net/publication/292643555
CITATIONS READS
8 3,404
2 authors:
Some of the authors of this publication are also working on these related projects:
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.
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
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.
1.2 Elasticity 6
1.4 Plasticity 34
1.4.1 Rigid-Plastic Models 34
1.4.2 Elastic-Plastic Models 38
1.5 Viscoplasticity 40
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.
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
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) dln : =
l
so that
l
dl l
ln = l
= ln( ) .
l0
l0
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)
1.1 Deformations, Stresses, and Work 3
(1.1.7) v := – .
3
Siméon Denis Poisson (1781-1840)
4 1 One-Dimensional Material Theory
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
F F A d2
0 = = = 2 = ( +1)2 (2 + 1)
A0 A A0 d0
4
Blaise Pascal (1623-1662)
1.1 Deformations, Stresses, and Work 5
= (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
6 1 One-Dimensional Material Theory
1.2 Elasticity
x0
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)
8 1 One-Dimensional Material Theory
HOOKE element
(1.2.3) = E d = ½ E 2 = : w () = ½
0
(1.2.4) = E –1 d = ½ E –1 2 = : w *() = ½
0
NEWTON element
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.
7
Isaac Newton (1643-1727)
1.3 Viscoelasticity (Rheology) 11
E1
E2
E1 E2
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.
12 1 One-Dimensional Material Theory
MAXWELL model
KELVIN or
VOIGT model
POYNTING model
BURGERS model
1.3 Viscoelasticity (Rheology) 13
E D
8
James Clerk Maxwell (1831-1879)
14 1 One-Dimensional Material Theory
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
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).
1.3 Viscoelasticity (Rheology) 15
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.
16 1 One-Dimensional Material Theory
r
= E r
( tr )
tr t
relaxation test
Hence, the relaxation time is an inverse measure for the speed of the relaxation.
9
Franz Dischinger (1887-1953)
1.3 Viscoelasticity (Rheology) 17
KELVIN model
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)
18 1 One-Dimensional Material Theory
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
1.3 Viscoelasticity (Rheology) 19
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
20 1 One-Dimensional Material Theory
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 CK
(1.3.18) C D
K K
This leads to the following explicit forms
11
John Henry Poynting (1852-1914)
1.3 Viscoelasticity (Rheology) 21
(1.3.19)
K2
t
K K
t C K
t 0 exp d exp t
D D D
0
t
t
CK
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
22 1 One-Dimensional Material Theory
C D
R BURGERS model
K = K K
R = K D = D D
= C + D + K,R R = R R
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)
1.3 Viscoelasticity (Rheology) 23
H
VN VN
+
creep curve C K
VN
C
CHU
relaxation curve
t
24 1 One-Dimensional Material Theory
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).
13
Ludwig Boltzmann (1844-1904)
14
Oliver Heaviside (1850-1925)
26 1 One-Dimensional Material Theory
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.
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)
In particular the NORTON law has gained much popularity. For other creep
functions see SZKRYPEK p. 363 ff.
Betten, J.: Kontinuumsmechanik. Springer, Berlin (1993), 2nd ext. edt. (2001)
Betten, J.: Creep Mechanics. Springer, Berlin (2002)
Besson, J.; Cailletaud, G.; Chaboche, J.-L.; Forest, S., Blétry, M.: Non-Linear
Mechanics of Materials. Springer, Dordrecht (2010)
28 1 One-Dimensional Material Theory
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.
1.3 Viscoelasticity (Rheology) 29
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.
F
(1.3.25) e
Ae
larger than the true stresses
F
.
A
A Ae
undamaged damaged
16
Lazar M. Kachanov (1914-1993)
30 1 One-Dimensional Material Theory
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
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
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)
32 1 One-Dimensional Material Theory
(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.
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)
1.3 Viscoelasticity (Rheology) 33
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
For these reasons, fatigue prediction is still rather difficult and highly empirical.
