
after (2.2.31).
For all differentiable scalar fields  (r) ,  (r) , vector fields u(r) , v(r) , and
tensor fields T(r) , S(r) the following rules hold.
(2.2.30) grad( + ) = ( + ) =  +  = grad  + grad 

(2.2.31) grad( ) =  grad  +  grad 
(2.2.32) grad(u + v) = (u + v)   = u   + v  
= grad u + grad v
(2.2.33) div(u + v) = (u + v)   = u   + v   = div u + div v
(2.2.34) curl(u + v) =   (u + v) =   u +   v = curl u + curl v
(2.2.35) grad( v) = ( v)   =  (v  ) + v  ()
=  grad v + v  grad 
(2.2.36) div( v) = ( v)   =  (v ) + v  ()
=  div v + v  grad 
(2.2.37) grad(u  v) = (u  v)  = (  u)  v + (  v)  u
= v  (u  ) + u  (v  ) = v  grad u + u  grad v
= grad T(u)  v + grad T(v)  u
(2.2.38) div(u  v) = (u  v)   = u (v  ) + (u  )  v
= u div v + grad(u)  v
(2.2.39) div(T  v) =   (T  v) = (T  v)  
= T  (  v) + (  T)  v
= tr(T  grad v) + div(TT)  v = TT  grad v + div(TT)  v
(2.2.40) div(v  T) = (v  T)   =   (v  T)
= T  (v  ) + v  (T  )
= tr(TT  grad v) + div(T)  v = T  grad v + div(T)  v
(2.2.41) div( T) = ( T)   =  (T  ) + T  ()
=  div T + T  grad 
(2.2.42) div(T + S) = (T + S)   = T   + S   = div T + div S
With the nabla notation one sees the following identities for all differentiable
scalar fields  (r) , vector fields v(r) , and tensor fields T(r) .
(2.2.43) curl grad  =   ( ) = o

(2.2.44) curl grad T v =   (v  ) = (  v)   = grad curl v
(2.2.45) curl grad v =   (  v) = 0
(2.2.46) div curl v =   (  v) = (  )  v = 0
(2.2.47) curl curl v =   (  v) = (v  )  – (v  )  
= grad div v – div grad v
(2.2.48) div grad T v = (  v)   = (  v)  = grad div v
(2.2.49) div grad grad  = [()  ]   = [()  ] 
= grad div grad 
(2.2.50) div grad grad v = [(v  )  ]  = [(v  )  ]  
= grad div grad v
(2.2.51) div div grad v =   [(v  )  ] = [(v  ) ]  
= div grad div v
(2.2.52) div grad curl v = [(  v)  ]  =   [(v  )  ]
= curl div grad v
(2.2.53) div (curl T)T = – T     = o
The LAPLACE36 operator is defined as

(2.2.54)  := ,
which equals the operations div grad. Examples: Let  be a scalar field, then
 
(2.2.55)   =     = div grad  = ei  ek  =  , ii
xi xk
is a scalar field. Let v be a vector field, then
(2.2.56)  v = v  = v (  ) = v     = div grad v
is also a vector field.
36
Pierre Simon Laplace (1749-1827)
2.2.3 Cylindrical Coordinates
If one uses coordinates to describe the geometrical space, the simplest

coordinate system is a Cartesian one. However, if the problem under consideration
comprises a rotational symmetry, which is quite often the case, it is not
recommended to use Cartesian coordinates, but instead cylindrical ones. We will
therefore derive the most important representations in cylindrical coordinates.
Here, a point is described by three real numbers r ,  , z , where r is the
distance of the z–axis, z the height above the base plane, and  the angle of the
projection into this plane, which can take values between 0o and 360o. Note that
for both endpoints the angle is not unique. Therefore we use the open interval and
leave out points with   0 or with r  0 .
Cylindrical coordinates are rectangular and curved. If {x , y , z} are Cartesian
coordinates with associated ONB {ex , ey , ez} , then we obtain the transformations
x = r cos  y = r sin 
y
r2 = x2 + y2  = arctan
x
for r  0 , while z is identical in both systems.
ez
ey
 r
ex
In a point with coordinates {r ,  , z} we can introduce an associated local ONB

by
er ( ) : = cos  ex + sin  ey
e ( ) : = – sin  ex + cos  ey
ez
This gives for the differentials
de r
der = d = d e
d
de
de = d = – d er
d
The position vector is
r = x ex + y ey + z ez
= r er ( ) + z ez
and its differential
dr = dx ex + dy ey + dz ez
= dr er ( ) + r der ( ) + dz ez
= dr er ( ) + r d e ( ) + dz ez
The gradient of a scalar field  (r ,  , z) can be obtained by the differential
d (r ,  , z) = grad   dr
=  , r dr +  ,  d +  , z dz
= grad   [dr er ( ) + r d e ( ) + dz ez]
1
=  , r er  dr er +  ,  e  r d e  +  , z ez  dz ez
r
and by comparison
1
(2.2.57) grad  =  , r er +  ,  e +  , z ez .
r
The gradient of a vector field v(r,  , z) is obtained by its differential
dv(r , dr) = v , r dr + v ,  d + v , z dz
= grad v  dr
with the partial derivatives
v , r = (vi ei) , r = v r, r er + v , r e + v z, r ez
v ,  = (vi ei) ,  = v r,  er + v r er ,  + v,  e + v e ,  + v z,  ez
= v r,  er + v r e + v,  e – v er + v z,  ez
v , z = (vi ei) , z = v r, z er + v, z e + v z, z ez
By inserting them and comparison we obtain the matrix of components of grad v
as
 r
v , r 
1 r
r
v ,  v  
vr , z 
 
(2.2.58)  v ,
 r
r

1 
v ,  v r  v , z  .

 1 z 
vz , r v , vz , z 
 r 
The divergence is its trace
1
(2.2.59) div v = v r, r + (v ,  + v r) + v z, z
r
and the curl is

1 z
(2.2.60) curl v = ( v ,  – v, z ) er + (v r, z – v z, r ) e
r
1
+ [v, r + ( v  – v r,  )] ez .
r
The divergence of a tensor field is
1 r T rr  T 
(2.2.61) div T = (T rr, r + T , + + T rz, z ) er
r r
1  T r  T  r
+ (T r, r + T , + + T z, z ) e
r r
1 z T zr
+ (T zr, r + T , + + T zz, z) ez .
r r
Nabla has the following representation in cylindrical coordinates
  
(2.2.62)  = er + e + ez
r r  z
and the LAPLACE operator
2  2 2
(2.2.63)  :=  = + + 2 2 + 2 .
r 2 rr r  z
Problem 8. Cylindrical Coordinates I

The sketched windmill with only one rotor blade rotates with a constant
angular velocity  . The mass of the blade is m , and we approximate it by a
homogeneous bar with moment of inertia J with respect to the rotor axis.
We consider the load-free case. Determine the moment at the foundation A
caused by the imbalance of the blade.
Solution
First we cut the windmill free. At the foundation point we introduce the
torque mA(t) . This is the only torque acting on the mill. The angular
momentum with respect to A is
d(t) = m rAM (t)  rAM (t) + J  (t) ez .
The first term represents the motion of the centre of mass of the blade (after
STEINER´s37 theorem). The second term comes from the rotation of the
blade with respect to its centre of mass. The moment of inertia of a
37
Jakob Steiner (1796-1863)
homogeneous bar with respect to some axis perpendicular to its own axis
through the centre of mass is
J = m l 2/ 12 = m (2R)2/ 12 = m R2/ 3 .
The axis of rotation points in the ez–direction. After the balance of angular
momentum, the resulting moments equal the rate of moment of momentum
mA(t) = d(t) = [m rAM (t)  rAM (t)] + J (t) ez .
ey
m
 = t
ex
H rAM(t)
If the angular velocity is constant, then  (t) = 0 and only

mA(t) = m rAM (t)  rAM (t) + m rAM (t)  rAM (t)
(P8.1) = m rAM(t)  rAM (t)
remains. The position vector is assigned with respect to a basis in the hub
of the mill. Its distance from the foundation is most easily represented by a
fixed ONB, while for the blade a cylindrical system is preferable
rAM (t) = H ey + R er( ) .
The angle  results from the initial condition  (0) = 0 as  =  t . We
determine the acceleration rAM(t) by the chain rule using
de r ( ) d e (  )
= e ( ) = – er( )
d d
so that (Chapter 2.2.3)
rAM (t) = [H ey + R er ( )] = R  e ( )
rAM (t) = – R  2 er ( ) .
Now we use P8.1
mA(t) = m rAM (t)  rAM (t)
= – R  2 m [H ey + R er ( )]  er ( )
= – R  2 m H ey  er( ) ,
and with er = cos  ex + sin  ey
mA(t) = R  2 m H cos ez .
This must be supported by the foundation.
Problem 9. Cylindrical Coordinates II

Calculate the angular momentum with respect to the axis of rotation of a
homogenous circular plate rotating with angular velocity  around ez
with a thickness D , radius R , and mass density  .
Solution
We represent the angular momentum after (2.1.110) as a volume integral in
cylindrical coordinates
D R 2
d =    r  r  r d dr dz .
0 0 0
The position vector is r = r er( t) . We apply the chain rule and

de r ( )
= e ()
d
to obtain
r = r  e ( t) .
Then the angular momentum is
R 2
d =  D   r3 er  e d dr .
0 0
With er  e = ez , independent of r and  , we find

d = ½   D R4 ez .
2.2.4 Integral Transformations
are needed to transform volume integrals into surface integrals, or vice versa.
Divergence Theorem (GAUSSOSTROGRADSKI38) Let V be a three-
dimensional regular volumetric region in the EUCLIDean space, and A its
surface with outer normal n . Let further U be a tensor field of arbitrary order,
and  an arbitrary product between U and n . Then
(2.2.64)  U  n dA =  U   dV
A V
The following special cases will be important for us.
 U is a scalar field  and  the scalar multiplication of a scalar and vector.
Then
(2.2.65)   n dA =    dV =  grad  dV.
A V V
 U is a vector field u and  the scalar product between vectors. Then
(2.2.66)  u  n dA =  u   dV =  div u dV.
A V V
We give a sketch of the proof of this form, from which the other forms can be
easily derived. We write down the component form with respect to some ONB
 u1 u2 u3 
 div u dV =      dx1 dx2 dx3 .
V V  x1 x2 x3 
x2
A 23 A '' A'
dA23 dA'' dA'

projection ray
n'' P'' P' n'
x1
x3
38
Carl Friedrich Gauß (1777-1855), Mikhail Vasilevich Ostrogradski (1801-1862)
We first assume that the region V is convex. If we project this region into the
x2–x3–coordinate plane, we obtain the area A23 . We next decompose the surface
of V into its positive part A ' with respect to x1 and its negative part A ''. A
projection ray parallel to the x1–axis penetrates the surface A of V twice at the
points P' and P''. In these points we notate the elements of area as dA' and dA''
and the normal vectors as n' and n'' , respectively. The projections of these area
elements into the x2–x3–plane shall be dA23 = dx2 dx3 . For them we obtain the
relation
dA23 = dA' cos(n', e1) = dA'' cos(n'', – e1)
wherein the cosine of the angle spanned by n' and e1 can be calculated by the
scalar product
cos(n', e1) = n'  e1
and analogously
cos(n'', – e1) = – n''  e1
so that
dA23 = n'  e1 dA' = – n''  e1 dA''.
The first term of the integral is by partial integration
P'
u1 u1

V x1
dx1 dx2 dx3 = 
A23
 x1
dx1 dA23
P' 
=  [u1 (P' ) – u1 (P'' )] dA23

A23
=  u1 n'  e1 dA' –  u1 (– n''  e1) dA''

A A 
=  u1 e1  n dA .
A
For the other two terms of the integral we obtain analogously

u2
 dx1 dx2 dx3 =  u2 e2  n dA
V x2 A
and
u3
 dx1 dx2 dx3 =  u3 e3  n dA
V x3 A
and for the sum of the three
 div u dV =  u  n dA .
V A
If the region is not convex, the projection ray penetrates the surface possibly more
than twice. Then we can also project these parts of A into the e2 e3plane.
Finally we will obtain again the same formula. If the region contains internal
voids, then one must also integrate over the internal surfaces.
For u  v  T with a vector field v and a tensor field T we obtain by the
divergence theorem with (2.2.40)
 v  T  n dA =  div (v  T) dV
A V
(2.2.67) =  [T  (v  ) + v  (T  )] dV
V
=  (T  grad v + v  div T) dV.

V
 U is a vector field u and  the cross-product between vectors. Then
 u  n dA =  u   dV = –  curl u dV
A V V
or
(2.2.68)  n  u dA =  curl u dV
A V
hold.
 U is a vector field u and  the tensorial product between vectors. Then
(2.2.69)  u  n dA =  u   dV =  grad u dV
A V V
holds.
 U is a tensor field of arbitrary order and  a simple contraction. Then
(2.2.70)  U  n dA =  U   dV =  div U dV.
A V V
3 FOUNDATIONS OF CONTINUUM MECHANICS
Literature
Altenbach, H.: Kontinuumsmechanik. Springer Vieweg, 2. edt., Berlin (2012)

Betten, J.: Kontinuumsmechanik. Springer, Berlin (1993), 2. edt. (2001)
Gonzalez, O.; Stuart, A. M.: A First Course in Continuum Mechanics.
Cambridge University Press (2008)
Karasudhi, P.: Foundations of Solid Mechanics. Kluwer Acad. Publ., Dordrecht
(1991)
Lai, W. M.; Rubin, D.; Krempl, E.: Introduction to Continuum Mechanics.
Pergamon Press, Oxford (1993)
Parisch, H.: Festkörper-Kontinuumsmechanik: Von den Grundgleichungen zur
Lösung mit Finiten Elementen. Teubner (2003)
Shames, I. H.; Cozzarelli, F. A.: Elastic and Inelastic Stress Analysis. Prentice
Hall, Englewood Cliffs (1992)
Stark, R.: Festigkeitslehre - Aufgaben und Lösungen. Springer (2006)
3.1 Kinematics
In Continuum Mechanics we consider material bodies which continuously

occupy a region in the three-dimensional EUKLIDean space. On its boundary and
in its interior, one defines physical quantities as fields. Its points are called
material points.
Xi xi
u u
e2
X x
reference placement e1 current placement

e3

DOI 10.1007/978-3-319-19566-7_3
122 3 Foundations of Continuum Mechanics
In the referential description one considers the body in its current placement
with respect to some arbitrarily chosen reference placement at some reference
time t0 . So let  be that particular mapping which maps the position vector X
of some material point into its position vector x at the time t in the current
placement
(3.1.1) x = (X , t) .
At the reference time t0 we have particularly
(3.1.2) X = (X , t0) .
The difference vector
(3.1.3) u(X , t) : = (X , t) – (X , t0) = (X , t) – X
is called displacement vector. After choosing some ONB we obtain the
representations of the components
X = Xi ei
x = xi ei
u = ui ei = (xi – Xi ) ei
and in components
(3.1.4) xi = (X , t)  ei = : i (X1 , X2 , X3 , t)
with
(3.1.5) Xi = i (X1 , X2 , X3 , t0) .
Since the Cartesian coordinates {Xi} ( components of the position vector X
with respect to the ONB {ei}) for a material point in the reference placement are
time-independent, they describe at all times the same material point and are thus
called material coordinates. In contrast to them, the Cartesian coordinates {xi}
( components of the position vector x with respect to an ONB) fix at all times
the same spatial point. {xi} are called spatial coordinates.
If we keep the position vector X in (X , t) fixed and let the time t run, then
the value describes the path of the material point X moving through the space.
The function  must be invertible in the region occupied by the body at a fixed
time t since two material points cannot occupy the same spatial point at the same
time. We write for its inverse at a fixed time
(3.1.6) X = –1(x , t) if x = (X , t)
or for the components
(3.1.7) Xi = i–1(x1 , x2 , x3 , t) if xi = i (X1 , X2 , X3 , t) .
Let  be a physical quantity defined in each material point and being
occasionally time-dependent. The representation of this quantity as a function of
the material coordinates
(3.1.8)  L(X , t) =  L(X1 , X2 , X3 , t)
3.1 Kinematics 123
is called the material or LAGRANGE39an representation of  . If we transform

the material coordinates into the spatial coordinates of the same material point,
then we obtain the spatial or EULERian representation of  as
(3.1.9)  L(X1 , X2 , X3 , t)
=  L(1–1(x1 , x2 , x3 , t) , 2–1(x1 , x2 , x3 , t) , 3–1(x1 , x2 , x3 , t) , t)
= :  E(x1 , x2 , x3 , t)
=  E(1(X1 , X2 , X3 , t) , 2(X1 , X2 , X3 , t) , 3(X1 , X2 , X3 , t) , t) .
We have in particular
x = L(X , t) = E(x , t)
so that the last function is trivially the identity in the spatial argument.
In the spatial description, the spatial coordinates {xi} describe a point in space,
which can be occupied by different material points in the course of time.  E is
therefore the value of a quantity  somewhere in space, regardless of the material
point which happens to be there at the instant of time. Since we can invert i or
i–1 , we can invert the spatial coordinates into material ones, and vice versa, so
that both descriptions are essentially equivalent. It depends only on the particular
application which representation becomes more practical.
If we consider the rate of  L for some material point X , then we obtain the
material time derivative
(3.1.10)  L(X , t) : =  L(X , t) / t =  L(X1 , X2 , X3 , t) / t =  L , t
as the partial derivative of the LAGRANGEan function  L with respect to the
time-variable. With the transformation (3.1.9) into the EULERian representation,
this leads by the chain rule to
 L(X , t)
: =  E(1(X1 , X2 , X3 , t) , 2(X1 , X2 , X3 , t) , 3(X1 , X2 , X3 , t) , t)
=  E(x1 , x2 , x3 , t) / t +  E(x1 , x2 , x3 , t) / xi i / t
=  E(x1 , x2 , x3 , t) / t +  E(x1 , x2 , x3 , t) / xi i
=  E(x1 , x2 , x3 , t) / t +  E(x1 , x2 , x3 , t) / xi vi
=  E , t +  E , i vi
=  E , t +  E , j ej  vi ei =
E
(3.1.11)  = + grad  E  vE
t
with the velocity field
39
Jean Louis Lagrange (1736-1813)
i
(3.1.12) v = vi ei : = (X , t) = i (X , t) ei = ei = ui ei .
t
E
Accordingly, the substantial rate  E of  E consists of the local rate
t
and the convective rate grad  E  v due to the flux of material points with
different values of  .
 If  E is a scalar field, then grad  E is a vector field, the product of which
with v gives again a scalar field.
 If  E is a vector field, then grad  E is a tensor field, and a simple
contraction with v gives again a vector field.
 If  E is a tensor field of K–th-order, then grad  E is a tensor field of K+1–
th-order, and its simple contraction with v gives again a K–th-order tensor
field.
In all cases, the convective rate is a tensor field of the same order as  E .
In the LAGRANGEan representation the derivatives with respect to space and to
time can be interchanged. In the EULERian description the derivatives with
respect to space can only be interchanged with the local rate.
Examples
1.) Let    be the current mass density in some material point
dm dm dV0 dV0
 := = = 0
dV dV0 dV dV
where 0 : = dm / dV0 is the mass density in the reference placement. We

assumed conservation of mass
dm = dm0 .
The substantial rate of the density becomes
 E
(3.1.13)  E = + grad  E  v E .
t
2.) Let    be the field of the position vectors. Its time-derivative is the velocity
field v with its representations
χ L χ E
 = v L = = + grad E  vE
t t
χ E
= + I  vE = vE .
t
Since E(x , t) is that particular mapping which maps x into itself at all times, its
partial time-derivative is zero.
3.) Let   v be the velocity field. Its time-derivative is the acceleration field a
with the representations
3.1 Kinematics 125
 vE
a = v = + grad vE  vE .
t
The gradient of the displacement vectors is after (2.2.11) the field of the
displacement gradient
ui
(3.1.14) H(X , t) : = Grad u(X , t) = ei  ek
X k
with its matrix of components
 u1 u1 u1 
 
 X 1 X 2 X 3 
 u2 u2 u2 
 
 X 1 X 2 X 3 
 u3 u3 u3 
 
 X 1 X 2 X 3 
with respect to some ONB. After
(3.1.15) H(X , t) = Grad ((X , t) – X) = Grad (X , t) – Grad X
= F(X , t) – I
it is related to the deformation gradient
 i
(3.1.16) F(X , t) : = Grad (X , t) = ei  ek .
X k
Herein we denote by Grad the derivative with respect to the material coordinates
{Xi} , the same as the other differential operators Div, Curl, Def, and Inc. The
corresponding material or LAGRANGEan nabla is then

(3.1.17) L = ek .
X k
This operator can only be applied to fields in the LAGRANGEan description.
This must be distinguished from grad, div, curl, def, and inc, which refer to the
derivatives with respect to the spatial coordinates {xi} . The spatial or
EULERian nabla is

(3.1.18) E = ek
xk
which can only be applied to fields in the EULERian description.
Between the two nablas we have the relation after the chain rule
   i  j 
L = ek = ek = ( ek  ej )  ( ei)
X k xi X k X k xi
(3.1.19) = FT  E = E  F .
For a fixed time t , we obtain after (2.2.10) the relation between the material and
the spatial vectorial line element
(3.1.20) dx = d(X , t , dX) = F(X , t)  dX (EULER 1762)
F maps the line elements dX of the reference placement into those of the current
placement dx . In direct notation we can write this as
dx d d dx
F = and L = =  = E  F .
dX dX dx dX
The square of the length of a line element dx is
dx2 = dx  dx = (F  dX)  (F  dX) = dX  FT  F  dX
and its difference with the one in the reference placement is
dx2 – dX2 = dx  dx – dX  dX
(3.1.21) = dX  FT  F  dX – dX  dX = dX  (FT  F – I)  dX .
FT  F – I is therefore a measure for the change in length of the line element
between reference and current placement.
If we consider the scalar product between two line elements
dx  dy – dX  dY = dX  (FT  F – I)  dY
it reflects also the change of the angle between the two line elements dX and
dY , which can be described by the (symmetric) GREEN40´s strain tensor
(3.1.22) EG : = ½ (FT  F – I) = ½ [(I + H)T  (I + H) – I]
= ½ ( H + HT + HT  H)
which is non-linear in H and therefore also non-linear in the displacements u . If
we linearise it for small deformations, we obtain the displacement-strain
relations (DSR) with the (infinitesimal) strain tensor
(3.1.23) E : = Def u = ½ (H + HT ) = ½ (u  L + L  u)
ui uk
= ½( + ) ei  ek = : ik ei  ek ,
X k X i
which is the symmetric part of the displacement gradient. With respect to some
ONB this has the component matrix
40
George Green (1793-1841)
3.1 Kinematics 127
 u1 1  u1 u2  1  u1 u3  

      
 X 1 2  X 2 X 1  2  X 3 X 1  
  u3  
 1  u2  u1  u2 1  u2
    .
 2  X 1 X 2  X 2 2  X 3 X 2  
 
 1  u3 u1  1  u3 u2  u3 
 2  X  
X 3 

2  X 2
 
X 3  X 3 
  1 
If the motion is a pure translation, then the displacement field is at all times
constant in space
u(X , t)  uo(t)
so that
H  0 and F  I
G
and E  E  H  0.
For the case of a pure (rigid) rotation neither lengths nor angles between line
elements change, and therefore EG  0 and FT  F  I , so that F turns out to be
orthogonal in this case. By an appropriate choice of an ONB with e3 as axis of
rotation and  its angle, we obtain the matrices
 cos   sin  0 
[F ij] =  sin  cos  0 

 0 0 1
 cos   1  sin  0 
[H ij] =  sin  cos   1 0 

 0 0 0 
 cos   1 0 0

[ij] =  0 cos   1 0 

 0 0 0 
Only for small angles of rotation with cos   1 one obtains

E0,
a peculiarity due to the linearisation. The skew part of the displacement gradient is
here
 : = ½ (H – HT ) = ½ (F – I – FT + I) = ½ (F – FT )
with the antisymmetric matrix of components
 0  sin  0  0   0
[ ij] =  sin  0 0  

 
0 0

 0 0 0   0 0 0 
which can be non-zero even for small rotations. It describes the (linear part of)
rigid rotations.
An (infinitesimal) rigid body motion consists of a translation and a rotation (see
EULER´s velocity formula (2.1.114)). The corresponding (infinitesimal)
displacement field can be represented as
(3.1.24) u(X , t ) = (t)  (X – X0) + c(t)
with (t) time-dependent skew tensor
X0 position vector of some reference point
c(t) time-dependent vector.
In the theory of small deformations or the geometrically linear theory we use
the infinitesimal or linearised GREEN´s strain tensor E for the local description
of deformations, thus neglecting all terms of higher-order than the linear ones.
As for any symmetric tensor, we can also find a spectral form for
3
(3.1.25) E =   pi e pi  e pi
i 1
with  p
i : principal strains
e p
i : principal axes of strain (ONB).
If we represent the volume element by a triple product with respect to the principal
strain axes, and assume H  E (since rotations do not alter the volume), we obtain
dV – dV0 = dx1 dx2 dx3 – dX1 dX2 dX3
= [dx1 , dx2 , dx3] – [dX1 , dX2 , dX3]
= [F  dX1 , F  dX2 , F  dX3] – [dX1 , dX2 , dX3]
= [(I + H)  dX1 , (I + H)  dX2 , (I + H)  dX3] – [dX1 , dX2 , dX3]
= [(I + E)  dX1 , (I + E)  dX2 , (I + E)  dX3] – [dX1 , dX2 , dX3]
= [(1 +  p1) dX1 , (1 +  p2) dX2 , (1 +  p3) dX3] – [dX1 , dX2 , dX3]
= (1 +  p1) (1 +  p2) (1 +  p3) dX1 dX2 dX3 – dX1 dX2 dX3
 ( p1 +  p2 +  p3) dX1 dX2 dX3
(3.1.26) = tr(E) dV0 = tr(H) dV0 = Div u dV0 .
In the linear theory, the trace of the strain tensor represents dilatations or
compressions of the volume element
3.1 Kinematics 129
ui dV  dV0 dV
(3.1.27) Div u = = tr H = tr E  = –1
X i dV0 dV0
0
= –1.

Consequently, the spherical part of E stands for dilatations, and its deviatoric part
E' for isochoric deformations or distortions.
Example: Compression or Dilatation

A pure compression ( < 0) or dilatation ( > 0) is characterized by a spherical
tensor
H  E  I and   0 with tr H = tr E  3  .
Example: Simple Shear

is defined as a displacement field that allows for a representation with respect to
some ONB as
ux =  Y , uy = uz = 0
with the shear number  . As a result we obtain the displacement gradient
H =  ex  ey
and the strain tensor
E = sym( ex  ey) =  sym(ex  ey) = ½  (ex  ey + ey  ex)
both being constant in space, and traceless or deviatoric
tr E = 0 .
Y

1
Thus, simple shear is volume-preserving or isochoric. In contrast to this,

GREEN´s strain tensor for simple shear contains a quadratic term
EG = ½ ( ex  ey +  ey  ex +  2 ey  ey) .
Of course, these simple forms hold only for a particular ONB. For others they may
look quite different. In principle, one can apply the following theorem to all
simple shears.
Theorem. BLINOWSKI/ RYCHLEWSKI (1998)

The strain tensor E describes a simple shear if and only if one of the following
equivalent conditions is fulfilled.
 There exists an ONB with respect to which E has the following matrix of
components
 s 0 0
0  s 0  .
 
0 0 0 
 There exists an ONB with respect to which E has the following matrix of
components
0 s 0 
 s 0 0 .
 
0 0 0 
 tr E = 0 and det E = 0 .
 tr E = 0 and tr(E3) = 0 .
 There exists an orthogonal tensor Q such that E =  Q E QT holds.
 E is plane41 and deviatoric.
All distortions can be interpreted as a superposition of only two simple shears
according to the following theorem.
Theorem. Each strain tensor E can be additively decomposed into
(3.1.28) E = 1/3 tr(E) I + sym(a1  b1) + sym(a2  b2)
with
 a dilatoric part 1/3 tr(E) I and
 two simple shears ai  bi with ai  bi = 0 , i = 1 , 2 .
Proof. First of all, we can decompose any symmetric tensor into its spherical part
 I and its deviatoric part E' . We can bring the latter into its spectral form
3
E' =   ' pi e pi  e pi
i=1
with an arbitrary order of the eigenvalues. Because of its tracelessness we have

 ' p3 = –  ' p1 –  ' p2
and therefore
E' =  ' p1 e p1  e p1 +  ' p2 e p2  e p2 – ( ' p1 +  ' p2) e p3  e p3
41
see next page.
3.1 Kinematics 131
=  ' p1 (e p1  e p1 – e p3  e p3) +  ' p2 (e p2  e p2 – e p3  e p3)

=  ' p1 sym[(e p1 + e p3)  (e p1 – e p3)] +  ' p2 sym[(e p2 + e p 3)  (e p2 – e p3)].
The terms in brackets are pairwise orthogonal and thus simple shears; q.e.d.
So all states of strain are superpositions of dilatations and simple shears.
However, a superposition of simple shears is in general not a simple shear again.
In some cases we can reduce the dimension of the problems.
Definition. A state of plane strain is such that all displacements are parallel to
some plane and independent of the distance to this plane.
If we choose a Cartesian coordinate system such that the X1 –X2 –plane is this
plane, then the state of plane displacements can be represented by the two
functions
u1(X1 , X2) u2(X1 , X2) u3 = 0 .
This gives a reduced form for the strain tensor
 u1 1  u1 u2  
    0
 X 1 2  X 2 X 1  
 1  u u2  u2 
(3.1.29) [ik] =   1   0 .
 2  X 2 X 1  X 2 
 
 0 0 0
 
 
Some materials like viscous fluids are sensitive to the rate at which the
deformation process is performed. In order to find appropriate variables to
describe such a behaviour, we take the material time-derivative of EULER´s
equation (3.1.20)
dx = F   dX
= F   F –1  dx
= (Grad (X , t))  F –1  dx
= Grad((X , t))  F –1  dx
= Grad vL  F –1  dx
= (vL  L)  F –1  dx
with (3.1.19) = vE  E  dx
= grad vE  dx
= L  dx
with the spatial velocity gradient
(3.1.30) L(x , t) : = grad vE = vE  E = F   F –1

which has a matrix-representation of its components
 v1 v1 v1 
 
 x1 x2 x3 
 v2 v2 v2 
 
 x1 x2 x3 
 v3 v3 v3 
 
 x1 x2 x3 
with respect to some ONB. We decompose this tensor into its symmetric part, the
rate of deformation tensor
vi vk
(3.1.31) D = ½ (L + LT ) = ½ ( + ) ei  ek = def (v)
xk xi
with the components
 v1 1  v1 v2  1  v1 v3  
      
 x1 2  x2 x1  2  x3 x1  
 
 1  v2  v1  v2 1  v2 v3  
   
 2  x1 x2  x2 2  x3 x2  
 
 1  v3 v1  1  v3 v2  v3 
 2  x  
x3 

2  x2
 
x3  x3 
  1 
and its skew part, the spin tensor or vorticity tensor
vi v
(3.1.32) W = ½ (L – LT) = ½ ( – k ) ei  ek
xk xi
with the components
 1  v1 v2  1  v1 v3  
 0      
 2  x2 x1  2  x3 x1  
 
 1  v2  v1  1  v2 v3  
 0    .
 2  x1 x2  2  x3 x2  
 
 1  v3 v1  1  v3 v2  
 2  x  
x3 

2  x2
 
x3 
0 
  1 
For the interpretation of the two parts, we remember that for a rigid rotation the
deformation gradient is orthogonal F  Q , and
dx = Q(t)  dX = Q(t)   Q(t)T  dx = W(t)  dx .
3.1 Kinematics 133
Therefore, for rotations, L  W is skew, and D  0 . Thus, the spin tensor

describes rigid rotations, which do not affect the rate of deformation tensor.
We have already seen that the scalar product contains the deformation of the
lengths and angles between line elements. Its time-derivative is
(dx  dy) = dx  dy + dx  dy
= L  dx  dy + dx  L  dy
= dx  (L + LT )  dy
= dx  2 D  dy .
If we consider as a special case two line elements which coincide with the first
base vector dx  e1  dy , then we get the first diagonal component of D
(dx  dy) = e1  2 D  e1 = 2 D11 .
Thus, the diagonal components Dii of D can be interpreted as the stretching rates
of line elements currently aligned to ei . On the other hand, if we set dx  e1 and
dy  e2 , then in
(dx  dy) = (e1  e2)  = cos(1,2) = e1  2 D  e2 = 2 D12
the rate of change of the (currently orthogonal) angle between line elements in e1
and e2 direction is also involved. This is known as the shear rate.
In fact, in the example of simple shear, we have
v =   y ex
and thus
L =   ex  ey
D = ½   (ex  ey + ey  ex)
W = ½   (ex  ey – ey  ex)
and
div v = div (u) = tr L = tr D = 0
since shear is isochoric.
If  is an arbitrary differentiable field, then its gradient is after (3.1.19)

Grad  L =  L  L = ( E  E)  F = ( E  E)  (I + H)
= grad  E  (I + H) .
By assuming small deformations: H << I , the difference between Grad  L
and grad  E becomes negligible
(3.1.33) Grad  L  grad  E
the same as
Div  L  div  E
Def  L  def  E
Curl  L  curl  E .
In these cases we can notate the partial derivatives with respect to the i-th
coordinate by ", i " since a distinction between Xi and xi is not necessary.
As an example, we obtain for the velocity gradient
L(x , t) = grad vE  Grad vL = vi , k ei  ek = Grad (u)
= (Grad u) = H
with the matrix of components with respect to some ONB
 v1 ,1 v1 ,2 v1 ,3 
v , v2 , 2 v2 ,3 
 21
 v3 ,1 v3 ,2 v3 ,3 
and
D  E .
3.1.1 Compatibility Conditions
If one knows the displacement field u = ui ei , one can determine

E = Def u = ½ (ui /Xk + uk /Xi) ei  ek = ik ei  ek
with its six independent components ik . Sometimes the inverse problem has to
be solved, namely to determine six component fields of E such that they can be
integrated to obtain a displacement field. However, this is not possible for all
symmetric tensor fields, but only for compatible ones. These are the ones for
which the compatibility condition is fulfilled
(3.1.34) Inc E = 0 = L  E  L = – Curl Curl E
after (2.2.27). Since this tensor is symmetric and with zero divergence, this is
equivalent to three independent equations. With respect to some ONB we obtain
 
0 = L  E  L = ei  kl ek  el  em
X i X m
= kl , im ei  ek  el  em = kl , im ikp lmr ep  er .
One has to discriminate between the components ij of E (3.1.23) with two
indices, and the permutation symbol kij after (2.1.2) with three indices. We
obtain the following compatibility equations componentwise
3.1 Kinematics 135
11 , 22 + 22 , 11 = 2 12 , 12

22 , 33 + 33 , 22 = 2 23 , 23
33 , 11 + 11 , 33 = 2 31 , 31
11 , 23 + 23 , 11 = 31 , 12 + 12 , 13
22 , 31 + 31 , 22 = 12 , 23 + 23 , 21
33 , 12 + 12 , 33 = 23 , 31 + 31 , 32
Theorem. (DE ST. VENANT42, BELTRAMI43)
The compatibility condition Inc E = 0 is necessary and sufficient for the existence
of a displacement field u for simply connected regions.
Proof. a) necessity
Let u be a displacement field and
E = ½ (u  L + L  u) .
Then we have because of (2.2.45)
L  L  u = 0
and
u  L  L = 0
and therefore
Inc E = 0 .
b) sufficiency (BELTRAMI 1886)
Let us assume that for some symmetric tensor field E
– Inc E = Curl Curl E = 0
holds, or with A : = Curl E
Curl A = 0 .
Then because of the symmetry of E we have with respect to some ONB
tr A = tr(Curl E) = tr(L  ET)
= tr(L  E) = tr( /Xk ek  ij ei  ej) = tr(ij , k ek  ei  ej)
= tr(ij , k kim em  ej) = ij , k kij = ji , k kij = ij , k kji = 0 .
Accordingly, A is a traceless and irrotational tensor field. We set ai : = ei  A for
i = 1, 2, 3, so that A = ei  ai . Then we obtain with (2.2.17)
Curl ai = Curl(ei  A) = Curl(AT  ei) = Curl(A)  ei = o
42
Adhémar Jean Claude Barré de Saint-Venant (1797-1886)
43
Eugenio Beltrami (1835-1900)
so that all vector fields ai are irrotational or conservative fields. Such fields allow
for a potential vi
ai = Grad vi .
Then we get with (2.2.35)
A = ei  ai = ei  Grad vi = Grad(vi ei) = Grad v
with v : = vi ei . Since with (2.2.13)
tr A = 0 = Div v
v is divergence-free. Let  be the antimetric tensor field, the axial vector of
which is v after (2.1.53) so that for all constant vectors x
vx =  x
holds, then
(Curl )  x = Curl(T  x)
= – Curl(  x)
= – Curl(v  x)
= – L  (v  x)
and by the rule of double-cross-products
= – v (L  x) + x (L  v)
= – v  L  x + x (L  v)
= – (Grad v)  x + (Div v) x
= – (Grad v)  x
so that
Curl  = – Grad v = – A .
We get
Curl(E + ) = Curl E + Curl  = A – A = 0 .
We have already seen that for each irrotational tensor field we can find a vector
field u such that
E +  = Grad u
holds. The symmetric part of this equation is the wanted strain tensor
E = Def u
q. e. d.
It is interesting that in the above proof no linearisation has been made. So the
theorem holds for arbitrary large deformations. Only the use of the linear strain
tensor E is limited to small deformations.
3.1 Kinematics 137
An analogue compatibility problem arises if one wants to determine the three

components of the velocity field by the six components of the rate of deformation
tensor. Here the compatibility conditions are analogously
D11 , 22 + D22 , 11 = 2 D12 , 12
etc., or
(3.1.35) E  D  E = 0 = inc D .
Problem 10. Compatibility Conditions

Let the components of the strain tensor E be given with respect to some
ONB {ei} as
 3x 2 /2  ye  x /4 0
 
(P10.1) [ij] =   ye  x /4 1/2  e  x /2 0 
 
 0 0 0
 .
Try to find a corresponding displacement field with components u, v, w,

fulfilling the condition u(x  o) = o .
Solution
If the compatibility conditions (3.1.34) are fulfilled, the existence of a
displacement field u is ensured such that E = def u holds. Since E does
not depend on the zcoordinate and describes a plane strain state, only one
compatibility condition of (3.1.34) remains non-trivial
xx , yy + yy , xx = 2 xy , xy .
Insertion of the components from P10.1 and differentiation shows
immediately that compatibility holds. We will next determine the
displacement field. First we see
u =  xx dx = x3/2 .
The integration constant must be zero because of u(o) = o . In the same
way we determine
v =  yy dy = y (e x –1) / 2 .
This gives the desired component of the strain
xy = ½ (u , y + v , x) = – y e x / 4 .
We will now visualize this displacement field. We depict the undeformed
and the deformed placement of a representative region {0 < y < 1,
0 < x < 1 , z arbitrary} in the xyplane.
Problem 11. Torsion

