Lecture Notes

Eective Properties of Micro-Heterogeneous

Materials
Dr.-Ing. Daniel Balzani
Institut f ur Mechanik
Universitat Duisburg-Essen
Fakultat f ur Ingenieurwissenschaften
Abteilung Bauwissenschaften
Summer term 2011
Eective Properties, summer term 2011, c _ Daniel Balzani 1
1 Introduction
The design and modeling of micro-heterogeneous materials is more and more based on
the computer simulation of how the individual components will behave along the process
chain. As an example, the study of how metals react to rolling, deep drawing and welding
and how the nal components mechanical properties will be, plays the major role in case
of car body parts. Today numerical methods and material models are able to simulate
many materials, such as aluminum and conventional steel, appropriately. Unfortunately,
some other materials as e.g. high-strength steels, ber-matrix composites, lled polymers,
etc. are not standard to be simulated due to their special properties, especially at smaller
scales. In order to overcome these diculties such materials have to be analized circum-
stantially. In this context the properties of the individual components or crystallites are
computed and results herefrom are used to conclude in the behavior of the entire ma-
terial. This problem of scales turns out to be one of the most challenging tasks with
respect to micromechanics. Depending on the type of micro-heterogeneity of materials
dierent lengthscales have to be considered: the macroscale (measured in e.g. meters),
the mesoscale (measured in millimeters), the microscale (measured in micrometers) and
the nanoscale (measured in nanometers). The macroscale characterizes systems and struc-
tures as e.g. a cantilever beam or a pillar, whereas the mesoscale usually describes the
material at the level of e.g. inclusions in a matrix. At the microscale we dierentiate be-
tween individual grains and crystals and the nanoscale considers individual molecules or
even atoms. However, if only two scales are to be analyzed then usually these two scales
are referred to as micro- and macroscale. The main idea behind micromechanics is that
the real micro-heterogeneous material can be treated as a homogenized one at a larger
scale.
A variety of materials is obviously micro-heterogeneous, as for instance concrete due to
the existence of supplements embedded in a cementoid matrix phase. A further exam-
ple is wood where a kind of microstructure composed of rays and longitudinal cells is
already visible to the naked eye. But there exist many further materials where the micro-
heterogeneity is not as obvious. One example is high-strength steel, whose advantageous
elasto-plastic behavior is mainly governed by a complex microstructure at the meso- and
at the microscale.
For the task of scale transition there exist a variety of approaches. One (self-consistent)
approach observes a single test crystallite in an environment matching the mean properties
of all other crystallites. Another possibility is to understand metal properties by use of
improved Taylor models. For the modeling of the macroscopic behavior in the framework of
the continuum mechanics it is clearly stated in [2], that hereby problems can be calculated
if the properties of the mesoconstituents are given. Conversely, the macroscopic behavior
can be investigated to determine some of the properties of the microconstituents. Due to
the nature of continuum mechanics it is however not possible to determine these properties
by fundamental physical and chemical information; there are still experiments necessary.
Eective Properties, summer term 2011, c _ Daniel Balzani 2
As an example continuum mechanics is not able to predict the yield strength from carbon
content in the ferrite of pearlitic or spheroidized steel without any tests.
One important application area of micro-macro modeling results from the fact that crys-
tallographic orientation has been observed to have signicant eect in most aluminum
alloys on crack initiation, which is quite important in the context of probability-of-failure
estimate analysis. In [11] an elasto-viscoplastic model and corresponding Finite-Element
implementation is presented for Al 7075-T651, based on microstructural information. In
[15] a dynamic explicit Finite Element Analysis is introduced by using a crystallographic
homogenization method to estimate the polycrystalline sheet metal formability. There,
the velocity in the homogenized macro- and the micro crystal structure is considered
additionally. A further step not only considering the micro- and macroscale is realized
in [5], where even the smallest scale at the molecular level is taken into account by us-
age of quantum physics. Furthermore, ne scale deformations are described by a particle
dynamics method extending the molecular dynamics to multi-atom aggregates.
Due to the fact that the consideration of both the micro- as well as the macroscale in
one calculation, is computationally very expensive, parallel computing algorithms seem
to become more important especially when large macrostructures have to be analyzed. In
[8] a component template library is used which is suitable to adapt for the parallelization
easily. Another example for a realization of parallel computing in the context of Finite-
Element problems at dierent scales synchronously is given in [10].
Literature Recommendation With respect to an introduction to the homogenization
and localization theory in the context of small strains we refer to [4]. Another important
reference is [17], where also the issue of computational aspects is discussed.
Other references used in these lecture notes are [1], [16], [2], [3], [5], [6], [7], [8], [9], [10],
[11], [12], [13], [14], [15].
Eective Properties, summer term 2011, c _ Daniel Balzani 3
2 Defects and Fundamental Solutions
In elastic materials defects are characterized by inhomogeneous stress or strain elds.
Basically, defects can be subdivided into the two main groups
Defects that are themselves source of eigenstrains or eigenstresses (dislocations,
inclusions)
Defects that arise due to the eect of external loadings (inhomogeneous materials,
wholes, cracks).
In the case of material inhomogeneities it is reasonable to decouple the stress and strain
elds into a homogeneous part , of the virtually undisturbed material (without defects),
and a uctuation part , , which is also referred to as eigenstrain or eigenstress, then
we obtain
= + and = + . (1)
2.1 Eigenstrains or Eigenstresses
2.1.1 Dilatation Centers The idealization of a punctiform region which is character-
ized by a radial expansion (eigenstrain) is referred to as dilatation center. In an isotropic
material radial-symmetric strain- and stress elds with tension in circumferential direction
and compression in radial direction are obtained. As an example we refer to a DP-steel,
where due to the production process an austenitic inclusion transforms to martensite re-
sulting in a volumetric jump. A dilatation center can be interpreted as a spherical region
with radius a where a pressure p exists, cf. Fig. 1.
p
r
a
Figure 1: Idealization of a dilatation center.
For this idealization an analytical solution can be derived in spherical coordinates (r, , )
for the displacements
u
r
= p
a
3
4r
2
, u