34 1 One-Dimensional Material Theory
1.4 Plasticity
Literature
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
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
Y
2Y
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.
38 1 One-Dimensional Material Theory
B Y
Y
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
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)
40 1 One-Dimensional Material Theory
1.5 Viscoplasticity
Literature
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. ext. edt. (2001)
Christescu, N.; Suliciu, I.: Viscoplasticity. M. Nijhoff Publ., Den Haag (1982)
Francois, D.; Pineau, A.; Zaoui, A.: Mechanical Behaviour of Materials, vol. II,
Kluwer, Dordrecht (1998)
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)
Skrzypek, J. J.: Plasticity and Creep. CRC Press, Boca Raton (1993)
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
22
Eugene Cook Bingham (1878-1945)
42 1 One-Dimensional Material Theory
= 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
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)
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
2.1 Vector and Tensor Algebra 45
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)
46 2 Introduction to Tensor Calculus
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
( ) a = ( a) (associative)
24
Tullio Levi-Civita (1873-1941)
2.1 Vector and Tensor Algebra 47
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
ab = ba (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 = ab cos .
If the vectors a and b are mutually orthogonal we have
ab = 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
48 2 Introduction to Tensor Calculus
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
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 Vector and Tensor Algebra 49
Problem 1. Linearity
Check the real function
y = f (x) : = m x + n
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 .
50 2 Introduction to Tensor Calculus
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 Vector and Tensor Algebra 51
(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 = Tx.
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
52 2 Introduction to Tensor Calculus
{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
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 = Tx
= 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)]
54 2 Introduction to Tensor Calculus
= 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.
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).
56 2 Introduction to Tensor Calculus
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
2.1 Vector and Tensor Algebra 57
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
58 2 Introduction to Tensor Calculus
= 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 .
For some applications one needs the cross-product between a vector (from the
left) and a tensor (from the right)
vT
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
60 2 Introduction to Tensor Calculus
= 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
Tv = (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 Vector and Tensor Algebra 61
(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) vI = Iv.
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 .
62 2 Introduction to Tensor Calculus
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} .
2.1 Vector and Tensor Algebra 63
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 e2e3plane 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
Qv
v
v
25
from lat. vertere = to turn
64 2 Introduction to Tensor Calculus
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
2.1 Vector and Tensor Algebra 65
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) Ta = 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
ao.
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
66 2 Introduction to Tensor Calculus
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
26
Francois Viète (1540-1603)
2.1 Vector and Tensor Algebra 69
Examples
For the zero tensor 0 every vector is an eigenvector with the (triple)
eigenvalue 0 since
0a = o = 0a.
For the identity tensor I every vector is an eigenvector with the (triple)
eigenvalue 1 since
Ia = 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 .
70 2 Introduction to Tensor Calculus
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
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
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)
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
2.1 Vector and Tensor Algebra 73
0 0.6 0
[A ij] = 0.6
ee
0 0.8 .
0 0.8 0
We now compute
b : = A v2
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
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).
74 2 Introduction to Tensor Calculus
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
– mk jl in – ik ml jn – jk il mn) vi Tjk vl em en .
27
Pierre Frédéric Sarrus (1798-1861)
2.1 Vector and Tensor Algebra 75
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.
(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
76 2 Introduction to Tensor Calculus
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
0.6 0.8 0
[Qvvij] = 0 0 1 = [Qeeij] .
0.8 0.6 0
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
2.1 Vector and Tensor Algebra 77
10 5 0
[Aij] = 5 10 0 .
0 0 20
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– 402+475 –1500) / ( – 20) = 2 – 20 + 75
3– 202
– 202+475
– 202+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}
2.1 Vector and Tensor Algebra 79
5 0 0
[Avvij] = 0 15 0 .