Find the displacement gradient H and the strain fields E and EG of a
torsional shaft with constant twist D along the e3direction . The boundary
conditions for the displacements are u(X1 , X2 , 0) = o .
e1
e2 e3
Solution
First of all we determine the motion x = (X) . Obviously the motion of
the cross-sections with different coordinates X3 can only differ from the
initial placement by a rigid rotation (no warping). We make the linear
ansatz
x = (X) = Q  X
with an orthogonal tensor Q(X3) , which describes the rotation around the
e3axis. The twist angle is supposed to be linear in X3 so that
Q = cos(D X3) (e1  e1 + e2  e2)
+ sin(D X3) (e2  e1 – e1  e2) + e3  e3 .
3.1 Kinematics 139
The displacement field is then

u = x–X
= [X1 cos (D X3) – X2 sin (D X3) –X1] e1
+ [X1 sin (D X3) + X2 cos (D X3) –X2] e2 .
Its gradient is with the material coordinates Xi
[Hij] =
 cos( DX 3 )  1  sin( DX 3 )  X 1Dsin( DX 3 )  X 2 Dcos( DX 3 )

 sin( DX ) cos( DX 3 )  1 X 1 Dcos( DX 3 )  X 2 Dsin( DX 3 ) 
 3 
 0 0 0 
with respect to the basis {ei  ej} . For E = sym(H) we find the
component matrix
 cos( DX 3 )  1 0 (  X 1 Dsin( DX 3 )  X 2 Dcos( DX 3 )) / 2 

 cos( DX 3 )  1 ( X 1 Dcos( DX 3 )  X 2 Dsin( DX 3 )) / 2 
 
 sym 0 
which still depends on the axial coordinate X3 .
This finding contradicts the assumed constant twisting. The reason for this
contradiction is the rigid rotation depending on X3 , which enters the
linear strain tensor and disappears only for small rotations. For X3  0 one
can clearly see the two shears in the X3plane
E = – D X2 /2 (e1  e3 + e3  e1) + D X1 /2 (e2  e3 + e3  e2) .
The non-linear GREEN´s strains become after (3.1.22) and multiple use of
the trigonometric identity sin2(a) + cos2(a) = 1
 0 0  DX 2 /2 
 
[EijG] =  0 0 DX 1 /2 .
 2 2 2 
  DX 2 /2 DX 1 /2 D ( X 1  X 2 )/2 
We find a stretch in the e3direction, which depends on the radius but not
on X3 , as one would probably expect. For small D we can neglect D2
compared to D , so that axial stretch in the e3direction is also negligible
and EG converges to E .
With respect to cylindrical coordinates materially {R ,  , Z} and spatially
{r ,  , z} , we obtain for the position vectors
x = r er ( ) + z ez
X = R eR ( ) + Z eZ
Torsion is described by r = R , z = Z , and  =  + DZ , which is locally

like simple shear. The nabla operator with respect to cylindrical
coordinates and a normed basis is
 1  
L = eR + e + eZ
R R Θ Ζ
so that
H = (x – X)  L
 1  
= R [er( ) – eR( )]  ( eR + e + eZ )
R R Θ Ζ
= [er( ) – eR( )]  eR( )
 
+ (er( ) ,  – eR( ) ,  )  e ( ) + R er( ) ,   eZ
Θ Z
= [er( ) – eR( )]  eR( )
+ [e ( ) – e ( )]  e ( ) + RD e ( )  eZ
using
er ,  = e e ,  = – er  =  + DZ .
For    , the remaining term R D e ( )  eZ stands for a simple shear
proportional to R in the tangential direction in the eZ plane.
3.2 Stress Analysis
We will apply the Principle of Cuts after which we can cut every body out of
the universe, and substitute the mechanical influence of the outer world by
applying forces. In continuum mechanics one understands all forces acting on a
body as resulting from continuously distributed forces (force densities), which
either act on the interior of the body as body forces like, e.g., gravity, or on the
surface as contact forces such as the air pressure.
Let us first consider the body forces. Let m be the mass of a subbody and
fm the body force acting on it. The specific body force is defined as the limit
dfm fm
b = := lim
dm m0 m
so that vice versa
(3.2.1) fm =  b dm =  b  dV
V V
is the resulting body force acting on a body with volumetric region V . The most
important example is gravitation for which b is the gravitational field pointing to
the centre of the earth and being almost constant for small bodies. Another
example is given by magnetic attraction.
The other class of forces contains the contact or surface forces fc . Let A be a
part of the surface of size A , and fc the contact force acting on it. Its density
is the traction
dfc fc
t = := lim .
dA A0 A
The resulting contact force is the integral over the tractions over the entire body
surface A
(3.2.2) fc =  t dA .
A
The balance of linear momentum (2.1.111) states that the rate of linear
momentum of the body
p =  a dm
V
equals the resulting force acting on the body
p = f = f c + f m .
The integral form of this balance is then
(3.2.3)  a dm =  t dA +  b dm .
V A V
In order to further specify the concept of tractions, we assume that t depends on
place and time, but also on the orientation of the surface in space expressed by the
normed outer normal vector n .
We next cut a small tetrahedron out of the interior of the body at an arbitrary but
fixed point. For this purpose we put an ONB into the point and consider the
tetrahedron spanned by the three axes with lengths x1 , x2 , x3 .
If t is the mean value of the tractions of the triangle spanned by x1 and x2
with surface area A3 = ½ x1 x2 , it depends on the outer normal n  – e3 .
The resulting contact force on this triangle is after the mean value theorem of
integration calculus
t(– e3) A3 = t(– e3) ½ x1 x2
acting somewhere within this triangle. For brevity, we did not note the point and
time argument.
e2 x2
t(– e1) t(– e3)
t(n4)
e3 e1
x3 x1
t(– e2)
Analogously we obtain for the contact forces on the other sides

t(– e2) A2 = t(– e2) ½ x1 x3
t(– e1) A1 = t(– e1) ½ x2 x3 .
The fourth side has the normal n4 and the area A4 . For a closed surface, the
divergence theorem (2.2.65) gives with   1
(3.2.4)  n dA = o = – e1 A1 – e2 A2 – e3 A3 + n4 A4

A
and thus
n4 = (e1 A1 + e2 A2 + e3 A3) / A4 .
We now apply the balance of linear momentum to this tetrahedron
t(– e1) A1 + t(– e2) A2 + t(– e3) A3 + t(n4) A4 +  b V =  a V
with the volume
V = 1/6 x1 x2 x3
also using mean values for  b and a . We now consider the limit for xi  0 , i
= 1 , 2 , 3 . The mass-specific parts are of higher-order and vanish in the limit if
compared to the surface parts
t(– e1) A1 + t(– e2) A2 + t(– e3) A3
+ t((e1 A1 + e2 A2 + e3 A3 ) / A4) A4 = o
or
(t(– e1) A1 + t(– e2) A2 + t(– e3) A3) / A4
=  t( (e1 A1 + e2 A2 + e3 A3 ) / A4) .
This is fulfilled for arbitrary values of xi and arbitrary areas Ai if and only if
t is a linear function of n . We introduced linear functions between vectors as
tensors.
Theorem of CAUCHY44 (1823)
The traction vector at a point X at a time t on a surface with normal n is given
as
(3.2.5) t(X , t , n) = T(X , t)  n
with CAUCHY´s stress tensor T .
The Latin origin tendere = to pull of the word tensor already relates to the traction.
If we consider the opposite surface with normal – n , we will also obtain the
opposite traction after the theorem. This fact is known under the label
Lemma of CAUCHY
On opposite surfaces the tractions are of equal size in opposite directions
(3.2.6) t(X , t , – n) = – t(X , t , n) .
The component representation of the stress tensor with respect to some ONB is
T = ij ei  ej .
When T  e1 is the traction vector acting on a surface with normal e1 , then
11 : = e1  T  e1
is its normal part and
21 : = e2  T  e1
31 : = e3  T  e1
the two tangential parts (shear stresses). In general, ik is the stress acting on an
ek–oriented surface in the direction of ei , particularly for i  k a normal stress,
and for i  k a shear stress.
If we determine by the Theorem of CAUCHY the resulting contact force, we
can apply the divergence theorem to transform the surface integral into a volume
integral
fc =  t dA =  T  n dA =  div T dV.
A A V
Inserting this into the balance of linear momentum (3.2.3) gives
 div T dV +   b dV =   a dV
V V V
44
Augustin Louis Cauchy (1789-1857)
being valid for the body and each subbody. Then for smooth arguments we obtain
the
Local balance of linear momentum
(3.2.7) div T +  b =  a (C1)
or CAUCHY´s 1st equation of motion (1823), the most important equation in
continuum mechanics. Its component form with respect to some ONB is
 ik
+  bi =  ai i = 1, 2, 3 .
xk
In the static case, the accelerations are zero, and one obtains the local equations
of equilibrium of forces
div T +  b = o .
After the assumption that neither in the interior nor on the surface of the body
pure couple stresses (distributed torques) exist, we can conclude the symmetry of
the stress tensor as a consequence of the balance of angular momentum (2.1.112).
This can be shown by an infinitesimal volume element by summing up the
moments with respect to its centroid. For this purpose, we expand the stresses into
a TAYLOR series, as can be seen in the figure.
In the e3direction the balance of angular momentum balance with respect to the
origin of the coordinate system in the centroid of the cube gives
2 21 dx2 dx3 (dx1 / 2) – 2 12 dx1 dx3 (dx2 / 2)
= (21 – 12) dx1 dx2 dx3 = 0 ,
neglecting higher-order terms, so that we obtain
21 = 12
22 + 22 , 2 dx2 /2 + ...
12 + 12 , 2 dx2 /2 + ...
x2
21 – 21 , 1 dx1 /2 + ... 21 + 21 , 1 dx1 /2 + ...
dx2 x1
11 – 11 , 1 dx1 /2 + ... 11 + 11 , 1 dx1 /2 + ...
dx1
12 – 12 , 2 dx2 /2 + ...
22 – 22 , 2 dx2 /2 + ...
and analogously
31 = 13 and 23 = 32 .
An alternative procedure to obtain the same result uses the balance of angular
momentum of continuum mechanics (2.1.112), after which the rate of the angular
momentum of a body with respect to a fixed point O with position vector rO
dO =  rO  v dm
V
equals the resulting moment of all forces acting on the body
d O = ( 
 rO  v dm) =  rO  a dm
V V
= mO =  rO  t dA +  rO  b  dV
A V
=  rO  T  n dA +  rO  b  dV
A V
=  (rO  T)  E dV +  rO  b  dV
V V
applying the divergence theorem. Next we multiply the balance of linear
momentum in the local form (3.2.7) by the position vector
rO  a  = rO  (T  E) + rO  b  .
The difference between the volume integral over this equation and the above
balance of angular momentum is
 [(rO  T)  E – rO  (T  E)] dV = o .
V
This must hold for all bodies, so that the integrand must be zero. By the product
rule we conclude

o = (rO  T)  E – rO  (T  E) = r O  T  E .
For the evaluation of this expression we use some ONB

o = r O  T  E
=

x i ei  kl ek  el 
 e
m
xm
xi
= ei  km ek
xm
= im km ei  ek
= ki ikp ep
= (32 – 23) e1 + (13 – 31) e2 + (21 – 12) e3 .

This is the same result as already found.
Theorem. BOLTZMANN´s axiom, 2nd equation of CAUCHY (1823)
The balance of angular momentum is locally fulfilled if the stress tensor is
symmetric
(3.2.8) T = TT (C2)
We can bring the stress tensor in a spectral form
3
(3.2.9) T =   pi e pi  e pi
i 1
with the
 pi principal stresses
p
e i axes of principal stresses.
So with respect to the axes of principal stresses, all shear stresses are zero, and
two of the normal stresses are extreme.
If we number the principal stresses such that
 p1   p2   p
3,
the maximal shear stress is found in the plane with normal n =  1/2 (e1 + e3)
(3.2.10) max = ½ ( p1 –  p3) .
For some purposes one decomposes the stress tensor into its spherical part, the
pressure
(3.2.11) – p I = 1/3 tr(T) I = 1/3 (11 + 22 + 33) I
= 1/3 ( p1 +  p2 +  p3) I
and the stress deviator after (2.1.102)
(3.2.12) T' : = T – 1/3 tr(T) I = T + p I .
In many examples, the state of stress is not fully three-dimensional, but only two
dimensional, which reduces the number of unknowns from six to only three.
Definition. For a state of plane stress, the traction vector in a plane is zero, and
the other stresses do not depend on the distance from this plane.
If we choose a Cartesian coordinate system such that the X1 X2 plane is this
plane, the matrix of the stress components obtains this form
 11  12 0
 0  with ij (X1 , X2) .
 21  22
 0 0 0 
3.2.1 The Principles of Mechanics
We have already introduced the (global) balance of linear momentum (2.1.111)

p =  a dm = f
V
and the angular momentum balance with respect to some fixed point O (2.1.112)
or - in the same form - with respect to the centre of mass M (2.1.113)
dO, M =  rO, M  a dm = mO, M
V
as two independent axioms. In these forms, the balances hold for both rigid and
deformable bodies, independently of their material. We can further specify the
right-hand sides of these balances to obtain
 the global balance of linear momentum
 a dm =  b dm +  t dA
V V A
 the global balance of angular momentum
 rO, M  a dm =  rO, M  b dm +  rO, M  t dA .

V V A
If these two balances hold for the body and all of its subbodies, we obtain their
equivalent local forms as field equations, assuming the integrands to be
sufficiently smooth:
 local balance of linear momentum (3.2.7)
(C1) div T +  b =  a
 local balance of angular momentum (3.2.8)
(C2) T = TT (symmetry of the stress tensor).
If we multiply C1 by an arbitrary differentiable vector field v (test function)
 (a – b)  v = div(T)  v
and integrate over the volume, we can apply the divergence theorem and the
product rule
 (a – b)  v dm =  div(T)  v dV
V V
  
=  ( v  T  E – T   v  E) dV
V
=  v  T  n dA –  T  grad v dV
A V
=  t  v dA –  T  L dV
A V
with
L : = grad v .
If the stress tensor is symmetric (C2), only the symmetric part of L enters in
the last term
D : = ½ (L + LT) = def v .
If we interpret the vector field v as a fictitious or virtual velocity field, the above
integrals would be virtual powers. This gives the
Principle of Virtual Power (PVP)
The laws of motion (2.1.111) and (2.1.113) are fulfilled for a body if and only if
(3.2.13)  a  v dm +  T  def v dV
V V
=  b  v dm +  t  v dA
V A
holds for all virtual (velocity) fields v .
Proof. The necessity of the PVP has already been shown. In order to show the
sufficiency, we reformulate the last term
 t  v dA =  (T  n)  v dA
A A
=  div(T)  v dV +  T  grad v dV.

V V
After decomposing the gradient into its symmetric and skew parts, (3.2.13) gives
 a  v dm –  T  W dV
V V
=  div(T)  v dV +  b  v dm
V V
with the skew part of the virtual velocity gradient
W : = grad v – def v = ½ (grad v – grad T v) .
This must hold for arbitrary vector fields v , which leads to the local equation
(div T +  b –  a)  v = – T  W
as a necessary and sufficient condition. The first term in brackets must be zero
(C1) if we insert for v three linear-independent constant vector fields. The right-
hand side is zero for arbitrary skew fields W if and only if T is symmetric
(C2); q. e. d.
In the literature the Principle of Virtual Power is sometimes labelled differently
as
- virtual work
- virtual velocities
- virtual displacements
or the like.
The benefit of this principle becomes clear if one uses parametric ansatz
functions for the velocity field in the form
v = v(q1 ,  , qn) .
The variation of this is
v
v = qi .
qi
Inserting this into (3.2.13), the equation must hold for all v and thus for all
values of qi . This leads to n scalar equations for the parameters q1 ,  , qn .
Such a procedure is used for the RITZ method or the Finite Element Method. The
field equations are for these methods not exactly fulfilled (strong formulation), but
only in an optimal mean (weak formulation). Please note that the stress field in the
formulation of the PVP does not need to be differentiable, in contrast to C1.
If we insert in the PVP for v the (real) velocity field v , then we obtain the
mechanical work balance.
Therorem. Work Balance

For every body and every motion at each instant, the global work balance holds
(3.2.14) Πa = Πi + K  (WB)
with the power of the forces
(3.2.15) Πa : =  b  v dm +  t  v dA ,
V A
and the stress power
(3.2.16) Πi : =  T  L dV =  T  D dV =  T  def v dV
V V V
and the kinetic energy
 v dm .
2
K := ½
V
In fact, the rate of the kinetic energy is
K = 
 v  v dm =  a  v dm .
V V
The energy balance is no longer sufficient but only necessary for the laws of
motion to hold because of the special choice v  v .
If we insert in the PVP (3.2.13) instead of v a time-dependent infinitesimal
virtual displacement field
u(x , t) = v dt
in analogy to the differential of the displacement du = v dt , then
 b  u dm +  t  u dA –  T  def u dV
V A V

=  a  u dm =  u  u dm
V V
   
= (  u  u dm) –  u  u dm .
V V
If we integrate this equation between two arbitrary times t1 and t2 , and choose
only functions u that start from a zero field and end at a zero field, then we
obtain
HAMILTON´s45 Principle
The laws of motion are fulfilled between two instants of time t1 and t2 if and only
if
t2
(3.2.17)  (Aa – Ai + K) dt = 0

t1
holds with
Aa : =  b  u dm +  t  u dA
V A
Ai : =  T  def(u) dV
V
  2
K : =  u  u dm =  ½ (u ) dm
V V
for all time-dependent differentiable vector fields u(x , t) with initial and end
conditions
u(x , t1) = u(x , t2) = o xV.
In all these different laws of motion we obtain equilibrium conditions as

special cases if we set the velocities to zero. As consequences we have
ao and po and dO  o
L  D  0
K  Πa  Πi  0
and we obtain the equilibrium conditions
 globally
 t dA +  b dm = o
A V
 rO, M  t dA +  rO, M  b dm = o
A V
 or locally
div T +  b = o
T = TT.
In the case of equilibrium, the PVP turns into the
45
William Rowan Hamilton (1805-1865)
Principle of Virtual Displacements

The equilibrium conditions are fulfilled for a body if and only if
(3.2.18)  b  u dm +  t  u dA =  T  def u dV
V A V
holds for all differentiable vector fields u(x) .
The virtual displacements u are interpreted as
 time-independent
 infinitesimal
 fictitious
displacements. If the displacements for certain body points are constrained
because of layers, then the forces in such layers are called reaction forces, in
contrast to active forces. If the virtual displacements are meant to fulfil these
constraints, then the reactions do not work. In this case, the forces in the PVP can
be reduced to the active forces.
In the case of non-equilibrium (motion), if one interprets the negative
accelerations as inertial forces and adds them to the body forces, one obtains the
generalised forces
bgen : = (b – a) .
These are equilibrated even in the case of non-equilibrium, so that the equilibrium
conditions are fulfilled
 either for the entire body and all its subbodies
fgen : =  t dA +  bgen dm = o
A V
mgen =  r  t dA +  r  bgen dm = o
A V
with respect to some fixed point or to the centre of mass
 or equivalently locally
div T +  bgen = o
T = TT.
If we insert the generalised forces into the PVP, we obtain the
Principle of D'ALEMBERT46 in the LAGRANGEan Form

The laws of motion are fulfilled for a body if and only if
(3.2.19)  bgen  u dm +  t  u dA =  T  def(u) dV
V A V
holds for the body for all differentiable vector fields u .
Again, if the virtual displacements are compatible with the external constraints,
the reaction forces do not produce any virtual power.
All these principles hold for bodies made of arbitrary materials (fluids, solids,
etc.) and can be generalised to finite deformations.
This is not the case for the following principle, which is restricted to small
deformations. We consider the complementary work that a virtual force group
does on the real displacements u(x , t) . We use a tensor field T and a vector
field b on the body which are equilibrated
div T +  b = o
and
T = TT
everywhere in V , but are otherwise arbitrary and do not have to be compatible
with boundary conditions (external constraints). We call such a couple T , b
a virtual force group. If we multiply the first equilibrium condition with the
displacement field u(x , t) and integrate over the body, we obtain
0 =  u  div T dV +  u  b dm
V V
  
=  u  δ T  E dV –  u  T  E dV +  u  b dm
V V V
=  u  T  n dA –  grad u  T dV +  u  b dm
A V V
=  u  T  n dA –  E  T dV +  u  b dm
A V V
by using the symmetry of T and the compatibility of E = Def u  def u . This
gives the
46
Jean le Rond dÁlembert (1717 - 1783)
Principle of Virtual Forces (PVF)

The conditions of compatibility between u and E are fulfilled if and only if for
all virtual force groups T , b
(3.2.20)  u  T  n dA +  u  b dm =  E  T dV
A V V
holds.
This principle can be applied in the same form to motions (non-equilibrium),
i.e., for time-dependent fields u and E , since the compatibility conditions are not
restricted to equilibrium.
If the body contains evolving incompatibilities like cracks or fissures or voids,
the PVF does not hold in this form and must be modified.
For comparison with the PVP we have the following scheme.
The PVP PVF

balances the virtual work complementary work
of the virtual displacements, force groups,
which fulfil the compatibility equilibrium
conditions upon the forces displacements
and ensure the conditions of equilibrium. compatibility.
In both cases, only infinitesimal or incremental works are balanced, but not
finite works.
While the PVP can be easily enlarged to finite deformations (with L =
grad v), the PVF is only valid for small displacements (geometrically linear:
E  def u).
3.2.2 Stress Functions
We will next try to determine the stresses in a body only by means of the
equilibrium conditions. One constructs representations of the stresses called stress
functions which lead to stress fields that shall automatically satisfy the equations
of equilibrium. If one uses them, one can only prescribe traction boundary
conditions. Such equilibrium solutions are not unique in two or three dimensions.
In many practical problems inertia and body forces are negligible compared to
contact forces. In such cases, the equations of motion C1 and C2 are reduced to
equilibrium conditions
div T = o and T = TT.

In other words, one is looking for a divergence-free symmetric tensor field T(x)
in the interior of the body domain, which satisfies prescribed tractions after
t = T  n on its surface. It is obvious that every homogeneous stress state is
divergence-free
T(x) = Thom
with Thom being a symmetric tensor constant in space.
More general is the
Stress function of BELTRAMI (1892)
Let A(x) be a differentiable and symmetric tensor field in V and
(3.2.21) T(x) : = – inc A = –   A   = curl curl A .
Then the equilibrium conditions hold
div T = o and T = TT .
Proof. For the curl of a (not necessarily symmetric) tensor field B the following
identities hold after (2.2.26)
 curl B =   BT = – (B  )T
 div curl B = (  BT )   =   (BT  ) = curl div BT
 div (curl B)T = (  BT )T   = – B    
= – B  (  ) = o
 (curl curl B)T = curl curl (BT ) .
Consequently
div T = div curl curl A = curl div (curl A)T = o .
For a symmetric tensor field A we get
TT = (curl curl A)T = curl curl(AT ) = curl curl A = T
q. e. d.
The component-form of BELTRAMI´s solution with respect to some ONB is
ij = imn jpq Amp , nq .
It contains the following special cases.
 MAXWELL´s stress function (1868)
(3.2.22) A(x1 , x2 , x3)  A1 e1  e1 + A2 e2  e2 + A3 e3  e3
with Ai (x1 , x2 , x3) so that
11 = A2 , 33 + A3 , 22
22 = A1 , 33 + A3 , 11
33 = A1 , 22 + A2 , 11
12 = – A3 , 12
23 = – A1 , 23
31 = – A2 , 31 .
This describes a three-dimensional stress state.
 MORERA47´s stress function (1892) with  i (x1 , x2 , x3)
(3.2.23) A   3 (e1  e2 + e2  e1)
+  2 (e1  e3 + e3  e1)
+  1 (e2  e3 + e3  e2)
so that
11 = – 2  1 , 23
22 = – 2  2 , 31
33 = – 2  3 , 12
12 = ( 1 , 1 +  2 , 2 –  3 , 3), 3
23 = ( 2 , 2 +  3 , 3 –  1 , 1), 1
31 = ( 3 , 3 +  1 , 1 –  2 , 2), 2 .
It can be shown that BELTRAMI´s solution does not describe all divergence-
free symmetric tensor fields if one only allows for continuous stress functions.
The completion is due to SCHAEFER48 (1953).
BELTRAMI-SCHAEFER´s stress function
Each divergence-free symmetric tensor field in V can be represented as
(3.2.24) T = curl curl A + (  h + h  ) – div(h) I
with
A : symmetric tensor field in V
h : harmonic vector field in V , i.e.
 h = div grad h = o .
Vice versa, every differentiable symmetric tensor field and each harmonic vector
field give in the above representation a divergence-free symmetric tensor field.
Proof. See GURTIN (1972) p. 58.
47
Giacinto Morera ((1856-1909)
48
Herrmann Schaefer (1907-1969)
 AIRY49´s stress function (1863)

(3.2.25) A   (x1 , x2) e3  e3
produces a state of plane stress with
11 =  , 22
22 =  , 11
(3.2.26) 12 = 21 = –  , 12
13 = 23 = 33 = 0 .
Moreover, one can show that the compatibility conditions for the strain field
resulting from the isotropic HOOKE´s law from this stress field apply if the stress
function fulfils the bipotential equation50
 = 0.
AIRY´s stress function can also be used if a constant body force b = b1 e1 + b2 e2
is applied in the plane. In this case one uses the ansatz
(3.2.27) 12 = 21 = –  , 12 –  b1 x2 –  b2 x1 .
The following homogeneous states of stress can be easily represented by AIRY´s
stress function.
1) uniaxial tension in the x1direction
 (x1 , x2) = ½ 11 x22  11 =  , 22
2) biaxial tension
 (x1, x2) = ½ (11 x22 + 22 x12 )
 11 =  , 22 and 22 =  , 11
3) pure shear
 (x1 , x2) = – 12 x1 x2  12 = 21 = –  , 12
while all other stress components are zero.
Example for AIRY´s stress function51. We consider a state of plane stress in a

triangular prismatic dam under dead weight and one-sided water pressure, which
increases linearly with the depth of the water.
The local equilibrium condition (3.2.7) with the specific weight  D of the dam is
(C1) div T   D ey = o .
This can be fulfilled by AIRY´s stress function after (3.2.26) and (3.2.27)
49
George Biddell Airy (1801-1892)
50
see SZABO, I.: Höhere Technische Mechanik. Springer (1964) p. 153
51
see SZABO (1964).
xx =  , yy yy =  , xx xy = yx = –  , xy +  M x .
air front
water front
n
h
 x
l
As an ansatz for the stress function, we choose a polynomial of degree 3 in the

two coordinates of the plane
 (x , y) = a x2 + b x3 + c y2 + d y3 + e x y + f x y2 + g x2 y
in which we have already suppressed all terms which vanish upon two derivations.
This gives the stress fields after (3.2.26-27)
xx = 2 c + 6 d y + 2 f x
yy = 2 a + 6 b x + 2 g y
xy = yx =  e  2 f y  2 g x +  M x .
The constants a to g can be determined by the traction boundary conditions.
On the water front a hydrostatic pressure is acting, so that the shear stresses have
to vanish
xyx  0 = 0  e = 0 f =0.
Only normal forces act as a pressure, being linearly distributed
xxx  0 =   W (h  y) = 2 c + 6 d y
 c =  ½ W h d = 1/6  W .
with the specific weight of the water  W .
The other non-zero stress fields are
xx =  W (y  h)
yy = 2 a + 6 b x + 2 g y
xy = ( D  2 g) x .
On the air face of the dam we neglect the air pressure and assume zero traction.
The outer normal on the air face of the dam is
n = sin  ex + cos  ey with tan  = h / l .
We determine the traction using CAUCHY´s Theorem (3.2.5)
t = T  n = (xx sin  + xy cos) ex + (xy sin  + yy cos) ey .
This must be zero at all surface points with y = h  (h/l ) x . With this coordinate
relation we can express the y-coordinate by the x-coordinate, and obtain two scalar
equations with linear terms in x and constant ones. By comparison of the
coefficients we can determine the other constants as
a = ½ (h3/l 2  W  h  M)
b = 1/6 (h/l  M  2 h3/l 3  W)
g = ½ ( M  h2/l 2  W)
This solution of the (incomplete) boundary value problem is a partial part of the
stress field, which only satisfies the stress boundary conditions on two faces of the
dam, but no boundary conditions on the base surface of it. If we pose displacement
boundary conditions there, then they can only be considered after introducing a
material law for the dam like HOOKE´s law.
Problem 12. Stress Functions

Find by an appropriate stress function the stress field in an infinite disc
with a circular hole with radius a due to remote unidirectional tension
loads 0 . Because of the symmetry, it is sufficient to consider only one
quarter of the disc.
0
ey
a er
ex
Solution
For a plane stress state, we can use AIRY´s stress function after (3.2.21)
with (3.2.25)
T = –   A   = –    ez  ez  .
Because of the circularity of the hole we use cylindrical coordinates
{r,  , z} with an associated ONB as we already did in Problem 11. Since
 does not depend on z , we can neglect all derivatives with respect to z .
Thus
 1   1 
T = –( er + e )   ez  ez  ( er + e )
r r  r r 
1  1 
= ( , r e –  ,  er)  ( e – er)
r r r 
1 1
=  , r r e  e + ( 2  ,  –  ,  r) er  e
r r
1 1
– ( , r e) ,   er + 2 ( ,  er) ,   er .
r r
For the evaluation of these terms one has to apply the product rule and
consider the dependence of er and e on  . One obtains
1 1
rr =  ,  + , r
r2 r
1 1
r = T r = 2
, – ,r
r r
 =  , r r .
We are now trying to find an appropriate ansatz for  . At a sufficient
distance from the hole, the state of stress must be uniaxial. If the direction
of the tension is in ex , we would need in Cartesian coordinates
xx = 0 = ∞ , yy .
By twice integrating we get
∞ = 0 y2 / 2 .
With y = r sin and sin2 = (1 – cos(2)) /2 we find
∞ = 0 r2(1 – cos(2)) /4 .
We must add additional terms to our ansatz in order to describe the
distortion of this uniaxial stress state due to the hole. Since the term that
leads to a constant state of stress is of order r2, any term with lower order
in r would vanish with r  ∞ . If we added higher-order terms, we would
distort the uniaxial stress state. Because of the symmetry it is obvious that
the additional terms must appear with a period of 2 . In general, one
could introduce arbitrary ansatz functions. However, it seems to be
reasonable to only use solutions of the bipotential equation  = 0 since
only these functions lead to a compatible strain field in the case of
HOOKE´s law. This is a strong restriction to our functions. Principally, one
could take any linear combination of such solutions of the bipotential
equation and would filter out those which would fulfil the boundary
conditions. However, this would mean a lot of work. To avoid this, we only
use a finite choice of such functions like, e.g.,
 = 0 {r2 [1 – cos(2 )] /4 + C1 ln r + C2 cos(2 ) / r2
+ C3 cos (2 )} .
This ansatz leads to the stresses
1 6C2 4C3  1 C1
rr = 0 (cos (2)    +  )
2 r 4
r2  2 r2
 1 C  6C 1 
 = 0   21 + cos (2)  42   
 2 r  r 2 
 1 6C2 2C3 
 r = 0 sin (2)    4
 .
 2 r r2 
We can adjust the constants C1 , C2 and C3 to the boundary condition that
the edge of the hole shall be stress-free
t = T  n =  T  er = o .
This means that for r  a that the stress components are rr = 0 and
r = 0 . We obtain two equations for the three unknowns C1 , C2 , C3 .
With the additional condition that C1, C2 and C3 shall not depend on 
we obtain by a comparison the following three equations
1 C1 1 6C2 4C3
 = 0   2 = 0
2 r2 2 a4 a
1 6C2 2C3
  2 = 0.
2 a4 a
This leads to
C1 = – a2 /2 C2 = – a4 /4 C3 = a2 /2
This solution for the stresses exactly satisfies the equilibrium conditions
and the (dynamical) boundary conditions at the edge of the hole and in
infinity.
Let us next consider the stress distribution at the edge of the hole. For
   /2 and r  a we find  = 30 , a significant stress increase. Also
the stress  at   0 is interesting. We get  = 0 , which means a
pressure of the same size as the remote tension. We see that in all cases (for
0 positive or negative) a tension force at the edge of the hole occurs,
either with a factor 3 (for 0 > 0) or 1 (for 0 < 0).
We will now compare the solution with the result of a Finite Element
calculation. For this purpose, we look at the components of T with respect
to the basis {ex , ey , ez} . We calculate the components  eij with
 eij = Qki kl Qlj
as we did in Problem P4 with an orthogonal tensor that describes the
change of the basis. Its components with respect to {ei  ej} are
 cos θ sin θ 0 
 
[Qij] =   sin θ cos θ 0  .
 0 0 1 
In the Figure we see on the left side the solution by the stress function and
on the right side the FE-solution with 0 < xx < 300 MPa (upper), –100
MPa < yy < 56.25 MPa (middle), –88.01 MPa < xy < 18.45 MPa (lower).
These ranges refer to the solution by the stress function. The FE-
calculations have been made using an isotropic elastic material law. The
difference to our exact solution is small. For the FE model we considered a
finite region only. We used the values a = 10 mm and 0 = 100 MPa.
4 THREE-DIMENSIONAL MATERIAL THEORY
The fundamental field variables are up to now

 the displacement field u (3 scalar variables)
 the stress tensor field T (9 scalar variables)
 the field of the spec. body force b (3 scalar variables).
While the body forces are often prescribed (like the gravitation field), we have for
the other fields with 3 + 9 real unknowns only the laws of motion, i.e., the balance
of linear momentum (3 equations) and of angular momentum (3 equations). So 6
more equations are needed. The introduction of the strain tensor (6) does not really
help us since we have 6 displacement strain relations.
We cannot even expect that the mechanical behaviour of material bodies can be
fully described by these general equations, since the individuality of the particular
material is not at all taken into account so far. We need 6 material laws, which
constitute the dependence of the kinematical variables (u , E , ...) on the
dynamical ones (b , T , ...) in an appropriate way for the particular material under
consideration.
Since nature offers us an infinite manifold of different material behaviour, the
framework for material laws must be rather large. Nevertheless, there are certain
statements or principles, which seem to hold for almost all materials and already
lead to an important specification of the structure of material laws. As soon as
these principles of material theory have been fully exploited, only restrictions to
some particular material class can lead to further specifications of the material
laws. From a methodological point of view, however, it is important to clearly
announce all assumptions and to motivate them physically.
First of all, we have to postulate a deterministic relation or causality between the
kinematical and the dynamical variables. For the independent variables we choose
the (time-dependent) displacement field of the body. By the displacements we can
determine arbitrary strain tensors, rate of stretch tensors, etc.
As the dependent variable we choose the present local stress tensor of some
material point in the interior of the body.
Principle of Determinism
The stresses in some material point of a body are determined by the displacement
field of the body in the past and the present.
By this principle we exclude any influence of the future on the presence. On the
other hand, we do not exclude the influence of past events on the present
behaviour (material memory), no matter how long ago they took place.
Besides of this temporal specification, we also need a spatial one.