= u

= 0 (2)
Eective Properties, summer term 2011, c _ Daniel Balzani 4
and for the stresses

r
= p
a
3
r
3
,

= p
a
3
2r
3
,
r
=
r
=

= 0 . (3)
Remark: a dilatation center can be used as a simple model for the description of an
inter-atomic-lattice atom (punctiform defect).
2.1.2 Straight Edge and Screw Dislocations Dislocations are linear defects in
crystalline solids, which can be continuum-mechanically characterized by a constant jump
b (Burgers vector), cf. Fig. 2 where x
3
is the dislocation line.
a) b)
Figure 2: Schematic illustration of an a) edge dislocation and b) screw dislocation, taken
from [4].
Let us now distinguish between the two dislocation types
Edge dislocations, where the Burgers vector is transverse to the dislocation line, and
Screw dislocations, where the Burgers vector follows the dislocation line.
For a straight edge dislocation following Fig. 2a we are able to provide analytical equations
with b = [b[ and r
2
= x
2
1
+ x
2
2
for the displacements
u
1
=
D
2
_
2(1 ) +
x
1
x
2
r
2
_
and u
2
=
D
2
_
(1 2)lnr +
x
2
2
r
2
_
, (4)
Eective Properties, summer term 2011, c _ Daniel Balzani 5
wherein the abbreviation D = b/2(1 ) is used, and for the stresses

11
= Dx
2
3x
2
1
+ x
2
2
r
4
,
12
= Dx
1
x
2
1
+ x
2
2
r
4
and
22
= Dx
2
x
2
1
x
2
2
r
4
. (5)
More simply the displacement- and stress elds can be analytically computed for the
screw dislocations by
u
3
=
b
2
,
13
=
b
2
x
2
r
2
and
23
=
b
2
x
1
r
2
. (6)
2.1.3 Inclusions Here, we focus on spatial distributions of eigenstrains
t
(x), which
result e.g. from phase transformations in solids, where the distances between the atoms
change and rearrange the lattice. Since these eigenstrains are not associated with stresses
they are also referred to as stress-free transformation strains. Other examples for such
eigenstrains are all eigenstrains existing in stress-free states as e.g. thermal or plastic
strains. In the context of small strains the total strains and resulting stresses are additively
obtained by
=
e
+
t
and = C : (
t
) . (7)
Herein, the elastic strains and the elastic tangent moduli are denoted by
e
and C. If
only a certain area is governed by such eigenstrains
t
,= 0 then this area is referred
to as inclusion and the surrounding region with
t
= 0 as matrix. Usually, these
two regions are associated with the same elasticity, otherwise it would be referred to as
inhomogeneity. It is remarked that often we speak of inclusions even if dierent material
properties exist if there is no danger of confusion.
Please note that in general it is not possible to derive analytical solutions for the distri-
butions of stress or total strains. However, there exist closed-form solutions for several
special cases which play an important role in the eld of multiscale mechanics and which
are discussed in the following sections.
2.1.4 Eshelby-Solution Here, we focus on an ellipsoidal inclusion which is located
in an innite matrix. With the semi axis a
i
the geometry of the ellipsoid is described by
_
x
1
a
1
_
2
+
_
x
2
a
2
_
2
+
_
x
3
a
3
_
2
1 , (8)
where the semi axis a
i
coincide with the x
i
-axis. If a constant eigenstrain
t
= const is
acting on the ellipsoidal inclusion, then an interesting nding of J.D. Eshelby (1916-1981)
can be proven that also the total strains are constant inside the inclusion . These can
be computed in terms of the Ehelby-tensor S based on the linear relationship
= S :
t
= const in or
ij
= S
ijkl

t
kl
in . (9)
Eective Properties, summer term 2011, c _ Daniel Balzani 6
We conclude with (7) that the stresses have to be constant in the inclusion , too. These
can then be computed by
= C : (S I) :
t
= const in , (10)
wherein I characterizes the fourth-order identity tensor
I
nmkl
=
1
2
(
mk

nl
+
ml

nk
) . (11)
The symmetry with respect to the rst and second pair of indices does not hold (S
ijkl
,=
S
klij
) although the symmetries S
ijkl
= S
jikl
= S
ijlk
hold. For isotropic materials the
components of S depend only on the
the Poissons ratio ,
the semi axis a
i
and
their orientation in the x
i
-coordinate system.
Interestingly, the coecients do not depend on the elasticity modulus. Due to the com-
plexity of the individual components we refer to [14] and focus on some more simple
special cases. However, we conclude several properties:
the strain- and stress elds are not constant in the matrix,
they behave indirectly proportional to the cubic distance from the inclusion, i.e.
, r
3
for r ,
the result of Eshelby holds generally also for anisotropic materials
closed-form representations of the components of S and strain- and stress elds in
the matrix are only possible for isotropic materials .
Let us now consider two simple special cases. If we consider a cylinder of innite length as
the inclusion geometry, i.e. a
3
, as shown in Fig. 3, then we are able to write down
a two-dimensional analytical solution.
In this case the matrix strain- and stress elds in the x
1
x
2
-plane behave indirectly
proportional to the square distance from the inclusion (, r
2
for r ). The
Eective Properties, summer term 2011, c _ Daniel Balzani 7