0 0 20
The representation with respect to {ei ej} can be most easily calculated
by a change of the basis
A1 = [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 A1 = [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
80 2 Introduction to Tensor Calculus
an 1 0 1 an 2
a = .
n 1 1 an 1
Solution
With the given initial values we can write
n2
an 1 0 1 1 n2 1
(P6.1) a = 1 = [ M ] 1 .
n 1 1
We obtain
an = [M n2] 21 + [M n2] 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]n2 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 Vector and Tensor Algebra 81
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
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 .
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
2.1 Vector and Tensor Algebra 83
x
M
rO
rM
V
29
Leonhard Euler (1707-1783)
84 2 Introduction to Tensor Calculus
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) .
2.1 Vector and Tensor Algebra 85
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
= 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
(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
86 2 Introduction to Tensor Calculus
(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 .
2.1 Vector and Tensor Algebra 87
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 rM2 m + ½
( x) dm + rM x dm
2
V V
(2.1.119) = ½ rM2 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
88 2 Introduction to Tensor Calculus
30
Louis Poinsot (1777-1859)
2.1 Vector and Tensor Algebra 89
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)
31
James McCullagh (1809-1847)
90 2 Introduction to Tensor Calculus
z,e3
s ds
e1, s S
x y, e2
ds(1+ )
x P
dA
ds
rs
ds
d
t = (u t) = u t + u t
ds
= (s) x t + (s) x t + (s) x t
d
(2.1.124) = xt 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
z dA y dA y z dA
2 2
Jyy = Jzz = Jyz = Jzy = –
A A A
92 2 Introduction to Tensor Calculus
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
1
= ½ (Jyy + Jzz) ( J yy J zz )2 J 2yz
4
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
2.1 Vector and Tensor Algebra 93
The scalar product of two vectors (which we will consider from now on as
1st-order tensors) gives the real number
vx.
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
94 2 Introduction to Tensor Calculus
as linear combinations
(2.1.127) C = Ci1 i2 ...iK ei1 ei2 ... eiK
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
96 2 Introduction to Tensor Calculus
= e i1 e i2 ... e iK e i1 ... e 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
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 66 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)
102 2 Introduction to Tensor Calculus
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.
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
= const grad
r dr
(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 = –Kg.
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)
106 2 Introduction to Tensor Calculus
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
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
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)
108 2 Introduction to Tensor Calculus
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 Vecter and Tensor Analysis 109
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
110 2 Introduction to Tensor Calculus
x1 ,x2 ,x3
(2.2.21) grad (r) = ei = , i ei = = .
xi
With the nabla notation one sees the following identities for all differentiable
scalar fields (r) , vector fields v(r) , and tensor fields T(r) .
112 2 Introduction to Tensor Calculus
36
Pierre Simon Laplace (1749-1827)
2.2 Vecter and Tensor Analysis 113
ez
ey
r
ex
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
2.2 Vecter and Tensor Analysis 115
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 rr r z
37
Jakob Steiner (1796-1863)
116 2 Introduction to Tensor Calculus
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)
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.
are needed to transform volume integrals into surface integrals, or vice versa.
Divergence Theorem (GAUSSOSTROGRADSKI38) 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'
x1
x3
38
Carl Friedrich Gauß (1777-1855), Mikhail Vasilevich Ostrogradski (1801-1862)
2.2 Vecter and Tensor Analysis 119
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 e1 n dA .
A
and
u3
dx1 dx2 dx3 = u3 e3 n dA
V x3 A
div u dV = u n dA .
V A
120 2 Introduction to Tensor Calculus
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 e3plane.
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
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
3.1 Kinematics
Xi xi
u u
e2
X x
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
39
Jean Louis Lagrange (1736-1813)
124 3 Foundations of Continuum Mechanics
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
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
126 3 Foundations of Continuum Mechanics
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
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
dx2 = dx dx = (F dX) (F dX) = dX FT F dX
and its difference with the one in the reference placement is
dx2 – dX2 = 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
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
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.
Y
1
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
41
see next page.