DOI 10.1007/978-3-319-19566-7_4
4 Three-Dimensional Material Theory 165
Principle of Local Action

The stresses in a material point of a body are determined by the motion of only
some finite neighbourhood of this point.
By this principle we exclude the influence of arbitrarily remote events on the
local behaviour. While the body forces may contain the gravitation of the earth,
we can apply this principle to the stresses in a much stronger sense.
The third and last principle shall express the plausible assumption that the
stresses do depend on deformations, but not on rigid body modifications
(translations and rotations) of the motion.
Principle of Invariance under Superimposed Rigid Body Motions
Stresses in a body are not directly caused by translations or rotations of the body.
Of course, indirectly such translations and rotations can lead to inertial effects
which do cause stresses. Here we only exclude a direct dependence of the stresses
on such rigid body motions.
Before we start exploiting these principles, we will restrict our considerations on
a specific material class, which will allow for further concretisations of the
material laws.
4.1 Elasticity
Literature
Atanackovic, T. M.; Guran, A.: Theory of Elasticity for Scientists and

Engineers. Birkhäuser, Boston (2000)
Barber, J. R.: Elasticity. Kluwer Acad. Pub., Dordrecht (1992)
Becker, W.; Gross, D.: Mechanik elastischer Körper und Strukturen. Springer,
Berlin (2002)
Eschenauer, H.; Schnell, W.: Elastizitätstheorie. BI, Mannheim. 3rd edt. (1993)
Göldner, H.: Lehrbuch Höhere Festigkeitslehre, Band 1. Grundlagen der
Elastizitätstheorie. 3rd edt. Fachbuchverlag Leipzig (1991)
Gould, P. L.: Introduction to Linear Elasticity. Springer, New York (1989)
Gurtin, M. E.: The Linear Theory of Elasticity. In: Encyclopedia of Physics. Edt.
S. Flügge. Vol. VIa/2, Springer (1972)
Hahn, H. G.: Elastizitätstheorie. Teubner, Stuttgart (1985)
Hetnarski, R. B.; Ignaczak, J.: The Mathematical Theory of Elasticity. CRC
Press (2004, 2011)
Kienzler, R., Schröder, R.: Einführung in die Festigkeitslehre. Springer,
Dordrecht (2009)
Kreißig, R.; Benedix, U.: Höhere Technische Mechanik. Lehr- and Übungsbuch.
Springer, Vienna, New York (2002)
166 4 Three-Dimensional Material Theory
Leipholz, H.: Theory of Elasticity. Noordhoff, Leyden (1974)

Love, A. E. H.: A Treatise on the Mathematical Theory of Elasticity. Dover, New
York (1944)
Maceri, A.: Theory of Elasticity. Springer (2010)
Mußchelischwili, N. I.: Einige Grundaufgaben zur mathematischen Elastizitäts-
theorie. Fachbuchverlag Leipzig (1971)
Sneddon, I. N.; Berry, D. S.: The Classical Theory of Elasticity. In:
Encyclopedia of Physics. Edt. S. Flügge. Vol. VI, Springer (1958)
Sokolnikoff, I. S.: Mathematical Theory of Elasticity. McGraw-Hill, New York
(1956)
Stark, R.: Festigkeitslehre - Aufgaben und Lösungen. Springer (2006)
Timoshenko, S. P.; Goodier, J. N.: Theory of Elasticity. McGraw-Hill, New
York, (new edt. 1970)
See also all books on Continuum Mechanics.
The strongest possible restriction of the Principle of Determinism in a temporal

sense leads to elasticity.
Definition. A material is called elastic if the present stresses depend only on the
present deformations.
Accordingly, elastic materials show absolutely no aftermath of past events.
Thus, for elastic materials the stresses T(X , t) depend only on the simultaneous
displacement field u(Y , t) of the other body points Y . Due to the Principle of
Local Action only those body points Y in some finite neighbourhood of X need
consideration.
We assume that the displacements can be expanded into a TAYLOR series
within such a neighbourhood
u(Y, t) = u(X , t) + Grad u(X , t)  (Y – X)
+ ½ Grad Grad u(X , t)  (Y – X)  (Y – X) + ... .
Consequently, the stresses T(X , t) can only depend on the local quantities
u(X , t)
Grad u(X , t) = : H(X , t)
Grad Grad u(X , t)
etc.
Since the first term of this series u(X , t) stands for a rigid translation, we can
exclude its influence on the stresses after the Principle of Invariance under
Superimposed Rigid Body Motions.
The next assumption is the strongest non-trivial form of the Principle of Local
Action.
4.1 Elasticity 167
Definition. A material is called simple if the stresses depend only on the first (or
simple) gradient of the displacements at this point.
If we restrict our considerations for the rest of this treatise to simple materials,
we definitely exclude gradient materials for which also higher gradients have an
influence on the stresses.
For simple elastic materials there must exist a material function f that assigns a
symmetric stress tensor to some displacement gradient
T = f (H) .
f is the elastic material law.
Within the present format, inelastic simple materials, some classes of which will
be considered later, are those where the present stresses can also depend on the
past H–process.
In this chapter we consider simple elastic materials. We decompose the
displacement gradient into its symmetric and skew part after (3.1.31) and (3.1.32)
H = E+
with E : = ½ (Grad u + Grad T u) = Def u
 : = ½ (Grad u – Grad T u) .
E describes (infinitesimal) deformations and  (infinitesimal) rotations. For
pure translations the displacement field is constant in space u(X , t) = u0(t),
which does neither contribute to E nor to  . After the Principle of Invariance
under Superimposed Rigid Body Motions the stresses cannot depend on the
rotations (X , t) . Only a dependence on the strains is possible. Thus the elastic
law can be reduced to
T = f (E) .
Since we restricted our concern in the entire book to small deformations, it is
reasonable to approximate the generally non-linear elastic law by a linear one. We
do this in the reference placement, which we assume to be stress-free (T  0 for
E  0). We already know that every linear function between tensors can be
represented by a 4th-order tensors or a tetrad. Therefore, the general form of a
linear elastic simple material is
HOOKE´s law
(4.1.1) T = C  E (HOOKE)
The tetrad C is the elasticity or stiffness tensor. If this tensor is invertible

(4.1.2) S : = C –1
such that
(4.1.3) E = S  T
then S is the (elastic) compliance tensor. In the realm of linear elasticity, it will
be our task to further specify these two tensors C and S .
If we understand C as acting on the displacement gradient T = C  H , then
because of C   = 0 it must have the right subsymmetry, and hence cannot be
invertible.
If we understand C as acting on the strain tensor T = C  E , then it is a
mapping between symmetric tensors, since its image must also be symmetric after
C2. Its action on antisymmetric tensors is not involved in this mapping, so that we
can arbitrarily normalize this part of the tetrad. Two choices are canonical.
1.) We propose that C maps each skew tensor into itself. The restriction of C to
the antisymmetric tensors is then the identity
C  (ei  ej – ej  ei) = ei  ej – ej  ei
for i, j = 1, 2, 3. The inverse of it on the antisymmetric tensors always exists and
maps also
S  (ei  ej – ej  ei) = ei  ej – ej  ei .
If the symmetric part is also invertible, then so is the tetrad as a whole.
2.) We propose that C maps each antisymmetric tensor into the (symmetric) zero
tensor. In this case C possesses the right subsymmetry. Since the image of every
symmetric tensor under C shall be symmetric as well, it must have also the left
subsymmetry. With these subsymmetries C cannot be invertible. Only its
restriction to the space of all symmetric tensors may (or may not) be still invertible
to give
E = S  C  E  T = C  S  T.
For S one would then also postulate both subsymmetries.
The advantage of the second alternative is that we can bring C and S into
VOIGT representations (2.1.142) understood as tensors on a 6-dimensional vector
space. This choice is broadly made in the literature and shall also be used in the
rest of this book. In this case, the elastic tensors have only 6  6 = 36 independent
components.
Moreover, we will also assume that the elasticities possess the (major) symmetry
(4.1.4) C = CT  S = ST
and therefore allow for a symmetric VOIGT notation. The physical rationale for
this assumption will be given in Chap. 4.1.5. This symmetry further reduces the
number of independent components to 21.
Accordingly, we will assume the following symmetry properties of the stiffness
and the compliance tensors:
4.1 Elasticity 169
Cijkl = Cjikl left subsymmetry

(4.1.5) Cijkl = Cijlk right subsymmetry
Cijkl = Cklij symmetry
with respect to every ONB.
4.1.1 Material Symmetry
Literature
Böhlke, T.; Brüggemann, C.: Graphical representation of the generalized
Hooke's law. Techn. Mech. 21,2, 145-158 (2001)
Forte, S.; Vianello, M.: Symmetry classes for elasticity tensors. J. Elast. 43, 81-
108 (1996)
Hearmon, R.: The elastic constants of anisotropic materials. Rev. Mod. Phys. 18,
409-440 (1946)
Lekhnitskij, S. G.: Theory of Elasticity of an Anisotropic Elastic Body. Holden-
Day (1963)
Nye, J. F.: Physical Properties of Crystals. Oxford University Press, Ely House,
London (1957)
Simmons, G.; Wang H.: Single Crystal Elastic Constants and Calculated
Aggregate Properties: A Handbook. The M. I. T. Press (1971)
Ting, T. C. T.: Anisotropic Elasticity. Oxford Univ. Press (1996)
Voigt, W.: Lehrbuch der Kristallphysik. Teubner, Leipzig (1910)
If the properties of a body are the same in all directions, the material is called
isotropic. This may be a reasonable and practical assumption for steel, rubber,
(unreinforced) concrete, and many other materials. Other materials, however, have
this directional dependence, like timber, reinforced concrete, reinforced rubber,
mineral or metallic crystals. These materials are called anisotropic. Among them
one distinguishes several types of anisotropy. All these notions shall now be made
precise for elastic materials.
The starting point is the linear elastic material law in the form (4.1.1) or (4.1.3).
Let u(X) be an arbitrary displacement field of the body, and Q some proper
orthogonal tensor. We now create a second displacement field
u*(X*) : = Q  u(X)  u(X) = QT  u(X*) ,
which is simply the result of a rotation of u(X) by Q around the origin O using
the transformation from X to X*
OX* = Q  OX  OX = QT  OX*
= X* = Q  X  = X = QT  X *
u*
X*
X* Q
u
O X X
Because of
du
H : = Grad u =
dX
and the chain rule, we obtain for the second displacement field
du * du * du dX
H* = Grad u* = =  
dX * du dX dX *
= Q  H  QT
and the two strain tensor fields
E(X) = Def u(X)
E*(X*) = Def u*(X*) = Q  E(X)  QT.
So one strain tensor can be rotated into the other, and vice versa. These relations
remain valid if the turning point is X , i.e., if O  X  X*.
The elastic law is considered as symmetric with respect to some rotation if the
stresses T* in X  X* caused by E* are also only the rotated stresses of T = ij
ei  ej in X  X* caused by E
T* = ij (Q  ei)  (Q  ej) = Q  ij ei  ej  QT = Q  T  QT .
This leads to the following
Definition. An orthogonal tensor Q is called a symmetry transformation of the
linear elastic law T = C  E if
(4.1.6) C  (Q  E  QT) = Q  (C  E)  QT
holds for all deformations E . The set of all symmetry transformations of C is
called the symmetry group G of the material.
If one considers only solid materials, as we do here, only orthogonal symmetry
transformations are important. This is not the case for fluids, where also non-
4.1 Elasticity 171
orthogonal transformations are involved. Elasticity deals mainly with solids, so

that this restriction is justified.
The symmetry group G satisfies the axioms of a group from algebra, as one can
easily verify.
(G1) Q1 , Q2  G  Q1  Q2  G
(G2) Q1 , Q2 , Q3  G  (Q1  Q2)  Q3 = Q1  (Q2  Q3)
(G3) IG
(G4) Q  G  Q–1  G
Q occurs twice on both sides of the symmetry conditions. Therefore a minus
sign would annihilate on both sides. Thus
QG  –QG.
Accordingly, with each pure rotation a combination with a reflection is also
contained in the symmetry group. However, this is only true for material functions
relating even-order tensors.
If the elastic law is invertible, then the elasticities and the compliances have the
same symmetries. This can easily be shown if we multiply (4.1.6) by S and
insert the result into (4.1.1)
Q  G  Q  (S  T)  QT = S  (Q  T  QT) .
So there is no need to distinguish between the symmetry groups of C and of S .
If we want to see the effect of a symmetry transformation on the components of
the tetrads with respect to some ONB, we use the following representations
C = Cijkl ei  ej  ek  el
and
S = Sijkl ei  ej  ek  el .
Because of (4.1.6) we achieve
C  E = QT  [C  (Q  E  QT )]  Q
for all symmetric tensors E , and therefore
(4.1.7) C = Cijkl ei  ej  ek  el
= Cijkl (Q  ei)  (Q  ej)  (Q  ek)  (Q  el)
and analogously
S = Sijkl ei  ej  ek  el
= Sijkl (Q  ei)  (Q  ej)  (Q  ek)  (Q  el) .
So both tetrads have the same components with respect to some ONB and with
respect to the one rotated by a symmetry transformation.
We can also formulate the symmetry condition with respect to an arbitrary ONB
equivalently as
sym(ei  ej)  C  sym(ek  el)
(4.1.8) = sym[(Q  ei)  (Q  ej)]  C  sym[(Q  ek)  (Q  el)]
for i, j, k, l = 1, 2, 3.
The smallest possible symmetry group consists of I and – I . If G  {I , – I} ,
the material is called triclinic. This group is contained in each other symmetry
group as a subgroup.
We call C (and S ) isotropic if its symmetry group contains all orthogonal
tensors (orthogonal group). In contrast to this, if there are orthogonal tensors not
contained in G , then the material is called anisotropic.
All other symmetry groups of solids can be found between these two extremes,
namely the triclinic group as a minimum, and the orthogonal group as the
(isotropic) maximum.
Algebraically one can construct infinitely many subgroups of the orthogonal
group. In crystallography 32 crystal classes are known. In the mechanical theory,
with each symmetry transformation also its negative is contained in the symmetry
group. Therefore only 6 symmetry classes remain. Additionally we have to add the
isotropic and the transversely isotropic (hexagonal) groups (see below), so that
within the orthogonal group only seven subgroups matter.
In the following table, the 11 crystal classes are defined by their generators. In it
{i , j , k} refers to an ONB and d : = (i + j + k) /3 to the normed diagonal. Qn
is that particular orthogonal tensor that causes a –rotation around the n–axis with
some   [0 , 2) . The symmetry group G consists of the generators and their
reflections and combinations according to the group axioms.
For all of these groups, only rotations of discrete angles are considered. These
groups are therefore called discrete groups. Only the isotropic and the transversely
isotropic groups are continuous.
The transversely isotropic group is generated by rotations Qk around a fixed
axis k with arbitrary angles  .
In the linear elastic case, there are only seven anisotropic and one isotropic
symmetry classes, which we will show below. All the representations are with
respect to some positively oriented ONBs {ei} , which is in most cases not
arbitrary, but induced by the crystal structure (the same as {i , j , k} before).
4.1 Elasticity 173
Table of the symmetry groups for different crystal classes52
No. crystal class generators order

1. triclinic system I 2
2. monoclinic system Qk 4
3. rhombic system Qi , Qj 8
4.1 tetragonal system Qk / 2 8

tetragonal-disphenoidal
tetragonal-pyramidal
tetragonal-dipyramidal
4.2 tetragonal-scalenohedral Qk / 2 , Qi 16
ditetragonal-pyramidal
tetragonal-trapezohedral
ditetragonal-dipyramidal
5.1 cubic system Qi , Qj , Qd2 / 3 24
tetratoidal
diploidal
5.2 hextetrahedral Qi / 2 , Qj / 2 48
gyroidal
hexoktahedral
6.1 trigonal system Qk2 / 3 6
trigonal-pyramidal
rhombo-hedral
6.2 ditrigonal-pyramidal Qk2 / 3 , Qi 12
trigonal-trapezohedral
hexagonal-scalenohedral
6.3 trigonal-dipyramidal Qk / 3 12
hexagonal-pyramidal
hexagonal-dipyramidal
6.4 ditrigonal-dipyramidal Qk / 3 , Qi 24
dihexagonal-pyramidal
hexagonal-trapezohedral
dihexagonal-dipyramidal
52
after COLEMAN, B. D.; NOLL, W.: Material symmetry and thermostatic
inequalities in finite elastic deformations. Arch. Rational Mech. Anal. 15, 87-111
(1964)
The VOIGT representations of C are given by 66-matrices after (2.1.142).

C C1122 C1133 2C1123 2C1113 2C1112 
 1111 
 C2222 C2233 2C2223 2C2213 2C2212 
 C3333 
(4.1.9)  2C3323 2C3313 2C3312 
 2C2323 2C2313 2C2312 
 
 symmetric 2C1313 2C1312 
 2C1212 

We will now consider the restrictions of the eight symmetry classes on the
components of C .
1.) triclinic elasticity. There are no restrictions, and we have 21 independent
elastic constants. Here all angles of the crystallographic elementary cell can be
oblique.
2.) monoclinic elasticity. G is generated by rotations Qe completed by

3
reflections – Qe at the e1e2plane. The elementary cell has two right angles and
3
one oblique one. Here,
(– Qe )  e1 = e1 (– Qe )  e2 = e2
3 3
(– Qe )  e3 = – e3 .
3
Consequently, all components that contain the index 3 once or three times must
vanish. The VOIGT matrix is then
C C1122 C1133 0 0 2C1112 
 1111 
 C2222 C2233 0 0 2C2212 
 C3333 
 0 0 2C3312 
 2C2323 2C2313 0 
 
 symmetric 2C1313 0 
 2C1212 

Only 13 independent constants remain.
The procedure to reduce the other elasticity tensors is similar.
4.1 Elasticity 175
3.) rhombic elasticity (orthotropy) generated by Qe , Qe with 9 independent

2 3
constants
C1111 C1122 C1133 0 0 0 
 C2222 C2233 0 0 0 
 
 C3333 0 0 0 
 
 2C2323 0 0 
 
 2C1212 
The elementary cell is a cuboid.
4.) tetragonal elasticity is generated by Qe / 2 and Qe with 6 independent

3 1
constants
C1111 C1122 C1133 0 0 0 
 C1111 C1133 0 0 0 
 
 C3333 0 0 0 
 
 2C2323 0 0 
 
 2C1212 
The elementary cell is a cuboid with two equal side lengths.
5.) cubic elasticity is generated by Qe / 2 , Qe / 2 with 3 independent constants

1 2
C1111 C1122 C1122 0 0 0 

 C1111 C1122 0 0 0 
 
 C1111 0 0 0 
 
 2C2323 0 0 
 
 2C2323 
The elementary cell is a cube.

6.) trigonal elasticity is generated by Qe2 / 3 , Qe with 6 independent constants

3 1
 2C1123 0 0 
C1111 C1122 C1133 
 C1111 C1133  2C1123 0 0 
 
 C3333 0 0 0 
 2C2323 0 0 
 
 symmetric 2C2323 2C1123 
 
 C1111  C1122 
7.) transversely isotropic elasticity generated by Qe ,  arbitrary, with 5

3
independent constants
C1111 C1122 C1133 0 0 0 
 C1111 C1133 0 0 0 
 
 C3333 0 0 0 
 
 2C2323 0 0 
 
 C1111  C1122 
The above representations of the stiffness matrix hold only for particular bases.
For other bases the matrix can have a different structure and can occasionally be
fully occupied. In such cases it can be difficult to find the corresponding
symmetry group, in particular if the data is not exact but scatters due to
measurements.
8.) isotropic elasticity generated by Qe with ei and  arbitrary with 2

i
independent constants
C1111 C1122 C1122 0 0 0 
 C1111 C1122 0 0 0 
 
 C1111 0 0 0 
 
 C1111  C1122 0 0 
 symmetric C1111  C1122 0 
 
 C1111  C1122 
The following figures demonstrate the anisotropy of different cubic crystals. The
stiffness under uniaxial tension is depicted in different directions. The more this
figure deviates from the (isotropic) sphere, the larger is the anisotropy for tension.
4.1 Elasticity 177
Aluminium Iron(78)-Aluminium(22)
Copper Lithium
Elasticity body of different cubic single crystals

from Böhlke/ Brüggemann (2001)
The following tables have been taken from

W. F. Hosford: The Mechanics of Crystals and Textured Polycrystals. Oxford
Science Pub. (1993)
We note that some of these values differ from those given by other authors, in
particular those for copper.
Elastic constants of different cubic crystals
Crystal c11 c12 c44/2 s11 s12 s44/2

(GPa) (GPa) (GPa) (TPa–1) (TPa–1) (TPa–1)
339.8 58.6 99.0 3.10 – 0.46 10.10
Cr
231.4 134.7 116.4 7.56 – 2.78 8.59
Fe
3.70 3.14 1.88 1223.9 56.19 53.19
K
Li 13.50 11.44 8.78 332.8 – 152.7 113.9
Mo 441.6 172.7 121.9 2.90 – 0.816 8.21
Na 6.15 4.96 5.92 581.0 – 259.4 168.9
Nb 240.2 125.6 28.2 6.50 – 2.23 35.44
Ta 260.2 154.5 82.6 6.89 – 2.57 12.11
W 522.4 204.4 160.8 2.45 – 0.69 6.22
Ag 122.2 90.7 45.4 22.26 – 9.48 22.03
Al 107.3 60.9 28.3 15.82 – 5.73 35.34
Au 192.9 163.8 41.5 23.55 – 10.81 24.10
Cu 166.1 199.0 75.6 15.25 – 6.39 13.23
Ni 248.1 154.9 124.2 7.75 – 2.98 8.05
Pb 49.5 42.3 14.9 94.57 – 43.56 67.11
Pd 227.1 176.0 71.7 13.63 – 5.95 13.94
Pt 346.7 250.7 76.5 7.34 – 3.08 13.07
C 949.0 151.0 521.0 1.10 – 1.51 1.92
Ge 128.4 48.2 66.7 9.80 – 2.68 15.00
Si 166.2 64.4 79.7 7.67 – 2.14 12.54
MgO 287.6 87.4 151.4 4.05 – 0.94 6.60
MnO 223.0 120.0 79.0 7.19 – 2.52 12.66
LiF 114.0 47.7 63.6 11.65 – 3.43 15.71
KCl 39.5 4.9 6.3 26.00 – 2.85 158.6
NaCl 49.0 12.6 12.7 22.80 – 4.66 78.62
NaB 40.4 10.1 10.2 27.54 – 5.53 98.52
NaI 30.1 9.12 7.33 38.72 – 9.011 136.4
NaF 97.0 23.8 28.2 11.41 – 2.29 35.43
ZnS 103.2 64.8 46.2 18.77 – 7.24 21.65
InP 102.2 57.6 46.0 16.48 – 5.94 21.74
GaAs 118.8 53.7 59.4 11.72 – 3.65 16.82
4.1 Elasticity 179
Elastic compliances of different hexagonal crystals in TPa–1
Material s11 s12 s44 /2 s33 s13
Cadmium 12.9 – 1.5 64.0 36.9 – 9.3
Hafnium 4.25 0.398 17.95 6.13 – 1.56
Magnesium 22.15 – 7.7 60.3 19.75 – 4.93
Titan 4.41 0.507 21.41 6.97 – 1.89
Zinc 8.40 1.1 26.4 28.7 – 7.75
Zirconium 5.32 0.719 31.25 7.96 – 2.39
4.1.2 Isotropic Elasticity
In the case of isotropic materials, the symmetry group contains all orthogonal
transformations, i.e., all rotations and reflections. We decompose the stress and
strain tensors into their spherical and deviatoric parts after (2.1.102)
(3.2.12) T = –p I + T' with –p : = 1/3 tr T (hydrostatic pressure)
(4.1.10) E =  I + E' with  : = /3 tr E
1
(dilatation)
and
tr T' = 0 and tr E' = 0
(deviators are traceless). For symmetric tensors both parts are again symmetric.
Theorem. An invertible stiffness tetrad C and the corresponding compliance
tetrad S are isotropic if and only if there are two constants k and G so that we
have for all stresses and strains
(4.1.11) C  [ I + E'] =  3 k I + 2 G E'
(4.1.12) S  [–p I + T'] = –p / (3 k) I + 1/(2 G) T'
Both tetrads are invertible on the symmetric tensors if and only if k and G are
non-zero.
Proof. Because of the linearity we have
T = C  E = C  [ I + E'] =  C  I + C  E'.
Now we assume that C maps a spherical tensor  I in both a spherical tensor

 I and a deviator T'
C  ( I) =  I + T'.
Because of the isotropy condition we find
C  [Q  ( I)  QT ] = Q  ( I + T')  QT
= C  ( I) =  I + Q  T'  QT.
The difference with the above equations is
T' = Q  T'  QT.
We bring T into its spectral form (3.2.9)
T =  p
1 e p1  e p1 +  p
2 e p2  e p2 +  p3 e p3  e p3 .
If the three principal stresses  pi are equal, then the tensor would be a spherical
one (without deviatoric part). If a non-zero deviatoric part exists, two principal
stresses must be different. So let, e. g.,  p1   p2 . Then
Q  Qe / 2 = e p2  e p1 – e p1  e p2 + e p3  e p3
3
is a rotations, which maps

Q  e p1 = e p2 Q  e p2 = – e p1
Q  e p3 = e p3 .
Thus,
Q  T'  QT  T'.
Consequently, the above equation can only be valid for all orthogonal Q if
T' = 0 . The image of a spherical tensor under an isotropic mapping is again a
spherical tensor. Since with C also S must be isotropic, the reverse also holds:
The pre-image of a spherical tensor under an isotropic mapping is also a
spherical tensor. This has the immediate implication: the (pre)image of a deviator
under an isotropic mapping is again a deviator.
The mapping of the spherical part gives
C  ( I) =  C  I =  I
 C  I = / I = : 3 k I
and can therefore be reduced to one real constant k . If C is invertible, then
necessarily k  0 .
We will next show the following fact: If C is isotropic, then all deviators are
eigentensors of C , i.e., E' and C  E' are parallel. We will use the finding
from (3.1.28) that every deviator can be combined by simple shears, and that for
4.1 Elasticity 181
each ONB {ei} the three tensors sym(ei  ej) , i < j, are mutually orthogonal and
deviatoric and describe simple shears. So let
E' = sym(e1  e2)
be one of them. We assume that
C  E' = r E' + s E'
with some E' orthogonal to E' like, e.g.,
E' = sym(e2  e3) .
Then the orthogonal tensor
Q  Qe = – e1  e1 – e2  e2 + e3  e3
3
maps this particular E' like

Q  E'  QT = E'
and
Q  E'  QT = – E' .
By the isotropy condition we obtain
C  [Q  E'  QT] = C  E' = r E' + s E'
= Q  (C  E')  QT = Q  (r E' + s E')  QT = r E' – s E' .
This is only possible if s = 0 . Analogously the same holds for all deviators being
orthogonal to E' . Since every deviator can be composed by simple shears, we can
conclude that every deviator is an eigentensor of C
C  E' = r E' .
We finally show that the corresponding eigenvalue r is the same for all deviators.
It is sufficient to prove this for an arbitrary deviator E' and another non-parallel
one E'. So let E' and E' be as before and
C  E' = r E'
and
C  E' = r E' .
Then a rotation around the diagonal for 2 /3
Q  e2  e1 + e3  e2 + e1  e3
gives
Q  E'  QT = Q  sym(e1  e2)  QT = sym(e2  e3) = E' .
The isotropy condition gives
C  (Q  E'  QT) = C  E' = r E'
= Q  (C  E')  QT = Q  (r E')  QT = r E' ,

which is only possible if r = r . Also for every linear combination of deviators the
eigenvalues must be equal. We set G : = r /2 . The mapping is invertible for
deviators if G  0 . This completes the proof; q. e. d.
Accordingly, an isotropic linear function maps spherical tensors into spherical
tensors, and deviators into deviators. Spherical tensors and deviators are the
eigentensors of both C and S .
Since spherical parts do not alter the eigendirections of tensors, we also have the
following result:
Theorem. For isotropic elastic materials the directions of principal stresses and
principal strains coincide.
We will now apply the representation of the isotropic law to particular cases.
1.) dilatation or compression
With
E  I
the isotropic law (4.1.11) produces a hydrostatic stress state
T = C  ( I) = 3 k  I = –p I
and with (3.1.27)
dV  dV0
(4.1.13) –p = 3 k  = k tr E  k
dV0
with the bulk modulus k  0 of dimension [stress]. If pressure causes a

compression, k must be positive.
2.) simple shear
Here
E  ½ xy (ex  ey + ey  ex) = E'
so that
tr E = 0 .
Then we have with (4.1.11)
T = C  [½ xy (ex  ey + ey  ex)]
= G xy (ex  ey + ey  ex)
= xy (ex  ey + ey  ex)
or
(4.1.14) xy = G xy = 2 G xy .
G is the shear modulus with dimension [stress].
4.1 Elasticity 183
3.) uniaxial tension

T  11 e1  e1
= 1/3 11 I + 2/3 11 e1  e1 – 1/3 11 (e2  e2 + e3  e3)
and therefore
–p = 11 /3
and
T' = 1/3 11 (2 e1  e1 – e2  e2 – e3  e3) .
The strains are
1 1
E = S  T = 11 I+ 1
/3 11 (2 e1  e1 – e2  e2 – e3  e3)
9k 2G
1 1 1 1
= 11 ( + ) e1  e1 – 11 ( – ) (e2  e2 + e3  e3)
9k 3G 6G 9k
3k +G 3k  2G
= 11 e1  e1 – 11 (e2  e2 + e3  e3)
9kG 18 k G
= 11 e1  e1 + 22 e2  e2 + 33 e3  e3 .
The ratio of axial stress to strain 11 / 11 is the YOUNG´s modulus
9kG EG
(4.1.15) E :=  k = ,
3k +G 9 G 3 E
while the (negative) ratio of transversal strain to axial strain is POISSON´s ratio
(1.1.7)
 22  33 3 k  2G
(4.1.16)  := = =
11  11 6 k + 2G
E 2 G  1+ 
 G =  k =
2  1   3  1  2 
which has the same value in all directions in the isotropic case.
The isotropic linear elastic law has the general representation

T = 3 k  I + 2 G E'
= (3 k – 2 G)  I + 2 G E
(4.1.17) T =  tr(E) I + 2 E =
E
[E +  tr(E) I]
1  1  2
with  : = G and  : = k – 2/3 G .  and  are called LAMÉ53´s constants.

The spherical part of the law is
 3 p = tr T = (3  + 2 ) tr E = 3 k tr E
and the deviatoric one
T' = 2 E' = 2G E'.
In components this gives
11 =  (11 + 22 +33 ) + 2 11
22 =  (11 + 22 +33 ) + 2 22
33 =  (11 + 22 +33 ) + 2 33
12 = 2 12 =  12
13 = 2 13 =  13
23 = 2 23 =  23
The inverse form of the law is
1  1
(4.1.18) E = [T – tr(T) I] = [(1 + ) T –  tr(T) I]
2 2   3 E
The isotropic elastic tetrads have the following representations
(4.1.19) C =  I  I + 2 I
1 
(4.1.20) S = [I – I  I] .
2 2   3
Instead of the 4th-order identity I one can also use the symmetriser I S after
(2.1.143). The eigenvalues of C , sometimes called KELVIN moduli, are 3 k =
2 + 3 (one) and 2 = 2G (five). The eigenvalues of S are their reciprocal
values.
The components of C and S are, e.g.,
1
C1111 =  + 2 S1111 =
E

C1122 =  S1122 = –
E
1
C1212 =  = G S1212 =
4
53
Gabriel Lamé (1795-1870)
4.1 Elasticity 185
The VOIGT representation of the stiffness matrix is

  2    0 0 0
   2  0 0 0 

   2 0 0 0
 
 2 0 0 
 symmetric 2 0 
 
 2  
and for compliance matrix
1 / E  / E  / E 0 0 0
 1/ E  / E 0 0 0 
 
 1/ E 0 0 0 
 
  2G 1 0 0 
 
 symmetric  2G 1 0 
 

  2G 1 
The important result is that isotropic linear elasticity is determined by two
material constants, the choice of which is due to individual preference.
In the following table, the transformations between the most important elastic
constants are given (after GURTIN, 1972 p. 79) as well as some examples (after
LAI/ RUBIN/ KREMPL, 1993). We note that some of these values differ from
those given by other authors, in particular those for steel.
POISSON´s ratio  is dimensionless as a ratio of lengths. All other constants in
the table are stiffnesses and have the dimension [stress = force / area].
 G= E  k
, G G (3  2G )  3  2G
 G 2   G 3
, E (E–3)/4+ –(E+)/(4)+ (E+3)/6+
( E  3 )2  8  E (E +  )2 + 8  2 ( E  3 ) 2  4 
4 4
6
,    1  2   1    1  2    1  

2 3
, k 3k    9k  k    
2 3k   3k  
G, E  2G  E  G E  2G GE
E- 3G 2G 3(3G  E )
G,  2Gν 2G(1+) 2G  1   
1  2ν 3  1  2 
G, k 3k  2G 9kG 3k  2G
½
3 3k  G 3k +G
E,  E E E
1 1 2  2 1    3  1  2 
E, k 3k  3k  E  3kE ½ 3k  E
9k  E 9k  E 3k
, k 3k 3k  1  2  3k(1–2)
1  2  1  
4.1 Elasticity 187
Elastic constants for isotropic materials at room temperature
Material Composition
E [GPa]  [-] G  [GPa] k
[GPa] [GPa]
Aluminium pure and alloy 68.2 – 78.5 0.32 – 0.34 25.5 –26.53 46.2 – 62.7 63.4 –
80.6
Brass 60 – 70 % Cu 99.9 – 109.6 0.33 – 0.36 36.6 – 41.3 73.0 – 103.4 97.1 –
40 – 30 % Zn 130.9
Copper 117 – 124 0.33 – 0.36 40.0 – 46.2 85.4 – 130.9 112.3 –
148.1
Cast iron 2.7 – 3.6 % C 90 – 145 0.21 – 0.30 35.8 – 56.5 26.9 – 83.4 51.0 –
121.3
Steel carbon and low 106.1 –114.4 0.34 41.3 84.1 – 90.9 111.6 –
alloy 118.5
Stainless 18 % Cr, 8 % Ni 193 – 207 0.30 73.0 111.6 – 160.5 –
steel 119.2 168.1
Titanium pure and alloy 106.1 – 114.4 0.34 41.3 84.1 – 111.6 –
90.9 118.5
Glass various 49.6 – 79.2 0.21 – 0.27 26.2 – 32.4 15.2 – 32.4 –
36.5 57.9
Methyl 2.41 – 3.45 - - - -
methacrylate
Polyethylene 0.14 – 0.38 - - - -
Rubber 0.00076 – 0.50 0.00028 –  
0.00413 0.00138
4.1.3 Projection Method

Literature
Halmos, P.: Finite-Dimensional Vector Spaces. D. van Nostrand Comp., New
York (1958) § 41 ff.
Bertram, A.; Olschewski, J.: Zur Formulierung linearer anelastischer
Stoffgleichungen with Hilfe einer Projektionsmethode. Z. ang. Math. Mech. 73, 4-
5, T401-3 (1993)
Sutcliffe, S.: Spectral decomposition of the elasticity tensor. J. Appl. Mech. 59,
762-773 (1992)
We will next show another way to represent isotropic laws and try to generalise
it for the construction of anisotropic laws.
We have already shown that the isotropic elastic tetrads C and S map
spherical tensors into spherical ones and deviators into deviators. The tetrad
(4.1.21) P1 : = 1/3 I  I
maps each tensor into its spherical part
P1  E = P1  [1/3 tr(E) I + E'] = 1/3 tr(E) I .
The tetrad
(4.1.22) P2 : = I – P1
maps each tensor into its deviatoric part
P2  E = P2  [1/3 tr(E) I + E'] = E'.
These tetrads have the following properties:
(P1) Pi  Pi = Pi i = 1, 2 projective
(P2) Pi  Pj = 0 i  j = 1, 2 orthogonal
(P3) P1 + P2 = I complete
Thus, P1 and P2 constitute a complete orthogonal system of projectors. With its
aid we can represent each isotropic stiffness tetrad as
(4.1.23) C = 1 P1 + 2 P2
with the eigenvalues
1 = 3 k = 2  + 3  and 2 = 2  = 2 G
and the compliance tetrad as
1 1
S = P1 + P2 .
1 2
Thus, the inversion of a tetrad in this form can be reduced to the inversion of reals.
Representation Theorem. Each isotropic linear mapping between symmetric
tensors E and T can be represented as
2
(4.1.24) T =   i Pi  E
i 1
with two scalar constants  1 and  2 .

Obviously, each spherical tensor is an eigentensor with eigenvalue  1 , and
each deviator is an eigentensor with eigenvalue  2 . The eigenspace belonging to
the eigenvalue  1 is one-dimensional, that of  2 is 5-dimensional. The above
representation is an eigenprojector representation. Note that the two eigenspaces
are identical for all isotropic tetrads, whatever the elastic constants are. This fact is
important if we want to use such representations also for anisotropic tetrads.
In fact, one can find for each symmetric VOIGT matrix (isotropic or anisotropic)
a spectral form
4.1 Elasticity 189
N
C =   i Pi
i 1
with the orthogonal eigenspace projectors Pi and N  6 different real

eigenvalues i . The problem is that in general the projectors also depend on the
material constants. Only in the isotropic case with N  2 and the cubic cases with
N  3 does such a representation generally exist. In the cubic case, the projectors
are defined in the following way. Let {ei} be an ONB which coincides with the
cubic axes of the material. Then the projectors are
P1 : = 1/3 I  I the same as before
3
P2 : =  ei  ei  ei  ei – P1
i 1
P3 : = I – P2 – P1
They also constitute a complete orthogonal system of projectors.
We now represent a symmetric 2nd-order tensor with respect to the same ONB
E = ij ei  ej .
Then we obtain
1 0 0 
P1  E  /3 tr E 0 1 0 
1
0 0 1 
 11 0 0 1 0 0 
P2  E   0  22 0  – /3 tr E 0 1 0 
1
 0 0  33  0 0 1 
 2ε11  ε22  ε33 0 0 

= 1
/3  0 2ε22  ε33  ε11 0 

 0 0 2ε33  ε11  ε22 
 0  12  13 
P3  E   21 0  23  with tr E = 11 + 22 + 33 .
 31  32 0 
In the cubic case, the eigenspaces do not depend on material constants. This leads
to the general representations with respect to a crystallographic basis
(4.1.25) C =  1 P1 +  2 P2 +  3 P3
and its inverse
1 1 1
S = P1 + P2 + P3
1 2 3
with three material constants  1 ,  2 ,  3 . For  2   3 this contains the
isotropic case.
Such representations are particularly practical in linear viscoelasticity (see
Chapt. 4.3).
4.1.4 Identification of the Elastic Constants
Literature
Förster, F.: Ein neues Messverfahren zur Bestimmung des Elastizitätsmoduls und
der Dämpfung. Z. für Metallkunde, 29, 109-115 (1937)
Han, J.: Identifikation der elastischen Kennwerte anisotroper Hochtemperatur-
legierungen mittels Resonanzmessungen and Finite-Elemente-Simulation. VDI
Vlg. Reihe 5, Nr. 404, Düsseldorf (1995)
Bertram, A.; Han, J.; Olschewski, J.; Sockel, H.-G.: Identification of elastic
constants of single crystals by resonance measurements and FE-analysis. Int. J.
Comp. Appl. Techn. 7,3/4, 285-292 (1994)
If one has obtained a representation of the elastic law, one still has to determine
the remaining material constants and the axes of anisotropy by experiments. This
procedure is called material identification. In the triclinic case, 21 constants have
to be determined, which is by no means easy. Even in the isotropic case, where
only two constants remain, this can be a challenging task. Theoretically, the
measurement of the straining and the transversal straining in a unidirectional
tensile test would be sufficient. For anisotropic materials, this is no longer
sufficient to completely determine the stiffness tensor. Part of the problem of the
material identification is the selection of appropriate tests and measurements.
A rather elegant method for the elastic material identifications is FÖRSTER54´s
resonance method (1937). One uses samples with a simple shape like a circular
cylinder, excites them, and measures some of the lowest resonance frequencies
(eigenfrequencies).
In the isotropic case analytic formulae for the eigenfrequencies are known if
standard assumptions hold (BERNOULLI´s hypothesis for bending of bars, de St.-
VENANT´s assumptions on the torsion of shafts, etc.). The i-th eigenfrequency of
 a longitudinal vibration of a slender bar with length L and mass density  is
54
Friedrich Förster (1908-1999)
4.1 Elasticity 191
E  i 
(4.1.26) fi =
  2L 
 for a transversal vibration a good approximation is
2
E I § 2 i 1 · S
(4.1.27) fi =
U A ¨© L ¸¹ 8
 and for a torsional vibration of a solid circular cylinder it is
G  i 
(4.1.28) fi =
  2L 
These formulae can be used to determine E and G if the corresponding
frequency has been measured.
In the anisotropic case things are more complicated. In general, neither the
eigenforms are known, nor do we have analytical formulae for the
eigenfrequencies. Instead, one can use approximation methods like the Finite
Element Method to determine the eigenmodes and eigenfrequencies, if only one
knew the elastic law including the elastic constants.
Our problem is an inverse problem: Determine the axes of anisotropy and the
elastic constants such that they render eigenfrequencies that coincide with the
measured ones. One can do this by minimising the square of the differences
between the measured and calculated eigenfrequencies R (1 , 2 , ... , n) . They
depend indirectly on the wanted parameters i . This leads mathematically to a
non-linear optimisation problem, for which a number of strategies have been
suggested like, e.g., the conjugate gradient method, the simplex method, or
evolution strategies.
The following picture of the eigenforms of a cylindrical bar made of a single
crystal has been taken from
J. OLSCHEWSKI: Die Ermittlung der elastischen Konstanten kubisch
einkristalliner Stoffe als Beispiel einer Systemidentifikation. Aus: Mechanik -
Beiträge zu Theorie und Anwendungen, edts. A. Bertram, M. Nasseri, R. Sievert,
TU Berlin, p. 162 (1988)
Most of the methods of material identification lead to non-linear optimisation
problems. These are considered in, e.g.,
Schwefel, H.-P.: Numerische Optimierung von Computer-Modellen mittels der
Evolutionsstrategie. Birkhäuser Vlg. Basel, Stuttgart (1977)
first transversal eigenform
second transversal eigenform
first torsional eigenform
third transversal eigenform
first longitudinal eigenform
second torsional eigenform
4.1.5 Elastic Energy
In the one-dimensional theory in Chapt. 1.2 we have already introduced the

concepts of stress power and elastic energy. In what follows, we generalise these
concepts to the three-dimensional case.
In the balance of power, the stress power (3.2.16) already appears
i =  πi dV
V
with the stress power density
4.1 Elasticity 193
πi : = T  D .
In the linear theory we can write
D = sym(grad v) = def v
= def u  (Def u) = (Def u) = E.
If we insert HOOKE´s law (4.1.1), we can cast the stress power in the form
(4.1.29) πi = T  E = E  T = E  C  E .
This equals the time-derivative of the function
(4.1.30) w(E) : = ½ E  C  E + w0
with some constant w0 only if C is symmetric. In fact, only in this case
E  C  E = E  C  E
holds for all E and E. Then
dw
(4.1.31) w(E) =  E = ½ E  C  E + ½ E  C  E
dE
= E  C  E = T  E = πi .
If this applies for all rates E, then the potential relation
dw  E 
(4.1.32) T =
dE
holds.
If E(t) , t0  t  t1 , is a strain process, and T(t) the corresponding stress process,
then the stress work between t0 and t1 is defined after (1.1.14) as the path
integral
t1 t1
a | tt1 =  πi dt =  T(t)  E(t) dt .
0
t0 t0
By HOOKE´s law we can eliminate the stresses or the strain rates

t1 t1
a | tt1 =  [C  E]  E(t) dt =  T  [S  T] dt
0
t0 t0
and, in case of a symmetric stiffness tetrad, we can integrate

t1
(4.1.33) a | tt1 =  w(E) dt = w(E(t1)) – w(E(t0)) .
0
t0
Definition (GREEN 1839): The (potential) elastic energy or strain energy is a

scalar function of the strain tensor w(E) such that along an arbitrary strain
process the stress work is
a | tt1 = w(E(t1)) – w(E(t0)) .