a
1
a
2
x
1
x
2
x
3
Figure 3: Schematic illustration of cylinder-shaped inclusion.
non-vanishing components of S can then be written down for isotropic materials by
S
1111
=
1
2(1 )
_
a
2
2
+ 2a
1
a
2
(a
1
+ a
2
)
2
+ (1 2)
a
2
a
1
+ a
2
_
S
2222
=
1
2(1 )
_
a
2
1
+ 2a
1
a
2
(a
1
+ a
2
)
2
+ (1 2)
a
1
a
1
+ a
2
_
S
1122
=
1
2(1 )
_
a
2
2
(a
1
+ a
2
)
2
(1 2)
a
2
a
1
+ a
2
_
S
2211
=
1
2(1 )
_
a
2
1
(a
1
+ a
2
)
2
(1 2)
a
1
a
1
+ a
2
_
S
1212
=
1
2(1 )
_
a
2
1
+ a
2
2
2(a
1
+ a
2
)
2
+
1 2
2
_
S
1133
=

2(1 )
_
2a
2
a
1
+ a
2
_
S
2233
=

2(1 )
_
2a
1
a
1
+ a
2
_
S
1313
=
a
2
2(a
1
+ a
2
)
S
2323
=
a
1
2(a
1
+ a
2
)
.
(12)
Eective Properties, summer term 2011, c _ Daniel Balzani 8
Example:
Consider a cylinder-shaped inclusion with the semi-axis a
1
= 10 and a
2
= 5 length units
and the Poissons ratio = 0.2. Then apply heating to the inclusion of with a thermal
expansion coecient of k

in the inclusion and compute the total normal strains into the
long semi-axis
11
. Assume that only strains in the x
1
-x
2
-plane dier from zero.
The resulting coecients of the eigenstrain tensor in the inclusion are then computed by

t
ij
= k

ij
. (13)
By taking into account the general relationship
ij
= S
ijkl

t
kl
the total strains can be
computed by

11
= S
1111

t
11
+S
1122

t
22
+S
1133

t
33
= (S
1111
+S
1122
+S
1133
) k

. (14)
With 1/(2 2) = 0.625 we evaluate the three coecients of the Eshelby tensor
S
1111
= 0.625
_
5
2
+ 2 10 5
(10 + 5)
2
+ (1 2 0.2)
5
10 + 5
_
= 0.472
S
1122
= 0.625
_
5
2
(10 + 5)
2
(1 2 0.2)
5
10 + 5
_
= 0.0556
S
1133
= 0.625 0.2
2 5
10 + 5
= 0.0833 ,
(15)
and obtain therewith the result

11
= (0.472 0.0556 + 0.0833) k

= 0.4997 k

. (16)
Let us now consider a spherical inclusion where a
i
= 1, then we obtain a geometrical
isotropy and the dependency of the orientation in the x
i
-coordinate system vanishes. In
this case the Eshelby tensor reduces to
S = 1 1 + (I
1
3
1 1) or S
ijkl
=
1
3

ij

kl
+ (I
ijkl

1
3

ij

kl
) , (17)
wherein we have the abbreviations
=
1 +
3(1 )
and =
2(4 5)
15(1 )
. (18)
Please note, that 1 denotes the second-order identity tensor. The interpretation of these
two parameters becomes clear when taking into account the volumetric- and deviatoric
split which ends up in

ii
=
t
ii
and dev[] = dev[
t
] . (19)
Eective Properties, summer term 2011, c _ Daniel Balzani 9
If we apply a thermal expansion in this spherical inclusion, we are able to provide ana-
lytical formulae in polar-coordinates for the strains in the inclusion

r
=

=
1 +
3(1 )
k

(20)
and outside of the inclusion

r
= 2
1 +
3(1 )
_
a
r
_
3
k

and

=
1 +
3(1 )
_
a
r
_
3
k

. (21)
2.2 Inhomogeneities
With contrast to the case of eigenstrains we concentrate now on micro-heterogeneous
materials. This means that the material properties vary with the position in the body.
2.2.1 Concept of Equivalent Eigenstrains Main idea: nd an equivalent eigen-
strain in a homogeneous replacement material which represents the inhomogeneity, in
order to be able to apply the Eshelby results.
The body V with the elasticity tensor C := C(x) is considered, where an external dis-
placement u is applied at the boundary V , cf. Fig. 4.
a) b) c) d)
Figure 4: a) Heterogeneous material, b) homogeneous replacement material, c) equivalent
eigenstrains and d) homogenized initial problem, taken from [4].
By neglecting body forces the boundary value problem is described by
div[] = 0 with u[
V
= u, (22)
wherein the general constitutive law for elasticity is considered
= C(x) : . (23)
Eective Properties, summer term 2011, c _ Daniel Balzani 10
Now we take into account the same physical body with the same boundary conditions
made of a homogeneous replacement material with the constant elastic properties C
0
,
cf. Fig. 4b. Then we are able to write down the boundary value problem
div[
0
] = 0 with u
0
[
V
= u, (24)
where we consider the constitutive law