3.1 Kinematics 131
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
132 3 Foundations of Continuum Mechanics
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 .
42
Adhémar Jean Claude Barré de Saint-Venant (1797-1886)
43
Eugenio Beltrami (1835-1900)
136 3 Foundations of Continuum Mechanics
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
vx = 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
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
e3axis. 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
with respect to the basis {ei ej} . For E = sym(H) we find the
component matrix
We find a stretch in the e3direction, 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 e3direction 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
140 3 Foundations of Continuum Mechanics
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 m0 m
so that vice versa
(3.2.1) fm = b dm = b dV
V V
3.2 Stress Analysis 141
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 A0 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.
142 3 Foundations of Continuum Mechanics
e2 x2
t(n4)
e3 e1
x3 x1
t(– e2)
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)
144 3 Foundations of Continuum Mechanics
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 e3direction 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
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 + ...
3.2 Stress Analysis 145
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
146 3 Foundations of Continuum Mechanics
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 Stress Analysis 147
a dm = b dm + t dA
V V A
the global balance of angular momentum
(a – b) v dm = div(T) v dV
V V
148 3 Foundations of Continuum Mechanics
= ( 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
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) .
3.2 Stress Analysis 149
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.
150 3 Foundations of Continuum Mechanics
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
3.2 Stress Analysis 151
HAMILTON´s45 Principle
The laws of motion are fulfilled between two instants of time t1 and t2 if and only
if
t2
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 xV.
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)
152 3 Foundations of Continuum Mechanics
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
3.2 Stress Analysis 153
= 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 d´Alembert (1717 - 1783)
154 3 Foundations of Continuum Mechanics
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).
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
3.2 Stress Analysis 155
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)
3.2 Stress Analysis 157
49
George Biddell Airy (1801-1892)
50
see SZABO, I.: Höhere Technische Mechanik. Springer (1964) p. 153
51
see SZABO (1964).
158 3 Foundations of Continuum Mechanics
air front
water front
n
h
x
l
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.
0
ey
a er
ex
160 3 Foundations of Continuum Mechanics
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
3.2 Stress Analysis 161
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.
162 3 Foundations of Continuum Mechanics
Let us next consider the stress distribution at the edge of the hole. For
/2 and r a we find = 30 , 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.
3.2 Stress Analysis 163
4 THREE-DIMENSIONAL MATERIAL THEORY
4.1 Elasticity
Literature
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)
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
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*
170 4 Three-Dimensional Material Theory
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
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. Qn
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 Qk 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
52
after COLEMAN, B. D.; NOLL, W.: Material symmetry and thermostatic
inequalities in finite elastic deformations. Arch. Rational Mech. Anal. 15, 87-111
(1964)
174 4 Three-Dimensional Material Theory
reflections – Qe at the e1e2plane. The elementary cell has two right angles and
3
one oblique one. Here,
(– Qe ) e1 = e1 (– Qe ) e2 = e2
3 3
(– Qe ) 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
2C1123 0 0
C1111 C1122 C1133
C1111 C1133 2C1123 0 0
C3333 0 0 0
2C2323 0 0
symmetric 2C2323 2C1123
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.
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
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'.
180 4 Three-Dimensional Material Theory
Q Qe / 2 = e p2 e p1 – e p1 e p2 + e p3 e p3
3
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 Qe = – e1 e1 – e2 e2 + e3 e3
3
E 2 G 1+
G = k =
2 1 3 1 2
which has the same value in all directions in the isotropic case.
(4.1.17) T = tr(E) I + 2 E =
E
[E + tr(E) I]
1 1 2
184 4 Three-Dimensional Material Theory
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
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 3k 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
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
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
188 4 Three-Dimensional Material Theory
(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
N
C = i Pi
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
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
190 4 Three-Dimensional Material Theory
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).
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)
192 4 Three-Dimensional Material Theory
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
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
"(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.