0
Since only differences of the energy appear in such expressions, the absolute value
of the energy remains undetermined. We can norm the energy by postulating
w(E = 0) = 0  w0 = 0 .
Existence theorem for the elastic energy
The following statements are equivalent for each linear elastic material.
1.) The stress work is zero for every cyclic strain process (E(t1) = E(t0)) .
2.) The stress work for every strain process depends only on its initial and final
value.
T
3.) The stiffness tetrad is symmetric: C = C
T
4.) The compliance tetrad is symmetric: S = S
5.) There exists a strain energy
w(E) = ½ E  C  E
dw  E 
as a potential for the stresses T =
dE
Proof.
"(3)  (5)": It is known from potential theory that a field T is conservative if and
only if its curl vanishes. The curl corresponds to the skew part of the gradient
d 2w E
C = dT = 2
.
dE dE
"(5)  (3)": The second differentiation of a function w(E) always leads to a
symmetric tetrad C .
"(3)  (4)": C is symmetric if and only if its inverse S is symmetric (assuming
its invertibility).
"(2)  (5)": If a material possesses an elastic energy, then we construct it by
calculating the work for an arbitrary symmetric E1
t1
w(E1) : =  T  E dt
t0
4.1 Elasticity 195
for proportional processes E(t) : = t E1 , t  [0 , 1] , which shall only depend on

the final value E(1)  E1 after our assumption. We see that the representation for
w(E) from (5) exactly leads to a linear elastic law.
"(5)  (2)": If the potential relation with w(E) from (5) holds for T , then the
power is integrable and depends only on the final value of the strain process.
"(2)  (5)": If the initial and final value of the strain process are equal (cyclic
strain process), then we obtain by (2)
a | tt1 = w(E(t1)) – w(E(t0)) = w(E) – w(E) = 0 .

0
"(1)  (2)": If (1) holds, and if E1 and E0 are arbitrary strains, then there surely
exists a cyclic strain process E(t) that starts from E(t0) = E0 and goes via
E(t1) = E1 again to E(t2) = E0 with t0 < t1 < t2 . The work done during this
process is
t2 t1 t2
t
a | t2 = 0 =  T  E dt =  T  E dt +  T  E dt
0
t0 t0 t1
t1 t1
=  T  E dt –  T  E dt .
t0 t2
Thus
t1 t1
 
 T  E dt =  T  E dt
t0 t2
for all processes with the same initial and final value, i.e., (2); q. e. d.
By using the compliance tetrad, we can also introduce a potential for the strains
in the stress space
w(E) = w(S  T) = ½ T  S  T = : w*(T)
so that
dw  T 
E = .
dT
w*(T) is the complementary elastic energy or stress energy.
The potential property of the stresses or, equivalently, of the strains is an
additional postulate for elastic materials. If such potentials exist, we call such
materials hyperelastic (sometimes also called perfect elastic or GREEN-elastic).
We see the implication
hyperelastic  elastic ,
but not vice versa. Many authors, however, do not distinguish between
hyperelasticity and elasticity.
In the isotropic case, the strain energy can be expressed by LAMÉ´s constants
as
 
(4.1.34) w(E) = (E  I)2 +  E  E = tr2(E) +  tr(E2)
2 2
so that
dw  E 
=  (E  I) I + 2  E = T .
dE
This corresponds to the decomposition (2.1.102)
 1 2
w(E) = tr2(E) +  tr[ tr (E) I + E'2]
2 9
 
(4.1.35) = ( + ) tr2(E) +  tr(E'2)
3 2
= wC(tr E) + wD(E')
into the compression energy wC and the distortional energy wD . In the
isotropic case, such an energy always exists. Thus, isotropic linear elastic
materials are also hyperelastic.
Apart from the symmetry of the elastic tetrads, another property is also
important. A (not necessarily symmetric) tetrad C is called positive definite if
A  C  A  0
holds for all A  0 . Clearly, in this definition only the symmetric part of the
tetrad enters. If C is symmetric, then the positive definiteness is equivalent to the
positivity of all of its eigenvalues.
One may assume that one has to do positive work for any deformation out of
some stress- and strain-free initial state. The following theorem gives us a simple
criterion for this property.
Theorem. For a linear elastic material, the work done on an arbitrary strain
process beginning at E = 0 but not being circular, is positive if and only if the
stiffness tetrad is positive definite.
The proof of this theorem can be found, e.g., in GURTIN (1972) p. 84.
In the isotropic case, the stiffness tetrad is positive definite if and only if the
following equivalent conditions are fulfilled
 G0  k  0
 0  3 + 2  0
 E0  –1 < v < ½
 G>0  –1 < v < ½
4.1 Elasticity 197
This can easily be shown by the positivity of the eigenvalues of C . Note that
v  0 is not required. Materials with negative POISSON´s ratio do exist in reality
and are called auxetic.
4.1.6 Boundary Value Problems in Elastostatics
Literature
Press (2004, 2011)
Fichera, G.: Existence Theorems in Elasticity. In: Encyclopedia of Physics. Edt.
Knops, R. J.; Payne, L. E.: Uniqueness Theorems in Linear Elasticity. Springer,
Berlin (1971)
In this chapter we restrict our concern to

 static or quasi-static problems. For this class of problems, the inertia term  a
is negligible in C1 if compared to the other terms
 isotropic or anisotropic linear elastic bodies, so that HOOKE´s law is valid
 this shall also be hyperelastic, so that an elastic energy exists
 homogeneous bodies, for which the elastic law is the same in all points
 and we choose the reference placement for all points as being stress free
(u  o  T  0 )
Accordingly, for all body points, the following equations hold
 displacement-strain relations (3.1.23)
(DSR) E  def u : = ½ (grad u + grad T u)
and as a consequence
 the compatibility conditions (3.1.34)
(INC) inc E : =   E   = 0
 the equilibrium conditions (3.2.7)
(C1) div T +  b = o
 and HOOKE´s law (4.1.1)
(HOOKE) T = C  E
or in the isotropic case (4.1.17)

T =  tr(E) I + 2 E
The field of the body forces b shall be known.
The balance of angular momentum is fulfilled by the symmetry of the stress
tensor resulting from HOOKE´s law and will not be mentioned anymore.
Two alternative methods for the treatment of elastostatic problems are used.
1.) Elimination of stresses

We insert HOOKE into C1 and DSR into HOOKE and obtain the displacement
equations
(4.1.36) div (C  def u) +  b = o .
In the isotropic case, this results in
div [ tr(def u) I + 2 def u] +  b = o .
With
tr E = tr(def u) = tr(grad u) = div u
and the chain rule for each scalar field  and each tensor field T
(2.2.41) div ( T) = T  grad +  div T
we obtain
div [tr(def u) I] = I  grad tr(def u) = grad div u .
Because of (2.2.47) we get
div grad u =  u = grad div u – curl curl u
and due to (2.2.48)
div grad T u = (  u)   =  (u  ) = grad div u .
Consequently
div def u = ½ (div grad u + div grad T u)
= ½ (div grad u + grad div u)
= grad div u – ½ curl curl u .
55
This results in NAVIER ´s displacement equations (1821) (ND)
(4.1.37) ( + ) grad div u +  div grad u +  b = o
= ( + ) (u  )  +   u +  b
= ( + 2) grad div u –  curl curl u +  b
or equivalently
55
Claude Louis Marie Henri Navier (1785-1836)
4.1 Elasticity 199
1 
div grad u + grad div u + b = o.
1  2 
If the body forces are negligible (b  o), or  b is constant in space, the
divergence of ND with the identity (2.2.51) gives
( + ) div grad div u +  div div grad u
= ( + 2) div grad div u = 0 .
This means that the trace of the strain tensor is a harmonic function
(4.1.38) div grad div u =  tr E = 0 .
If we instead apply the LAPLACE operator to ND, then we obtain with the
identity (2.2.49)
( + ) div grad grad div u +  div grad div grad u
(4.1.39) = ( + ) grad div grad div u +  div grad div grad u .
With the previous results, this leads to the bipotential equation
  u = o.
Thus, for constant body forces, the displacements must be biharmonic functions.
These field equations are called displacement equations and are solved by
appropriate ansatz functions for the displacements.
2.) Elimination of the displacements

We use the inverse form of HOOKE´s law and insert it into INC (3.1.34)
inc E = inc (S  T) = 0 .
For the isotropic case, this gives
1
inc E = {(1 +) inc T –  inc[tr(T) I]} = 0 .
E
If no body forces act (b  o), or if they are constant in space, this can be
formulated as
1
(4.1.40) div grad T + grad grad tr T = 0
1 
known as BELTRAMI´s equation (1892). For constant body forces we see after
twice applying the LAPLACE operator, that the stresses are biharmonic functions
 T = 0
while the trace of the stress tensor is harmonic
 tr T = 0 .
If one uses ansatz functions for the stresses, it must be ensured that the resulting
stress fields satisfy the above conditions.
If one wants to gain a unique solution of the field equations, one needs
boundary conditions. We distinguish between two types of boundary conditions.
1.) geometrical or displacement boundary conditions or DIRICHLET56
conditions: particular components or the entire displacement u is prescribed for
boundary points;
2.) dynamic or traction boundary conditions or NEUMANN57 conditions:
particular components or the entire traction vector t = T  n is prescribed for
boundary points.
It is also possible that both types of conditions are given for the same boundary
points. In any case, one needs precisely three linear-independent components for
the tractions or displacements.
We talk about a displacement boundary value problem, if displacements are
prescribed on the whole boundary of the body. If only dynamic boundary
conditions are given, it is a traction boundary value problem. In all other cases,
we have a mixed boundary value problem. This is the most important case,
which we introduce in the following way.
Prescribed are
 the specific body force b in the interior of the body V
 the tractions tpr on one part Ad of the surface
 the displacements upr on the complementary part Ag of the surface
Both parts of the surface shall be almost everywhere disjoint and shall form the
entire surface
Ad  Ag = A .
We will next consider some important properties of such boundary value
problems.
Since all field equations of the classical elasticity theory are linear, we have the
Principle of Superposition
Let {ui , Ei , Ti} for i = 1, 2 be solutions of two mixed boundary value problem for
given body forces bi , surface tractions ti on the part Ad of the surface, and
displacements ui on the complementary part Ag .
Then for all reals 1 , 2  R , the same equations are fulfilled by
{1 u1 + 2 u2 , 1 E1 + 2 E2 , 1 T1 + 2 T2} for the body forces 1 b1 + 2 b2 ,
surface tractions 1 t1 + 2 t2 on Ad , and displacements 1 u1 + 2 u2 on Ag .
The following Theorem holds for all linear hyperelastic materials.
56
Johann Peter Gustav Lejeune Dirichlet (1805-1859)
57
Carl Gottfried Neumann (1832-1925)
4.1 Elasticity 201
Reciprocal Theorem of BETTI58 (1872)

Let {ui , Ei , Ti} for i = 1, 2 be two solutions of the boundary value problem for
given body forces bi and surface tractions ti . Then for all hyperelastic materials
the following equation holds
(4.1.41)  t1  u2 dA +  b1  u2 dm =  t2  u1 dA +  b2  u1 dm
A V A V
Proof. For linear hyperelastic materials, the stiffness tetrad is symmetric, as we
have already seen. For two arbitrary states, we have
T
T1  E2 = (C  E1)  E2 = (C  E2)  E1
= (C  E2)  E1 = T2  E1 .
By the PVP (3.2.18) for both states we obtain (4.1.41); q. e. d.
Theorem of CLAPEYRON59 (LAMÉ 1852)
For a body consisting of some homogeneous linear hyperelastic material loaded
by surface tractions t and body forces b ,
(4.1.42) ½  b  u dm + ½  t  u dA =  w(E) dV
V A V
holds with the resulting displacement field u .
Proof. By the PVP (3.2.18) we conclude for u  u
 b  u dm +  t  u dA =  T  E dV
V A V
=  E  C  E dV =  2 w dV
V V
q.e.d.
For the interpretation of this equation we present the following rationale. Since
the principle of superposition holds, we can conclude for every boundary value
problem of a linear elastic body:
 if the loads (t , b) cause a displacement field u ,
 then the loads ( t ,  b) cause a displacement field  u for each real
factor  .
Now let  (t) be an arbitrary differentiable function of time in the unit interval
starting at  (0) = 0 and ending with  (1) = 1. After (3.2.15), the power of the
external loads during the loading process ( t ,  b) is
 
Πa(t) =  b(t)  u dm +  t(t)  u dA
V A
58
Enrico Betti (1823-1892)
59
Benoit Paul Emile Clapeyron (1799-1864)
 
=   (t) b  u  (t) dm +   (t) t  u  (t) dA
V A

= [  b  u dm +  t  u dA]  (t)  (t) .
V A
The time integral over the power is the work done on a quasistatic load process
starting at the unloaded state and being proportionally augmented
1
 Πa(t) dt
t 0
1
= [  b  u dm +  t  u dA]   (t)  (t)dt
V A t 0
1
= [  b  u dm +  t  u dA]  ½  (t)2  dt
V A t 0
= ½  b  u dm + ½  t  u dA .
V A
This is the left side of equation (4.1.42). The right-hand side is the elastic energy
stored during this process. Thus, (4.1.42) balances the internal and external works.
Uniqueness Theorem (KIRCHHOFF60 1859)

For a linear elastic material with a positive definite stiffness tetrad, two solutions
u1 and u2 of the mixed boundary value problem differ only by an infinitesimal
rigid body motion
u1(X) = u2(X) + u0 +   (X – X0)
with two constant vectors u0 and X0 and a constant antisymmetric tensor  . If
the displacements are prescribed for at least three points that do not lie on a
straight line, then the solution is unique
u1(X) = u2(X) .
Proof. After the assumptions of the Theorem, the principle of superposition
renders the following statement: u : = u1 – u2 solves the mixed boundary value
problem for b  o , t  o on Ad and u  o on Ag . After the PVP (3.2.18) we
obtain for the zero-boundary value problem
 b  u dm +  t  u dA +  t  u dA =  T  E dV = 0
V Ag Ad V
with E = def u , since at least one factor under the integrals on the left-hand side
is zero in the region of integration. Because of the assumed positive definiteness
of the stiffness tetrad,
60
Gustav Robert Kirchhoff (1824-1887)
4.1 Elasticity 203
T  E = (C  E)  E
must be non-negative everywhere. The integral over it can only be zero, if the
integrand is zero everywhere
E = 0 .
Thus, the displacement field u can only be strain-free, and with (3.1.24)
u(X) = u0 +   (X – X0) = u1 – u2 .
This concludes the proof of the first part of the Theorem. If Ag contains at least
three points that are not collinear, for which the displacements are prescribed, then
u must vanish in these points, so that u is the zero field everywhere;
q. e. d.
Problem 13. Elastostatic Boundary Value Problem

Formulate the equilibrium conditions C1 without body forces and the
isotropic HOOKE´s law (4.1.1) in cylindrical coordinates for the case that
all fields show a rotational symmetry. Eliminate the stresses in C1 by
HOOKE´s law. Simplify the equations further by excluding a rotation or
torsion of the body and by assuming an independence of the fields with
respect to the axial coordinates  and z . Find a general solution of the
resulting PDE.
Solution
We begin with div T = o , the only remaining term of C1. In Cartesian
coordinates, this gives

div T = (ij ei  ej)  ek .
xk
In cylindrical coordinates, we would identify x1  r, x2   , x3  z . In each

derivative with respect to  we have to adjoin the factor 1/r , and to take
into account the relations e1 , 2 = e2 and e2 , 2 = – e1 . The product rule
gives
div T = ij , k jk ei + ij jk ei , k + ij ei  ej , k  ek
= ij , j ei + Tij ei , j + ij ei  ej , k  ek .
Among the derivatives of the base vectors only e1 , 2 and e2 , 2 are non-
zero, so that
div T = ij , j ei + 12 e1 , 2 + 22 e2 , 2 +i1 ei  e1 , 2  e2 + i2 ei  e2 , 2  e2 .
The last term is zero because of the orthogonality
e2 , 2  e2 = – 1/r e1  e2 = 0
so that the sums give only
div T = (rr , r + 1/r r ,  + rz , z) er

+ ( r , r + 1/r  ,  + z , z) e
+ (zr , r + 1/r z ,  + zz , z) ez
+ 1/r r e – 1/r  er + 1/r (rr er +  r e + zr ez)
= [rr , r + 1/r r ,  + rz , z + 1/r (rr – )] er
+ ( r , r + 1/r  ,  + z , z +2/r  r) e
+ (zr , r + 1/rz ,  + zz , z + 1/r zr) ez
In the presence of a rotational symmetry, this can be drastically reduced.
Firstly, all derivatives of ij with respect to  vanish. Secondly, all
planes ( constant, r and z arbitrary) are symmetry-planes for the
strains and therefore free of shears. For an isotropic material also the
corresponding shear stresses must vanish, i.e., r = r = z = z = 0.
The only remainders of the equilibrium conditions are
(P13.1) div T = [rr , r + rz , z + 1/r (rr – )] er
+ (zr , r + zz , z + 1/r zr) ez .
Starting with a displacement field u = ui ei , one obtains the strain tensor

E = sym(ui ei  ej)
x j
= sym(ui , j ei  ej + ui ei , j  ej)
= sym(ui , j ei  ej + u1 e1 , 2  e2 + u2 e2 , 2  e2)
= sym(ui , j ei  ej + u1/r e2  e2 – u2/r e1  e2) .
By assuming rotational symmetry, again all derivatives with respect to 
vanish. Moreover, we can assume u = 0 if we exclude torsion and
rotation of the body. Then
E = ur , r er  er + uz , z ez  ez + ur /r e  e
+ ½ (uz , r+ ur , z) (er  ez + ez  er) .
By means of HOOKE´s law we can determine the strains using the form
(4.1.17) with LAMÉ´s constants. With
tr E = ur , r + uz , z + ur /r
we obtain the stresses
(P13.2) rr =  (ur , r + uz , z + ur /r) + 2 ur , r
 =  (ur , r + uz , z + ur /r) + 2 ur /r
zz =  (ur , r + uz , z + ur /r) + 2 uz , z
4.1 Elasticity 205
rz =  (uz , r+ ur , z) .
If we insert P13.2 into P13.1, we obtain two coupled PDEs for the two
fields ur and uz . For the component in the erdirection, this is
0 =  (ur , rr + uz , zr + ur,r /r– ur /r2) +  (uz , rz+ ur , zz)
+ 2 (ur , rr + ur , r /r – ur /r2)
and in the ezdirection
0 =  (uz , rr + ur , zr) +  (ur , rz + uz , zz + ur,z /r)
+ 2 uz , zz +  (uz , r /r +ur , z /r) .
In many technologically interesting problems, no dependence of the
geometry, boundary conditions, stresses, and strains on the ezdirection
occurs. Then
E,z = 0
or
rr , z = ur , rz = 0
 , z = ur , z / r = 0
zz , z = uz , zz = 0 .
Furthermore, the ezplane is a symmetry-plane for E . Consequently, the
last shear component must also vanish
rz = 0 =  (uz , r+ ur , z) .
Therefore,
ur , z = uz , r = ur , rz = uz , zz = 0 .
This allows for an ansatz
(P13.3) uz = c + d z
with two constants c and d . We want to exclude rigid body motions in
the direction of ez by assuming that the origin of the coordinate system is
fixed. This leads to c = 0 . The remaining linear part with the factor d
stands for a constant strain in the ezdirection. The equilibrium condition
leads to the non-trivial equation
0 = rr , r + (rr – ) / r
or in terms of displacements
(P13.4) 0 = ( + 2) (ur , rr + ur , r /r – ur /r2)
with only one unknown function ur(r) . It is interesting that the elastic
constants drop out of the final differential equation. However, by traction
boundary conditions they can indirectly influence the solution. The above
ODE can be solved by the polynomial ansatz

ur =  ai r i
i  
so that
 
ur , r =  i ai r i–1 ur , rr =  i (i –1) ai r i–2
i   i  
If we insert this into the ODE, then the terms with i 1 and i  –1 must
vanish. Accordingly, a1 = a and a –1 = b may have arbitrary values, while
all other ai must be zero. Since we have identified two parameters of a
second-order ODE, the solution
(P13.5) ur = a r + b / r
is complete.
The special cases of plane stress or plane strain in the ezplane are obtained
by zz = 0 and zz = 0 = d , respectively. In each case, one needs one
equation for the complete determination of the stress and displacement
field, which either relates zz to zz or prescribes one of them.
Due to the assumption that neither T nor E depend on z , the PDEs for
ur(r , z) and uz(r , z) become uncoupled, so that two ODEs in ur(r) and
uz(z) remain, the solutions of which we have already found with P13.3 and
P13.5.
Problem 14. Pull-out of a Bar

Use the results of Problem 13 for an approximate calculation of the force
needed to pull a bar (undeformed radius r0) out of a sleeve (outer radius
ri , inner radius ra , length l ). The sleeve is made of a soft material (the
LAMÉ´s constants  and  are known). In contrast to this, we assume
rigidity of the bar. The coefficient of friction µ0 is also known. The sleeve
is fixed against moving in the axial direction by design.
r0
ra
l ri
4.1 Elasticity 207
We adopt the solution P13.4 from Problem 13 ur = a r + b/r for the sleeve
by imposing boundary conditions
ur (ri) = r0 – ri
rr (ra) = 0
since the inner radius is extended to the radius of the bar r0 , while the
outer mantle of the sleeve remains load-free. The first boundary condition
gives
(P14.1) r 0 – ri = a ri + b / r i .
The second boundary condition is inserted into P13.2. This gives with
ur , r = a – b/r2
rr =  (ur , r + uz , z + ur / r) + 2 ur , r
= 2  a + 2 µ(a – b / r2) +  uz , z .
By impeding axial strains of the sleeve, we get uz , z = 0 . With rr(ra) = 0
we have
(P14.2) 0 = 2 a  +2 (a – b/ra2) .
P14.1 and P14.2 form a system of linear equations for the unknown
coefficients a and b with the solutions
a = ri (r0 – ri) / [ra2(1 +  / µ) + ri2 ]
b = ri (r0 – ri) / [1+ µ (ri / ra)2 / (µ + )] .
With this we can determine ur and rr . For the radial stresses at the inner
core we find
2µ(r0  ri )(ri 2  ra 2 )
rr(ri) = .
ra 2 ri  µri 3 /( µ   )
By using the values ri = 10 mm, ra = 20 mm, r0 = 11 mm, E = 50 MPa,

and  = 0.49, we can show the functions rr (r) ,  (r) and ur (r) . E
and  are close to the elastic constants of rubber polymers. This gives for
LAMÉ´s constants the values µ = 16.8 MPa and  = 822 MPa (see the
table in Chapt. 4.1.2).
Expansion in mm (above), and radial (solid line) and tangential

stresses (dashed line) in MPa versus the radius of the sleeve in mm.
The force Fz needed for pulling out the bar, can be estimated by use of
COLOUMB´s law of friction, |Fz| < µ0 |Fn| . The critical value is
|Fz| = µ0 |Fn| . The force Fn results from the radial stress at the inner
surface of the sleeve times the area
Fn = rr (ri) Ai = rr (ri) 2 ri l .
This, however, is only an estimation. By pulling out the bar, the
homogeneous state in the axial direction will be distorted, so that the real
value for Fz might be different from ours.
We should note that stress functions are not helpful in this example since
displacement boundary conditions have to be fulfilled.
Problem 15. Rotating Compact Disc

Enlarge the solution from Problem 13 to describe a rotating disc. Apply the
solution to a freely rotating CD made of polycarbonate with the following
data: n = 15,000 rounds/min,  = 1.2 g /cm3, E = 2.2 GPa,  = 0.37, ri =
7.5 mm, ra = 60 mm, max. tensile stress 55 to 75 MPa , and compute the
radial and tangential stresses and the expansion versus the radius.
4.1 Elasticity 209
Solution
If we want to use the results from Problem 13, we can implement the
inertial effect by appropriate centrifugal (body) forces. This can be made in
the following way.
The centripetal acceleration directs radially towards the centre. It is
quadratic in the angular velocity and linear in the radius
r =  r  2 er .
C1 gives in this case
div T =  r = –  r  2 er .
We have already calculated div T in Problem 13. A comparison in the
direction of er renders
–  r  2 = ( + 2) (ur , rr + ur , r / r – ur / r2)
in analogy to P13.3. This can be reformulated into
0 = ur , rr + ur , r / r – ur /r2 + C r
with the abbreviation
C =   2/( + 2) .
In analogy to the procedure in Problem 13, we solve this ODE by the
polynomial ansatz function
ur = a r + b/r – C r3/8 .
By inserting this into the radial and tangential stresses we obtain
(P15.1) rr =  (ur , r + uz , z + ur /r) + 2 ur , r
=  (2a – C r2 / 2 + uz , z) + 2 ( a – b / r2 – 3 C r2 / 8 )
 =  (2a – C r2 / 2 + uz , z) + 2 ( a + b / r2 – C r2 / 8) .
In the z-direction we assume either uz , z = 0 or zz = 0 . This corresponds
to the assumption of a plane strain or plane stress state, respectively. The
last assumption renders
zz =  (ur , r + uz , z + ur /r) + 2 uz , z = 0
which gives with the solution for ur from above
uz , z = (C r2/2 – 2 a ) / ( + 2) .
For the boundaries, we assume the conditions rr (ri) = 0 and rr (ra) = 0 .
We again obtain a system of linear equations for the parameters a and b ,
the solution of which depends on the above choices (plane strain or plane
stress). The constant C has the value 0.761 m–2, again by transforming the
elastic constants E and  by using the table in Chapt. 4.1.2 , and putting
 = 2 n . For the plane strain case, we obtain a = 0.786 10–3 and
b = 0.167 mm2, and for the plane stress state a = 0.855 10–3 and b = 0.103
mm2. This can be inserted into the equations for rr and  .
Radial displacement in µm (left) and radial and tangential stresses in

MPa (right) over the radius in mm. The dotted lines correspond to the
plane stress solution, the solid lines correspond to the plane strain
solution.
The radial stresses have to vanish on the edges according to the boundary
conditions, while the tangential stresses obtain their maxima at the inner
edge. The stresses are lower in case of a plane stress state. This is not
surprising, since the hindrance of transversal strains is a constraint. The
expansion is relatively constant in the radial direction.
CD drivers run with up to 11,200 rounds/min. If polycarbonate has a
tension resistance of 50 MPa, we can apply this limit for the tangential
stress at the inner edge and thus determine the critical spin as n  34,300
rounds/min in the case of plane strain, and n  43,700 rounds/min for
plane stress. Experiments give in most cases values already between 23,000
and 25,000 rounds/min. The difference to our calculation can have many
reasons. One is that we ignored micro fissures. In addition, the boundary
condition rr(ri) = 0 is not realistic. But, more importantly, we did not
take into account vibrations of the disc.
4.1 Elasticity 211
4.1.7 Variational Principles in Elastostatics
Literature
Press (2004, 2011)
Washizu, K.: Variational Methods in Elasticity and Plasticity. Oxford (1968)
For applying approximation methods like the Finite Element Method or the
Boundary Element Method, it is helpful to introduce variational forms of the
equilibrium conditions.
Variational principles deal with extremal properties of functionals. A functional
is a scalar-valued function defined on fields on the body, like integrals. A typical
example of such a variational principle is the
Principle of Minimum Potential Energy
Let a body consist of a hyperelastic material with a positive definite stiffness
tensor C in each point. Let U be the set of all (compatible) vector fields u(X)
on the body which fulfil the displacement boundary conditions.
We define the following functional  : U  R
(4.1.43) (u) : =  w dV –  b  u dm –  tpr  u dA

V V Ad
with w = ½ E  C  E and E = def u

for prescribed b in V and tpr in Ad .
Then the functional  obtains a minimum for the solution u0 of the mixed
boundary value problem, i.e.,
(u0)  (u) uU .
If equality holds, then u0 and u differ only by a rigid body displacement
u – u0 = c +   (X – X0)
with c , X0 constant vectors and  an antimetric tensor.
Proof. Let u0 be the solution, u  U arbitrary, and u : = u – u0 . Then u
fulfils zero-displacement boundary conditions on Ag . Let
E : = def u and T : = C  E .
Then
 w(def u) dV –  w(def u0) dV

V V
= ½  def u  C  def u dV – ½  def u0  C  def u0 dV

V V
= ½  (def u – def u0)  C  (def u – def u0) dV

V
+  (def u0)  C  (def u – def u0) dV

V
= ½  E  C  E dV +  T0  E dV
V V
=  w(E) dV +  T0  E dV with T0 : = C  def u0 .

V V
By the divergence theorem (2.2.67) we have with C2 and the chain rule for zero-
displacement boundary conditions in Ag = A \ Ad
 T0  E dV =  T0  grad u dV
V V
=  (u  T0)   dV   u  (T0  ) dV

V V
=  u  (T0  n) dA –  u  div(T0) dV
A V
=  t0  u dA –  u  div(T0) dV with t0 = T0  n
A V
=  t0  u dA –  u  div(T0) dV .
Ad V
Thus
(u) – (u0) =  w(def u) dV –   b  u dV –  tpr  u dA
V V Ad
–  w(def u0) dV +   b  u0 dV +  tpr  u0 dA

V V Ad
=  w(E) dV +  t0  u dA –  u  div(T0) dV
V Ad V
4.1 Elasticity 213
–   b  u dV –  tpr  u dA
V Ad
=  w(E) dV +  (t0 – tpr)  u dA

V Ad
–  u  (div T0 +  b) dV .
V
Since u0 solves the mixed boundary value problem, the two last integrals are
zero. Because of the positive definiteness of C we have w(E)  0 and therefore
(u)  (u0) .
Equality holds if and only if
def u = def u0 .
So the two displacement fields can only differ by an (infinitesimal) rigid body
motion; q. e. d.
The above principle tells us that the difference between the elastic energy and
the final work of the body forces and the (prescribed) surface loads takes a smaller
value for the solution of the mixed boundary value problem than for any other
displacement field. By the way, the uniqueness theorem can also immediately
concluded by this principle.
One can show that the reverse of the principle of minimum potential energy also
holds: Let u0  U . If (u) is larger than (u0) for all other u  U , then u0
is the solution of the mixed boundary value problem.
Complementary to the above energy principle is the following.
Principle of Minimum Complementary Energy
Let a body consist of a hyperelastic material with positive definite compliance
tensor S in each point. Let T be the set of symmetric stress fields that obey the
equilibrium conditions (C1)
div T +  b = o
and the stress boundary conditions tpr = T  n on Ad .
Then the functional
(4.1.44) (T) : =  w*(T) dV –  upr  t dA
V Ag
with the complementary elastic energy w*(T) obtains for the solution T of the
mixed boundary value problem a minimum in T .
The reverse of this principle has been shown yet only for simply connected and
convex (for Ag) bodies (see GURTIN, 1972 p. 117).
While minimal principles can be used to construct upper bounds for

(generalised) energies like  and  , maximum principles give us lower bounds.
Here we give one example for such an energy principle.
Principle of Maximum Potential Energy
The body consists in each point of a hyperelastic material with symmetric and
positive definite stiffness tensor C . Let u0 be the solution of the mixed boundary
value problem with b in V , tpr in Ad , and upr in Ag prescribed. Let u be
another displacement field which solves the same boundary value problem with
the exception of the geometrical boundary conditions. For the latter the following
condition is assumed to hold
(4.1.45)  t  (upr – u) dA  0 with t = T  n = [C  (def u)]  n .
Ag
Then the functional (4.1.43) has a maximum for the solution

(u0)  (u) .
Equality holds if and only if u and u0 differ only by an (infinitesimal) rigid body
motion.
Proof. Let u : = u0 – u (differently from before). We have already shown in the

proof of the minimal principle that
 w (def u0) dV –  w (def u) dV =  w (def u) dV +  T  E dV

V V V V
holds. After the principle of virtual power (PVP) we have
 T  E dV =  b  u dm +  t  u dA .
V V A
Since u0  upr on Ag and t  tpr on Ad , we conclude
(u0) – (u)
=  w (def u0) dV –  w (def u) dV
V V
–  b  (u0 – u) dm –  tpr  (u0 – u) dA

V Ad
=  w (def u) dV +  b  u dm +  t  u dA
V V A
–  b  u dm –  tpr  u dA
V Ad
4.1 Elasticity 215
=  w (def u) dV +  t  (upr – u) dA .

V Ag
After our assumptions, both terms cannot be negative. Therefore, (u0) is the
maximum; q. e. d.
Analogously one can show that those displacement fields u , for which
u = upr auf Ag and  (tpr – t)  u dA  0
Ad
hold, the functional  has a maximum for the solution u0 of the mixed
boundary value problem (Principle of Maximum Complementary Energy).
For all previously mentioned principles, the admissible displacement fields have
to fulfil certain field equations or boundary conditions. For practical reasons,
however, one is interested in principles with functions acting on fields with fewer
or even no restrictions. For such principles we will give two examples.
Principle of DE VEUBEKE61 (1951)  HU (1955)  WASHIZU (1955)

The body consists of a hyperelastic material and is due to mixed boundary
conditions: b prescribed in the interior, upr on Ag , tpr on Ad . For the
equilibrium solution {u0 , E0 , T0} the functional
(4.1.46) (u , E , T) : = ½  E  C  E dV –  T  E dV
V V
–  (div T +  b)  u dV +  t  upr dA +  (t – tpr)  u dA

V Ag Ad
with t = T  n has a stationary value among all smooth fields

u : vector field
E : symmetric tensor field
T : symmetric tensor field;
and vice versa.
Proof. A necessary and sufficient condition for a stationary value is that the
derivatives at {u , E , T} in the directions of arbitrary fields {u , E , T} after
(2.2.1) vanish
(u , E , T , u , E , T)
d
:= (u +  u , E +  E , T +  T) = 0.
d  0
61
Baudouin M. Fraeijs de Veubeke (1917-1976), Hu Haichang (1928-2011),
Kyuichiro Washizu (1921-1981)
Due to the symmetry of the stiffness tetrad C we conclude for the first term in
the variation of 
½  (E +  E)  C  (E +  E) dV = ½  E  C  E dV
V V
+  E  C  E dV +  ½  E  C  E dV .
2
V V
After differentiation with respect to  at   0 , this is
 E  C  E dV .
V
Thus
(u , E , T , u , E , T)
=  [(C  E – T)  E – T  E – (div T +  b)  u – u  div T] dV
V
+  t  upr dA +  [t  u + (t – tpr)  u] dA
Ag Ad
with t = T  n . Using the chain rule, the divergence theorem, and the
symmetry of T , we find
 (div T)  u dV
V
=  div (T  u) dV   T  grad u dV

V V
=  t  u dA –  T  (def u) dV
A V
and therefore
(u , E , T , u , E , T)
=  [(C  E – T)  E – (div T +  b)  u + (def u – E)  T] dV
V
+  t  (upr – u) dA +  (t – tpr)  u dA .
Ag Ad
If {u , E , T} solves the mixed boundary value problem, then  = 0 for all

u , E , and T . If in turn  = 0 , then because of the arbitrariness of u , E
and T, all expressions in brackets must vanish. This is only the case if
{u , E , T} is a solution; q. e. d.
4.1 Elasticity 217
While we did not pose any restrictions on the fields u , E , T for the previous
variational principle (only differentiability), for the one below we will assume the
DSR.
Principle of HELLINGER – REISSNER62 (1950)
The body consists of hyperelastic material and is submitted to mixed boundary
conditions (the same as before). Then the solution of the boundary value problem
is a stationary value for the functional
(4.1.47) (u , T) : =  (½ T  S  T – T  E +  b  u) dV
V
+  t  (u – upr) dA +  tpr  u dA
Ag Ad
with t = T  n and E = def u among all smooth fields

u : vector field
T : symmetric tensor field,
and vice versa.
Proof. We take the directional differential of the functional in the direction of

{u , T} at {u , T} (as we already did in the previous proof), and obtain by the
divergence theorem
(u , T , u , T)
=  (T  S  T – T  E – T  E +  b  u) dV
V
+  t  (u – upr) dA +  t  u dA +  tpr  u dA
Ag Ag Ad
=  [(S  T – E)  T + (div T +  b)  u] dV

V
–  t  u dA +  t  (u – upr) dA
A Ag
+  t  u dA +  tpr  u dA
Ag Ad
=  (S  T – E)  T dV +  (div T +  b)  u dV
V V
+  (u – upr)  t dA +  (tpr – t)  u dA .
Ag Ad
62
Ernst David Hellinger (1883-1950), Eric Reissner (1913-1996)
These variations are all zero if and only if {u , T} is the solution; q. e. d.
Principle solution conditions boundary

of for for the conditions
admissible fields
for the admissible
fields
Minimum of the strain displacements DSR, HOOKE displacement
energy boundary cond.
Minimum of the stress stresses C1 traction
energy boundary cond.
Maximum of the strain displacements DSR, HOOKE, traction
energy C1 boundary cond.
displacement b. c.
in weak form
Maximum of the stress displacements DSR, HOOKE, displacement b. c.
energy C1 traction b. c.
in weak form
DE VEUBEKE- displacements - -
HU-WASHIZU stresses
strains
HELLINGER- displacements DSR -
REISSNER stresses
The table gives an overview over the variational principles and their conditions
(after GURTIN, 1972).
4.1.8 Displacement Functions

If we eliminate the stresses in the equilibrium conditions C1 by HOOKE´s law
and the strains by the displacements with DSR, then we obtain PDEs in the
displacements. In the isotropic case these are NAVIER´s displacement equations
ND. These also hold in non-equilibrium if one adds inertia to the body forces
b – a , which renders this field dependent on time and place. The resulting
displacement equations thus imply all field equations. They reduce the problem to
find a displacement field as its solution, which has to be adopted to the boundary
conditions.
In the general (anisotropic) linear elastic case, such general representations
hardly exist. We thus limit our concern from the beginning to isotropy. The
displacement equations ND can be given the following form
1 
(ND) div grad u + grad div u = – b
1  2 
4.1 Elasticity 219
with given body forces b . We further assume that for the elastic constants the
stability requirements  > 0 and ½ >  > –1 hold, and thus the stiffness tetrad is
positive definite. We are looking for solutions u(X) of ND in V which are twice
continuously differentiable. For this purpose some general ansatz functions shall
be represented.
First of all, we apply the following representation theorem to the field of the
body forces b .
Representation Theorem of HELMHOLTZ63

Each smooth vector field b on V can be represented as a sum of a potential field
and a curl
(4.1.48) b = grad  + curl g
with  scalar field on V
g divergence free vector field on V .
Proof. For each scalar field  (x) the NEWTON potential

1  y
 (x) : = –  dV
4 xy
V
gives  (x) =  (x) . Analogous potentials exist also for vector and tensor fields.
Thus, for a vector field b there also exists a potential field v , and with (2.2.47)
we have
b = v = div grad v = grad div v – curl curl v .
With  : = div v and g : = – curl v we obtain the representations of the
theorem; q. e. d.
Theorem. HELMHOLTZ´s displacement function
Let
(4.1.49)  b = grad  + curl g
be the HELMHOLTZ representation of the body force. Then the following function
solves the ND
u = grad  + curl w
with

div grad  = –
  2
and
g
div grad w = – .