0
= C
0
:
0
. (25)
If we compute the dierences of the kinematic quantities
u = u u
0
and =
0
, (26)
then we are also able to compute the dierence of the stresses by
=
0
= C(x) : C
0
: ( )
= C
0
: C
0
: +C(x) :
= C
0
: + [C(x) C
0
] :
= C
0
: + (C
0
)
1
[C(x) C
0
] : .
(27)
For the case of Fig. 4c, where we focus on the dierences of the elds, we write down the
associated boundary value problem
div[ ] = 0 with u[
V
= 0 . (28)
Therein, the constitutive law reads
= C
0
:

(29)
with the equivalent eigenstrains

= (C
0
)
1
: [C(x) C
0
] : . (30)
Due to the formal analogy with (7
2
) we notice that in this case the boundary value problem
describes a homogeneous material with the elasticity tensor C
0
, where an eigenstrain eld

is applied and no diplacements occur at the boundary V . Therewith the complex

heterogeneous problem shown in Fig. 4a can be reduced to the less complex problem
depicted in Fig. 4d, where we only treat a homogeneous material with a distribution
of eigenstrains. Now we are generally able to use the Eshelby results also for material
inhomogeneities. The in Equation (30) included quantity
(x) = [C(x) C
0
] : (31)
Eective Properties, summer term 2011, c _ Daniel Balzani 11
is referred to as stress polarization. It describes the deviation of the true stresses =
C : from the stresses that would result from the true strains applied only to the
homogeneous replacement material.
If a true eigenstrain eld
t
exists in addition to the material inhomogeneities, then the
above described proceeding leads to the equivalent eigenstrains

= (C
0
)
1
: [(C(x) C
0
) : C(x) :
t
] . (32)
2.2.2 Ellipsoidal Inhomogeneities As an important special case we consider an el-
lipsoidal inhomogeneity in an innite matrix. The elasticity tensors of the inhomogeneity
and the matrix are denoted by C
I
and C
M
. We chose the matrix material as a replacement
material and obtain C
0
= C
M
. Here, we consider a given strain eld
0
= const acting at
an innite distance from the inhomogeneity. Now we use = +
0
and the nding from
Equation (30), then we obtain the equivalent eigenstrains

(x) = C
1
M
: [C
I
C
M
] : ( (x) +
0
) . (33)
We know that the equivalent eigenstrains should be zero

= 0 outside of the inhomo-

geneity. Thus, by using the Eshelby result we nd that
= S :

= const . (34)
We insert (34) into (33) and solve the resulting equation with respect to the equivalent
eigenstrains; then we obtain the expression

= [S + (C
I
C
M
)
1
: C
M
]
1
:
0
in . (35)
If we take again into account =
0
+ , then we are able to compute the total strains in
the inhomogeneity by
= A

I
:
0
= const . (36)
Herein, the so-called inuence tensor is identied by
A

I
= [I +S : C
1
M
: (C
I
C
M
)]
1
(37)
and describes the relation between the strains in and the external strain eld
0
.
Therewith, and by using the relation

0
= C
1
M
:
0
(38)
we are able to compute the stresses
= C
I
:
= C
I
: A

I
:
0
= C
I
: A

I
: C
1
M
:
0
,
(39)
Eective Properties, summer term 2011, c _ Daniel Balzani 12
as a function of an external stress eld acting at an innite distance from the inhomo-
geneity. Please note, that the stresses are also constant in .
Example:
As an example we focus on a spherical isotropic inhomogeneity, which is located in an
isotropic matrix and concentrate on the hydrostatic part. This means that for = 1/3
we only need to take into account
C
I

1111
and C
M

1111
S
1111
= =
2
3
, see Eq. (18) ,
wherein K
I
and K
M
denote the compression moduli for the inhomogeneity and the matrix.
Then we are able to compute the stresses

ii
= C
I

1111
A

I

1111
C
1
M

1111
,
0
ii
= 3 K
I
_
1 +
3 K
I
3 K
M
3 K
M
_
1
1
3 K
M

0
ii
=
K
I
K
M
_
2
3
K
I
K
M
+
1
3
_
1

0
ii
=
K
I
K
M
3
2
_
K
I
K
M
+
2
9
_
1

0
ii
.
(40)
If we analyze the case of a hard inhomogeneity, i.e. K
I
K
M
, then we are able to
neglect the constant 2/9 and obtain for the hydrostatic stresses

ii

3
2
K
I
K
M
K
M
K
I

0
ii
= 1.5
0
ii
. (41)
In case of a soft inhomogeneity, i.e. K
I
K
M
, we nd that the hydrostatic stresses
ii
are much smaller than the applied hydrostatic stresses
0
ii

ii

K
I
K
M
3
2
9
2

0
ii

0
ii
. (42)
Eective Properties, summer term 2011, c _ Daniel Balzani 13
2.2.3 Cavities and Cracks A special case of inhomogeneities are given by cavaties
and cracks since there the elasticity can be interpreted as zero. For a graphical illustration
of dierent types of cavaties and cracks see Fig. 5.
a) b) c)
Figure 5: a) Voids, b) straight cracks and c) penny-shaped crack, taken from [4]).
Voids in 2D
If we apply an external stress eld
0
at an innite distance of a spherically-shaped void
of radius a embedded in a plate of innite size (Fig. 5a), the displacements at the void
boundary (r = a) can be computed by
u
r
(a, ) =
a
E
_

0
11
(3cos
2
sin
2
) +
0
22
(3sin
2
cos
2
) + 8
0
12
sincos

(43)
for the radial part and for the circumferential part by
u

(a, ) = 4
a
E
_

0
11
sincos +
0
22
sincos +
0
12
(cos
2
sin
2
)