196 4 Three-Dimensional Material Theory
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
G0 k 0
0 3 + 2 0
E0 –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.
Literature
Gurtin, M. E.: The Linear Theory of Elasticity. In: Encyclopedia of Physics. Edt.
S. Flügge. Vol. VIa/2, Springer (1972)
Hetnarski, R. B.; Ignaczak, J.: The Mathematical Theory of Elasticity. CRC
Press (2004, 2011)
Fichera, G.: Existence Theorems in Elasticity. In: Encyclopedia of Physics. Edt.
S. Flügge. Vol. VIa/2, Springer (1972)
Knops, R. J.; Payne, L. E.: Uniqueness Theorems in Linear Elasticity. Springer,
Berlin (1971)
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.
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
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)
202 4 Three-Dimensional Material Theory
= (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.
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.
= 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 erdirection, 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 ezdirection
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 ezdirection
occurs. Then
E,z = 0
or
rr , z = ur , rz = 0
, z = ur , z / r = 0
zz , z = uz , zz = 0 .
Furthermore, the ezplane 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 ezdirection. 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
206 4 Three-Dimensional Material Theory
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 ezplane 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.
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 )
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.
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
210 4 Three-Dimensional Material Theory
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 .
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
Literature
Gurtin, M. E.: The Linear Theory of Elasticity. In: Encyclopedia of Physics. Edt.
S. Flügge. Vol. VIa/2, Springer (1972)
Hetnarski, R. B.; Ignaczak, J.: The Mathematical Theory of Elasticity. CRC
Press (2004, 2011)
Slaughter, W. S.: The Linearized Theory of Elasticity. Birkhäuser, Boston (2002)
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
= ½ E C E dV + T0 E dV
V V
T0 E dV = T0 grad u 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(E) dV + t0 u dA – u div(T0) dV
V Ad V
4.1 Elasticity 213
– b u dV – 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).
214 4 Three-Dimensional Material Theory
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
= 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
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.
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)
216 4 Three-Dimensional Material Theory
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
= 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
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
+ 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)
218 4 Three-Dimensional Material Theory
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 .
63
Hermann Ludwig Ferdinand von Helmholtz (1821-1894)
220 4 Three-Dimensional Material Theory
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
div grad u + grad div u
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 = xb
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)
222 4 Three-Dimensional Material Theory
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 .
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
f + T eZ dA = o .
A
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
Finally we get
GZ 3C1 3 R GZ 3C1 3 GC1
– f = (1 ) [ D ] R 0 = (1 ) (0 Z ) = – (1 )
f Z 2 2 2
(P16.7) uZ = 4 G 3 D
D
Z
4.1 Elasticity 229
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
1 2
w(R) = 2 G p() d d
0 1
a2 = 2 + ( /2)2 =2 a 2 2 .
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) .
1 1 1 2
w0 = a2 +
G1 G2
1 1 1 1 1 2
+ R = G + G
R
1 2 1 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
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.
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
234 4 Three-Dimensional Material Theory
(u) = w dV – b u dm – tpr u dA
V V Ad
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
( 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.
236 4 Three-Dimensional Material Theory
The starting point for the applications of the Principle of Minimum Stress
Energy is the functional (4.1.44)
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
= 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
(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
238 4 Three-Dimensional Material Theory
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
i0 i0
3 3
= i p i ,x = i pi, x
i0 i0
with p : = x / l.
3 3
u = ui pi = i pi
i0 i0
3 3
= i pi , x = i pi, x .
i0 i0
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) + 33 x6 / (6 l 6) + g x4 / (4 l 3)] 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.
u(x 0, y) = o u(x l, y) = d ex
ey
ex
n n
uy = x (l – x) Cy ij xi y j
i 0 j0
n n
ux = x d / l + x (l – x) Cx ij xi y j.
i 0 j0
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.