Proof. We insert the ansatz into ND
63
Hermann Ludwig Ferdinand von Helmholtz (1821-1894)
1
div grad u + grad div u
1  2
= div grad grad  + div grad curl w
1
+ (grad div grad  + grad div curl w)
1  2
with (2.2.46), (2.2.47) and (2.2.52)
= grad div grad  – curl curl grad  + curl div grad w
1
+ (grad div grad  + grad 0)
1  2
1  g
with (2.2.43) = (1 + ) grad(– ) + curl(– )
1  2   2 
2  2 1 1
= – grad  – curl g .
1  2   2 
2  2   2
A calculation with the elastic constants gives = , and therefore
1  2 
1 1 
= – grad  – curl g = – b
  
and thus fulfils the ND; q. e. d.
However, this ansatz is not general, in contrast to the following.
Theorem. Displacement function of BOUSSINESQ (1885)  SOMIGLIANA
(1889)  GALERKIN (1930) 64
Let g be a vector field on V which fulfils the bi-potential equation

(4.1.50) g = – b.

Then the displacement function
1
(4.1.51) u = div grad g – grad div g (BSG)
2  2
completely solves the ND.
Proof. We insert the BSG function into ND
1
1  2
64
Valentin Joseph Boussinesq (1842-1929), Carlo Somigliana (1860-1955), Boris
Grigorievich Galerkin (1871-1945)
4.1 Elasticity 221
1
= div grad div grad g – div grad grad div g
2  2
1 1
+ (grad div div grad g – grad div grad div g)
1  2 2  2
with (2.2.50), (2.2.51)
1
= div grad div grad g – grad div grad div g
2  2
1 1
+ (grad div grad div g – grad div grad div g)
1  2 2  2
1 1 1
We can easily verify that – + (1 – ) = 0 , thus
2  2 1  2 2  2

= g = – b

and ND. We now show the completeness of BSG. If we substitute in BSG
 ' : = 3/2 –  and  b' : = –  u
1 1
we obtain with – =
2  2 1  2 
1 
u = div grad g + grad div g = – b' .
1  2  
If g solves the ND for an artificial material under the body force b' , then u in
BSG solves ND with respect to b ; q.e.d.
Theorem. Displacement function of BOUSSINESQ (1885) 
PAPKOVITSCH65 (1932)  NEUBER (1934)
Let

  a scalar field with div grad  = xb


 w a vector field with div grad w = – b

(inhomogeneous LAPLACEan PDE)
with the position vector x and the body force field b . Then the displacement
function
1
(4.1.52) u := w – grad(x  w +  ) (BPN)
4  4
completely solves ND.
65
Pyotr Feodorovich Papkowitch (1887-1946), Heinz Neuber (1906-1989)
Proof. If we insert BPN into ND, we obtain

1
1  2
1
= div grad w + grad div w
1  2
1
– [div grad grad(x  w + )
4  4
1
+ grad div grad (x  w + )]
1  2
1
with (2.2.50) = div grad w + grad div w
1  2
1
– grad div grad (x  w + ) .
2  1  2 
With the reformulation

 
div grad (x  w) = (x  w)    = (xi wi)
x x k
k
 2 xi  2 wi xi wi
= k k
wi + k k
xi + 2 k
x x x x x x k
wi
= x  (w    ) + 2
x i

= x  div grad w + 2 div w = – x  b + 2 div w

= – div grad  + 2 div w
we see that the above term is

div grad w = – b

and thus fulfils ND. To prove the completeness of the ansatz, we transform the
BSG, the completeness of which has already been demonstrated, into BPN. We set
 = 2 div g – x   g
w = g.
Then with

  g =  w = div grad w = – b

and with BSG, we obtain
4.1 Elasticity 223
1
u = g – grad div g
2  2
1
= w – ½ grad( + x   g)
2  2
1
= w – grad( + x  w)
4  4
and thus BPN; q. e. d.
BPN is perhaps easier to handle than BSG, and therefore more popular.
The following ansatz functions are also complete solutions of the equilibrium
problem in the case of vanishing body forces (b  o).
Theorem. Displacement function of FREIBERGER (1949)
Let
w = o
and
  = div w .
Then
1
(4.1.53) u = w – grad 
2  2
is a complete solution of ND with b  o .
Theorem. Displacement function of KORN (1915)

Let
w = o
and
 c = curl w .
Then
1
(4.1.54) u = w – curl c
1  2
is a complete solution of ND with b  o .
Further concretisations of the displacement functions can be obtained under
restrictions like plane states of stress (LOVE/ NOLL, GOURSAT) or rotational
symmetry (BOUSSINESQ).
Problem 16. Force on an Elastic Half-Space

Compute the displacement field which is caused by a singular force after
the BOUSSINESQ-SOMIGLIANA-GALERKIN ansatz (4.1.51) on an
elastic half space. Check the compatibility of the solution.
D
Z
Solution
With b = o , the remainders of the BSG solution (4.1.50) are
g = o.
Since the problem shows a rotational symmetry with respect to the force,
we use cylindrical coordinates. The LAPLACE operator is
 = 
 1    1  
= ( eR + e + eZ )  ( eR + e + eZ )
R R Θ Ζ R R Θ Ζ
2 1  1 2 2
= + + +
2R R R R 2  2 Θ  2 Ζ
where the mixed scalar products vanish, and one term appears due to
eR ,  = e . With the ansatz g = g(R, Z) eZ the problem can be reduced to
the bi-potential equation
(P16.1)  g = 0 .
Since we cannot give a general solution of this equation, we use the ansatz
of BOUSSINESQ
(P16.2) g = C1 D + C2 Z ln(Z + D) , D = R2  Z 2 .
We insert this into (4.1.51)
1
u = div grad (g) – grad div(g)
2  2
1
= (g eZ  )   – (g eZ  ) 
2  2
1
=  g eZ – g, Z 
2  2
= (g , RR + g , ZZ + g , R /R + g ,  /R2) eZ
4.1 Elasticity 225
1
– (g , ZR eR + g , Z /R e + g , ZZ eZ) .
2  2
Because of the rotational symmetry, all derivatives with respect to 
vanish. This gives
1 1  2
(P16.3) u = – g , ZR eR + (g , RR + g , R /R + g , ZZ) eZ .
2  2 2  2
Next we provide the derivatives of g . For differentiation of the function D
P16.2 we apply the chain rule
D R D Z
  .
R D Z D
We obtain
g , R = R[C1 + C2 Z /(Z +D)] /D
g , Z = Z(C1 + C2) / D + C2 ln(Z +D)
g , RR = (1– R2/D2) [C1 + C2 Z /(Z +D)] / D
– R2C2 Z /[(Z +D)2D2]
g , ZZ = (1– Z2/D2) (C1 + C2) / D + C2 / D
g , ZR = C2 R / (Z D +D2) – Z R(C1 + C2) / D3.
We insert this into P16.3 and obtain the displacement field u . By use of
the DSR in cylindrical coordinates from Problem 13 we obtain for the non-
zero strain components
RR = uR , R  = uR / R ZZ = uZ , Z
RZ = (uZ , R + uR , Z) /2 .
We can insert these into HOOKE´s law (4.1.17) in order to determine the
corresponding stresses. For the calculation a representation with the elastic
constants G and v is most practical. With these stresses and
displacements we can determine the constants C1 and C2 through the
boundary conditions. Obviously, there are no displacement boundary
conditions. The traction boundary conditions are
(P16.4) RZ = 0 for Z  0
(P16.5)  ZZ dA =  f for Z  0

A
with the surface A and its normal eZ .

P16.4 ensures a stress-free surface because of 0 = (T  eZ)  eR and
0 = (T  eZ)  e . Because of Z = 0 , only P16.4 remains to fulfil. P16.5
results from the cut principle at some Z and evaluating the equilibrium
condition
f +  T  eZ dA = o .
A
We begin with P16.4

0 = 2 G RZ
= G (uZ , R + uR , Z )
1  2 G
= G (g , RR + g , R / R + g , ZZ) , R – g , ZRZ .
2  2 2  2
In the first bracket we can already set Z  0 , while the last term must be
previously differentiated with respect to Z . For Z  0 we have g , RR = 0 ,
g , ZZ = (C1 +2 C2) / R , and g , R = C1 . The derivative is then
g , ZRZ = (C1 + C2) (3 R Z 2 / D2 – R) / D3
– C2 (R + R Z / D) / (Z D2 + D3) .
Taking into account that D = R at Z  0 , we can write the equilibrium
condition as
2
0 = (C1 + 2 C2) / R2 – C1 / R2 .
2  2
This leads to
1  2
C2 = C1 .
2
For the exploitation of P16.5

ZZ = 2G [uZ , Z + ( uR , R +uR / R + uZ , Z)]
1  2
1  
= 2G [ uZ , Z + (uR , R + uR / R)]
1  2 1  2
is needed. The evaluation is lengthy, but straightforward since uR and uZ
are given by P16.3. Taking the derivatives leads to
3G Z3
ZZ = C1 5 .
2 (1   ) D
One can easily see that the normal stresses ZZ vanish for R > 0 at
Z  0 , and tend to infinity at R  0 where the force is applied. We insert
ZZ in the condition of equilibrium P16.5
2 
3GZ 3C1
– f =  ZZ dA =
A 2 (1   )   R / D5 dR d .
0 0
4.1 Elasticity 227
Since D = (R2 + Z2)1/2 , we have dD/dR = R/D and dR = D/R dD, so

that we can write for the integral

3GZ 3C1
–f =
 (1   )  1/ D4 dD .
0
Finally we get
GZ 3C1 3 R  GZ 3C1 3 GC1
– f =  (1   ) [ D ] R  0 =  (1   ) (0  Z ) = –  (1   )
so that C1 = f v (1–v) / (G ) is determined. The constants C1 and C2 are

now known and the problem is solved.
f  RZ R( 1  2 ) 
(P16.6) uR = 4 G  3  D( D  Z ) 
D 
f Z 2 2  2 
(P16.7) uZ = 4 G  3  D 

D 
At X  o the displacement uZ goes to infinity due to the singularity of the

force.
In the following graphic, the highest layer (Z < 0,25) has been suppressed
since the solution is singular at the force application point. We want to
estimate the size of the zone inside which the solution becomes unrealistic.
There are several ways to do this, for example by checking the
compatibility conditions (Chapter 3.1.1). This is somewhat laborious, since
we used cylindrical coordinates.
Close to the application point of the force, the volume is inverted, which
involves unrealistic large volumetric strains. More specifically, the
deformed volume is considerably smaller than the undeformed volume. A
measure for the volume change is the trace of E after (3.1.26). We can use
this property to estimate the size of the zone of unrealistic deformations by
introducing a lower limit for tr E , beyond which we consider the
volumetric strains as too large
tr E = uR , R + uR /R + uZ , Z = crit .
This equation can be solved for R(Z)

2/3
 f (2ν  1)Z 
2
R  Z    .
 2Gπ crit 
By setting crit = – 0.1, we consider a volume decrease of approximately

10% as critical.
We can now show the zone in which the volume change exceeds this
critical value. Here we used the following values G = 81 GPa, f = 1000 N,
v = 0.3, and crit = – 0.1. For this example, the zone of unrealistic
displacements is only a fraction of a millimeter thick. Even for
crit = – 0.01 the depth of the critical zone is only 0.28 mm.
Z
4.1 Elasticity 229
Problem 17. HERTZean Contact

We want to use the solution of BOUSSINESQ to determine the pressure
between two spheres (given radius R1,2 , elastic constants v1,2 and G1,2). The
centres of the spheres shall be moved towards one another by the amount
of w0 . We expect a contact zone of radius a . Determine a , the resulting
force, and the displacement field in the contact zone.
Solution
We use cylindrical coordinates because of the rotational symmetry of the
problem. The contact condition is the following. The mutual approach of
the centres w0 equals the initial distance z1(R) and z2(R) plus the two
indents w1(R) and w2(R) . In the following, z1,2 and w1,2 are absolute
values
z1 + z2 + w1 + w2 = w .
For the contact condition z1,2 (R) and w1,2 (R) are needed.
Z1,2
R
From the geometry of a rectangular triangle we know that

R2 = Z1,2 (2R1,2 – Z1,2) = 2R1,2 Z1,2 – Z1,22.
For Z1,2 being rather small compared to the radii, we can neglect the
quadratic term Z1,22 . This leads to the approximation
Z1,2 = R2 / (2R1,2) .
We use the solution of BOUSSINESQ P16.7 to determine the indentations

w1 and w2 . We consider p(R) dA = dF(R) as a small incremental force
acting on the elastic half space with the pressure p(R). By the principle of
superposition, we obtain the displacements as the integral of all
displacement fields of the BOUSSINESQ solution over the contact area,
with the indenting force dF(R) depending on the location inside the
contact area. The displacement in the Z-direction is
F  Z 2 2  2 
w(R,Z) = 4 G  D 3  D 
 D = R2  Z 2 .
 
We are interested in the solution at Z  0 . For the force F we take

p(R) dA . This gives at some R the displacement
p(R)dA  2  2 
dw(R) = 4 G  R  .
The resulting total displacement can be achieved by integration over all

differential forces in the contact zone. For the displacement in the centre
we obtain with the surface element dA = R dR d Θ
1   2 a
 
w = 2 G p ( R ) dR dΘ
0 0
where R cancels out once. If we want to determine the deformation in the

entire contact zone, we shall first transform the (central) coordinate system
to some other point in the contact zone with radius R , in which we
introduce an angle  and a radius  (in the figure you see a view upon the
contact zone).
Now we can write for the integral at R

4.1 Elasticity 231
1    2
w(R) = 2  G   p() d d
0  1
containing the dependence on R through the functions 1,2 (R , ) . We

substitute the integrand by the mean pressure p multiplied by the length of
the section  (R , ) = 1(R ,  ) + 2(R ,  )
1  
w(R) = 2  G  p  (R , ) d .
0
By the Theorem of PYTHAGORAS we can write  as a function of 
a2 =  2 + ( /2)2   =2 a 2   2  .
If we use the relation  = R sin  we obtain

1   2 2 2
w(R) = 2  G  p a  R sin (  )d .
0
We need an ansatz for p . It is clear that the mean pressure p can only
depend on the perpendicular  to the centre. p must vanish at the edge (at
  a) and obtain its maximum in the centre (at   0). We use a spherical
distribution, parameterised by a scalar  , p   a 2   2 . The integral can
now be solved
1    a 2 - R 2 sin 2 ( )  d 
w(R) =  G    
0
1 
= G  ( a2 – R2/2) .
The displacements w1,2 and the approximations of the surfaces of the

spheres Z1,2 can be implemented into the contact condition. Following the
reaction principle, the pressure distributions on both spheres must have
opposite signs and equal value. Thus, there exists only one parameter 
and
 1 1   1  1 1  2 
R2  2R + 2R  +  ( a2 – R2/2)  G + G  – w0 = 0 .
 1 2  1 2 
This must hold in the entire contact zone, independent of R . Comparison

in terms of R2 leads to the two equations
 1  1 1  2 
w0 =  a2  + 
 G1 G2 
 1 1   1  1 1  2 
 + R  =   G + G 
R
 1 2  1 2 
which give for  and a

R1  R2
 = R R ((1   )/G  (1   )/G )
1 2 1 1 2 2
a = w0 R1 R2 /( R1  R2 ) .
It is interesting to note that for a given w0 the size of the contact zone does
not depend on the elastic constants.  depends both on the geometry and
on the material. The reaction force is
a a
F =  p ( )  d  = 2  ( a 2   2 ) d 
a a
1
8 w03 R1 R2  1   1 1  2 
= 8 a3  /3 =  +  .
3 R1  R2  G1 G2 
Although the material law is linear, we find a non-linear dependence

between force and displacement. This is a purely geometrical effect, due to
the change of the contact area. We depict the contact zones and the
approximation of the spherical surface for two spheres with R1 = 1 mm,
R2 = 2 mm, v1 = v2 = 0.3, E1 = 200 GPa, E2 = 70 GPa (which corresponds
roughly to steel and aluminium), which have been pressed by 10 kN, 20 kN
and 30 kN against each other
4.1 Elasticity 233
8 kN
experiment
calculation
4 kN
0 kN
0,1 0,2 0,3 0,4
The force/displacement curve for a steel/steel indentation has been
experimentally measured, where both radii where 5 mm. The unknown
stiffness of the machine has been accounted for by a proportional scaling to
a best fit with the calculated result. Qualitatively, the coincidence between
the measurement (solid line) and calculation (dashed line) is rather good, in
spite of all the assumptions and simplifications we have made. The
differences for larger displacements (for w0 > 0.4 mm) is probably due to
plastic deformations, which can be finally observed on the spheres.
Problem 18. Comparison of Variational Principles
We consider a vertically suspended bar of length l with cross-section area

A , mass density  , and YOUNG´s modulus E . The upper end is fixed,
while the lower one is displaced by an amount of u0 .
FB
B
x Fg
l, A, , E N
u0
Determine the displacements u(x) and the normal force N(x) in the bar.
Use (1) the ODE of the one-dimensional boundary value problem, (2) the
the Principle of Minimum Strain Energy, (3) the Principle of Minimum

Stress Energy, and (4) the Principle of DE VEUBEKE – HU – WASHIZU.
Solution
In the case of a uniaxial model of the bar, the displacement strain relations
DSR and HOOKE´s law are simply
(P18.1)  = u,x  = E .
The normal force N results from the equilibrium condition in vertical
direction
N = FB – Fg = FB – g  A x
where FB is the force at the upper end, and Fg is the weight. It is assumed
that the stresses are homogenously distributed in the cross-sections. We
obtain with or without taking into account the transversal strains
 = N/A or  = N / [A(1 – u, x )2] ,
respectively. With P18.1 we get for the two cases
(P18.2) u , x = (FB – g  A x) / (EA)
or u , x = (FB – g  A x) / [EA(1 – u , x )2] .
In the first case it is a simple linear ODE, which can be easily integrated. In
the second case the ODE is non-linear. We continue with the geometrically
linear theory using P18.2-1, hence neglecting the influence of the
transversal strains. The integration over x gives us
u = (FB x – ½ g  A x2) / (EA) + c
with a constant c . The two unknowns FB and c can be determined by
the two boundary conditions u(0) = 0 and u(l) = u0
c = 0
FB = u0 E A / l +  l g A / 2
so that the final solution is
(P18.3) u (x) = u0 x / l + ½  g (l x – x2) / E .
We will first use the Principle of Minimum Strain Energy. It states that
the functional from (4.1.43)
(u) =  w dV –  b  u dm –  tpr  u dA
V V Ad
with w = ½ E  C  E obtains a minimum for the solution. We now

specialise this for the one-dimensional bar. The last term vanishes due to
zero-traction boundary conditions
4.1 Elasticity 235
 E  dV –  g u dm
2
( u ) = ½
V V
l l
(P18.4) = ½  E A u , x2 dx –  g u  A dx .
0 0
For a prismatic bar we can thus reduce the volume integrals to line
integrals. The minimum principle requires a parameterised ansatz for u (x)
that already fulfils the boundary conditions. Moreover, it must be linear in
the (unknown) parameters. This leads to a system of linear equations in the
parameters. We assume
u (x) = a x2 + b x + c
with three parameters a, b, and c. After applying the boundary conditions
u (0) = 0 and u (l) = u0 only one parameter (a) remains unknown. Thus
u (x) = a x2 + (u0 / l – a l ) x
u (x) , x = 2a x + u0 / l – a l .
We insert this into the functional P18.4
l
( u ) = ½  E A (2a x + u0 / l – a l ) 2 dx
0
l
–  g  A (a x2 + (u0 / l – a l ) x) dx
0
= ½ E A [4 a2 l 3/3+2 a l 2 (u0 / l – a l ) + l (u0 / l – a l ) 2]

– g  A [a l 3 /3 + l 2 (u0 / l – a l ) / 2] .
A necessary condition for  to obtain a stationary value (minimum or
maximum), is that its derivative with respect to a is zero
( u ) , a = 2 E A a l3/6 + g  A l3/6 = 0
so that
a = – ½ g  /E .
We insert this into our ansatz for u (x) and obtain the solution
u (x) = u0 x / l + ½ ( l g x – g  x2 ) / E .
This equals the exact solution which we already determined by the ODE
P18.3. This is due to the fact that the exact solution is contained in our
ansatz. If this is the case, the variational problem generally yields the exact
solution.
The starting point for the applications of the Principle of Minimum Stress
Energy is the functional (4.1.44)
(T) =  w*(T) dV –  upr  t dA

V Ag
with w*(T) = ½ T  S  T .
For the one-dimensional case this is
1 l
(P18.5) (  ) =   A dx – u0 A  (l) .
2
2E 0
Thus, the volume integral has been transformed into a line integral, and the
surface integral can be reduced to the evaluation at one point where the
displacement boundary conditions is given. For the minimum strain energy
principle, we have to make an ansatz for the stresses  that must satisfy
the equilibrium conditions. Thus, we are restricted to approaches that fulfil
the balance of linear momentum, see Chapter 3.2.2. In the one-dimensional
case, the equilibrium condition is
 ,x = – g 
which gives by integration
 = – g x + d
with a constant d . By substituting this into P18.5 we obtain
(  )
A l
 [(g  x) – 2 d g  x + d ] dx – u0 A d + A u0 g  l
2 2
=
2E 0
A
= [(g)2 l 3/3 – d g  l 2 + l d 2] – A u0 d + A u0 g  l .
2E
At the minimum of  its derivative with respect to d must vanish
A
0 = (– g  l 2 + 2 l d ) – A u 0
2E
which renders for d and  finally

d = E u0 / l + g  l / 2
 = E u0 / l + g  l / 2 – g  x .
We will now compare this result with the preceding one. Obviously, it is
easier to compute  by u than vice versa. In this direction we only have
to differentiate, and not to integrate and to determine an integration
4.1 Elasticity 237
constant. This is one reason why people prefer the displacements as the
independent variables. The differentiation of the solution P18.3 with
respect to x gives us the strains  = u , x and a multiplication by E
(HOOKE´s law  = E  ) the stresses
 = E u0 / l + g  l / 2 – g  x .
So  is the exact solution. This is of no surprise, since the restrictions for
the ansatz did not leave any other choice. In the one-dimensional case the
equilibrium condition is rather simple. In the general three-dimensional
case things become much more complicated.
For the Principle of DE VEUBEKE – HU – WASHIZU, we must evaluate
the functional
(u , E , T) = ½  E  C  E dV –  T  E dV
V V
–  (div T +  b)  u dV +  t  upr dA +  (t – tpr)  u dA

V Ag Ad
(4.1.46). Therefore we need ansatz functions for the displacement field, the
strain field, and the stress field. At first glance this may appear rather
awkward since E and T are related by HOOKE´s law, and with u by
the DSR. The advantage of this principle, however, lies in the fact that
these functions are not constrained. We obtain solutions for all these fields
simultaneously. The application of this principle can be generally
performed by some standard algorithm.
In the one-dimensional case the functional becomes
l l
( u ,  ,  ) = ½ A   2 E dx – A    dx
0 0
l
– A  (  , x +  g) u dx + A  (l) u0 .
0
The integral over the boundary with zero-stresses vanishes again. We must
find ansatz functions for the fields u ,  , and  . The functional 
obtains a stationary value if its derivatives with respect to these
independent functions vanishes. At the beginning we consider also  , x as
an independent ansatz function. The variation of ( u ,  ,  ) is its
directional differential with respect to the independent fields. It must be
zero at the solution
l
(P18.6)  [(  E –  )  – (  , x +  g)  u –    – u   , x] dx
0
+   (l) u0 = 0 .
The factor A has already been suppressed. For the functions  ,  , and u
we choose polynomial functions
3 3
u =  ui pi  =  i pi
i0 i0
3 3
 =  i p i  ,x =  i pi, x
i0 i0
with p : = x / l.
The variations of   ,  u ,   , and   , x are then
3 3
u =  ui pi  =  i pi
i0 i0
3 3
 =  i pi  , x =  i pi, x .
i0 i0
In P18.6 we can separate the twelve parameters u0,..,3 , 0,..,3 , and 0,..,3 .
Under the integral we obtain sums of twelve terms, each of which is a
product of the variation and a factor. We integrate each term separately and
pull each parameter u0,..,3 , 0,..,3 and 0,..,3 out of the integral. For
getting  = 0 all integrals must vanish, since the variations u0,..,3 ,
0,..,3 , and 0,..,3 are independent of each other. We demonstrate this
procedure in detail for u3 .
l
0 = – [(  , x +  g) u3 p3] dx
0
l
= – u3  [(0 p0 , x + 1 p1 , x + 2 p2 , x + 3 p3 , x +  g) p3] dx
0
l
= – u3  [(1 / l + 2 2 x / l 2 + 3 3 x2 / l 3 +  g) x3 / l 3] dx
0
l
= – u3  [1 x3 / l 4 +2 2 x4 / l 5 + 3 3 x5 / l 6 +  g x3 / l 3] dx
0
4.1 Elasticity 239
l
= – u3 [1 x4 / (4 l 4) +2 2 x5 / (5 l 5) + 33 x6 / (6 l 6) + g x4 / (4 l 3)] 0
= – u3 (1 / 4 +22 / 5 +33 / 6 + g l /4) .

Thus
0 = 1 / 4 + 22 / 5 + 33 / 6 +  g l / 4
which gives the first equation for the coefficients u0,..,3 , 0,..,3 and 0,..,3 .
The evaluation of all variations renders the following system of linear
equations presented in a matrix form
0 0 0 0 0 0 0 0 0 1 1 1   u 0   ρgl 
0 0 u  
 0 0 0 0 0 0 0 1/2 2/3 3/4   1  ρgl/2

0 0 0 0 0 0 0 0 0 1/3 2/4 3/5   u 2  ρgl/3
     
0 0 0 0 0 0 0 0 0 1/4 2/5 3/6   u 3  ρgl/4
0 0 0 0  El  El/2  El/3  El/4 l l/ 2 l/ 3 l/ 4   ε0   0 
    = 
0 0 0 0  El/2  El/3  El/4  El/5 l/ 2 l/ 3 l/ 4 l/ 5   ε1   0 
0 0 0 0  El/3  El/4  El/5  El/6 l/ 3 l/ 4 l/ 5 l/ 6  ε   0 
   2  
0 0 0 0  El/4  El/5  El/6  El/7 l/ 4 l/ 5 l/6 l/7   ε3   0 
     
 0 0 0 0 l l/2 l/3 l/4 0 0 0 0   σ 0   u0 
1 1/2 1/3 1/4 l/2 l/3 l/4 l/5 0 0 0 0   σ 1   u0 
     
1 2/3 2/4 2/5 l/3 l/4 l/5 l/6 0 0 0 0   σ 2   u0 
1 3/4 3/5 3/6 l/4 l/5 l/6 l/7 0 0 0 0   σ   u 
  3  0 
After a proper numbering, the matrix of coefficients can be made

symmetric. Its rank is 11, so that we cannot expect a unique solution. In the
present example, the four parameters of the displacement field u appear in
only three equations. This is a result of the fact that the approach for the
displacement field in P18.6 appears only through  , x in the functional .
Because of the derivative with respect to x , the constant u0 vanishes.
Thus, we need to choose appropriate ansatz functions for u and for  . In
our example, we need to either remove one parameter in u , or add one
parameter in the ansatz for  . Thus, the ansatz for u can not contain
more parameters than the ansatz for  . In the opposite direction, this
restriction does not appear. In conclusion, when using the Principle of DE
VEUBEKE – HU – WASHIZU, the ansatz functions need not fulfil any
boundary conditions a priori, but it may happen that they have to be chosen
in accordance to each other.
If we complete the system of equations by the boundary condition
u (0) = 0 , we eliminate the parameter u0 , so that we can find the solution
u0 = 0 u1 =  g l 2 / (2E) + u0
u2 = –  g l 2/ (2E) u3 = 0
0 =  g l / (2E) + u0 / l 1 = –  g l / E
2 = 0 3 = 0
0 =  g l / 2 + E u 0 / l 1 = –  g l
2 = 0 3 = 0 .
We thus obtain the exact solution. As we have seen in this example, the
application of the Principle of DE VEUBEKE – HU – WASHIZU causes
more work than the other principles. The reason is that we had to choose
three ansatz functions at the same time, and we choose them with higher
orders as for the other principles. However, for an algorithmic treatment
the absence of constraints for the ansatz functions is quite practical.
Problem 19. Principle of Minimum Strain Energy
We want to demonstrate the algorithmic construction of approximative

solutions by means of variational principles. A strip of size l  l /2 (0 < x <
l, 0 < y < l / 2) shall be strained in xdirection.
u(x  0, y) = o u(x  l, y) = d ex
ey
ex
The end planes shall be rigidly displaced relative to each other by an

amount of d . In zdirection we assume a plane state of strain, so that we
Strain gradient & nonclassical approacheshave uz = 0 everywhere. The
strip is notched in the middle at x  l / 2 . The Principle of Minimum
Elastic Energy shall be used.
Solution
We make an ansatz for the displacements ux and uy as a double series
already taking into account the boundary conditions u = o at x  0 and
u = d ex at x  l . This can be easily implemented by multiplying the series
by appropriate factors
4.1 Elasticity 241
n n
uy = x (l – x)   Cy ij xi y j
i 0 j0
n n
ux = x d / l + x (l – x)   Cx ij xi y j.
i 0 j0
We have to determine 2(n +1)2 parameters Cx ij and Cy ij , in which we

proceed similarly to what we did in Problem P18. The quality of the
solution rises with n . This can be executed by a computer-algebra system.
Here we used MATHEMATICA.
Remove["Global`*"]
n = 9;(* maximal polynomial degree *)
l = 1;(* length of the strip *)
G = 40;(* shear modulus *)
K = 120;(* bulk modul *)
disp = l/5;(* displacement *)
(* shape of the notch, centered at x=l/2, width = l/5 *)
ynotch = 20*(-(l/2 - x)^2 + (l/10)^2);
Print["Building the integrand w for polynomial approach of
degree ", n, " with ", 2*(n + 1)^2, " coefficients ..."]
(* Construct arrays that hold the coefficients *)
CX = Array[cxval, {n + 1, n + 1}];
CY = Array[cyval, {n + 1, n + 1}];
eqb1 = Array[e1, {n + 1, n + 1}];
eqb2 = Array[e2, {n + 1, n + 1}];
(* Ansatz for ux and uy *)
ux = x/l*disp + x*(x - 1)*Sum[CX[[i + 1, j + 1]]*xî*y^j,
{i, 0, n}, {j, 0, n}];
uy = x*(x - l)*Sum[CY[[i + 1, j + 1]]*xî*y^j,
{i, 0, n}, {j, 0, n}];
(* Calculate strain from displacement *)
exx = Simplify[D[ux, x]];
eyy = Simplify[D[uy, y]];
exy = Simplify[(D[ux, y] + D[uy, x])/2];
spe = exx + eyy;
(* Calculate stresses from strains *)
txx = Simplify[3*K*spe/3 + 2*G*(exx - spe/3)];
tyy = Simplify[3*K*spe/3 + 2*G*(eyy - spe/3)];
txy = Simplify[2*G*exy];
integrand = Expand[exx*txx + eyy*tyy + 2*exy*txy];
(* Elastic energy = T .. E *)
Print["Integrate over region 1..."];
w1 = Integrate[Expand[Integrate[integrand, {y, 0, l/2}]],
{x, 0, 2/5*l}];
w2 = Integrate[Expand[Integrate[integrand, {y, ynotch, l/2}]],
{x, 2/5*l, 3/5*l}];
w3 = Integrate[Expand[Integrate[integrand, {y, 0, l/2}]],
{x, 3/5*l, l}];
w = w1 + w2 + w3;
Print["Building and solving linear system..."]
For[i = 1, i < n + 2, i++, For[j = 1, j < n + 2, j++,
eqb1[[i, j]] = Simplify[D[w, CX[[i, j]]]] == 0;
eqb2[[i, j]] = Simplify[D[w, CY[[i, j]]]] == 0;
]];
(* Summarize eqs and vars in one list *)

alleqs = Flatten[Union[eqb1, eqb2]];
allvars = Flatten[Union[CX, CY]];
erg = Solve[alleqs, allvars];(* Calculate coefficients *)
Print["Assigning results..."]
For[i = 1, i < n + 2, i++, For[j = 1, j < n + 2, j++,
cxval[i, j] = CX[[i, j]] /. erg[[1]];
cyval[i, j] = CY[[i, j]] /. erg[[1]];
]];
Print["Plot deformed strip..."]
xe = Simplify[x + ux];
ye = Simplify[y + uy];
part1 = ParametricPlot[{xe, ye}, {y, 0, l/2}, {x, 0, 2/5*l}];
part2 = ParametricPlot[{xe, ye}, {x, 2/5*l, 3/5*l},
{y, ynotch, l/2}];
part3 = ParametricPlot[{xe, ye}, {y, 0, l/2}, {x, 3/5*l, l}];
Show[part1, part2, part3, PlotRange -> {{0, l + disp},
{0, l/2}}]
x = 0.5; y = ynotch; txxmax = N[txx]; Clear[x]; Clear[y];
Print["Txx at notch center: ", txxmax]
txxplot = DensityPlot[txx, {x, 0, l}, {y, 0, l/2},
AspectRatio -> Automatic, ColorFunctionScaling -> False,
ColorFunction -> (ColorData["Rainbow"][Rescale[#, {0, txxmax}]]
&),PlotRange -> {0, txxmax}];
tyyplot = DensityPlot[tyy, {x, 0, l}, {y, 0, l/2}, AspectRatio
-> Automatic, ColorFunction -> "Rainbow"];
txyplot = DensityPlot[txy, {x, 0, l}, {y, 0, l/2}, AspectRatio
-> Automatic, ColorFunction -> "Rainbow"];
hidenotch = RegionPlot[y < ynotch, {x, 0, l}, {y, 0, l/2},
AspectRatio -> Automatic];
Show[txxplot, hidenotch]
Show[tyyplot, hidenotch]
Show[txyplot, hidenotch]
The script produces the following output:

Building the integrand w for polynomial approach of degree 9
with 200 coefficients ...
Integrate over region 1...
Building and solving linear system...
Assigning results...
Plot deformed strip...
Txx at notch center: 41.7511

4.1 Elasticity 243
From upper left to lower right: deformed shape, stresses xx , yy , and xy .
Stress xx at the notch centre in MPa (left) and elastic energy w (right)
versus degree of polynomial for the notched (dark gray) and unnotched
specimen (light gray).
One can see that for the unnotched specimen already for a degree of 3 no
more essential change can be expected, while for the notched specimen the
convergence can not yet be seen. This demonstrates the advantage of FEM
compared to such global ansatz functions. FEM allows for a local
discretisation appropriate for the problem, while global ansatz functions do
not offer this opportunity.
4.1.9 Wave Propagation in Elastic Media
Literature
Achenbach, J. D.: Wave Propagation in Elastic Solids. Elsevier, North Holland
(1984)
Chen, P. J.: Growth and Decay of Waves in Solids. In: Encyclopedia of Physics.
Edt. S. Flügge. Vol. VIa/3, Springer (1973)
Chen, P. J.: Selected Topics in Wave Propagation. Noordhoff Int. Pub. (1976)
Dieulesaint, E., Royer, D.: Elastic Waves in Solids. John Wiley & Sons,
Chichester, New York, Brisbane, Toronto (1980)
Fedorov, F. I.: Theory of Elastic Waves in Crystals. Plenum Press, New York
(1968)
Graff, K. F.: Wave Motion in Elastic Solids. Oxford University Press (1975)
S. Flügge. Vol. VIa/2, Springer (1972) Sect. 70, p. 243 ff.
Harker, A. H.: Elastic Waves in Solids. Adam Hilger, Bristol and Philadelphia
(1988)
Press (2004, 2011)
Pollard, H. F.: Sound Waves in Solids. Pion, London (1977)
Stephan, W.; Postl, R.: Schwingungen elastischer Kontinua. B.G.Teubner,
Stuttgart (1995)
Wauer, J.: Kontinuumsschwingungen. Springer (2008)
As an example for elastodynamics, we consider the propagation of plane

periodic waves in a homogeneous elastic material. We try to solve C1 with
HOOKE´s law neglecting body forces (b  o)
div(C  def u) =  u
by the following special ansatz for the displacements
(4.1.58) u(x , t) =  ( (x , t)) u
with
 ( ) the wave function, a scalar function of
 (x , t) : =  t – n  x the phase, which depends on place and time
u the (constant) amplitude vector
n the (constant) wave vector
n : =n the wave number
 the (constant) angular frequency
c := /n the (phase) speed
4.1 Elasticity 245
The wave function determines the form of the wave like, e.g., a harmonic one. The
phase determines the motion of the wave through the body. It depends on the
locus through n  x , which stands for the distance to a plane through the origin.
The amplitude vector plays a double role. Its direction stands for the directions of
the excitation of the material points, its value for the amplitude of the wave.
Similarly, the wave vector stands for both the direction of propagation of the wave
and for the wave length. We consider here only plane wave fronts with
displacements in the direction of u , which propagate by a speed c in the
direction of n .
In particular, a harmonic wave is described by the wave function
t x e
(4.1.59)  ( ) = sin ( t – n  x) = sin [2 (  )]
T l
with
e :=n/n a unit vector in direction of n
T = 2 / the period
l = 2 /n the wave length
f = 1/T =  / 2 the frequency
By the above ansatz we obtain the spatial derivative by the chain rule
grad  ( (x , t)) =  ( )' grad ( t – n  x) = –  ( )' n
and
grad  ( )' = –  ( )'' n
and the derivative with respect to time
 ( (x , t)) =  ( )' ( t – n  x) =   ( )'
   =  2  ( )''
and for the acceleration
u(x , t) =  ( ) ''  2 u .
The displacement gradient is with (2.2.35)
H(x , t) = grad u = u  grad  = –  ( )' u  n
a simple dyad, and with (2.1.92)
div u = tr H = –  ( )' u  n .
The curl is the axial vector of the antimetric part of the gradient after (2.1.53) and
(2.2.14)
curl u = –  ( )' n  u =  ( )' u  n .
Taking into account the usual subsymmetries of the stiffness tetrad, the left side of
C1 becomes
div (C  def u) = div (C  grad u) = div (–  ( )' C  u  n)

with (2.2.41) = – (C  u  n)  grad( ( )' ) =  '' n  C  u  n
=  '' n2 (e  C  e)  u =  '' n2 A  u
with the acoustic tensor A defined by
(4.1.60) A(e) : = e  C  e .
This tensor inherits the following properties from the stiffness tetrad:
 C is symmetric  A is symmetric for all e
 C is strongly elliptic  A is positive definite for all e
 Q is a symmetry transf. of C  Q  A(e)  QT = A (Q  e)
A tetrad C is called strongly elliptic if the following holds for all non-zero
vectors a and b
a  b  C  a  b > 0 .
This means that all components of the stiffness tetrad of type Cijij (which
correspond to the diagonal elements of the VOIGT matrix) must be
positive. This condition is weaker than the positive definiteness for which the
inequality is postulated for all tensors, while here only for all simple dyads. For
these being a proper subset of all tensors, the following implication holds
C is positive definite  C is strongly elliptic
but not vice versa.
We obtain from C1 (3.2.7)
 '' n2 A  u =   ''  2 u
and thus the FRESNEL66HADAMARD67 condition of wave propagation
(4.1.61) A  u =  c2 u
after which our displacement ansatz fulfils the equations of motion for all wave
functions if and only if u is an eigenvector of the acoustic tensor A
corresponding to the eigenvalue  c2 . For A being symmetric in the hyperelastic
case, we can find three orthogonal eigendirections with real eigenvalues. The
strong ellipticity of the elastic tensor is a necessary and sufficient condition that
the eigenvalues are positive and thus the speed of propagation of waves is real.
The following are important specific cases:
66
Augustin Jean Fresnel (1788-1827)
67
Jacques Salomon Hadamard (1865-1963)
4.1 Elasticity 247
1) longitudinal waves (compression waves)
n
u
Here u is parallel to n . Therefore the direction of excitation coincides with that

of the propagation. The longitudinal wave is irrotational since
curl u =  ( )' u  n = o .
2) transversal waves (shear waves)
Here u is perpendicular to n . So the excitation is transversal to the direction of

propagation. Shear waves are isochoric since
div u = –  ( ) ' u  n = 0 .
The following theorem is a direct consequence of the spectral theorem of positive
definite symmetric tensors.
Theorem of FEDOROV68 (1964)
In every anisotropic homogeneous elastic medium, transversal and longitudinal
waves are possible if C is symmetric and strongly elliptic.
Moreover we can see that the symmetry directions of the material are always
directions of the propagation of transversal and longitudinal waves.
Isotropic Wave Propagation

In the isotropic case, the acoustic tensor is
A(e) = e  C  e = e  ( I  I + 2 I S )  e
(4.1.62) = ( + ) e  e +  I
= ( + 2) e  e +  (I – e  e) ,
since for every vector v
68
Fedor Ivanovich Fedorov (1911-1994)
[e  2 I S  e]  v
= e  2 I S  v  e
= e  (v  e + e  v)
= ( e  e + I)  v
holds. The speeds of the compression wave and the shear wave are, respectively,
  2 
(4.1.63) cp : = cs : = .
 