. (44)
Straight Crack in 2D
Let us consider a straight crack of length 2a which is located in a plate of innite size and
apply an external stress at an innite distance
0
under plain strain conditions (Fig. 5)b,
then we observe a jump of the displacements u. These can be computed by
u
i
(x
1
) =
4
0
i2
E
_
a
2
x
2
1
for i = 1, 2 . (45)
Penny-Shaped Crack in 3D
The displacement jumps in the x
1
x
2
-plane of a spherically shaped crack of radius a
whose normal coincides with the x
3
-coordinate axis (Fig. 5c) can be described by
u
i
(r) =
16(1
2
)
E(2 )

0
i3

a
2
r
2
for i = 1, 2 . (46)
Eective Properties, summer term 2011, c _ Daniel Balzani 14
The jump in x
3
-direction can be computed by
u
3
(r) =
8(1
2
)
E

0
33

a
2
r
2
, (47)
wherein we used the abbreviation r =
_
x
2
1
+ x
2
2
.
Eective Properties, summer term 2011, c _ Daniel Balzani 15
3 Theoretical Concepts with Respect to Homogenization
Figure 6: The idea of homogenization, taken from [4].
3.1 Representative Volume Element (RVE)
Considering micro-heterogeneous materials the continuum mechanical properties at the
macroscale are characterized by the geometry and by the properties of the particular con-
stituents at the microscale. For the description of such materials we apply the concept
of representative volume elements (RVE), see e.g. [7] and [6]. We dene a partial volume
of the material, which is macroscopically considered to be statistically homogeneous, as
the representative volume element. We refer a partial volume of a microstucture to as
statistically homogeneous, if each arbitrary section of the microstructure with equal di-
mensions as the partial volume leads to the same macroscopic quantities. This induces
that the choice of an RVE is not unique and it should be noted that some geometric
structures are more applicable for the implementation than others. In general an RVE
at the microscale has a complex structure, which is characterized by a large amount
of micro-heterogeneities, as e.g. inclusions, phase interfaces between the particular con-
stituents, cracks or cavities. An important requirement for the application of the concept
of representative volume elements is the existence of two length scales: the length scale
of the macrostructure, which denes the innitesimal vicinity, and the length scale of the
microstructure, which is characterized by the smallest signicant dimension of the micro-
heterogeneities. Denoting a typical characteristic length at the macroscale by L and at
the microscale by L, then we require L/l 1 to hold. Therefore, the ratio of lengthscales
Eective Properties, summer term 2011, c _ Daniel Balzani 16
is important and not absolute values of the lengthscales.
In order to obtain a volume element, which has a representative character, the represen-
tative volume element (RVE) should be much larger than the characteristic size of an
inhomogeneity. Otherwise it may happen, that the RVE consisted only of the material
of the matrix or of the inhomogeneity, which would obviously lead to unreasonable ef-
fective properties. Therefore, we conclude that the RVE should satisfy the geometrical
requirement
l d L. (48)
There exist various denitions for a RVE, which show that a unique mathematically exact
denition seems not to be possible. Some important denitions are provided below:
Hill (1963): Overall moduli have to be independent of the surface value of traction
and displacement, so long as these values are macroscopically uniform
Hashin (1983): RVE should be large enough to contain sucient microstructural
information, but much smaller than macroscopic body
Drugan and Willis (1996): RVE is smallest volume element for which the overall
eective modulus is suciently accurate to represent the mean constitutive response
Ostoja-Starzewski (2001): RVE is i) unit cell of periodic microstructure, ii)
volume possessing statistically homogeneous and ergodic properties
Stroeven, Askes, and Sluis (2002): Determination of RVE size is not straight-
forward! It depends on the material under consideration and on the structure sen-
sitivity of the physical quantity that is measured
3.2 Concept of an Ensemble
Another approach for the description of the macroscopic material behavior of micro-
heterogeneous materials is based on statistic considerations of the microstructure, see
the basic literature in e.g. [1] and [9]. Let o be the collection of samples of random
microstructures and p the probability density of an individual sample in o, then the
ensemble average of a material response F is
F(x) =
_
S
F(x, )p() d.
If the number of samples is suciently high, then the material response can be interpreted
as representative at the macroscale and thus as a suitable eective property F

= F.
Eective Properties, summer term 2011, c _ Daniel Balzani 17
3.3 Ergodic Hypothesis
The transition from ensemble average computations to the above dened representative
volume elements is enabled by further physical assumptions strongly associated with the
ergodic hypothesis. In this contex the ensemble average is replaced by simple volumetric
averages over one RVE. Basically, the ergodic hypothesis can be rephrased in words by
All states available to an ensemble are also available to every sample in the ensemble.
The main message behind is that if one considers only one particular sample which is
suciently large and computes the volume average of the material response
F(x, ) =
1
[V [
_
V
F(x +y, ) dy ,
then we are independent of the particular sample. This means that the volume average is
identical to the ensemble average
F(x, ) = F(x, ) for [V [
and a suitable eective material response is obtained by computing the volumetric average
F

= F of a suciently large volume.