244 4 Three-Dimensional Material Theory
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)
Gurtin, M. E.: The Linear Theory of Elasticity. In: Encyclopedia of Physics. Edt.
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)
Hetnarski, R. B.; Ignaczak, J.: The Mathematical Theory of Elasticity. CRC
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)
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
246 4 Three-Dimensional Material Theory
(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
n
u
68
Fedor Ivanovich Fedorov (1911-1994)
248 4 Three-Dimensional Material Theory
[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 249
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)
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
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)
4.2 Thermomechanics 251
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
70
Rudolf Julius Emmanuel Clausius (1822-1888), Pierre Maurice Marie Duhem
(1861-1916)
71
Max Karl Ernst Ludwig Planck (1858-1947)
252 4 Three-Dimensional Material Theory
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
72
Josiah Willard Gibbs (1839-1903)
4.2 Thermomechanics 253
= (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) qg 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
254 4 Three-Dimensional Material Theory
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)
4.2 Thermomechanics 255
= +
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
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 Eθ 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)
4.2 Thermomechanics 257
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).
gKg 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
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 = –Kg
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
258 4 Three-Dimensional Material Theory
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 ).
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.
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.
4.2 Thermomechanics 263
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
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
t
(t) = J(t – ) ( ) d
0
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.
268 4 Three-Dimensional Material Theory
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
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).
270 4 Three-Dimensional Material Theory
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,
Kluwer, Dordrecht (1998)
Fung, Y. C.; Tong, P.: Classical and Computational Solid Mechanics. World
Scientific, Singapore (2003)
Hill, R.: The Mathematical Theory of Plasticity. Clarendon Press, Oxford (1950)
Irgens, F.: Continuum Mechanics. Springer (2008)
Krawietz, A.: Materialtheorie. Springer, Berlin (1986)
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)
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)
Skrzypek, J. J.: Plasticity and Creep. CRC Press, Boca Raton (1993)
Yang, W.; Lee, W. B.: Mesoplasticity and its Applications. Springer, Berlin
(1993)
Yu, M.-H.: Generalized Plasticity. Springer, Berlin (2006)
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.
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
n n
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)
274 4 Three-Dimensional Material Theory
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)
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)
276 4 Three-Dimensional Material Theory
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
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
p3
n
p2
H1
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
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.
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.
282 4 Three-Dimensional Material Theory
d n.
The symmetric part of this tensor stands for the rate of plastic deformation
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)
284 4 Three-Dimensional Material Theory
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
83
Ludwig Prandtl (1875-1953)
84
Endre Reuss (1900-1968)
4.4 Plasticity 285
TR
T
T0 0
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)
286 4 Three-Dimensional Material Theory
= [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) + 2Y 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 + 26 K(p)] Y2.
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
convex non-convex
dEp dEp
T TA T
TA
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
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) .
292 4 Three-Dimensional Material Theory
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 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 + (2rr – – ) / 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
296 4 Three-Dimensional Material Theory
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 = 2Y (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 – 2Y / r
and an integration over r to
(P22.6) rad = 2Y 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 = 2Y ln ra + c
– pi = 2Y ln ri + c .
Elimination of c gives the critical internal pressure
(P22.7) pi crit pl = 2Y 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) = 2Y 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
298 4 Three-Dimensional Material Theory
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.
300 4 Three-Dimensional Material Theory
u(r)
tang
rad
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)
Negahban, M.: The Mechanical and Thermodynamical Theory of Plasticity. CRC
Press, Boca Raton (2012)
Taylor, G. I., Quinney, H.: The latent heat remaining in a metal after cold
working. Proc. Roy. Soc. London A163, 157-181 (1937)
1 qg
+ – 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
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
304 4 Three-Dimensional Material Theory
= ( 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
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
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
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.
308 4 Three-Dimensional Material Theory
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
t1
T TB T TB
(4.4.89) wp : = Ep dt wp = Ep.
t0
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
310 4 Three-Dimensional Material Theory
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
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