These two eigenvalues are positive if and only if C is strongly elliptic. The two
speeds thus depend only on the material, but not on the form or the amplitude of
the wave.
The quotient of the propagation speeds of these two types of waves is
cs  1  2
  .
cp   2 2  2
If we have for example  = 0.3 , then the compression wave is twice as fast as the
shear wave. For steel, the speed of the shear wave is around 5 km/s .
4.2 Thermomechanics
Literature
Bermudez de Castro, A.: Continuum Thermomechanics. Birkhäuser, Basel
(2005)
Boley, B. A.; Weiner, J. H.: Theory of Thermal Stresses. Wiley, New York
(1960)
Carlson, D. E.: Linear Thermoelasticity. In: Encyclopedia of Physics. Edt. S.
Flügge. Vol. VIa/2, Springer-Verlag (1972)
Dafermos, C. M.: On the existence and asymptotic stability of solutions to the
equations of linear thermoelasticity. Arch. Rat. Mech. Anal. 29, 241-71 (1986).
Ieşan, D.: Thermoelastic Models of Continua. Kluwer Acad. Pub. (2004)
Melan, E.; Parkus, H.: Wärmespannungen. Springer-Verlag, Vienna (1953)
Müller, I.: Grundzüge der Thermodynamik. Springer-Verlag, Berlin (1999)
Müller, I.: A History of Thermodynamics. Springer-Verlag, Berlin (2007)
Nowacki, W.: Thermoelasticity. Addison-Wesley (1962)
Nowacki, W.: Dynamic Problems of Thermoelasticity. Noordhoff Pub., Leyden
(1975)
Parkus, H.: Thermoelasticity. Springer-Verlag, Vienna (1976)
Sneddon, I. N.: The Linear Theory of Thermoelasticity. CISM Course 119,
Springer (1974)
4.2.1 Thermodynamic Balances
In order to include thermo-mechanical interactions and couplings in our theory,

we need some additional (field) variables:
(x , t) the specific internal energy (a scalar field)
r (x , t) the specific heat source (a scalar field)
q(x , t) the heat flux per unit time and area (a vector field)
(x , t) the absolute temperature (a positive scalar field)
(x , t) the specific entropy (a scalar field).
In all cases, specific means per unit mass.
In analogy to the stress principle of CAUCHY, we assume that the heat supply
Q of a body takes place in two forms:
 by heat sources (r > 0) or sinks ( r < 0) in the interior of the body due to phase
changes, chemical processes, inner absorption of radiation, supply of OHM´s
heat, etc.;
 by heat fluxes through the surface, since the body is in direct contact with its
surrounding.
Then one can apply a rationale similar to the tetrahedron argument for the stress
analysis, that the heat flux must be a linear function of the (outer) normal on the
surface. Therefore, a vector field q(x , t) exists which is called the heat flux
vector. If n is the outer normal on the surface, then – q  n stands for the heat
supply through the surface (Theorem of FOURIER).
In total, the current heat supply to the body per unit time is
(4.2.1) Q =  r dm –  q  n dA .
V A
The surface integral can be transformed by the divergence theorem into a volume
integral, so that we have
(4.2.2) Q =  ( r – div q) dV .
V
The thermodynamic balance of energy (1st law of thermodynamics69) tells us
that the supply of heat Q and mechanical power Πa go into a change of the
kinetic energy K or into its internal energy Ei
(4.2.3) Q + Πa = K  + Ei (1LT)
with the (global) internal energy

Ei : =   dm .
V
The internal energy can be stored as heat, which gives rise to a change of the
temperature of the body. It can also be absorbed or set free in chemical bonds,
phase changes, or changes of the microstructure. With the mechanical work
balance (3.2.14)
(WB) Πa = Πi + K 
one can eliminate the kinetic energy in 1LT, so that we obtain
(4.2.4) Q + Πi = Ei
or locally
(4.2.5)  r – div q + T  E =   (1LT)
In analogy to the heat supply, we define the entropy flux (-vector) per unit time
q r
and area as , and the specific entropy source as . After the 2nd law of
θ θ
thermodynamics (2LT) the entropy supply to the body is never larger than its
entropy change
69
Julius Robert Mayer (1814-1878), James Prescott Joule (1818-1889)
 r q
(4.2.6) (   dm)   dm –   n dA
θ θ
V V A
r q
=  (  – div ) dV (2LT)
θ θ
V
or locally
r  q
(4.2.7)    – div .
θ θ
Together with the 1LT and the quotient rule
q div q q  grad θ
div = –
θ θ θ2
and after introducing the specific free (HELMHOLTZ) energy
(4.2.8)  :=  – 
we obtain the local form of the 2LT called the CLAUSIUS-DUHEM70 inequality
1 q g
(4.2.9)  +   – T  E +  0 (CDI)
ρ θ
Herein, g : = grad  is the temperature gradient after (2.2.5). It tells us that

the specific dissipation, which consists of the mechanical dissipation
1 1
(4.2.10) m : = T  E –  –   = T  E –  +  
ρ ρ
1
with (4.2.5) =   – r + div q
ρ
and the thermal dissipation
q g
th : = – ,
θ
is non-negative (dissipation inequality)
(4.2.11) m + th  0 .
In the isothermal case (  const.) the CDI reduces to the CLAUSIUS-
PLANCK71 inequality
1
(4.2.12) T  E  .
ρ
70
Rudolf Julius Emmanuel Clausius (1822-1888), Pierre Maurice Marie Duhem
(1861-1916)
71
Max Karl Ernst Ludwig Planck (1858-1947)
One can also introduce the specific enthalpy

1
(4.2.13) h := T  E – 
ρ
and the specific free enthalpy or free GIBBS72 energy
1
(4.2.14)  := T  E – 
ρ
so that
(4.2.15) h +  =  +
holds.
All the above findings hold for all materials (under small deformations). To
further proceed, we particularise them to a special material class.
4.2.2 Thermoelasticity
For a complete thermomechanical description of the behaviour, one needs both
general balance laws and material laws. Here we will first focus our attention on
materials without memory, i.e., on thermoelastic materials.
For the independent variables we choose in each point the thermo-kinematical
state consisting of
 the deformations E = def u
 the temperature 
 the temperature gradient g = grad 
The specific body force b and the heat source r are considered as given. For the
dependent variables we choose the local caloro-dynamical state consisting of
 the stress tensor T
 the heat flux vector q
 the specific free energy 
 the specific entropy 
Principle of Determinisms for Simple Thermoelastic Materials

The thermo-kinematical state in each point determines its caloro-dynamical state.
Consequently, there exist four material functions
T = T (E ,  , g )
q = q (E ,  , g )
72
Josiah Willard Gibbs (1839-1903)
 =  (E ,  , g )
 =  (E ,  , g ) .
This choice of local variables already fulfils the local action assumption.
We will next study the consequences of the CDI for such thermoelastic material
laws. We assume that the thermo-kinematical process {E(t) , (t) , g(t)} of each
point is continuous and piecewise continuously differentiable, but otherwise
arbitrary. With
   
 (E ,  , g ) =  E +  +  g
E θ g
we obtain by the CDI
  T    q g
(4.2.16)     E + ( + )  + ( )  g +  0.
 E  θ g  θ
Since E, , g are arbitrary, and the terms in brackets are independent of these
rates, all the terms in the brackets must be zero.
Theorem. The CLAUSIUS-DUHEM inequality is fulfilled for thermoelastic
materials for all thermo-kinematical processes if and only if the following
conditions hold:
 the free energy is a potential for the stresses

(4.2.17) T = 
E
 as well as for the entropy

(4.2.18)  = –
θ
 the free energy, the stresses, and the entropy do not depend on the temperature
gradient
 the heat flux vector and the temperature gradient do not include an acute
angle
(4.2.19) qg  0 (heat condition inequality)
Consequently, the mechanical dissipation (4.2.10) is zero in elasticity: m = 0 .
By taking the second time derivative of the free energy, we obtain the
MAXWELL relation
 T 
(4.2.20)   = – .
θ    E
If we insert the potential relations into the definition of the free energy, we obtain
the GIBBS equality
1
(4.2.21)  = T  E –  
ρ
or with (4.2.8)
1
(4.2.22)  = T  E +  
ρ
so that the 1LT (4.2.5) gives
(4.2.23)    =  r – div q .
The internal energy can also be represented as a function of the strains and the
temperature
(4.2.24)  =  (E , ) +   (E , ) = :  (E , ) .
Its derivative with respect to the temperature is called the specific heat capacity
  
(4.2.25) c (E , ) : = = +  (E , ) + 
θ θ θ
  2
=  = –  2 .
θ θ

Alternative. If the entropy  (E , ) is for all E in  invertible, then
θ
cannot change its sign. Thus

> 0  c(E , ) > 0
θ
for all thermo-kinematical states. In this case, the roles of temperature and entropy
as independent and dependent variables can be interchanged (LEGENDRE73
transformation). This would lead to an alternative set of thermoelastic material
laws
T = T (E ,  )
q = q (E ,  , grad )
 =  (E , )
 =  (E ,  ) .
Here,
(4.2.26)  (E , ) =  (E ,  (E , )) = :  (E , )
(4.2.27)  (E , ) : =  (E , ) +   (E , )
and therefore
73
Adrien Marie Legendre (1752-1833)
   
= +
E E θ E
  
= + 
E E E
  
= +  (E ,  ) +  .
η η η
This gives the potential relations for the stresses
T       
(4.2.28) = = – = + =
 E E θ E E E E
and after (4.2.26)
     
= = – = + –
η θ η η η η
those for the temperature

(4.2.29)  (E , ) = .
η
4.2.3 Linear Thermoelasticity

We obtain the linear theory of thermoelasticity if we assume that
u, b,  – 0, r
as well as all needed derivatives of these variables with respect to time and space
are small, where 0 is an arbitrary, but fixed reference temperature. Small means
that we neglect higher-order terms compared to linear terms in these variables.
The difference temperature
(4.2.30)  : =  – 0
is thus small, but it can be positive or negative. After a series expansion and
truncation after the linear term, we obtain
1 1 1 1 1 
= =  (1 – ).
θ 0   θ0 1   / 0 θ0 0
1
So in a product with a small quantity we do not have to distinguish between
θ
1
and .
θ0
Also the mass density  is close to the density in the reference placement 0 . So
1 1
in a product with a small quantity we can substitute  by 0 and by .
ρ 0
We will further assume that the reference placement is stress-free at the
reference temperature
T(E  0 ,   0) = 0 .
We will now linearise the material laws at {E  0 ,   0 , g  o} . For the
stresses we obtain the DUHAMEL74-NEUMANN equality
T T
T = T(E , )   E + 
E θ
( T 0 /  ) ( T  0 /  )
(4.2.31)   E + 
E θ
= C  E +  M
with
( T /  )  2
C : = 0 = 0 (elasticity tetrad)
 E E 2
and with the MAXWELL relation (4.2.20)
( T /  )  2 
M : = 0 = 0 = – 0 (stress-temperature tensor)
θ Eθ E
Here all derivatives have to be evaluated at E  0 ,   0 ,   0 . Both of these
tensors are considered as constant within the linear theory. The elasticity tetrad is
symmetric due to the CDI, the stress-temperature tensor is symmetric as a
consequence of the balance of angular momentum C2 (3.2.8).
We will next investigate the dependence of the heat flux on the temperature
gradient. We expand q(E ,  , ) at E  0 , 0   ,   0 into a series in g at g
 o , which we truncate after the linear term
q(0 , 0 , g)  q0 – K  g
with q0 : = q(0 , 0 , o) and the constant heat conduction tensor
q
K := – g  o ,
g
which is assumed to be symmetric (CASIMIR (1945)  ONSAGER´s (1931)
reciprocal relations). If we multiply the series by g , we can apply the heat flux
inequality (4.2.19)
q  g  q0  g – g  K  g  0 ,
74
Jean Marie Constant Duhamel (1797-1872)
which must hold for all g . We can derive the following necessary and sufficient
conditions for this inequality to hold
 q0 = q(0 , 0 , o) = o
Without a temperature gradient there is no heat flux (non-existence of the
piezocaloric effect).
 gKg  0
(positive semi-definiteness of the heat conduction tensor).
As a result, in the linear theory of thermoelasticity FOURIER´s law holds in the
form
(4.2.32) q = –Kg (FOURIER)
with a constant positive semi-definite heat flux tensor K , which is assumed to be

symmetric.
By the energy balance 1LT (4.2.23) with (4.2.25) and (4.2.31) we conclude
1 1   
r – div q  r – div q =   =   +  E
ρ 0 θ E
θ θ
(4.2.33) = c  – M  E  c  – 0 M  E
ρ0 ρ0
or
1
(1LT) c  = ( M  E – div q) + r .
0
This equation can be used to determine the temperature change due to
deformations and heat supply by integration.
We obtain the following field equations of linear thermoelasticity
(DSR) E = def u
(2.2.5) g = grad 
(C1) div T + 0 (b – a) = o
(1LT) 0 r – div q + 0 M  E = 0 c 
(HOOKE) T = C  E +  M
(FOURIER) q = –Kg
with the material constants C , M , K , and c . If we insert the two material laws
into the balances we get
(C1) div(C  grad u) + div ( M) + 0 b = 0 u
(1LT) 0 r + div (K  grad ) + 0 M  grad u = 0 c 

for the balance of linear momentum and of energy, both being linear in u and
 . Evidently, the mechanical and the thermodynamical equations are coupled. If
we multiply the 1LT by  , the product rule renders
div ( K  g) = g  K  g +  div (K  g)
which gives the heat supply equation
0  r +  div (K  g) + 0  M  E = 0 c  
(4.2.34) = 0  r – div ( q) – g  K  g + 0  M  E
= ½ 0 c (2).
If C is invertible on the symmetric tensors, then the inverse form of HOOKE´s
law is
(4.2.35) E = S  T +  A
with S : = C –1 (compliance tensor)
A : = – S  M (temperature strain tensor)
in which S and A are symmetric.
By integration of the potential relations, we obtain the free energy with some
additive constant (which we set to zero)
1 1 c
(4.2.36)  (E , ) = E  C  E +  M  E – 2
2 0 0 2θ0
and the entropy
 1 c
(4.2.37)  (E , ) = – = – M  E +  .
θ 0 θ0
Using HOOKE (4.2.31) we have
 M  E = T  E – (C  E)  E
= T  E – (½ E  C  E)
and with the mechanical work balance WB (3.2.14)
 
 t  u dA +  b  u dm
A V
   
=  T  E dV + (½  u  u dm)
V V
and the heat supply equation (4.2.34)
1
 M  E = {½ 0 c (2) + div( q) –  0 r + g  K  g}.
θ0
Dividing it by 0 and integrating over the body, we obtain the thermoelastic

energy balance in the form
 =  
 t  u dA +  b  u dm
A V
1
(4.2.38) + {–   q  n dA +  r  dm –  g  K  g dV }
θ0 A V V
with the total energy
1 1
(4.2.39)  : = ½  ( u  u + E  C  E + c 2) dm .
V
 0 θ 0
4.2.4 Isotropic Linear Thermoelasticity

Further concretisations of the constitutive equations can be gained by assuming
particular symmetries. An orthogonal symmetry transformation Q renders
E* = Q  E  QT T* = Q  T  QT
(4.2.40) g* = Q  g q* = Q  q
* =  * =  .
Applying these transformations to the material laws, leads to the following
Definition. An orthogonal tensor Q is a symmetry transformation of the linear
thermoelastic material if
(4.2.41) Q  (C  E +  M)  QT = C  (Q  E  QT ) +  M
and
(4.2.42) QKg = KQg  Q  K  QT = K
hold for all E ,  , and g . They form the symmetry group G of the material.
For the important isotropic case, G consists of the entire orthogonal group. This
gives rise to the following representation.
Theorem. A linear thermoelastic material is isotropic if and only if
(4.2.43) T =  tr(E) I + 2 E +   I (HOOKE)
and
(4.2.44) q = –g (FOURIER)
hold with the LAMÉ constants  and  , the stress-temperature coefficient 
and the heat conduction coefficient  .
Proof. If we set in HOOKE´s law   0 , we obtain the mechanical law, for

which we already demonstrated the isotropic representation (4.1.17). If we instead
set E  0 , the isotropic symmetry conditions makes M an isotropic tensor
Q  M  QT = M .
M has at least one real eigenvalue  . Let {ei} be an ONB and e1 the
corresponding eigenvector. Then
Q  e2  e1 + e3  e2 + e1  e3
is orthogonal and
M  e2 = Q  M  QT  e2 = Q  M  e1 =  Q  e1 =  e2 .
So { , e2} also solves the eigenvalue problem of M , the same as { , e3} does,
so that all eigenvalues must be equal and all vectors are eigenvectors. This holds
only for a spherical tensor
M = I.
In the same way one shows that the heat conduction tensor is spherical
K = I
with a real heat conduction coefficient  ; q.e.d.
 is non-negative due to the heat conduction inequality (CDI).
For   0 and 3 + 2  0 an inverse form of HOOKE´s law exists
1 
(4.2.45) E = [T – tr(T) I] +   I
2 3  2
with the thermal expansion coefficient

 := – .
3  2
The free energy is in the isotropic case after (4.2.36) and (4.1.34)
1  c
(4.2.46)  (E , ) = [ tr2(E) +  E  E +   tr(E)] – 2
0 2 2θ0
and the entropy after (4.2.37)
 1 c
(4.2.47)  (E , ) = – = –  tr E +  .
θ 0 θ0
With the isotropic representations we eliminate the stresses and heat fluxes in
C1 and 1LT and obtain as an extension of ND (4.1.37) for homogeneous linear
isotropic thermoelastic materials the
Temperature-displacement relations
 div grad u + ( + ) grad div u +  grad  + 0 (b – ü) = o
(4.2.48)
 div grad  + 0  div u + 0 r = 0 c .
This is a set of four coupled linear PDEs in u and  . The thermodynamical and
the mechanical problems are only decoupled for   0 (  0 ).
4.2.5 Boundary and Initial Value Problems of

Thermoelasticity
For the boundary and initial value problems of linear thermoelasticity we

prescribe the following fields:
 the body force b and the heat source r in the interior of the body
 initial values for the following fields
displacements u(x, t  0)
velocity v(x, t  0)
temperature difference (x, t  0)
in the interior body region V
and at all times the boundary conditions for
 the displacements upr on a part Ag of the surface A
 the tractions tpr on another part Ad of A
 the temperature difference pr on a part A of A
 the heat flux qpr  n on another part Aq of A
with A = Ad  Ag Ad  Ag = 
A = A  Aq A  Aq = 
The solution {u , } of this problem must further fulfil the balances of linear
momentum and energy at all times. In the homogeneous and isotropic case these
are the temperature-displacement relations (4.2.48).
Existence and Uniqueness Theorem. If C is positive semidefinite and c > 0 ,
then the mixed boundary and initial value problem has exactly one solution.
For the proof see DAFERMOS (1986).

In the thermostatic case we assume
  0 and u  o
for the entire body at all times. The field equations are then reduced to the
thermostatic equilibrium conditions
(4.2.49) div (C  grad u +  M) + 0 b = o (C1)
(4.2.50) div (K  grad ) + 0 r = 0 (1LT)
The first condition can be formally turned into the elastic equilibrium condition
div T + 0 b = o
by the introduction of a reduced stress
T : = C  def u = T –  M
and a reduced body force
1
b := b+ div ( M) .
0
With this analogy we can use the results of elastostatics for thermoelastostatics.
Problem 20. Thermoelastic Boundary Value Problem

A hollow sphere is submitted to a stationary radial temperature gradient.
The material is isotropic and thermoelastic. Determine the temperature
field for given outer and inner temperature assuming an isotropic
FOURIER heat flux. Determine the induced stress field assuming traction-
free boundary conditions on both surfaces.
Ra , a Ri , i
e3 e1
e2
Solution
We will use Cartesian coordinates. Then the associated base vectors do not
depend on the coordinates, which simplifies the calculation. On the other
hand, the spherical symmetry has to be accounted for by other means. Even
with respect to Cartesian coordinates, we can implement the spherical
symmetry of the problem by letting all fields depend only on the radius R
with
R2 = x12 + x22 + x32.
Using FOURIER´s isotropic law, (4.2.48) gives for the static case (u = o)
without heat sources
 = a div grad 
with the conduction coefficient a : =  / (0 c) , i.e., the heat conduction
divided by the heat capacity density 0 c . In a stationary case we have
 = 0 , so that
(P20.1) 0 = div grad  =  , ii .
In what follows we use a prime (') for the derivative with respect to R .
The derivatives with respect to the coordinates xi can be transformed by
the chain rule into those with respect to R using
R xi
 .
 xi R
We now exploit P20.1

xi
x x x
 ii R  xi
0 =  , ii = (' i ) , i = '' i 2i + ' R
R R R2
= '' + 2 ' / R .
After the convention of sums we have  ii = 3 and xi xi = R2. By

substitution of ' = P(R) , we obtain the first-order ODE
0 = P' + 2 P / R .
Its solution is P(R) = – 1 / R2. Integration gives
 = 1 / R + 2
with parameters 1,2 which have to be adopted to the boundary conditions.
With i = (Ri) and a = (Ra) we get
1 = – Ri Ra (a – i) / (Ra – Ri)
2 = (Ra a – Ri i) / (Ra – Ri)
For a linear isotropic thermoelastic material after (4.2.43) the stresses can
be determined with respect to a Cartesian basis as
(P20.2) ij =  kk ij + 2  ij +   ij
with the LAMÉ constants  and  and the stress-temperature coefficient
 , which is related to the compression modulus and the thermal expansion
coefficient by  = – 3K . The strains are
 ij = (ui , j + uj , i) / 2
and we need an ansatz for displacements. Since the sphere will expand
isotropically, the displacement field u(x) must be parallel to the position
vector x , so that an ansatz
ui = F(R) xi
is justified. We insert this into P20.2
ij =  ur , r ij +  (ui , j + uj , i) +   ij
=  ( F' xr xr /R + F rr) ij
+  (F' xj xi /R + F ij + F' xj xi /R + F ij) +   ij
(P20.3) = ( F' R + (2 +3) F ) ij + 2 F' xj xi /R +   ij .
We will next use the equilibrium condition
ij , j = 0
= [ F' R + (2 +3) F ] , i + 2 (F' xj xi /R) , j +   , i
= [ F'' xi +  F' xi / R +(2 +3) F' xi / R]
+ 2 (F'' xj xi xj / R + F' jj xi + F' ij xj) / R
– 2 F' xj xj xi / R3 +  ' xi / R
= [ F'' xi +  F' xi / R + (2 +3) F' xi / R]
+ 2 (F'' xi R + 3 F' xi + F' xi) / R – 2 F' xi / R +  ' xi / R
= xi [( + 2)(F'' +4 F' / R) +  ' / R] .
This equation is fulfilled everywhere if the term in brackets vanishes. Thus
we have to solve the ODE
F'' + 4 F' / R = C / R3 with C : =  1 / ( + 2) .
By the substitution F' = P we obtain
P' = C / R3 – 4 P(R) / R = A(R) + B(R) P(R)
with
A(R) = C / R3
B(R) = – 4 / R .
The general solution is
P = P1 + P2 with P1 = P0 e 
B dR P1 AP0
and P2 =  dR .
P0 P1
We do not have to introduce an integration constant, since we already have

P0 as a parameter. We get
P1 = P0 / R4 P2 = C / (2 R2) .
In order to obtain F we have to integrate P1 + P2 with respect to R ,

which leads to an integration constant K2 . For convenience we replace
K1 = – P0 / 3 . So
F = K1 / R3 – C / (2R) + K2 .
K1 and K2 have to be determined by the boundary conditions. We assume
traction-free surfaces
ij nj = 0
at R  Ri and R  Ra . The normal vector n is again parallel to the
position vector. There is no need to normalise n in the above equation.
We obtain
0 = ij xj
= {[F' R + (2 +3 ) F ] ij + 2 F' xj xi / R +   ij} xj
= [F' R + (2 + 3 ) F + 2  R F' +  ] xi .
So the term in square brackets must be zero for R  Ri and R  Ra . This
gives two linear equations with the solutions
K1 = Ri2Ra2 (1 – C) / [4 (Ri2 + Ri Ra + Ra2)]
K2 = [ {1 (Ra + Ri) + 2(Ri2 + Ri Ra + Ra2 )} – C (Ra + Ri)] /
[(2 + 3 ) (Ri2 + Ri Ra + Ra2 )] .
For Ri =100 mm and Ra = 200 mm , the material constants E = 210 GPa,
v = 0.3 ,  = –2.44 MPa/K , and the thermal loading i = 400K , a = 0 K
we obtain the results presented below. The stresses have been determined
by P20.3 at x1  R , x2  x3  0 . This leads to normal stresses and
tangential stresses
11 = ( F' R + (2 +3 ) F ) + 2 F' R +  
22 = 33 = ( F' R + (2 +3 ) F ] +  
while all shear stresses are zero here.
Radial displacements (left) in mm, radial (starting and ending at zero)

and tangential stresses (right) in MPa over R in mm.
4.3 Linear Viscoelasticity
From the one-dimensional theory of linear viscoelasticity in Chapt. 1.3 we know

already the following representations:
 as a convolution after BOLTZMANN
t
 (t) =  R(t – )  ( ) d
0
t
 (t) =  J(t – )  ( ) d
0
 as an ODE of (p, q)–type

M0  + M1   +  + Mp  (p) = Q0  + Q1   +  + Qq  (q)
where the coefficients Mi and Qi can be calculated from the elementary elastic
constants E of the springs and viscosity constants D of the dampers.
For the three-dimensional generalisation, we replace
 the uniaxial stress  by the stress tensor T
 the same for its time-derivatives  ,  
,  by T , T , 
 the uniaxial strain  by the strain tensor E
 the same for its time-derivatives  ,  ,  by E , E , 
 and all proportionality factors Mi , Qi , E , and D by tetrads Mi , Qi , E ,
and D as the general linear mappings between tensors.
Since all these 2nd-order tensors are symmetric, we can assume both
subsymmetries for the tetrads without loss of generality, and use VOIGT´s
representation.
Moreover, by another assumption we can drastically simplify the material laws
by use of the projection method.
Assumption of Coaxiality
All material tetrads of the material have the same eigenspaces.
For isotropic material tensors this is automatically fulfilled, since spherical
tensors and deviators constitute the (two) eigenspaces of an isotropic tetrad. For
anisotropic materials these eigenspaces are determined by the anisotropy
directions like the lattice directions for crystals. In the particularly important case
of cubic symmetry this assumption is also automatically fulfilled, if the anisotropy
axes coincide. For all other anisotropy classes the assumption of coaxiality,
4.3 Linear Viscoelasticity 267
however, is not a matter of course, but a real restriction, even if it appears to be

plausible in many cases.
Based on this assumption, we can represent the (linear) material laws of the
elements of the rheological models as follows.
 linear spring ( = E ) :
(4.3.1) T = C  E = C i Pi  E = C i E i
with the projections of E into the eigenspaces of C
E i : = Pi  E
and the elastic constants C i  R , i = 1,  , N and the according spectral
form
C = C i Pi ;
 linear damper ( = D  )
(4.3.2) T = D  E = D i Pi  E = D i E i
with the viscosity constants D i  R , i = 1,  , N and the spectral form
D = D i Pi .
Here, N is the number of eigenspaces of the tetrads and depends on the material
(isotropy: N  2 ; cubic: N  3). We will as usual assume that all elastic constants
C i of all springs and all viscosity constants D i of all dampers are positive.
Then C and D are positive definite and invertible on the symmetric tensors.
The specific dissipation of the dampers is then
1 1
 =
 
 T  E =
 
 (D  E)  E
1 1
(4.3.3) =
 
 E  D  E =
 
 D i E i  E i
which is positive for all non-zero strain rates E , as we would expect for
thermodynamic consistency.
If we apply a projector Pi to the elastic or viscous law, we achieve
T i : = Pi  T = C i E i i = 1,  , N
and
T i = D i E i i = 1,  , N
(no sum), respectively. The advantage of this projection method is that we could
reduce tensorial material constants to scalar ones, identically fulfilling the material
symmetry conditions.
As an example for the projection method we reconsider the BURGERS model

from Sect. 1.3.5. The elementary material laws for the two springs and two
dampers are written in analogy to the one-dimensional case as
(4.3.4) TC = C  EC = C i Pi  EC  TCi = C i ECi (no sum)
(4.3.5) TK = K  EK = K Pi  EK i
 TKi = K EKi i
(no sum)
(4.3.6) TD = D  ED = D i Pi  ED  TDi = D i EDi (no sum)
(4.3.7) TR = R  ER = R i Pi  ER  TRi = R i ERi (no sum)
with material constants C , C , ... , C , K , ... , K , D , ... , D , R 1, ... , R N,
1 2 N 1 N 1 N
which are altogether positive.
C D
R BURGERS model
Due to the coaxiality assumption, all material tetrads in these laws are coaxial.
Consequently, the composition of these tensors commutes, e.g.,
–1 –1 –1 –1 –1 –1
K  C  R = C  K  R = R  C  K
etc. The equations of the BURGERS model are
ERi = EKi
Ei = ECi + EDi + ERi i = 1 , ... , N
Ti = TCi = TDi = TKi + TRi
and thus (no sum)
1 1 1
Ei = i
Ti + i
Ti + (Ti – TKi)
C D Ri
TKi = K i EKi = K i ERi = K i(Ei – ECi – EDi)
Ki Ki
= K i Ei – i
Ti – Ti
C Di
1 1 1 Ki Ki Ki
 Ei = i
Ti + i
Ti + i
Ti – Ei + i i
Ti + Ti
C D R R i
C R Di Ri
Ki 1 Ki 1  1  Ki 
 Ti+( + + ) Ti + Ti = Ei + Ei
Di Ri Di C i Ri Ri Ci Ri
4.3 Linear Viscoelasticit 269
which is a system of ODEs of (2,2)-type

M0i Ti + M1i Ti + M2i Ti = Q0i Ei + Q1i Ei + Q2i Ei (no sum)
with the constants (no sum)
Ki 1 Ki 1 1
M0i : = M1i : = i
+ + i
M2i : =
D Ri i
D i
C R i
R Ci
Ki
Q 0i : = 0 Q 1i : = Q2i : = 1
Ri
for i = 1, ... , N . If we superimpose these equations in the N eigendirections, we
obtain (sum over i)
M0i Pi  T + M1i Pi  T + M2i Pi  T
= Q0i Pi  E + Q1i Pi  E + Q2i Pi  E
(4.3.8) = M0  T + M1  T + M2  T = Q0  E + Q1  E + Q2  E
with
Mj : = Mji Pi (sum over i)
Qj : = Qji Pi .
A comparison renders the relations
–1 –1
M0 = K  D  R
–1 –1 –1 –1
M1 = D + K  C  R + R
–1
(4.3.9) M2 = C
Q0 = 0
–1
Q1 = K  R
Q2 = I
Due to the projection method, we can handle the materials laws (after their
projection into eigenspaces) in the three-dimensional case in the same way as in
the one-dimensional one, i.e., like scalar equations. And one can, vice versa,
transform the one-dimensional equations directly into tensorial ones. This method
works for all rheological models in linear viscoelasticity. So there is no need for
us to reconsider all the findings from one-dimensional theory. Instead, we can
directly translate them into three dimensions.
This method is applicable to all linear coaxial laws. All these models lead to
tensorial ODEs of (p, q)-type
(4.3.10) M0  T + M1  T +  + Mp  T(p)
= Q0  E + Q1  E +  + Qq  E(q).
A tensorial form of the convolution integrals after BOLTZMANN is

t

(4.3.11) T(t) =  F(t – )  E() d
0
t

(4.3.12) E(t) =  G(t – )  T() d
0
with (sum over i)

F(t) = fi (t) Pi
G(t) = gi (t) Pi
where fi and gi are one-dimensional relaxation and creep functions, respectively.
For non-linear laws, however, the generalisation to three dimensions becomes
much more complicated, as we will see for plasticity in the sequel.
Literature on the projection method:

Bertram, A.; Olschewski, J.: Zur Formulierung linearer anelastischer
Stoffgleichung mit Hilfe einer Projektionsmethode. Z. ang. Math. Mech. 73,4-5,
T401-3 (1993)
and on the non-linear generalisation
Bertram, A.; Olschewski, J.: A phenomenological anisotropic creep model for
cubic single crystals. In: Handbook of Materials Behaviour Models. Edt.: J.
Lemaitre. Academic Press. Vol. 1, 303-307 (2001)
4.4 Plasticity 271
4.4 Plasticity
Literature
Betten, J.: Kontinuumsmechanik. Springer, Berlin (1993), 2. edt. (2001)
Burth, K.; Brocks, W.: Plastizität. Vieweg, Braunschweig/ Wiesbaden (1992)
Chakrabarty, J.: Theory of Plasticity. McGraw-Hill (1987)
Chen, W. F.; Han, D. J.: Plasticity for Structural Engineers. Springer-Verlag,
New York, Berlin (1988)
Francois, D.; Pineau, A.; Zaoui, A.: Mechanical Behaviour of Materials, vol. I,
Hill, R.: The Mathematical Theory of Plasticity. Clarendon Press, Oxford (1950)
Lubliner, J.: Plasticity Theory. Macmillan, New York (1990)
Pawelski, H. u. O.: Technische Plastomechanik. Verlag Stahleisen, Düsseldorf
(2000)
Prager, W.; Hodge, P. G.: Theorie ideal plastischer Körper. Springer, Vienna
(1954)
Rösler, J.; Harders, H.; Bäker, M.: Mechanisches Verhalten der Werkstoffe. B.
G. Teubner, Stuttgart (2003)
Yang, W.; Lee, W. B.: Mesoplasticity and its Applications. Springer, Berlin
(1993)
Yu, M.-H.: Generalized Plasticity. Springer, Berlin (2006)
The classical theory of plasticity has the following ingredients:

1.) an assumption on the decomposition of the strains E into an elastic part Ee
and a plastic part Ep in analogy to (1.4.4)
(4.4.1) E = Ee + Ep
2.) an elastic law, which describes the behaviour within the elastic ranges or, more
precisely, which describes the relation between elastic strains and stresses
(4.4.2) T = C  Ee .
This elastic law can be either isotropic or anisotropic. The assumption that plastic
deformations have no influence on C is useful and important. For many materials
like metals, this assumption is confirmed by experiments even for large plastic
deformations (which we do not consider in the present context).
3.) a yield criterion, which determines the limit of the current elastic range of the
material. In the one-dimensional theory of Chapt. 1.4 this could be described by
some scalar yield stress Y such that the elastic range is characterised by
   Y . In the three-dimensional theory, however, such a criterion must take
into account the multiaxiality of the stress state. We make the following ansatz for
the yield criterion as a scalar-valued function
 (T , Z)
of the stress tensor T and additional internal variables Z , which can be scalars
or tensors (here notated as tensors of second-order), which describe the hardening
of the material, i.e., the influence of previous deformations on the yield limit. The
yield limit is the kernel of this criterion
(4.4.3)  (T , Z) = 0
while we postulate for the interior of the elastic range (and only there)
 (T , Z) < 0 .
4.) a flow rule, which describes the evolution of the plastic deformation during
yielding. Its general form is assumed as a first-order ODE (as is customary for
evolution equations)
Ep = F(E , Ep , Ee , T , Z , E) .
Since we can eliminate the total strain E and the stresses T by (4.4.1) and
(4.4.2), the following ansatz would be equipollent
(4.4.4) Ep = f (E , T , Z , E) .
5.) a hardening rule for Z , which describes the evolution of hardening or
softening during yielding, here assumed in the same form as the flow rule
(4.4.5) Z = z(E , T , Z , E) .
The most important examples of hardening models are isotropic and kinematic
hardening, which are described by a scalar parameter and a symmetric tensor,
respectively, as we will show later in more detail.
4.4.1 Yield Criteria
Most solid materials consist of crystals on a micro scale. From crystal physics
two facts are well-known and well-substantiated.
 The stresses in a crystal result from deformations of the crystal lattice and can
therefore be identified with the elastic strains Ee .
 Plastic deformations occur under conservation of the lattice. Ep therefore
describes the deformation of the material in relation to the lattice.
4.4 Plasticity 273
The most important contribution of plastic deformations in crystals results from

crystallographic slips (shears) in distinct planes (slip planes) and distinct
directions (slip directions), which are characterised by the densest alignment of
atoms in the lattice. If we describe the slip plane by its normal vector  n and
the slip direction by a vector  d , then the pair {d , n} constitutes a slip
system. Since the slip direction must lie in the slip plane, we always have the
orthogonality
(4.4.6) d  n = 0 .
One can norm both vectors to one:  d  = 1 =  n . The positive sense of d
and n is irrelevant for most slip systems. The index  is a label standing for the
particular slip system, the number N of which depends on the crystallographic
system of the material under consideration.
stress free stress free not stress free

d d
n n
initial lattice plastically deformed material elastically deformed material
Such a slip system can be activated after SCHMID75´s law if the shear stress in it
(4.4.7)   : = d  T  n = T  d  n = T  sym(d  n) ,
called the SCHMID stress (1924) or resolved shear stress, has reached a critical
value, the critical resolved shear stress
   =  c.
Algebraically, the calculation of the resolved shear stress is a projection of the
stress tensor into the particular slip system. However, the projections of the
different resolved shear stresses are not linearly independent, since the number of
slip systems N is usually larger than the dimension 6 of the stress space.
 c can change its value due to slips of the same slip system (self-hardening) as
well as the other slip systems (cross-hardening or latent hardening). Thus, the
different critical shear stresses  c are only initially material constants, and
become later hardening variables, which we collect in the vector
Z = { c1,  c2, ... ,  cN}.
The SCHMID yield criterion can be represented in the following form
75
Erich Schmid (1896-1983)
(4.4.8)  (T, Z) = max (   –  c )


= max (T  sym(d  n ) –  c ) = 0 .