Remark: If periodic composites are considered this is automatically satised for the peri-
odic unitcell Y
lim
|V |
1
[V [
_
V
F(x +y, ) dy =
1
[Y [
_
Y
F(x +y, ) dy .
3.4 Statistical Homogeneity
A material is statistically homogeneous, if the ensemble average of a material response
F(x
1
, ...x
n
) is invariant with respect to translation
F(x
1
, ...x
n
) = F(x
1
y, ...x
n
y) for arbitrary y
and if the translation is y = x
1
, then we obtain the alternative representation
F(x
1
, ...x
n
) = F(x
12
, ...x
1n
) with x
ij
= x
j
x
i
.
3.5 Statistical Isotropy
A material is statistically isotropic, if the ensemble average of a material response F
is invariant with respect to translation and rotation. In this case the ensemble average
depends only on absolute values of the vectors x
ij
F(x
12
, ...x
1n
) = F(r
ij
)
for arbitrary r
ij
= [[x
ij
[[, i = 1, ...n, j = (i + 1), ...n.
Eective Properties, summer term 2011, c _ Daniel Balzani 18
3.6 Notation
Due to the importance of the ensemble average on the eective properties of a material at
the macroscale, we denote quantities at the macroscale by ()
macro
= () if the separation
of scales is considered. Exemplarily, the stresses and strains at the macroscale are denoted
by and .
Eective Properties, summer term 2011, c _ Daniel Balzani 19
4 Homogenization of Linear Elastic Materials
4.1 Eective Stresses and Strains
It can be shown that the average of traction forces acting on the boundary of a RVE is
the eective stress at the macroscale
=
1
[V [
_
V
t xda . (49)
For the strains we obtain analogously the macroscopic eective representation
=
1
[V [
_
V
sym[u n] da . (50)
Let us take into account the property
(x
i

jk
),
k
= x
i,k

jk
+ x
i

jk,k
div[x ] = +x div (51)
and include the balance of linear momentum, then we obtain
div[x ] = +x (b u) . (52)
Neglecting body forces and acceleration terms one obtains
div[x ] = . (53)
Then an alternative representation for the eective stresses is possible by applying the
Gauss theorem
=
1
[V [
_
V
t xda
=
1
[V [
_
V
div[x ] dv
=
1
[V [
_
V
dv ,
(54)
which represents the volumetric average over the microscopic stresses. If we do not neglect
body forces and accelerations and by assuming integrability of the elds we arrive at
=
1
[V [
_
V
t xda
=
1
[V [
_
V
div[x ] dv
=
1
[V [
_
V
[ +x (b u)] dv .
(55)
Eective Properties, summer term 2011, c _ Daniel Balzani 20
For the macroscopic strains we obtain in an analogous manner
=
1
[V [
_
V
sym[u n] da
=
1
[V [
_
V
grad
sym
udv
=
1
[V [
_
V
dv ,
(56)
which represents the volumetric average of the microscopic strains. Please note that in
this representation no cavaties or cracks are taken into account.
In order to obtain a general way to derive the macroscopic stress and strain quantities,
we have to take into account cavities and cracks. For this purpose we start from a rep-
resentative microstructure V with the boundary V . Furthermore, the microstructure is
characterized by cavities with the boundary S and singular areas
S
. Such a singular area

S
splits the vicinity A
X
of a point x V into the sections A
x
+ and A
x
.
n
+
V

S
n
A
n

V
S
Figure 7: Microscopic body V with cavity and cavity boundary S and singular area
S
.
The normal n of the singular area
S
aims from A
x
into the direction of A
x
+, i.e. n =
n

= n
+
.
If the divergence theorem is applied attention has to be paid to the fact this theorem is
only applicable in sections where the considered quantity is smooth. As an example, if
singular areas exist, then the partial integration of the gradient of a vector eld y over
the volume of V yields
_
V
grady dv =
_
V
y nda
_

S
[[y]] nda , (57)
with the jump [[y]] := y
+
y

. Herein, y
+
and y

denote the thresholds from the right

and from the left of y at the singular area
S
, i.e.
y
+
:= lim
x
+
xs
y(x
+
) and y

:= lim
x

xs
y(x

) (58)
Eective Properties, summer term 2011, c _ Daniel Balzani 21
with x
+
A
x
+ and x

A
x
. For a general representation we consider rst the two
tensor elds K and G in V with
div[K] = 0 in V
[[K]]n = 0 on
S
Kn = 0 on S
G = grad[y]
_

_
, (59)
wherein we have the smooth function y and piecewise smooth G. Then we obtain the
volumetric average of KG by
KG :=
1
[V [
_
V
KGdv =
1
[V [
_
V
(Kn) y da . (60)
For the representation of the macroscopic Cauchy stress tensor we set K = and
G = grad[x] = 1 and obtain
div[] = 0 in V
[[]]n = 0 on
S
n = 0 on S
_

_
. (61)
Inserting this in (60) and taking into account the Cauchy theorem t = n we receive the
expressions for the macroscopic stresses
:= =
1
[V [
_
V
(n) xda =
1
[V [
_
V
t xda =
1
[V [
_
V
dv . (62)
Hereby, it can be seen that the macroscopic stress tensor can be either computed by the
volumetric average over V or it can also be dened by boundary tractions.
By setting K = 1 and G = =
sym
u we obtain for the average of the strain tensor
=
1
[V [
__
V
sym[u n] da +
_
S
sym[u n] da
_

S
sym[[[u]] n] da
_
. (63)
In the absence of singularities the jump of the displacement eld is
[[u]] = 0 on
S
(64)
and the last term in (63) vanishes. This means that by solving with respect to the
macroscopic strains are given by
:=
1
[V [
_
V
sym[u n] da =
1
[V [
_
S
sym[u n] da . (65)
Eective Properties, summer term 2011, c _ Daniel Balzani 22
Table 1: Denition of macroscopic strains and stresses.
Strain tensor =
1
[V [
__
V
dv
_
S
sym[u n] da
_
=
1
[V [
_
V
sym[u n] da
Stress tensor =
1
[V [
_
V
dv
=
1
[V [
_
V
t xda
Therefore, is only dened as the volumetric average over the representative microstruc-
ture V if no cavities and cracks occur. Thus, we conclude that for the general case the
macroscopic quantities are not governed by volumetric averaging. In Table 1 the important
macroscopic quantities are summarized.
In many cases the volume V consists of n partial volumes V

( = 1...n) with the volume

fractions
c

= V

/V and
n

=1
c

= 1 , (66)
wherein the elastic constant material properties C

are observed. In this case the mi-

crostructure is said to consist of discrete phases and the macroscopic stresses and strains
can be computed by
= =
n

=1
c

and = =
n

=1
c

. (67)
Herein, the phase averages are given by

=
1
[V

[
_
V
dv and

=
1
[V

[
_
V
dv . (68)
Inside the discrete phases we have the relations

= C

in V

, (69)
since the elastic properties do not vary within one discrete phase.
Eective Properties, summer term 2011, c _ Daniel Balzani 23
4.2 Eective Elasticity Tensor and Hill Condition
The eective coecients of the elasticity tensor are dened by
= = C(x) : (x) = C : . (70)
The Hill condition (1963) states that the average microscopic strain energy density should
be equal to the macroscopic strain energy density
= or (C : ) =
_
C :
_
. (71)
Now we consider the uctuations of the stresses and strains
= and = . (72)
By inserting the relations for the macroscopic stress and strain quantities we nd that
the volumetric averages of the uctuations have to vanish. For this purpose we focus on
the calculation rule for averages X +Y = X +Y and obtain
= = = 0
= = = 0 .
(73)
It is emphasized that this holds for the case when the macroscopic strains can be rep-
resented by the volumetric averages of the microscopic strains, which is considered here.
Then we obtain an additional relation for the volumetric average of the uctuation strain
energy density
=
_
( + ) ( + )
_
= + + +
= + .
(74)
By inserting the Hill condition (71) we nd that
= 0 . (75)
By using the Gauss theorem and the equilibrium equation div = 0 we are able to
reformulate the Hill condition in terms of quantities that are expressed at the boundary
of the RVE
1
[V [
_
V
(u x)
. .

w
(t n)
. .

t
da = 0 (76)
with the uctuations of the displacements w and the uctiations of the tractions

t. In this
form we observe that the uctiation terms vanish energetically at the boundary and have
therefore no inuence. This means that the Hill condition can be interpreted in the sense
that the uctuating elds at the boundary of a heterogeneous material are energetically
equivalent to their volumetric averages. As already mentioned, this can of course only be
expected if the considered RVE is suciently large.
Eective Properties, summer term 2011, c _ Daniel Balzani 24
4.3 Average Strain- and Average Stress Theorem
For the computation of the locally distributed stress and strain elds (x) and (x) in
a given volume V at the microscale, we have to solve the microscopic boundary value
problem
div[] = 0 , (77)
with suitable boundary conditions. The main goal is to replace the real heterogeneous
volume by a homogeneous (eective) material which represents a point at the macroscale
and which only notices homogeneous strains and stresses. Enabled by the Hill-condition
we apply homogeneous strain states
0
or stress states
0
at the boundary and end up in
two dierent types of boundary conditions:
a) Linear displacements: rst we apply a uniform strain eld at the boundary which
then leads to linear displacements at the boundary, i.e.
u =
0
x on V with
0
= const . (78)
By applying the Gauss-theorem we obtain the property
_
V
x nda =
_
V
gradxdv = [V [1 (79)
and nd that the macroscopic strain is equal to the homogeneous strain at the
boundary, which is constant over the surface, i.e.
=
1
[V [
_
V
u nda =
1
[V [
_
V
(
0
x) nda =
1
[V [

0
[V [ 1 =
0
. (80)
b) Uniform stresses: second, a uniform stress eld is applied at the boundary and
we obtain
t =
0
n on V with
0
= const . (81)
Here, we nd analogously that the macroscopic stress is equal to the homogeneous
stress at the boundary
=
1
[V [
_
V
t xda =
1
[V [
_
V
(
0
n) xda =
1
[V [

0
[V [ 1
T
=
0
. (82)
Since for many cases the macroscopic strains and stresses are given by the volumetric
averages, the relations (80) and (82) are referred to as average strain theorem and
average stress theorem. Taking a look on (76) again we notice that the Hill criterion
is satised independently by each of the boundary conditions a) and b). In addition, we
Eective Properties, summer term 2011, c _ Daniel Balzani 25
conclude that therefore the Hill condition can be generalized to the case of independent
stress
(1)
and strain elds
(2)

(1)

(2)
=
(1)

(2)
. (83)
Due to the fact that in the case of linear elasticity the solutions of the boundary value
problems are unique and independent from the history, the total strains and stresses can
be computed by
a) (x) = L

(x) :
0
for u =
0
x on V
b) (x) = L

(x) :
0
for t =
0
n on V .
(84)
Herein, the localization or inuence tensors L

and L

represent the complete solution

of the boundary value problem and depend on the microstructure in the whole volume
V . By taking into account the average strain theorem and computing the volume average
on both sides of (84
1
) we nd that
(x) = L

(x) :
0

0
= L

:
0

= I .
(85)
Analogously, we obtain for the stresses when focussing on the average stress theorem
L