If it is fulfilled for some slip system {d  n} , then we have

 
(4.4.9) =  sym(d  n ) and = – 1.
T  c
If more than one slip system fulfil this criterion simultaneously, then  (T, Z) is
not differentiable at this stress state. In the case of    being smaller than
 c , this particular slip system is currently not activated.
The SCHMID law is perhaps the physically best understood yield criterion. Its
application, however, is limited since the size of the grains with crystallographic
structure is for many materials rather small if compared to machine parts etc. It
would mean a prohibitive amount of computational costs to reduce the description
of plasticity to glide mechanisms in grains. Nevertheless, at least two important
fields of applications of crystal plasticity shall be mentioned
 single crystals, which are nowadays produced in rather large parts as, e.g.,
turbine blades;
 micro-macro investigations to study the behaviour of grains within
polycrystalline structures.
A polycrystal possesses a crystalline structure or lattice within each grain,
which constitutes the micro-level. On the macro level, the behaviour of a
polycrystal is dominated by mean values. These two levels have to be carefully
distinguished, since their behaviour can be rather different. The crystalline
behaviour, for example, will be anisotropic in most cases. The polycrystal,
however, can be isotropic, since the anisotropy may rule out in the mean. Or
another type of anisotropy can appear on the macro scale due to a crystallographic
texture, as it is introduced by, e. g., forming processes (strain induced anisotropy).
Let us next look at (macro-scale) isotropic yield criteria.
One of the first suggestions for an (isotropic) yield criterion is due to
COULOMB (1773) in analogy to the friction law of AMONTONS76 and
COULOMB. Here it is assumed that the plastic deformation can occur as a slip in
some slip plane as soon as the shear stress reaches a critical value after the ansatz
(4.4.10) | c | =  0 +  =  0 +  tan 
with the normal stress  in the slip plane, the angle of friction  , the friction
coefficient  = tan  and some constant  0 .
77
MOHR (1900) generalised this to a non-linear dependence on the normal
stress
W c = W 0 + f (V) .
76
Guillaume Amontons (1663-1705)
77
Otto Mohr (1835-1918)
4.4 Plasticity 275
Such yield criteria are applied in soil mechanics, but not in metals for which the
critical shear stress hardly depends on the normal stress in the slip plane (non-
SCHMID effect).
| c | =  0 +  tan 
 
  0
 H3  H1 0 

½ (1+ 3)
COULOMB´s yield criterion
More successful for metal plasticity was the suggestion by TRESCA78 (1865)
to relate the maximal shear stress to a critical value
(4.4.11) max =  c .
If we express this by the principal stresses, it is equivalent to three criteria
13  = ½  p1 –  p3 =  c
23  = ½  p2 –  p3 =  c
12  = ½  p1 –  p2 =  c
or in a more compact form
(4.4.12)  (T ,  c)
= [( p
1 –  3) – 4  c2] [( p2 –  p3)2 – 4  c2] [( p1 –  p2)2 – 4  c2] = 0 .
p 2
If we do not want to use the principal stresses, then we obtain for a state of plane
stress
2
  11   33 
(4.4.13) 13max =     13 2 .
 2 
In the case of a uniaxial stress state only
78
Henri Edouard Tresca (1814-1885)
2
  xx   xx
(4.4.14) max =  2  = 2
 
remains.
The main disadvantage of TRESCA´s yield criterion is that the yield loci, if
understood as a hyper-plane in the stress space, is continuous, but not
differentiable. In fact, if two principal stresses have the same value, this surface
has a kink at such a state.
An important property of this yield criterion and also of SCHMID´s criterion is
their pressure-independence, since max is determined by the difference of two
principal stresses. If we decompose the stress tensor into its spherical part
(pressure) and its deviator T = – p I + T', then pressure-independent yield
criteria must not depend on p but only on the deviator T'. In the isotropic case, a
scalar tensor function can only depend on the invariants of the tensor
iso(IT , IIT , IIIT)
the first of which is proportional to the pressure. Thus, IT can be dropped from
the list of arguments for pressure-independent yield criteria. We could instead
introduce the invariants of the stress deviator as
J1 : = IT ' = 0
(4.4.15) J2 : = – IIT '  0
J3 : = IIIT ' .
If only a dependence on the second invariant is taken into account, one calls such
a theory a J2 -theory. This leads to the yield criterion after HUBER79 (1904) and
v. MISES80 (1913), which can be brought into the form
(4.4.16) equ2 : = 3 J2   Y2
or as
(4.4.17)  (T , Y) = equ2 –  Y2 = 3 J2 –  Y2 .
equ is called the v. MISES equivalent stress. Yielding can be induced if
equ  Y holds.
Applying this to a uniaxial state of stress (with all ij  0 except xx) gives
J2 = 1/3 xx2 = 1/3 equ2  equ = xx .
The equivalent stress thus stands for the critical tension stress in a uniaxial stress
test.
For a state of shear (only xy = yx non-zero stress components) we obtain
J2 = xy2 = 1/3 equ2  equ = 3 xy.
79
Maksymilian Tytus Huber (1872-1950)
80
Richard M. E. von Mises (1883-1953)
4.4 Plasticity 277
Thus, a shear stress appears with a factor of 3 in the equivalent stress.

For a general state of stress we have
(4.4.18) J2 = – IIT '
= –  'xx  'yy +  'xy2 –  'xx  'zz +  'xz2 –  'yy  'zz +  'yz2
with the components of the stress deviator
 'xx = xx – 1/3 IT = 2/3 xx – 1/3 (yy + zz)
 'yy = yy – 1/3 IT = 2/3 yy – 1/3 (xx + zz)
 'zz = zz – 1/3 IT = 2/3 zz – 1/3 (xx + yy)
 'xy = xy  'yz = yz  'xz = xz
After insertion of this into J2 we get
(4.4.19) J2 = 1/6 [(xx – yy)2 + (yy – zz)2 + (zz – xx)2] + xy2 + yz2 + zx2.
Note that J2 is always non-negative. With respect to the principal stress axes, this
gives
(4.4.20) J2 = 1/6 [( p1 –  p2)2 + ( p2 –  p3)2 + ( p3 –  p1)2] .
Using the condition that the deviator is traceless
J1 =  'xx +  'yy +  'zz = 0
we get
J2 = 1/3 equ2 = – IIT ' = ½ tr(T'2)
= ½  'ij  'ij = ½ T'  T' = ½ T'2
= ½ ( 'xx2 +  'yy2 +  'zz2 +  xy2 +  yz2 +  zx2 +  yx2 +  zy2 +  xz2)  0 .
J2 is proportional to the norm of the stress deviator. The equivalent stress can
therefore be cast into the form
equ = (3 J2) = (3/2 T'  T') = (3/2) T'.
The HUBER-v. MISES yield criterion is
(4.4.21)  (T , Y) = 3/2 T'  T' –  Y2 = 3/2 T'2 –  Y2
and consequently
 V 
(4.4.22) = 2equ = 2equ (3/2 T'  T')
T T T
= 2equ ½ (3/2 T'  T')1/2 3/2 2 T' = 3 T'
because of

d(T'  T') = 2 T'  dT' = 2 T'  dT = (T'  T')  dT.
T
Another interpretation of this equivalent stress comes up if we decompose the

elastic energy of an isotropic material after (4.1.35) into the distortional and the
compression energies. The distortional energy is
1 J
(4.4.23) wD(E') =  tr(E' 2) = tr(T'2) = 2 .
4 2
Thus, the HUBER-v. MISES yield criterion limits the distortional energy for
isotropic materials, while it does not depend of the dilatational/compression
energy.
In contrast to TRESCA´s, this yield criterion has the advantage that it is
differentiable everywhere.
Generalisations of the HUBER-v. MISES criterion to pressure-sensitive
materials can be made by assuming that the critical equivalent stress depends on
the pressure: Y (p) . Other generalisations go in the direction of anisotropic
behaviour. There are general anisotropic yield criteria suggested by v. MISES
(1928) and in particular for orthotropic materials by HILL81 (1948) as a square
form of the stress deviator
(4.4.24)  (T ,  Y) = T'  K  T' –  Y2.
K is a symmetric and positive semidefinite material tetrad submitted to the
symmetry of the corresponding symmetry group of the material. Representations
of such tetrads are well-known from elasticity. If we set K as a multiple of the
fourth-order identity I , we would again obtain the isotropic HUBER-v. MISES
criterion. If we instead use an orthotropic tetrad, then we get HILL´s criterion. In
general, we have

(4.4.25) = 2 K  T'.
T
Graphical representations of yield criteria are not easy since they act in the 6-
dimensional space of symmetric tensors or its 5-dimensional subspace of the
deviators. If the yield criterion is isotropic, one can reduce the space to the three-
dimensional space of the principal stresses  p1 ,  p2 , and  p3 . Moreover, such
representation must be invariant under any changes of them. If we denote the
diagonal of this tripod as
1
n := { p1  1,  p2  1,  p3  1}
3
then the yield surface must be invariant under rotations of the type Rn2/3 due to
isotropy. If the criterion is pressure-independent, then the surface must be
prismatic in the direction of the diagonal n . Examples for such criteria are the
circular cylinder of HUBER-v. MISES and a hexagonal prism of TRESCA. In
these cases no information gets lost by the projection of the prism in the plane
81
Rodney Hill (1921-2011)
4.4 Plasticity 279
with normal n , the plane of principal stresses (sometimes called the octahedral
plane, HAIGH-WESTERGAARD plane, or -plane).
 H3 n
 H2
-plane
 H1
v. MISES circular cylinder
In this plane the yield surface of HUBER-v. MISES is a circle, and TRESCA´s is
a regular hexagon.
 p3
TRESCA v. MISES
 p1
 p2
Projection into the octahedral plane

 p3
n
 p2
 H1
Representation in the space of principal stresses
If we look at the cut section of the yield surface with one of the planes spanned
by two principal stress axes, we find an ellipse for HUBER-v. MISES´ and a
hexagon for TRESCA´s criterion.
2 – 3
v. MISES
TRESCA
1 – 3
Such a representations allow for a comparison of these important yield criteria.

If we calibrate both of them by a shear test, then the HUBER-v. MISES circle is
enclosed by the TRESCA hexagon. If we instead calibrate them for a uniaxial
tension or compression test, then the TRESCA hexagon is enclosed by the
HUBER-v. MISES circle. The largest differences between the two criteria occur
for a pure state of shear such as
 p1  –  p2  p3  0 .
We obtain for the
4.4 Plasticity 281
1
 HUBER-v. MISES criterion  p1 =  equ
3
 TRESCA criterion  p1 =  ½ equ
which is a maximal difference of approximately 15 %.
Experiments for isotropic metals give values which lie in most cases in between
the two surfaces.
4.4.2 Flow Rules
The (current) elastic range of the material point is bounded by the yield limit.
Within the elastic range, the stresses are given by the elastic law T = C  Ee . If
the current state is on the yield surface, then the material is prone to plastic flow or
yielding. Such events are characterised by the loading condition: If the
deformation increment were purely elastic, then this would lead to stress states
outside of the current elastic range. This, however, is not possible. Therefore, the
contrary must hold: The total strain increment must contain plastic parts, i.e.,
yielding must occur to ensure that the state of stress remains on the yield surface.
This leads to the following loading condition (in the strain space)

 (T , Z) =  C  E > 0 for E  Ee and Z  0 .
T
We have to distinguish three cases, when the yield condition is fulfilled:

1. unloading  C  E < 0
T

2. neutral loading  C  E = 0
T

3. loading  C  E > 0
T
While the first and the second case can be realised elastically, i.e., without
yielding, the third case necessarily requires yielding. Consequently we have two
necessary conditions for the material to yield, namely the yield condition and the
loading conditions

(4.4.26)  (T , Z) = 0 and  C  E > 0 .
T
The reverse also holds: If these two conditions hold simultaneously, then yielding
must occur.
If the increment of the total strain E is given, a constitutive description is

needed to decide how this increment splits into elastic and plastic parts. This leads
to the concept of a flow rule.
In the case of the slip system theory of crystal plasticity, one thinks of plastic
deformations taking place as shears in the activated slip systems {d , n} with a
shear rate
 d  n
where  is the shear number. If more than one slip system is active
simultaneously, then we have to add the shear rates
  d  n.

The symmetric part of this tensor stands for the rate of plastic deformation
(4.4.27) Ep =  ½  (d  n + n  d)


which is generally traceless or isochoric. Traceless symmetric tensors like Ep

have five independent components. Therefore we need in principle at least five
active slip systems. The partial shear rates  have to be determined by the
consistency conditions, which ensure that the resolved shear stress of a slip system
remains critical during slip. If these conditions are not sufficient to uniquely
determine all shear rates, one needs in addition a criterion for selecting the active
set (TAYLOR problem).
By integration of the plastic deformation rates along the deformation process,
one obtains the final plastic deformation
t
(4.4.28) Ep =  Ep dt .
0
The dependence of this variable on the deformation process is characteristic for

plasticity (path-dependence).
If we consider the case of only one active slip system (simple glide), then the
direction of the plastic strain rate Ep equals the direction of the derivative of
SCHMID´s yield criterion with respect to the stress tensor
  T, Z 
 sym(d  n)  Ep.
T
Such a relation is called an associated flow rule. If more than one slip system is
active (multiple glide), then the associated flow rule holds for each of them
separately, but not as a superposition.
4.4 Plasticity 283
Let us next consider the phenomenological plasticity theory. The most important
flow rule is that of a plastic potential (MELAN82 1938), a real-valued function of
the stress tensor  (T) , after which the plastic strain increment Ep goes into the
direction of the gradient of  (T)
d
Ep = 
dT
with a positive real  called the plastic parameter, which we will determine
later. If we interpret the loci for which the potential has equal levels
 (T) = const.
as 5-dimensional hypersurfaces in the 6-dimensional stress space, then the
gradient of  goes in the direction of the normal on this surface.
d
dT
 (T) = const.
If we choose for the plastic potential  (T) the flow criterion  (T , Z) , then
we obtain the associated flow rule after v. MISES (1928)

(4.4.29) Ep =  .
T
Such associated flow rules play an important role in metal plasticity, while they
are less appropriate in other branches of plasticity like soil mechanics.
In each case, the KUHN-TUCKER conditions hold
 =0 with   0 and   0.
The associated flow rule of the HUBER-v. MISES criterion is with (4.4.22)
(4.4.30) Ep =  3 T'
and together with the isotropic HOOKE´s law
1 1 
(4.4.31) E = Ep + Ee = 3 T' + T' – p I
2G 3k
82
Ernst Melan (1890-1963)
This is the PRANDTL83-REUSS84 equation (1924, 1930). Its spherical part

equals the dilatational part of HOOKE´s law (4.1.13), while the deviatoric part is
1
E' = 3 T' + T'.
2G
4.4.3 Hardening Rules
Before we can completely establish the flow rule (not only the flow direction),
we must consider hardening. In the one-dimensional theory in Chapt. 1.4 we
already introduced isotropic and kinematic hardening. This shall now be
generalised to three dimensions.
In the case of isotropic hardening, the yield surface is blown up in an affine
way. If we describe the flow criterion as in (4.4.17), then Y is no longer a
material constant (as it is in perfect plasticity), but rather evolves during plastic
processes, possibly in a linear way in analogy to (1.4.3)
(4.4.32) Y = K p K : material constant
or non-linear
(4.4.33) Y = K(p) p K(p) : material function
or with saturation
(4.4.34) Y = K ( Y –  Y) p K,  Y : material constants
with the plastic equivalent strain
t
p : =  Ep d  p = Ep .
0
Another ansatz for hardening is plastic work hardening

Y = K T  Ep K : material constant
If K is positive, hardening occurs. If K is negative, it would induce softening.
For the isotropic hardening models, we have the identification Z  Y , which
describes an expansion of the flow surface being equal in all directions of the
stress state (therefore isotropic).
83
Ludwig Prandtl (1875-1953)
84
Endre Reuss (1900-1968)
4.4 Plasticity 285
TR
T
T0 0
isotropic hardening kinematic hardening
In the case of kinematic hardening, the yield surface is shifted rigidly in the
stress space. This can be described by a displacement of its centroid by means of a
tensorial back stress TB as
(4.4.35)  (T , Z  TB) = [equ (T' – TB)] 2 – Y2.
T' – TB is called the effective stress. TB is initially zero and evolves during
yielding. It shall be generally symmetric and deviatoric for pressure-independent
plasticity. The following suggestion for such a hardening rule is due to PRAGER85
(1955)
(4.4.36) TB = c(p) Ep
after which the increment of the back stress is parallel to the plastic strain rate.
The scalar material function c(p) determines the size of the hardening or
softening.
Later ZIEGLER86 (1959) suggested the rule
(4.4.37) TB =  C (T' – TB)
with a material constant C and the plastic parameter  from the flow rule.
All these hardening rules allow for an unbounded growth of the back stress. In
contrast to this somehow unrealistic behaviour, the ansatz of ARMSTRONG-
FREDERICK (1966)
(4.4.38) TB = C [Ep – g(p) TB p]
with some appropriate function g(p) will tend to a saturation value.
For many materials a combination of isotropic and kinematic hardening will be
closest to reality. In this case, the identification Z  {TR , Y} holds. The yield
criterion would then be of the form
85
William Prager (1903-1980)
86
Hans Ziegler (1910-1985)
(4.4.39)  (T, TB , Y) = equ(T' – TB) 2 – Y2

and
  T ,TB , Y   equ  T'  TB    T ,TB , Y 
(4.4.40) = 2equ = –
T T TB
are deviatoric, and
  T ,TB , Y 
(4.4.41) = – 2Y
 Y
This is what we will assume further on.
4.4.4 Consistency Condition
We consider an elastic-plastic material during yielding (equ  Y) with

 HOOKE´s law (4.4.2)
T = C  Ee = C  (E  Ep)
 a yield criterion of the form (4.4.39)
 (T, TR , Y) = equ (T' – TR) 2 – Y2 = 0
 with isotropic hardening (4.4.33)
Y = K(p) p
 and kinematic hardening after ZIEGLER (4.4.37)
TR =  C (T' – TR)
 and an associated flow rule (4.4.29)
  T  TB , Y 
Ep = 
T
We can calculate the plastic parameter  , which determines the size of the plastic
strain increment, by the consistency condition (PRAGER 1949), after which the
stress must remain on the yield surface during yielding
(4.4.42) 0 =  (T – TB , Y)
 
=  (T   TR) + Y 
T  Y
   
=  [C  (E –  )   C (T' – TR)] + K(p)  | |
T T  Y T
4.4 Plasticity 287
which gives

(4.4.43)  =  –1
 C  E
T
with the abbreviation
   
(4.4.44) :=  C  + C  (T' – TR) + 2Y K(p) | |
T T T T
for the consistent flow rule. Within the associated and pressure-independent

isotropic plasticity, is deviatoric after (4.4.40), and so is Ep . With the
T
deviatoric part of the isotropic HOOKE´s law (4.1.17), we have

 C = C   = 2 G 
T T T
and thus for the flow rule (4.4.29)
 
(4.4.45) Ep =  –1 2 G   E.
T T
If we use the HUBER-v. MISES yield criterion and substitute the stress by the
effective stress, then we obtain with (4.4.22) and (4.4.21)
(4.4.46) Ep =  –1 18 G (T' – TR)  (T' – TR)  E
with
(4.4.47)  : = [12 G + 2 C + 26 K(p)] Y2.
Sometimes it is convenient in associated elastoplasticity to use a rate-form for

the stresses. This is with (4.4.29) and (4.4.43)
T = C  Ee = C  (E – Ep)
(4.4.48) = H  E
 
with H:= C– –1
 C  C  .
T T
Here H is the symmetric elasto-plastic tangential stiffness tetrad, which is
generally path-dependent, since it depends on the elastic and plastic deformations,
the hardening variables, and the direction of the strain rate. In the isotropic case,
the deviatoric part of this equations is
 
(4.4.49) T' = 2 G (I –  –1
2G  )  E'
T T
while its spherical part evolves purely elastically. If occasionally no yielding takes
place, then  is zero, the same as the second term, and only the deviatoric
isotropic elastic relation remains
(4.4.50) T' = 2 G E'.

So in the interior of the elastic range and under unloading, the stiffness tetrad H
coincides with the elastic tetrad C . In contrast, under loading at the yield surface,
also the hardening behaviour influences this rate form. H is in such cases a
function not only of the current state variables, but also on the strain increment
E , which governs the decision of whether loading or unloading takes place.
In general, one can use an ansatz for the infinitesimal stiffness tetrad of the form
H(E , T, Z , E) .
An incremental constitutive law of the form
(4.4.51) T = Hh(E , T)  E
is called hypoelastic. Here, the tetrad does not depend on E anymore, so that the
case of classical plasticity is not included as a special case.
A special case of hypoelasticity is the rate-form
(4.4.52) T = C(E)  E
of non-linear elasticity. If C does not depend on E , then the material is linear
elastic.
4.4.5 DRUCKER´s Postulate
If we look at the one-dimensional  –  –diagram of an elastic-plastic material,

then we see that
d d > 0
holds within the elastic range and the (stable) hardening regime, while the
(unstable) softening regime is characterised by the inequality
d d < 0 .
The product of stress increment times strain increment is an indicator of the
material stability of an elastic-plastic material.
As a generalisation of this, DRUCKER87 (1951) has postulated the following
stability condition.
87
Daniel C. Drucker (1918-2001)
4.4 Plasticity 289
DRUCKER´s Postulate
For each cyclic stress process, starting and ending at some arbitrary stress TA in
the current elastic range and only moving from it a little, the integral
(4.4.53)  (T – TA)  dE
TA
shall be non-negative.
If this holds, then it is not possible to extract work under such closed process. If
the elastic law is invertible, the accompanying strain process would also be cyclic.
To exploit this postulate, we decompose the strain increment into its elastic and
plastic parts
dE = dEe + dEp
and also the integral
 (T – TA)  dEe +  (T – TA)  dEp

TA TA
=  T  dEe – TA   dEe +  (T – TA)  dEp

EA EA TA
with EA : = S  TA . The first term on the right-hand side of the equation is zero if
the stress law is hyperelastic, the same as the second term. The last term is only
non-zero if yielding occurs. Since the postulate shall hold for all cyclic stress
processes not going too far beyond the initial yield surface, the integrand must be
non-negative
(4.4.54) (T – TA)  dEp  0
or
(4.4.55) T  dEp  TA  dEp
for all stresses within and on the boundary of the elastic range. In the interior it is
always fulfilled since here dEp  0 holds. The assumption that Ep shall not
grow arbitrarily ensures that TA remains in the initial elastic range.
One can conclude three necessary and sufficient conditions for the postulate to
hold:
1.) the convexity of the yield surface in the stress space
2.) the normality rule: the flow rule must be associated

dEp 
T
convex non-convex
dEp dEp
T TA T
TA
3.) the inequality

dT  dE  0 .
In fact, if we consider a proportional process segment of duration t starting with
TA and EA on the yield surface, and ending with TE and EE after
T(t) = TA + t/t (TE TA) and E(t) = EA + t/t (EE  EA) ,
then, in the beginning, the two increments are
dT = (TE  TA) dt/t and dE = (EE  EA) dt/t .
Finally, this gives for the integral of the postulate
EE t
 (T – TA)  dE =  t/t (TE TA)  (EE  EA) dt/t

EA t 0
t
=  t dt (TE TA)  (EE  EA) /t 2
t 0
2
= t 2/2 (TE TA) /t  (EE  EA) /t = ½ (t/dt) dT  dE ,
from which the inequality follows.
Less strong as DRUCKER´s postulate is ILYUSHIN88´s postulate (1961),
which demands that the stress work for arbitrary cyclic strain processes, which
contain plastic parts, is always positive. From it one can also conclude the
normality rule, but not the convexity of the yield surface. Only if one additionally
assumes that the elastic behaviour does not change due to yielding, then also
convexity is ensured, as has been shown by KRAWIETZ (1981, 1986).
We shall, however, emphasise that both postulates shall not be considered as
universal natural laws, since counter-examples do exist. For many real materials,
normality is violated. An example is the plasticity of granular materials like dry
sand.
88
Alexey Antonovich Ilyushin (1911-1998)
4.4 Plasticity 291
Problem 21. SCHMID´s Law

A tensile part of a face centred cubic (fcc) single crystal with slip systems
 001  {001} with a critical resolved shear stress c = 100 MPa (primary
slip systems) and  011  {011} with c* = 150 MPa (secondary slip
systems) shall be designed. ({} brackets contain crystallographic equivalent
slip planes,   equivalent slip directions). SCHMID´s law (4.4.8) shall be
valid for all slip systems. We assume a uniaxial stress state within the
specimen. Determine the optimal orientation of the specimen with respect
to the tensile direction so that the tensile load can be maximal.
Solution
At first we determine all slip systems. In all cases, the slip directions must
be perpendicular to the slip planes d  n = 0 . This gives six
combinations for both variants. The resolved shear stresses are then
 = T  M after (4.4.7) with the SCHMID tensors M = d  n. Due
to the symmetry of T only the symmetric part of M matters. Since d
and n are crystallographically equivalent in both classes of slip systems,
we obtain pairs of slip systems, whose SCHMID tensors M differ only
by a transposition. Their symmetric parts are then identical, which leads to
pairwise equal resolved shear stresses. So we have to take into account
only 3 slip systems instead of 6. For the primary slip systems we obtain
M12 = e1  e2 , M13 = e1  e3 , M23 = e2  e3
and for the secondary ones
M12* = ½ (e1 + e2)  (e1 – e2)
M13* = ½ (e1 + e3)  (e1 – e3)
M23* = ½ (e2 + e3)  (e2 – e3) .
The factor ½ comes from the normalisation of the vectors d and n .

Next we parameterise the tensile direction indicated by a unit vector z
with the angles  and  on the unit sphere
z = cos  cos  e1 + sin  cos  e2 + sin  e3 .
The state of uniaxial stress can then be described as
T = zz
 cos 2θ cos 2 sin cos cos 2θ cos cosθ sinθ 

 
=   cos 2θ sin 2 sin cosθ sinθ  ei  ej .
 
 symmetric sin 2θ 
 
Now we can calculate the resolved shear stresses in all slip systems
12 = T  M12 = 12 =  cos2 cos  sin 
13 =  cos  cos sin 
23 =  sin  cos  sin 
12* = T  M12* = (11 + 21 – 21 –22) / 2
= (11 – 22) / 2
=  cos 2θ (cos 2 – sin 2 ) / 2
13* =  (cos2 cos2 – sin2 ) / 2

23* =  (sin2 cos2 – sin2 ) / 2
The factors right of  are called SCHMID factors. We will write for
them | fij ( ,  ) | . If we insert the six resolved shear stresses in the
SCHMID law | |   c , we obtain six inequalities of the form
  c / | f12 ( ,  )|
  c / | f13 ( ,  )|
  c / | f23 ( ,  )|
  c* / | f12* ( ,  )|
  c* / | f13* ( ,  )|
  c* / | f23* ( ,  )| .
The most restrictive condition applies for that particular slip system for
which the right-hand side is smallest. So we must find those angles for
which the smallest of the these six right-hand sides becomes maximum.
4.4 Plasticity 293
max {min [ c / | cos 2θ cos sin | ,

 ,
 c / | cos cosθ sinθ | ,

 c / | sin cosθ sinθ | ,
2 c* / | cos 2θ( cos 2 – sin 2 )| ,
2 c* / |cos2 cos2 – sin2 | ,

2 c* / |sin2 cos2 – sin2 | ]} .
It is easiest to find if we implement the inner function min [...] into a
computer algebra system. Because of symmetry, it is sufficient to
investigate this function for  and  in a range of 0 to  / 2.
In the figure, we see the triple symmetry with three global maxima. We can
fix    / 4 and maximise over  . The maximum can be found for
   / 4 and   66.82° , having a value of approximately 390.83 MPa.
This is the maximally possible tensile stress for this orientation of the
single crystal with respect to the load axis. For comparison, the minimum
value is 200 MPa if we pull in  110  – direction. For a load in  111  –
direction (like, e.g.,   45°,   35.12°) we would get 300 MPa.
0< </2 0< </2
Problem 22. J2Plasticity

A spherical shell (inner radius ri , outer radius ra) is submitted to an
internal pressure pi . The outer pressure is negligible. The material of the
shell is isotropic elastic with LAMÉ´s constants  and µ and plastic after
the J2criterion with a yield stress Y . For which internal pressure does
yielding initiate? Determine the radius rp at the limit between the elastic
and plastic parts as a function of pi . For which pi does the shell yield
completely?
Solution
We introduce spherical coordinates
x1 = r cos  cos 
x2 = r cos  sin 
x3 = r sin 
We will need the divergence and the deformator in these coordinates. The
corresponding vector basis is
er = (xi ei) / r e = (xi ei) /  e = (xi ei) / 
which gives after normalisation the ONB
er = cos  cos  e1 + cos  sin  e2 + sin  e3
(P22.1) e = – sin  e1 + cos e2
e = – sin  cos  e1 – sin  sin  e2 + cos  e3 .
The derivatives of these are zero except
er ,  = cos  e
er ,  = e
(P22.2) e ,  = – cos  er + sin  e
e ,  = – sin  e
e ,  = – er
We obtain the nabla operator by solving P22.1 for e(1,2,3) and substituting
ei by e(r, , ) in (2.2.20). The partial derivative must be determined after
the chain rule
       r          
= + + .
x i r xi  xi  xi
The partial derivatives (r,  ,  ) /  xi are the components of the inverse

of the JACOBI matrix
Jij =  xi / (r,  ,  ) .
We use for the calculation a computer algebra system (MATHEMATICA).
4.4 Plasticity 295
Remove["Global`*"]; (* clear cache *)

(* Invert normalized basis for spherical coordinates *)
sol=FullSimplify[Solve[{er==Cos[t]*Cos[p]*ex+Cos[t]*Sin[p]*ey+
Sin[t]*ez,ep==-Sin[p]*ex+Cos[p]*ey,et==-Sin[t]*Cos[p]*ex-
Sin[t]*Sin[p]*ey+Cos[t]*ez},{ex,ey,ez}]];
ex=ex/.sol[[1]];
ey=ey/.sol[[1]];
ez=ez/.sol[[1]];
(* the change of coordinates *)
x=r*Cos[t]*Cos[p];
y=r*Cos[t]*Sin[p];
z=r*Sin[t];
(* determine Jacobian and its inverse *)
J=FullSimplify[{{D[x ,r],D[x ,p],D[ x,t]},{D[y ,r],D[y ,p],
D[y ,t]},{D[z ,r],D[z ,p],D[ z,t] }}];
Jinv=FullSimplify[Inverse[J]];
(* multiply partial derivatives for r,phi,theta as vector from
the left with the inverse of the Jacobian to determine the
partial derivatives for x,y,z *)
partdiffs=FullSimplify[{ddr,ddp,ddt}.Jinv];
(* now we have the base vectors ex, ey, ez in terms of er,
ephi, etheta and the partial derivatives w.r.t x, y, z in terms
of r, phi, theta: we can plug everything into the cartesian
nabla operator and summarize *)
FullSimplify[
ex*partdiffs[[1]]+ey*partdiffs[[2]]+ez*partdiffs[[3]]]
The result is
(ddt et+ddr er r+ddp ep Sec[t])/r,
that is
 = () / r er + 1/r () /  e + 1 / (r cos ) () /  e .
Because of the spherical symmetry of the problem, we have the mean
stresses rad = rr and tang =  =  , so that all shear stresses with
respect to this basis vanish. With nabla and the rules P22.2 for the
derivatives, we can determine the divergence of the stress tensor
div T = [rr , r + (2rr –  – ) / r] er + [ ,  / (r cos)] e
+ [ ,  + tan ( – )] / r e .
Because of the spherical symmetry, the derivatives with respect to  and
 vanish, and  =  = tang , so that the only non-trivial equilibrium
condition in the absence of body forces is
(P22.3) rad , r + 2(rad – tang) / r = 0 .
We assume a displacement field with spherical symmetry
u = u(r) er .
With E = Def u we obtain
E = u , r er  er + u / r (e  e + e  e) .
We first investigate the elastic part. We insert the strains into the elastic
law (4.1.17) and calculate the stress tensor. We put the stresses into P22.3
and obtain (in analogy to Problem 13) an ODE for u
0 = ( + 2µ) (u , rr + 2u , r / r – 2u / r2) .
Its solution is
(P22.4) u = a r + b / r2.
The stresses are then
rad = a (3 + 2µ) – 4bµ / r3
tang = a (3 + 2µ) + 2bµ / r3.
The boundary conditions are rad (ra) = 0 and rad (ri) = – pi . This gives
a = pi ri3 / [(3 + 2µ) (ra3– ri3)]
b = pi ri3 ra3 / [4µ (ra3– ri3)] .
Because of pi , ra , ri ,  , µ > 0 and ra > ri we have a, b > 0 , so that
tang > rad and u > 0 . This is what we expect, since a sphere under
internal pressure becomes larger.
We will next determine the load and loci for which the HUBER-v. MISES
yield criterion is fulfilled. For the shear stresses being zero, J2 is after
(4.4.20)
J2 = 1/3 (rad – tang)2.
After (4.4.16) the limit is
3 J2 = Y2
so that we have Y = |rad – tang | . We already found out that tang > rad
so that
Y = tang – rad = 6 b µ / r3.
The TRESCA criterion would give the same equation if Y is calibrated in
a tensile test. The critical radius is
rpl = (6 b µ / Y)1/3 = k pi1/3
with
k = ra ri {3 / [2 Y (ra3–ri3)]}1/3.
rpl grows monotonously for increasing internal pressure starting from 0 .
The elastic limit case is rplast  ri . This leads to
(P22.5) pi crit el = 2Y (1 – ri3 / ra3) / 3.
We now consider the case when pi > pi crit el . Obviously there will be
yielding for ri < r < rpl , while for rpl < r < ra no yielding occurs. The latter
region has been already completely determined. In the other part the local
equilibrium condition P22.3 must hold. This leads with Y = tang – rad
to
4.4 Plasticity 297
0 = rad , r – 2Y / r
and an integration over r to
(P22.6) rad = 2Y ln r + c
with some constant c which can be determined by the boundary condition.
Let us first consider the limit case rpl = ra . In analogy to the elastic case,
we can adopt the last equations to the boundary conditions rad = 0 at
r  ra and rad = –pi at r  ri , which gives
0 = 2Y ln ra + c
– pi = 2Y ln ri + c .
Elimination of c gives the critical internal pressure
(P22.7) pi crit pl = 2Y ln (ra / ri)
for which the hollow sphere will be completely plastically deformed. The
plastic buffer is defined as
R = pi crit pl / pi crit el = 3 ln(ra / ri) / [1 – (ri / ra)3] .
After considering the extreme cases for rpl  ri and rpl  ra we look at
the case ri < rpl < ra . The strains will be decomposed into elastic and
plastic parts after (4.4.1). Ep is coaxial to the stress tensor and deviatoric,
so that there is only one free parameter  p
Ep =  p er  er – ½ p (e  e + e  e) .
The stress state can be determined by the elastic law
(P22.8) rad =  (u , r + 2u / r) +2 µ (u , r – p)
(P22.9) tang =  (u , r + 2u / r) +2 µ (u / r + p /2) .
If we set this equal to P22.6 and P22.8, we get the first ODE for the
unknown functions  p and u
(P22.10)  (u , r + 2u / r) +2 µ (u , r – p) = 2Y ln r + c .
Finally we have to apply the consistency condition Y = tang – rad within
the entire plastic region
(P22.11) Y = 2 µ (u / r + p / 2 – u , r +  p) .
This equation can be brought into the form
 p = Y / (3µ) + 2 (u , r – u / r) / 3
and inserted into P22.10, which gives an ODE for u
(3 +2 µ) (u , r + 2u / r) / 3 = Y (2 ln r + 2 / 3) + c .
Its solution is
(P22.12) u = d / r2 + r (c + 2Y ln r) / (3 +2µ)

which gives further
(P22.13)  p = Y ( +2µ) / (3µ +2µ2) – 2d / r3.
We finally have two functions for the displacement field, P22.12 for ri < r
< rplast and P22.4 for rplast < r < ra with altogether five unknowns a, b, c, d,
and rplast . So we need five boundary and transition conditions. These are
the boundary conditions
radEP (ri) = – pi
radE (ra) = 0
already needed in the purely elastic case, the kinematic compatibility
condition
uEP(rplast) = uE(rplast) ,
the equilibrium condition at the transition zone between the two regions
radEP (rplast) = radE(rplast) ,
and the condition
 p(rplast) = 0
at rplast . The latter condition is due to the fact that the plastic strain is
initially zero.
We used the indices EP for variables in the elastoplastic region r < rplast ,
and E for the purely elastic region rplast < r . Together with the solutions
for uE(r) , uEP(r) , and  p(r) (P22.4, P22.12 and P22.13) these give
(P22.14) c + 2 Y ln ri = – pi
(P22.15) a (3 + 2 µ) = 4bµ / ra2
(P22.16) (d – b) / rp2 + rp (c + 2 Y ln rp) / (3 + 2 µ) = a rp
(P22.17) c + 2 Y ln rp = a (3 + 2 µ) – 4bµ / rp3
(P22.18) – 2 d / rp3 + Y ( / µ + 2) / (3 + 2 µ) = 0 .
This is a system of linear equations for a, b, c, d. P22.14 can be directly
solved for c
c = – pi – 2 Y ln ri .
d results from P22.18
d = rp3Y ( / µ + 2) / (6 + 4 µ) .
We insert c and d into P22.16, which gives together with P22.15 a linear
system for a and b with solutions
4.4 Plasticity 299
a = 2 rp3 (Y( + 2µ + 4 µ ln (rp / ri)) – 2µ pi) /

{(3 + 2 µ) [(3 + 2 µ) ra3+4µ rp3]}
b = rp3 ra3 (Y ( + 2µ + 4 µ ln (rp / ri)) – 2µ pi) /
{2 µ[(3 + 2 µ) ra3+4µ rp3]} .
We replace a, b, c, and d in P22.17
0 = ( + 2µ) [3 pi ra3 – 2Y (ra3 – rp3) – 6Y ra3 ln (rp / ri)] /
[(3 + 2 µ) ra3+ 4µ rp3] .
This can not be explicitly solved for rp , but we can determine pi (rp)
pi (rp) = Y [2 / 3 – 2rp3 / (3ra3) + 2 ln(rp / ri)] .
We also find the limit cases rp  ri (P22.5) and rp  ra (P22.7) again. The
latter function has been depicted for the parameters ra  2 ri and
Y = 400 MPa.
Internal pressure pi vs. rp
The limit cases are: pi = 233.3 MPa at the beginning of yielding in the
interior, and pi = 554.5 MPa for full yielding. The plastic buffer is
554.5/233.3  2.37. The largest plastic strain is  p  – 0.019 , occurring at
the inner radius at full yielding.
Finally we look at the states of strain and stress. For ri  ra / 2 = 100 mm,
E = 200 GPa,  = 0.3, Y = 400 MPa, and pi = 500 MPa we obtain the
following curves.
u(r)
equ = tang – rad
tang
rad
We depicted the radial displacement in mm (above) and the equivalent

stress equ , the radial and the tangential stresses (below) over the radius in
mm. For the stresses we clearly see the transition from elastic to plastic at
rplast = 158 mm.
4.4 Plasticity 301
4.4.6 Thermoplasticity
Literature
Bertram, A., Krawietz, A.: On the introduction of linear thermoplasticity. Acta
Mechanica 223,10, 2257-2268, (2012)
Farren, W. S., Taylor, G. I.: The heat developed during plastic extension of
metals. Proc. Royal Soc. London, Series A. 107, 422-451 (1925)
Lemaitre, J., Chaboche, J.-L.: Mechanics of Solid Materials. Cambridge
University Press (1990)
Maugin, G. A.: The Thermomechanics of Plasticity and Fracture. Cambridge
University Press (1992)
Taylor, G. I., Quinney, H.: The latent heat remaining in a metal after cold
working. Proc. Roy. Soc. London A163, 157-181 (1937)
When plastic yielding occurs in a material, a dominant part of the mechanical

work will be dissipated. This has been demonstrated in the historical experiments
by FARREN/ TAYLOR (1925) and TAYLOR/ QUINNEY (1937). Dissipation
means a transformation into heat or consumed for changes of the microstructure.
The dissipation can lead to large local changes of the temperature, which in turn
effect the mechanical behaviour. For these reasons it is necessary to embed the
plasticity theory into a thermodynamical format. This would also offer the
opportunity to study the restrictions imposed by the 2nd law of thermodynamics in
the form of the CLAUSIUS-DUHEM inequality on the constitutive model, thus
assuring a thermodynamically consistent plasticity theory.
In contrast to thermoelasticity, in thermoplasticity we will need, beside the
thermo-kinematical state (temperature and deformation), other internal variables.
We will assume that all internal variables are determined by the thermo-
kinematical process in a path-dependent manner. The internal variables in
elastoplasticity are the plastic deformations Ep = E  Ee after (4.4.1) and the
vector of the hardening variables Z (here noted as a 2nd-order tensor).
The set of constitutive equations for a thermoplastic material is assumed to be of
the form
T = T (Ee ,  , g , Ep , Z) for the stresses
(4.4.56) q = q (Ee ,  , g , Ep , Z) for the heat flux
 =  (Ee ,  , g , Ep , Z) for the entropy
 =  (Ee ,  , g , Ep , Z) for the internal energy
or  =  (Ee ,  , g , Ep , Z) for the free energy
We assume that the material possesses thermoelastic ranges with a corresponding
(4.4.57)  yield criterion  (Ee ,  , Ep , Z) .