= I . (86)
For the eective elasticity tensor the relation
C : = = = C : , (87)
holds, thus, we are able to transform this equation by using relation (84
1
) to
C : = C : L

:
0
= C : L

: . (88)
This leads to an expression for the eective elasticity tensor as a result of boundary
condition a)
C
(a)
= C : L

. (89)
By applying boundary condition b) we transform analogously
C
1
: = C
1
: L

:
0
= C
1
: L

: , (90)
and the associated eective elasticity tensor is given by
C
(b)
= C
1
: L

1
. (91)
Eective Properties, summer term 2011, c _ Daniel Balzani 26
If we insert these results into the representation of the Hill condition (71), then we obtain
(C : ) =
_
C :
_
(L

:
0
) (C : L

:
0
) =
0
(C :
0
)
C
(a)
= (L

)
T
: C : L

(92)
and analogously for the boundary conditions b)
C
(b)
= (L

)
T
: C
1
: L

1
. (93)
From these expressions we directly notice the symmetries of the eective elasticity tensors
with respect to the rst and second pair of indices
C
(a)
ijkl
= C
(a)
klij
and C
(b)
ijkl
= C
(b)
klij
. (94)
It should be noted that the two dierent eective moduli C
(a)
and C
(b)
can be generally
computed for arbitrary heterogeneous volumes, although they depend on the type of
boundary condition. Therefore, strictly speaking, these elasticity moduli are no eective
properties since they can be also computed for a volume which does not satisfy some
conditions dening a reasonable RVE. Such a denition could be that if
C
(a)
= C
(b)
= C (95)
holds for a considered volume V , then this volume can be interpreted as a RVE. In this
case C characterizes the unique eective elastic properties even if larger volumes are
considered that contain V .
Eective Properties, summer term 2011, c _ Daniel Balzani 27
References
[1] M. Beran. Statistical continuum theories. Wiley, 1968.
[2] D.C. Drucker. Material response and continuum relations; or from microscales to
macroscales. Journal of Engineering Materials and Technology, 106:286289, 1984.
[3] M.G.D. Geers, V. Kouznetsova, and W.A.M. Brekelmans. Multi-scale rst-order and
second-order computational homogenization of microstructures towards continua. In-
ternational Journal for Multiscale Computational Engineering, 1:371386, 2003.
[4] D. Gross and T. Seelig. Fracture Mechanics. Springer, 2006.
[5] S. Hao, B. Moran, W.K. Liu, and G.B. Olson. A hierarchical multi-physics model
for design of high toughness steels. Journal of Computer-Aided Materials Design,
10:99142, 2003.
[6] Z. Hashin. Analysis of composite materials - a survey. Journal of Applied Mechanics,
50:481505, 1983.
[7] R. Hill. Elastic properties of reinforced solids: some theoretical principles. Journal
of the Mechanics and Physics of Solids, 11:357372, 1963.
[8] A. Ibrahimbegovic, D. Markovic, H.G. Matthies, R. Niekamp, and R.L. Taylor. Multi-
scale modelling of heterogeneous structures with inelastic constitutive behavior. In
D.R.J. Owen, E. Onate, and B. Suarez, editors, Computational Plasticity, COMPLAS
VIII. CIMNE, Barcelona, 2005.
[9] E. Kroner. Statistical continuum mechanics. In CISM Courses and Lectures, vol-
ume 92. Springer-Verlag, Wien, New-York, 1971.
[10] H. Kuramae, K. Okada, M. Yamamoto, M. Tsuchimura, H. Sakamoto, and E. Nakam-
chi. Parallel performance evaluation of multi-scale nite element analysis based on
crystallographic homogenization method. In D.R.J. Owen, E. Onate, and B. Suarez,
editors, Computational Plasticity / COMPLAS VIII. CIMNE, Barcelona, 2005.
[11] D.J. Littlewood and A.M. Maniatty. Multiscale modeling of crystal plasticity in
al 7075-t651. In D.R.J. Owen, E. Onate, and B. Suarez, editors, Computational
Plasticity, COMPLAS VIII. CIMNE, Barcelona, 2005.
[12] C. Miehe, J. Schotte, and J. Schr oder. Computational micro-macro-transitions and
overall moduli in the analysis of polycrystals at large strains. Computational Mate-
rials Science, 16:372382, 1999.
Eective Properties, summer term 2011, c _ Daniel Balzani 28
[13] C. Miehe, J. Schr oder, and C. G. Bayreuther. Homogenization analysis of linear com-
posite materials on discretized uctuations on the micro-structure. Acta Mechanica,
155:116, 2002.
[14] T. Mura. Micromechanics of Defects in Solids. Martinus Nijho Publishers, Dor-
drecht, 1982.
[15] E. Nakamachi, H. Kuramae, Y. Nakamura, and H. Sakamoto. Crystallographic ho-
mogenization elastoplastic nite element analyses of polycrystal sheet material. In
D.R.J. Owen, E. Onate, and B. Suarez, editors, Computational Plasticity / COM-
PLAS VIII. CIMNE, Barcelona, 2005.
[16] R.J.M. Smit, W.A.M. Brekelmans, and H.E.H. Meijer. Prediction of the mechanical
behaviour of nonlinear heterogeneous systems by multi-level nite element modelling.
Computer Methods in Applied Mechanics and Engineering, 155:181192, 1998.
[17] T.I. Zohdi and P. Wriggers. Introduction to Computational Micromechanics.
Springer, 2005.