The elastic ranges are changed if and only if
(4.4.58)  the yield condition  (Ee ,  , Ep , Z) = 0
 
(4.4.59)  and the loading condition  E +    > 0
Ee θ
are simultaneously fulfilled. The yield criterion can be transformed into the stress
space by the elastic law (4.4.561) if required. A typical example for such a yield
criterion is the generalisation of the HUBER-v. MISES yield criterion (4.4.21)
 (T ,  F) = 3/2 T'  T' –  Y ()2
with a temperature-dependent yield stress  Y () .
If both conditions are not simultaneously fulfilled, yielding cannot occur such
that Ep  0 and Z  0 . Otherwise, yielding must occur. Therefore the evolution
equations for Ep and Z contain a switcher to switch from elastic to plastic
events, or vice versa.
We will next consider the consequences of the CDI for the thermoelastic
constitutive equations. With the above ansatz (4.4.565) for the free energy we get
from the CDI (4.2.9) with the decomposition (4.4.1)
1 qg
0   (Ee ,  , g , Ep , Z) +   – T  E +
 ρ
     
(4.4.60) =  Ee +  +  g +  Ep +  Z
Ee θ g E p Z
1 qg
+   – T  (Ee + Ep) + .
 ρ
Let us first consider an elastic event with Ep  0 and Z  0 below the yield
limit ( (Ee ,  , Ep , Z) < 0). From the inequality only the reversible parts remain
in analogy to (4.2.16)
 T    q  g
(4.4.61) 0  ( – )  Ee + ( + )  + g + .
Ee  θ g ρ
We can again conclude the necessary and sufficient conditions of thermoelasticity
like in (4.2.17-19), namely the independence of the free energy of the temperature
gradient  =  (Ee ,  , Ep , Z) and the potential relations

(4.4.62) T =  = T (Ee ,  , Ep , Z)
Ee

(4.4.63)  = – =  (Ee ,  , Ep , Z)
θ
4.4 Plasticity 303
and the heat conduction inequality

(4.4.64) q (Ee ,  , g , Ep , Z)  g  0 .
Because of continuity, these relations must also hold on the yield surface and
during yielding. In addition to them, the irreversible parts of (4.4.60) must fulfil
the residual dissipation inequality (RDI) during yielding
T   qg
0  (– + )  Ep +  Z + . (RDI)
 E p Z ρ
This inequality must hold for all values of the temperature gradient g . Since only
the last term depends on g , we can conclude that RDI holds for arbitrary
processes only if the heat flux inequality (4.4.64) and, independent of it, the RDI
in the form
T  
(4.4.65) m : = (  )  Ep   Z  0 (RDI)
 E p Z

hold. The term  can be interpreted as a back stress (see (4.4.35)), and the
E p
term in brackets as the effective stress. The restrictions of the RDI will be studied
later.
For many materials and, in particular, for metals it is a both theoretically and
experimentally well-substantiated finding that the thermoelastic behaviour within
the elastic ranges hardly alters, even under very large plastic deformations. This is
why it is usually assumed to be invariant.
Assumption:
In all thermoelastic ranges, the thermoelastic behaviour is identical.
We make this assumption more precise. We assume that in all elastic states, for
given body forces b and given heat supply Q , all measurable quantities depend
only on the elastic strains and the temperature, but not on the plastic internal
variables Ep and Z . Measurable means that they occur in the balance laws C1
and 1LT like
 the stresses T
 the rate of the internal energy 
 the heat flux q .
In contrast, the free energy or the entropy are not considered as measurable
quantities.
For elastic events, the mechanical dissipation (4.4.65) is zero, so that we have by
(4.2.10)
(4.4.66)  r  div q =   .
With (4.2.18) is the right-hand side for elastic events
  
   =   (  Ee + ),
Ee θ
which must be independent of Ep and Z after our assumption. This is the case if
and only if the following decomposition of the entropy in elastic and a plastic
parts exists
(4.4.67)  = e (Ee , ) + p (Ep , Z) .
For the stresses, the assumption means that they can neither depend on Ep nor
Z , so that T = T (Ee , ) remains, and with (4.4.62) we get

(4.4.68) T (Ee , ) =  .
Ee
With (4.4.63) and (4.4.67) this leads to the split of the entropy

(4.4.69) e (Ee , ) +p (Ep , Z) = –
θ
and of the free energy
(4.4.70)  (Ee ,  , Ep , Z) = e (Ee , )   p (Ep , Z) + p (Ep , Z)
 e  Ee ,
with e (Ee , ) = –
θ
or e (Ee , ) : =   e (Ee , ) d
with an up-to-now arbitrary function p of the plastic variables. The internal

energy can also be split after (4.2.8)
(4.4.71)  (Ee ,  , Ep , Z) = e (Ee , ) + p (Ep , Z)
with e (Ee , ) : = e (Ee , ) +  e (Ee , ) .
Neither can the heat flux depend on Ep or Z after our assumption, which leads
to the reduced version of (4.4.562)
(4.4.72) q = q (Ee ,  , g) .
We state these findings in the following
4.4 Plasticity 305
Theorem. The assumption of identical thermoelastic behaviour in all

thermoelastic ranges is fulfilled for thermoplastic materials, if and only if the
following representations hold for the
 e  Ee ,
 entropy  = – + p (Ep , Z)
θ
 e  Ee ,
 internal energy  = e (Ee , ) –  + p (Ep , Z)
θ
 free energy  = e (Ee , )   p (Ep , Z) + p (Ep , Z)
 heat flux q = q (Ee ,  , g)
 e  Ee , 
 stresses T =  .
Ee
Accordingly, the following constitutive functions completely constitute a

thermoplastic material:
 the elastic free energy e (Ee , )
 the plastic entropy p (Ep , Z)
 the plastic internal energy p (Ep , Z)
 the heat flux q (Ee ,  , g)
 the yield criterion  (Ee ,  , Ep , Z)
completed by the flow rule and the hardening rule(s).
With (4.4.70), the RDI (4.4.65) can be brought into the form
 p  p  p  p
(4.4.73) 0  (T –  + )  Ep + ( + )  Z
E p E p Z Z
= T  Ep +   p (Ep , Z)   p (Ep , Z). (RDI)

If we want to determine the change of the temperature of a material point, we use
the 1LT (4.2.5) together with (4.4.71)
(4.4.74)  r – div q
T T T
=    E = e(Ee , )    Ee + p (Ep , H)   Ep.
  
According to this, we can split the local heat supply
 r – div q = qe + q p
into the elastic part with (4.4.71)
T
(4.4.75) qe(Ee ,  , Ee, ) : = e(Ee , )    Ee =  e(Ee , )

M
= –  Ee + c 

because of (4.4.66) with
 the specific heat in analogy to (4.2.25) with (4.4.70)
e  2 e  Ee , 
(4.4.76) c (Ee , ) : =  = – 
θ θ 2
 and the stress-temperature tensor
 e
(4.4.77) M(Ee , ) : = – 
E e
and into a plastic part

T
(4.4.78) q p : = p (Ep , H)   Ep.

(4.4.75) can be solved after
M
(4.4.79) c  = q – qp +   Ee.

By this balance equation, one can determine the current temperature by integration
along the process. Changes of the temperature are thus due to
1.) the heat supply from the surroundings,
2.) the heat – qp generated by plastic yielding and hardening, and
3.) thermoelastic changes after the last term in (4.4.79).
By assuming that the stresses and the heat supply are measurable during elastic
processes, we can determine the rates of the internal energy and the entropy by
(4.4.75). e and e themselves are only determined up to an additive constant,
which is not detectable in any experiment, and therefore does not have any
influence on the thermomechanical behavior. If we substitute e by e + c1 and
e by e + c2 , then we must substitute the elastic part of the free energy by
e + c1 – c2  after (4.4.71). The elastic part is thus only unique up to an additive
constant and a linear function of the temperature.
If the heat production due to plastic yielding qp is also measurable, then the rate
of the plastic energy is determined by (4.4.78), while p itself is unique only up
to an additive constant.
4.4 Plasticity 307
We will now make an ansatz for the rate-independent evolution equations of the
plastic variables, namely for
(4.4.80) the flow rule Ep =  P(Ee ,  , Ep , Z)
(4.4.81) the hardening rule Z =  Z(Ee ,  , Ep , Z)
with some non-negative plastic parameter  . It is zero if and only if no yielding
takes place. So the KUHN-TUCKER conditions are always fulfilled
(4.4.82)  =0 with   0 and   0.
During yielding, the plastic parameter can be determined by the consistency
condition in analogy to (4.4.42)
0 =  (Ee ,  , Ep , Z)
    
(4.4.83) =  Ee +  +  Ep +  Z
Ee θ E p Z
  
=  Ee + 
Ee θ
 
+   P(Ee ,  , Ep , Z) +   Z(Ee ,  , Ep , Z)
E p Z
which can be solved for

  
(4.4.84)  =  –1
(  Ee + )
Ee θ

with  (Ee ,  , Ep , Z) : = –  P(Ee ,  , Ep , Z)
E p

  Z(Ee ,  , Ep , Z) .
Z
Because of the loading condition (4.4.59),  is positive during yielding. After
(4.4.82)  alone is also positive, so that  must also be positive. This is a
restriction for the functions P and Z after (4.4.84).
Another restriction comes from the dissipation inequality (4.4.73) for a positive 
 p  p
(4.4.85) (T –  + )  P(Ee ,  , Ep , Z)
E p E p
 p  p
+ ( + )  Z(Ee ,  , Ep , Z)  0 . (RDI)
Z Z
Its consequences will be demonstrated by the following example.
Example
K C
Ep Ee
We generalise the shown PRAGER model with two springs and a frictional
element to three dimensions. The strains in the spring C are Ee , and the strains
in the other spring K and the parallel frictional element are called Ep .
For the elastic part of the free energy, we use the usual ansatz from linear
thermoelasticity as a quadratic form similar to (4.2.36)
1 1 c
e (Ee , ) = Ee  C  Ee +  M  Ee – 2
2  2θ0
with an isotropic or anisotropic positive definite stiffness tetrad C of the spring

C , a constant symmetric stress-temperature tensor M , a constant specific heat
c , a reference temperature 0 , and the temperature difference  : =  – 0 . If
we want to guarantee that c is in fact the specific heat, (4.4.76) must hold, which
is not the case for the foregoing ansatz. We use instead
1 1 
(4.4.86) e (Ee , ) = Ee  C  Ee +  M  Ee + c ( –  ln ).
2  θ0
The difference between the two functions, however, is negligible for small  .
We obtain with (4.4.68)
(4.4.87) T = C  Ee +  M
and with (4.4.69)
1 
e (Ee , ) = – M  Ee + c ln
 θ0
and after (4.4.71)
1 1
e (Ee , ) = Ee  C  Ee – 0 M  Ee + c  .
2 
We will next consider the plastic parts of the constitutive equations of the model.
The stress in the spring K stands for the back stress
(4.4.88) TB = K  Ep
with some elasticity tetrade K assumed to be positive definite. TB is assumed to
be deviatoric, the same as Ep .
We define the specific plastic work in the time interval [t0 , t1] as the work of the
frictional element, on which the effective stress T – TB acts,
4.4 Plasticity 309
t1
T  TB T  TB
(4.4.89) wp : =   Ep dt  wp =  Ep.
t0
 
We identify this plastic work as a (scalar) hardening variable Z .

If we use the HUBER-v. MISES yield criterion (J2-theory) (4.4.17) in the form
 = (T' – TB)  (T' – TB) – Y (wp , )2
with the stress deviator T', then the associated flow rule is like (4.4.30)
(4.4.90) Ep =  (T' – TB) .
The consistency condition (4.4.83) demands with (4.4.87), (4.4.88), (4.4.89)
    
0 =  =  Ee +  +  Ep +  wp
Ee θ E p w p
= 2 (T' – TB)  (C  Ee + M  – K  Ep )
 B T  TB  B 
– 2 Y (  Ep + )
 wp  θ
or
 B 
(T' – TB)  (C  E + M ) – F 
θ
 B
= (T' – TB)  (C + K + F I )  Ep
  wp
 B
= (T' – TB)  (C + K + F I )   (T' – TR) .
  wp
During yielding, the left-hand side is positive due to the loading condition
(4.4.59). If the material hardens, we have
 Y
 0
 wp
and the tetrad on the right-hand side of the above equations is positive definite, so
that the condition  > 0 is fulfilled. Please note that (4.4.89) and (4.4.90) imply
that the plastic power and the plastic work are positive during yielding
(4.4.91) wp > 0  wp > 0 .
This is a consequence of the loading condition, and not of the dissipation
inequality, as it is sometimes stated.
The plastic work is only partly transformed into heat. Another part can also be
consumed for changes of the microstructure like phase changes, bonding or
debonding, etc. After the seminal experiments of FARREN/ TAYLOR (1925) and
TAYLOR/ QUINNEY (1937), for metals the part of the plastic work that is turned
into heat lies typically in the order of 85-95 %. Such a percentage can be
implemented into our theory by a TAYLOR-QUINNEY factor  with values
between 0 and 1 in most cases, which we consider here as a material constant.
There can also be energy stored in the spring K . We make the following ansatz
for the plastic part of the internal energy
1
p (Ep , wp) = Ep  K  Ep + (1 –  ) wp .
2
The heat production caused by plastic yielding is then after (4.4.78), (4.4.88), and
(4.4.91) in fact
T
(4.4.92)  qp =  Ep  p (Ep , wp)

T  TB
=  Ep – (1 –  ) wp =  wp > 0 .

The residual dissipation inequality (4.4.73) becomes with (4.4.78) and (4.4.92)
(4.4.93) – q p +  p =  wp +  p  0 (RDI)
and is a restriction to the plastic entropy. This can be fulfilled by the trivial ansatz
p  0 . However, also the ansatz
p (wp)  γ wp
with a positive real constant γ would be a possible choice to satisfy the
dissipation inequality
( + γ ) wp  0 . (RDI)
Further concretisations for the plastic entropy based on measurements are not
possible. So we clearly see that the plastic entropy remains rather undetermined,
the same as the plastic free energy. In order to prove the thermodynamic
consistency, it is sufficient to find at least one plastic entropy which satisfies the
RDI. Its choice, however, does not have any influence on the measurable
thermomechanical behavior.
By (4.4.70) we obtain for the complete free energy of our model
1 1 
(78)  = Ee  C  Ee +  M  Ee + c ( –  ln )
2  θ0
1
+ Ep  K  Ep + (1 –    γ) wp .
2
Together with FOURIER´s law for the heat flux (2.2.6), all constitutive equations
for the model have been suggested.
4.4 Plasticity 311
For this model, yielding against the stress is possible, i.e., T  Ep can be
negative without contradiction with the dissipation inequality. There are
experimental verifications for this effect for materials with a strong
BAUSCHINGER effect. In the literature, the positive semi-definiteness of this
expression is sometimes declared as a sufficient condition for the dissipation
inequality to hold. However, we see that this is neither sufficient nor necessary for
the second law to hold.
The End
5 INDEX
acceleration................................. 124 CASIMIR-ONSAGER´s reciprocity
acoustic tensor ............................ 246 ........................................ 105, 256
AIRY´s stress function ....... 157, 160 CAUCHY´s equation of motion 144,
AMONTONS friction ................... 274 146
amplitude vector ......................... 244 CAUCHY´s stress tensor............ 143
angular momentum ....................... 83 CDI ..................................... 253, 302
angular velocity ............................ 85 centre of mass ............................... 82
anisotropic .................................. 172 CHABOCHE model ..................... 42
anti(sym)metric tensor .................. 57 CHOI-KREMPL model ................ 42
ARMSTRONG-FREDERICK´s CLAPEYRON´s theorem ........... 201
hardening rule ......................... 285 CLAUSIUS-DUHEM inequality
associated flow rule .................... 282 ........................................ 253, 302
auxetic ........................................ 197 CLAUSIUS-PLANCK inequality
axial vector ........................... 57, 108 ................................................ 251
axis of principal stress ................ 146 coaxiality .................................... 266
back stress .................... 37, 285, 303 collinear dyad ............................... 49
balance of angular momentum ..... 83 compatibility............................... 134
balance of energy ............... 250, 259 complementary energy ................... 8
balance of linear momentum83, 141, complementary power .................... 5
144 complementary work ...................... 5
balance of power ........................ 150 complete dyad .............................. 51
BAUSCHINGER effect ....... 36, 311 complex compliance ..................... 25
BELTRAMI ............... 135, 156, 199 complex stiffness .......................... 25
BELTRAMI´s stress function .... 155 complex viscosity ......................... 25
BELTRAMI-SCHAEFER stress compliance tensor ....................... 168
function .................................. 156 component of a tensor .................. 52
bending ......................................... 89 composition of tensors.................. 53
BERNOULLI´s hypothesis .......... 89 compression........................ 128, 182
BETTI´s reciprocal theorem ....... 201 compression energy .................... 196
bilinear form ................................. 56 conservation of energy ................. 88
BINGHAM model ........................ 41 conservation of mass .................. 124
BODNER-PARTOM model......... 42 consistency condition 282, 286, 297,
body force................................... 140 307, 309
BOLTZMANN´s axiom ............. 146 consistency parameter ................ 283
BOLTZMANN´s superposition contact force ............................... 141
principle.................... 25, 266, 270 contraction ........................ 54, 93, 94
boundary condition ............. 200, 261 convective rate............................ 124
boundary value problem ............. 200 convexity .................................... 289
BOUSSINESQ ........... 220, 221, 229 coordinate, material .................... 122
BRUHNS...................................... 42 coordinate, spatial....................... 122
bulk modulus .............................. 182 COULOMB .................... 34, 39, 208
BURGERS model .......... 12, 22, 268 COULOMB friction ................... 274
creep ....................................... 15, 23

DOI 10.1007/978-3-319-19566-7
Index 313
creep function ...............................26 DUHAMEL-NEUMANN equality

creep time .....................................18 ................................................256
cross-hardening ..........................273 dummy index................................44
cross-product ..........................48, 59 dyad ........................................49, 93
crystal ................................. 178, 272 dyadic product ..............................49
cubic ........................... 173, 175, 178 effective stress ..............................29
curl..............................................108 eigenbasis .....................................70
cylindrical coordinates................113 eigendirection ...............................65
D'ALEMBERT...........................153 eigenspace ....................................69
damage parameter.........................30 eigentensor.............................. 98, 99
DE VEUBEKE ...........................215 eigenvalue...............................65, 98
deformation gradient ..................125 eigenvector ...................................65
deformation method .....................40 elastic..........................................166
deformator ..................................107 elastic energy..........................8, 194
density ........................................124 elastic limit ...................................38
determinant...................................66 elasticity ................................. 6, 165
determinant rule............................67 elimination of displacements ......199
determinism ........................164, 252 elimination of stresses ................198
deviator................................. 67, 146 ellipsoid ........................................88
diagonal form ...............................70 endurance strength........................32
differential ..................................104 energy balance ....................250, 259
differential type ............................24 energy ellipsoid ............................88
dilatation.............................128, 182 energy, complementary .................. 8
direct notation.........................64, 74 energy, elastic .........................8, 194
DIRICHLET conditions .............200 energy, free................................. 304
DISCHINGER law .......................16 energy, internal ...................249, 304
displacement...............................122 energy, kinetic ......................87, 150
displacement equations...............198 engineering stress ........................... 4
displacement function.................218 enthalpy ......................................252
displacement function of entropy................................249, 310
BOUSSINESQ-PAPKOVITSCH- entropy flux ................................250
NEUBER ................................221 entropy source ............................250
displacement function of equilibrium condition .................262
BOUSSINESQ-SOMIGLIANA- equilibrium of forces ..................144
GALERKIN............................220 equivalent stress .........................276
displacement gradient.................125 EULER ................................. 83, 126
displacement-strain relation........126 EULER´s equations of gyroscopes
dissipation...................................251 ..................................................87
dissipation inequality.. 251, 303, 305 EULER´s velocity formula...........85
distorsion ....................................129 EULERian representation...........123
distortional energy......................196 EURATOM model .......................42
divergence ..................................107 existence theorem .......................261
divergence theorem ....................118 fading memory .......................14, 26
double contraction ........................96 fatigue...........................................31
DRUCKER´s postulate...............289 FEDOROV´s theorem ................247
DSR ....................................126, 218 FIBONACCI series ......................80
dual vector basis ...........................47 final work ....................................... 6
314 Index
finite element method . 149, 191, 243 heat supply ................................. 249
first law of thermodynamics.......250 heat supply equation...................258
flow rule ............................. 272, 307 HEAVISIDE function ..................25
forces, generalised ...................... 152 HELMHOLTZ energy................ 251
forces, virtual.............................. 154 HELMHOLTZ´ displacement
FÖRSTER´s resonance method..190 function ..................................219
FOURIER´s law 105, 257, 259, 262, HELMHOLTZ´ representation
310 theorem................................... 219
FOURIER´s theorem..................250 HENCKY .....................................40
free energy.................. 251, 304, 308 HENCKY strain ............................. 2
free index......................................45 HERTZean pressure ...................229
FREIBERGER´s displacement hexagonal ................................... 179
function ..................................223 HILL........................................... 278
frequency....................................244 history functional.......................... 27
FRESNEL-HADAMARD condition HOLLOMAN law ........................ 40
................................................ 246 HOOKE element ............................ 8
friction ........................................208 HOOKE´s law .................. 7, 90, 167
GALERKIN ............................... 220 HU .............................................. 215
GAROFALO law .........................27 HUBER-v. MISES yield criterion
GAUSS´ integral transformation 118 ................................ 276, 302, 309
generalised forces ....................... 152 hyperelastic ................................ 195
geometrically linear.................... 128 hypoelasticity ............................. 288
GIBBS energy ............................ 252 hysteresis ......................................19
GIBBS equation ......................... 253 identity tensor......................... 54, 98
gradient ............................... 104, 106 ILYUSHIN´s postulate............... 290
gradient material............................. 7 incompatibility ........................... 109
GREEN elastic ........................... 195 incompressible................................ 4
GREEN´s strain tensor ............... 126 indefinite .............................. 59, 102
group ..........................................171 inertia............................................85
gyroscope ............................... 82, 87 inertia of area................................91
HADAMARD ............................246 inner product .................... 47, 62, 96
HAIGH-WESTERGAARD plane instantaneous elasticity.... 10, 15, 21,
................................................ 279 22
half space.................................... 223 internal energy.................... 249, 304
HAMILTON´s principle............. 151 internal variable............................22
hardening...................................... 35 invariants of a tensor ....................66
hardening rule..................... 272, 307 inverse tensor................................ 55
hardening, isotropic .............. 37, 284 isotropic.............................. 172, 259
hardening, kinematic ............ 37, 285 isotropic hardening ............... 37, 284
harmonic vibration .......................19 isotropy............................... 176, 179
harmonic wave ...........................245 J2-theory .....................................276
HART-MILLER model................42 JENKIN model............................. 39
heat capacity...............................254 KACHANOV damage..................29
heat conduction coefficient ........260 KACHANOV-RABOTNOV
heat conduction tensor................256 damage ..................................... 30
heat flux...................... 105, 249, 259 KELVIN model ...................... 12, 17
heat source..................................249 KELVIN module ........................ 184
Index 315
KIMBALL-LOVELL law ............27 maximum of potential energy.....214

kinematic hardening .............37, 285 MAXWELL model.................12, 13
kinetic energy .......................87, 150 MAXWELL relation ..................253
KIRCHHOFF .............................202 MAXWELL´s stress function.....155
KORN´s displacement function . 223 McCULLAGH ellipsoid ...............89
KRONECKER symbol.................45 minimum of potential energy .....211
KUHN-TUCKER conditions.....283, minimum of stress energy ..........213
307 MOHR´s yield criterion..............274
LAGRANGE ..............................153 moment of inertia .........................85
LAGRANGEan presentation......123 momentum, angular......................83
LAMÉ.........................................201 momentum, linear.........................83
LAMÉ´s constants ......................184 monoclinic ..........................173, 174
LAPLACE operator...112, 115, 199, MORERA´s stress function ........156
224 nabla ...................................109, 125
LAPLACEan PDE......................221 NAVIER´s displacement equations
latent hardening ..........................273 ................................................198
LCF ..............................................33 ND ......................................198, 218
left subsymmetry ........................100 negative definite ...........................59
left-product ...................................57 negative semidefinite....................59
LEGENDRE transformation ......254 NEUBER ....................................221
LEMAITRE..................................30 NEUMANN conditions ..............200
lemma of CAUCHY ...................143 NEWTON.....................................83
LEVI-CIVITA symbol .................46 NEWTON element .......................10
line element ................................126 NEWTON potential....................219
linear mapping..............................49 nominal stress ................................. 4
linear momentum..........................83 norm .............................................96
linear momentum balance...141, 144 norm of a vector ...........................47
linear space ...................................46 normal stress...............................143
live time prediction.......................32 normality rule .............................289
loading condition................281, 302 NORTON-BAILEY law...............27
local action ................. 165, 166, 253 obliviator function ........................27
local rate .....................................124 octahedral plane..........................279
logarithmic strain............................ 2 ONB .......................................48, 97
longitudinal wave .......................247 orthogonal tensor ..........................62
loss factor .....................................20 orthonormal basis ...................48, 97
low cycle fatigue ..........................33 orthotropy ...................................175
LUDWIK law ...............................40 OSTROGRADSKI´s integral
MALVERN model .......................20 transformation ........................118
MANSON-COFFIN rule ..............34 overstress ......................................41
MASING model ...........................39 PAPKOVITSCH ........................221
mass conservation.......................124 path .............................................122
mass density ...............................124 permutation symbol ......................46
material coordinate .....................122 PERZYNA model.........................42
material law ................................164 phase...........................................244
material presentation ..................123 piezocaloric effect ......................257
material symmetry......................169 planar dyad ...................................51
material time derivative..............123 plane strain ................................. 131
316 Index
plane stress ................................. 146 principle of virtual displacements

plastic potential ..........................283 ................................................ 152
plastic work ................................308 principle of virtual forces ...........154
plasticity ............................... 34, 271 principle of virtual power ........... 148
POINSOT ..................................... 88 projection method....................... 187
POISSON´s ratio ............ 3, 183, 197 proper-orthogonal......................... 63
polar coordinates ........................115 PVF ............................................ 154
polecone .......................................88 PVP ............................................ 148
polehode .......................................89 RABOTNOV ansatz..................... 30
PONCELET fatigue .....................31 RAMBERG-OSGOOD law .........40
positive definite ............ 59, 102, 196 rate of deformation tensor .......... 132
positive semidefinite ............ 59, 102 rate-independence......................... 10
potential energy ...................... 8, 194 RDI............................. 303, 305, 310
power...................................... 5, 148 reaction forces ............................152
POYNTING model ................ 12, 20 reciprocal relations ............. 105, 256
p-q-type ........................................ 24 reciprocal theorem......................201
PRAGER law ...............................40 reference placement.................... 122
PRAGER model ......................... 308 reflection ......................................63
PRAGER´s hardening rule .........285 relaxation......................................15
PRANDTL model ........................ 39 relaxation function........................ 26
PRANDTL-NADAI law............... 27 relaxation time.............................. 16
PRANDTL-REUSS equations ...284 representation theorem of
pressure ......................................146 HELMHOLTZ ....................... 219
primary creep................................ 23 resonance method .......................190
principal axis of inertia................. 86 retardation time ............................18
principal invariants ....................... 66 rheological model ......................... 11
principal moment of inertia .... 86, 92 rheology.......................................... 9
principal stress............................146 rhombic .............................. 173, 175
principle of cuts.......................... 140 right subsymmetry......................100
principle of D'ALEMBERT .......153 rigid body ............................... 82, 84
principle of DE VEUBEKE – HU – rigid body motion ....................... 128
WASHIZU ..................... 215, 237 rigid plasticity............................... 34
principle of determinism .... 164, 252 RITZ method ..............................149
principle of HAMILTON ...........151 ROBINSON model ...................... 42
principle of HELLINGER- rotation ................. 62, 127, 167, 208
REISSNER .............................217 rotational energy...........................87
principle of invariance under rigid scalar product ......................... 47, 96
body motions .............. 7, 165, 167 SCHMID´s law................... 273, 291
principle of local action..... 165, 166, SCHWEDOFF model...................41
253 second law of thermodynamics ..250
principle of maximum energy ....214 secondary creep ............................ 23
principle of minimum energy .....211 self-hardening.............................273
principle of minimum strain energy shear ................................... 129, 182
........................................ 234, 240 shear stress ................................. 143
principle of minimum stress energy shear wave .................................. 247
................................................ 213 simple dyad ..................................49
principle of superposition ...........200 simple material ....................... 7, 167
Index 317
simple shear........................129, 182 symmetry transformation....170, 259

singular tensor ..............................56 TAYLOR problem .....................282
skew tensor ...................................57 TAYLOR-QUINNEY factor ......310
slip system ..................................273 temperature................................. 249
small deformation.......................128 temperature gradient..105, 251, 252,
SODERBERG law ....................... 27 257
softening .......................................35 temperature strain tensor ............258
SOMIGLIANA...........................220 temperature-displacement relation
spatial coordinate........................122 ................................................261
spatial representation ..................123 tension ........................................183
specific heat................................306 tensor ............................................51
spectral form................................. 70 tensor analysis ............................104
spherical tensor.............................55 tensor basis ...................................52
spin tensor ..................................132 tensor component .........................52
square form...................................59 tensor of inertia.............................85
ST. VENANT.............................135 tensor of inertia of area.................91
ST. VENANT element .................34 tensor product .........................49, 93
stability .......................................288 tensor surface................................59
steady creep ..................................23 tertiary creep...........................23, 28
STIELTJES integral .....................26 tetrad.......................................95, 98
stiffness tensor............................168 tetragonal ............................173, 175
stiffness tetrad.............................287 theorem of CAUCHY.................143
strain tensor ................................126 theorem of FEDOROV...............247
strain, linear .................................... 1 thermal expansion.......................260
strain, logarithmic........................... 2 thermoelasticity ..........................252
stress ............................................... 4 thermoplasticity ..........................301
stress deviator .............................146 time-derivative..............................81
stress energy ................................... 8 trace ..............................................66
stress equivalence .........................30 traction........................................141
stress function.............................154 traction boundary conditions ......200
stress power .................... 5, 150, 193 translation ...................................127
stress tensor ........................143, 146 translational energy ......................87
stress work...................................... 5 transpose of a tensor .....................56
stress, effective .............................29 transposed tetrad...........................99
stress-temperature tensor ....256, 306 transversal deformation .................. 3
strongly elliptic...........................246 transversal wave .........................247
substantial rate............................124 transversely isotropic..........172, 176
subsymmetry ......................100, 168 TRESCA´s yield criterion ..........275
summation convention .................44 triad...............................................93
superposition principle 25, 200, 266, triclinic ...............................172, 174
270 trigonal ...............................173, 176
SWIFT law ...................................40 triple product ................................48
symbolic notation .........................64 true stress........................................ 4
symmetric tensor ..........................57 uniqueness theorem ............202, 261
symmetriser ................................101 unit tensor .....................................54
symmetry group..................170, 173 v. MISES´ equivalent stress .......276
symmetry of a tetrad.....................99 VARIGNON´s principle...............84
318 Index
vector basis................................... 47 VOIGT representation................168

vector components........................ 47 vorticity tensor............................132
vector norm .................................. 47 WASHIZU ................................. 215
vector product............................... 48 wave ........................................... 244
vector space ..................................46 WB ............................. 150, 250, 258
velocity....................................... 123 WÖHLER curve ...........................32
velocity gradient .........................132 work................................................ 5
versor............................................63 work balance .............................. 150
VIETA.......................................... 68 work hardening...........................284
virtual displacement ................... 152 yield criterion ..................... 272, 302
virtual power .............................. 148 yield limit ............................... 34, 38
viscoelasticity ......................... 9, 266 yield stress .................................... 34
viscoplasticity............................... 41 yielding......................................... 38
VOCE law ....................................36 YOUNG´s modulus................ 7, 183
VOIGT .......................................101 zero tensor .............................. 55, 98
VOIGT model .............................. 12 ZIEGLER´s hardening rule ........285
View publication stats

Solid Mechanics: Theory, Modeling, and Problems: January 2015

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Solid Mechanics: Theory, Modeling, and Problems: January 2015

Diunggah oleh

Hak Cipta:

Format Tersedia

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Solid mechanics: Theory, modeling, and problems

Book · January 2015

Albrecht Bertram R. Glüge

SEE PROFILE SEE PROFILE

Dissertation View project

The user has requested enhancement of the downloaded file.

First German edition Febr. 2013

ISBN 978-3-319-19565-0 ISBN 978-3-319-19566-7 (eBook)

Library of Congress Control Number: 2015948416

Springer Cham Heidelberg New York Dordrecht London

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media

Do not say anything against Mechanics!

Honoré de Balzac (The Magic Skin, 1831)

Albrecht Bertram, Rainer Glüge Magdeburg, 2015

1 ONE-DIMENSIONAL MATERIAL THEORY 1

1.1 Deformations, Stresses, and Work 1

1.3 Viscoelasticity (Rheology) 9

2 INTRODUCTION TO TENSOR CALCULUS 43

2.1 Vector and Tensor Algebra 44

2.1.12 Time-Dependent Vectors and Tensors 81

2.2 Vector and Tensor Analysis 104

3 FOUNDATIONS OF CONTINUUM MECHANICS 121

3.1 Kinematics 121

3.2 Stress Analysis 140

4 THREE-DIMENSIONAL MATERIAL THEORY 164

4.1 Elasticity 165

4.2 Thermomechanics 249

4.3 Linear Viscoelasticity 266

4.4 Plasticity 271

C1 1st equation of CAUCHY (3.2.7)

Frequently used abbreviations

COOS coordinate system

List of frequently used symbols

A , A0 content of area (of cross-section), current and initial

 coefficient of thermal expansion

Besseling, J. F.; van der Giessen, E.: Mathematical Modelling of Inelastic

1.1 Deformations, Stresses, and Work

Ó Springer International Publishing Switzerland 2015 1

The transformation between these two strain measures is

If we expand this relation into a TAYLOR series

It is also dimensionless. For linear-elastic materials this ratio is a material constant

In addition, measurement of the stresses cannot be achieved directly, but only

The name is due to the fact that

Mechanical material theory addresses the relationships between stresses and

 (x0) = F(u(x) xx 0l ) .

We expand u(x) into a TAYLOR series at x0

u(x) = u(x0) + u(x0)' (x – x0) + ½ u(x0)'' (x – x0)2 + ...

The symbol of a linear elastic material is a spring, called a HOOKE element

The specific stress power of this material possesses a potential w

which is stored as elastic or potential energy or strain energy w ().

with the complementary stress power

1.3 Viscoelasticity (Rheology)

Bert, C. W.: Material damping: An introductory review of mathematical models,

Tschoegl, N. W.: The Phenomenological Theory of Linear Viscoelastic

Among the inelastic materials, we find rate-independent (scleronomous) ones,

1.3.1 The NEWTON Element

with a positive viscosity or damper constant D of dimension [stress  time]. This

called NEWTON7 element. Under constant load, this material creeps at a

Other linear-viscoelastic models can be constructed by parallel arrangements or

 for parallel arrangements:

(1.3.3)  = 1 + 2 The stresses are additive.